Syntax: 2. Halbband 9783110203301, 9783110142631

Citation preview

Syntax HSK 9.2

Handbücher zur Sprach- und Kommunikationswissenschaft Handbooks of Linguistics and Communication Science Manuels de linguistique et des sciences de communication Mitbegründet von Gerold Ungeheuer

Herausgegeben von / Edited by / Edités par Hugo Steger Herbert Ernst Wiegand

Band 9.2

Walter de Gruyter · Berlin · New York 1989

Syntax Ein internationales Handbuch zeitgenössischer Forschung An International Handbook of Contemporary Research Herausgegeben von / Edited by Joachim Jacobs · Arnim von Stechow Wolfgang Sternefeld · Theo Vennemann 2. Halbband / Volume 2

Walter de Gruyter · Berlin · New York 1995

Gedruckt auf säurefreiem Papier, das die US-ANSI-Norm über Haltbarkeit erfüllt.

Die Deutsche Bibliothek — CIP-Einheitsaufnahme Handbücher zur Sprach- und Kommunikationswissenschaft / mitbegr. von Gerold Ungeheuer. Hrsg. von Hugo Steger; Herbert Ernst Wiegand. — Berlin; New York: de Gruyter. Teilw. mit Parallelt.: Handbooks of linguistics and communication science. — Teilw. mit Nebent.: HSK. — Früher hrsg. von Gerold Ungeheuer und Herbert Ernst Wiegand NE: Ungeheuer, Gerold [Begr.]; Steger, Hugo [Hrsg.]; PT; NT Bd. 9. Syntax: ein internationales Handbuch zeitgenössischer Forschung / hrsg. von Joachim Jacobs ... Halbbd. 2 (1995) ISBN 3-11-14263-5 NE: Jacobs, Joachim [Hrsg.]

© Copyright 1995 by Walter de Gruyter & Co., D-10785 Berlin. Dieses Werk einschließlich aller seiner Teile ist urheberrechtlich geschützt. Jede Verwertung außerhalb der engen Grenzen des Urheberrechtsgesetzes ist ohne Zustimmung des Verlages unzulässig und strafbar. Das gilt insbesondere für Vervielfältigungen, Übersetzungen, Mikroverfilmungen und die Einspeicherung und Verarbeitung in elektronischen Systemen. Printed in Germany Satz und Druck: Arthur Collignon GmbH, Berlin Buchbinderische Verarbeitung: Lüderitz & Bauer, Berlin

Vorwort zum zweiten Band Der hiermit vorliegende zweite Halbband des Handbuchs Syntax konzentriert sich auf die die Syntax betreffenden Teile der Typologie und der Sprachgeschichtsforschung sowie auf das Verhä ltnis der Syntax zu einigen ihrer Nachbardisziplinen. Außerdem wird der Satzbau einer Reihe von Sprachen skizziert, die unter verschiedenen Gesichtspunkten interessant erscheinen. (Nä here Erlä uterungen im Vorwort zum ersten Band.) Den Abschluß bilden ein Personen-, ein Sach- und ein Sprachenregister für beide Halbbände. Zwei Hinweise für die Benutzung der Register: (a) Aus technischen Gründen konnten im Personenregister Namensmehrdeutigkeiten nicht aufgelöst werden; gelegentlich beziehen sich also verschiedene Seitenangaben unter demselben (Familien-)Namen auf verschiedene Personen, was sich in der Regel durch Nachschlagen an den entsprechenden Textstellen oder in der jeweiligen Artikelbibliographie klären läßt. (b) Im Sachregister ist bei Verweisen, die sich auf das Thema größerer Abschnitte (etwa ganzer Kapitel) beziehen, die Seite angegeben, auf der der Abschnitt beginnt. Auf eine weitere Kennzeichnung mit „ff“ oder „passim“ wurde in solchen Fällen verzichtet. Für vielfältige Hilfe bei der Arbeit am zweiten Band haben wir (außer den schon im Vorwort zum ersten Band genannten Personen) Nürten Öztürk zu danken sowie auf Seiten des Verlags Christiane Graefe und Heike Plank. Unser besonderer Dank gilt darüber hinaus Birgit Kaiser, die das Register konzipiert und erstellt hat. Im August 1995

Joachim Jacobs Arnim von Stechow Wolfgang Sternefeld Theo Vennemann

Preface to Volume 2 This second volume of the Handbook on Syntax concentrates on the parts of typology and research on language history that deal with syntax, as well as the relationship of syntax to some related disciplines. In addition, the syntax of a number of languages is outlined that have some particularly interesting aspects. (This is explained in more detail in the preface to the first volume. ) An index of names, subjects, and languages for both volumes can be found at the back of this volume. Here are two comments to assist you in using the index: (1) for technical reasons it was not possible to clarify certain ambiguities in the name index; different pages listed for one (last) name sometimes refer to different persons that have the same last name. Any confusion can generally be eliminated by turning to the respective pages or to the article bibliography directly. (2) for references pertaining to the subject matter of extensive sections (such as an entire chapter), only the first page is listed in the subject index. No further reference was made by means of “ ff” or “ passim.” We would like to thank (in addition to those already mentioned in the preface to the first volume), Nürten Öztürk, and from the publisher, Christiane Graefe and Heike Plank, for their extensive assistance in preparing the second volume. Particular thanks are also due to Birgit Kaiser, who designed and prepared the index. August 1995

Joachim Jacobs Arnim von Stechow Wolfgang Sternefeld Theo Vennemann

VII

Inhalt/Contents 2. Halbband/Volume 2 Vorwort ................................................................................................................................... Preface .....................................................................................................................................

XVI.

Syntaktische Typologie I: Allgemeines Syntactic Typology I: General

53.

Christian Strömsdörfer/Theo Vennemann, Ziele der syntaktischen Typologie (Objectives of Syntactic Typology) ......................................................... Gustav Ineichen, Geschichte der syntaktischen Typologie (History of Syntactic Typology) ...............................................................................

54.

V VI

1031 1043

XVII. Syntaktische Typologie II: Ansätze und Übersichtsdarstellungen Syntactic Typology II: Approaches and Surveys 55. 56. 57. 58.

Matthew S. Dryer, Word Order Typology (Wortstellungstypologie) ....................... Hans-Jürgen Sasse, Prominence Typology (Prominenztypologie) .......................... Beatrice Primus, Relational Typology (Relationstypology) .................................... Dietmar Zaefferer, Comparative Surveys in Syntactic Typology: A Critical Guide (Sprachvergleichende syntaxtypologische Übersichtsdarstellungen: Ein kritischer Führer) .....................................................................

1050 1065 1076 1109

XVIII. Die Erforschung des Syntaxwandels I: Allgemeines Research into Syntactic Change I: General 59. 60.

Winfred P. Lehmann, Objectives of a Theory of Syntactic Change (Ziele einer Theorie des Syntaxwandels) ................................................................ Christian Strömsdörfer/Theo Vennemann, Das Verhältnis des Syntaxwandels zur Theorie der Sprachzustände (The Relationship of a Theory of Language Change to a Theory of Language States) ...........................................

XIX.

Die Erforschung des Syntaxwandels II: Ansätze Research into Syntactic Change II: Approaches

61.

Lyle Campbell, History of the Study of Historical Syntax (Geschichte der Erforschung der historischen Syntax) ........................................... Jürgen Lenerz, Transformationsgrammatische Ansätze (TG Approaches) .....................................................................................................

62.

1116 1126

1136 1166

VIII

63.

Inhalt/Contents

John A. Hawkins, Typology-based Research into Syntactic Change (Typologisch fundierte Erforschung des Syntaxwandels) ........................................

XX.

Die Erforschung des Syntaxwandels III: Phänomene Research into Syntactic Change III: Phenomena 64. Frans Plank, Ergativity (Ergativität) 1184

65. 66. 67. 68. 69. 70. 71. 72.

Christian Lehmann, Relativsätze (Relative Clauses) .............................................. Martin Harris, Word Order (Wortstellung) ............................................................. Susan Steele, Clisis (Klitisierung) ........................................................................... Greville G. Corbett, Agreement (Kongruenz) ......................................................... Bernard Comrie, Tense and Aspect (Tempus und Aspekt) ...................................... Christian Lehmann, Synsemantica (Synsemantika) ................................................ Jürgen Lenerz, Klammerkonstruktionen (Bracketing Constructions) ..................... Ariane von Seefranz-Montag, Impersonalien (Impersonal Constructions) ....................................................................................

XXI.

Syntaktische Skizzen Syntactic Sketches

73. 74. 75. 76. 77. 78. 79. 80. 81.

Günther Grewendorf, German (Deutsch) ................................................................ James D. McCawley, English (Englisch) ................................................................ Giuliana Giusti, Italian (Italienisch) ....................................................................... Takao Gunji, Japanese (Japanisch) ......................................................................... Alice C. Harris, Georgian (Georgisch) ................................................................... Manfred Krifka, Swahili (Swahili) .......................................................................... Paul Schachter, Tagalog (Tagalog) .......................................................................... Kenneth L. Hale/Mary Laughren/Jane Simpson, Warlpiri (Warlpiri) .................... Derek Bickerton, Creole Languages (Kreolsprachen) ............................................

1176

1199 1216 1226 1235 1244 1251 1266 1277

1288 1319 1347 1363 1377 1397 1418 1430 1451

XXII. Syntax und Nachbardisziplinen Syntax and Neighbouring Disciplines 82. 83. 84. 85. 86.

Julius M. Moravcsik, Syntax and Philosophy (Syntax und Philosophie) ........................................................................................ Josef Bayer, Syntax and Psychology (Syntax und Psychologie) ............................. Richard Hudson, Syntax and Sociolinguistics (Syntax und Soziolinguistik) .................................................................................... Hans-Werner Eroms, Syntax und Stilistik (Syntax and Stilistics) ........................... Wolfgang Sucharowski, Syntax und Sprachdidaktik (Syntax and Language Didactics) ...........................................................................

1463 1484 1514 1528 1545

XXIII. Service-Artikel Service-Article 87.

Ursula Klenk, Formale Sprachen (Formal Languages) ..........................................

1576

Inhalt/Contents

IX

Register Indexes Namenregister (Name Index) .................................................................................................. Sprachenregister (Language Index) ........................................................................................ Sachregister (Subject Index) ....................................................................................................

1607 1623 1629

1. Halbband/Volume 1 Vorwort ................................................................................................................................... Preface .....................................................................................................................................

I.

Grundlagen Foundations

1.

Arnim von Stechow, Die Aufgaben der Syntax (The Objectives of Syntax) ...................................................................................... Peter Matthews, Central Concepts of Syntax (Zentrale Begriffe der Syntax) ................................................................................. Hans-Heinrich Lieb, Syntax and Linguistic Variation: Orientation (Syntax und sprachliche Variation: Orientierung) ..................................................

2. 3.

II.

Geschichte der Syntaxforschung The History of Syntactic Research

4. 5.

Wolf Thümmel, Westliche Entwicklungen (Western Developments) ...................... Peter Raster, Die indische Grammatiktradition (The Indian Grammatical Tradition) ....................................................................... Jonathan Owens, Arabic Syntactic Theory (Arabische Syntaxforschung) ...................................................................................

6.

III.

Ansätze syntaktischer Theoriebildung I: Traditionelle Syntax Approaches to a Theory of Syntax I: Traditional Syntax

7. 8. 9.

Franz Hundsnurscher, Grundideen (Basic Ideas) ................................................... Wladimir Admoni, Funktionale Grammatik (Functional Grammar) ..................... Jürgen Dittmann, Inhaltbezogene Grammatik (“ Inhaltbezogene Grammatik”) .............................................................................

IV.

Ansätze syntaktischer Theoriebildung II: Strukturalistische Syntax Approaches to a Theory of Syntax II: Syntax in Structuralism

10.

Wolf Thümmel, Der europäische Strukturalismus (European Structuralism) ........................................................................................ Wolf Thümmel, Der amerikanische Strukturalismus (American Structuralism) ........................................................................................

11.

V IX

1 89 118

130 199 208

216 232 242

257 280

X

Inhalt/Contents

V.

Ansätze syntaktischer Theoriebildung III: Dependenzsyntax Approaches to a Theory of Syntax III: Dependency Syntax

12.

Hans Jürgen Heringer, Basic Ideas and the Classical Model (Grundideen und das klassische Modell) ................................................................ Hans Jürgen Heringer, Formalized Models (Formale Modelle) ............................. Richard A. Hudson, Recent Developments in Dependency Theory (Neuere Entwicklungen der Dependenztheorie) ......................................................

13. 14.

VI.

Ansätze syntaktischer Theoriebildung IV: Funktionale Syntax Approaches to a Theory of Syntax IV: Functional Syntax

15.

A. Machtelt Bolkestein, General Ideas of Functionalism in Syntax (Grundgedanken der funktionalen Syntax) ............................................................. Petr Sgall, The Czech Tradition (Die tschechische Tradition) ................................ Simon C. Dik, Functional Grammar (Funktionale Grammatik) .............................

16. 17.

VII.

Ansätze syntaktischer Theoriebildung V: Semantisch motivierte formale Syntax Approaches to a Theory of Syntax V: Semantically Motivated Formal Syntax

18. 19.

Mark Steedman, Categorial Grammar (Kategorialgrammatik) .............................. Helmut Frosch, Montague-Grammatik (Montague Grammar) ...............................

VIII.

Ansätze syntaktischer Theoriebildung VI: Syntax in der Integrativen Sprachwissenschaft Approaches to a Theory of Syntax VI: Syntax in Integrational Linguistics

20.

Hans-Heinrich Lieb, Integrational Linguistics (Integrative Sprachwissenschaft) .............................................................................

IX.

Ansätze syntaktischer Theoriebildung VII: Syntax in der Generativen Grammatik Approaches to a Theory of Syntax VII: Syntax in Generative Grammar

21.

Gisbert Fanselow, Die frühe Entwicklung bis zu den „Aspekten“ (Early Developments up to “ Aspects”) .................................................................. Frederick J. Newmeyer, Diverging Tendencies (Divergente Entwicklungen) .................................................................................... James McCloskey, Constraints of Syntactic Processes (Beschränkungen für syntaktische Prozesse) ........................................................... Noam Chomsky/Howard Lasnik, The Theory of Principles and Parameters (Die Theorie der Prinzipien und Parameter) ........................................

22. 23. 24.

298 316 329

339 349 368

395 413

430

469 486 496 506

Inhalt/Contents

25. 26. 27.

Robert D. Borsley, Phrase Structure Grammar (Phrasenstrukturgrammatik) ................................................................................... Tibor Kiss, Lexical-Functinal Grammar (Lexikalisch-funktionale Grammatik) ..................................................................... Wilhelm Oppenrieder, Relationale Grammatik (Relational Grammar) ..................

X.

Syntaktische Präferenztheorie Theories of Syntactic Preferences

28.

Willi Mayerthaler/Günther Fliedl, Natürlichkeitstheoretische Syntax (Syntactic Naturalness) ........................................................................................... Hubert Haider, „Markiertheit“ in der Generativen Grammatik (“ Markedness” in Generative Grammar) ..............................................................

29.

XI.

Syntaktische Phänomene in den Sprachen der Welt I: Kategorien und Relationen Syntactic Phenomena in the World’s Languages I: Categories and Relations

30.

Hans-Jürgen Sasse, Syntactic Categories and Subcategories (Syntaktische Kategorien und Subkategorien) ........................................................ Beatrice Primus, Syntactic Relations (Syntaktische Relationen) ............................ Edith A. Moravcsik, Government (Rektion) ............................................................ Christian Lehmann, Kongruenz (Agreement) .........................................................

31. 32. 33.

XII.

Syntaktische Phänomene in den Sprachen der Welt II: Konstruktionstypen Syntactic Phenomena in the World’s Languages II: Types of Construction

34. 35. 36. 37.

Dieter Wunderlich, Diathesen (Valency Changing) ................................................ Robert R. van Oirsouw, Coordination (Koordination) ............................................ Wolfgang Klein, Ellipse (Ellipsis) .......................................................................... Rose-Marie Déchaine, Serial Verb Constructions (Serialverb-Konstruktionen) ....................................................................................

XIII.

Syntaktische Phänomene in den Sprachen der Welt III: Wortstellung Syntactic Phenomena in the World’s Languages III: Word Order

38.

Anna Siewierska, On the Interplay of Factors in the Determination of Word Order (Das Zusammenwirken von Wortstellungsfaktoren) ............................ Brigitta Haftka, Topologische Felder und Versetzungsphänomene (Topological Fields and Dislocation) ..................................................................... Jürgen Pafel, Scope and Word Order (Skopus und Wortstellung) ...........................

39. 40.

XI

570 581 601

610 635

646 686 705 722

730 748 763 799

826 846 867

XII

41.

42.

Inhalt/Contents

Beatrice Primus, Word Order and Information Structure: A Performance Based Account of Topic Positions and Focus Positions (Wortstellung und Informationsstruktur: Eine performanzbezogene Erklärung von Topik- und Fokus-Positionen) .............................................................. John Haiman, Iconicity (Ikonizität) ........................................................................

XIV.

Syntaktische Manifestationen semantischer Bezüge Syntactic Encoding of Semantic Aspects of Meaning

43. 44. 45. 46. 47. 48. 49.

Bernard Comrie, Argument Structure (Argumentstruktur) ..................................... Östen Dahl, Negation (Negation) ............................................................................ John Haiman, Conditionals (Konditionalsätze) ...................................................... Kjell Johan Sæbø, Causality and Finality (Kausalität und Finalität) ..................... Wolfgang Sternefeld, Anaphoric Reference (Anaphorische Referenz) ................... Claudia Gerstner/Manfred Krifka, Genericity (Generizität) .................................. Ekkehard König, Focus Particles (Fokus-Partikeln) ...............................................

XV.

Syntaktische Manifestationen pragmatischer Bezüge Syntactic Encoding of Pragmatic Aspects of Meaning

50. 51. 52.

Christa Hauenschild, Definitheit (Defìniteness) ..................................................... Josef Taglicht, Focus and Background (Fokus und Hintergrund) ........................... Hans Altmann, Satzmodus (Sentence Mood) ..........................................................

880 896

905 914 923 930 940 966 978

988 998 1006

1031

XVI. Syntaktische Typologie I: Allgemeines Syntactic Typology I: General

53. Ziele der syntaktischen Typologie 1. 2. 3. 4. 5.

1.

Was ist syntaktische Typologie? Methoden und Ergebnisse der syntaktischen Typologie Zur Erklärung typologischer Parameter Zusammenfassung Literatur

Was ist syntaktische Typologie?

Unter „Typologie“ versteht man in der Wissenschaft die Ordnung eines Forschungsbereiches nach Typen. Unter einem „Typus“ versteht man in diesem Zusammenhang eine mehreren Elementen gemeinsame Merkmalskonstellation, die aus den empirischen Daten konstruiert wird. Eine Typologie erklärt diese Merkmalskonstellationen noch nicht, sie beschreibt sie nur. Andererseits setzt das Auffinden von Merkmalen bereits einen wissenschaftlichen Analyseapparat und damit theoretische Vorannahmen voraus. Typologie ist also gleichzeitig „Input“ wie auch „Output“ von Theorien. In diesem Spannungsfeld bewegt sich auch die linguistische Typologie (zum Verhältnis von Typologie und linguistischer Theorie vgl. Vennemann 1985). Im Gegensatz zur Klassifikation, die jedes Phänomen (also auch Merkmalskonstellationen) einer bestimmten Klasse zuordnet, kennt die Typologie Übergangsphänomene, d. h. es gibt zwar Kernbereiche von Typen, aber Phänomene, auf die nur ein Teil der Typus-Eigenschaften zutreffen, werden nicht sofort in andere Typen eingeordnet, sondern sie werden (je nach Grad der Verschiedenheit) als Randbereich eines Typs oder als Übergang zwischen zwei Typen verstanden (zur wissenschaftstheoretischen Unterscheidung von „Typologie“ und „Klassifizierung“ siehe z. B. Sandkühler 1990, 624 f). Wie man in sinnvoller Weise Merkmale eines zu untersuchenden Forschungsgegenstandes gewinnt, hängt vom jeweiligen Forschungsbereich ab. So ordnet man Sprachen

in der Linguistik z. B. nicht nach der Hautfarbe ihrer Sprecher, sondern nach Merkmalen, die den Sprachen selbst eigen sind, etwa nach der Zusammensetzung ihres Lautinventars oder nach der dominanten Stellung bestimmter grammatischer Elemente. Diese Merkmalskonstellationen sollten nicht einfach wahllose Zusammenstellungen unzusammenhängender Einheiten sein, sondern sie sollten aufzeigen, daß bestimmte Merkmale immer gemeinsam auftreten. Die Merkmale bilden also ein gegliedertes Ganzes, anders gesagt: eine Struktur. Es geht in der Typologie also um strukturelle Typen. Im weitesten Sinne versteht man unter “ linguistischer Typologie“ daher eine Klassifizierung linguistischer Merkmale unterschiedlicher Sprachen in strukturelle Typen. Dies schließt auch die Möglichkeit ein, ganze Sprachen oder Gruppen von Sprachen in Typen einzuteilen. Die meisten zeitgenössischen Linguisten verbinden mit Typologie vor allem sprachübergreifende Vergleiche bestimmter (vorher definierter) sprachlicher Konstruktionen, wobei die Betonung auf dem Wort „sprachübergreifend“ liegt. Um die Frage nach linguistischer (und damit auch der syntaktischen) Typologie zu klären, müssen vorher zwei andere Probleme gelöst werden. 1) Da Typologie selbst noch keine Theorie ist, muß man als Typologe klarstellen, in welchem theoretischen Beschreibungsrahmen man sich bewegt, d. h. wie man z. B. die relevanten sprachlichen Merkmale identifiziert oder welche grammatischen Elemente man überhaupt annimmt. 2 ) Im Rahmen eines solchen theoretischen Konzepts muß geklärt sein, was eine linguistische Struktur ist und wie sie zu beschreiben ist. Es existiert in der Linguistik heutzutage eine Fülle verschiedener Beschreibungs- und Theoriekonzepte, die teilweise erheblich voneinander abweichen. Ihnen allen gemeinsam ist die Frage „Wie muß eine mögliche menschliche Sprache beschaffen sein?“ .

1032

Für die meisten Linguisten, die im Rahmen der Typologie arbeiten, ist diese Teildisziplin ihrer Wissenschaft gleichbedeutend mit dem Studium sprachlicher Muster („Muster“ im Sinne von „strukturierter Merkmalskonstellation“ ), die sich nur durch den Vergleich verschiedener Sprachen finden lassen (vgl. z. B. Comrie 1990, Givón 1984, Art. Nr. 63 von Hawkins in diesem Band). Insofern wird die Typologie zur Gegenspielerin der generativen Grammatik, die ja gerade davon ausgeht, daß man alle (universal-) grammatischen Muster bereits durch intensive Erforschung einer einzelnen Sprache finden könne (vgl. z. B. Chomsky 1970, 12 5 ff). Das soll nicht heißen, daß linguistische Typologie selbst zu einer Theorie wird (sie setzt ja eben eine voraus), aber Linguisten, die im Rahmen einer so verstandenen Typologie arbeiten, fordern von einer linguistischen Theorie, daß sie in der Lage sein muß, solcherart (durch Sprachvergleich) gefundene grammatische Muster adäquat zu beschreiben und zu erklären (vgl. dazu Croft 1990, 4 ff). Es spricht jedoch im Prinzip nichts dagegen, daß eine solche Erklärung auch im Rahmen einer generativen Grammatik erfolgen kann. Wir wollen daher im folgenden vermeiden, die Typologie als methodischen oder gar theoretischen Gegenpol zur generativen Schule zu behandeln. Die beiden angesprochenen linguistischen Forschungsrichtungen lassen sich viel besser als formal (für die generative Schule) und funktional (für die „typologische“ Schule) charakterisieren. Diese beiden Termini bezeichnen dabei die Art und Weise der Erklärungsstrategien für sprachliche Phänomene, die Linguisten der einen oder anderen Richtung jeweils bevorzugen. So gibt es auch viele Funktionalisten, die ihre Ergebnisse ohne oder kaum mit sprachvergleichenden Untersuchungen gefunden haben bzw. formulieren (z. B. Heath 1984), ebenso wie Typologen, die ihre komparativ gefundenen Ergebnisse im Rahmen generativer linguistischer Theorien formulieren (z. B. in Hawkins 1983). Viele Typologen betonen ihrerseits auch, daß ihre Forschungen „neutral“ hinsichtlich linguistischer Theoriemodelle seien (z. B. Stassen 1985, 21 und Mallinson/Blake 1981, 35 f). Typologie als linguistische Disziplin im zuletzt genannten Sinne, d. h. als komparative (= sprachvergleichende) Methode zur Auffindung grammatischer Muster ist eine verhältnismäßig junge Erscheinung, deren Grundla-

XVI. Syntaktische Typologie I: Allgemeines

gen von Greenberg 1960 (siehe Greenberg 1966) durch seine Arbeit über implikative Universalien (zur Explikation dieses Begriffes siehe 2 .1.1.) in der Morphologie und Wortstellung gelegt wurden. Greenberg begründete damit nicht nur die moderne linguistische Typologie überhaupt, sondern auch speziell die moderne syntaktische Typologie, da sich seine implikativen Universalien auch mit der Wortstellung von (und in) Sätzen beschäftigten. Diese “ Wortstellungstypologie“ sollte die syntaktische Typologie bis in die siebziger Jahre dominieren (siehe dazu 2.2.). Linguistische Typologie gab es selbstverständlich lange vor Greenberg. Die meisten Bücher über die Geschichte der Linguistik lassen die Typologie mit Friedrich Schlegel beginnen (z. B. Robins 1979, 176), der die Sprachen der Welt in zwei Typen unterteilte: 1) Sprachen, die grammatischen Gebrauch von Veränderungen im Wortinneren machen, und 2 ) Sprachen, die nur linear geordnete Elemente aneinanderhängen (F. Schlegel 1808). August Wilhelm Schlegel fügte dem einen weiteren Typ von Sprachen hinzu: 3) Sprachen „ohne Grammatik“ , d. h. ohne morphologische Affixe (A. Schlegel 1818). Schließlich fand Wilhelm von Humboldt durch seine Forschungen auf dem Gebiet der Indianersprachen noch einen weiteren Typ (Humboldt 1836): 4) einverleibende Sprachen. Diese vier Typen werden heute mit den Begriffen 1) ‘flektierend’, 2 ) ‘agglutinierend’, 3) ‘isolierend’ und 4) ‘inkorporierend’ (manchmal auch ‘polysynthetisch’) bezeichnet. Obwohl Typologie in dieser Form erst im 19. Jahrhundert entstand, reichen ihre Ursprünge viel tiefer (vgl. dazu z. B. den Artikel Nr. 61 von Lyle Campbell in diesem Band). Diese morphologisch motivierte Typologie, wie auch alle anderen Typologien vor Greenberg, unterscheiden sich von letzterer vor allem dadurch, daß immer eine ganze Sprache als Repräsentant eines Typs angesehen wird. Es handelt sich also um Typologien von Sprachen und nicht um Typologien sprachlicher Strukturen. Für eine Weiterentwicklung der morphologisch motivierten Typologie in moderner Sichtweise vgl. z. B. Vennemann 1982 b. Doch trotz dieses Unterschieds sind viele Fragen der modernen Typologie auch früher gestellt worden und haben Antworten gefunden, die noch heute von großer Bedeutung sind. Im Bereich der syntaktischen Typologie sind hier z. B. die fünf Behaghelschen Gesetze zu nennen, die Prinzipien formulieren,

53.  Ziele der syntaktischen Typologie

nach denen sich die Syntax einer Sprache organisiert. Die Typologie muß auch deutlich von einer anderen Art der Klassifizierung in der Linguistik unterschieden werden: der Ordnung von Sprachen in verschiedene Gruppen aufgrund gemeinsamer ‘Ursprachen’. Diese Klassifizierung wird häufig als ‘genetisch’ oder ‘genealogisch’ bezeichnet. In diesem Zusammenhang stellt sich vor allem die Frage, inwieweit die beiden Disziplinen voneinander unabhängig sind, oder anders gesagt: ob die eine Disziplin nicht einfach eine Teildisziplin der anderen ist. Methodisch lassen sich die beiden nicht klar unterscheiden. Die ‘genetisch’ orientierten Linguisten arbeiten (soweit möglich) vornehmlich diachron, d. h. sie untersuchen die Veränderung einer oder mehrerer Sprachen im Laufe der Zeit (Dies ist nicht immer möglich. Zur Methodik genealogischer Linguistik aufgrund zumeist synchroner Daten vgl. Greenberg 1987). Insbesondere vergleichen sie also verschiedene Sprachen miteinander. Typologen arbeiten zwar in erster Linie ‘synchron’, d. h. sie untersuchen eine bestimmte Sprache nur für einen bestimmten Zeitabschnitt, aber im Grunde genommen macht es für die Typologie methodisch keinen Unterschied, ob man nun Althochdeutsch und Neuhochdeutsch oder Neuhochdeutsch und Nordfriesisch vergleicht (vgl. z. B. Croft 1990, 2 03 ff und unseren Artikel über Syntaxwandel in diesem Band). In beiden linguistischen Disziplinen wird also zuvorderst ‘komparativ’ gearbeitet. Die Frage nach der Unabhängigkeit von typologischer und genetischer oder diachroner Sprachwissenschaft läßt sich aber trotzdem uneingeschränkt mit ‘ja’ beantworten. So ist etwa das südsemitische Ge’ez eine VSO-Sprache. Unter den Tochtersprachen nun finden sich Sprachen mit VSO-Wortstellung (Tigre) wie SOV-Wortstellung (u. a. Amharisch & Harare) (zur Wortfolgetypologie vgl. 2 .2 .). Sprachen, die genetisch zu einer Gruppe zusammengefaßt werden, können also unterschiedliche linguistische Typen repräsentieren. Die modernen keltischen Sprachen dagegen, die zur indoeuropäischen Gruppe gehören, stehen in syntaktischer Hinsicht den altsemitischen Sprachen typologisch wesentlich näher als irgendeiner indoeuropäischen Sprache (VSO-Wortstellung). Sowohl die keltischen als auch die (alt-) semitischen Sprachen sind wiederum syntaktisch spiegelbildlich zur OSV-Sprache

1033

Haida, einer Indianersprache im südlichen Grenzgebiet von Alaska und Kanada (siehe dazu Strömsdörfer 1992 , 2 5 ff). Sprachen gleichen Typs können also gentisch unterschiedlich klassifiziert sein. Die Gruppierung von Sprachen aufgrund gemeinsamer Ursprachen ist also unabhängig von ihrer typologischen Einordnung und umgekehrt. Eine weitere Art der Gruppierung in der Linguistik, die durch komparative Methoden entsteht, ist die areale Gliederung von Sprachen. Sie untersucht gemeinsame grammatische Eigenheiten von Sprachen, die nicht von einer gemeinsamen Ursprache herrühren, sondern durch Sprachkontakt entstanden sind. Ein hier häufig genanntes Beispiel ist der sog. ‘Balkan-Sprachbund’ (Comrie 1988, 448), zu dem z. B. Serbo-Kroatisch, Bulgarisch, Griechisch, Rumänisch und Albanisch gehören. Ein typisches syntaktisches Charakteristikum dieses Sprachbundes ist z. B. der Verlust des Infinitivs und sein Ersatz durch finite Nebensätze. Vergleiche etwa Albanisch a-më të pi, Rumänisch dá-mi sá beau und Griechisch δως μου νά πιῶ /ðs mu na pj/, die sich alle als ‘gib mir zu trinken’, wörtlich ‘gib mir, daß ich trinke’ übersetzen lassen. Areale Linguistik geht also über die reine linguistische Typologie insofern hinaus, als sie die Ursache gemeinsamer sprachlicher Merkmale (also einer typologischen Klassifizierung) gleich beim Namen nennt und sie damit auf ihre Erklärung im Sprachkontakt verweist. Die so gefundenen Typen sind aber natürlich wieder Teil der Typologie, deren Aufgabe darin besteht, diese spezifische Merkmalskonstellation auch in anderen Sprachen bzw. Sprachgruppen aufzufinden (oder auch nicht). Eine sehr interessante Verknüpfung von ursprünglich typologisch motiviertem Sprachvergleich und Erklärung der gefundenen Gemeinsamkeiten durch Sprachkontakt findet sich bei Pokorny 192 7—30: Dort werden die syntaktischen Übereinstimmungen des Irischen mit dem Afro-Asiatischen durch ein vorgeschichtliches Substrat erklärt. Bleibt noch das Adjektiv ‘syntaktisch’ zu klären. In der Linguistik herrscht keine Einigkeit darüber, was der Untersuchungsbereich der Syntax umfaßt. Er muß sowohl nach „unten“ (zur Morphologie und Phonologie) als auch nach „oben“ (zur Semantik und Pragmatik) abgegrenzt werden. Dies geschieht in unterschiedlichen linguistischen Theorien auf verschiedene Art und Weise. In der traditionellen Grammatik, d. h. in der

XVI. Syntaktische Typologie I: Allgemeines

1034

griechisch-lateinischen Grammatiktradition, beherrscht die Morphologie den linguistischen Beschreibungsapparat. ‘Syntax’ bezieht sich hier nur auf die lineare Abfolge einzelner Wörter und Wortgruppen (vgl. z. B. Cherubim 1976). In der generativen Transformationsgrammatik und ihren Nachfolgemodellen steht dagegen die Syntax in Form einer hierarchisch strukturierten Analyse im Mittelpunkt der grammatischen Beschreibung (vgl. z. B. Chomsky 1988, 4 ff oder Sells 1985, 3). Die genaue Bestimmung des Begriffs ‘Syntax’ muß also im theoretischen Vorfeld einer typologischen Untersuchung von dem Linguisten expliziert werden. Die von uns in diesem Artikel getroffene Auswahl an Beispielen und Untersuchungen zum Thema ‘syntaktische Typologie’ ist nicht an einer bestimmten Theorie orientiert. Sie umfassen einen gewissen Kernbereich sprachlicher Erscheinungen und Konstruktionen, die von den meisten linguistischen Theorien im Rahmen der Syntax behandelt werden.

2.

Methoden und Ergebnisse der syntaktischen Typologie

In der modernen Typologie gibt es eine große Anzahl von Methoden und Begriffen, die in einem kurzen Artikel wie diesem nur stichwortartig erwähnt werden können. Wir möchten die wichtigsten herausgreifen, um die Konzeption und Methodik dieser linguistischen Disziplin, speziell hinsichtlich syntaktischer Verhältnisse, zu beleuchten. 2.1. Beschreibungsmittel typologischer Parameter 2.1.1. Implikative Universalien Linguistische Universalien beschreiben Eigenschaften, die allen menschlichen Sprachen zukommen, d. h. sie formulieren einschränkende Antworten auf die Frage, wie eine mögliche menschliche Sprache beschaffen sein muß. Dabei können Universalien verschiedene Formen annehmen. Die stärkste Form ist die des uneingeschränkten Universals: „Alle Sprachen haben die Eigenschaft Y.“ Beim derzeitigen Stand der Forschung scheint es jedoch kein wirklich gesichertes Universal dieser Art zu geben. Ein häufig zitiertes uneingeschränktes Universal ist z. B.

(Croft 1990, 50): „Alle Sprachen haben orale Vokale.“ Selbst wenn man die Gebärdensprachen gehörloser Menschen ausklammert, die sich gerade dadurch auszeichnen, daß sie keine Sprachlaute haben, bei denen es sich aber ebenso wie bei beliebigen Lautsprachen um vollständige „mögliche menschliche Sprachen“ handelt, gibt es Lautsprachen, bei denen die Existenz von Vokalen bezweifelt wurde, z. B. für das Kabardische (Kuipers 1960 und 1976). Auch alle anderen Kandidaten für uneingeschränkte Universalien werden immer wieder neu diskutiert. So wird häufig behauptet, alle Sprachen unterschieden Nomina und Verben (siehe dazu z. B. Shopen 1985). Sasse (1988) zeigt dagegen z. B. für das Irokesische, daß dies keineswegs sicher ist. Dem uneingeschränkten Universal am nächsten kommt das eingeschränkte Universal bzw. Präferenzgesetz: „Eine Sprache L hat bevorzugt die Eigenschaft Y, d. h. fast alle Sprachen haben die Eigenschaft Y.“ Ein typisches Beispiel dafür ist Greenbergs erstes Universal: „In declarative sentences with nominal subject and object, the dominant order is almost always one in which the subject precedes the object.“ (Greenberg 1966, 77) Greenberg war sich dabei der Tatsache bewußt, daß es Sprachen gibt, in denen das Objekt üblicherweise dem Subjekt vorangeht (vgl. dazu Pullum 1981). Tomlin (1986) zeigte, daß die Anzahl derartiger Sprachen etwa 5% aller natürlichen Sprachen ausmacht. Aber auch eingeschränkte Universalien, die für fast alle Sprachen gelten sollen, gibt es nur in sehr beschränkter Anzahl, zumindest nach gegenwärtigem Stand der Forschung. Linguistische Universalien haben daher am häufigsten die Form bedingter bzw. implikativer Universalien: „Für alle Sprachen gilt: Wenn eine Sprache L die Bedingung X erfüllt, so hat sie die Eigenschaft Y.“ Oder kurz und prägnant in logischer Form (mit den obigen Bezeichnungen): ∀L(X(L) → Y(L)) Implikative Universalien formulieren also eine Abhängigkeit zwischen logisch voneinander unabhängigen linguistischen Parametern (wären sie nicht logisch voneinander unabhängig, bräuchten wir sie ja nicht zu formulieren, da ja alles bereits — logischerweise

53.  Ziele der syntaktischen Typologie

— gesagt wäre und die Universalien keinen empirischen Gehalt mehr hätten). Nimmt man zwei beliebige linguistische Parameter A und B, so gibt es vier logische Möglichkeiten für deren Verteilung (‘’ = „nicht“ ): A&B A & B A & B A & B Sei A z. B. „VSO-Satzstellung in Deklarativsätzen“ und B „Existenz von Präpositionen“ . A entspricht dann allen anderen Satzstellungen (SVO, SOV, usw.), B entspricht der Existenz von Postpositionen (von den extrem seltenen Zirkumpositionen wie in deutsch von mir aus sehen wir hier einmal ab). Greenberg (1966, 78) fand heraus, daß Sprachen mit dominanter VSO-Stellung immer nur Präpositionen haben. Von den vier möglichen Konstellationen kommt eine also nicht vor: Sprachen mit dominanter VSO-Stellung und Postpositionen, d. h. A & B. Damit ist eine klare Einschränkung für mögliche menschliche Sprachen festgestellt. Ein bedeutender Teil typologischer Ergebnisse ist heute in Form implikativer Universalien formuliert. Einige davon stellen wir unter 2.2. zur Wortfolgetypologie vor. Implikative Universalien können auch mit mehr als zwei linguistischen Parametern formuliert werden. Dabei können vor wie nach dem Folgepfeil mehrere Parameter vorkommen. Der Stern ‘*’ steht als Platzhalter für irgendeine sinnvolle logische Verknüpfung wie ‘und’ oder ‘oder’ (die Verknüpfung muß nicht immer die gleiche sein). Ebenso können die Xi und die Yi (≦ i ≤ n, m) bereits selber implikative Universalien sein. (X1 * ... * Xn) → (Y1 * ... * Ym) Ein typisches Beispiel eines solchen komplexen implikativen Universals ist ein Universal von Hawkins (1983, 64): „If a language is SOV, then if the adjective precedes the noun, then the genitive precedes the noun.“ In logischer Notation: SOV → (AN → GN) Dies ist logisch äquivalent mit der Aussage: (SOV & AN) → GN In Worten: In einer SOV-Sprache, in der (attributive) Adjektive dem Nomen vorangehen, gehen auch (attributive) Genitive dem Nomen voran.

1035

An diesem Beispiel wird deutlich, daß der Logik eine gewisse Rolle bei der Formulierung und Behandlung von implikativen Universalien zukommt. Ferner nehmen mit der Statistik (siehe 2 .1.2 .) auch mathematische Verfahren in der modernen Typologie einen wichtigen Platz ein. 2.1.2. Markiertheit Der Begriff der Markiertheit eines sprachlichen Elements wurde ursprünglich in der Prager Schule von Trubetzkoy (1931) entwikkelt. Er spielt in vielen modernen linguistischen Theorien eine Rolle, allerdings in sehr unterschiedlichen Ausprägungen (zur Rolle dieses Terminus in der generativen Grammatik vgl. z. B. Chomsky/Halle 1968, Kap. 9). Mit „Markiertheit“ wird grundsätzlich eine Asymmetrie zwischen zwei ansonsten gleichartigen sprachlichen Elementen bezeichnet. Dieser sehr vage Begriff soll im folgenden expliziert werden. Eine typologische Markiertheitstheorie besteht aus zwei Abteilungen: — den Kriterien zur Identifizierung von markierten und unmarkierten Elementen und — den sprachübergreifend gefundenen Markiertheitsstrukturen. Die wichtigste Grundvoraussetzung bei der Kennzeichnung markierter und unmarkierter Elemente ist, daß sie in einem paradigmatischen Verhältnis zueinander stehen. Es ist z. B. unsinnig zu fragen, ob nasale Vokale „markierter“ sind als Passivkonstruktionen. Sinnvoll dagegen ist die Frage, in welchem Markiertheitsverhältnis nasale und orale Vokale oder Aktiv- und Passivkonstruktionen zueinander stehen. Die wichtigsten auch heute noch gültigen Kriterien für Markiertheit wurden bereits in der Prager Schule formuliert (Jakobson 1936, Trubetzkoy 1939). In Greenberg (1966b) findet sich eine Liste verschiedener Kriterien, die bis heute weitgehend als vollständig gelten kann. Croft (1990, 72 ff) hat diese Kriterien in einer neuen Form strukturiert, die wir im folgenden als Grundlage benutzen, wobei wir uns auf die für die syntaktische Typologie wichtigsten unter ihnen beschränken. Das strukturelle Kriterium ist das älteste, und es wurde bereits in der Prager Schule in seiner heutigen Form entwickelt: „The marked value of a grammatical category will be expressed by at least as many morphemes as is the unmarked value of that category.“ (Croft 1990, 73) Als Beispiel sei hier die morphologische Markierung von Numeri ge-

XVI. Syntaktische Typologie I: Allgemeines

1036

nannt, bei der (wenn überhaupt) der Plural syntagmatisch durch mindestens ebenso viele Morpheme markiert wird wie der Singular. Die Aussage „Bei der Bezeichnung des Numerus ist der Singular der unmarkierte, der Plural der markierte Wert“ (in üblicher Notation: „Singular < Plural“ ) formuliert eine Einschränkung für mögliche menschliche Sprachen: Es kann Sprachen geben, in denen der Numerus überhaupt nicht bezeichnet wird (wie z. B. das Chinesische) und solche, in denen der Singular mit weniger (0 oder mehr) Morphemen gekennzeichnet wird als der Plural (wie z. B. das Deutsche). Doch kann es keine Sprachen geben, in denen z. B. der Plural (normalerweise) unbezeichnet bleibt und der Singular durch zusätzliche Morpheme gekennzeichnet wird. „Normalerweise“ soll hier bedeuten, daß es in manchen Sprachen eine mehr oder weniger große Gruppe von Wörtern gibt, deren unmarkierte Form die Gesamtheit bezeichnet (also pluralisch zu verstehen ist), während ein individuelles Mitglied durch Morpheme markiert ist, z. B. walisisch coed ‘Holz, Bäume’ vs. coeden ‘(ein) Baum’. Man spricht in diesem Fall von „Kollektiv“ vs. „Singulativ“ , um diese Fälle von der üblichen Singular-Plural-Dichotomie zu unterscheiden. Von besonderer Bedeutung für die Markiertheit syntaktischer Konstruktionen sind die Kriterien des grammatischen Verhaltens („behavior“ ) und der Häufigkeit („frequency“ ), sowohl sprachintern wie sprachübergreifend. Insgesamt ergeben sich durch Kombination dieser vier Parameter („behavior“ , „frequency“ , „language-internal“ , „cross-linguistic“ ) vier Kriterien. Wir greifen im folgenden zwei davon heraus. „Behavior (cross-linguistic): if the marked value occurs in a certain number of distinct language types (...), then the unmarked value will occur in at least the language types that the marked value occurs in.“ (Croft 1990, 839) Betrachten wir noch einmal das bereits angeführte Beispiel von VSO-Sprachen und Prä- bzw. Postpositionen in Form einer Tabelle: VSO VSO

Präpos. ✓ ✓

Postpos. ✓

Während also Postpositionen nur in Sprachtypen vorkommen, deren normale Satzstellung nicht VSO ist, kommen Präpositionen in allen möglichen Sprachtypen vor. Postpo-

sitionen sind daher gegenüber Präpositionen „markiert“ (in bezug auf die dominante Satzstellung). Dies gilt trotz Greenbergs Universal 4, das besagt, daß SOV-Sprachen weitaus häufiger Postpositionen haben als Präpositionen. Wenn wir hier von „Sprachtypus“ reden, so verstehen wir darunter, daß eine einzelne Sprache als Repräsentant einer bestimmten Konstellation definierter typologischer Parameter zu verstehen ist. Nur diese Konstellation selber stellt den Typus dar. Keinesfalls ist damit gemeint, daß eine Sprache an sich einen Typus habe oder gar selber einer sei. „Frequency (textual): if a marked value occurs a certain number of times in frequency in a given text sample, then the unmarked value will occur at least as many times in a comparable text sample.“ (Croft 1990, 85) Dieses Kriterium macht deutlich, daß auch statistische Methoden in der Typologie Anwendung finden. Zum Beispiel lautet Greenbergs Universal 6: „Alle Sprachen mit dominanter VSO-Anordnung haben SVO als alternative oder als einzige alternative Grundanordnung.“ (Greenberg 1966, 79) Es gibt zwar auch VSO-Sprachen, die tatsächlich keine SVO-Serialisierung zulassen (z. B. das Gälische), bei den meisten VSO-Sprachen trifft dieser Sachverhalt jedoch zu. In einer „echten“ VSO-Sprache müßten die SVOKonstruktionen seltener vorkommen als die VSO-Konstruktionen. Für eine nahezu konsistente VSO-Sprache wie das Althocharabische ist dies fraglos richtig. Auch das Bretonische wird normalerweise zu den VSO-Sprachen gezählt (siehe z. B. Gregor 1980, 146). Dagegen ergibt bereits ein kurzer Blick auf moderne bretonische Texte, daß die Verhältnisse hier keineswegs so klar sind wie in anderen VSO-Sprachen. Die Anzahl der SVOSätze (die aus Relativsätzen entstanden sind) ist häufig größer als die der VSO-Sätze. Tatsächlich legt dies die Frage nahe, ob das Bretonische dann überhaupt noch als VSO-Sprache bezeichnet werden sollte. Die Verwendung implikativer Universalien im Zusammenhang mit „Markiertheit“ (siehe das obige Beispiel zu VSO-Sprachen und Präbzw. Postpositionen) macht deutlich, daß zwischen diesen beiden typologischen Beschreibungsparametern ein enger Zusammenhang besteht. Ähnlich wie im obigen Fall lassen sich alle implikativen Universalien auch als Markiertheitsstrukturen formulieren. Dies wird in den folgenden Abschnitten noch deutlicher werden, wenn das Konzept der

53.  Ziele der syntaktischen Typologie

Markiertheit grammatischer Konstruktionen in Form von „grammatischen Hierarchien“ und „Prototypen“ weiter differenziert wird. 2.1.3. Grammatische Hierarchien Im vorhergehenden Abschnitt haben wir das Konzept der „Markiertheit“ in binärer Form entwickelt, d. h. von zwei (paradigmatischen) grammatischen Elementen ist jeweils eines „markiert“ und eines „unmarkiert“ . Im Anschluß daran stellten wir einige Kriterien vor, mit denen diese Unterscheidung getroffen werden kann. Nun gibt es aber für grammatische Parameter nicht immer nur zwei Werte, sondern häufig drei oder mehr. Eine wichtige grammatische Markiertheitsrelation besteht etwa zwischen den beiden Termini „Subjekt“ und „Objekt“ , bei der letzterer den „markierten“ Fall bezeichnet: Subjekt < Objekt Tatsächlich kann man aber den Bereich der Objekte weiter unterteilen in direkte Objekte, indirekte Objekte, Präpositionalobjekte usw. Die binäre Markiertheitsrelation wird mit dieser Unterscheidung weiter differenziert (der Terminus „obliques Objekt“ faßt alle Objekte zusammen, die gegenüber indirekten Objekten markiert sind). Subjekt < direktes Objekt < indirektes Objekt < obliq. Objekt Die „absolute“ Markiertheit eines Elements wird so zu einer „relativen“ . Direkte Objekte sind zwar in bezug auf Subjekte markiert, in bezug auf indirekte Objekte und alle anderen obliquen Objekte dagegen sind sie unmarkiert. Die genannte Hierarchie ist in der Typologie nach Keenan/Comrie (1977) unter dem Namen „NP-Zugänglichkeits-Hierarchie“ (NP accessibility hierarchy) bekannt geworden. In ihrer Arbeit untersuchten Keenan und Comrie das grammatische Verhalten von Relativsatz-Konstruktionen. Diese bestehen im weitesten Sinn aus einem Referenzausdruck (normalerweise einem Nomen) und einer syntaktischen Konstruktion, in der ein Bezug auf denselben Referenten vorkommt (z. B. in Form eines Relativpronomens) und die ihn gleichzeitig spezifiziert, wie z. B. der Mann, dessen Tochter gestern heimkehrte. Keenan und Comrie fanden heraus, daß es in allen Sprachen Beschränkungen für die Relativierung von Nomina gibt. Diese Beschränkungen lassen sich in einfacher Form mit der zuvor erwähnten Hierarchie formulieren, d. h. für jede Sprache läßt sich indivi-

1037

duell sagen, daß sich Nomina in allen syntaktischen Rollen links von einer bestimmten Grenze auf der Hierarchie relativieren lassen, die rechts davon grundsätzlich nicht. Im Deutschen beispielsweise können fast alle Nomina relativiert werden, nicht jedoch Vergleichsobjekte in komparativen Sätzen (die folg. Liste ist nicht vollständig!). Subjekt:

Der Mann kommt. → der Mann, der kommt Dir. Obj.: Hanna sieht den Mann. → der Mann, den Hanna sieht Hanna gibt dem Mann das Ind. Ob.: Buch. → der Mann, dem Hanna das Buch gibt Dat.-obj.: Hanna fährt zu dem Mann. → der Mann, zu dem Hanna fährt Gen.-obj.: Hanna erinnert sich des Mannes. → der Mann, dessen Hanna sich erinnert Gen.-attr.: Hanna besucht die Schwester des Mannes. → der Mann, dessen Schwester Hanna besucht Präp.-attr.: Hanna besucht den Bruder von dem Mann. → ?der Mann, von dem Hanna den Bruder besucht Vergl.-obj.: Hanna ist größer als der Mann. → *der Mann, größer als der Hanna ist Im Toba Batak (einer austronesischen Sprache, die auf Sumatra gesprochen wird) können hingegen z. B. nur Subjekte relativiert werden (Beispiel aus Keenan/Comrie 1977, 68 f). Subjekt:

Boru-boruna manussi woman that wash.ACTIVE abit i clothes the ‘the woman who is washing the clothes’ Dir. Obj.: *abit na manussi clothes that wash.ACTIVE boru-boru i woman the ‘the clothes that the woman is washing’ In der typologischen Literatur findet sich eine ganze Reihe unterschiedlicher grammati-

1038

scher Hierarchien. Wir wollen im folgenden nur einige der häufig erwähnten wiedergeben. Numerushierarchie Singular < Plural < Dual < Trial/Paukal Hierarchie der grammatischen Relationen (s. o.) Subjekt < direktes Objekt < indirektes Objekt < obliq. Objekt Definitheitshierarchie definit < referentiell indefinit < nicht-ref. indefinit Kasushierarchie Nominativ < Akkusativ < Dativ < and. Kasus bzw. Absolutiv < Ergativ < and. Kasus Tatsächlich sind die verschiedenen Hierarchien miteinander verknüpft und lassen sich nicht einfach nur so aufzählen, wie obige Liste suggerieren mag. So hängen z. B. die Kasushierarchie und die der grammatischen Relationen eng miteinander zusammen: Subjekte werden bevorzugt im Nominativ (in Ergativsprachen im Absolutiv) kodiert, direkte Objekte im Akkusativ usw. Diese Einsicht führt zur Theorie der grammatischen Prototypen (siehe Kap. 2.1.4.). Neben dem hier dargestellten Ansatz, grammatische Hierarchien als eine Differenzierung des Markiertheitskonzepts zu entwikkeln, gibt es auch Ansätze, die die zweiwertige Markiertheitsrelation beibehalten und den jeweils markierten Wert in zwei weitere Werte aufteilen (siehe z. B. Silverstein 1976, 122): + 2. Person — 2. Person: + 1. Person — 1. Person: + Pronomen — Pronomen: + menschlich — menschlich: + belebt — belebt Auf diese Art werden verschiedene relationale Hierarchien miteinander verknüpft, im vorstehenden Fall die pronominale Hierarchie (2 ./1. Person < 3. Person), die Hierarchie über NP-Typen (Pronomen < Eigennamen < Nomen) und die Belebtheitshierarchie (menschlich < belebt < unbelebt). Darüber hinaus besteht ein enger Zusammenhang zwischen der Belebtheits- und der Definitheitshierarchie, der für die transitiven Prototypen von entscheidender Bedeutung ist (siehe

XVI. Syntaktische Typologie I: Allgemeines

Kap. 2.1.4.). Diese verschiedenen Ansätze bei der Behandlung grammatischer Hierarchien widersprechen einander nicht, sie ergänzen sich vielmehr. 2.1.4. Prototypen Wie bereits am Ende des vorangegangenen Teils über grammatische Hierarchien deutlich wurde, wirken einzelne grammatische Kategorien wie Numerus, Genus, Belebtheit, Modus usw. nicht isoliert, sondern gemeinsam auf die sprachlichen Formen, die wir empirisch feststellen können. In diesem Zusammenhang stellt sich die Frage, ob sich Muster typischer Relationen zwischen verschiedenen grammatischen Parametern feststellen lassen. Oder mit einem anderen Ausdruck: Gibt es grammatikalische Prototypen? Der einfachste Fall eines solchen Prototyps liegt vor, wenn zwei oder mehr unmarkierte Werte verschiedener Parameter zusammentreffen, wie z. B. „Singular“ , „Nominativ“ und „menschlich“ . Dies wäre in vielen Sprachen der Welt der prototypische Fall eines Nomens (als Zitierform, als Referenznomen bei Kongruenz usw.). Andere Nomina, z. B. eines im Dativ Dual, besitzen nicht mehr all die Eigenschaften des Prototyps. Prototypen sind keine klar abgegrenzten Klassen grammatischer Ausdrücke, sie besitzen vielmehr eine Reihe von protoypischen Mitgliedern, die alle Eigenschaften der Klasse erfüllen, und anderen Mitgliedern, die nur noch einen Teil dieser Eigenschaften besitzen. Verbalnomina z. B. gehören in der Klasse der Nomina eher einem Randbereich als dem Kern der Klasse der Nomina an (sie haben oft nicht die gleiche Vielfalt an verschiedenen Kasusformen, ihre syntaktische Verwendung ist stark eingeschränkt, usw.). In den achtziger Jahren gab es eine intensive Forschung auf dem Gebiet „transitiver“ Prototypen, die von einer Arbeit von Hopper/Thompson (1980) ausgelöst wurde (siehe auch Hopper/Thompson 1982 ). Hopper und Thompson bearbeiteten die Frage, ob sich grammatische Bedingungen formulieren lassen, die erfüllt sein müssen, um morphosyntaktisch transitive Syntagmen bzw. Sätze bilden zu können. Dazu legten sie zuerst Kriterien dafür fest, was transitive Konstruktionen in Akkusativ- und Ergativsprachen überhaupt sind (z. B. Kriterien, um den Status ei-

53.  Ziele der syntaktischen Typologie

nes nominalen Arguments eines Verbs als direktes Objekt feststellen zu können). Als Ergebnis ihrer Untersuchung formulierten sie ein Bündel aus zehn prototypischen Eigenschaften der Transitivität (Hopper/Tompson 1982 , 2 52 ), von denen wir hier nur einige nennen wollen, da sie ohne weitere Erklärung nicht verständlich wären: — zwei oder mehr Aktanten müssen beteiligt sein — es liegt ein Handlungsverb vor — das Verb hat perfektiven Aspekt — der Agens handelt selbst — die Handlung ist in besonders starkem Maß auf das direkte Objekt gerichtet (und nicht etwa auf andere Objekte oder das Subjekt) — das direkte Objekt ist klar identifizierbar („highly individuated“ ) Die zuletzt genannte Eigenschaft ist wiederum ein Bündel aus Eigenschaften, die sich aus den Hierarchien ergeben, die wir am Ende des Kapitels zu grammatischen Hierarchien nach Silverstein (1976) zu einer Art treppenförmiger Hierarchie umgeformt haben. Danach ist ein Objekt am besten identifizierbar, wenn es einen einzelnen Menschen bezeichnet, idealiter mit einem Eigennamen oder gar einem Pronomen usw. (siehe Hopper/Thompson 1980, 253). Der dritte der obigen Punkte läßt sich z. B. im Finnischen demonstrieren. Hat das Verb perfektiven Aspekt, so steht das direkte Objekt im Akkusativ, hat es dagegen imperfektiven Aspekt, so steht das direkte Objekt im Partitiv (Beispiel aus Hopper/Thompson 1980, 271): kirjoitti kirjee-n liikemies businessman wrote letter.ACC valiokunalle committee-to ‘The businessman wrote a letter to the committee.’ kirjoitti kirjet-tä liikemies businessman wrote letter.PART valiokunnalle committee-to ‘The businessman was writing a letter to the committee.’ Die Charakterisierung des Agens läßt sich in vielen Punkten im Gälischen beobachten. Das Gälische kennt eine spezielle unpersönliche Form von Verben, die häufig mit dem Passiv übersetzt wird. Tatsächlich kann sie aber nur verwendet werden, wenn der (nicht

1039

genannte) Agens eine tatsächlich handelnde Person ist. bhualadh Màiri le cloich man-erschlug Màiri mit Stein ‘Jemand erschlug Màiri mit einem Stein.’ bhualadh Màiri le Seumas man-erschlug Màiri mit Seumas ‘Màiri wurde von Seumas erschlagen.’ Der erste Satz kann in keinem Fall ‘Màiri wurde von einem Stein erschlagen.’ heißen, d. h. die transitive Interpretation des Präpositionalobjekts ist nur möglich, wenn es einen genuinen Agens bezeichnet. Wir haben in den letzten Kapiteln gesehen, daß die Beschreibungsmethoden der Typologie miteinander zusammenhängen. Wir begannen mit der Formulierung „implikativer Universalien“ und zeigten, daß diese auf dem Konzept der „Markiertheit“ grammatischer Elemente beruhen. Dieses Konzept erweiterten wir zu dem der „grammatischen Hierarchien“ . Aus diesen wiederum läßt sich das Beschreibungsmittel der typologischen „Prototypen“ gewinnen. Im folgenden wollen wir noch einmal kurz zum Ausgangspunkt der syntaktischen Typologie zurückkehren, zu Greenbergs syntaktischen (und morphologischen) Universalien (siehe Greenberg 1966). 2.2. Wortfolgetypologie Die Geschichte der modernen syntaktischen Typologie beginnt mit Greenbergs Forschungen zur Wortstellung (Greenberg 1966). Eine eingehende Diskussion dieses Ansatzes findet sich in Artikel 55 in diesem Band. Hier soll er wegen seiner methodischen und historischen Bedeutung behandelt werden. Eine grundsätzliche Erkenntnis der Wortfolgetypologie ist, daß es einen Zusammenhang zwischen der normalen Wortfolge der Hauptkonstituenten in einem Aussagesatz (Verb, Subjekt und Objekt) und der Struktur anderer syntaktischer Konstruktionen gibt. Die drei genannten Konstituenten Verb (V), Subjekt (S) und Objekt (O) können logisch auf sechs Arten angeordnet werden: SOV, SVO, VSO, VOS, OVS und OSV. Die ersten drei Typen, in denen das Subjekt immer dem Objekt vorangeht, finden sich am häufigsten in den Sprachen der Welt. Der Anteil der Sprachen, deren Grundkonstruktion das Objekt dem Subjekt vorangehen läßt, beträgt nur ca. fünf Prozent (vgl. Tomlin 1986).

XVI. Syntaktische Typologie I: Allgemeines

1040

Ainu: SOV kamuy umma rayke Bär Pferd tötet ‘Ein Bär tötete ein Pferd.’ Englisch: SVO wife the man kissed his der Mann küßte seine Frau ‘Der Mann küßte seine Frau.’ Gälisch: VSO Dh’fhosgail Calum an doras. öffnete Calum die Tür ‘Calum öffnete die Tür.’ Malagassisch: VOS ny zazavavy manasa lamba wäscht Kleider das Mädchen ‘Das Mädchen wäscht Kleider.’ Hixkaryana: OVS yanmno bryekomo kana Junge Fisch er-fing-ihn ‘Der Junge fing den Fisch.’ Haida: OSV laa l tiigan sie er tötete ‘Er tötete sie.’ Greenberg formulierte seine Korrelationen in Form von 45 implikativen Universalien bzw. Präferenzgesetzen. Greenbergs syntaktische Universalien lassen sich verkürzt mit zwei Haupttypen für die Serialisierung syntaktischer Elemente wiedergeben (von der Position des Subjekts sehen wir hier ab). (i)

(ii)

Objekt Adverbial Adjektiv Genitiv Nomen Verb Verb Nomen Nomen Präposition

— — — — — — — — — —

Verb Verb Nomen Nomen Postposition Objekt Adverbial Adjektiv Genitiv Nomen

In einigen späteren Arbeiten wurde dieses Ergebnis besonders hervorgehoben. So reduziert beispielsweise Vennemann (1972 ) — ähnlich Lehmann (1973) — die obigen Typen im Rahmen der Kategorialgrammatik auf zwei mögliche Serialisierungen von Operator und Operand. Unter dem Operator ist in diesem Zusammenhang das modifizierende Element (bzw. der Spezifikator) zu verstehen. In obigen Beispielen ist dies unter (i) die linke, unter (ii) die rechte Spalte. Unter dem Operanden versteht man das modifizierte Element (den „Kopf“ der Konstruktion, der die Kategorie des Resultatausdrucks bestimmt).

Das Besondere an dieser „Idealisierung“ bestimmter syntaktischer Konstruktionen ist der Anspruch, mit einer Aussage („Operand vor Operator“ = „postspezifizierend“ bzw. „Operator vor Operand“ = „präspezifizierend“ ) die gesamte Serialisierung des Spezifikationsverhältnisses einer Sprache über alle Kategorien hinweg zu erfassen. Man spricht in diesem Fall von „holistischen“ Typologien (vgl. dazu z. B. Ramat 1987). In den meisten Sprachen der Welt sind die syntaktischen Verhältnisse jedoch keineswegs so konsistent. Das Gälische z. B. ist eine konsistente VSO-Sprache, die auch Präpositionen besitzt. Diese sind aber keine Operanden, sondern vielmehr präfigierte nominale Operatoren (vgl. z. B. Borgstrøm 1968). Im Haida, einer Sprache, die häufig unter den Kandidaten für OSV-Sprachen genannt wird (siehe Pullum 1981), liegt syntaktisch genau der umgekehrte Fall vor, d. h. die Postpositionen des Haida sind suffigierte nominale Operatoren (siehe Lawrence 1977, 70 ff). Man kann sich dann allerdings die Frage stellen, ob es sich bei den gälischen und den Haida-Adpositionen überhaupt um solche handelt, da Adpositionen sich eben genau dadurch auszeichnen, daß sie stets und ausnahmslos Operanden sind (dies ist eine Frage der Definition). Insgesamt ist die Wortfolgetypologie heute an einem Punkt angelangt, an dem viele Forscher davon ausgehen, daß Wortfolgeuniversalien, die von irgendeiner heuristischen Bedeutung sein sollen, so speziell sind, daß sie für eine Typologisierung nicht mehr handhabbar sind (vgl. hierzu auch Hawkins 1983 und Artikel 29 in Bd. I). Darüber hinaus ist auch das grammatische Konzept der „grammatischen Relationen“ in den letzten zwanzig Jahren neu problematisiert worden (vgl. z. B. Keenan 1976, Sasse 1978, Vennemann 1982 a, Primus 1987, Artikel Nr. 57 über Relationstypologie in diesem Band). Ähnlich holistische Ansätze, d. h. Ansätze, die mit Hilfe eines Parameters alle anderen erklären möchten, sind in der Typologie immer wieder aufgetreten, wie z. B. die Prominenztypologie (vgl. Li/Thompson 1976, Edward 1979, Artikel Nr. 56 über Prominenztypologie in diesem Band). Im allgemeinen werden sie jedoch heute nicht mehr in gleichem Maße vertreten. Die Forscher konzentrieren sich zumeist auf die Formulierung grammatischer Muster im Rahmen der unter 2 .1. besprochenen Beschreibungsmethoden.

53.  Ziele der syntaktischen Typologie

Einige der bedeutendsten Untersuchungen zur syntaktischen Typologie (u. a. die Zugänglichkeitshierarchie für Nominalphrasen, die transitiven Prototypen) sind bereits an entsprechender Stelle von uns vorgestellt worden.

3.

Zur Erklärung typologischer Parameter

Typologen interessieren sich natürlich nicht nur für das „Wie“ sprachübergreifender sprachlicher Strukturen, sondern auch für das „Warum“ . Wie wir bereits eingangs verdeutlicht haben, kann die Typologie diese theoretische Interpretationsarbeit nicht selbst leisten. Die Frage nach dem „Warum“ geht über die eigentliche Typologie hinaus. Gleichwohl liefert die Typologie zum einen das empirische Material, das zu interpretieren ist, zum anderen liefert sie eine heuristische Methode, die auch in der theoretischen Arbeit gewinnbringend eingesetzt werden kann. Die Typologie ist „empiristisch“ orientiert, im Gegensatz zur „rationalistischen“ Vorgehensweise der generativen Grammatik. Dieser Ansatz bringt es mit sich, daß Linguisten, die im Rahmen typologischer Forschung linguistische Theorien entwerfen, empirisch feststellbare oder meßbare Ursache-WirkungsZusammenhänge suchen. Sie versuchen daher, sprachliche Eigenschaften aus dem zu erklären, was Sprache „tut“ , aus ihren „Funktionen“ also. Der rationalistische Theoretiker wird eher versuchen, durch Verbesserung und Differenzierung seines theoretischen Apparates die sprachlichen Daten besser zu erfassen. Dies ist eine „formale“ Vorgehensweise. Es gibt allerdings keinen irgendwie zwingenden logischen Zusammenhang zwischen einer rationalistischen und einer formalen bzw. einer empiristischen und einer funktionalen Methodik. Es ist gewissermaßen ein historischer Zufall, daß diese Begriffe in der angesprochenen Paarung zumeist zusammen genannt werden. Um das „funktionale Paradigma“ in Ansätzen vorzustellen, wollen wir im folgenden einige der Erklärungsmechanismen vorstellen, die in funktional-typologischen Arbeiten verwendet werden. Die zwei wichtigsten Termini in diesem Zusammenhang sind die Prinzipien der „Ökonomie“ und der „Ikonizität“ . Das Ökonomie-Prinzip besagt, daß ein Ausdruck um so kürzer wird, je häufiger er gebraucht wird.

1041

Für Markiertheitsrelationen bedeutet dies insbesondere, daß die häufig gebrauchten unmarkierten Werte eben deswegen grammatisch unmarkiert sind, weil eine Anhäufung grammatischer Informationen nur ein Übermaß an grammatischen Elementen zur Folge hätte. Und dies wiederum widerspräche dem Ökonomie-Prinzip, das davon ausgeht, daß menschliche Sprache irgendeine Funktion hat, die außerhalb der Sprache selbst liegt. Demnach sollte ein Minimum an sprachlicher Aktivität ein Maximum an sprachlicher Funktion erfüllen. Die sprachliche Äußerung sollte also möglichst „ökonomisch“ sein. Aus diesem Grund finden sich Unregelmäßigkeiten von Sprachen normalerweise besonders bei häufig vorkommenden Elementen. Die unregelmäßige Form läßt sich als Verschmelzung verschiedener regelmäßiger Elemente in ein verkürztes Element verstehen, das sowohl ökonomisch als auch optimal in bezug auf die Vermittlung grammatischer Information ist. Im Isländischen z. B. wurde in einer früheren Sprachstufe der Plural der Neutra auf -u gebildet: barn ‘Kind’ — *barnu ‘Kinder’. Durch das -u wurde -a- zu -ɔ- umgelautet, wodurch *bɔrnu ‘Kinder’ entstand. Das -u ist im Lauf der Sprachentwicklung geschwunden. Der Plural wird also nur noch durch den sog. „Labialumlaut“ angezeigt: neu-isl. börn ‘Kinder’ (siehe dazu Kress 1982, 44). Das Prinzip der Ikonizität beruht auf der Annahme, daß die Struktur von Sprachen in irgendeiner Form die Struktur menschlicher Erfahrung reflektiert. Sprache ist also gewissermaßen eine „Ikone“ der Struktur menschlicher Erfahrung bzw. der Struktur, die Menschen der Welt um sich herum geben. Ikonizität läßt sich in zwei Teile aufteilen: (i) die Ikonizität sprachlicher Elemente und der Elemente der menschlichen Erfahrungswelt und (ii) die Ikonizität zwischen den grammatischen Relationen zwischen sprachlichen Elementen und den Relationen zwischen den Erfahrungsgegenständen. In der linguistischen Literatur gibt es für diese beiden Formen von Ikonizität verschiedene Begriffe. So wird (i) z. B. von Haiman (1985) als „Isomorphismus“ bezeichnet, von Croft (1990) als „ikonische Motivation“ usw. Die Forschung in diesem Bereich ist noch zu neu, als daß es bereits eine einheitliche Terminologie gäbe. Den idealen ikonischen Fall beschreibt die Regel: eine Form — eine Bedeutung. In der Morphologie verwirklichen diesen Fall die

XVI. Syntaktische Typologie I: Allgemeines

1042

sog. agglutinierenden Sprachen, z. B. das Mongolische. Die beiden oben unter (i) und (ii) genannten Prinzipien sind keineswegs die einzigen, die heute in typologisch motivierten Sprachtheorien verwendet werden, doch sie verdeutlichen die Richtung linguistischer Argumentation, die hier angestrebt wird. Ein weitergehender Ansatz linguistischer Theorienbildung, der weitgehend auf typologischer Forschung aufbaut, wird in Artikel 60 vorgestellt.

4.

Zusammenfassung

Dieser Artikel gibt einen Überblick über die Methoden und Ergebnisse der syntaktischen Typologie. Im ersten Teil wurde der allgemeine Beschreibungsrahmen von Typologie beschrieben. Im zweiten Teil wurden einige Methoden der Typologie vorgestellt. In diesem Rahmen gingen wir auf verschiedene Beispiele für syntaktische Typologie ein. Anschließend gaben wir einen kurzen Rückblick auf die Anfänge der modernen syntaktischen Typologie, um die Unterschiede zu neueren Ansätzen zu verdeutlichen. Wir schlossen im dritten Teil mit einem Ausblick auf typologisch motivierte Sprachtheorien im Rahmen funktionaler Beschreibungsmodelle.

5.

Literatur

Behaghel, Otto. 192 3—1932 . Deutsche Syntax: eine geschichtliche Darstellung. Heidelberg. Borgstrøm, Hjalmar. 1968. Notes on Gaelic grammar. In: James Carney & David Greene: Celtic Studies. Essays in memory of Angus Matheson. London. Cherubim, D. 1976. Grammatische Kategorien. Das Verhältnis von „traditioneller“ und „moderner“ Sprachwissenschaft. Tübingen. Chomsky, Noam. 1970. Sprache und Geist. Frankfurt am Main. —. 1988. Lectures on government and binding. Dordrecht. —, and M. Halle. 1968. The sound pattern of English. New York. Comrie, Bernard. 1988. Linguistic typology. In: Frederick J. Newmeyer: Linguistics. The Cambridge survey. Vol. 1: Linguistic theory: Foundations, 447—461. Cambridge. —. 1990. Language universals and linguistic typology. Chicago. Croft, William. 1990. Typology and universals. Cambridge.

Edwards, Elizabeth A. 1979. Topic marking in Haida. In: IJAL, Vol. 45, 149—156. Givón, Talmy. 1984. Syntax: a functional-typological introduction. Amsterdam. Greenberg, Joseph. 1966 a. Universals of language. Cambridge. —. 1966 b. Language universals, with special reference to feature hierarchies. (Janua Linguarum, Series Minor, 59) The Hague. —. 1987. Language in the Americas. Stanford. Gregor, D. B. 1980. Celtic. A comparative study. New York. Haiman, John. 1985. Natural Syntax. Cambridge. Hammond, Michael, Edith A. Moravcsik & Jessica R. Writh (eds.) 1988. Studies in syntactic typology. Amsterdam/Philadelphia. Hawkins, John A. 1983. Word order universals. Quantitative analysis of linguistic structures. New York. —. 1986. A comparative typology of English and German. London. Heath, Jeffrey. 1984. Functional grammar of Nunggubuyu. Canberra. Humboldt, Wilhelm von. 1836. Über die Verschiedenheit des menschlichen Sprachbaues und ihren Einfluß auf die geistige Entwicklung des Menschengeschlechts. Berlin. Jakobson, Roman. 1936. Beitrag zur allgemeinen Kasuslehre. Gesamtbedeutungen der russischen Kasus. In: TCLP 6, 240—288. Keenan, Edward L. 1976. Towards a universal definition of subject. In: Charles N. Li: Subject and topic. New York. Kress, Bruno. 198 2 . Isländische Grammatik. Leipzig. Kuipers, Aert H. 1960. Phoneme and morpheme in Kabardian. Den Haag. —. 1976. Typologically salient featurs of some north-west Caucasian languages. Lisse. Lawrence, Erma. 1977. Haida Dictionary. Fairbanks. Lehmann, Winfried P. 1973. A structural principle of language and its implications. In: Language 49, 47—66. —. (ed.) 1990. Language typology 1987. Systematic balance in language. Amsterdam/Philadelphia. Li, Charles N. & Sandra A. Thompson. 1976. Subject and topic. A new typology of language. In: Charles N. Li: Subject and topic. New York. Mallinson, Graham & Barry Blake. 1981. Language typology. Amsterdam. Pokorny, Julius. 192 7—1930. Das nicht-indogermanische Substrat im Irischen. In: Zeitschrift für celtische Philologie, Bd. 16 (95—144, 2 31—2 66, 363— 394), Bd. 17 (373—388), Bd. 18 (233—248).

54.  Geschichte der syntaktischen Typologie

Primus, Beatrice. 1987. Grammatische Hierarchien. München. Pullum, Geoffrey K. 1981. Languages with object before subject: a comment and a catalogue. In: Linguistics 19, 147—155. Ramat, Paolo. 1987. Linguistic typology. Berlin. Robins, R. H. 1979. A short history of linguistics. Harlow. Sandkühler, H. J. (Hrg.) 1990. Europäische Enzyklopädie zu Philosophie und Wissenschaften. Bd. IV (R—Z). Hamburg. Sasse, Hans-Jürgen. 1978. Subjekt und Ergativ: Zur pragmatischen Grundlage primärer grammatischer Relationen. In: Folia Linguistica XII, 3—4, 219—252. Den Haag. —. 1982 . Subjektprominenz. In: Sieglinde Heinz & Ulrich Wandruszka (Hrg.): Fakten und Theorien. Festschrift für Helmut Stimm zum 65. Geburtstag. Tübingen. —. 1988. Der irokesische Sprachbau. Arbeitspapier Nr. 9 (Neue Folge). Inst. für Sprachwissenschaft der Universität zu Köln. Schlegel, August Wilhelm. 1818. Observations sur la langue et la littérature provençales. Paris. Schlegel, Friedrich. 1808. Über die Sprache und Weisheit der Indier. Heidelberg. Sells, Peter. 1985. Lectures on contemporary syntactic theories. Stanford. Silverstein, Michael. 1976. Hierarchy of features and ergativity. In: R. M. W. Dixon (ed.): Grammatical categories in Australian languages, S. 112—171. Canberra. Shopen, Timothy (Hrg.) 1985. Language typology and syntactic description. Cambridge.

1043

Stassen, Leon. 1985. Comparison and universal grammar. Oxford. Strömsdörfer, Christian. 1982 . Ist das Gälische eine polysynthetische Sprache? MA Thesis. München. Thompson, Sandra A. & Robert E. Longacre. 1983. Adverbial clauses. In: Shopen (1985), 171—234. Tomlin, Russel. 1986. Basic word order. Functional principles. London. Trubetzkoy, Nicholas. 1931. Die phonologischen Systeme. In: Travaux du Cercle Linguistique de Prague 4, S. 96—116. —. 1939. Grundzüge der Phonologie. Göttingen. Vennemann, Theo. 1972 . Analogy in generative grammar. The origin of word order. In: Proceedings of the 11th International Congress of Linguistics, Bologna, Aug—Sep 1972. —. 1982 a. Remarks on grammatical relations. In: The Linguistic Society of Korea (ed.): Linguistics in the Morning Calm: Selected papers from SICOL 1981. Seoul. —. 1982 b. Isolation—Agglutination—Flexion? Zur Stimmigkeit typologischer Parameter. In: Sieglinde Heinz & Ulrich Wandruszka (Hrg.): Fakten und Theorien. Festschrift für Helmut Stimm zum 65. Geburtstag. Tübingen. —. 1985. Universals, preferences, typologies: Definitions and deliminations. In: Ursula Pieper und Gerhard Stickel (eds.): Studia Linguistica et Diachronica. Festschrift für Werner Winter zum 60. Geburtstag. Berlin.

Christian Strömsdörfer, Nürnberg (Deutschland)Theo Vennemann, München (Deutschland)

54. Geschichte der syntaktischen Typologie 1. 2. 3. 4. 5. 6.

1.

Vorbemerkung Erste Ansätze. Vorläufer Die Typologie der Korrelationen Die Wortfolgetypologie Die Funktionstypologie Literatur

Vorbemerkung

Die syntaktische Typologie ist die modernste Form der typologischen Linguistik. Es ist deshalb nicht so einfach, bereits von einer Geschichte derselben zu reden, zumal die Abgrenzung ihres Objektbereichs ebenso unscharf ist wie ihre methodologische Gestalt, die von den Konzeptionen verschiedener

Schulen der Linguistik abhängt. Man kennt auch die schwierige Gratwanderung zwischen Allgemeiner Grammatik und Typologie, d. h. zwischen theoretisch-deduktiven Verfahren aufgrund eines intuitiven Vorwissens (z. B. nach Keenan 1982 ) und empirisch-induktiven Verallgemeinerungen aufgrund der interlingual verfügbaren Vergleiche (wie zumeist; vgl. auf der Linie von Greenberg z. B. Comrie 1981, ferner Dezső 1982 , Givón 1984). Zu erwähnen ist auch der Beitrag idiosynkratischer Typologien, wie z. B. die sogenannte integrale Typologie von Coseriu (vgl. Ineichen 1988). Sieht man von einer Intonationstypologie, die stark syntaktisch gebunden wäre, ab,

XVI. Syntaktische Typologie I: Allgemeines

1044

dann ist man sich dem gegenüber einig, daß es — z. B. nach Dressler (1979); vgl. auch Bartsch/Vennemann (1982 , 30) — eine phonologische Typologie als umschriebene Disziplin gibt. Ebenso anerkannt ist der Rahmen der morphologischen Typologie, die oft dargestellt wurde (z. B. Ineichen 1979). Doch hat Ulrike Kölver (1985) im Falle des Birmanischen nachgewiesen, daß zwischen Wortstruktur und Syntax signifikante Abhängigkeiten bestehen können. Dieses Problem ist im Rahmen der syntaktischen Typologie bislang noch kaum untersucht worden.

2.

Erste Ansätze. Vorläufer

Unter dem Gesichtspunkt der Geschichte der Disziplin stellen wir nun zunächst die Frage nach jeweils maßgeblichen Vorläufern. Wir beschränken uns dabei auf die Situation in Deutschland. 2.1.  Das Interesse an einer typologisch orientierten vergleichenden Syntax findet man explizit formuliert erstmals bei Hans Georg von der Gabelentz (1840—1893), und zwar (1869, 376): „Die Frage, die ich mir gestellt habe, lautet: ‘Auf welchen allgemeinen Principien beruhen die Wort- und Satzstellungsgesetze der einzelnen Sprachen?’“ Vermutlich dachte Gabelentz an eine Syntax, die der Morphologie von Schleicher (1865) vergleichbar gewesen wäre. Doch waren die Vorstellungen von der Gabelentz’ damals noch ohne theoretische Stütze im modernen Sinn. Seine Position war auch ideologisch nicht ungefährlich, und zwar deshalb, weil er der Auffassung war, „le style c’est l’homme“ , oder anders: „Wie ein Volk seine Begriffe, seine Gedanken ordnet, so ordnet es seine Sätze“ (1869, 377). Mit anderen Worten: er ging in die Richtung der damaligen Völkerpsychologie. (Unter diesem Gesichtspunkt war Saussure — „la langue envisagée en elle-même et pour elle-même“ — vollkommen ideologiefrei.) Man denke hier auch an Franz Nikolaus Finck (1867—1910), dessen Sprachauffassung letztlich doch sehr „deutsch“ war. Der seinerzeitigen Völkerpsychologie maßgeblich verbunden war auch Wilhelm Wundt (1832 —192 0). Hier gilt zunächst die Gebärde zwischen Schrift und Sprache als eine „Äußerung der Gedanken durch sichtbare, aber nicht hörbare Bewegungen“ . Nach Wundt (1900, 2 06) gibt es deshalb „bestimmte Ge-

setze der Wortfügung“ , die durch die Zeichenfolge in den Gebärden bestimmt sind. Die heutige Sprachwissenschaft spricht von Genese, wenn überhaupt, unter ganz anderen Gesichtspunkten: Sprache war immer schon da; vom Sprachursprung kann in der systematischen Linguistik deshalb nicht geredet werden. Spekulationen, z. B. von Adelung (1732 —1806) über Relikte von Lallsprachen in Ostasien, gehören zur Anekdotik. 2.2.  Mit Bezug auf das Deutsche liest man bei Gabelentz (1875, 144): „Grundgesetz scheint mir nach dem heutigen Stande unserer Sprache zu sein, daß in der mittheilenden Redeweise das verbum finitum des Hauptsatzes immer die zweite Stelle einnimmt“ .

Man erinnert sich dabei an die Beobachtungen von Erich Drach (1885—1935) bzw. ante litteram an das sogenannte Drach’sche Gesetz (nach Drach 1940, 16) und an die Forschungen von Behaghel (1854—1936). Dort findet man (1932 , 4) bereits Hinweise auf die Diskursfunktionalität und insbesondere das bekannte „Gesetz der wachsenden Glieder“ . Im Romanischen konnte Gabelentz (1975, 161 bzw. 300) eine solche Präzision nicht gelingen, da die komplexe syntaktische Organisation der romanischen Sprachen erst in den letzten Jahren deutlicher geworden ist. Andererseits erscheint das Lateinische (1875, 311) als SVO. Im Japanischen kennt von der Gabelentz (1875, 319) auch bereits die topikalisierende Funktion der Partikel wa. Gabelentz (1875, 141) erkannte auch das Problem der Kategorisierung. Er sagt dazu: „Das Problem ist, ein Gesetz zu finden, nach welchem für alle Sprachen die Frage, was ist Wort, was ist Satz, — was ist Hilfswort, was Wortbestandtheil, sich beantworten liesse. Es soll ein gemeingültiges Gränzmerkmal errichtet werden: giebt es aber eine gemeingültige Gränze?“

2.3.  Entscheidend für die Erfahrung von der Gabelentz’ ist vermutlich seine Beschäftigung mit dem Chinesischen. Man liest in seiner noch heute verwendeten Grammatik (1881, 113): „Die ganze nun folgende Grammatik ist Syntax, und diese ganze Syntax beruht auf wenigen, mehr oder minder unverbrüchlichen Gesetzen der Wortstellung. Vom Satze ist also auszugehen; er ist zugleich erstes Objekt der analytischen Untersuchung und erster Zweck der synthetischen Sprachanwendung“ .

Dabei werden im Rahmen der traditionellen Grammatik auch stringente Festlegungen ge-

54.  Geschichte der syntaktischen Typologie

troffen. So sind z. B. die Präpositionen „sowohl ihrem Ursprunge als auch ihrer syntaktischen Behandlung nach Verba“ (S. 2 56), mit anderen Worten das, was man heute als CoVerb bezeichnet (vgl. Chiang 1977). 2.4.  Im Hinblick auf die Satzfunktionalität findet man bei Gabelentz (1875, 137), also lange vor der „Sprachwissenschaft“ von 1891, bereits den Terminus Thema, nämlich „das Thema, worüber ich reden will, also mein psychologisches Subjekt“ . Nach den Stellungsregeln des Deutschen — im Gegensatz also zum Mandschu, zum Japanischen und zum Chinesischen — war dann „jedes folgende Satzglied psychologisches Prädikat“ (1875, 150). Man findet diese psychologisierende Position in der Folge auch bei Hermann Paul (1846—192 1) (1880, Kap. 16: Verschiebung der syntaktischen Gliederung), und sie reicht bis in die Anfänge der Prager Schule. Verheerend wirkte das „psychologische Subjekt“ bei den Erklärungsversuchen des Konjunktivs im Französischen, z. B. bei Eugen Lerch (1888—1952). 2.5.  Im Jahre 1898 veröffentlichte Hermann Reckendorf (182 5—1875?) eine umfassende Syntax des Arabischen, wo von „syntaktischen Verhältnissen“ die Rede ist. Ebenfalls einzelsprachlich orientiert sind vier neuere Arbeiten zur Wortfolge, die man bei Bakalla (1983) verzeichnet findet. Das Phänomen, um das es im Arabischen zunächst geht, ist die Organisation der Nominal- und der Verbalphrase. Doch findet man bei Reckendorf auch bereits Hinweise auf das Verfahren, „die Aufmerksamkeit des Zuhörers zu erregen“ (S. 12 8), z. B. im Falle von inna, das traditionell in Konstruktionen mit „Wahrlich ...“ aufgefangen wird, auch wenn von einer topikalisierenden Funktion hier nicht eigentlich die Rede sein kann, da diese Partikel (wie lat. ecce und seine Nachfolger im Romanischen) über ganze Sätze operiert. 2.6.  Übergehend zur Problemstellung im einzelnen könnte man den Versuch machen, die Geschichte der Disziplin anhand einer bestimmten Kategorie aufzurollen. Besonders geeignet dazu wäre die Behandlung des Passivs. Man stößt auch hier auf einen Gabelentz, und zwar auf Hans Conon, Georgs Vater, der 1861 mit der „Betrachtung eines Sprachtheils durch alle bekannten Sprachen des Erdboden“ (1861, 415) eine sprachver-

1045

gleichende Abhandlung verfaßte, die man als morphologisch bezeichnen könnte. Vater Gabelentz schließt mit „Syntaktischen Bemerkungen“ (1861, 540), wo er u. a. feststellt: „Es würde nun noch übrig sein, auch seine (des Passivs) Stellung im Satz und sein Verhalten zu den anderen Satzgliedern nachzuweisen, oder eine allgemeine Syntax des Passivums aufzustellen.“

Über eine weitverzweigte Bibliographie gelangt man zu verschiedenen wechselnden Positionen — Ableitbarkeit, Konverse; NichtAbleitbarkeit; Konstruktionen; Funktionen u. a. — in einem sehr allgemeinen Rahmen zur Zeit wohl zu Chomsky (1981, 117) und Keenan (1985). Man spricht andererseits aber auch von einer Leningrader Schule (vgl. Chrakovskij 1978). Nach dem Gesagten stellt man fest, daß diese ersten Ansätze, auch wenn sie vorwiegend einzelsprachlich gebunden sind, die Probleme einer typologisch vergleichenden Syntax zumindest aufwerfen. Bemerkenswert ist, daß die europäischen Neuphilologien in dieser Phase dazu fast keine Beiträge geleistet haben.

3.

Die Typologie der Korrelationen

3.1.  Wie man heute im nachhinein feststellen kann, bildet das Jahr 1958 einen Markstein in der Geschichte der neueren Typologie. In diesem Jahr sind drei wichtige Arbeiten in die Welt gestellt worden: der Osloer Kongreßvortrag von Jakobson (1896—1982 ) über Implikationsgesetze, der Aufsatz von Benveniste (1902 —1976) über den Relativsatz und derjenige von Martinet über den Ergativ. Die Korrelationen werden heute implikativ begriffen, und zwar allgemein, während Jakobson seinerzeit nur an die Phonologie und an die Morphologie dachte. Man formuliert: „Wenn eine Sprache eine bestimmte Eigenschaft P hat, dann hat sie auch Q (aber nicht notwendigerweise umgekehrt).“ Diese Form ist jedoch nicht zwingend, und es können auch Eigenschaften als solche in Betracht gezogen werden. Eigenschaften in diesem Sinne sind Merkmale des Sprachsystems. Weitere Ausführungen zum Begriff des Merkmals findet man bei Ineichen (1991, 37). 3.2.  Im Rahmen der Syntax stellt sich nach dem Gesagten zunächst das Problem des Vorhandenseins oder Nicht-Vorhandenseins bestimmter Kategorien bzw. deren Organisation und Funktion. Nach Bartsch/Venne-

XVI. Syntaktische Typologie I: Allgemeines

1046

mann (1982 , 32 ) würde man in diesem Fall von Kategorialtypologie sprechen. Ein erfolgreiches und typologisch instruktives Kapitel dieser Art war in den letzten Jahrzehnten die Problematik des Relativsatzes. Man bezieht sich in diesem Zusammenhang auf die umfassende Arbeit von Lehmann (1984). Nach dem einleitend Gesagten, d. h. im Hinblick auf die mögliche Gestalt einer Allgemeinen Grammatik, ist die Liste jedoch noch weitgehend offen: Determination, Tempus bzw. Aspekt und Adverb, Vollständigkeit des Satzes, Modalisierung, Abtönung, Textualität u. a. Dabei bleibt zu bedenken, daß Untersuchungen dieser Art oft auch im Bereiche der kontrastiven bzw. konfrontativen Grammatik anzusiedeln sind. 3.3.  In den vorliegenden Zusammenhang gehört auch das Problem der Organisation des nuklearen Satzes. Gemeint ist damit nicht, wie in der Wortfolge oder in der Funktionstypologie, die Abfolge von Verb (V) mit Subjekt (S) und ev. Objekt (O), sondern die Tatsache, daß Sätze grundsätzlich anders konstruiert werden können. Das bekannteste Beispiel dafür ist die Ergativität, die in verschiedenen Ausprägungen über die Sprachen der Welt sehr verbreitet ist. Man erkennt, daß die Beziehung zwischen Nom./Acc. und Erg./Abs. (bzw. die Typologie von Akkusativ- vs. Ergativsprachen) nur über ein Tertium hergestellt werden kann. Diese sog. relationelle Typologie arbeitet deshalb mit zugrundeliegenden Strukturen, d. h. mit semantischen Rollen, die beim Ausdruck mit einer Kasuszuweisung versehen werden müssen. Man orientiert sich sehr gut in der Aufsatzsammlung von Plank (1985), die auf ein Symposion zurückgeht, das 1982 mit internationaler Besetzung in Hannover stattfand. Für andersgeartete Initiativen zumal aus dem sowjetischen Bereich verweisen wir auf Klimov (1985). Diese Erkenntnisse sind geistesgeschichtlich und sprachphilosophisch von großem Interesse, weil damit festgestellt wird, daß die syntaktische Form von Ausdrücken mit ihrer logischen Form — die man gewöhnlich auf europäische Sprachen bezieht („from translation into a European language“ — Harbsmeier 1983/85, 78) — nicht übereinzustimmen braucht. 3.4.  Für die Korrelativität in einem strikten Sinne gibt es im europäischen Bereich neuerdings das Beispiel des Romanischen. Diese

Gruppierung wird für gewöhnlich als Einheit aufgefaßt und typologisch nicht weiter spezifiziert. Wir zitieren hier deshalb exemplarisch die Studien von Körner (1987): Ergebnis: Unter Berücksichtigung des Toskanischen (ital. Schriftsprache) als typologisch inkonsistente Zwitterform, umfaßt die Peripherie (Typ II, dem in gewisser Hinsicht auch das Russische angehört) Spanisch, Portugiesisch, Gaskognisch, Süditalienisch mit Sardisch. Das Zentrum (Typ I) bilden Französisch, Norditalienisch, Okzitanisch, Katalanisch (?). Diese Typen sind — wie oft über die Welt — auch geolinguistisch abstufbar.

Die dazu zugrundeliegenden nen sind sieben (Körner 1987, 175):

Korrelatio-

1. Eine Sprache ohne obligatorische Subjektspronomina und deswegen auch ohne Wortarten wie dt. man oder frz. on (und engl. one oder people) kann nicht eine reine SVO-Sprache sein. 2 . Eine Sprache mit SV und VS bei intransitiven Verben neigt eher zur Verbalkategorie Aspekt als eine Sprache ohne solche VS. 3. Sprachen mit Objektspronominal-Redundanz („Objektskonjugation“ ), die ja Objektmarkierung (z. B. „präpositionaler Akkusativ“ ) voraussetzt, haben eine Wortstellungsfreiheit, die auch Objekt und Subjekt die Plätze tauschen läßt. 4. Sprachen, bei denen der Passivausdruck nicht an eine einzige und eindeutige Form gebunden ist, d. h. besonders Sprachen, die eine vitale Reflexivkonstruktion haben, können keine reinen SVOSprachen sein. 5. In Sprachen, in denen der dativische Agens-Ausdruck vital ist („mihi est“ -Sprachen) muß auch die Wortstellung VS verbreitet sein. 6. Sprachen mit „Negationsredundanz“ und ohne Unterscheidung zwischen einer „kein“ - und einer „nicht“ -Negation können nicht durchgehend Subjektspitzenstellung haben. 7. Solche Sprachen, in denen die Formen „zwischen“ Nomen und Verb (z. B. Infinitiv und Gerundium) zur selbständigen Prädikatsbildung neigen, können gut Subjekte den Verben nachfolgen lassen.

Weiter auszuloten wären in diesem Zusammenhang die Begriffe Konsistenz, Kompatibilität und Skalarität („mehr“ oder „weniger“ ). Dies gilt insbesondere auch für die Wortstellungstypologie, auf die wir im folgenden Abschnitt zu sprechen kommen.

4.

Die Wortfolgetypologie

4.1.  Die Wortfolgetypologie, die gelegentlich auch als positionelle Typologie bezeichnet wird, geht im wesentlichen auf Greenberg (1963, 21966) zurück. Eine Darstellung der weiteren Entwicklung der Disziplin findet

54.  Geschichte der syntaktischen Typologie

man bei Ineichen (1991, Kap. 9 und 10), eine mehr problemorientierte bei Bartsch/Vennemann (1982 , 32 —36). Im Hinblick auf die Indogermanistik berücksichtigt man vor allem Lehmann (1972 ). Für zahlreiche Autoren ist die Wortfolge- oder Wortstellungstypologie zur Zeit die syntaktische Typologie schlechthin (vgl. Ramat 1984). Im Falle des Neufranzösischen, wo sich die SVO-Struktur immer stringenter durchsetzt (Ineichen 1984, 17) sprach schon Foulet (3192 8) von dem selben „canevas monotone: sujet, verbe, objet“ . Die Entwicklung läßt sich vom Altfranzösischen her genau verfolgen. Man liest dazu bei Foulet (a. a. O.), der allerdings mehr unter stilistischen Gesichtspunkten schreibt: „Et tout l’art de l’écrivain consiste précisément à inventer des variations sur cette formule unique que lui impose la langue. Il n’en est pas de même au moyen âge. Les variations sont bien moins ingénieuses, mais aussi la formule n’est pas du tout immuable; les trois termes sont encore les mêmes, bien entendu, mais leur enchaînement n’est pas fixé une fois pour toutes. Entre trois éléments donnés, il y a six combinaisons possibles, et ces six combinaisons nous les trouvons toutes en ancien français. Elles n’y sont pas également privilégiées, mais quatre au moins sont parfaitement courantes.“

Bemerkenswert ist im Falle des Slavischen, daß Berneker (1900) in seiner stark indogermanistisch orientierten Habilitationsschrift die Untersuchung auf die „Stellung des Verbums im Satze“ konzentriert. (Untersucht werden Prosatexte in Russisch, Kleinrussisch, Bulgarisch, Serbisch, Slovenisch, Tschechisch, Polnisch, Litauisch.) Berneker (1900, 155) findet sechs Stellungsregeln, betreffend Verbum, Enklitika, Dativ/Akkusativ, Attributstellung, Infinitiv, Hervorhebung. Man liest (a. a. O.) zum Verb: „1) Das Verbum steht ursprünglich am Anfang des Satzes (vor dem Subj.) oder am Ende desselben. Die Folge: Subj.-Verb-Obj., von mir als MSt [Mittelstellung] bezeichnet, ist in den ältesten Sprachperioden nur occasionell; sie tritt nur ein, wenn das Subj. betont ist. Allmählich gewinnt sie jedoch immer größere Ausdehnung, bis sie in den heutigen Sprachen als mehr oder weniger gleichberechtigt neben der ASt [Anfangsstellung] und der ESt [Endstellung] erscheint“ .

4.2.  Die 45 Universalien Greenbergs sind empirisch durch multilingualen Vergleich ermittelt; sie sind teils vorkommenshäufig und stark implikativ, teils zumindest mit „more than chance frequency“ nachweisbar. Sie ha-

1047

ben dieselbe Form wie die Regularitäten, die wir oben zur Abwechslung aus dem Bereich des Romanischen zitierten. Es gibt in diesem Bereich mittlerweile viele einzel- und übereinzelsprachliche Untersuchungen mit bemerkenswert explikativer Kraft, darunter im Hinblick auf den typologischen Wandel auch historische (vgl. dazu Li 1975). Auf theoretischer Ebene maßgeblich sind die Studien zur sog. natürlichen Seralisierung von Vennemann (Bibliographie bei Bartsch/Vennemann 1982 ) und die Arbeiten von Keenan (z. B. 1982 ), der sich mit dem gegenseitigen Verhältnis von syntaktischer (SF) und logischer Form (LF) befaßt. Es kommt auch vor, daß die wissenschaftliche Stringenz der Wortfolgetypologie in Abrede gestellt wird. Als Beispiel dafür nehme man Abraham/Scherpenisse (1983, wo die Überprüfung anhand von Daten aus dem Deutschen und dem Niederländischen erfolgt).

5.

Die Funktionstypologie

5.1.  Im Umkreis der Prager Schule — historisch gewöhnlich punktuell bezogen auf Mathesius (192 8) — entwickelte sich die Analyse der funktionalen Satzperspektive. Mathesius soll durch ein Buch des klassischen Philologen Weil (1844) zu seinen Studien angeregt worden sein. Es geht dabei in einem pragmatischen Sinne um die Verfahren, mit deren Hilfe die Aufmerksamkeit der jeweiligen Adressaten (Hörer, Leser) gelenkt wird. Die einschlägigen Begriffe, die teilweise individuell nuanciert werden, sind bekannt: Thema vs. Rhema, Notum bzw. Datum vs. Novum, in Amerika seit Hockett (1958, 2 0) Topic vs. Comment. Daraus resultieren verschiedene Arten der Gewichtung, deren Syntax sich in den Einzelsprachen sehr unterschiedlich gestaltet. Man denke an die Topikalisierung, die neue Information hervorhebt, oder an die Dislokation, die sich auf bekannte Information bezieht, an die Passivierung und an verschiedene Regularitäten der Inversion. Letztlich geht es dabei um die Gestaltung von Textualität und deren kommunikativer Dynamik. (Vgl. dazu Halliday 1970, Daneš 1974). Eine Dokumentation des hier skizzierten Problemkomplexes würde eine eigene Bibliographie erfordern. In diesem Rahmen ergäbe sich eine Typologie der Verfahren. Dies gilt auch diachron, wo die Diskursfunktionalität — entsprechend dem sog. Subjekt-Topik-

XVI. Syntaktische Typologie I: Allgemeines

1048

Zyklus — den Stellungswandel beeinflussen kann. (Vgl. z. B. Li-Thompson 1975, 169, im Falle der Definitheit des Nomens im Chinesischen). 5.2.  Die eben besprochene Satzperspektive, d. h. die Möglichkeit der Verwendung von adressatenbezogenen Diskursstrategien, gehört in den Bereich von Sprachen mit Subjekt-Prädikat-Struktur. Sätze dieser Art sind in ihrem propositionalen Gehalt (Sprop) objektiv, solange sie nicht sprecherbezogen modalisiert sind (Smod). Also z. B.: „Ich sage [Smod], daß der Sohn des Vaters [Subj.] krank ist [Präd.]“ . Andererseits gibt es Sprachen ohne Subjekt-Prädikat-Struktur, von denen behauptet wird, sie seien mehr diskursorientiert. Dementsprechend besitzen solche Sprachen z. T. andere Kodierungseigenschaften. Mit einem europäischen Grammatikersatz illustriert: „Der Sohn des Vaters ist krank“ heißt dann etwa „Hinsichtlich des Vaters [Topik], der Sohn ist krank [Komment].“ Die beiden Extreme der Skalarität sind deshalb entweder subjektprominent oder topikprominent. In der Geschichte der syntaktischen Typologie ist diese Unterscheidung der bislang letzte wichtige Punkt. Die einschlägigen Eigenschaften und deren Typologie findet man bei Li/Thompson (1976). (Vgl. auch Art. 56 in diesem Band.) Der Sammelband von Li (1976) dokumentiert ein Symposium, das 1975 in Santa Barbara, Kalifornien, stattgefunden hat. Unnötig zum Schluß anzumerken, daß im Rahmen der Funktionstypologie auch die Korrelative und die Wortfolgetypologie ihrerseits wieder zu Worte kommen.

6.

Literatur

Abraham, Werner, und Wim. Scherpenisse. 1983. Zur Brauchbarkeit von Wortstellungstypologien mit Universalanspruch. Sprachwissenschaft 8, 291—355. Bakalla, M. H. 1983. Arabic Linguistics. An Introduction and Bibliography. London. Bartsch, Renate, und Theo Vennemann. 1982 . Grundzüge der Sprachtheorie. Eine linguistische Einführung. Tübingen. Behaghel, Otto. 1932 . Deutsche Syntax. Eine geschichtliche Darstellung. Bd. IV: Wortstellung. Periodenbau. Heidelberg. Benveniste, Emile. 1957—58. La phrase relative, problème de syntaxe générale. Bulletin de la Société de Linguistique 53, abgedruckt in: Ders., Problèmes de linguistique générale. Paris: Gallimard

1966, 208—22. Berneker, Erich. 1900. Die Wortfolge in den slavischen Sprachen. Berlin. Chiang-Jen Chang, Roland. 1977. Co-Verbs in Spoken Chinese. Taiwan. Chomsky, Noam. 1981. Lectures on Government and Binding. Studies in Generative Grammar 9. Dordrecht. Chrakovskij, V. S. 1978. Problemy teorii grammatičeskogo zaloga. Leningrad. Comrie, Bernard. 1981. Language Universals and Linguistic Typology. Syntax and Morphology. Oxford. Danes, F. 1974. Papers on Functional Sentence Perspective. Janua Linguarum-Series Minor 147. Den Haag, Paris. Dezső, Lásló. 1982 . Studies in Syntactic Typology and Contrastive Grammar. Budapest. Dirven, René, und Günter Radden (Hrsg.). 1982 . Issues in the Theory of Universal Grammar. Tübingen. Drach, Erich.31940. Grundgedanken der deutschen Satzlehre. Frankfurt am Main: Diesterweg. Nachdruck Darmstadt: Wissenschaftliche Buchgesellschaft 1963. (1. Aufl. 1937). Dressler, W. U. 1979. Reflections on Phonological Typology. Acta Linguistica Academiae Scientiarum Hungaricae 29, 259—73. Foulet, Lucien.3192 8. Petite syntaxe de l’ancien français. Paris. 11919. von der Gabelentz, Hans Conon. 1861. Über das Passivum. Eine sprachvergleichende Abhandlung. Abhandlungen d. philos.-hist. Cl. d. Königlich Sächsischen Gesellschaft d. W., Bd. 3, 449—546. von der Gabelentz, Georg. 1869. Ideen zu einer vergleichenden Syntax. Wort- und Satzstellung. Zs. f. Völkerpsychologie und Sprachwissenschaft 6, 376—84. —. 1875. Weiteres zur vergleichenden Syntax. Wort- und Satzstellung. Zs. f. Völkerpsychologie und Sprachwissenschaft 8, 129—65, 300—37. —. 1881. Chinesische Grammatik, mit Ausschluß des niederen Stils und der heutigen Umgangssprache. Berlin (Nachdruck: Deutscher Verlag der Wissenschaften 1953). —. 1891. Die Sprachwissenschaft, ihre Aufgaben, Methoden und bisherigen Ergebnisse. Leipzig: T. O. Weigel Nachfolger. (Durchgesehener Nachdruck der zweiten Auflage von 1901, Tübingen: Narr 1969). Givón, Talmy. 1984. Syntax. A Functional-Typological Introduction. Vol I. Amsterdam/Philadelphia. Greenberg, Joseph H. (Ed.). 1963. Universals of Language. . Cambridge, MA Press. Man zitiert Greenberg

54.  Geschichte der syntaktischen Typologie

selbst gewöhnlich nach 21966, 73—113. Der Titel des einschlägigen Aufsatzes lautet: „Some Universals of Grammar with Particular Reference to the Order of Meaningfull Elements“ . Halliday, M. A. K. 1970. Language Structure and Language Function. In Lyons 1970, 140—65. Harbsmeier, Christoph. 1983/85. Where Do Classical Chinese Nouns Come From? Early China 9/10, 77—163. Hockett, Charles F. 1958. A Course in Modern Linguistics. New York, 101966. Ineichen, Gustav. 1991. Allgemeine Sprachtypologie. Ansätze und Methoden. Darmstadt, 11979. —. 1984. Pour une caractérisation typologique du français. In Linguistica 14, 11—25. —. 1988. Die Sprachtypologie in der Linguistik von Coseriu. In Energeia und Ergon, Festschrift für Eugenio Coseriu, Tübingen, Bd. 3, 31—37. Keenan, Edward. 1982 . Parametric Variation in Universal Grammar. In Dirven/Radden 198 2 , 11—74. —. 1985. Passive in the world’s languages. In Shopen 1985, I, 243—81. Klimov, Georgij A. 1985. Zur Kontensiven Typologie. In Plank 1985, 175—90. Kölver, Ulrike. 1985. Kasusrelationen im Birmanischen. In Plank 1985, 195—212. Körner, Karl-Hermann. 1987. Korrelative Sprachtypologie. Die zwei Typen romanischer Syntax. Stuttgart. Lehmann, Christian. 1984. Der Relativsatz. Typologie seiner Strukturen. Theorie seiner Funktionen. Kompendium seiner Grammatik. Tübingen. Lehmann, Winfred P. 1972 . Contemporary Linguistics and Indo-European Studies. Publications of the Modern Language Association 87, 976—93. Li, Charles N. (Ed.). 1975. Word Order and Word Order Change. Austin and London. —. (Ed.). 1976. Subject and Topic. New York—San Francisco—London.

1049

—.; Thompson, Sandra A. 1976. In: Li 1976, 457—89. Lyons, John (Ed.). 1970. New Horizons in Linguistics. Hardmondsworth. Martinet, André. 1958. La structure ergative. Journal de psychologie normale et pathologique 55, 377—92 ; abgedruckt in: Ders., La linguistique synchronique. Etudes et recherches. Paris: Presses Universitaires de France 1965, 206—22. Mathesius, Vilém. 192 8. On linguistic characterology. Actes du Premier Congrès International de Linguistes à la Haye. Paul, Hermann. 1980. Prinzipien der Sprachgeschichte. Tübingen 81968. Plank, Frans (Ed.). 1985. Relational Typology. Trends in Linguistics, Studies and Monographs 2 8. Berlin—New York—Amsterdam. Ramat, Paolo. 1984. Linguistica Tipologica. Studi linguistici e semiologici 2 1. Bologna: Il Mulino. (Vgl. die Rezension von Gustav Ineichen, in: Kratylos 37. 1987, 170—72). Reckendorf, Hermann. 1898. Die syntaktischen Verhältnisse des Arabischen. Leiden (Photomechanischer Nachdruck Leiden 1967: Brill). Shopen, Timothy (Ed.). 1985. Language typology and syntactic description. 3 vol., Cambridge—London—New York—New Rochelle—Melbourne— Sydney. Schleicher, August. 1865. Die Unterscheidung von Nomen und Verbum in der lautlichen Form. Abhandlungen d. philos.-hist. Cl. d. Königlich Sächsischen Gesellschaft d. W., Bd. 4, 497—587. Weil, Henri. 1844. De l’ordre des mots dans les langues anciennes comparées aux langues modernes. Paris, 21869, 31879. Wundt, Wilhelm. 1900. Völkerpsychologie. Eine Untersuchung der Entwicklungsgesetze von Sprache, Mythus und Sitte. Erster Band. Die Sprache. Erster Theil. Leipzig: Verlag von Wilhelm Engelmann. [Des ersten Bandes zweiter Teil wird im Herbst 1900 erscheinen].

Gustav Ineichen, Göttingen (Deutschland)

1050

XVII. Syntaktische Typologie II: Ansätze und Übersichtsdarstellungen Syntactic Typology II: Approaches and Surveys

55. Word Order Typology 1. 2. 3. 4. 5. 6.

1.

History The Basic Pattern Explanations for the Basic Pattern Other Patterns Further Issues References

History

Although an awareness of a relationship between the order of v erb and object and other word order characteristics dates back to at least the nineteenth century, it is the work of Greenberg (1963) — sometimes cited as Greenberg (1966), the apparently unrev ised second edition — that is generally v iewed as marking the beginning of an interest in word order typology. The basic idea of word order typology is that there is an association or correlation between a number of word order characteristics, so that giv en a single word order characteristic of a language, like the order of v erb and object, one can predict, at least in a statistical sense, a v ariety of other characteristics of the language. Greenberg proposed 45 linguistic univ ersals, based primarily on a sample of 30 languages, though clearly formulated with an awareness of the properties of other languages as well. 28 of these univ ersals refer to the order or position of grammatical elements, the others dealing exclusiv ely with morphological characteristics. The majority of Greenberg’s univ ersals are implicational, taking the basic form “If P then Q”. Another important feature of many of his univ ersals is that they are statistical, being stated to hold “with ov erwhelmingly greater than chance frequency”. The majority of them are not formulated in statistical terms, but many of these do in fact hav e exceptions outside his 30-language sample.

Many of Greenberg’s univ ersals refer to the basic order of subject, object, and v erb. His typology assumes that languages belong to one of three types, SVO (e. g. English), SOV (e. g. Japanese) and VSO (e. g. Welsh). He described the other three possibilities (VOS, OVS, OSV) as not occurring at all or being excessiv ely rare, but all three of these are now claimed to exist. Though not common, VOS languages are found in v arious parts of the world (cf. Keenan 1978; Tomlin 1986). The two types of object-initial languages, OVS and OSV, are so far primarily attested in South America (cf. Derbyshire/ Pullum 1981; 1986), though a conv incing case for an OSV language has yet to be presented in detail. Because VOS languages pattern like VSO languages in their characteristics, it is common to refer to these together as v erb-initial languages. This term is potentially misleading since certain elements like negativ e words or question particles often precede the v erb in such languages; the term should be understood in the sense that among the set of elements consisting of the v erb, subject, and object, the v erb occurs first. Though OVS and OSV languages seem to pattern like SOV languages, ev idence is so far too meagre to draw firm conclusions. Because of the rarity of OSV languages, it is common to refer to SOV languages as v erbfinal. The notion of v erb-medial language, howev er, does not seem useful, since SVO and OVS languages do not seem to form a natural class. The terms predicate-initial and predicate-final would be more appropriate than verb-initial and verb-final, since nonv erbal predicates (e. g. nouns, locativ es) normally occur initially in v erb-initial languages and finally in v erb-final languages, but the latter terms are well-established. Implicit in many of Greenberg’s univ ersals and in some of his discussion is the idea that

55.  Word Order Typology

many of his univ ersals reflect a more fundamental typology of two “ideal” language types, OV and VO. This idea is made explicit in the later work of Lehmann (1973; 1978) and Vennemann (1973; 1974a; 1974b; 1976). The basic idea is that, at least in the ideal case, VO languages are the mirror image of OV languages, all word order characteristics being rev ersed. This basic pattern is discussed in 2. The work of Lehmann and Vennemann (as well as v arious others during the 1970’s) was also characterized by attempts to explain this fundamental pattern, to find a single principle which would account for the full range of pairs of elements whose order correlates with that of the object and v erb. Some of these possible explanations are discussed in 3. The work of Lehmann and Vennemann differed from that of Greenberg in two other important respects. First, although Greenberg’s study was really only a pilot study (being based primarily on a 30-language sample), little attempt was made during the 1970’s to expand the empirical basis for the univ ersals or the basic pattern. A number of correlations or patterns not proposed by Greenberg were claimed to be true and hav e since become widely assumed by linguists. Yet little systematic ev idence was presented in support of these claims. Unfortunately, many of these assumptions are not supported by the av ailable empirical ev idence. Among these is the idea that OV languages tend to place the adjectiv e before the noun. In fact, as illustrated by Greenberg’s 30-language sample, and as is more fully discussed in Dryer (1988b), both orders of adjectiv e and noun are common among OV languages. A second difference between the work of Greenberg and that of Lehmann and Vennemann is that the basic pattern of the two ideal types assumed by the latter two linguists, consistent OV and consistent VO, represents a statistical generalization of a far weaker sort than Greenberg’s statistical implicational univ ersals. While Greenberg’s univ ersals admitted at most a few exceptions in his sample, it is clear ev en from his sample that there are far more exceptions to the basic pattern, that there are many languages that are not one of the two ideal types. This can be v iewed, howev er, as an inev itable consequence of isolating a more fundamental principle. If the array of existing types reflects the interaction of such a principle with a number of competing principles, one would expect many exceptions to the fundamental

1051

principle. An important aspect of the work of Lehmann and Vennemann was to bring out its diachronic implications. These are discussed elsewhere in this handbook in the article on word order change. The more recent work by Hawkins (1979; 1980; 1983) contrasts with that of Lehmann and Vennemann in a number of ways. First, he argues for the use of absolute (nonstatistical) implicational univ ersals, achiev ing this in general by formulating more complex multiterm implicational univ ersals of the form “If P then if Q then R”. He also argues for distinguishing these nonstatistical implicational univ ersals from what he calls distributional principles, which predict the relativ e frequency of different language types. An example of the latter is his principle of Cross-Category Harmony, discussed in 2. He also argues against Greenberg’s three-way SOV/SVO/ VSO typology because SVO languages are supposedly v aried in their properties (though cf. Dryer 1991) and because adposition type — whether a language uses prepositions or postpositions — seems to be a better predictor of other word order characteristics. Hawkins’ work is the first since Greenberg’s article to examine the relev ant empirical ev idence, based on an expansion of the data in Greenberg’s appendix.

2.

The Basic Pattern

The work of Lehmann and Vennemann focuses on the fact that many of Greenberg’s univ ersals seem to reflect a single fundamental underlying principle whereby a large number of pairs of grammatical elements tend to be ordered in a way that is predictable from the order of the object and v erb. For example, genitiv es tend to precede the noun in OV languages, and follow in VO languages. Similarly, OV languages tend to employ postpositions, while VO languages tend to employ prepositions. In other words, objects of adpositions tend to precede the adposition in OV languages, but to follow in VO languages. Hawkins (1983) follows Vennemann (1974a; 1974b) in referring to the elements whose order correlates with that of the object and v erb as operators and operands respectiv ely, but since this terminology reflects Vennemann’s earlier explanation for the correlations, and since the application of this terminology is sometimes found confusing (often being the opposite of what many

1052

XVII. Syntaktische Typologie II: Ansätze und Übersichts-darstellungen

semantic theories claim), the terms object patterner and verb patterner will be used here for those elements whose order tends to correlate with that of the object and v erb respectiv ely. Thus genitiv es are object patterners, while the nouns they modify are v erb patterners, and adpositions are v erb patterners while the objects of adpositions are object patterners. The central question of this section is therefore: What is the set of pairs of object patterners and v erb patterners? The answers giv en here are the results of a project by the author of this article that has inv olv ed the collection of word order data from ov er 600 languages. For reasons of space, the ev idence supporting the conclusions stated is not giv en here. For details, see Dryer (1992). Before we can ask what pairs of elements are object patterners and v erb patterners, we must first address the question of what it means to say that pairs of elements tend to occur in the same order as the object and v erb. In his discussion of Vennemann’s work, Hawkins (1980) interprets Vennemann as claiming that the majority of languages of the world fall into one of his two ideal types, either consistently ordering object patterners before v erb patterners, or consistently ordering v erb patterners before object patterners. It is easy to show, at least for the set of pairs of elements Vennemann claims to be operators (object patterners) and operands (v erb patterners), that such a claim would be false: a clear majority of the languages of the world are inconsistent in at least one respect. Howev er, this seems to be a rather ungenerous interpretation of what Vennemann is claiming, although it does point to the fact that Vennemann (like others) is not clear as to exactly what he is claiming. An alternativ e interpretation of Vennemann suggested by Hawkins (1983) is that consistent languages are more common than languages with (exactly) one inconsistency, and that languages with one inconsistency are more common than languages with two inconsistencies, and so on. A rather different kind of interpretation is that for each pair of elements, object patterners more often precede v erb patterners in OV languages, while they more often follow in VO languages. The following pairs of elements can be shown to be object patterners and verb patterners by this criterion: object genitive noun

patterner

verb patterner noun phraseadposition

adverb verb adpositional phraseverb main v erbauxiliary v erb main v erbmodal v erb v e auxiliary main v erbnegati verb verb want verb verb cause standard of comparisonadjective standard of comparisonmarker sentence question particle clause v ad erbial subordinator noun plural word By plural word is intended a word that serves the same function as plural affixes in other languages, but which is a separate word modifying the noun. By auxiliary v erb is intended one indicating tense or aspect. Negativ e auxiliaries are negativ e words distinct from the main v erb that exhibit v erbal characteristics (cf. Dahl 1979). The abov e list inv olv es pairs of elements that exhibit bilateral correlations with the order of object and v erb, where two pairs of elements X1 Y1 and X2 Y2 are said to exhibit a bilateral correlation if languages which are X1 Y1 tend to be X2 Y2 and languages which are X2 Y2 tend to X1 Y1. A number of other pairs of elements exhibit unilateral correlations with the order of object and v erb, where two pairs of elements X1 Y1 and X2 Y2 are said to exhibit a unilateral correlation if languages which are X1 Y1 tend to be X2 Y2 or languages which are X2 Y2 tend to be X1 Y1, but not both. An example of a pair of elements that exhibits a unilateral correlation is the order of relativ e clause and noun. Although VO languages almost nev er place relativ e clauses before the noun, both orders of relativ e clause and noun are common among OV languages (cf. Dryer 1992). Such unilateral correlations, or asymmetries as Hawkins (1985) calls them, are naturally described by unilateral implicational univ ersals, e. g. “If the language is VO, then the relativ e clause follows the noun.” Giv en the unilateral nature of the correlation, one might ev en question whether it is appropriate to describe the situation as a correlation at all. Howev er, the term is appropriate because a plausible interpretation of the facts is that there is a principle by which prenominal relativ e clauses are preferred in OV languages, but which is in conflict with a principle whereby postnominal relativ e clauses are univ ersally preferred (cf. Hawkins 1988, 1990).

55.  Word Order Typology

1053

Furthermore, it remains the case that prenominal relativ e clauses occur significantly more often in OV languages than they do in VO languages. It will be assumed here that two elements X and Y are object patterner and v erb patterner if X precedes Y more often in OV languages than in VO languages, i. e. if the proportion of languages in which X precedes Y is greater among OV languages than it is among VO languages. By this criterion, relativ e clauses are object patterners and the nouns they modify are v erb patterners, despite the fact that OV languages with postnominal relativ e clauses are as common as OV languages with prenominal relativ e clauses. The following is a list of pairs of elements which can be shown to be object patterners and v erb patterners in this weaker, unilateral, sense (though not in the bilateral sense): object patterner relative clause noun noun clause stem subject

verb patterner noun article numeral complementizer affix verb

VO languages tend to hav e prenominal articles and numerals, and clause-initial complementizers, while OV languages tend to employ suffixes and to be SV rather than VS (i. e. OVS is rare). But the rev erse tendencies do not hold: both orders of article and noun and of numeral and noun and both positions of complementizers are common among OV languages (though complementizers are not common in OV languages); both prefixes and suffixes are common among VO languages; and both orders of subject and v erb are common among VO languages. The tendency for numerals modifying nouns to precede the noun in VO languages is not a strong one, and exhibits a strong areal split: VO languages in Africa generally place the numeral after the noun, while VO languages elsewhere in the world generally place it before the noun (cf. Dryer 1989). Affixes in general behav e as v erb patterners (i. e. suffixing is more common in OV languages than it is in VO languages), although different classes of affixes behav e differently. In particular, pronominal object affixes on v erbs exhibit the opposite correlation (they are prefixes more often in OV languages than they are in VO languages), and pronominal possessiv e af-

fixes on nouns exhibit a similar though weaker trend (they are also prefixes more often in OV languages than they are in VO languages). It turns out that ev en with this rather weak interpretation of what it means for a pair of elements to correlate in order with the object and v erb, a number of pairs of elements that are often assumed to be object patterners and v erb patterners are not in fact such. An important example of such is adjectiv e and noun. Vennemann and Lehmann both assume that adjectiv es precede the noun more often in OV languages than they do in VO languages, but, as discussed by Dryer (1988, 1992), this is not so. Two facts apparently led people to this mistaken belief. First, there is a strong tendency in a large area of Asia, stretching from Turkey to Japan, and from southern India to Siberia, for OV languages to place the adjectiv e before the noun (cf. Masica 1976). But this is apparently due to remote areal or genetic factors: outside this area, it is considerably more common for the adjectiv e to follow the noun in OV languages. Second, one might be led to conclude that the order of adjectiv e and noun correlates with the order of object and v erb if one extrapolates from Greenberg’s Univ ersal 17: “With ov erwhelmingly more than chance frequency, languages with dominant order VSO hav e the adjectiv e after the noun.” But in fact it turns out that this univ ersal is incorrect, and simply an accidental property of Greenberg’s sample, in which all six VSO languages happened to place the adjectiv e after the noun. It is somewhat more common for VSO languages to place the adjectiv e after the noun, but the same is true for SOV languages as well. In short, adjectiv es are not object patterners and the nouns they modify are not verb patterners. Similar comments apply to a number of other pairs of elements that are widely assumed to correlate in order with the order of object and v erb. It is widely believ ed that modifiers in general tend to precede the noun in OV languages, follow in VO languages. But numerals, articles, and plural words exhibit a trend in the opposite direction, preceding the noun more often in VO languages than they do in OV languages. (Cf. Dryer (1992) for details.) Demonstrativ es, like adjectiv es, do not exhibit any significant correlation at all. In short, the only modifiers of nouns that are object patterners are genitiv es and relativ e clauses. Auxiliaries are v erb pat-

1054

XVII. Syntaktische Typologie II: Ansätze und Übersichts-darstellungen

terners and their accompanying main v erbs are object patterners if one restricts attention to clearly v erbal auxiliaries, ones like auxiliaries in English and German which exhibit v erbal characteristics and are presumably best treated as a subclass of v erbs. But if one examines nonv erbal auxiliaries, i. e. particles or clitics indicating tense/aspect, one finds no correlation. Similar comments apply to negativ e words. If they are v erbal, they tend to be v erb patterners, but if not, there is no correlation; they tend to precede the v erb in both OV and VO languages. Finally, the order of qualifier (adv erbial modifiers of adjectiv es) and adjective exhibits no correlation either. Hawkins (1980; 1983) approaches the basic pattern in a rather different way. His principle of Cross-Category Harmony claims (loosely speaking) that the most frequent language types are those in which the proportion of dependents preceding one category of head will be the same as the proportion of dependents preceding other categories of heads. In some ways, this makes similar claims to those of Vennemann, predicting that among languages in which all dependents precede the v erb, the most common type will be that in which all dependents precede the noun. The results reported abov e might lead one to expect this prediction to be false for my data, since articles, numerals, and plural words precede the noun less often in OV languages than they do in VO languages, and since adjectiv es and demonstrativ es exhibit no correlation. In fact, howev er, this language type, in which all modifiers precede the noun, is one of the most common types of SOV languages, outnumbering most (and perhaps all) indiv idual “inconsistent” language types. In other words, it represents a minority of SOV languages, but outnumbers most other possible types. Howev er, most of these SOV languages in which all modifiers precede the noun are spoken in Europe and Asia; relativ ely few SOV languages outside this area are of this type. Hawkins’ principle also predicts that the most common type among v erb-initial languages will be that in which all modifiers follow the noun. This prediction is definitely not borne out: among v erb-initial languages, the types with articles and/or numerals preceding the noun are definitely more common (cf. Dryer 1991). Hawkins’ principle differs from prev ious work in predicting that in languages in which the v erb comes in the middle of the clause,

the most common type should be that in which the noun comes in the middle on the noun phrase, that it should be less common for SVO languages to place all modifiers before the noun or all of them after. It thus places SVO languages between v erb-final and v erb-initial languages, in contrast to Lehmann and Vennemann, who treated SVO languages like v erb-initial languages as belonging to the general category of VO languages. Although the ev idence seems to bear this out to a certain extent, this may be largely due to the order of noun and genitiv e: while the genitiv e normally precedes the noun in v erbfinal languages and follows in v erb-initial languages, both orders of noun and genitiv e are common among SVO languages. For other characteristics, howev er, SVO languages clearly pattern more like v erb-initial languages (cf. Dryer 1991). For example, SVO languages are ov erwhelmingly prepositional rather than postpositional. In so far as SVO languages are less consistent than v erbinitial languages, this may simply be due to the number of SVO languages which were recently v erb-final languages (cf. Finnish). Questions also arise regarding the representativ eness of Hawkins’ sample. Despite being larger than Greenberg’s 30-language sample, it suffers from genetic and areal biases (cf. Dryer 1989). In a number of cases it is not clear whether the greater frequency in his sample of one language type relativ e to another is not simply due to certain genetic groups being well-represented in his sample. A further issue largely ignored in the literature is whether the v arious pairs of object patterners and v erb patterners all correlate with each other. The early literature assumes a special status for the order of object and v erb, and the definitions of object patterner and v erb patterner employed here follow that tradition. But no ev idence for such a special status has ev er been prov ided, and Hawkins (1983) argues that adposition type is in fact a better predictor, at least for implicational univ ersals. It is quite possible, of course, that no pair of elements has a priv ileged status. It is also important to distinguish priv ileged status as a predictor of word order properties from a common trigger of word order change. Adposition type might be a good predictor of word order properties, because it seems to be more stable than the order of object and v erb, but an infrequent trigger of word order changes, for precisely the same reason. It has ev en been suggested (cf. Anti-

55.  Word Order Typology

nucci/Duranti/Gebert 1979) that the order of relativ e clause and noun is a major trigger for word order change; but though prenominal relativ e clauses are a good predictor of OV order, postnominal relativ e clauses do not predict anything. Ev en if some special status should be assigned to one pair of elements, like the object and v erb, the question arises as to whether the v arious pairs of object patterners and v erb patterners correlate not only with the order of object and v erb but also with each other. The v arious attempts to isolate a single underlying principle assume that they do. For example, if v arious pairs of elements correlate in order with the order of object and v erb because they all inv olv e pairs of dependents and heads (cf. 3.1.), then we would expect each pair to correlate with each other for the same reason. On the other hand, Justeson/Stephens (1987) argue that this is not the case, that the basic pattern actually consists of chains of correlations, where A correlates with B, and B with C, but where there is no direct correlation between A and C, where the apparent correlation between A and C is due entirely to the correlations between A and B and between B and C. This has important implications for explaining the basic pattern. Justeson and Stephens’ conclusions are based on a sample of questionable quality, howev er, and it remains to be inv estigated what results a more careful sample would provide.

3.

Explanations for the Basic Pattern

3.1. Syntactic Explanations Considerably more attention has been giv en in the literature to attempts to explain the basic pattern than to determining which pairs of elements are in fact object patterners and v erb patterners. Unfortunately, many of the proposals for explaining the basic pattern suffer from incorrect assumptions as to which pairs of elements do pattern with the object and v erb, as discussed in 2. Perhaps the most popular explanation is that the basic pattern reflects a tendency to consistently order grammatical heads with respect to their dependents (or modifiers). Greenberg himself suggested that some such principle underlies the basic pattern, and the idea is widely assumed in the literature. Hawkins’ principle of Cross-Category Harmony is formulated in terms of such notions. As noted in 2., howev er, there are many categories of dependent

1055

whose order with respect to their head does not correlate with the order of object and v erb, at least under the most common assumptions as to what elements are heads and what are dependents. Among modifiers of nouns, only genitiv es and relativ e clauses are object patterners. Adjectiv es, demonstrativ es, articles, plural words and numerals do not precede nouns more often in OV languages than they do in VO languages, and in fact the last three of these categories exhibit precisely the opposite tendency. The order of negativ e particles and tense-aspect particles with respect to v erbs does not correlate with the order of object and v erb. Nor does the order of qualifier (adv erb modifying adjectiv e) and adjectiv e. Conv ersely, a number of v erb patterners are not heads, at least under the most common assumptions, such as articles, complementizers, question particles, numerals, plural words, and affixes. There are, howev er, theories in which these are heads; Vennemann (1976) and Vennemann/Harlow (1977) offer an account of the basic pattern in terms of a categorial grammar in which the first three of these are heads. There are other v erb patterners whose status as heads is at least controv ersial; this includes adpositions and auxiliary v erbs. Again, only theories which analyse these as heads could account for the basic pattern in terms of consistent ordering of heads and dependents. But the number of categories of dependents which fail to be object patterners presents a problem for any claim that there is a general tendency for dependents to precede heads in OV languages, and to follow in VO languages. There are a number of ways in which one might distinguish those dependents which are object patterners from those which are not. One possibility is the distinction between complements and noncomplements, though this distinction is heav ily theory-dependent (cf. 3.2.). Another possibility (cf. Dryer 1992) is the distinction between phrasal dependents and nonphrasal dependents. It is also possible that certain categories, like numerals, are heads in some languages, dependents in others. (Babby (1987) argues that numerals in Russian hav e changed from being heads to being dependents. This suggests that the correlations with the order of numeral and noun can only be understood if we hav e ev idence as to whether numerals are heads or dependents in a given language. Lehmann (1973) offers a rather different kind of syntactic explanation for the basic

1056

XVII. Syntaktische Typologie II: Ansätze und Übersichts-darstellungen

pattern. According to his principle, “modifiers are placed on the opposite side of a basic syntactic element from its primary concomitant” (Lehmann 1973, 48). One of the putativ e correlations that this is intended to account for is the presumed correlation between the order of adjectiv e and noun and the order of object and v erb. The idea is that the primary concomitant of an object is the v erb, so an adjectiv e modifying the object noun is on the opposite side of the noun from the v erb if it precedes the noun in an OV language. Howev er, as noted in 2., there is in fact no correlation between the order of adjectiv e and noun and the order of object and v erb. Similarly, Lehmann’s principle is intended to account for the presumed fact that negativ es follow the v erb in OV languages. Howev er, although such a tendency exists for negativ e affixes and negativ e auxiliaries, it is not true for negativ e particles. Ev en with some pairs of elements whose correlation in order with that of object and v erb has long been widely-assumed, such as adv erbials (cf. Vennemann 1974a; 1974b), Lehmann’s principle seems to make the wrong prediction. Since objects are the primary concomitant of v erbs (on his own assumptions), his principle would seem to predict that adv erbials should tend to occur on the opposite side of the v erb from the object. But exactly the opposite is the case. The discussion here assumes that the notions consistent OV language and consistent VO language are tied to a pattern that is observ able among the languages of the world, to a language type for which there is some sort of linguistic preference. An alternativ e v iew (cf. Vennemann 1981) is that the notions are simply a conv enient way to characterize certain languages, quite apart from what word order patterns are common among the languages of the world. There is no doubt that certain languages can be characterized as consistently ordering heads before dependents or v ice v ersa. And it is quite plausible that the grammars of these languages inv olv e a single rule gov erning word order. This has not, howev er, been the standard v iew in word order typology. Rather, the standard v iew is that certain word order characteristics are expected or predictable (in a statistical sense) giv en other word order characteristics of the language. 3.2. Semantic Explanations An alternativ e approach to explaining the basic pattern has been in terms of semantic no-

tions. In his earlier work, Vennemann (1973; 1974a) proposes an explanation in terms of consistent ordering of functor and argument, or in his terminology, operator and operand. His account assumes a semantic system in which the assignment of functor and argument is different from that in many other semantic theories. For example, he assumes in the combination of object and v erb that the object is the functor, the v erb the argument. In later work, Vennemann (1976) and Vennemann/Harlow (1977) offer a rev ised account of the basic pattern that inv olv es more conv entional assumptions as to what is functor and argument, but the new theory, though formulated in terms of categorial grammar and thus tied to semantic notions, is essentially a theory of consistent ordering of head and dependent of the sort discussed in 3.1. Keenan (1979) offers an alternativ e semantic account in which the object is argument, the v erb functor. Keenan’s account requires a dissimilation principle, whereby the position of functors combining with “determined noun phrases”, like the tall man, is the opposite of that of functors taking “common noun phrases”, like tall man. For example, a VO language on his theory is one in which functors combining with determined noun phrases precede their arguments (e. g. v erb preceding object), while functors combining with common noun phrases follow them (e. g. adjectiv e following noun). As Hawkins (1983) points out, a dissimilation principle like this is less elegant than a single principle, like consistent ordering of head and dependent, or of functor and argument. A further weakness with Keenan’s account is that it makes no prediction as to what should happen with functors that combine with arguments that are neither determined noun phrases nor common noun phrases, such as negativ es, auxiliaries, or adv erbs combining with v erbs (or v erb phrases), or qualifiers combining with adjectives. Hawkins (1983; 1984) argues for an account in terms of head and dependent rather than one in terms of functor and argument, arguing against the approaches of both Vennemann and Keenan, but all three linguists assume essentially the same set of object patterners and v erb patterners. Howev er, these semantic approaches need to be re-ev aluated in light of the new ev idence. Keenan’s dissimilation principle was motiv ated by the assumption that noun modifiers like adjectiv es, numerals, and articles are object patterners. But since they are not, the possibility exists

55.  Word Order Typology

of a semantic account of the basic pattern within a theory that treats the v erb as functor, but without any need for a dissimilation principle. The fact that articles (and to a lesser extent numerals) tend to precede the noun in VO languages is precisely what a theory claiming that functors tend to precede arguments in VO languages would predict. Nev ertheless a number of problems still exist for an approach which claims that OV languages consistently place arguments before functors. First, adv erbs and many adpositional phrases modifying v erbs are most naturally treated as functors on the v erb (or v erb phrase), but they tend to precede the v erb in OV languages. Second, prenominal relativ e clauses occur with v ery few exceptions only in OV languages. But relativ e clauses are most naturally treated as functors on the head noun. Finally, there are a number of pairs of elements whose order does not correlate with the order of object and v erb, but which seem to inv olv e a relationship of functor and argument. Among these are the adjectiv e and noun, demonstrativ e and noun, qualifier and adjectiv e, negativ e particle and v erb (or v erb phrase), and tense-aspect particle and v erb (or v erb phrase). It may be significant, howev er, that four of the functors in these fiv e pairs of elements — all but demonstrativ es — are attributes, in the sense of Vennemann (1976): they are dependents which combine with their heads to form constituents of the same syntactic and semantic type as the head. Conv ersely, many of the object patterners are complements in Vennemann’s sense: they are dependents that serv e as arguments rather than functors. It is possible that both syntactic and semantic notions are relevant to accounting for the basic pattern. 3.3. Processing Explanations The underlying assumption of the approaches to explaining the basic pattern that appeal to syntactic notions or semantic notions is that consistent languages inv olv e simpler grammars while inconsistent languages inv olv e more complex grammars. Other approaches assume that the higher frequency of consistent languages is to be explained instead in terms of extragrammatical notions. Kuno (1974) was the first to suggest that at least some of the correlations with the order of object and v erb might be due to the nature of sentence processing. He argued that certain combinations of inconsistent types would lead to structures that are more diffi-

1057

cult to process. He argues, for example, that prenominal relativ e clauses in VO languages and postnominal relativ e clauses in OV languages lead more often to centre-embedded structures than the rev erse situations. Dryer (1992) argues that the consistent language types inv olv e consistent left-branching or consistent right-branching, while inconsistent word order types, at least in certain cases, also leads to centre-embedding. Frazier (1979) argues that some of the correlations reflect a tendency for heads to be adjacent to heads of complements, a tendency for which she offers a further explanation in terms of processing. Hawkins (1990) offers further explanations in terms of processing. 3.4. Diachronic Explanations All of the explanations discussed so far are synchronic in the sense that they assume that inconsistent languages present some small burden on speakers of the language, either in terms of greater complexity in the grammar, or in terms of more difficult sentence processing, and that such languages will be under some internal pressure to become consistent. An alternativ e line of explanation, suggested for some of the correlations, both by Greenberg (1963) and by Vennemann (1973; 1974b), but pursued at greater length by Giv ón (1971; 1975; 1984), is an entirely diachronic one. According to this approach, inconsistent language types present no burden on speakers, but rather are less common simply because the conditions under which they arise occur less often than the conditions under which consistent languages occur. Consider for example the fact that case suffixes correlate with postpositions, while case prefixes (though not common) correlate with prepositions. The most plausible diachronic source for case affixes is adpositional words that become attached to nouns. On the approach of Giv ón (1971), the position of affixes relativ e to stems reflects the position of the words from which they arise. Hence the normal diachronic source for case suffixes will be postpositions rather than prepositions. Prepositions that become attached to nouns will become case prefixes rather than suffixes. There is no natural way in which a preposition could become a suffix, unless it passed through an intermediate stage as a postposition. Consider the inconsistent language type with prepositions but case suffixes. On the diachronic approach, such a type would arise only if a language had

1058

XVII. Syntaktische Typologie II: Ansätze und Übersichts-darstellungen

changed its adposition type from postposition to preposition, since the case suffixes would hav e arisen, by assumption, from postpositions. The greater frequency of consistent languages with postpositions and case suffixes is thus simply due, on this account, to the fact that languages do not often change their adposition type. If languages with postpositions tend to remain postpositional, then the account predicts that languages with case suffixes will more often be postpositional. Under this approach, the inconsistent type with prepositions and case suffixes is not dispreferred synchronically, nor will it be under any internal pressure to change, except in the sense that if the case suffixes are ev entually lost by erosion, and new case affixes arise, the new affixes will be prefixes (since the language is prepositional). Giv ón offers this type of explanation for the correlation between adposition type and the order of object and v erb. One of the diachronic sources for adpositions is v erbs; in an OV language, the v erb will become a postposition, while in a VO language the v erb will become a preposition. The English preposition including is an apparent example of this process. Since the v erb include preceded its object, when including became reanalysed as an adposition, its order with respect to its object remained the same as when it was a v erb, and so it became a preposition. Another source for adpositions is head nouns in genitiv e constructions. English because of is an apparent example of this. It became a preposition because it arose from the head noun cause that preceded a genitv e (by cause of X). This predicts the correlation of adposition type with the order of noun and genitiv e. A further example of this line of explanation is the explanation for auxiliary v erbs being v erb patterners: they arise from main v erbs, while the new main v erb arises from a complement. The English v erbals be going to, used to and supposed to all appear to be in the midst of such a transition. The tendency for the numeral to precede the noun in VO languages may reflect its original status as head in a noun-genitiv e construction, ve en in languages in which it is now a dependent. Finally, on such an approach the correlation between affix type and stem can be v iewed as a reflection of the fact that the kinds of words which serv e as the source of affixes tend to be ones which follow the word they go with in OV languages, but precede in VO languages, such as adpositions becoming case

affixes and auxiliaries becoming tense-aspect affixes. It is significant that the two classes of affixes which are more often prefixes in OV languages than they are in VO languages are pronominal object prefixes on v erbs and possessiv e prefixes on nouns. Both presumably arise from separate pronoun words (cf. Giv ón 1976) that more often precede their heads in OV languages. Some of the correlations in the basic pattern seem less susceptible to this line of explanation. For example, there is no obv ious diachronic account for the fact that adv erbials tend to occur on the same side of the v erb as the object. Rather this correlation seems to be due to the (synchronic) grammatical parallelism between the relation of objects to v erbs and the relation of adv erbials to v erbs. Significantly, howev er, this correlation is an intra-categorial correlation, since it inv olv es the order of two elements with respect to the same head, the v erb. Whether there are cross-categorial correlations that are not susceptible to this kind of diachronic explanation is less clear. While it is clear that the cross-categorial correlations are due in part to diachronic factors of the sort discussed, there is v ery little ev idence in the literature bearing on the question of whether the correlations are due entirely or largely to such diachronic factors, or whether synchronic factors of the sort assumed in 3.1. to 3.3. also play a significant role as well. Nor is there any reason to believ e that the basic pattern reflects a single principle. Rather, the correlations may be due to a combination of factors (both synchronic and diachronic), and the indiv idual correlations may be due to different factors: some may be largely due to diachronic factors, while others may be due more to synchronic grammatical parallels. It was noted at the end of 2. that an unanswered question is whether the basic pattern involv es a set of pairs of elements all of which correlate in order with each other, or whether there instead exist chains of correlations. The explanations which appeal to a single underlying principle, such as consistent ordering of heads and dependents, predict the former. The kind of explanation proposed by Giv ón is consistent with the latter since it inv olv es specific explanations for each pairwise correlation. Justeson/Stephens (1987), as noted in 2., argue that there do indeed exist such chains of correlations, and that not all the pairs of elements correlate in order with each other. One of the chains they argue for

55.  Word Order Typology

is one by which the order of genitiv e and noun correlates with adposition type and adposition type correlates with the order of object and v erb; but they argue that there is no direct correlation between the order of genitiv e and noun and the order of object and v erb, and that the apparent correlation is simply due to the fact that each of these correlates with adposition type. Significantly this is precisely what Giv on’s account of adposition type would lead us to expect. His argument that adpositions arise either from v erbs or from head nouns in noun-genitiv e constructions predicts the two correlations with adposition type, but does not predict any correlation between the order of object and verb and the order of genitive and noun.

4.

Other Patterns

4.1. Deviations in the Basic Pattern Although the basic pattern may characterize consistent languages, it represents at best a tendency, a statistical generalization about an ideal around which languages cluster. As Hawkins (1983) notes, simply characterizing this ideal fails to do justice to the large number of languages which do not conform to it. And not only are there many languages which do not conform; there are many additional generalizations to be made about those which do not. Hawkins’ distinction between implicational and distributional univ ersals is an attempt to address this problem. On the one hand, his principle of Cross-Category Harmony makes predictions about the relativ e frequency of languages that fail to conform to the basic pattern, predicting that among such languages language frequency will correlate with degree of consistency: those which are more consistent will be more common than those which are less consistent. And on the other hand, his attempt to formulate a set of exceptionless implicational univ ersals is designed to distinguish those inconsistent types which are attested from those which are not. Hawkins (1983, 41) computes that of the languages in Greenberg’s 30-language sample, only 7 are consistent. But in making this computation, he follows Lehmann and Vennemann in his assumptions about the set of object patterners and v erb patterners. In particular, he assumes that adjectiv es, demonstrativ es, and numerals modifying nouns and qualifiers (adv erbs) modifying adjectiv es are

1059

object patterners, that these elements precede the word they modify in consistent OV languages and follow in consistent VO languages. But these elements do not exhibit any such tendency (cf. 2. and Dryer 1986), and numerals in fact exhibit a weak tendency in the opposite direction. OV languages outside of Asia more often place the adjectiv e after the noun, and VO languages outside of Africa ov erwhelmingly place the numeral before the noun. Hence it is hardly surprising that only 7 of the 30 languages are consistent by these criteria. Nor are the 7 a random subsample: the 4 consistent OV languages are all spoken in Asia, and the 3 consistent VO languages are Niger-Congo languages, spoken in Africa. If we restrict our criteria to pairs of elements whose order does correlate with that of the object and v erb (cf. 2.), it turns out that 20 of the 30 languages are consistent. (The pairs assumed here are genitiv e and noun, relativ e clause and noun, noun phrase and adposition, main v erb and inflected auxiliary, main v erb and subordinate v erb in a construction indicating v olition and purpose, and standard of comparison and adjectiv e.) Obv iously the frequency of inconsistent languages depends on one’s assumptions as to what constitutes consistency. But ev en with a more restricted notion of consistency, there are clear patterns to the inconsistencies. In fact, they primarily inv olv e what Hawkins (1988) calls asymmetries, and what are called unilateral correlations in 2. For example, 5 of the 10 inconsistent languages in Greenberg’s sample are inconsistent with respect to the order of main v erb and subordinate v erb in expressions of v olition and purpose. But all 5 exceptions are OV languages in which the subordinate v erb follows the main v erb. Thus of the four possible types of languages v arying for these two parameters, order of object and v erb, and order of main v erb and subordinate v erb in purpose constructions, three types are wellattested in Greenberg’s sample, and one is unattested, VO languages in which the subordinate v erb precedes the main v erb. As noted in 2., these unilateral correlations are naturally described by the use of implicational univ ersals, in this case “If a language is VO, then it will place the subordinate v erb after the main v erb in expressions of purpose” (or equiv alently “If a language places the subordinate v erb before the main v erb in expressions of purpose, the language is OV”). This is essentially Greenberg’s Univ ersal 15. Simi-

1060

XVII. Syntaktische Typologie II: Ansätze und Übersichts-darstellungen

larly two of the inconsistent languages in Greenberg’s sample are OV languages which place the relativ e clause after the noun. In fact, OV languages of this sort are more common than Greenberg’s sample might suggest (cf. Hawkins 1990, Dryer 1992). But VO languages with prenominal relativ e clauses are quite rare. Note that both of these asymmetries inv olv e OV languages placing subordinate v erbs to the right of their heads. This is reminiscent of a conclusion of Dryer (1980): there are OV languages in which object clauses normally follow the v erb, but not VO languages in which such object clauses normally precede the v erb. These asymmetries seem to inv olv e an ov erall preference for subordinate v erbs (or clauses) to follow their heads, which conflicts with the basic pattern in OV languages. The remaining four inconsistent languages in Greenberg’s sample are SVO languages in which the genitiv e precedes the noun. Three of these are also inconsistent in being postpositional. In fact, postpositional SVO languages are somewhat less common than this sample might suggest. On the other hand, SVO languages in which the genitiv e precedes the noun are quite common. This asymmetry apparently reflects the fact that in many languages the order of genitiv e and noun parallels the order of subject and v erb rather than the order of object and v erb. This is not surprising in light of the existence of other grammatical similarities between genitiv es and subjects. The inconsistent languages in Greenberg’s sample thus fall into one of two categories: OV languages in which subordinate v erbs follow their heads and SVO languages in which the genitiv e precedes the noun. A further example of an asymmetry inv olv es the position of affixes (cf. Hawkins 1988). While on the basic pattern suffixes should occur in OV languages, prefixes in VO languages, the exceptions tend to be VO languages with suffixes. This seems to reflect an ov erall preference for suffixes (cf. Greenberg 1957; Cutler/Hawkins/Gilligan 1986; Bybee/ Pagliuca/Perkins 1990). A final asymmetry is the order of subject and v erb. Although the earlier v iew (cf. Vennemann 1974a) was that the subject-v erb relation lay outside the basic pattern, Keenan (1979) suggests that SVO order can be v iewed as an inconsistent type whose not infrequent occurrence is due to a competing principle whereby the subject tends to precede other major constituents of the sentence. This v iew of the subject-v erb relation as falling within the basic pattern is

also followed by Vennemann/Harlow (1977) and Hawkins (1983). The order of subject and v erb does exhibit a unilateral correlation with that of the object and v erb: the subject precedes the v erb significantly more often in OV languages than it does in VO languages, since OVS languages are rare. 4.2. Other Correlations with the Basic Pattern Not all correlations with the order of object and v erb inv olv e the order of two elements. For example, VO languages exhibit a stronger tendency than OV languages to place interrogativ e words or phrases in content questions at the beginning of sentences. But it is difficult to describe this in terms of the order of two elements since the alternativ e type inv olv es placing such phrases in the position of other phrases of their grammatical or semantic type. A number of OV languages place such phrases immediately before the v erb (cf. Kim 1988), but this tendency seems restricted to languages in Eurasia. There are also typological characteristics which correlate with the order of object and v erb but which do not inv olv e word order at all. For example case affixes are more common in OV languages than in VO languages (cf. Greenberg’s Univ ersal 41). This is partly because VO languages often use prepositional words where OV languages use case suffixes (cf. 5.2.). Howev er, OV languages employ case affixes or adpositions to distinguish subjects from objects more often than VO languages do. SVO languages are particularly lacking in such case marking for subjects and objects, presumably because the word order is a reliable guide, ev en in clauses containing only one independent nominal. There also seem to be correlations between word order type and morphological type. Isolating languages tend to be SVO. Whether v erb-final and v erb-initial languages differ in this respect is unclear. Lehmann (1973) suggests that OV languages are more often agglutinativ e, VO languages inflectional, but this too remains to be tested. Finally, articles appear to occur more commonly in VO languages than in OV languages, but this tendency has yet to be documented. 4.3. Correlations outside the Basic Pattern Although a number of pairs of elements often thought to correlate with the order of object and v erb in fact do not (see 2.), some of them

55.  Word Order Typology

do correlate with other pairs of elements that correlate with the order of object and v erb. For example, the order of adjectiv e and noun may not correlate with the order of object and v erb, but it does correlate with the order of relativ e clause and noun. This correlation is a unilateral one: while languages with prenominal adjectiv es and postnominal relativ e clauses are common, the rev erse situation is not. The order of adjectiv e and noun also correlates with the order of article and noun, though here the correlation is with a v erb patterner rather than with an object patterner, since articles are v erb patterners (in contrast to relativ e clauses which are object patterners). Again the correlation is a unilateral one: languages with prenominal articles but postnominal adjectiv es are common while the rev erse situation is not. Curiously, the order of adjectiv e and noun does not correlate with the order of genitiv e and noun, despite the fact that both are modifiers of the noun. All four possible combinations are common, and in fact the most common type is that in which the adjectiv e follows the noun while the genitiv e precedes. There are also correlations between pairs of elements neither of which correlate with the order of object and v erb. For example, the orders of adjectiv e and demonstrativ e with respect to the noun correlate with each other, though the correlation is unilateral; cf. Greenberg’s Univ ersal 18: “When the descriptiv e adjectiv e precedes the noun, the demonstrativ e and the numeral, with ov erwhelmingly more than chance frequency, do likewise.” 4.4. Complex Patterns A few of Greenberg’s implicational univ ersals, and the majority of Hawkins’, are complex in the sense that they refer to more than two terms. An example is Greenberg’s Univ ersal 5: “If a language has dominant SOV order and the genitiv e follows the gov erning noun, then the adjectiv e likewise follows the noun.” Some multi-term implicational univ ersals, like this one, characterize patterns among inconsistent languages. The languages defined by the antecedent clause in the univ ersal just mentioned are themselv es inconsistent. The effect of the univ ersal is thus to say that one class of inconsistent languages is more common than another. A number of Hawkins’ multi-term implicational univ ersals are generalizations about noun modifier order in languages of a giv en adpositional type. By restricting attention to

1061

prepositional languages he is able to arriv e at generalizations that do not hold for postpositional languages as well. He argues that there is a hierarchy among noun modifiers such that, in prepositional languages, modifiers higher on the hierarchy will not follow the noun while modifiers lower precede the noun: demonstrativ es, numerals > adjectiv es > genitiv es > relativ e clauses. This predicts, for example, that one will not find prepositional languages in which adjectiv es follow the noun while genitiv es precede. There are a few exceptions, but most languages conform to the hierarchy. Hawkins offers an explanation for the hierarchy in terms of the relativ e heaviness of the different modifiers.

5.

Further Issues

5.1. The Notion of Basic Word Order Word order typology often seems to assume that ev ery language can be assigned a basic order, not only among the six possible orders of subject, object, and v erb, but also for the v arious other word order parameters, like the order of adjectiv e and noun, and adposition type. But there are a number of possible problems with this assumption. First, it is not always clear what is meant by basic order. Hawkins (1983) lays out three criteria: higher (text) frequency; higher frequency within the grammatical system (e. g. more adjectiv e lexemes occurring on one side of the noun); and grammatically unmarked status. To these, a fourth criterion is often appealed to: pragmatically unmarked status. In practice these notions generally conv erge; cases in which different criteria point to different conclusions are not common. The notion of basic order is sometimes thought to be equiv alent to deep structure order in some v ersion of transformational grammar. But in fact there is little reason to believ e the latter notion is relev ant to word order typology. The assumptions of transformational grammar point to SOV as the deep structure order of Dutch and German (cf. Koster 1975). But there is little reason to say that the basic order of these languages is SOV for the purposes of word order typology. It is often not realized that the question as to what is the proper definition of basic word order is an empirical question: Which definition allows the strongest generalizations about word order correlations? For example, as is implicit in univ ersals of both Greenberg and

1062

XVII. Syntaktische Typologie II: Ansätze und Übersichts-darstellungen

Hawkins, languages which place the adjectiv e before the noun and the genitiv e after the noun are rarely, if ev er, SOV. If one were to classify Dutch as SOV, it would be an exception to this generalization. Yet the reason it would be an exception is clearly that it is not SOV in the standard sense; SOV order in Dutch satisfies none of Hawkins’ criteria. The prominent role that SVO (or at least VO) order plays as a surface order in Dutch would seem to be linked to the order of the modifiers with respect to the noun: the structure of the Dutch noun phrase is one which is common among VO languages, but uncommon among OV languages. Because of cases like these it seems that the notion of deep structure order, whatev er its theoretical merit, is irrelevant to word order typology. But the cases of Dutch and German do point to problems identifying basic order. If they are not SOV, are we to say their basic order is SVO? SVO may hav e the highest text frequency, and it may be grammatically unmarked, since SOV order either requires an auxiliary or occurs in a subordinate clause, but classifying these languages as SVO misses the obv ious point that they are more accurately described as basically v erb-second. In other words, they fall outside the SVO/SOV taxonomy. In these and other instances in which assignment of basic order is problematic, it may be best to leav e them unclassified and not force them into some single basic order. The problem of assigning basic order is perhaps most frequently encountered in languages with flexible or so-called free word order. In all such cases, the word order is probably not literally free, but rather is gov erned by discourse factors (cf. Giv ón 1983; Payne 1987; Mithun 1987). This type of word order flexibility is especially common among clause-lev el constituents, but in some languages is found at other lev els as well. It might be argued that such languages should be left unclassified as to basic order. Howev er, the criteria discussed abov e, particularly text frequency, frequently point to one order as basic, ev en in such languages. It is sometimes questioned whether that is enough. Should a language with flexible order in which SOV is most common be classified the same as a language with fairly rigid SOV order? Again, the question is an empirical one: do these two kinds of languages pattern the same with respect to the word order correlations? And although the question has yet to

be systematically examined, the answer appears to be that they do. For example, flexible SOV languages seem to prefer postpositions ov er prepositions, just as more rigid SOV languages do. There are languages in which word order is so flexible that they are not easily assigned to one of the six basic orders, such as Cayuga, an Iroquoian language (cf. Mithun 1987). But the six-way typology, despite its historical prominence, is not crucial to the basic pattern of word order correlations (see 2.). All that is necessary is whether the language is OV or VO, and perhaps whether it is SV or VS. And some languages which are difficult to assign to one of the six orders can be classified with respect to one or both of these. Some of the literature on word order typology employs notions like OV language in a sense that is dependent on the other word order characteristics of the language: on this usage, a language with characteristics typical of OV languages is described as an OV language, ev en if the criteria for basicness fail to point to OV as the basic order of object and v erb. Such a notion presupposes that it has already been established what characteristics are typical of OV languages. But as already discussed, this has not been settled, and certain properties widely assumed to be characteristic of OV languages, such as adjectiv enoun order, turn out not be such. Furthermore, such a notion of OV language, leads to circularity in testing univ ersals and correlations. 5.2. Identifying the Relevant Categories Another problem in assigning languages to a particular word order type, like SOV, or adjectiv e-noun, is that of identifying in each language grammatical functions like subject and object and grammatical categories adjectiv e and noun. The problems identifying subjects in some languages hav e been widely discussed (cf. Keenan 1976), but in practice they do not present a serious problem for word order typology. For example there is a legitimate question as to what, if anything, should be called subjects in ergativ e languages. But most ergativ e languages are either v erb-initial or v erb-final. Hence these languages can be assigned to the categories v erb-initial or v erbfinal, ev en if they cannot easily be assigned to more specific categories like VSO. In general, semantic criteria hav e played a primary role in identifying subjects and objects in word order typology: agents of transitiv e v erbs are

55.  Word Order Typology

treated as subjects while patients of transitiv e v erbs are treated as objects. Thus in practice an OV language has been identified as one in which patients of transitiv e v erbs precede the v erb. Again it must be stressed that whether this practice is acceptable is independent of questions about how best to define grammatical relations: rather, the question is whether identifying objects in this way leads to strong generalizations about word order. The fact that this practice was used by Greenberg in formulating the original univ ersals suggests it does. Hence the practice is justified unless there is ev idence that alternativ e approaches lead to stronger generalizations. The problem presented by identifying adjectiv es is in many ways similar. As with other categories, Greenberg assumed an essentially semantic notion of adjectiv e, namely as a word that modifies a noun and has a meaning corresponding to adjectiv es in English. But in many languages, such words do not belong to a separate word class, being either v erbs or nouns. While there is rarely any problem in assigning a language to the adjectiv e-noun category or the noun-adjectiv e category, the question arises as to whether the class of languages that are classified by this method as, say noun-adjectiv e, is a meaningful class, since the words being called adjectiv es actually belong to different grammatical categories in the different languages. Again, howev er, the question is an empirical one: do the v arious types of languages that belong to this class pattern the same way with respect to word order correlations? The answer to this question remains unanswered. The primary problem identifying postpositions and prepositions surrounds the question of whether affixes expressing the meaning of adpositions in languages like English should be treated as adpositions. Greenberg treated such morphemes as adpositions, again employing essentially semantic criteria. Giv ón, giv en his assumptions, also treats the distinction as irrelev ant, as does Vennemann (1974b). Hawkins, howev er, excluded such languages in constructing a sample that expanded on Greenberg’s. There exist a number of exceptions to Hawkins’ univ ersals if we follow Greenberg’s practice. Whether this is because the class of languages examined is simply larger, or whether adpositional words and adpositional affixes behav e differently with respect to the word order correlations remains to be inv estigated. It should be noted that if we assume the narrower notion of ad-

1063

position, excluding affixes, and if we use adposition type as the fundamental word order parameter, as Hawkins (1983) suggests, then many languages will be unclassifiable since there are many languages which lack adpositional words. Similar questions about identifying categories in different languages, and how best to define these categories cross-linguistically, apply to a number of other elements that are relev ant to word order correlations. Lehmann (1973) suggested that negativ e markers follow the v erb in OV languages, and precede in VO languages. The position of negativ e markers depends, howev er, on their properties within each language (cf. Dahl 1979). Lehmann’s claim is true (as a tendency) for negativ e affixes on v erbs and negativ e auxiliaries, but not for negativ e particles (like English not). Such particles tend to precede the v erb in both VO and OV languages. The fact that such particles exhibit different properties from affixes suggests that the question of whether an element is a separate word or not may be important in general. To what extent attention to such grammatical characteristics — as opposed to the largely semantic notions assumed by Greenberg — will lead to new generalizations and insights remains to be seen. Despite the period of time that has elapsed since the appearance of Greenberg’s seminal paper, few issues in word order typology hav e been resolv ed. There is no agreement as to what pairs of elements correlate in order with that of the object and v erb, or how to best characterize that basic pattern, or how to explain it. It has been generally assumed that there is a basic pattern, and that assumptions has been maintained here. But ev en that assumption might be questioned. While there is clear ev idence that certain pairs of elements do correlate with each other in order, it is possible to propose a notion of consistent OV (or VO) language that is div orced from any observ able correlations we find among the world’s languages (cf. 3.1., Vennemann 1981), and thus distinct from the basic pattern assumed here. On this approach we might define a consistent OV language as one that consistently places dependents before heads, regardless of whether this is common among OV languages. But of course alternativ e notions of consistent OV language are equally possible, such as one that consistently places arguments before functors, or one that consistently places complements be-

1064

XVII. Syntaktische Typologie II: Ansätze und Übersichts-darstellungen

fore heads (without any principle for other dependents), or one that is consistently leftbranching. It is quite possible that all of these are useful notions and that consistent languages differ as to what notion of consistency they operate according to. And the basic pattern described in 2. may simply reflect the fact that these different notions of consistency coincide for many pairs of elements, so that we find an observ able correlation among the languages of the world for these pairs of elements. On such an approach, the notion of inconsistent language would hav e to be reassessed. A common v iew is that inconsistent languages are in transition from consistent OV to consistent VO or v ice v ersa. But a language that is consistent on one notion of consistency may be inconsistent relativ e to another notion, so the notion of inconsistent language would not be well-defined. Hence this alternativ e v iew of consistent word order has important implications for understanding word order change. But whether such an approach is warranted remains to be inv estigated.

6.

References

Antinucci, Francesco; Alessandro D uranti; and Lucyna Gebert. 1979. Relativ e clause structure, relativ e clause perception, and the Change from SOV to SVO. Cognition 7. 145—76. Babby, Leonard H. 1987. Case, prequantifiers, and discontinuous agreement in Russian. Natural Language and Linguistic Theory 5. 91—138. Bybee, Joan; William Pagliuca; and Revere Perkins. 1990. On the asymmetries in the affixation of grammatical material. Studies in typology and diachrony: Papers presented to Joseph H. Greenberg on his 75th birthday, ed. by William Croft, Keith Denning, & Suzanne Kemmer 1—42, Amsterdam. Cutler, Ann; John A. Hawkins; and Gary Gilligan. 1986. The suffixing preference: a processing explanation. Linguistics 23. 723—58. D ahl, Osten. 1979. Typology of sentence negation. Linguistics 17. 79—106. D erbyshire, D esmond, and Geoffrey Pullum. 1981. Object-initial languages. International Journal of American Linguistics 47. 192—214. —, —. 1986. Introduction. Handbook of Amazonian languages, v ol. 1, ed. by Desmond Derbyshire, & Geoffrey Pullum. Berlin. D ryer. Matthew S. 1980. The positional tendencies of sentential noun phrases in univ ersal grammar. Canadian Journal of Linguistics 25. 123—95. —. 1988. Objectv erb order and adjectiv e-noun order: dispelling a myth. Lingua 74. 77—109.

—. 1989. Large linguistic areas and language sampling. Studies in Language 13. 257—92. —. 1991. SVO languages and the OV:VO typology. Journal of Linguistics 27. 443—482 —. 1992. The Greenbergian word order correlations. Language 68. 81—138. Frazier, Lynn. 1979. Parsing and constraints on word order. Univ ersity of Massachusetts Occasional Papers in Linguistics 5. 177—98. Givón, Talmy. 1971. Historical syntax and synchronic morphology: an archaeologist’s field trip. Proceedings from the Sev enth Regional Meeting of the Chicago Linguistic Society. 394—415. —. 1975. Serial v erbs and syntactic change: NigerCongo. Word order and word order change, ed. by Charles Li, 47—112. Austin. —. 1976. Topic, pronoun and grammatical agreement. Subject and topic, ed. by Charles Li, 149—188. Austin. Givón, Talmy (ed.). 1983. Topic continuity in discourse: A quantitativ e cross-linguistic study. Amsterdam. —. 1984. Syntax: A functional-typological introduction, vol. 1. Amsterdam. Greenberg, Joseph H. 1957. Order of affixing: a study in general linguistics. Essays in linguistics. Chicago. —. 1963. Some univ ersals of grammar with particular reference to the order of meaningful elements. Univ ersals of language, ed. by Joseph Greenberg, 73—113. Cambridge, MA. Hawkins, John A. 1979. Implicational univ ersals as predictors of word order change. Language 55. 618—48. —. 1980. On implicational and distributional univ ersals of word order. Journal of Linguistics 16. 193—235. —. 1983. Word order universals. New York. —. 1984. Modifier-head or function-argument relations in phrase structure? Lingua 63. 107—38. —. 1988. On explaining some left-right asymmetries in syntactic and morphological univ ersals. Studies in syntactic typology, ed. by Michael Hammond, Edith Morav csik & Jessica Wirth. Amsterdam. —. 1990. A parsing theory of word order univ ersals. Linguistic Inquiry 21. 223—61. Justeson, John S., and Laurence D . Stephens. 1987. Explanations for word order univ ersals: a log linear analysis. Paper presented to the XIV International Congress of Linguists. Berlin. Keenan, Edward. 1976. Toward a univ ersal definition of subject. Subject and topic, ed. by Charles Li, 305—333. New York. —. 1978. The syntax of subject-final languages. Syntactic typology, ed. by Winfred P. Lehmann, 267—327. Austin.

56.  Prominence Typology

—. 1979. On surface form and logical form. Linguistics in the sev enties: Directions and prospects, Studies in the Linguistic Sciences 8:2, ed. by Braj B. Kachru. Urbana, Illinois. Kim, Alan Hyun-Oak. 1988. Prev erbal focusing and Type XXIII languages. Studies in Syntactic Typology, ed. by Michael Hammond, Edith Morav csik & Jessica Wirth, 147—169. Amsterdam. Koster, Jan. 1975. Dutch as an SOV language. Linguistic Analysis 1. 111—36. Kuno, Susumo. 1974. The position of relativ e clauses and conjunctions. Linguistics Inquiry 5. 117—36. Lehmann, Winfred P. 1973. A structural principle of language and its implications. Language 49. 42—66. —. 1978. Introduction. Syntactic typology, ed. by Winfred P. Lehmann. Austin. Masica, Colin P. 1976. On defining a linguistic area. Chicago. Mithun, Marianne. 1987. Is basic word order univ ersal? Coherence and grounding in discourse, ed. by Russell S. Tomlin, 281—328. Amsterdam. Payne, D oris L. 1987. Information structuring in Papago narrativ e discourse. Language 63. 783—

1065

804. Siewierska, Anna. 1988. Word order rules. London. Tomlin, Russell. 1986. Basic constituent orders: functional principles. London. Vennemann, Theo. 1973. Explanation in syntax. Syntax and Semantics. Vol. 2, ed. by John Kimball, 1—50. New York. —. 1974a. Analogy in generativ e grammar: the origin of word order. Proceedings of the Elev enth International Congress of Linguists (1972), 19—83. Bologna. —. 1974b. Theoretical word order studies: results and problems. Papiere zur Linguistik 7. 5—25. —. 1976. Categorial grammar and the order of meaningful elements. Linguistic studies offered to Joseph Greenberg on the occasion of his sixtieth birthday, ed. by A. Juilland, 615—634. Saratoga, CAL. —. 1981. Typology, univ ersals, and change of language. Presented at the International Conference on Historical Syntax, Poznan. —, and Ray Harlow. 1977. Categorial grammar and consistent basic VX serialization. Theoretical Linguistics 4. 227—54.

Matthew S. Dryer, Buffalo, New York (USA)

56. Prominence Typology 1. 2. 3. 4. 5.

Topic-Prominence vs. Subject-Prominence: The Original Model Earlier Critique and Elaboration of the Li-Thompson Model Syntactic Pivot Theory Focus-Prominence and “Configurationality” References

The phenomena described by this expression hav e been studied since the mid-sev enties. The term is unfortunate; “Topic-Comment Typology” or “Typology of Topic-Comment Grammaticalization” would probably be a better choice. The basic idea is that languages may differ as to whether their fundamental syntactic organization reflects the discoursepragmatically determined bipartiteness of ‘topic’ and ‘comment’ directly, or whether it is mediated through the syntactic relations of ‘subject’ and ‘predicate’. Recently, howev er, the concept of “prominence typology” has been broadened by the introduction of a third type, the “focus-prominent” or “focusconfigurational” languages. In the following, we will first deal with the original concept and then address more recent approaches.

1.

Topic-Prominence vs. SubjectProminence: The Original Model

The question first receiv ed attention in a brief note by Hockett (1958, 202—3), who observ ed that “the most general characterization of predicativ e constructions is suggested by the terms ‘topic’ and ‘comment’ for their ICs: the speaker announces a topic and then says something about it. [...] In English and the familiar languages of Europe, topics are usually also subjects, and comments are predicates: [...] When the topic and comment of a predicativ e constitute are not also the subject and predicate, then usually the comment in turn is a predicativ e constitute consisting of subject and predicate. [...] In Chinese the preceding generalization does not hold. The fav orite sentence-type of Chinese is different from that of English. If we delete the subject from a simple English sentence, [...] the lone predicate cannot function as a sentence of the fav orite type, but only as a subjectless sentence [...] If we delete the topic

1066

XVII. Syntaktische Typologie II: Ansätze und Übersichts-darstellungen

from a simple Chinese sentence that has one, the comment still can stand, in most cases, as a sentence of the fav orite type. [...] Furthermore, many Chinese comments consist in turn of a topic and comment, so that one can hav e a sentence built up of predications within predications, Chinese-box style. [...] The tie in Chinese between topic and comment is to us unusually loose, particularly if we compare it only with the usual tie between subject and predicate in English.” This idea was taken up by Charles Li and Sandra A. Thompson in a series of articles and papers between 1974 and 1976. In a comprehensiv e study of the matter published in Li (1976) they dev elop a general typology of language according to the parameter of topic-comment v s. subject-predicate orientedness. On the basis of ev idence gathered from sev eral members of the Sino-Tibetan language family, they claim that there are languages in which the basic syntactic constructions manifest a topic-comment relation rather than a subject-predicate relation. The ev idence suggests not only that the notion of topic as a grammatical relation in the sentential structure of a language may be as basic as that of subject, but also that languages may differ in their strategies of constructing sentences according to the dominance (“prominence”) of the notion of topic and subject. In a “subject-prominent language” the basic sentence structure would be like (1): (1)

John SUBJECT

hit Mary PREDICATE

In a “topic-prominent” languages the basic sentence structure would be similar to (2): (2) As for education, TOPIC John prefers Bertrand Russell’s ideas. COMMENT Li and Thompson distinguish four basic types of languages: (a) Languages that are subject-prominent (Sp). Examples: Indoeuropean, NigerCongo, Semitic ... (b) Languages that are topic-prominent (Tp). Examples: Chinese, Lahu (Lolo-Burmese), Lisu (Lolo-Burmese) ... (c) Languages that are both Sp and Tp. Examples: Japanese, Korean ... (d) Languages that are neither Sp nor Tp. Examples: Philippine languages ... The decision to classify a language as either Sp or Tp depends on the distinction

between the notions of subject and topic. According to Li and Thompson, the following characteristics are distinctive: (i) D efiniteness. Topics are always definite, while subjects need not be definite. The topics of Tp languages, then, are inev itably definite expressions, while Sp languages may have indefinite subjects. (ii) Selectional relations. Topics need not have a selectional relation with any v erb in a sentence, i. e. the topic need not be an argument of a predicativ e constituent. The following examples from Lahu and Mandarin, two Sino-Tibetan languages classified as highly Tp, demonstrate this point: (3) LAHU hε chi tê pê  dà field this one CLASSIFIER rice very jâ good ‘This field (topic), the rice is very good’ (4) LAHU hɔ  na-qh y ve long PRT elephant TOPIC nose MARKER yò DECLARATIVE MARKER ‘Elephants (topic), noses are long’ (5) MANDARIN neì-chang huǒ xìngkui that-CLASSIFIER fire fortunate xīaofang-duì laí de fire-brigade come ADV.PARTICLE kuài quick ‘That fire (topic), fortunately the fire-brigade came quickly’ (6) neì-xie shùmu shù-shēn dà those tree tree-trunk big ‘Those trees (topic), the trunks are big’ (iii) Case frame and valence. Subject selection is determined by the case frame of the v erb v ia ‘subjectiv alization rules’ in the sense of Fillmore (1968). “Thus in English, if a v erb occurs with an agent as well as other noun phrases, the agent will become the subject unless a ‘special’ construction is resorted to, such as the passiv e” (Li & Thompson 1976, 463). Similarly, if the v erb is intransitiv e, its sole argument will be the subject; if it is causativ e, the causer will be the subject; etc. Such characteristic generalization can not be made in ev ery case, but the fact remains that giv en a v erb, the subject is predictable on the basis of its syntactic “v alence”. “The topic, on the

56.  Prominence Typology

other hand, is not determined by the v erb. Discourse may play a role in the selection of the topic, but within the constraints of the discourse, the speaker still has considerable freedom in choosing a topic noun phrase regardless of what the v erb is” (Li & Thompson 1976, 463). (iv ) Functional role. “What the topics appear to do is to limit the applicability of the main predication to a certain restricted domain [...] Typically, it would seem, the topic sets a spatial, temporal, or indiv idual framework within which the main predication holds” (Chafe 1976, 50 on “topics, Chinese style”). The topics of this sort are not only comparable to English as for or left-dislocation constructions, but also to frame-setting temporal adverbs, as in (7) Tuesday I went to the dentist This function of specifying the domain within which the predication holds is a pragmatic one, insofar as it is related to the structure of the discourse in which the sentence is found: topics are intimately connected with the “theme of the discourse”. Subjects, on the other hand, can either hav e a purely syntactic function (such as the “dummy” subjects found in some European languages), or one which is at least definable within the confines of the sentence rather than within the larger context of the discourse. (v ) Verb-agreement. As a result of the difference in selectional status between subjects and topics, v erb-subject agreement is a v ery common phenomenon (often indicativ e of subject-hood), while topic-predicate agreement is v ery rare and there is no case attested so far where it is obligatory. (v i) Sentence-initial position. Topics obligatorily occur sentence-initially, while subjects are not confined to this position. (vii) Grammatical processes. “The subject but not the topic plays a prominent role in such processes as reflexiv ization, passiv ization, Equi-NP deletion, v erb serialization, and imperativization” (Li & Thompson 1976, 465). The authors then go on to demonstrate the typological difference between Tp and Sp languages in terms of surface coding (in Tp languages there may be surface coding for the topic (“topic marker”), but not necessarily for the subject), voice (Tp languages do not tend to hav e passiv es, or if they do, the passiv e is functionally quite different from that of Sp languages), “dummy” subjects (only occur in Sp languages), position of the verb (Tp languages tend to be v erb-final),

1067

constraints on topic constituent (in Sp languages, but not in Tp languages). The most striking characteristic of Tp languages, which has already been discussed in the literature, is the so-called “double subject (DS) construction”: (8)

JAPANESE ga sakana wa tai fish TOPIC red snapper SUBJECT oisii delicious ‘Fish (topic), red snapper is delicious’

There hav e been attempts to explain DS constructions as deriv ed from some “normal”, subject-oriented, basic sentence structure. As there is often a part-whole relationship between the topic and the subject of DS constructions, it has been claimed in early generativ e work that DS sentences arise from an (obligatory?) transformation of possessor exbraciation, such as noses of elephants are big ⇒ elephants, (their) noses are big. Teng (1974) has taken a different approach in suggesting that these sentences inv olv e a “sentential predicate”. Li and Thompson conv incingly show that there is no reason to v iew topic-comment sentences in Tp languages as “deriv ed”, “marked” or otherwise “special”. On the contrary, the topic-comment structure constitutes a basic sentence type: “The topic function, which is highly marked and set off from the rest of the sentence in Sp languages, has in Tp languages been integrated into the basic syntax of the sentence. The topic notion must be reckoned with in constructing an adequate grammatical description of these languages, and topic-comment sentences must be counted among the basic sentence types prov ided by the language” (Li & Thompson 1976, 483). Synchronic as well as diachronic observ ations suggest that subject and topic are not unrelated notions. Subjects are essentially topics that hav e become integrated into the case frame of a v erb. The four-way typological distinction made by Li and Thompson thus has certain diachronic implications; it represents a complete cycle with four idealtypological fix-points (see next page).

2.

Earlier Critique and Elaboration of the Li-Thompson Model

The original approach depicted thus far has been refined in v arious directions. First of all, there hav e been a number of studies of Tp

1068

XVII. Syntaktische Typologie II: Ansätze und Übersichts-darstellungen

Fig. 56.1: The Topic-Subject Cycle (from: Li & Thompson 1976, 485)

languages (Fujii 1988 on Chinese, Sasse 1981 and Sasse forthcoming on Boni, Saeed 1985 on Somali). These studies hav e rev ealed that the difference between “topic-prominence” and “subject-prominence” cannot be v iewed as a selfcontained, autonomous, typological distinction, but is embedded in the larger context of a deeper-rooted typological differentiation which relates to role-assignment strategies. We can distinguish between primarily discourse-pragmatically oriented syntax, characterized by a basic inv entory of operations sensitiv e to text categories such as topic, focus, reference, definiteness, giv en/new etc., and primarily semantically oriented syntax, in which the notion of semantic roles such as actor, undergoer, instrument, etc. has precedence ov er discourse categories (Foley and Van Valin’s “reference-dominated” v s. “roledominated” languages, cf. 1984, 123). There are different types and degrees of amalgamation between the two (“subject-prominence” being just one of them), and there is also the possibility that the two systems work independently of each other. In other words, prominence typology, at its present lev el of dev elopment, represents an interesting first step in the direction of a general theory of

the interaction of discourse and syntax, but is in itself hardly sufficient, since the whole array of phenomena connected with the “pragmaticization” of syntax has not yet been taken into account. Another question that has been raised in some of the critical reactions to Li and Thompson’s 1976 paper (Sgall 1977, Myhill 1982, Sasse 1982) concerns the basic notions of subject and topic themselv es. Sasse (1982) is an attempt to characterize “subject-prominence” in terms of a more rigorous definition of “grammatical subject”, concluding that the specific Indo-European type of pragmaticization v ia a system of semanto-pragmatically gov erned and hierarchically structured v alence (“grammatical relations”), with an increasing tendency towards the actor/topic as the principal grammatical relation (formally defined in terms of case, agreement and/or word order), is a comparativ ely rare type of syntactic organization and perhaps restricted, in a narrow sense at least, to European and certain adjacent languages. He suggests reserv ing the term “subject” for the principal grammatical relation of languages of this type, i. e. to the traditional notion of “grammatical subject” as a formally defined syntactic relation, and consequently av oiding

56.  Prominence Typology

the term for all languages that do not function in a way similar to European languages. The other alternativ e would be to use the broader concept of “Aristotelian subject”, defined as the referential base of a bipartite predication, the “indiv idual a predication is about”. The danger inv olv ed in such a broad definition of “subject” is its potential ov erlapping with the topic notion. I would therefore recommend using the term “predication base” for “Aristotelian subjects” in general (including Indo-European subjects where they are predication bases) and restricting the term “subject” for the cases defined abov e. This would in end result lead to abandoning the term “subject” for Chinese, Tagalog, and many other languages for which it is now frequently used (for some discussion of Tagalog cf. Himmelmann 1987, for Quechua Pustet 1986). As for the notion of topic, the question still remains to what extent it can be defined as a syntactic category or relation. What is the difference and the connection between text topic, paragraph topic (cf. Hinds 1979), and sentence topic? Is the notion of “sentence topic” identical with that of “predication base”? How are the topics of Tp languages tied to the discourse? Moreov er, topics, ev en in Tp languages, must normally be described as clause-external (Foley-Van Valin 1984, 124 ff; Saeed 1984). In Somali, for instance, an obv ious example of a pragmatically oriented language, topics do not contribute in any significant way to the structure of the clause, while focus-marking on the other hand plays a considerable role: it is obligatory, it is inv olv ed in a number of spectacular morphosyntactic operations, it determines word order, and it ev en ov errides subject marking. Giv en this significant contribution of focus-marking to the syntax of the Somali clause in contrast with the relativ ely little syntactic significance of topics, it would be more adequate to speak of a “focus-prominent language” rather than of a “topic-prominent” language. (On “focus-prominence” see section 4. below.) Moreov er, is the “topic” of such a language a syntactic category at all? A number of further questions arise with respect to the typological status of the dichotomy “subject-prominence”/“topic-prominence” itself. The relationship between topiccomment structure and subject-predicate structure is imbalanced insofar as v ery many languages hav e the topic-comment construction without necessarily being Tp, whereas

1069

the subject-predicate construction is less frequent and typologically more specialized. In other words, the question arises if we are really dealing with a genuine typological opposition at the same level. Finally, there has been considerable criticism of the diachronic implications of Li and Thompson’s typology. Breiv ik (1984) finds the historical claims they make “not particularly well supported” (p. 3), because “they prov ide no documentation at all of mov ement between any two adjacent positions in their diagram” (p. 14). In fact, this is a serious shortcoming of the entire approach, especially in v iew of the fact that they had claimed a cycle of word order change SOV → SVO → SOV in Chinese in 1974, which has no connection to the typological schema proposed in their 1976 article. Similar doubts hav e been expressed by Munro 1979, 376: “Most disturbingly, they do not seriously address the problem that, although unrelated subject-prominent languages are found all ov er the world, the other groups draw their representativ es from one genetic group each (assuming that Japanese and Korean are related). How can this be consistent with the idea that all four types represent natural points around a cycle of development?”

3.

Syntactic Pivot Theory

In order to av oid some of the problems of Li and Thompson’s v ersion of prominence typology, Foley and Van Valin (1984) hav e proposed a theory of “syntactic piv ot”. The basic idea behind piv ot theory is that we might want to operate, metalinguistically, with a syntactically defined notion of “primary grammatical relation” or “central NP”. In Sp languages such as English, the syntactic piv ot coincides with the grammatical subject, but the piv ot notion is broader than the traditional subject notion and takes phenomena such as ergativ ity, activ e languages, etc. into account. The piv ot of a syntactic construction is defined as “the NP which is crucially inv olv ed in it; i. e., it is the NP around which the construction is built” (Foley and Van Valin 1984, 110). The role a piv ot plays in a syntactic construction can be exemplified by English subjects in operations such as “participai relativ ization”, “raising to subject”, “raising to object”, “deletion in complements”, and “ellipsis in coordinate structures”:

1070

XVII. Syntaktische Typologie II: Ansätze und Übersichts-darstellungen

Participal relativization (9a) The woman scolding the policeman is my mother (9b) *The policeman the woman scolding is my father (9c) The policeman being scolded by the woman is my father Raising to subject (10a) It seems that Paul caught the wombat (10b) Paul seems to have caught the wombat (10c) *The wombat seems Paul to have caught (10d) The wombat seems to have been caught by Paul Raising to object (11a) John expects that Paul will catch the wombat (11b) John expects Paul to catch the wombat (11c) *John expects the wombat Paul to catch (11d) John expects the wombat to be caught by Paul Deletion in complements (12a) Fred wants to see Marsha (12b) *Fred wants Marsha to see [him] (12c) Fred wants to be seen by Marsha Ellipsis in coordinate structures (13a) Oscar went to the store and spoke to Bill (13b) *Oscar went to the store and Bill spoke to [him] (13c) Oscar went to the store and was spoken to by Bill From these examples it is clear that the subject is the central NP for much of the crossclause syntax of English. In participai relativ ization only subjects may be relativ ized. In raising constructions only the subject of the embedded clause may occur as the subject or the direct object of the matrix clause. Finally, subjects are the targets of deletion in complements and in coordinated clauses. Piv ots need not be subjects in the traditional sense. In Dyirbal, an Australian ergativ e language, the controller and target of equi-NP deletion, participal relativ ization, ellipsis in coordinate structures, etc. is the absolutiv e NP, which denotes the undergoer or patient of a transitiv e v erb rather than the actor, as with the subject in European languages (Dixon 1972). It is characteristic of languages hav ing an unmarked choice for piv ot to possess a syntactic mechanism (such

as the passiv e in English and the so-called antipassiv e in certain ergativ e languages (Silv erstein 1976) in order to indicate a marked pivot NP. Foley and Van Valin differentiate between pragmatic pivots, whose choice is determined (or at least strongly influenced) by discourse factors (e. g. cross-clause reference and topicality) and which represent the syntactization of these discourse factors in clause-internal grammar, and semantic pivots, which are determined in strictly semantic and lexical terms. The choice of the latter is fixed, on the basis of a giv en v erb and its associated arguments. For instance in Choctaw, an American Indian language of the Muskogean family, piv ot choice is determined by the semantic hierarchy actor > undergoer > dativ e (Foley and Van Valin 1984, 115—6). Such languages are unlikely to hav e v oice distinctions, since there is no need for “repiv otization”, as in the case of pragmatic pivots. Generally speaking, piv ot theory surpasses the notion of subject prominence in that it cov ers the case of ergativ e languages, some of which, like Dyirbal, possess a primary grammatical relation which cannot be equated with the subject of accusativ e languages, but resembles it in acting as a controller and target of syntactic operations. It also tackles the problem of purely semantically determined controllers of syntactic rules (the Choctaw case), not dealt with at all by Li and Thompson. Finally, it has the adv antage of allowing for different piv ots in different constructions. Yet, it also still leav es many questions unanswered. It is doubtful whether a clear-cut distinction can be made between pragmatic piv ots and semantic piv ots. Both are essentially actant function-marking systems or parts thereof, so that the role of the participant which occurs as piv ot is always unambiguously deducible. In the case of the semantic piv ot it can be inferred on the basis of some role hierarchy directly; in the case of the pragmatic piv ot it follows from a hierarchy which determines the unmarked choice, and a v oice system which determines the marked choice. There are also languages where there is no unmarked choice at all (e. g. Tagalog), and where the semantic role of the piv ot is always indicated by some affix on the predicate. “Pragmatic piv ots” are thus basically no less role-indicating than semantic piv ots and differ from the latter only in the degree of discourse-sensitiv ity that gov erns their choice.

56.  Prominence Typology

The most fundamental objections against piv ot theory in the form presented by Foley and Van Valin (1984) concerns their definition of piv ots in terms of their property of being the target or controller of a syntactic operation. Generativ istic heritage has led the authors to an a priori assumption of deletion rules, raising, and similar transformationally influenced notions, without looking for functionally more interesting properties of piv ots. For instance, it is doubtful whether “ellipsis in coordinate structures” can really be adequately described as a deletion process. Could it not be described more simply and with more insight as an example of a double (coordinated) predicate ascribed to a single predication base, so that there is no ellipsis at all? Why should one assume (unless one is forced to do so by some theory-immanent principle such as transformations) that there are two coreferential predication bases, the second of which is deleted? Similar doubts arise with regard to participai relativ ization. One can only speak of target and control from a transformational point of v iew; on the basis of purely surface-oriented considerations all we can say is that the participles are oriented in a way suited to fulfilling their function as proper attributes of the predication base. “The woman scolding” is semantically different from “the woman being scolded”; this is not so much a question of syntactic organization but of what the sentence is about. To sum up, the whole question of target and control has been approached from the wrong angle in that the functional syntactic approach has not entirely been freed of arguments deriv ing from an autonomous syntactic approach. In order to tackle the piv ot problem more adequately, one first has to consider the fact that piv ots are “Aristotelian subjects”, i. e. predication bases of a bipartite predicativ e structure. Most of their syntactic behav iour, in particular the so-called “ellipsis” phenomena, can be deriv ed directly from this basic property. According to this v iew, there would be no deletion at all; a single predication base can be assumed, to which complex predicates or attributes are added. Secondly, piv ots participate in semantic role-marking systems, so that the degree of control they exert in syntactic constructions must primarily be considered a semantic phenomenon, e. g. a matter of the control agents hav e ov er actions. Consequently, the typology of piv otness

1071

could perhaps be dealt with more properly by considering the following four parameters: (i) how far predication bases are grammaticalized; (ii) whether the semantic role of a predication base is fixed or variable; (iii) whether the variability of the semantic role of the predication base correlates with a system of voice distinctions on the verb; (iv) how far predication bases are discoursesensitive (i. e. how far they refer to discourse or paragraph topics, “things spoken about”). Considerations similar to those abov e lead to a continuum of bipartite sentence structures: from clear discourse-sensitiv e, nonrole-marking predication bases (“topics, Chinese style”), v ia discourse-sensitiv e, but role-marking predication bases (“pragmatic piv ots” or “subjects” of v arious types), to chiefly role-indicating predication bases. In order to arriv e at a deeper functional explanation of the entire complex, the typology of predication base/piv ot should be embedded in a typology of what is usually called “basic clause structure”, one which includes parameters such as the general predication type (whether bipartite or non-bipartite), and the general structure of the role-assignment system. Some preliminary remarks along these lines may be found in Sasse (1991).

4.

Focus-Prominence and “Configurationality”

Prominence Typology, in its original sense as env isaged by Li and Thompson, was obsolescent in the mid-80s and would be purely a matter of linguistic historiography today were it not for the fact that the “configurationality” debate in Generativ e Grammar stirred new interest in the influence of discourse categories on syntactic structure. The configurationality parameter was introduced in the works of Ken Hale (1979, 1980, 1982, 1983) and others (e. g., Farmer 1980) to incorporate the traditional typological distinction between “fixed word order” languages and “free word order” languages into the theory of Generativ e Grammar. It is based on the assumption that the relev ant distinction has to do with a difference in the phrase structure configurations utilized by these two language types. The “configurational” type is

1072

XVII. Syntaktische Typologie II: Ansätze und Übersichts-darstellungen

characterized by a rich, multilev eled hierarchical phrase structure (exemplified by English), while the “non-configurational” type exhibits a “flat”, hierarchically undifferentiated phrase structure (exemplified by Warlpiri). For a brief historical account cf. Horvath 1986. It was predominantly European scholars who raised the issue of a connection between non-configurationality and topic-focus structure. The chief source of inspiration was research on Hungarian carried out by Julia Horv ath (esp. Horv ath 1981), Anna Szabolcsi (1983) and, most prominently, Katalin É. Kiss. Working on a generativ e approach to Hungarian syntax since the late 70s, É. Kiss came out with an influential paper on “Structural relations in Hungarian, a ‘free’ word order language” in 1981 (listed in the bibliography as 1981 a, cf. also 1981 b). Her main argument maintains that topic and focus are syntactically definable positions in Hungarian, an analysis already well-established outside generativ ist circles (cf. Behrens 1982, which includes further references). Roughly speaking, focus occupies the position immediately before the finite v erb. Topic precedes focus if the focus position is filled by a constituent other than the v erb. In her early work, É. Kiss still considers topic and focus as positions outside of S and hence, as landing sites for Mov e α, of which “Topic Raising” and “Focus Raising” are particular instances. She then addresses the typological problem and its consequences for the configurationality discussion in her 1984 paper, which presents a brief elaboration of Li and Thompson’s typology, with special reference to focus phenomena. Configurationality in Hungarian is discussed in É. Kiss (1987), with the conclusion that “the term ‘configurational’ v s. ‘non-configurational’, assigned to the language types represented by English and Hungarian, respectiv ely, do not seem to be felicitous: English and Hungarian rather seem to represent two different types of configurationality” (1987, 250). Hungarian was not the only language examined for topological restrictions on topic and focus. Further European languages with syntactic focus positions similar to that of Hungarian had come to light. One of them was Basque, cf. Ortiz de Urbina 1983 and later works; another case was Armenian as studied by Comrie 1984. The fixed positions focalized constituents occupy in these languages are usually those immediately in front

of the finite v erb. This fact lent further support to the hypothesis that there are systematic relations between so-called “non-configurationality” and discourse prominence, for what appeared to be “free word order” turned out to be “fixed word order”, determined by a factor other than prev iously looked for. Meanwhile, focus had become a fav orite topic in German linguistics thanks to the work of Werner Abraham, Tilman Höhle, Joachim Jacobs and Arnim v on Stechow. Since the status of German with respect to the configurationality question was still a matter of debate, it was of particular interest to explore to what extent v arious aspects of “focus projection” (a subject whose inv estigation goes back to Chomsky 1971 and which was rev italized by Culicov er & Rochemont 1983 and Rochemont 1986) had a bearing on this issue. Moreov er, a systematic relationship was claimed to exist between topicalization and v erb-second, in that the “initial field” (“Vorfeld”) could be regarded as a structural theme or topic position. In this respect, German would then be parallel to Hungarian with its two prev erbal topic and focus “fields” as described by É. Kiss. At the 6th Groningen Grammar Talks in 1984, Werner Abraham brought Hungarian and German specialists together in a colloquium dev oted to “Topic, Focus, and Configurationality” (Abraham & de Meij 1986). The main goal of the conference was the exploration of the systematic connections between “configurationality” on the one hand, and “topic prominence” or “focus prominence” on the other: “The term ‘focus prominence’ is used here to echo Li & Thompson’s term ‘topic prominence’. Hungarian, aside from being a free word order language, is a focus prominent language in the sense that Hungarian sentences hav e a fixed, syntactically marked focus position. Roughly speaking, the position immediately preceding the finite v erb is the focus position. Focus, when filled, accommodates only one constituent. Also, focus is typically the position where WHphrases must appear. [...] Hungarian is not the only language to hav e such a syntactic focus position: many more languages are known to hav e one, and it inv ariably is the position either immediately preceding or following the finite v erb (or INFL, if that is an independent constituent). As these are all free word order languages, the correlation leads to the assumption that focus prominent

56.  Prominence Typology

languages tend to be nonconfigurational” (Abraham & de Meij 1986, 4—5). The fact that the focus position and its influence on the topology of the other syntactic constituents determines fixed ordering rules comparable to those determined by grammatical relations in “configurational” languages, though triggered by a different factor (namely, discourse categories rather than grammatical relations), has led to notions such as “focusconfigurationality” or, more generally, “discourse-configurationality”, which hav e become one of the central issues of word-ordertypological inv estigation within the generativ e paradigm. Research undertaken in the framework of Theme Group 2 (“Constituent Order”) of the EUROTYP project has recently tackled the question from a broader European typological perspectiv e. É. Kiss, in a summarizing paper (É. Kiss 1993 MS), distinguishes between two subtypes of “discourse-configurational” languages which may ov erlap, and do so to a large extent: topic prominent languages in which “constituents extracted from the v erbal projection into a syntactic subject-of-predication position inv ariably function as topics/notional subjects”, and focus prominent languages which hav e a “structural focus”, i. e. which “employ a predicate-internal structural relation to express the discourse-semantic function ‘focus’”. European languages claimed to be “clear cases” of both topic and focus prominence are, among others, Basque, Catalan, Italian, Hungarian, Lov ari (a Roma dialect spoken in Hungary), Polish, Russian, Rumanian, Greek, Turkish, Armenian, and certain Caucasian languages. Howev er, not all topic-prominent languages hav e a structural focus position, (e. g. Czech and Slov ak do not hav e one), and not all focus-prominent languages are also topic-prominent, e. g. Welsh, Breton and Finnish are claimed to hav e a structural focus position but not to be topic prominent. The new generativ ist approach, including the notions of “focus prominence”, “focus configurationality” and “discourse configurationality”, has contributed considerably to an elaboration of the original v ersion of “prominence typology”. It has stimulated a lot of (still ongoing) research, uncov ering a considerable number of interesting new details concerning the impact of discourse categories on syntactic structure. In particular, the recognition of the central role of a focus position (much more central than that of the

1073

topic position) in determining linear order in a large number of languages, certainly amounts to a major progress in generativ e syntax. Nev ertheless, the main points of critique from section 2 abov e can be repeated here. Too much weight has notoriously been lent in generativ e grammar to alleged correlations of linearity, hierarchical order, and representability by some sort of X bar dev ice. This has led, in this case, to echoing the Li & Thompson dichotomy and incorporating it into the otherwise scarcely modified theoretical apparatus simply in order to keep the hierarchical notion of configurationality intact. Adv ocates of the discourse-configurationality approach v iew as their major achiev ement the discov ery (in some languages) of topic and focus as syntactically definable positions. This pride is justified, but only to a certain extent. It does not follow from the syntactic definability of a phenomenon that it can reasonably be represented in the form of hierarchically organized phrase structure trees. Considerable ev idence still exists that a language such as Hungarian is not configurational in the usual sense employed by generativ ists, but displays a type of “configurationality” of a quite different make-up, namely one that has nothing to do with a hierarchically organized multi-layered X bar structure but with a flat, “positionally“ organized chain of syntactic slots. It is not at all clear how this can be compared and represented in a way similar to the hierarchical structure usually posited for English. Closely related to this is a further critical issue. It has been repeatedly pointed out (cf. already section 2. and 3. abov e) that discourse-oriented and relations- or role-oriented syntax cannot so easily be handled by means of a simple plus/minus-parameter. In a recent paper, A. Lötscher (1992) has again drawn attention to the fact that “the contrast between subject-prominence and topic-prominence need not result in mutually exclusiv e coding systems for the surface coding of different grammatical and/or pragmatic functions. Subject-prominence and topic-prominence can well liv e together in one grammatical system, as can be seen from Russian and German” (Lötscher 1992, 104). Ev en É. Kiss admits that there are “intermediate cases”. On closer inspection, howev er, a tremendous number of languages rev eal themselv es as “intermediate cases”. In most of the Slav onic languages, Modern Greek, and Albanian, for instance, grammatical relations do not deter-

1074

XVII. Syntaktische Typologie II: Ansätze und Übersichts-darstellungen

mine word order as they do in English, although they play an important role morphosyntactically: Nominati v e/Accusati v e morphology, transitiv ity, middle v oice/passiv e morphology, etc., are main determinants of morphosyntactic organization. Word order is discourse-pragmatically determined, including topicalization strategies, but is by no means “discourse-configurational” like that in Hungarian, in that there is no fixed focus position. These languages are usually characterized by an extraordinary freedom in exploiting “focus-in-situ”. This leav es us with two unsolv ed questions: (1) To what extent are “topic” and “focus” syntactic notions, and in what sense are they considered to be syntactic? (2) Is it adequate to treat these notions on the same lev el of linguistic analysis as “subjects” and “objects” and hence, as opposite types of determinants for “phrase structure”? The original Li & Thompson dichotomy was clearly English- and Chinesebiased, while the new dichotomy is Englishand Hungarian-biased. In both cases English plays a central role, but, as Lötscher puts it, “grammatical functions and theme-rheme articulation pertain to distinct content lev els, and it appears as rather unpractical to hav e to encode these two dimensions by one and the same semiotic dev ice, i. e. word order. In such a perspectiv e, configurational languages appear rather as an operational accident in diachrony with respect to the efficient and economic exploitation of possible resources for coding propositional and pragmatic content, although, as the example of English shows, the consequences are not necessarily as sev ere as might be expected theoretically” (Lötscher 1992, 104—5). Thus, the dichotomy remains a simplification. It is still doubtful whether its two opposites are based on the same criteria, and it is also doubtful whether the languages used to exemplify it are really representativ e of a v alid typological opposition div iding the linguistic univ erse into two clear-cut sets. Most languages do not display the type of subjectprominence characteristic of English, which turns out to be the exception rather than the rule, and most languages do not display the radical type of “discourse-configurationality” characteristic of Hungarian. The polarization inv oking these two languages as “ideal types” thus creates an incorrect and misleading picture of the typological state of affairs. Further refinement is certainly necessary in the framework of an approach that treats gram-

matical relations, phrase-structural configurationality, topic-comment articulation, focus-background articulation and perhaps other types of discourse-sensitiv e highlighting strategies as separate dev ices of grammatical organization whose coincidence or independence of each other v aries along a parameter of basic clause structure inv olv ing a considerable number of different factors. There can be no doubt, howev er, that the phenomenon of “focus” plays an important role in the syntactic organization of many more languages than prev iously assumed. More attention should therefore be paid to focus phenomena in future syntactic research, not only with respect to word order, but also with respect to intonation, morphological dev ices, particles, and perhaps other means of expression that serv e as markers of discourse-sensitive highlighting.

5.

References

Abraham, W. and S. D e Meij (eds.) 1986. Topic, focus, and configurationality. Amsterdam. Behrens, L. 1982. Zur funktionalen Motiv ation der Wortstellung: Untersuchungen anhand des Ungarischen. München. Breivik, L. E. 1984. On the typological distinction between subject-prominence and topic-prominence. Folia Linguistica Historica 5. 3—23. Chafe, W. L. 1976. Giv enness, contrastiv eness, definiteness, subjects, topics, and point of v iew. Subject and Topic, ed. by Ch. N. Li, 25—55. New York. Chomsky, N. 1971. Deep structure, surface structure and semantic interpretation. Semantics: An Interdisciplinary Reader in Philosophy, ed. by D. Steinberg and L. Jakobov its, 183—216. Cambridge. Comrie, B. 1984. Some formal properties of focus in Modern Eastern Armenian. Annual Armenian Linguistics 5. 1—21. Culicover, P. W. and M. S. Rochemont. 1983. Stress and focus in English. Language 59. 123—65. D ixon, R. M. W. 1972. The Dyirbal language of North Queensland. Cambridge. Farmer, A. 1980. On the interaction of morphology and syntax. Doctoral Dissertation, MIT, Cambridge, MA. Fillmore, Ch. 1968. The case for case. Univ ersals in Linguistic Theory, ed. by E. Bach and R. Harms, 1—88. New York. Foley, W. and R. D . Van Valin Jr. 1984. Functional Syntax and Universal Grammar. Cambridge. Fujii, F. 1988. The typological status of Chinese and its implications. München.

56.  Prominence Typology

Hale, K. 1979. On the position of Walbiri in a typology of the base”. Mimeographed, MIT. —. 1980. Remarks on Japanese phrase structure: Comments on the papers on Japanese syntax. Theoretical Issues in Japanese Linguistics, ed. by Y. Otsu and A. Farmer. MIT Working Papers in Linguistics 2. —. 1982. Preliminary remarks on configurationality. Proceedings of NELS 12 ed. by J. Pustejov sky and P. Sells. Amherst, MA. —. 1983. “Warlpiri and the grammar of non-configurational languages”. Natural Language and Linguistic Theory 1. 5—47. Himmelmann, Jr., N. 1987. Morphosyntax and Morphologie — Die Ausrichtungsaffixe im Tagalog. München. Hinds, J. 1979. Organizational pattern in discourse. Syntax and Semantics, v ol. 12 (Discourse and Syntax), ed. by T. Givón, 135—57. New York. Hockett, Ch. F. 1958. A Course in Modern Linguistics. New York. Horvath, J. 1981. Aspects of Hungarian syntax and the theory of grammar. Doctoral dissertation, University of Los Angeles. —. 1986. Remarks on the configurationality-issue. Topic, Focus, and Configurationality, ed. by W. Abraham and S. de Meij, 65—87. É. Kiss, K. 1981a. Structural relations in Hungarian, a ‘free’ word order language. Linguistic Inquiry 12. 185—213. —. 1981b. Topic and focus: The operators of the Hungarian sentence. Folia Linguistica 15. 305— 330. —. 1984. Topic-focus languages: A dev elopment of Li and Thompson’s typology. The origins and orginality of American culture, ed. by F. Tibor, 773— 82. Budapest. —. 1987. Configurationality in Hungarian. Budapest and Dordrecht. —. 1993 MS. Discourse-configurationality in the languages of Europe. Unpublished manuscript. Li, Ch. N. (ed.) 1976. Subject and Topic. New York. Li, Ch. N. and S. A. Thompson. 1974. Historical change of word order: A case study in Chinese and its implications. Historical Linguistics. Proceedings of the First International Conference on Historical Linguistics, v ol. 1, ed. by J. M. Anderson and

1075

Ch. Jones, 199—217. —. 1976. Subject and Topic: A new typology of language. Subject and Topic, ed. by Ch. N. Li, 457— 89. Lötscher, A. 1992. The relativ ity of subject/topic prominence in German, English and Russian. Folia Linguistica 26. 95—105. Munro, P. 1979. Rev iew article on Li 1976. Language 55. 372—80. Myhill, J. 1982. Word order and marginal ergativ ity. Papers from the Eighteenth Regional Meeting, Chicago Linguistic Society, 377—89. Chicago. Ortiz de Urbina, J. M. 1983. Empty categories and focus in Basque. Studies in the Linguistic Sciences 13. 133—56. Pustet, R. 1986. Zur Frage der Univ ersalität des “Subjekts”: Das Ayacucho-Quechua. Arbeitspapiere des Instituts für Sprachwissenschaft der Universität zu Köln, Nr. 51. Köln. Rochemont, M. S. 1986. Focus in generativ e grammar. Amsterdam. Saeed, J. I. 1985. The syntax of focus and topic in Somali (= Cushitic Language Studies 3). Hamburg. Sasse, H.-J. 1981. ‘Basic word order’ and functional sentence perspectiv e in Boni. Folia Linguistica 15. 253—90. —. 1982. Subjektprominenz. Fakten und Theorien. Festschrift für Helmut Stimm zum 65. Geburtstag, hrsg. v . S. Heinz und U. Wandruszka, 267—86. Tübingen. —. (forthcoming). A Grammar of Boni. Mouton Grammar Library. —. 1991. Predication and sentence constitution in univ ersal perspectiv e. Semantic univ ersals and univ ersal semantics, ed. by D. Zaefferer, 75—95. Berlin etc. Sgall, P. 1977. Rev iew of Li 1976. The Prague Bulletin of Mathematical Linguistics 27. 65—70. Silverstein, M. 1976. Hierarchies of features and ergativ ity. Grammatical categories in Australian languages, ed. by R. M. W. Dixon, 112—71. Canberra. Szabolcsi, A. 1983. Focussing properties, or the trap of first order. Theoretical Linguistics 10. 125—46. Teng, S.-H. 1974. Double nominativ es in Chinese. Language 50. 455—73.

Hans-Jürgen Sasse, Köln (Deutschland)

1076

57. 1. 2. 3. 4. 5.

1.

XVII. Syntaktische Typologie II: Ansätze und Übersichts-darstellungen

Relational Typology Ergative vs. Accusative Rules and Languages Active Rules and Languages Other Systems of Mapping Conclusion References

Ergative vs. Accusative Rules and Languages

1.1. Advances in Past Research The typological distinction which will be discussed in the present article is concerned with different types of mapping between thematic roles, such as agent and patient, syntactic cases, such as nominativ e and accusativ e, structural-topological positions and sentence topic. This parameter is also known as contentiv e typology (cf. Klimov 1979, 1985), relational typology (cf. Plank 1985) or alignment typology (cf. Nichols 1990). The major language types of this typology are ergativ e, accusative, and active. We will start our discussion with the ergativ e-accusativ e distinction. This distinction is not consistent within all sentence types of a language, but it has to manifest itself in basic transitiv e and intransitiv e clauses in order to classify this language as ergativ e or accusativ e. The concept of basic clause will also help to distinguish the basic ergativ e construction of an ergativ e language from the non-basic passiv e construction of an accusativ e language. The most conv enient concept of basic clause for our purposes is formulated in (1): (1) A clause type S is basic with respect to argument mapping in a language L if: (a) the predicate of S is morphosyntactically in the unmarked voice, and (b) the predicate of S selects an agent (and a patient, if the predicate is multivalenced). An agent is defined by most researchers as a participant which the meaning of the predicate specifies as doing or causing something, usually intentionally. A patient is characterized as a participant that has something happen to it and that is affected by what happens to it, for instance, because it is mov ed or changed in a particular way. Let us illustrate first the more familiar accusativ e, or nominativ e, language type

with examples from German, an Indo-European language, and Hungarian, a FinnoUgric language (I will only separate and analyse morphological markers which are relev ant for the present discussion. Standard European languages will not be glossed in general, unless the case marking or other morphosyntactic details are at issue. The abbrev iations are the following: 1, 2, 3 — 1st, 2nd, 3rd person respectiv ely, A, B — agreement prefix set A or B (Guarani); ABS — absolutiv e (used interchangeably with NOM); ACC — accusativ e; ALL — allativ e; APASS — antipassiv e; AUX — auxiliary; CLT — clitic; Cn — nominal class; DAT — dativ e; DEM — demonstrativ e/deictic pronoun; DIR — directiv e; ERG — ergativ e; F — feminine; GEN — genetiv e; IMPF — imperfect: IND — indicativ e; INSTR — instrumental; M — masculine; NOM — nominativ e; NFUT — non-future; RFL — reflexiv e; RN — relational noun; PAR — partitiv e; PASS — passive; PF — perfect; PL — plural; PPRS — present participle; PRS — present; PROG — progressive; SG — singular). (2a) Ein Mann kommt. a(NOM) man(NOM) comes ‘a man is coming’ (2b) Das Mädchen schlägt ein-en the(NOM) girl(NOM) hits a-ACC Mann. man(ACC) ‘the girl is hitting a man’ jön. (3a) Egy ember a(NOM) man(NOM) comes ‘a man is coming’ (3b) A lány üt egy ember-t. a girl(NOM) hits a man-ACC ‘the girl is hitting a man’ An accusativ e language uses the most unmarked case, the nominativ e, for the sole argument of a basic intransitiv e sentence and for the agent of a basic transitiv e sentence. A more marked case, the accusativ e, codes the patient of a basic transitive sentence. Linguistic tradition explicitly or implicitly uses the correlation between nominativ e and agent in an accusativ e language as a defining property of the syntactic case nominative and of an allegedly univ ersal concept of subject (cf. Keenan 1976). This explains why many characterizations of an accusativ e type of rule and language are v ery similar or iden-

57.  Relational Typology

tical to the formulation in (4), cf. Martinet (1958), Comrie (1978), Mallinson/Blake (1981, chap. 2), Dixon (1979, 1987), Croft (1990, chap. 5): (4) A rule R of a language L is accusative if and only if R treats the subject of an intransitive clause in the same way as the subject of a transitive clause, and differently from the object of a transitive clause. L is consistently accusative if and only if all the rules of L are accusative. (4) is incomplete with respect to a major property of v erb arguments across languages. Verb arguments are, in general, hierarchically organized and this manifests itself in the fact that some arguments are more readily inv olv ed in the rules of a language than others. Giv en the fact that nominativ e agents outrank accusativ e patients in accusativ e languages, nominativ e agents will be more readily inv olv ed in the rules of this language type than accusativ e patients (cf. Keenan/Comrie 1977, Johnson 1977, and (18) below). This dominance of nominativ es and/or agents in accusativ e languages is illustrated in the abov e examples (2) and (3) with respect to v erb agreement and basic linear order. In German and Hungarian the predicate agrees with nominativ e arguments, which also precede accusativ e arguments in the basic word order. In Hungarian, the predicate also agrees with accusativ e arguments, but only under certain conditions (cf. section 1.3.2. below). The most important accusativ e rule, which is sometimes the only one mentioned in a formulation such as (4), is the accusativ e mapping of cases to thematic roles, i. e. the fact that the sole argument of an intransitiv e v erb and the agent of a transitiv e v erb are coded by the same case, the nominativ e. This is often referred to as morphological accusativ ity. In accusativ e languages agents are also the preferred sentence topics (in the sense of pragmatic aboutness, cf. Reinhart 1981). These mapping preferences between nominativ e, agent and topic hav e a special status, since they are part of the definition of subject. Correspondingly, the mapping default between accusativ e and patient characterizes the notion of (direct) object. There is massi v e empirical ve idence against the v iability of grammatical relations such as subject and object in univ ersal grammar. As repeatedly shown in the literature,

1077

these notions are not applicable to ergativ e languages (cf. Lyons 1968, Van Valin 1977, Sasse 1978, Shaumyan 1985). More than that, they are not ev en applicable to standard accusativ e languages such as German and other allegedly ‘subject’-oriented languages (cf. Vennemann 1982, Primus 1987, 1993). The elimination of the traditional grammatical relations and the reformulation of (4) in terms of thematic roles, such as in (4'a) is suggested more or less explicitly, for instance, by Van Valin (1977) and Sasse (1978). The bracketing of the case names indicates the common belief that they hav e a secondary and derivative status. (4') A rule R of a language L is accusative if and only if (a) R treats the only argument of a basic intransitive clause in the same way as the agent of a basic transitive clause, and differently from the patient of a basic transitive clause, and (b) the (nominative) agent dominates the (accusative) patient with respect to the applicability and conditioning of R. L is consistently accusative if and only if all the rules of L are accusative. (4'b) incorporates the insight of the hierarchy approaches to univ ersal grammar and restricts the definition to basic sentences. The dominance relation mentioned in (4') will be made more precise in section 1.3. As a first approximation, a v erb argument A can be assumed to dominate another v erb argument B with respect to a rule R, if R applies to A and not to B or if A has fewer restrictions with respect to R than B. In basic order rules, if A dominates B, then A precedes B in the unmarked case. On the basis of (4'), one can not only correctly describe but also explain the v erb agreement and basic order illustrated in the German and Hungarian examples above. Let us proceed by showing typical traits of ergativ e languages, which can be found for instance in Av ar, a Caucasian language, and Dyirbal, an Australian language. Compare (5)—(6) to (2)—(3): Avar (Charachidzé 1981, 144 f) (5a)yas y-orč’ana. girl(ABS,C2)C2-woke-up ‘a/the girl woke up’ (5b)di-cca y-osana yas. I-ERG,C1 C2-tookgirl(ABS,C2) ‘I took a/the girl’ Dyirbal (Dixon 1972, 59)

1078

XVII. Syntaktische Typologie II: Ansätze und Übersichts-darstellungen

yaa baniɲu. (6a) bayi DEM(ABS) man(ABS) come(NFUT) ‘man is coming/came’ yaa baŋgun (6b) bayi DEM(ABS) man(ABS) DEM(ERG) ugumbiu balgan. woman(ERG) hit(NFUT) ‘woman is hitting/hit man’ An ergativ e language uses the most unmarked case, the absolutiv e, for the only argument of a basic intransitiv e sentence and for the patient of a basic transitiv e sentence. A more marked case, the ergativ e, codes the agent of a basic transitiv e sentence. Very common characterizations of an ergativ e type of rule and language are quite similar or identical to the formulation in (7), cf. Martinet (1958), Comrie (1978), Mallinson/Blake (1981, chap. 2), Dixon (1979, 1987), Croft (1990, chap. 5): (7) A rule R of a language L is ergative if and only if R treats the subject of an intransitive clause in the same way as the object of a transitive clause, and differently from the subject of a transitive clause. L is consistently ergative if and only if all the rules of L are ergative. One conspicuous ergativ e rule, the ergativ e mapping of syntactic cases to thematic relations, in particular that of absolutiv e to patient and ergativ e to agent in basic transitiv e sentences, defines what is commonly called the ergativ e construction, surface ergativ ity or morphological ergativ ity. Languages with syntactic ergativ e rules are also called deep ergative. The formulation in (7) has sev eral disadv antages which hav e been eliminated as linguistic research about ergativ e languages adv anced. The first disadv antage of (7) is that it does not restrict the characterization of the ergativ e construction to basic sentences. Without this qualification, howev er, the ergativ e construction can hardly be distinguished from the passiv e construction of an accusativ e language, as repeatedly pointed out in the literature, v ery explicitly for instance by Comrie (1988). Note that in the man is hit by the girl, a passiv e v ariant of the German example (1 b), the patient is coded by the unmarked case as is the subject of an intransitiv e clause, and the agent is coded differently by a marked form. The older treatment of the ergativ e construction as a passiv e construc-

tion (cf. Grierson 1903 f, Boas 1911 f, Meillet 1921 f, Schuchardt 1928) has been abandoned by most linguists. The second disadv antage of (7) is that it does not take the hierarchical organization of v erb arguments into consideration. Dixon (1979, 1987), for instance, pointed out that an absolutiv e patient exhibits the same behav iour in an ergativ e rule as the subject, i. e. nominativ e agent in an accusativ e rule. Dixon coins the term syntactic pivot for the v erb argument in both ergativ e and absolutiv e languages that exhibits the syntactic properties listed by Keenan (1976) as characteristic for subjects. Similarly, Sasse (1978) treats the absolutiv e patient as the primary grammatical relation in ergativ e languages. Finally, Mel’čuk (1979) defines grammatical subject as that argument noun phrase that is most similar to the only argument of an indicativ e clause headed by a v erb hav ing exactly one thematic argument. This definition allows us to treat an absolutiv e argument in an ergativ e language as a grammatical subject, since it is syntactically similar to the only argument of an intransitive verb. The dominance of the absolutiv e patient in ergativ e languages is illustrated in the abov e examples (5) and (6) with respect to v erb agreement and basic linear order. In Av ar, the predicate agrees only with absolutiv e arguments in nominal class and number. In Dyirbal absolutiv e patients precede ergativ e agents in the unmarked order. The dominance of the absolutiv e patient ov er the ergativ e agent in ergativ e languages is implicitly assumed by many researchers dealing with this language type and is explicitly stated, for instance, in Van Valin (1977), Sasse (1978), Wierzbicka (1981), Mallinson/ Blake (1981). These authors establish this dominance by treating the absolutiv e patient as the grammaticalized sentence topic in ergative languages. The third disadv antage of (7) is the use of subject and object as in accusativ e languages, and this masks the ‘subject’-like behav iour of absolutiv e patients. A reformulation of (7) as in (7') eliminates the abov e-mentioned disadv antages by capturing v aluable insights from the various researchers quoted above. (7') A rule R of a language L is ergative if and only if (a) R treats the only argument of a basic intransitive clause (S) in the same way as the patient of a basic transitive clause (P),

57.  Relational Typology

and differently from the agent of a basic transitive clause (A); (b) The (absolutive) S and P dominate the (ergative) A with respect to the applicability and conditioning of R. L is consistently ergative if and only if all the rules of L are ergative. S, A, and P (or O) are v ery popular abbrev iations for grammatical relations in both accusativ e and ergativ e languages. A is reserv ed for the argument that receiv es the treatment normally accorded to the agent of a basic transitiv e clause. P (or O) names the function of the argument that receiv es the treatment normally accorded to the patient of a basic transitiv e clause. S abbrev iates the function of the only argument of a basic intransitive clause. The bracketing of the cases in (7'b) indicates the common belief about the secondary, deriv ativ e nature of cases. As noted abov e, linguistic tradition explicitly or implicitly uses the correlation between the nominativ e and the agent in a basic transitiv e clause in an accusativ e language as a defining property of nominativ e and subject. Correspondingly, the correlation between the ideally zeromarked case and the patient of basic transitiv e clauses in ergativ e languages is used as a defining property of the absolutiv e. This thematic functional v iew also explains the common usage of the terms accusativ e, ergativ e and dativ e as relating to the cases that code as a default the patient, the agent, and respectiv ely the recipient of multi-v alenced v erbs. Therefore, absolutiv e and nominativ e are usually termed differently by researchers dealing with the ergativ e-accusativ e typology despite their obv ious common property of hav ing the least complex allomorphism, ideally the zero allomorph, and of hav ing the least restricted subcategorization behaviour. The formulation in (4') and (7') makes a further explicit distinction which has been neglected or only implicitly assumed in past research. It is that between an ergativ e rule and an ergativ e language. Past research has concentrated on capturing the typological distinction between ergativ e and accusativ e languages as a whole. The problem is that, as with other typological classifications, language types are not always consistent. Consistent ergativ e languages, in particular, are rare. According to Dixon (1979, 63), it appears that there are no languages that are fully ergativ e, at either the syntactic or the morphological level.

1079

The syntactic split behav iour of ergativ e languages can be illustrated on the Av ar examples in (5). Whereas case marking and v erb agreement match the definition of an ergativ e rule in (7'), the basic word order rule of Av ar is an accusativ e rule, since an agent tends to precede a patient. Dyirbal has a morphological split, which is also v ery common among ergativ e languages. A systematic cross-linguistic account of the patterns of morphological split in ergativ e languages is Silv erstein (1976). In Dyirbal, nouns and determiner-like 3rd person pronouns are case marked ergativ ely and 1st and 2nd person pronouns pattern accusativ ely. The ergativ e case marking is illustrated in (6), the accusativ e case marking on pronouns is presented in (8): Dyirbal (Dixon 1972, 60) baniɲu (8a) ɲaa I(NOM) come(NFUT) ‘I’m coming/came’ ŋinuna balgan (8b) ɲaa I(NOM) you(ACC) hit(NFUT) ‘I’m hitting/hit you’ ŋayguna balgan (8c) ŋinda you(NOM) me(ACC) hit(NFUT) ‘you’re hitting/hit me’ The case marking in (8) is termed accusative, since the pronominal form for the patient is more complex than the form for the agent and since the latter is generalized ov er intransitiv e v erbs. Most of the syntactic rules of Dyirbal are ergativ e, so that the accusativ e case marking rule for pronouns is syntactically irrelev ant according to Dixon (1972, 49 f). Since Dyirbal seems to be one of the rare instances of a syntactically rather consistent ergati v e language, the ergati v e-accusati v e parameter has not been considered to be a major syntactic parameter within different lines of research. Within Generativ e Grammar for instance, Lev in (1983) and Marantz (1984) are the first systematic accounts of syntactic ergativ e languages such as Dyirbal and Eskimo. The formulation in (4') and (7') allows a language to hav e both ergativ e and accusativ e syntactic rules in principle. The more basic concept in (4') and (7') is that of an ergativ e or accusativ e rule, which then characterizes the respective language type. A fact that cannot be ignored by typological research is that ergativ e syntactic rules as defined abov e in (7') are well attested and dis-

1080

XVII. Syntaktische Typologie II: Ansätze und Übersichts-darstellungen

tributed across the languages of the world. The rule orientation of the parameter makes it immune against the criticism that consistent syntactic ergativ e languages are exceptional. An ov erv iew of the accusativ e rules in ergativ e languages, as presented by S. Anderson (1976) in order to sav e the univ ersality of the subject concept, cannot refute the parameter as defined in (4') and (7'), but it would seriously challenge its v iability in terms of consistent language types. The following information about the distribution of languages with morphological ergativ ity giv es a first impression of their widespread distribution. Nichols (1992, 90, and Appendix 2) offers information about 168 languages with ov ert pronominal, nominal or v erbal inflection. Ca. 25% of these language are ergativ e with respect to at least one of these ov ert inflectional categories. Ov ert ergativ e inflection is fav oured on nouns in Nichols’ sample and is found in 29 of the 74 languages with ov ert noun inflection (39%). The remaining 45 (61%) are languages with ov ert accusativ e inflection on nouns. Similar results are offered by Siewierska (1994) on the basis of 124 languages from Europe (including the Caucasus). 122 languages hav e ov ert inflection on nouns, pronouns or v erbs and can be classified as ergativ e or accusativ e, and since 5 of them hav e both ergativ e and accusativ e alignment, they hav e been counted twice, yielding a total of 127 languages. 41 of these (32%) hav e ergativ e marking for at least one of the categories mentioned abov e and 86 (68%) hav e accusative marking. Morphological ergativ ity is a genetically and diachronically quite stable feature (cf. Nichols 1992, chap. 5) so that it tends to accumulate within certain language groups and families. Morphologically ergativ e languages are more dominant than av erage among the Australian languages, the Indic languages, the Caucasian languages, the Mayan, TupiGuarani and the Carib languages within the Amerind phylum, the Eskimo languages, and the Chukchi languages. Some language isolates are also ergativ e, such as Basque, Burushaski, Ket and the extinct Hurrian and Sumerian. There are also language groups and families where ergativ e coding of thematic roles is rarely found. These are the European languages of the Indo-European branch, the language groups within Africa (Khoisan, Niger-Kordofanian, Nilo-Saharan, Afro-Asiatic

including Semitic), the Altaic languages, the Uralic languages, the Elamo-Drav idian languages and the Austro-Tai languages. It has been claimed that languages with accusativ e case marking may also be syntactically split. This assumption is often referred to as the ergativ e or unaccusativ e hypothesis for accusativ e languages. Some intransitiv e v erbs in accusativ e languages are often analysed as syntactically unaccusativ e or ergativ e (cf. Perlmutter 1978, den Besten 1985, Burzio 1986, Grewendorf 1989, Dowty 1991). This assumption is illustrated on the following examples from German, Dutch, Italian and French: (9a) Also hat das Kind heute im Garten den Zaun gebrochen. ‘Therefore the child broke the fence today’ (9b) Also ist heute im Garten der Zaun gebrochen. ‘Therefore the fence broke in the garden today’ (10a) Ich bin weggegangen. (10b) Ik ben vertrokken. (10c) Sono partita. (10d) Je suis partie. (11a) Ich habe gelacht. (11b) Ik heb gelachen. (11c) Ho riso. (11d) J’ ai ri. (12a) der geschriebene Brief, der weggegangene Gast ‘the written letter’, ‘the visitor who left’ (12b) *das gelachte Kind, *der gearbeitete Linguist ‘the child who laughed’, ‘the linguist who worked’ The examples show that there are two types of intransitiv e v erbs in these languages: the v erbs and sentence patterns in (9b) and (10) were called ergativ e or unaccusativ e, and the v erbs and sentence patterns in (11) were analysed as accusativ e or unergativ e. The different types of intransitiv e v erbs select different auxiliaries in the languages illustrated in (10) and (11). The term ergativ e was justified by the definition of ergativ ity giv en in (7), for instance, in Grewendorf (1989, 1). The subjects of the ergativ e intransitiv e v erbs share some properties with the objects of transitiv e v erbs. This is quite obv ious in (9), where a lexical deriv ation rule connects a transitiv e v erb with its intransitiv e v ariant (described as ergativ ity

1081

e. g. by Martinet (1958) and Lyons (1968)). (9) also shows that den Zaun and der Zaun share the same basic position after temporal and locativ e modifiers, whereas subjects of transitiv e v erbs, das Kind in (9a), tend to precede these modifiers. (12) shows that the participial form of a v erb can be used as a noun modifier in German and that the modified noun has to be the subject of an ergativ e v erb (e. g. weggegangen ‘left’) or the object of a transitiv e v erb (e. g. geschrieben ‘written’). Within Relational Grammar (Perlmutter 1978) and Generativ e Grammar (den Besten 1985, Burzio 1986, Grewendorf 1989) these facts were treated syntactically by analysing the surface subjects of ergativ e v erbs as deep objects. Linguists working with ergativ e languages hav e criticized the use of the ergativ ity parameter for the phenomena mentioned in connection with (9)—(12), since the facts presented abov e are different from the morphological and syntactic ergativ ity found in ergativ e languages (cf. Comrie 1978, 391 f, Dixon 1987, 5 f). Let us look at morphological ergativ ity first. The case coding exhibited in (9)—(11) does not match the definition in (7) or (7'). The other properties fall under the definition in (7) but not under that in (7'). The terminological misunderstanding originates from the flaw in the definition giv en in (7), which disregards the dominance of absolutiv e patients in an ergativ e rule. Only (7'b) captures the insight that absolutiv e patients show subject-like behav iour in an ergativ e rule. The syntactic treatment of the allegedly ergativ e intransitiv e v erbs in accusativ e languages is explicit and v ery successful in showing the object-like behav iour of the allegedly ergativ e intransitiv e arguments. Ironically, the syntactic approaches cited abov e supply the crucial ev idence against their own terminological usage in showing that the relev ant arguments are underlying objects, and not syntactic piv ots or primary arguments as expected from an ergativ e rule according to Dixon (1979) and Sasse (1978). Note for instance that an ergativ e basic order rule, as illustrated in Dyirbal in (6b), will posit the absolutiv e patient in sentence initial and v erb phrase external position (cf. Dixon 1972, 137 f). By contrast, the basic position of the allegedly ergativ e subject in (9b) is v erb phrase internal. The semantic analyses of the two types of intransitiv e v erbs mentioned abov e are neutral with respect to the ergativ e hypothesis.

The ov erv iew in Van Valin (1990) and Dowty (1991) clarifies the matters typologically: the two types of intransitiv e v erbs illustrated in (9)—(11) abov e characterize what is now commonly called split intransitiv ity. Split intransitiv ity is the defining property of the active language type (cf. section 2. below). 1.2. Problems with Past Approaches The major disadv antage of past approaches to ergativ ity is the assumption that in ergativ e rules the patient or object dominates the agent or subject. Note that due to the functional v iew of syntactic cases the dominance of the patient is implicitly assumed also in terms of object dominance or absolutiv e dominance. This hypothesis has serious empirical disadv antages and leads to further problematic assumptions, which are summarized in (13): (a) An ergative rule formulated in terms of patient prominence or object prominence cannot capture the data appropriately. (b) Languages with ergative mapping of cases to thematic roles are apparently never fully consistent. There are many languages in which case mapping is the only ergative rule, all other rules being accusative. By contrast, there is apparently no language with an accusative case mapping and ergative syntactic rules (cf. Comrie 1978, Dixon 1987, Croft 1991). (c) The assumption that thematic roles are inconsistently ranked within split ergative languages and across languages is very problematic. (d) The assumption that there are distinct syntactic case systems including nominative-accusative and absolutiveergative can only be maintained within a semantic-pragmatic approach to cases, but not within an autonomous morphosyntactic approach. Let us illustrate the empirical disadv antage mentioned in (13a) with predicate agreement in Av ar. The agreement category is nominal class (Cn in the glosses), which includes information about the gender and number of the agreement trigger. See (14): Avar (Charachidzé 1981, 35, 144 f, 156, 166) y-ač’un y-igo (14a) ebel mother(NOM,C2) C2-come C2-is ‘mother has come’

1082

XVII. Syntaktische Typologie II: Ansätze und Übersichts-darstellungen

(14b) aka b-ač’un b-ugo cow(NOM,C3) C3-come C3-is ‘the/a cow has come’ (14c) y-osana yas di-cca C2-took girl(NOM,C2) I-ERG,C1 ‘I (male) took the/a girl’ di-ye (14d) yas y-oλula girl(NOM,C2) I-DAT,C1 C2-love ‘I (male) love a/the girl’ b-ugo ču (14e) di-r I-GEN,C1 C3-be horse(NOM,C3) ‘I (male) have a horse’ du-ye (14 f) di-cca I-ERG,C1 you-DAT,C1 ču λole-b b-ukana horse(NOM,C3) given-C3 C3-was ‘I (male) had given you (male) a/the horse’ (14g) qartay aka sorceress(ABS,C2) cow(ABS,C3) b-ečč’ule-y y-igo C3-treat(PPRS)-C2 C2-is ‘the sorceress treats the cow’ Kibrik’s analysis of v erb agreement in Av ar (1985, 318) refers to the patient-like semantic macrorole Factitiv e, which resembles the macrorole Undergoer of Role and Reference Grammar (cf. Foley/Van Valin 1984). Such descriptions do not capture the facts appropriately. The greatest problems crop up with intransitiv e v erbs since they select a wide range of thematic roles including agents, as in (14a). Kibrik’s treatment is also problematic for multi-v alenced v erbs. The semantic interpretation of the Factitiv e has to be wide enough to include an absolutiv e stimulus and an absolutiv e possessed as in (14d) and (14e), but at the same time it must exclude the recipient in (14 f). A thematic approach also fails to capture the double agreement pattern triggered by the double absolutiv e construction in (14g). Keenan (1976, 316) treats v erb agreement as a property of grammatical relations such as subject and object and cites Av ar, Hindi and Kala Lagaw Ya (= Mabuiag) as one of the languages exhibiting v erb agreement with objects and not with subjects. The object dominance of the predicate agreement in Av ar is refuted by the fact that the arguments of intransitiv e v erbs are agreement triggers although they are nev er treated as objects ev en in the approaches using the traditional relational terms. An additional problem is that Keenan’s analysis suggests a v iolation of a univ ersal restriction of v erb agreement: sub-

jects may agree with the v erb to the exclusion of objects, but objects nev er agree with the v erb to the exclusion of subjects. This univ ersal is v ery reliable as a statistical generalization for accusativ e languages and has only one systematic source of exceptions: ergativ e languages. Some of the empirical disadv antages are av oided within approaches using conglomerate terms such as S, A and O or P. Predicate agreement in Av ar is assumed to be restricted to S and O (or P). Such approaches are unnecessarily complicated, since they inv olv e two pairs of terms (i. e. subject and object or patient as well as transitiv e and intransitiv e). The description inv olv ing the two terms S and O masks the unitary behav iour of agreeing arguments and shows that the common formal expression of these terms in an ergativ e language, the absolutiv e, was neglected in fav our of the different thematic functions they preferably code. All three types of approaches suggest that ergativ e languages with this type of agreement either v iolate the abov e mentioned univ ersal for v erb agreement or fall outside the scope of this universal. A description of predicate agreement in Av ar in terms of the absolutiv e as well as the formulation of univ ersal restrictions in terms of a generalized hierarchy notion that allows reference to case hierarchies (cf. section 1.3. below) av oids the abov e mentioned problems. The examples in (14) clearly show that predicate agreement in Av ar is triggered by any absolutiv e syntactic argument of the predicate to the exclusion of other case arguments. An analysis of v erb agreement in Av ar in terms of ‘cas zéro’ is offered by Charachidzé (1981), who adequately classifies a zero-marked argument as ‘le prime actant’ and an ergativ e argument as ‘le second actant’. What is missing in Charachidzé’s approach is a principled account of the univ ersal parallel properties of absolutiv e and nominative arguments. Let us turn now to the second disadv antage of past approaches mentioned in (13b). The asymmetry in the split behav iour of ergativ e and accusativ e languages cannot be explained under the assumption that ergativ e languages hav e their own thematic hierarchy and their own syntactic case system. Patient orientation is a rather unmarked option for case assignment, as shown by Nichols’ sample (1992, Appendix 2), in which 74 (= 39%) of the languages with ov ert inflection on

57.  Relational Typology

nouns hav e an ergativ e case pattern. It is therefore puzzling that it shows up so rarely in the syntax of ergativ e languages. This observ ation is also corroborated by the fact that there are apparently no morphological accusativ e languages with a syntactic patient dominance (cf. the abov e discussion of the alleged ergativ ity of some intransitiv e v erbs in accusativ e languages, as illustrated in (9)— (11)). The assumption of inconsistent thematic ranking (see (13c)) is also v ery problematic, particularly in v iew of the syntactic split behav iour of (nearly) all languages with an ergativ e mapping of thematic roles to cases. For these languages some phenomena are assumed to work along an agent dominance, while other phenomena are treated as patient dominated. This is v ery unsatisfactory under the standard plausible assumption that agent and patient are universal semantic notions. The semantic-pragmatic approach to cases mentioned in (13d) is not the only approach that neglects the relev ance of syntactic cases in univ ersal grammar. Within Generativ e Grammar the case of a syntactic argument is fully determined by its structural relation (cf. Chomsky 1981). One major aim of the next section is to show that both accusativ e and ergativ e grammatical rules are v ery often case determined. An autonomous morphosyntactic approach to cases achiev es by far better data predictions for both ergativ e and accusativ e languages. Furthermore, as soon as ergativ e grammatical rules are formulated in terms of cases, they turn out to be identical to accusativ e grammatical rules. This leads to the conclusion that no grammatical rule, whether syntactic, semantic or pragmatic in nature, has to include an ergativ e-accusativ e parameter. 1.3. Ergative Rules as Case-Determined Rules: A Generalized Hierarchy Approach 1.3.1. The Case Hierarchy An approach to ergativ e and activ e languages that av oids the problems mentioned in (13) is presented in Primus (1994) and is based on prev ious work on accusativ e languages (cf. Primus 1987, 1993). It is called Generalized Hierarchy Grammar and is distinguished from work within both functionalist and generativ e approaches to univ ersal grammar in so far as it does not try to reduce its descrip-

1083

tiv e inv entory to a single type of relational concept. The relational concepts at issue are thematic relations such as agent and patient, structural relations as conceiv ed in recent Generativ e Grammar, and case relations such as nominativ e argument or accusativ e argument of a predicate. Pragmatic relations such as topic and predication will also be briefly mentioned. Most of the approaches to univ ersal grammar hav e tried to limit their descriptiv e inv entory to just one of the relational systems mentioned abov e. The other relational systems hav e been neglected, or considered to be deriv ed from the chosen set of relations. As is well known, Functional Grammars (cf. Dik 1978, 1989, Giv ón 1984, Foley/Van Valin 1984) concentrate on semantic and pragmatic relations and disregard structural relations and case relations. Generativ e Grammars, by contrast, formulate rules of grammar solely in terms of structural relations. This relational reductionism is most clearly manifested in the use of the traditional grammatical relations such as subject and object, and therefore also characterizes traditional descriptiv e approaches to v arious languages. No matter how subject and object are defined, each definition ov ergeneralizes a correlation between two relational concepts belonging to different systems (e. g. agent-topic, nominati v e-topic, nominati v e-agent, nominativ e-least embedded argument of a sentence, etc.). This reductionsism is v ery problematic ev en for accusativ e languages such as German or Rumanian (cf. Primus 1987, 1993). For ergativ e or activ e languages it is untenable, because it ev en fails to capture the grammar of basic sentences in these languages. The descriptiv e and explanativ e adv antage of Generalized Hierarchy Grammar is not achiev ed by reducing the descriptiv e inv entory, but rather by organizing the relational concepts on hierarchies in such a way as to permit adequate and simple rule formulations. The first relational concepts to be presented here are the step-children of both functionalist and generativ ist approaches to cross-linguistic phenomena: morphosyntactic cases. Cases are, in general, not a random set of formal categories with little relev ance for cross-linguistic research. Cases form hierarchically organized systems and the univ ersal Case Hierarchy proposed within Generalized Hierarchy Grammar is given in (15):

1084

XVII. Syntaktische Typologie II: Ansätze und Übersichts-darstellungen

nominative/absolutive argument P&Q (1 c) -P&-Q > P&Q The uniformitarianist principle rules out (1 a), but p ermits (1 b) and (1c). Sequence (1 a) would involve p ostulating an intermediate language typ e that goes against all the current synchronic evidence, and for which there is no indep endent sup p ort. I. e. the acquisition of P can occur either after that of Q, or simultaneously with it, but not before it. And if P is acquired, this guarantees either the p rior p resence, or the simultaneous acquisition, of Q. This method is used in Hawkins (1983) to make p redictions for word order changes in certain families of Indo-Europ ean. In fact, given that word order changes are gradual and p roceed via “ doubling stages” (e. g. the shift from Adj + N to N + Adj does not hap p en overnight, but involves an intermediate stage in which both orders occur, as in the modern Romance languages, with p ossibly different grammatical and/or p ragmatic functions associated with each), Hawkins formulates more fine-grained p redictions in the same vein, such as the Doubling Acquisition Hyp othesis: “ Given a set of synchronic universal imp lications of the form ‘if P then Q’, where P and Q are basic word orders of certain sp ecified typ es; then, at two successive stages in the growth of a language, IF P is acquired as a doubling structure from the earlier uniquely -P stage; THEN EITHER Q

XIX. Die Erforschung des Syntaxwandels II: Ansätze

must already be p resent at the earlier stage (whether as a doublet with -Q or not), OR, if it is not p resent, Q must be acquired as a doubling structure simultaneously with P. But P will not be acquired in the total absence of Q.” Imagine a language with SOV & Postp ositions at some stage in its history, in relation to Greenberg’s (1966) universal “ if VSO, then Prep ositions”. If either VSO or p rep ositions were to develop in this language at a later stage, initially as doublets with SOV and p ostp ositions, the Doubling Acquisition Hyp othesis p redicts that p rep ositions will be acquired first, or else VSO and p rep ositions will arise simultaneously. The acquisition of VSO will not p recede that of p rep ositions since the co-occurrence of VSO & p ostp ositions is not sanctioned by the universal. Similarly, within the context of Keenan/Comrie’s Accessibility Hierarchy, we exp ect languages to acquire or lose relativization p ossibilities in accordance with the synchronic co-occurrence facts, and to avoid a *P&-Q combination of a p rimary relativization strategy (i. e. one which op erates on subjects) which ap p lies to oblique NPs but not to indirect objects or direct objects. Similar p redictions can be made for language acquisition, as illustrated in Hawkins (1987). Consider the following cross-language universal in the area of conditional constructions: “ If a language has a construction ex p ressing p rototy p ical conditional meaning (cf. Comrie 1986) by means of exp licit conditional markers (as in if S1 then S2), then that language will also exp ress p rototyp ical conditional meaning by means of conjunctions of S1 & S2 or juxtap ositions of S1, S2 (comp are if you wake up late, then you’ll miss the bus with wake up late and you’ll miss the bus or wake up late, you’ll miss the bus).” This imp licational statement enables us to formulate the following order-ofacquisition p rediction for child language and second language acquisition: Conditionality will be exp ressed first either via S1 & S2 or S1, S2 constructions, or it will be exp ressed by these conjunctions or juxtap ositions at the same time as it is exp ressed by exp licit (ifthen) conditional constructions, but exp licit conditional constructions will not be acquired before conditionality is exp ressed by conjunctions or juxtap ositions. Data from Bowerman (1986), cited in Hawkins (1987), provides supporting evidence.

63.  Typology-based Research into Syntactic Change

Imp licational universals have also been widely used in historical reconstruction, i. e. in reconstructing the co-occurring p rop erties of some p roto-language for a given family, or in filling in missing details of some p artially attested language. For examp le, the runic inscrip tions of Late Common Germanic (AD 200—600, cf. J. Smith 1971) p rovide no examp les of relative clauses, yet other word orders in the noun p hrase and the p rep ositional p hrase p ermit us to infer that relative clauses were p ostp osed to the right of the head rather than p renominal, cf. Hawkins (1983, 263—4). The reconstruction of a whole p roto-language, however, may be more p roblematic. Friedrich (1975), W. Lehmann (1974) and Hawkins (1983) reach rather different conclusions for the reconstruction of ProtoIndo-Europ ean word order using the typ ological method. It is imp ortant to ap p reciate why: the synchronic universals up on which each relies are p artially different; and the weighting given to the various reconstruction criteria (such as the age of the resp ective daughters, geograp hical location and contact borrowing, etc) also differs. But the logic of actually reconstructing some p roto-language whose comp onent p rop erties are in accordance with imp licational co-occurrence regularities, and which is maximally comp atible with attested daughter language variants, is quite valid (contra Watkins 1976). As we imp rove our understanding of the synchronic facts, more reliable reconstructions will become p ossible. And the resp ective weighting to be attached to the different reconstruction criteria will, in any case, have to be resolved for p rop erties that do not figure in imp licational universals, just as it must be for those that do (e. g. is some earlier recorded word order in a given subfamily necessarily closer to that of the Ursp rache, or can it be the result of contact borrowing?). A typ ological ap p roach to reconstruction is therefore imp ortant, and the existence of different arguments for p roblematic cases such as ProtoIndo-Europ ean word order is no more damning to the enterp rise than is the existence of different generative analyses for modern English syntax. At any one stage of investigation, the conclusions to be drawn from our reconstruction logic will only be as reliable as the synchronic universals up on which they are based, and will need to be fine-tuned with the evidence of other reconstruction criteria and language-particular data. These “ if P then Q” universals, when ap -

1179

p lied to historical change, language acquisition and historical reconstruction, amount, as we have seen, to a simp le p rediction: a language without P or Q may acquire Q first, or both together, but not P first. This is what current synchronic facts require. One does find in the literature, however, two other interp retations of these imp licational statements in a diachronic context, both of them erroneous in my view. First, some scholars have argued that the p rior acquisition of the antecedent P p rop erty may serve as a “ trigger” setting in motion a chain of subsequent changes as the language acquires Q in accordance with the imp licational universal. I have argued against this “ trigger-chain” logic in Hawkins (1983, 234—6). Notice that there is no evidence for a *P&-Q co-occurrence ever arising diachronically whenever “ if P then Q” is synchronically excep tionless. And if the imp lication is statistical only, this theory of change immediately encounters p roblems of logical contradiction. Since “ if P then Q” is logically equivalent to “ if -Q then -P”, the evolution of *P&-Q should never arise in the first p lace. Or at least, one cannot ignore the p ressure in favor of imp licationally dep endent co-occurrences one minute, and then invoke it the next when it is convenient in order to “ explain” a given change. Second, there is an imp licit claim in much of the literature on language acquisition stemming from Jakobson (1968), that the consequent p rop erty, Q, in these imp licational dep endencies will necessarily be acquired before P, so that for examp le a subject relativization must be acquired before an object relativization in accordance with Keenan/Comrie’s Accessibility Hierarchy. I have argued against this interp retation in Hawkins (1987). Although the data may frequently sup p ort this view, sometimes they do not, and both antecedent and consequent are acquired simultaneously, as one would p redict given the synchronic facts. Hence, in language acquisition as in historical change, the consequent may be acquired first, or both may be acquired simultaneously, and all that we can rule out is the acquisition of the antecedent p rior to the consequent, since there are no languages of this typ e currently attested. 2.2. What Causes Change? We have concentrated so far on the p redictions made by imp licational universals for p ossible and imp ossible p athways of change.

1180

In essence, we are simp ly claiming that successive diachronic states of individual languages will conform to the descrip tive and exp lanatory p rincip les of variation that have been derived from the synchronic comp arison of many languages. But why should languages change from one state to another? Conformity with indep endently established synchronic laws is not, in itself, necessarily a reason for change. The typ ological ap p roach does contribute insights into this question. These insights need to be p laced within a fuller context of the general causes of change, however. It is useful to distinguish the following typ es of causes, only the last three of which are elucidated by a typological approach. 2.2.1. Language-External Causes of Change N. Smith (1981) has summarized the crosslinguistic evidence for contact borrowing as a major determinant of change in the area of word order. Weinreich (1953) is an early study of contact borrowings in other areas of grammar, while Masica (1976) draws attention to the Sp rachbund or areal convergence phenomenon in different parts of the globe. 2.2.2. Grammatical Reorganization The kind of work illustrated by Lightfoot (1979) is an attemp t to motivate syntactic change in terms of generative p rincip les, coup led with a general evaluation metric on evolving grammars that motivates the elimination of derivational comp lexity. Certain reanalyses, or changes, are thereby claimed to be exp lained as a consequence of others, in accordance with these p rincip les. Much research on the role of “ analogy” in syntactic change can also be included in this category, since it ap p eals ultimately to a grammatical basis for reorganization in resp onse to smaller successive changes, e. g. Stockwell (1977). 2.2.3. Preference-Based Explanations for Change Typ ological comp arison reveals clear p references for certain construction typ es or language typ es over others. E. g. there is an interesting frequency or p reference gradient in VO languages with resp ect to the p osition of the subject: SVOX > VSOX > VOSX > VOXS (cf. Hawkins 1983, 157); certain crosscategorial word order combinations are more

XIX. Die Erforschung des Syntaxwandels II: Ansätze

frequent across languages than others; and so on. Now, these p references can be exp loited in two ways in a diachronic context. First, we exp ect that there will be a corresp ondence between synchronic language quantities and diachronic time amounts, all things being equal. I. e. the less frequent a p rop erty or p rop erty cluster synchronically, the less its survival time in the history of individual languages and in the collective history of all languages. This corresp ondence is referred to as the Relative Time Hyp othesis in Hawkins (1983, 256). What this means, secondly, is that infrequent typ ological p henomena will be diachronically unstable, and when an infrequent p henomenon changes into its more frequent counterp art, we can ap p eal to the synchronic exp lanation for the p reference to actually motivate and exp lain the change. There are imp ortant issues raised in this context about the p recise typ e of causality to which we are ap p ealing. Vennemann (1983) contains an interesting discussion of this matter in the area of p honology. But I agree with him in wishing to assert that the motive for the synchronic p reference is also the motive for the historical change in such cases. (Throughout this discussion I am, of course, assuming that the synchronic p references in question are theoretically motivated and not accidental, as discussed above.) In the event that a language changes its basic word order in resp onse to the p rocessing p references documented in Hawkins (1990), I, like Vennemann, would wish to assert that the p rocessing considerations exp lain the change. And in the event that a language develop s, in the marked case, a more comp lex word order for p rocessing, I would have to ap p eal to indep endent counteracting causes (such as the language-external ones in 2.2.1.) and I would regard the synchronic p references as motivating a subsequent change to a more p referred order for p rocessing. I. e. languages tolerate disp referred states for certain p eriods only, and the reasons for the disp reference constitute an omnip resent motive for change that will be resisted only when there are good, independent reasons for doing so. 2.2.4. Implicationally-Based Explanations for Change I mentioned above that imp licational universals constrain the p ermissible p athways of change from one language state to another. In certain cases, however, they also p rovide answers to the why of change. Consider Greenberg’s (1966) (statistical) universal (41)

63.  Typology-based Research into Syntactic Change

“ if in a language the verb follows both the nominal subject and the nominal object as the dominant order, the language almost always has a case system”. This statement allows for three p roductive language typ es, and excludes one, as follows: (2a) OV & +Case (P&Q) (2b) VO & +Case (-P&Q) (2c) VO & —Case (-P&-Q) (2d) *OV & —Case (P&-Q) Imagine that indep endent p honological p rocesses are eroding the case system of an OV language (as hap p ened p roductively in earlier Romance and Germanic). This universal then p rovides an exp lanation for the concomitant shift of SOV to SVO, since “ if OV then +Case” is logically equivalent to “ if -Case then VO”. More generally, whenever there is a logical dep endency between p rop erties of this typ e, and some indep endent exp lanation for a change in the logically antecedent p rop erty, a change in the consequent p rop erty is both p redicted and exp lained by the synchronic dep endency and whatever exp lanation underlies it. (Notice in this p articular examp le that the existence of VO & +Case languages, such as Lithuanian, means that the loss of the case system is not a necessary cause of the OV to VO shift, since OV can shift to VO whether or not the case system is eroded, but it is certainly a sufficient cause.) 2.2.5. Diachronic Universals There have been a number of interesting attemp ts to define and exp lain various diachronic universals, i. e. regular diachronic drifts whereby some p rop erty X changes into another, Y, rather than vice versa. Within p honology, for examp le, only certain sounds can merge into others: a voiceless consonant may become voiced between vowels, but not vice versa (cf. Foley 1977 for many such examp les). In morp hology, a bound morp heme derives historically from a free lexical or grammatical morp heme and generally inherits the same order relative to its stem that was assigned to the earlier free morp hemes by the syntax — the converse shift (bound to free morp heme) does not occur (cf. Hall 1988). Morp hological devices marking syntactic dep endents may gravitate diachronically towards their syntactic heads of p hrases, but not vice versa (cf. Nichols 1986). And within semantics it has been p rop osed that root modals may acquire ep istemic senses later in time, but not vice versa (cf. Sweetser 1984),

1181

and that sp eech act verbs will emerge diachronically from descrip tive verbs that do not p erform sp eech acts, whereas sp eech act meanings do not develop into descrip tive ones (cf. Traugott 1986). The causes of these drifts are various and constitute p art of the theory of language change. Their consequences for typ ology will be p ursued in the next subsection. 2.3. The Diachronic Dimension in Typology Diachronic universals of change can make p redictions for (and contribute to the exp lanation of) synchronic variation p atterns as the typ e X p rop erties discussed above merge into typ e Y. Co-occurrences of X and Y are p redicted in the relevant p honological, morp hological and syntactic environments, and in co-occurring word meanings, whereas cooccurrences of, for examp le, typ e Z and Y are not. Notice the logic of this kind of diachronic exp lanation for synchrony. At any one historical stage in a language, X may occur unshifted, or in combination with Y, or else Y may occur alone up on comp letion of the shift. What is crucially imp ortant, therefore, is the identity of X and Y. The frequent co-occurrence of these p rop erties is p redicted, as is some p ossible vestige of features of X in Y, even when X is no longer p resent as such synchronically, as in the bound morphology example. The diachronic dimension in typ ological exp lanation raises some p rofound issues, however. Consider an instructive examp le taken from Mallinson/Blake (1981, ch. 6). These authors summarize a number of attemp ts that have been made to set up diachronic links between the word order p rop erties that were first correlated in the imp licational universals of Greenberg (1966). For examp le, N+Gen and Pr+NP are highly correlated, as are Gen+N and NP+Po. Mallison/ Blake observe that a head noun before genitive construction (the bac k of the hut) may shift diachronically into a p re p ositional p hrase (bac k of the hut) in a number of languages, whereas a genitive before head would shift into a p ostp ositional p hrase. Such considerations are then invoked to exp lain the implicational universals themselves. But there are p roblems with this line of reasoning. While diachronic links can reasonably be established between N+Gen and Pr+NP, etc, the number of word orders for which such links can be established is far less than the total number of word order corre-

XIX. Die Erforschung des Syntaxwandels II: Ansätze

1182

lations that exist synchronically. Diachronic links of this typ e are only p lausible for p airs of word orders that are sufficiently closely related semantically so that the one structure can merge gradually into the other, which is not many. Nor is it terribly convincing even with p rep ositions and genitives. The simp le p rep ositions of English are not derived from genitives, and for most languages of the world we either have no record of earlier synchronic states or no comp elling arguments for reconstructing a genitive source for adp ositions in many cases at least. Thus, the diachronic link p rop osed here is not a diachronic universal and does not have sufficient generality to p rovide an exp lanation for a synchronic universal. This is not to deny that diachronic changes of these typ es occurred. They did, in certain languages. But this is no exp lanation for a universal. And where there are no such diachronic universals, successive language states will be constrained only by the wellformedness requirements of synchronic universals. Nor will there be a diachronic dimension to exp lanation if there are certain diachronic tendencies (such as the genitive-adp osition shift) which cover only a p rop er subset of the range of synchronic data to be exp lained. Where these p roblems can be avoided, however, then diachronic universals do p rovide exp lanations for p atterns of synchronic typ ological variation, as in the examples of 2.2.5.

3.

Finale

What I have tried to do in this brief p ap er is p rovide a critical overview of typ ology-based research into syntactic change. I began by asking (section 1.) what typ ology actually is, and argued that it is a research methodology whose goals are to describe and exp lain the variant construction typ es, language typ es, and their relative frequencies, that are currently found across the globe. I then considered some p redictions for diachrony made by imp licational universals (section 2.1.). The role of typ ological considerations in exp laining what causes change was then outlined (section 2.2.). And finally I considered some attemp ts to exp lain typ ological variation in terms of diachronic universals or drifts (section 2.3.).

4.

References

Bell, Alan. 1978. Language samp les. Universals of human language. Vol. 1, Method and theory, ed. by Josep h Greenberg, Charles Ferguson & Edith

Moravcsik. Stanford. Bowerman, Melissa. 1986. First step s in acquiring conditionals. In Elizabeth Traugott et al. (eds.), 1986. Chomsky, Noam. 1965. Asp ects of the theory of syntax. Cambridge, MA. —. 1981. Lectures on government and binding. Dordrecht. Comrie, Bernard. 1986. Conditionals: a typ ology. In Elizabeth Traugott et al. (eds.), 1986. —. 1989. Language universals and linguistic typ ology, 2nd edn. Oxford. Foley, James. 1977. Foundations of theoretical phonology. Cambridge. Frazier, Lyn. 1979. Parsing and constraints on word order. University of Massachusetts occasional p ap ers in linguistics, Vol. 5, ed. by Jean Lowenstamm, 177—198. Friedric h, Paul. 1975. Proto-Indo-Europ ean syntax. Journal of Indo-Europ ean Studies, Monograph 1, Butte. Greenberg, Joseph. 1966. Some universals of grammar with p articular reference to the order of meaningful elements. Universals of language, ed. by Joseph Greenberg, 2nd edn. Cambridge, MA. Hall, Christopher. 1988. Integrating diachronic and p rocessing p rincip les in exp laining the suffixing preference. In John Hawkins (ed.), 1988b. Hawkins, John. 1983. Word order universals. New York. —. 1987. Imp licational universals as p redictors of language acquisition. Linguistics 25. 453—473. —. 1988a. On generative and typ ological ap p roaches to universal grammar. In John Hawkins & Heather Holmback (eds.), 1988. —. (ed.) 1988b. Exp laining language universals. Oxford. —. 1990. A p arsing theory of word order universals. Linguistic Inquiry 21. 223—261. —. and Heather Holmbac k. (eds.) 1988. Pap ers in universal grammar: generative and typ ological ap proaches. Lingua Special Issue 74. 2/3. Jac kendoff, Ray. 1977. X-bar syntax: a study of phrase structure. Cambridge, MA. Jakobson, Roman. 1968. Child language, ap hasia and phonological universals. The Hague. Keenan, Edward. 1979. On surface form and logical form. Studies in the Linguistic Sciences Sp ecial Issue 8.2. 1—41. [ap p ears also in Edward Keenan, Universal grammar: 15 essays, 1987, London]. —. and Bernard Comrie. 1977. Noun p hrase accessibility and universal grammar. Linguistic Inquiry 8. 63—99. Lass, Roger. 1980. On exp laining language change. Cambridge. Lehmann, Christian. 1984. Der Relativsatz. Tübingen.

63.  Typology-based Research into Syntactic Change

Lehmann, Winfred. 1974. Proto-Indo-Euro p ean syntax. Austin. Lightfoot, David. 1979. Princip les of diachronic syntax. Cambridge. Mallinson, Graham, and Barry Blake. 1981. Language typ ology: cross-linguistic studies in syntax. Amsterdam. Masic a, Colin. 1976. On defining a linguistic area. Chicago. Nic hols, Johanna. 1986. Head-marking and dep endent-marking grammar. Language 62.1. 56—119. Pulleyblank, Douglas, and Akinbiyi Akinlabi. 1988. Phrasal morp hology in Yoruba. In John Hawkins & Heather Holmback (eds.), 1988. Pullum, Geoffrey. 1981. Languages with object before subject: a comment and a catalogue. Linguistics 19. 147—155. Skalička, Vladimir. 1979. Typ ologische Studien: mit einem Beitrag von Petr Sgall, ed. by Peter Hartmann. Braunschweig. Smith, Jesse. 1971. Word order in the older Germanic dialects. Urbana-Champ aign: University of Illinois dissertation. Smith, Neil. 1981. Consistency, markedness and language change: on the notion ‘consistent language’. Journal of Linguistics 17.1. 39—54. Stoc kwell, Robert. 1977. Motivations for exbraciation in Old English. Mechanisms of syntactic change, ed. by Charles Li. Austin.

1183

Sweetser, Eve. 1984. Semantic structure and semantic change: a cognitive linguistic study of modality, p ercep tion, ps eech acts, and logical relations. Berkeley: University of California dissertation. Thompson, Sandra. 1988. A discourse ap p roach to the cross-linguistic category ‘adjective’. In John Hawkins (ed.), 1988b. Traugott, Elizabeth. 1986. From p olysemy to internal semantic reconstruction. Proceedings of the Twelfth Annual Meeting of the Berkeley Linguistics Society; Pap ers from the Parasession on semantic typology, 539—550. —. Alic e ter Meulen; Judith Reilly; and Charles Ferguson. (eds.) 1986. On conditionals. Cambridge. Vennemann, Theo. 1983. Causality in language change: theories of linguistic p references as a basis for linguistic exp lanations. Folia Linguistica Historica 4.1. 5—26. —. 1984. Typ ology, universals and change of language. Historical syntax, ed. by Jacek Fisiak, 593— 612. Berlin. Watkins, Calvert. 1976. Toward Proto-Indo-Europ ean syntax: p roblems and p seudop roblems. Diachronic syntax, ed. by S. Steever, C. Walker & S. Mufwene. Chicago. Weinreic h, Uriel. 1953. Languages in contact: findings and problems. The Hague.

John A. Hawkins, Los Angeles, University of Southern California (USA)

1184

XX. Die Erforschung des Syntaxwandels III: Phänomene Research into Syntactic Change III: Phenomena

64. Ergativity 1. 2. 3. 4. 5.

1.

Changeable Alignment Away from Ergative Alignment Towards Ergative Alignment Reasons for Realignments References

Changeable Alignment

1.1.  In terms of the valency of verbs (or, more generally, predicates) and the semantic roles associated with them we may distinguish four major kinds of core arguments: active and inactive arguments of intransitives (as in The soldiers marched and The ancient oak fell), and similarly, although with the two kinds of arguments co-occurring in one clause, active and inactive arguments of transitives (as in The ducks ate the old bread) — or ACTi, INACTi, ACTt, and INACTt for short. While perhaps not equally appropriate for all semantic classes of predicates, the characterizations of arguments as denoting active participants and participants acted upon, or performers and undergoers, should enable us to refer summarily to the most common relational oppositions. When these kinds of arguments are involved in grammatical rules and regularities, they may group in several different ways — if indeed they group at all, for one possibility is that each one behaves differently from every other. ACTi may be aligned with ACTt and INACTi with INACTt, with relational semantic similarities thus taking priority over valency distinctions, yielding a pattern known as ‘active (-inactive)’, shown schematically in (1). Or ACTi and INACTi may fail to be distinguished from one another, in spite of their semantic difference, with both of them then aligning either with INACTt or with ACTt, either choice partly motivated and partly arbitrary on semantic grounds, yielding respectively an ‘ergative(-absolutive)’

or a ‘(nominative-)accusative’ pattern, as seen in (2) and (3). ACTt and INACTt may likewise fail to be distinguished from one another, despite their semantic difference and their co-occurrence in a single clause, and then behave like or unlike the uniform intransitive argument, yielding patterns, shown in (4) and (5), that have been called ‘neutral’ and ‘double-oblique’ respectively. In the ‘tripartite’ pattern shown in (6 ), ACTi joins INACTi, but ACTt and INACTt behave differently from one another as well as from this uniform intransitive argument. (See Kibrik 1979 for an even fuller catalogue of patterns of alignment.) It will be convenient to use ‘active’, ‘inactive’, ‘ergative’, ‘absolutive’, ‘nominative’, and ‘accusative’ also as general terms for those relations which behave uniformly in the patterns known by these names — viz. for ACTi/ACTt and INACTi/INACTt in (1), ACTt and ACTi/INACTi/INACTt in (2), ACTi/INACTi/ACTt and INACTt in (3), respectively. Such alignments hold for individual rules and regularities or, more specifically still, for particular domains of their application (as defined, for example, by tense or aspect or nominal reference distinctions). Despite the once common typological practice of labelling entire languages ‘ergative’, ‘accusative’ etc., language-particular rules and regularities need not be consistent in this respect. While one or the other alignment pattern may predominate within a language, and to some extent thus justify global type assignments, it is a safe bet that, strictly speaking, all languages are ‘split’ insofar as their rules and regularities alternatively follow two or more patterns of alignment rather than only a single one. Obviously, then, if we want to study how alignments change in time, it will be those manifesting themselves in particular rules and regularities, or even particular domains of rule applications, that need to be

64.  Ergativity

taken into account primarily. It can only be determined secondarily whether any such particular changes occur in unison, thus supporting the idea of languages converting from one to the other alignment type in their entirety. 1.2.  There are rules and regularities which appear to operate on essentially the same alignment patterns in all languages where they are attested. The formation of imperatives has often been mentioned, not quite accurately, as an example of invariably nominative-accusative alignment, making reference to intransitive arguments and ACTt to the exclusion of INACTt. Normally imperatives consist of predicates denoting kinds of behaviours that may be initiated, or not initiated, at will and are addressed to beings considered capable of choosing to comply, or not to comply, with the directive issued; their addressees, implicit (Shut up!) or explicit (Don’t you worry! Somebody open the door please!), will therefore normally be arguments, of transitives as well as of intransitives, of the ACT but not of the INACT kind, resulting in an active, rather than accusative, alignment. There are instances, especially in passive, i. e. non-basic constructions, where the addressees of imperatives might seem to be INACT’s; but this need not argue against an active and for an

1185

accusative (or perhaps neutral) patterning of genuine imperative formation as long as such clauses do not really have the force of directives (cf., in English, Be warned!), hence may be disregarded, or the seemingly inactive participant really is the one considered primarily responsible for initiating an action, hence is more of an ACT than of an INACT argument after all (cf. Get dressed everybody!). An example of a derivational or lexical regularity whose patterning likewise seems cross-linguistically invariable has to do with causatives; what its pattern is, however, depends on how one is looking at it. If we consider the range of arguments of non-causatives which may appear as (not necessarily overtly expressed) causees of corresponding causatives, it typically includes ACTi, INACTi, and ACTt, but not INACTt (cf. The soldiers marched — The sergeant marched the soldiers; The ancient oak fell — They felled the ancient oak; The ducks eat old bread — They feed the ducks with/on old bread or They feed old bread to the ducks) — which is a nominative-accusative grouping. On the other hand, if we consider which arguments of non-causatives correspond to which arguments of causatives, we need to relate non-causative ACTi, INACTi, and ACTt to causative INACTt (and perhaps also to causative indirect or oblique objects, if the non-causative is already transi-

XX. Die Erforschung des Syntaxwandels III: Phänomene

1186

tive), to the exclusion of causative ACTt, the argument reserved for the causer — and these relation-ships define a different pattern, almost looking ergative, were it not for the relatability of non-causative ACTt to causative ACTt in pairs such as eat — feed. For present purposes we need not adduce further examples of rules and regularities arguably showing cross-linguistically invariable alignment (among these, ergative and also active ones are no less common than accusative ones); nor need we go into the whys and wherefores of such universal preferences (presumably explicable in terms of natural semantic or pragmatic classes of argument relations). The point here is that we must reckon with rules and regularities whose alignments, on the evidence of their cross-linguistic invariance, are resistant to change. The rules which indeed are changeable in this respect typically come from the areas of relational encoding, especially by means of case (or adpositional) marking and agreement, and of clause combination, pertaining in particular to the ellipsis of shared arguments. 1.3.  There are three general questions which any systematic account of changes of alignment will need to raise: Which patterns may develop from and into which others? How are the transitions effectuated? What are the reasons for such developments? The focus is here primarily on the ergative pattern, and there is an extra difficulty facing its historians. In many languages and language families where there is reason to believe that relevant changes have occurred, their documentation (in written records) is sparse or nonexistent, and recourse must thus be had to hypothetical, and rarely uncontroversial, reconstructions. If ergative histories were to be based exclusively on hard and fast facts covering none-too-short spans of time, they would virtually be confined to only two families of typological repute, viz. Indo-Iranian and Kartvelian (where there are also controversies, thought less about the facts than about their interpretation). The following account of scenarios for the emergence and demise of ergative alignments will draw on evidence from these families, but in the interest of comprehensiveness much use will also be made of informed, if not always unanimous speculation about others whose past is less well documented. With the ergative pattern consisting in an identification of arguments across clause types, viz. ACTi/INACTi and INACTt, it

may appear as well as disappear owing to developments manifesting themselves in intransitive or in transitive clauses (or in both): the argument(s) diachronically realigned may be ACTi/INACTi on the one hand or ACTt and INACTt on the other. In the following survey, developments both away from and towards ergative alignment will thus be distinguished according to the actual locus of change. We will further distinguish three general mechanisms of change: direct realignments in basic constructions, realignments ensuing from exchanges of construction, and realignments due to revaluations of non-basic constructions as basic and vice versa. (For lack of space references to specialist literature are omitted from this survey; many of them are supplied in Plank 1985, an article covering partly similar ground, or also in Plank 1979. Other general discussions of ergativity in diachrony include Kuryłowicz 1946 ; Anderson 1977; Comrie 1978, § 3; Dixon 1979, § 4; Trask 1979; Plank (ed.) 1979, part 6 ; Dik 1980; Klimov 1983; Estival/ Myhill 1988; Harris 1990.)

2.

Away from Ergative Alignment

2.1. Intransitive Realignment 2.1.1.  A transition from ergative to accusative alignment in basic constructions comes about when a rule or regularity ceases to align the intransitive core argument (subsuming non-distinct ACTi and INACTi) with INACTt and aligns it with ACTt instead, with nothing happening in transitive clauses themselves. Patternwise this kind of development thus amounts simultaneously to an extension of the ergative, originally comprising only ACTt, and a restriction of the absolutive, originally comprising INACTt and ACTi/INACTi. In terms of the resulting pattern the extended ergative, now comprising ACTt and ACTi/INACTi, is a nominative, and the restricted absolutive, now comprising only INACTt, an accusative. If such realignments of ACTi/INACTi take place gradually rather than across-the-board, they may be expected to produce active patterns: if only some intransitive core arguments are to be realigned with ACTt, while others continue to align with INACTt, these should be ACTi’s rather than INACTi’s, finding encouragement in relational semantic similarities. Such full and partial extensions of ergatives/restrictions of

64.  Ergativity

absolutives in basic constructions are widespread, especially for rules of case marking, but also for verb agreement (and cross-reference). Here are some characteristic examples, presented in bare outline. In Kartvelian languages and dialects an originally ergative case marker was extended either to all intransitive core arguments (as in Mingrelian) or, more frequently, only to ACTi (at least according to the Ancestral Ergative Hypothesis; the rival Ancestral Active Hypothesis posits a reverse development, with an active case marker, i. e. one originally covering ACTi as well as ACTt, being restricted to ACTt, except in Mingrelian, where it was allegedly extended to INACTi instead). In Udi, of the Lezgian subgroup of Northeast Caucasian, a similar intransitive-centred development was accompanied by a change in transitive clauses, where dative replaced absolutive as a marker, patternwise now accusative, of INACTt. In various Pamir languages, as a late episode in the story of the rise and fall of ergativity in Iranian, the oblique case of ACTt was extended to the core arguments of a small group of intransitive verbs, including ‘weep’, ‘laugh’, and ‘cough’ — i. e. to some ACTi’s but to no INACTi’s. Indo-European would also have to be included here on the not uncontroversial assumption that a case marker *-s, historically encoding ACTi/ INACTi as well as ACTt (though not with nouns of all classes), and thus qualifying as a nominative in an essentially accusative-type pattern, had earlier been limited to ACTt or ACTt plus ACTi in an ergative or active pattern. A similar extension of an ergative, or active, case, purportedly reconstructible for Proto-Afroasiatic, seems to have occurred in Akkadian and perhaps further Afroasiatic languages. In the course of the collapse of the earlier inflectional case system and its partial renewal by means of postpositions, modern Indo-Aryan languages frequently generalized the ergative (formally instrumental) case forms of personal pronouns, especially of 1st and 2nd person, to intransitive core arguments, previously marked by the absolutive case; for these pronouns, case assignment then operated, if only temporarily, on a nominative-accusative pattern, with ACTt and ACTi/INACTi being in what used to be the ergative and INACTt in what used to be the absolutive case forms. In Wappo, apparently of Yukian affiliation, an (extrapolated) ergative case marker encroached on intransitive terrain, replacing the absolutive in main

1187

clauses, but not in subordinate and equational ones, and thus producing a split between traditional ergative alignment (in subordinate and equational main clauses) and innovative accusative alignment (in nonequational main clauses). In Sherpa, a Tibeto-Burman language, there is a tendency for ergative case markers, originally confined to perfective verbs in a typical aspectual split, to be extended to ACT arguments especially of verbs of experience which have been intransitivized, or at any rate detransitivized, by the incorporation of their INACTt arguments, with these extended ergatives then appearing also in the imperfective aspect. In the Polynesian language Samoan an ergative case marker of ACTt was likewise generalized to less transitive, if not strictly intransitive, clauses, replacing the absolutive on ACT arguments of ‘middle’ verbs, whose other arguments are oblique objects. In Burushaski, a language isolate of northern Pakistan, verbal suffixes which originally agreed only with ACTt in person, number, and class, were extended from transitive to intransitive verbs, with verb agreement, now triggered by ACTt and ACTi/INACTi, thus switching from ergative to nominative alignment. (Verbal prefixes, apparently more archaic, are often present as well, agreeing with ACTi/INACTi and INACTt.) For TibetoBurman, a verb agreement system has been reconstructed whose alignment was split up into a predominant absolutive and a more circumscribed nominative pattern, with ACTi/INACTi triggering essentially the same person-number markers on intransitive verbs as INACTt did on transitive verbs, except for 3rd person INACTt, where ACTt was the agreement trigger; and precisely such ACTtagreeing affixes subsequently spread to intransitive verbs in some languages, replacing the former affixes for the same personnumber triggered by ACTi/INACTi. (A further development in some languages was to drop the pronominal affixes of transitive verbs agreeing with INACTt, which further weakened the remaining absolutive ingredients of verb agreement; in many Tibeto-Burman languages pronominal verb agreement has been lost entirely.) In Lummi, a Coast Salish language with ergative and absolutive sets of pronominal markers agreeing with, or rather cross-referencing, ACTt on the one hand and INACTt and ACTi/INACTi on the other, the ergative-set markers appear to have been extended also to ACTi/INACTi in

1188

subordinate clauses, where they optionally replace absolutive-set markers, thus dividing the domains of ergative and innovated accusative alignment between main and subordinate clauses. In the Siouan language Dakota there are phonological grounds for suspecting that the cross-reference markers for ACTi (identical with those for ACTt) have become associated with intransitive verbs later than those for INACTi (identical with those for INACTt), which suggests that the active agreement pattern derives from an ergative one. If ACTi/INACTi arguments in an ergative pattern, instead of realigning with ACTt, would discontinue their alignment with transitive arguments altogether, the resulting pattern would be of the tripartite type. This, however, is not what they commonly do. Tripartite patterns are comparatively rare anyhow, being mostly found in languages where one subset of noun phrases follows an ergative, another an accusative pattern in case marking or other relational encoding, and yet another small subset, often including pronouns, combines the two patterns, behaving ergative-like when in ACTt, accusative-like when in INACTt, and absolutive/nominativelike when in ACTi/INACTi function. Wangkumara, a member of the Pama-Nyungan family of Australia, is one of the extremely few languages, if not the only one, where all noun phrases follow the tripartite pattern in their case marking, with bound forms of 3rd person singular pronouns apparently serving as case markers. Now, while perhaps never arising as the result of intransitive-centred developments away from ergative alignment, tripartite patterns may themselves be transformed into accusative ones by means of intransitive realignments, just like ergative patterns. Thus, Proto-Australian 1st and 2nd person personal pronouns have been reconstructed as having had three distinct case forms in the singular for ACTi/INACTi (non-suffixed, hence monosyllabic), ACTt, and INACTt functions (with ergative and accusative suffixes respectively), and in some languages (such as Western Desert and Dhalandji) the attested ACTi/INACTi forms can then be understood as former pure ACTt forms, with the extension motivated by the ancestrally accusative alignment of non-singular pronouns (and triggered by an innovated prohibition against monosyllabic words). In the upper dialect of Wakhi, a Pamir language, 1st

XX. Die Erforschung des Syntaxwandels III: Phänomene

and 2nd person singular pronouns are, in past tense clauses, in the oblique case in ACTt function, optionally in the accusative case in INACTt function, and in the absolute form in ACTi/INACTi function; but 1st and 2nd person singular pronouns in this last function, especially in that of ACTi, are, again only in the past tense, also acquiring oblique case marking, and thus reapproach an accusative pattern like that traditionally obtaining in the present tense. 2.1.2.  The alignment of intransitive core arguments may also be altered in a more indirect manner, as a by-product of an exchange of constructions. It is not uncommon for verbs to alternate between various constructions, basic and non-basic ones (such as derived voices) or ones with mutually exclusive domains of occurrence (such as main and subsidiary predications), and for these alternative constructions to differ in verbal valency (e. g. in derived voices valency is commonly reduced by one) as well as in the mapping of semantic relations onto syntactic ones. Nominalizations are one kind of nonbasic construction within the reach of both intransitive and transitive verbs (see sections 2.2.3. and 3.2.3. for others often reserved for transitives). Depending on the syntactic and semantic properties of nominalizations (e. g. on their transitivity, voice-neutrality, actional or resultative meaning), the alignment of arguments may differ from that of the corresponding verbal constructions — witness nominalizations in English, where ACTi/INACTi, ACTt, and INACTt are all eligible for genitive case marking and preposing, and thus form a neutral pattern (cf. his death — his discovery of them — their discovery by him), whereas an accusative case marking and constituent ordering pattern prevails with finite verbs (He died — He discovered them). A realignment will then be brought about automatically when one construction happens to be replaced, for whatever reasons, by another that used to differ from it in alignment. This is what happened in Mayan languages such as Chol and Jacaltec. In Mayan Set A person-number markers of nouns (including nominalizations) refer to ACTt and ACTi/ INACTi ‘possessors’, patterning nominatively; but when used with verbs they exclusively referred to ACTt, with Set B markers referring to INACTt and ACTi/INACTi, yielding an ergative-absolutive pattern. No-

64.  Ergativity

minalizations then replaced verbal constructions in the present (but not past) tense in Chol and in various kinds of subordinate clauses in Jacaltec. Although both transitive and intransitive verbs were thus replaced, the nominalizations corresponding to transitive verbs continued to take Set A markers to refer to ACTt; it was only with nominalizations corresponding to intransitives that cross-reference markers were exchanged in this process, with those of Set A replacing those of Set B. In effect, though not in mechanism, this resembles extensions of the ergative/restrictions of the absolutive dealt with in section 2.1.1. 2.2. Transitive Realignment 2.2.1.  When the locus of change away from ergative alignment is in basic transitive clauses, two arguments are candidates for realignment, individually or jointly, and the results are potentially fairly diverse. If INACTt ends its association with intransitive core arguments and joins its transitive partner, ACTt, an ergative pattern is transformed into a double-oblique one; the ergative is here extended syntagmatically rather than paradigmatically. If INACTt parts company with intransitive core arguments without simultaneously teaming up with ACTt, the pattern becomes tripartite. If ACTt, the only loner in the ergative pattern, joins the other core arguments in transitive and intransitive clauses, already in company with each other, the pattern ensuing from this all-encompassing extension of the absolutive is neutral. If both transitive arguments change simultaneously, with ACTt coming to behave as INACTt used to behave (and ACTi/INACTi still does) and with INACTt going its own way (perhaps that which ACTt used to go), the alignment pattern ends up accusative. There is obviously no way an ergative pattern can turn active in exclusively transitive realignments of any kind. All possible developments are attested, especially again for rules of case marking and verb agreement (or cross-reference). Individually and perhaps also collectively, they are less frequent than intransitive realignments, though. It is often a pattern in other domains within the same language which provides a model for such realignments. While nouns in ACTt function are in the ergative (identical with locative or instrumental) in other Chukotko-Kamchatkan lan-

1189

guages, they have lost this case marker in basic transitive constructions in Kamchadal, and appear in the absolutive instead, sharing this case with INACTt and ACTi/INACTi nouns, which thus pattern neutrally. The uniquely ergative forms of personal pronouns, on the other hand, are retained, but are used for emphasis, regardless of the relation of the argument to be emphasized, in what is thus effectively also a neutral pattern. In Lardil, of the Tangkic subgroup of PamaNyungan, ergative case markers have been claimed to have been eroded in the course of phonetic truncations, with ACTt thus becoming as unmarked as INACTt and ACTi/INACTi, likewise resulting in a neutral pattern. In certain varieties of Burushaski there seems to have been a tendency to drop the ergative case from nouns or 3rd person pronouns (1st and 2nd person pronouns do not have this case) in the present, future, and imperfective, with this optional neutral pattern thus complementing the ergative one in the other tenses or aspects. In the Pamir subgroup of Iranian, already mentioned above, the ergative patterns of case marking and verb agreement in past tenses are generally de-ergativized, under the influence of accusative patterns in the present tense, with somewhat different intermediate and final results in different languages and dialects. Concerning case marking, the absolutive on INACTt was commonly exchanged for an oblique case, also used for ACTt, resulting (in languages such as Rushan) in a double-oblique pattern. The oblique form of INACTt was subsequently often differentiated (e. g. in Yazgulyam, Munji, Bartang, Rushan, Upper Wakhi) from the general oblique case, especially if such arguments were definite or had specific reference, by means of additional, specifically accusative prepositions, postpositions, or suffixes, rendering the alignment pattern tripartite. ACTt, on the other hand, commonly (e. g. in Oroshor, Shugn, Ishkashim, Lower Wakhi) switched from oblique to absolutive, the case also of all INACTi’s and all or most ACTi’s, transforming doubleoblique or tripartite patterns into accusative ones, or ergative into neutral ones. The neighbouring Dardic languages too appear to have run this whole gamut of transitivecentred realignments. Modern Indo-Aryan languages have likewise tended to eliminate an ergative case marking pattern, once general in the participial tenses, by means of replacing the nominative of INACTt (patternwise an absolutive but called nominative on

1190

the strength of its patterning in non-participial tenses) by an oblique case (accusative or dative, especially if definite or also animate) and/or the ergative of ACTt by the nominative case (sometimes only for 1st and 2nd person pronouns), leading to tripartite, neutral, or accusative patterns. The ergative pattern of gender-number agreement of past participles used in periphrastic verb forms with essere/être and avere/avoir in Romance languages such as Italian and French, in the literary languages already limited in transitive clauses to INACTt preceding the past participle, is transformed into a double-oblique one colloquially and dialectally, where there is a tendency for past participles not to agree with INACTt at all, with ACTi/INACTi thus being left as the sole agreement trigger. Exactly the same pattern of (continuing) gender or gender-number agreement with ACTi/INACTi but nonagreement (i. e. no-longer-agreement) with either transitive argument is found in the Iranian dialect Danesfani and in the ShugnRushan subgroup of the Pamir languages. In Persian the intransitive past-tense agreement markers were tranferred to transitive pasttense verbs, here agreeing in accusative fashion with ACTt. Elsewhere in Iranian the ergative agreement pattern of verb forms based on past-tense stems (deriving from perfect participles) has been reorganized in different manners, often by way of introducing clitic pronouns functioning as, or cross-referencing, ACTt and ACTi/INACTi but not INACTt arguments. In the interrogative and imperative moods verb agreement in transitive clauses in Asiatic Eskimo does not follow the ergative pattern as consistently as in the indicative mood, on the basis of which they had originally been formed, which suggests that accusative alignment has been gaining ground since. Realignments in basic transitive constructions, here described in terms of essentially semantic relations and verbal valency, can often be made sense of by taking into account syntactic relations as well. On the assumption that INACTt once was the syntactic subject in basic transitive constructions (see section 3.2.3. on how this state of affairs may arise), but is in the process of being stripped of this rank, developments from ergative to accusative or neutral alignment can be seen as transfers of particular subject properties — such as those of taking an unmarked case and triggering verbal agreement,

XX. Die Erforschung des Syntaxwandels III: Phänomene

as do (all or some) intransitive core arguments, themselves subjects almost by default — from INACTt to ACTt. 2.2.2.  Clear cases of transitive-centred realignments following from the wholesale replacement of a construction by one with a divergent alignment pattern are rare. Perhaps some of the examples of the preceding section, where transitive core arguments were apparently realigned on the model of coexisting constructions in other tenses or aspects etc., should be analysed in this manner. In the case of the optionally omitted ergative in Burushaski, where only 1st and 2nd person pronouns could be invoked as models, matters may have been more complex, too: the present tense, where this happened, was a periphrastic formation homonymous with, or actually deriving from, an intransitive nominalization where an ergative case was unwarranted; this intransitive-inspired construction where ACT could shed the ergative may then have been taken over by transitive verbs in the future tense, which was not periphrastic but used the same verb stem as the present. Less equivocal are a reconstructed development in Lardil and perhaps further members of the Tangkic subgroup of Pama-Nyungan (excepting Yukulta), and one documented for the early modern phase of many Indo-Aryan languages. Upon the phonetic effacement of tense markers in main clauses in Lardil, these were here mostly replaced by ‘infinitival’ subordinate clause constructions, where tense distinctions had survived, partly in the form of markers appended both to verbs and their INACTt arguments; since these peculiar tense-cum-case markers were barred from ACTi/INACTi and ACTt, all case-marked identically (in subordinate clauses originally dative or genitive, in main clauses then nominative), an ergative pattern of nominal case marking had turned accusative in the domain of this exchange of constructions. In early modern Indo-Aryan transitive verbs used to agree with INACTt in the participial (past or perfective) tenses, and the argument thus agreed with was in the nominative case (patternwise an absolutive); but this construction was frequently replaced, sometimes only when INACTt was definite or animate, by an impersonal one which originally had only been used with intransitive verbs: here verbs were invariably neuter or masculine singular, i. e. did not agree with INACTt in absolutive

64.  Ergativity

fashion, INACTt was in an oblique case (patternwise an accusative), and ACTt as well as ACTi/INACTi were in the ergative case (patternwise a nominative). 2.2.3.  Changes of alignment may further come about as the consequences of revaluations of non-basic constructions as basic and vice versa, with these alternative constructions possibly differing in transitivity and in the mapping of semantic relations onto syntactic ones. In a basic transitive construction where INACTt is subject, to a higher degree at any rate than ACTt, INACTt should align with the only candidates for subjecthood in intransitive clauses, ACTi/INACTi, in an ergative (or, more appropriately, absolutive) fashion for purposes of all kinds of rules operating in terms of this syntactic relation — e. g. rules of noun-phrase ellipsis in clause combination, which often require relational syntactic as well as referential identiy. If there are coexisting non-basic constructions of transitive verbs involving antipassivization or retopicalization, where subjecthood is instead conferred on ACTt (which is the topicworthier argument especially if verbs are imperfective or progressive and the INACTt referent is less than fully affected and on the verge of losing its status as a core argument), and it is these that are matched with basic intransitive constructions, rules sensitive to syntactic relations are bound to manifest an accusative alignment. If the verbs in such non-basic constructions are detransitivized in respect of INACTt, ACT, their single remaining argument of impeccable core status, should also for this reason naturally group with the single core argument of basic intransitives, especially with ACTi. The revaluation of such non-basic constructions as basic and vice versa, or perhaps even the complete supplanting of previously basic by previously non-basic constructions, eventually accompanied by a reintegration of somewhat peripheral INACTt arguments into the core of verbal valency, then automatically entails switched alignment of transitive arguments. The accusative pattern of case marking in the present-tense system of Georgian, with the dative as the INACTt case, has been assumed to originate in an antipassive construction; this was supposedly once a non-basic option highlighting imperfectivity and incomplete affectedness of INACTt, but was reanalysed as basic in non-perfective aspects, whereas the originally basic construction,

1191

with case marking on an ergative or active pattern, subsists in the aorist. It is controversial whether the alignment of case marking and perhaps further syntactic rules in ProtoPolynesian is to be reconstructed as ergative or accusative; according to the Ancestral Ergative Hypothesis many East Polynesian languages replaced the basic transitive construction, where ACTt had been marked by an (ergative) preposition and INACTt unmarked (like ACTi/INACTi), by an originally nonbasic imperfective construction with ACTt unmarked (like ACTi/INACTi) and INACTt marked prepositionally, itself purportedly an early innovation for transitive verbs modelled on the pattern of semi-transitive middle verbs (denoting relations where one participant does not really affect the other). Perfective transitive clauses, whose verbs allegedly generalized the suffix -Cia, for a while continued the ergative pattern, with ACTt marked by the original preposition and INACTt unmarked, but are claimed to have been revaluated as non-basic passives, with unmarked INACT as subject, completing the triumph of accusative alignment for basic constructions. In the history of the more consistently accusative Pama-Nyungan languages, hailing from the Ngayarda and Tangkic groups, intransitive antipassive constructions, once productively derived from basic transitive constructions and in certain circumstances probably even obligatory (e. g. in negative and some subordinate clauses or with certain person-number combinations of ACTt and INACTt pronouns), seem to have been reanalysed as basic, too. In Lardil only those original main clause constructions which had not been replaced by subordinate constructions (see above, 2.2.2.) were available for this reanalysis. The frequent presence in PamaNyungan of a structurally similar, basic or derived ‘middle’ construction, where ACT is subject for all syntactic and inflectional purposes and INACT is syntactically rather peripheral and generally in the dative, is likely to have been conducive to the revaluation of antipassives as basic and the complete or partial displacement of the former basic construction. Wargamay is (or rather was) a Pama-Nyungan language retaining an ergative pattern of nominal case marking (pronominal case marking being tripartite) and of argument ellipsis in clause combination. Nevertheless, owing to a reanalysis of intransitive verbs derived from transitives, as occurring in antipassive constructions, as morphologi-

XX. Die Erforschung des Syntaxwandels III: Phänomene

1192

cally basic, most transitive verbs could equally well be used in intransitive constructions, with ACT behaving in all respects like, and INACT unlike, ACTi/INACTi; and if it had not been for the retention of the original transitive construction of these verbs, alignments would thus have turned completely rather than only partly accusative. In a variation on this theme, in Kamchadal, mentioned earlier for its extension of the absolutive case to ACTt, the ergative case, formally identical with the locative, actually survived, but the basic transitive clauses where it occurred on ACTt appear to have been reanalysed as non-basic passives, influenced perhaps by an impersonal construction in Russian; basic transitive constructions are those where ACTt, like ACTi/INACTi and INACTt, is in the absolutive. In the Vakh dialect of Khanty, an Uralic language, two basic constructions coexist, one of which shows an ergative case marking pattern insofar as ACTt is in the locative and INACTt in the nominative (which is not differentiated from accusative for nouns — pronouns, however, are in the accusative); this construction seems to have been absorbed by the non-basic passive in the other dialects of Khanty, which is structurally not very different (even pronominal INACT is here in the nominative, and INACT rather than ACT is agreed with by the verb). It has also been conjectured that the passive of Khanty and related Mansi actually is this very construction with ergative features reanalysed as non-basic. In Kurdish the basic transitive construction of Middle Iranian provenance, where INACTt (like ACTi/INACTi) is in the direct case and triggers agreement and ACTt is in the oblique case, appears to have been crowded out by means of a non-basic construction where ACTt is topicalized, characterized by resumptive pronominal clitics cross-referencing the topic (retained in Southern but lost in Northern Kurdish). 2.3. Joint Intransitive and Transitive Innovations When changes affect intransitive and transitive clauses simultaneously, an ergative alignment may be superseded by a non-ergative one differing from its predecessor both in the intransitive and the transitive components of the pattern. This typically happens, though on the whole not very frequently, when rules of relational encoding are innovated or extended from one domain into another, rather

than being merely reorganized. Thus, in Tabasaran, an East Caucasian language with verbs traditionally agreeing in class with ACTi/INACTi and INACTt, i. e. with absolutives, person agreement was in addition introduced in intransitive and transitive clauses, and the arguments agreed with here were ACTi/INACTi (as in class agreement) and ACTt, i. e. nominatives. In a comparable example the point of departure, however, was not strictly an ergative pattern: in Maithili, a modern Indo-Aryan language where the impersonal construction mentioned in section 2.2.2. had become predominant, personal suffixes recently created from optional pronominal enclitics were extended from non-participial (present or imperfective) tenses to intransitive and transitive impersonal verbs in this construction, introducing nominative agreement (with ACTi/INACTi and ACTt) where there had been no agreement with any core argument (a kind of neutral pattern). If Kapingamarangi once shared the typical case marking pattern of SamoicOutlier languages — ACTi/INACTi and INACTt unmarked (the latter marked by the prepositions i/ki in semi-transitive middle constructions), ACTt with preposition e —, it must have innovated markers both for INACTt in canonical transitive constructions (viz. i, taken from middles or perhaps also retained from a former active construction, cf. 3.2.3.) and for ACTi/INACTj (among others e, originally confined to ACTt), arriving at an accusative pattern.

3.

Towards Ergative Alignment

3.1. Intransitive Realignment 3.1.1.  It is far less common for an ergative pattern to be the result of the realignment of intransitive core arguments in a basic construction than to be its victim, even though there are three possible sources: accusative patterns, turning ergative, perhaps via an active stage, by extensions of the accusative relation to (some) intransitive clauses and, equivalently, limitations of the nominative relation to transitives; active patterns, turning ergative by extensions of the inactive relation to ACTi and, equivalently, limitations of the active relation to ACTt; and tripartite patterns, turning ergative by aligning ACTi/INACTi, so far unaligned, with INACTt. The very few instances actually on record of such

64.  Ergativity

changes towards ergative, or ergativish, realignment involve rules of case marking and linear ordering and the distribution of case alternations. In a few Pama-Nyungan languages an accusative case marker, typically used on personal pronouns and perhaps certain nouns from the top of the animacy scale in INACTt function in this family (other nominals being in the absolutive case in this function), was extended to apparently all kinds of nominals (Dhalandji) or only to proper names (Warluwarra, Western Desert) in ACTi/INACTi function. Insofar as proper names are likelier than nouns lower on the animacy scale to occur as ACTi, the alignment thus innovated in Warluwarra and Western Desert could almost be characterized as counter-active, with INACTt tending to be paired with ACTi and ACTt with INACTi for a subset of nominals. In Medieval Latin, especially as written prior to the mid-8th century, and perhaps in other varieties of Late Latin, the accusative case, generally competing for survival with the nominative in the transition from Latin to the Romance vernaculars, was frequently extended in a conspicuously non-random manner: the nominatives replaced by accusatives were mostly markers of INACT in passive, i. e. derived intransitive constructions, and sometimes of INACTj in basic intransitive constructions, but virtually never of INACTt nor of ACTt. This is in fact an active pattern, and there was no way the extension of the accusative to all intransitive core arguments could be completed, since the paradigmatic contrast between accusative and nominative was about to collapse. According to the controversial Ancestral Active Hypothesis, Kartvelian languages (except Mingrelian) witnessed a development Late Latin might have been expected to undergo if this case contrast had survived, viz. the withdrawal of an active case, marking ACTt and ACTi, from intransitive clauses, transforming it patternwise into an ergative. Kungkari is one of the very few Pama-Nyungan languages where the old tripartite case system for personal pronouns was reduced to two cases by means of extending the INACTt rather than the ACTt form to ACTi/INACTi, resulting in an ergative rather than an accusative pattern, unusual for pronouns but standard for nouns. Although significant ergative alignments are not usually recognized in modern Uralic languages other than Khanty, the Finnic group (comprising Finnish, Karelian, Veps,

1193

Vote, Estonian, and Livonian) shows a development that is interpretable as an extension of an ‘accusative’. The distributional pattern of the partitive case in early Proto-Finnic is reconstructible as accusative, alternating in basic constructions with the accusative and nominative cases on INACTt, but never on ACTt or ACTi/INACTi. From INACTt, the first foothold in the process of its grammaticization from a local case of separation, the partitive then extended its distribution into intransitive clauses, especially those affirming the existence, the coming into existence, the cessation of existence, or an essential change in the state of their core argument — probably more often an INACTi than an ACTi. (Since verbs do not agree in person and number with core arguments marked partitively in the modern languages, this extension of an accusative presumably pertained to agreement as well.) It is questionable, however, whether the inactive or absolutive distribution of a mature partitive is typologically significant: such patterns are recurrent in languages with comparable devices for distinguishing between partial and total involvement, between quantified and unquantified, or indefinitely and definitely quantified, reference, and the like. In 17th century Spanish a statistical preference has been observed for intransitive core arguments, like non-clitic INACTt’s, to follow their verbs and for ACTt’s to precede them. The earlier pattern of preferred constituent ordering — ACTt + INACTt + Vb, ACTi/INACTi + Vb — was not unequivocally accusative: ACTi/INACTi could be paired with ACTt on grounds of clause-initial position, but with INACTt on grounds of preverbal position. Assuming that the former pairing is more appropriate, the frequent postverbal positioning of ACTi/INACTi, once INACTt had become postverbal, amounted to an acquisition of an accusative property by ACTi/INACTi. The resulting statistically predominant ordering pattern was perhaps again activish rather than strictly ergative: it was primarily non-topical intransitive arguments which were postverbal, and these are very common with verbs of existence, i. e. verbs taking an INACTi rather than an ACTi. As to the typological significance of this pattern, it should be noted that if linear ordering is sensitive to topic-comment structures, arguments of intransitive verbs of existence and perhaps other intransitive verbs of low semantic specificity are al-

1194

most bound to be aligned with INACTt, typically forming part of the comment in basic transitive constructions in most languages. It has been suggested that intransitive core arguments, subjects virtually by default, will tend to realign with transitive arguments holding the same syntactic relation, and that ACTt rather than INACTt is the subject of basic transitive constructions in the great majority of languages. On these assumptions it is not surprising that ergatives are frequently, but accusatives only rarely, extended to intransitive clauses. Where accusatives are so extended, however, an analogous reasoning is hardly cogent: the relevant languages are not among those where INACTt is the subject of basic transitives. Stronger motivation for realignments of ACTi/INACTi, or more typically only a subset of intransitive arguments, with INACTt seems generally to be provided by relational semantic similarities and common pragmatic potentials. 3.1.2.  Intransitive-centred changes introducing ergative alignment as the by-product of an exchange of constructions, analogous to those with the reverse effect dealt with in section 2.1.2., do not seem to be frequent. A pertinent example is perhaps the abandonment of the ancient intransitive impersonal construction in some dialects of Hindi while the innovated transitive impersonal construction (cf. 2.2.2.) was retained. There had been an accusative pattern here insofar as ACTi/INACTi and ACTt were both in the ergative case (and definite INACTt was in an oblique case). In the personal construction which took over from the impersonal intransitive one ACTi/INACTi was in the nominative case, which reinforced the ergative patterning of case marking in the past tense (although an accusative ingredient remained, viz. the oblique case on INACTt in impersonal transitives). 3.2. Transitive Realignment 3.2.1.  When the locus of change is in basic transitive constructions, an ergative pattern could evolve from four sources: from a neutral pattern, by way of ACTt separating from INACTt and ACTi/INACTi; from an accusative pattern, by way of ACTt separating from, and INACTt associating with, ACTi/ INACTi; from a tripartite pattern, by way of INACTt teaming up with ACTi/INACTi; and from a double-oblique pattern, by way of IN-

XX. Die Erforschung des Syntaxwandels III: Phänomene

ACTt separating from ACTt and joining ACTi/INACTi. Only the first and, more rarely, the second of these transitions seem attested, both involving realignments of ACTt. The rules concerned are primarily ones of case marking, rarely of verb agreement. In Papuan languages, noun phrases in core relations, unlike those in peripheral relations, are assumed originally not to have been casemarked; peripheral case markers such as ablative, causal, or instrumental were then optionally used also for ACTt, especially in situations where a transitive clause might otherwise be relationally ambiguous owing to insufficiently distinctive verbal agreement. (In other Papuan languages an oblique case, especially the dative, was instead used for INACTt, especially if animate, converting a neutral into an accusative rather than an ergative pattern; in Waris, such an extended dative, patternwise an accusative, was further extended from INACTt, of verbs expressing a change of state, to INACTi, creating an active pattern.) In Lardil, where subordinateclause constructions with accusative-style case marking have been installed in main clauses, non-infinitival subordinate clauses themselves seized on the genitive case as a new marker specifically for ACTt, perhaps first for personal pronouns (whose oblique case forms, some of which had already been employed in similar functions, are based on the genitive), which precluded relational ambiguities which might have arisen with transitive verbs one of whose arguments was omitted under identity with an argument in the main clause; when INACTt in such subordinate clauses is in the accusative, the resulting pattern is, however, tripartite rather than ergative. Among Indo-Iranian languages Shina (belonging to the Dardic group) is almost unique in having fully generalized the ergative case-marking pattern from past tenses or perfective aspects to all tenses or aspects: in the non-past, where ACTi/INACTi and ACTt used to be in the direct case, ACTt acquired a new ergative suffix, borrowed from the neighbouring Tibetan language Balti; INACTt continued to be in the oblique case with a minority of transitive verbs, giving rise to a tripartite pattern, but mostly changed over to the direct case, patternwise now an absolutive. Tibetan influence has also been claimed for a similar generalization in colloquial Nepali, where it was the indigenous instrumental, rather than a borrowed ergative case form, that gained currency as a marker of ACTt also in non-past tenses, with IN-

64.  Ergativity

ACTt being alternatively oblique (especially if animate) or direct. In Chinook, a Penutian language, pronominal clitics on verbs have originally had single forms common to all core arguments; as a result of phonological changes applying only in certain environments, special ACTt forms then happened to be differentiated for masculine and feminine 3rd person singular pronouns, and a special ACTt formative was added to others except those of 1st and 2nd person singular, which continued the neutral pattern. In a development that is more difficult to evaluate typologically, Gujarati and some dialects of Rajasthani and Pahari added number and gender markers to impersonal transitive verbs thus agreeing with INACTt. On the face of it this looks like an accusativization of a neutral alignment, since ACTi/INACTi and ACTt continued not to be agreed with in the impersonal construction. On the other hand, insofar as verb agreement is with absolutives, i. e. ACTi/INACTi and INACTt, in personal constructions of participial (past) tenses in these Indo-Aryan languages, the innovated agreement of former impersonal verbs with INACTt can also be seen as a partial reinforcement of absolutive agreement, with intransitive impersonals becoming the odd one out. Comparing realignments in transitive clauses which effect ergative patterns with those that affect them (2.2.1.), it seems that models from other domains in the same language play a less important role here (they are a factor only in the relevant Indo-Aryan languages), and that not much motivation is to be found in syntactic relations either, as there are no indications in the relevant languages that INACTt rather than ACTt has to be paired with ACTi/IACTi as subject. 3.2.2.  Although not a logical impossibility, ergative alignment does not seem to come about as a consequence of the wholesale replacement of transitive constructions patterning non-ergatively. 3.2.3.  It does come about rather frequently, however, as a result of the revaluation of non-basic constructions as basic, complementing or ousting, at least from some domains, the original basic transitive constructions — a mirror image of the transformation of ergative into accusative alignment dealt with in section 2.2.3. In a basic transitive construction where ACTt is subject, to a

1195

higher degree at any rate than INACTt, ACTt should align with the only candidates for subjecthood in basic intransitive clauses, ACTi/INACTi, for purposes of rules operating in terms of this syntactic relation. Now passives or other derived voices, nominalizations or other deverbalizations, and retopicalizations are possible non-basic constructions of transitive verbs where subjecthood is likely to be conferred on INACTt instead — a particularly strong rival of ACTt for topicworthiness if predicates are perfective, resultative, or stative, the INACTt referent is thoroughly affected by or is itself partly responsible for what has happened, and/or the ACTt referent is relegated to the communicative background or ranks lower than INACTt on the referential hierarchy or in definiteness. If intransitives are matched with such nonbasic constructions, rules sensitive to syntactic relations are bound to follow an ergative pattern — unless intransitives also appear in a non-basic construction where ACTi/INACTi is not subject. If the verbs or deverbal predicates in non-basic constructions of transitives are detransitivized in respect of ACTt, INACT is their single argument of indubitable core status, and this should be another reason for it to group with the core arguments of basic intransitives, in particular with INACTi. The revaluation of such nonbasic constructions as basic, with the obsolescence of previously basic constructions as a likely concomitant, and the eventual reintegration of somewhat peripheral ACT arguments into the syntactic core of bivalent verbs, then automatically entails switched alignment of transitive arguments. The ergative case marking and agreement patterns in Indo-Iranian are thus to be attributed to a perfective participial construction of transitive verbs used as a periphrastic passive or possessive construction in Old IndoAryan and Old Iranian (‘something is done by me’ or ‘my doing’, cf. for the latter Latin mihi factum est), where the participle agreed in gender, number, and case with nominative INACT, functioning as subject for other purposes as well, and ACT, syntactically a more peripheral argument, was in the genitive(-dative) or instrumental case. This nonbasic construction then took the place of the active perfect or past of transitives (initially unpassivizable), with the participle reintegrated into the finite verb paradigm and ACT reintegrated into the syntactic core (though not at first as subject), accounting

1196

for the ergative alignment in these aspectual or temporal domains. Old Armenian and, incipiently, Modern Irish may be further IndoEuropean examples of such reanalyses of respectively passive and possessive constructions with stative deverbal adjectives as basic. If Proto-Polynesian is to be reconstructed as accusative in case (or rather adpositional) marking and elsewhere in syntax (see section 2.2.3. for the contrary view), the most plausible origin of ergative alignments in Tongic and many Samoic-Outlier languages, by now essentially only a matter of case marking, is the reanalysis of a frequently used passive construction of transitives — itself intransitive, with suffix -Cia as a verbal passive marker, and with INACT unmarked and non-core ACT accompanied by an agentive preposition — as basic and again transitive. In this view the original passive suffix -Cia took on new functions (e. g. those of an inflectional transitivity marker or of derivational aspect markers), and the original active construction was lost or subsisted, perhaps merged with the semi-transitive middle construction. Assuming that not only Eastern Austronesian (especially Polynesian) languages, but also Western Austronesian ones, especially from the West Indonesian and Philippine groups, evince patterns that qualify as ergative, these seem best accounted for by deriving them historically from passives, or passive-like constructions where the focus is on INACT rather than ACT, that have come to be the norm for transitives. Passive sources of basic ergative alignments have likewise been claimed for Basque, Ancient Egyptian and some modern Aramaic dialects, and, more speculatively, Enga and other Papuan languages, Mayan, Salish languages such as Squamish, and Lardil (here for subordinate clauses only). The ‘middle’ and the passive have become obligatory in Rural and Urban Ozark English respectively whenever there is no true volitional and responsible agent (as in Sue’s new dress tore because of Billy / was torn by Billy — but he didn’t mean to); when there is one, transitive verbs must be used in the active construction. Since their obligatoriness, conditional on the nature of the ACT relation, distinguishes the Ozark middle and passive from typical derived voices, with the middle moreover being on morphological grounds almost as basic as the active, the ergative alignments of relational coding and syntax ensuing from them are no longer as clearly secondary vis-à-vis the accusative alignments ensuing from the active

XX. Die Erforschung des Syntaxwandels III: Phänomene

transitive construction as they are in Standard English. In Aleut, a nominalization of transitive verbs where INACT, i. e. the ‘possessed’ rather than the ‘possessor’, is assigned the same status as an ACTi/INACTi, has been held responsible for incipient basic ergative alignment. The Carib language Makúsi provides another example where the newly installed basic construction is not a formerly non-basic voice: here the reconstructed preferred order ACTt + INACTt + Vb has been changed to INACTt + Vb + ACTt, apparently by grammaticization of the right-dislocation of afterthought constituents; since ACTi/INACTi continued to occur preferably in preverbal or rather clause-initial position, the ordering pattern ended up as unequivocally ergative. 3.3. Joint Intransitive and Transitive Innovations Ergative alignment may appear in the course of changes affecting both intransitive and transitive clauses, but this seems to happen as infrequently as the reverse development. In Päri, a Western Nilotic language, the basic constituent order used to be Vb + ACTi/INACTi and Vb + ACTi + INACTt, and ACTi/INACTi and ACTt were casemarked identically while INACTt was unmarked. The ordering pattern was thus ambiguously accusative or ergative, depending on the pairing of ACTi/INACTi with ACTt as postverbal or with INACTt as clause-final; case marking conformed to the accusative pattern, if in a comparatively unusual manner, with the nominative as the marked and the accusative as the unmarked case. When the basic constituent order changed in both intransitive and transitive independent indicative clauses to ACTi/INACTi + Vb and INACTt + Vb + ACTt, apparently grammaticizing the topicalization of ACTi/INACTi and INACTt, ACTi/INACTi was unequivocally associated with INACTt in basic constructions (imperatives and subordinate clauses continued the old order). Since ACTi/ INACTi further dropped the nominative, case marking ended up as ergative too, by way of an extension of the (unmarked) accusative to, or withdrawal of the (marked) nominative from, intransitives. In certain nonPama-Nyungan languages of Australia all arguments used to be accompanied by nounclass markers (patternwise a neutral distribu-

64.  Ergativity

tion) but only some are now. In Burarra noun phrases in ACTi/INACTi and INACTt function have lost these class markers but those in ACTt function have retained them (elsewhere they were kept only on definite noun phrases, but these tend to be ACTt’s as well); these former pure class markers are thus exploited also for relational coding, distinguishing ergatives from not class-marked absolutives.

4.

Reasons for Realignments

4.1.  As ergative and other alignments commonly manifest themselves in the morphological encoding of grammatical relations, they may be altered fortuitously by phonological changes which affect inflectional exponents and create or obliterate paradigmatic contrasts, with the same forms perhaps faring differently in different contexts. Realignments due to such extraneous interferences are rare, though; normally they are a matter of genuinely morphosyntactic or also lexical change, and thus call for explanations of their own. A brief sketch of some general explanatory considerations, in particular of constraints on and incentives to realignments, must suffice here. 4.2.  The assumption that formal economy is at a premium in language change accounts for replacements of tripartite alignment, especially in case marking and verb agreement, by ergative, accusative, active, or doubleoblique patterns, because these reduce a ternary contrast to binary ones. Neutral alignment is of course the ultimate in economy, and should thus be the preferred target on this count. On the assumption that the foremost task of relational coding devices merely is to distinguish the relations of arguments which co-occur in one clause, ergative or accusative patterns should gain the upper hand on neutral and double-oblique ones, which fail in this task, on tripartite ones, which are unnecessarily uneconomical, and also on active ones, which are overdoing it too, there being no need of syntagmatic distinction in intransitive clauses. It follows that transitive clauses ought to be the locus of change if realignments are to do with the introduction of a relational distinction. On the assumption that relational encoding also ought to ensure the clear identifiability of the relation of each argument, realignments

1197

should be aiming at the active pattern, which best takes care of relational semantics, or perhaps the tripartite one, which best reflects verbal valency. It is the ergative, accusative, and active patterns which get the best overall grades on these three criteria. On each of them the ergative pattern finishes level with the accusative one, so that there should be no particular reason for them to interchange diachronically; but this is what they are wont to do. Otherwise there should be possibilities for ergative patterns to develop from tripartite and into neutral ones if the first criterion proves decisive, from neutral and into tripartite and active ones if things go as the third criterion would have it, and from all other patterns except the accusative one if improvement is according to the second criterion. Of all these possibilities only that of an ergative pattern evolving from a double-oblique one seems unrealized; the transition from ergative to double-oblique, on the other hand, is attested, if rarely, but cannot be considered an improvement on any count. 4.3.  As to rules and regularities of variable alignment, the patterning encountered most commonly at least in present-day languages is the accusative one, followed at some distance by the ergative one, with the active one coming third. Going by these unequal crosslinguistic frequencies, one might conclude that the currently less favoured alignments, representing the somehow costlier options, will generally be at a disadvantage, with their rivals always poised to advance at their expense. Some such view has been held most prominently by adherents of a ‘stadial’ theory of language development, for whom changes of alignment were essentially unidirectional and tied up with sociocultural progress. The idea of ergative alignment being inherently inferior to, hence less fit to survive than, its accusative competitor tends to seek support from the supposed failure of this pattern to give due recognition to the syntactic relation of subject, pairing ACTi/INACTi in basic constructions with INACTt rather than with ACTt. Much of the evidence surveyed in sections 2 and 3 is plainly inconsistent with any hypothesis that realignments are necessarily a one-way affair. 4.4.  A major, and more tangible, force that keeps alignments changing is the tension between diverse patterns coexisting in particu-

XX. Die Erforschung des Syntaxwandels III: Phänomene

1198

lar languages, providing each other with models and actual forms for analogical reorganization. No language is completely homogeneous in its alignments, and if this heterogeneity is due to rules and regularities which admit of variation, rather than only to such as are invariably of one or another type, the stage is set for some to-ing and fro-ing. Once alignments have come to be diverse, for whatever reason, a motive for realignments then is to reduce their diversity and thus to simplify grammars. To some extent this may be done randomly, but there are apparently also general principles determining winners and losers in such analogical levellings. Thus, if different alignments obtain for the syntactic behaviour of arguments and the morphological encoding of their relations, the latter will be adjusted to the former but not vice versa; or if constructions differing in alignment also differ in their expressive potential, perhaps owing to the loss of a morphological contrast in one of them, the more expressive one will succeed. The ways alignments are split within languages are largely non-random in the first place, reflecting different affinities, motivated by the semantics or pragmatics of types of arguments and constructions, to different patterns. Thus, whenever such alignments coexist, an ergative one is likely to appear for instance in past tenses or perfective aspects, with nominals not predestined to be agents and especially with non-pronouns, in relational coding in general and especially in case marking and agreement in terms of class, and in certain kinds of subordinate clauses, while present tenses and imperfective aspects, nominals predestined to be agents and especially 1st and 2nd person pronouns, syntactic rules of noun-phrase ellipsis in clause combination, relational coding by means of verbal agreement or cross-reference, agreement in terms of person, and main clauses are liable to pattern accusatively. Changes of alignment, occurring individually or in combination, should not lead to splits contravening such universal preferences. On the evidence of sections 2 and 3 only the main/subordinate parameter is not fully effective, insofar as accusative patterns are found to replace ergative ones selectively in main but also in subordinate clauses. Changes of alignment due to the borrowing of inflectional forms or entire constructions ought to be subject to similar constraints as those regulating language-inter-

nal trading. 4.5.  Non-basic constructions are unlikely to be revalued as basic — a common source of changes of alignment — unless they are used relatively frequently. In Standard English, as in other languages with predominantly accusative morphosyntax, there may be particular genres (e. g. scientific prose), verbs (e. g. bury), and tenses or aspects (past, perfect) where the passive is actually favoured over the active, but it is improbable that this will ever suffice for the passive to be reranked over the active even for these particular domains. One reason for massive increases in the frequency of non-basic constructions is the curtailment of the expressive potential of basic ones, perhaps caused accidentally by the phonological obliteration of morphological contrasts. Another, relevant in particular to voice alternations, are changes in the cultural milieu, in the practices, norms, and beliefs of the language community that are reflected in lexical and morphosyntactic structures relating to the notions of topic and subject. The (nominative) alignment of ACTt with ACTi/INACTi on the grounds of their being subjects, i. e. preferred topics, in basic constructions would seem to betray a certain cultural bias towards active arguments of transitive verbs as being inherently topicworthier than those acted upon; and vice versa for the (absolutive) alignment of INACTt with ACTi/INACTi. The conceptions of basic predicates as either perfective/resultative/stative or imperfective/progressive/dynamic, naturally focussing the main attention on ACTt or on INACTt respectively, may be a corresponding factor. Reversals of such cultural attitudes responsible for ACTt having the edge on INACTt or vice versa in respect of inherent topicworthiness are likely also to reverse the frequencies of their being chosen as actual topics and subjects, and concomitantly of basic and non-basic voices — a precondition for their revaluation and for appropriate modifications of the lexical semantics of predicates.

5.

References

Anderson, Stephen R. 1977. On mechanisms by which languages become ergative. Mechanisms of syntactic change, ed. by Charles N. Li, 317—36 3. Austin.

65.  Relativsätze

Comrie, Bernard. 1978. Ergativity. Syntactic typology: Studies in the phenomenology of language, ed. by Winfred P. Lehmann, 329—394. Austin. Dik, Simon C. 1980. On the subject of ergative languages. Studies in Functional Grammar, by Simon C. Dik, chapter 5. London: Academic Press. Dixon, R. M. W. 1979. Ergativity. Language 55. 59—138. Estival, Dominique, and John Myhill. 1988. Formal and functional aspects of the development from passive to ergative systems. Passive and voice, ed. by Masayoshi Shibatani, 441—491. Harris, Alice C. 1990. Alignment typology and diachronic change. Language typology 1987: Systematic balance in language, ed. by Winfred P. Lehmann, 67—90. Amsterdam. Kibrik, Aleksandr E. 1979. Canonical ergativity and

1199

Daghestan languages. In Plank (ed.) 1979, 61—77. Klimov, Georg ij A. 1983. Principy kontensivnoj tipologii. Moskva. Kuryłowicz, Jerzy. 1946 . Ergativnost’ i stadial’nost’ v jazyke. Izvestija Akademii Nauk SSSR, Serija literatury i jazyka 5. 387—393. Plank, Frans. (ed.) 1979. Ergativity: Towards a theory of grammatical relations. London. —. 1979. Bibliography on ergativity. In Plank (ed.) 1979, 511—554. —. 1985. The extended accusative/restricted nominative in perspective. Relational typology, ed. by Frans Plank, 269—310. Berlin. Trask, Robert L. 1979. On the origins of ergativity. In Plank (ed.) 1979, 385—404.

Frans Plank, Konstanz (Germany)

65. Relativsätze 1. 2. 3. 4. 5.

1.

Vorbemerkung Grammatik des Relativsatzes Der Relativsatz im Sprachwandel Zusammenfassung Literatur

Vorbemerkung

Die Grammatik des Relativsatzes gehört in den meisten Hinsichten zu den am besten erforschten Gebieten der Syntax. Umfassende Behandlungen sind Touratier 1980 und Lehmann 1984 (aus dem letzteren sind die meisten Beispiele ohne weitere Quellenangabe entnommen). Lediglich über die Diachronie des Relativsatzes ist vergleichsweise wenig Allgemeines bekannt. Verhältnismäßig zahlreichen Detailstudien, insbesondere über indogermanische Sprachen (vor allem Kurzová 1981), stehen wenige diachron-typologische Arbeiten gegenüber, aus denen Verallgemeinerungen über Wandel von Relativsätzen abgeleitet werden könnten. Man ist weitgehend in der methodischen Lage, aus wohlerforschten indogermanischen Phänomenen induktiv zu verallgemeinern oder die beobachtbare synchrone Variation in die Diachronie zu projizieren. Der Relativsatz wird in Abschnitt 2 als interlinguale grammatische Kategorie konzipiert. D. h. es wird ein Prototyp definiert, mit dem Strukturphänomene einzelner Sprachen verglichen werden können. Dabei kann sich für eine gegebene Sprache herausstellen, daß

sie eine Konstruktion besitzt, die zweifelsfrei unter die Definition fällt, oder eine Konstruktion, die nur marginal als Relativsatz zu betrachten ist, oder daß sie überhaupt keinen Relativsatz hat. Die im folgenden angesetzten definitorischen Bedingungen sind spezifisch genug, daß in der Tat nicht alle Sprachen einen Relativsatz haben, aber andererseits so schwach, daß doch die meisten Sprachen einen haben. Projiziert man diese Situation in die Diachronie und berücksichtigt die relativ kurze Dauer, über welche die meisten Sprachen der Welt historisch bezeugt sind, so wird verständlich, daß die meisten Sprachen während ihrer gesamten dokumentierten Geschichte einen Relativsatz haben. Folglich sind die Chancen, einen Relativsatz in statu nascendi oder moriendi zu beobachten, ziemlich gering. Dies ist der Aspekt des Objektbereichs, der für die genannte Forschungssituation verantwortlich ist. Empirisch gut untersucht sind lediglich Wandelprozesse, die den Relativsatz als solchen voraussetzen und unangetastet lassen. Verhältnismäßig wenig ist bekannt über Prozesse, die zu Relativsätzen führen oder sie eliminieren. Mehr hierzu in Abschnitt 3.1.

2.

Grammatik des Relativsatzes

2.1. Wesentliche Funktion des Relativsatzes An (1) läßt sich zunächst der Begriff des Relativsatzes vortheoretisch erläutern.

1200

(1) Wir kennen einen Arzt, dem Anna vertraut. (1) identifiziert einen Gegenstand — hier eine Person — durch begriffsbildende Operationen. Der Begriff ‘Arzt’ wird als Basis genommen. Seine Extension wird eingeschränkt auf Ärzte, die dadurch charakterisiert sind, daß sie in bestimmter Weise an einem anderen Sachverhalt beteiligt sind. Es wird somit der Begriff, ‘Arzt, dem Anna vertraut’ gebildet. In (1) wird genau ein unter den Begriff fallendes Individuum ins Redeuniversum eingeführt. Dies wird nicht durch den Relativsatz ausgedrückt, sondern durch den Singular und den indefiniten Artikel in (1). Die Möglichkeit jedoch, auf der Basis von Klassenbegriffen wie ‘Arzt’ so spezifische Begriffe zu bilden, daß sie bestimmte Individuen identifizieren, wird durch die Relativsatzbildung gewährleistet. 2.2. Subordination und Nominalisierung

XX. Die Erforschung des Syntaxwandels III: Phänomene

„Wo ist der Brief, den der Direktor geschrieben hat?“ (c) Büyük mektupnerede? groß Brief wo „Wo ist der große Brief?“ Ein infiniter Relativsatz wie der in (4b) heißt Relativpartizip. Wiewohl das Englische auch finite Relativsätze hat, spricht nichts dagegen, auch das Attribut in (3) als Relativpartizip zu bezeichnen. Es ist ganz gewöhnlich, daß eine Sprache über mehr als eine Konstruktion verfügt, die unter den Begriff des Relativsatzes fallen. Sie pflegen sich dann im Grad der Sententialität zu unterscheiden. Ein Relativsatz ist ein subordinierter (untergeordneter) Satz, traditionell auch Nebensatz genannt. Dadurch unterscheidet er sich von dem zweiten Satz in (5). (5) Wir kennen einen Arzt, dem vertraut Anna.

Die Eigenschaften, durch die ein subordinierter sich von einem selbständigen Satz unterDie definierenden Bedingungen für den Relascheidet, variieren von Sprache zu Sprache, tivsatz sollen nun durch Abgrenzung gegen gründen aber jedenfalls in der Tatsache, daß verwandte Konstruktionen zusammengestellt Subordination einen Ebenenwechsel (engl. werden. Ein Relativsatz ist ein Satz (im Sinne rank shift) bedingt: der Satz wird auf die von engl. clause, nicht sentence; s. Art. 2). Ebene eines Satzglieds herabgedrückt. Dieses Mithin ist das Attribut vertrauenswürdig en in ist im einfachsten Fall nominaler oder adver(2) kein Relativsatz. bialer Natur. Der subordinierte Satz verliert (2) Wir kennen einen vertrauenswürdigen also seine Sententialität und erwirbt stattdesArzt. sen die Nominalität bzw. Adverbialität mit ih(3) We know a doctor trusted by Anna. ren Korrelaten. Die Überführung eines prädikativen in einen nominalen Ausdruck heißt Die hier relevante Eigenschaft der SententiaNominalisierung. In der traditionellen Gramlität (Satzhaftigkeit) ist freilich gradueller matik ist dies ein Wortbildungsprozeß, so Natur. Das Attribut in (3) basiert auf einem daß als seine Basis nur ein Stamm, im einPartizip, ist jedoch um nichts finiter als das fachsten Falle ein Verb, infrage kommt. In in (2). Lediglich seine verbale Basis verleiht der Konstituentenstrukturgrammatik wird ihm einen höheren Grad an Sententialität. der Begriff jedoch weiter gefaßt, so daß auch Ein typischer Relativsatz hat ein finites verdie Umkategorisierung eines Satzes zum Nobales Prädikat. Viele Sprachen haben allerminal(syntagma) Nominalisierung heißt (vgl. dings keine Attribute auf verbaler Basis, die Art. 22). Methodisch ist es in jedem Falle anfinit wären. (4b) zeigt einen Relativsatz aus gemessen, die Syntax in diesen Zusammendem Türkischen, dessen Verb infinit ist und hang einzubeziehen, weil die Nominalisiedessen Subjekt im Genitiv steht. Die eckigen rung verschiedene syntaktische Eigenschaften Klammern, die mehrere Beispiele enthalten, des zugrundeliegenden prädikativen Ausschließen stets einen Nebensatz und, wo vordrucks verändert. (6 ) zeigt, wie die syntaktihanden, einen Relativsatz ein. schen Funktionen der Aktanten des finiten yaz-dı. (4) (a) Müdür mektup Verbs von (a) in der Nominalisierung (b) anDirektor Brief TÜRK schreib-PRÄT gepaßt werden. „Der Direktor schrieb einen Brief.“ (6) (a) Anna vertraut einem Arzt. yaz-diğ-ı (b) Müdür-ün (b) Annas Vertrauen zu einem Arzt [Direktor-GEN schreib-NR-POSS.3] mektup nerede? Andere Eigenschaften des Satzes (6a), die auf Brief wo dem Weg nach (b) verloren gehen, sind die

65.  Relativsätze

Kategorien des finiten Verbs wie Person, Numerus, Tempus, Modus. Die gleiche paradigmatische Beziehung besteht zwischen dem selbständigen Satz (4a) und dem Relativsatz in (b): Während in (4a) Präteritum spezifiziert wird, wird dies durch den Subordinator -diğin b neutralisiert; und während das Satzsubjekt in (a) im Nominativ steht, bewirkt die Nominalisierung in (b) seine Überführung in einen Genitivus subjectivus und damit seine Markierung durch das Genitivsuffix sowie das Auftreten des mit diesem nominalen Attribut kongruierenden Possessivsuffixes am Schluß des Relativsatzes. In diesem Sinne ist Nominalisierung gleichzeitig Desententialisierung, also Verlust an Sententialität. Subordination/Nominalisierung ist eine konstitutive Operation der Relativsatzbildung. Relativsätze verschiedener Sprachen und Sprachstadien unterscheiden sich darin, wie weitgehend sie nominalisiert sind. Die Subordination eines Satzes wird oft durch eine Konjunktion markiert, die im einfachsten Fall an einer seiner Grenzen steht. Eine subordinierende Konjunktion, die nicht zugleich eine besondere semantische Relation des Nebensatzes zum Hauptsatz ausdrückt, ist ein bloßer Subordinator (engl. complementizer). Häufig ist der Relativsatz durch denselben Subordinator untergeordnet, der auch in Komplementsätzen auftritt. Das englische (7) ist ein typisches Beispiel. (7) We know a doctor (that) Anna trusts. Auch der suffixale Subordinator -diğ- des Türkischen ist in Komplement- und Relativsätzen derselbe. Und ebenso wie in Komplementsätzen ist auch in Relativsätzen häufig gar kein segmentaler Subordinator vorhanden. Auch dies verdeutlicht (7). Während in der Konstituentenstrukturgrammatik der Terminus Nomen meist ‘Substantiv’ bedeutet, hat er in der traditionellen Grammatik eine allgemeinere Bedeutung und umfaßt neben dem Substantiv alle Wortarten, die diesem grammatisch ähnlich sind (s. Art. 2). Dazu gehört neben dem Pronomen vor allem das Adjektiv. Die Bildung eines Adjektivs, etwa des Partizips trusted in (3), heißt Adj ektivierung. Ein Adj ektival ist ein Syntagma, das dieselbe Distribution wie ein Adjektiv hat. So ergibt der Vergleich von (4b) mit (c), daß der Relativsatz ein Adjektival ist. Adjektivalisierung ist die Bildung eines Adjektivals. Wo nicht anders vermerkt, werden die Termini ‘Nomen’, ‘Nominal’ und ‘Nominalisierung’ im folgenden in dem weiten Sinne

1201

verwandt, in dem sie Adjektiv, Adjektival und Adjektivalisierung einschließen. Dies ist sinnvoll, erstens weil in zahlreichen Sprachen diese substantivischen und adjektivischen Kategorien und Prozesse einander ähnlich oder gleich sind, und zweitens, weil der Unterschied zwischen ‘substantivisch’ und ‘adjektivisch’ auch für Begriff und Funktion der Relativkonstruktion nur von sekundärem Belang ist. Im Hinblick auf Sprachen wie das Deutsche ist freilich festzustellen, daß der Relativsatz in (1) nicht dieselbe Stellung zu Arzt wie das Adjektiv vertrauenswürdig en in (2) und mithin nicht dieselbe Distribution hat, folglich kein Adjektival in dem definierten Sinne ist; Entsprechendes gilt für das Partizipial in (3). Es ist dies ein Indiz dafür, daß rein strukturelle Kriterien bei der Definition eines interlingualen Begriffs wie ‘Relativsatz’ gegenüber funktionalen zurückzutreten haben. 2.3. Attribution und Nukleusbildung Ein Vergleich von (1) mit (5) ergibt weiter, daß in (1) ein Nominalsyntagma (engl. noun phrase) einen Arzt, dem Anna vertraut gebildet ist, während die Kette einen Arzt, dem vertraut Anna in (5) kein Syntagma ist. Wie eingangs gesagt, wird in (1) auf der Basis eines Gegenstandsbegriffs derart ein spezifischerer Gegenstandsbegriff gebildet, daß sein Referent als Teilnehmer an einer bestimmten Situation charakterisiert wird. Syntaktisch betrachtet, geht es um das Verhältnis zwischen dem den Kernbegriff bezeichnenden nominalen Ausdruck, dem Nukleus (Arzt in (1)), und dem Restsatz. Dieses kann auf zwei Weisen gestaltet werden. Entweder der Nukleus wird, als Bezugsnomen (bzw. gegebenenfalls Bezugsnominal), aus dem Verband des Satzes, der die Situation bezeichnet, verselbständigt, und der Restsatz wird ihm als Attribut beigegeben. In diesem Sinne ist ein Relativsatz wie der in (1) ein Attributsatz. Das aus Bezugsnomen und Relativsatz bestehende komplexe Nominal, also Arzt, dem Anna vertraut in (1), heißt Relativkonstruktion. (8) Ammuk-maANA ŠEŠ-IA HETH [ich-aber DAT Bruder-mein kwi-n DUMU.SAL REL-AKK.SG.HUMTochter nepis-as KI-ass-a Himmel-GENErde-GEN-aber pihh-i, geb-l.SG] n-an-kan kwedan-i handa-mi? KONN-sie-ab wer-DAT ordn-l.SG

XX. Die Erforschung des Syntaxwandels III: Phänomene

1202

„Die Tochter des Himmels und der Erde [d. h. Prinzessin], die ich aber meinem Bruder geben werde, wem soll ich sie gleichstellen?“ (KUB XXI 38 Vs. 12’ f) Oder der Nukleus wird in seinem Satz belassen und nur besonders gekennzeichnet. In dem hethitischen (8) geht der Relativsatz dem Hauptsatz voran und ist nicht seine Konstituente. Er ist auch nicht Attribut zu dem Nukleus DUMU.SAL nepis-as KI-as, sondern enthält diesen in der Funktion und an der Position des direkten Objekts. Jedoch ist der Nukleus durch das Relativpronomen kwi- als solcher markiert. Dadurch wird der Relativsatz subordiniert und implizit nominalisiert, so daß das anaphorische Pronomen -an im Hauptsatz in der Tat auf eine Person und nicht auf einen Sachverhalt verweist. Der Nukleus dieses Relativsatzes ist also nicht sein Bezugsnomen, sondern vielmehr seine Konstituente. Es handelt sich daher um einen Relativsatz mit internem Nukleus (engl. internalhead relative clause), und zwar dessen angeschlossene Variante. Die englische Terminologie hält übrigens im allgemeinen den Nukleus als Kernbegriff der Relativkonstruktion und das Bezugsnomen als syntaktischen Kopf einer endozentrischen Konstruktion nicht auseinander, sondern nennt beides head. Da ein Relativsatz mit internem Nukleus kein Bezugsnomen hat, heißt er auch gelegentlich headless relative clause. Vom Relativsatz mit internem Nukleus gibt es die durch (9) aus dem Navaho illustrierte eingebettete Variante. (9) (shí) łééchąą’í b-aá hashtaał-ígíí NAV [ich Hund 3-für IMPF:1:sing-NR] nahał’in. IMPF:3:bell „Der Hund, für den ich singe, bellt.“ Hier erschöpft der Relativsatz die Relativkonstruktion, bildet also ein komplexes Nominal, das, als Nominalsyntagma, Konstituente des Hauptsatzes ist; in (9) ist es Subjekt. Wenn der Relativsatz mit internem Nukleus, wie in (8) und (9) und auch sonst meist, an der Spitze des Hauptsatzes steht, sind die angeschlossene und die eingebettete Variante nur noch durch das Stehen vs. Fehlen eines Anaphorikums im Hauptsatz unterscheidbar. Die zweite konstitutive Operation der Relativsatzbildung stellt also eine besondere Beziehung zwischen dem Nukleus und dem Restsatz her. Dafür gibt es zwei alternative

Verfahren, die Attribution, die den Restsatz dem Nukleus syntaktisch unterordnet und mit ihm zu einem komplexen Nominal vereinigt, und die Nukleusbildung, die den Nukleus innerhalb des ihn enthaltenden Satzes hervorhebt. Hinsichtlich der syntaktischen Relation des Relativsatzes zum Hauptsatz gibt es zwei Möglichkeiten. Der Relativsatz kann eingebettet sein. Er ist dann, zusammen mit seinem Nukleus, Konstituente eines Nominalsyntagmas, und dieses ist Konstituente des Hauptsatzes. Oder er kann angeschlossen sein. Dann ist er nicht Konstituente eines Nominalsyntagmas und überhaupt nicht Konstituente des Hauptsatzes, sondern bildet gemeinsam mit diesem einen komplexen Satz. Strukturell sind diese beiden Haupttypen klar unterschieden dadurch, daß die eingebettete Relativkonstruktion im einfachsten Falle die ihrer syntaktischen Funktion entsprechende Position im Hauptsatz einnimmt, während der angeschlossene Relativsatz ausschließlich am Rand des Hauptsatzes steht. Hinsichtlich der syntaktischen Relation des Relativsatzes zum Nukleus sind Relativsätze mit internem von solchen mit externem Nukleus zu unterscheiden. Die beiden Einteilungskriterien kreuzklassifizieren in der in (10) gezeigten Weise miteinander und definieren so die Stellungstypen des Relativsatzes. (10) Stellungstypen des Relativsatzes Relation zum Hauptsatz angeschlossen eingebettet Relation zum Nukleus vorangestelltzirkuminterner Nukleus nominal nachgestellt adnominal externer Nukleus Den vorangestellten Relativsatz illustriert (8), den nachgestellten (11) aus dem homerischen Griechisch. (11) autàr épeit’ epéthēke AGR aber dann dran:setzte:er mégan thureiòn groß:AKK.SG Türstein:AKK.SG hupsós’ aeírās, empor hebend:NOM.SG óbrimon, ouk àn gewaltig:AKK.SG [NEG wohl ge tón dúō kaì REL:AKK.SG doch zwei und eíkos’ ámaxai ... zwanzig Wagen:NOM.PL okhlísseian hebeln:würden]

65.  Relativsätze

„Aber dann setzte er einen großen Türstein davor, ihn hoch aufhebend, einen gewaltigen, den nicht zweiundzwanzig Wagen ... hätten wegwuchten können.“ (Hom. Od. 9, 240—242) Den zirkumnominalen Relativsatz illustriert (9). Der adnominale Relativsatz kann vor oder hinter seinem Bezugsnomen stehen und ist entsprechend pränominal, wie (4 b), oder postnominal, wie (1) und (7). 2.4. Anapher und Leerstellenbildung Ein Nebensatz, der Attribut zu einem Bezugsnomen ist, kann auch ein Komplementsatz sein, wie in (12). (12) Wir kennen das Problem, daß Anna ihm vertraut. Der wesentliche syntaktische Unterschied zwischen (1) und (12) besteht darin, daß eine im Nebensatz von (1) vorhandene syntaktische Funktion — die des indirekten Objekts — als vom Nukleus besetzt zu denken ist, während alle vorhandenen syntaktischen Funktionen im Komplementsatz von (12) intern besetzt sind. Wie ein jeder Modifikator eröffnet auch ein Relativsatz eine Leerstelle, in die der Nukleus eintritt. Eine Aufgabe der Relativsatzbildung besteht also darin, die Art der Beteiligung des Nukleus an der vom Nebensatz bezeichneten Situation, also die relativierte Funktion, zu verdeutlichen. Die Lösung dieser Aufgabe fällt ganz verschieden aus, wenn der Relativsatz Attribut zum Nukleus ist, als wenn er einen internen Nukleus hat. Im letzteren Falle ist die Aufgabe schon gelöst. Nur im ersteren Falle treten besondere syntaktische Mittel zur Markierung der relativierten Funktion ein. Wenn der Relativsatz hochgradig nominalisiert ist, wie der in (3), besteht die Strategie häufig darin, daß die relativierte Funktion auf die des Subjekts (bzw. in ergativischen Sprachen auf die des Absolutivs) beschränkt wird. Hierin verhält sich ein solcher Relativsatz dann ganz wie ein einfaches Attribut, also etwa das in (2). Es gibt auch Strategien wie die durch (4b) und (7) illustrierte, wo die relativierte Funktion variieren kann und dennoch unbezeichnet bleibt. Sowohl die türkische Konstruktion mit dem Nominalisator -diğ- als auch die englische Konstruktion ohne Relativpronomen oder Subordinator schließen die Subjektsfunktion als relativierte Funktion aus, lassen aber verschiedene oblique Funktionen zu. Die syntaktische Funktion der Leerstelle

1203

im Relativsatz ist dann nur ex negativo eindeutig, d. h. dadurch, daß alle anderen in ihm eröffneten Leerstellen besetzt sind. Die relativierte Funktion wird häufig auch durch ein Personalpronomen bezeichnet, das den Nukleus im Relativsatz vertritt. (13) i kopela NGR DEF.NOM.SG.F Mädchen:NOM.SG pu káθisa kondá tis [SR sitz:PRÄT:1.SG neben sie.GEN] „das Mädchen, neben dem ich saß“ Der neugriechische Relativsatz in (13) wird durch den Subordinator pu eingeleitet. Die relativierte Funktion ist die des Komplements zu einem Präpositionaladverb. Sie wird markiert durch das klitische Personalpronomen, das in dieser Funktion auch Resumptivum heißt. Es markiert die Leerstelle als solche, drückt durch seinen Kasus deren syntaktische Funktion mit aus und weist schließlich durch Kongruenz mit dem Bezugsnomen in Genus und Numerus darauf hin, daß eben der Nukleus in dieser Funktion zu denken ist. Da (13) ein Nominalsyntagma ist, ist die Beziehung zwischen dem Resumptivum und dem Bezugsnomen nicht eigentlich Anapher (vgl. Art. 47). Die Leerstellenbildung wird jedoch strukturell der Anapher angeglichen in dem Maße, in dem die in der Relativkonstruktion bestehenden syntaktischen Verhältnisse den von der Anapher vorausgesetzten ähneln. Das gilt uneingeschränkt nur für den angeschlossenen und in gewissem Maße noch für den postnominalen Relativsatz und ist darüber hinaus desto mehr der Fall, je weniger stark der Relativsatz desententialisiert ist. Ob die Leerstelle unbezeichnet bleibt oder durch ein Resumptivum markiert wird, hängt schließlich auch von der relativierten Funktion ab. Zentrale syntaktische Funktionen wie die des direkten Objekts bleiben eher unmarkiert als periphere wie die des Komplements einer Präposition (s. Keenan & Comrie 1977, Lehmann 1984, Kap. IV.3). (4b) ohne Resumptivum und (13) mit Resumptivum repräsentieren die gemäß diesen Generalisierungen normalen Verhältnisse. 2.5. Relativsatz und Relativpronomen Aufgrund der vorangegangenen Übersicht kann nun der Begriff des Relativsatzes definiert werden. Gegeben sei eine Proposition p und ein Term t mit einer Argumentfunktion in p. Dann kann ein komplexer Term t’ von einem t, das durch seine Rolle in p spezifiziert ist, gebildet werden. Wenn diese begriffliche

1204

Operation grammatikalisiert ist (vgl. Art. 70 zur Grammatikalisierung), so ist der p repräsentierende Satz ein Relativsatz und der t repräsentierende nominale Ausdruck dessen Nukleus. Der syntaktische Repräsentant von t’ ist die Relativkonstruktion. Die syntaktischen Operationen, die diese begriffliche Operation auf grammatischer Ebene umsetzen, sind die folgenden: Durch Subordination wird der Relativsatz zum Nebensatz, durch Nominalisierung wird er in einem gewissen Grade desententialisiert. Seine modifikative Beziehung zum Nukleus kann auf zwei Arten syntaktisch umgesetzt werden. Entweder es wird durch Leerstellenbildung eine bestimmte syntaktische Funktion im Relativsatz freigehalten und mit Strukturmitteln markiert, die den Mitteln der Anapher in dem Maße ähneln, wie die ganze Konstruktion die Bedingungen der Anapher erfüllt. Sodann wird der Relativsatz als Attribut mit dem Nukleus verbunden. Oder der Nukleus wird durch eine nominale Konstituente des Relativsatzes repräsentiert, die mit Strukturmitteln als Kern einer syntaktisch nicht expliziten Modifikation markiert ist. Die konstitutiven Operationen der Relativsatzbildung heißen daher Subordination/Nominalisierung, Attribution/Nukleusbildung und Leerstellenbildung/Anapher. Einige durch diese Operationen eingeführte Strukturmittel wurden schon erwähnt. Ein Mittel, das mindestens zweien, manchmal auch allen drei konstitutiven Operationen dient, ist das Relativpronomen. Das in (1) auftretende Pronomen dem leistet folgendes: 1. Es subordiniert den Relativsatz. Im Zusammenhang damit steht es, wie alle unterordnenden Konjunktionen des Deutschen, am Satzanfang. 2. Es kongruiert, vergleichbar einem Adjektivattribut, mit dem Nukleusnominal in Genus und Numerus, kennzeichnet also den Relativsatz als Attribut. 3. Es besetzt stellvertretend für den Nukleus die Leerstelle im Relativsatz und bezeichnet durch seinen Kasus deren syntaktische Funktion. Das Relativpronomen wird also auf der Basis des Relativsatzes bzw. der ihn konstituierenden Operationen definiert. In der traditionellen Grammatik wird gewöhnlich das Umgekehrte versucht. Dann lassen sich freilich die Relativsätze der meisten Sprachen der Welt einschließlich des englischen in (7) nicht mehr als solche bezeichnen. Die typologisch verschiedenen Verfahren der Relativsatzbildung heißen in der Fachliteratur auch Strategien. Der Terminus hebt den

XX. Die Erforschung des Syntaxwandels III: Phänomene

Aspekt der Problemlösung hervor, der ein sprachliches Mittel als solches konstituiert. Abschnitt 3. zeigt, daß eine Sprache im Laufe ihrer Geschichte sowohl ihre Strategie der Relativsatzbildung ändern als auch nur innerhalb einer Strategie die spezifischen grammatischen Zeichen verändern kann. 2.6. Determination in der Relativkonstruktion In einer deutschen Relativkonstruktion wie der von (1) geht der Determinator — hier der indefinite Artikel — dem Bezugsnominal voran und scheint dieses zu determinieren. Semantisch betrachtet ist dies jedoch bei der prototypischen Relativkonstruktion nicht der Fall. (14) Der Rektor, der der Ministerin gefallen will, verhält sich anders. (14) ist homograph in bezug auf den Skopus des ersten Worts (vgl. Art. 40). (Diese Homographie ist übrigens eine Besonderheit der deutschen Orthographie). Erstreckt der Skopus des Determinators sich auf das von der Relativkonstruktion gebildete komplexe Nominal, so werden Referenten für den mithilfe des Relativsatzes gebildeten spezifischen Begriff (Rektor, der der Ministerin g efallen will) gesucht. Es gibt einen optionalen Intonationsgipfel auf dem Artikel, und die Intonation bleibt an der Grenze zwischen Bezugsnominal und Relativsatz erhoben. Diese Verwendung eines Relativsatzes heißt restriktiv. Erstreckt der Skopus des Determinators sich bloß auf das Bezugsnomen, so hat der Relativsatz mit der Identifikation des Referenten nichts zu tun, sondern macht nur eine zusätzliche Aussage über den schon unabhängig aufgrund des Bezugsnominals identifizierten Referenten (der Rektor will der Ministerin g efallen). Das Bezugsnominal trägt den Intonationsgipfel, die Intonation kann dahinter steil abfallen, und der Relativsatz bildet eine eigene Intonationseinheit. Diese Verwendung eines Relativsatzes heißt appositiv. Im restriktiven Falle bilden also Bezugsnominal und Relativsatz ein Syntagma, im appositiven Falle nicht. Dieser Unterschied wird durch die suprasegmentale Ausdrucksstruktur ikonisch wiedergegeben. (15) (a) Der Rektor verhält sich anders, der der Ministerin gefallen will. (b) Der Rektor verhält sich anders, der will der Ministerin gefallen.

65.  Relativsätze

In (15a) liegt ein restriktiver Relativsatz vor, in (15b) dagegen eine Satzreihe. Eine appositive Relativkonstruktion hat die gleiche semantische Interpretation wie eine Satzreihe. Die Nutzung des zweiten Teilsatzes zur Identifikation eines Referenten setzt dagegen einen Relativsatz, nämlich einen restriktiven, voraus. Nach diesem semantischen Kriterium kann man Relativsätze auch in methodischen Situationen erkennen, wo man keine strukturellen Anhaltspunkte wie die Verbstellung hat, die im Deutschen selbständige von abhängigen Sätzen unterscheidet. Angenommen ferner ein Sprachstadium, das nicht über postnominale oder nachgestellte Relativsätze wie in (1) und (15a) verfügt. Dann können die Sprecher den zweiten Satz in (15b) dadurch als nachgestellten Relativsatz reanalysieren (vgl. Art. 70, Abschnitt 4.2. zur Reanalyse), daß sie das einleitende Demonstrativum kataphorisch intonieren und verstehen, also weiten Skopus für es erzwingen. Die Intension des Kernbegriffs kann leer sein oder sich auf allgemeine klassifikatorische Merkmale beschränken. Dies ist der Fall des sogenannten Relativsatzes ohne Bezugsnomen. Er konstituiert exhaustiv ein Nominal. Es gibt zwei Hauptvarianten, den Relativsatz mit pronominalem Nukleus wie in 6 (1 a) und den substantivierten Relativsatz wie in (16b). (16) (a) Wem Anna vertraut, dem kannst auch du trauen. (b) Der das getan hat, wird es büßen. Da der Nukleus eines restriktiven Relativsatzes nicht definit sein kann, manifestiert sich der pronominale Nukleus in vielen Sprachen als Fragepronomen. Ein substantivierter Relativsatz dagegen hat im allgemeinen keine besonderen Struktureigenschaften; ihm fehlt lediglich das Bezugsnomen.

3.

Der Relativsatz im Sprachwandel

3.1. Von benachbarten Konstruktionen zum Relativsatz Die Bildung komplexer Sätze, einschließlich der Relativsatzbildung, ist ein allgemeines Verfahren, das ungleich auf die Sprachen der Welt, nämlich auf Schriftsprachen gegenüber bloß oralen Sprachen, verteilt ist. Zum Teil beruht dies auf dem historischen „Zufall“ der über Jahrtausende wirkenden und inzwischen über viele Sprachen vermittelten Vorbild-

1205

funktion des klassischen Lateins. Im allgemeinen aber hängt es damit zusammen, daß die gesprochene Sprache einerseits suprasegmentale Mittel zur Verfügung hat, um die Zusammengehörigkeit und Abhängigkeit von Gedanken zu signalisieren, und insofern syntaktische Mittel in geringerem Umfang benötigt, und daß andererseits syntaktische Mittel das Kurzzeitgedächtnis und die Konzentration auf sprachliche Form stärker fordern und daher von der gesprochenen Sprache nicht begünstigt werden. Bei der Schriftsprache verhalten sich diese beiden Faktoren genau umgekehrt. Die Folge davon ist, daß Schriftsprachen ausgiebigen Gebrauch von komplexen Sätzen machen, während orale Sprachen oft nur ansatzweise darüber verfügen (Mithun 1984: 507—509). Alle Sprachen, in denen bisher keine Relativsätze entdeckt wurden, sind schriftlose Sprachen. Hier wie auch sonst korrelieren Sprachsystematik und Sprachgebrauch: Sprachen, in deren Texten komplexe Sätze einen großen Anteil haben, haben auch in ihrem System die komplexe Satzbildung stark ausdifferenziert. Sprachen, in deren Texten nur wenige komplexe Sätze vorkommen, haben auch nur wenige grammatische Verfahren zu ihrer Bildung. Was hier für Sprachen gilt, gilt übrigens auch für Sprecher. Bei einer Reihe von Sprachen, wo sich der Übergang von der Mündlichkeit zur Schriftlichkeit historisch beobachten läßt, ist auch ein Ausbau der syntaktischen Komplexität und eine Zunahme komplexer Sätze in Texten zu beobachten. Schon seit dem letzten Jahrhundert ist in der Sprachwissenschaft die folgende Auffassung von dem Ablauf dieses Vorgangs verbreitet: Auf einer primitiven Stufe werden lediglich selbständige Sätze juxtaponiert. Sie können, abgesehen von der Intonation, dadurch semantisch aufeinander bezogen sein, daß sie Proformen oder Partikeln enthalten, die auf einen Nachbarsatz verweisen. Dann findet eine Reanalyse statt, wodurch der eine Satz als Konstituente des anderen und das verbindende Zeichen als Operator dieses Ebenenwechsels interpretiert wird. Auf der zweiten Stufe ist somit ein komplexer Satz entstanden. Diese Art von syntaktischem Wandel kommt tatsächlich vor und wird in Abschnitt 3.1.1. f besprochen. Zu korrigieren ist allerdings zunächst die Auffassung von primitiven Sprachen, die überhaupt keine Subordination hätten. Solche Sprachen sind nicht bekannt. Ferner kann der Relativsatz — und ebenso auch andere komplexe Sätze — noch auf an-

1206

dere Weise entstehen. Die Möglichkeiten der Entstehung werden durch die Kombination der drei konstitutiven Operationen der Relativsatzbildung vorgegeben. Eine Relativkonstruktion entsteht aus einer anderen Konstruktion nicht dadurch, daß auf einmal alle drei konstitutiven Operationen angewandt werden. Vielmehr sind die Basis solche Konstruktionen, die sich von der Relativkonstruktion nur im Fehlen einer oder zweier der drei Operationen unterscheiden. Der Übergang zum Relativsatz besteht dann nur noch im Hinzutreten der fehlenden Operationen. Die traditionelle Auffassung der Entstehung eines Relativsatzes aus einer Satzreihe ist in dieser Sicht nur eine von drei Möglichkeiten, nämlich die, die auf einer vorhandenen anaphorischen Beziehung basiert. 3.1.1. Über den vorangestellten zum postnominalen Relativsatz Die Entstehung von Subordination in einer Satzreihe setzt deren textuelle Einheit und Gliederung nach der funktionellen Satzperspektive voraus (vgl. Art. 41 und 51). Im folgenden wird der Terminus Exposition verwendet für den situativen oder gegenständlichen Rahmen, der am Anfang eines komplexen Satzes steht und dessen geläufigste strukturelle Manifestation die sogenannte Linksversetzung ist (vgl. den Terminus topic in Chafe 1976 ). Eventum soll der Restsatz heißen, der innerhalb des gesteckten Rahmens bzw. mit Bezug auf die dort eingeführten Entitäten gilt. (17) Von den Feldzügen hatte ich Beutegut mitgebracht; damit schmückte ich sie aus. Im ersten Satz von (17) wird ein Referent eingeführt, der im zweiten Satz anaphorisch aufgenommen wird. Es gibt zwei Möglichkeiten, eine solche textuelle Einheit nach der funktionellen Satzperspektive zu gliedern. Wird sie in einer Sprache gebildet, die das Prädikat normalerweise am Satzende hat, so wird der erste Teilsatz als Exposition für den zweiten aufgefaßt werden, der das Eventum abgibt, wie in (18a). In einer Sprache, die das Prädikat normalerweise am Satzanfang hat, wird der erste Teilsatz das Eventum enthalten, und der zweite wird, wegen seiner Stellung, zwar nicht gerade als Exposition, aber jedenfalls als textsemantisch untergeordnet aufgefaßt werden, wie in (18b).

(18) (a) Welches Beutegut ich von den Feldzügen mitgebracht hatte, damit schmückte ich sie aus. (b) Von den Feldzügen hatte ich Beutegut mitgebracht, mit dem ich sie ausschmückte. Im Urindogermanischen war die Wortstellung zwar frei, aber die geläufigste Stellung des Prädikats war doch am Satzende. In einer Satzreihe wie (17) war daher die funktionelle Satzperspektive ‘Exposition—Eventum’ und damit die Möglichkeit einer syntaktischen Hierarchisierung der Struktur ‘Nebensatz— Hauptsatz’ vorgegeben. Die Funktion dieser Struktur war, im Vordersatz einen Referenten auf der Basis eines komplexen Begriffs einzuführen, der dann im Eventum eine Rolle spielt. Dies geschieht dadurch, daß der Kernbegriff des Vordersatzes durch das indefinitspezifische Pronomen *kwi- markiert wird, dem im Nachsatz ein Demonstrativum entspricht, das gleichzeitig die syntaktische Funktion des so implizit gebildeten komplexen Nominals im Nachsatz bezeichnet. Auf diese Weise entsteht der Typus des indogermanischen Relativsatzes, wie er durch das hethitische (19), ebenso aber auch durch (8) repräsentiert wird. (19) KASKAL-z-a kwi-t H [Feldzug-ABL-aber REL-N.SG ETH assu utahhun Gut(N.SG) mitbrachte:ich] n-at ape-danda jen-INSTR.SG KONN-sie halissiyanun. ausschmückte:ich „Mit Beutegut aber, das ich von den Feldzügen mitgebracht hatte, schmückte ich sie aus.“ (KBo III 22 Rs. 58, mit Ergänzung) Ein komplexer Satz, der in einem angeschlossenen Relativsatz ein Relativpronomen und im Hauptsatz ein ihm entsprechendes Demonstrativum enthält, heißt korrelatives Diptychon (s. Haudry 1973). In dessen Grundform ist der Relativsatz vorangestellt. Sobald die Konstruktion grammatikalisiert ist, fungiert das ehemalige Indefinitum als Relativpronomen, und dann gibt es auch eine invertierte Variante, in der der Relativsatz dem Hauptsatz folgt, wie in (20). (20) natta apū-nGEŠTINa-n H ETHNEG jen-AKK.SG Wein-AKK.SG piēr LUGALu-s gaben:sie[König-NOM.SG

65.  Relativsätze

kwi-n austa. REL-AKK.SGsahst] „Jenen Wein haben sie nicht gegeben, den du, König, gesehen hast.“ (KBo III 34 Vs. II 5d/KBO III 36 Vs. Beide Varianten des korrelativen Diptychons enthalten den Nukleus im ersten Teilsatz, was beweist, daß hier noch normale anaphorische Verhältnisse herrschen. Obwohl der Relativsatz in (20) nicht eingebettet ist, ist er eindeutig restriktiv. Auch die Grammatik des Satzes setzt die Bildung eines komplexen Begriffs voraus, denn der Skopus des Kataphorikums im Hauptsatz schließt den Relativsatz ein (vgl. Abschnitt 2.6 .). Diese semantischen Verhältnisse können in einem weiteren Schritt auch strukturell manifestiert werden. Wenn der Relativsatz okkasionell seinem Nukleus unmittelbar folgt, kann er als dessen Kokonstituente reanalysiert werden. Es resultiert ein postnominaler Relativsatz, wie in (21). (21) nu apēl É- ZU H ETHKONNjen:GEN.SG Haus-POSS.3 GIŠ eyan kwēl-a [REL:GEN.SG-aber Eyan ask-i-ss-i sakuwān Tor-LOK-POSS.3-LOKsichtbar] araūwan abgabenfrei „dann ist dessen Haus, an dessen Tor der Eyanstrauch sichtbar ist, abgabenfrei.“ (HG I, § 50)

3.1.2. Über den nachgestellten zum postnominalen Relativsatz Auf dem Wege vom Urindogermanischen zum Altgriechischen wurde eine Satzreihe wie die in (17) nach dem Muster von (18b) interpretiert. Das hierfür wesentliche Strukturmittel ist das demonstrative Anaphorikum im Nachsatz. Ein Beispiel wie (11) kann noch als Satzreihe oder schon als Relativkonstruktion interpretiert werden. Die semantische Voraussetzung für die Reanalyse als Relativkonstruktion ist auch hier geschaffen, wenn der Relativsatz, wie in (22), restriktiv ist. Agamémnōn (22) oud’ AGR NEG:aber Agamemnon(NOM.SG.M) lêg’ éridos abließ Streit:GEN.SG.F ten prȏton epēpeílēs’ [DEM:AKK.SG.F zuerst androhte Akhilêï Achilles:DAT.SG.M]

1207

„Aber Agamemnon ließ nicht ab von dem Streit, den er zuvor dem Achilles angedroht hatte. (Hom. Il. 1, 318 f) Damit ist auf anderem Wege als bei (20) ein nachgestellter Relativsatz entstanden. Auch hier führt die Grammatikalisierung zur Herausbildung eines Relativpronomens, so daß in derselben Struktur Relativpronomina indefiniter und definiter Herkunft fungieren können. Das weitere Schicksal des nachgestellten Relativsatzes im Griechischen verläuft dann ebenso wie im Hethitischen. Auch der Relativsatz in (22) kann, wenn er seinem Bezugsnomen unmittelbar folgt, mit diesem zu einer Konstituente zusammengefaßt und somit zum postnominalen Relativsatz werden, wie ihn (23) zeigt. (23) ts dè AGR DET:AKK.PL.F aber stḗlās ts Säule:AKK.PL.F [REL:AKK.PL.F hístā ... Sésōstris ... aufstellte Sesostris:NOM.SG.M] „die Säulen aber, die Sesostris aufstellte, ...“ (Hdt. II, 106) 3.1.3. Vom einfachen Attribut zum adnominalen Relativsatz In zahlreichen Sprachen können Partizipien attributiv konstruiert werden. Der Übergang zur Schriftlichkeit bringt gelegentlich den Ausbau solcher Partizipien zum komplexen Partizipial, wie in (3), mit sich. Dieser Vorgang ist fürs Türkische wahrscheinlich und führt zu dem in (4b) gezeigten Relativsatz. Der Vorgang ist für den Übergang vom Altlatein zum klassischen Latein historisch belegt, führt dort allerdings nicht zur Bildung eines Relativsatzes, denn die Partizipialien werden überwiegend prädikativ, nicht attributiv verwendet. Die Entstehungsgeschichte des deutschen Relativsatzes so wie durch (1) repräsentiert ist kompliziert (vgl. Johansen 1935). Aus dem Urgermanischen war bereits ein postnominaler Relativsatz ererbt, der mit einem optionalen Subordinator eingeleitet wurde, wie in (24). (24) then weg [th’er faran wolle] AHD „den Weg, den er ziehen wollte“ (Otfr.) Dieser Typ existiert auch noch im Mittelhochdeutschen, wird jedoch durch die Neubildung verdrängt. Diese hat ihrerseits zwei

1208

Voraussetzungen. Die erste besteht in einer attributiven Konstruktion, die sonst auch im Altgriechischen und Altakkadischen vorkommt und in der das Attribut durch ein Demonstrativum mit seinem Bezugsnomen verbunden wird, wie in (25) und (26). (25) Ih meinu sancta Mariun, kuningin AHD thia richun. „Ich meine St. Maria, die mächtige Königin.“ (Otfr. 1, 3, 31) (26) ich bim Gabriel thie azstantu fora AHD gote „Ich bin Gabriel, der vor Gott steht“ . (Tat. 2, 9) Die zweite Voraussetzung sind Satzreihen wie (27), die die oben anhand von (17) illustrierte Struktur haben und textsemantisch nach dem Muster von (18b) interpretiert werden. (27) ingegin liofun imo zehen man riobe, AHD thie gistuontun ferro „entgegen liefen ihm zehn aussätzige Männer, die blieben in der Ferne stehen“ (Tat. 111, 1) Jede der beiden Konstruktionen gibt die Grundlage für den neuen postnominalen Relativsatz ab. Die Reanalyse der Satzreihe von (27) als nachgestellte Relativkonstruktion und deren Reanalyse als postnominale Relativkonstruktion verläuft ebenso wie im Altgriechischen ((22) f); die Demonstrativa, die in den beiden Sprachen die Relativa ergeben, sind übrigens etymologisch verwandt. Der Ausbau der attributiven Konstruktion zur Relativkonstruktion erfordert, daß der ererbte Relativsatz mit optionalem Subordinator in die Position des Attributs, also etwa des Partizipials in (26 ), eingesetzt wird. (28) zeigt dies für die Variante mit Subordinator, (29) für die ohne Subordinator. (28) gisah ther heilant ... iungiron ... AHD [then-de her minnota] „sah der Heiland den Jünger, den er liebte“ (Tat. 206, 2) (29) sandida mih ... zi dheodom [dhem euAHD wih biraubodon] „sandte mich zu den Völkern, die euch beraubten“ (Is. 11, 5) Ebenso wie bei den nicht satzartigen Attributen von (25) und (26 ) kongruiert hier das Demonstrativum nicht nur in Genus und Numerus, sondern auch im Kasus mit seinem Bezugsnomen. Es leistet von den drei konstitutiven Operationen also nur die Attribution. Diese ist also sehr aufwendig, die Leerstelle

im Relativsatz dagegen überhaupt nicht markiert. Dasselbe Pronomen dient jedoch in Sätzen wie (27) als Anaphorikum. In Anlehnung an solche Strukturen findet nun folgende Reanalyse statt. Gegeben ist eine Konstruktion, die, wie (29), keinen Subordinator hat und wo die relativierte Funktion, wie in (28), zufällig denselben Kasus des Relativpronomens erfordert wie die Kongruenz mit dem Bezugsnomen. Dann wird das Relativpronomen als Leerstellenbesetzer und gleichzeitig als Subordinator interpretiert. Fortan richtet sich der Kasus des Relativpronomens nach der relativierten Funktion, und es resultiert ein Relativsatz wie in (30), das ist der auch noch in (1) repräsentierte Typ. (30) sê mîn sunu [den ih gachôs] AHD „sieh, mein Sohn, den ich erwählt habe“ (Mth. [Mondsee]) 3.2. Wandel im Subtyp 3.2.1. Vom Relativpronomen zur Konjunktion Ein Relativpronomen, das zur Subordination dient, kann sich auf einen Subordinator reduzieren. Dies ist aus der Geschichte des Altakkadischen, der modernen arabischen Dialekte, des Übergangs vom Altgriechischen zum Neugriechischen sowie des Lateinischen zu den romanistischen Sprachen bekannt. Auch das that, das im Englischen Relativsätze wie (7) einleitet, ist auf diese Weise entstanden. Der Relativsatz des Lateinischen ist auf dieselbe Weise wie der hethitische (vgl. 3.1.1.) entstanden (Einzelheiten in Lehmann 1979); sein Relativpronomen qui geht ebenfalls auf idg. *kwi- zurück. Während in den Stadien des archaischen und Altlateins das korrelative Diptychon sowohl in der vorangestellten als auch in der nachgestellten Variante häufig ist, wird es bis zur klassischen Periode weitgehend durch den postnominalen Relativsatz, mit einem Relativpronomen so wie in Abschnitt 2.5. beschrieben, verdrängt. legione (31) ea LAT DEM:ABL.SG.F Legion:ABL.SG.F quam secum habebat ... [REL:AKK.SG.F sich:mit hatte] ad montem Iuram ... zu Berg:AKK.SG.M Iura:AKK.SG.F murum ... perducit. Mauer:AKK.SG.M aufführte „Mit der Legion, die er bei sich hatte, führte er eine Mauer bis zum Juragebirge auf.“ (Caes. B. G. I, 8, 1)

65.  Relativsätze

Prototypische Beispiele dieser Konstruktion wie (31) haben die Theorie des Relativsatzes über Jahrtausende bis auf den heutigen Tag geprägt. In der Geschichte des Lateins ist sie jedoch nur für die klassische Periode typisch. Bereits in den ersten Jahrhunderten nach Christus wird die nominale Morphologie durch eine Konspiration von phonologischen, morphologischen und syntaktischen Faktoren radikal vereinfacht. Das Relativpronomen verliert Genus und Numerus gänzlich. Die Kasusflexion geht im Nomen ganz, im Pronomen bis auf einen geringen Rest verloren. Auf der Stufe des Spät- bzw. Vulgärlateins, dem (32)—(36 ) entstammen, wird für das Relativpronomen zunächst eine Dreikasusflexion eingeführt: Nominativ qui ((32)), Akkusativ que ((33)), andere oblique Funktionen cui ((34)). (32) Primae Florentiae LAT Prima:DAT.SG.F Florentia:DAT.SG.F filiae carissimae ... Tochter:DAT.SG.F teuer:DAT.SG.F qui ab [REL:NOM.SG.M von Orfeu maritu Orpheus:ABL.SG.M Gatte:ABL.SG.M in Tiberi in Tiber:ABL.SG.M decepta est. betrogen:NOM.SG.Fist] „Für Prima Florentia, die geliebte Tochter, die von ihrem Ehemann Orpheus im Tiber im Stich gelassen wurde.“ (Thylander A 210) (33) de illo que fecerat LAT von jen:ABL.SG.N [REL getan:hatte] „von dem, was er getan hatte“ (Rhytm. de Ludovico II, 9, 1) (34) episcopum de LAT Bischof:AKK.SG.M [von cui parrochia REL:DAT.SG Gemeinde:NOM.SG.F fuit war] „den Bischof, von dem die Gemeinde war“ (Leg. Alam. 12) Es gibt auch schon vulgärlateinische Belege dafür, daß que in diesem Paradigma die unmarkierte Form ist und die Funktionen der anderen Formen mitübernehmen kann. In (35) fungiert es als Komplement einer Präposition, also anstelle des in (34) erscheinenden cui.

1209

(35) M. Aurelius LAT M. Aurel:NOM.SG.M Hermes fecit Hermes:NOM.SG.M machte Casperiae Casperia:DAT.SG.F coiugi Zosime Zosima:DAT.SG.F Gattin:DAT.SG.F bene merenti con wohl verdient:DAT.SG [mit quem vixit REL(AKK.SG.M) lebte annos XVI Jahr:AKK.PL.M sechzehn] „M. Aurelius Hermes machte es für Casperia Zosima, seine treue Gattin, mit der er 16 Jahre lebte.“ (CIL VI 13122) In den romanischen Schriftsprachen erhalten sich Varianten dieses Systems. Da das Relativpronomen gegenüber dem Zustand des klassischen Lateins an Funktionalität eingebüßt hat, wird das Pronomen „welcher“ (ital. il quale, frz. lequel, span. el cual usw.) zusätzlich als Relativpronomen genutzt. Auf umgangssprachlichem Niveau geht der durch (33) angedeutete Vorgang dagegen weiter. In dem Maße, in dem das Relativpronomen die Leerstelle nicht mehr markiert, wird ein Personalpronomen als Resumptivum eingeführt. Damit wird bei den im Sinne von Abschnitt 2.4. peripheren syntaktischen Funktionen angefangen ((36 )), und bereits im 13. Jahrhundert wird gelegentlich sogar die direkte Objektsfunktion durch ein Resumptivum bezeichnet ((37)). Gleichzeitig setzt sich als Form des Relativums que (ital. che) gegen alle anderen durch. (36) hominem LAT Mensch:AKK.SG.M ego quem [REL(AKK.SG.M) ich beneficium ei Wohltat:AKK.SG.N er:DAT.SG feci machte] „einen Mann, dem ich einen Gefallen getan habe“ (Form. And. 48) (37) Tu hai un’altra cosa che ITAL du hast eine andere Sache REL non la ho io. NEG sie habe ich „Du hast etwas anderes, was ich nicht habe.“ (Dec. 3, 10) Ähnliche Beispiele ließen sich aus anderen romanischen Sprachen anführen (s. z. B. Gui-

1210

raud 196 6 für das Französische). Da das den Relativsatz einleitende Wort nicht mehr flektiert (nur im Relativsatz ohne Bezugsnomen tritt das Fragepronomen auf) und auch nicht von Präpositionen regiert werden kann, ist es kein Pronomen mehr, sondern erfüllt nur noch die Funktion der Subordination. Die Attribution kommt durch die nunmehr obligatorische postnominale Stellung des Relativsatzes zum Ausdruck. Für die Leerstellenbildung wird, außer in der Subjektsfunktion und mit gewissen Unterschieden zwischen den romanischen Sprachen, ein Personalpronomen (normalerweise dessen klitische Variante) eingesetzt. Auch in den romanischen Umgangssprachen ist also der Subordinator des Relativsatzes derselbe wie im Komplementsatz. Ebenfalls in verschiedenen Sprachen findet sich die Verwendung von ‘wo’ als relativischer Subordinator. Beispiele sind oberdeutsch wo, neugriech. pu und Tok Pisin we (Romaine 1992). An umgangssprachlichen Beispielen wie der Krug , wo der Henkel abg ebrochen ist sieht man, daß diese Ausbreitung von ‘wo’ ihren Ausgang nimmt an Fällen, wo die relativierte Funktion alternativ lokal (an dem der Henkel ...) oder nicht-lokal (dessen Henkel) aufgefaßt werden kann. 3.2.2. Ersatz des Subordinators Ein Wandel der Strategie auf noch niedrigerer Ebene ist der Ersatz der einleitenden Konjunktion eines postnominalen Relativsatzes. Allerdings wird ein Morphem mit syntaktischer Funktion nicht ohne weiteres gegen ein anderes ausgetauscht. Das Beispiel des Isländischen zeigt, daß dabei diachrone Beziehungen zwischen ganzen syntaktischen Konstruktionen im Spiel sind (vgl. Kemmer 1984). Wie das Althochdeutsche hatte auch das Altisländische einen postnominalen Relativsatz mit einleitendem Subordinator aus dem Urgermanischen geerbt. Der Subordinator lautet er und tritt nicht nur in Relativsätzen wie (38), sondern auch in Temporalsätzen und anderen Adverbialsätzen auf. (38) Eirīkr inn rauđi hēt mađr ISL Erik der rote hieß Mann breiđfirzkr er fōr ūt hēđan Breitfjorder [REL fuhr aus von-hier Þangat dorthin] „Erik der Rote hieß der Breitfjorder Mann, der von hier dorthin ausfuhr“ (Libellus Islandorum 127)

XX. Die Erforschung des Syntaxwandels III: Phänomene

Ebenfalls bereits im Altisländischen gibt es die Partikel sem in der Bedeutung „wie“ : (39) ISL (40) ISL (41) ISL

rauđr sem blōđ „rot wie Blut“ svā vitran mann sem Þū ert „so ein weiser Mann wie du bist“ Þat var et sama das war der selbe sem hann hafe honum greitt [wie er hatte ihm bezahlt] „das war derselbe [Preis], wie/den er ihm bezahlt hatte“

Wo sem ein Demonstrativum aufnimmt, wie in (40) f, liegt ein invertiertes korrelatives Diptychon (s. 2.3.) vor. Wenn die verglichene Einheit nominaler Natur ist, wie in (41), ist die Struktur grundsätzlich schon die der Relativkonstruktion. Wenn der Vergleich nun noch auf Identität hinausläuft, wie in demselben Beispiel, ist die Vergleichspartikel einem Relativum völlig äquivalent, wie auch die alternative Übersetzung verdeutlicht. Wie die Abschnitte 2.3. f zeigten, kann ein nachgestellter Relativsatz als postnominaler, mithin als Attribut des Bezugsnomens reanalysiert werden, wenn er diesem unmittelbar folgt. Dasselbe geschieht bei Vergleichssätzen. Der Nebensatz in (41) kann als Attribut des (in diesem Falle zufällig leeren) Bezugsnomens aufgefaßt werden. Unter diesen Bedingungen ist der mit sem eingeleitete Komparativsatz einem mit er eingeleiteten Relativsatz semantisch und syntaktisch völlig äquivalent. Noch zu altisländischer Zeit wird sem als Subordinator eines Relativsatzes reanalysiert und fängt auch an, in echten Relativsätzen aufzutreten, wie in (42). (42) Þat fōdr sem hrossonom var ISL das Futter [REL Rössern war ǣtlat ādr bestimmt vor] „das Futter, das für die Pferde bestimmt war“ (Hoensa-Thóris-Saga 13, 9) Während sich die Variation zwischen er und sem in der Schriftsprache bis heute hält, kommen in der neuisländischen Umgangssprache seit Jahrhunderten nur noch durch sem eingeleitete Relativsätze vor. 3.3. Vom Relativsatz zu benachbarten Konstruktionen Wir Ebene tion p der in

nehmen auf funktionalsemantischer eine Konstellation aus einer Proposiund einem ausgezeichneten Term t an, ihr eine Argumentfunktion wahrnimmt

65.  Relativsätze

und gleichzeitig gegenüber allen anderen Elementen von p hervorgehoben ist. Diese Prominenz von t besagt, daß die Konstellation für den umgebenden Kontext nicht einfach eine Proposition ist, sondern eher t, insoweit er an p teilnimmt. Präzisiert man die Konstellation zwischen t und p, so ergeben sich mindestens drei Varianten. In der ersten wird auf der Basis von t ein komplexer Begriff gebildet bzw. ein Gegenstand identifiziert dadurch, daß p ihn modifiziert. Das Resultat ist ein komplexer Term, ausgedrückt durch eine Relativkonstruktion. In der zweiten Variante wird die durch t bezeichnete Entität kontrastiert gegen alle anderen, die dieselbe Stelle besetzen könnten, auf die p aber nicht zutrifft. Das Resultat ist eine komplexe Proposition, ausgedrückt durch eine Fokuskonstruktion. In der dritten Variante gibt es ein übergeordnetes Prädikat, dessen Argument ebensowohl p wie t sein kann, aus dessen Perspektive also die durch p bezeichnete Situation an der durch t bezeichneten Entität auftritt. Das Resultat ist ein Argument zu dem höheren Prädikat, dessen unmittelbarer Repräsentant t ist und zu dem p prädikativ ist, ausgedrückt z. B. durch ein Perzeptionsverb mit einer AcI-Konstruktion. In allen drei genannten Fällen gibt es auf struktureller Ebene die Alternative, t in p an Ort und Stelle zu belassen und mit segmentalen oder suprasegmentalen Mitteln zu markieren, oder t aus p herauszunehmen und den syntaktischen Repräsentanten von p dem syntaktischen Repräsentanten von t zu subordinieren. Es ergeben sich die folgenden sechs Konstruktionen: 1. Relativkonstruktion (a) Relativsatz mit internem Nukleus (vorangestellt oder zirkumnominal) (b) Relativsatz mit externem Nukleus (nachgestellt oder adnominal) 2. Fokuskonstruktion (a) einfacher Satz mit Fokuskonstituente (b) Spaltsatz (oder Pseudospaltsatz) 3. (a) Komplementsatz (dessen Subjekt t repräsentiert) (b) AcI und andere Konstruktionen, die aus einem Nominalsyntagma und auf einer darauf bezogenen prädikativen Konstruktion bestehen. Innerhalb jeder der drei Gruppen gibt es systematische Beziehungen zwischen den beiden Varianten. Davon interessiert hier nur die Variation in der Relativkonstruktion, die

1211

in 3.1.1 f dargestellt wurde. Ferner gibt es Beziehungen über die Gruppen hinweg zwischen den a-Varianten und zwischen den bVarianten. Diese werden im folgenden auszugsweise behandelt. 3.3.1. Vom Relativsatz zum Spaltsatz Ein Spaltsatz ((43)) ist ein komplexer Satz, bestehend aus einem selbständigen Satz, genannt Fokussatz, und einem abhängigen Satz, genannt extrafokaler Satz (vgl. Art. 51), und zwar in dieser Reihenfolge. Der Fokussatz ist ein Kopulasatz mit semantisch leerem und höchstens pronominal repräsentiertem Subjekt und einer nicht-finiten Konstituente im Prädikat, die der Fokus des Gesamtsatzes ist. Der extrafokale Satz ist ein unmarkierter Nebensatz ohne semantische Rolle oder Satzgliedfunktion im Hauptsatz. Er enthält eine Leerstelle, die vom Fokus besetzt gedacht wird. Ein Pseudospaltsatz ((44)) ist ein komplexer Satz, bestehend aus einem Relativsatz ohne Bezugsnomen und einem Hauptsatz, in dieser Reihenfolge. Der Hauptsatz ist eine Gleichsetzung in Form eines Kopulasatzes, dessen Subjekt der Relativsatz ist und dessen Prädikat der Fokus des Gesamtsatzes ist. Beide Konstruktionen sind im Französischen und Englischen sehr beliebt, und besonders der Spaltsatz kommt in zahlreichen Sprachen vor. C’est en marchant que j’écris mes (43) FRANZ lettres. „Ich schreibe meine Briefe beim Marschieren.“ (44) Ce qu’on voit ce sont des rats. FRANZ „Was man sieht, sind Ratten.“ Auf der Basis der in Abschnitt 3.3. konstatierten funktionellen Ähnlichkeit zwischen Relativkonstruktion und Spaltsatz gibt es auch Beispiele, die zwischen den beiden Interpretationen strukturell zweideutig sind, wie (45). (45) What John did was annoying the children. In einer Interpretation ist der Nebensatz Subjekt eines Verbalsatzes mit progressivem Aspekt in der Verbform. In der Interpretation als Pseudospaltsatz ist (45) ein Kopulasatz mit einem Gerundial im Prädikat. Eine vom Satzfokus bevorzugte Position ist der Satzanfang. Sprachen mit freier Wortstellung wie Altindisch und Altgriechisch können ohne weitere syntaktische Anpassung jede beliebige Konstituente dorthin bringen und haben daher keinen Bedarf am Spaltsatz.

1212

Das Lateinische jedoch büßt während seiner Geschichte, mit Ausnahme der klassischen Kunstsprache, seine Wortstellungsfreiheit ein und bildet gleichzeitig seit altlateinischer Zeit den Spaltsatz aus (vgl. Löfstedt 196 6 für das Folgende). (46 a) zeigt einen einfachen Satz mit Fokuskonstituente in Initialposition, (46b) den entsprechenden Spaltsatz. (46) (a) In Scythiam in Skythien:AKK.SG LAT mittimur. geschickt:werden:wir. „Nach Skythien schickt man mich.“ (b) Scythia est quo Skythien:NOM.SGist [wohin mittimur. geschickt:werden:wir] „Skythien ist es, wohin man mich schickt.“ (Ov. Trist. 1, 3, 61) Die Konstruktion in (46 b) ist noch restlos analysierbar als Kopulasatz mit einem (in (46 b) adverbialen) Relativsatz mit pronominalem Nukleus als Subjekt und einem nichtverbalen Prädikat, das, da es Satzfokus ist, am Anfang steht. Der Spaltsatz ist also noch keine eigenständige Konstruktion. In dem Maße, in dem die Wortstellungsfreiheit abnimmt, sind Sätze wie (46 a) nur noch eingeschränkt möglich. Da auf höheren grammatischen Ebenen immer größere Freiheit herrscht als auf niedrigeren, wird die für die funktionelle Satzperspektive notwendige Stellungsfreiheit durch Rekurs auf die Ebene des komplexen Satzes (wieder-)gewonnen. Der Spaltsatz gewinnt an Bedeutung, löst sich von der Grammatik des Relativsatzes ab und entwickelt sich zu einer eigenständigen Konstruktion. Dazu ist es erforderlich, daß die Konstruktion von (47b) systematisch auf den einfachen Satz (47a) bezogen wird. (47) (a) Id suspicabar. „Das vermutete ich.“ LAT (b) Id est quod es ist [REL:AKK.SG.N suspicabar. vermutete:ich] „Das ist es, was ich vermutete.“ (Pl. Men. 774) Wo, wie in (47), die Fokuskonstituente im einfachen Satz dieselbe Form hat wie in dem komplexen Satz, kann der letztere reanalysiert werden. Die Fokuskonstituente verdankt ihre Form dann nicht mehr der Tatsache, daß sie Prädikatsnomen ist, sondern ihrer Satzgliedfunktion im extrafokalen Satz. Dieser Schritt findet allerdings erst in roma-

XX. Die Erforschung des Syntaxwandels III: Phänomene

nischer Zeit statt; Beispiele von der Struktur von (48) sind im Französischen erst seit dem 13. Jahrhundert belegt. Nunmehr braucht die syntaktische Funktion der Leerstelle im extrafokalen Satz dort nicht mehr bezeichnet zu werden, da sie an der Fokuskonstituente selbst ersichtlich ist. Dies ist dasselbe Entwicklungsstadium, auf dem die Flexion des Relativpronomens ohnehin fast vollständig verloren geht (vgl. 3.2.1.). Der allgemeine Subordinator quelche, der in den romanischen Sprachen Relativsätze nur in nichtstandardsprachlichen Soziolekten einleitet, setzt sich als Subordinator des Spaltsatzes weitgehend durch. Im Ergebnis hat man den fertigen Spaltsatz (48) als französische Entsprechung zu (46b). (48) C’est en Scythie que je suis envoyé. Die Ausbreitung des Spaltsatzes als allgemeine Fokuskonstruktion in einer Sprache bedeutet eine Verdrängung des einfachen Satzes mit fokussierter Konstituente in Initialposition. Diese Konstruktion ist am zähesten in der Satzgliedfrage etabliert. Obwohl Satzgliedfragen zu den am häufigsten und frühesten bezeugten lateinischen Spaltsätzen gehören, lassen alle romanischen Sprachen noch Satzgliedfragen in der Form eines einfachen Satzes mit dem Fragepronomen in Initialposition zu. Sie gehen verschieden weit in dem Ausmaß, in dem eine Satzgliedfrage die Form eines Spaltsatzes annimmt. (49) PORT (50)

(a) Aonde te mandam? (b) Aonde é que te mandam? Où est-ce que tu es envoyé?

So kann die Frage „Wohin schickt man dich?“ im Portugiesischen entweder — wie im Deutschen — in Form eines einfachen Satzes mit dem Fragepronomen in Initialstellung, wie in (49a), oder als Spaltsatz, wie in (b), realisiert werden. Im gesprochenen Französischen dagegen ist allein die zweite Möglichkeit, wie in (50), üblich. Die Fixierung der Satzgliedstellung erfaßt auch den Fokussatz selbst. Die Stellung des Fokus wird nach zwei widerstreitenden Prinzipien geregelt. Das erste besagt, daß das Komplement der Kopula, einschließlich des Prädikatsnomens, dieser zu folgen hat. Das zweite besagt, daß der Fokus am Satzanfang zu stehen hat. Das erste Prinzip gilt im Französischen in allen nicht-interrogativen Spaltsätzen wie z. B. (43), (48). Das zweite setzt sich meist in der Satzgliedfrage durch, erlaubt im Portugiesischen aber auch z. B. die

1213

Ersetzung des Fragepronomens in (49b) durch ein nicht-interrogatives Adverbial wie a Escítia „nach Skythien“ . Wichtig für den gegebenen Zusammenhang ist, daß in solchen Fixierungen Eigenschaften des Spaltsatzes zu sehen sind, die ihn als eigenständige Konstruktion etablieren und also auch von der Relativkonstruktion absondern. Zu diesen Eigenschaften gehört auch die Möglichkeit, ein Gerundial im Fokus des Spaltsatzes zu haben, wie in (43), das weder als Bezugsnominal eines Relativsatzes noch als Prädikat eines gewöhnlichen Kopulasatzes infrage kommt. Dazu gehört drittens die zuvor erwähnte Subordination des extrafokalen Satzes, die, von bestimmten Ausnahmen abgesehen, nicht vom Relativpronomen, sondern vom allgemeinen Subordinator übernommen wird. Diese Fallstudie zeigt also, wie der Relativsatz als Ausgangspunkt für die Entwicklung einer funktionell ähnlichen Konstruktion genommen werden kann, die in dem Maße, in dem sie grammatikalisiert wird, sich von ihrer Basis fortentwickelt und somit die Syntax der Sprache um eine neue Konstruktion bereichert. 3.3.2. Vom Relativsatz zum Komplementsatz Während ein Relativsatz mit externem Nukleus sich von einem Komplementsatz durch seine offene Leerstelle unterscheidet, hat ein Relativsatz mit internem Nukleus dieselbe Struktur wie ein Komplementsatz. Wenn der Kontext den Unterschied zwischen einer Gegenstands- und einer Sachverhaltsbezeichnung nicht disambiguiert, so kann es, auf der Basis der in Abschnitt 3.3. konstatierten funktionellen Ähnlichkeit, zu einer Strukturambiguität zwischen einer Interpretation als Komplementsatz und als Relativsatz kommen, wie sie die alternativen Übersetzungen von (51) verdeutlichen. Solche Strukturambiguitäten finden sich in mehreren Sprachen mit zirkumnominalem Relativsatz, unter anderem im Navaho und Diegueño. (51) hastiin kinłání-d’ N [Mann Flagstaff-ABL AV yóó’anáálwod-ígíí b-aa áhonisin. weg:PRF:3:lauf-NR] 3-bzgll:wiss „Ich weiß, daß der Mann von Flagstaff weglief / von dem Mann, der von Flagstaff weglief.“ Systematische Beziehungen zum Komplementsatz geht jedoch auch der Relativsatz mit externem Nukleus ein. Wahrnehmungs-

verben nehmen als direktes Objekt i. a. ebensowohl Bezeichnungen für Gegenstände wie für Sachverhalte. Der wahrgenommene Sachverhalt kann konzipiert werden als an einem Gegenstand auftretend. In syntaktisch akkusativischen Sprachen wird dieser vom Subjekt des Komplementsatzes bezeichnet. Daher haben viele Sprachen als sententiales Objekt zu Wahrnehmungsverben eine AcI-Konstruktion, wie in (52). eum fumare. (52) Vidi LAT sah:ich ihn rauch:INF „Ich sah, daß er rauchte. / Ich sah ihn rauchen.“ Der AcI wird in der Entwicklung vom Latein zu den romanischen Sprachen eingeschränkt. Im Gegensatz zum Latein tritt er überhaupt nur noch nach Wahrnehmungsverben auf ((53b)) und steht auch dort in Opposition zum finiten Komplementsatz ((53a)). (53) (a) J’ai vu qu’il fumait. FRANZ „Ich habe gesehen, daß er rauchte.“ (b) Je l’ai vu fumer. „Ich habe ihn rauchen sehen.“ (c) Je l’ai vu qui fumait. „Ich habe ihn gesehen, wie er rauchte.“ Als Alternative zum AcI hat sich im Französischen der prädikativeRelativsatz herausgebildet, wie in (53c) (vgl. van der Auwera 1985). Die Konstruktion besteht auch in so verschiedenen Sprachen wie dem Rumänischen, Serbokroatischen und Dyirbal. Sie ist weder als restriktive noch als appositive Relativkonstruktion analysierbar, denn im Französischen kann ein klitisches Pronomen wie in (53c) nicht Bezugsnomen einer solchen Relativkonstruktion sein. Daß prädikative Relativkonstruktionen für den Kontext eher die Funktion eines Komplementsatzes als die einer Relativkonstruktion haben, beweisen auch die Regeln der Katapher: auf eine prädikative Relativkonstruktion wie in (54b) verweist man nicht mit dem Personalpronomen, wie auf die Relativkonstruktion in (54a), sondern mit dem sächlichen Pronomen ça, wie in (54c). (54) (a) Tout le monde la déteste, la femme FRANZ qui habite au rez-de-chaussée. „Jedermann verabscheut sie, die Frau, die im Erdgeschoß wohnt.“ (b) Je n’ai jamais vu ça, Marie qui joue du piano.

XX. Die Erforschung des Syntaxwandels III: Phänomene

1214

„Das habe ich nie gesehen, daß Maria Klavier spielt.“ (c) Vous ne m’avez jamais dit ça, que vous êtes communiste. „Sie haben mir das nie gesagt, daß Sie Kommunist sind.“ Prädikative Relativsätze kommen in einer Reihe von Kontexten vor, die thetische vor kategorischen Prädikationen bevorzugen (s. Sasse 1987, bes. Abschnitt 3.5.). Das gilt für Wahrnehmungs- und Entdeckungsverben wie in (53), für Präsentationen wie in (55) und für Existenzprädikationen wie in (56) f (a) Il est là-bas qui arrose. (55) FRANZ (b) Le voilà qui arrose. „Er ist dort am Blumengießen.“ Il y a mon frère qui travaille. (56) FRANZ „Mein Bruder ist da am arbeiten.“ Il a les cheveux qui tombent. (57) FRANZ „Ihm fallen die Haare aus.“ Thetische Prädikationen können nicht negiert oder modalisiert sein und beziehen ihr Tempus vom Kontext; dasselbe gilt für prädikative Relativsätze. Eine thetische Prädikation hebt nicht darauf ab, daß auf ein gegebenes Subjekt ein bestimmtes Prädikat zutrifft, sondern präsentiert das Subjekt als Träger des Prädikats. Daher ist die relativierte Funktion im prädikativen Relativsatz die des Subjekts. Auch hierin ähnelt die prädikative Relativkonstruktion dem AcI. Die prädikative Relativkonstruktion hat sich in der französischen Sprachgeschichte aus der restriktiven entwickelt. Diese stand zunächst als Variante zum AcI zur Verfügung, wenn das Subjekt indefinit war, wie in (58a/b). (a) J’ai vu un policier fumer. (58) FRANZ (b) J’ai vu un policier qui fumait. „Ich habe einen Polizisten rauchen sehen.“ (a) J’ai vu le policier fumer. (59) FRANZ (b) J’ai vu le policier qui fumait. „Ich habe den Polizisten rauchen sehen.“ Auf der Basis der Analogie (58a) : (b) = (59a) : (b) (vgl. Art. 70) wurde der prädikative Relativsatz dann auch zu einem determinierten Bezugsnomen, wie in (59b) gebildet. 3.3.3. Vom Relativsatz zum einfachen Attribut In der Geschichte des Persischen wird eine Relativkonstruktion zur einfachen Attributkonstruktion grammatikalisiert (Haider &

Zwanziger 1984). In der folgenden Darstellung vertritt das Avestische (vgl. Seiler 196 0, Teil II) das nahe verwandte Altpersische, für das keine analysierten Daten vorliegen. Das Avestische hat, wie andere indogermanische Sprachen auch, ein Relativpronomen ya- demonstrativen Ursprungs geerbt (vgl. Benveniste 1957). Damit werden u. a. postnominale Relativsätze wie in (60) gebildet. dma (60) bvat AVESTgeschieht Wesen:GEN.PL ašava rechtschaffen:GEN.PL yaoždāθrǝm Läuterung:NOM.SG hǝnti y [REL:NOM.PL sind mainyuš spǝntahe heilig:GEN.SG Geist:GEN.SG] „es geschieht die Läuterung der rechtschaffenen Wesen, die des heiligen Geistes sind“ (Yt. 6, 2) yazamaide ... miθrǝm ... (61) AVESTMithra:AKK.SG verehren:wir āsunm yō [REL:NOM.SG schnell:GEN.PL āsuš schnell:NOM.SG] „den Mithra verehren wir, der unter den Schnellen schnell ist“ (Yt. 10, 65) Eine häufige Variante sind nominale Relativsätze, wie in (6 1). Sie unterscheiden sich von verbalen nur durch das Fehlen der optionalen Kopula. Sie bestehen neben einfachen nominalen Attributen, die unmittelbar pränominal stehen. Im Vergleich zu diesen hat ein nominaler Relativsatz mehr Eigengewicht. Während der verbale Relativsatz im Mittelpersischen mithilfe des Pronomens ki erneuert wird, das mit dem hethitischen und lateinischen Relativpronomen etymologisch verwandt ist, wird die nominale Relativkonstruktion weiter grammatikalisiert. Bereits im Avestischen können Relativpronomen und nominales Prädikat mit dem Bezugsnomen im Kasus kongruieren, wie in (62). yat miθrǝm (62) āat A damals SR Mithra:AKK.SG VEST yim vouru-gaoyaoitīm [REL:AKK.SG weit-Trift:AKK.SG] frādaðm azǝm erschuf ich „als ich den Mithra, den mit den weiten Triften, erschuf“ (Yt. 10, 1) In diesem Falle steht der sonst ziemlich bewegliche Relativsatz postnominal und das

65.  Relativsätze

Relativum an seinem Anfang. Dieses wird hier zur „Gelenkpartikel“ , zum bloßen Attributor. Seine Flexion wird somit überflüssig. Bereits zu jungavestischer Zeit erscheint statt der konkreten Kasus des Relativums die neutrale Form yat, die schon als allgemeiner Subordinator dient (vgl. (6 2)). Sie schließt Adjektivattribute, wie in (6 3), und Genitivattribute an. aŋhvō yat (63) ahmi AVESTdies:LOK.SG Leben:LOK.SG [SR astvainti leiblich:LOK.SG] „in diesem irdischen Leben“ (V. 5, 39) Dieser Subordinator wird im weiteren auf eine enklitische Partikel ya, in der persischen Grammatik Ezāfe (auch Izāfat) genannt, reduziert. Durch Ezäfe angeschlossene postnominale Genitiv- und Adjektivattribute beginnen im Mittelpersischen, die ererbten pränominalen Attribute zu verdrängen. In dem Maße, in dem die Deklination reduziert wird und attributive Verhältnisse nicht mehr durch den Genitiv bzw. durch Kongruenz bezeichnet werden, wird das postnominale, mit Ezāfe eingeleitete Attribut zum einzigen Attribut neben dem erneuerten Relativsatz. garm (64) (a) āb-i PERS Wasser-AT warm „warmes Wasser“ (b) kitāb-i Hasan Buch-AT Hassan „Hassans Buch“ (c) ru-ye u Gesicht-AT er „sein Gesicht“ (d) mughic-i raft-an Zeit-AT geh-INF „Zeit zu gehen“ Die neupersischen Beispiele in (6 4) zeigen verschiedenartige Attribute, in denen die Ezāfe obligatorisch ist. Durch Ezāfe plus Personalpronomen wird, wie man an (6 4c) sieht, sogar das Possessivpronomen ersetzt. Hier ist also die Grammatikalisierung dieser attributiven Konstruktion vollendet.

4.

Zusammenfassung

Die Relativkonstruktion hat funktionelle Ähnlichkeiten zu drei Arten von Konstruktionen: Erstens mit solchen, die weniger stark grammatikalisiert sind, also insbesondere Satzreihen. Zweitens mit solchen, die sich im Grammatikalitätsgrad nicht von ihr unterscheiden, also insbesondere anderen Nebensätzen. Drittens mit stärker grammatikalisierten, insbesondere einfachen attributiven Konstruktionen. Zu allen solchen Konstruk-

1215

tionen unterhält die Relativkonstruktion synchrone und diachrone Beziehungen. Sie kann prinzipiell auf drei Arten entstehen, durch Grammatikalisierung einer Satzreihe, durch Reanalyse einer funktional und strukturell ähnlichen Konstruktion und durch analogische Expansion einer attributiven Konstruktion. Ferner kann ein einmal vorhandener Relativsatz durch eine neue Konstruktion ersetzt werden, wobei die alte Strategie erneuert oder eine neue eingeführt werden kann. Ebenso wie die Relativkonstruktion sich aus funktionell ähnlichen Konstruktionen entwickeln kann, können schließlich auch solche sich aus der Relativkonstruktion entwickeln. Der allgemeinste Nenner, auf den solche Vorgänge sich bringen lassen, sind die Mechanismen des grammatischen Wandels, nämlich Grammatikalisierung und analogischer Wandel. Die erstere ist gerichtet; es führt also kein Grammatikalisierungsprozeß von der Relativkonstruktion zur Satzreihe oder vom einfachen Attribut zum Relativsatz. Die Analogie dagegen operiert in beiden Richtungen. Daher kann man sowohl, wie in Abschnitt 3.1.3. gezeigt wurde, einfache Attribute zu Relativsätzen expandieren, als auch, wie in 3.3.3. gezeigt wurde, Relativsätze zu einfachen Attributen reduzieren. Im letzteren Fall wirken Grammatikalisierung und Analogie zusammen.

5.

Literatur

Benveniste, Emile. 1957. La phrase relative, problème de syntaxe générale. Bulletin de la Société de Linguistique de Paris 53. 39—53. Chafe, Wallace L. 1976 . Givenness, contrastiveness, definiteness, subjects, topics, and point of view. Subject and topic, ed. by Charles N. Li, 25—55. New York etc. Guiraud, Pierre. 196 6 . Le système du relatif en français populaire. Langages 3. 40—48. Haider, Hubert, and Roland Zwanzig er. 1984. Relatively attributive: the ‘ezāfe’-construction from Old Iranian to Modern Persian. Historical syntax, ed. by Jacek Fisiak, 137—172. (Trends in Linguistics, 23). Berlin etc. Haudry, Jean. 1973. Parataxe, hypotaxe et correlation dans la phrase latine. Bulletin de la Société de Linguistique de Paris 68 (1). 147—186. Johansen, Hol g er. 1935. Zur Entwicklungsgeschichte der altgermanischen Relativsatzkonstruktion. Kopenhagen. Keenan, Edward L., and Bernard Comrie. 1977. Noun phrase accessibility and universal grammar. Linguistic Inquiry 8. 63—99.

XX. Die Erforschung des Syntaxwandels III: Phänomene

1216

Kemmer, Suzanne E. 1984. From comparative to relativizer: the case of Icelandic sem. Proceedings of the Annual Meeting of the Berkeley Linguistics Society 10. 296—306. Kurzová, Helena. 1981. Der Relativsatz in den indoeuropäischen Sprachen. Hamburg. Lehmann, Christian. 1979. Der Relativsatz vom Indogermanischen bis zum Italienischen. Eine Etüde in diachroner syntaktischer Typologie. Die Sprache 25. 1—25. Lehmann, Christian. 1984. Der Relativsatz. Typologie seiner Strukturen — Theorie seiner Funktionen — Kompendium seiner Grammatik. (LUS, 2). Tübingen. Löfstedt, Beng t. 196 6 . Die Konstruktion c’est lui qui l’a fait im Lateinischen. Indogermanische Forschungen 71. 253—277. Mithun, Marianne. 1984. How to avoid subordination. Annual Meeting of the Berkeley Linguistics

Society 10. 493—509. Romaine, Suzanne. 1992. The evolution of complexity in a creole language. Acquisition of relative clauses in Tok Pisin. Studies in Language 16 . 139—182. Sasse, Hans-Jürg en. 1987. The thetic/categorical distinction revisited. Linguistics 25. 511—580. Seiler, Hansjakob. 196 0. Relativsatz, Attribut und Apposition. Wiesbaden. Touratier, Christian. 1980. La relative. Essai de théorie syntaxique (à partir de faits latins, français, allemands, anglais, grecs, hébreux, etc.) (Collection Linguistique, 72). Paris. van der Auwera, Johan. 1985. The predicative relatives of French perception verbs. Predicates and terms in functional grammar, ed. by A. Machtelt Bolkestein et al., 219—272. (Functional Grammar Series, 2). Dordrecht/Cinnaminson.

Christian Lehmann, Bielefeld (Deutschland)

66. Word Order 1. 2. 3. 4. 5. 6. 7. 8. 9. 10.

Introductory Remarks Unidirectional Serialization Causes of Word Order Change Causes and Mechanisms of Change Phonetic Erosion Local Nature of Syntactic Change External Influences Linguistic Reconstruction Conclusion References

1.

Introductory Remarks

Much though by no means all recent work in the field of word order change derives from the enhanced interest in word order patterns themselves that is inherent in the approach to linguistic typology most closely associated with the name of Joseph Greenberg, and subsequently developed in a variety of ways by numerous scholars as discussed elsewhere in this volume. Once it is accepted that there are word order configurations which co-occur with more than chance frequency — and the detailed reasons for such co-occurrences, which are far too widely documented simply to be discounted, lie outside the scope of this article — then it follows that a change in one of the relevant word order characteristics may well have, or tend to have, a consequen-

tial effect on one or more other characteristics also. Twin aims of the present article are to see whether such a claim can be empirically demonstrated and, to the extent that it can be, to suggest explanations for what has been observed.

2.

Unidirectional Serialization

The principal claims made within Greenbergian word order typology centre around the relative order of a finite verb and its complement, which tends to correlate with a number of other features, as noted below. I say “centre around” because there has been much discussion in the literature as to whether a VO/OV order is to be taken as definitional of a particular type of language rather than merely a convenient label: the general conclusion is that it is no more central than any one other single feature (cf. Harris 1984, 182 and the references cited there, and also the allusion to Mixe-Zoquean languages in section 7.). What has eventually emerged as the most sustainable position within this perspective is well summarised in Vennemann (1983): “Everything else being equal, having unidirectional serialization is preferred to not having unidirectional serialization.” What this means, of course, is that it is the unmarked situation in a language for

66.  Word Order

categories and their respective modifiers to appear in a consistent order vis à vis each other, because, in the words of Hawkins (1982, 9), “grammars with more cross-categorial generalisations will be simpler than, and hence preferred over, those with fewer.” Thus in a VO language, for example, there will be a strong preference for prepositions, the relationship between V and O being parallel to that between a preposition and an associated nominal: conversely, postpositions are more likely in OV languages. Equally, for the same reason, main verbs will tend to precede auxiliaries in OV languages and follow them in VO languages. The synchronic patterns which are found were first tabulated and labelled by Bartsch/Vennemann (1972) and Winfred Lehmann (1973), while the initial insight has been substantially refined in later work by John Hawkins, whose principle of Cross-Category Harmony “makes the explicit prediction that the more a language type departs from the most harmonic ordering, the fewer exemplifying languages there will be” (Hawkins 1983, 134). (See article 6 3 for further discussion of these principles.)

3.

Causes of Word Order Change

3.1. Preference for Consistent Word Order The implications of such a position for the study of word order change seemed at first to be both clear and straightforward: consistency, in the sense just described, was a goal to which all languages aspired, and all inconsistency was to be attributed to the fact that a given language was “in transition” between one favoured configuration and another. (The most influential proponent of this view is Winfred Lehmann, e. g. 1973, 1974.) There are, however, two fundamental objections to so simple a view. Firstly, why should languages remain inconsistent, as some undoubtedly do, over very long periods and/or begin to change when they are, or are close to, word order consistency? Secondly, to talk of the goal-orientation of “languages” only makes sense if it is short-hand for “languageusers”; as soon as this is conceded, as it must be, then one has to acknowledge that any (unconscious) pressure on speakers towards unidirectional serialization may be (wholly or partly) offset by any one (or more) of the myriad of other pressures operating on speakers who have grammars with their own clear internalized structures and who operate

1217

in real social contexts with real communicative needs. Notice that the “quantifiable preference” approach of Hawkins is inherently more equipped to take this on board than earlier more absolute claims such as those made by Lehmann. It is an examination of word order change in the light of these conflicting pressures which forms the core of what follows. 3.2. Drift and Unconscious Rationality That there is a tendency for word order change to move in particular directions, noone seriously doubts. The insight dates back at least to Edward Sapir, who wrote (1921; 196 3, 150): “Language moves down time in a current of its own making. It has a drift”, continuing later (154/5): “The drift of a language is constituted by the unconscious selection on the part of its speakers of those individual variations which are cumulative in some special direction”. One notes the modernity of the formulation which, I shall argue, remains in essence the bare bones of the claim which later scholars have been seeking, consciously or otherwise, to flesh out. It will be observed that this view is the antithesis of that advanced by Ferdinand de Saussure, namely that “diachronic events are always accidental and particular” (1959, 93; italics added). Preference for undirectional serialization in the domain of word order may be seen as one particular instance of drift and it is accordingly for the latter more general concept that we must most urgently seek to account. And here, it seems to me, we do have an explanation, one which derives from the cognitive structure of individual human beings and their tendency to impose order on the data of experience. (Recall in this context Vennemann’s (1983) observation that “we can arrive at explanations for the regularities within a certain domain by turning to theories that are not theories for that particular domain” cf. also Harris (1983).) Speakers prefer to order the major constituents of their language “consistently”, that is, in accordance with what Itkonen (1978, 45; 1983, 185 ff) calls “unconscious rationality”; indeed, Itkonen specifically points out that “it is ... quite probable ... that the internalizations of rules exhibit — at various levels — a means-end structure of their own, and must therefore be subsumed under the concept of unconscious rationality.” If this view is accepted, then we do have at least the outline of a cognitive explanation for those instances

1218

of word order change which do increase synchronic word order consistency of the type under discussion here. But we have already observed, in connection with Lehmann’s claims over a decade ago, that this can only be part of the answer. “Preference for consistency” is a non-nomic explanation (in the sense of Lass (1980)); we are, of course, in the realm only of “dominant tendencies” rather than “universal laws” (Hagège 1982, 10) and hence not in a position to make (precise) predictions. We are faced in fact with constant interplay between one major characteristic of human cognitive structure which we have just described, and a whole variety of other factors which need to be examined in their turn, and which may account for exceptions to the expected patterns or indeed provoke a change within what was hitherto an apparently stable and “consistent” situation. (It is after all the case both that there are apparently no exceptionless patterns in the domain of word order change, and also that unexpected changes do occur quite frequently.) Our brief must be to identify some of these factors, and to show how they might explain particular “unexpected” developments. From all of this it follows that the theoretical apparatus we develop can never be predictive in any absolute sense, but can merely indicate what is statistically most likely. It may however be able to indicate retrospectively with some plausibility what factor or factors caused a departure from a typologically optimal pattern in the case of a particular attested change in a given language. 3.3. Cognitive, Functional and Historical Causes What other factors, then, may over-ride speakers’ preference for consistency? Essentially, they may be of three kinds. Firstly, they may lie within the same broad domain as the “preference for unidirectional serialization”, that is, they may reflect (variant or invariant) aspects of human cognitive structure or at least of the “language module” in the sense of Fodor (1983), Chomsky (1986 ). Secondly, they may be termed “functional”, reflecting the communicative needs of speakers and their responses to such difficulties as they may encounter. Thirdly, they may be historical, attributable to circumstances wholly external to the individual language user: changes due to prolonged periods of language contact are a clear example of what we

XX. Die Erforschung des Syntaxwandels III: Phänomene

have in mind here. The first two of these factors are presented rather differently in Hawkins (forthcoming, esp. 1—2 and n1) where he talks of [explanatory] principles drawn from the domains of competence and of performance, the former relating exclusively to features of a “genetically transmitted human linguistic endowment”, and the latter either to “certain innate predispositions” in humans, or to “the various functions that humans use language for”. Even the competence:performance distinction is not absolute, since “it is not inconceivable that the human language faculty encodes grammatical universals that have evolved, in part at least, in response to performance (particularly processing) considerations” and there may thus be no clear-cut distinction between those parts of the human endowment “unique to the processing of language” and others “which may hold more generally across other cognitive and perceptual domains.” (Itkonen’s “unconscious rationality” springs at once to mind in this connection.) Various suggestions that have been made in each of these areas will now be examined in turn. 3.4. Word Order Change and Universal Grammar The first serious claim that syntactic change may be due to features of our linguistic competence in the very narrow sense defined above is due to David Lightfoot (1979, 1981a), who argues convincingly that intolerable derivational opacity, howsoever this may arise, will in due course lead to the restructuring of the relevant segment of a language, the adaptive rules eventually becoming too complex and abductive change in the sense of Andersen (1973) then taking place. One brief (hypothetical) example, due originally to Jespersen, must suffice: the Old English OVS structure “ðam cynge [dat.] licodon peran” lit. “to the king were pleasing pears” is said to have been reinterpreted as an SVO structure “the king liked pears” (with a consequential change in the meaning of “like”) because of the intolerable opacity of the earlier pattern once SVO had become overwhelmingly the most common word order and once case endings had been lost. The new dimension to Lightfoot’s analysis of this familiar example is his claim that derivation of the original meaning from the original structure given the relevant change in word order became impossible for a highly specific reason,

66.  Word Order

that it would violate the Trace Erasure Principle (Dresher/Hornstein 1979, Friedin 1978), such a violation being impossible because of the genetically determined characteristics of Universal Grammar. (“Intolerable opacity” of course has a (nomic) explanatory value only if formally characterizable principles of this type can be specified.) It should be noted that apparent counter-arguments to claims of this type are at least as likely to be due to the inadequacy of the empirical data (as strongly suggested in this case by Allen 1986 ) or of current formulations of particular principles as to the possibility that linguistic competence has no role to play in language change. The most recent work in this vein adduces the principle of parametric variation to account for both current divergence (the reflection of earlier reanalyses) and ongoing change. Particular attention has been focused, for instance, on word-order in the Germanic languages, especially the widespread preference for V/2 (i. e. verb second) order in languages other than English and its abandonment in the latter. If one considers for example such proposals as those advanced by Haider (1985) (without in any way entering into a detailed evaluation of the version of GB theory on which he relies), one can readily accept his view (1985, 72—3, citing Lenerz 1984) that V/2 originally came about in Germanic by means of the reanalysis of an initial Topic/Focus slot frequently filled in Old High German by a verb as being in fact the INFL position, releasing the T/F slot to be filled by other material, and hence creating a V/2 language. (For further discussion of the analogous Topic/Subject cycle, see below, section 3.6 .) Haider presents this reanalysis thus:

1219

ceptional property” of being V/2, there is then the question of why English develops further. Here, the well-attested development of the special class of English modals (cf. Lightfoot 1979) appears to be the determinant factor. Leaving aside the vexed question of exactly how, when and why this category emerged (for a recent detailed discussion, cf. Goossens 1987) it appears that the development of this syntactically unique class of verbs triggered, for quite specific reasons, the loss of the V/2 property in English in contrast to the other Germanic languages, where modals to a much greater extent “still function as plain verbs”. (Haider notes wryly that “the paradigm language of linguistic research turns out to be the exceptional Germanic language.”) What matters is not the detailed formulation, which is clearly still highly contentious: indeed Platzack (1986 ), reviewing essentially the same material, comes to markedly different conclusions, although agreeing that the creation of a special class of modals, which cause lexical material to be generated under the node INFL (Platzack 1986 , 192), is crucially relevant to the loss of the V/2 pattern in English. What does matter is that parametric variation within either or both of two categories, INFL and COMP, may apparently account for specific instances of word order divergence and change; as Lightfoot (1981a, 237—238) puts it, such changes “will arise for reasons of necessity and not by chance, in short, they will be explained.” It remains important, however, to remember that unless the cause of the initial variation, the “small differences in presented evidence” which “could lead to the growth of very different systems” (Chomsky 1981, 42), is clearly and independently attested, then one is merely describing, in however sophisticated a manner, rather than explaining why, the “presented evidence” came to be different in the first place. Haider’s proposals in respect of the initial V/2 order seem to come close to meeting this condition, though of course in a wholly non-nomic fashion. (We examine briefly below (section 7.) a quite different way in which a V/2 pattern may enter a language.) 3.5. Processing Factors

Having thus accounted for what Haider calls (1985, 72) Germanic’s “typological ex-

What causes of word order change might be attributable to aspects of human cognitive structure other than our innate linguistic endowment? Apart from the tendency towards unidirectional serialization itself, there is, for

XX. Die Erforschung des Syntaxwandels III: Phänomene

1220

instance, the widely discussed difficulty of processing centre-embedded material (cf. Kuno 1974, Dryer 1980). The general tendency to avoid centre-embedding cannot be due to an inherent feature of Universal Grammar innate in all speakers since if it were, as Hawkins (forthcoming) points out, it would be exceptionless, which it isn’t (cf. Japanese). Vincent (1976 ) describes one particular instance in which the conflict (itself an example of cross-categorial disharmony) between a basic SOV order in Latin and postnominal relatives can be, and is, most readily resolved by Heavy NP shift, this producing in many cases an SVO sentence — accentuating a drift towards a new consistency of serialization — and also moving rightwards a complex or ‘heavy’ constituent, a frequent (but again by no means ubiquitous) tendency in many languages (Hawkins 1983, 92). (Consider in this connection the preference for post verbal sentential complements even in verb-final languages, section 6 .) We may note also, briefly, that since in the case of VO languages with post-nominal relatives all the relevant pressures are operating in the same direction, while in the case of OV languages with pre-nominal relatives, other factors just noted vie with Cross-Category Harmony, then it is reasonable to predict that the basic word order of a language may not become VO unless its relative clauses are postnominal, whereas a change to OV does not necessarily involve prenominal relatives. The virtually total absence on VO/Rel N languages in Table 2 of Hawkins (forthcoming) precisely bears out what he aptly calls the “left-right asymmetry” of languages, resulting from the interaction of the grammatical principle of serialization and the performance notion of language processing in real time. 3.6. The Subject-Topic Cycle There are also functional reasons for language change, reflecting the use made of languages in communicative situations. Perhaps the best-known of these involves the subjecttopic cycle, first fully described by Li/Thompson (1975). Given that “subject” is a syntactic notion and that “topic” is a discourse notion, there is no necessary reason why the organisational principles which operate in the two domains should always point in the same direction. Specifically, “there is an iconic relation between the primary role of the topic as establishing the focus of the discourse and its sentence-initial position” (Vincent 1975, 5), a

situation which causes absolutely no problems when the topic is encoded as the subject in a subject-initial language. In all other circumstances, however, there may well be a conflict between purely syntactic rules — as Hawkins (forthcoming) puts it, “cross-categorial rules of head placement ... have nothing to do with pragmatics” — and the preferred way of organising discourse, summarised as “attend first to the most urgent task” by Givón (1983, 20) and recently labelled “The Theme First Principle” by Tomlin (1986 ). When the topic is not the subject it will very often tend to be fronted; at the same time, the criteria influencing the choice of topic will often also be directly relevant to the choice of subject (cf. Allerton 1980 for a justification of this observation). Subject-initial languages therefore often allow the fronting of (non-subject) topics (cf. the discussion of verb-fronting in Old High German, section 3.4.), with consequential loss of syntactic consistency. Conversely, nominal topics often become subjects: as Harris (1984, 190) puts it, “there is a marked tendency ... for the looser discourse constraints operating on the topic to give way to the much tighter grammatical constraints operating on (surface) subjects”, while Harris (1978) surveys both the emergence of topic-fronting in Old French (cf. section 7. below) and the subsequent passage of the standard language to subject-initial, with a consequential increase in unidirectional serialization. (Much recent work, e. g. Lambrecht 1981, suggests that certain registers of popular French once again favour initial topics.) It will be readily apparent that each of these stages may (but need not) provoke significant changes in word order, including the disturbance of the unmarked preference for unidirectional serialization. At any given moment when this preference is set aside, then pressure for its re-establishment will be one of the factors that may lead to a further change in due course. And again the conflict between competence and performance considerations is clear.

4.

Causes and Mechanisms of Change

4.1. Grammaticalization of Afterthoughts This discussion of what is in effect the grammaticalization of originally left-dislocated structures leads naturally to the consideration of what seems at first sight to be the

66.  Word Order

converse, namely the absorption of right-dislocated nominals within the sentential core, with a consequential change in the unmarked or basic word order. The grammaticalization of afterthoughts, first explicitly discussed in Hyman (1975), arises if and when structures such as, for instance, French ildéteste son frère, Paul (SVO, S) (where Paul is referred to as an “antitopic” in Lambrecht 1981) comes to be reinterpreted as il-déteste (V) son frère (O) Paul (S), new structures then being needed to fulfil the original discourse function. There is, however, one important difference between the grammaticalization of leftand right-dislocated nominals. Whereas there are clear pragmatic reasons why topics tend to appear initially and why afterthoughts appear finally, and while there is empirical evidence that nominals playing these roles tend to come to be grammaticalized within the sentences to which they were previously adjoined, there is no ovious functional reason to explain this development in the case of afterthoughts corresponding to the reasons for topic > subject assimilation noted in section 3.6 . The grammaticalization of afterthoughts is very much a mechanism of syntactic change; what the underlying reason for such a process might be remains to be determined, unless one is prepared simply to accept a general tendency for “pragmatic unmarking” as an explanation in itself. 4.2. SVO to SOV in Mandarin This seems an appropriate point to highlight very briefly the general importance of distinguishing at all times between “mechanisms” or “pathways” of change on the one hand and “explanations” for change on the other (cf. Harris 1984, 188). One very clear example arises in the classic discussion of the emergence of SOV order in Mandarin Chinese to be found in Li/Thompson (1974), where the process whereby the cange came about is clearly shown to be the development of prepositions from earlier verbs ba and bèi via a serial verb construction, thus: SVO > S V/prep O V > SOV (where O is in fact a PP). After a perceptive analysis leading to the statement (1974, 208) “we suggest that the presence of [other] OV properties may have had a catalytic effect, inducing the language to shift toward the OV word order”, Li/ Thompson (1974, 209) continue “a more significant and direct factor explaining (italics added) the SVO-SOV change is that the verb

1221

in an SVO language can develop into a case marker” It seems to me that their first observation is (at least potentially) causal, but that the second is not. Just as with the grammaticalization of afterthoughts, we don’t as yet have an explanation for the particular word order changes which are observed to occur, or at least not an explanation why. (For an elaboration of the important distinction between “explanations how” and “explanations why”, see Harris 1982, 12 ff.) Hawkins (1983, 244) puts it well when he says that the SOV (ie. S+PP+V) structure “appears to be a direct output of the change from one lexical category into another” but that only if such a pattern turns out empirically to be generalizable can causality be established, an outcome which seems unlikely in this instance.

5.

Phonetic Erosion

There is a further cause of word order change which has been much discussed in the literature, namely the effects of reductive phonological change. The most obvious situation where such an explanation is in principle plausible is when a language changes from SOV to SVO at more or less the same time as suffixed nominal case markers are lost, or substantially weakened, through phonetic erosion. At a relatively early stage in contemporary word order studies, when changes of this type had been disproportionately well analysed, for instance in Romance or English, the view was advanced that “phonological change [is] the ultimate cause of drift” (Vennemann 1975, 301). The most powerful counterweight to such a point of view is that there are many other types of word order change — for instance, topic-fronting or the passage from SVO to SOV (Li/Thompson 1974; cf. also section 4.2.) — for which an explanation based predominantly on phonological criteria is simply not available. This does not mean, however — pace Antinucci (1977, 133) — that phonetic erosion can never be a factor in word order change. It seems clear, for instance, that any reduction in the efficiency of a set of suffixes may favour the use of originally more marked alternative structures which, once pragmatic unmarking occurs, may well be “inconsistent” with other patterns of serialization within a given language, a situation in which various of the pressures operating on speakers which have been described in this paper may come into conflict.

XX. Die Erforschung des Syntaxwandels III: Phänomene

1222

The question now arises as to how to classify “phonetic erosion” in terms of the three types of causal factor outlined in section 3.3. The answer seems to be twofold. The explanation for the propensity towards phonetic erosion, part of the general tendency towards economy fully discussed from Meillet (1912/ 1958) to Haiman (1985, 157 ff), clearly lies in the cognitive domain. As Frei (1929/1971, 107) puts it at the start of his detailed analysis of the consequences for popular French of this particular tendency: “le besoin d’économie, bien qu’il puisse être tenu en échec par des influences opposées, est un facteur indéniable dans la vie du langage.” (Note this early reference to the multiplicity of relevant factors.) The explanation for this aspect of the principle of least effort clearly reflects one of the innate predispositions of human beings, albeit one which is presumably not part of the language-specific component of our endowment, i. e. a performance rather than a competence factor for Hawkins (cf. section 3.3.). Reactions to the effects of phonetic erosion, however, clearly have a functional motivation as speakers seek to minimize the consequences of information loss and maximise the effectiveness of their act of communication. The loss of case endings can, for instance, be made good by the rigidification of a previously flexible word order and/or by the introduction of adpositions; these latter will in turn tend strongly to be pre- or postposed in accordance with the demands of the principle of Cross Category Harmony.

6.

Local Nature of Syntactic Change

One further “language-internal” comment is in order at this point. Firstly, just as, in Nigel Vincent’s apt words (1978, 416 ), “sound change can only proceed remorselessly on, leaving the speaker to do the best he can to mend any pieces of language that get broken in the process” — consider the effect of the loss of final [-s] on the system of plural marking in (spoken) French, Harris (1978, 74 ff) — so too the essentially local nature of linguistic change can have the same effect, the solution to one problem leading to the creation of new difficulties elsewhere. Although Roger Lass (1980, 6 6 ) spoke scathingly of the “perpetual dialectic of teratogenesis and cosmetic surgery”, there does seem to be convincing evidence to this effect. To take just one example in the field of word order, consider the case

of sentential complements in verb-final languages. Processing considerations disfavour the interruption of main clause material (Bever 1970), this tending to lead to VO structures in such instances. Insofar as such structures remain marginal, there is no knock-on effect; but insofar as these VO structures provide a model for further changes in verb-object ordering, then uniformity may be reestablished in one area of the language at the price of serious cross-category disharmony elsewhere, with possible further changes in word order as a result. Indeed, almost all of the scenarios described earlier can be seen as ones in which a therapeutic, or at least functional, change in one part of the language may lead, directly or indirectly, to an enhanced likelihood of change elsewhere. Another process which demonstrates both the effects of phonological factors and the “local” nature of change is the move from synthetic to analytic structures perhaps best documented in the history of Romance but widely attested elsewhere also. As Vincent (1987) puts it in a recent study: “Given ... universal tendencies, and given the fact that whatever state it is in, a language will tend to change, it follows automatically that a predominantly ‘synthetic’ language ... can only stay as it is or engender ‘analytic’ structures. These in turn can either persist as separable periphrastic formations or ‘resynthesise’, and if the latter, they may do so at different rates.” The “universal tendencies” are of course for unstressed affixes to be lost, leading to new analytic syntagms and/or to changes in word order, and conversely for periphrases to come progressively to be treated as single phonological units. In each case the resultant or replacement structure may be disharmonic with the rest of the language in respect of its internal ordering or of its ordering vis-a-vis other constituents, and consequentially a potential stimulus for further change. The history of the “future tense” in Latin and Romance, analysed in detail in Fleischman (1982), well describes changes from a synthetic from (faciet) to a syntagm involving a post-posed auxiliary (facere habet) to a new synthetic form (fera) to a new analytic form with a pre-posed auxiliary (va faire), and the question of the inter-relationship between ordering of the auxiliary and of other constituents at various periods in the history of Romance has been much discussed.

66.  Word Order

7.

External Influences

There remains the question of external influence on a language, recently claimed by Neil Smith (1981, 52) to be the major cause of word order change. It seems beyond dispute that contact between one language and another can lead to change, including word order change, in either on both. A major chapter in Masica (1976 ) is devoted to the question of converging word orders within a linguistic area, as a result of the interaction of speakers of genetically unrelated languages and the consequential development of a Sprachbund, while a recent detailed analysis of the criteria for establishing the boundaries of such areas is to be found in Campbell et al. (1986 ). The authors demonstrate convincingly that “the isogloss demarcating nonverb-final languages is ... diagnostic of [the Meso-American linguistic area]” and discuss the possible loss of earlier verb-final orders in certain languages under the influence of other languages of the region (Campbell et al. 1986 , 548). Particularly interesting is the case of the Mixe-Zoquean languages, which show many features of verb-final languages (e. g. postpositional phrases), but where “the order of the main constituents — the verb and its Noun Phrase arguments — is not basically SOV” (Campbell et al. 1986 , 547). This is thus a clear apparent instance of areal pressures inducing categorial disharmony. As a second illustration, consider our earlier argument (section 3.6 .) that there are general functional reasons why languages might become topic-initial. It has been plausibly suggested that language contact may also bring about word order changes of this type, as in the case of Romansch (Haiman 1974) and French (Harris 1978), the two Romance languages most influenced by Germanic (itself topic-initial: cf. section 3.4.) and which were most clearly topic-prominent for a significant period of their history. (For an evaluation of this claim, cf. Harris 1984, 198.) As always, factors of various kinds may interact, as particularly with this last example; that wholly external factors arising from language contact may contribute to word order change, however, seems demonstrable beyond risk of contradiction.

8.

Linguistic Reconstruction

From all that has gone before, it should be apparent that attempts to reconstruct word

1223

order configurations for unattested earlier états de lang ue, given data for later periods and our knowledge of the causes and mechanisms of word order change, should be viewed with caution, not to say scepticism. Margaret Langdon puts the point aptly when, as early as 1977, she calls “disturbing” “the temptation to apply the tentative findings of word order typologies in a categorial manner to less empirically verifiable fields, such as historical reconstruction” (Langdon 1977, 255). The reasons for hesitating to use probabilistic models, with so many conflicting parameters, for a venture of this kind are particularly apparent in the reactions to Lehmann (1974), a major attempt to reconstruct the syntax (including various aspects of the word order) of Proto-Indo-European. Not only have his specific conclusions been attacked (e. g. by Friedrich 1975, Miller 1975) but the whole enterprise has been viewed as misconceived (e. g. by Watkins 1976 , Lightfoot 1979). Broadly speaking, one is forced to share this view; reliable reconstruction depends on exceptionless patterns, and almost everything we have said tends to militate against the occurrence of these in the domain of word order change. Even chapter 7 of Hawkins (1983) (“Language Universals and the Logic of Historical Reconstruction”), while certainly more rigorous than earlier attempts, can only lead us to “inductive inferences about parent language word order co-occurrences, with degrees of probability reflecting current language quantities” (Hawkins 1983, 279). There are many who think that such a goal is not really worth pursuing.

9.

Conclusion

Everything that has been said in this brief presentation points to the same conclusion: only a multi-factorial approach to the causes of word order change (as of grammatical change more generally) can even begin to handle the very complex issues in this fascinating field of study. There are scholars — of whom Roger Lass is perhaps the most prominent recent example — for whom the absence of nomic explanations for language change renders hopeless the whole task of searching for such explanations: “We haven’t a clue what (if anything) causes or constrains linguistic change” (Lass 1980, 75). But that simply isn’t true, as he implicitly admits by the modality of “linguistic change should ... in principle be no more predictable than change

1224

in art styles” (Lass 1980, 132). But word-order change is more predictable than fashion, and we do have (more than) a clue what causes and constrains it: a number of recurrent patterns of attested change have been examined and explained here, while Hawkins’ concept of cross-category harmony makes it quite clear (as in the example of relative clause ordering discussed in section 3.5.) that certain synchronic word order configurations appear to rule out, on principled grounds, particular changes of order, at least on the evidence of an ever-widening sample of attested languages. The fact that many of the statements we can make are probabilistic and hence have only a general rather than a specific predictive value should not lead us to despair — rather, it reflects the inevitable complexity of explanation and prediction within any social science. We have seen, of course, that there may indeed be nomic explanations in the domain of word order, to the extent that parametric variation of a particular specifiable kind within the framework of Universal Grammar can be shown to lead necessarily to re-ordering elsewhere in the language: the filling of INFL by the lexical material of the modal auxiliaries in English (section 3.4.) is said, for instance, to have led directly to the abandonment of V/2 ordering in that language, in clear opposition to the rest of the Germanic family. Explanations of this type, however, if they are in due course found to be sustainable, will never be able alone to account for the myriad of attested changes, given that, as Lightfoot (1981b, 36 2) rightly notes, “a grammar is just one of our cognitive capacities, interacting with perceptual mechanisms and other aspects of our mental make-up”, and that there are yet further functional and external pressures operating on speakers at any given time. John Haiman (1985, 8) makes a similar point when he says that “the impairment of iconicity” in languages — iconicity being seen as a desideratum akin, for instance, to unidirectional serialization — is due not to “totally novel and mysterious human genetic predispositions” but to “relatively familiar principles such as economy, generalization and association.” We have pointed here, as on many previous occasions, to the view that there most probably is an interaction between language-specific predispositions (or even absolutes) and more general aspects of human cognitive structure in causing changes of word order, and indeed language change more generally,

XX. Die Erforschung des Syntaxwandels III: Phänomene

so to that extent we would not make quite the same juxtaposition as Haiman — nevertheless, the point about many interacting principles is well-taken. Ironically, Lass (1980, 16 1) gradually retreats from the “puritanical” position outlined earlier to a more or less tolerant acceptance of what he calls “epistemological pluralism”, exactly what is being advocated here. Hawkins, in his most recent work (forthcoming), adopts as three of his working assumptions, that “we must adopt a more interactive approach to description and explanation whereby principles of one type are not pursued in isolation from others”, that “there is often a (partial) conflict between principles” and that “many causal principles are gradient rather than absolute”. All of these assumptions are crucially relevant to the successful understanding of word order change. Finally, it seems fitting to cite the last paragraph of Elizabeth Traugott’s “concluding remarks” to the 5th International Conference on Historical Linguistics (Traugott 1982, 46 6 ) which, while it has much wider import than the subject matter of this paper, can be taken to sum up the only defensible view on the causes of word order change. “The principle of imposing order on chaotic data which Harris (1982) proposes as an explanatory principle is of course a psychological one. Explanation of language change, it appears, may be possible only in so far as language change is ... embedded not only in the kind of socio-cultural setting I referred to earlier but also in a psychological setting. If such multi-faceted embedding can be achieved, the possibility of developing an explanatory theory of language change looks encouraging; but as in the case of most theories in the social, behavioural, and cognitive sciences, the predictive power that explanatory theories should ideally have (Lass 1980) will necessarily remain elusive.”

10. References Allen, Cynthia. 1986 . Reconsidering the history of like. Journal of Linguistics 22, 2. 375—409. Allerton, David. 1980. Grammatical subject as a psycho-linguistic category. Transactions of the Philological Society (1980). 40—61. Andersen, Henning . 1973. Abductive and deductive change. Language 49. 765—791. Antinucci, Francesco. 1977. Fondamenti di una tipologia del linguaggio. Bologna.

66.  Word Order

Bartsch, Renate, and Theo Vennemann. 1972. Semantic Structures. Frankfurt. Bever, Thomas. 1970. The Influence of Speech Performance on Linguistic Structures. Advances in Psycholinguistics, ed. by G. B. Flores d’Arcais & W. J. M. Levelt. Amsterdam. Campbell, Lyle; Terrence Kaufman; and Thomas Smith-Stark. 1986 . Meso-America as a Linguistic Area. Language 62, 3. 530—570. Chomsky, Noam. 1981. Principles and parameters in syntactic theory. Explanation in Linguistics: the logical problem of language acquisition, ed. by Norbert Horntein & David Lightfoot. Chomsky, Noam. 1986 . Knowledge of Language: its Nature, Origin and Use. New York. Dresher, Bezalel, and Norbert Hornstein. 1979. Trace Theory and NP Movement Rules. Linguistic Inquiry 10, 1. 65—82. Dryer, Matthew. 1980. The Positional Tendencies of Sentential Noun Phrases in Universal Grammar. Canadian Journal of Linguistics 25, 2. 123—195. Fleischman, Suzanne. 1982. The Future in Thought and Language: Diachronic Evidence from Romance. Cambridge. Fodor, Janet. 1983. The Modularity of Mind. Cambridge. Frei, Henri. 1929/1971. Le grammaire des fautes. Paris, Geneva. Friedin, Robert. 1978. Cylicity and the Theory of Grammar. Linguistic Inquiry 9, 4. 519—549. Friedrich, Paul. 1975. Proto-Indo-European Syntax. Journal of Indo-European Studies, Monograph I. Butte, Montana. Givón, Talmy. (ed.) 1983. Topic Continuity in Discourse: a Quantitative Cross-Language Study. Amsterdam. Goossens, Louis. 1987. The auxiliarization of the English modals: a functional view. The Historical development of auxiliaries, ed. by Martin Harris & Paolo Ramat. 111—143. Berlin. Hag èg e, Claude. 1982. La structure des langue. Paris. Haider, Hubert. 1985. V-second in German. Verb Second Phenomena in Germanic Languages, ed. by Hubert Haider & Martin Prinzhorn. 49—75. Dordrecht. Haiman, John. 1974. Targets and Syntactic Change. The Hague. —. 1985. Natural Syntax: Iconicity and Erosion. Cambridge. Harris, Martin. 1978. The Evolution of French Syntax: a comparative approach. London. —. 1982. On Explaining Language Change. Papers from the 5th International Conference on Historical Linguistics, ed. by Anders Ahlqvist, 1—14. Amsterdam.

1225

—. 1983. On the strengths and weaknesses of a typological approach to historical syntax. Historical Syntax, ed. by Jacek Fisiak, 183—197. Berlin. —. 1984. On the causes of word order change. Lingua. 63. 175—204. Hawkins, John. 1982. Cross-Category Harmony, X-bar and the Predictions of Markedness. Journal of Linguistics 18, 1. 1—35. —. 1983. Word Order Universals. New York. —. Forthcoming. Competence and Performance in the Explanation of Language Universals. To appear in Syntactic Theory: A Festschrift for David Kilby, ed. by Jacques Durand & Martin Atkinson. Oxford. Hyman, Larry. 1975. On the change from SOV to SVO: Evidence from Niger-Conger. Word order and word order change, ed. by Charles Li. Austin. Itkonen, Esa. 1978. Short-term and long-term teleology in linguistic change. Papers from the Conference on General Linguistics, Seili, 24—25. 8. 1978. Turku. —. 1983. Causality in Linguistic Theory. London, Bloomington. Kuno, Susumu. 1974. The Position of Relative Clauses and Conjunctions. Linguistic Inquiry 5. 117—136. Lambrecht, Knud. 1981. Topic, antitopic and verb agreement in non-standard French. Amsterdam. Lang don, Marg aret. 1977. Syntactic Change and SOV Structure: the Yuman case. Mechanisms of Syntactic Change, ed. by Charles Li, 255—290. Austin. Lass, Ro g er. 1980. On Explaining Language Change. Cambridge. Lehmann, Winfred. 1973. A structural principle of language and its implications. Language, 49. 47— 66. —. 1974. Proto-Indo-European Syntax. Austin. Lenerz, Jurg en. 1984. Syntaktischer Wandel und Grammatiktheorie. Tübingen. Li, Charles, and Sandra Thompson. 1974. An explanation of word order change SVO SOV. Foundations of Language 12. 201—214. —, and —. 1975. Subject and topic: a new typology of language. Subject and Topic, ed. by Charles Li, 457—489. New York. Lig htfoot, David. 1979. Principles of Diachronic Syntax. Cambridge. —. 1981a Explaining Language Change. Explanation in Linguistics: the logical problem of language acquisition, ed. by Norbert Hornstein & David Lightfoot. London. —. 1981b A reply to some critics. Lingua 55. 351—368. Masica, Colin. 1976 . Defining a linguistic area: South Asia. Chicago, London.

XX. Die Erforschung des Syntaxwandels III: Phänomene

1226

Meillet, Antoine. 1912/1958. Linguistique historique et linguistique générale. Paris. Miller, D. Gary. 1975. Indo European: VSO, SOV, SVO or all three? Lingua 37. 31—52. Platzack, Christer. 1986 . COMP, INFL, and Germanic Word Order. Topics in Scandinavian Syntax, ed. by Lars Hellan & Kirsti Koch Christensen. 185—234. Dordrecht. Sapir, Edward. 1921/1963. Language. London. de Saussure, Ferdinand. 1959. Course in General Linguistics, trans. Wade Baskin. New York. Smith, Neil. 1981. Consistency, markedness and language change: on the notion ‘consistent language’. Journal of Linguistics, 17, 1. 39—54. Tomlin, Russell. 1986 . Basic Word Order: Functional Principles London. Traug ott, Elizabeth. 1982. Concluding Remarks. Papers from the 5th International Conference on Historical Linguistics, ed. by Anders Ahlqvist, 460—466. Amsterdam. Vennemann, Theo. 1975. An explanation of drift. Word order and word order change, ed. by Charles

67. 1. 2. 3. 4. 5. 6. 7.

1.

Martin Harris, Salford (Great Britain)

Clisis Introduction Case Study Theories A New Proposal Diachrony Conclusion References

Introduction

Where do clitics come from? Do they have a single diachronic source? What kinds of change over time are possible within an established clitic system? These questions about the diachrony of clitics have no widely accepted answers, although a few reasonably well-stated hypotheses have been advanced. Discussion in this article of the major proposals is bracketed on one side by a brief diachronic case study as an example of the problems to be solved and on the other by a proposal and its implications for diachronic studies of this phenomenon.

2.

Li. Austin. —. 1983. Causality in language change: theories of linguistic preferences as a basis for linguistic explanations. Folia Linguistica Historica 4, 1. 5—26. Vincent, Nig el. 1976 . Perceptual factors and word order change in Latin. Romance Syntax: synchronic and diachronic perspectives, ed. by Martin Harris, 54—69. Salford. —. 1978. Is sound change teleological? Recent Developments in Historical Phonology, ed. by Jacek Fisiak, 409—430. The Hague. —. 1979. Word order and grammatical theory. Linear order and generative theory, ed. by J. M. Meisel & M. D. Pam. 1—22. Amsterdam. —. 1987. Latin. The Romance Languages, ed. by Martin Harris & Nigel Vincent. London, New York. Watkins, Calvert. 1976 . Towards Proto-Indo-European Syntax: problems and pseudo-problems. Diachronic Syntax, ed. by Sanford Steever, Carol Walker & Salikoko Mufwene, 305—326. Chicago.

A Case Study

The problems besetting an adequate diachronic theory of clitics are best revealed by

a case study. Although many languages have clitics, their history has not been a major focus in the historical literature. We consider here some of the relevant facts of a language family where the diachrony of clitics has received a reasonable amount of attention — Takic, the Southern California branch of Uto-Aztecan. The generally accepted internal organization of Takic is given in (1). (See Jacobs 1975 for discussion.)

We focus also on only one clitic type, pronominal clitics marking person and number. Consider the Luiseño examples in (2). (Except for Luiseño, the source of example sentences is indicated in parentheses. All Luiseño examples are taken from my work with Mrs. Villiana Hyde, a native speaker of the language.) The contrast among (2a), (2b), and (2c) illustrates that the clitic varies in

67.  Clisis

person and number. The contrast between (2) and (3) illustrates that this clitic is part of a sequence marking other notions — here specifically tense.

Of the four most well-documented Takic languages (Cahuilla, Cupeño, Luiseño, and Serrano), three have second position person and number clitics. That is, the clitics appear in sentential second position. To illustrate, the order of elements in the Luiseño sentences in (2) and (3), except for the clitics, is free. (4a) is simply another, semantically non-distinct, way of executing (2a), but (4b) — where the clitic sequence is not second — is unacceptable.

1227

clitic would be m and if a third singular subject were acting on a third singular object there would be no clitic. Takicists have generally assumed that Luiseño has clitics for subject only. (We modify this assumption in Section 4. below.) The Cupeño second position clitic sequence exhibits aspects of both these possibilities; while it includes a single person/number value, the form varies for transitive and intransitive cases. Jacobs (1975, 49) offers the contrast between (5a) and (6 ), identifying the latter as containing a transitive verb even though there is no object in the example.

Cahuilla also has a set of bound pronominal elements, but as the sequence is obligatorily preverbal, these elements are generally taken to be prefixes rather than clitics. The examples in (7) illustrate that this set, like the Serrano second position clitics, can include two person/number markings — one for subject and another for object — although the analysis of the sequence seems more transparent.

Cupeño also has prefixed forms. (8) illustrates what is described by both Jacobs (1975) and J. Hill (196 6 ) as an optional object prefix, as well as a prefix for the number and person of the subject.

One difference among the Takic second position person/number clitics is illustrated in a contrast between the Serrano and the Luiseño examples. In Serrano but not Luiseño the clitic sequence includes two markings for person and number. According to Crook (1973) and Langacker (1977) the form mi in (5b) indicates both a third plural subject and a third plural object. If a third plural subject were acting on a third singular object the

Unlike the Cahuilla situation, these cooccur as illustrated with a second position clitic sequence, but not one which marks person and number. ǝp in (8) is characterized by Jacobs as marking a + realized situation. The choice of prefixal or second position pronominal forms depends on the tense of the verb. In sum, although each of these four reasonably closely related languages has a system of bound pronominal elements, none is identical to any of the others. The systems differ both in position and in argument types

XX. Die Erforschung des Syntaxwandels III: Phänomene

1228

marked. Jacobs (1975), Steele (1977), and Langacker (1977) agree that proto-Takic had a second position clitic sequence and that the prefixal person/number marking in Cupeño and Cahuilla is an innovation. In addition, Langacker (1977) argues that the proto sequence included object marking; thus, the absence of second position object marking in Luiseño and Cupeño is also an innovation. In short, Serrano, having a second position clitic sequence which includes object marking, is generally taken to be closest to the Proto-Takic situation — although the amalgamation of subject and object marking occurred in Serrano. The presumed development sketched in (9) is a slight modification of Langacker’s diagram.

(SM and OM abbreviate Subject Marking and Object Marking respectively.) This reconstruction identifies how some of the synchronic states give rise to others, but it doesn’t consider why. Why should prefixes have been introduced in Proto Cahuilla-Cupeño, for example? And on the change from Proto Takic to Proto Cupan, it is silent on both questions: Given a beginning point with a second position clitic sequence marking both subject and object, what changes yield Proto-Cupan? How and why does a language lose object marking? The point here is to illustrate the issues confronting a study of the diachrony of clitics, not to answer all the questions (9) raises. After an examination of the more general theoretical questions, we return to one, that which receives the least attention in (9). Put most succinctly, how does a proto-language like Serrano yield a language like Luiseño?

3.

Theories

The diachronic model in (10) has a number of adherents particularly in the typological

literature. (10) Independent Word → Clitic → Affix Givón (1971), for example, puts forward the relatively strong claim that clitics reflect the reductions in situ of previously independent forms. In (10) clitics are merely a transition between words and affixes, a consequence presumably of phonological reduction and semantic bleaching. Givón’s claims and the model in (10) have not gone unchallenged. Chafe (1977) argues that affixes can be the source of other affixes in Iroquian, and Langacker’s reconstruction of Takic requires the refurbishing of old clitics. That is, clitics are not the sole source of affixes, and clitics can be maintained. But the most direct challenge is that of Steele (1977). Based also on a reconstruction, this paper proposes that the Uto-Aztecan second position clitics do not reflect the position of their diachronic source. A second position clitic marking the person and number of the subject does not indicate that the preceding stage had subjects following, for example, objects. Rather their appearance in second position is a consequence of the fact that they are clitics; clitics have positional properties peculiar to them. Thus, even if independent words are the source of clitics, cliticization is not reduction in situ. The diachronic argument reduces to a dispute over whether grammatical elements have syntactic properties distinct from lexical elements. In fact, in the recent flurry of interest in clitics as a synchronic phenomenon two distinct strands can be isolated, analogues of the sides drawn in the diachronic dispute. One strand, represented most clearly in Borer (1986 ), is concerned with clitics because of the issues they raise for a theoretical paradigm that has to place them either in the word formation component of a grammar or in the syntactic component. If they are subsumed in the former, the fact that they have syntactic effects is a problem for the strict lexicalist hypothesis, the strongest formulation of which is due to LaPointe (1981): Syntactic operations may not refer to parts of words. If they are considered part of the syntactic component, however, the fact that they are clearly distinct from their independent counterparts becomes an issue. In the other strand (see Kaisse (1982), Zwicky/Pullum (1983), Zwicky (1985), and Klavans (1985), in particular), this problem is non-existent. This research begins with the idea that clitics are a theoretical construct —

67.  Clisis

“ ... languages contain ... only words belonging to syntactic categories, clitics, and (inflectional or derivational) affixes” (Zwicky 1985, 194). Given this status for clitics, the issue is to define their properties. Kaisse argues that Wackernagel’s position (sentential second position) for sentential clitics is a special case of the more general attachment to edges required in the generalization: (11) = Kaisse’s (11) A clitic originating as an immediate daughter of X may be adjoined only to the leftmost or rightmost node of X. So, an NP clitic (e. g. a possessive) attaches to the edge of the NP, a VP clitic (e. g. a pronominal object clitic) attaches to the edge of a VP, and a sentential clitic (e. g. a question marker) attaches to the edge of an S. (12) illustrates the possibilities considered by Kaisse in regard to (11). (12a) clitic [X...] [X clitic ...] (12b) [...X] clitic [... clitic X] Klavans presents the most elaborate proposal in regard to clitic properties. She offers three binary parameters (initial/final, before/after, proclitic/enclitic) yielding eight clitic types. For example, the Luiseño clitic sequence is initial because it is dominated by the initial element in S; after because it follows the initial element; and enclitic because it is phonologically dependent on the preceding element. As this brief review suggests, this approach to clitics has concentrated heavily on their positional properties, and even here some problems remain. As Kaisse acknowledges, the asymmetries between beginning edge and final edge and between enclitic and proclitic — the first of which are strongly favored cross-linguistically — is not accounted for by (11). Similarly, all eight of Klavans types are equally valued, but are not equally distributed. The important point for our purposes, however, is the parallel between the two sides to the diachronic argument and the two synchronic approaches just outlined. This parallel has gone without comment, a result perhaps of the place of reconstruction in the diachronic argument. Givón’s proposal and the challenges to it are based on reconstructions. This has put the entire discussion beyond the pale for some. Lightfoot’s (1979) comments about the value of reconstruction to dia-

1229

chronic theory are the most condemnatory. For example, he says: “ ... we can never discover anything new about the nature of change by examining the relationship between attested languages and our reconstructed abstraction ... Therefore the mapping between a reconstructed language and its attested daughters ... is not an appropriate basis for illustrating types of change, much less for acquiring insight into the nature of change” (Lightfoot 1979, 387). Lightfoot’s point is that, because a child learning a language must construct a grammar anew from primary data, there is no connection between grammars in succeeding stages of a language. A syntactic reconstruction is a (fragment of a) grammar hypothesized on the basis of a synchronic grammar or set of grammars; thus, it involves the kind of relationship which doesn’t exist. We will return to this point. For now, we note only that we needn’t depend on reconstruction for the two diachronic views of clitics. Rivero’s (1986 ) and Wanner’s (1987) competing recent work on Romance clitics parallels the earlier argument. Rivero argues that Old Spanish “ nontonic pronouns” are lexical categories occupying an NP position. The sentence Ellos lo entendieron ‘They understood it.’ has the structure in (13).

Although Wanner doesn’t supply a tree structure for his analysis, he counters that the positional properties of such pronouns are to be distinguished from NP’s — they are restricted to sentential second position or to a verb-adjacent position. Thus, they aren’t NP’s; they are clitics.

4.

A New Proposal

This parallel between diachronic and synchronic debates argues that the two types of investigation are not inherently different.

XX. Die Erforschung des Syntaxwandels III: Phänomene

1230

However, the implications of the most recent synchronic work for the diachrony of clitics have gone almost entirely without comment. Klavans’ brief consideration of the diachronic implications of her classification is one exception. If there are eight clitic types, defined by three parameters, change in a clitic system results from a change in a single parameter. For example, the clitic type exemplified in (2) and (3) above could change from an enclitic to a proclitic. (14)

initial after enclitic

→

initial after proclitic

Klavans doesn’t address how a parameter might change; her point is simply that the diachronic relationships among the eight clitic types are severely constrained. The initial state in (14) has only three, not seven, possible succeeding states — that in (14) and the other two in (15): (15a) initial before enclitic

(15b) final after enclitic

These ideas are very much in the spirit of Greenberg (1978). Given a number of different types, the diachronic problem is determining their logical progression. This view of language change sees it not as a psychological problem, as Lightfoot would have it, but rather as a logical problem. Where there are proposals as to the sorts of grammars that can exist, it is possible to consider their sequential relationships. However elegant, Klavans’ classification does little to resolve the historical questions posed by Takic in Section 2. above. Recall that three of the four Takic languages discussed there (Serrano, Luiseño, and Cupeño) maintain a second position clitic sequence of the initial/after/enclitic type, like the hypothesized proto-language. Yet only Serrano marks the person and number of two arguments in this clitic sequence, also like the hypothesized protolanguage. The prefixal forms are assumed to have a diachronic source distinct from the second position forms. Leaping out, then, is not a change in position or attachment as in (14), but rather how a language might lose clitics marking the number and person of an object. We focus, as suggested above, on the subpart of (9) represented in 6 (1 ).

To address this question, a theory of the linguistic types available is necessary. This is a logical, not a psychological, problem for language change. We introduce a typology which, unlike Klavans’, distinguishes among two functionally distinct clitic types. As has been argued in a number of recent works (e. g. Jelinek 1984, Manandise 1984, Nishida 1987, or Bresnan/Mchombo 1987), a bound person/number form (clitic or affix) may serve to satisfy, or partially satisfy, the argument requirements of the verb. If every verb is lexically specified for the set of arguments it requires, the presence of such a form fulfills these requirements. The forms themselves, then, don’t “ agree” with the arguments; they are the arguments. That is, they are fully anaphoric pronouns. This type of person/number marking is not limited to clitics, rather clitics are but one instantiation. There is, I would argue, a second possibility available to a clitic sequence containing person and number marking: It may be the element which maps a verb and its arguments to a sentence. (17) Clitic Sequence: [Verb, Arguments] → Sentence A clitic sequence as in (17) applies to the combination of a verb and its arguments to yield a sentence. A person/number clitic of the type in (17), then, is not anaphoric, does not satisfy the verb’s argument requirements. Rather it is a sort of “ agreement” with whatever arguments it indexes. (Bresnan/ Mchombo also do not take the position that all person/number marking is anaphoric. But they do not analyze the non-anaphoric cases as in (17).) Both these types are found in Takic. The Luiseño second position clitic sequence is of the second type (hereafter Type 2); the second position clitic sequences in Serrano and Cupeño — and the pronominal affixes in Cahuilla and Cupeño — are of the first (hereafter Type 1). In support of this claim, we note one striking difference between Luiseño on the one hand and the other three Takic languages on the other. The descriptions of Serrano, Cahuilla, and Cupeño all mention that ques-

67.  Clisis

tions and statements have different intonational properties. Sauvel/Munro (1981, 21), for example, say of the two Cahuilla sentences in (18): (18a) Níchill kúktashqa woman is:talking ‘The woman is talking.’ (18b) Níchill kúktashqá woman is:talking ‘Is the woman talking?’ “ The same words exactly are used to make the statement and to ask the question; the only difference is that in the question [the] voice goes up at the end of the verb ...” In Luiseño, there is no intonational difference between questions and statements; rather the form of the clitic sequence indicates the distinction. Compare, for example, the two sentences in (19) and (3a) above. In (19a) the clitic sequence contains the form u in addition to nil ‘lsg past’ and the sentence involves modality. In (19b) the clitic sequence contains su in addition to n ‘lsg’ and the sentence is a question.

su is not a question-marker; rather a question in Luiseño must contain a clitic sequence with this form and person/number marking and nothing else. (See Steele et al. 1981 for further discussion.) In sum, the Luiseño clitic sequence is functionally equivalent to the intonation contour in the other three Takic languages: Both distinguish between questions and statements. Assuming that a speech act is one of the critical properties distinguishing between a combination of a verb and its arguments and a sentence, the identification of the Luiseño clitic sequence as in (17) is clear. In the absence of a clitic sequence, there is neither a statement nor a question, but rather a combination of a verb and its arguments only. Equally, the person/number marking in the other Takic languages, even those in Serrano and Cupeño that share the same sentential position, cannot be so identified. It might appear from the above examples that the Luiseño second position person/ number clitics mark the person and number of the subject. Recall the difference among the three sentences in (2), where the clitics

1231

vary for person and number with what is intuitively the subject; in fact, as noted in Section 2., Luiseño person/number clitics have been generally identified as subject marking. But Type 2 clitics are claimed not to satisfy the argument requirements of the verb. Fortunately, it can be shown that the Luiseño second position person/number clitics do not perform this function. I have argued in other work (Steele 1987, for example) that the argument requirements of a verb in Luiseño are met by the obligatorily present non-clitic forms and a person/ number value dependent on them. Consider the sentences in (20).

’ariqu ‘was kicking’ requires both a kicker and a kickee. The latter may appear as a reflexive as in (20a) or as an object-marked form as in (20b), but it must be present. The former has no obligatory lexical instantiation; it is rather a person/number value determined by the obligatory forms. So, in (20a) the kicker is ‘3sg’ because of the value of the reflexive. Compare (21). Like ariqu, miyqu ‘was’ requires two arguments. One must appear as a number-marked form — in (21) yawaywich-um ‘pretty-pl’; the other has no obligatory instantiation (and is not lexically present in (21).

The lexically absent argument in (21) is open for person but is specifically ‘pl’, because of the number of yawaywichum; that is, this argument is the value ‘Ppl’. The kicker in (20b) is also a value given by the obligatory objectmarked argument; the value is open for both person and number — i. e. ‘PN’. The second position person/number marking is compatible in these three cases with this value; ‘3sg’ is compatible with ‘3rd sg’ and ‘PN’ in (20a) and (20b) respectively, while ‘nonlpl’ is compatible with ‘Ppl’ in (21). This compatibility between second position clitic and argument is not equivalent, of course, to satisfying the argument requirements of the verb. Most telling in support of this analysis and this claim is the behavior of clitics in a

XX. Die Erforschung des Syntaxwandels III: Phänomene

1232

particular kind of sentence. Consider the contrast between the two sentences in (22).

’aw’ qu and qalqu are like ’ariqu and miyqu in requiring two arguments, only one of which has obligatory lexical form. (While somewhat less preferred, semantically equivalent sentences lacking noo ‘I’ are acceptable, e. g. nowaamay pil ’aw’qu and nowaamayum pahchum mil qalqu respectively.) Like miyqu the one obligatory argument must be marked for number — nowaamay in (22a) is ‘sg’ and nowaamayum in (22b) is ‘pl’. In addition, however, ’aw’qu and qalqu require that the number-marked argument also have a possessive prefix, in (22) the prefix no ‘my’. As the glosses in (22) suggest, the argument dependent on the obligatory form has whatever value is specified by the possessive. Compare (23) which replaces no ‘my’ with cham ‘our’.

Of interest for our purposes is the character of the second position clitic; it varies in (22) and (23) not with the person/number value — ‘1 sg’ and ‘1 pl’ — but with the number of the obligatory argument. If the clitic were analyzed as an argument, these sentences present considerable difficulties. However, if the clitic “ agrees” with the set theoretic complement of the clitic sequence, they have a simple account. In (20) and (21), the set theoretic complement presents one value with which the clitic can agree (‘3sg’ in (20a), ‘PN’ in (20b), and ‘Ppl’ in (21)); in (22) and (23) it presents two (‘1 sg’ and ‘sg’ in (22a), ‘1 sg’ and ‘pl’ in (22b), and ‘1pl’ and ‘sg’ in (23)). An alternative, less highly valued, execution of such sentences has the clitic agreeing with the other value.

In sum, the role of the Luiseño clitic sequence in distinguishing between questions and statements is consistent with its characterization as Type 2. Thus, we predict that the person/number clitics in this sequence may not satisfy the argument requirements of the verb. The sketch above supports this prediction. Assuming that the bifurcation between Type 1 and Type 2 is exhaustive, the role of intonation in distinguishing between questions and statements in the other Takic languages is consistent with characterizing their bound pronominal forms as anaphoric.

5.

Diachrony

The preceding has argued that the second position person/number clitics in Serrano, the prefixal forms in Cahuilla, and both types in Cupeño are functionally different from those in Luiseño. The first three are instances of Type 1 and the last, an instance of Type 2. (25) plugs these identifications into the genetic tree.

The types of the person/number forms in the proto languages in (25) are left unidentified. Our attention is focused on the type of Proto Takic and Proto Cupan. One observation about the daughters is suggestive toward resolving this issue. Note that the difference between Type 1 and Type 2 forms correlates with the distribution of object marking. Cahuilla, Cupeño, and Serrano have object marking; Luiseño doesn’t. One hypothesis to be entertained, then, is clear: Object marking is always anaphoric. If this is correct, it follows necessarily that no Type 2 sequence

67.  Clisis

1233

could contain object marking. The mapping from a verb and its arguments to a sentence presumes a fully specified argument set. Conversely, it should be clear that the absence of object marking does not yield Type 1. Type 1 clitic sequences do not obligatorily contain both subject and object arguments; they could contain only one person/number set, whether it mark the subject argument or the object argument. (26 ) represents the three possibilities in a table. (26) Type 1

Type 2

2 person/number 1 person/number sets set subj/obj subj or obj (a) (b) 1 person/number set neither subj nor obj (c)

This classification defines the diachronic possibilities. (26 a) differs from (26 c) in both ways in which a difference is possible, while (26 a) and (26 c) differ from (26 b) in only one of the two possibilities. Thus, using Greenberg’s criterion of the logical progression of types, (26 a) cannot develop directly into (26 c), but (26 a) could develop into (26 b) which, in turn, could develop into (26 c). Although we focus on this direction of development because of the proto-state assumed, we should note that the reverse developments are also possible. (27) imposes the diachronic paths on the typology in (26).

The identification of Proto Takic and Proto Cupan can now be given relatively straightforwardly. Proto Takic must be of the type identified in (26 a). Proto Cupan, as its descendent, must be of the type identified in (26b) — not the type in (26c).

In short, from the hypothesized proto-Takic

state to the synchronic Luiseño state cannot be a single unitary change; it must involve two steps. And this necessity follows from the classification into Type 1 and Type 2 sequences, once we consider what properties these types entail. While (27) satisfies certain diachronic questions, it raises others. The arrows in (27) indicate only that some synchronic states are directly connected to others and indirectly connected to still others; the arrows say nothing about how one state could develop into another. I touch here only — and only briefly — on the possible cause of the change between (27a) and (27b), i. e. in our case study on the change between Proto Takic and Proto Cupan. As discussed above, in Luiseño the argument set required by a verb involves a particular form or set of forms and a single person/ number value. Recall the differences between ’ariqu, miyqu, and ’aw’qu/qalqu. While all require one obligatory lexical argument, the form varies — object-marked or reflexive for ’ariqu, number-marked for miyqu, and possessive- and number-marked for ’aw’qu and qalqu. The argument which appears as a person/number value is the single member of the argument set whose form is not specified by the verb. In Serrano, Cahuilla, and Cupeño, however, it appears that the argument set involves maximally two members whose form is not specified by the verb — the two which appear as bound pronominals. A startling similarity among these three languages — and an equally striking difference from Luiseño — offers confirmation of this conjecture. Whether a bound object pronominal is possible or not varies not simply with the verb itself, but rather with the form of the verb as well. In each of these languages, we have verbs that take two arguments (subject and object) where the form of the object argument is a function of the inflections on the verb. When tensed, such a verb occurs with a bound person/number object form, but the same predicate inflected in the imperative occurs rather with an independent argument — specifically, one inflected for number. (This is the Takic analogue of the Finnish “ nominative object” discussed by Timberlake (1975).) The contrast between the Cahuilla sentences in (29) is illustrative. (29a) tevinmallm-i me-sexqa beans-object 3sg/3pl-cook:pres ‘She is cooking the beans.’

XX. Die Erforschung des Syntaxwandels III: Phänomene

1234

(29a) contains a tensed form of ‘cook’; (29b), the imperative form. The object in (29a) is a bound pronominal form (accompanied by the object-marked tevinmallmi); the object in (29b) is the independent (and non-objectmarked) tevinmallem. tevinmallmi contains a plural suffix, like tevinmallem, but it is the presence of the object suffix that is critical in (29a); the number internal to the object suffix is unimportant. In contrast, the presence of a number-marked form is essential in (27b). A comparison with the Luiseño sentences in (30) where the form of the object does not vary with the form of the verb should drive the point home.

and Type 2 person/number marking has farreaching consequences. In requiring two changes between proto-Takic and synchronic Luiseño, it allows a new perspective on Takic diachrony. More importantly, and perhaps surprisingly, it provides a framework within which certain apparently unconnected properties are shown to be closely connected. It is not an accident that Serrano, Cahuilla, and Cupeño exhibit object marking, an intonational distinction between statements and questions, and maximally two arguments outside the control of the predicate, while Luiseño lacks object marking, does not distinguish statements and questions by intonation, and has maximally one argument outside the control of the predicate. We saw in Section 3. that the issues raised in the study of clitics and diachrony are right in the mainstream of synchronic theory. The discussion of Sections 4. and 5. argues that the properties of clitics are not isolated from other grammatical properties. Future study of the diachrony of clitics will, we hope, play more heavily on such connections.

7. In short, in Luiseño, the form of one argument may be external to the requirements of the verb; in Cahuilla, Serrano, and Cupeño, the form of two arguments may be external, i. e. may be determined by other properties of the sentence. In light of this difference among the Takic languages, it seems reasonable to consider the hypothesis that bound person/ number forms marking object are possible only when the verb does not govern the form of the argument. The development from (27a) to (27b) — the change from Proto Takic to Proto Cupan — is a decrease (from maximally two to maximally one) in the number of argument forms outside the control of the verb. Conversely, the development from (27b) to (27a) is an increase (from maximally one to maximally two).

6.

Conclusion

The typological distinction between Type 1

References

Borer, Hag it. (ed.) 1986 . The Syntax of Pronominal Clitics. Syntax and semantics, vol. 20. Orlando. Bresnan, Joan, and Sam Mchombo. 1987. Topic, pronoun, and agreement in Chichewa. Language 63. 741—82. Chafe, Wallace. 1977. The evolution of third person verb agreement in the Iroquian languages. Mechanisms of syntactic change, ed. by Charles Li, 493—524. Austin. Crook, Donald. 1973. Informal Serrano notes. MS. Givón, Talmy. 1971. Historical syntax and synchronic morphology: an archaeologist’s field trip. Papers from the 7th Regional Meeting, Chicago Linguistic Society, 394—415. Greenberg , Joseph. 1978. Diachrony, synchrony, and language universals. Universals of human language, vol. 1, ed. by Joseph Greenberg, 6 1—92. Stanford. Hill, Jane. 196 6 . A grammar of the Cupeño language. University of California at Los Angeles dissertation. —, and Rosinda Nolasquez. 1973. Mulu’wetam: The first people. Banning, California. Hill, Kenneth. 196 7. A grammar of the Serrano language. University of California at Los Angeles dissertation. —. 1969. Serrano clitics. MS. Jacobs, Roderick. 1975. Syntactic change: a Cupan (Uto-Aztecan) case study. (University of California publications in linguistics 79). Berkeley.

68.  Agreement

Jelinek, Eloise. 1984. Empty categories, case, and configurationality. Natural language and linguistic theory 2. 39—76. Kaisse, Ellen. 1982. Sentential clitics and Wackernagel’s law. Proceedings of the first West Coast Conference on Formal Linguistics, 1—14. Stanford. Klavans, Judith. 1985. The independence of syntax and phonology in cliticization. Language 6 1. 95— 120. Lang acker, Ronald. 1977. Reconstruction of pronominal elements in Takic. Proceedings of the third annual meeting of the Berkeley Linguistics Society, ed. by Kenneth Whistler, Robert D. Van Valin, Jr., Chris Chiarello, et al., 85—102. Berkeley. Lapointe, Steven. 1981. General and restricted agreement phenomena. The scope of lexical rules, ed. by Michael Moortgat, Harry van der Hulst & Teun Hoekstra, 125—59. Dordrecht. Lig htfoot, David. 1979. Review of Mechanisms of Syntactic Change. Language 55. 381—95. Manandise, Esmeralda. 1984. Some aspects of Basque morphology and syntax. University of Arizona dissertation. Nishida, Chiyo. 1987. Interplay between morphology and syntax: a lexical analysis of inflection and cliticization. University of Arizona dissertation.

1235

Rivero, Maria-Luisa. 1986 . Parameters in the typology of clitics in Romance and Old Spanish. Language 62. 774—807. Sauvel, Katherine S., and Pamela Munro. 1981. Chem’ivillu’ (Let’s speak Cahuilla). Los Angeles. Steele, Susan. 1977. Clisis and Diachrony. Mechanisms of Syntactic Change, ed. by Charles Li, 539—82. Austin. —. 1987. Constituency and Luiseño argument structure. Syntax and semantics, vol. 20, ed. by Geoffrey Huck & Almerindo Ojeda, 341—75. Orlando. —; Adrian Akmajian; Richard Demers; Chisato Kitag awa; Eloise Jelinek; Richard T. Oehrle; and Thomas Wasow. 1981. An encyclopedia of AUX: A study in cross-linguistic equivalence. (Linguistic Inquiry Monograph Series 5). Cambridge. Timberlake, Alan. 1975. The nominative object in Finnish. Lingua 35. 201—30. Wanner, Dieter. 1987. Unstressed object pronouns in Old Spanish: NP vs. clitic status, paper presented at LSA meeting. Zwicky, Arnold. 1985. Clitics and particles. Language 61. 283—305. —, and Geoffrey Pullum. 1983. Cliticization vs. inflection: English n’t. Language 59. 502—13.

Susan Steele, Tucson, Az (USA)

68. Agreement 1. 2. 3. 4. 5. 6. 7. 8.

1.

Introduction The Rise of Verb Agreement Systems Assessment of the Accepted View The Development of Other Types of Agreement The Loss of Agreement Finer-Grained Changes in Agreement Systems Conclusion References

Introduction

Agreement is still poorly understood. Changes in agreement systems are therefore of particular interest, since they may provide insights into the function of agreement as well as into more general mechanisms of language change. We shall concentrate on major changes, beginning with the way in which agreement systems arise — starting with the accepted view on the rise of verb agreement

(section 2.) and giving an assessment of it (section 3.). Then we move to the development of other types of agreement (section 4.) and the loss of agreement (section 5.). We also consider briefly finer-grained work on minor changes within agreement systems (section 6.).

2.

The Rise of Verb Agreement Systems

The most influential paper concerning the rise of agreement is Givón (1976 ), where it is claimed that verb agreement markers develop from anaphoric pronouns. This idea was not completely new; there are various suggestive observations going back to early in the last century (see Claudi 1985, 56 —7; Bresnan and Mchombo 1987, 742); work in typology also offered valuable clues. But it was Givón who presented an explicit hypothesis and based extensive claims upon it (an early sketch is found in Givón 1971, 402—3). His account

XX. Die Erforschung des Syntaxwandels III: Phänomene

1236

still provides a good framework for a discussion of what is known at present about the rise of agreement. Givón claims that subject-verb agreement develops from topic-verb agreement, more specifically from topic-shifting constructions where the noun phrase which is topicalized is coreferential to one of the verb’s arguments. This is shown schematically as follows; in (1) we have a shifted topic: (1) the man, he TOPIC PRONOMINAL SUBJECT came VERB In this marked construction, the pronoun represents topic agreement. This construction may become reanalyzed as a neutral sentence type, as has occurred in many non-standard dialects of American English: (2) the man SUBJECT he-came AGREEMENT.MARKER-VERB Why should this reanalysis occur? It results from the overuse of a powerful discourse device, which is a reasonable strategy, particularly if conditions for communication are difficult. Two interesting predictions follow. First, languages which use zero-anaphora in place of anaphoric pronouns will not develop corresponding verb agreement. And second, since the subject noun phrase typically has more features favouring topicalization than have other noun phrases, it is more frequently topicalized and so subject agreement is a precondition for agreement with other arguments (Moravcsik 1974, 27—8). Givón claims that evidence for pronouns becoming subject agreement markers can be found in English and French dialects and in related pidgins and creoles. Various Bantu languages also provide confirming data, as is illustrated from Swahili: ki-li-vunjika (3) kikopo cup AG-PAST-break ‘the cup broke’ The subject agreement markers retain their older anaphoric function: (4) ki-li-vunjika it-PAST-break ‘It (the cup) broke’ Swahili also helps us to understand the rise of object agreement. This is not obligatory,

as shown by the fact that it does not occur in the following example: (5)

ni-li-vunja I-PAST-break ‘I broke

kikopo cup a

cup’

The object may be topicalized, as in example (6): 6( )

kikopo, cup, ‘the

ni-li-ki-vunja I-PAST-it-break cup, I broke

it’

Here we do find the object marker ki. Provided the object is definite, it may in Swahili occur in its usual place after the verb and the marker ki may be retained (thus the verb form would be as in (6 )). This represents (definite) object agreement. Various Bantu languages are at different stages in the development of object agreement. As is implicit in the account above, Givón claims that agreement and pronominalization are ‘fundamentally one and the same phenomenon’. This seems right. It is supported by the fact that most recent attempts to define agreement include within it the determination of the form of pronouns (though they may not be linked syntactically to the antecedent). And those who try to split agreement from pronominalization (anaphora) put the dividing line at different points. Moreover data on agreement options show that attributive modifiers and pronouns are linked as poles of a single hierarchy (the Agreement Hierarchy, Corbett 1979). After a review of recent work on the subject Barlow (1988, 134—52) concludes that there is no good basis for distinguishing between agreement and antecedent-anaphora relations (even when cross-sentential); see also Lehmann (1982, 237—8) on this point. Givón’s account (which is restated briefly in Givón 1984, 36 0—72) is intuitively satisfying and many have accepted it as basically correct. Perhaps as a result, even though Givón’s data were somewhat patchy, rather less has been done to fill the gaps than might have been hoped. Nevertheless some interesting work has resulted. Duranti and Ochs (1979) give a good deal of data on Italian discourse and suggest that clitics are moving towards becoming agreement markers in the way predicted. In another detailed study of discourse data, this time from Swahili, Wald (1979) tries to improve on Givón’s analysis of the rise of object marking. The role of the object marker in various Bantu languages is

68.  Agreement

taken up by Hyman and Duranti (1982), and in great detail in Chichewa by Bresnan and Mchombo (1986 , 1987). Claudi (1985, 105— 14) documents the early stages of the development of predicate agreement in Zande. The different stages of grammaticalization of the categories of person and number, and their interaction with case relations, are documented in Mithun (1991). Finally, there is a useful overview of the development of agreement in general in Lehmann (1982, 251—7; see also Lehmann 1988).

3.

Assessment of the Accepted View

In this section we review three areas of Givón’s account which are open to question. 3.1. Topic-Shifting There is considerable evidence to support the view that agreement results from topicalization, part of which we have already considered. More recently, Harris (forthcoming) suggests that in the North-East Caucasian languages Tabassaran and Udi, new agreement markers have arisen from clitics which were pronominal copies of the topic (here there is evidence for topicalization but no positive evidence for any movement). What is particularly interesting is that the new markers show agreement in person and number, while the old ones marked agreement in class (gender) and number. Moreover, the old agreement system was an ergative one (showing agreement with the absolutive) while the new system shows agreement with the subject. The claim that agreement (and especially agreement in person) arises from shifting constructions gains considerable support from typological data. Thus Steele (1978, 6 10—14) found a correlation between certain departures from canonical word order and the existence of person agreement, in particular person agreement in which the agreement markers are similar to the independent pronouns. And in a more detailed study, Foster and Hofling show that: “ In languages which are not typologically regular with regard to word order, irregularly ordered constituents are marked for agreement significantly more frequently than are regularly ordered constituents. [...] Similarly, at a language-specific level, when constituents which appear contiguously in unmarked constructions in the

1237

language are rearranged or separated, agreement tends to appear” (1987, 476). Clearly if synchronically agreement is more likely when constituents have been ‘moved’, then this adds weight to the suggestion that shifting constructions provide the source for agreement. (Note that the evidence presented by Foster and Hofling goes well beyond verb agreement.) More generally, the evidence that agreement is more likely with constituents which occur other than in the position expected does not necessarily support topic-shifting as the source. Steele (1977) shows that in various Uto-Aztecan languages clitic pronouns move to second position as they become clitics and that from there they may be attracted on to the verb. Here then we have a potential source of verb agreement, with pronouns and movement as essential elements, but with no evidence that a topic-comment construction was involved (1977, 550). Givón (1984, 357 n) appears to accept this possibility. And somewhat earlier, Hale (1973, 339—44) in his discussion of the development of person and number marking in Warlpiri, suggested that independent pronouns gave rise to unstressed clitics, which attached to the auxiliary (found in second position). He did not invoke topic-shifting to account for the development. At this point it is worth mentioning that many, including Givón, believed that the position of agreement markers (as prefixes or suffixes) was a clear indication of the language’s basic word order at an earlier stage (‘today’s morphology is yesterday’s syntax’). The strong version of this hypothesis is no longer tenable (Comrie 1980 a); for one thing, the positional properties of pronouns frequently differ from those of lexical noun phrases. We conclude that while it seems clear that agreement can arise from topic-shifting, this may not be the only mechanism. 3.2. Anaphoric Pronouns as the Sole Source Givón is quite specific in his claims. Besides postulating topic-shifting as the mechanism by which agreement arises, he claimed that “ verb agreement paradigms always arise from anaphoric pronoun paradigms” (1976 , 180). There is ample evidence to show that pronouns can be the source, the question is whether they are the only source. Givón pointed to the problems of the Indo-European and Semitic conjugations, and sketched possible solutions. Russell (1984) demon-

XX. Die Erforschung des Syntaxwandels III: Phänomene

1238

strates, however, that the problems with Semitic are indeed severe, and suggests that some agreement forms have non-pronominal source. Another case where a non-pronominal source for agreement markers is proposed is in Chafe’s reconstruction of the evolution of verb agreement in Iroquoian (1977). He claims that the agreement markers arose “ from a reinterpretation of affixes that were already being used for other purposes” (1977, 495). Typically forms which originally marked number were specialized to other categories within the third person (see also Mithun 1993). This analysis might cast doubt on Givón’s exclusive claim; but the question remains as to where, in the remote past, the number markers came from. Perhaps Givón’s hypothesis can be saved by the claim that originally they must have developed from pronouns; subsequent realignment of agreement markers (or re-grammaticalization) is not excluded by the hypothesis. Another account involving various reanalyses, which lead to complex patterns of agreement markers, is found in the same collection. Haas (1977) shows how in Muskog ean languages a conjugated auxiliary used with intransitives gave rise to a general set of conjugational suffixes. Haas refers to the variable part of the auxiliary as a pronoun; her account does not therefore constitute a challenge to Givón’s hypothesis since a pronominal origin to the agreement markers is implied. It is significant in that it shows how rather different verb conjugations can develop from a common pronominal origin. Although there are cases of agreement where a pronominal source seems unlikely, there are to my knowledge no clear-cut counter-examples. That is, there are no generally accepted cases of languages which had no form of agreement of the verb and which gained it via a route other than that proposed by Givón. Conversely, as mentioned earlier, the quality of data supporting the general view could be improved. Many examples actually involve the renewal of agreement systems — the development of new agreement morphology to maintain an already functioning agreement system. We would hope to be able to compare a range of instances of the genuine birth of agreement systems, that is, cases where agreement arises in languages which previously had no agreement system. If, however, we take seriously the claim that agreement and pronominalization are part of one larger phenomenon, then the real begin-

ning of agreement must be taken one stage further back, to the rise of pronouns with category distinctions of the type which can lead to agreement systems. These it has been suggested have their origins in classifiers (Greenberg 1978, 78).

4.

The Development of Other Types of Agreement

Givón dealt specifically with verb agreement. But there are several further agreement targets whose development of agreement must be accounted for. As we shall see, there are whole areas of agreement other than verb agreement where we find little more than plausible suggestions. Givón (1984, 375) suggests that head-modifier agreement develops from verb agreement, but gives no detail. Lehmann (1982, 254) states that the possibility for this is ‘severely restricted’. Greenberg (1978) shows how agreeing articles can arise from demonstratives; this appears to be a frequent occurrence. He also suggests (1978, 76 ) that phrases like the g ood one may be a source for the agreement of other noun-phrase elements. If one is a pronoun, marked for category such as gender, then adjectives with potential agreement forms arise. If these are then used redundantly even when the noun is present, then noun-modifier agreement results. A different source is suggested by evidence from the Daly languages of north-west Australia, whose significance was noted by Greenberg (1978, 51—2; 74). Comparison of languages within the group suggests that a small number of nouns have come to be used as classifiers (data from Tryon 1974). In languages of the Maranunggu subgroup, like Ami, the words for ‘meat’, ‘vegetable food’ and ‘tree/stick’, which occur as free forms, also occur as prefixes on nouns denoting animals hunted for meat (awa-wanka ‘shark’), vegetable food and plants (miya-mimi ‘round yam’) and weapons and wooden implements (yili-mitiwur ‘nullanulla’) Other nouns have no prefix. In the Brinken subgroup, the prefix is found in a reduced form in some cases, not just on the noun but also on the adjective or possessive adjective, both of which stand after the noun: for example, in Mirityabin we find yeli-meltem yeli-yikin ‘my digging-stick’. Some might argue that this is already an agreement system; others would prefer to analyse it as having repeated classifiers. A sub-

68.  Agreement

1239

sequent stage, which certainly looks like an agreement system is found in the Tyemeri group, in Ng ang ikurrung g urr for example. Here we find that agreement markers do not always match the prefix on the noun (as in Øyawul ali-kere ‘big spear’) and they may differ according to whether a possessive adjective or some other adjective is involved. These languages therefore give a clear picture of how agreement within the noun phrase can arise from classifiers. (For a more recent account see Reid 1990.) Dixon (1982, 171—3) also considers the Daly evidence. He proposes a slightly different scenario for the development of gender (class) agreement in Dyirbal (North Queensland). Dyirbal has noun markers (a type of demonstrative which indicate location and visibility). He suggests that classifiers like the widely attested mayi ‘nonflesh food’ stood between the noun marker and the noun and were subsequently reduced: mayi (7) bala NOUN → visible, there CLASSIFIER bala + m NOUN However, these noun markers can also function rather like pronouns and it could be that the classifier was first attached in this use. These sources all deserve further investigation. None of them imply that verb agreement need exist as a prerequisite. When we turn to other types of agreement the picture is even less clear. However, one unusual case can be handled readily within the scheme of agreement developing from clitic pronouns: while the verb is the most likely place for permanent attachment (giving rise to agreement) there are other possible sites. Thus in West Flemish clitics have attached to complementizers, a situation which Bennis & Haegeman (1984) consider to be agreement of the complementizer. Another type of agreement for which a pronominal source is probable is the agreement of possessive heads with possessors, as in this example from the ‘more literary level’ of Modern Hebrew (Givón 1979, 216—7): (8)

bet-o shel house-his that-to ‘Yoav’s house’

Yoav Yoav

Such constructions could plausibly arise from afterthought topics. They may in turn give rise, according to Keenan (1978, 175), to a relatively unusual type of agreement: that of prepositions and postpositions with the noun. Adpositions can arise from possessive

constructions; thus a phrase like behind it may be expressed as at back of it. If the language in question also had agreement of possessive heads with possessors, then agreement of such prepositions (or postposition) would result: at-back-his of-John → at-back+ MASC. of-John. (For agreement of the possessor see Lehmann 1982, 255—7.) A large gap in our understanding is the way in which verb agreement systems not based on topic or subject arise. In ergative agreement systems, the verb may agree with the noun in the absolutive case (the agent of intransitives and the patient of transitives). Givón (1984, 370) takes up the suggestion that such systems arise from the reanalysis of passive constructions (in an earlier nominative-accusative system). There is some justification for treating nominative-accusative constructions as more basic, in that we find languages with case marking which works according to ergativity and with agreement on the nominative-accusative model, but not the converse (Comrie 1978, 340; Moravcsik 1978, 259, 26 3). Nevertheless, ergative constructions can also develop into a nominativeaccusative system (Payne 1980, 147—8); it is therefore at least reasonable to look for an independent source for agreement systems based on ergativity. Whatever their original source, such systems can renew themselves without any appeal to reanalysis. This can be seen in Khinalug (Xinalug ), a member of the Lezgian subgroup of North-East Caucasian (data from Kibrik, Kodzasov & Olovjannikova 1972, 117—28; Kibrik personal communication; and Magometov 1976 ). Khinalug has a set of old agreement markers, which have various allomorphs conditioned by complex phonotactic constraints; Ø is a frequent allomorph, and this leads to a weakening of the system. There is also a new set of agreement markers, which are clearly based on the demonstrative/personal pronoun; they are not subject to the same range of variation, and their presence indicates tense (past or future). Both types are illustrated in the example below (cf. table 68.1). The preverb k’úr and the root qχin together form a verb meeting ‘forget’. It has an internal agreement slot, and the markers show agreement with the direct object according to the indications on the left (the Roman numerals represent the four noun classes, or genders, of Khinalug). The second root khu ‘do’ makes the verb causative. This root takes a prefixed agreement marker; it is of the same

XX. Die Erforschung des Syntaxwandels III: Phänomene

1240

Table 68.1: ‘Cause to forget’ in Khinalug (past concrete)

sg pl

I II III IV I—II III—IV

k’úr k’úr k’úr k’úr k’úr k’úr preverb

Ø s ph Ø ph Ø AG

qχin qχin qχin qχin qχin qχin root

-

Ø s Ø Ø Ø Ø AG

khu khu khu khu() khu khu() root

type as the previous one, though the allomorph is different in some instances (it depends on the following consonant). After the second root and before the indicative marker (a form of ‘be’) there is a third agreement marker, of the new type, clearly derived from the demonstrative/personal pronoun. Besides marking agreement, it also indicates the tense (past definite). The important point is that these new markers behave just like the old ones in that they mark agreement with the same noun phrase — the one in the absolutive case, which in this instance will be the direct object. Thus an agreement system based on ergativity can renew itself without recourse to reanalysis of a different construction. Lehmann (1982, 253—4) implies that ergative agreement systems develop directly from pronouns, though he does not give an explicit mechanism for the development. He does, however, incorporate ergative type agreement in a hierarchy giving the steps of evolution of what he calls ‘external agreement’.

Fig. 68.1: External agreement (Lehmann 1982,

254) A language can acquire agreement forms only along downward paths from the top.

5.

The Loss of Agreement

Givón states that “ Agreement systems meet their predictable demise via phonological

-

d - mæ dæ - mæ dæ - mæ ʒ - mæ dur - mæ - mæ ʒith AG/tenseINDIC

attrition much like other bound affixes” (1976 , 172). But there are patterns in the loss of agreement systems which are much more interesting and illuminating than this quotation suggests. Attrition may give rise to a situation in which some agreement targets of a particular type mark agreement while others do not. The loss of French ‘mute e’ means that some adjectives agree in gender: court (masculine singular) ‘short’ [ku:r] courte (feminine singular) [ku:rt], while others, in the spoken language, do not: vrai masculine singular ‘true’, vraie (feminine singular) pronounced identically [vrε]. The situation is often not so straightforward. Some Lower Cross languages have a small set of adjectives which retain number agreement: generally the same adjectives are involved in different languages, but it is not clear what factor determines whether an adjective will agree or not. A necessary condition is that they precede the noun (Faraclas 1986 , 48; Connell 1987, 121). And in the Caucasus we find great diversity between related languages as to the number of adjectives which mark agreement. More significantly, different types of agreement targets lose agreement at different times. Again in Cross River languages, numerals and adjectives lose agreement first, while it is preserved in verbs and pronouns (Demuth, Faraclas and Marchese 1986 , 459; Connell 1987, 121). More detailed data are available on nine Kru languages (Marchese 1988). Here again agreement within the noun phrase is being lost while being preserved outside it. Numerals do not show agreement — this probably represents loss of agreement. Within the noun phrase, adjectives lose agreement in these languages before determiners, and adjectives preserve agreement in number longer than agreement in gender (class). Priestly (1983, 343—6 ) provides an interesting comparison: he is concerned with the gender agreement in Indo-European and shows how it is usually preserved best in the pronoun and less well in the adjective. There

68.  Agreement

is fruitful work to be done in this area, investigating other language groups to find the order in which different types of target lose agreement, the order in which different agreement categories (gender, number, person, case) are lost, and the factors which determine which items within a target type (e. g. which attributive adjectives) will retain agreement longest. Even at the level of the loss of a single agreement opposition within a particular target type there may be a great deal going on. In a detailed study Naro (1981) shows how the loss of verb agreement in spoken Brazilian Portug uese starts from a low level phonetic rule, which makes third singular and third plural identical for one class of verbs, but that then non-agreement diffuses through the system, mainly in environments where the change is least noticeable in surface form.

6.

Finer-Grained Changes in Agreement Systems

There are various types of change which, though not as dramatic as the birth or death of agreement, affect the system and by doing so give insights into how it operates. Harris (1985, 395—412) considers six changes which involve alternations in agreement markers in four Kartvelian languages. Most represent complications of the grammar, but reinforce other changes relating to verb type. Payne (1980) documents the unification of agreement markers from different sources in the eastern Iranian Pamir languages. Comrie (1980 b) considers the relation of animacy and voice to agreement, and (1984, 19—23) investigates change in the possessive construction in various Arabic dialects. Possessives are particularly prone to reanalysis (Givón 1976, 175—8). Several constructions regularly cause problems for agreement and so are often involved in change. Conjoined noun phrases frequently allow agreement either with just one conjunct (usually that nearest to the target but sometimes the first even when not nearest) or with all the conjuncts. There is then an agreement option: agreement with one conjunct represents strict or syntactic agreement, while agreement with all conjuncts (involving resolution of gender, number, person — whichever are involved) is agreement with a higher degree of semantic justification. Comitative expressions (like

1241

Russian Ivan s Petrom ‘Ivan with Peter’) allow a similar option. And quantified expressions may be problematic: for example, in the Russian phrase pjat’ knig ‘five books’, pjat’ is not marked for gender and number, while knig is in the genitive plural, and controllers of verb agreement in Russian must be nominative. Both singular and plural agreement are possible. Certain types of lexical item also give rise to agreement options: again these are cases where meaning and form clash. German Mädchen ‘girl’ is feminine in meaning, but has the neuter diminutive suffix -chen: the anaphoric pronoun may be neuter or feminine. Collectives too may have plural meaning but singular morphology; thus Serbo-Croat deca ‘children’ declines like a singular noun but controls singular and plural agreement. While agreement options are often found with the constructions and lexical items of the types listed, the choice is severely constrained. From the point of view of the agreement controller, controllers denoting animates and those preceding the target are more likely to take agreement forms with a higher degree of semantic justification than those which denote inanimates and/or follow the controller. One possibility for change is that the importance of animacy and precedence relative to each other may vary over time (Corbett 1983, 125—31). And, contrary to some suggestions, there is no necessity that change should be in favour of semantically justified agreement forms: one form may gain over time only to lose again (Corbett 1988, 29—30; 50). From the point of view of the target, the distribution of alternative agreement forms is constrained by the Agreement Hierarchy: attributive < predicate < relative pronoun < personal pronoun. As we move rightwards along the hierarchy, so the likelihood of agreement forms with greater semantic justification will increase monotonically (see Corbett 1991, 225—6 0 for data). We would predict that change favouring forms with greater semantic justification would proceed from the right of the hierarchy. This has happened, for example with Russian words like vrač ‘doctor’, which though they can denote a female remain formally like masculines. The personal pronoun is normally feminine (though the masculine is not completely excluded), the relative almost always feminine; the feminine (semantic

XX. Die Erforschung des Syntaxwandels III: Phänomene

1242

form) is also common in the predicate, though the masculine is also found, and the feminine is beginning to make headway even in attributive position (though here the masculine is the more likely). A development which appears to have spread along the hierarchy in the opposite direction is found in some southern Polish dialects (Zaręba 1984— 85). Hypochoristics and patronymics used for girls and unmarried women (like Zusię ‘Zuzia’ follow a declension whose nouns are usually neuter. Neuter agreements have spread right along the hierarchy so that girls and unmarried women are referred to with neuter pronouns, and they use the first singular pronoun with neuter agreements to refer to themselves. In these dialects the effect is considerable; the meaning of the genders has been changed. The feminine is now for married females, the neuter for unmarried females (both also include inanimates). Thus agreement fluctuations with constructions or particular lexical items may remain as minor irregularities, but they may spread to give rise to more significant changes, as in the Polish dialects just discussed.

7.

Conclusion

Givón’s hypothesis according to which verb agreement arises from pronouns through topic-shifting has considerable evidence in its favour and is widely accepted. As we have seen, however, there is some question as to whether the exclusive claims of the hypothesis are valid: it is not at all clear that verb agreement arises only in this way, nor whether topic agreement is always involved. Part of the problem is that some evidence concerns the renewal of agreement systems and it is not self-evident that pristine cases of the rise of agreement necessarily occur in the same way. Agreement of targets other than predicate verbs is a major problem. For agreement within the noun phrase classifiers are a likely source. Our understanding of the relation between agreement categories (notably number, gender and person) is patchy. Thus a promising start has been made, and evidence on the loss of agreement is helping to fill in the picture. But we require data on more languages before we can properly complete or replace Givón’s hypothesis.

8.

References

Barlow, Michael. 1988. A Situated Theory of Agreement. Doctoral dissertation, Stanford University. (Published 1992, New York.)

Bennis, Hans and Liliane Haeg emann. 1984. On the status of agreement and relative clauses in WestFlemish. Sentential Complementation: Proceedings of the International Conference held at UFSAL, Brussels June, 1983 (Linguistic Models 5), ed. by W. de Geest and Y. Putseys, 33—53. Dordrecht. Bresnan, Joan and Sam Mchombo. 1986 . Grammatical and anaphoric agreement. CLS 22: Part 2: Papers from the Parasession on Pragmatics and Grammatical Theory at the Twenty-Second Regional Meeting, Chicago Linguistic Society, ed. by A. M. Farley, P. T. Farley and K.-E. McCullough, 278—297. Chicago. —. 1987. Topic, pronoun, and agreement in Chichewa. Language 63. 741—742. Chafe, Wallace L. 1977. The evolution of third person verb agreement in the Iroquoian languages. Mechanisms of Syntactic Change, ed. by C. N. Li, 493—524. Austin. Claudi, Ulrike. 1985. Zur Entstehung von Genussystemen: Überlegungen zu einigen theoretischen Aspekten, verbunden mit einer Fallstudie des Zande. Hamburg. Comrie, Bernard. 1978. Ergativity. Syntactic Typology: Studies in the Phenomenology of Language, ed. by W. P. Lehmann, 329—394. Sussex. —. 1980a. Morphology and word order reconstruction: problems and prospects. Historical Morphology (Trends in Linguistics, Studies and Monographs 17), ed. by J. Fisiak, 83—96. The Hague. —. 1980b. Agreement, animacy and voice. Wege zur Universalienforschung: Sprachwissenschaftliche Beiträge zum 6 0. Geburtstag von Hansjakob Seiler, ed. by G. Brettschneider and C. Lehmann, 229—234. Tübingen. —. 1984. Agreement as a research tool ESCOL ’84: Proceedings of the First Eastern States Conference on Linguistics, ed. by G. Alvarez, B. Brodie and T. McCoy, 13—26. Columbus, Ohio. Connell, Bruce. 1987. Noun classification in Lower Cross. Journal of West African Languages 17. 110—125. Corbett, Greville G. 1979. The agreement hierarchy. Journal of Linguistics 15. 203—224. —. 1983. Hierarchies, Targets and Controllers: Agreement Patterns in Slavic. London. —. 1988. Agreement: a partial specification, based on Slavonic data. Agreement in Natural Language: Approaches, Theories, Descriptions, ed. by M. Barlow and C. A. Ferguson, 23—53. Stanford. —. 1991. Gender. Cambridge. Demuth, Katherine, Nicholas Faraclas and Lynell Marchese. 1986 . Niger-Congo noun class and agreement systems in language acquisition and historical change. Noun Classes and Categorization: Proceedings of a Symposium on Categorization and Noun Classification, Eugene, Oregon, October 1983, ed. by C. Craig, 453—471. Amsterdam.

68.  Agreement

Dixon, Robert M. W. 1982. Where Have All the Adjectives Gone? and other essays in Semantics and Syntax. Berlin. Duranti, Alessandro and Elinor Ochs. 1979. Leftdislocation in Italian conversation. Discourse and Syntax, ed. by T. Givón, (Syntax and Semantics 12), 377—416. New York. Faraclas, Nicholas. 1986 . Cross River as a model for the evolution of Benue-Congo nominal class/ concord systems. Studies in African Linguistics 17. 39—54. Foster, Joseph F. and Charles A. Hofling . 1987. Word order, case, and agreement. Linguistics 25. 475—499. Givón, Talmy. 1971. Historical syntax and synchronic morphology: an archeologist’s field trip. CLS 7: Papers from the Seventh Regional Meeting, Chicago Linguistic Society, April 16 —18, 1971, 394—415. Chicago. —. 1976 . Topic, pronoun and grammatical agreement. Subject and Topic, ed. by C. N. Li, 149—188. New York. —. 1979. On Understanding Grammar. New York. —. 1984. Syntax: A Functional-Typological Introduction: Volume I. Amsterdam. Greenberg , Joseph H. 1978. How does a language acquire gender markers? Universals of Human Language: III: Word Structure, ed. by J. H. Greenberg, C. A. Ferguson and E. A. Moravcsik, 47—82. Stanford. Haas, Mary, R. 1977. From auxiliary verb phrase to inflectional suffix. Mechanisms of Syntactic Change, ed. by C. N. Li, 525—537. Austin. Hale, Kenneth. 1973. Person marking in Walbiri. A Festschrift for Morris Halle, ed. by S. Anderson and P. Kiparsky, 308—344. New York. Harris, Alice C. 1985. Diachronic Syntax: The Kartvelian Case (Syntax and Semantics 18). Orlando. —. forthcoming. Ergative-to-accusative shifts in agreement in Lezghian languages. To appear in Grammatical Categories and Grammaticalization ed. by N. Vincent and S. Romaine. London. Hyman, Larry M. and Alessandro Duranti. 1982. On the object relation in Bantu. Studies in Transitivity ed. by P. J. Hopper and S. A. Thompson (Syntax and Semantics 15), 217—239. New York. Keenan, Edward L. 1978. On surface form and logical form. Linguistics in the Seventies: Directions and Prospects, ed. by B. B. Kachru (= Studies in the Linguistic Sciences 8, 2), 16 3—203. Urbana, Ill: Department of Linguistics, University of Illinois. Kibrik, A. E., S. V. Kodzasov and I. P. Olovjannikova. 1972. Fragmenty grammatiki xinalugskogo jazyka (Publikacii otdelenija strukturnoj i prikladnoj lingvistiki 9). Moscow. Lehmann, Christian. 1982. Universal and typologi-

1243

cal aspects of agreement. Apprehension: Das sprachliche Erfassen von Gegenständen: II: Die Techniken und ihre Zusammenhänge in Einzelsprachen, ed. by H. Seiler and F. J. Stachowiak, 201—267. Tübingen. —. 1988. On the function of agreement. Agreement in Natural Language: Approaches, Theories, Descriptions, ed. by M. Barlow and C. A. Ferguson, 55—65. Mag ometov, A. A. 1976 . Sistema grammatičeskix klassov v xinalugskom jazyke. Annual of IberoCaucasian Linguistics 3. 232—59. Marchese, Lynell. 1988. Noun classes and agreement systems in Kru: a historical approach. Agreement in Natural Language: Approaches, Theories, Descriptions, ed. by M. Barlow and C. A. Ferguson, 323—341. Mithun, Marianne. 1991. The development of bound pronouns. Language Typology 1988: Typological Models in Reconstruction (Current Issues in Linguistic Theory 81), ed. by W. P. Lehmann and H.-J. J. Hewitt, 85—104, Amsterdam. —. 1993. Reconstructing the unidentified. Historical Linguistics 1989: Papers from the 9th International Conference on Historical Linguistics (Current Issues in Linguistic Theory 106 ), ed. by H. Aertsen and R. J. Jeffers, 329—347. Amsterdam. Moravcsik, Edith A. 1974. Object-verb agreement. Working Papers on Language Universals 15. 25— 140. —. 1978. On the distribution of ergative and accusative patterns. Lingua 45. 233—279. Naro, Anthony J. 1981. The social and structural dimensions of a syntactic change. Language 57. 63—98. Payne, John R. 1980. The decay of ergativity in Pamir languages. Lingua 51. 147—186. Priestly, Tom. M. S. 1983. On ‘drift’ in Indo-European gender systems. Journal of Indo-European Studies 11, pt. 3—4. 339—363. Reid, Nicholas J. 1990. Ngan’gityemerri: A Language of the Daly River Region, Northern Territory of Australia. Unpublished PhD thesis, Australian National University, Canberra. Russell, Robert A. 1984. Historical aspects of subject-verb agreement in Arabic. ESCOL ’84: Proceedings of the First Eastern States Conference on Linguistics, ed. by G. Alvarez, B. Brodie and T. McCoy, 116—127. Columbus, Ohio. Steele, Susan. 1977. Clisis and diachrony. Mechanisms of Syntactic Change, ed. by C. N. Li, 539— 579. Austin. —. 1978 Word order variation: a typological study Universals of Human Language: IV: Syntax, ed. by J. H. Greenberg, C. A. Ferguson and E. A. Moravcsik, 585—623. Stanford. Tryon, D. T. 1974. Daly Family Languages, Australia (Pacific Linguistics C 32). Canberra.

XX. Die Erforschung des Syntaxwandels III: Phänomene

1244

Wald, Benji. 1979. The development of the Swahili object marker: a study of the interaction of syntax and discourse. Discourse and Syntax, ed. by T. Givón, (Syntax and Semantics 12), 505—24. New York.

Zaręba, Alfred. 1984—85. Osobliwa zmiana rodzaju naturalnego w dialektach polskich. Zbornik Matice srpske za filologiju i lingvistiku 17—18. 243—247.

Greville G. Corbett, Surrey (Great Britain)

69. Tense and Aspect 1. 2. 3. 4.

1.

Theoretical Background Historical Change in Tense and Aspect Conclusions References

Theoretical Background

In order to understand the mechanisms of diachronic change as they apply to tenseaspect systems, it is necessary to have an understanding of the nature of a synchronic tense-aspect system. The summary given below follows the main lines of Comrie (1976 , 16 —40, 52—6 5; 1985, 1—101). Both tense and aspect refer to grammatical categories, and must be sharply distinguished from lexical categories that express similar cognitive oppositions. 1.1. Tense Tense is the grammaticalized expression of location in time. The best-known tense systems are those in which situations are located in time relative to the present moment, giving rise to the traditional trichotomy of Present (including the present moment, e. g. English he sing s), Past (located prior to the present moment, e. g. English he sang ), and Future (located after the present moment, e. g. English he will sing ). Whether the Future is properly a tense or a mood is a point of controversy both in general linguistic theory and in the analysis of particular categories in individual languages, including the English periphrastic construction with will; this point will recur in section 2.4.1. Tense using the present moment as its deictic enter in this way is called absolute tense. Relative tense takes a contextually given reference point as its deictic center, giving rise to, for instance, Relative Present (Simultaneous), Relative Past (Anterior), Relative Future (Posterior). In English, many nonfinite verb forms have relative time reference, so that the Present Participle in the patients waiting for the doctor were then al-

lowed to enter is taken as simultaneous with the contextually given reference point of when the patients were allowed to enter. Semantically more complex tenses combine absolute and relative time reference; they are traditionally subsumed under relative tenses, though a better term might be absolute-relative tenses. For instance, the Pluperfect, as in Alfred had already arrived when Betty sang , locates Alfred’s arrival before a contextually given reference point (Betty’s singing) which is itself located before the present moment (anterior relative to the past); the Future Perfect, as in Charles will have arrived before Donna sing s, locates Charles’s arrival before a contextually given reference point (Donna’s singing) which is itself located after the present moment (anterior relative to the future). A number of languages distinguish in their tense systems not only whether a situation is to be located before or after the present moment (or other reference point), but also how remote from the present moment the situation is located; thus Haya, a Bantu language of Tanzania, has three past tenses: Today Past twákôma [‘we tied up (earlier today)’], Yesterday Past tukomíle [‘we tied up (yesterday)’], Remote Past túkakôma [‘we tied up (before yesterday)’] (Ernest Byarushengo, personal communication). 1.2. Aspect Aspect refers to the grammaticalized expression of the internal temporal contour of a situation. A primary distinction is made between Perfective and Imperfective aspect. Perfective aspect treats a situation as being bounded in time, i. e. as an event. Imperfective aspect treates a situation as being unbounded in time, i. e. as a state or process. Russian on pročital knig u [‘he read (Perfective) the book’] refers to a single, complete event of reading the book, including the beginning and end of this event, whereas on čital knig u [‘he used to read/was reading (Imperfective) the book’] refers to an ongoing

69.  Tense and Aspect

situation without specification of its bounds (beginning or end). The French Passé Simple/ Imparfait distinction is semantically one of Perfective versus Imperfective aspect, though restricted to the past tense, so a similar opposition is provided by il lut le livre versus il lisait le livre. Important subcategories of Imperfective are Habitual and Progressive. Habitual refers to a situation that is prolonged or iterated through so long a period of time that it becomes a characteristic attribute, as in English Edmund used to drink tea every day; being a daily tea-drinker was a characteristic of Edmund even at those times during the day when he was not actually drinking tea. Progressive refers to the combination of (non-Habitual) Imperfective aspect with dynamic (as opposed to stative) semantics, i. e. refers to an ongoing process, as in English Fanny was writing a letter. Another category that is usefully discussed under aspect is the Perfect, although its precise classification under tense, aspect, and other verbal categories is controversial. Perfect is distinguished from non-Perfect in that the former refers explicitly to the continuing relevance of an earlier situation at a later point. The English Present Perfect, as in Georg e has broken his leg , refers to the continuing relevance at the present moment of the past event of George’s breaking his leg. In dealing with aspect it is particularly important to distinguish grammatical and lexical expressions of the internal temporal contour of a situation; such lexical expressions are often referred to as aktionsarten, though this term is also used in other senses. The distinction between drink and drink up in English, or between trinken and austrinken in German, is not one of Imperfective versus Perfective, since the latter can be combined with explicit markers of Imperfective aspect, e. g. the English Progressive in Henrietta is drinking up the cocoa. The aktionsart represented by drink up is often referred to as telic: a telic situation is one which, if allowed to proceed, will eventually reach an end-point, whereas an atelic situation may in principle continue indefinitely; but a telic lexical item or expression can still be used imperfectively, thus excluding reference to this built-in endpoint. The distinction between telic and Perfective will recur in section 2.3.2.

2.

Historical Change and Tense-Aspect

The historical development of tense-aspect

1245

systems has been examined in detail in relatively few languages. The examples presented below should therefore be taken as illustrations of the kinds of changes that are well documented in this area, though by no means as exhaustive, given the rather small sample on which we have had to draw. 2.1. Changes Within Aspect The main change to be investigated in this section is the semantic extension of Progressive to encompass the whole of Imperfective. One other change that deserves mention is the shift of meaning from Perfect to Perfective; since this change involves, in some languages, a greater shift, namely from Perfect to Past, i. e. from aspect to tense, the relevant materials are presented in section 2.4.2. 2.1.1. From Progressive to Imperfective The shift from Progressive to Imperfective involves the loss of the constraint on the Progressive that it applies only to dynamic verbs. This change is well illustrated by the comparison of the Irish and Scots Gaelic forms containing the verb ‘be’ as auxiliary followed by the preposition ag [‘at’] and a verbal noun. In Irish, this is semantically a Progressive, as seen in the contrast between tá sé ag dúnadh an dorais [‘he is closing the door’], literally ‘is he at closing the door-GENITIVE’, and dúnann sé an doras [‘he closes the door’]. (The latter would be used, for instance, in a habitual interpretation.) In Scots Gaelic, by contrast, the periphrastic form is used to cover the whole of this semantic area, so that tha e a’ dunadh an doruis corresponds to both ‘he is closing the door’ and ‘he closes the door’. If this kind of extension of the semantic range of an originally Progressive form is gradual, one might expect to find intermediate cases, and it is possible that Modern English represents such an intermediate case. In comparison with the canonical definition of Progressive aspect and with other languages that have a distinct category of Progressive aspect, the semantic range of the English Progressive is wide, encompassing, for instance, lexically stative verbs used dynamically (e. g. Ian is being g ood) and verbs with restricted habitual interpretation (i. e. expressing a habit which is restricted in terms of the period, of time to which the habit applies, e. g. Janet is skating more often these days). Generally, where we find variation between dialects or closely related languages in the use of a given form, here as Progressive, there as Imperfective, this is good prima facie

1246

evidence that the latter is the result of a shift from Progressive to Imperfective. For instance, for some speakers of Yoruba, ó ń-lọ can be freely interpreted as ‘he is going’ or ‘he goes (is in the habit of going)’, while for others it has primarily only the former interpretation (Rowlands 1969, 60—61). 2.2. Changes Within Tense Most of the well attested diachronic changes involving tense concern the development of tense out of other categories, and are therefore treated in section 2.4. In this section, only one kind of change is examined. 2.2.1. From Relative to Absolute Tense A shift of the meaning of tense categories from relative to absolute tense is documented for the historical development from Classical Arabic to vernacular Arabic, here represented by Maltese. In Classical Arabic (Wright 1898, 1—22), there is a basic morphological distinction between two sets of forms, traditionally called Perfect and Imperfect, though since these terms are thereby used in a rather idiosyncratic way we will replace them by the arbitrary I and P. The distinction between I and P combines aspect and relative tense: P indicates Perfective aspect combined with Relative Past tense; I covers all other possibilities, i. e. all of Imperfective aspect irrespective of time reference and all of Relative non-Past (Present, Future) irrespective of internal temporal contour. Crucially, in Classical Arabic the I form can be used with past time reference, though only where there is independent indication of the past time reference, e. g. by an adverbial. (This restriction is presumably pragmatic in origin: the I form is taken to have non-past time reference unless this inference is cancelled.) This is illustrated by wa ’ttabaū (P) mā tatlū (I) ’š-šayāṭīnu alā mulki sulaymāna [‘and they followed (P) what the demons used to recite (I) in Solomon’s reign’], where the crucial item is the I form tatlū with past time reference. Conversely, the P form can be used with future time reference, provided the context gives some reference point even further into the future, as in ajīu-ka (I) iðā ’ḥmarra (P) ’l-busru [‘I will come (I) to you when the unripe date ripens (P)’], where the crucial item is the P form ’ḥmarra, referring to the future ripening of the dates which will, however, precede the speaker’s coming. In Classical Arabic, if it is necessary to mark the absolute time reference of one of these verb forms, this can be done

XX. Die Erforschung des Syntaxwandels III: Phänomene

by attaching the auxiliary verb ‘be’, which distinguishes Past kāna, Present yakūnu, Future sa-yakūnu: thus yaktubu (I) is interpreted out of context as ‘he writes/is writing’, but kāna yaktubu is unequivocally ‘he was writing/used to write’; kataba (P) is interpreted out of context as ‘he wrote’, but sa-yakūnu kataba is ‘he will have written’. In Maltese, the descendent of the P form is unequivocally a Past tense, e. g. kiteb [‘he wrote’], while the descendent of the I form is unequivocally non-Past, e. g. jikteb [‘he writes’], se jikteb [‘he will write’]. Where in Classical Arabic the auxiliary ‘be’ could optionally be added to give explicit indication of a reference point other than the present moment, in Maltese inclusion of the auxiliary is obligatory for this interpretation, giving rise to compound tenses such as kien jikteb [‘he used to write’], sa jkun kiteb [‘he will have written’]. Thus, the relative tense of Classical Arabic has been reinterpreted as absolute tense. 2.3. Origins of Aspect While in many instances we have no idea of how a particular aspectual opposition arose in a given language, there are several instances where either direct historical attestation, comparison with related dialects or languages, or internal reconstruction (in terms of plausible etymologies) allows us to show an aspectual opposition arising out of some other linguistic element. In well documented examples, aspect arises not from other verbal categories, but rather from distinct constructions (section 2.3.1.) or from lexical categories (sections 2.3.2.—3.). 2.3.1. From Locative to Progressive In a surprisingly large number of languages, Progressive aspect has a locative origin, i. e. ‘he is working’ deriving from a construction like ‘he is in/at work(ing)’, i. e. in the process of working (Comrie 1976 , 98—106 , and more generally Traugott 1978); even in the history of English there was, for a while, a construction of this type, namely he is a’ working , with a’ a reduced form of at. One example of such a construction has already been introduced in section 2.1.1., namely the Irish Progressive tá sé ag dúnadh an dorais [‘he is closing the door’], more literally ‘he is at [the] closing of the door’. Essentially the same construction, but using ‘in’, is found in Welsh, e. g. y mae ef yn canu [‘he is singing’], literally ‘PARTICLE is he in singing’. (In Welsh, as in Scots Gaelic, this has general

69.  Tense and Aspect

Imperfective rather than just Progressive meaning, by the extension discussed in section 2.1.1.). Further afield, locative expressions are used for Progressive aspect in a number of African languages (Welmers 1973, 311—312, 314—315, 323—324), e. g. Yoruba, e. g. ó ń-lọ [‘he is going’], literally ‘he in go’, and Shona, in the form of a verb ri meaning ‘be in’, e. g. tì-rí kù-fúndá ChìMànyíkà [‘we are studying ChiManyika [a dialect of Shona]’], literally ‘we-be:at INFINITIVEstudy ChiManyika’. In Mandarin, one way of indicating Progressive aspect is by using zài [‘in’] before the verb, e. g. tā zài niàn shū-ne [‘he is reading’], literally ‘he in read bookPARTICLE’. Some varieties of Georgian can use the postposition -ši [‘in’] to form Progressives, e. g. Soso era-ši ikneba [‘Soso will be writing’], literally ‘Soso writing-in will-be’ (Tschenkéli 1958, 83—84). Hindi uses the verb ‘remain’ in its Progressive, e. g. my cǝl rǝha hũ [‘I am going’], literally ‘I go havingremained am’. And coming back to northwestern Europe, Icelandic forms its Progressive with the preposition að [‘at’], e. g. ég er að lesa [‘I am reading’]. In some languages, locative expressions with parallel meaning are not fully grammaticalized, as in the Dutch construction hij is aan het tuinieren [‘he is gardening’], literally ‘he is at the gardening’. Even where the form of the Progressive is clearly of locative origin, this does not mean that the two constructions are to be identified synchronically, and in some instances differentiation has been introduced between locative and Progressive aspect expressions. In Welsh, for instance, the preposition yn [‘in’] nasalizes a following consonant, while the yn used in the expression of Progressive aspect does not; thus, the citation form Caerdydd [‘Cardiff’] corresponds to the locative yng Ng haerdydd [‘in Cardiff’], while the citation form canu [‘singing’] corresponds to yn canu, as in the example cited earlier in this section. Sometimes, the locative expression is used not just for Progressive aspect, but more generally for Imperfective aspect. In many cases, including some of those discussed in this section, this reflects the superimposition of the Progressive to Imperfective shift (section 2.1.1.) on a Progressive of locative origin. However, conceptually, just as one can ‘be in’ a process, so one can also ‘be in’ a state. In both Irish and Scots Gaelic, for instance, certain stative verbs are constructed localisti-

1247

cally, e. g. Irish tá sé ina shuí, Scots Gaelic tha e ’na shuidhe [‘he is sitting’], literally ‘is he in-his sitting’. 2.3.2. From Telic to Perfective Although telic aktionsart and Perfective aspect are conceptually distinct, there are nonetheless close relations between them which manifest themselves in the historical development from telic to Perfective, especially in the Slavic languages. One way of forming Perfective aspect in the Slavic languages, exemplified here by Russian, is by prefixing a simple verb that is otherwise Imperfective, e. g. čitat’ [‘to read (IMPERFECTIVE)’], pro-čitat’ [‘to read (PERFECTIVE)’] (Forsyth 1970, 18— 20). For most such simple Imperfective verbs, there is one prefix that changes the lexical meaning minimally or not at all, while other prefixes also change the lexical meaning in addition to assigning Perfective aspect, e. g. pere-čitat’ [‘to re-read’], s-čitat’ [‘to count’]. Formally, these prefixes are very similar to the particle of English drink up or the prefix of German austrinken, and the general construction (though not necessarily individual prefixes/particles) have the same etymology within Indo-European. Even in an example like pro-čitat’, the prefix can be identified with a prefix elsewhere having the lexical meaning ‘through’, as in pro-lezt’ [‘to climb through’], cf. lezt’ [‘to climb’]. In the earliest Slavic texts, such prefixes did not necessarily cause perfectivization, but had a purely lexical function. By the modern Slavic languages, these prefixes have been fully integrated into a Perfective/Imperfective opposition (which has a number of other formal exponents in addition to prefixing, such as suffixing to form Imperfectives from otherwise Perfective verbs, e. g. pere-čit-yvat’ [‘to re-read (IMPERFECTIVE’] (Forsyth 1970, 27—28)). One trace of the former telic aktionsart is that the Slavic Perfective still necessarily implies the reaching of the natural endpoint of the action in question, i. e. on pro-čital knig u [‘he read the book’] implies that he finished the book (Dahl 1985, 73—79). Many other languages with a Perfective/Imperfective opposition do not impose this restriction on the interpretation of their Perfective, e. g. French il lut le livre [‘he read the book’] can be interpreted, especially in a context favorable to this interpretation, to mean that he ‘had a read of’ the book, without necessarily finishing it.

1248

2.3.3. From Lexical Item to Perfective

XX. Die Erforschung des Syntaxwandels III: Phänomene

While many tenses found in languages have no obvious etymology, it is rather more common for tense than for aspect to be able to point to an origin either somewhere else in the verbal system or elsewhere in the language.

reference will had its origin as a verb meaning ‘want’. One of the Future tenses in Rumanian has a similar origin, namely voi lucra [‘I will work’], literally ‘I-want to-work’, where voi is etymologically a form of the verb ‘to wish’. Often, once the form in question becomes primarily an expression of future time reference, its form changes (is abbreviated) from that found in modal use; the form of the auxiliary found in Rumanian voi lucra is now distinct from the main verb vreau [‘I want’]. The Modern Greek Future θa rápso [‘I will write’] has the same etymological origin as θélo na γrápso [‘I want to write’], literally ‘Iwant that I-write’, but in the former the verb θélo (and its various other conjugational forms) and the particle na have coalesced into the invariable particle θa. Outside IndoEuropean, a desiderative origin of Future tense is clearly established for the Swahili Future tense prefix ta-, cf. taka- [‘want’] (Givon 1971, 148—149). While ‘want’ is perhaps the most widely cited modal origin for Future tense, it is not the only one. The Vulgar Latin Future formed by adding the verb ‘to have’ after the Infinitive was in origin a modal expression of obligation; thus Spanish cantar-é (< cantare habeo) was originally ‘I have to sing’ (Fleischman 1982). In Sardinian, Latin debere [‘must’] likewise provides a Future tense with its origin in a debitive construction (i. e. one expressing obligation).

2.4.1. From Modal to Future

2.4.2. Perfect to Past

As noted in section 1.1., there is considerable controversy over whether Future tense should be recognized as such, or whether future time reference should rather be considered a mood (either in general, or at least in particular languages). Aprioristically, both positions seem defensible: future time reference, if grammaticalized, clearly satisfies the definition of tense; but the future is necessarily less certain than the present or the past, and uncertainty is a modal notion. Without entering into the general debate here, we will rather concentrate on some examples where a given morpheme has changed diachronically from being a marker of some mood (other than future time reference) to being primarily, or at least to a much greater extent, a marker of future time reference (Ultan 1978, 112—114). One cross-linguistically common source of future tense markers is the auxiliary ‘want’, a development that is attested in English, where the basic marker of future time

The long attested history of the Indo-European languages provides a number of instances of development from Perfect to Past, or at least to Perfective Past, often as part of a longer chain that has its origin in a stative form rather than a strict Perfect. For Ancient Greek (Perel’muter 1988), there is evidence that the Perfect had only recently been grammaticalized as part of the verbal paradigm. In Homeric Greek, the (later) Perfect was primarily a stative or resultative form, and many transitive verbs are used intransitively in this form, e. g. Present peíthō [‘I persuade’] but Perfect pépoitha [‘I believe’] (i. e. am in a state of having been persuaded). This finds a natural explanation if these forms were originally lexical stative derivatives: it is hard to conceive of a stative that would involve both agent and patient, as the transitive verb peíthō does, but easy to imagine a lexical relation between this transitive verb and a stative derivative that would express only the state of the patient. Originally then, such forms expressed a present state, and only subse-

A number of claims have been made about markers of Perfective aspect in various languages deriving from lexical items with such meanings as ‘finish’, but the data here are unfortunately less secure than those discussed in the rest of this article, so they must be treated with greater caution. Perhaps the clearest example is the Mandarin Perfective marker, the verbal suffix -le (not to be confused with the homophonous sentence suffix -le, which seems rather to be a marker of current relevance), whose etymology seems to be the verb liǎo [‘finish’] (Chao 196 8, 246 ). A number of pidgins and creoles use reflexes of English finish (or Portuguese acabar [‘to finish’], etc.) in a way that is somewhat similar to Perfective aspect, e. g. Tok Pisin, the English-based pidgin of Papua New Guinea, em i g o pinis [‘he has gone’], literally ‘he PARTICLE go finish’, but the precise semantic characterization of these constructions is not always clear. 2.4. Origins of Tense

69.  Tense and Aspect

quently was more attention focused on the earlier situation giving rise to this present state. Thus, we have an initial development from stative (focus only on current state) to resultative (focus on current state, but as result of earlier situation) to Perfect (focus on earlier situation, but with current relevance). A similar development underlies the development of the Romance Perfect with ‘have’ (Salvi 1982). In Latin, habeo litteras scriptas originally meant ‘I have letters in a state of writtenness’, where the Present tense of ‘to have’, namely habeo, is to be taken literally. By semantic shift, greater focus came to be placed on the writing, so that by we get to the Spanish construction he escrito las cartas [‘I have written the letters’] there is no longer any implication that I still have the letters, rather the crucial points are that the letters were written and that this still has continuing relevance. This forms a preamble to the discussion of the development from Perfect to (Perfective) Past. A number of Indo-European languages, in particular, show a gradual semantic change whereby an earlier perfect comes to be reinterpreted as a Past tense, or sometimes as a Perfective Past tense (where there is already a distinct Imperfective Past tense, such as the Romance Imperfect). The Romance development (Harris 1982) is particularly interesting here, because of the wealth of historical attestation and the number of languages and dialects involved. At the earliest relevant stage, a distinction is made between Perfect (e. g. Spanish he cantado, French j’ai chanté, Italian ho cantato [‘I have sung’] and Preterite (e. g. Spanish canté, French je chantai, Italian cantai [‘I sang’]) that corresponds to that established in section 1 between Present Perfect and Perfective Past. Castilian Spanish is still close to this, but with one innovation: the Perfect, rather than the Preterite, is used for situations that held earlier today, even if they no longer have continuing relevance; French seems to have passed through this stage, with the Perfect also encoding recent past meaning, in the seventeenth century. The most extreme development is found, for instance, in spoken French and in many northern varieties of Italian, where the Preterite has been completely replaced by the Perfect (which is therefore more properly called Passé Composé, since it is no longer semantically a Perfect): the component of current relevance has been completely lost from this form. A similar development is found in

1249

southern varieties of German, though here, because of the absence of any original Perfective/Imperfective distinction in the Past tense, the earlier Perfect takes over as the only Past tense. In most Slavic languages, including Czech, which is used as the illustration here, the Aorist (Perfective Past) and Imperfect (Imperfective Past) inherited from ProtoSlavic have been replaced by compound forms using the verb ‘to be’ and a Past Participle in -l (although an aspectual opposition has been reintroduced by means such as those discussed in section 2.3.2.). This is illustrated by Czech já jsem koupil [‘I bought’], literally ‘I am having-bought’. In earlier stages of the Slavic languages, and still in Bulgarian and Macedonian, equivalent constructions have Perfect, rather than Past meaning, the semantic component of current relevance having been lost. Incidentally, when we turn from the Present Perfect to the Past Perfect (Pluperfect) and Future Perfect, loss of perfect aspectual meaning in favor of a component of anteriority (relative time reference) is much more prevalent. Even in English, which maintains, especially in its British variety, a rather strict separation of Present Perfect from Past, this distinction is neutralized in the Pluperfect and the Future Perfect, so that Keith had g one can represent either ‘Keith has gone’ or ‘Keith went’ relative to a reference point in the past. 2.4.3. From Perfective to Past or Future Although the Perfective/Imperfective distinction is in principle completely distinct from tense oppositions, in practice the combination of Perfective and Present is rare, because a situation encompassing the present moment is almost necessarily ongoing rather than bounded. And indeed, many languages have perfectivity restricted to the past and future (or, indeed, only the past). This tenseaspect correlation also has diachronic reflections. The Ancient Greek Aorist (Goodwin 1889, 16 —18, 22—31, 48—53) derives from a Proto-Indo-European Perfective, with no particular tense connotations, and is indeed still so used in the classical language in nonfinite forms and in the Subjunctive and Optative moods. In the Indicative, however, by the classical period it was by and large restricted to past time reference and required the so-called augment, e-, a distinctive marker of Past tense, thus giving rise to the system Present l-omen [‘we free’], Imperfect e-l-omen, Aorist e-l-s-amen, with no form

XX. Die Erforschung des Syntaxwandels III: Phänomene

1250

*l-s-amen. In the East and West Slavic languages, where there is independently a clear morphological distinction between Past and non-Past, the Perfective non-Past has been almost completely commandeered to express future time reference, to the extent that many descriptions of languages like Russian describe this as the Perfective Future, even though it is morphologically parallel to the Imperfective Present, e. g. Perfective nonPast on pro-čitaet [‘he will read’], Imperfective non-Past on čitaet [‘he reads’]. (There is a compound Imperfective Future, on budet čitat’ [‘he will read, will be reading’], with no formal Perfective counterpart.) In Czech this has even been taken a stage further with a few verbs, the form that is morphologically a Perfective non-Past functioning semantically as a Future with no distinction of aspect, e. g. pů-jdu [‘I will go’] (cf. jdu [‘I go’]. 2.4.4. Lexical Origins of Tense As a general rule, the kinds of temporal reference expressed grammatically tend to be distinct from those expressed lexically, the former being much more general (e. g. ‘past’, ‘remote future’) than the latter (e. g. ‘yesterday’, ‘tomorrow’, Czech loni [‘last year’]). However, there is some overlap, especially with some tenses expressing rather specific degrees of remoteness (such as the Haya Yesterday Past tense noted in section 1.1.), and recently a number of instances have become attested of tense markers deriving from lexical items. One such example is the Tok Pisin Future tense marker bai (Sankoff/Laberge 1974, 77). This derives from the adverb baimbai, itself from English by and by, and there are three indications of its increasing grammaticalization: the phonetic reduction from baimbai to bai, the possibility of its occurring more closely bound to the verb complex, immediately preceding the verb, as in dispela meri bai i g o [‘this woman will go’] (cf. *dispela meri baimbai i g o), and the possibility of its redundant cooccurrence with other adverbials of future time reference, e. g. tumora em bai i g o [‘he will go tomorrow’]. The development of tenses with rather specific time reference from the corresponding adverbs is attested in a number of languages. In Kala Lagaw Ya, the suffix for the Yesterday Past tense -ng ul, though grammaticalized on a par with other tense suffixes, clearly derives from the adverb ng ul [‘yesterday’] (Bani/Klokeid 1972, 98). A number of such developments have recently been documented for the Kru

languages of West Africa by Marchese (1984). An interesting development here is that some such tense markers have extended their semantic range after grammaticalization, in keeping with the observation at the beginning of this section that tense meanings tend to be more general than the meanings of lexical time reference (Marchese 1984, 205— 207): in Neyo, a general Past tense has been derived from the adverb ‘yesterday’. In some of the Grassfields Bantu languages of Cameroon, tense markers derive not from time adverbials but rather from verbs which are grammaticalized (and often extend their semantic range) (Hyman 1980).

3.

Conclusions

Although the data available to us for the historical study of tense-aspect systems are quite restricted, they are nonetheless sufficiently rich to enable us to identify a number of natural, cross-linguistically recurring patterns of semantic shift involving tense and aspect categories. Not surprisingly, these semantic shifts involve semantically or pragmatically closely related notions, and one of the goals of an overall theory of tense-aspect (and semantically and pragmatically adjacent areas) is to ensure that the naturalness of these shifts finds a natural expression within the theory.

4.

References

Bani, Ephraim, and Terry J. Klokeid. 1972. Kalau Lagau Langgus: the Western Torres Strait language: a report to the Australian Institute of Aboriginal Studies, Canberra. Chao, Yuen Ren. 196 8. A grammar of Spoken Chinese. Berkeley. Comrie, Bernard. 1976 . Aspect: an introduction to the study of verbal aspect and related problems. (Cambridge textbooks in linguistics). Cambridge. —. 1985. Tense. (Cambridge textbooks in linguistics). Cambridge. Dahl, Östen. 1985. Tense and aspect systems. Oxford. Fleischman, Suzanne. 1982. The future in thought and language development: diachronic evidence from Romance. (Cambridge Studies in Linguistics 36). Cambridge. Forsyth, J. 1970. A grammar of aspect: usage and meaning in the Russian verb. (Studies in the modern Russian language, Extra volume). Cambridge.

70.  Synsemantika

Givón,Talmy. 1971. On the verbal origin of Bantu verbal suffixes. Studies in African Linguistics 2. 145—163. Goodwin, W. W. 1889. Syntax of the moods and tenses of the Greek verb. New York. Harris, Martin. 1982. The ‘past simple’ and the ‘present perfect’ in Romance. Studies in the Romance verb, ed. by Nigel Vincent & Martin Harris, 42—70. London. Hyman, Larry M. 1980. Relative time reference in the Bamileke tense system. Studies in African Linguistics 11. 227—237. Marchese, Lynell. 1984. Tense innovation in the Kru language family. Studies in African Linguistics 15. 189—213. Perel’muter, Il’ja A. 1988. Stative, resultative, passive, and perfect in Ancient Greek (Homeric Greek). Typology of resultative constructions (resultative, stative, passive, perfect), ed. by V. P. Nedjalkov, 277—287. Amsterdam. (Russian original, Leningrad, 1983.) Rowlands, E. C. 196 9. Yoruba. (Teach yourself books). London. Salvi, Giampaolo. 1982. Sulla storia sintattica della

1251

costruzione romanza habeo + part. perf. RRom 17. 118—133. Sankoff, Gillian, and Suzanne Laberg e. 1974. On the acquisition of native speakers by a language. Pidgins and creoles, ed. by David De Camp et al., 73—84. Washington, DC (Originally published in Kivung 6.1. (1973), 32—47.) Traug ott, Elizabeth. 1978. On the expression of spatio-temporal relations in language. Universals of human language, Vol. 3, word structure, ed. by Joseph H. Greenberg, 369—400. Stanford. Tschenkéli, Kita. 1958. Einführung in die georgische Sprache, Band 1, theoretischer Teil. Zürich. Ultan, Russell. 1978. The nature of future tenses. Universals of human language, Vol. 3, word structure, ed. by Joseph H. Greenberg, 83—123. Stanford. Welmers, Wm. E. 1973. African language structures. Berkeley. Wrig ht, W. 1898. A grammar of the Arabic language. Third edition. Cambridge.

Bernard Comrie, Los Angeles, University of Southern California (USA)

70. Synsemantika 1. 2. 3. 4. 5. 6. 7. 8. 9.

1.

Die Begriffe ‘autosemantisch’ und ‘synsemantisch’ Grammatizität Grammatikalisierung Analogischer Wandel Grammatikalisierung und Analogie Grammatikalisierung und Lexikalisierung Evolutive Typologie und historische Sprachwissenschaft Sprachtheoretische Aspekte Literatur

Die Begriffe ‘autosemantisch’ und ‘synsemantisch’

Ein sprachliches Zeichen hat ein Significatum, d. h. eine Bedeutung, die innerhalb der betreffenden Sprache im Schnittpunkt seiner paradigmatischen und syntagmatischen semantischen Relationen liegt. Da Bedeutung in diesem Sinne nichts mit außersprachlichen kognitiven Größen, geschweige mit realen Gegenständen und Sachverhalten zu tun hat, kommt sie grammatischen (z. B. ist) ebenso wie lexikalischen (z. B. vernünftig) Zeichen zu. Es ist also nicht erforderlich, mit Bezug auf grammatische Zeichen von Funktion statt von Bedeutung zu sprechen; vielmehr

kann man den Terminus ‘Funktion’ für andere Begriffe (im Bereich der Grammatik) reservieren. Die Bedeutung eines Sprachzeichens kann als aus semantischen Merkmalen zusammengesetzt gedacht werden. Daneben hat es grammatische (strukturelle) Eigenschaften, die durch grammatische Merkmale dargestellt werden können. Die semantischen Merkmale sind ihrerseits entweder grammatisch-semantisch oder lexikalisch-semantisch, je nachdem, ob sie für die Grammatik relevant sind oder nicht. Z. B. hat das Wort Kuh u. a. die semantischen Merkmale PHYSIKALISCHES OBJEKT, SÄUGETIER, NUTZTIER, BELEBT, NICHTMENSCHLICH, WEIBLICH. Von diesen sind einige wie SÄUGETIER, NUTZTIER rein lexikalisch-semantisch, denn keine Regel der deutschen Grammatik nimmt auf sie Bezug. Andere wie NICHT-MENSCHLICH, WEIBLICH dagegen sind grammatisch-semantisch, denn sie sorgen dafür, daß man nach Kuh nicht mit wer fragen, das Wort aber durch sie anaphorisch aufnehmen kann. Andere Merkmale von Kuh wie [Substantiv], [femininum] oder

1252

[umlautfähig] dagegen sind in erster Linie oder ausschließlich für die Grammatik relevant, also grammatische Merkmale. Nichtsdestoweniger besteht ein Zusammenhang zwischen einem grammatisch-semantischen Merkmal wie PHYSIKALISCHES OBJEKT oder WEIBLICH und einem grammatischen wie [Substantiv] oder [femininum]. Da alle Zeichen irgendwie unter grammatischem Aspekt funktionieren, impliziert das Vorhandensein von lexikalisch-semantischen Merkmalen das von grammatisch-semantischen Merkmalen und das Vorhandensein der letzteren wiederum das von grammatischen Merkmalen. Manche Zeichen haben wenige oder keine lexikalisch-semantischen Merkmale. Dazu gehören Wörter wie schon, sind, wir oder da, aber auch Affixe wie -st (sei es in sing st, sei es in höchst) und erst recht grammatische Prozesse wie der Ablaut (in singst vs. sangst), falls diese überhaupt noch zu den Zeichen zu rechnen sind. Wörter, deren Bedeutung überwiegend durch lexikalisch-semantische Merkmale charakterisiert ist, wie Kuh, sind häufig mit außersprachlichen Designata assoziiert, können z. B. Gegenstand einer Enzyklopädie sein oder können jedenfalls für sich referieren und haben insofern eine gewisse semantische Selbständigkeit. Sie heißen daher auch Autosemantika. Zeichen, die keine lexikalisch-semantischen Merkmale haben, sind nie mit außersprachlichen Designata assoziiert, stehen in keiner Enzyklopädie, sondern beziehen ihre Bedeutung ausschließlich aus dem Sprachsystem. Sie können also nur Aspekte referierender Ausdrücke bezeichnen, aber nicht selbst referieren und sind insofern semantisch unselbständig. Sie heißen auch Synsemantika (s. Brauße 1994, Kap. 1 zur Begriffsgeschichte). Die Termini ‘autosemantisch vs. synsemantisch’ werden in Marty 1908, 205 ff eingeführt, allerdings mit einer völlig anderen Bedeutung. Autosemantika sind dort im wesentlichen referenzfähige Ausdrücke (Nominalsyntagmen im Nominativ und Sätze), Synsemantika sind alle anderen Ausdrücke (u. a. oblique Kasusformen, S. 224). Eine solche Unterscheidung ist möglicherweise sinnvoll; z. B. macht Langacker 1987, Kap. 8.3. mit den Termini ‘autonomous vs. dependent’ eine ganz ähnliche. Es dürfte aber kaum das Verständnis der grammatischen Formative fördern, wenn man sie unter die Synsemantika in diesem Sinne subsumiert.

XX. Die Erforschung des Syntaxwandels III: Phänomene

Neben diesen beiden Termini existieren auch die bedeutungsverwandten ‘kategorematische vs. synkategorematische Zeichen’. Sie beziehen sich auf die aristotelischen Kategorien. Ein Zeichen, dessen Bedeutung in eine der Kategorien fällt, ist demnach kategorematisch; die synkategorematischen treten dagegen an den ersteren auf. Keines der beiden Terminipaare ist sonderlich stark verbreitet. Mit der Unterscheidung von Autosemantika und Synsemantika ist gelegentlich auch die Vorstellung verbunden, daß erstere eine Bedeutung für sich haben, letztere jedoch nur eine Bedeutung innerhalb eines Kontexts oder sogar keine Bedeutung. Dies ist nicht haltbar. Die Bedeutung eines hochgradig polysemen Autosemantikums wie recht läßt sich, in einer gewissen Hinsicht, nur mit Bezug auf einen Kontext angeben, während die Bedeutung eines Synsemantikums wie wir kontextfrei eindeutig angebbar ist (als „Kollektiv von Personen, das den Sprecher einschließt“ ). Wegen dieser irreführenden Implikation wird im folgenden meist von lexikalischen Zeichen vs. grammatischen Zeichen statt von Autosemantika vs. Synsemantika die Rede sein. Im Artikel 30, „Syntactic Categories and Subcategories“ , werden Synsemantika mit Partikeln zusammengebracht. Da dort unter Synsemantika ebenso wie hier grammatische Formative verstanden werden, ist darauf hinzuweisen, daß in diesem Artikel grammatische Formative (Synsemantika) und Partikeln nichts miteinander zu tun haben. Partikeln können lexikalisch (stracks) oder grammatisch (von) sein, und grammatische Wörter können flektieren (haben) oder Partikeln sein (von). Freilich sind bei weitem nicht alle Zeichen, die nicht mit einem außersprachlichen Designatum assoziiert sind, grammatische Zeichen. Es genügt, hier auf Partikeln wie freilich, immerhin, überhaupt hinzuweisen. Falls der Begriff ‘Zeichen, das nur im syntagmatischen Zusammenhang mit anderen eine Bedeutung hat’ expliziert werden könnte, wäre es denkbar (so wie in der Literatur gelegentlich angenommen wird), daß solche Partikeln zu den Synsemantika in diesem Sinne zählen. Die in diesem Artikel vorgenommene Zurückführung von ‘autosemantisch vs. synsemantisch’ auf ‘lexikalisch vs. grammatisch’ wäre dann unangemessen. Allerdings ist über die Entstehung solcher Wörter ohnehin fast nichts Allgemeines bekannt.

70.  Synsemantika

2.

1253

Grammatizität

Der Unterschied zwischen lexikalisch-semantischen, grammatisch-semantischen und grammatischen Merkmalen ist gradueller Natur. Entsprechend ist ein Zeichen den Kategorien ‘lexikalisches’ vs. ‘grammatisches Zeichen’ nicht notwendigerweise eindeutig zuzuordnen. Z. B. kann die benefaktive semantische Relation durch Zeichen ausgedrückt werden, die verschiedene Positionen zwischen dem lexikalischen und dem grammatischen Pol einnehmen. (1 a) Erna kaufte ein Grundstück zugunsten von Erwin. (1 b) Erna kaufte ein Grundstück für Erwin. (1 c) Erna kaufte Erwin ein Grundstück. Die komplexe Präposition in (1 a) ist ein lexikalisches Zeichen, der Dativ in (1 c) ist ein grammatisches Zeichen, die einfache Präposition in (1 b) steht zwischen beiden. Die Position eines Zeichens zwischen dem lexikalischen und dem grammatischen Pol bzw. der Grad, zu dem es ein grammatisches Zeichen ist, ist seine Grammatizität. Das Attribut ‘grammatisch’ ist mithin zweideutig, je nachdem ob sein Träger die Eigenschaft der Grammatizität, also der Zugehörigkeit zum grammatischen System, oder die der Grammatikalität besitzt, also gemäß den grammatischen Regeln gebildet ist. Grammatizität ist eine komplexe Eigenschaft, für die mehrere Kriterien gelten. (1) zeigt, daß es keine allgemeinen semantischen Kriterien gibt, wonach etwa Benefaktivität zu den grammatischen Bedeutungen gehörte. Es ist also nicht möglich, den Begriff der grammatischen Bedeutung intensional so zu definieren, daß sich daraus ein Katalog der grammatischen Bedeutungen der Sprachen der Welt ableiten ließe. Das einzige, was sich allgemein sagen läßt, ist, daß bestimmte kognitive Domänen, etwa die Kochterminologie oder die Kosmologie, sich in keiner Sprache in grammatischen Bedeutungen niederschlagen, während andere kognitive Domänen,

etwa die Possession oder die Partizipation (Beteiligung von Mitspielern an einer Situation), sich in den meisten Sprachen in grammatischen Bedeutungen manifestieren. Die Kriterien der Grammatizität sind formaler Natur. Jedes sprachliche Zeichen unterliegt gleichzeitig der Selektion auf der paradigmatischen und der Kombination auf der syntagmatischen Achse (Jakobson 1956 ). Die Freiheit, mit der ein Zeichen auf den beiden Achsen manipuliert, also selektiert und kombiniert wird, bestimmt seine Autonomie. Mit steigender Grammatizität verliert ein Zeichen an Autonomie, d. h. es wird stärker in das Netz paradigmatischer und syntagmatischer Relationen integriert und grammatischen Regeln unterworfen. Autonomie und Grammatizität sind also komplementäre Pole derselben Eigenschaft. Diese ist auf den in Tab. 70.1 dargestellten Parametern meßbar. Die Präposition zug unsten gehört zu der großen, offenen und heterogenen Klasse der Präpositionen, die den Genitiv bzw. von regieren. Sie kann von ihrem Rektum durch von getrennt sein. Sie unterliegt keinerlei Beschränkungen der Auswahl. Es gibt die Stellungsvariante zu Erwins Gunsten. Der Dativ gehört zu dem kleinen, geschlossenen und homogenen Paradigma der Kasus. Er ist ein flexivisches Suffix an seinem Träger. Er hat in vielen Kontexten keine Alternative, z. B. wenn er von Verben regiert wird, und häufig ist er sogar obligatorisch (z. B. nach ähneln). Das Kasusparadigma als ganzes ist an Substantiven obligatorisch. Der Dativ hat gegenüber seinem Träger keine Stellungsfreiheit. Die Präposition für steht in allen diesen Beziehungen zwischen den beiden Polen. Ein Zeichen ist demnach desto grammatischer, je größer seine Kohäsion und je geringer seine Variabilität ist. Mit diesen kriterialen Eigenschaften korreliert eine weitere, das Gewicht des Zeichens. Das ist in erster Linie seine Integrität, d. h. sein Umfang, gemessen in semantischen und phonologischen Merkmalen. Grammatische Zeichen haben durchschnittlich eine geringere Integrität als lexika-

Tab. 70.1: Parameter der Grammatizität

Relationenachse Autonomieaspekt Kohäsion Kohäsion Variabilität

paradigmatisch

syntagmatisch

Integrität Paradigmatizität Wählbarkeit

struktureller Skopus Fügungsenge Stellungsfreiheit

XX. Die Erforschung des Syntaxwandels III: Phänomene

1254

lische. Z. B. besteht das Significans des Dativsuffixes im Singular bestenfalls aus einem Phonem, das von für aus drei Phonemen. Das Significatum des Dativs hat ausschließlich grammatische Merkmale, für hat zusätzlich grammatisch-semantische Merkmale (z. B. der Zuwendung). Das Gewicht ist in zweiter Linie der strukturelle Skopus des Zeichens, d. h. die grammatische Ebene, der seine Kokonstituente angehört. Grammatische Zeichen haben gewöhnlich einen kleineren Skopus. Z. B. geht der Skopus von für über ein Nominalsyntagma, der des Dativsuffixes über einen Substantivstamm. Hiernach kann man ‘lexikalisch vs. grammatisch’ für Zeichen (einschließlich Wörter, Morpheme usw.) definieren durch Rückgriff auf den Grad der Grammatizität, der seinerseits durch die Kriterien der Autonomie auf der paradigmatischen und syntagmatischen Achse bestimmt wird. Grammatizität ist in erster Linie eine Eigenschaft bestimmter Zeichen, dann aber, wegen der paradigmatischen Beziehungen, auch eine Eigenschaft ganzer Klassen und Kategorien von Zeichen und, wegen der internen syntagmatischen Beziehungen, auch einer Konstruktion, etwa einer periphrastischen Verbform. In einem abgeleiteten Sinne läßt sich ferner von der Grammatizität bestimmter Eigenschaften solcher Gegenstände, also etwa von deren Bedeutung oder deren Wortstellung, sprechen. Schließlich erscheint in einer dynamischen Sprachbetrachtung der Einsatz eines grammatischen Elements als eine Operation, die Bildung einer Konstruktion als Strategie zur Lösung eines Ausdrucksproblems. Auch solche Operationen und Strategien können mehr oder minder grammatisch genannt werden.

3.

Grammatikalisierung

3.1. Grammatischer Wandel Grammatische Zeichen unterliegen, wie alles in der Sprache, der Variation. Diese findet in der Synchronie und der Diachronie statt. Im letzteren Falle ist es grammatischer Wandel. Hierbei entstehen grammatische Zeichen, werden umstrukturiert und verschwinden wieder. Lexikalische Zeichen werden gelegentlich, durch sog. Urschöpfung, aus dem Nichts neugeschaffen. Die Möglichkeit hängt an ihrem autosemantischen Charakter. Bei gram-

matischen Zeichen geschieht dies nie. Sie entwickeln sich ausschließlich aus vorhandenen Zeichen oder, seltener, durch Morphologisierung einer phonologischen Alternation. Unter Entstehung grammatischer Zeichen sei die Schaffung eines grammatischen Zeichens aus einem nicht-grammatischen Zeichen, unter Umstrukturierung die Schaffung eines Zeichens von gegebener Grammatizität aus einem Zeichen annähernd derselben Grammatizität verstanden. In diesem Sinne entsteht die Präposition zug unsten aus dem nichtgrammatischen Zeichen zu Gunsten, während nebst im 17. Jahrhundert aus neben umstrukturiert wird (vgl. Abschnitt 4.1.). Der wesentliche Prozeß, in dem Synsemantika entstehen, ist die Grammatikalisierung. Daher ist hier hauptsächlich von ihr zu handeln. Prozesse der Umstrukturierung grammatischer Zeichen sind davon jedoch nicht ganz zu trennen und kommen in Abschnitt 4. zur Sprache. Die Idee, daß grammatische Zeichen aus lexikalischen entstehen, geht bereits auf die Vorstufen der historischen Sprachwissenschaft im 18. Jahrhundert zurück. Der französische Linguist Antoine Meillet machte sie 1912 erstmals explizit und führte auch den Terminus ‘Grammatikalisierung’ (frz. ‘grammaticalisation’, engl. ‘grammaticalization’ oder ‘grammaticization’) ein. In der Linguistik des ausgehenden zwanzigsten Jahrhunderts ist die Grammatikalisierungsforschung hauptsächlich in die Domäne der typologischvergleichenden Sprachwissenschaft übergegangen. 3.2. Die Grammatikalisierungsskala (2) ist ein Schulbeispiel für Grammatikalisierung, an dem die wichtigsten Aspekte erläutert werden können. Das Symbol ‘>’ ist zu lesen als „wird (grammatikalisiert) zu“ . (2a) vulgärlat. cantare habet „hat zu singen“ > ital. canterà „wird singen“ (2b) vulgärlat. cantare habuit „hatte zu singen“ > ital. canterebbe „würde singen“ Grammatikalisierung betrifft in erster Linie sprachliche Zeichen, einfache wie habet und zusammengesetzte wie cantare habet. Sowohl theoretisch als auch methodisch wichtig ist, daß ein Zeichen nicht in Isolation betroffen wird, sondern im Schnittpunkt von paradigmatischen und syntagmatischen Relationen steht. So schließt die Grammatikalisierung von habet seine paradigmatische Rela-

70.  Synsemantika

1255

Tab. 70.2: Prozesse der Grammatikalisierung

Grammatizitätniedrig Parameter Zeichen gehört zu Paradigmatizität losem Wortfeld Zeichen ist nach komWählbarkeit munikativen Absichten frei wählbar Bündel semantischer Integrität Merkmale; evtl. mehrsilbig Zeichen ist unabhängig Fügungsenge juxtaponiert Stellungsfreiheit struktureller Skopus

Zeichen ist frei umstellbar Zeichen bezieht sich auf Syntagma beliebiger Komplexität

tion zu habuit und seine syntagmatische Relation zu cantare ein. Die Grammatikalisierung von lat. cantare habet und cantare habuit ist nicht zu verstehen ohne deren (lose) paradigmatische Beziehung zu cantat „singt“ , cantabat „sang“ usw. bzw. die (enge) paradigmatische Beziehung zwischen ital. canterà, canterebbe und canta „singt“ , cantaba „sang“ usw. Da Grammatikalisierung das dynamische Gegenstück zu Grammatizität ist, läßt sich der Begriff auf dieselben Gegenstände wie dieser anwenden. Man spricht daher von der Grammatikalisierung von Zeichen, Kategorien, Konstruktionen, Strategien usw. Der von der Grammatikalisierung betroffene Aspekt eines Sprachzeichens ist seine Autonomie; sie nimmt ab. Mit Bezug auf Tab. 70.1 läßt sich präzisieren, daß die Kohäsion zunimmt und daß die Variabilität, ebenso wie das mit beiden korrelierende Gewicht des Zeichens, abnimmt. Jeder dieser Aspekte der Grammatizität hat ein prozessuales Gegenstück, wie in Tab. 70.2 dargestellt. So steht cantare habet zunächst in Opposition zu allen Syntagmen, die aus einem Verb und einem regierten Infinitiv bestehen (z. B. cantare potest/debet/cog itat „kann/muß/beabsichtigt zu singen“ , wächst dann aber in das Paradigma der verbalen Kategorien hinein und wird diesem semantisch und im Ausdruck angepaßt. Habet ist unter allen Verben, die einen Infinitiv regieren können, frei wählbar, während die Wahl des Futurs durch Kontextfaktoren wie temporale Adverbialien

→ hoch Prozeß Paradigmatisierung Zeichen gehört zu hochintegriertem Paradigma Obligatorisierung Wahl des Zeichens ist beschränkt bzw. obligatorisch grammatische Merkmale; Erosion oligo- oder monosegmental Koaleszenz

Zeichen ist Affix oder bloß phonologische Eigenschaft des Trägers Fixierung Zeichen besetzt feste Position Kondensierung Zeichen modifiziert Stamm

oder übergeordnete Prädikate beschränkt bzw. erzwungen wird. Das Significans von habet erodiert von fünf zu einem Segment, von zwei zu einer Silbe, und das Significatum verliert die semantische Komponente des Verfügens über etwas. Während zwischen cantare und habet Konstituentengrenze ist, an der keine phonologischen Prozesse stattfinden, ist -à vom Stamm nur durch eine Morphemgrenze getrennt, die nicht hindert, daß es mit dem Endkonsonanten des Futursuffixes zu einer Silbe zusammengefaßt wird. Während cantare und habet frei umstellbar und trennbar sind, ist das Morphem -à an seine suffixale Position gebunden. Schließlich nimmt habet als Komplement eine Infinitivsyntagma, während sich -à mit einem Tempusstamm verbindet. Grammatikalisierung ist eine Erscheinung der sprachlichen Variation. Als solche äußert sie sich sowohl in der Synchronie als auch in der Diachronie. Auf der synchronen Achse hat der Sprecher die Wahl zwischen isofunktionalen Strategien verschiedenen Grammatizitätsgrads. Außerdem kann er den Grammatizitätsgrad eines gegebenen Zeichens variieren und so die Art von Polysemie erzeugen bzw. nutzen, die etwa in der lokalen vs. genitivischen Funktion der Präposition von vorliegt. Auf der diachronen Achse verleiht der Sprecher einem Zeichen einen höheren Grad an Grammatizität. Grammatikalisierung ist ein kontinuierliches Phänomen. Es gibt eine abstrakte Skala ihrer Phasen wie in Tab. 70.3.

1256

XX. Die Erforschung des Syntaxwandels III: Phänomene

Tab. 70.3: Grammatikalisierungsskala

Stadien, denen Zwischenprodukte des Prozesses angehören, können als Paare aus einer grammatischen Ebene und einer Technik aufgefaßt werden. Der graduelle Charakter der Grammatikalisierungsskala läßt sich dann im Hinblick auf Tab. 70.3 so bestimmen, daß der Unterschied zwischen zwei aufeinanderfolgenden Stadien A und B von derselben Art wie der Unterschied der aufeinanderfolgenden Stadien B und C ist. Die Skala ist darüber hinaus kontinuierlich in dem Sinne, daß die Stadien lediglich prototypischen Charakter haben und sprachliche Phänomene durchaus auf Punkten der Skala zwischen den benannten Stadien liegen können. 3.3. Degrammatikalisierung Grammatikalisierung ist ein zielgerichteter Prozeß. Die von ihm erfaßten Zeichen werden immer mehr grammatischen Regeln unterworfen, wobei im Einzelfall die betreffenden Regeln schon vorher existieren oder erst mit dem Prozeß geschaffen werden können. Der Prozeß ist gerichtet (unidirektional), also nicht umkehrbar. Dies hat mehrere Konsequenzen. Es gibt keinen systematischen Prozeß der Degrammatikalisierung, der als das gerade Spiegelbild von Grammatikalisierung zu fassen wäre. Es kommt also nicht in systematischer Weise vor, daß ein grammatisches Zeichen durch Zunahme an Autonomie, also eine Bewegung von rechts nach links in Tab. 70.2 und Tab. 70.3, zu einem selbständigeren lexikalischen Zeichen würde. Mögliche Gegenbeispiele zu dieser empirischen Generalisierung kommen hauptsächlich aus dem Bereich der Wortbildung. So wurde aus dem Derivationssuffix englischer Abstrakta wie communism und capitalism ein Substantiv ism „Ideologie“ abstrahiert. Auch die Substantivierung grammatischer Wörter, etwa von if, das auch Basis der Ableitung iffy „reich an Vorbehalten“ ist, kommt der Idee der Degrammatikalisierung in einigen Punkten nahe. Dies sind jedoch Einzelfälle, die im

wesentlichen beweisen, daß bei einer kreativen Tätigkeit wie Sprache keine Regel ohne Ausnahme bleibt. Dagegen sind hier einige systematisch auftretende Phänomene, die gelegentlich als Fälle von Degrammatikalisierung präsentiert werden, näher zu beleuchten. In der Geschichte des Englischen sind die ererbten Kasusssuffixe verlorengegangen. Dadurch verloren sie zwar ihren grammatischen Charakter; aber da sie auch alles andere verloren, ist dies nicht ein Beispiel für Degrammatikalisierung, sondern einfach die letzte Stufe auf der Skala Tab. 70.3. Entsprechendes gilt für Fälle wie den folgenden aus dem Hausa: Fast alle Substantive enden auf Langvokal. Dieser läßt sich als ehemaliger definiter Artikel rekonstruieren. Es ist also ein flexivisches Morphem zu einem Merkmal einer lexikalischen Klasse geworden. Auch dies ist nicht Degrammatikalisierung — und auch nicht Lexikalisierung; vgl. Abschnitt 6 . —, denn es wird nicht ein grammatisches zu einem lexikalischen Zeichen, sondern ein grammatisches Morphem wird zu einem grammatischen Merkmal. Dies ist die in Tab. 70.3 als Demorphologisierung bezeichnete Phase der Grammatikalisierung. Aus der Geschichte zahlreicher Sprachen, insbesondere der romanischen, ist die Entwicklung der Morphologie vom synthetischen zum analytischen Typ bekannt. (3a) dono libros Marco „ich gebe Marcus die Bücher“ (3b) je donne les livres à Marc (dito) Anstatt des lateinischen (3a) hat das Französische (3b). Statt des Personalsuffixes der 1. Pers. Sg. -o hat es also je, statt der nominalen Pluralendung -os den pluralischen definiten Artikel les, statt des Dativsuffixes -o die Präposition à. Die französischen Strukturmittel sind in allen Fällen weniger grammatisch als die lateinischen. Dennoch liegt hier keine De-

70.  Synsemantika

grammatikalisierung vor. Denn diese Ausdrucksmittel des Französischen sind ja nicht die historischen Fortsetzer der entsprechenden lateinischen, sondern sind aus ganz anderen Quellen (lat. eg o „ich“ bzw. illos „jene“ bzw. ad „zu“ ) entstanden. Dies ist also im Gegenteil ein klarer Fall von Grammatikalisierung. Ob bei solchen diachronen Vergleichen der spätere Zustand einen geringeren Grad an Grammatizität aufweist als der frühere, hängt einfach von der Wahl der verglichenen Stadien ab. 3.4. Desemantisierung Die in Tab. 70.2 angeführte Erosion betrifft in gleicher Weise Significans und Significatum des grammtikalisierten Zeichens. Semantische Erosion heißt auch Desemantisierung. Soweit dabei lediglich semantische Merkmale verlorengehen, resultiert notwendigerweise Bedeutungsverallgemeinerung, die als solche auch im rein lexikalischen Wandel vorkommt (vgl. vulgärlat. arripare „landen“ > frz. arriver „ankommen“ ). Das Spezifikum der semantischen Vorgänge bei der Grammatikalisierung liegt darin, daß gerade die grammatisch-semantischen und zum Schluß vielleicht nur noch die grammatischen Merkmale des Zeichens übrigbleiben. Eine stark vereinfachte Beschreibung der Entwicklung der grammatischen Präposition de des Französischen und Spanischen (= ital. di) aus lat. dē, die sich auf ein paar Funktionen konzentriert, soll dies verdeutlichen. In einer lateinischen Konstruktion ‘x dē y’ ist x zunächst eine dynamische Situation, y ein räumlicher Fixpunkt, dē bezeichnet x’ dynamische Relation, d. h. seine Bewegung von y herunter. Erste Schritte zur Grammatikalisierung werden bereits im Lateinischen gemacht. Der Höhenunterschied als semantische Komponente wird fakultativ, so daß eine ablativische Relation („x von y weg“ ) verbleibt. Ferner kann x auch ein Gegenstand sein, dessen Herkunft angegeben wird; syntaktisch ist die Konstruktion dann ein Nomi-

1257

nal mit x als Kern und ‘dē y’ als Attribut. Hier wird also auf die dynamische Komponente verzichtet. Für andere Relationen nominaler Attribution, einschließlich partitiver Verhältnisse, wird allerdings noch der Genitiv eingesetzt. In der Entwicklung zum Französischen wird de auf partitive Verhältnisse, Materialangaben, possessive Verhältnisse und schließlich auf alle syntaktischen Beziehungen zwischen einem nominalen Kern und einem abhängigen Nominalsyntagma ausgedehnt. Sämtliche semantischen Merkmale von de einschließlich seiner Selektionsrestriktionen über x und y gehen verloren, es bleibt lediglich die syntaktische Relation, in der x der die Kategorie des Syntagmas bestimmende Kern und y ein von ihm abhängiges Nominalsyntagma ist. Einer finalistischen Sicht des Vorgangs kann es fast scheinen, daß lat. dē zielstrebig rekrutiert wurde, weil es letztlich gerade auf diese schon in ihm stekkenden grammatischen Eigenschaften ankam. Natürlich folgen die zunehmend grammatikalisierten Bedeutungen von dē nicht in der Geschichte nahtlos aufeinander. Das Gegenstück diachroner Variation ist immer synchrone Variation. Insofern erzeugt die mit Grammatikalisierung einhergehende Desemantisierung eine spezifische Art von Polysemie. Sie liegt auch vor in frz. avoir „haben“ , das ebenso wie im Deutschen Vollverb und Hilfsverb ist. Seine semantische Entwicklung aus dem Lateinischen ist in Tab. 70.4 schematisiert. Auf ein Stadium, in dem die Grundbedeutung mit der grammatischen Bedeutung eines Zeichens koexistiert, kann dann eines folgen, in dem die ursprüngliche Bedeutung aufgegeben wird. So repräsentiert das Spanische in bezug auf die Grammatikalisierung von lat. habere ein diachron weiter fortgeschrittenes Stadium als frz. avoir, denn span. haber hat die possessive Bedeutung nicht mehr, sondern ist in dieser von tener verdrängt worden.

Tab. 70.4: Semantische Entwicklung in der Grammatikalisierung

Stadium 1 Bedeutungsstruktur lexikalische Bedeutung romanisches Beispiel lat. habere „besitzen“

2 lexikalische neben grammatischer Bedeutung frz. avoir „besitzen“ , PERFEKT

3 grammatische Bedeutung span. haber PERFEKT

1258

Der aktuelle Sinn eines Zeichens wird auch durch den Kontext bestimmt. Je mehr ein Zeichen desemantisiert ist, desto anfälliger ist es für Übernahme von Merkmalen aus dem syntagmatischen Kontext und für die Konventionalisierung von Implikaturen (Traugott & König 1991). Die englische Konjunktion since und die italienische Konjunktionalfügung dal momento che, beide „seit“ , erwerben — die englische sehr viel früher — aufgrund des Implikaturprinzips, das vereinfacht als ‘post hoc, ergo propter hoc’ bezeichnet wird, eine kausale Komponente und bedeuten dann „da“ . Solche semantischen Veränderungen sind bei Grammatikalisierung nicht selten, jedoch für sie nicht konstitutiv. 3.5. Grammatikalisierungskanäle Es sei die Möglichkeit vorausgesetzt, interlinguale (übereinzelsprachliche) Begriffe einzelner Zeichen und Zeichenklassen zu bilden, also etwa das Verb ‘gehen’ oder das Futur bzw. die Bewegungsverben oder Tempora in verschiedenen Sprachen. Damit ist die — mit strengem Strukturalismus unvereinbare — Möglichkeit impliziert, auf einer gewissen Abstraktionsebene das deutsche und das lateinische Futur miteinander zu identifizieren. In bezug auf die Beteiligung solcher interlingualer Entitäten an Grammatikalisierungsvorgängen in den Sprachen der Welt ist folgendes festzustellen: 1. Bestimmte lexikalische Zeichen werden immer wieder zur Grammatikalisierung herangezogen, andere eignen sich dazu nicht. So werden Hilfsverben immer wieder aus Bewegungsverben, lokale Adpositionen immer wieder aus Körperteiltermini gewonnen. Witterungsverben oder Tierbezeichnungen dagegen sind nirgends eine fruchtbare Quelle grammatischer Zeichen (vgl. auch Abschnitt 2. zur „Ungrammatizität“ bestimmter kognitiver Domänen). Gewisse Eigenschaften lexikalischer Zeichen bestimmen also ihre Grammatikalisierbarkeit. 2. Ein gegebenes lexikalisches Zeichen kann Quelle für verschiedenartige Grammatikalisierungen sein. Dies gilt zunächst auf interlingualer Ebene. So ist venir „kommen“ im Französischen ein Hilfsverb zur Bildung des Präteritum Recens (je viens d’arriver „ich bin soeben angekommen“ ); im Italienischen dagegen bildet venire eine Form des Passivs (viene detto „es wird gesagt“ ). Es gilt aber sogar auf einzelsprachlicher Ebene. So ergibt das lateinische ‘haben’ in den romanischen

XX. Die Erforschung des Syntaxwandels III: Phänomene

Sprachen sowohl das synthetische Futur, wie in (2a), als auch das Perfekt, wie in (4). (4) vulgärlat. habet cantatum „hat gesungen“ > ital. ha cantato dito Bei dieser Polygrammatikalisierung ist freilich darauf zu achten, daß am Ursprung der verschiedenen grammatischen Formative zwar ein identisches Lexem, jedoch in unterschiedlichen Konstruktionen steht. Polygrammatikalisierung ist daher, analog zu Polysemie, ein aufs isolierte Zeichen bezogener Begriff. Bezieht man, wie in Abschnitt 2. gefordert, die paradigmatischen und syntagmatischen Relationen ein, so gibt es Polygrammatikalisierung in einem strikt grammatischen Sinne nicht. 3. Die Desemantisierung führt zwar zum Verlust konkreter Bedeutungsmerkmale. Nichtsdestoweniger tragen Zeichen mittleren Grammatizitätsgrades oft Spuren ihrer lexikalischen Herkunft in Form spezifischer Kombinationsbeschränkungen an sich, die sie von anderen, unter denselben interlingualen Begriff fallenden, aber vielleicht auf andere Weise entstandenen grammatischen Zeichen unterscheiden (Bybee et al. 1994). Z. B. benutzen mehrere europäische Sprachen das Verb ‘haben’ als Hilfsverb des Perfekts. Während jedoch im Englischen und Spanischen dies das einzige perfektische Hilfsverb ist, alterniert es im Deutschen und Französischen mit dem Verb ‘sein’. In beiden letzteren Sprachen gilt die Regel, daß alle transitiven Verben das Perfekt mit ‘haben’ bilden und daß ‘sein’ ausschließlich bei intransitiven Verben vorkommt. Hier besteht eine offensichtliche Beziehung zur Transitivität bzw. Intransitivität der zugrundeliegenden Vollverben ‘haben’ bzw. ‘sein’ selbst. 4. Bei hohem Grammatizitätsgrad konvergieren Grammtikalisierungsprozesse, die an verschiedenen Quellen ihren Anfang nahmen. Das vom Vulgärlateinischen aufgegebene ererbte Futur auf -bi- geht auf ein indogermanisches Verb ‘sein’ zurück. Das neue synthetische Futur der romanischen Sprachen basiert auf dem Verb ‘haben’, ersetzt jedoch im Ergebnis das Futur des Lateinischen vollständig und ist ihm nicht nur funktional, sondern sogar strukturell so ähnlich wie möglich. Dieses Futur wird seinerseits in den modernen romanischen Sprachen durch Periphrasen verdrängt. Einige Varianten des lateinamerikanischen Spanisch benutzten die „synthetisches Futur“ genannte morphologische Kategorie

70.  Synsemantika

1259

Tab. 70.5: Verbaler Grammatikalisierungskanal des Futurs

1 Vollverb

2 Hilfsverb

3 klitisches Hilfsverb

‘gehen’/ intentionale/deontische/ immediates/obligatives/ ‘haben’/ volitive Modalität volitives Futur ‘wünschen’ etc. nur noch für Vermutungen (cantará „wird (wohl) singen“ ) und bilden im übrigen die funktionale Kategorie des Futurs mit ir „gehen“ als Hilfsverb (va a cantar „wird singen“ ). Auch das Resultat dieses Prozesses ist wieder der ersetzten Kategorie funktional maximal ähnlich. Dasselbe gilt auf zwischensprachlicher Ebene. Das englische Futur beruht auf dem Verb will „wollen“ , aber auch seine Funktion ähnelt sich der eines neutralen Futurs immer mehr an. Verben verschiedener Herkunft geraten also im Laufe ihrer Grammatikalisierung in einen Kanal, der für ihr zukünftiges Schicksal nur wenige Alternativen läßt. Auch dies ist eine Konsequenz der Zielgerichtetheit von Grammatikalisierung. Ein Grammatikalisierungskanal ist eine Instantiierung von Tab. 70.3 in einem bestimmten funktionalen Bereich, derart daß die Positionen auf der Skala durch bestimmte interlinguale lexikalische bzw. grammatische Bedeutungen repräsentiert sind, die sich typischerweise auf der betreffenden Ebene bzw. durch die betreffende Technik manifestieren und durch Grammatikalisierungsprozesse miteinander verbunden sind. So stellt Tab. 70.5 einen Kanal (unter mehreren bekannten) für die Entstehung eines Futurs durch Grammatikalisierung dar. Beispiele für die Stadien sind: 1. engl. g o, span. querer „wollen“ ; 2. engl. be g oing to, will; 3. engl. ’ll; 4. türk. -ecek; 5. lat. -bi-/-a-/-e- FUTUR. Die Existenz solcher Grammatikalisierungskanäle steht außer Frage. Es bleibt jedoch empirisch zu klären, inwieweit Grammatikalisierungsvorgänge durch ihre Quelle bestimmt sind und wieweit sie auf bestimmte Ziele zulaufen. Schließlich ist es ein sprachtheoretisches Problem, den zielgerichteten Charakter der Grammatikalisierung explizit zu machen. Auf empirischer Grundlage scheint die Annahme plausibel, daß es auf interlingualer Ebene eine zwar große, aber doch endliche Menge grammatischer Bedeu-

4 agglutinative Flexion prädiktives Futur

5 fusionierende Flexion neutrales oder vermutendes Futur

tungen vom Typ „Futur“ gibt. Moderne Theorien des Sprachwandels, z. B. Keller 1990, gestehen Sprachgemeinschaften jedoch keine sprachinternen Ziele zu. Insofern bleibt die Frage offen, wieso Grammatikalisierungsvorgänge allenthalben auf die gleichen Ziele konvergieren. 3.6. Neuerung, Erneuerung und Verstärkung Eine Sprache kann in einem gegebenen funktionalen Bereich über ein oder mehrere Strukturmittel bzw. Strategien verfügen. Wenn es mehrere sind, unterscheiden sie sich im allgemeinen im Grammatizitätsgrad. So haben zahlreiche Sprachen in dem der Referenz auf die Sprechaktteilnehmer gewidmeten funktionalen Bereich sowohl Personalpronomina als auch Personalaffixe (meist am Verb) zur Verfügung; und einige wie Französisch unterscheiden sogar noch emphatische und klitische Personalpronomina voneinander. Freilich streben hier die Personalsuffixe dem rechten Pol von Tab. 70.3 zu in dem Maße, in dem die Klitika zu Affixen und die emphatischen zu gewöhnlichen Personalpronomina werden. Dieses Beispiel zeigt ebenso wie das des lateinisch-romanischen Futurs, daß während eine maximal grammatikalisierte Kategorie verlorengeht, eine weniger grammatische zu ihrem Ersatz bereitsteht. Theoretisch ist es denkbar, daß entweder der Verlust einer Kategorie die Grammatikalisierung ihres Ersatzes auslöst oder daß eine frisch grammatikalisierte Kategorie eine schon stark grammatische verdrängt. Seit Beginn der historischen Sprachwissenschaft versucht man immer wieder, diese Alternative zu entscheiden. Grammatikalisierung ist jedoch ein ganz allgemeiner Prozeß, der Strukturmittel verschiedenen Grammatizitätsgrades in gleicher Weise betrifft. Isofunktionale Strukturmittel sind paradigmatisch und oft auch syntagmatisch interdependent. Sie befinden sich gleichzeitig

XX. Die Erforschung des Syntaxwandels III: Phänomene

1260

in einem Drift nach rechts auf Tab. 70.3. Die Alternative braucht daher nicht entschieden zu werden. Die Erneuerung einer grammatischen Kategorie ist der Einsatz neuer Strukturmittel zur Erfüllung der gleichen Funktion. Zahlreiche historische Fälle von Grammatikalisierung sind Erneuerungen. Neben dem lateinisch-romanischen Futur und den in (3) gegebenen Beispielen ist etwa die Erneuerung des Relators des indirekten Objekts (des Dativs in einem allgemeinen Sinne) im Portugiesischen zu nennen. Nachdem der lateinische Dativ, ebenso wie in dem in Abschnitt 3.3. erwähnten Fall des Französischen, durch die Präposition lat. ad > port. a erneuert worden ist, wird dieses derzeit durch para „für“ ersetzt, etwa in eu dou os livros para o Marco „ich gebe Marcus die Bücher“ . Die Verstärkung eines Strukturmittels ist seine syntagmatische Kombination mit einem zusätzlichen, isofunktionalen oder Emphase verleihenden Ausdrucksmittel. Ein bekanntes Beispiel ist die lateinisch-französische Negation. Lateinisches präverbales non „nicht“ wird auf dem Wege zum Französischen durch ein nominales Komplement verstärkt, das einen minimalen Umfang bezeichnet. Verwendet werden Wörter wie pas „Schritt“ , point „Punkt“ , mie „Krume“ usw. in Konstruktionen wie ne va pas „geht keinen Schritt“ , ne mang e mie „ißt keinen Bissen“ . Bei Grammatikalisierung bekommt man diskontinuierliche Negation vom Typ (5a) (vgl. dt. nicht die Bohne). (5a) elle ne chante pas/point „sie singt kein bißchen“ > „singt nicht“ (5b) elle chante pas „sie singt nicht“ Durch Paradigmatisierung überleben nur zwei der Alternativen, nämlich ne pas und ne point. Von diesen wird die erstere weiter grammatikalisiert. Der erste Teil des Negators geht verloren, und in der gesprochenen Sprache resultiert (5b). Ein Prozeß, der als Verstärkung einer grammatischen Kategorie begann, kann also im Ergebnis zu ihrer Erneuerung führen. Erneuerung und Verstärkung sind konservativer Sprachwandel. Man kann annehmen, daß bestimmte Grammatikalisierungskanäle in einer Sprache wohl eingeschliffen sind und zu ständiger Erneuerung genutzt werden. Der Sprachtyp ändert sich durch solchen Wandel nicht. Typologischer Wandel setzt vielmehr die Einführung neuer Kategorien voraus. Auch dies geschieht durch Grammatikalisie-

rung, und zwar nur durch sie. Unter Neuerung soll die diachrone Einführung einer grammatischen Kategorie in ein Sprachsystem verstanden werden, das bislang nicht über diese verfügte. Ein prominentes Beispiel ist der Artikel, der dem Urindogermanischen noch fehlt, jedoch in mehreren indogermanischen Sprachen eingeführt worden ist. Dies geschieht überall durch Grammatikalisierung. Normalerweise wird aus einem Demonstrativum der definite und aus dem Zahlwort ‘eins’ der indefinite Artikel gewonnen. Allerdings führen nicht alle beteiligten Sprachen beide ein. Z. B. beginnt Altgriechisch zu homerischer Zeit mit der Herausbildung eines definiten Artikels aus dem Demonstrativum hó „dér“ , ohne je einen indefiniten Artikel zu entwickeln. Im Vulgärlatein finden sich die ersten Belege einer Verwendung von ille „jener“ als definiter und von unus „einer“ als indefiniter Artikel. Daraus resultiert dann ital. il/la, frz. le/la usw. „der/die“ bzw. ital. un(o) /una, frz. un/ une usw. „ein(e)“ . Ähnliches gilt im Althochdeutschen für das schwach deiktische ther/ thiu/thaz und das Zahlwort ein, das zunächst im Sinne von „ein gewisser“ gebraucht wird. Erst im weiteren Verlaufe entwickeln sich der definite bzw. indefinite Artikel des Neuhochdeutschen.

4.

Analogischer Wandel

4.1. Grundbegriffe Analogie ist eine konzeptuelle Beziehung zwischen einem Vorbild A, das als verstanden vorausgesetzt wird, und einem Operanden B, den es kognitiv zu bewältigen gilt. Auf der Basis einer vorausgesetzten Ähnlichkeit von A und B werden B weitere Eigenschaften von A zugeschrieben. Analogie ist auch der Name für diese kognitive Operation. Im analogischen Sprachwandel besteht die Ähnlichkeit zwischen A und B im allgemeinen darin, daß beide an bestimmten paradigmatischen oder syntagmatischen Beziehungen teilhaben. Die Beziehungen von A werden als gegeben angenommen, und die von B werden nach diesem Vorbild gestaltet. Der paradigmatische Ausgleich nimmt im einfachsten Falle die Form einer Proportion an. Die erste Zeile von Tab. 70.6 stellt die allgemeine Form dar, die zweite Zeile verdeutlicht die Rolle der Merkmalzusammensetzung der an der Proportion beteiligten Glieder.

70.  Synsemantika

1261

Tab. 70.6: Paradigmatischer Ausgleich

A : C = B : x; am : an = bm : x;

x=D x = bn

Nach diesem Schema funktioniert z. B. die Überführung von Verben sowohl in die schwache Konjugation, wie in (6 a), als auch in die starke Konjugation, wie in (6b). (6a) lebe : lebte = webe : x; (6b) trage : trug = frage : x;

x = webte x = frug

Es versteht sich, daß das Sprachsystem gewöhnlich schon vor Anwendung der Analogie über Elemente verfügt, die Partner des Operanden in der paradigmatischen Relation sind. Dies sind wob im Falle von (6 a) und frag te in (6 b). Deren Existenz spielt jedoch für Anwendung und Resultat der Analogie keine Rolle. Auf entsprechende Weise funktioniert der syntagmatische Ausgleich. Bei der Bildung neuer Adpositionen geben im Deutschen die Präpositionen hinsichtlich der Stellung das Vorbild ab. Sekundäre Postpositionen haben daher eine Tendenz, präpositional konstruiert zu werden. Da infinite Verben ihr Komplement linksläufig regieren, sind departizipiale Adpositionen wie entsprechend und betreffend zunächst Postpositionen, wie (7a) illustriert. (7a) deinen Vorschlägen entsprechend (7b) entsprechend deinen Vorschlägen In Analogie zu ähnlichen Adpositionen, in diesem Falle etwa g emäß und nach, die präpositional konstruiert werden (können), werden solche Postpositionen sekundär zu Präpositionen, wie in (7b). Analogie durchzieht das ganze Sprachsystem. Abstrakte Verhältnisse wie Abstammung werden mit der Präposition von bezeichnet in Analogie zu konkreten Verhältnissen wie der räumlichen Herkunft (vgl. Abschnitt 3.4. zu lat. dē). Im Frühneuhochdeutschen wird die Präposition neben umgestaltet. Sie bekommt ein Superlativsuffix in Analogie zu dem Vorbild nahe — nächst. Sekundär wird dann, immer noch nach demselben Vorbild, nebenst zu nebst verkürzt. Wenn man von einem monovalenten Verb wie trinken ein Kausativ wie tränken ableitet, wird das untergeordnete Agens als direktes Objekt realisiert. Der sogenannte doppelte Akkusativ bei trivalenten Verben wie lehren (semantisches Kausativ zum bivalenten lernen) versteht sich

in Analogie zu diesen Verhältnissen. Analogie dehnt die Anwendung des Musters, dem das Vorbild folgt, im Sprachsystem aus. Sie hat daher das Potential zur Regelverallgemeinerung und zur Vergrößerung der Regelmäßigkeit im Sprachsystem. Indem allerdings die analogische Umgestaltung Eigenschaften und Beziehungen des Operanden durch solche des Vorbilds ersetzt, nimmt sie auf etwa bestehende Regelmäßigkeiten in den ersetzten Eigenschaften und Beziehungen keine Rücksicht. Die durch (6 ) ersetzten Formen wob und frag te und die durch (7b) ersetzte postpositionale Konstruktion sind ja nicht unregelmäßig, gelegentlich sogar regelmäßiger als die Vorbilder. Bei der Analogie kommt es also nicht auf größere oder geringere Regelmäßigkeit an, sondern auf die Absicht, den Operanden dem Vorbild anzugleichen. Die Rolle der Analogie im Sprachsystem muß auch im Zusammenhang des Verhältnisses von operativer Bildung von Zeichen vs. Verwendung inventarisierter Zeichen gesehen werden. Durch Analogie wird notwendigerweise (paradigmatische oder syntagmatische) Struktur übertragen. Das Ausmaß von Strukturierung in der Sprache insgesamt wird also mindestens konstant gehalten, wenn nicht erhöht. Somit steht Analogie im Gegensatz zu Tendenzen, die den Anteil an ganzheitlichen, direktem Zugriff im Sprachsystem erhöhen. Dazu gehören z. B. suppletive Bildungen in der Morphologie, Phraseologismen in der Syntax und Lexikalisierung im Lexikon (vgl. Abschnitt 6.). 4.2. Reanalyse Reanalyse (auch Reinterpretation, Umdeutung oder Gliederungsverschiebung) ist eine Operation über einem Zeichen (bzw. einer Klasse von Zeichen), die ihm eine neue morphologische oder syntaktische Struktur verleiht, ohne sein Significans anzutasten. Das Significatum kann sich in dem Maße ändern, wie die neue Struktur neue grammatisch-semantische Merkmale involviert. Sicher diagnostizierbar ist die stattgehabte Reanalyse allerdings erst in dem Moment, wo das Zeichen neue Ausdruckseigenschaften erwirbt, die seine veränderte interne Struktur voraussetzen. Es gibt im Deutschen perfektische Absolutkonstruktionen wie die in (8). (8a) von diesen Problemen (ganz) abgesehen, (8b) die Anwesenden (einmal) ausgenommen

XX. Die Erforschung des Syntaxwandels III: Phänomene

1262

Tab. 70.7: Reanalyse des Stammsuffixes als Pluralsuffix

Numerus Kasus Nom./Akk. Genitiv Dativ

Germanisch Plural Singular *lamb-iz *lamb-iz-ō *lamb-iz-iz *lamb-iz-ōn *lamb-iz-e *lamb-iz-um

Das Partizip regiert hier sein Komplement auf die aktivische Weise. Solche Kontruktionen können als Postpositionalsyntagmen interpretiert werden; die eingeklammerten Adverbien in (8) wären dann nicht mehr möglich. Nachdem diese Reanalyse vollzogen ist, können die neuen Postpositionen sekundär zu Präpositionen umgestaltet werden (vgl. Abschnitt 4.), wie in (9). Vgl. auch Art. 6 5 für Beispiele von Reanalyse in der Syntax des komplexen Satzes. (9a) abgesehen von diesen Problemen (9b) ausgenommen die Anwesenden Ein Schulbeispiel für Reanalyse in der Flexionsmorphologie ist die Entstehung des deutschen Plurals auf -er bei den Neutra, wie in Tab. 70.7 am Beispiel von lamb „Lamm“ gezeigt. Noch auf germanischer Stufe handelt es sich um Stämme, die mit einem -es-Suffix gebildet sind, welches in der Deklination unverändert bleibt. Auf dem Weg zum Althochdeutschen wird das intervokalische /z/ zu /r/, und weiterer Lautwandel führt zur Kürzung und zum Verlust zahlreicher Deklinationsendungen. Es resultiert das in Tab. 70.7 rechts angeführte Paradigma. Hier kommt nun das ehemalige Stammsuffix nur noch im Plural vor, so daß sich seine Interpretation als Pluralsuffix nahelegt. Als solches breitet es sich in der Folge auf andersstämmige Neutra (z. B. Haus) und letztlich sogar auf Maskulina (z. B. Mann) aus. Es bleibt die Frage nach der Motivation einer Reanalyse. Im allgemeinen gibt es für sie ein analogisches Vorbild. So setzt die anhand von (8) vorgeführte Reanalyse das Vorhandensein von Postpositionen in der Sprache voraus. Auch bei der Reanalyse von -er als Pluralsuffix dürfte es eine Rolle gespielt haben, daß die germanischen -i-Stämme im Althochdeutschen im Plural, und nur dort, stets ein -i-Suffix plus Wurzelumlaut hatten (Typ gast—gesti). Auch bei den -n- Stämmen, der nachmaligen schwachen Deklination, bahnt sich wenig später ein allgemeines Pluralsuffix -en an. Gleichzeitig sind dies jedoch

Althochdeutsch Singular Plural lamb lemb-ir lamb-es lemb-ir-o lamb-e lemb-ir-um

alles Beispiele für Grammatikalisierung. Es ist nicht auszuschließen, daß Reanalysen auch ohne analogisches Vorbild vorkommen. Z. B. wird im Französischen das Morphem pas als Negator reanalysiert (vgl. (5). Zwar geschieht dies gleichzeitig mit der Reanalyse der Indefinita personne, aucun und weiterer polaritätssensitiver postverbaler Wörter als Negativa. Für einen postverbalen Negator als solchen gibt es jedoch kein analogisches Vorbild. Anscheinend kann Reanalyse, im Verein mit Grammatikalisierung, auch ohne analogisches Vorbild funktionieren.

5.

Grammatikalisierung und Analogie

Grammatikalisierung ist ein gerichteter Prozeß. Die Variation geht immer in Richtung auf größere Grammatizität. Analogischer Wandel ist ein ungerichteter Prozeß. Zwar gilt tendentiell, daß ein Zeichen von undurchschaubarer oder unproduktiver Struktur eher in Analogie zu einem Zeichen offensichtlicher, produktiver Struktur umgestaltet wird. Wenn aber zwei Muster konkurrieren, kommt analogische Umgestaltung in beiden Richtungen vor, wie an (6 ) zu sehen war (s. Ramat 1987). Grammatikalisierung führt neue grammatische Zeichen und gelegentlich sogar grammatische Kategorien ins System ein. Die meisten Theoretiker sind der Ansicht, daß alle grammatischen Zeichen aller Sprachen letztlich — d. h. mit Einräumung möglicher Umwege und Beteiligung anderer Prozesse — durch Grammatikalisierung entstanden sind. Grammatikalisierung ist in diesem Sinne der grammatikschaffende Prozeß par excellence. Analogischer Wandel ist dagegen verhältnismäßig konservativ. Das Vorbild eines analogischen Wandels ist immer bereits im System vorhanden, d. h. der Prozeß führt für sich nie zur Schaffung neuer grammatischer Zeichen oder Kategorien. (6 ) illustriert analogischen Wandel ohne Grammatikalisierung. Die in Abschnitt 3.6 .

70.  Synsemantika

besprochene Einführung eines Artikels illustriert Grammatikalisierung ohne unmittelbares analogisches Vorbild. Die beiden Prozesse sind also prinzipiell voneinander unabhängig. Aber sie interagieren in vielen historischen Fällen. Schon die in (2) veranschaulichte Erneuerung des synthetischen Futurs bzw. des Konditionals nimmt offensichtlich die so verdrängten Kategorien zum Vorbild. Man kann annehmen, daß Grammatikalisierung erleichtert und beschleunigt wird, wenn ihr Ziel im selben Sprachsystem vorgegeben ist. Nur so ist die ungeheure Produktivität komplexer Präpositionen vom Typ zugunsten, infolg e im Deutschen und komplexer Konjunktionen in den romanischen Sprachen vom Typ frz. parce que „weil“ , après que „nachdem“ zu verstehen. Während hier offensichtlich durch die ständige Nutzung des Kanals der Grammatikalisierungsvorgang selbst analogische Vorbilder hat, kommt es auch vor, daß ein Kanal Jahrhunderte stilliegt und dann wieder genutzt wird. (10a) Das braucht/beabsichtigt er nicht zweimal zu sagen. (10b) Das brauch/muß er nicht zweimal sagen. In der Hochsprache wird brauchen als Vollverb behandelt. Es hat, wie an (10a) zu sehen, die entsprechende Flexion und regiert den abhängigen Infinitiv mit zu. In der Umgangssprache wird es als Modalverb behandelt. Es hat, wie (10b) zeigt, die Modalverbflexion und regiert den Infinitiv direkt. Brauchen wird also in der Umgangssprache zum Modalverb grammatikalisiert. Seine Eingliederung in das Paradigma der Modalverben ist als der mit Grammatikalisierung verbundene Vorgang der Paradigmatisierung zu verstehen. Gleichzeitig ist klar, daß dies in Analogie zu den schon länger bestehenden Modalverben geschieht, so daß das Ziel dieses Grammatikalisierungsvorgangs vorgegeben ist.

6.

Grammatikalisierung und Lexikalisierung

Der Ausdruck Lexikalisierung hat eine Reihe von Verwendungen in der Linguistik. Hier bezeichnet er den Prozeß, durch den ein sprachliches Zeichen zum Lexikoneintrag wird. Dies impliziert, daß es einen holistischen Zugriff auf das Zeichen gibt und daß es nicht mehr kompositionell gebildet werden

1263

muß oder sogar kann. Wiewohl bereits der holistische Zugriff, etwa auf Phrasen wie mit freundlichen Grüßen, Lexikalisierung genannt werden kann, ist die Lexikalisierung doch erst sicher diagnostizierbar, wenn die Zuordnung von Significans und Significatum des Zeichens nicht mehr kompositionell ist. In vielen Phrasen wie eine Frag e stellen ist ein Significatum ad hoc mit einem strukturell komplexen Significans gepaart, ohne daß vollständige Kompositionalität bestünde (im Spanischen „macht“ man eine Frage). Im Sprachwandel kann es zur Demotivierung kommen dadurch, daß Significans und Significatum sich getrennt entwickeln. (11a) *hiu tagu „an diesem Tage“ > ahd. hiutu > nhd. heute (11b) *hiu jâru „in diesem Jahr“ > ahd. hiuro > nhd. heuer Die in (11) rekonstruierten voralthochdeutschen Ausdrücke sind völlig regelmäßige temporale Adverbialien im Instrumental. Während sich auf dem Wege zum Neuhochdeutschen die Bedeutung nur unwesentlich ändert, wird der Ausdruck schon im Althochdeutschen univerbiert. Das Significans entwickelt sich unregelmäßig, die Bildungen werden vollständig demotiviert. Es entstehen zwei neue Lexikoneinträge. Lexikalisierung ist in ihren Auswirkungen auf das Significans der Grammatikalisierung ähnlich; beides sind reduktive Prozesse. Doch während ein Zeichen, das zum Teil eines Lexikoneintrags wird, sein Potential zur Bildung komplexer Zeichen einbüßt, vergrößert sich dieses Potential gerade, wenn das Zeichen grammatikalisiert wird. Lexikalisierung steht also im Dienste des ganzheitlichen Zugriffs auf Zeichen, Grammatikalisierung dagegen im Dienste der analytischen Bildung von Zeichen. Grammatikalisierung und Lexikalisierung wirken in vielen Fällen zusammen. Wenn Präpositionalfügungen wie zu Gunsten oder Konjunktionalfügungen wie frz. par ce que zu morphologisch komplexen Präpositionen (zugunsten) bzw. Konjunktionen (parce que [paskǝ]) univerbiert werden, entstehen erstens neue Lexikoneinträge — und insoweit handelt es sich im Lexikalisierung — und zweitens Formative höheren Grammatizitätsgrades — und insoweit handelt es sich um Grammatikalisierung. Dieses ist auf frühen Stufen der Grammatikalisierung, wo die betroffenen Zeichen noch syntaktisch komplex sein können, geradezu der Regelfall. Erst wenn das

XX. Die Erforschung des Syntaxwandels III: Phänomene

1264

Zeichen zu einem Wort reduziert ist, geht weitere Grammatikalisierung ohne Lexikalisierung vor sich.

7.

Evolutive Typologie und historische Sprachwissenschaft

Die in Tab. 70.3 versammelten Techniken sind eben diejenigen, die die klassischen morphologischen Typen (, die eigentlich grammatische Typen heißen sollten) definieren: isolierender, analytischer, synthetisch-agglutinativer, synthetisch-flektierender (fusionierender) Typ. Wir sahen, daß die Techniken zunächst nur den Grammatizitätsgrad von Zeichen, Zeichenklassen und Konstruktionen definieren. Es kann aber vorkommen, daß die meisten grammatischen Mittel einer Sprache sich einer dieser Ebenen zuordnen. Dann kann das Vorherrschen der betreffenden Technik ein ganzes grammatisches System charakterisieren. Grammatikalisierung schiebt grammatische Kategorien und Konstruktionen die Ebenen von Tab. 70.3 hinunter. So kommen die morphologischen Typen in eine diachrone Abfolge. Ist das Ende der Skala erreicht, beginnt der Zyklus von neuem. In dieser Spiralform präsentiert sich die evolutive Typologie des 19. Jahrhunderts. Soweit sie etwas Reales beschreibt, basiert sie auf der Grammatikalisierung. Freilich ist das Maß an Idealisierung erheblich. Erstens verfügen die meisten Sprachen in vielen funktionalen Bereichen über isofunktionale Strategien, die sich im Grammatizitätsgrad unterscheiden (s. Abschnitt 3.5.). Keine Sprache ist durchaus isolierend, agglutinierend usw. Zweitens liegt es nicht im Konzept der Grammatikalisierungsskala oder des Grammatikalisierungskanals, daß ein jedes zu grammatikalisierende Zeichen seinen Kanal am linken Pol betreten und am rechten verlassen muß. Seiteneinstiege, insbesondere aus der Derivation, sind möglich. Und zahlreiche Ausdrucksmittel geraten außer Mode, lange bevor sie das Ende ihres Kanals erreicht haben. Z. B. abundieren im Spanischen die Aspekt- und Aktionsartauxiliarien vom Typ seg uir „weitermachen“ , acabar „fertig werden“ , quedar „bleiben“ . Die Sprache hat davon auch schon zahlreiche verschlissen, etwa uviar „gelangen“ , fincar „bleiben“ , ohne seit altspanischer Zeit auch nur ein einziges weiter als bis zum Hilfsverbstatus zu grammatikalisieren. So kann man die morphologische Entwicklung der germa-

nischen, romanischen und slavischen Sprachen zum Teil dadurch charakterisieren, daß der indogermanische flektierende Typ durch einen analytischen erneuert wurde. Die isolierende Stufe haben jedoch alle diese Sprachen übersprungen. Wenn die Prinzipien der Grammatikalisierung nicht auf ganze Sprachen, sondern auf einzelne Zeichen angewandt werden, können sie methodisch sehr fruchtbar sein. Da Grammatikalisierung ein gerichteter Prozeß ist, läßt sich ein synchroner Befund, in dem verschieden stark grammatikalisierte Varianten eines etymologisch identischen Zeichens koexistieren, dynamisieren: die stärker grammatischen müssen aus den weniger grammatischen entstanden sein. Dies Prinzip wird z. B. in der Lexikographie bei der inneren Anordnung eines polysemen Lexems genutzt (vgl. Abschnitt 3.4. zur durch Grammatikalisierung geschaffenen Polysemie). Die historische Sprachwissenschaft nutzt es in methodischen Sitationen, wo die untersuchte Sprache über keine lange Geschichte verfügt, für die innere Rekonstruktion. So braucht man keine Daten über die deutsche Sprachgeschichte, um zu wissen, daß das Hilfsverb haben sich aus dem Vollverb entwickelt hat und nicht umgekehrt. Auch beim historischen Vergleich genetisch verwandter Sprachen steht man methodisch vor einem Zustand, den man zu dynamisieren hat. So ist es angesichts von lat. sē „sich“ und altgriech. he-autón „sich (selbst)“ nicht schwierig, nach Grundsätzen der Grammatikalisierungstheorie ein idg. *swe- „sich“ zu rekonstruieren, das im Lateinischen fast unverändert vertreten ist, im Griechischen aber durch autós „selbst“ verstärkt und zu dessen Präfix grammatikalisiert wurde. Zahlreiche Grammatikalisierungsforschungen über Sprachen ohne schriftliche Tradition gehen bei ihrer inneren oder vergleichenden Rekonstruktion methodisch so vor. Daneben läßt sich in begrenztem Umfang der Grundsatz der Analogie, daß im allgemeinen eher das Durchsichtigere, Produktivere als Vorbild des Unmotivierten, Unproduktiven dient denn umgekehrt, für die Rekonstruktion nutzen: fossile Reste haben eine größere Chance, archaisch zu sein, produktive Bildungsmuster sind vermutlich jung. Nach diesem Grundsatz würde man die starke Konjugation des Deutschen für älter als die schwache halten und würde darin durch historisch-vergleichende Evidenz bestätigt werden.

70.  Synsemantika

8.

Sprachtheoretische Aspekte

Grammatik ist immer einzelsprachliche Grammatik. Die Gesetze, nach denen sprachliche Zeichen zum Diskurs zusammengefügt werden, sind weitgehend keine grammatischen Regeln, sondern universale Ikonizitätsprinzipien. Das betrifft z. B. die Stellung von Syntagmen und Teilsätzen in der funktionellen Satzperspektive, aber auch die Möglichkeit, eine grammatiklose Sequenz wie Brunnen Seite Kirche in einer beliebigen Sprache als „Brunnen neben der Kirche“ zu interpretieren. Auf dieser Ebene unterscheiden sich Sprachen sehr wenig voneinander. Auf den niedrigeren grammatischen Ebenen nehmen die sprachspezifischen Strukturen zu, die nicht mehr allein mit universalen Prinzipien interpretierbar sind. Die Morphologie schließlich ist hochgradig arbiträr. Da Grammatikalisierung ein Zeichen die grammatischen Ebenen hinunterbefördert (vgl. Tab. 70.3), vergrößert sie die Arbitrarietät. Ein modaler Operator hat z. B. eine Proposition im semantischen Skopus. Wird er auf relativ hoher grammatischer Ebene, z. B. als Modalverb, manifestiert, so bildet sein struktureller Skopus den semantischen einigermaßen ab. Wird er freilich zum verbalen Modus grammatikalisiert, so betrifft er strukturell den Verbstamm, was nicht unmittelbar semantisch interpretierbar ist. Ein zweites Beispiel ist der Wechsel von ergativischer vs. akkusativischer Kasusflexion an Aktanten in Abhängigkeit von syntaktischen Bedingungen. Auch hier liegt stark grammatikalisierte Morphologie vor, die nicht unmittelbar semantisch interpretiert werden soll. Dasselbe gilt schließlich für das Genussystem in Sprachen wie dem Deutschen. Wiewohl zum Teil aus einem Sexussystem entstanden, ist es im neuhochdeutschen Sprachzustand so stark grammatikalisiert, daß es nicht in erster Linie Designata klassifiziert, sondern eine weitgehend arbiträre grammatische Klassifikation über Substantiven bietet, anhand deren sie in syntaktische und anaphorische Relationen gesetzt werden können. Sprachtätigkeit unterliegt der Antinomie zwischen Freiheit und Regelgebundenheit. Hätten wir nur die universalen ikonischen Diskursprinzipien zur Verfügung, so wären wir zwar durch keine Grammatik eingeengt. Wir hätten aber auch eine unendliche Mühe, uns auszudrücken und verständlich zu machen. Konventionen, in diesem Falle grammatische Mittel und Regeln, sorgen für den Anteil der Sprachtätigkeit, der automatisier-

1265

bar ist und Kreativität nicht erfordert. Insofern hat Grammatik eine Entlastungsfunktion. Wir sahen anhand von Grammatikalisierungskanälen in Abschnitt 3.5., daß eine Sprache häufig in einem gegebenen funktionalen Bereich über isofunktionale Alternativen verschiedenen Grammatizitätsgrades verfügt. Dies läßt dem Sprecher im aktuellen Falle die Wahl, ob er eine sprachliche Funktion in freier Kreativität erfüllen oder dem grammatischen Mechanismus überlassen will. Sprache und somit auch Grammatik wird immerfort geschaffen. Die zielorientierte Kreativität des Sprechers setzt freilich an den oberen grammatischen Ebenen an, wo er die Freiheit zum Manipulieren hat. Des Sprechers unmittelbares Ziel ist es, expressiv zu sein. Dadurch überlagert er immer wieder schon vorhandene Ausdrucksmittel, deren Einsatz sich automatisiert. Dadurch entsteht Grammatik. Den Bodensatz dieses Durchschubs von Ausdrucksmitteln auf die niedrigeren Ebenen bilden die fossilierten Strukturen, die als nächste absterben. Die Schaffung von Synsemantika wird also nicht als eigenes Ziel angestrebt, sondern ist das notwendige Ergebnis der Neugestaltung von Ausdrucksformen auf den höheren Ebenen.

9.

Literatur

Brauße, Ursula. 1994. Lexikalische Funktionen der Synsemantika. (Forschungsberichte des Instituts für Deutsche Sprache, 71). Tübingen. Bybee, Joan, Revere Perkins, and William Pag liuca. 1994. The evolution of grammar. Tense, aspect and modality in the languages of the world. (Perspectives in Neurolinguistics and Psycholinguistics). Chicago, London. Givón, Talmy. 1979. On understanding grammar. New York etc. Heine, Bernd, Ulrike Claudi, and Friederike Hünnemeyer. 1991. Grammaticalization. A conceptual framework. Chicago, London. —, and Mechthild Reh. 1984. Grammaticalization and reanalysis in African languages. Hamburg. Hopper, Paul J., and Elizabeth C. Traug ott. 1993. Grammaticalization. (Cambridge Textbooks in Linguistics). Cambridge. Jakobson, Roman. 1956 . Two aspects of language and two types of aphasic disturbances. Jakobson, Roman, and Morris Halle. Fundamentals of language. (Janua linguarum, 1). ‘s-Gravenhage. 53— 87. Keller, Rudi. 1990. Sprachwandel. Von der unsichtbaren Hand in der Sprache. (UTB, 156 7). Tübingen.

XX. Die Erforschung des Syntaxwandels III: Phänomene

1266

Kuryłowicz, Jerzy. 196 5. The evolution of grammatical categories. Diogenes 51. 55—71. Lang acker, Ronald W. 1987. Foundations of cognitive grammar. I: Theoretical prerequisites. Stanford, Cal. Lehmann, Christian. 1982. Thoughts on grammaticalization. A programmatic sketch. Vol. I. (akup, 48). Köln. Rev. ed.: Unterschleissheim, 1995. —. 1985. Grammaticalization: Synchronic variation and diachronic change. Lingua e Stile 20. 303—318. —. 1989. Grammatikalisierung und Lexikalisierung. Zeitschrift für Phonetik, Sprachwissenschaft und Kommunikationsforschung 42. 11—19. Lessau, Donald A. 1994. A dictionary of grammaticalization. (Bochum-Essener Beiträge zur Sprachwandelforschung, 20). Bochum. Marty, Anton. 1908. Untersuchungen zur Grundlegung der allgemeinen Grammatik und Sprachphilosophie. Erster Band. Halle. Nachdruck: Hildesheim, New York, 1976..

71. 1. 2. 3. 4. 5.

1.

Meillet, Antoine. 1912. L’évolution des formes grammaticales. Scientia (Rivista di Scienza) 12(26 ). Abgedr.: Meillet, Antoine. 1921. Lingiustique historique et linguistique générale. [Tome I]. Paris. Pag liuca, William (ed.). 1994. Perspectives on grammaticalization. (CILT, 109). Amsterdam, Philadelphia. Ramat, Paolo. 1987. Verbi forti e verbi deboli in Germanico. Festschrift for Henry Hoenigswald. On the occasion of his seventieth birthday, ed. by George Cardona and Norman H. Zide. (Ars Linguistica, 15). Tübingen. 319—327. Traug ott, Elizabeth C., and Bernd Heine (eds.). 1991. Approaches to grammaticalization. 2 vols. (Typological Studies in Language, 19). Amsterdam, Philadelphia. —, and Ekkehard König . 1991. The semantics-pragmatics of grammaticalization revisited. Traugott & Heine (eds.) 1991, I. 189—218.

Christian Lehmann, Bielefeld (Deutschland)

Klammerkonstruktionen Einleitung Klammerkonstruktionen in den germ. Sprachen Die historische ‘Entwicklung’ von Klammerkonstruktionen Erklärungsversuche Literatur

Einleitung

In fast allen germ. Sprachen steht das finite Verb (Vf) in Nebensätzen (NS) später als in Hauptsätzen (HS) (zum Engl. s. 2.2.). Diese Erscheinung ist im Neuhochdeutschen (Nhd.) besonders stark ausgeprägt. Man spricht von einer Satzklammer, die aus den in (1) hervorgehobenen Elementen besteht: (1a) Paul hat gestern keine Zeitung gelesen. (1b) Gestern hat Paul keine Zeitung gelesen. (1c) Hat Paul gestern keine Zeitung gelesen? (1d) weil Paul gestern keine Zeitung gelesen hat. Da sich eine vergleichbare Klammerbildung in der gesamten Geschichte der germ. Sprachen findet, stellte die Erforschung ihrer Entstehung und einzelsprachlich unterschiedlichen diachronen Differenzierung von Anfang an eine zentrale Aufgabe historischer Syntaxforschung dar.

Im folgenden sollen zunächst die syntaktischen Verhältnisse im Nhd. (2.1.) und in anderen germ. Sprachen (2.2.) kurz dargestellt werden. Die historische Entwicklung vom Althochdeutschen (Ahd.) bis zum Nhd. (3.1.) wird mit der in anderen germ. Sprachen verglichen (3.2.). Schließlich werden unterschiedliche Erklärungsversuche vorgestellt (4.).

2.

Klammerkonstruktionen in den germ. Sprachen

2.1. Die Verbstellung im Ndh. Die nhd. Verbstellung läßt sich einfach beschreiben (s. o. (1a—d)): — Vf kann am Satzanfang stehen (V1); — Vf kann an der zweiten Stelle im Satz stehen (V2), d. h. es geht ihm ein Satzglied voraus (Vorfeld); — Vf kann am Satzende stehen (VEnd). — Infinite Elemente (Vi) des verbalen Komplexes (VK) (trennbare Präfixe, Partizip Präteritum, Infinitiv) stehen am Satzende, wobei Vf unter bestimmten Bedingungen vor die von ihm abhängigen Vi treten kann (weil du die Zeitung nicht hast lesen können) — Auf die am Satzende stehenden verbalen Elemente können noch weitere nicht-verbale

71.  Klammerkonstruktionen

Elemente folgen (Nachfeld). (weil du die Zeitung nicht gelesen hast am letzten Freitag) — Das Vorfeld kann scheinbar mehrfach besetzt sein. (Die Zeitung am Freitag hast du nicht gelesen). Mit diesen wenigen Bemerkungen sind im wesentlichen alle Verbstellungsmöglichkeiten des Deutschen erfaßt. Sie werden gewöhnlich zu bestimmten Satzmodi in funktionale Beziehung gesetzt: V1 = Entscheidungsfrage oder Imperativ, V2 = Aussagesätze, VEnd = eingeleitete Nebensätze. Das Verhältnis von Form und Funktion ist aber keineswegs eindeutig: V1 kommt auch in vorangestellten Konzessiv- oder Konditionalsätzen vor, V2 in uneingeleiteten Nebensätzen und Fragesätzen (als Echofragen), VEnd in Fragesätzen etc.: (2a) Kommst du heut nicht, kommst du morgen. (2b) Ich glaube, heute kommt er nicht mehr. (2c) Paul kommt doch morgen? (2d) Ob Paul morgen kommt? Abweichende Möglichkeiten treten auch in poetischen Texten auf (vgl. Maurer 1924): Sah’ ein Knab’ ein Röslein stehn. Mit Staunen auf die Frau sie wiesen. (Uhland) Damit ergibt sich insgesamt folgendes Bild (Abb. 71.1) (vgl. auch Art. 39 in Bd. I). Vorfeld Vf Konj. linke Satzklammer

Mittel- VK Nachfeld (Vf) feld rechte Satzklammer

Abb. 71.1: Wortstellungsmöglichkeiten im Deut-

schen Zwischen Vf und Mittelfeld können Personalpronomina und bestimmte Adverbien/Satzadverbien/Modalwörter stehen: Wackernagel-Position (s. u. „Leichtgliederfeld“ ).

1267

kernagel-Elemente unmittelbar hinter Vf umfaßt: dän.: Bog en ville jeg jo ikke have sendt H. tilbag e ig år (‘Das Buch würde ich ja nicht haben geschickt H. zurück gestern’) (Diderichsen 1966, 370). Seit Diderichsen (1941) wird ein Nexusfeld (ville jeg jo ikke) und ein Inhaltsfeld (have sendt H. (bog en) tilbag e ig år) unterschieden (vgl. Clausen 196 9 und Zemb 1986 für das Nhd). Das Engl. zeigt kaum noch Reste einer Klammerbildung: Vf und VK stehen in HS und NS an der gleichen Stelle, nämlich unmittelbar hinter dem Subjekt: Last nig ht Peter may have read the newspaper. Lediglich die sog. Subjekt-Aux-Inversion und das Auftreten bestimmter Adverbien zwischen Vf und VK sind Reste einer Klammerbildung (vgl. Stockwell 1984 und s. u. 3.2.). (3a) Did John actually read the paper? (3b) What did John actually read? (3c) Never will John actually read such a paper. (3d) John will actually never read ... 2.3. Zur Forschungsgeschichte Die grundlegenden Einsichten lassen sich bis zu Herling (1821) und Erdmann (1886 ) zurückverfolgen (vgl. Höhle 1986 ). Über Drach (1937), Erben (1954), Griesbach (196 0) und Engel (1970) hat sich die heute geläufige Terminologie entwickelt (zur skand. Tradition vgl. Thümmel 1980 und Sætningsskema 1986 ). Die empirische Nützlichkeit und theoretische Tauglichkeit der Feldereinteilung hat Reis (1980) nachgewiesen; Scherpenisse (1985) zeigte, daß sie auch in einer generativen Darstellung zugrundegelegt werden kann (vgl. Olsen 1985), eine Feststellung, die wohl auch für die Darstellung der diachronen und komparativen Verhältnisse gilt (vgl. Lenerz 1984, 1985b; Platzack 1983, Weerman 1989).

2.2. Andere germ. Sprachen

3.

Die historische ‘Entwicklung’ von Klammerkonstruktionen vom Ahd. bis zum Nhd.

Niederländisch (Nl.) und Friesisch zeigen im wesentlichen das gleiche Bild wie das Nhd., weisen aber im VK eine andere (Nl.: umgekehrte) Reihung auf. (ndh.: ... sing en lernen will, nl.: ... wil leren zing en). In den nordgerm. Sprachen (Dän., Schwed., Norw.) stehen Vi (+ VK) vor den vom Hauptverb abhängigen Satzgliedern, so daß eine verbale Klammer (Vf—VK) lediglich die sog. Wak-

(4a) Warun tho hirta in thero lantsceffi wahhante (T. 6,1) (‘waren da Hirten in dieser Landschaft wachend’) (4b) Bruthlauft ist gawisso garo (MFr 15,18) (‘Hochzeit ist gewiß bereitet’)

1268

(4c) erino portun ih firchnussu (Isid. 6,2) (‘eherne Pforten ich zerschmettere’) Die Endstellung begegnet dabei vor allem in Nebensätzen, z. T. auch ohne konjunktionelle Einleitung (Behaghel 1932, 46 ): daz mir g ot alle mine schulde virg ebe (MSD. 307,20). er chit, wola so tuon muosi (‘er spricht, (daß er) zu Recht so handeln dürfe’; N. I 58,27). Die Möglichkeit, das Vf vor infinite Elemente des VK zu stellen oder Satzglieder auszuklammern, ist stärker als im Nhd. ausgeprägt (vgl. Behaghel 1932, 88 ff; 108 ff; 78; 14): dher dhar scoldii chiboran werdan (‘der da sollte geboren werden’; Isid. 24,13). dhasz dhu firstandes heilac chiruni (‘daß du verstehest heiliges Geheimnis’; Isid. 6 .5). Damit ist die Bildung einer Satzklammer schon im Ahd. belegt (Bolli 1974). Im HS überwiegt V2, in Fragesätzen, Konditional- und Konzessivsätzen V1; V1 tritt gelegentlich (Emphase, Themawechsel) auch in Aussagesätzen auf. VEnd liegt bei Nebensätzen vor, ist jedoch aufgrund von Umstellungen in VK oder Ausklammerungen häufig kaum erkennbar, weshalb die meisten Autoren die Bezeichnung „Spät(er)stellung“ oder „Nicht-Zweitstellung“ bevorzugen. Dies zeigt ein großes Maß an Übereinstimmung zwischen der Verbstellung des Ahd. und des Nhd. Die keineswegs zu leugnenden Unterschiede sind dabei durch verschiedene Ausnutzung der Umstellungsmöglichkeiten im VK und ins Nachfeld zu erfassen. Im Verlauf des Ahd. (Notker) ist eine deutliche Festigung des beschriebenen Musters zu beobachten: V1 und VEnd in Aussagesätzen gehen zurück. Die Klammerbildung wird weniger durch Umstellungen in VK und Ausklammerungen verwischt als noch in früheren Textzeugnissen. Das Bild im Mittelhochdeutschen (Mhd.) ist ähnlich. Es wird noch stark von der Möglichkeit der Umstellung in VK Gebrauch gemacht, häufig auch von der Ausklammerung (vgl. Behaghel 1900 und 1932, 108 f; 79 f): daz drie tag e solte han g ewert (Bert. I; 10,20). als der sun ein ist mit dem vater nach dem wesen (Myst. 158,35). Die Ausklammerungsmöglichkeiten nehmen jedoch seit dem Ahd. ab, eine Tendenz, die im Frühnhd. (Fnhd.) fortwirkt, offenbar unterstützt durch den stark normierenden Einfluß der Kanzleisprache. Vermuteter lat. Einfluß (vgl. Maurer 1926 , Behaghel 1932, 17; 21; 87) ist nicht sicher nachzuweisen (Fleischmann 1973, vgl. auch Scaglione 1981, 122 f und Ebert 1980a, 388). VK tritt immer häufiger in absoluter Endstellung auf (Behaghel 1932,

XX. Die Erforschung des Syntaxwandels III: Phänomene

106 ff). Eine weitere typische Eigenart des Fndh. ist die Zunahme der Anfangsstellung von verba dicendi (Maurer 1924, 16 1; 16 3; Behaghel 1932, 37 ff): sprichet sant Paulus ... (St. Georg. Pred. 16 4, 17) (lat. Einfluß, später auch bei anderen Verben). Als vermeintlich volksliedhafte Eigentümlichkeit findet sich V1 weiter in der Dichtung (Sturm und Drang, Expressionismus, Maurer 192 6 , 206 ff), aber auch umgangssprachlich (exklamativ oder in Witzen) (Behaghel 1932, 38 f): Kommt ein Vog el g eflog en, ... War das ein Wetter! Kommt ’ne Frau zum Arzt .... VEnd in HS oder V2 in NS ist in poetischen Texten bisweilen auch im Nhd. möglich (Dal 196 2, 175): Keiner den Becher g ewinnen will (Schiller). Und als er g ing im finstern Wald ... (Uhland, Siegfrieds Schwert). Mit der Festigung von VEnd in NS bzw. mit der Endstellung der Vi des VK im HS ist im wesentlichen der oben skizzierte Zustand des Nhd. erreicht. 3.2. Andere germ. Sprachen Über die Entwicklung in den anderen germ. Sprachen geben Gerritsen (1984) und Stockwell (1984) Auskunft. Gerritsen unterscheidet zwischen Klammerbildung (S Vf O V im HS, S O Vf im NS) und Inversion im HS (X Vf S). Aufgrund dieser beiden Kriterien ergibt sich eine Dreiteilung der neueren germ. Sprachen (vgl. Abb. 71.2). Englisch:

—Umklam- — obl. Inmerung version Skand. Spr.: — + Nhd., Friesisch, Nl.: + — Abb. 71.2: Klammerbildung und Inversion in germanischen Sprachen

In den frühesten Zeugnissen liegen im Prinzip ähnliche Verhältnisse, aber mit etwas größerer Variation vor. (Die strikteren Regelungen der neueren (Standard)-Sprachen mögen auf normierenden Einfluß zurückzuführen sein.) Dabei zeigten die frühesten Belege der skand. Sprachen relativ selten Umklammerung (also selten VEnd), aber stets Inversion. Die im Altenglischen (Ae.) noch mögliche Umklammerung geht völlig verloren, während sich die in den ältesten Zeugnissen gelegentlich anzutreffende Inversion zunächst obligatorisch verfestigt (100% um 958—1001, vgl. Bean 1976 , 145), um dann wieder zurückzugehen; das Neuenglische (Ne.) zeigt

71.  Klammerkonstruktionen

dabei im wesentlichen die gleichen (restringierten) Möglichkeiten zur Inversion wie das früheste Ae. (vgl. Stockwell 1984).

4.

Erklärungsversuche

4.1. Traditionelle Theorien Die wesentlichen traditionellen Theorien zur Entstehung der nhd. Verbstellung sind mit den Namen Delbrück und Behaghel verbunden (vgl. Fleischmann 1973, 33 ff; Scaglione 1981, 117). Aufgrund der Verhältnisse im Altindischen (Ai.) ging Delbrück (1878; 1911) von ursprünglicher VEnd-Stellung im Germ. aus. Das Vf habe sich jedoch im HS an das betonte erste Satzglied angeschlossen, so daß sich V2 ergab. Wackernagel (1882) wies nach, daß im Ai. das Vf im HS unbetont (V2), im NS aber betont (VEnd) war. Bei der Entstehung der NS im Germ. soll dann nach Delbrück Vf deshalb wieder an das Satzende gerückt sein, weil es durch neu entstandene, aus dem übergeordneten Satz an die Spitze des NS übertretende Konjunktionen aus der Anfangsstellung gewissermaßen verdrängt worden sei (vgl. auch Biener 1922, 173). Trotz naheliegender Kritik („gedanklicher Salto mortale“ , Fleischmann 1973, 6 8) enthält diese Erklärung einen wesentlichen Kern (Scaglione 1981, 131 f), der sich auch in neueren generativen Analysen wiederfindet (Lenerz 1984 und s. u. 4.4.). Tomanetz (1879) erschien Delbrücks Vorschlag als zu kompliziert. Er und Erdmann (1886 ) gingen stattdessen von ursprünglichem V2 aus, während sich VEnd erst später bei der Entstehung von NS entwickelt habe. Eine Erklärung für VEnd konnten sie allerdings nicht geben. Eine Zwischenstellung nimmt Behaghel (1929) ein. Er stellte fest, daß sich schon im Indogermanischen (Idg.) HS und NS durch die Verbstellung unterschieden hätten: „Mittelstellung“ (im wesentlichen V2) im HS und „Späterstellung“ im NS. So bleibt lediglich die absolute Endstellung des Vf zu erklären, die sich im Fnhd. als Norm herausgebildet hat. Behaghel sah hier den Einfluß des Humanistenlateins als Ursache an. Entsprechend versuchten Maurer (1926 ) und Biener (1926 ) direkten lat. Einfluß auf frühe dt. Schulgrammatiken nachzuweisen. Die Kritik (Fleischmann 1973) hob hervor, daß humanistische Lateingrammatiken keine absolute VEnd lehrten, sondern SV-Abfolge. Auch aus einer Untersuchung von Luthers Tischreden (Stolt 196 4) geht hervor, daß sich dort in den lat. Passa-

1269

gen weitaus seltener VEnd findet als in den deutschen. Eine neuere Untersuchung fnhd. Texte aus Nürnberg (Ebert 1980) deutet ebenfalls auf geringen lat. Einfluß hin. Deutlich erkennbar ist aber die Wirksamkeit dt. Schulgrammatiken, die absolutes VEnd in NS forderten (Fleischmann 1973, 325 ff). Es bleibt festzuhalten: In den frühesten germ. Quellen ist sowohl V2 in HS wie eine Späterstellung in NS zu beobachten (vgl. Braunmüller 1982), Sequenzen wie „AdverbVerb“ im Ahd. durch faran (‘durchfahren’) belegen VEnd. Ein erklärender Zusammenhang zwischen beiden Stellungsmöglichkeiten sowie für die funktionale Differenzierung (V2 = HS; VEnd (Späterstellung) = NS) wird von Delbrück und Wackernagel aufgrund satzrhythmischer Gründe angedeutet. Absolute VEnd-Stellung im Nhd. ist eine späte (wohl normierende) Entwicklung, wobei stets Ausklammerungsmöglichkeiten bleiben, die auch historisch anzunehmen sind: („echt deutsche Art“ : Behaghel 1932, 267). Neben der Wirkung des Rhythmus, die auch für Umstellungen im VK herangezogen wird (Maurer 1926 , Behaghel 1932, 88—108, Ebert 1980), wird „Analogie“ wiederholt als Begründung genannt (Biener 1922, 170). Beide Konzepte sind aber nicht expliziert und bleiben notwendigerweise vage (vgl. Mayerthaler 1980 und 1984). 4.2. Fourquet In einem strukturalistisch orientierten Ansatz untersucht Fourquet (1938, vgl. 1974) frühe ae., altsächsische (as.) und ahd. Zeugnisse (vgl. Fleischmann 1973, 230 ff, Ebert 1978, 34 ff, Scaglione 1981, 135 ff). Die urtümlichste uns erreichbare Stufe des Germ. sieht er im ae. Beowulf repräsentiert (System I). Hier steht das Verb am Ende. Umstandsangaben oder das Subjekt können voranstehen (Thema), aber auch nachfolgen, wenn sie besonders gewichtig sind. VEnd ist die unmarkierte Stellung, V1 dient der markierten Hervorhebung des gesamten Satzinhaltes. Pronominale Satzglieder (pp) stehen immer vor nominalen (nn): pp ... nn ... v (unmarkiert); v pp ... nn (markiert). VEnd zeigt dabei „einen klaren Fall „zentripetalen“ Aufbaus im Sinne Tesnières“ (Fourquet 1974, 318): die hierarchische dependentielle Gliederung wird durch eine (einheitliche) lineare Abfolge dargestellt (vgl. 4.3.: OV-Serialisierung). In einer auf das „System I“ folgenden Zwischenstufe (erster Teil der ae.-Sachsenchronik (bis 891) und as. Heliand) wird V2 im HS vorherrschend, wäh-

1270

rend im eingeleiteten NS Späterstellung des Vf vorliegt. Pronominale Elemente können noch vor Vf stehen, so daß noch keine strikte V2-Stellung vorliegt: se papa hine heht petrus (‘der Papst ihn nannte Petrus’). V2 tritt nur bei nominalen Satzgliedern (se cyning ), aber nicht bei pronominalen Elementen (he) auf: (þy ilcan g eare for se cyning to Rome (‘in diesem Jahr fuhr der König nach Rom’) vs.: þy ilcan g eare he for to Rome). Der markierten Konstruktion mit V1 im „System I“ entspricht auf dieser Zwischenstufe zum „System II“ die Voranstellung einer satzeinleitenden Partikel pa/paer (‘da’) (þa/þaer for he to Rome). Erst im „System II“ (zweiter Teil der ae. Sachsenchronik (891—925), altnordische Edda und ahd. Isidor) ist V2 im HS so weit durchgeführt, daß auch pronominale Elemente meist auf das Verb folgen (ih sag hem dhir dhasz ... (‘ich sage dir, daß ...’; Isid. 37, 8)); allerdings finden sich Reste der früheren Wortstellung (34b), die sogar im Isidor noch überwiegen: ih inan infahu (‘ich ihn empfange’; Isid. 18, 18) (Vorlage: suscipiam eum). Insgesamt aber sind die ahd. Texte im „System II“ zu beschreiben; dies ist nach Fourquet Ergebnis eines Wandels, durch den das Verb in Aussagesätzen aus einer Späterstellung in die Zweitstellung übergetreten war. Dieser Wandel scheint in allen germ. Sprachen seit dem 6 . Jahrhundert aufzutreten und ist bis ca. 900 abgeschlossen. Fourquets Darstellung vermag allerdings keinen Grund dafür anzugeben, wieso das Verb im Übergang vom System I zu System II gerade an die zweite Position im Satz tritt. 4.3. Der typologische Ansatz Nach Greenberg (196 3) werden Sprachen gemäß der Abfolge von Subjekt (S), Objekt (O) und Verb (V) im normalen Aussagesatz klassifiziert (SOV, SVO, VSO). SOV-Sprachen weisen häufig folgende Eigenschaften auf: Sie haben fast immer ein Kasussystem und Postpositionen; wenn ein Genitiv auf das regierende N folgt, dann stehen auch attributive Adjektive hinter N; dem Hauptverb folgen die ihm übergeordneten Hilfs- und Modalverben. (Vgl. Art. 55.) Auf der Grundlage solcher typologischen Universalien rekonstruierte Lehmann (1972) das Idg. und das Urgerm. als SOV-Sprachen. Erscheinungen, die nicht mit Greenbergs Universalien übereinstimmen (z. B. Präpositionen), versuchte Vennemann (1973, 1974) aus Wortstellungsveränderungen zu erklären. In der sog. „Natürlichen Generativen Gram-

XX. Die Erforschung des Syntaxwandels III: Phänomene

matik“ (Bartsch/Vennemann 1972) wurden verschiedene, z. T. schon von Behaghel (1932, 3—9) entdeckte Gesetzmäßigkeiten in einem kategorial-grammatischen Rahmen formalisiert. Von zentraler Bedeutung ist dabei das sogenannte „Prinzip der natürlichen Serialisierung“ . Es besagt, daß semantisch ähnliche Verhältnisse zwischen zwei zusammengehörigen Satzgliedern vorwiegend durch die gleiche Abfolge dargestellt werden. So steht z. B. das spezifizierende Element (Operator) entweder stets vor dem spezifizierten (Operand) oder umgekehrt. OV-Sprachen bevorzugen die Abfolge Operator-Operand, VO-Sprachen die entgegengesetzte Abfolge (Vennemann 1974, 347): (Äpfel essen: im Deutschen (OV); eat apples: im Englischen (VO); Äpfel = Operator, spezifiziert den Operanden = essen, also: O(V)). Abweichungen von einer konsistenten Serialisierung im Sinne dieser allgemeinen Gesetzmäßigkeit sind nun nach Vennemann als Ergebnisse von Sprachwandel zu betrachten. Auf den uns speziell interessierenden Fall bezogen stellt sich seiner Ansicht nach die Entwicklung wie folgt dar: Das Idg. als SOV-Sprache verfügt über ein Kasussystem, das eine ausreichende morphologische Differenzierung von Subjekt und Objekten erlaubt. Wenn nun ein Satzglied vorangestellt wird (als Thema oder um es kontextuell besonders hervorzuheben), so kommt es in SOV-Sprachen zur Ambiguität, wenn S und O nicht morphologisch unterschieden sind (NP NP V = SOV oder OSV). In diesem Falle tritt das Verb zwischen ein topikalisiertes Satzglied T und das Subjekt („Inversion“ ), um diese markierte Abfolge von SXV unterscheidbar zu machen (TVX-Sprache) (X = beliebiges Satzglied): (a) SXV mit Subjekt als Thema, (b) TVSX sonst. Wenn sich nun V2 in (b) analogisch durchsetzt und auch thematisierte (vorangestellte) Subjekte erfaßt, wird generell SVX als typische Abfolge erreicht: SXV hat sich also über TVX zu SVX gewandelt. Dabei ist der erste Schritt (SXV → TVX) funktional erklärt, während der zweite (TVX → SVX) als analoger Wandel verstanden wird. Gegen diese Theorie sind verschiedene grundsätzliche Einwände erhoben worden (Klein 1975, Sasse 1977). So scheint insbesondere die Disambiguierungshypothese unhaltbar, nach der das Verb zwischen ein topikalisiertes Satzglied und das Subjekt tritt, um bei Verlust einer eindeutig unterscheidenden Kasusmarkierung eine Disambiguierung zwischen XSV und SXV zu ermöglichen. Im

71.  Klammerkonstruktionen

Nhd. ist derartige Ambiguität durchaus vorhanden (Die Mutter küßte die Tochter: SVO und OVS). Auch zeigte sich, daß es recht viele SOV-Sprachen ohne Kasusmorphologie gibt, die gleichwohl Voranstellung (TSV) erlauben (Sasse 1977). Zudem läßt sich beobachten, daß sich SOV-Sprachen zu SVOSprachen entwickelt haben, bevor Kasusverlust eintrat (vgl. Sasse 1977, 89 und SeefranzMontag 1983). Weitere Kritik ist an den unscharfen Definitionen der grundlegenden Begriffe der SOV- bzw. SVO-Sprache sowie am Prinzip der Kategorienkonstanz geübt worden (Klein 1975). Eine andere funktionale Deutung des Übergangs von SOV zu SVO bietet die sog. „afterthought“ -Hypothese (Hyman 1975). Es ist in einer SOV-Sprache zunächst als reine Performanz-Erscheinung zu bewerten, wenn man nach dem Verb noch ein Satzglied nachträgt. Wenn ein solcher stilistischer Nachtrag besonders häufig auftritt, kann er grammatikalisiert werden (zur textgrammatischen Funktion vgl. Sasse 1977, 128 ff). Entsprechend hatte Vennemann (1974) die Ausklammerung als wesentliches Mittel bei der Umstellung von SOV zu SVO angesehen. Dies erweist sich jedoch am Beispiel der Sprachgeschichte des Deutschen nicht als überzeugend, da hier die Ausklammerung seit dem Ahd. abnimmt (vgl. Kavanagh 1970), so daß man wohl eher von einer entgegengesetzten Tendenz zur Verfestigung einer SOV-Abfolge ausgehen müßte. Auch die Einbeziehung nachgestellter Sätze in den Hauptsatz (Subordination) ist als Ursache für eine funktional ausgenutzte Unterscheidung der Früher- bzw. Späterstellung des Vf in Haupt- und Nebensatz erwogen worden (Vennemann 1984a, b). Arealtypologische Untersuchungen (Bechert 1976 , Sasse 1977) lassen den Gedanken aufkommen, Sprachkontakt könne für die Entstehung von V2 in den germ. Sprachen verantwortlich sein (vgl. Bean 1976, 67 ff). 4.4. Die generative Analyse Aus der Übertragung der traditionellen Begriffe der Satzklammer und der Satzfelder in die generative Grammatik ergibt sich ein Erklärungsansatz für die synchronen Wortstellungsverhältnisse im Nhd. (4.4.1.), der sich weitgehend auf die diachronen Verhältnisse übertragen läßt (4.4.2.). 4.4.1.  Je nach dem theoretischen Modell wa-

1271

ren in der generativen Grammatik verschiedene Abfolgen als für das Dt. zugrundeliegend angenommen worden (vgl. Kohrt 1976 ). Im Rahmen der sog. „Extended Standard Theory“ hat Koster (1975) gezeigt, daß für das Nl. zugrundeliegend SOV anzunehmen ist. Diese Basisabfolge (die nicht mit der typologischen in 4.3. gleichzusetzen ist) erlaubt es, die erforderlichen Umstellungen im Einklang mit generellen Prinzipien zu formulieren, durch die die generative Kraft der Grammatiktheorie in erwünschter Weise auf nur wenige Optionen reduziert wird. Entsprechend ist auch für das Nhd. zugrundeliegende VEndstellung angenommen worden (vgl. den Besten 1977, Thiersch 1978, Lenerz 1981, Olsen 1982, 1985; als Überblick: Platzack 1985; zu unterschiedlichen Vorschlägen in Detailfragen insbesondere zur Begründung von V2 etc. vgl. Grewendorf 1986 , s. auch Haider/ Prinzhorn 1986) (vgl. Abb. 71.3).

Abb. 71.3: Basisabfolge im Nhd.

COMP stellt die Position dar, die in Nebensätzen durch eine subordinierende Konjunktion oder durch ein Relativpronomen besetzt wird. Ist das der Fall, so verbleibt Vf am Ende des Satzes. Fehlt aber eine derartige Besetzung von COMP, so tritt Vf an diese Stelle (subordinierende Konjunktionen und Vf in V2 sind komplementär verteilt!). In diesem Falle (V1) kann zudem fakultativ ein Satzglied vorangestellt werden, so daß sich V2 ergibt. Zusätzlich ist Adjunktion eines Satzgliedes oder eines Gliedsatzes an S möglich (Ausklammerungen ins Nachfeld). Schließlich muß für Umstellungen in VK eine lokale Umstellungsregel angenommen werden (den Besten/Edmondson 1981). Diese — hier nur in ihren Grundzügen skizzierte — Analyse weist eine interessante Übereinstimmung mit der diachronen Analyse von Delbrück (s. 4.1.) auf, die ja ebenfalls einen inne-

1272

ren Zusammenhang zwischen (neben)satzeinleitender Position und VEnd-Position annimmt und Vf durch nebensatzeinleitende Konjunktionen von seinem satzeinleitenden Platz verdrängt sieht. 4.4.2.  Entsprechend läßt sich diese Analyse auf die diachronen Verhältnisse übertragen (Lenerz 1984, 1985b). Nach Fourquet (4.2.) besteht die entscheidende Veränderung der syntaktischen Struktur im frühesten Ae. (Beowulf) gegenüber dem Urgerm. darin, daß das Verb nun auch in Spitzenposition auftreten kann. Eine derartige stilistisch bedingte und markierte Voranstellung (vgl. Dressler 196 9) wäre zunächst als Adjunktion zu analysieren: (t markiert als Spur (trace) die „ursprüngliche“ Position von Vf):

XX. Die Erforschung des Syntaxwandels III: Phänomene

ren: eine stilistische Voranstellung eines Satzgliedes wird als grammatisch gegebene Strukturmöglichkeit re-analysiert. Festzuhalten ist, daß sich V1- und V2-Strukturen seit dem Ahd. auf die oben beschriebene Weise analysieren lassen. Die Besetzung des Vorfeldes wie die Ausklammerung gehorchen seit dem Ahd. den gleichen Bedingungen wie im Nhd. (Zur Ausklammerung vgl. Kavanagh 1970, Bolli 1975, für das Ahd. Notkers, Ebert 1980, für das Fnhd., Schildt 1972, 1976 ). Die Möglichkeiten zur Umstellung im VK sind dialektal und umgangssprachlich im Nhd. größer als in der Standardsprache. Ältere Sprachstufen zeigen hier die gleiche Variationsbreite wie heutiger nicht-standardisierter Gebrauch: (7a) In dhes dagum scal iuda uuerdhan chihaldan (Isid. 651 f) In poetischen, archaisierenden Texten hat es daneben stets uneingeleitete Sätze mit VEnd gegeben: (7b) .../Ihm längst von böser Schadenlust/ die schwarze Seele schwoll (Schiller)

Die so erzeugte Struktur läßt sich nun in Übereinstimmung mit einer — universell gegebenen — Regel für Satzstrukturen re-analysieren, die eine satzeinleitende Position COMP definiert. (Dabei ist es offenbar ein sprachspezifisch festzulegender Parameter, ob COMP realisiert wird oder nicht, ob es also (neben)satzeinleitende Konjunktionen gibt oder nicht). Entsprechend wird (5b) gemäß 6( a) als 6( b) re-analysiert:

Aufgrund des in (6 a) definierten Knotens COMP wird die Voranstellung von Vf in COMP durch „move α“ von allgemeinen Prinzipien erzwungen, wenn COMP nicht anderweitig lexikalisiert ist. Entsprechend ist die Entstehung von V2-Strukturen zu erklä-

Sie scheinen einen alten Strukturtyp zu repräsentieren und gehören wohl schon im Mhd. nicht mehr zum Kernbereich grammatischer Strukturen (vgl. Horacek 19 6 4, Lenerz 1985 b). 4.4.3.  Die skizzierte generative Analyse scheint im Prinzip dazu in der Lage zu sein, das Phänomen der Satzklammer und der verschiedenen Verbstellungen aufgrund genereller Prinzipien nicht nur synchron für alle germ. Sprachen zu erklären (Haider/Prinzhorn 1986 ), sondern auch die diachronen Verhältnisse in wesentlichen Zügen zu erfassen (Weerman 1989). Dennoch ist auf die Probleme hinzuweisen, die sich mit der skizzierten Analyse des syntaktischen Wandels ergeben: So ist zunächst völlig unklar, inwiefern Re-Analysen (5b) wirklich nachweisbar sind. Das liegt u. a. daran, daß eine explizite und detaillierte Theorie parametrischer Variation (COMP/INFL-Parameter, vgl. Platzack 1983, Lenerz 1985) nur in Grundzügen erkennbar ist. Auch eine genauere Untersuchung dessen, was synchron und diachron als Re-Analyse angesehen wird, steht bislang aus. Einen Fortschritt bringt hier die Arbeit von Weerman (1989). Die traditionelle Vorstellung, daß Flexionsverlust und feste Wortstellung funktional äquivalent sind und daß Flexionsverlust für die Wortstellungsveränderungen in den germ. Sprachen verantwortlich

71.  Klammerkonstruktionen

ist, versucht Weermann in der GovernmentBinding Theorie in folgender Weise zu explizieren: In der D-Struktur definierte und für LF relevante Information muß auf der SStruktur „sichtbar“ sein: So werden etwa die Θ-Rollen der Argumente des Verbs durch die Kasus der entsprechenden NPs erkennbar. Kasus kann nun morphologisch oder strukturell identifiziert werden. Im letzteren Fall wird er durch die Wortstellung (OV oder VO) ausgedrückt. Entsprechendes gilt nach Weerman auch für die V-Projektion (Vmax = S): Der Modus wird entweder durch morphologisch ausreichend differenzierte Flexion des Verbs oder durch die Position des Vf in COMP erkennbar. Daraus ergibt sich nun auch eine Erklärung für die historische Entwicklung. Bei ausreichend differenzierter Kasusmorphologie kann ein Argument vor oder nach dem regierenden Verb auftreten (VO und OV in den älteren germ. Sprachen, vgl. Braunmüller 1982). Erst bei Kasussynkretismus wird zur S-Identifizierung die Abfolge sprachspezifisch auf VO oder OV festgelegt. Anstelle einer Umstrukturierung oder Re-Analyse nimmt Weerman (1989, 173) also nur an, daß der Parameter verändert wird, der festlegt, auf welche Weise Kasus auf der S-Struktur identifiziert wird. Kinder haben dabei lediglich zu lernen, ob Kasus morphologisch oder syntaktisch identifiziert werden. Ein solcher Parameter erscheint lerntheoretisch sinnvoll (vgl. auch Clahsen 1988) und ist ohne Mühe auf die Verhältnisse beim Verb übertragbar: Der Modus ist durch die Konjugation des Verbs auf der S-Struktur „inhärent“ identifizierbar, wenn die Verbflexion z. B deutlich zwischen Indikativ und Konjunktiv unterscheidet. Gilt das (durch den phonologisch zu erklärenden Flexionsverlust) in den rezenten germ. Sprachen nicht mehr, so muß das Vf in genau den Fällen in COMP auftreten, in denen sich dort kein eigenständig modustragendes Element (Subjunktion) befindet. Sätze mit Endstellung des Vf (vgl. 7a) in früheren Stufen des Germ. sowie in poetischen Texten wären demnach „inhärent“ modusmarkiert. Der Wandel wird durch die Festlegung des gleichen Parameters erklärt wie bei der S-Identifizierung der Kasus. Diese Deutung kommt ohne unbefriedigende „Re-Analysen“ aus, die frühere Ansätze prägten (Lightfoot 1979, Lenerz 1984). Gleichwohl bleiben einige Probleme, insbesondere mit einer dritten Position, an der (im Engl., Jidd., Nisl.) finite Verben auftreten können (vgl. Lenerz (1991)). So muß Weer-

1273

man für einfache Sätze des Engl. inhärente SIdentifizierung annehmen, da Vf hier nicht in COMP auftritt. Diese Lösung erscheint jedoch weniger überzeugend: Die Flexion des Engl. ist nur wenig ausgeprägt, so daß man ihr die Möglichkeit, den Modus „inhärent“ auszudrücken, an sich nicht zusprechen möchte. Trotz dieses Einwands aber geht Weermans Ansatz einen Schritt weiter in dem Versuch, syntaktischen Wandel auf der Grundlage allgemeiner Prinzipien zu erklären, die synchron gelten und deren spezifische Ausprägung durch entsprechend einfach zu erlernende und unabhängig zu motivierende Parameterfestlegung zu verstehen ist. 4.5. Zur Entwicklung in anderen germanischen Sprachen Unter dem Blickwinkel der oben skizzierten generativen Analyse erweist sich die Unterscheidung von Gerritsen (1984) in Klammerbildung und Inversion (s. o. 3.2.) als wenig hilfreich. Vielmehr lassen sich auch die oberflächlich als „Inversion“ erscheinenden Fälle in den skand. Sprachen (und auch im Engl., vgl. Chomsky 1986 ) durch die Voranstellung des Vf nach COMP oder Infl deuten. S Vf O V und X Vf S O V unterscheiden sich demnach nur darin, ob ein Subjekt oder ein anderes Satzglied vor Vf gestellt wird: in beiden Fällen liegt Voranstellung von Vf und Klammerbildung vor. Der entscheidende Unterschied der skand. Sprachen zum Dt. liegt lediglich in der Abfolge der nominalen Elemente. Hier weisen die skand. Sprachen (und das Engl.) VO-Abfolge auf, während das Dt. OV-Abfolge zeigt. Daß allein das Engl. eine entscheidende Umstrukturierung erfahren hat, während alle anderen germ. Sprachen im wesentlichen ihre angestammten Wortstellungsmöglichkeiten beibehielten (Gerritsen 1984), wirft zudem ein bezeichnendes Licht auf die Frage nach der Möglichkeit syntaktischen Wandels überhaupt. Die engl. Entwicklung zur SVO-Sprache ist wohl im wesentlichen als Kreolisierung (oder Entlehnung) zu deuten (Gerritsen 1984), während die verbleibenden Möglichkeiten zur „Inversion“ (also Reste der alten Klammerbildung) im wesentlichen gleiche Bedingungen erfüllen wie im Ae. (Stockwell 1984).

5.

Literatur

Altmann, Hans. 1981. Formen der Herausstellung im Deutschen. Rechtsversetzung, Linksversetzung,

1274

Freies Thema und verwandte Konstruktionen. Tübingen. Bartsch, Renate, und Theo Vennemann. 1972. Semantic structures. Frankfurt a. M. Bean, Marian Callaway. 1976 . A study of the development of word order patterns in Old English in relation to theories of word order change. PhD. UCLA. Bechert, Johannes. 1982. Abduktiver Wandel und Sprachkontakt. Sprachtheorie und angewandte Linguistik. Festschrift für A. Wollmann, hrsg. von W. Welte, 203—210. Tübingen. Behag hel, Otto. 1900. Zur deutschen Wortstellung. Wissenschaftliche Beihefte zur Zeitschrift des allgemeinen deutschen Sprachvereins 17/18. 233—251. —. 1923, 1924, 1928, 1932. Deutsche Syntax, 4 Bde. Heidelberg. —. 1929. Zur Stellung des Verbs im Germanischen und Indogermanischen. Zeitschrift für vergleichende Sprachforschung auf dem Gebiet der indogermanischen Sprachen 56. 276—281. den Besten, Hans. 1983. On the interaction of root transformations and lexical deletive rules. On the formal syntax of the West Germania, ed. by W. Abraham, 47—131. Amsterdam. —, und Jerold Edmondson. 1981. The verbal complex in Continental West Germanic. Groninger Arbeiten zur Germanistischen Linguistik 19. 11—61. Biener, Clemens. 1922. Wie ist die nhd. Regel über die Stellung des Verbums entstanden? Zeitschrift für deutsche Altertumsforschung 59. 165—179. —. 1926 . Die Stellung des Verbums im Deutschen. Zeitschrift für deutsches Altertum und deutsche Literatur 63. 225—256. Bolli, Ernst. 1975. Die verbale Klammer bei Notker. Untersuchungen zur Wortstellung in der Boethius-Übersetzung. Berlin, New York. Braunmüller, Kurt. 1982. Syntaxtypologische Studien zum Germanischen. Tübingen. Chomsky, Noam. 1986. Barriers. Cambridge, MA. Clahsen, Harald. 1988. Normale und gestörte Kindersprache. Amsterdam, Philadelphia. Clausen, Ove K. 196 9. Ein deutsches Satzschema. Kopenhagener germanistische Studien 1. 118—126. Dal, Ing erid. 196 2. Kurze deutsche Syntax auf historischer Grundlage. 2. Auflage. Tübingen. Delbrück, Berthold. 1878. Syntaktische Forschungen, III: Die altindische Wortfolge aus dem Çatapathabramana. Halle. —. 1911. Germanische Syntax. II: Zur Stellung des Verbums. Abhandlungen zur Königl. Sächs. Gesellschaft der Wissenschaften. Phil.-hist. Klasse XXVIII. Leipzig. Diderichsen, Paul. 1941. Sætningsbygningen i skaanse lov. København. —. 196 6 . Helhed og Struktur. Darin insbes.: Logische und topische Gliederung des germanischen

XX. Die Erforschung des Syntaxwandels III: Phänomene

Satzes. 1943. 52—6 3. und: Saetningsleddene og deres stilling- 30 ar efter. 196 6 . 36 4—386 . København. Drach, Erich. 1937. Grundgedanken zur deutschen Satzlehre. Frankfurt a. M. Dressler, Wolfg ang . 196 9. Eine textsyntaktische Regel der idg. Wortstellung. (Zur Anfangsstellung des Prädikatsverbums). Zeitschrift für vergleichende Sprachforschung 83. 1—25. Ebert, Robert Peter. 1978. Historische Syntax des Deutschen. Stuttgart. —. 1980. Social and stylistic variation in Early New High German word order: The sentence frame („Satzrahmen“ ). Beiträge zur Geschichte der deutschen Sprache und Literatur (PBB(W)) 102. 357— 398. fortgesetzt als: —. 1981. Social and stylistic variation in the order of auxiliary and nonfinite verb in dependent clauses in Early New High German. Beiträge zur Geschichte der deutschen Sprache und Literatur (PBB) 103. 203—237. Eng el, Ulrich. 1970a. Regeln zur Wortstellung. Forschungsberichte des Instituts für dt. Sprache. Bd. 5. März 1970. 3—148. —. 1970b. Studie zur Geschichte des Satzrahmens und seiner Durchbrechung. Studien zur Syntax des heutigen Deutsch. Paul Grebe zum 6 0. Geburtstag, hrsg. von Hugo Moser, 45—61. Düsseldorf. Erben, Johannes. 1954. Grundzüge einer Syntax der Sprache Luthers. Vorstudie zu einer Luther-Syntax, zugleich ein Beitrag zur Geschichte der deutschen Hochsprache und zur Klärung der syntaktischen Grundfragen. Berlin. Erdmann, Oskar. 1886 , 1897. Grundzüge der deutschen Syntax. Bd. I. Stuttgart 1886 . Bd. II bearb. von O. Mensing. Stuttgart. Fleischmann, Klaus. 1973. Verbstellung und Relieftheorie. Ein Versuch zur Geschichte des deutschen Nebensatzes. München. Fourquet, Jean. 1938. L’ordre des éléments de la phrase en germanique ancien — Etudes de syntaxe de position (Publications de la Faculté des Lettres de l’Université de Strasbourg, fasc. 86). Paris. —. 1974. Genetische Betrachtungen über den deutschen Satzbau. Studien zur deutschen Literatur und Sprache des Mittelalters. Festschrift für Hugo Moser zum 6 5. Geburtstag, hrsg. von Werner Besch, Günther Jungbluth, Gerhard Meissburger et al., 314—323. Berlin. Gerritsen, Marinel. 1984. Divergent word order developments in Germanic languages: A description and a tentative explanation. Historical syntax, ed. by Jacek Fisiak, 107—135. Berlin, New York, Amsterdam. Greenberg , Joseph. 196 3. Some universals of grammar with particular reference to the order of meaningful elements. Universals of Language, ed. by J. Greenberg, 73—113. Cambridge, MA.

71.  Klammerkonstruktionen

Grewendorf, Günther. 1986 . Deutsche Syntax. Teil II. KLAGE 13. Griesbach, Heinz. 196 0. Neue Grundlagen für den fortgeschrittenen Deutschunterricht I. Deutschunterricht für Ausländer 10. 97—109. Grubacic’, Emilija. 19 6 5. Untersuchungen zur Frage der Wortstellung in der dt. Prosadichtung der letzten Jahrzehnte. Zagreb. Haider, Hubert, and Martin Prinzhorn (eds.). 1986. Verb second phenomena in Germanic languages. Dordrecht. Herling , S. H. A. 1821. Über die Topik der deutschen Sprache. Abhandlungen des frankfurtischen Gelehrtenvereines für deutsche Sprache. Drittes Stück. Frankfurt a. M. 296—362; 394. Höhle, Tilmann. 1986 . Der Begriff „Mittelfeld“ . Anmerkungen über die Theorie der topologischen Felder. Kontroversen, alte und neue. Akten des VII. Internationalen Germanisten-Kongresses Göttingen 1985, hrsg. von Albrecht Schöne, Bd. 3: Textlinguistik contra Stilistik? Wortschatz und Wörterbuch. Grammatische oder pragmatische Organisation von Rede?, hrsg. von Walter Weiss, Herbert Ernst Wiegand, Marga Reiss, 329—340. Tübingen. Horacek, Blanca. 196 4. Kunstprinzipien der Satzgestaltung. Studien zu einer inhaltsbezogenen Syntax der deutschen Dichtersprache. Wien. (Österr. Akad. d. Wiss., ph.-hist. Kl., Sitzungsberichte, 243. Bd., 5. Abhandlung). Hyman, Larry M. 1975. On the change from SOV to SVO: Evidence from Niger-Congo. Word order and word order change, ed. by Charles N. Li, 115— 146. Austin, London. Kavanag h, Joseph Hayes. 1970. The verb-final Rule in High German: A diachronic analysis of surface structure constraints. Michigan: PhD dissertation. Klein, Wolfg ang . 1975. Eine Theorie der Wortstellungsveränderung. Einige kritische Bemerkungen zu Vennemanns Theorie der Sprachentwicklung. Linguistische Berichte 37. 46—57. Kohrt, Manfred. 197 6 . Koordinationsreduktion und Verbstellung in einer generativen Grammatik des Deutschen. Tübingen. —. 1978. The Long and Winding Road to German SOV, Aller — Retour, Taking the Short Cut. Linguistische Berichte 57. 69—73. Koster, Jan. 1975. Dutch as an SOV language. Linguistic Analysis 1. 111—136. Lehmann, Winfred P. 1972. Proto-Germanic syntax. Towards a grammar of Proto-Germanic, ed by F. v. Coetsem & H. L. Kufer, 239—268. Tübingen. Lenerz, Jürg en. 1981. Zur Generierung der satzeinleitenden Positionen im Deutschen. Akten des 15. Linguistischen Kolloquiums, Münster 1980, Bd. 1, hrsg. von Manfred Kohrt & Jürgen Lenerz, 171— 182. Tübingen.

1275

—. 1984. Syntaktischer Wandel und Grammatiktheorie. Eine Untersuchung an Beispielen aus der Sprachgeschichte des Deutschen. Tübingen. —. 1985a. Zur Theorie syntaktischen Wandels: das expletive es in der Geschichte des Deutschen. Erklärende Syntax des Deutschen, hrsg. von W. Abraham, 99—136. Tübingen. —. 1985b. Diachronic syntax: Verb position and COMP in German. Studies in German Grammar, ed. by J. Toman, 103—132. Dordrecht. —. 1991. Rezension von Fred Weerman: The V2 Conspiracy. Linguistische Berichte 131. Lig htfoot, David. 1979. Principles of diachronic syntax. Cambridge. Maurer, Friedrich. 1924. Zur Anfangsstellung des Verbs im Deutschen. Beiträge zur germanischen Sprachwissenschaft (Festschrift Behaghel). 141— 184. Heidelberg. —. 1926 . Untersuchungen über die deutsche Verbstellung in ihrer geschichtlichen Entwicklung. Heidelberg. Mayerthaler, Willi. 1980. Aspekte der Analogietheorie. Kommunikationstheoretische Grundlagen des Sprachwandels, hrsg. von H. Lüdtke, 80—130. Berlin, New York. Olsen, Susan. 1982. On the syntactic description of German: Topological fields vs. X-bar Theory. Sprachtheorie und angewandte Linguistik. Festschrift für A. Wollmann, hrsg. von W. Welte, 29— 45. Tübingen. —. 1985. On deriving V-1 and V-2 structures in German. Studies in German grammar, ed. by J. Toman, 133—163. Dordrecht. Platzack, Christer. 1983. Germanic word order and the COMP/INFL parameter. Working Papers in Scandinavian Syntax 2. Trondheim. —. 1985. A survey of generative analyses of the verb second phenomenon in Germanic. Nordic Journal of Linguistics 8. 49—73. Reis, Marg a. 1980. On justifying topological frames: „Positional Fields“ and the order of non-verbal elements in German. Documentation et Recherche en Linguistique Allemande. Vincennes (DRLAV). Papers 22. 59—85. Sætning sskema og dets stilling — 50 år efter. 1986. NyS 16 & 17. København. Sasse, Hans Jürg en. 1977. Gedanken über Wortstellungsveränderung. Papiere zur Linguistik 13/14. 82—142. Scag lione, Aldo. 1981. Komponierte Prosa von der Antike bis zur Gegenwart. Bd. II, Die Theorie der Wortstellung im Deutschen. Stuttgart (Übers. v. The theory of German word order from the renaissance to the present. Minnesota. 1981). Scherpenisse, Wim. 1985. Die Satzstrukturen des Deutschen und Niederländischen im Rahmen der GB-Theorie. Eine Reaktion auf Marga Reis’ Dop-

1276

pelkopfkritik. Erklärende Syntax des Deutschen, hrsg. von W. Abraham, 313—334. Tübingen. Schildt, Joachim. 1972. Die Satzklammer und ihre Ausbildung in hoch- und niederdeutschen Bibeltexten des 14. bis 16 . Jahrhunderts. Studien zur Geschichte der deutschen Sprache, hrsg. von Günter Feudel, 231—242. Berlin. —. 1976 . Zur Ausbildung des Satzrahmens. Zur Ausbildung der Norm der deutschen Literatursprache auf der syntaktischen Ebene (1470—1730), hrsg. von Kettmann & Schildt, 235—284. Berlin. Seefranz-Montag , A. v. 1983. Syntaktische Funktionen und Wortstellungsveränderung. Die Entwicklung „subjektloser“ Konstruktionen in einigen Sprachen. München. Stockwell, Robert P. 1984. On the history of the verb-second rule in English. Historical Syntax, ed. by Jacek Fisiak, 575—592. Berlin, New York, Amsterdam. Stolt, Birg it. 196 4. Die Sprachmischung in Luthers Tischreden: Studien zum Problem der Zweisprachigkeit. Acta Universitatis Stockholmiensis. Stockholmer Germanistische Forschungen 4. Uppsala. Thiersch, Craig . 1978. Topics in german syntax. Cambridge, MA: PhD dissertation. Thümmel, Wolf. 1980. Une conception skandinave de la grammaire positionelle: Les travaux de Paul Diderichsen. Documentation et Recherche en Linguistique Allemande. Vincennes (DRLAV). 22/23, 87—104. Tomanetz, K. 1879. Die Relativsätze bei den ahd. Übersetzern des 8. und 9. Jahrhunderts. Wien. van de Velde, Marc. 1979. Zum Vorfeld im heutigen Deutsch. Diss. Gent. Vennemann, Theo. 1973. Explanation in syntax. Syntax and Semantics, vol. II, ed. by John P. Kimball, 1—50. New York, San Francisco, London. —. 1974. Topics, subjects, and word order: From SXV to SVX via TVX. Historical Linguistics, vol. 2, ed. by John Anderson & Charles Jones, 339— 376. Amsterdam, New York, Oxford. —. 1984a. Typology, universals and change of language. Historical Syntax, ed. by Jacek Fisiak, 593—612. Berlin, New York, Amsterdam. —. 1984b. Verb-second, verb-late, and the brace construction. Historical Syntax, ed. by Jacek Fisiak, 627—636. Berlin, New York, Amsterdam.

XX. Die Erforschung des Syntaxwandels III: Phänomene

Wackernag el, Jacob. 1882. Über ein Gesetz der indogermanischen Wortstellung. Indogermanische Forschungen 1. 333—436. Weerman, Fred. 1989. The V2 conspiracy. A synchronic and a diachronic analysis. Dordrecht. Zemb, Jean Marie. 1986 . Beschreibung und Erklärung: Oder — oder oder Oder — und? Kontroverses zu Feldermodellen in deutschen Satzlehren. Kontroversen, alte und neue. Akten des VII. Internationalen Germanisten-Kongresses Göttingen 1985, hrsg. von Albrecht Schöne, Bd. 3: Textlinguistik contra Stilistik? Wortschatz und Wörterbuch. Grammatische oder pragmatische Organisation von Rede?, hrsg. von Walter Weiss, Herbert Ernst Wiegand, Marga Reiss, 320—328. Tübingen. Quellenverzeichnis Bert. = Berthold von Regensburg. Vollständige Ausgabe seiner Predigten mit Anmerkungen und Wörterbuch hrsg. v. Franz Pfeiffer und Joseph Strobl, 2 Bde., Wien 186 2 und 1880, neu herausgegeben von Kurt Ruh, Berlin 1965. Isid. = Der althochdeutsche Isidor. Nach der Pariser Handschrift und den Monseer Fragmenten neu hrsg. von Hans Eggers. Tübingen (ATB) 1964. MFr. = Des Minnesangs Frühling. Unter Benutzung der Ausgaben von Karl Lachmann und Moritz Haupt, Friedrich Vogt und Carl von Kraus bearbeitet von Hugo Moser und Helmut Tervooren, Stuttgart 361977. MSD = Denkmäler deutscher Poesie und Prosa aus dem 8.—12. Jahrhundert, hrsg. v. Karl Müllenhoff und Wilhelm Scherer, dritte Ausgabe von E. Steinmeyer, Berlin 1892. Myst. = Deutsche Mystiker des 14. Jahrhunderts, hrsg. v. Franz Pfeiffer, 2 Bde., Leipzig 1845 und 1857. N. = Die Werke Notkers des Deutschen, nach den Handschriften neu hrsg. v. Edward H. Sehrt und Taylor Starck, fortgesetzt von James C. King und Petrus W. Tax, Bde. 1 ff, Halle (Saale) und Tübingen (ATB) 1933 ff. St. Georg. Pred. = Der sogenannte St. Georgener Prediger, hrsg. v. Karl Rieder. Berlin 1908. T. = Tatian. Lateinisch und altdeutsch mit ausführlichem Glossar hrsg. v. Eduard Sievers, Paderborn 1892, Nachdruck ebda. 1966.

Jürgen Lenerz, Köln (Deutschland)

72.  Impersonalien

1277

72. Impersonalien 1.

2. 3. 4. 5. 6.

1.

Strukturtypen, Distribution und sprachtypologische Einordnung subjektloser Konstruktionen Zur morphosyntaktischen Entwicklung des Englischen und Deutschen Überblick über die Entwicklung englischer Impersonalien mit obliquen Ergänzungen Die Entwicklung deutscher Experiencer-Impersonalien Vergleichende Evidenz aus dem Nordgermanischen: Impersonalien im Isländischen Literatur

Strukturtypen, Distribution und sprachtypologische Einordnung subjektloser Konstruktionen

Der Begriff Impersonalien bzw. subjektlose Konstruktionen bezeichnet im folgenden verbale Fügungen ohne referentielle Nominativbzw. äquivalente Satz- oder Infinitivergänzung. (Vgl. zu den teils beträchtlichen Definitionsproblemen v. Seefranz-Montag 1983, Kap. 2.) 1.1. Strukturtypen unpersönlicher Ausdrücke Zum subjektlosen Konstruktionsmuster zählen (a) nullwertige Impersonalien zur Bezeichnung von Naturvorgängen u. ä. („Wetterverben“ ) wie lat. gelat, got. rigneiþ, dt. es wird Tag; (b) einwertige Impersonalien mit einem obliquen DAT/AKK/GEN/LOK-Komplement in Experiencer-Funktion wie got. mik huggreiþ; dt. mich friert sowie (c) zweiwertige Impersonalien mit obliquem Experiencer und einem (fakultativen) GEN/AKK/PP/Satz/Infinitiv-Komplement in Patiensfunktion wie lat. me taedet (alicuius rei), me. him remembreth of sth. zur Bezeichnung physischer, emotionaler und mentaler Erfahrungen (Hunger, Schmerz, Freude, Scham, Mangel usw.); (d) unpersönliche „Passiva“ und Reflexiva wie dt. mir wurde geholfen, span. se dice. (Vgl. zum Dt. z. B. Abramov 1967.) Der syntaktische Stellenwert von nicht-referentiellen Formalsubjekten wie dt. es, engl. it, frz. il bei Impersonalien (es schneit, mir graut (es)) ist strittig. Während obligatorische Formalsubjekte bei Wetterverben synchron als valenzdeterminierte Ergänzungen klassifiziert werden können (Reis

1982; Lenerz 1984), ist dies angesichts der variablen, nur zum Teil stellungsabhängigen Distribution von es bei den Konstruktionstypen (b)—(c) und insbesondere für eine historisch-vergleichende Analyse problematisch. Im folgenden wird daher zwischen referentiellen Subjekten einerseits und (i) obligatorischen, (ii) stellungsbedingten sowie (iii) korrelativen Formalsubjekten andererseits unterschieden. Der folgende Entwicklungsüberblick konzentriert sich auf Impersonalien mit obliquen Experiencern ((b) und (c)), die den historisch aufschlußreichsten Typus darstellen. 1.2. Zur Distribution von Impersonalien Innerhalb des Ide. waren subjektlose Experiencer-Konstruktionen weit verbreitet, z. B. im Altnordischen, Gotischen, Altsächsischen, Keltischen, Altslawischen und Lateinischen (Miklosich 186 5; Brugmann 1925, 26 ff; Wagner 1959, 54 ff). Sie sind noch heute gebräuchlich, v. a. im Slawischen (Vondrák 1928, 420 ff; Scholz 1973; Haiman 1976 ; Mrazović 1978; Nichols 1984, 200 ff), im Rumänischen und anderen romanischen Dialekten außer dem Französischen (Whalen 1978), im Keltischen (Vendryes 1956 ; Wagner 1959; Hartmann 1977), in einigen germanischen Sprachen wie dem Deutschen, Niederländischen, Dänischen, Schwedischen, v. a. aber im Isländischen (vgl. 4., 5. und z. B. Lenerz 1984), sowie im Indoiranischen (Masica 1976 , 16 1 ff; Klaiman 1981 und Mukherjee 1985 für Bengali; Andrews 1985, 101 f für Hindi). — Außerhalb des Ide. kommen Fügungen mit obliquen Experiencern im Finnougrischen (Fromm-Sadeniemi 1956 , 112 ff), im Hebräischen (Berman 1980), in kartvelischen Kaukasussprachen (Tschenkéli 1958, 446 ff; Wagner 1978; Boeder 1979, 455 f; Cole/Harbert/ Hermon u. a. 1980, 736 ff), im Dravidischen (vgl. Sridhar 1976 für Kannada), im Japanischen (Kuno 1973, 79 ff; 93; Shibatani 1977, 799 f) und anderen Sprachen vor. 1.3. Ansätze zur sprachtypologischen Einordnung von Impersonalien Die overte Markierung der ExperiencerFunktion durch einen obliquen Kasus setzt offenkundig ein Sprachsystem voraus, in dem Verbergänzungen bevorzugt aufgrund ihrer semantischen Relationen zum Verb kodiert werden und nicht aufgrund einer verbspezi-

XX. Die Erforschung des Syntaxwandels III: Phänomene

1278

fischen Valenz und Rektion mit obligatorischen, hierarchisch geordneten syntaktischen Funktionen wie Subjekt und Objekt, und das mittels einer leistungsfähigen Morphologie auch in der Lage ist, solche semantischen Differenzierungen formal zu repräsentieren. Aufgrund dieser Korrelation wurde gelegentlich eine typologische „Affinität“ zwischen subjektlosen Konstruktionen und aktivischer (teils auch ergativischer) Syntax postuliert, die ja gleichfalls durch eine vergleichsweise lockere, „quasi-adverbiale“ Verb-Argument-Beziehung gekennzeichnet sind (Moravcsik 1978, 254; 257; Wagner 1978, 42; 6 9 Fn. 125; Klimov 1979, 329; Schmidt 1979, 336 ). — Entsprechende historische Hypothesen über eine gemeinsame Abkunft unpersönlicher und aktivischer (evtl. ergativischer) Satzmuster aus einer Art verbnominaler Struktur der Form (es g ibt) Liebe/ Furcht (von-mir) (zu-dir) ‘ich liebe/fürchte dich’ mit funktional bestimmter Kasusselektion ohne Grammatikalisierung von Verbergänzungen, wie sie z. B. für das Vor-Ide. angesetzt wird (Velten 1931, 230; Vaillant 1936 , 107; Watkins 196 9, 49; Wagner 1978, 57 u. a.), sind bisher nicht über den Status von — gleichwohl plausiblen — Vermutungen hinausgelangt.

Unvereinbar sind Experiencer-Impersonalien andererseits mit den syntaktischen Prinzipien von konsequent „subjektprominenten“ Sprachen, in denen Form und Vorkommen von Verbergänzungen einheitlich geregelt sind (s. Artikel 56 : Prominenztypologie). Als syntaktisches Problem werden sich subjektlose Konstruktionen daher nur in Sprachen manifestieren, die irgendwo zwischen diesen beiden „typologischen Polen“ angesiedelt sind; und hierzu gehören die meisten der belegten ide. Einzelsprachen. Die folgende Skizze stellt die wichtigsten Veränderungsmechanismen subjektloser Ausdrücke anhand der gut dokumentierten Entwicklungen im Englischen und Deutschen sowie im Isländischen dar (vgl. ausführlich, auch zum Französischen und einigen nichtide. Sprachen, v. Seefranz-Montag 1983). Im Mittelpunkt stehen dabei die Wechselbeziehungen zwischen morphologischen und wortstellungssyntaktischen Veränderungen und der Ausbildung syntaktischer Funktionen.

2.

Zur morphosyntaktischen Entwicklung des Englischen und Deutschen

Die morphosyntaktische Entwicklung der heutigen ide. Einzelsprachen aus dem Pie. läßt sich in großen Zügen charakterisieren als

Übergang von einem Sprachtypus mit (a) überwiegend flexivisch (synthetisch) kodierten grammatischen Kategorien (z. B. Kasus), (b) vorwiegend präspezifikativen und für diskurspragmatische Zwecke veränderlichen Wortstellungsmustern („freie“ SXV-Stellung) und (c) einem System teilweise grammatikalisierter (polyfunktioneller und teils kongruentiell eingebundener), jedoch noch nicht uneingeschränkt obligatorischer und einheitlich markierter adverbaler Relationen (kombiniert semantisch-syntaktische Funktionen) zu einem Sprachtypus mit (a') überwiegend analytisch kodierten grammatischen Kategorien (z. B. Adpositionen und Wortstellungsmerkmale), (b') vorwiegend postspezifikativen, auch bei Veränderungen der Diskursstruktur aufrechterhaltenen Wortstellungsmustern (SVX bzw. TVX/SVX) und (c') einem verfestigten System weitgehend obligatorischer und einheitlich markierter syntaktischer Funktionen. (Vgl. Kurylowicz 196 4, 31; 179; Lehmann 1974; 1976 , 454; Dik 1978, 16 1; Kurzová 1986, 53 f u. v. a.) Am konsequentesten vollzog sich dieser Prozeß im Englischen: Aufgrund von Funktionsüberlappungen und Teilsynkretismen leistete schon das ae. Kasussystem weder eine eindeutige Identifizierung der semantischen noch eine verläßliche Distinktion der syntaktischen Funktionen von Verbargumenten. Als zunehmend dysfunktionales und intransparentes Kodierungsmittel wurden flexivische Kasus bis auf wenige Reste bis zum Ne. eliminiert und durch analytische Funktionsträger — Adpositionen und distinktive SVX-Stellung — ersetzt; präverbale Position gilt seit dem 14./15. Jh. als prinzipiell subjektspezifisches Charakteristikum. (Vgl. z. B. Fries 1940; MacLeish 196 9, 214 ff; Bean 1983. Zur umstrittenen relativen Chronologie von Kasusabbau und Wortstellungswandel s. z. B. Koch 1974; Vennemann 1975; 1984; Sasse 1977; Hawkins 1983, Kap. 5.5; Harris 1984 und in diesem Band Artikel 6 6 : Wortstellungswandel.) Im Deutschen wird dagegen die SubjektObjekt-Unterscheidung bis heute fast vollständig von den — wenngleich seit dem Ahd. reduzierten — Kasusmarkierungen getragen. Die relativ „synthetische“ Morphosyntax

72.  Impersonalien

stützt ihrerseits die im Deutschen sehr stabile Verbzweitstellung in Deklarativsätzen ab, während die residuale SXV-Stellung als Merkmal der Subordination von eingeleiteten abhängigen Sätzen funktionalisiert wurde (vgl. z. B. Russ 1980, Kap. II; Vennemann 1984). Wortstellungssyntaktisch entspricht das Gegenwartsdeutsche damit etwa dem Stadium des Frühmittelenglischen.

3.

Überblick über die Entwicklung englischer Impersonalien mit obliquen Ergänzungen

3.1. Die me hungreth-Konstruktion Im Ae. und Frühme. waren ein- und zweiwertige Impersonalien wie need, ail, think, shame, behove, list, hung er, chance, rue, like, long + DAT/AKK-Experiencer (gewöhnlich in Topic-Position) + GEN/PP/NOM/Satz/Infinitiv-Patiens (teils fakultativ) ein produktives syntaktisches Muster, das sogar Entlehnungen erfaßte (me dremeth, me semeth; me pleseth, me remembreth + Patiens) (Gaaf 1904, 4—25; Visser 196 3, 20—29). Bis zum 15./ 16 . Jh. ging dieser Konstruktionstypus bis auf vereinzelte idiomatisierte Relikte, die sich bis ins 19. Jh. hielten (meseems, methinks, melists), vollständig unter, und zwar auf drei Wegen: (a) durch lexikalischen Verlust von ca. 30 Verben (thynk, boot, rue); (b) durch die Einführung eines Formalsubjekts (h) it und/oder die Subjektivierung eines obliquen Patiens-Komplements (it/ sth. needs/likes/pleases me); (Unmarkierte Patiensterme wurden in ae./me. subjektlosen Konstruktionen i. a. nicht als Subjekte interpretiert, wie aus ihrer grundsätzlichen Tilgbarkeit und postverbalen Stellung, ihrer Substituierbarkeit durch oblique Terme und fehlenden Kongruenzauslösung hervorgeht, vgl. him deaueþ pa æren, him dimmeþ þa eiʒen (Worcester Fragm. I, B, 17).) (c) durch die Nominativierung des obliquen Experiencers (I need sth., I like sth., I think sth.). Zwischen dem 14. und 16. Jh., zwischen einer TVX- und einer bereits relativ stabilen SVXSyntax, koexistierten die strukturellen Alternativen für unpersönliche Ausdrücke als Varianten in einem im Umbruch befindlichen Sprachsystem: subjektlose Konstruktionen (1a, 2a), it-Konstruktionen mit obliquen (1b,

1279

2b) oder neueren adpositionellen Experiencern (1c), Konstruktionen mit nominativischen Patiens-Komplementen (1d, 2d) und mit Nominativ-Experiencern (1f, 2f). Bei französischen Lehnwörtern wie deynen, repenten und remembren wurde sogar die ursprüngliche persönliche Reflexivform (2e) zugunsten größerer semantisch-syntaktischer Einheitlichkeit aufgegeben (vgl. Gaaf 1904: 143—154). (1a) if thee like to be myn (Gower, Confessio Amantis, I, 950) (1b) it lykede him to take flesch (Mandeville’s Travels, I, 6 f) (1c) it lyketh to youre fader and to me þat I yow wedde (Chaucer, Canterbury Tales, E 345) (1d) this message lyketh me wel (Caxton, Le Morte D’Arthur, 187, 19) (1f) som what she likede hym the bet (Chaucer, Legend of Good Women, 1075 f) (2a) certes, thee remembreth wel (Chaucer, Boethius II, Prose VI, 11) (2b) it ne remembreth me (Chaucer, Boethius III, Prose III, 33) (2d) and what his compleynt was, remembreth me (Chaucer, Complaint of Mars 150) (2e) a man shal remembre hym of his synnes (Chaucer, C. T. I, 133) (2f) thou remembrest wel (Chaucer, Boethius I, Prose IV, 236) 3.2. Formalsubjekte und die Subjektivierung obliquer Patiens-Ergänzungen Das „oberflächlichste“ Verfahren zur Anpassung von Impersonalien an die Verbzweitstellungssyntax des Me., das bereits im Ae. konsequent auf nullwertige Verben angewendet worden war (die ja gewöhnlich keine topikalisierbare Ergänzung haben), ist die Besetzung der Topic-Position durch ein formales „TVX-Subjekt“ : Die Stellung vor dem Verb wird besetzt, der Valenzrahmen des betroffenen Verbs bleibt unangetastet (1a vs. b, c, 2a vs. b). Bei zweiwertigen Verben steht (h)it häufig in Korrelatfunktion zu nachfolgenden Patienstermen (1b, c) und damit in Konkurrenz zur regulären Subjektivierung dieser Ergänzungen (1d, 2d). — Bestand hatten Patiens-Subjekte bei englischen Impersonalien nur in sth. behoves/becomes/boots me, sth. happens/seems to me, und als Strukturvariante zur personalisierten Form in sth. ails/ smarts/grieves/pleases/delig hts me. Bei den übrigen Verben wurden it/Patiens-SubjektKonstruktionen um das 13. bis 16 . Jh. von

1280

Experiencer-Subjekt-Konstruktionen abgelöst. Sie sind daher als eine Art „Provisorium“ für ein historisches Übergangsstadium zu identifizieren, in dem (a) morphologische Kriterien (hier: objekttypischer Kasus) für die Zuordnung syntaktischer Funktionen höher bewertet werden als syntaktische (subjekttypische Topicposition, Kontrolle syntaktischer Prozesse, s. 3.3), so daß oblique Experiencer mithilfe der it-Insertion auch noch die objekttypische Position zugewiesen bekamen und gelegentlich sogar durch Adpositionen morphologisch „restauriert“ wurden (1c), und in dem andererseits (b) Subjekte so „prominent“ geworden sind, daß sie bei Impersonalien als formaler Topic-Ersatz eingesetzt werden: ein schon den SVX-Regeln entsprechendes TVXVerfahren. 3.3. Die Subjektivierung obliquer Experiencer-Ergänzungen (Personalisierung) Mit der Personalisierung wurde der Valenzrahmen subjektloser Verben umorganisiert von DAT/AKK (Experiencer) — Verb — GEN/PP/NOM/Satz/Infinitiv(Patiens) zu NOM (Experiencer) — Verb — AKK/PP (Patiens) wie in (3) me (DAT/AKK) remembreth of the day of dome (PP) (Chaucer, Canterbury Tales, I, 159) > I (NOM) remember sth. (AKK) Dieser Prozeß ist rekonstruierbar als kumulativer Erwerb zunächst (im Ae./Me.) der syntaktischen Kontroll- und Verhaltenseigenschaften (4a—e), dann (im Me.) der morphosyntaktischen Kodierungseigenschaften (5a, b) von Subjekten durch Experiencer-Argumente (vgl. Keenan 1976 ; Visser 196 3, 31; Butler 1977; Lightfoot 1979, 233 ff; Cole/ Harbert/Hermon u. a. 1980, 729 f; Elmer 1981, 49 ff). (4a) Subjekt-zu-Subjekt-Anhebung ae. þā ongan hine eft langian on his cȳþþe (Blickling Homilies, 113, 15) (4b) Subjekt-zu-Objekt-Anhebung me. ye me cause so to smerte (Chaucer, Troilus and Criseyde, IV, 1448) (4c) Kontrolle von Subjekt-zu-Objekt-Anhebung me. me þuncheð me unschriuen (Ancrene Riwle XIV, M 332)

XX. Die Erforschung des Syntaxwandels III: Phänomene

(4d) Koreferenz-Tilgung me. lewed meni (NOM) leued hym well and —i(DAT/AKK) liked his wordes (Langland, Piers the Plowman, prol. 72) (4e) Tilgung indefiniter Subjekte me. good is, quaþ Joseph, to dremen of win (Gen. & Ex. 2067) (5a) Subjektkongruenz me. sum men ... wondren what hem eilen (Chastising of God’s Children, 103, 15) (5b) Nominativ me. him ofte scomede (1205 Layamon, A 13763) > he ofte scamede (C 1250 Layamon, B 13763) Gleichzeitig assimilierten sich die betroffenen Verben morphologisch, z. B. durch Passivierung (I am ashamed/annoyed/disgusted/pleased), und semantisch an das neue Fügungsmuster. So wandelte sich die Bedeutung von like ‘gefallen’ (ne. please) zu ‘mögen’, die von need, lack und want ‘fehlen, nötig sein’ zu ‘brauchen, nötig haben’ usw. Die ursprünglich von der unpersönlichen Sonderkonstruktion vermittelte Bedeutung ‘unwillkürliche Erfahrung’ wurde in die lexikalische Verbstammbedeutung inkorporiert. 3.4. Erklärungsansätze zum Strukturwandel subjektloser Konstruktionen 3.4.1. Morphologisch begründete Hypothesen Der „Flexionsverfall“ und daraus entstehende morphologische Subjekt-Objekt-Ambiguitäten, die im Mittelpunkt der klassischen Untersuchungen englischer Impersonalien standen (Spies 1897, 278 ff; Gaaf 1904, 25—40; Jespersen 1927, 208; Fries 1940, 203; Mustanoja 196 0, 112; Visser 196 3, 30 u. a.), mögen für den Strukturwandel eine zumindest unterstützende Rolle gespielt haben. (6a) ae. him (DAT) behofaþ, þam cyninge (DAT) behofaþ > me. þe kyng (DAT/ NOM) behoveth, he (NOM) behoveth (6b) ae. þam eorlum (DAT Pl.) licode (Sg.) > me. þe erles (DAT/NOM Pl.) likode (Sg./Pl.) > ne. they (NOM Pl.) liked Allerdings können Kasusambiguitäten v. a. bei nominalen Experiencern der 3. P. Sg. bzw. im Präteritum (wegen der Verbkongruenz) Bedeutung erlangt haben. Gerade bei Verben für subjektive Empfindungen dürften jedoch Komplemente in der 1. und 2. P. sowie präsentische Verbformen vorherrschend gewesen sein (vgl. McCawley 1976 , 198); ferner wären

72.  Impersonalien

Experiencer bei wirklich kommunikationsstörender Ambiguität wohl konsequenter adpositionell abgestützt worden, als dies in Wirklichkeit der Fall war. 3.4.2. Syntaktisch begründete Hypothesen Vorrangig syntaktisch argumentierende Forschungsansätze führten die Nominativierung von Experiencern auf die Verfestigung der SVX-Stellung zurück — Fries’ (1940, 203) „pressure of position“ —, die, wieder v. a. bei formaler Ambiguität nach 3.4.1., eine Uminterpretation der obliquen Komplemente nach dem dominanten Serialisierungsmuster ausgelöst habe (Gaaf 1904, 3; Jespersen 1927, 208; Fries 1940, 203; Mustanoja 196 0, 112; vgl. Vennemann 1974, 280; Lightfoot 1979, 230; Elmer 1981, 135 ff). — Hierfür spricht die auffallende zeitliche Korrelation beider Prozesse im Englischen, wenngleich der subjektlose Konstruktionstyp bei der Durchsetzung von SVX im 14./15. Jh. in Form von analogischen Neubildungen aus persönlichen Verben noch durchaus produktiv war und insgesamt bis ins 16 . Jh. fortbestand. Dies mag man dem bekannten zeitlichen „Nachhinken“ morphologischer gegenüber syntaktischen Veränderungen (Givón 1971, 413) zuschreiben. Doch ist die SVX-Hypothese ebensowenig wie die Ambiguitäts-Hypothese in der Lage zu erklären, weshalb auch der Konstruktionstyp mit it-Subjekt bzw. referentiellem Patiens-Subjekt vor dem Verb — der doch den SVX-Regularitäten entspricht — zugunsten von Experiencer-Subjekt-Fügungen auf der Strecke blieb (vgl. 1d, f, 2d, f): (7) me. non other good of thee (NOM) me (DAT/AKK) nedeth (Gower, Confessio Amantis, III, 1311) > ne. I (NOM) need no other good of yours (AKK) 3.4.3. Semantisch-lexikalisch begründete Hypothesen Semantisch-lexikalisch ausgerichtete Erklärungsversuche leiten die Strukturveränderung mehrwertiger Impersonalien aus einer semantischen „Koaleszenz“ von Verb und Patiens-Argument ab (Marchand 1951) bzw. aus einer — gleichfalls nicht näher begründeten — Tendenz zur Durchsetzung derjenigen semantisch-syntaktischen Variante der betroffenen Verben, die Experiencer in „Perspektive“ setzt (Elmer 1981, 135 ff; Kovatcheva 1981; vgl. Trnka 1930, 56 ; McCawley 1976).

1281

3.4.4. Außersprachlich begründete Hypothesen Schließlich wurde der Verlust von Impersonalien als letztlich außersprachlich motiviertes Phänomen gedeutet, nämlich als Folge von Veränderungen in „Kulturstufe und Weltanschauung“ (Havers 1928; 1931, 105 ff; Wagner 1959, 50 f; Tripp 1978) bzw. als Auswirkung „psychologischer“ Veränderungen: „the greater interest taken in persons than in things“ (Jespersen 1927, 208).

4.

Die Entwicklung deutscher Experiencer-Impersonalien

4.1. Die mich friert-Konstruktion Dem aus heutiger Sicht sehr zielstrebig verlaufenen Abbau subjektloser Konstruktionen im Englischen steht im Gegenwartsdeutschen eine ziemlich uneinheitliche Menge konkurrierender Strukturformen gegenüber, wie sie für Phasen sich vollziehenden Sprachwandels typisch ist. — Ein Teil der im Ahd. und Mhd. noch fast uneingeschränkt verbreiteten Impersonalien mit obliquen Ergänzungen (vgl. Grimm 1898, 26 7 ff; Brugmann/Delbrück 1900, 29 ff; Behaghel 1923, 571 f; 6 14 ff; 6 94 ff; 1924, 128 ff u. v. a.) wurde lexikalisch (8a) oder in der alten Konstruktionsform verdrängt (8b). Im Gegensatz zum Englischen setzte diese stufenweise Reduktion im Ahd./ Mhd. erst sehr zögernd ein; umgekehrt blieb das subjektlose Muster noch bis ins 16 ./ 18. Jh. produktiv (Paul 1919, 34 ff). Der heutige Restbestand ist fast ausnahmslos archaischen Stilformen oder regionalen Dialektvarianten vorbehalten (8c) oder wurde während der letzten Jahrzehnte auf die „gehobene“ Schriftsprache beschränkt (8d). (8a)ahd. mir egisot, mir (un)willot u. a. mhd.mir/mich gerinnet, mir gollet, mich betrâget, mir gerîset u. a. fnhd.mich lüstert, mich lächert, mich schläfert, mich tanzert (8b)nhd. mir zweifelt, mich begehrt, mich leidet, mich schämt, mir/mich denkt (8c)nhd. mich hungert/dürstet/verlangt/ jammert/gelüstet/dauert/dünkt, mir beliebt/ahnt/bangt/träumt/ schwant; mich schwitzt/juckt/gruselt (reg.) (8d)mich fröstelt, mir/mich ekelt/schaudert/ schauert, mir (selten: mich) schwindelt

1282

Nur eine Handvoll Verben ist in der alten Form nach wie vor uneingeschränkt geläufig (9a). Erhalten sind insbesondere Konstruktionen mit prädikativen Adjektiven und Nomina (9b) sowie unpersönliche Passiva und Passivvarianten (9c). Diesen unpersönlichen Relikten gemeinsam ist jeweils mindestens eine topikalisierte Konstituente, also eine relativ große Konformität mit der VerbzweitSyntax. (9a) mich friert, mir graut (vor etwas), mir liegt an etwas (9b) mir ist/wird schlecht/übel/schwindlig/komisch/angst usw. (9c) mir wurde verziehen, nach mir wurde geschickt (vgl. nullstelliges hier wird geschlafen); ihm gehört gekündigt 4.2. Formalsubjekte und die Subjektivierung obliquer Patiens-Ergänzungen TVX-Subjekte etablierten sich bei ein- und mehrwertigen Impersonalien erst seit dem Spätmhd. (Brugmann 1917, 22 f; Maurer 1926 , 185—94; Henning 1957, 182 ff), nämlich nachdem um das 13. Jh. die Verbzweitstellung generalisiert worden war. Nicht stellungsbedingtes es, auch in korrelativer Funktion v. a. zu extraponierten Subjektsinfinitiven (11) wurde erst im Laufe des Nhd. regulär (vgl. auch Lenerz 1984, 104 ff). — Produktiv wurde die es-Konstruktion im heutigen Deutsch als Mittel zur Valenzminderung v. a. zur analogischen Neubildung ergänzungsloser Impersonalien (Lehmann/Spranger 196 8, 315 ff; Pütz 1975, 29 f): es stinkt; es kommt zu; neue es-Konstruktionen mit obliquem Experiencer sind dagegen fast ausnahmslos keine Impersonalien (so Wegener 1985), sondern reguläre Patiens-Subjekt-Fügungen (mir stinkt’s/dein Benehmen). (10) mir schwindelt > es schwindelt mir, mir schwindelt es; mir grauet vor der Götter Neide > es graut mir, mir graut es; mich friert > es friert mich, mich friert es (11a) Mir gefällt/mich wundert/verdrießt, daß du dich so schnell erholt hast/ (?) dich so munter zu sehen; mir gelang nicht, (?) ihn zu erreichen; (11b) es gefällt mir/wundert mich/ ... + S/Inf; es gelang mir + Inf (11c) mir gefällt/mich wundert/ ... es, + S/Inf mir gelang es + Inf Parallel zur es-Subjektivierung wurden v. a. seit dem 16 . bis 18. Jh. oblique Patiensergänzungen nominativiert (Paul 1919, 36 f).

XX. Die Erforschung des Syntaxwandels III: Phänomene

(12)mhd.umbe dîne swester ist mir ... leit (Nibelungenlied, 625, 2) daz ist mir groezlîchen leit (Nib. 1791, 4) mhd.mich reizet vaste darzuo (Hartmann v. Aue, Der Arme Heinrich, 1167) > nhd. das reizt mich Die häufigste Kodierungsform für NichtExperiencer-Argumente mehrwertiger Impersonalien war bis ins Frühnhd. der Genitiv (13a). Der seit dem 13. Jh. belegte Zusammenfall der Pronominalformen von mhd. NOM/AKK-ez und Genitiv-es in nhd. NOM/ AKK/GEN-es mag daher als morphologischer Faktor auf die im Deutschen auffallende Verbreitung von Patiens-Subjekten bei ehemaligen Impersonalien Einfluß genommen haben (vgl. Brugmann 1917, 32 f; Paul 1919, 36 ; Behaghel 1924, 139; Lockwood 1968, 172). (13)mhd.mich jamert + GEN/in/nâch, mir/ mich anet + GEN, mich dunket + GEN, mir/mich grûset + GEN, mir gelinget + GEN/an u. v. a. Wann immer diese Genitivkomplemente mit s pronominalisiert wurden, waren sie seit dem Nhd. in bezug auf ihre Subjekt- oder Objektfunktion ambig, um so mehr als die Stellungsregularitäten von genitivischen wie nominativischen es „in allen wesentlichen Punkten von jeher dieselben gewesen sind“ (Brugmann 1917, 34). In subjektlosem Kontext wurde daher für ambiges es nicht die eindeutige suppletive GEN-Form sîn eingesetzt, sondern der entsprechende Nominativ (14a), der die alten Impersonalien-Genitive schließlich generell ablöste (14b), solange die Konstruktion nicht personalisiert wurde (vgl. ich g edenke seiner, ich erfreue mich dessen/an/ über). (14a)ahd. nu ist es (GEN) not (Otfried IV, 30, 31) > nhd. jetzt ist es (NOM) nötig; ahd. sô thih es (GEN) wola lustit (Otfried I, 1, 14) > nhd. es (NOM) gelüstet dich (arch.) (14b)mhd.untz mir der speis (GEN) abgie (Gundacker 3709) > nhd. *dessen/das geht mir ab; vgl. das wundert/verdrießt/freut/reut/ jammert/erbarmt/wurmt mich; das genügt/gelingt/gefällt/geschieht mir

72.  Impersonalien

4.3. Die Subjektivierung obliquer Experiencer-Ergänzungen Vereinzelt im Ahd., in größerer Zahl seit dem Mhd. wurden subjektlose Konstruktionen durch die Nominativierung von DAT/AKKExperiencern personalisiert (vgl. Bernhardt 1903, 36 0; Lockwood 196 8, 170) (15) und dabei teilweise morphosyntaktisch umgeformt (16). (15)ahd. sie wuntarôtun (Tatian 78, 2) mhd.miner sorgen ich vergaz (Walter 94, 27) nhd. ich bange, ich schaudere, ich schwitze, ich fröstele, ich träume, ich zweifle, ich ahne usw. (16a) Periphrastische haben- und sein-Konstruktionen mich gelüstet > ich habe Lust auf; ich habe Hunger/Durst; ich bin hungrig/ durstig/schwindlig (16b) Präfigierung mich dauert > ich bedauere; ich bereue, ich benötige (16c) Reflexivierung mich wundert > ich wundere mich über; ich ekle mich vor, ich sehne mich nach, ich schäme/erinnere/erbarme mich Während die im Deutschen (bisher?) dominanten es- bzw. Patiens-Subjekt-Konstruktionen wie auch die mit morphosyntaktischen Veränderungen gekoppelten Personalisierungen (16 ) die ursprüngliche semantische Charakteristik unpersönlicher Ausdrücke konservieren, bewirkten die vergleichsweise wenigen Fälle „direkter“ Personalisierung auch im Deutschen entweder den Verlust der zuvor unpersönlich kodierten Bedeutungsvariante eines Verbs (17a) oder aber die Modifikation der lexikalischen Verbbedeutung nach der neuen syntaktisch-semantischen Valenz (17b). (17a) mich verlangt ‘ich sehne mich nach’ > ich verlange ‘ich fordere’; mich jammert ‘mir tut leid’ > ich jammere ‘ich klage’ (17b) mich betriuget ‘etwas verlockt mich’ > ich betrüge; mir schwindelt vs. ich schwindle; vgl. mich hungert (unwillkürlich) vs. ich hungere (willkürlich oder unwillkürlich) Neben dem Übergewicht von es- und PatiensSubjekt-Konstruktionen, die die oblique Markierung von Experiencern bewahren, zeigt sich der im Vergleich zum Englischen

1283

bedeutend schwächere Trend zur Experiencer-Subjektivierung im Deutschen an dem nur zurückhaltenden Erwerb subjekttypischer syntaktischer Eigenschaften durch die noch bestehenden obliquen Experiencer. Zwar zeigen sie, abgesehen vom Kasus, auch keine objekttypischen Eigenschaften wie die Erhaltung bei der „Zitierform“ von Verben (18a) oder Tilgung in Koreferenz zu Objekten (18b), sie können aber auch nur gelegentlich Reflexivierungen (18c) und einen Typ der koreferentiellen Tilgung von Nominativsubjekten (18d) kontrollieren (vgl. Reis 1982): Sie sind insgesamt noch „intaktere“ Subjekte als ihre me. Gegenstücke (vgl. (4), (5)). (18a) *jemandem grauen, *jemanden frieren (18b) *er verzieh miri und —i graute davor, ihn noch einmal zu enttäuschen (18c) Faust graute vor sich; ihm liegt nur an sich und seiner Bequemlichkeit (18d) ihm wurde schlecht, ohne von dem Hering gegessen zu haben; (?) ihm wurde geholfen, ohne darum gebeten zu haben 4.4. Zusammenfassung: Der Strukturwandel von Impersonalien im Englischen und Deutschen Insgesamt entspricht der Entwicklungsstand der nhd. subjektlosen Konstruktionen annähernd dem der mittelenglischen um das 14. bis 16. Jh. Betrachtet man andererseits (a) den Grad an morphologischen Reduktionen, (b) den Grad an Wortstellungsverfestigung und (c) die Form und Funktion grammatischer Relationen in beiden Sprachen, die im Englischen zeitlich und kausal relativ eng miteinander korreliert waren, ist es offensichtlich, daß die historische Entwicklung in Richtung auf eine einheitliche Kodierung der Relation Subjekt (c) im heutigen Deutschen Priorität gegenüber (a) und (b) besitzt: Während die morphosyntaktische Gesamtstruktur dieser Sprache erst etwa der des Ae./Frühme. entspricht, sind Nominativsubjekte — wenn auch nicht notwendig vor dem Verb — in beinahe allen Sätzen obligatorisch geworden. Immerhin gestatten die Regeln für die Subjektselektion im Deutschen mit einer Vielzahl von Konversen (ich wundere mich über — etwas wundert mich) noch einen relativ weiten Spielraum für die

XX. Die Erforschung des Syntaxwandels III: Phänomene

1284

Kennzeichnung unterschiedlicher semantischsyntaktischer Verb-Argument-Relationen. An der Strukturveränderung deutscher Impersonalien wird daher deutlich, daß die Etablierung syntaktischer Funktionen wie des Subjekts selbst in einer als notorisch „subjektprominent“ geltenden ide. Sprache — trotz einer gewissen Wechselbeziehung — keineswegs stringent mit dem Abbau flexivischer Morphologie und der Verfestigung der SVXSyntax zu erklären ist.

5.

Vergleichende Evidenz aus dem Nordgermanischen:

Impersonalien im Isländischen Unser Befund wird auch anhand des Isländischen erhärtet, der morphosyntaktisch konservativsten germanischen Sprache mit vorherrschender Verbzweitstellung und Verberststellungsresten (Einarsson 1945, 172 ff). Das unpersönliche Konstruktionsmuster wurde hier am geringsten von allen germanischen Sprachen restringiert. (a) Das Formalsubjekt það — als Exponent der TVX-Syntax (wie im Ae./Me. und Mhd./Nhd.) — ist nach wie vor auf satzeröffnende Stellung bei nullwertigen und passivischen Impersonalien beschränkt; bei Experiencer-Impersonalien haben sich weder (b) nominativierte Patiens-Argumente — als Exponenten einer fortgeschrittenen Grammatikalisierung von Subjekten (wie im Me. und Nhd.) — noch (c) nominativierte Experiencer-Argumente — als Exponenten der für die vollständige Grammatikalisierung von Subjekten charakteristischen Vereinheitlichung der Subjektselektion (wie im Ne. und teils im Nhd.) — durchgesetzt (vgl. Thráinsson 1979, Kap. 7). Vielmehr wird die oblique Markierung von Experiencern sogar durch den Dativ morphologisch aufgebaut (Collberg 1986 , 89; 91 f). Gleichzeitig werden jedoch diese Ergänzungen syntaktisch zu Subjekten umgepolt: Sie kontrollieren Reflexivierungen und Koreferenztilgungen und sind in Koreferenz zu Nominativsubjekten tilgbar (Kossuth 1976 , 30 ff; Cole/Harbert/Hermon u. a. 1980, 724; Collberg 1986 , 81 ff): Die vollständige Grammatikalisierung von Subjekten, die in der sukzessiven Subjektivierung obliquer Ergänzungen von Impersonalien manifest wird,

kommt auch im Isländischen lange vor Flexionsabbau und SVX in Gang. Anhand der vergleichenden Analyse von Flexions- und Wortstellungswandel einerseits und der historischen Veränderung subjektloser Konstruktionen andererseits ließen sich wiederkehrende Mechanismen bei der Konstitution von Subjekten nachweisen. Entgegen der herrschenden Forschungsmeinung ergaben die Fakten jedoch nur eine relativ grobe und teils widersprüchliche Wechselbeziehung zwischen der morphosyntaktischen Struktur einer Sprache und der jeweiligen Ausformung syntaktischer Funktionen, die die folgenden Diagramme (unter Einbeziehung analoger Entwicklungen in skandinavischen Festlandsprachen und dem Französischen, s. v. Seefranz-Montag 1983, Kap. 7.3 und 8) veranschaulichen.

Abb. 72.1: Skala relativer syntaktischer „Konservativität“ vs. „Progressivität“ germanischer Sprachen (v. Seefranz-Montag 1983, 210)

Eine über die Feststellung (oder Widerlegung) von Korrelationen hinausgehende Erklärung für die Ausbildung und die verschie-

72.  Impersonalien

1285

Abb. 72.2: Zur Korrelation von Wortstellungswandel und Impersonalien im Englischen, Französischen und Deutschen (v. Seefranz-Montag 1983, 241)

denen historischen Erscheinungsformen syntaktischer Relationen, wie sie in der Entwicklung von Impersonalien deutlich wurden, ist eine letztlich noch ungelöste Aufgabe für künftige sprachwissenschaftliche Forschungen.

6.

Literatur

Abramov, B. A. 196 7. Modelle der subjektlosen Sätze im Deutschen. Deutsch als Fremdsprache 4. 361—74. Andrews, A. 1985. The major functions of the noun phrase. Language typology and syntactic description, ed. by Th. Shopen, 62—154. Cambridge. Bean, M. C. 1983. The development of word order patterns in Old English. London.

Behag hel, O. 1923. Deutsche Syntax: eine geschichtliche Darstellung. Band I: 1923, Band II: 1924, Band IV: 1932. Heidelberg. Berman, R. A. 1980. The case of an (S)VO language: subjectless constructions in Modern Hebrew. Language 56.4. 759—76. Bernhardt, E. 1903. Beiträge zur mittelhochdeutschen syntax. I. Vom fehlen des subjectpronomens beim persönlichen zeitwort. II. Vom unpersönlichen zeitwort. Zeitschrift für deutsche Philologie 35. 145—56. 343—62. Boeder, W. 1979. Ergative syntax and morphology in language change: the South Caucasian languages. Ergativity: towards a theory of grammatical relations, ed. by F. Plank, 435—80. London. Brug mann, K. 1917. Ursprung des Scheinsubjekts ‘es’ in den germanischen und den romanischen

1286

Sprachen. Berichte über die Verhandlungen der Königl. Sächs. Ges. der Wissenschaften V. 1—57. Leipzig. —. 1925. Die Syntax des einfachen Satzes im Indogermanischen. Leipzig. —. und B. Delbrück. 1900. Grundriß der vergleichenden Grammatik der indogermanischen Sprachen. Band III. Straßburg. Butler, M. Ch. 1977. Reanalysis of object as subject in Middle English impersonal constructions. Glossa 11.2. 155—70. Cole, P., W. Harbert, S. Hermon und S. N. Sridhar. 1980. The acquisition of subjecthood. Language 56.4. 719—43. Collberg , Sh. D. 1986 . Oblique subjects in Icelandic passive constructions. Studia Linguistica 40.1. 67—95. Dik, S. C. 1978. Functional grammar. Amsterdam. Einarsson, S. 1945. Icelandic grammar, texts, glossary. Baltimore. Elmer, W. 1981. Diachronic grammar: the history of Old and Middle English subjectless constructions. Tübingen. Fries, Ch. 1940. On the development of the structural use of word-order in Modern English. Language 16. 199—208. Fromm, H. und M. Sadeniemi. 1956 . Finnisches Elementarbuch. Heidelberg. Gaaf, W. v. d. 1904. The transition from the impersonal to the personal construction in Middle English. Heidelberg. Givón, T. 1971. Historical syntax and synchronic morphology: an archaeologist’s field trip. Papers from the 7th regional meeting of the Chicago Linguistic Society. 394—415. Chicago. Grimm, J. 1898. Deutsche Grammatik. Neuer vermehrter Abdruck, hg. von W. Scherer, G. Roethe und E. Schröder. Hildesheim. 1967. Haiman, J. 1976 . Agentless sentences. Foundations of language 14.1. 19—54. Harris, M. B. 1984. On the strengths and weaknesses of a typological approach to historical syntax. Historical syntax, ed. by J. Fisiak, 183—97. Berlin. Hartmann, H. 1977. Das Impersonale im Keltischen und Indogermanischen: Probleme der Dominanz. Indogermanisches und keltisches Kolloquium der Indogermanischen Gesellschaft 1976 . 159—203. Wiesbaden. Havers, W. 1928. Primitive Weltanschauung und Witterungsimpersonalia. Wörter und Sachen 11. 75—111. —. 1931. Handbuch der erklärenden Syntax. Heidelberg. Hawkins, J. A. 1983. Word order universals. New York. Henning , J. D. 1957. Studien zum Subjekt imperso-

XX. Die Erforschung des Syntaxwandels III: Phänomene

nal gebrauchter Verben im Althochdeutschen und Altniederdeutschen unter Berücksichtigung altwestnordischer Zeugnisse. Diss. Göttingen. Jespersen, O. 1927. A modern English grammar on historical principles. Part III: Syntax (2nd volume). Heidelberg. Keenan, E. 1976 . Towards a universal definition of ‘subject’. Subject and topic, ed. by Ch. Li, 303—34. New York. Klaiman, M. H. 1981. Volitionality and subject in Bengali. Bloomington. Klimov, G. A. 1979. On the position of the ergative type in typological classification. Ergativity: towards a theory of grammatical relations, ed. by F. Plank, 327—32. London. Koch, M. 1974. A demystification of syntactic drift. Montreal Working Papers in Linguistics 3. 6 3— 114. Kossuth, K. C. 1976 . Subjectless sentences, word order change and non-nominative subjects in Icelandic. MS Pomona College und Westfälische Wilhelms-Universität Münster. Kovatcheva, M. 1981. Middle English impersonal constructions and their equivalents in Modern English. MS Universität Sofia. Kuno, S. 1973. The structure of the Japanese language. Cambridge, MA. Kurylowicz, J. 196 4. The inflecional categories of Indo-European. Heidelberg. Kurzová, H. 1986 . Typologie und die Rekonstruktion der indogermanischen Morphosyntax. Folia Linguistica 20. 49—80. Lehmann, D. und K. Spranger. 196 8. Zum Problem der subjektlosen Sätze. Zeitschrift für Phonetik, Sprachwissenschaft und Kommunikationsforschung 21. 304—22. Lehmann, W. P. 1974. Proto-Indo-European syntax. Austin. —. 1976 . From topic to subject in Indo-European. Subject and topic, ed. by Ch. Li, 445—56 . New York. Lenerz, J. 1984. Zur Theorie syntaktischen Wandels: das expletive es in der Geschichte des Deutschen. Erklärende Syntax des Deutschen, hrsg. von W. Abraham, 99—136. Tübingen. Lig htfoot, D. W. 1979. Principles of diachronic syntax. Cambridge. Lockwood, W. B. 196 8. Historical German syntax. Oxford. MacLeish, A. 196 9. The Middle English subjectverb cluster. Den Haag. Marchand, H. 1951. The syntactical change from inflectional to word order system and some effects of this change on the relation ‘verb/object’ in English. Anglia 70. 70—89. Masica, C. P. 1976 . Defining a linguistic area: South Asia. Chicago.

72.  Impersonalien

Maurer, F. 1926 . Untersuchungen über die deutsche Verbstellung in ihrer geschichtlichen Entwicklung. Heidelberg. McCawley, N. 1976 . From OE/ME ‘impersonal’ to ‘personal’ constructions: What is a ‘subject-less’ S?. Papers from the Parasession on diachronic syntax of the Chicago Linguistic Society, 192—204. Chicago. Miklosich, F. 186 5. Die verba impersonalia im Slawischen. Denkschrift der Kaiserl. Akad. d. Wiss., Philos.-hist. Kl. XIV. Wien. Moravcsik, E. A. 1978. The distribution of ergative and accusative patterns. Lingua 45. 233—79. Mrazović, P. 1978. Verben ohne Nominativergänzung im Serbokroatischen und ihre Entsprechungen im Deutschen. Linguistische Arbeitshefte 21. 189—95. Mukherjee, Sh. 1985. Kasus und Diathese im Bengalischen. Heidelberg. Mustanoja, T. 196 0. A Middle English syntax. Band I. Helsinki: Soc. néophilologique. Nichols, J. 1984. Direct and oblique objects in Chechen-Ingush and Russian. Objects: towards a theory of grammatical relations, ed. by F. Plank, 183—209. London. Paul, H. 1919. Deutsche Grammatik. Band III. iv. 1. Halle. Plank, F. 1977. Typologische Überlegungen zu subjektlosen Sätzen. Salzburger Beiträge zur Linguistik 3. 83—105. Pütz, H. 1975. Über die Syntax der Pronominalform es im modernen Deutsch. Tübingen. Reis, M. 1982. Zum Subjektbegriff im Deutschen. Satzglieder im Deutschen, hrsg. von W. Abraham, 171—211. Tübingen. Russ, Ch. V. J. 1980. Historical German phonology and morphology. 2. Aufl. Oxford. Sasse, H.-J. 1977. Gedanken über Wortstellungsveränderung. Papiere zur Linguistik 13.14. 82— 142. Schmidt, K. H. 1979. Reconstructing active and ergative stages of Pre-Indo-European. Ergativity: towards a theory of grammatical relations, ed. by F. Plank, 333—45. London. Scholz, F. 1973. Russian impersonal expressions used with reference to a person. Paris. v. Seefranz-Montag , A. 1983. Syntaktische Funk-

1287

tionen und Wortstellungsveränderung: die Entwicklung ‘subjektloser’ Konstruktionen in einigen Sprachen. München. Shibatani, M. 1977. Grammatical relations and surface cases. Language. 53.4. 789—809. Spies, H. 1897. Geschichte des englischen Pronomens im XV. und XVI. Jahrhundert. Halle. Thráinsson, H. 1979. On complementation in Icelandic. New York. Tripp, R. P. 1978. The psychology of impersonal constructions. Glossa 12.2. 177—88. Trnka, B. 1930. On the syntax of the English verb from Caxton to Dryden. Prag. Tschenkéli, K. 1958. Einführung in die georgische Sprache. Band I. Zürich. Vaillant, A. 1936 . L:ergatif indo-européen. Bulletin de la Société de Linguistique de Paris 37. 93—108. Velten, H. V. 1931. On the origin of the categories of voice and aspect. Language 7.4. 229—41. Vennemann, Th. 1974. Zur Theorie der Wortstellungsveränderung: von SXV zu SVX über TVX. Zur Theorie der Sprachveränderung, hg. von G. Dinser, 265—314. Kronberg. —. 1975. An explanation of drift. Word order and word order change, ed. by Ch. Li, 26 9—305. Austin. —. 1984. Verb-second, verb late, and the brace construction in Germanic: a discussion. Historical syntax, ed. by J. Fisiak, 627—36. Berlin. Visser, F. Th. 196 3. A historical syntax of the English language. Band I. Leiden. Vondrák, W. 1928. Vergleichende slawische Grammatik. Band II. Göttingen. Wag ner, H. 1959. Das Verbum in den Sprachen der britischen Inseln. Tübingen. —. 1978. The typological background of the ergative construction. Proceedings of the Royal Irish Academy 78.C.3. 37—74. Watkins, C. 196 9. Indogermanische Grammatik. Band III. 1. Heidelberg. We g ener, H. 1985. Der Dativ im heutigen Deutsch. Tübingen. Whalen, S. 1978. The impersonal sentence in Russian and Romanian. Papers and studies in contrastive linguistics 8. 5—68.

Ariane v. Seefranz-Montag, München (Deutschland)

1288

XXI. Syntaktische Skizzen Syntactic Sketches

73. German 1. 2. 3. 4. 5. 6. 7.

1.

Position of the Verb Infinitives The Clausal Periphery Clause-Internal Structures Subject-Drop, Topic-Drop, Object-Drop Reflexivization References

Position of the Verb

According to the standard v iew, German is a “v erb second” language whose basic (Dstructure) constituent order is v erb-final. Opinions on the origin of the SOV order div erge. Lehmann (1971) and (1978) suggests that standard German adopted the OV-ordering in the period immediately following 1500; others (cf. e. g. Lenerz 1984) maintain that Old High German already displayed the same v erb placement characteristics as New High German, but that v arious reordering rules affecting v erbal and nonv erbal constituents had the effect of camouflaging the “final position” of the finite verb. The basic descriptiv e generalizations standardly made about German word order are the following. The finite v erb occupies final position (but precedes any extraposed constituents) in clauses introduced by a complementizer (1) and in indirect interrogativ es (2):

In finite sentences not introduced by a complementizer (which are usually but by no means always main clauses), the finite v erb appears in either second or first position:

Descriptiv e grammars treat the position of the finite v erb as standing in a functional relation with some specified sentence type. So the occurrence of the finite v erb in initial position usually indicates either a yes-no question (6) or an imperativ e (7), but may also mark concessiv e and conditional clauses (8), while second position typifies declarativ es (3) or direct wh-questions (4), and also occurs in clauses gov erned by certain v erbs that license complementizer-less complements (5). In infinitiv es, as (9) shows, the v erb appears in final position (again, preceding extraposed constituents); in finite sentences with v erb-fronting, the same is true of other non-finite members of the v erb complex, such as for example the separable prefix (10), the past participle (11) or bare infinitiv e form (12):

73.  German

1289

(cf. Grewendorf 1988, v on Stechow/Sternefeld 1988):

Under certain conditions, traditionally linked with the “substitute infinitiv e” phenomenon (“Ersatzinfinitiv”) — whereby a bare infinitiv e takes on the role of a past participle in present perfect tense forms (cf. Bech 1955, Edmondson 1980), a finite v erb otherwise located in final position can precede the non-finite v erb forms in its domain (cf. Edmondson 1980, Lenerz 1984):

In the German tradition of descriptiv e grammar, these generalizations on the position of the v erb are captured in terms of the “topological” sentence model (cf. e. g. Reis 1980). The clause is structured around a left “bracket” (linke Satzklammer = LS), which hosts the v erb in initial or second position, and a right “bracket” (= RS), which is the position taken by clause-final v erbs; together, these “brackets” define an “initial field” (Vorfeld), a “middle field” (Mittelfeld), and a “final field” (Nachfeld):

The initial field is the location for topicalized constituents and the final field hosts extraposed constituents, while the middle field contains constituents that hav e not been extracted. Within a more explanatory syntactic theory, the facts concerning v erb placement in German can be represented in a clause structure conforming to the X'-schema as follows

The v erb that is raised to I to pick up the inflectional affixes is forced to mov e into an empty C-position in finite clauses (this being the “verb-second-strategy”). The customary description of German as possessing a “basic” or “underlying” SOVstructure (cf. e. g. Reis 1974, Thiersch 1978) needs to be made more precise. Recent dev elopments in parametric theory (Borer 1984, Chomsky 1989, Ouhalla 1991) indicate that only functional categories are affected by syntactic parametrization. It should follow from this either that non-functional lexical heads and their complements show no fixed serialization at D-structure (notwithstanding apparent ev idence to the contrary stemming from language acquisition research — cf. Ouhalla 1991), or that the serialization of lexical heads and complements is determined by semantic factors or factors relating to informational structure, for which ev idence can likewise be adduced from research into language acquisition (Wanner/Gleitman 1982). On the basis of (15), it would then be possible to say that the syntactic SOV-serialization in German represents a derived structure resulting from the process of v erb raising to the head I, giv en the right-headed orientation of I' but not V'. The syntactic SVO-structure of a language such as English would then arise in analogous fashion from the left-headed nature of I' (cf. Ouhalla 1990 a).

2.

Infinitives

As mentioned abov e, German infinitiv es are v erb-final. Sev eral types are recognized in the generativ e literature on German: subject- und object-control infinitiv es (1), infinitiv es with accusativ e subject (accusativum-cum-infinitivo = AcI) (2), and raising infinitiv es (3):

1290

XXI. Syntaktische Skizzen

as in (7), or represented by an “implicit argument”(8):

It has sometimes been suggested that the passiv izability of control v erbs and control infinitiv es does not depend solely on structural factors, but that thematic considerations appear to play a role, as may be seen in the following paradigm: Control and raising infinitiv es contrast with AcI’s in that the prepositional affix zu which combines with the infinitiv e form of the v erb in the former, does not appear in the latter. Control infinitiv es furthermore contrast with AcI’s and raising infinitiv es in being able to undergo extraposition:

When a control infinitiv e functions as a complement clause, its subject (PRO) underlies a requirement that it be controlled by a c-commanding antecedent in the matrix clause, the choice of antecedent depending upon lexical properties of the matrix v erb. The controller may be a “split antecedent”,

Howev er, the fact that the English counterpart to (11 b) is grammatical, although ungrammatical in German, indicates that there are language-specific factors at work here, which means that the passiv izability of control structures is not gov erned by thematic considerations alone (cf. v on Stechow/ Sternefeld 1988).

73.  German

In subject clauses and adjunct infinitiv es, the determination of the reference of PRO is free, or at least, not lexically determined (Sternefeld 1985):

1291

bedded v erb is intransitiv e; but the result of passiv izing an AcI-v erb whose complement gov erns an object is generally unacceptable:

Although the reference of PRO in subject clauses is subject to some restrictions, as indicated by the contrast in (13) (from v on Stechow/Sternefeld 1988), it is doubtful whether these restrictions pertain to the structural link between PRO and its antecedent:

This contrast might be attributed to a restriction to the effect that a NP needs to be coindexed with PRO to function as an antecedent, under the assumption that the index of the specifier of NP may only percolate to the NP when the N-head is not a referential element. AcIs occur in German in constructions with perception v erbs (hören (hear), sehen (see), fühlen (feel)) and the causativ e-permissiv e v erb lassen. In the latter case, a morphologically unmarked v ariant of the passiv e is possible (the “lassen-passiv e”, the German equiv alent of the French faire-par construction), while in the former, passiv es surface in participial form:

Höhle (1978, 170) suggests an alternativ e generalization whereby the passiv izability of lassen-constructions correlates with the possibility for the past participle form of the causativ e v erb to be used in the (activ e) perfect tense construction with haben (hav e), instead of the infinitiv e form, which is usually used here:

Raising infinitiv es with zu occur with a small class of v erbs including scheinen (seem), pfle g en (tend to), drohen (threaten): It appears as if the AcI-v erbs themselv es may be passiv ized fairly freely when the em-

1292

Raising infinitiv es are also possible without zu when gov erned by a member of the class of modal v erbs (Wilder 1989):

In (20c), the empty expletive subject has been raised to the subject position. Bech (1955) proposed a typology of German infinitiv es that distinguishes “coherent” and “incoherent” constructions. Incoherent infinitiv es behav e as normal full clauses (CPs): they form barriers with respect to clause-bound processes, can be extraposed and topicalized and may undergo the process known as “pied-piping” (23):

XXI. Syntaktische Skizzen

Secondly, they themselv es are unable to be extraposed (25) or scrambled (26):

Thirdly, scope-bearing elements inside the infinitiv e are able to take scope ov er the matrix clause (27):

Fourthly, the v erb complex containing infinitiv e and matrix v erbs behav es like a constituent, being able for instance to undergo topicalization (28):

Coherent infinitiv e constructions may be formed with certain subject-control v erbs (a necessary condition being that the v erb hav e no additional complements other than the infinitiv e itself, but see Haider (1986) on possible exceptions to this), as well as with AcIand raising v erbs. The notion “coherence” is employed as a cov er term for a collection of syntactic properties. Firstly, coherent infinitiv es are transparent for scrambling of phrases from the embedded clause into the matrix clause:

Finally, when the matrix v erb is passiv ized, the object of a coherent infinitiv e may in certain cases undergo passiv ization to the matrix subject position (29):

Among the v arious proposals that hav e been made in the generativ e literature to account for these characteristic properties of coherent infinitiv es, it is possible to distinguish two broad strategies. The first assumes a bisentential D-structure, on the basis of the apparent thematic properties of the construction, and seeks to make the properties described abov e follow from a subsequent restructuring (or “structure-pruning”) process,

73.  German

cf. Ev ers’ (1975) v erb-raising theory or Rizzi’s (1982) restructuring proposal for Italian. The alternativ e strategy took the surface facts as a starting point, and used the v arious types of infinitiv es to justify different underlying structures. Examples are Fanselow’s (1989) analysis of transparent control infinitiv es as IPs, Haider’s (1986) proposal that extraposable infinitiv es are CPs, AcIs are IPs, and coherent control structures formed on a monosentential base structure. In Grewendorf (1987), it is shown that coherent infinitiv es do not form a homogenous class with respect to scrambling and passiv ization, these facts being captured in terms of a restructuring analysis proposed for coherent control infinitiv es and a v erb raising analysis with no restructuring, i. e. without clause union, for AcIs. The strategies underlying these approaches in one way or another contrav ene univ ersal principles. Restructuring v iolates the Theta Criterion; the postulation of differing underlying structures for identical thematic grids conflicts with Baker’s (1988) Uniformity of Theta Assignment Hypothesis (= UTAH). Reasons such as these led Sternefeld (1990) to seek a unified analysis of coherent infinitiv es as CPs by applying to German Baker’s (1988) proposal for the analysis of certain causativ e constructions in terms of VP-mov ement to Spec of CP. Sternefeld suggested that satisfaction of selectional properties in coherent control infinitiv es is only achiev able by means of obligatory VP-mov ement into the specifier of the embedded (infinitiv e) CP, on the grounds that the C°-position (and therefore also the INFL-position) in coherent infinitiv es is not selected, a consequence of some lexical property of the releant matrix v erbs:

This hypothesis made it possible to deriv e the transparency of these infinitiv es within the framework of Baker’s (1988) theory of barriers (for discussion, see Grewendorf/Sabel 1994). The C°-position of incoherent infinitiv es is according to Sternefeld a selected position. C° thus counts as “filled”, so that the “Doubly-Filled-COMP-Filter” prohibits VP-mov ement to Spec of CP and the concomitant transparency effects do not appear. Clearly, the embedded subject position in AcI and raising infinitiv es must be accessible to

1293

gov ernment by the matrix v erb, which can be achiev ed by stipulating C°-deletion, or C-toV incorporation, in the sense of Baker (1988) (though this must be ruled out in control infinitiv es to prev ent the PRO subject from falling under the gov ernment domain of the matrix verb).

3.

The Clausal Periphery

3.1. Left Dislocation In left dislocation constructions, some constituent — generally an NP or PP — is generated in a position left of the CP node, either adjoined to CP, or located in some specially stipulated position. The left-dislocated expression is resumed with progredient intonation by a coreferential pro-form bearing identical Case — usually a demonstrativ e pronoun — located in the specifier of CP (cf. Altmann 1981):

A theoretical analysis of this pattern might run along the following lines: the resumptiv e category is mov ed into the CP-specifier position from some position inside CP, and is coindexed there with the base-generated leftdislocated phrase, so that a chain results which includes the dislocated phrase, the operator in CP-Spec and the trace of the mov ed element (cf. Cardinaletti 1986). Sev eral types of phrasal categories turn out not to permit left dislocation (cf. Haider 1990), namely, lexical anaphors, as in (4); quantifiers (5) (although here an exception must be made for “referential” quantifier

1294

XXI. Syntaktische Skizzen

phrases as in (5')); indefinite pronouns (6); sentential adv erbs (7) and negation (8); idiom expressions (9); and subconstituents of NPs (10):

Cyclic wh-mov ement out of infinitiv e complements is possible, irrespectiv e of whether the infinitiv e stands in its base position or in extraposed position:

Unlike languages such as Italian or English, German seems to permit “parasitic gaps” only as a highly marginal phenomenon (cf. Felix 1985):

Left dislocation structures can be distinguished by intonational characteristics (i. e. no progredient intonation) as well as syntactic factors (e. g. absence of Case agreement) from other types of construction where a constituent is adjoined to the left of the clause — for example “Hanging Topic” and Nominativus Pendens (cf. Altmann 1981). 3.2. Wh-Movement and Topicalization Apart from echo questions, wh-questions in German, as in English, underly the requirement that a wh-phrase must occupy the CPspecifier position at the lev el of S-Structure. Mov ement of the phrase in question out of its base position into the relev ant CP-Spec position proceeds either directly, or v ia intermediate landing sites, as determined by independent grammatical restrictions (Chomsky 1986). Unlike languages such as Polish, German allows only one wh-phrase to be mov ed in multiple questions:

By contrast with English, German allows no interrogativ e infinitiv e CPs containing the marker zu; wh-infinitiv es are only possible if this marker is absent:

With respect to wh-extractions out of finite complements in German, a number of generalizations can be made. Cyclic wh-movement out of v erb-second-complements is generally possible with no subject-object asymmetries (ECP-effects) discernible:

73.  German

Possibilities for wh-extraction out of complement clauses introduced by a complementizer v ary across dialects: north German speakers generally judge examples inv olv ing such extractions to be unacceptable, and are ev en able to detect subject-object asymmetries — such as in the contrast (21) v s. (22):

South German speakers on the other hand tend to regard such examples as generally acceptable, and to report no differences between subject extraction and object extraction. Such examples are howev er only acceptable where the domain of extraction is a complement to one of a specified class of v erbs (so-called “bridge v erbs”) — this being the same class whose members license v erb-second complements in standard German (cf. Haider 1984). In general, possibilities for wh-extraction are more restricted in German than, for instance, in English or Italian. Extraction out of prepositional phrases in German is only possible where an element — usually a member of the class of r-pronouns (Riemsdijk 1978) — occurs as complement to a postposition:

This restriction holds howev er only at Sstructure. The distribution of wh-phrases in situ indicates that PPs are nev er islands for

1295

LF-movement (cf. Fanselow 1989a). Extraction out of NPs, discussed in Fanselow (1987), Grewendorf (1989) and Webelhuth (1989), seems to be restricted along the following lines. With the exception of ergativ e and passiv e subjects, extraction out of subject NPs is not possible (26), while the possibility of extraction out of object NPs appears to be lexically gov erned, inasmuch as grammaticality judgements v ary with choice of v erb — cf. (27—28):

These observ ations might be accounted for along the following lines. The ungrammaticality of extraction from a subject NP is interpreted as the result of a v iolation of the ECP, since nouns cannot function as proper go v ernors (cf. Fanselow (1987), Müller (1989) on German) and the NP node constitutes a minimality barrier. The possibility of extraction from object NPs has been attributed to the existence of restructuring rules: a reanalysis rule that breaks up complex object NPs and makes NP and PP structurally equiv alent complements of the v erb in Fanselow (1987); or a lexically gov erned N-V-reanalysis formulated in the terms of Baker’s (1988) theory of abstract incorporation in Müller (1989). Restrictions on the av ailability of the relev ant extractions is then ascribed to the lexically gov erned nature of the reanalysis process. The presence of a prenominal genitiv e generally renders extraction from NPs impossible; this holds also for extraction of the (prenominal or postnominal) genitiv e itself:

1296

XXI. Syntaktische Skizzen

contrasts discernibly with the completely ungrammatical result of extracting the subject (36b). This contrast does not appear with (declarativ e) v erb-second complements whose CP-Spec position is filled (Müller 1989) — both examples in (37) are equally ungrammatical:

The data on extraction from NP hav e already demonstrated that “island effects” in German are far stricter than in English or Italian. This is also confirmed directly by examination of wh-island effects. Unlike in English or Italian (cf. Cinque 1990), so-called “weak” islands, such as wh-complement islands, or factiv e islands, form barriers ev en for the extraction of referential wh-elements. Extraposed CPs, howev er, do not form islands at all, being transparent for extraction generally: In v iew of the possibility that long mov ement of NPs might in fact represent an instance of a non-mov ement strategy, whereby a base-generated empty resumptiv e pronoun is bound by a base-generated wh-operator (“A'-bound pro” — cf. Cinque (1990)), the contrast between (36) and (37) might turn out not to be due to the ECP, which constrains only nonpronominal empty categories, after all. Instead, such an account would predict that the long extraction of a complement prepositional phrase (for which an A'-boundpro option is not av ailable) does not yield a grammaticality contrast corresponding to that between (36a) and (37a). This prediction appears indeed to be borne out:

Recent studies hav e established the existence among wh-island effects of a subtle subject-object asymmetry ov er and abov e those attributable to the Subjacency condition: i. e. this one is directly attributable to the ECP (Fanselow 1987). The extraction of an object from a (v erb-final) wh-complement (36a)

73.  German

Topicalization is usually analyzed, following Chomsky (1977), as an instance of whmov ement. With respect to German, this v iew is not uncontrov ersial, though. Haider (1990) cites a wide range of ev idence to argue against a deriv ational analysis of German topicalization and in fav our of a base-generation approach. One case he notes concerns constituents topicalized together with extraposed categories in a manner which appears to v ha e no deri v ational source:

Other cases — e. g. (41b) v s. (41c) — appear to demand conflicting structural analyses for the unmov ed v ariant (41a), and there are ev en cases where what has been topicalized appears to be a non-constituent (42):

The arguments for a non-deri v ational treatment of topicalization structures based on (40) and (41) lose much of their force in v iew of the possibilities for prov iding deriv ational analyses using concepts from Baker’s (1988) theory of incorporation in conjunction with an extended clause structure in the sense proposed in Chomsky (1989). In the first case, v erbs in base position underly an obligatory adjacency requirement. In the second, the existence of a scrambling rule permits the deriv ation of the examples in (41) without the

1297

need to assign inconsistent structures. The phenomenon of NP-splitting (or “split topicalization”) illustrated in (42) represents a problem independent of the question of topicalization (since it appears that NP-subconstituents may also be scrambled — cf. example (49) below), and one which still awaits adequate resolution; as such, it can only be used in an argument about topicalization under the assumption of a specific and independent analysis. Assuming now that topicalization in German is indeed an instance of wh-mov ement, the question arises as to whether the landing site of the topicalized category is, as in the case of wh-phrases, the specifier position of CP. Cinque (1983) and Cardinaletti (1986) hav e attempted to show that topicalization in German inv olv es both mov ement and basegeneration: base-generation of the topicalized element in a special Topic position independent of the CP-specifier position, and whmov ement of an empty operator into the CPspecifier position. Such an analysis would mimic the structure assumed for left dislocation in German, and as such, faces the objection raised by Haider (1990), that all the elements shown in 3.1. to resist left dislocation (see examples (4)—(10)) can be topicalized:

The conclusion to be drawn is that topicalization in German is to be analyzed in the same manner as wh-movement(although there still remain empirical differences between wh-mov ement and topicalization, as discussed in Müller (1989)). The special form of topicalization sometimes known as “split topicalization” gav e

1298

XXI. Syntaktische Skizzen

rise to the v iew that German apparently allows non-maximal N-projections to occupy the CP-specifier position, as in (42) or (48). Howev er, this phenomenon (referred to here as “NP-splitting”, following Fanselow (1988)) is not exclusiv e to topicalization, as witnessed by the following instance of scrambling:

This phenomenon, which is also found elsewhere, for instance, in certain Australian languages (cf. Dixon 1972), has been used to demonstrate the existence of “nonconfigurational” languages which are characterized by freely ordered word sequences, rather than constituent orders generated in conformity with the X'-schema and other principles of UG (cf. Hale 1983). Although this v iew is generally no longer adhered to, the question of how NP-splitting is to be analyzed still awaits resolution. Arguments can be found that speak both for and against a mov ement analysis (cf. v an Riemsdijk 1989). The fact that NP-splitting is sensitiv e to islands speaks in fav our of a mov ement analysis (50), (51), as does the fact that the CP-specifier position in a v erb-second complement straddled by NP-splitting is obligatorily left empty (an effect usually attributed to the presence of an intermediate trace — cf. (52), (53)):

The main argument against mov ement is a conceptual one, namely that non-maximal projections may neither occupy the CP-specifier position, nor be affected by mov e-alpha. Additionally, a mov ement analysis has difficulties coping with the fact that some instances of split NPs appear to hav e no (wellformed) deri v ational source (54)—(56):

Although a mov ement analysis as “DPsplit” with an intermediate landing site in DP-specifier position, such as is proposed in den Besten/Webelhuth (1990), is able to av oid the conceptual problem, it fails to deliv er an explanation for the morphosyntactic phenomena illustrated. Van Riemsdijk (1989) assumes a mov ement analysis (mov ement of N'), and proposes to solv e the conceptual problem by recourse to a process of “syntactic regeneration”, whereby intermediate phrase-structural projections are able under certain circumstances to be conv erted into maximal projections at S-structure. The lack of independent motiv ation for v an Riemsdijk’s “regeneration” process led Fanselow (1988) to reject this solution, and to attempt instead to demonstrate that NP-splitting is in reality the result of mov e-alpha separating a base-generated cluster of two complete NPs. Following Olsen’s (1987) analysis of nominalized adjectiv es, he proposes that the NP lacking an ov ert N-head is in possession of an empty pronominal head (pro) that is capable of being identified v ia the inflectional morphology of adjectiv es or determiners. Ac-

73.  German

cording to Fanselow (1988), this analysis yields an explanation for why NP-splitting is only found in languages like German which display rich agreement morphology in the NP-domain:

This view is confirmed by the fact that adjectiv es which hav e uninflected v ariants may only occur in a headless NP in their inflected forms:

Although this analysis is able to explain the morphosyntactic restrictions illustrated in (54)—(56), the question remains as to how the assumption of two NPs in NP-splitting constructions is to be reconciled with the Theta Criterion, i. e. how a single theta-role of the v erb may be distributed across a cluster of two NPs. Fanselow suggests that in a fashion similar to that observ ed in clitic-doubling constructions in Spanish, the NP containing an empty pronominal head is A'bound by the other NP, which is assumed to be base-generated in a position adjoined to V0 (hence in a non-theta position) in a structure like (59):

This analysis giv es rise to a series of objections. The first concerns the fact that all other known instances of the empty category pro (subject-pro, object-pro, A'-bound pro) display the distribution of maximal projections rather than N', as here (this objection might howev er be neutralized within a DPanalysis of nominal phrases). Secondly, the adjunction of a maximal projection to a category of the X0-lev el conflicts with univ ersal restrictions on adjunction (cf. Chomsky 1986), and the extraction of a category out of a position adjoined to an X0-category is not compatible with the condition (cf. Baker 1988) forbidding the occurrence of a trace inside a (complex) X0-category if it is not the head of the complex, as in the case of the trace left by v erb-fronting of a particle v erb in German.

1299

These objections indicate that a more adequate analysis should be sought. One clue to how such an alternativ e might look is to be found in Fanselow’s (1988) observ ation that the NP-splitting construction to a certain degree parallels ordinary predication structures, in that the NP with ov ert nominal head in the splitting construction may only be interpreted as a property expression. This suggests that maybe a small clause configuration of the form (60) is v in v ol ed:

Since the small clause itself is of the category NP, the analysis presents no problems with respect to selection restrictions. Restrictions on possibilities for extraction would then need to be formulated along the same lines as for a small clause analysis of particlev erb constructions (cf. Kayne 1984, Grewendorf 1990). 3.3. Relative Clauses and S'-Pied-Piping The syntax of relativ e clauses is similar to that of embedded wh-questions: the v erb occurs in final position and a phrase containing a relativ e pronoun agreeing in number and gender with the head noun undergoes obligatory mov ement to the specifier position of CP. This mov ement is subject to much the same restrictions as wh-mov ement in questions. A special feature of relativ e clause formation which German shares with Italian, but which is not found in English, French, or Dutch, is the possibility of fronting the clause containing the relativ e pronoun, in the “inv finiti e pied-piping” construction:

Riemsdijk (1985) sought to justify an analysis of this construction in which the relativ e pronoun is mov ed to the specifier position of the infinitiv e and the whole infinitiv e subsequently raised to the specifier position of the finite matrix CP:

XXI. Syntaktische Skizzen

1300

trix IP prior to successiv e-cyclic mov ement of the relativ e pronoun to the specifier position of the matrix CP (cf. Grewendorf 1986). As pointed out by Haider (1985), this analysis conflicts with the generalization that infinitiv al relativ e clauses do not otherwise exist in German. Haider proposes an alternativ e analysis that meets this objection, and which enables the possibility for pied-piping of infinitiv al CPs in German as opposed to English and Dutch to be deriv ed from independent properties of the grammar of German not shared by the other languages. The relev ant property according to Haider is the fact that in German, subjects are not structurally restricted to precede objects: in particular, not only nominal, but also sentential objects may precede the subject, as in (63):

Infinitiv al pied-piping is in Haider’s conception nothing more than mov ement of a relativ e pronoun out of an infinitiv al object clause that is base-generated in pre-subject position, and into the specifier position of the matrix CP. Howev er, not only is it based upon the untenable assumption of a nonconfigurational clause structure for German, this analysis is further affected by certain empirical problems. Specifically, it is unable to explain why in a configuration such as (64), pied-piping of CP3 is possible ev en when CP2 is in extraposed position, as in (65) (cf. Grewendorf 1986):

An alternativ e which av oids the problems affecting both Riemsdijk’s and Haider’s analyses would inv olv e mov ement and adjunction of the infinitiv e to the left of the ma-

4.

Clause-Internal Structures

4.1. Configurationality German displays v irtually ve ery surface property which Hale (1982) identifies as being characteristic of “nonconfigurational” languages. As well as possessing a relativ ely free word order and rich Case system, German permits discontinuous syntactic constituents, forms complex v erb clusters, and appears not to require obligatory NP-mov ement (Grewendorf 1989). It comes thus as no surprise that German has been classified as a nonconfigurational language — e. g. by Sternefeld (1982). Howev er, despite the apparent nonconfigurational nature of languages like Warlpiri, Latin, or German, the question of whether nonconfigurationality is to be recognized as an option permitted by UG is v ery much open to debate. It has furthermore been demonstrativ ely argued by sev eral authors (cf. e. g. Webelhuth 1984/85, Fanselow 1987) that when examined under a more abstract perspectiv e, German in particular fails to fulfil the criteria for nonconfigurationality. The configurationality debate with respect to German centred largely on the question of whether German contains a VP-node, in particular, whether predictions concerning subject-object asymmetries that arise under the assumption of a maximal projection separating subject and object are actually fulfilled. Those arguing for the existence of such a projection include Thiersch (1982), Webelhuth (1984/85), den Besten/Webelhuth (1987), Fanselow (1987) and Grewendorf (1988); arguments against are to be found in Sternefeld (1982), Tappe (1982) and Haider (1983; 1983a; 1986a). An intermediate position, according to which infinite clauses of German contain a VP while finite clauses do not, is argued for by Kratzer (1984). In considering this issue, it first needs to be noted that constituent tests of the type traditionally used to justify VP in English fail to deliv er decisiv e results for German. Pronominalization data such as (1) do not prov ide relev ant ev idence since modal v erbs in German in all likelihood embed clausal (subject control or raising) complements, as noted already by Bech (1955).

73.  German

1301

Notice that data such as (7) (cf. Höhle 1983) could equally be cited as ev idence against a VP:

Such ev idence might be forthcoming from examples containing auxiliary v erbs, which gov ern VPs (or in terms of an extended clause structure, TPs — cf. Grewendorf (1990)). Howev er, in this case, pronominalization turns out to yield v irtually unacceptable results:

Such data do not demonstrate the absence of a VP-constituent, howev er, as the impossibility of the pronominalization in (2) and (3) may be attributed to the special incorporation requirements existing between auxiliary and participle. The contrast (4) v s. (5) also fails to demonstrate the existence of VP-pronominalization in (5), as it may well be that the pronoun in this example substitutes for a subject control CP instead.

Coordination data as in (6) cannot be considered as ev idence for VP, as long as the role played by other interv ening grammatical processes such as deletion and “Right node raising” remains unclear.

In a similar v ein, before topicalization data can be brought to bear on the issue, the effects of other properties of German syntax, for example, incorporation, ergativ ity or scrambling, need to be clarified. It is possible, for instance, for a v erb to be topicalized together with its subject — if the v erb belongs to the v ergati e class:

The v arying possibilities concerning the incorporation of v erbal elements distort the situation as regards the determination of constituent structure (9), just as does the possibility of scrambling in the “middle field” (10):

Once all these interfering factors hav e been taken into consideration, it becomes possible to construct contrasting topicalization pairs such as (11/12) which appear to support the hypothesis of a VP-constituent:

1302

Clear that-trace effects such as those found in English are not to be found in German: as mentioned abov e, either both object and subject extraction are judged grammatical, as in southern German dialects, or both types of extraction are held to be unacceptable (with somewhat worse results for subject extraction), as in standard German. ECP effects attributable to subject-object asymmetries hav e on the other hand been diagnosed ev en in German, in the case of extraction out of whislands (Fanselow 1987, Grewendorf 1989): assuming that the contrast between (13b) and (14b) is due to the presence, ov er and abov e the subjacency v iolation found in both examples, of a v iolation of the ECP in the latter:

A problem with this diagnosis is that it is not clear whether island v iolations inv olv ing NPs inv olv e extraction (mov ement) in the first place, since such cases might be analyzed as instances of the strategy of deploying an A'-bound pro (Cinque 1990). In that case, the distinction between (13b) and (14b) would no longer be explainable v ia the ECP. These examples could then only be brought to bear on the VP-debate by appealing to the existence of some subject-object asymmetry within the confines of the A'-bound prostrategy. While opinions differ concerning the possi-

XXI. Syntaktische Skizzen

bilities for extraction out of infinite subject clauses (cf. Haider 1983, Sternefeld 1985, Fanselow 1987, Grewendorf 1989), extraction from finite subject clauses introduced by a complementizer yield a clear subject-object asymmetry, illustrated in (15) v s. (16), that is detectable ev en by those speakers who generally reject examples of extraction out of such finite complements:

Ev en these data can only be considered relev ant for the VP-debate when other issues hav e been clarified: for instance, the question of whether subject clauses are dominated by a NP node, whether the same is true of object clauses, what the exact circumstances are that allow the barrierhood of such an NP to be neutralized; and so forth. A further set of data displaying the classic ECP asymmetry concerns the “superiority” effect. The German data present a quite different picture to that familiar from English. Disregarding so-called “D-linked”wh-phrases (i. e. those that refer to entities salient in the discourse), which appear not to trigger superiority effects in English (Pesetsky 1987), it appears that in German, unlike in English or Dutch (Haider 1986 a), ev en “non-referen-

73.  German

1303

tial” wh-phrases fail to trigger superiority effects:

Howev er, in v iew of the fact that the relev ance for the ECP of superiority data has been called into question ev en for English (Huang 1982) and Dutch (Koster 1987), this phenomenon does not deserv e the central role in the German VP-debate accorded to it for instance by Haider (1986 a). Considerations stemming from the binding theory rev eal the existence in German of further subject-object asymmetries typical of a configurational structure. Adopting a “flat” structure would be inconsistent with a theory incorporating the PRO theorem, ev en in its weaker from (Chomsky 1986 a, Kayne 1990); and would create mutual binding in examples such as (18)—(19), causing them to be analysed as Principle C v iolations:

Most Binding theoretic discussions of subject-object asymmetries contain reference to “weak crossov er” phenomena. In English, these are triggered by quantifier expressions, focused phrases and phrases that hav e undergone wh-mov ement (also, in weakened form, by mov ement of relativ e pronouns). In German, weak crossov er effects are triggered by quantifiers and focused elements, but not, it appears, by wh-phrases (cf. Grewendorf 1988):

Ev en disregarding such fundamental problems as these, there are a number of (subtler) coreference restrictions which might permit an account in terms of a structure containing a VP. Supposing, with Reinhart (1983), that extraposed subject clauses are situated under VP (and abstracting away from the problems posed for this idea by the assumption of Vto-I mov ement in German) while clauses that hav e been extraposed out of a NP are adjoined to IP, then the different coreference possibilities associated with the two types of extraposition structure, illustrated in examples (20)—(21), might be accounted for on the basis of a subject-object asymmetry induced by the presence of a VP node:

If it is assumed that a scope-bearing element is mov ed at LF to a position which determines its scope, then (23a) and (24a) both inv olv e an LF representation in which two v ariables are locally bound by a scope-bearing element, in v iolation of the “Bijection Principle” of Chomsky (1982), Koopman/

1304

XXI. Syntaktische Skizzen

Sportiche (1982). In (23b) and 24b), syntactic a functional PASS-projection, as proposed in adjunction (i. e. scrambling) of the object to Ouhalla (1990) — one consequence which this effect seems to hav e is that of enforcing NPIP precedes LF mov ement of the scope-bearing element. The v ariable created in subject mov ement, as languages like English demonposition by LF mov ement of the operator is strate. This conclusion appears howev er not thus able to locally bind the v ariable (proto be confirmed by the data from German noun) inside the phrase adjoined to IP, so passiv e constructions, as these giv e ev ery appearance of not inv olv ing obligatory NPthat the operator itself stands in a local binding relationv with only one v ariable. Crucially mo ement: for the VP-debate, an explanation for why, following LF raising, the v ariable in object position is unable to (locally) bind the v ariable contained in the subject phrase in (23a) and (24a) is only possible if a VP node is assumed to be present. Of course, this account leav es open the question of why weak crossov er effects are not triggered in German by syntactic whmov ement (or alternativ ely, why the weak crossov er effects found with wh-mov ement are so considerably weaker than in the other cases of (LF) operator movement). Howev er, if the hypothesis of optional NPmov ement in German passiv e sentences is to 4.2. NP-movement be maintained, then alternativ e deriv ations for examples like (28) inv olv ing for instance NP-mov ement — mov ement out of a caseless NP-mov ement of the object and scrambling theta-position into a non-theta position with (adjunction to IP) of the indirect object, need Case — is made obligatory by the fact that to be excluded. It turns out in fact that (28) the relev ant NP would receiv e no Case in its has a deriv ation in which the object does not base position, and made possible by the fact lea v e its D-structure position (cf. Grewenthat the target Case-position is not a thetadorf 1989). position. A typical syntactic construction fulThe question thus arises as to how the obfilling these conditions is the syntactic pasject can be assigned nominativ e inside VP. siv e. According to Chomsky (1981), the One possible answer might lie in the examiessential properties of passiv e structures are nation of the impersonal passiv e in German, the absorption of the Case of the object and particularly in the observ ation that unlike the dethematizing of the subject position. The Dutch for example, German does not require former is v iewed as the effect of neutralizing a lexical expletiv e element to occupy the subthe [+/—N] feature of the (passiv ized) v erb, ject position: so that it is no longer able to function as a Case-assigning category of the type [+V,—N]. The latter property is attributed to the passiv e morphology, although the mechanism inv olv ed is not made precise. An account is offered in recent theories of passiv ization (Jaeggli 1986, Baker/Johnson/Roberts 1989), where the passiv e morphology is treated as an argument morpheme generated under INFL, which, just like any other argument, requires a theta-role and Case to be assigned to it, and thus creates the absorption effect typical of passives. It is reasonable to suppose that German Howev er the Case-absorption effect in possesses an empty expletiv e pronominal elepassiv es is accounted for — as the result of ment which is inserted in non-thematic subfeature neutralization as in Chomsky (1981), ject positions, and which is able in personal v ia a Case-attracting passiv e morpheme as in passiv e constructions to transmit the nomiBaker et al. (1989), as a by-product of incornativ e assigned to it by INFL to an NP coinporation, as in Baker (1988) and v on dexed with it inside VP. This mechanism of Stechow (1990), or as a specific property of

73.  German

“indirect” nominativ e case-marking of VPinternal subjects is presented and justified in detail in Grewendorf (1989). As noted by Höhle (1978), some of the v erbs which embed coherent infinitiv e complements permit the object of the embedded v erb to be promoted to the matrix subject position when the matrix v erb is passiv ized (more specifically, this object is either mov ed into the subject position of the matrix v erb or receiv es nominativ e in situ from the matrix (finite) INFL):

1305

question:

This generalization giv es rise to the prediction that the participle form of an intransitiv e v erb may not be used attributiv ely, as confirmed by cases like the following:

Interestingly, and in accordance with the “Ergati v e Hypothesis”, intransiti v e v erbs which are syntactically ergativ e allow their participles to be used attributi v ely:

The apparent problems which this “long distance” passiv e presents with respect to UG principles — for instance the apparent v iolation of the binding theory it represents — can be ov ercome in an analysis which employs the dev ice of VP-mov ement to Spec of CP (as first proposed by Baker (1988) for the analysis of causativ e constructions) in the deriv ation of coherent infinitive constructions. A phenomenon related to the passiv e (insofar as it inv olv es structures with non-thematic subjects and Case-less objects) is ergativ ity. Like Italian (Burzio 1986), the German lexicon contains ergativ e v erbs (ankommen “arriv e”, einschlafen “fall asleep”, passieren “happen”, auffallen “strike”) which display the properties typical of ergativ es (no external argument, Caseless internal argument, inability to assign accusativ e case). Just as in Italian, the surface subject in German sentences containing ergativ e v erbs behav es syntactically — in accordance with the “Ergativ e Hypothesis” — like a D-structural object. Nominativ e case is assigned to this NP in the same way as to the object of a passiv ized transitiv e v erb: either v ia NP-mov ement, or mediated by a pro-chain. The syntactic indicators for the presence of an ergativ e surface subject with object properties that are specific to German include the possibility for attributiv e use of the past participle. An attributiv ely used participle is usually predicated of the object of the v erb in

Further properties which symptomize deep object status of nominativ e subjects of ergativ e v erbs include the impossibility of forming passiv es on ergativ e v erbs (in contrast with genuine intransitiv e v erbs, which are passiv izable in German); the fact that their subjects display extraction behav iour typical of direct objects; and that these v erbs enter into certain object-oriented word formation processes. The auxiliary selection test, which in Italian appears to prov ide an unambiguous test for ergativ e status (ergativ e and only ergativ e v erbs form their perfect with essere), seems not to function in such a clear fashion in German. Although it seems true that all ergativ e v erbs form the perfect with sein (“be”), it appears that there are sein-selecting v erbs which fail to meet other criteria for ergativ e status:

XXI. Syntaktische Skizzen

1306

What is more, there are a number of seinselecting v erbs that hav e the ability to assign accusati e case:

Depending on the analysis adopted for adjectiv e phrases, adjunction to AP can also be considered as an instance of scrambling (cf. v on Stechow/Sternefeld 1988):

Howev er, the status of auxiliary selection as a test for ergativ ity is only weakened to the extent that the phenomena in question find no independently motiv ated explanation (some possible accounts are discussed in Grewendorf 1989). A short glance at the raising constructions of German shows that there is syntactic ev idence in this case, too, for two optional deriv ations (with NP-mov ement or indirect nominati v e assignment v ia pro subject):

The existence of these twin options would follow from the av ailability of the same two modes of nominativ e assignment as found in passive. 4.3. Scrambling In the “middle field” of the German clause, non-v erbal constituents can be relativ ely freely “scrambled”, i. e. mov ed to positions adjoined to IP or VP:

The actual implementation of the syntactic possibilities for scrambling is generally considered to be dependent on pragmatic factors. Lenerz (1977) demonstrates that scrambling of a direct object to a position preceding an indirect object is blocked when the former is more “rhematic” than the latter. Hoberg (1981) discusses the significant role played by the feature [+/— animate] in the realization of scrambling options, showing that constituents denoting animates tend to precede those denoting inanimates. Lastly, it has been suggested by Webelhuth (1989) that scrambling may only affect non-focused constituents. With respect to its syntactic licensing, scrambling is subject to a series of restrictions that are not found with topicalization. It appears for instance that constituents containing unbound traces of scrambled categories may nev er undergo scrambling, although they may be topicalized:

73.  German

Data such as (48) hav e giv en rise to the impression that a VP may nev er be scrambled (see v on Stechow/Sternefeld (1988), Grewendorf/Sternefeld (1990) for discussion):

A more thoroughgoing inv estigation of the facts shows howev er that the ungrammaticality of (48) is not directly attributable to constraints on VP-scrambling per se, but rather has to do with the fact that scrambling interrupts a necessary incorporation relation between the head of the VP and the gov erning auxiliary (cf. Grewendorf/Sabel 1994). The claim that IPs may not be scrambled is commonly illustrated with examples inv olv ing AcI or raising complements, or by means of examples of the pied-piping construction constructed with “bare” (i. e. zuless) infinitiv es (v on Stechow/Sternefeld 1988, Fanselow 1989):

The assessment of the data does not seem to be without problems, though, in v iew of unobjectionable examples like (51):

1307

Furthermore, it is questionable whether the constituent scrambled in (49) and (51) is really an IP anyway. It is also not clear what reasons could be giv en for assuming that the projection adjoined to IP in (50) is different from the zu-infinitiv e in pied-piping constructions (otherwise, grammatical examples of adjunction of a zu-infinitiv e could equally well be used to argue that IPs can be adjoined to IP). It therefore seems reasonable to seek an account for the ungrammaticality of (50) that does not rely on the purely stipulativ e generalization that VPs and IPs may not be scrambled. The hypothesis suggested in Grewendorf/Sabel (1994) is that the ungrammaticality of (50) is due to the same factor that excludes (48). The standard analysis of AcI and raising constructions v ia the stipulation of S'-deletion (Chomsky 1981) is hardly explanatory. More recent approaches, such as that of Baker (1988), base themselv es on the assumption that these infinitiv es are full CPs just like any other infinitiv e types, and seek to account for the specific properties of ECM and raising in other ways (cf. for example Baker 1988, Grewendorf 1994). Under such an approach, clearly, data like (49) and (50), ev en assuming the judgements giv en, cannot be used to support the claim that IPs (or VPs) cannot be scrambled. Local (clause-internal) scrambling out of NPs is subject to much the same conditions as wh-mov ement out of NPs, being possible only out of NP objects to a certain class of v erbs (cf. Webelhuth 1989, Müller 1989):

Scrambling out of PPs also conforms to the general pattern of restrictions already illustrated with respect to wh-extraction. Giv en the independently motiv ated prohibition on adjunction to NPs and PPs (cf. Chomsky 1986), it is possible to giv e a general account for the ungrammaticality of examples like:

1308

XXI. Syntaktische Skizzen

Although many infinitiv es also disallow “long” scrambling, the prohibition in this case is not general. As already mentioned (see section 2.), long distance scrambling is possible from infinitiv e complements gov erned by members of the class of coherent v erbs (which includes versuchen “try”, but not bedauern “regret”): As has already been shown for wh-movement, howev er, extraction out of PP is possible when the PP is composed of a postposition and a so-called r-pronoun — that is, the r-pronoun itself can be scrambled (an explanation for this is proposed in Koster 1987):

Unlike such languages as Japanese or Russian, there is a general and absolute prohibition on “long” scrambling out of finite clauses in German:

As mentioned abov e, the class of coherent infinitiv es does not prov ide a uniform picture with respect to the scrambling data. The possibilities for scrambling out of AcI complements are more limited than is the case with control infinitiv es. While long scrambling of an AcI-subject is perfectly possible (58), the result of scrambling the object out of an AcI infinitiv e is clearly less acceptable than is the case with control infinitiv es (cf. the contrast between (57b) and (59):

The judgements for (55b) and (55c) are shared by all speakers, but are of particular Similarly, long distance scrambling of an significance with respect to speakers who indirect object is possible out of control infingenerally accept extraction out of comple-v iti es but not AcIs: mentizer-initial finite complements. As (56b) and (56c) show, scrambling out of v erb-second complements is also generally prohibited: in this case, the impossibility of scrambling contrasts with successiv e cyclic whmov ement which is allowed by all speakers:

The question of which mov ement type scrambling is to be analysed as, is a matter

73.  German

of much controv ersy in the literature. In his examination of the shared properties of scrambling and wh-mov ement, Webelhuth (1989) concludes that scrambling is essentially a v ariety of wh-mov ement. Fanselow (1990), on the other hand, seeks to defend an analysis of scrambling as NP-movement. Fanselow’s argumentation consists in emphasising the differences between scrambling and wh-mov ement, and exploiting parallels between scrambling and anaphor-antecedent dependencies. It appears that the constituents which can be extracted out of AcI-complements are exactly those which can host an anaphor bound from outside the AcI:

Fanselow also points out that unlike scrambling, wh-mov ement — at least according to the theory of Chomsky (1986) — does not permit adjunction to IP, an assumption which leads to correct predictions concerning the impossibility of scrambling in situ whphrases and focused phrases:

1309

into a non-thematic Case-marked A-Position) but it also appears that treating scrambling-positions as A-positions hardly opens new possibilities for an alternativ e NP-mov ement analysis, giv en the problems that would arise with data like (65) with respect to the binding theory.

Nor does it appear that the arguments against treating scrambling as wh-mov ement can be readily maintained. It seems for instance that the prohibition against scrambling of in situ wh-phrases does not hold without exception:

Moreov er, the v iew that wh-mov ement does not permit adjunction to IP is faced with numerous empirical and conceptual problems that hav e been highlighted in recent studies (see for instance Cinque (1990)) and no longer counts as the canonical view. It would be rashed, howev er, to leap from these considerations to the conclusion that scrambling is nothing more than wh-movement. As already shown at the opening of this section, the restrictions on scrambling and those gov erning topicalization clearly do not produce the same effects. The conclusion to be drawn is either that the restrictions on wh-mov ement hav e not yet been sufficiently explored or that there are special (as yet illunderstood) restrictions go v erning scrambling whose interaction with the known restrictions on wh-mov ement permit the deriv ation of the special effects displayed by scrambling. 4.4. Negation

Considerations such as these lead Fanselow to conclude that scrambling (in German) is to be analysed as NP-movement. Clearly, as v on Stechow/Sternefeld (1988) hav e noted, scrambling not only fails to fulfill standard criteria for NP-mov ement (mov ement from a Case-less thematic A-Position

According to Jacobs (1982), the negation element nicht (in its non-contrastiv e use) can be classified as an Adsentential, as in (68), as an Adv erbial, as in (69), as an Ad-NP as in (70), as Ad-Adv erb as in (71) and as the Modifier of a Modifier of a Common Noun, i. e. as Ad-AdCN as in (72):

1310

XXI. Syntaktische Skizzen

(73), under a normal reading, the negation must occupy a position following the NP objects.

Within the categorial syntax adopted by Jacobs, it is possible to deriv e all the categorial options for nicht (with the exception of its function as Ad-AdCN) from the basic category Adsentential. Jacobs thus assumes the occurrence of nicht in (68) to represent its basic function (despite distributional differences with respect to other adsentential elements such as ständig (“constantly”), or notwendig erweise (“necessarily”) and generates nicht as an expression of the category S/S. Variation in the placement of nicht is then analysed with the aid of ad hoc “adjacency restrictions” and “scope restrictions”. As argued in Grewendorf (1990), there are no strong syntactic arguments for categorizing the negation as an adsentential element, and in fact a considerable body of syntactic ev idence speaks against this v iew. The arguments Jacobs’ giv es for adopting this v iew of the category of the negation are semantic in nature, and only motiv ate this syntactic categorization within the framework of a categorial syntax. In the modular syntax of the principles and parameters approach of Chomsky (1982), arguments concerning semantic scope can be accommodated in terms of the representation of negation at the lev el of LF. In work within the generativ e tradition spanning the range from Bartsch/Vennemann (1972) to Webelhuth (1989), the German negation element has been analysed as an adv erbial category, which — in its guise as sentential negation — is generated in a position left-adjoined to VP. Under this approach, it comes as unexpected that in sentences like

This observ ation has been ev aluated in different ways in the literature. Retaining the traditional v iew of the position of nicht, Webelhuth (1989) assumes that in sentences like (73), the VP-constituents preceding the negation hav e been scrambled out of the VP. Zanuttini (1988; 1989), dev iating somewhat from the traditional v iew, assumes that the German negation element is generated inside the VP. Both these analyses render it necessary to adopt an additional mechanism to account for the sentential scope of the negation in examples like (73), which generally takes the form of an LF operation of NEG-raising (cf. Zanuttini 1988). Additionally, the assumption of scrambling for sentences like (73) made necessary by Webelhuth’s analysis faces a range of counterarguments discussed in Grewendorf (1990). The assumption that scrambling of the objects in such examples is obligatory is for example inconsistent with Webelhuth’s (1989) general assumption concerning the nature of scrambling as an A'-mov ement process that may only affect non-focused constituents. In negated clauses, it appears that focused constituents can freely precede the negation:

Following the analysis of negation in French and English dev eloped in Pollock (1989) and Chomsky (1989), v on Stechow (1990) proposes an analysis for German in which the negation element is generated in a position preceding the objects, as in Webelhuth’s proposal, but this time as the head of an autonomous maximal projection c-selecting the VP:

This analysis permits a treatment of the German “cohesiv e” negation elements (such as kein (“no, not ... any”) along the lines

73.  German

1311

sketched in Kratzer (1988). Howev er, as well (76) [IP NP [I' [NegP NEG VP] INFL]] as suffering from the same problem with reThis process is optional for singular negativ e spect to scrambling already noted for Webelindefinites, and applies obligatorily to plural huth’s proposal, this solution also raises the negativ e indefinites and negated mass nouns. question of why the NEG-element in GerOn the basis of this theory of cohesion, toman should not block mov ement of the Verb gether with the independent assumption that to Infl. V-raising to I in a structure such as “stage lev el predicates” theta-mark their sub(75) would either imply a straightforward v ijects inside their maximal projections, while olation of the Head Mov ement Constraint, “indiv idual lev el predicates” theta-mark their or necessitate further assumptions to the efsubject outside their maximal projections fect that the NEG in German, as in French, (Kratzer 1988), it becomes possible to deriv e is an affixal element which permits the v erb certain contrasts in the formation of cohesiv e to raise to INFL v ia the NEG-position. This elements with arguments of stage-lev el predilatter alternativ e can be shown to be empiricates such as verfüg bar (“av ailable”) and incally untenable: the v erb would only be able div idual-lev el predicates such as barmherzig to mov e out of the NEG-Position into INFL, (“merciful”) — see v on Stechow (1990) for if it were to take the negation element with it, further discussion: which would result in a v iolation of a central constraint on head mov ement (cf. Baker 1988) under subsequent mov ement of the verb to C in verb-second clauses. There are two ways of escaping this dilemma, the choice between them to be decided by independent empirical arguments. The first would be to adopt a proposal made by Pesetsky (1989) for English, whereby the 5. Subject-Drop, Topic-Drop, negation element occupies not the head posiObject-Drop tion but the specifier position of the NEGPlatzack (1987), Koster (1986) and Grewenphrase. The second would be to return to the dorf (1989) use the term “semi pro-drop lanhypothesis that the negation element has adguage” to characterize German. The use of v erbial character, this time linking it with an this term expresses the claim that German idea formulated by Lehmann (1974; 1978) permits phonetically null expletiv e pronomi(but rejected in the face of apparently ov ernal subjects (for the opposing v iew, see whelming counterev idence by a number of Haider 1988). The subject of impersonal paslinguists, including Dahl 1979, Jacobs 1982 siv e sentences, for example, may not host an and Dryer 1988): namely that the German o ert subject: negation element is v generated in postv erbal position. Current analyses of clause structure and v erb mov ement permit the adoption of this hypothesis in a form in which it remains unaffected by the alleged counterev idence, namely that the NEG-phrase is base-generated in a position right-adjoined to VP or TNSP (cf. Grewendorf 1990). This hypotheThe existence of null expletive subjects is also sis is supported both by historical ev idence supported by the optionality of NP-mov e(cf. Lehmann 1978) and by a range of syntacment in (transitiv e) passiv es, in raising contic arguments (cf. Platzack 1986; Grewenstructions, and in sentences with ergativ e dorf 1990). v erbs: The phenomenon of “negativ e cohesion” in German — instantiated by elements such as nichts (nothing), niemand (no one), and the negativ e indefinite determiner kein, has been analysed in Kratzer (1988) and v on Stechow (1990). On the basis of a structure like (76), it is assumed that the negation element merges with a right-adjacent indefinite article (ein) or pronoun (jemand (“someone”), etwas (“something”)) to form a single cohesiv e element at the level of Phonetic Form:

1312

XXI. Syntaktische Skizzen

It seems that middle constructions are possible ev en when the subject position is not lexicalized, as illustrated in (8b), or in examples like (10) (from Höhle 1978):

Although it is possible for a subject to surface in its D-structure object position, German does not permit v isible free subject inv ersion of the type associated with null expletiv es in Italian, for example. The possibility for a subject to appear between any objects and the v erb, ev en in non-passiv e or non-ergativ e structures is the result of scrambling of the objects and not of free inv ersion of the subject, as can be independently shown (cf. Grewendorf 1989). Unlike full pro-drop languages, German does not permit argument subjects to be null in finite clauses, and unlike other “semi prodrop” languages such as Icelandic, neither does it permit quasi-argument subjects to be null:

The es in the topic (CP-specifier) position in (9) cannot be analysed as an expletiv e subject, since the subject position itself is a position where this element may not appear:

Rather, in cases like (9) this pronoun takes on the function of a placeholder for the topic position, guaranteeing the declarativ e reading (for an alternativ e v iewpoint, see Cardinaletti 1990). Without this placeholder, the corresponding structure takes on the reading of a yes/no question: In middle constructions, the presence of a lexical (ov ert) subject represents the usual case (cf. (7)); but exceptions do not seem to be ruled out (cf. (8)):

Neither the lexicalization of the subject in middle constructions, nor the obligatory presence of the pronoun es in declarativ e clauses of the type in (9) justify the v iew that expletiv e subjects in German must be lexicalized.

It is possible in German to delete a pronoun in the topic position. This process, known as “Pronoun Zap” or “Null Topic” (Ross 1982, Huang 1984, Fries 1988) is howev er only licensed when the pronoun standing in the specifier position of CP is an argument. As the absence of the declarativ e reading in (12) shows, the placeholder es cannot be dropped. Either the nominativ e subject (14), or the accusativ e object (16) of the clause may be deleted, but only when they stand in CPspecifier position. Dativ es may not undergo “topic drop” at all, as seen in (18):

73.  German

1313

The theory of empty categories dictates that this null object must be an instance of pro.

6.

Reflexivization

German possesses a reflexiv e pronoun that is morphologically distinct from the personal pronouns, namely the pronoun sich, which is used in the third person (singular and plural) in the dativ e and accusativ e cases. This reflexiv e element also appears in construction with so-called “inherent reflexiv e” v erbs, where it lacks any independent grammatical function, forming instead a component of the v erb itself: Rizzi (1986) adduces ev idence from Italian data to argue that Chomsky’s (1982) v iew that only languages with object agreement license pro in object position is in need of rev ision. German data are similar to Rizzi’s Italian data and can also be used to show that object-drop is a productiv e syntactic phenomenon justifying the inclusion of an empty In its argumentai function, sich corefers category in the syntactic structure. A null obwith the subject of its clause (in a pattern ject is not only able to act as antecedent in similar to that of other anaphoric expresobject control structures as in (19), it is also sions, such as the reciprocal element einable to act as antecedent to an anaphor (20): ander):

Moreov er, such null objects can be modified by adjunct small clauses, as in (21):

Some cases of null “object” are actually to be analysed as subjects of selected small clauses, — although this possibility appears to be restricted to causativ e structures, and not to be possible for epistemic v erbs:

The generalization gov erning the distribution of sich, in accordance with Principle A of the binding theory, is that an NP occurring in a clause Sk takes the form of a reflexiv e iff it is coreferent with the subject of Sk, where Sk is the S-node immediately dominating the NP in question. Examples can be found howev er of simple main clauses which appear to contain no antecedent for the reflexi v e pronoun:

XXI. Syntaktische Skizzen

1314

function

of

the

coreferent

NP:

To bring these data in line with the generalization just sketched, it is necessary to assume that the coreferent element may remain In non-finite constructions, the regularities unrealized at the surface. This is unproblemto be captured with respect to reflexiv ization atic for (5), since a phonetically empty eleconcern control structures on the one hand, ment in the form of PRO, which also funcand AcI and raising structures on the other. tions as antecedent to anaphoric elements in In control structures, NPs coreferent with a other contexts (i. e. control infinitiv es), is local PRO subject undergo obligatory reflexiav ailablev here. In impersonal passiv e conization: structions like (6) and (7), howev er, the empty expletiv e subject is not able to function as antecedent to the anaphor. This case requires therefore either that an implicit argument, which can act as the antecedent to the anaphor, is assumed, or, following the theory of passiv e dev eloped in Baker et. al. (1989), that it is the passiv e argument itself that functions as the antecedent in such cases. In middle constructions (which can be In AcI constructions, the lexical (accusativ e) identified by generic or non-ev entiv e intersubject constitutes a local subject, thus inducpretation and the obligatory appearance of ing reflexiv ization; and in raising construcan adv erbial modifier), the reflexiv e pronoun tions, it is the empty (NP-trace) subject of the sich appears obligatorily, both in examples infiniti v e that takes on this role: formed on transitiv e v erbs, in which case the original direct object argument surfaces as the subject, and in middles formed on intransitiv e v erbs, where the surface subject is the impersonal es:

Some dialects, such as Bav arian, contain what has been described as a personal v ariant of the middle construction, in which the personal substitute for the first reflexiv e occurs:

Unlike other European languages, German displays obligatory reflexiv ization in ev ery case where a non-subject noun phrase corefers with its local subject NP, independently of the argument status and syntactic

There are a number of exceptions to these generalizations, discussed in Grewendorf (1983). Firstly, non-thematic subjects, as found in the lassen-passiv e, fail to block reflexi v ization to the matrix subject:

The same goes for the unoccupied subject position of AcI infinitiv es with ergativ e v erbs, where the “subject” (i. e. the D-structure ob-

73.  German

ject of the ergativ e v erb) remains in situ inside the VP:

Reflexiv ization of the dativ e object of the raising v erb scheinen to the raised infinitiv e subject is not possible (for an account of this, see Rizzi (1986 a), and Belletti/Rizzi (1988)):

In ordinary AcI-infinitiv es, an NP inside a free prepositionial phrase can reflexiv ize to the matrix subject across the interv ening local v accusati e subject:

The reflexiv e in (22) is thus an instance of what is commonly called a “long distance” anaphor. Apart from this case, instances of “long distance” anaphora are otherwise only to be found in German in complex noun phrases; howev er, the data is far from uniform. Long distance binding does not appear to be a general option in complex NPs:

1315

Even if it is by no means clear what the generalizations to be made here are, at least the examples in which the option of long distance binding of an anaphor actually exists accord with the general restriction that only subjects form possible antecedents of “long distance” anaphors (Giorgi 1984, Wexler/Manzini 1987):

One v ided where to

exception to this restriction is proby examples of the following type, an object apparently acts as antecedent a “long distance” anaphor:

The complex NPs in these examples, howev er, contain no lexical subject, and it is possible that the anaphor is actually not a “long distance” anaphor at all, the complex NP not forming a local domain for the anaphor. German reflexiv es are not otherwise restricted to taking subjects as their antecedents: this is demonstrated by the following examples, where coreference with the object forces v reflexi ization:

XXI. Syntaktische Skizzen

1316

Reflexiv ization is howev er not possible when the coreferent NP is itself contained in a prepositional phrase:

For this case, it seems reasonable to attribute the impossibility of reflexiv ization to the lack of c-command by the antecedent. Howev er, examples can be found where reflexiv ization is possible despite the absence of a c-command relation between antecedent and anaphor:

In cases where the coreferent NP is contained in a free prepositional phrase, reflexiv ization to an object antecedent is likewise impossible, as in the following case (abstracting away from the irrelev ant “small clause” reading):

A final restriction on reflexiv ization to an object concerns the observ ation that a direct object may not reflexiv ize to a preceding indirect object (reflexiv ization of the indirect object is likewise impossible under this constituent order):

An account for this restriction is offered in

Grewendorf (1988) in which the attempt is made to relate the restriction to a hierarchy of grammatical functions. Howev er, since the question of why neither principle C nor principle B of the binding theory is v iolated in (36a) can only be giv en an answer in terms of structural factors, it seems more reasonable to replace this functional restriction on anaphoric relations with a purely structural account of the type offered in Larson (1988), or Ouhalla (1994).

7.

References

Altmann, H. 1981. Formen der ‘Herausstellung’ im Deutschen. Rechts v ersetzung, Links v ersetzung, Freies Thema und v erwandte Konstruktionen, Tübingen. Baker, M., 1988. Incorporation: A Theory of Grammatical Function Changing. Chicago. —. Johnson, K., and L. Roberts. 1989. Passiv e Arguments Raised. Linguistic Inquiry 20. 219—251. Bartsch, R., and Th. Vennemann. 1972. Semantic Structures. Frankfurt. Bech, G. 1955. Studien über das deutsche Verbum infinitum, Bd. 1. Kopenhagen. [2., unv eränderte Auflage, Tübingen, 1983.] Belletti, A., and L. Rizzi. 1988. Psych-Verbs and Th-Theory. Natural Language and Linguistic Theory 6. 291—352. Besten, H. den, and G. Webelhuth. 1987. Remnant Topicalization and the Constituent Structure of VP in the Germanic SOV Languages. Talk giv en at the GLOW conference, Venice. —, and —. 1990. Stranding. Scrambling and Barriers, ed. by G. Grewendorf & W. Sternefeld, 77—93. Borer, H. 1984. Parametric Syntax. Dordrecht. Burzio, L. 1986. Italian Syntax. Dordrecht. Cardinaletti, A. 1986. Topicalization in German: Mov ement to COMP or Base-Generation in TOP. Groninger Arbeiten zur Germanistischen Linguistik (GAGL) 28. 202—231. —. 1990. Pronomi nulli e pleonastici nelle lingue germaniche e romanze: Saggio di sintassi comparata. Tesi di Dottorato, Univ ersità di Padov a e Venezia. Chomsky, N. 1981. Lectures on Gov ernment and Binding. Dordrecht. —. 1982. Some Concepts and Consequences of the Theory of Gov ernment and Binding. Cambridge, MA. —. 1986. Barriers. Cambridge, MA. —. 1986a. Knowledge of Language: Its Nature, Origin and Use. New York. —. 1989. Some Notes on the Economy of Deriv ation and Representation. Functional Heads and Clause Structure, ed. by I. Laka & A. Mahajan,

73.  German

43—74. Massachusetts Institute of Technology Working Papers in Linguistics 10. Cinque, G. 1983. Topic Constructions in Some European Languages and ‘Connectedness’. Connectedness in Sentence, Discourse and Text, ed. by K. Ehlich & H. van Riemsdijk, 7—41. —. 1990. Types of A'-Dependencies. Cambridge, MA. Dahl, Ö. 1979. Typology of Sentence Negation. Linguistics 17. 76—106. Dixon, R. 1972. The Dyirbal Language of North Queensland. Cambridge. Dryer, M. 1988. Univ ersals of Negativ e Position. Studies in Syntactic Typology, ed. by M. Hammond, E. Morav csik & J. Wirth. Amsterdam, Philadelphia. Edmondson, J. 1980. Gradienz und die doppelte Infiniti v -Konstruktion im Deutschen. Papiere zur Linguistik 22. 59—82. Evers, A. 1975. The Transformational Cycle in German and Dutch. Utrecht: Univ ersity of Utrecht dissertation. Fanselow, G. 1987. Konfigurationalität. Untersuchungen zur Univ ersalgrammatik am Beispiel des Deutschen. Tübingen. —. 1988. Aufspaltung v on NPs und das Problem der ‘freien’ Wortstellung. Linguistische Berichte 114. 91—113. —. 1989. Coherent Infinitiv es in German: Restructuring v s. IP-Complementation. Syntactic Phrase Structure Phenomena in Noun Phrase and Sentences, ed. by Ch. Bhatt, E. Löbel & C. Schmidt, 1—16. Amsterdam. —. 1989a. Kasusminimalität und L-Markierung. University of Passau. MS. —. 1990. Scrambling as NP-Mov ement. Scrambling and Barriers, ed. by Grewendorf & Sternefeld, 113—140. Felix, S. 1985. Parasitic Gaps in German. Erklärende Syntax des Deutschen, ed. by W. Abraham, 173—200. Tübingen. Fries, N. 1988. Über das Null-Topik im Deutschen. Sprache und Pragmatik 3. University of Lund. Giorg i, A. 1984. Toward a Theory of Long Distance Anaphors: a GB Approach. The Linguistic Review 3. 307—361. Grewendorf, G. 1983. Reflexiv ierung in deutschen A. c. I.-Konstruktionen — kein transformationsgrammatisches Dilemma mehr. Groninger Arbeiten zur Germanistischen Linguistik (GAGL) 23. 120—196. —. 1986. Relativ sätze im Deutschen: Die Rattenfänger-Konstruktion. Linguistische Berichte 105. 409—434. —. 1987. Kohärenz und Restrukturierung. Zu v erbalen Komplexen im Deutschen. Neue Forschun-

1317

gen zur Wortbildung und Historiographie der Linguistik. Festgabe für H. E. Brekle zum 50. Geburtstag, ed. by B. Asbach-Schnitker & J. Roggenhofer, 123—144. Tübingen. —. 1988. Aspekte der deutschen Syntax. Eine Rektions-Bindungs-Analyse. Tübingen. —. 1989. Ergativity in German. Dordrecht. —. 1990. Verbbewegung und Negation im Deutschen. Groninger Arbeiten zur Germanistischen Linguistik (GAGL) 30. 57—125. —. 1994. Kohärente Infinitiv e und Inkorporation. Zur Satzwertigkeit v on Infiniti v en und Small Clauses, ed. by A. Steube & G. Zybatow, 31—50. Tübingen. —, and J. Sabel. 1994. Long Scrambling and Incorporation. Linguistic Inquiry 25. 263—308. —, and W. Sternefeld. 1990. Scrambling Theories. Scrambling and Barriers, ed. by G. Grewendorf & W. Sternefeld, 3—37. —, and —. (eds.) 1990. Scrambling and Barriers. Amsterdam, Philadelphia. Haider, H. 1983. Connectedness Effects in German. Groninger Arbeiten zur germanistischen Linguistik (GAGL) 23. 82—119. —. 1983a. The Case of German. Groninger Arbeiten zur germanistischen Linguistik (GAGL) 22. 47—100. —. 1984. Topic, Focus and V-Second. Groninger Arbeiten zur Germanistischen Linguistik (GAGL) 22. 47—100. —. 1985. Der Rattenfängerei muß ein Ende gemacht werden. Wiener Linguistische Gazette 35—36. 27—50. —. 1986. Nicht-sententiale Infinitiv e. Groninger Arbeiten zur Germanistischen Linguistik (GAGL) 28. 73—114. —. 1986a. Deutsche Syntax, generativ — Parameter der Deutschen Syntax. University of Wien. MS. —. 1988. Studies on Missing Subjects. Linguistik in Stuttgart 1. University of Stuttgart. —. 1990. Topicalization and Other Puzzles of German Syntax. Scrambling and Barriers, ed. by G. Grewendorf & W. Sternefeld, 93—112. Hale, K. 1982. Preliminary Remarks on Configurationality. IN J. Pustejov sky & P. Sells (eds.), NELS 12, 86—96. —. 1983. Warlpiri and the Grammar of Non-Configurational Languages. Natural Language and Linguistic Theory 1. 5—48. Hoberg , U. 1981. Die Wortstellung in der geschriebenen deutschen Gegenwartssprache. München. Höhle, T. 1978. Lexikalistische Syntax. Die Aktiv Passi v -Relation und andere Infinitkonstruktionen im Deutschen. Tübingen. —. 1983. Subjektlücken in Koordinationen. University of Köln. MS.

1318

Huang , C. 1982. Logical Relations in Chinese and the Theory of Grammar. Massachusetts Institute of Technology dissertation. —. 1984. On the Distribution and Reference of Empty Pronouns. Linguistic Inquiry 15. 531—574. Jacobs, J. 1982. Syntax und Semantik der Negation im Deutschen. München. Jaeg g li, O. 1986. Passiv e. Linguistic Inquiry 17. 587—622. Kayne, R. 1984. Principles of Particle Constructions. Grammatical Representation, ed. by J. Guéron, H.-G. Obenauer & J.-Y. Pollock, 101— 140. Dordrecht. —. 1990. Romance Clitics and PRO. Proceedings of NELS 20. Koopman, H., and D. Sportiche. 1982. Variables and the Bijection Principle. The Linguistic Rev iew 2. 139—160. Koster, J. 1986. The Relation between pro-drop, Scrambling and Verb Mov ements. Groningen Papers in Theoretical and Applied Linguistics, TTT Nr. 1. University of Groningen. —. 1987. Domains and Dynasties: The Radical Autonomy of Syntax. Dordrecht. Kratzer, A. 1984. On Deriv ing Syntactic Differences between German and English. TU Berlin. MS. —. 1988. Stage-Lev el and Indiv idual-Lev el Predicates. Univ ersity of Massachusetts at Amherst. MS. Larson, R. 1988. On the Double Object Construction. Linguistic Inquiry 19. 335—391. Lehmann, W. 1971. On the Rise of SOV Patterns in New High German. Grammatik, Kybernetik, Kommunikation, ed. by K. Schweisthal, 19—24. Bonn. —. 1974. Proto-Indo-European Syntax. Austin, London. —. 1978. Changes in the Negativ e Sentence Pattern in German. Sprache in Gegenwart und Geschichte. Festschrift für H. M. Heinrich zum 65. Geburtstag, ed. by D. Hartmann, 94—109. Köln. Lenerz, J. 1977. Zur Abfolge nominaler Satzglieder im Deutschen. Tübingen. Müller, G. 1989. A'-Bewegung, Minimalitätsbarrieren und Kopfreanalyse. Fachgruppe Sprachwissenschaft, Arbeitspapier Nr. 13. Univ ersity of Konstanz. Olsen, S. 1987. Zum ‘substantiv ierten’ Adjektiv im Deutschen: Deutsch als eine pro-Drop-Sprache. Studium Linguistik 21. 1—35. Ouhalla, J. 1990. Sentential Negation, Relativ ised Minimality and the Aspectual Status of Auxiliaries. The Linguistic Rewiew 7. 183—231. —. 1991. Functional Categories and Parametric Variation. London. —. 1994. Verb Mov ement and Word Order in Ara-

XXI. Syntaktische Skizzen

bic. Verb Mov ement, ed. by D. Lightfoot & N. Hornstein, 41—72. Cambridge. Pesetsky, D. 1987. Wh-in-Situ: Mov ement and Unselectiv e Binding. The Representation of (In)definiteness, ed. by E. Reuland & A. ter Meulen, 98— 129. Cambridge, MA. —. 1989. Language-particular Processes and the Earliness Principle. Massachusetts Institute of Technology. MS. Platzack, C. 1986. COMP, INFL and Germanic Word Order. Topics in Scandinav ian Syntax, ed. by L. Hellan & K. Christensen, 185—234. Dordrecht. —. 1987. The Scandinav ian Languages and the Null Subject Parameter. Natural Language and Linguistic Theory 5. 377—401. Pollock, J.—Y. 1989. Verb Mov ement, Univ ersal Grammar and the Structure of IP. Linguistic Inquiry 20. 365—424. Reinhart, T. 1983. Anaphora and Semantic Interpretation. London. Reis, M. 1974. Syntaktische Hauptsatz-Priv ilegien und das Problem der deutschen Wortstellung. Zeitschrift für Germanistische Linguistik 2. 299— 327. —. 1980. On Justifying Topological Frames. ‘Positional Field’ and the Order of Nonv erbal Constituents in German. Documentation et Recherche en Linguistique Allemande contemporaine Vincennes. Revue de Linguistique 22/23. 59—85. Riemsdijk, H. van. 1978. A Case Study in Syntactic Markedness. Dordrecht. —. 1985. Zum Rattenfängereffekt bei Infinitiv en in Deutschen Relativ sätzen. Erklärende Syntax des Deutschen, ed. by W. Abraham. Tübingen. —. 1988. Mov ement and Regeneration. Dialect Variation in the Theory of Grammar, ed. by P. Beninca, 105—136. Dordrecht. Rizzi, L. 1982. Issues in Italian Syntax. Dordrecht. —. 1986. Null Objects in Italian and the Theory of pro. Linguistic Inquiry 17. 501—557. —. 1986a. On Chain Formation. The Syntax of Pronominal Clitics, Syntax and Semantics Vol. 19, ed. by H. Borer, 65—95. London. Ross, J. 1982. Pronoun Deleting Processes in German. Annual Meeting of the Linguistic Society of America. San Diego. Stechow, A. von. 1990. Kompositionsprinzipien und grammatische Struktur. Univ ersity of Konstanz. MS. — and W. Sternefeld. 1988. Bausteine syntaktischen Wissens. Opladen. Sternefeld, W. 1982. Konfigurationelle und nichtkonfigurationelle Aspekte einer modularen Syntax des Deutschen. SFB 99. University of Konstanz. —. 1985. Deutsch ohne grammatische Funktionen: Ein Beitrag zur Rektions- und Bindungstheorie. Linguistische Berichte 99. 394—439.

74.  English

—. 1990. Scrambling and Minimality. Scrambling and Barriers, ed. by G. Grewendorf & W. Sternefeld, 239—57. Tappe, H. T. 1982. VP and Coherent infinitiv es in German. University of Göttingen. MS. Thiersch, C. 1978. Topics in German Syntax. Massachusetts Institute of Technology dissertation. —. 1982. A Note on Scrambling and the Existence of VP. Wiener Linguistische Gazette 27—28. 83— 95. Wanner, E., and L. Gleitman. 1982. The State of the State of the Art. Language Acquisition: the State of the Art, ed. by E. Wanner & L. Gleitman, 3—48. Cambridge, MA. Webelhuth, G. 1984/85. German is Configurational. The Linguistic Review 4. 203—246.

1319

—. 1989. Syntactic Saturation Phenomena and the Modern Germanic Languages. Amherst: Univ ersity of Massachusetts dissertation. Wexler, K., and M.-R. Manzini. 1987. Parameters and Learnability in Binding Theory. Parameter Setting, ed. by T. Roeper and E. Williams, 41—76. Dordrecht. Wilder, C. 1989. The Syntax of German Infinitiv es. University College London dissertation. Zanuttini, R. 1988. Two Strategies for Negation: Ev idence from Romance. Univ ersity of Pennsylv ania. MS. —. 1989. The Structure of Negativ e Clauses in Romance. University of Pennsylvania. MS.

Günther Grewendorf, Frankfurt a. M. (Deutschland)

74. English 1. 2. 3. 4. 5. 6. 7. 8. 9. 10.

Parts of Speech Constituent Order Grammatical Relations Complements Modifiers Coordination Anaphora and Ellipsis Quantification and Syntax Illocutionary Syntax References

1.

Parts of Speech

Part-of-speech distinctions play an important role in the grammatical sketch that follows. The identification of parts of speech is based on implicational univ ersals, combined with the syntactic and morphological pecularities of the giv en language (McCawley 1992). Morphology prov ides a sufficient but not a necessary condition for membership in a part of speech, e. g. an English word that is inflected for tense is ipso facto a v erb, though a word may hav e such a defectiv e morphological paradigm (e. g. must) as to prov ide no ov ert morphological clue as to its part of speech. (Must, though, can be identified as a v erb on the basis of the generalization that independent declarativ e Ss and certain types of complements require a tense, and in such a sentence as There must have been some trouble, must has to be recognized as bearing present tense if that generalization and general-

izations about where tense can be manifested are not to be violated.) The parts of speech in English whose members can often be identified through morphology are Noun (N), Verb (V), Adjectiv e (A), Adv erb (Adv ). A singular/plural distinction, with plural typically marked by -s, prov ides a sufficient condition for identifying an item as a N; marking for tense (with present tense 3. sg. typically hav ing a suffix -s and past tense a suffix -d) and for present participle (-ing) and past participle (most commonly a form identical with the past tense, though often with a suffix -en) prov ides a sufficient condition for identifying an item as a V; and comparativ e and superlativ e forms (with suffixes -er and -est respectiv ely) prov ide a sufficient condition for identifying an item as either an A or an Adv. The deriv ational suffix -ly prov ides a sufficient condition for identifying the form to which it is added as an A and the composite form as an Adv . There is a clear syntactic basis for recognizing Preposition (P) as a part of speech; other syntactic category notions that it may be reasonable to take as defining parts of speech are (coordinating) Conjunction (Conj), Determiner (Det), and Complementizer (Comp), though it is difficult here to keep the supposed part-of-speech notions separate from the corresponding notions of syntactic function. (By contrast, it is easy to keep the part-of-speech notion “Adv erb” sep-

XXI. Syntaktische Skizzen

1320

arate from the syntactic function notion “adv erbial”, notwithstanding linguists’ frequent failure to keep them straight; most adv erbials are not adv erbs but rather prepositional phrases.) “Pronoun” is not a part of speech but rather a feature distinguishing pronominal members of v arious categories from non-pronominal members. The widespread belief that personal pronouns are pronominal nouns has no factual support: the external syntax of I, you, he, etc. is not that of a noun but that of a NP, and these elements will be treated as (pronominal) NPs below. A number of other words (all of them in some sense “function words”) belong not to parts of speech as normally understood but to “phrasal” categories such as NP, N′, and S, e. g. who is a NP, pronominal one is an N′, and the pronominal so of I think so is a S. It will be worthwhile to state at the outset certain policies about parts of speech that will figure in the discussion below. (i) The part of speech of a word, or of the head of a phrasal unit, is one of the factors that can affect its syntactic behav ior, and category symbols such as A′ (to be interpreted as: “phrasal unit whose head is an A”) will be used, in which symbols for parts of speech figure; howev er, the widespread belief that the head of a “NP” has to be a noun will be rejected here, and the symbol NP should not be taken as conv eying “maximal projection of N” or the like. Multiple “bars” play no role in the category labels used here. (ii) Part-of-speech notions can cross-cut syntactic function notions, e. g. the Japanese complementizer to is a P, while koto and no are Ns. (iii) So-called “subordinating conjunctions” other than complementizers are Ps with S objects (e. g. before is a P both in before the g ame and in before the g ame beg an; cf. Jespersen 1924). (iv ) Units that belong to no part of speech (e. g. tense markers, which are not Vs nor Ns nor ...) will be labeled 0 when a label is needed; phrasal units that they head are thus labeled 0′. (v ) Some parts of speech hav e phonologically zero members, e. g. “adv erbial NPs”, as in I went to Boston last week, will be treated as P′s with a zero P, in v iew of their hav ing the semantics and external syntax of P′s and not of, say, Adv s. (v i) For the understanding of syntactic reanalysis (a widespread phenomenon in the history of English), it is important to recognize words of indeterminate part of speech.

2.

Constituent Order

2.1. Order of Elements in a Clause With unusual rigidity, predicate elements in English precede their objects and complements, with the obv ious qualification that topicalized or Wh-mov ed objects and complements occupy (respectiv ely) a topic or a complementizer position: (1 a) Susan sent the computer to the repair shop. (1 b) The computer, Susan sent to the repair shop. (1 c) The computer which Susan sent to the repair shop needed a new power supply. Subjects precede their predicates, except in sev eral kinds of “inv ersion” constructions. A tense-bearing auxiliary v erb (with do used as tense-bearer when there would otherwise be no auxiliary v erb) precedes the subject (i) in independent Yes-no questions, and in independent Wh-interrogativ e constructions in which a Wh-mov ed interrogativ e element precedes the subject, (ii) when a negativ e expression is put before the subject, a tensebearing auxiliary precedes the subject, irrespectiv e of whether the clause is main or subordinate, and (iii) optionally in Ss introduced by than or as, but only if the remainder of that S is deleted: (2a) Have you received their letter? (2a′) Did you receive their letter? (corresponding declarative: You received their letter.) (2a″) I wonder whether you received their letter. (no inversion in embedded questions) (2b) When did you last hear from Janet? (2b′) Who put the flowers in this vase? (2b″) On which days will which visitors be giving lectures? (2b‴) I wonder when they last heard from Janet. (3a) Under no circumstances will I talk to him. (3b) He said that under no circumstances would he talk to me. (4a) Mary has published more papers than has Ted. ... than Ted has. (4a′) *Mary has published more papers than has Ted published. (4b) Mary has published as many papers as has Ted. ... as Ted has. (4b′) *Mary has published as many papers as has Ted published.

74.  English

It is not because the subject of (2b′) is interrogativ e that there is no inv ersion but rather because it is not preceded by an interrogativ e expression (cf. (2b″)). The “tags” of both rev ersal tag questions and reduplicativ e tag “questions” (which really are not questions but a type of “echo”) hav e inv erted word order, even in the limited class of cases in which a tag has an embedded clause as its host: (5a) You’ve received my letter, haven’t you? (Reversal) (5a′) I don’t suppose you’ve seen my briefcase, have you? (5b) So you’ve won the lottery, have you? Well then let’s go out and celebrate. (Reduplicative) (iv ) In presentativ e clauses, and (v ) in (not necessarily main) clauses that bring a new entity onto the scene in narrativ es, it is possible for a locativ e or directional expression to precede the main v erb and the subject to follow it, prov ided that the subject is not a pronoun: (6a) Here comes the bus. (6a′) *Here comes it. (6b) Out of the bushes rushed a snarling dog. (6c) I was about to go home when through my door came a friend that I hadn’t seen in years. The head of any phrasal unit precedes the object(s) and/or complement(s) of that head. A V generally is immediately followed by any objects that are not marked by Ps, with indirect object preceding direct object, then P′s and then any V′ or S complements, except that Heavy Constituent Shift (often called “Heav y NP Shift”, though the “shifted” constituent need not be a NP) can often mov e to the end of the V′ a constituent that is sufficiently long or complex: (7a) John put the flowers in the vase. (7a′) *John put in the vase the flowers. (7a″) John put in the vase the flowers that Julie had sent him. (7b) Phil was ordered by the colonel to shine his shoes. 2.2. Order of Elements in a Noun Phrase “NP” is a fuzzy syntactic category generated by the logical category “argument of predicate”. The core of this category consists of items that are arguments of predicates, occupy an argument position (unlike e. g. extra-

1321

posed complements, which occupy an adjunct position), and hav e the internal syntax most typical of arguments of predicates: they consist of a pronoun, or of a proper name, or of a (possibly zero) determiner followed by an N′. The periphery of this category includes among other things clausal arguments such as will be discussed in 3. “Determiner” here includes articles, quantifiers, possessiv es, and the demonstrativ es this/these and that/those. Not included under “determiner” are numerals and numeral substitutes such as many and several, which are rather As modifying an N′ — they can cooccur with true determiners in such expressions as his three children and the many problems that we encountered. As left-modifiers of N′s, one encounters A′s and infinitiv al V′s, with the qualification that left-modifiers hav e to be head-final: (8a) a soundly sleeping child (8b) *a sleeping soundly child Modifiers as in an easy to remember name, which end not with the head A but with the head of a complement of that A, can be argued (Nanni 1980) to inv olv e reanalysis of an [A to V] combination as a complex A, on the grounds that they exclude all of the extra material that A′s such as easy to remember normally allow, e. g. *an easy for you to remember name, *a hard to stop humming melody. (This reanalysis still leav es some left-modifiers with non-final heads, as in a larg er than usual honorarium, unaccounted for). As rightmodifiers of N′s one finds restrictiv e relativ e clauses, both in their full forms and reduced to a predicate phrase: a person (who is) eag er to make friends. Some left-modifiers of N′s are reduced relativ e clauses (those which are themselv es modified by S-modifiers, e. g. an until recently unsuspected possibility, are the ones whose status as reduced relativ es is clearest), while others ev idently are not, particularly those interpreted like adv erbial modifiers of a proposition inherent in the noun (John’s recent bride = the woman who John recently married). Modifiers can be stacked to arbitrary depths: the two currently recog nized potential serious obstacles to world peace that have received the most attention. Numerals most often precede other left-modifiers, but for a semantic rather than a syntactic reason: they describe not indiv iduals but rather sets of indiv iduals, and N′s referring to indiv iduals are embedded in whatev er N′s refer to sets of those indiv iduals (cf. para-

1322

phrases such as “the two-member set consisting of obstacles to world peace which ...”). Left-modifiers that describe sets rather than indiv iduals can in fact precede numerals: the worst three days that I’ve ever experienced (≠ the three worst days that I’ve ever experienced). The minimal N′ of a NP consists of a N, possibly accompanied by P′s or sentential complements that represent objects or complements or the subject of the N: (9a) the proposal by the committee to dismiss the manager (9b) his lack of common sense (9c) any suggestion that our actions are unjustified NPs whose determiner is followed by a P′ with of can be assimilated to a [NP Det N′] structure only by treating them as hav ing a phonologically zero N: [NPall [N′ ØN [P′of [NPthe members]]]]. The N′ that such a N would head is the only plausible host for a relativ e clause in (10a), as contrasted with (10a′), where the N′ of the embedded NP must be the host, in v irtue of the number agreement (Jackendoff 1977, 184): (10a) each Ø of the candidates who was elected (modifies Ø) (10a′) each Ø of the candidates who were elected (modifies candidates) The claim that restrictiv e relativ e clauses modify N′s rather than NPs is sometimes disputed on the grounds that an “indefinite pronoun” such as someone or nothing, which is a single word and has the distribution of a NP rather than that of a N or N′, can host a restrictive relative: (11a) someone who has been trying to reach you for days (11b) nothing that any of you is likely to object to Alternativ ely, one could reject the idea that parts of words cannot host syntactic modifiers and take -one in (11a) and -thing in (11 b) as the hosts of the relativ e clauses. The latter proposal will be accepted here, in v irtue of differences between the morphologically opaque interrogativ e pronouns and the morphologically transparent indefinite pronouns in their capacity to serv e as hosts of modifiers. Indefinite pronouns allow the full gamut of full and reduced relativ e clauses (ev en As that only appear as left modifiers when combined with a morphologically “free” noun),

XXI. Syntaktische Skizzen

whereas with interrogativ e pronouns, reduced relativ es are sharply limited in their range and are subject to great indiv idual v ariation in acceptability, aside from a few combinations (e. g. (12c)) that are generally accepted: (12a) We should cast someone tall as Hamlet. (12a′) %Who that is tall can we get to play Hamlet? (12a″) *Who tall can we get to play Hamlet? (12b) You’d better use something even as a value for x. (12b′) an even number (12b″) *a number even (12c) Where in London can you buy Malay spices? Both interrogativ e and indefinite pronouns hav e semantic constituents that support restrictiv e relativ es (i. e. a restrictiv e relativ e is just as compatible semantically with who as with what person), but only the indefinite pronouns hav e surface forms inv olv ing an N′ and thus a normal host for a restrictiv e relativ e; the indiv idual v ariation in the acceptability of examples such as (12a′) reflects v ariation in whether the speaker has dev eloped a “patch” (Morgan 1972) that extends the cov erage of their grammars to a class of cases that, taken strictly, it does not prov ide for, because the morphology wipes out the implicit N′ host of the relative clause. 2.3. Complementizer and Topic Positions The that that introduces many finite complements and the for that introduces many infinitiv al complements occupy a position as left sister of the S that they are combined with. Note that for, notwithstanding its historical origins as a P, does not combine with the subject into a P′; rather, it is outside the complement S, as can be seen from the possibility of combining it with a conjunction of Ss: (13) It would be disastrous for [Mary to quit and John to replace her]. In interrogativ e and relativ e clauses in English, the relativ e or interrogativ e expression (one of the interrogativ e expressions, if there is more than one) is mov ed into a position at the beginning of the clause that is arguably the position of an underlying complementizer. For restrictiv e relativ e clauses, a particularly strong case can be made that the relativ e expression replaces a complementizer: finite relativ e clauses allow an al-

74.  English

ternativ e form with that rather than a true relativ e pronoun, the relativ e expression, like both the that of relativ e clauses and the that of complements, is a left sister of the (full or truncated) S that it is combined with, and that and a true relativ e expression cannot cooccur in the surface form; howev er, the that cannot be part of a complex relativ e expression (such as to whom or for whose sake) the way that who and which can. That is the situation that would result if finite relativ e clauses hav e an underlying that complementizer, and the relativ e expression either replaces the that or, if it is a simple relativ e pronoun (rather than a complex expression inv olv ing also a P and/or a N), is deleted, allowing the that to surface or to be itself deleted. The grounds for holding that interrogativ e expressions and the relativ e expressions of nonrestrictiv e clauses occupy a “Comp position” are not nearly as strong (for example, nonrestrictiv e clauses do not hav e alternativ e forms with that); but at least the hypothesis that these items all occupy a Comp position is consistent with the facts, prov ides some unity to the different kinds of “Wh-mov ement”, and serv es as the basis for an interesting typology of languages. The Wh-expressions of infinitiv al interrogativ e and relativ e clauses then presumably would replace an underlying for complementizer: (14a) They’re trying to decide who to offer the job to. (14b) Valerie is the best person to whom to give this job. One noteworthy difference between infinitiv al interrogativ e and relativ e clauses is that infinitiv al relativ es allow only P′s as relativ e expressions: (15) Valerie is the best person (*who(m)) to give this job to. Independent declarativ e Ss allow a “topic” constituent as a left adjunct, either preceding or following any S-modifiers: (16a) “War and Peace” for some reason he didn’t find interesting. (16a′) For some reason, “War and Peace” he didn’t find interesting. That-clauses can also contain topics, while nonfinite Ss not only exclude them but do not allow anything other than the Comp before their subjects:

1323

(17a) I’m sure that this book you’ll find fascinating. (17b) *For [your latest book Knopf to reject] is outrageous. The “topic” position can be filled either by a “topicalized” constituent (one corresponding to a “gap” in the S) or by a “dislocated” constituent (one corresponding to a pronoun in the S): (18) Your brother, I can’t stand him. There is also a S-final position for “right-dislocated” constituents (called antitopics by Chafe 1976, 53); the use of this position in nonfinite Ss is marginal: (18a) I can’t stand him, your brother. (18b) ?For him to be fired, your brother, would be outrageous. 2.4. Adverbials; Focusing elements; Extraposition; Parentheticals, Vocatives S-modifiers can occur either to the left or to the right of their hosts, with the qualification that an S-modifying Adv to the right of its host must be destressed and pronounced as a separate intonational phrase: (19a) On many occasions, John has acted foolishly. (19a′) John has acted foolishly on many occasions. (19b) Probably there’ll be a hailstorm. (19b′) There’ll be a `hailstorm, probably. (`indicates fall in pitch) (19b″) *There’ll be a hailstorm `probably. An Adv modifying a predicate phrase can generally occur on either side of it, while a P′ can usually only follow it: (20a) Susan willingly took a pay cut. (20a′) Susan took a pay cut willingly. (20b) ??Bill in an angry voice complained to the boss. (20b′) Bill complained to the boss in an angry voice. Adv s modifying a V either immediately precede the V or follow it and any direct object: (21a) Vera severely reprimanded Ted. (21a′) Vera reprimanded Ted severely. (21 b) He completely covered the wall with paint. (21 b′) He covered the wall completely with paint. (21 b″) ??He covered the wall with paint completely.

XXI. Syntaktische Skizzen

1324

(See 5.1. for syntactic and semantic differences among S-modifiers, V′-modifiers, and V-modifiers.) Focusing elements such as only appear either adjacent to their focus or adjacent to a predicate phrase containing the focus. Such elements differ from one another with respect to whether they precede or follow the focus — only and even normally precede the focus, too follows the focus, and also can appear on either side of the focus: (22a) Edith drinks only/even cognac. (intended focus underlined) (22a′) Edith only/even drinks cognac. (22b) Tom sent flowers too/also to his mother. (22b′) Tom sent flowers also to his mother. (22b″) Tom sent flowers to his mother also/ too. (22b‴) Tom also/*too sent flowers to his mother. When the focus is the object of a P, a focus marker usually comes before rather than after the P: (23a) Janet lets Phil talk only to priests. (23a′) *Janet lets Phil talk to only priests. Focusing elements have not only foci but also scopes, and a focusing element that is adjacent to its focus can be ambiguous with regard to scope, e. g. (24a) allows both widescope and narrow-scope interpretations (paraphrasable respectiv ely as “Wine is the only thing that Linda allows Joe to drink” and as “Linda allows wine to be the only thing that Joe drinks”, i. e. she doesn’t make him drink other things), while corresponding sentences in which only is separated from its focus are unambiguous with regard to scope: (24a) Linda lets Joe drink only wine. (24b) Linda only lets Joe drink wine. (wide scope) (24b′) Linda lets Joe only drink wine. (narrow scope) In this respect, the Only-float of (24b—b′) differs from the separation of only from the object of a P in (23a), since the latter example has the same scope ambiguity as (24a). Extraposed constituents of sev eral sorts appear in S-final position: (25a) It was clear to everyone that something was wrong. (25b) Something happened that took us by surprise.

(25c) More people came to the party than I had expected would come. (25d) So many people came to the party that we ran out of beer. Extraposed complements are adjuncts to a predicate phrase, while extraposed relativ es remain within their host NPs in a discontinuous structure. In some cases, extraposed clauses of different types may co-occur, although such possibilities are sev erely limited by the general tendency to av oid Ss in the middles of Ss: (26) Iti caused soj many people problems thati the show was canceled thatj there was a riot. Parenthetical and v ocativ e expressions can occupy v arious niches in their host S (the beginning or end of the host S or of v arious major constituents of it), though they are not allowed to separate an unstressed unit from a unit that it normally must be adjacent to: (27a) To put it bluntly, your behavior is disgraceful. (27a′) Your behavior, to put it bluntly, is disgraceful. (27a″) Your behavior is disgraceful, to put it bluntly. (28a) *It, to put it bluntly, is unlikely that you will be invited back. (28a′) It is, to put it bluntly, unlikely that you will be invited back. (28a″) ??It’s, to put it bluntly, unlikely that you will be invited back. There are some differences among parentheticals with regard to what niches they can occupy; for example, frankly can appear in final position, but confidentially cannot.

3.

Grammatical Relations

3.1. Grammatical Relations and Constituent Structure With relativ ely few exceptions, surface grammatical relations in English match surface syntactic configurations; for example a NP that combines directly with a predicate phrase (not necessarily a V′, in v iew of examples such as With Mary sick, we’ll have to g et someone else to g ive the lecture) to yield a S is the subject of that predicate phrase, ev en if the NP is a highly atypical one, as in (29): (29a) It was raining. (29b) There was a gun in his hand.

74.  English

(29c) My leg was pulled. (with pull X′ s leg = “play jokes on X”) The claim that there and not a gun is the surface subject of (29b) is confirmed by the fact that agreement is controlled by whatev er there replaces (here, the g un), irrespectiv e of whether that item is the underlying subject of the v erb; there is plural agreement not only in (30a), where be appears to agree with its underlying subject, but also in (30b), where the plural NP is the underlying subject only of the subordinate V: (30a) There were several books on the table. (30b) There are/*is said to be some problems with this proposal. Discrepancies between grammatical relations and syntactic configurations emerge when one considers NPs within a V′. A NP that combines with a V (and possibly other material) into a V′ is not necessarily its direct object (or its indirect object, in examples such as John g ave Mary the money, where two NPs combine with a V). While there is no consensus as to what exactly constitutes a direct object, it is at least generally agreed that the NP that follows the V in the surface form of “There-insertion” sentences is part of the V′ but is not a (deriv ed) direct object of the v erb. Montague grammarians such as Bach (1980) hav e argued that postv erbal NPs that cannot become subjects of passiv es are not direct objects, with e. g. the NP following reg ard but not the one following strike being a direct object: (31a) Everyone regards Bill as demented. (31a′) Bill is regarded by everyone as demented. (31b) Bill strikes everyone as demented. (31b′) *Everyone is struck by Bill as demented. 3.2. Relation-Changing Constructions There are a number of syntactic constructions in English in which a grammatical relation in surface structure is filled by a constituent that has some other status in underlying structure. In the passiv e construction, a constituent of a V′ (not necesssarily a direct object) takes on the role of subject (strictly speaking, not the subject of that V′ but the subject of be + the V′ in its past participle form); the underlying subject is either deleted or mov ed into a by-phrase adjunct to the V′. The NP

1325

that is adv anced to the role of subject can be a direct object, an indirect object, or an object of a P, with direct objects allowing the adv ancement quite freely, while objects of Ps allow it only when there is semantic motiv ation for the adv ancement (e. g. the NP is presented as denoting an “affected” entity) and there is either no direct object or a highly nonreferential direct object (Ø indicates a “gap”, here the gap left by the adv ancement of the NP to subject position): (32a) Mary was given Ø the money (by John). (passive of (John) gave Mary the money) (32b) The money was given Ø to Mary. (passive of X gave the money to Mary) (32b′) *Mary was given the money to Ø. (32c) I hate being looked at Ø by people. (32d) Kissinger is often stolen (*jokes) from Ø. (passives of X steals (jokes) from Kissinger) (32e) The house was set fire to Ø by some vandals. Passiv es in English do not in general require that the adv anced NP denote an affected entity or that the underlying subject denote an agent: (33) That fact is known Ø by everyone. The NP cannot be adv anced out of a V′ contained in the passivized V′: (34a) *You were tried to warn Ø. (passive of X tried to warn you) (34b) *Bill was happened to notice Ø by Mary. (passive of Mary happened to notice Bill) In the simple Tough-mov ement construction (henceforth, TM), the deriv ed subject is an underlying constituent of a nonfinite V′, indeed, an infinitiv e V′ with to (except perhaps for worth V-ing ), and may be embedded arbitrarily deeply in it: (35a) John is easy for us [to talk to Ø]. (35b) Linguistics is hard to get people [to stop saying stupid things about Ø]. (35c) An assumption as implausible as that is impossible [to prevent people from taking Ø as grounds for laughing at you]. (35d) This book is worth [buying a copy of Ø]. There is also a “circumnominal” TM construction, in which an A and an infinitiv e V′ with a gap surround an N′:

XXI. Syntaktische Skizzen

1326

(36) John is an easy person for us to talk to Ø. Circumnominal TM expressions are popularly treated as reduced relativ e clauses inv olv ing a simple TM construction, e. g. (36) would hav e the same underlying structure as John is a person who is easy for us to talk to. Howev er, that analysis is untenable, since many As that are excluded in the simple TM construction can appear in the circumnominal construction: (37a) Ed was a stupid person to invite Ø. (37a′) *Ed was stupid to invite Ø. (37b) That was an ingenious remark to make Ø. (37b′) *That remark was ingenious to make Ø. The As (also some predicate NPs, such as a breeze or a pain in the ass) that allow simple TM all express the ease or difficulty of what they are predicated of, while the circumnominal construction is not subject to that restriction. Other constructions that inv olv e change of grammatical relations are extraposition of complements, as in (38a—a′), Quantifierfloat, as in (38b), and the There-insertion construction that was taken up in 3.1.: (38a) It amazes me that Fred was so rude to you. (38a′) I took it for granted that I would run into you eventually. (38b) The guests all left before 10:00. The it of (38a—a′) marks the underlying position of the complement S, which appears instead at the end of its host S as an adjunct to the predicate phrase. A “floated” quantifier such as the all of (38b) occupies the position of an Adv (here, modifying the V′ left before 10:00) but has as antecedent a NP with which it is combined semantically: all is interpreted here as in All the g uests left before 10:00. In English, only all, both, and each (and “hedged” v ersions of all such as almost all) can be floated; by contrast, Japanese allows a much broader range of floated quantifiers. Leav ing aside sentences such as I g ave the children a dollar each, which are clearly instances of a different construction, English allows only a subject to be the antecedent of a floated quantifier, unlike German and Japanese, where objects too can be antecedents. A quite different kind of relation-changing construction is represented by the cleft and

pseudo-cleft constructions, in which a constituent is “focused”, i. e. is explicitly presented as contrasting with alternativ es, by mov ing it into a predicate position: (39a) It was a bicycle that John wanted Ø. (cleft) (39b) What John wanted Ø was a bicycle. (pseudo-cleft) In both constructions, the embedded S contains a gap matching the focused constituent. The two constructions are identical semantically but are subject to v ery different syntactic restrictions. The focus of the cleft construction can be either a NP or a P′, though some regional v arieties of English (particularly, Irish) allow other things (e. g. A′s) in that role. In the pseudo-cleft construction, there is considerable dialect and indiv idual v ariation in what the wh-pronoun can be, with the focused constituent limited only by the requirement that it match the wh-pronoun; all v arieties of English allow it to be what, in which case the focused constituent can be not only a NP but a predicate A′ or P′ or a bare infinitive V′: (40a) %When Bob left was on Tuesday. (40a′) It was on Tuesday that Bob left. (40b) What Tom has always been is afraid of snakes. (40b′) %It’s afraid of snakes that Tom is. (40c) What he did was climb a tree. (40c′) *It was climb a tree that he did. The internal structure of the embedded S of the cleft construction is that of a restrictiv e relativ e clause; it can be introduced by a relativ e pronoun, by that, or by nothing at all, with the latter possibility subject to the same dialect variation as in restrictive relatives: (41a) It’s Judy who/that/Ø I was talking to. (41b) It was on Tuesday when/that/Ø Bob left. (41c) It was Bob who/that/%Ø left on Tuesday. The form of the embedded S in the pseudocleft, by contrast, is that of an embedded question, in that what and not which is used as the wh-pronoun.

4.

Complements

The term “complement” here refers to a S that serv es as subject or object of a higher predicate, commonly in an altered form in

74.  English

which it is marked by a complementizer or by a non-finite form of the v erb or is reduced in v arious ways, most commonly through omission of the subject. 4.1. Full Complements English has an unusually broad range of forms for full complements: (42a) Mary told me that she had taken a new job. (42b) It’s illegal for liquor to be sold to minors. (42c) Your agreeing to help us means a lot to me. (42d) I disapprove of children being disrespectful to their parents. (42e) It seems as if they took us at our word. (42f) He keeps raving about how all the Cabinet members are communists. The that, for-to, ’s-ing , and Accusativ e-ing complements of (42a—d) are fairly unrestricted in their usage: they can appear as subjects and direct objects and also as objects of prepositions, with the qualification that prepositions are deleted before that and forto complements. The occurrence of a P in the underlying structure of such sentences as (43a, b) is confirmed by the ov ert occurrence of a P in corresponding pseudo-cleft sentences: (43a) I insist that you join us for dinner. (43a′) What I insisted on was that they join us for dinner. (43b) I’m thankful to be alive. (43b′) What I was thankful for was to be alive. The as if complement of (42e) occurs only as an extraposed subject complement of v erbs such as seem, look, and sound, and the how declarativ e complement of (42f) (which is syntactically and semantically distinct from the how interrogativ e complement of such sentences as I don’t know how she found me) occurs only as the object complement of certain verbs of saying. There are both finite and infinitiv al interrogative complements: (44a) He asked me if I would lend him $5. (44a′) He asked me whether I would lend $5 to a total stranger. (44b) I wonder what they’ll do to us. (44c) I wish I could tell you what to do.

1327

(44d) I don’t know whether to help him or to tell him to leave us alone. Both if and whether introduce dependent yesno questions, though (as Bolinger 1978 points out) they are not interchangeable, whether being restricted to reports of questions in which both an affirmativ e and a negativ e answer are regarded as possible; thus, whether often sounds odd in reports of rhetorical questions: (45) When Fred told Elaine that he preferred Bruckner to Brahms, she asked him if/ ??whether he was out of his mind. The complementizer that can often be omitted from finite complements that are not in subject position, though omission often results in substantially lowered acceptability when the complement is distant from its host: (46a) He told us (that) he was sorry. (46b) He told us with tears in his eyes that/ ?Ø he was sorry. The complementizer for is deleted when the subject of a for-to complement is extracted or deleted, as in examples such as (49) in 4.2. Subject to considerable dialect v ariation, it is also deleted when it immediately follows want and some other v erbs, though it is retained when separated from the verb: (47a) Mary wants (%for) John to go home. (47b) Mary wants very much for/*Ø John to go home. (47b′) What Mary wants is for/*Ø John to go home. 4.2. Reduced Complements The most common forms of reduced complements in English are infinitiv e V′s (with or without to) and -ing V′s: (48a) It’s illegal to sell liquor to minors. (48a′) Let them do anything they want to. (48b) Working in a factory isn’t fun. These nonfinite V′s have no overt subject but correspond in one of three ways to underlying structures in which they hav e subjects: (i) in some cases the subject of the V′ undergoes Raising and takes on the role of subject or object of the host S; (ii) in some cases the subject of the V′ is deleted by Equi-NP-deletion under the control of a NP in the host S (the deleted NP will be represented in the underlying structures by a personal pronoun with the controller as antecedent, since its contribution to the meaning of the S is that which such a pronoun would make); and (iii)

XXI. Syntaktische Skizzen

1328

in some cases a “generic” subject of the V′ is allowed to have a null realization: (49a) [everyonei expect [hei get rich]] > Everyone expects to get rich. (49b) [[you be angry] seem] > You seem to be angry. (49c) [[onei nominate onei] be against the rules] > Nominating oneself is against the rules. The need to allow some predicate elements to participate in deriv ations with Equi-NP Deletion and others to participate in deriv ations with Raising can be seen from examples in which the subject is something that is licensed by material in the dependent V′: (50a) *There expected to be an error in your argument. (50a′) There seems to be an error in your argument. (50b) *It expected to be raining. (50b′) It seems to be raining. (50c) *My leg expects to have been pulled. (50c′) My leg seems to have been pulled. With the underlying structure posited for seem (49b), there is nothing to prev ent the embedded S from being the structure that underlies There is an error in your arg ument, It is raining , or My leg has been pulled, and the subject of the embedded S (specifically, the deriv ed subject: My leg has been pulled is deriv ed from a structure corresponding to X has pulled my leg ), no matter what it is, is free to become the deriv ed subject of seem. By contrast, if the corresponding Ss were used as the underlying object complement of expect (49a), no application of Equi-NP Deletion would be possible, since there is no way that expect could license existential there, “ambient it”, or a piece of the idiom pull X’s leg as its underlying subject. In some cases, complements undergo more sev ere reductions, as when the v erb is deleted. That the apparent object of want in (51a) is a reduced complement can be seen from the fact that the S of which it is a reduced form serv es as a locus of modification (in (51b) the adv erbial modifies the understood “Mary has the apartment”) and of conjoining ((51c) is ambiguous with regard to whether it refers to one desire that can be fulfilled in two ways or expresses uncertainly as to what Nancy wants) and can serv e as the scope of a quantifier (as in the “nonreferential” interpretation of (51a), in which it does not imply that there is a specific lollipop that he wants):

(51a) John wants a lollipop. (< Johni wants [hei have a lollipop]) (51b) Mary wants the apartment until October. (51c) Nancy wants either a Cadillac or a Lincoln. Whether past participle complements, as in (52), need to be recognized as a separate complement type depends on how one treats the past participle that accompanies be, have, and g et and whether one can justify positing deletion of a passive be in (52): (52) The king ordered the prisoners drawn and quartered. 4.3. Auxiliary Verbs The v arious auxiliary v erbs fill the V position of a [V′ V V′] combination; the dependent V′ is in the -ing form for progressiv e be, the bare infinitiv e form for modal auxiliary v erbs other than oug ht, which takes an infinitiv e with to, and the past participle form for perfect have and passiv e be. Except for the “be able” sense of can and the “be obliged” sense of must, which behav e like Equi-NP deletion predicates, and passiv e be, which requires a treatment sui g eneris, the v arious auxiliary v erbs behav e like raising predicates such as seem. A treatment of English auxiliary v erbs as Raising or Equi-NP-deletion predicates yields the correct deriv ed constituent structure, in which each auxiliary V combines with a non-finite V′ into a larger V′. Have replaces the past tense when a past tense clause (such as Mary left an hour ag o) is embedded as a non-finite complement (Mary appears to have left an hour ag o). This alternation can be dev eloped into an argument (McCawley 1971, 1988) that tenses are themselv es raising predicates and that all instances of auxiliary have are underlying past tenses that are in nonfinite contexts and thus undergo conv ersion into have. The fact that, with one rare but important exception, English auxiliaries come in the sequence (Modal)(hav e)(beprog) (bepass) can be explained on the basis of morphological, syntactic, and semantic properties of the indiv idual auxiliary v erbs plus the restriction on the passiv e construction that was noted in 3.2.: (i) each auxiliary v erb imposes a particular non-finite form on its complement; (ii) modal auxiliaries in English (unlike the other Germanic languages) hav e only finite forms, and thus a modal that followed another auxiliary v erb would require a form

74.  English

that its morphology does not prov ide for; (iii) the complements of auxiliary v erbs and of tenses are non-finite, and thus a Past in the complement of an auxiliary is conv erted into have (and a Pres is deleted); (iv ) in most (though not all!) v arieties of English, multiple occurrences of auxiliary have are reduced to one occurrence (thus, a past perfect clause such as John had already left when Mary arrived has a nonfinite counterpart John seems to have already left when Mary arrived rather than *John seems to have had already left when Mary arrived, though for some speakers the acceptability of these examples is rev ersed); (v ) progressiv e be requires a complement that denotes a process or activ ity; and (v i) in passiv ization, as was noted abov e, a NP cannot be adv anced from a dependent V′. These characteristics rule out almost all of the combinations of auxiliary v erbs that div erge from the template (Modal)(hav e) (beProg)(bePass); for example, passiv e be could precede an auxiliary v erb only through passiv ization of a S whose top predicate was an auxiliary V, which would v iolate (v ), and (iv ) will generally prev ent progressiv e be from being followed by an auxiliary v erb other than the semantically transparent passiv e be, since a V′ headed by an auxiliary V generally denotes a state rather than an activ ity or process. Howev er, there is one class of be having sentences that do not v iolate (iv ), namely those in which the have is combined with the iterativ e construction that conv erts state V′s into process V′s, and such sentences are reasonably acceptable despite their not conforming to the template: (53a) *John is knowing the answer. (53a′) You’re always knowing things I never expected you to know. (53b) *You’re having worked too hard. (53b′) Whenever I see you, you’re always having just returned from a vacation in Florida.

5.

Modifiers

5.1. Adverbials Ss, V′s, and Vs can be modified by Adv s or by P′s: (54a) Usually there is no problem. (54a′) In most cases, there is no problem. (54b) John offended Mary intentionally.

1329

(54b′) John offended Mary by intention. (54c) Edith turned the handle carefully. (54c′) Edith turned the handle with care. “Adv erbial NPs” such as last Tuesday are P′s in which the NP is the object of a zero P, a conclusion that can be justified both on the grounds that their distribution is that of P′s and not that of Adv s (for example, they can appear only as right and not as left sisters of V′s that they modify) and on the basis of their semantics, in which the meaning of the NP and that of its host are connected in the same way as in sentences with an overt P: (55a) John frequently got drunk. (55a′) *John many times got drunk. (55b) Frieda is arriving next Tuesday. (55b′) Frieda is arriving on Tuesday. Words such as today, tomorrow, here, there, now, then are commonly misclassified as Adv s but are in fact NPs; their adv erbial uses hav e the distribution of P′s but not of Adv s, except that now and then hav e dev eloped additional uses in which they are Advs: (56a) *He here washes the windows. (56b) ?Bill right now is busy. (56b′) I now have visited Japan six times. V-modifying Adv s differ from V′-modifying Adv s in not allowing focus-markers such as only to interv ene between the Adv and the V: (57a) John intentionally only reprimanded Mary. (V′-modifier; focus underlined) (57b) *John severely only reprimanded Mary. (V-modifier) When a focus-marker such as only is separated from its focus, it has to be an adjunct to a predicate phrase containing the focus; the unacceptability of the word order in (57b) is ev idence that the position where only appears in that example is not the beginning of a predicate phrase, and the only way for that condition to be met is for the Adv to modify the V. The possibilities for conjoining confirm that intentionally is a V′-modifier and severely a V-modifier — the one can apply jointly to all the conjuncts of a coordinate V′, while the other cannot: (58a) John intentionally [[offended Mary] and [made a fool of Ted]]. (58b) John [severely reprimanded Mary] and [criticized Ted]. This difference in syntactic structure matches a difference in possibilities for scope of a

XXI. Syntaktische Skizzen

1330

quantifier within the V′ (Thomason and Stalnaker 1973); with a V′-modifier such as intentionally, there is an ambiguity as to whether a quantified object has the Adv inside or outside its scope, whereas V-modifier such as severely has to be within its scope: (59a) John intentionally offended all of the guests. (59b) John severely reprimanded all of the employees. There is a corresponding semantic difference between S-modifiers and V′-modifiers — a quantified subject is ambiguous with regard to whether a S-modifier is in its scope, but a V′-modifier has to be in its scope: (60a) All of my friends sometimes spend Christmas in Hawaii. (60b) All of my friends intentionally treat Fred with contempt. Or at least, that difference distinguishes underlying S-modifiers from underlying V′modifiers. In the surface structures of both of these examples, the Adv occupies the position of a V′-modifier, though only one of these Adv s can occupy the S-initial position of a S-modifier: (61a) Sometimes he did crazy things. (61b) ??Intentionally he treated Fred with contempt. Indeed, S-modifiers generally allow the option of being conv erted into deriv ed V′-modifiers. A particularly common type of S-modifier is a P′ consisting of a P (traditionally called a “subordinating conjunction”) and a S object. The S object is usually finite, though some Ps allow or ev en require that their S object be non-finite: (62a) Before he went home, John phoned Linda. (62a′) Before going home, John phoned Linda. (62b) With the bus drivers (*are) on strike, we’ll have to form a car pool. 5.2. Modifiers of N′s and NPs N′s can be modified by restrictiv e relativ e clauses (in full or reduced form: see 2.2.) as well as by a v ariety of As that cannot plausibly be treated as reduced relativ es, notably semantically adv erbial As such as former and alleged:

(63a) Many prospective candidates for governor are unable to raise sufficient campaign funds. (63b) A one-time heavyweight boxing champion has been found murdered. Full restrictive relatives consist of either a relativ e expression or that or nothing, followed by a S that contains a gap. The possibilities for these three forms are those that correspond to an analysis in which the structure underlying the relativ e clause is a that-clause containing a pronoun that matches the host N′, the pronoun either is conv erted into the appropriate relativ e pronoun and mov ed (along with any material that the “Pied-piping” principle allows to accompany it) into the position of the that or is deleted, and the that, if not replaced by the relativ e expression, is optionally deleted, under conditions that are subject to much dialect variation: (64a) the violinist who/that/Ø I told you about (64a′) the violinist about whom I told you (64a″) the violinist whose recording of “Tzigane” we were listening to 5.3. Negation Finite and non-finite clauses in English are negated differently. A finite clause is negated by combining not with the tensed auxiliary v erb (an appropriate form of do if there otherwise is no auxiliary v erb), generally in a contracted form, while a nonfinite clause is negated by adjoining not as a left-modifier of its predicate phrase: (65a) I won’t reveal your secret. (65a′) I didn’t reveal your secret. (65b) You can count on my not revealing your secret. (65b′) He promised not to reveal your secret. It is clearly a V′ and not a V that not modifies in examples like (65b—b′), in v iew of the possibility of a single not applying to a conjoined V′: (66) You can count on my not [revealing your secret or risking your reputation]. The v arious nonfinite Ss that are posited in the analysis of auxiliary v erbs giv en in 4.3. conform to the above generalization: (66a) Mary has been [not saying hello to me] recently.

74.  English

(66b) He must have [not understood a word that you said]. To integrate the analysis of negation with that analysis of auxiliary v erbs, it is necessary to treat not as a deep-structure S-modifier, obligatorily undergoing the (usually optional) conv ersion of S-modifier into V′modifier, and in a finite S, being combined with the tense-bearing element. In the examples taken up so far, the negativ e element has been a modifier. (Ev en won’t is arguably V + modifier.) Negation can also be incorporated into v arious words. The transparent instances of incorporated negation are nothing , no one (a single word, despite the spelling), nobody, never, nowhere, no/ none, neither, and nor (and some others in v arious regional v arieties). None in none of them is the strong form of the determiner no (as in No money was left). Aside from nor, these words can be treated as combinations of not and an existential quantifier, the latter being expressed by a word of the any-series if incorporation does not take place. To a large extent, incorporated negativ es and not ... any- are interchangeable; howev er, incorporation must take place if the any-word would otherwise be higher in the structure than the not, and, except for the common use of never as a left modifier of the V′, not ... any- is preferred to an incorporated negativ e word in colloquial speech when the any-word is within the predicate phrase: (67a) No one paid any attention to him. (67a′) *Anyone didn’t pay any attention to him. (67a″) At no time did anyone pay any attention to him. (67b) He had no money. (67b′) He didn’t have any money. (67c) He never told me anything about his family. The any-words, in the uses in which they are “licensed” by a negativ e element, as in (67), are among the rather div erse negative polarity items (NPIs) found in English, which are normal only in combination with negative elements and other “licensers”: (68a) They wouldn’t/*would lift a finger to help us. (68b) The food there wasn’t/*was all that bad. NPIs are not, as is sometimes claimed, parts of idioms that also include a negativ e element, since it is immaterial how the negation

1331

is expressed: (69a) No one would lift a finger to help us. (69b) It’s unlikely that they’ll lift a finger to help us. (69c) I doubt that they’ll lift a finger to help us. (69d) Only Ann lifted a finger to help us. There in fact are a small number of idioms that include a negativ e element (not half bad, stop at nothing , would like nothing better), and in those expressions there is no freedom with regard to how the negation is expressed. (70a) Your brother will stop at nothing. (70a′) *Your brother won’t stop at anything. (70a″) *None of them will stop at anything. NPIs differ with regard to both what elements license them and how close they need to be to the licensing element, with the anywords requiring only that a possible licenser be higher in the structure (taken in the broad sense in which a negativ e V′ is higher in the structure than a subject complement) and the until of not until the most restricted: (71a) I’m surprised that anyone came. ... that John budged. *... that John left until 10:00. (71b) ?Only Peter budged when I asked them to make room. (71b′) *Only Bill left until 10:00. (= Except for Bill, no one left until 10:00) (71c) That anyone will lend him the money is unlikely. (71c′) *That Bill left until 10:00 isn’t likely. Must and should hav e the irregularity that a negation with which they are combined has the complement of the auxiliary v erb as its scope: (72a) You mustn’t bother him. (= must + you not bother him) (72b) I shouldn’t be so careless. (= should + I not be careless) Alternativ ely, mustn’t and shouldn’t could be treated as what Horn (1989, 271) calls “contraries in contradictories’ clothing”, i. e. as meaning the opposite, rather than the negation, of what they seemingly negate, just as not like expresses the opposite of like (i. e. “dislike”) rather than merely its negation. The existence of words that idiosyncratically combine with not/n’t to the yield an expression that means the opposite of what is ostensibly negated adds to the difficulty of giv ing

XXI. Syntaktische Skizzen

1332

a clear answer to the question of whether there is syntactic Negative raising (NR), e. g. whether sentences as in (73) hav e underlying structures in which the negativ e applies not to the main S but to the embedded S: (73a) Ed doesn’t think he’ll be finished until 10:00. (73b) We don’t want you to waste any more of our time. Certain v erbs denote relations whose opposites are for practical purposes equiv alent to what one gets by negating the complement of the v erb, e. g. the opposite of wanting something to happen is wanting it not to happen. For such v erbs, semantics could not prov ide any grounds for a choice between an analysis in terms of a NR transformation and one in which the v erb allows its formal negation to express the v erb’s opposite. The NPIs in (73) suggest one possible way that NR, if English has it, might manifest itself syntactically: do any possible cases of NR inv olv e NPIs in positions where they otherwise would not conform to the normal restrictions on their licensing? The any of (73b) is beside the point, since any can just as well be licensed by a negation in a higher clause as by one in its own clause, but NPI until (73a) normally cannot be licensed by a negation in a higher S (74), and its acceptability in (73a) prov ides ev idence for an underlying structure in which the not negates the embedded S: (74) *I didn’t say that Bill would be finished until 10:00. Horn (1979, 168) has also noted that the inv ersion that accompanies S-initial negativ e expressions applies (and indeed, is obligatory) ev en when the negativ e appears in a higher S in what could plausibly be regarded as a NR construction: (75a) I don’t think that ever before have the media played such a major role in a kidnapping. (75a′) ??I don’t think that ever before the media have played such a major role. Another case where an apparently raised negativ e leav es a mark of its negativ ity in the lower S is the tag-question complements discussed in R. Lakoff (1969): (76) I don’t suppose you’ve found the money, have/*haven’t you? The form of the tag reflects the auxiliary verb have of the embedded S, but the non-negativ e

form of the tag corresponds to a host S (you haven’t found the money) in which that auxiliary verb is negated.

6.

Coordination

6.1. Devices for “Reduction” of Coordinate Structures There are sev eral ways in which one can express the content of conjoined constituents (especially conjoined Ss) without needing to repeat identical parts of the conjuncts. (i) Conjunction reduction (henceforth, CR) allows one to replace a conjunction of items that contrast in only one place with a nonconjoined expression hav ing in that place a conjunction of the contrasting parts, as where conjoined Ss with contrasting Vs are converted into a simple S with conjoined Vs: (77) Bill washed the dishes, he dried them, and he stacked them > Bill [v washed, dried, and stacked] the dishes. Washed, dried, and stacked here is a V, not a V′: it is not syntactically a “phrasal” unit, ev en if it is phonologically a “phrase”. Unlike many languages in which one set of conjunctions conjoins NPs or P′s and another set conjoins V′s or Ss, English uses the same conjunctions irrespectiv e of the category of the conjuncts and allows units of v irtually any category to be conjoined, though since the conjuncts hav e to stand in some sort of contrast with one another, conjoining of items such as articles that normally do not participate in contrasts is usually excluded: (78) *I didn’t use the or a screwdriver. Not all conjoined parts of Ss can be deriv ed from conjoined Ss; well-known examples such as (79) show that when a predicate expresses a property of a set, a conjoined NP listing the members of the set cannot be derived through CR: (79a) The king and the queen are an amiable couple. (79a′) *The king is an amiable couple and the queen is an amiable couple. In most such cases, the conjunction is and, though or occurs when the conjuncts are a list of alternativ es from which a choice is to be made; and in such examples lacks the usual option of introducing the first conjunct with both:

74.  English

(80a) *Both the king and the queen are an amiable couple. (80b) (Either) Nixon or Humphrey was a horrible choice. The clearest cases in which CR needs to be inv oked are those in which one of the conjuncts is deriv ed by a transformation whose domain of application is a S, as where one conjunct of a conjoined V′ is passiv e or where one conjunct of a conjoined A′ is derived by TM: (81 a) Tom robbed a bank and was caught by the police. (81 b) Janet is both clever and easy to work with. In addition to the simple CR of the examples treated so far, there is a generalized CR that applies to conjoined Ss whose conjuncts are parallel but contrast in two or more positions, yielding deriv ed forms in which respectively marks a deriv ed coordinate structure that is treated as dependent on the coordinate structure with which its conjuncts are to be matched: (82) John and Mary applied to Harvard and Stanford, respectively, in linguistics and computer science, respectively. In v iew of the possibility of respective and respectively hav ing the same antecedent, the same mechanism must cov er the interpretation of respective as cov ers the interpretation of respectively: (83a) John and Mary applied to their respective mother’s alma maters in linguistics and computer science, respectively. (83b) John and Mary applied to Harvard and Stanford, respectively, in their respective fathers’ major fields. CR is not (as linguists who should hav e known better hav e often claimed) a deletion of repeated material but a factoring out of material that is shared by the conjuncts of a coordinate structure. (ii) Gapping, by contrast, is a deletion of repeated material: (84a) John wants to write a historical novel, Mary a guidebook to Madrid, and Lucy a treatise on probability. (84b) On Tuesdays John plays poker, and on Wednesdays bridge. The deleted material need not be a syntactic

1333

unit (note that neither wants to write nor John plays is a constituent in these examples); what is required is that the two residues that remain behind in each conjunct be constituents: one constituent outside the predicate phrase (usually a subject, but it can be a Smodifier, as in (84b)) and one non-predicate constituent within the predicate phrase. (iii) in Right-Node-Raising (RNR), identical final constituents of the v arious conjuncts are fused into one: (85a) John has written books about, and Mary has done extensive fieldwork on, the dialects of Northern Albania. (85b) Many linguists have dealt only superficially with, and a number have risked their reputations on claims that they should have known better than to make about, the syntax of coordinate structures. Note that the “shared” item need not hav e the same syntactic role in all the conjuncts. RNR must be treated as a fusion of identical items, and not (as is often thought) as an across-the-board extraction of the shared item, since with respect to all syntactic phenomena, the “shared” item behav es as if it remains within all the conjuncts; for example, in (85b) the syntax of coordinate structures cannot hav e been extracted from all of the conjuncts, since extractions from a complex NP (such as claims that they should ...) are normally excluded. (iv ) Unlike (i)—(iii), Stripping is restricted to two-term conjoining, reducing the second conjunct to a single constituent that contrasts with its counterpart in the first conjunct, along with an additional element such as too, also, or not: (86a) Susan contributes books to the library, but not money. (86b) John sometimes gets drunk, and Mary also. As with Gapping, what is deleted need not be a constituent, and it is of no significance that what is deleted is sometimes a constituent, as in (86b); ev en though a V′ is deleted, (86b) is not an instance of V′-deletion, which has to leav e a tense behind and is not restricted to coordinate structures: (87) If Susan contributes books to the library, John will/*Ø too.

1334

6.2. Differences among Conjunctions And, or, and (neither ...) nor allow any number of conjuncts from two on up, while all other coordinating conjunctions allow only two: (88a) Lithuania, Latvia, and Estonia regained their independence in 1991. (88b) You can make this dish with chicken, pork, or lamb. (88c) Neither Nixon nor Reagan nor Bush reduced the federal budget. This presumably reflects the fact that these conjunctions denote a symmetric relationship among the conjuncts, while the meanings of the other conjunctions (such as but and as well as) are not completely symmetric with regard to the conjuncts. The belief (especially popular among logicians) that multi-term and and or can be reduced to iterated twoterm conjoining (as in an analysis of (88a) as an inv olv ing a conjunction of Lithuania and Latvia with Estonia and deletion of one occurrence of and) is not v iable, since it makes it impossible to treat rules such as Gapping as simply applying across-the-boards in examples such as (84a) in which the third conjunct must undergo the same deletion as the second. This is not to say that two-term conjunction cannot be iterated; of course, it can, as in the structure that underlies (89): (89) John got drunk and wrecked his car, and his wife left him. The point is simply that [[S and S] and S] is a different syntactic structure from [S, S, and S], and neither can be reduced to the other. A coordinate structure in English normally requires a conjunction as an adjunct to the last conjunct, and if the conjunction is one of the three that take arbitrarily many conjuncts, it may (sometimes, must) also appear as an adjunct to the earlier conjuncts, with the qualification that and, or, and nor take the respectiv e forms both, either, and neither when they introduce the first conjunct. It is instructiv e to tabulate the options for four-term or-conjunctions, since or is the conjunction that exhibits the greatest freedom with regard to spreading it ov er the conjuncts (both is restricted to conjunctions of two conjuncts, and in modern English, nor has to be accompanied by neither on the first conjunct), and since one needs at least four conjuncts in order to see how the conjuncts between the first and last work. There are the

XXI. Syntaktische Skizzen

two independent options of spreading the conjunction onto the first conjunct or not, and of spreading it onto all of the interv ening conjuncts or not: (90a) They’re going to open a branch office in (either) Boston, Atlanta, Denver, or Seattle. (90b) They’re going to open a branch office in (either) Boston, or Atlanta, or Denver, or Seattle. (90c) *They’re going to open a branch office in (either) Boston, Atlanta, or Denver, or Seattle. These options apply in principle to and and nor too, though their special properties narrow the range of possibilities. But and plus are restricted to two conjuncts, as are sev eral items that hav e become coordinating conjunctions through the reanalysis of structures inv olv ing subordination: as well as, besides, rather than, instead of, ... There is considerable indiv idual v ariation with regard to what these “peripheral” coordinating conjunctions are allowed to coordinate, some of them generally serv ing to coordinate NPs, P′s, and nonfinite V′s but, except for more permissiv e speakers, not Ss or finite V′s: (91 a) Debbie keeps cassettes as well as CDs in this cabinet. (91 b) I often get letters from Russia as well as from Japan. (91 c) John asked Mary to lend him money as well as help him find a job. (91 d) Mary lent John money as well as helping/%helped him find a job. Or at least, less permissiv e speakers do not allow a finite V′ as the second conjunct with as well as. However, even the version of (91 d) with helping behav es syntactically as a coordinate structure in allowing across-the-board extraction and RNR: (92a) John is the only person that Mary lent money as well as helping to find a job. (92b) Bill often writes articles about, as well as maintaining a data base on, Emma Goldman. In sentences like (92a), the V′ as well as V′ combination is a coordinate structure syntactically but not morphologically. In such languages as Japanese and Tamil, syntactic coordination of Ss or of V′s is not morphologically coordinate: non-final conjuncts hav e a special non-finite form, and only the final

74.  English

conjunct bears the tense or other inflection required by the context in which the coordinate V′ or S occurs. English also has structures that are morphologically but not syntactically coordinate, as in the g o and V′ construction, in which both v erbs must be inflected alike, but extractions out of the supposed second conjunct are allowed freely: (93a) He went and told her the truth. (93b) What did he go and tell her? 6.3. Agreement with Coordinate Subjects When an and-conjunction is the subject of a finite S, the finite V takes an agreement form that corresponds to the denotation of the conjoined subject, i. e. usually a plural form, but singular when the conjuncts are coreferential and singular: (94a) Bill and I are/*is/*am going on a picnic. (94b) His closest friend and the committee’s choice as his successor was/were at his bedside. (singular if the friend and the successor are the same person, plural otherwise) With or, the coordinate subject does not have a determinate reference, and each conjunct makes its own demand on the agreement form. If a single form of the v erb can satisfy all of those demands simultaneously, that form is used; otherwise, many speakers find no form of the v erb acceptable, though many hav e dev eloped a patch (Morgan 1972) that enables them to say things for which their mental grammars prov ide deep structures but no determinate surface structure, such as a policy of making the v erb agree with the closest conjunct: (95) Either Bill or I *is/*are/%am going to be laid off. With other coordinating conjunctions, each conjunct likewise makes an independent demand on the agreement form: (96a) Sheila as well as Bill is/*are going to be laid off. (96a′) Sheila as well as I *is/*am/*are going to be laid off.

7.

Anaphora and Ellipsis

7.1. Essential and Derived Anaphora An anaphoric element is an item that is interpreted in terms of an antecedent elsewhere in

1335

the sentence or the discourse. Sense anaphora, such as the so of (97a), contribute the sense of the antecedent to the clause in which they occur, and reference anaphora, such as the personal pronoun of (97b), contribute only the reference of the antecedent, without regard to its sense: (97a) Mary is sure that the world will end soon, and Tom thinks so too. (97b) My 7-foot tall brother is under the delusion that he is a midget. As Lakoff (1968) and Hankamer and Sag (1976) hav e argued, some types of anaphoric elements are derived elements, i. e. they are deriv ed from a structure containing two identical constituents through a transformation that replaces one of them by a (perhaps null) pronoun hav ing the other constituent as antecedent, while other types of anaphoric elements are essential, i. e. they are not so deriv ed (though they might well replace other essential anaphoric elements, as in an analysis of else as other than (pronoun)). Lakoff and Hankamer and Sag hav e adduced a number of systematic distinctions between derived and essential anaphoric elements: (i) A deriv ed anaphoric element, since it replaces a repeated syntactic unit, requires a linguistic antecedent, while an essential anaphoric element allows a situational antecedent. The zero V′ of (98a) illustrates the behav ior of deriv ed anaphoric elements, while the it of (98b) illustrates that of essential anaphoric elements: (98) (while the speaker and addressee watch someone juggle three oranges) (98a) *I can Ø too. (98b) I can do it too. (ii) In v irtue of the cyclic principle, the antecedent of a deriv ed anaphoric element can be a constituent created by transformations; the antecedent of an essential anaphoric element cannot. Conv ersely, the antecedent of an essential anaphoric element can be a deep structure constituent that is broken up in the course of a deriv ation, while the antecedent of a derived anaphoric element cannot: (99a) John has been mistreated by the boss, and Lucy has Ø too. (Ø = been mistreated by the boss) (99b) I may be wrong, but I doubt it. (it = I am wrong) (99b′) *John is hated by his relatives, but his friends don’t Ø. (Ø = hate John)

1336

(iii) There are both essential reference anaphora (such as personal pronouns) and essential sense anaphora (such as the one of ... and I boug ht a blue one), but deriv ed elements hav e to be sense anaphora in v iew of being deriv ed from repetitions of an expression hav ing the sense of the antecedent. (iv ) Since a deriv ed anaphoric element can be deriv ed from a constituent containing the antecedent of a pronoun, deriv ed anaphoric elements can prov ide “missing antecedents” for pronouns elsewhere in the discourse; essential anaphoric elements, by contrast, cannot: (100a) Studs Terkel has never interviewed an astronaut, but Barbara Walters has Ø, and he was a real bore. (Ø = interviewed an astronaut; he = the astronaut that Walters interviewed) (100a′) ??Studs Terkel has never interviewed an astronaut, but Barbara Walters has done it, and he was a real bore. (v ) If one makes the (reasonable, though far from triv ial) assumption that linguistic structures must be finite, then in a Bach-Peters sentence (a sentence in which two anaphoric elements are contained in each other’s antecedents), the anaphoric elements cannot both be deriv ed, since otherwise there would hav e to be an infinite structure underlying the sentence. Howev er, not all essential anaphoric elements can yield Bach-Peters sentences — only reference anaphora can: (101a) The boy who wanted it got the prize that he deserved. (101b) *That Bill wants to Ø suggests that Mary expects to Ø. (101c) *Inexperienced writers of lurid ones often enjoy elegant novels about successful ones. 7.2. Restrictions on Anaphoric Relations Anaphoric elements in English are subject to constraints on their surface position in relation to their antecedents. An anaphoric element is not allowed to c-command its antecedent in surface structure when it is a clausemate of (102a) or precedes (102b) the antecedent: (102a) *Near Johni, hei saw a snake. (102a′) Near himi, Johni saw a snake. (102b) *Shei went home after Janei finished the report. (102b′) Janei went home after shei finished the report.

XXI. Syntaktische Skizzen

(102c) After Janei finished the report, shei went home. (102c′) *In Maryi’s manuscript, shei found several errors. (102c″) In the manuscript that Maryi was reading, shei found several errors. A simple requirement that no anaphoric element c-command its antecedent is not v iable in v irtue of the possibility of backwards pronominalization into a subordinate clause (102c, c″). (This statement assumes the constituent structure argued for in 5.1., in which after S is an adjunct to the S that it modifies, and the proposal of McCawley 1988 that modifiers are transparent with regard to command relations, which jointly imply that she c-commands Jane in (102c)). This restriction implies that backwards pronominalization with the antecedent in a S-modifier is possible when the pronoun is in the V′ but not when it is a subject, and that that possibility v anishes when the modifier containing the antecedent has to be interpreted as a surface V′-modifier; those implications are correct: (103a) Mary hit himi before Johni knew what was happening. (103b) *Hei was knocked unconscious before Johni knew what was happening. (103c) *Mary [hit himi before John knew what was happening and screamed for help]. These constraints appear to apply not only to personal pronouns but to anaphoric elements in general, though it is difficult to test this fully, since many anaphoric elements cannot occupy the crucial positions, e. g. sentential so cannot be a subject, and pronominal one is an N′ and thus does not c-command anything outside of its NP: (104a) Unless the pope says so, I won’t believe that human beings are related to apes. (104b) The young robin that the old one had pecked was bleeding. (104b′) *The young one that the old robin had pecked was bleeding. (one ccommands robin) The unacceptability of sentences such as (105a) cannot be attributed to an impermissible surface configuration of pronoun and antecedent, since it does not differ in surface constituent structure from acceptable examples such as (105b):

74.  English

(105a) *It was close to Johni that hei saw the snake. (105b) It was lucky for Johni that hei saw the snake. Since such pairs of sentences have underlying structures that differ in a respect that can plausibly be held to be responsible for the difference in (105), namely that the A′ is an underlying complement of see in (105a) but not (105b), Carden (1986) proposed that pronouns are also subject to a constraint on cyclic outputs, in this case, a constraint that would be v iolated by the cyclic output Hei saw the snake close to Johniof the S that is embedded in the cleft structure of (105a). Carden’s constraint on cyclic outputs amounts to a requirement that, under a certain structural condition, one of two coreferential NPs in a cyclic output be a pronoun with the other as antecedent. The structural condition, howev er, cannot be simply that the antecedent c-command the pronoun, since that would wrongly exclude such sentences as (102c); as a makeshift that fits the facts fairly well, let us say that a NP that is outranked by a coreferential NP must be a pronoun with the latter as antecedent, where a subject outranks all constituents of its predicate phrase and a bare (= without P) object outranks all other constituents of the predicate phrase. This condition allows one to distinguish cases where pronominalization is obligatory from cases where it is optional (the latter being cases where neither NP outranks the other in any cyclic structure nor ccommands the other in surface structure) and restricts the occurrence not only of personal pronouns but also of anaphoric epithets such as the bastard: (106a) Davei’s wife concealed the fact that Davei’s/hisi car was in the repair shop. (106a′) *Davei concealed the fact that Davei’s car was in the repair shop. (106b) Napoleoni’s nephew collected portraits of the bastardi. (106b′) *Napoleoni collected portraits of the bastardi. The choice between pronouns and full NPs is also influenced by considerations of logophoricity (Kuno 1987). When a clause represents someone’s thought or speech, pronouns are required in places where, in direct discourse, that person would hav e to use a pro-

1337

noun, as in (107a), where (unlike (107b)) the that-clause expresses Mary’s thought and thus requires a pronoun for references to Mary: (107a) *That Maryi would put up with John’s insults was expected by heri. (107b) That Maryi would put up with John’s insults was expected of heri. 7.3. Reflexives and Reciprocals English has distinct reflexiv e pronoun forms in all three persons, though (unlike Swedish, Russian, and Latin) no distinctiv ely reflexiv e possessiv e forms. The antecedent of a reflexiv e pronoun can be the subject, the direct object, or occasionally even the object of a P: (108a) John protected Mary from himself/ herself. (108b) John talked with Mary about herself. A reflexiv e pronoun contained within an N′ has an antecedent either within the same N′, or in the determiner position of its NP, or outside the NP, though the latter possibility is usually blocked if the NP is definite: (109a) Johni showed Mary a picture of himi/himselfi. (109b) Johni showed Mary that/Lucy’s picture of himi/*himselfi. Reflexiv es in English (unlike Japanese and Tamil) allow their antecedents to be in higher clauses only to a limited extent, typically cases in which the antecedent denotes a person who is conscious of the identity between him and the person denoted by the reflexive: (110a) I told Fred that we were disappointed with him/*himself. (110b) John was asked about all that scandalous gossip about him/himself. (Kuno 1987, 164—5) (110b′) John was oblivious to all the gossip about him/*himself. Expressions such as anyone other than X-self (discussed in Keenan 1988) allow reflexiv es to hav e antecedents in higher clauses quite freely: (111a) John thinks that Mary couldn’t love anyone other than him/himself. (111b) That no one but him/himself got a low grade worried John. Reciprocal pronouns can appear in roughly the same structural relations to their antecedents as can reflexiv es, though since the reciprocal has a possessiv e form, there are

XXI. Syntaktische Skizzen

1338

determinately reciprocal examples (112c) that have no determinately reflexive analog: (112a) They asked me about each other. (112b) I asked them about each other. (112c) Smith and White plagiarized each other’s work. A reciprocal can hav e an antecedent in a higher clause when the subject of the lower clause is deleted by Equi-NP-deletion; (113) allows not only an interpretation in which the antecedent of each other is the understood lower subject (both claim that they caused each other trouble) but also one in which it is the ov ert higher subject (each claims to have caused the other trouble): (113) Reagan and Mondale claimed to hav e caused each other trouble. 7.4. Other Anaphoric Elements English allows pronominalization of a peated V′ through deletion, subject to the striction that either an auxiliary V or a must be left behind; the usual restrictions backwards pronominalization apply:

rereto on

(114a) Unless you do Ø, I won’t sign the petition. (114b) If Tom has been to Lisbon, you can be sure that Edith has/*Ø too. (114c) I wouldn’t have come here if I hadn’t chosen to/*Ø. The constituent that is deleted need not be a V′ but can be an A′, P′, or NP that is a predicate complement of be: (115) Fred is a nice guy, and Bill is Ø too. Besides the zero V′, English has sev eral other zero anaphoric elements: (116a) Even though it’s obvious that Sam caused the accident, Ted accused Nancy Ø. (Ø = of causing the accident) (116b) Alice’s car is bigger than Jane’s Ø. (116c) Mary is younger than Tom, and Alice is older Ø. The possibility of a zero anaphoric complement, as in (116a), depends on the v erb; for example, prevent does not allow a zero complement, and criticize without an ov ert complement does not hav e an anaphoric interpretation, and probably should be treated as not having even an understood complement: (117a) *Bob prevented Carol from going to the meeting, and Alice prevented Ted.

(117b) Even though Edna criticized Sam for spending too much money, Jane didn’t criticize Edgar. (≠ criticize Edgar for spending too much money) A number of words are indirectly anaphoric, in that they hav e meanings that incorporate an anaphoric element, e. g. else, other, and different all mean “other than [pronoun]”, though they differ considerably in their syntax: (118a) If we’re going to Hong Kong and Singapore, we won’t have time to go anywhere else. (else = “other than Hong Kong and Singapore”) (118b) One person asked for a $10 donation and another person asked for a $50 donation. Similarly, therefore means “because of that” and nevertheless means “despite that”. One can speak of else, therefore, etc. as hav ing antecedents, namely the antecedent of the incorporated pronoun, if one keeps in mind that the interpretation of else, etc. is not that of the antecedent but something of which the antecedent is a proper part. Partee (1989) discusses many other cases of “concealed anaphora” (e. g. a foreigner is a person who is foreign in relation to the person(s) that are taken as a reference point) and notes that often no alternativ e form with an ov ert pronoun is possible: (119a) Citizens of every country tend to find foreigners attractive. (119a′) Citizens of every country tend to find persons foreign to them attractive. (≠ (119a)) That has sev eral anaphoric uses, including a use as a pronominal degree expression: (120a) Alex chose War and Peace and a Dostoevsky novel, and I chose that too. (Channon 1980, 103) (120b) Susan says her brother is so tall that he has to have his clothes specially made, but I really doubt that he’s that tall.

8.

Quantification and Syntax

8.1. Quantifier Scope and Syntactic Form The most straightforward way to represent the scope of a quantified expression is to represent the expression as an adjunct to the S that is its scope. Suppose that syntactic deep

74.  English

structures are taken as representing scopes directly in this way, so that e. g. an ambiguous sentence such as (121a) would hav e two deep structures, one with someone as an adjunct to the main S and one with it as an adjunct to the embedded S: (121a) John thinks that someone is persecuting him. (121b) (someone)x(John thinks (x is persecuting him)). (121 b′) John thinks ((someone)x (x is persecuting him)). Such deep structures require that one also posit a transformation that mov es the quantified expression into a position in the host S occupied by the v ariable that it binds; let us henceforth call such a transformation Q′lowering, in v irtue of the presumable status of the quantifier as the head of the quantified expression in logical structure. One then is forced to reformulate all transformations in which quantifiers play any role as applying to a structure in which the quantified expression has not yet been mov ed into its host S, since the principle of strict cyclicity would otherwise be v iolated: if Si consists of a quantified expression plus Sj, then once Q′lowering has applied with Si as its domain, ev erything that had been in Si is now in Sj, and any application of a transformation would be to a domain (Sj or one of its parts) that was properly contained in a domain (Si) to which cyclic transformations had already applied. For example, There-insertion will hav e to be formulated as applying to a structure consisting of an existentially quantified expression and a S of the form [x V′], where x is the v ariable that the existential quantifier binds. This explains why, although (121a) is ambiguous with regard to the scope of someone, (122) can only be interpreted with the embedded S as the scope: (122) John thinks that there is someone following him. Of the two deep structures that correspond to (121), only (121b′) meets the conditions for There-insertion (since only in that structure is someone combined with a S whose subject is x), and thus (122) has to hav e the embedded S as the scope of someone. More generally, all transformations in which quantifiers play any role will then be sensitiv e to the scope of the quantifier (since their effect will then be in the S that is the scope of the quantifier), and if their effect is

1339

at the top of the structure to which they apply (as in the case of There-insertion, Quantifier-float, and Only-float), they will disambiguate the scope of the quantifier, since differences in the domain to which the transformation applies will correspond to differences in where the effect of the transformation appears. This can be seen especially clearly with Only-float, where the scope ambiguity in (123a) is resolved in (123b—b′): (123a) Mary lets John drink only vodka. (underlines indicate the intended focus) (123b) Mary lets John only drink vodka. (123b′) Mary only lets John drink vodka. With narrow scope of only, (123a) says that it is all right with Mary if John drinks nothing but v odka (though she might prefer that he also drink wine and beer), which is the only meaning that (123b) allows, and with wide scope of only, (123a) says that v odka is the only thing that Mary allows John to drink (thus, she forbids him to drink wine or beer), which is the only meaning that (123b′) allows. The multiple positions of a floated quantifier in relation to auxiliary v erbs reflect multiple possibilities as to which of them are in its scope, and accordingly, which of the stack of V′s is the one to which the floated quantifier is adjoined. The position in which the floated quantifier appears matches its scope, with the qualification that, because of the way that Q-float interacts with Attraction-to-tense, a quantifier that immediately follows a tensed auxiliary v erb is ambiguous with regard to whether that v erb is in its scope: (124) The employees have all been stealing paperclips. While (124) can be interpreted with have outside the scope of all (= “There hav e been occasions when all the employees were stealing paperclips”), a more plausible interpretation is with it in the scope of all (= “Each of the employees has been stealing paperclips”), and the deriv ation of (124) in the latter interpretation inv olv es a step in which the have of [[the employees all have been stealing paperclips] Pres] is adjoined to Pres, thus mov ing from a position below all to a position above it. This treatment of quantifiers defuses one of the few plausible arguments that hav e been offered (Williams 1974) for a fixed ordering

1340

of transformations, namely the argument that Passiv e must be ordered before Quantifier-float in order to prev ent it from incorrectly applying to its output: (125a) The guests all praised the host. (125b) *The host was all praised by the guests. Under the treatment suggested here, in the deep structure all the g uests is combined with x praise the host, the domain to which Passiv e applies has to be the latter S, and Passiv e would thus “bleed” Quantifier-float, since Quantifier-float would require the v ariable to be the subject, in v irtue of the restriction in English that only quantifiers on subject NPs can float, and x is not the subject in [all the guests [the host be praised by x]] The once widespread belief that the position of quantifiers in the surface word order of a clause corresponds to their scope relations is grossly at odds with the facts. Different idiolects of English v ary considerably in what possibilities they allow for the scopes of quantifiers, and different quantifiers are subject to different restrictions and preferences for their scope possibilities (for example, each, as in Two secretaries assist each executive, usually takes wide scope). For those quantifiers whose scope possibilities are restricted in at least certain idiolects, the factor determining possible scopes is not word order but c-command. For example, while there are many speakers for whom no critics has only wide scope in (126a) and only narrow scope in (126a′), the relev ant factor is not which quantifier precedes the other but which Q′ (if either) c-commands the other, since (as Kuno (1971) has pointed out) such speakers generally allow no instructors to have wide scope in (126b′) just as in (126b): (126a) No critics reviewed many of his books. (126a′) Many of his books were reviewed by no critics. (126b) No instructors gave Fs to many students. (126b′) Fs were given to many students by no instructors. Passiv ization rev erses the word order of the two Q′s in both pairs, but only in (126a—a′) does it also rev erse their c-command relations. In conjunction with the cyclic principle, the treatment of quantified expressions as deep structure sisters of their scopes prov ides an explanation of why in some cases the

XXI. Syntaktische Skizzen

scope of quantifiers matches surface syntactic structure and why in other cases it does not. The input to any particular application of Q′lowering is a structure of the form [Q′ Si]. In v irtue of the cyclic principle, any transformations such as Passiv e that alter grammatical relations within Si will hav e applied with Si as domain. Thus, any constraints on the application of Q′-lowering will apply on the basis of the deriv ed structure of Si, not its deep structure. If the restriction that is in force in those idiolects of English in which sentences such as (126a—a′) are unambiguous with regard to scope is formalized as a constraint excluding deriv ational steps in which a negativ e Q′ (such as no ling uist or few politicians) rev erses its c-command relation with another Q′, only (127a) can be associated with the surface form (126a′), since the application of Q′-lowering that affects no critics in (127b) would be mov ing it into a position in the deriv ed structure many of his books be reviewed by x in which it would be c-commanded by a negativ e Q′ that it c-commanded in the input, and thus the constraint would be violated. (127a) not (some critics)x (many of his books)y (x review y) (127b) (many of his books)y not (some critics)x (x review y) (No other combination of not and the two quantified expressions comes into the picture, since not can be incorporated only into a Q′ that is in its immediate scope). Howev er, a transformation applying to a higher domain than one to which Q′lowering has applied can mov e the Q′ into a position higher than an element that is outside its scope, as in All the passeng ers appear to have escaped unharmed, which allows an interpretation in which the scope of all the passengers is the complement of appear, or can mov e to a position higher than the Q′ an element that is within its scope, as in the interaction between Q′-float and the rules for auxiliary v erbs noted abov e (yielding e. g. an interpretation of (124) in which have is in the scope of all). 8.2. Comparative Constructions In comparisons of inequality, the compared element is marked either with more or by inflection (usually the suffix -er, though a few adjectiv es hav e suppletiv e forms) and an expression introduced by than giv es the “stan-

74.  English

dard” of comparison. The comparison of equality, with as marking the compared element and as introducing the “standard” shares v irtually all of the syntax of the comparison of inequality and will accordingly be ignored in the rest of this section. The thanexpression exhibits sev eral degrees of reduction, ranging from a nearly full sentence to a single constituent: (128a) The USA exports more wheat to India than France exports beans to Vietnam. (128b) The USA exports more wheat to India than France exports to Vietnam. (128c) The USA exports more wheat to India than France does to Vietnam. (128d) The USA exports more wheat to India than France does. (128e) The USA exports more wheat to India than France. Underlining here indicates the host S counterparts of the material omitted from the than-expression. Normally at least the counterpart of the compared constituent (much, in (128)) is deleted, though ev en that item is retained when it contrasts with the compared constituent: (129) My table is wider than it is long. A constituent containing the counterpart of the compared element can be deleted if it does not contrast with the host S ([NP much wheat] in (128b), [V′ export much wheat to India] in (128d)), and one can retain a contrasting constituent of a V′ while deleting the remainder of the V′ (128c). The deletion in (128e) is of a v ery different nature from those in (128a—d), since (as in the coordinate stripping of (86)) the deletion (henceforth, Comparative Stripping) is not of a syntactic constituent but of ev erything but one constituent: (130) The USA exports more wheat to India than to Vietnam. As Russell (1905) noted, comparativ e constructions exhibit scope ambiguities, e. g. (131) has both a “narrow scope” reading, in which “Fred’s boat is longer than a’s boat is” is the content of Mary’s thought (where a is the speaker of (131)) and a “wide scope” reading, in which “Fred’s boat is more than x long” is the content of her thought (where x is the length of the speaker’s boat): (131) Mary thinks that Fred’s boat is longer than my boat is.

1341

The ambiguity is with regard to the scope of a definite description operator: “the x such that a’s boat is x long” can hav e either the main S or the subordinate S of “Mary thinks that Fred’s boat is more than x long” as its scope. If a deep structure is adopted in which the S of the than-expression is part of such a definite description operator, the deletion in (128d) can be assimilated to normal V′-deletion, since then the deleted V′ and its antecedent are disjoint in underlying structure, rather than one being contained in the other. Such an analysis predicts correctly that Comparativ e Stripping giv es rise to v ery different ambiguities from those created by V′-deletion (131). Comparativ e Stripping requires the S that undergoes reduction and the S to which it is adjoined to contrast in only one constituent. This means that the higher the S is adjoined in the structure, the more material it will hav e to contain for Comparativ e Stripping to be applicable. This implies that, with the present perfect analyzed as in 4.3., a sentence such as (132a) should be ambiguous among the three interpretations in (132b— b″): (132a) Bob has eaten more pizza than Ted. (132b) The amount of pizza that Bob has eaten exceeds the amount that Ted has eaten. (132b′) There was an occasione when the amount of pizza that Bob atee exceeded the amount that Ted atee. (132b″) There was an occasion when the amount of pizza that Bob ate exceeded the amount that Ted [tenseless] eat. The interpretation in (132b″) inv olv es an atemporal stereotype, here, a feat of pizzaeating for which Ted is renowned (say, a record in competitiv e pizza-eating, like the 60 home runs in a season for which Babe Ruth is renowned). The sentence does in fact hav e these three interpretations.

9.

Illocutionary Syntax

A number of syntactic constructions are intimately connected with particular types of illocutionary acts (Austin 1962), that is, acts in which saying something constitutes doing something, as when saying Open the door! constitutes ordering the addressee to open

1342

the door. Let us in fact begin an ov erv iew of such syntactic constructions with a sketch of imperative sentences. 9.1. Imperatives English imperativ e sentences usually lack an ov ert subject, though an ov ert subject is often possible: (133a) Be quiet. (133b) Somebody help me! (133c) Yóu wash the díshes, and yóu sweep the flóor. [pointing to two different persons] Somebody and you in (133b—c) are subjects and not v ocativ es, since they differ both intonationally and syntactically from analogs with a vocative: (134a) Somebodyi, take off youri/*hisi/*theiri coat. (vocative) (134b) Somebodyi take off youri/hisi/theiri coat. (subject) In (134a), only a second-person pronoun can hav e the (understood) subject as antecedent, while in (134b), a pronoun with the (ov ert) subject as antecedent can be either third-person (reflecting the form of the subject) or second-person (reflecting its ov erlap in reference with the addressee). In negativ e imperativ es, an auxiliary v erb do is required, bearing the negation and preceding any ov ert subject, except that when the negation is incorporated into the subject, do is not used: (135a) Don’t open the door yet! (135b) Don’t any of you leave the room! (135c) Nobody (*do) leave the room! Do can also appear in non-negativ e imperativ es that hav e the force of requests or pleas rather than orders; unlike the do that appears in constructions with inv ersion, this do can cooccur with auxiliary verbs: (136) Do open the door. (137a) Do be a bit nicer to your brother than you have been. (137b) Don’t be sleeping when they arrive. (137b′) Don’t have finished the job yet when we come back. For imperativ es with an (ov ert or understood) first-person plural subject, let’s (contracted except in v ery formal speech) is used, with both let’s not and don’t let’s used for corresponding negative imperatives:

XXI. Syntaktische Skizzen

(138a) Let’s go home. (138a′) Let’s you and me go to the movies tonight. (138b) Let’s not leave yet. (138b′) Don’t let’s forget about Bert. Imperativ es that serv e as requests or pleas can be followed by a “tag” (usually inv olv ing will, though would, can, and could are also possible); shall we (though not *shan’t we) is used in tagged let’s imperatives: (139a) Open the door, won’t/will you? (139b) Let’s go to the movies tonight, shall we? (139b′) Let’s not leave yet, shall we? The use of shall in the tags of first-person plural imperativ es mirrors a class of formally interrogativ e sentences that can plausibly be regarded as interrogativ es of imperativ es, since they elicit as answers imperativ es or their functional equivalents: (140a) Shall I open the window? Yes, please open it. (140b) What shall I do with this string? (Why don’t you) throw it out. These tags cannot be identified with the tags of tag questions, since won’t you and will you differ in interpretation in ways that div erge greatly from the difference in interpretation between a “rev ersal tag” and a “reduplicativ e tag”: when a speaker adds won’t you to an imperativ e, he is pleading, and when he adds will you, he is insisting, whereas a rev ersal tag (as in You”ll be at the party, won’t you?) presents the speaker as hav ing knowledge that merely requires confirmation and a rev ersal tag (as in You’ll be at the party, will you?) takes the addressee’s word for what has just been said. Imperativ e constructions can be blended with the pseudo-coordinate conditional construction of (141) in sentences (142) that simultaneously express conditional propositions and give orders or make requests: (141) Even mention Cuba and Harry starts talking like a lunatic. (= If you even mention Cuba ...) (142a) Open the door and I’ll give you a dollar. (142b) Don’t open the door, or I’ll kill you. Such sentences manifest the full range of special characteristics of imperative sentences: (143a) Open the door, will you, and I’ll give you a dollar.

74.  English

(143b) You be quiet, you little bastard, or I’ll kick you out. (143c) Please wait just a few minutes and Mr. Morton will be happy to see you. Imperativ es are reported using (for) ... to complements in which the only ov ert signs of imperativ eness are the optional use of please and the limited possibility of an embedded pseudo-coordinate imperative: (144a) Fred screamed for someone to please help him. (144b) Sam told us [to go home or he’d call the police]. 9.2. Interrogatives Much of the syntax of yes-no and Wh-interrogativ es was cov ered in the discussion of word order in 2. It is possible for more than one interrogativ e expression to figure in an interrogativ e clause, and a Wh-question that is embedded in a Wh-question can be ambiguous as to which interrogativ e expression goes with which question, as in (145a), which can elicit answers corresponding to either an interpretation in which which books is part of the embedded question (145b), or one in which it is part of the main question (145b′): (145a) Which students wanted to know which professors had assigned which books? (145b) Ted and Lucy wanted to know which professors had assigned which books. (145b′) Ted wanted to know which professors had assigned Syntactic Structures and Barriers, and Lucy wanted to know which professors had assigned Growing up Absurd and Living my Life. A question in which which professors goes with the main question and which books with the embedded question requires a different word order: (146) Which students wanted to know which books which professors had assigned? Ted wanted to know which books Smith and Morton had assigned, and Lucy wanted to know which books Gonzalez and Tucker had assigned. The reason for this correlation between word order and interpretation is that each (main or dependent) Wh-question requires Wh-mov ement of one interrogativ e expression associated with that question; thus, if which books

1343

is the only interrogativ e expression associated with the dependent question, it must be the one that is mov ed into the Comp position of the lower question. There often is an option as to which of the interrogativ e expressions associated with a giv en question is mov ed into its Comp position, and (as Kuno (1982) has noted) the one that is mov ed corresponds to the information around which the answer is organized: (147a) Which professors assigned which books? McCawley assigned Against Method and Analytic Syntax, Sadock assigned Barriers and Sapir’s Language, ... (147b) Which books did which professors assign? Analytic Syntax, McCawley and Aronson assigned, Sapir’s Language, Sadock and Dahlstrom assigned, ... In relev ant respects (148) is identical to (147b): (148) ??What did who say? Its lowered acceptability does not reflect syntactic ill-formedness but a pragmatic defect, namely the lack of any plausible organization of the answer around what: there rarely is any predetermined set of things that might hav e been said and are to be associated with speakers. Alternative questions take either a form looking like a conjunction of yes-no questions, with each conjunct giv ing one of the alternativ es, or a form looking like a Whquestion, but with a list of answers added: (149a) Was John able to catch the 2:00 plane, or will he be coming on the 3:00 plane? (149b) Which plane did John take, the 2:00 one or the 3:00 one? Yes-no questions, Wh-questions, and both v arieties of alternativ e questions hav e dependent forms, without inv ersion, and in the case of yes-no and the first kind of alternativ e question, with initial whether or if: (150a) Mary asked whether/if John was able to catch the 2:00 plane. (150b) Mary asked which books which professors had assigned. (150c) Mary asked whether John was able to catch the 2:00 plane or whether he would be on the 3:00 plane. (150d) Mary asked which plane John was coming on, the 2:00 one or the 3:00 one.

1344

In a rev ersal tag question, a declarativ e S is followed by a “tag” consisting of an auxiliary v erb matching that of the host (negativ e if the host S is affirmativ e, and v ice v ersa), followed by a NP (usually but not always a pronoun) matching the subject of the host: (151a) You’re happy, aren’t you? (151a′) You aren’t happy, are you? (151b) There hasn’t been any progress, has there? (151c) Fred’s been behaving peculiarly recently, hasn’t the bastard? The tag is arguably a reduced form of a full S, since its v erb agrees with its own subject, not with the subject of the host S (as can be seen from (152a), where the subject of the tag differs in number from its counterpart in the host S), and when the host has multiple auxiliary v erbs, the tag can hav e additional auxiliary verbs after the subject: (152a) IBM has laid off a lot of people recently, haven’t/*hasn’t they? (152b) John won’t have been released from jail yet, will he (have (been))? Unlike yes-no, Wh-, and alternativ e questions, tag questions do not hav e any dependent form: (153a) John is coming on the 2:00 plane, isn’t he? (153b) *Mary asked (whether) John is coming on the 2:00 plane, isn’t he. (153b′) *Mary asked (whether) John is coming on the 2:00 plane, he isn’t. There is a limited class of cases in which a tag is attached to (and takes its form from) a subordinate S, but such sentences are not reports of tag questions: (154a) I imagine you’re angry, aren’t you? (154b) I don’t suppose you’ve seen the flashlight, have you? Reduplicativ e tag “questions” such as He’s here, is he? and echo questions, neither of which is strictly speaking interrogativ e, are discussed in 9.3. 9.3. Echoes There are sev eral types of sentences, here grouped together as “echoes”, that are required to partially mimic something that has just been said:

XXI. Syntaktische Skizzen

(155a) John: There’s a fly in my soup. Mary: There’s a fly in your soup?? (pure echo) (155b) John: We need some more vinegar. Mary: We need some more what?? (wh-echo) (155c) John: Fred’s still in the hospital. Mary: He’s still in the hospital, is he? Then I guess he won’t be able to come along on the picnic. (reduplicative tag) (155d) John: Ted plans to quit his job. Mary: Ted quit his job?? Don’t be ridiculous! (incredulity response) (155e) John: There isn’t anything wrong with me. Mary: The hell there isn’t! (imprecative retort) The “stimulus” that is mimicked in pure and wh-echoes can be an utterance of any sort, not necessarily ev en a sentence; the stimulus for the other three kinds of echoes, however, must be a declarative sentence: (156a) Bob: Did I make enough pozole? Carol: Did you make enough what? (156b) Bob: Hooray for Andre Marrou! Carol: Hooray for who?? (157a) Ted: Who took my Perrier? Alice: *The hell who/someone did. (157b) Ted: Open the door. Alice: *Open the door, will I? Unlike rev ersal tags, which exhibit different degrees of reduction when there is more than one auxiliary v erb, reduplicativ e tags and imprecativ e retorts allow only the tensed auxiliary verb to appear in their surface form: (158a) So there have been a lot of problems, have there (*been)? (158b) In a pig’s eye there have (*been). Reduplicativ e tags and imprecativ e retorts are finite (i. e. they hav e to bear a tense, indeed, a tensed auxiliary v erb), whereas incredulity responses not only are nonfinite but do not allow any auxiliary verb: (159a) Linguistic Inquiry win a Pulitzer Prize?? (159a′) *Linguistic Inquiry have won a Pulitzer Prize?? The mimicry in all fiv e types of echo is not exact repetition of the words of the stimulus, since the choice of pronouns reflects the v antage point of the person uttering the echo, not the person that he is mimicking, e. g. in (156a), I is “repeated” as you.

74.  English

9.4. Exclamatives English has three forms for exclamativ e sentences, two of which ev idently reflect conv entionalization of exclamativ e uses of interrogative sentences: (160a) God, was I ever hungry! (inverted exclamative) (160b) How hungry we were! (Wh-exclamative) (160c) I was so hungry! (so-exclamative) Inv erted exclamativ es require two primary stresses, one on the subject and one on the focus of the construction: the expression denoting something whose degree or magnitude is presented as remarkable. The requirement of a stress on the subject ov errides the requirements that anaphoric and expletiv e elements be unstressed: (161a) Was hé ever tíred! (161b) Was it raining! (161c) Were thére a lot of próblems! The interrogativ e elements in the Wh-exclamativ e construction are limited to what and how, and ev en those elements hav e a different range of interpretations than in interrogativ e sentences — how is restricted to its degree sense, and what occupies not the determiner position but the position before the article where such occurs: (162a) *How Fred got away from them! (162a′) How did Fred get away from them? (162b) What I wouldn’t give to be in Mexico right now! (162c) What a nice guy Bill is! (162c′) *What a nice guy is Bill? (162c″) Bill is such a nice guy! (162d) *What book Bill bought! (162d′) What book did Bill buy? The possibilities for combining exclamativ es with other material are sharply limited. If- and when-clauses can modify all three kinds of exclamatives: (163a) When the Bears won, was Nancy ever surprised! (163b) If the Bears win, Nancy will be so surprised! (163c) When the Bears won, what a surprise it was to Nancy! There is a limited class of cases in which inv erted exclamativ es (and to some extent, also so-exclamativ es) can be embedded, and a non-ov erlapping class of cases that allow em-

1345

bedded Wh-exclamatives: (164a) I’d better leave now, because do I ever have a lot of work to do! (164a′) *I’d better leave now, because what a lot of work I have to do! (164b) I’ve just discovered that are we ever in for a hard time! (164b′) *I’ve just discovered that what a hard time we’re in for! (165a) It’s amazing what a jerk Ted is. (165a′) *It’s amazing that is Ted ever a jerk. 9.5. Nonrestrictive Clauses and Other Parentheticals Nonrestrictiv e relativ e clauses correspond to separate speech acts from those that one performs in uttering their host Ss. For example, in (166), the speaker informs or reminds the addressee that John was here a minute ago, while in the process of asking the addressee to find John and tell him to come back: (166) Find John, who was here a minute ago, and tell him to come back. Restrictiv e relativ e clauses, by contrast, figure in the same speech acts as their host N′s. While many languages (e. g. Japanese) do not draw a sharp distinction between restrictiv e and nonrestrictiv e relativ e clauses, English distinguishes between them phonologically (a nonrestrictiv e clause is always a separate intonational phrase, while a restrictiv e clause normally is not) and in many details of syntax, including the following. (i) Nonrestrictiv e relativ es require an ov ert relativ e expression, while restrictiv e relativ es hav e the option (discussed in 5.2.) of omitting a simple relativ e expression in fav or of that or of nothing at all. (ii) A restrictiv e relativ e can only hav e an N′ as its host, while nonrestrictiv e relativ es can hav e NPs (166), predicate phrases, or Ss as antecedents: (167a) Fred is very confident of himself, which I am not. (167b) Bill has been discharged from the hospital, which means he’ll be back at work soon. (iii) Restrictive clauses can be extraposed, but nonrestrictive clauses generally cannot: (168a) A man walked in who was dressed in black. (168b) ??Marcia has just arrived, who you wanted to meet. (iv ) Restrictiv e clauses can be stacked, but nonrestrictive clauses cannot:

1346

(169a) The man who works in her office who Mary is dating is a real jerk. (169b) *Sam, who works in Mary’s office, who she’s dating, is a real jerk. The latter two characteristics follow (as was noted in Emonds 1978) from an analysis in which a nonrestrictiv e relativ e and the constituent that it refers to do not make up a constituent but are simply required to be adjacent: neither an extraposed clause nor the second and subsequent items of a “stack” of nonrestrictiv e relativ es would be adjacent to the “antecedent”. The term “antecedent” is in fact appropriate, since the relationship between a nonrestrictiv e clause and the item that it refers to is anaphoric. Only if a constituent can be the antecedent of an anaphoric item in a following sentence can it be the constituent to which a nonrestrictive clause refers: (170a) The winner of the contest, who is the judge’s sister-in-law, sang dreadfully. (170a′) The winner of the contest sang dreadfully. She is the judge’s sisterin-law. (170b) *Each applicant, who survived the interview, was required to take a written examination. (170b′) Each applicant was required to take a written examination. *He survived the interview. Furthermore, the possibility of an anaphoric NP containing its own N is mirrored in the possibility of relativ e expressions containing that N in nonrestrictiv e (though not restrictive) relative clauses: (171a) Restrictive relatives can be extraposed. They share that property with than-clauses. (171a′) Restrictive relatives can be extraposed, which property they share with than-clauses. While nonrestrictiv e clauses often occur in the middles of their host Ss, they do not behav e like constituents of the host S. For example, when a nonrestrictiv e clause occurs in the middle of a V′, a zero V′ hav ing the latter V′ as antecedent is interpreted without regard to the nonrestrictiv e clause, e. g. the zero V′ in (172) means “ask Bill how he was going to get to work”, not “ask Bill, who wanted to sell her his car, how he was going to get to work”:

XXI. Syntaktische Skizzen

(172) Mary asked Bill, who wanted to sell her his car, how he was going to get to work, and Alice did Ø too. In addition, the embedding of Ss containing nonrestrictiv e clauses is not allowed, in the sense that a nonrestrictiv e clause whose antecedent is in a subordinate clause is interpreted as an interpolation by the speaker of the whole sentence, not as his report of another speaker’s interpolation: (173) Teresa says that Bill, who has just gotten out of jail, has threatened to kill her. These observ ations suggest an analysis in which the connection between nonrestrictiv e clauses and their host Ss is in the realm of discourse structure and the structure of complex actions, the condition for the formation of a nonrestrictiv e clause is an anaphoric relation between constituents of two sentences that represent separate speech acts, one of which is “incidental to” the other, and the formation of nonrestrictiv e clauses inv olv es mov ing the nonrestrictiv e clause to a position immediately after its antecedent, without making it a constituent of the host S. Nonrestrictiv e relativ es are one of a fairly div erse class of parenthetical constituents that are “interpolated” into a host S but do not strictly speaking become part of it. They include expressions that indicate the speaker’s estimate of the reliability of what he is saying (clearly, presumably), his reaction to information that he is con v eying (amazingly, strang ely enoug h, fortunately (for John)), remarks as to how the addressee should take what the speaker is saying (frankly, to put it bluntly, confidentially), or how the host S is supposed to fit into the discourse structure (in conclusion, on the other hand, incidentally, furthermore). One important class of parentheticals has the form of a truncated S, more specifically, a S that is missing a final thatcomplement: (174a) It probably can’t be fixed, I’m sorry to say. (174b) You’ve passed the exam, you’ll be happy to hear. In such examples, the host S functions simultaneously as main S and as the missing complement, e. g. in uttering (174a), one informs the addressee that it probably can’t be fixed and in that connection expresses one’s sorrow

75.  Italian

at hav ing to say that it probably can’t be fixed. Regarding the niches that parentheticals can occupy, see 2.4.

10. References Austin, J. L. 1962. How to do things with words. Oxford. Bach, Emmon. 1980. In defense of passiv e. Linguistics and Philosophy 3. 297—342. Boling er, Dwig ht. 1978. Yes-no questions are not alternativ e questions. In H. Hiz (ed.), Questions (Dordrecht), 87—105. Carden, Guy. 1986. Blocked forward coreference: implications of the acquisition data. In B. Lust (ed.), Studies in the Acquisition of Anaphora, v ol. 1 (Dordrecht), 319—57. Chafe, Wallace. 1976. Giv enness, contrastiv eness, definiteness, subjects, topics, and point of v iew. In C. Li (ed.), Subject and Topic (New York), 25—55. Channon, Robert. 1980. Anaphoric that: a friend in need. Pronouns and anaphora (Chicago), 98—109. Emonds, Joseph. 1979. Appositiv e clauses hav e no properties. Linguistic Inquiry 10. 211—43. Hankamer, Jorg e, and Ivan Sag . 1976. Deep and surface anaphora. Linguistic Inquiry 7. 391—428. Horn, Laurence R. 1979. Remarks on Neg-raising. In P. Cole (ed.), Pragmatics (Syntax and Semantics 9, New York), 129—220. —. 1989. A Natural History of Negation. Chicago. Jackendoff, Ray S. 1977. X-bar Syntax. Cambridge, MA. Jespersen, Otto. 1924. The Philosophy of Grammar. London. Reprinted 1992, Chicago. Keenan, Edward. 1988. Complex anaphors and bind alpha. CLS 24.1. 216—32.

1347

Kuno, Susumu. 1971. The position of locativ es in existential sentences. Linguistic Inquiry 2. 333—78. —. 1982. The focus of the question and the focus of the answer. In R. Schneider et al. (eds.), Nonde-. claratives (Chicago), 134—57. —. 1987. Functional Syntax. Chicago. Lakoff, Georg e. 1968. Pronouns and reference. Bloomington. Reprinted in McCawley (ed.), Notes from the Linguistic Underground (Syntax and Semantics 7, New York), 273—335. Lakoff, Robin. 1969. A syntactic argument for negative transportation. CLS 5. 140—49. McCawley, James D. 1971. Tense and time reference in English. In C. Fillmore and D. T. Langendoen (eds.) Studies in Linguistic Semantics (New York), 96—113. —. 1988. The Syntactic Phenomena of English, 2 vols. Chicago. —. 1992. Justifying part-of-speech assignments in Mandarin Chinese. Journal of Chinese Linguistics 20. 211—46. Morg an, J. L. 1972. Verb agreement as a rule of English. CLS 8. 278—86. Nanni, Deborah. 1980. On the surface syntax of constructions with easy-type adjectiv es. Language 56. 568—81. Partee, Barbara. 1989. Binding implicit v ariables in quantified contexts. CLS 25. 342—65. Russell, Bertrand. 1905. On referring. Mind n. s. 14. 479—93. Thomason, Richmond, and Robert C. Stalnaker. 1973. A semantic theory of adv erbs. Linguistic Inquiry 4. 195—220. Williams, Edwin. 1974. Rule ordering in syntax. M.I.T. Ph. D. thesis.

James D. McCawley, Chicago (USA)

75. Italian 1. 2. 3. 4. 5.

1.

The State of The Art Sentential Phenomena Noun Phrases Marked Word Orders References

The State of The Art

Italian does not hav e a tradition of detailed and exhaustiv e description on which theoretical research could base its inv estigation. Only in the last decade, important works of description hav e been entertained, such as the

one directed by Lorenzo Renzi (et al.) (1988, 1992, 1994). (For a rev iew of other Italian grammars recently appeared cf. Radtke 1991). The major part of work on Italian formal linguistics is done in the framework of Generativ e Grammar, as Graffi (1990) remarks, to which Italian has prov ided challenging fieldwork as well as important refinements (cf. also Kayne 1991). This will also be the theoretical background of this surv ey. — It is well-known that Italian has a relativ ely free word order. The challenge for the (generativ e) linguist is to giv e a principled account

XXI. Syntaktische Skizzen

1348

of all the grammatical word orders, while excluding all the ungrammatical ones. Here I will discuss some of the work dev eloped in this trend of research.

2.

Sentential Phenomena

2.1.  The main predicate classes. Well-known properties of Italian that help building tests to differentiate the basic structure of different v erb classes include, a. o., auxiliary selection, past participle agreement with the subject, ne-extraction, empty pro in subject position. For reasons of space, copular structures will not be dealt with here. (But cf. Longobardi 1985, Salvi 1992, Moro 1993). 2.1.1.  Verbs assigning no θ-role such as raising verbs like sembrare in (1 a), or a quasi-argumental θ-role such as weather verbs like piovere in (1b), allow no lexically realized expleti e subject:

2.1.2.  Verbs assigning a single θ-role can be div ided into intransitive, which assign an Agent θ-role, and erg ative which assign a Theme θ-role. This is reflected in their syntactic behav iour. Perlmutter (1978), Burzio (1981, 1986, 20 ff) and much subsequent literature hav e deriv ed the differences between ergativ e and intransitiv e v erbs from a single one, namely the D-structure position of the argument which receiv es the unique θ-role assigned by the v erb: ergativ e v erbs assign their θ-role to the internal argument position, while intransitiv e (unergativ e) v erbs assign their θ-role to the external argument position. This simple hypothesis captures the many similarities between ergativ e subjects, passive subjects and direct objects. 2.1.2.1.  Morphological means to signal ergativ ity. Some v erbs, such as affondare (“sink”) in (5), display no morphological difference between the transitiv e and the ergativ e form; others, such as rompere (“break”) in (6), signal ergativ ization by means of insertion of a reflexiv e clitic (cf. 2.4.); others still, such as uscire (“go out”) in (7), are intrinsically ergav ti e with no transiti v e counterpart:

The impossibility of hav ing a phonologically realized expletiv e or quasi-argument subject is related to the possibility of hav ing an empty pronoun in subject position (so-called pro-drop parameter cf. Rizzi (1982, ch. 4) and 2.3. below). — Verbs assigning no θ-role at all take only essere, while weather v erbs, which assign a quasi θ-role, can select both essere and avere as their auxiliary: Notice that in Southern Italian, verbs that express change of location are of the same class as those in (5)—(6), as exemplified in (8):

As argued for by Benincà/Cinque (1992), if weather v erbs hav e a lexical subject, they behav e like ergativ es (cf. 2.1.3.) in requiring essere as an auxiliary as in (3), in displaying past participle agreement as in (3)—(4), and in allowing ne-extraction as in (4):

This is also found in child language, (cf. Antelmi 1992, 427).

75.  Italian

2.1.2.2.  Auxiliary selection. Intransitiv es as in (9), on a par with transitiv es as in (5a), take avere in compound tenses. Ergativ es as in (5b), (6b), on a par with passiv es as in (10) take essere:

1349

ated as a sister of the VP headed by the past participle, as in (11):

This analysis is extended by Kayne to socalled absolute small clauses such as conosciuta me (lit.: met me, “once they met me”), or arrivata Maria (lit.: arriv ed Mary, “once Mary had arriv ed”) originally studied by Belletti (1981). His approach is further dev eloped by Belletti (1990, ch. 2) in a more recent framework where functional projections are very generally assumed (cf. 2.2.).

Burzio (1986, 55) suggests that essere assignment is triggered by the binding relation holding between the subject and the internal object position in passiv e and ergativ e predicates. His proposal, howev er, does not predict the v ariation that is found among languages with respect to this phenomenon. Kayne’s (1993a) alternativ e account is based on a wide range of Romance languages, among which many Italian dialects. First of all he unifies possessiv e avere with auxiliary avere and analyses both as essere plus an incorporated (null) preposition/determiner, which is the head of the predicate selected by essere. In the case of auxiliary avere, the PP/ DP complement embeds the VP, or better still the AgrOP, which triggers past participle agreement with the object under certain circumstances (cf. 2.1.2.3.).

2.1.2.4. Ne-extraction. Partitiv e and genitiv al ne can only be extracted from postv erbal ergativ e subjects (12a) and direct objects (12b). Ne-extraction from prev erbal ergativ e subjects (13a), prev erbal or postv erbal (in)transiv ti e subjects (13b—c) is ruled out:

2.1.2.5.  V-S-Complement order. A postv erbal 2.1.2.3.  Past participle agreement. Essere asergativ e subject (cf. 2.3.) can be followed by signment goes together with past participle other nuclear arguments such as the sentenagreement with the subject: Ergativ e subjects tial complement in (14a). This is not possible as in (5b), (6b), (7) agree with a past particifor the postv erbal subject of an intransitiv e, ple, when this is present, on a par with pasas in (14b). In this respect, ergativ e subjects siv e subjects (10a—b) and clitic objects (10c). behav e on a par with objects and postv erbal On the contrary, the subject of intransitiv es v passi e subjects (14c—d): (9a) and transitiv es (5a), (6a), does not. Burzio (1986, 55), relates this phenomenon to essere assignment and analyses it in terms of the binding relation that holds between the direct object and the clitic. Kayne (1989a) analyses past participle agreement with clitic objects (and ergativ e subjects) in Italian as the result of mov ement of the object to (or through) the Specifier of an Agr head gener-

1350

Cf. Belletti (1988, 7 ff) on the definite restriction that appears to interv ene on postv erbal subjects. — All the facts in (6)—(14) strongly support a theory that structurally distinguishes the argument projection of two classes of v erbs assigning one θ-role:

XXI. Syntaktische Skizzen

The following are some of the tests presented by Belletti/Rizzi to distinguish the temere class (parallel to transitiv e v erbs) from the preoccupare class, and the latter from the piacere class (which displays the same structure as the preoccupare class, but different caseassigning properties). 2.1.3.1  Reflexiv e anaphors. Rizzi’s (1986a) theory of chain formation prov ides a test to distinguish a subject base generated in the external argument position from a subject deriv ed from the internal argument position:

On ergative nouns cf. Giorgi (1991, 46 ff). On ergative adjectives cf. Cinque (1990b). 2.1.3.  Verbs assigning two θ-roles can be distinguished into transitive and psychological ones. The former assign an Agent and a Theme θ-role. The latter assign an Experiencer and a Theme θ-role. As before with intransitiv es, the Agent θ-role of transitiv e v erbs is assigned to the external argument position. According to Belletti/Rizzi (1988), Italian displays three types of psych-v erbs: the temere class, the preoccupare class, and the piacere class. The temere class has the same configuration as transitiv e v erbs in that it assigns the experiencer θ-role to the external subject position and the Theme to the internal object position. The preoccupare and the piacere class hav e the Experiencer θ-role in a position internal to VP, ev en if higher than the Theme (a sort of internal secondary object, cf. 2.1.3.3. below):

The structures proposed by Belletti/Rizzi (op. cit.) are in (18):

In (19a) the subject is generated in a position higher than the reflexiv e, the anaphoric reading (which is the only one av ailable to si in this case) is allowed and the sentence is ruled in. In (19b) the subject of the passiv e construction is deriv ed from the internal object position, which is lower than the anaphor. The structure is ruled out because the reflexiv e enters the mov ement chain of the subject, giv ing rise to an illegal chain with two θ-roles and two cases. — A reflexiv e anaphor can refer to the subject of the temere class but not of the preoccupare class:

(20b) is acceptable in an inchoativ e reading with preoccupare and interessare embedding an empty prepositional object with generic interpretation (cf. Gianni si preoccupa/interessa di questo, “G. worries/is interested about this”). On the anaphoric/pronominal system of Italian, which will not be dealt with here, cf. Burzio (1989), and Giorgi (1990). 2.1.3.2  Arbitrary empty subjects. For independent reasons, arbitrary interpretation can be assigned only to base generated subjects in Italian. Psych-v erbs of the temere class pattern with transitiv e v erbs in allowing an

75.  Italian

arbitrary subject, while psych-v erbs of the preoccupare class do not:

1351

2.1.3.4  The piacere class (27) has three main properties that are not shared by the preoccupare class (28): a) the NP bearing the Experiencer-role is marked with dativ e case; b) the aspectual auxiliary is essere; c) both orders ‘Experiencer V Theme’ and ‘Theme V Experiencer’ are possible.

2.1.3.3 Ne-extraction. The Theme of the temere class behav es on a par with the Theme of transitiv e and ergativ e predicates in allowing ne-extraction (22); the Experiencer of the preoccupare class does not (23):

The contrast between (22b) and (23b) shows that the Experiencer of the preoccupare class does not reside in the internal object position, where the deriv ed subject is base generated. The object Experiencer of preoccupare is not a configurational object, but a sort of secondary object. As a consequence, it does not allow extraction, contrary to the structural object of the temere class:

Belletti/Rizzi (op. cit.) propose that the two classes hav e the same basic structure (with a subject deriv ed from the internal object position), the difference between them being that the former assigns (inherent) accusativ e and the latter assigns dativ e to the Experiencer in the external object position. Inherent dativ e Case assignment to the external object not only allows the piacere class to behav e like an ergativ e v erb with respect to essere selection, but also the dativ e Experiencer, introduced by the preposition a, to be (optionally) mov ed to the empty subject position, as in (29b). That this is a subject position and not an extrasentential position is shown by the contrast with (29a):

The contrast in (29) shows that the dative experiencer in (29b) behav es more like the subject in (29c) than like the dativ e Goal in (29a). 2.2.  V°-mov ement. Emonds (1976) and, more recently, Pollock (1989) hav e deriv ed the dif-

1352

ference between English and French in the position of (finite and infinitiv e) v erbs with respect to other elements of the clause such as negativ e particles, adv erbials, floating quantifiers, from an independent mov ement of the v erb. Belletti (1990) dev otes a whole monograph on the study of this problem in Italian. Giorgi/Pianesi (1992), Cinque (1994) further inv estigate the phenomena relev ant to v erb mov ement to functional projections and the position of adv erbs. All of them assume and support what can be considered as the null hypothesis, namely that adv erbs are in fixed A′-positions (possibly the Specs of functional projections), v ariation among languages being reduced to the lev el of V-mov ement at S-structure. The proposals v ary considerably w. r. t. the number and typology of these functional projections, a problem which I will not deal with here. For the position of the v arious kinds of adv erbs in the Italian sentence structure cf. also Lonzi (1992). Abstracting away from independent differences, such as do support in English and the presence of the negativ e clitic ne in French, a strict parallelism is drawn between the negativ e morphemes not and pas which are taken to be in the same structural position. Their different location with respect to the v erb is analysed as the result of application or lack of application of v erb mov ement. In French, but not in English, the tensed v erb must mov e to a higher head in order to incorporate inflection. That this is a property of the inflectional morpheme is shown by the fact that mov ement does not apply to lexical v erbs in French untensed clauses:

In Italian, the verb tends to move higher than in French. This is most apparent in untensed clauses where mica must follow the v erb, and in compound tenses, where mica may appear after the past participle, while pas must precede it. (For the parallel status of mica and pas cf. Zanuttini 1992):

The position of certain adv erbs in tensed clauses may appear inconsistent with the afore mentioned V-mov ement in Italian:

XXI. Syntaktische Skizzen

Untensed clauses, howev er, behav e as expected:

The unexpected (33a) does not display lack of V-mov ement, but an instance of left-dislocation of the adv erb (cf. Belletti 1990, 42) as well as of the subject (cf. 4.2.). 2.3.  Null and postv erbal subjects. Well-established research, tracing back to Perlmutter (1978), Taraldsen (1978), Rizzi (1982), Burzio (1986, 85 ff), relates the possibility of postv erbal subjects in Italian (and in other Romance languages except French) to the possibility of a null-subject (pro-drop parameter). In other words, a structure like (36a) is related to one like (36b):

The correlation is drawn along the following lines: Giv en the possibility of null personal pronouns as in (36b), it is natural to assume a null expletiv e pronoun corresponding to there in English or il in French, in the parallel Italian structure:

If this assumption is correct, then the correlation between (36a) and (36b) must also hold, as argued by Burzio (1986, 85). Much consensus is found in the literature in relating the null subject property to the richness in morphological inflection. More specifically, Rizzi (1982, ch. 4) and Burzio (1986, ch. 2), following Taraldsen (1978), assume that the inflectional morphology on the finite v erb may function as a subject clitic and produce an empty category in subject position:

A stipulative difference between verbal inflection and an object clitic is that the former

75.  Italian

only optionally triggers an empty subject, while the latter forces an empty object:

The stipulation also holds for “real” subject clitics in the dialects studied by Brandi/Cordin (1981, 1989), Renzi/Vanelli (1982), Rizzi (1986b), Benincà (1994), Poletto (1993, ch. 2). — The post-v erbal subject constructions differ w. r. t. the class of v erb, as noted in 1.1. In particular, with ergativ e v erbs the subject can be in the basic position embedded into V′, while with unergativ e v erbs, according to the literature, the subject is mov ed from the prev erbal subject position and right adjoined to VP (Burzio 1986, 85 ff), or directly base generated in that position (Rizzi 1990):

Rizzi (1990, 76) proposes that VP-adjoined subjects are properly gov erned by the v erb in I°, hence a trace is well-formed in this position. This deriv es the fact that long extraction of the subject in Italian is allowed from the postv erbal position, contrary to English (as can be seen from the glosses), in which the subject mov es from the ungov erned prev erbal position giv ing rise to an ECP v iolation:

1353

Rizzi (1978), Belletti (1982), Burzio (1986), Manzini (1983). Cinque (1988a) is the most recent study in this tradition and is the one that will be reported on here. — It is descriptiv ely possible to distinguish sev eral constructions v v in ol ing si:

Impersonal passiv es are not allowed in Italian differently from e. g. French:

It is generally noticed that in tensed clauses, impersonal si can occur with all major v erb classes, while in control untensed complement clauses they are uniformly excluded:

The ungrammaticality of (46b) is standardly attributed to a v iolation of the Case Filter (cf. Cinque 1988a, 523, and the references quoted there). On a par with all clitics, si must be in a CHAIN with Case. This is not possible in control clauses, in which si is in a CHAIN with the subject PRO, which receiv es no case. This hypothesis predicts that impersonal si should be possible in infinitiv al Rizzi’s proposal, as it stands, is incompatible contexts in which a Case is assigned to the with Kayne’s (1993b) antisymmetric apsubject, such as infinitiv al complements to proach which rules out any right-adjunction raising v erbs and in Aux-to-COMP construcin UG. In current terms, it could be restated tions (cf. 2.5.). The prediction, howev er, is by assuming that the post-v erbal subject is in only fulfilled by transitiv e (47) and unergaa left branch that is crossed ov er by headv ti e predicates (48): movement of the verb (cf. 2.2.). 2.4. Si-constructions. The inv ariable reflexiv e clitic si appears in quite a wide range of constructions in Italian, apparently v ery different from each other. The literature in the generativ e framework has obv iously attempted a unification of (some of) them, with alternating results. Works on this subject include

1354

XXI. Syntaktische Skizzen

For criticism and an alternativ e approach see Dobrov ie-Sorin (1993, 175 ff) who bases her analysis on comparativ e ev idence between French and Romanian (among other languages). She maintains an ambiguous analysis of Italian si dorme, but claims that the ambiguity is between an impersonal (nominativ e) construction, parallel to the French on dort, and a middle-se construction, which she attributes to Romanian se doarme. She claims, in fact, that Romanian does not present any counterpart to the si corresponding to French on, and in French intransitiv es are illicit with the impersonal-middle si.

(50a) is out only in the relev ant reading of impersonal si. But it is not in the “inchoativ e” reading also mentioned for (20b), with the gloss “only one parent seems to be worrying”.

Cinque (1988a, sect. 2.) proposes that impersonal si is an argument that must be linked to an external θ-role. This accounts for the occurrence of impersonal si in all kinds of clauses and only with those predicates that assign an external θ-role. The occurrence of impersonal si in tensed clauses with predicates that do not assign an external θ-role, as in (46a), is treated as an instance of a [-arg] si. This amounts to saying that the si construction with transitiv e and unergativ e predicates in finite clauses is ambiguous between a [+arg] and a [-arg] interpretation. This ambiguity can be checked in the semantic interpretation of the arbitrary pro-subject in Chain with si, which can be either quasi-existential or quasi-universal. In fact, if the latter is excluded by independent syntactic means, the former is av ailable only with transitiv e and unergativ e predicates, as expected. For instance, specific time reference, is incompatible with quasi-uni v ersal arbitrary interpretation (cf. Cinque 1988a: sect. 3.); as a consequence, (51c—d) result pragmatically v de iant:

2.5.  Untensed Clauses. Italian control and raising complements do not display substantial differences to their counterparts in other well-studied languages. For this reason, these constructions will not be discussed here, cf. Manzini (1979, 1983) for a general theory. — Depending on the selectional properties of the embedding predicate, some infinitiv al constructions giv e rise to “clitic climbing”, as in (52), and switch their auxiliary selection (53) from avere to essere:

These phenomena are analysed by Rizzi (1978, 1982, ch. 1) and Manzini (1983) in terms of “restructuring”, Burzio (1986, 322 ff) in terms of VP-preposing. Kayne (1989b) focusses on the contrast between Italian and French w. r. t. the presence v s. absence of clitic climbing, reducing it to the “strength” of I(NFL) in Italian (which is also responsible for the possibility of a null subject, cf. 2.3.), and its “weakness” in French. This contrast appears to hold in infinitiv al clauses as well, where V mov es out of the VP crossing ov er adv erbials in Italian but not in French, cf. 1.2. Additional ev idence of Vmov ement in infinitiv al clauses in Italian and not in French is prov ided by the different position of clitics on untensed v erb forms:

75.  Italian

According to Kayne (1989b, 241), who formulates his analysis in the Barriers framework, if VP is L-marked by a strong I, then V can mov e out of the VP without passing through I, to which the clitic is left adjoined. (For the definition of L-marking cf. Chomsky 1986, 24). If a negation is present in the lower clause, the clitic cannot climb:

This is expected if negation is a non-L-marking head interv ening in the deriv ation of clitic climbing and if clitic mov ement takes place stepwise. That the latter is actually the case is supported by the observ ation that clitic climbing is blocked by an interv ening C° (56a, c) and not by something in SpecCP (56b), and by the fact that clitics climb in clusters (57):

1355

3.

Noun Phrases

Recent studies in the framework of principles and parameters hav e extended to noun phrases certain properties attributed to clauses, namely the fact that they are introduced by a (series of) functional projections (cf. Brame 1982, Abney 1987), and that nouns can hav e arguments in the same fashion as v erbs (cf. Giorgi/Longobardi 1990, Grimshaw 1990, Bottari 1990). The problem of relativ e and pseudo-relativ e clauses will not be discussed here for reasons of space. For a discussion cf. Cinque (1981, 1988b), Graffi (1980), Guasti (1988). 3.1.  Determiners. Longobardi (1992), proposes that Reference and Quantification are properties of the D° position and that all arguments hav e a DP projection. Nominal phrases with no article in argument position are therefore taken as displaying an empty D. In Italian, this empty category undergoes the following restrictions: a) it is restricted to plural or mass nouns:

b) is subject to lexical gov ernment requirement like other empty heads; c) receiv es indefinite (ne v er generic) interpretation: If clitics cluster onto the embedded I and clitic climbing is the result of the incorporation of the embedded I onto the matrix I (through the embedded C°), climbing of one clitic leav ing another in the embedded clause is excluded. — According to Kayne (op. cit.), negation interv enes between I and VP, and blocks mov ement out of the VP because it does not L-mark VP itself. In a more complex approach to INFL-features such as the one rev iewed in 2.2. abov e, negation does not immediately dominate VP, (cf. Zanuttini 1992). — A consequence of this analysis is coindexation of the two I’s which brings about the possibility of clitic climbing with subject control predicates and its incompatibility with object control predicates. Kayne suggests that the switch in auxiliary, found in cases like (64), can also be reduced to I-to-Cto-I mov ement, but he does not get into the details of the analysis, just observ ing that this mov ement appears to be more restricted than clitic climbing.

Crucially, (60c) is correct only in the indefinite, not in the generic, interpretation, as shown in (60d) where the generic interpretation is forced by the predicate. — The determiner may also be missing with no mass/indefinite interpretation in predicativ e position: cf. Gianni è amico di Maria (“G. is friend of

1356

Mary”); in v ocativ es, as in: caro amico, (“dear friend”); in certain prepositional expressions, such as in abito lung o, (lit.: in long dress); in proper nouns, e. g. Gianni, Maria. (But il Gianni, la Maria are grammatical in Northern v arieties, as well as family names referring to women in the Standard language, cf. Pavarotti, la Callas.) Predicates and v ocativ es are taken to lack the DP-projection tout court in that they are not arguments; idiomatic expressions are treated as idiosyncratic, ev en if certain subregularities can be found. Proper nouns apparently contradict generalization a) in that they do not hav e a mass/ indefinite interpretation. Howev er, their interpretation is particular in that the referential operator (assumed to be D) is identical with the restriction. For this reason, Longobardi proposes that Italian proper nouns are mov ed from their basic N position to the D position, ev idence for this being prov ided by the following contrast:

Italian proper nouns, therefore, prov ide non-triv ial ev idence for N-mov ement to D and ultimately the assumption of a DP projection. — Giusti (1991, 1993) argues, on ev idence from Romance, Germanic, and Balkanic, against a unification of the categorial status of Quantifiers, Demonstrativ es, Possessiv es and Articles under the same node D(eterminer). In particular she argues that the only element in D° is the article, while demonstrativ es and possessiv es are in Spec positions, and quantifiers are div ided into two classes: one external to the highest nominal projection and the other in an adjectiv al position. Italian prov ides ev idence for some of these claims. Possessiv es are clearly not determiners a. o. for the following reasons: a) they cooccur with a determiner and always follow it (cf. le sue belle lettere lit.: the his beautiful letters); b) they are adjectiv es, as they agree with the noun, marginally allow reordering with other adjectiv es (cf. le belle tue lettere lit. the beautiful your letters), can appear in predicativ e position (cf. queste lettere sono tue, lit.: these letters are your). Since adjectiv es are maximal projections it must be concluded that possessiv es also are. Furthermore, since c) the possessiv e position is used as an escape hatch position for extraction of arguments out of the noun phrase (cf. 3.3.), and arguments clearly are maximal projec-

XXI. Syntaktische Skizzen

tions, it must be so concluded for possessiv es as well. Interestingly, Italian displays possessiv e constructions that are in all respects parallel to the general case in English, in that they appear to be in complementary distribution with an article: la madre / (*la) mia madre (“my mother”). For these cases Giusti (op. cit.) proposes that the possessiv e mov es to SpecDP, dispensing with article insertion (for reasons parallel to those underlying the so-called doubly-filled COMP Filter of Chomsky/Lasnik 1977). — The univ ersal quantifiers tutti, entrambi, in Italian always appear to cooccur with DP: tutti/entrambi *(i) rag azzi (cf. English “all (the) boys”). Therefore they cannot be in D. Nor can they be in SpecDP. In fact, they can remain in situ when the noun phrase is mo v ed:

Since mov ement only applies to maximal projections, it is straightforward to take the empty category in the basic object position in (62b) to be a DP (not a D′) and the quantifier as external to it. — Partitiv e quantifiers, such as alcuni (“some”), molti (“many”), and numerals, although not co-occurring with articles, also prov e to be external to DP, according to Cardinaletti/Giusti (1992, 130 ff) who claim that an indefinite quantifier heads an autonomous QP-projection and selects two different complements: an indefinite partitiv e noun phrase and a definite partitiv e PP introduced by di, as represented in (63): (63) [QP[Q′[Q′[Q°molti] [DP [NPragazzi]]] [PPdi quelli che conosco]]] many boys of the boys I know Ev idence for the complement status of both is that none of them can appear when an indefinite quantifier is not present. The impossibility of a prepositional partitiv e with di contrasts clearly with the acceptability of a prepositional partitiv e with tra, which is a sort of adverbial adjoined PP:

75.  Italian

Less self-ev ident is the different status of the unarticulated noun in questi rag azzi/ i rag azzi and in molti rag azzi. Based on the fact that the latter, but not the former can be autonomously mov ed by cliticization, Giusti (1991, 450) proposes that the latter is a complete nominal complex, while the former is not, as supported by the following contrasts: (65a) Ne conosco molti. (65b) Li conosco tutti. (65c) *Ne/li conosco questi/quelli/i. [I] NE/LI know many/all/these/those/ the. Ev idence for the double selection of indefinite quantifiers is that the two partitiv e noun phrases can cooccur, as shown in (63). 3.2.  N°-mov ement (in common nouns). Cinque (1990c, 1993a), proposes to deriv e the postnominal position of certain attributiv e adjectiv es in Italian with leftward mov ement of the head noun. In so doing he generalizes to Italian the basic structure of noun phrases assumed for English (with all adjectiv es in left branching nodes), capturing the common linear hierarchy, and the fact that complements of the noun follow all attributiv e adjectives: (66a) Event noun phrases: (possessive) > quantificational > numeral > speakeroriented > manner > subject > object: e. g.: I suoi ultimi tre evidenti tentativii falliti ti di recupero His last three evident attempts unsuccessful of retrieval (66b) Non-event noun phrases: possessive > quantificational > numeral > affective > evaluating > colour > complement(s) e. g. I suoi ultimi tre grandi pannelli rossi di tela His last three big panels red of cloth At least the step that moves N across the thematic adjectiv e is obligatory. — According to Giorgi/Longobardi (1991), the thematic adjectiv e receiv es an agentiv e θ-role, in Spec NP: (67a) *L’ [NP italiana [N′ [N° invasione] [PP dell’Albania]]] (67b) L’invasionei [NP italiana [N′ [N° ti] [PP dell’Albania]]] The Italian invasion of Albania As pointed out for Catalan by Picallo (1991, 291), the order N SUBJ OBJ also appears when the subject is a di-PP:

1357

In (68a) di Gianni is in Spec NP where it receiv es the agentiv e θ-role. (68b), on the contrary is a passiv e noun phrase, with del teorema di Pitag ora in subject position and da parte di Gianni as an agentiv e adjunct, as the different preposition shows. — The binding phenomena that arise with anaphors inside NPs, as discussed by Giorgi/Longobardi (1991, ch. 3, sect. 8), are straightforwardly captured in this framework: (69a) La riconciliazionei [NP[PPdi Gianni]j [N′ [N° ti] [PPcon se stesso]j] The reconciliation of Gianni with himself (69b) *La riconciliazionei [NP[PPdi se stesso]j [N′ [N°ti] [PPCon Gianni]j] (70a) La letterai [NP [PPdi Gianni]j [N′ [N°ti] [PP alla propriaj madre]] The letter of Gianni to his own mother (70b) *La letterai [NP[PPdella propriaj madre] [N′ [N°ti] [PPa Gianni]j] The fact that di Gianni must c-command the anaphors se stesso in (69) and propria in (70), could be captured only in a complex way if mov ement of the noun was not assumed. Giorgi/Longobardi take the rev erse order to be basic, assuming a right-branching subject: (71a) La [NP [N′ [N° riconciliazione] [PP con se stesso] [PP di Gianni]]] (71 b) La [NP [N′ [N° lettera] [PP alla propria madre] [PP di Gianni]]] There are two empirical reasons and a theoretical one to fav or the analysis of a leftbranching subject ov er Giorgi/Longobardi’s analysis: one is that it permits a perfect parallelism between PP-subjects and thematic adjectiv es, which cannot appear after the PPobject: (72) *L’invasione dell’Albania italiana (cf. (67)) The other is that the order NSO is the unmarked one in most cases. Finally, such an analysis is compatible with Kayne’s (1993b) anti-symmetric approach to syntactic structures which rules out right-branches. It can

1358

be concluded that N° must cross at least the SpecNP position where PP-subject and thematic adjectiv es are inserted (Cf. also Cinque 1993b). 3.3.  Extraction from NPs. Cinque (1980) notices the following restrictions on extraction from Noun Phrases and analyses them in the framework of the Extended Standard Theory. Only PPs introduced by di can be extracted: (73a) Una persona di cui apprezziamo la grande generosità (è Giorgio). A person of whom we appreciate the great generosity (is Giorgio). (73b) *Una persona da cui non sopportiamo la lontananza (è Giuseppe). A person from whom we don’t stand the separation (is Giuseppe). cf. Una persona da cui stare lontani A person from whom to be separated (73c) *Una persona a cui approviamo le critiche (è Gianni). A person to whom we approve the criticisms (is Gianni). cf. Una persona a cui muovere critiche A person to whom address criticisms and only if the PP introduced by di is the subject of the noun phrase: (74a) L’icona di cui è stato scoperto il furto (*del custode). The icon, of which has been discovered the theft of the custodian. (74b) Il custode di cui è stato scoperto il furto (dell’icona). The custodian of whom has been discovered the theft of the icon (74a) is possible only if the Agent is not present. If the Theme is extracted, the noun phrase is interpreted as “passiv e”; in other words, the theme is interpreted as its subject. The same restriction is found in possessiv ization:

(75c) is ambiguous between an agent and a theme interpretation of the possessiv e, in other words between an activ e and a passiv e interpretation of the head noun. In a Gov ernment and Binding framework, Longobardi (1991, 66 ff) formulates the possessiv i-

XXI. Syntaktische Skizzen

zation principle as (76) and argues for the obligatory step in SpecNP in the extraction out of an NP. (76) Possessivization Principle. The unique phrase allowed to appear as a possessive is the hierarchically highest genitive argument of an NP. (77a) Quel famoso collezionista il cui celebre ritratto di Lord Snowdon della Regina Elisabetta è un capolavoro di fotografia. That famous collector whose famous portrait of Lord Snowdon of Queen Elizabeth is a masterpiece of photography. (77b) Lord Snowdon, il cui celebre ritratto della regina (??di quel collezionista) etc. (77c) La regina Elisabetta, il cui celebre ritratto (*di Lord Snowdon di quel collezionista) etc. In a more recent DP-framework such as the one presented in 3.1. abov e, it can be assumed that the possessiv e mov es from SpecNP where it receiv es the Agent/Possessiv e θ-role to the highest SpecAgrP where it is gov erned by the article and receiv es Case at S-structure. From that position, it can mov e out of the DP only through SpecDP for reasons that hav e not been discussed in the literature but that can be reduced to some requirement of proper gov ernment on the trace in SpecAgrP. Empirical ev idence in fav or of this adjustment of Longobardi’s proposal is prov ided by the fact that demonstrativ es block extraction, while quantifiers giv e milder v iolations:

If demonstrativ es are in SpecDP (cf. Giusti (1993, 59 ff) on Romanian data), while Quantifiers are the head of a separate projection, it is predicted that the former block while the latter allow XP-extraction out of the noun phrase. 3.4.  Pronouns. Romance languages display two series of personal pronouns: tonic and

75.  Italian

clitic. Kayne (1975, 72 ff) giv es a series of properties which distinguish clitics from tonic pronouns and full noun phrases in French. Most of them can be reproduced in Italian: clitics cannot bear contrastiv e stress (79), they cannot be conjoined either with a tonic pronoun or with another clitic (80), they appear in a fixed order (81), they can nev er appear in isolation (82), they cannot be separated from the v erb (83), they cannot be modified (84): (79) Non li/*LI ho visti. Cf. Non ho visto LORO/ loro/ i ragazzi. [I] didn’t see them/the boys (80) *li e le ho viste. [I] saw CL-pl.masc. and CL-pl.fem. cf. Non ho visto loro e loro/ i ragazzi e le ragazze. [I] didn’t see them and them/ the boys and the girls. (81a) Te ne ho dati. (81b) *Ne te ho dati. [I] to-you of-them have given many (82) Chi hai visto? Loro/ *li. Who did you see? Them. (83) *li, pare, ha licenziati subito. cf. li ha (pare) licenziati (pare) subito (pare). [he] CL-pl.masc. has (it seems) fired (it seems) immediately (it seems). (84) li (*due) ho visti. cf. ho visto loro due. [I] saw the two of them. The two series hav e been attributed a different internal structure by Cardinaletti (1993). The dativ e 3rd person plural pronoun loro, on the contrary, strikingly differs from the other pronouns (cf. Cardinaletti 1991): On the one hand, it has all properties of clitics except that it is bisyllabic. Like clitics, it must be mov ed from its basic position (85), but it cannot reach the prev erbal position (cf. (85c) and (86)):

Cardinaletti deriv es all the differences with clitic and tonic pronouns by assuming that dativ e loro is not of a class of unstressed pro-

1359

nouns different from clitics, namely that it is a weak pronoun parallel to unstressed pronouns in Germanic. Cardinaletti/Starke (1993), claim that the tripartition suggested by Cardinaletti (1991) is v ery general across languages.

4.

Marked Word Orders

With syntactically marked word orders, I intend those that are not expected in the sentence structure proposed for this language by a linguistic theory, following the definition giv en by Benincà (1988, 115). The framework adopted up to this point prov ides for the possibility to depart from the basic structure by means of additional operations independently allowed by general principles. Italian displays a consistent amount of extrasentential displacements that are discussed in Benincà (op. cit.). Some of them hav e prov ided a fruitful empirical domain of inv estigation of general properties of UG, carried ov er around the ’80s by Cinque and collected in his (1990a) monograph, on which I will base the following presentation. 4.1.  Clitic Left Dislocation v s. Left-Dislocation. Here are some of the properties that Cinque (1990a, 57 ff) discusses to characterize Italian Clitic Left Dislocation construction from the Left Dislocation of the type found in English. CLLD can apply to any phrase type (87) while LD in English can apply only to Noun Phrases, as the ungrammaticality of the glosses shows: (87a) Al mare ci siamo già stati. To the sea, we have already been there. (87b) Gentile non lo è mai stato. Nice he never was it. (87c) Che bevi lo dicono tutti. That you drink everybody says so. CLLD can apply to any clause type, LD in English cannot: (88a) Non so proprio chi questo libro potrebbe recensirlo. I really don’t know who this book could review it. (88b) Da quando gli articoli li recensisce lui la rivista è molto più selettiva. Since when the articles he reviews them the journal is much more selective. CLLD displays connectiv ity, while it is wellknown that LD does not (cf. Cinque 1977,

XXI. Syntaktische Skizzen

1360

and references quoted there for further discussion): (89a) A lei/*se stessa, Maria dice che non ci pensiamo mai. Of her/herself Maria says that we never think. (89b) A *lei/se stessa, Maria non ci pensa. of her/herself Maria never thinks. CLLD is sensitive to islands, LD is not: (90a) *Carlo ti parlerò solo delle persone che lo amano. (90b) John, I’ll talk to you only about the people who like him. 4.2.  Clitic Left Dislocation v s. Topicalization. A clearcut partition must be drawn between topicalization and Clitic Left Dislocation (CLLD) despite some common properties among which the sensitiv ity to island and to connectiv ity. Apart from their different informational import and intonation, two major differences are found: CLLD has a clause internal resumptiv e clitic pronoun while Topicalization does not (cf. (91)). CLLD can inv olv e more than one constituent, Topicalization cannot (cf. (92)): (91a) GIANNI ho già visto (non Maria). (91b) Gianni, l’ho già visto. Gianni I have already met. (92a) Di vestiti, a me, Gianni, in quel negozio, non mi ce ne ha mai comprati. (92b) *A ME, GIANNI, IN QUEL NEGOZIO, non ha mai comprato vestiti. Clothes, to me, Gianni, in that shop, [he] not medat CLloc CLpart has ever bought. This is taken by Cinque (op. cit. 1990, 60 ff) as ev idence to argue that Topicalization but not CLLD is the result of Wh-mov ement, since Wh-mov ement in Italian is incompatible with a resumptiv e clitic and can apply only to one element per clause: (93a) Con chi (*ci) hai parlato. With whom did you to-him talk. (93b) *Chi di che cosa ha parlato. Who about what spoke. Italian does not ev en allow a Wh-element in situ in a Wh-clause, unlike English: (94a) *chi ha visto che cosa? (94b) who saw what? The lack of this constraction is analysed by Calabrese (1984) to be at the base of the lack of two focused elements in the same clause:

(95a) *GIANNI ha scritto una LETTERA (95b) JOHN wrote a LETTER As a consequence, connectiv ity and sensitiv ity to islands, which are properties shared both by topicalization and CLLD in Italian, cannot be taken as a diagnostic criterion for Wh-mov ement. — Topicalization is analysed as requiring an empty Operator coindexed with the emarginated element and binding a v ariable in the basic position. CLLD is the result of coindexing of the emarginated element with an empty position. This position is optionally bound by a clitic throughout, except in case of an NP where the clitic becomes obligatory. In fact, NPs are the only elements subject to binding theory, according to which they should qualify for one of the four combinations for the features (±pronominal, ±anaphor]. The only possibility would therefore be for the NP to qualify as a pro, but improperly, giv en that it would not be identified. Thus the need of clitic insertion. According to Cinque (op. cit.) the two strategies are always av ailable in a language, Wh-mov ement being the default choice. 4.3.  Right dislocation. There is not much literature on this construction which is briefly described by Benincà (1988, 145 ff). Theoretical literature most often hints at this construction in negativ e terms. In fact, sometimes it is important to exclude that a clause final element is right dislocated in order to correctly identify its position inside the clausal structure. A test to exclude right-dislocation is contrastive stress: (96a) L’ho vista ieri, Maria. (96b) *L’ho vista ieri MARIA, non Carla. [I] saw her yesterday, Maria/*MARIA not Carla. In some respect, the right periphery of the clause seems quite parallel to the left periphery in that the focused element is incompatible with a stressed pronoun. It is proposed by Calabrese (1984) that because the last portion of the clause is informationally new, a focused element should come last.

5.

References

Abney, Steven. 1987. The English Noun Phrase in Its Sentential Aspect. PhD. Diss. MIT. Antelmi Donella. 1992. L’ipotesi maturazionale nell’acquisizione del linguaggio. Indagine longitudi-

75.  Italian

nale su una bambina italiana. Tesi di dottorato. Padov a/Venezia. [in press as: La prima grammatica dell’Italiano. Bologna.] Belletti, Adriana. 1981. Frasi ridotte assolute. Riv ista di Grammatica Generativa 6. 3—33. —. 1982. “Morphological” Passiv e and Pro Drop: The Impersonal Construction in Italian. Journal of Linguistic Research 2.4. 1—33. —. 1988. The Case of Unaccusativ es. Linguistic Inquiry. 19. 1—34. —. 1990. Generalized Verb Movement. Torino. —, and Luig i Rizzi. 1981. The syntax of ne: Some theoretical implications. The Linguistic Rev iew 1. 117—154. —. 1988. Psych-v erbs and θ-theory. Natural Language and Linguistic Theory 6. 291—352. Benincà, Paola. 1988. L’ordine degli elementi nella frase e le costruzioni marcate. In Renzi ed. 115— 196. —. 1994. La v ariazione Sintattica. Dialetti Romanzi e Teoria della Grammatica. Bologna. —, and Gug lielmo Cinque. 1992. Sur l’ambiguïté structurale des v erbes météreologiques en italien. De la musique à la linguistique. Homage à Nicolas Ruv eret, ed. by Liliane Tasmowski, and Anne Zribi Hertz. 155—162. Communication & Cognition. Bottari, Piero. 1990. Liv elli di rappresentazione lessicale: complementazione nominale e complementazione frasale. Tesi di dottorato. Padov a/Venezia. Brame, Michael. 1982. The General Theory of Lexical Specification and the Non-existence of Coarse Categories. Linguistic Analysis 8.4. 321—325. Brandi, Luciana, and Patrizia Cordin. 1981. Dialetti e italiano: un confronto sul parametro del soggetto nullo. Riv ista di Grammatica Generativ a 6. 33—87. —. 1989. Two Italian Dialects and the Null Subject Parameter. Jaeggli and Safir eds. 111—142. Burzio, Luig i. 1981. Intransitiv e Verbs and Italian Auxiliaries. PhD. Diss. MIT. —. 1986. Italian Syntax. A Gov ernment and Binding Approach. Dordrecht. —. 1989. On the non-existence of disjoint reference principles. Riv ista di Grammatica Generativ a. 14. 3—27. Calabrese, Andrea. 1984. Multiple Questions and Focus in Italian. Sentential Complementation, ed. by Wim de Geest and T. Putseys. Dordrecht 67—76. Cardinaletti, Anna. 1991. On Pronoun Mov ement: The Italian Dative loro. Probus. 3. 127—153. —. 1993. On the internal structure of pronominal DPs. ms. [To appear in] Proceedings of the 8th GCS, ed. by K. Tarald Taraldsen, and Ov e Lorenz. The Linguistic Review.] —, and Giuliana Giusti. 1992. Partitiv e ne and the

1361

QP hypothesis. A Case study. In Fav a ed. 121— 141. —, and Michal Starke. 1993. On Dependent Pronouns and Pronoun Mo v ement. GLOW talk. Lund. April 5—7, 1993. Chomsky, Noam. 1986. Barriers. Cambridge, MA. —, and Howard Lasnik. 1977. Filters and Control. Linguistic Inquiry 8. 425—504. Cinque, Gug lielmo. 1977. The mov ement Nature of Left Dislocation. Linguistic Inquiry 8. 397—411. —. 1980. On Extraction from NP in Italian. Journal of Italian Linguistics 5. 47—99. —. 1981. On the Syntax of relativ e Clauses and Markedness. The Linguistic Review 1. 247—296. —. 1988a. On Si Constructions and the Theory of Arb-Linguistic Inquiry. 19. 521—581. —. 1988b. La frase relativa. In Renzi ed. 443—503. —. 1990a. Types of A′-Dependencies. Cambridge, MA. —. 1990b. Ergativ e Adjectiv es and the Lexicalist Hypothesis. Natural Language and Linguistic Theory. 8. 1—39. —. 1990c. Head-to-Head Mov ement in the Romance NP. Talk deliv ered at the XXI Linguistic Symposium on the Romance Languages. Ottawa. April 1990. —. 1993a. On the Ev idence for Partial N-Mov ement in the Romance DP. Univ ersity of Venice Working Papers in Linguistics 3.2. ed. by Giuliana Giusti and Laura Brugè. —. 1993b. Posizione del Soggetto e Assegnazione di genitiv o nel DP italiano. Omaggio a F. Dimitrescu e A. Niculescu, ed. by Lorenzo Renzi and Coman Lupu. In press. Padova. —. 1994. Romance past participle mov ement and clause structure. ms. Univ. of Venice. Dobrovie-Sorin, Carmen. 1993. The Syntax of Romanian. Comparativ e Studies in Romance. Studies in Generative Grammar 40. Berlin, New York. Emonds, Joseph. 1976. A Transformational Approach to English Syntax. New York. Fava, Elisabetta ed. 1992. Proceedings of the XVII Meeting of Generativ e Grammar. Volume presented to Giuseppe Francescato on the occasion of his seventieth birthday. Torino. Giorg i, Alessandra. 1990. On the Italian Anaphoric Pronominal System. Padova. —. 1991. Ch. 1 of Giorgi/Longobardi. —, and Giuseppe Long obardi. 1991. The Syntax of Noun Phrases. Configuration, Parameters and Empty Categories. Cambridge. —, and Fabio Pianesi. 1992. On v erbal morphology in Germanic and Romance. In Fava ed. 211—233. Giusti, Giuliana. 1991. The Categorial Status of Quantified Noiminals. Linguistische Berichte 136. 438—452.

1362

—. 1993. La Sintassi dei Determinanti. Padova. Graffi, Gior g io. 1980. Su alcune costruzioni pseudo-relativ e. Riv ista di Grammatica Generativ a 5. 117—139. —. 1990. Generativ e grammar in Italy. Grammar in Progress. GLOW Essays for Henk v an Riemsdijk, ed. by Mascaró, Joan and Marina Nespor. 147—152. Dordrecht. Grimshaw, Jane. 1990. Argument Structure. Cambridge, MA. Guasti, M. Teresa. 1988. La pseudo-relativ e et les phénomènes d’accord. Ri v ista di Grammatica Generativa 13. 35—57. Jaeg g li, Osvaldo, and Ken Safir eds. 1989. The Null Subject Parameter. Dordrecht. Kayne, Richard S. 1975. French Syntax. The Transformational Cycle. Cambridge, MA. —. 1989a. Facets of Past Participle Agreement. Dialect Variation and the Theory of Grammar, ed. by Paola Benincà. Dordrecht. 85—103. —. 1989b. Null Subjects and Clitic Climbing. In Jaeggli and Safir eds. 239—261. —. 1991. La linguistica in Italia v ista dal M. I. T. La linguistica italiana, oggi, ed. by Alberto Varv aro. Società di Linguistica Italiana 29. 141—145. Roma. —. 1993a. Towards a Modular Theory of Auxiliary Selection. Studia Linguistica 47. 3—31. —. 1993b. Anti-symmetry in Syntax. ms. CUNY. Long obardi, Giuseppe. 1985. Su alcune proprietà della sintassi e della forma logica delle frasi copulari. Sintassi e morfologia della lingua italiana d’uso. Teorie ed applicazioni descrittiv e, ed. by Annalisa Franchi de Bellis and Leonardo Sav oia. Società di Linguistica Italiana 24. 211—223. Roma. —. 1991. Extraction from NP and the proper notion of head gov ernment. Ch. 2 of Giorgi/Longobardi. —. 1992. Proper Names and the theory of N-Mov ement in Syntax and Logical Form. Univ ersity of Venice Working Papers in Linguistics 9. [To appear in Linguistic Inquiry). Lonzi, Lidia. 1992. Il Sintagma Av v erbiale. In Renzi and Salvi eds. 341—412. Manzini, M. Rita. 1979. Una teoria del controllo. Rivista di Grammatica Generativa. 4. 139—163. —. 1983. On Control and Control Theory. Linguistic Inquiry 14. 421—446.

XXI. Syntaktische Skizzen

Moro, Andrea. 1993. I predicati nominali e la struttura della frase. Padova. Perlmutter, D. M.1978. Impersonal Passiv es and the Unaccusativ e Hypothesis. Berkeley Linguistics Society IV. 157—189. Picallo, M. Carme. 1991. Nominals and Nominalizations in Catalan. Probus 3. 279—316. Poletto, Cecilia. 1993. La sintassi del soggetto nei dialetti italiani settentrionali. Padova. Pollock, Jean-Yves. 1989. Verb Mov ement, Univ ersal Grammar, and the Structure of IP. Linguistic Inquiry 20. 424—635. Radtke, Edg ar ed. 1991. Le nuov e grammatiche italiane. Tübinger Beiträge zur Linguistik. Tübingen. Renzi, Lorenzo.ed. 1988. Grande grammatica italiana di consultazione. vol. 1. Bologna. —, and Laura Vanelli. 1982. I pronomi soggetto in alcune v arietà romanze. Studi linguistici in onore di G. B. Pellegrini. Pisa. —, and Giampaolo Salvi. eds. 1992. Grande grammatica italiana di consultazione. Vol. 2. Bologna. —, Giampaolo Salvi, and Anna Cardinaletti. eds. 1994. Grande grammatica italiana di consultazione. Vol. 3. Bologna. Rizzi, Luigi. 1978. ch. 2 of Rizzi 1982. —. 1982. Issues in Italian Syntax. Dordrecht. —. 1986a. On chain formation. The Grammar of Pronominal Clitics. Syntax and Semantics 19, ed. by Hagit Borer. New York. 65—95. —. 1986b. On the Status of Subject Clitics in Romance. Studies in Romance Linguistics. ed. by Osv aldo Jaeggli and Carmen Silv a-Corv alán. Publications in Language Sciences 24. 391—419. Dordrecht. —. 1990. Relativized Minimality. Cambridge. Salvi, Giampaolo. 1992. Le frasi copulativ e. In Renzi and Salvi eds. 163—189. Taraldsen, K. Tarald. 1978. On the NIC. Vacuous Application and the That-Trace Filter. Bloomington. IN: Indiana University Linguistic Club. Zanuttini, Raffaella. 1992. Syntactic Properties of Sentential Negation. A Comparativ e Study of Romance Languages. Ph. D. Diss. Univ ersity of Pennsylvania.

Giuliana Giusti, Venice (Italy)

76.  Japanese

1363

76. Japanese 1. 2. 3. 4. 5.

1.

Introduction Headed Structures Unbounded Dependencies Sentence Levels References

Introduction

Japanese is known to be an SOV-order language from the typological point of v iew. It is in fact one of the strictly head-final languages. For example, a v erb follows its complements, a v erb phrase follows its subject, an auxiliary follows its v erbal complement, etc. Moreov er, a noun follows adjectiv es and a postposition follows a noun, a sentential complement is marked by a postsentential complementizer, a conjunction follows the conjunct, and so on. Thus, we hav e the following examples:

except the last one. As for the case of (1d), the adjectiv e atui ‘thick’ precedes the noun hon ‘book’. In this case, the noun is the head of the noun phrase atui hon ‘thick book’, and the adjectiv e is the adjunct (modifier) of the head noun. (1 e) shows the case of an auxiliary following a v erb phrase. In this case, the auxiliary tai ‘want’ is a lexical head of the larger v erb phrase hon-wo yomi-tai ‘want to read a book’, in which the inner v erb phrase hon-wo yomi ‘to read a book’ is the complement. The “auxiliary” is a head of a v erbal phrase in the sense that the tense and/or the aspect of the sentence is carried by it. Moreov er, the polarity of the sentence is determined by whether the auxiliary is of the negativ e form or not. Here are more examples:

The structures of (2a) and (2b) is roughly as indicated in (3a) and (3b), respectively: (3a) [S [S Ken-ga [VP hon-wo yomi] takat] ta] (3b) [S [S Ken-ga [VP hon-wo yomi] taku] nai]

In (1 a), the v erb miru ‘see’ follows its object complement Naomi-wo, where wo is a postposition for accusativ e marking. In (lb), the subject complement Ken-ga (g a is a postposition for nominativ e marking) precedes the v erb phrase Naomi-wo miru ‘see Naomi’. In this case, the v erb phrase is a phrasal head of the sentence Ken-g a Naomi-wo miru. Unlike English, Japanese v erbal heads, whether lexical or phrasal, always follow their complements. In addition, the complementizer koto is the head of the entire phrase, making the sentence a nominal sentential complement. (1c) shows that the nominal conjunction to ‘and’ follows each nominal conjunct

Thus, sentence-final auxiliaries are v erbal heads taking either a v erb phrase or a sentence as their complements. Moreov er, the past tense morpheme and the negation morpheme are also heads taking a sentential phrase as their complements. In this regard, the order of SOV is more appropriately characterized as S(OV), where the substructure consisting of the object and the head v erb (OV) is also a phrasal head. The head-final nature of Japanese is further exemplified by the internal structure of both the subject and the object; they consist of a noun phrase complement and a postpositional head following it. This head-final nature is also demonstrated in the fact that the v erb may be followed by an auxiliary morpheme. Schematically, a typical Japanese sentence has the following fundamental structure:

XXI. Syntaktische Skizzen

1364

theory (Principles-and-Parameters approach) (cf. Chomsky 1981; 1986 see also IX.24), are mentioned for comparison. See Gunji (1987) for a more formal presentation of the phrase structure grammar analysis of Japanese sketched in this article. The final section concentrates on the ‘functional’ aspects of Japanese, rather than ‘formal’ aspects in the preceding sections. The description in this section is accordingly more informal than the other sections.

2. where C stands for a complement and H a head. Each complement (and head) may consist of its own head and complement or adjunct. Japanese sentences are distinguished as belonging to one of sev eral ‘lev els’ depending on the head morpheme, which contributes to add a communicativ e power to a sentence. For example, the question marker no indicates that the status of the sentential complement is a question:

In this article, I will sketch v arious structures in Japanese syntax. The next section concentrates on headed-structures; in particular, on complement-head structures, which includes causativ ization and passiv ization. Some specific cases of adjunct-head structures are discussed in the following section on unbounded dependencies, as both topicalization and relativ ization, which are typical examples of unbounded dependencies, are representativ e cases of adjunct structures. The case of reflexiv ization will be included in this section since, unlike English, Japanese reflexiv es are not subject to the ‘clause-mate’ condition, which restricts the domain of reflexiv ization within a clause. In the final section, the ‘lev els’ of sentences mentioned abov e are discussed with respect to the pragmatic status of some of the constituents. The theoretical apparatus used for the description of specific syntactic constructions in Japanese is based on generativ e grammar, though the presentation is generally informal. In particular, the concepts in phrase structure grammar (cf. Gazdar/Klein/Pullum et al. (1985), and Pollard/Sag (1987; 1994) see also IX.25) are mostly used, but occasionally analyses based on other generativ e frameworks, especially the Gov ernment-Binding

Headed Structures

2.1. Verbs and Complements Japanese v erbs always follow their complements. Thus, a typical sentence consists of a series of complements and the v erb at the sentence-final position:

In (6a), heya-ni ‘in the room’ is usually considered to be a locativ e adjunct (modifier) and not a complement of the v erb iru ‘be’). Howev er, exact distinction between a complement and an adjunct is sometimes v ery difficult. Superficially, both complements and adjuncts precede the head v erb and they often hav e the same internal structure; they consist of a noun phrase and a postposition.

76.  Japanese

Moreov er, sometimes the same postposition can be used for different case markings (e. g., ni for dativ e and locativ e). In this article, I will not go into the details of the distinction between complements and adjuncts, and follow the traditional distinction between transitive and intransitive verbs. The usual subject marker is the nominativ e postposition ga. Some v erbs, howev er, take a dativ e subject, headed by ni (cf. (6e)). The direct object is usually marked by the accusativ e postposition wo (often transcribed as o) (cf. (6b)), while the indirect object of a ditransitiv e v erb is headed by the dativ e postposition ni (cf. (6c)). There are some transitiv e v erbs that take a dativ e object (cf. (6d)), and there are some stativ e v erbs (cf. (6e)) and adjectiv es (cf. (6f)) that take a nominativ e object. The choice of the postposition is lexically determined by the v erbal head and is part of their subcategorization properties. Japanese adjectiv es of predicativ e use directly follow its subject, without a copula. Thus, they hav e their own inflectional pattern when followed by a postv erbal marker such as a tense marker. In this sense, Japanese adjectiv es actually consist a subcategory of verbs with a distinct inflectional pattern. As for inflection, Japanese lacks syntactic concepts of person, number, and gender, though, of course, it is possible to linguistically express these concepts in semantic and pragmatic terms. Thus, the v erbal inflection only occurs when v erbal categories are followed by some particular morphological markers; there is no agreement in the sense of European languages. There has been some controv ersy ov er whether the object complement and the transitiv e v erb form a constituent in Japanese phrase structure. Some hav e treated all the complements as sisters of the v erb. Hence the subject, the object, and possibly other complements and adjuncts, are all arranged in a ‘flat’ structure as follows:

(cf. Inoue (1976), Shibatani (1978), Farmer (1984), and Ishikawa (1985) among others, for this position). Recently, howev er, more attention is paid to the phrase structure of Japanese and a

1365

more hierarchical structure such as the following seems to be preferred:

(cf. Hasegawa (1980), Saito (1985), Hoji (1985), Fukui (1986), and Gunji (1987), among others, for such a position). Some of the arguments for the existence of a VP node in Japanese concerns the possibility of pronominal reference, where a structure-dependent notion such as ‘c-command’ plays a crucial role in specifying the condition of coreference (cf. Saito/Hoji 1983). Similar arguments concern with the possibility of reflexiv ization, on which I will return in Section 3.3. These arguments, of course, are based on the assumption in the approaches taken that the phrase structure is central to determining syntactic conditions. Such approaches include the Gov ernment-Binding theory (cf. Saito (1985), Hoji (1985), Fukui (1986), Nishigauchi (1990), among others, for Gov ernment-Binding analyses of v arious aspects of Japanese) and phrase structure grammar (cf. Gunji (1987) for a Japanese phrase structure grammar analysis). Other approaches, notably, that by Lexical-Functional Grammar (cf. Kaplan/Bresnan 1982), may not be required to assume a hierarchical phrase (constituent) structure (cf. Ishikawa (1985) for the case of Japanese). Howev er, since essentially the same hierarchical structure is reproduced in their functional structure and the similar notion of ‘f-command’ is used in the description of pronominalization in Lexical-Functional Grammar, the need for a hierarchical structure of some form seems to be unanimously acknowledged in the case of Japanese syntax, contrary to some claims that Japanese is one of the so-called ‘nonconfigurational’ languages like Warlpiri (cf. Hale 1980). A brief note before proceeding is in order. In some analyses, the subject and the object are assumed as a noun phrase, with the postposition being some kind of suffix. In these analyses, Ken-g a and Naomi-wo abov e are treated as NPs, while heya-de is treated as a PP. In this article, howev er, ev ery phrase

1366

headed by a postposition is treated as a PP, emphasizing the status of the postposition as the head. This also reflects the fact that the grammatical relation of such a phrase is determined solely by the head postposition; the complement noun phrase itself has no effect on the grammatical status of the postpositional phrase. In the phrase structure grammar analysis, complements of a v erb are represented by the v alue of the subcategorization feature subcat which consists of the complements of the v erb. For example, the transitiv e v erb mi ‘see’ has the following feature structure (only the v rele ant features are shown):

XXI. Syntaktische Skizzen

semantically corresponds to either coerciv e causation or permissiv e causation. The distinction essentially depends on the context. Thus, the following causativ e sentence can be ambiguous.

where the first consonant of sase is dropped following a v erb whose stem ends with a consonant. The following phrase structure tree is based on the assumption that causativ ization is a case of complement-head structure, with the causativ e morpheme as a head taking a v erb phrase as a complement (the tense marker is omitted for simplicity):

Each member in the subcat v alue becomes identical to a complement in a phrase structure, and according to the principle concerning the subcategorization feature, the subcat v alue of the mother node is identical to that of the head minus the complement. Thus, the v erb phrase Naomi-wo mi ‘see Naomi’ has the following feature structure: (10) Naomi-wo mi: v[subcat {p[sbj]}] Naturally, the v alue of the subcategorization feature of a full sentence like Ken-g a Naomi-wo mi will be nil (empty). It is often the case in Japanese that some of the complements are suppressed in the sentence. If a complement is obv ious from the context, it will be replaced by a ‘zero pronoun’, a phonologically null pronominal. In general, a ‘zero’ pronoun will appear in Japanese where an ov ert pronoun would be used in English.

‘Zero’ pronouns are often used in the place of the first and second pronouns. In this sense, Japanese is a highly context-dependent language. 2.2. Causativization Japanese causativ ization is characterized by the auxiliary morpheme sase ‘cause’, which

Some analyses treat a sequence like yomase as a complex v erb (cf. Miyagawa (1980), Ishikawa (1985)) and assume that it behav es as if it were a single lexical item. In this article, howev er, the causativ e morpheme sase is assumed to be a head taking a v erb phrase (VP) as its complement and forming a transitiv e v erb phrase (TVP). Part of the reason for the current treatment is the fact that the complement can be a complex v erb phrase consisting of coordinated VPs:

This sentence shows that the causativ e morpheme sase can be considered to attach not to a lexical verb but to a verb phrase. Some analyses, notably those within the Gov ernment-Binding theory, posit an empty pronoun PRO as the subject of the comple-

76.  Japanese

ment v erb phrase, thus making the complement a full sentence rather than a v erb phrase: (15) Ken-ga Naomii-ni [S PROi hon-wo yom] sase The PRO is assumed to be giv en the same ‘index’ as the object PP by the ‘Control Theory’ of the Gov ernment-Binding theory; hence it is made to be semantically identical to the object PP. That is, causativ ization is a case of object control. In the analysis within phrase structure grammar, object control is assumed to be the result of the subcategorization property of the causativ e morpheme sase. The v alue of the subcategorization feature of this morpheme consists of three elements: the subject, the object, and the v erb phrase. The v erb phrase is a subcategory of v (for verbal) whose v alue of the subcategorization feature consists of only one element: the subject. The fundamental idea is that the semantic v alue of the object in the subcat v alue of the causativ e is identical to the semantic v alue of the subject in the subcat v alue of the v erb phrase that is in the subcat v alue of the causativ e. To put this more straightforwardly, the following scheme will giv e the general idea:

where the v ariable after the colon following the category symbol represents the semantic v alue of the category. The postpositional phrases are distinguished by the v alue of the grammatical relation feature, which takes either a sbj (subject) or an obj (object) as the v alue. The lexical information of sase shown in (16) explicitly states the object control property of causativ ization without inv olv ing an abstract empty category. The object-marker for causativ ization has been the source of some controv ersy, since two distinct markers are usually possible: the dativ e ni and the accusativ e wo. The former has sometimes been associated with the permissiv e causation, while the latter with the coerciv e causation. Howev er, such a distinction, if any, is easily ov erridden by the context and not univ ersally acknowledged. Moreov er, the choice of the object-marker is limited to ni when the embedded v erb phrase has its own object marked by wo, which is known as the ‘Double wo Constraint’ (cf.

1367

Shibatani 1976). That is, we cannot use wo in the place of ni in (12):

Thus, the distinction between the ni-marked and the wo-marked objects is somewhat v ague and not so clear-cut as is sometimes claimed. 2.3. Passivization Like causativ ization, Japanese passiv ization is characterized by a specific morpheme rare, which takes a v erb phrase as its complement. The case of passiv ization is slightly more complicated since there are two syntactically and semantically different morphemes, although morphologically they are identical. One subcategorizes for an intransitiv e v erb phrase (VP). The other subcategorizes for a transitiv e v erb phrase (TVP). Of these two kinds of morphemes, the one subcategorizing for a TVP is a Japanese counterpart for the passiv e in English and many other languages. The semantic function of this morpheme is to ‘adv ance’ (in the sense of relational grammar) the object to the subject, with the subject becoming an (oblique) object (marked by ni). We ha v e the following example:

where in (19b), rare loses the first consonant following a consonant-ending verb stem. As with the causativ e, the fact that rare does not simply attach to a lexical transitiv e v erb, but instead subcategorizes for a transi-

XXI. Syntaktische Skizzen

1368

tiv e v erb following

phrase

is

exemplified

by the sentence:

Note that the causativ e sase subcategorizes for a VP. Since it further subcategorizes for a subject and an object, the phrase consisting of the VP and sase is a TVP, subcategorizing for a subject and an object. This TVP becomes the complement of rare, forming another TVP. Some analyses in the Gov ernment-Binding theory assume, just like the passiv ization in English, an application of the Mov e a transformation for Japanese passiv ization. Thus, a passiv e sentence like (18b) will be deriv ed from a d-structure like the following:

Thus, semantically, the subject of the complement TVP corresponds to the (oblique) object of rare, while the object of the complement TVP corresponds to the subject of the passiv e morpheme. In this analysis, no empty category like the ‘trace’ in (21b) is assumed to assure the semantic correspondence. The subcategorization property specified in the lexicon represents both syntactic and semantic information. Japanese is known to hav e a second type of passiv ization. The morphologically identical morpheme rare takes a VP, rather than a TVP, as its complement. The following is a typical example.

In this case, since the complement of the passiv e morpheme is an intransitiv e v erb phrase, which subcategorizes for only a subject, the syntactic function of the morpheme is only to ‘demote’ the subject of the complement VP to an (oblique) object. The subject of the passiv e sentence is supplied independently of what the complement VP subcategorizes for. The phrase structure of (23b) is as follows: In this type of analysis, the coindexing between the trace left by the application of Mov e a and the subject is assumed to assure the appropriate semantic role of the subject with respect to the verb. In the phrase structure grammar analysis, the semantic correspondence is assumed to be designated in the subcategorization property of the passiv e morpheme, just like the case of causativ e. The subcategorization frame of rare has the following structure:

The following examples are to show that the complement of this type of rare is a v erb phrase, not merely a lexical verb:

76.  Japanese

1369

interpretation is hardly present. Moreov er, ev en a ‘direct’ passiv e sometimes has an adv ersity connotation (cf. (19b)). Thus, the passiv es like (23b) are sometimes called ‘indirect’ passiv e without referring to any ‘adv ersity’ connotation. In this article, these two kinds of passiv e are distinguished by the subcategorization property as intransitv e and transitiv e passiv es, respectiv ely. (See Kuno (1983) for an attempt to explain the origin of ‘adv ersity’ connotation in intransitive passives.) 2.4. Adjuncts

As with causativ es, analyses in Gov ernment-Binding theory might assume an empty category PRO in this type of passiv es:

Japanese adjuncts are of two types: adnominal and adv erbial. The former adjoins to nominal heads and includes adjectiv es of attributiv e use, postpositional phrases headed by no, and a closed class of lexical adnominals. As with complements, adjuncts precede the head they adjoin to. Here are some examples:

The phrase structure grammar analysis of this type of passiv e morpheme is again based on the subcategorization property of this kind of rare:

which is exactly like the subcategorization frame for the causative morpheme. The existence of two kinds of passiv es in Japanese has been a source of controv ersy. Some hav e argued that the two kinds should hav e the same kind of underlying structure, namely, the one with an embedded sentence (cf. Kuroda (1965), Howard/Niyekawa-Howard (1976), among others). Other transformational analyses hav e assumed a mov ement transformation for the passiv e of the type of (18) or (19) (often called the ‘direct’ passiv e) (cf. McCawley (1972), Kuno (1973)). The former, so-called ‘uniform’, theory is known to hav e the defect that it fails to explain the nonambiguous interpretation of the reflexive in the ‘direct’ passive, to which I will return in Section 3.3. Ev en though the passiv e of the type of (23) has often been called the ‘adv ersity’ passiv e, the adv ersity connotation is sometimes missing. For example, (25a) is syntactically of the type of ‘adv ersity’ passiv e but the ‘adv ersity’

The form of adjectiv es of attributiv e use is indistinguishable from their form of predicativ e use. In this sense, they can be considered to be cases of relativ e clauses. In fact, v erbs can also be used attributiv ely in the prenominal position:

I will discuss relative clauses in Section 3.2. The genitiv e postposition no is unique in that it makes an adnominal phrase: all the other postpositions either make a postpositional phrase as a complement of a v erb or an adv erbial phrase as an adjunct to a v erbal head. The semantic content of no is quite general and cov ers any plausible relationship between the two noun phrases connected by it. The lexical adnominal ookina is semantically identical to the adjectiv e ookii but the former doesn’t hav e an inflected form and is never used predicatively:

XXI. Syntaktische Skizzen

1370

The class of lexical adnominals includes tiisana ‘small’, sono ‘that’, kono ‘this’, etc. The adv erbials include inflected forms of adjectiv es, postpositional phrases headed by v arious postpositions, and a small closed class of lexical v ad erbs.

The adnominal postposition no can conv ert some of the adv erbials to adnominals:

In the next section, two of the productiv e adjunct formation processes are discussed, namely, relati v ization and topicalization. Since they both inv olv e the existence of a gap in a sentential constituent and the gap can be related to a constituent separated by an unlimited number of sentence boundaries, they are treated as subcases of unbounded dependencies.

3.

Unbounded Dependencies

3.1. Topicalization Japanese topicalization is just like the counterpart in English in that the topic appears at the sentence-initial position as an adv erbial adjunct, with the v erbal (sentential) head following the topic. The topic is always a postpositional phrase headed by wa. We hav e the following examples:

As can be seen in (33a) and (33b), the usual postposition for the nominativ e subject (ga) and the accusativ e object (wo) are suppressed when they are followed by the topic marker wa. Other postpositions remain and appear before wa. (33c) is an example of such a case. In this case, the adv erbial adjunct phrase Ken-kara ‘from Ken’ is topicalized. The sentential heads in the abov e examples all hav e a gap somewhere: (33a) has a gap at the subject position, (33b) at the object position, and (33c) at an adv erbial adjunct position just before the verb. Thus, some analyses in the Gov ernmentBinding theory assume an application of Mov e a in the deriv ation of these kinds of topicalized sentences and hence the s-structures v ha e a trace somewhere:

The phrase structure grammar analysis treats a gap by designating it as the v alue of one of the features concerning binding: slash. Roughly speaking, a phrasal node dominating a gap has as the v alue of slash the category corresponding to the gap. Thus, for example, the node corresponding to Ken-g a kiratteiru in (33b) has the following feature structure (the tense and aspect are ignored here):

Since the v alue of the subcategorization feature is nil (empty), it is a kind of sentence, with the gap corresponding to the object. By

76.  Japanese

a gap introduction lexical rule, the head v erb has the following feature structure:

That is, one of the complements in the v alue of subcat is remov ed and put in the v alue of slash. The information about the gap as the v alue of slash is propagated from the lexical v erbal head kiratteiru to upper nodes by a principle concerning the inheritance of binding features. This principle essentially requires that the v alue of a binding feature of a mother node is identical to the union of the v alues of the binding feature of all its daughters. Thus, if a head has a gap, that information is propagated to the mother node and this process is repeated up to the sentential head of the topic. At the v ery top node, the topic semantically binds the v alue of the slash of the head. Though the treatment of topicalization in the Gov ernment-Binding theory and that in phrase structure grammar are alike in that a counterpart of a gap is assumed, they are different in one fundamental way: the latter allows the information about the gap to be propagated only locally. Here the ‘locality’ is used in its strict sense in that a node in the phrase structure tree is related with only its mother, sisters, and/or daughters. That is, a node cannot be directly related to an ‘aunt’ node or a ‘grandmother’ node, for example. In this sense, the Gov ernment-Binding theory allows the mov ement transformation to mov e a constituent nonlocally, to a sister of a ‘grandmother’ node, for example. The unbounded nature of topicalization is illustrated in the following examples:

These are natural consequences from the as-

1371

sumption that the mov ement transformation in the Gov ernment-Binding theory or the gap propagation mechanism in phrase structure grammar can cross the sentence boundary. There are notable examples of topicalization in which no gap is in v ol v ed:

In this type of topicalized sentences, the topic is generated at the sentence-initial position from the beginning ev en in transformational analyses. Since the semantic relationship between the topic and the sentential head in these sentences v aries depending on the sentence and its context, we can only say that there is some kind of ‘topic-comment’ relationship between the topic and the head. I will return to the status of the ‘topic’ in the final section. Japanese is often cited as a language of relativ ely free word order. In fact, complements of a v erb can generally be freely ordered.

The SOV order is generally considered to be fundamental in that other orders sound

XXI. Syntaktische Skizzen

1372

less natural unless there is a motiv ation for putting a complement other than the subject at the sentence-initial position. That is, nonsubject complements are put at the sentenceinitial position only when they bear some kind of ‘old information’. The rest of the complements bear ‘new information’ and relativ ely more important. I will return to the problem of old/new information in Section 4. Some analyses in the Gov ernment-Binding theory treat this kind of word order v ariation as a v ariant of topicalization, resulting from the application of Mov e α. Thus, the abov e sentences would hav e a trace or traces somewhere. (40a) Ken-ga Naomi-ni hon-wo ageta. (40b) Naomii -ni Ken-ga ti hon-wo ageta. (40c) Honj-wo Ken-ga Naomi-ni tj ageta. (40d) Ken-ga hon-woj Naomi-ni tj ageta. (40e) Naomii-ni honj-wo Ken-ga ti tj ageta. (40f) Honj-wo Naomii-ni Ken-ga ti tj ageta. The phrase structure grammar analysis distinguishes intrasentential word order v ariation and unbounded fronting and attributes the former to a special property of subcategorization in Japanese syntax. In the phrase structure grammar treatment of English subcategorization (Pollard/Sag 1994), the elements of the subcat feature are ordered according to their grammatical obliqueness. That is, the least oblique complement comes first, then a more oblique complement, and so on. The subject comes first and the direct object comes just after the subject. One of the functions of this order is to determine the surface order of sister complements. In the case of Japanese (Gunji 1987), the v alue of the subcat feature is assumed to be an unordered set of categories. That is, subcategorization plays no role in determining the constituent order. Since the information on grammatical relations is necessary in describing other grammatical phenomena, they are expressed by a separate feature. Thus, the categories subcategorized for by a v erbal head can appear in any order within a sentence. Of course, since subcategorization doesn’t extend beyond the sentence boundary, fronting across a sentence boundary is assumed to be a different phenomenon and handled just like topicalization. In fact, fronting across a sentence boundary is not a free process as compared with topicalization. Note the difference in acceptability between topicalization and unbounded fronting:

Without an explicit marker like wa, fronting seems to become increasingly difficult as it crosses sentence boundaries. As compared with cross-sentential fronting, intrasentential word order v ariation, or ‘scrambling’ as is often called, is a much freer process. This fact is expressed by assuming no ordering among subcategorized-for complements. 3.2. Relativization There is much parallelism in topicalization and relativ ization in Japanese. The only contrast is that, in topicalization, the topic is an adv erbial adjunct to a sentential head, while in relativ ization, the relativ e clause is an adnominal adjunct to a nominal head. Since Japanese does not hav e counterparts of relativ e pronouns, the relativ e clause is immediately followed by the nominal head.

In general, there is no syntactic distinction between the restrictiv e and nonrestrictiv e relativ e clauses. The nature of the nominal head and the context usually determines which is intended. A usual interpretation of (42b) would inv olv e a nonrestrictiv e relativ e clause, while that of (42c) would inv olv e a restrictiv e relative clause. The mechanisms to analyze relativ e clauses are exactly the same as that for topicalization: Mov e a in the Gov ernment-Binding theory and the slash feature in phrase structure grammar. Again, we can see the unbounded nature of these mechanisms.

76.  Japanese

1373

the antecedent subject of zibun need not be ov ertly expressed, either. This will result in apparent object control of the reflexiv e (see (49) below). Since zibun is morphologically an ordinary noun, it can be followed by any postposition, including the v geniti e no.

As with topicalization, a relativ e clause without a gap is also abundant in Japanese:

As has been mentioned in Section 2.4., a single adjectiv e or intransitiv e v erb can constitute a relativ e clause. In such cases, it will look exactly like a lexical adnominal adjunct:

Note that akai in (45a) and sinda in (45b) are considered to be sentences with subject gaps, rather than being a lexical adjectiv e or v erb. In this respect, an attributiv e adjectiv e and a relativ e clause whose head is an adjectiv e are hardly distinguishable. In fact, the former can be considered to be simply a case of relativization. 3.3. Reflexivization Reflexiv ization in Japanese is characterized by a special noun zibun, which corresponds to ‘self’ without a distinction in gender, person, or number. The antecedent of zibun is a subject in some representation of the phrase structure. It is not necessarily restricted to the subject in the same clause in which the reflexiv e appears. In this sense, the condition like the so-called ‘clause-mate’ condition in English does not apply in Japanese. Moreov er,

Note that in (46b), the object cannot be the antecedent of zibun. Since there are two subjects in (46c), one in the embedded clause and the other in the matrix clause, zibun is ambiguous in this sentence. The unbounded nature of reflexiv ization is handled in phrase structure grammar by positing a binding feature refl, which takes a p[sbj] (subject PP) as its v alue. The lexical specification of zibun is as follows: (47) zibun: n[refl {p[sbj]:X}]:X That is, the semantic v alue of zibun is identical to the semantic v alue of the subject in the v alue of the refl feature. The v alue of refl is propagated in the same way as slash and becomes identical to a subject in the v alue of subcat at a v erb phrase node. For example, the VP zibun-wo butta ‘hit self’ in (46a) has the following feature specification:

The p[sbj] in the v alue of subcat becomes identical to the p[sbj] in the v alue of refl propagated from zibun, and hence the two semantic v alues X and Y become identical. Since the p[sbj] in turn becomes identical to the subject of the sentence, Ken, the zibun ev entually becomes coreferential with the subject. Thus, it is assumed that the v alue of refl is not directly bound by the subject of the sentence but is indirectly bound by v irtue of a

1374

p[sbj] in the subcat v alue of a v erb phrase. This assumption is crucial in explaining the apparent object control of the reflexiv e in causativ es and one type of passiv es (intransiti e v passi e):

Note that in these examples, we hav e two v erb phrases with their own subcat features. For example, in (49a), whose phrase structure tree is shown below, the subcat v alue of the lower VP zibun-no hon-wo yom includes a p[sbj], which is semantically bound by the object Naomi due to the lexical semantics of the causati v e morpheme (cf. Section 2.2.).

On the other hand, the subcat v alue of the higher VP Naomi-ni zibun-no hon-wo yom-ase in (50) also includes a p[sbj], which becomes identical to the subject of the sentence Ken. Thus, since the v alue of refl becomes identical to either one of the p[sbj]s in these subcat values, there will be ambiguity of the reflexiv e in this kind of examples. Generally speaking, this kind of ambiguity appears whenev er a VP is embedded whose p[sbj] in the subcat value is controlled by the object. Analyses in the Gov ernment-Binding theory will explain the ambiguity of these constructions by allowing the empty category PRO to be an antecedent of the reflexiv e. Since the PRO is a subject in the embedded sentence and is controlled by the object, the reflexiv e can be controlled indirectly by the object, although it is only directly controlled by the subject of the sentence.

XXI. Syntaktische Skizzen

(51) Ken-ga Naomii-ni [S PROi zibun-no hon-wo yom] sase] As mentioned in Section 2.3., the so-called ‘uniform’ theory of passiv ization faces a problem when both kinds of passiv e are assumed to hav e an embedded sentence. Note that the embedded subject in these structures will hav e to be able to control the reflexiv e. Howev er, the reflexiv e is not ambiguous in the ‘direct’ v passi e.

Ev en though there has been a proposal to stipulate an additional constraint on the reflexiv e, such a constraint is known to be subject to counterexamples. One of the more recent attempts to explain the interpretation of the reflexiv e in ‘direct’ passiv es exploits ‘functional’ concepts (such as “inv olv ement”) in syntax (cf. Kuno 1983, 1987). Since Japanese is known to heav ily use ‘functional’ information in determining the syntactic and morphological structures (cf. Section 4.), such an attempt deserv es to be pursued further. Ev en though, at the current stage, these ‘functional’ concepts are often v ague and lack formal definitions, both ‘functional’ and ‘formal’ analyses will be used in compensatory situations to enrich the description of the grammar of Japanese. In the phrase structure grammar analysis, the nonambiguity of zibun in (52) simply comes from the fact that there is only one VP node in the phrase structure tree, namely, Naomi-ni zibun-no heya-de home-rare ‘praised by Naomi in self’s room’. Since the p[sbj] in the subcat feature of this VP becomes identical to only the subject, zibun can only be coreferential with Ken.

76.  Japanese

4.

Sentence Levels

Perhaps the most perspicuous phenomenon in Japanese syntax that demonstrates the head-final property of this language is the sentence-final clusters of markers. What follow the v erb stem include: v oice markers such as sase (causativ e) and rare (passiv e), aspectual markers such as i (progressiv e, resultativ e, or experiential) and simaw (perfectiv e), tense markers such as ru (present) and ta (past), modal markers such as daroo (supposition), etc. Each of these takes a distinct v le el of sentential complements. The following shows a schematic hierarchy of a sentential structure from the functional point of v iew (cf. Minami (1974), Takubo (1987)) (the names for lev els are somewhat tentativ e).

The mood markers include those for question ka or no (colloquial), for the speaker’s sex wa (female speakers), for confirmation ne, etc. They all contribute to make the sentence hav e some kind of communicativ e power. In addition, adjuncts may adjoin to some of the lev els. For example, manner adv erbials may adjoin to ‘action’, restrictiv e sentential adv erbials to ‘ev ent’, and nonrestrictiv e sentential adv erbials to ‘judgment’, etc. Each of these functional lev els is syntactically realized as a v erbal category. For example, the ‘action’ is realized as either a transitiv e or an intransitiv e v erb phrase, and the ‘state’ as a sentence. All the higher lev els including ‘process’, ‘ev ent’, ‘comment’, and ‘judgment’ are realized as sentences headed by respectiv e markers. Thus, if a sentence is headed by a tense marker, it is a sentence of the ‘ev ent’ lev el. As for the ‘communication’

1375

lev el, the categorial status of the ‘mood’ markers is somewhat unclear. Traditionally, they are classified as postpositions, but the functions of these markers are quite different from other typical postpositions such as nominativ e and accusativ e markers. In fact, some theoreticians in traditional Japanese grammar hav e stressed the importance of these markers (the so-called tinzyutu-zi ‘statement marker’) and put them in the center of their theories. Note the position of ‘topic’, headed by the topic marker wa, in this hierarchy. In general, the sentential head of the topic (‘comment’) includes the markers for the aspect, tense, and modality. On the other hand, the ‘agent’, usually headed by the nominativ e marker ga, appears inside the complement of an aspectual marker (‘state’). Thus, we hav e the following contrast:

The postposition node ‘because’ takes a sentential complement of the ‘ev ent’ lev el. Thus, in (55a), the complement of node is Ken-g a ki-ta ‘Ken came’. The same postposition in (55b), howev er, cannot take Ken-wa ki-ta as the complement, since the topic is outside the ‘ev ent’ lev el. In fact, Ken-wa ki-ta (55b) is not a constituent, though Ken-g a kita in (55a) is. We hav e roughly the following phrase structures for (55a) and (55b), respectively. (56a) [s Ken-ga ki-ta] node [Spro kaet-ta] (56b) Ken-wa [S [Spro ki-ta] node [Sgap kaet-ta]] The v erb kaetta ‘returned’ in (55a) has a ‘zero’ pronominal subject (designated by pro in (56a)). On the other hand, the topic Kenwa in (55b) binds the g ap subject of the v erb kaetta. The subject of kita ‘came’ in (56b) is also a ‘zero’ pronoun. The referents of these ‘zero’ pronouns are determined by the context. If the ‘zero’ pronominal positions are lexically filled, the sentences in (55) would have variants like the following:

XXI. Syntaktische Skizzen

1376

tion; (58d) is merely a case of intrasentential ‘scrambling’ and both (58c) and (58d) are grammatical. Thus, some transformational grammarians doubt the existence of ‘Whmov ement’ (or ev en the v ery existence of Mov e α, including ‘NP-mov ement’) at the lev el of s-structure. At the lev el of LF, howev er, the existence of ‘Wh-mov ement’ is sometimes hypothesized (cf. Nishigauchi 1990). The contrast exhibited in (58) can be generalized as a constraint that interrogativ es cannot appear inside the topic phrase: In this sense, the sentential head to which a topic adjoins is a much higher lev el of sentence, a lev el of ‘comment’ (about the topic) that constitutes some kind of ‘judgment’ together with the topic. The topic marker wa is usually associated with old, or giv en, information, namely, information already introduced into the discourse by the speakers. In the abov e sentence (57b), the topic ‘Ken’ corresponds to old information and what is commented on Ken, namely, the ‘ev ent’ that he went home because Naomi came, is new information. Since the subject appears inside an ‘ev ent’, the subject marked by the nominativ e g a naturally corresponds to new information. In this respect, interrogativ e nominals such as dare ‘who’ and nani ‘what’ are nev er used with wa, since they are nev er old information. These facts all follow from the assumption that the topic bears old information; a bearer of new information such as an interrogativ e cannot be included in it.

5.

(58b) and (58e) show that interrogativ es cannot be topicalized. Incidentally, Japanese interrogativ es need not be fronted to the sentence-initial posi-

References

Chomsky, Noam. 1981. Lectures on Gov ernment and Binding. Dordrecht. —. 1986. Barriers. Cambridge, MA. Farmer, Ann. 1984. Modularity in Syntax: A Study of Japanese and English. Cambridge, MA. Fukui, Naoki. 1986. A Theory of Category Projection and its Applications. Massachusetts Institute of Technology Ph. D. dissertation. Gazdar, Gerald; Ewan Klein; Geoffrey K. Pullum; and Ivan A. Sag . 1985. Generalized Phrase Structure Grammar. Oxford. Gunji, Takao. 1987. Japanese Phrase Structure Grammar. Dordrecht. Hale, Kenneth. 1980. Remarks on Japanese phrase structure: comments on the papers on Japanese

77. Georgian

syntax. Theoretical Issues in Japanese Linguistics (MIT Working Papers in Linguistics, v ol. 2), ed by Yukio Otsu & Ann Farmer, 185—203. Department of Linguistics and Philosophy. Haseg awa, Nobuko. 1980. The VP constituent in Japanese. Linguistic Analysis 6. 115—131. Hoji, Hajime. 1985. Logical Form Constraints and Configurational Structures in Japanese. Univ ersity of Washington Ph. D. dissertation. Howard, Irwin, and Ag nes M. Niyekawa-Howard. 1976. Passiv ization. Syntax and Semantics. Vol. 5, Japanese Generativ e Grammar, ed. by Masayoshi Shibatani, 201—237. New York. Inoue, Kazuko. 1976. Henkei Bunpoo to Nihongo (‘Transformational Grammar and Japanese’). Tokyo (in Japanese). Ishikawa, Akira. 1985. Complex Predicates and Lexical Operations in Japanese. Stanford Univ ersity Ph. D. dissertation. Kaplan, Ronald M., and Joan Bresnan. 1982. Lexical-Functional Grammar: a formal system for grammatical representation. The Mental Representation of Grammatical Relations, ed. by Joan Bresnan, 173—281. Cambridge, MA. Kuno, Susumu. 1973. The Structure of the Japanese Language. Cambridge, MA. —. 1983. Sin Nihon Bunpoo Kenkyuu (‘New Studies on Japanese Grammar’). Tokyo (in Japanese). —. 1987. Functional Syntax: Anaphora, Discourse, and Empathy. Chicago. Kuroda, Shig e-Yuki. 1965. Generativ e Grammatical Studies in the Japanese Language. The Massachusetts Institute of Technology Ph. D. dissertation [Published in New York, 1979.] McCawley, Noriko Akatsuka. 1972. On the treatment of Japanese passiv es. Papers from the 8th Regional Meeting, Chicago Linguistic Society. 259—

77.

Georgian

1. 2. 3.

Simple Sentences Complex Sentences References

Georgian is spoken in the Caucasus in the Sov iet Union; it is a member of the Kartv elian (South Caucasian) family. The main purpose of this chapter is to giv e an impression of the wealth of natural language by highlighting those typological facts that distinguish Georgian (and the Kartv elian family)

1377

270. Minami, Fujio. 1974. Gendai Nihongo-no Koozoo (‘The Structure of Modern Japanese’). Tokyo (in Japanese). Miyag awa, Shig eru. 1980. Complex v erbs and the Lexicon. Univ ersity of Arizona Ph. D. dissertation. [Av ailable as Coyote Papers: Working Papers in Linguistics from A → Z, vol. 1. Tucson.] Nishig auchi, Taisuke. 1990. Quantification in the Theory of Grammar. Dordrecht. [Rev ision of his Ph. D. dissertation. Univ ersity of Massachusetts, 1986]. Pollard, Carl J., and Ivan A. Sag . 1987. Information-based Syntax and Semantics. Vol. 1, Fundamentals. CSLI Lecture Notes No. 13, Center for the Study of Language and Information, Stanford University. —. 1994. Head-driv en Phrase Structure Grammar, Chicago. Saito, Mamoru. 1985. Some Asymmetries in Japanese and Their Theoretical Implications. Massachusetts Institute of Technology Ph. D. dissertation. —, and Hajime Hoji. 1983. Weak crossov er and mov e a in Japanese. Natural Language and Linguistic Theory 1. 245—259. Shibatani, Masayoshi. 1976. Causativ ization. Syntax and Semantics. Vol. 5, Japanese Generativ e Grammar, ed. by Masayoshi Shibatani, 239—294. New York. —. 1978. Nihongo-no Bunseki (‘Analysis of Japanese’). Tokyo (in Japanese). Takubo, Yukinori. 1987. Toogo Koozoo to Bunmyaku Zyoohoo (‘Syntactic Structures and Contextual Information’). Nihongo-gaku (‘Japanese Studies’) (Meijishoin), v ol. 6, No. 5, 37—48 (in Japanese).

Takao Gunji, Osaka (Japan)

from other languages. We may note from the outset that Modern Georgian is a typical SOV language, that it is non-configurational and is a candidate for ergativ ity, and that it has complex morphosyntax, partially inflectional, but predominantly agglutinativ e. An additional goal of section 1 of the chapter is to illustrate the application of relational grammar. Section 1 of this chapter treats the syntax of simple sentences, while section 2 discusses complex sentences.

1378

1.

in Series II, while the so-called nominativ e (Nom) marks the direct object in Series II and III and the subject in Series I.

Simple Sentences

1.1. Word Order While Georgian has SOV order, it is not a strict v erb final language. The most frequent orders are Subject—Indirect Object—Direct Object—Verb, and Subject—Direct Object— Verb—Indirect Object, but order of major constituents is v ery free (Apridoniʒe 1986). It has postpositions ( magida-ze ‘table on’), possessors, adjectiv es and numerals preceding head nouns ( čemi ori didi c’ig ni ‘my two big book(s)’), standard of comparison — comparativ e adjectiv e order ( šen-ze didi ‘bigger than you’, lit. ‘you-on big’) or comparativ e adjectiv e — standard of comparison ( upro didi vidre šen, lit. ‘more big than you’), and auxiliaries most often following main v erbs ( čamosuli var ‘I hav e arriv ed’, lit. ‘arriv ed I/ am’); there is some flexibility in the last characteristic. As discussed in section 2.1., relativ e clauses may precede or follow the head noun. 1.2. Subject, Direct Object, Indirect Object Perhaps the most salient problem in the syntax of Georgian is the identification of subject, direct object, and indirect object and the encoding of these grammatical relations. In section 1.2.1. I describe the encoding systems in a pretheoretical way using traditional terminology; v arious analyses of these facts are discussed in sections 1.2.2.—1.2.5. 1.2.1. Description Georgian has three sets of tense-aspect-mood paradigms, traditionally called “series”; each appears to hav e a distinct way of assigning cases. Some of the differences can be seen in the following examples: the tense-aspectmood paradigm (henceforth TAM) illustrated in (1) is in Series I, (2) in Series II, and (3) in Series III. The names used for the cases are traditional and should not be taken literally; for the transitiv e v erb, illustrated here, the narrativ e (Nar) case marks subject

Howev er, these sentences only partially represent the facts, for there is also v ariation according to the class to which the v erb belongs. Case marking v ariation is summarized in Table 77.1. The classes referred to in Table 77.1 are sets of finite v erb forms. They are defined morphologically (Harris 1981, 1985; Holisky 1981 a) and are characterized by clusters of morphological, syntactic, and semantic traits that correlate for most v erbs in the language. The correlations are summarized in Table 77.2; in the table the syntactic and semantic characteristics are not signalled specifically by the morphological trait on the same line, but are independent characteristics of the same set of verbs. Three morphological traits distinguish among the classes. Morphological trait (a) refers to the formation of the future, conditional, future subjunctiv e, and aorist TAMs. Morphological trait (b) giv es the suffixes used to mark subjects of the third person singular and plural in the future TAM. (c) represents the suffix used to mark third person plural subjects in the aorist. While there are additional morphological v ariations among

Table 77.1: Case Assignment in Series I, II and III.

Series I Series II Series III

Subject of Class 1 NOM NAR DAT

Subject of Class 3 NOM NAR DAT

Subject of Class 2 NOM NOM NOM

Direct Object DAT NOM NOM

Indirect Object DAT DAT -tvis

77.  Georgian

1379

Table 77.2: Class Correlations in Modern

Georgian. Class Morpholo- Syntactic gical a.preverb a. transitive 1 b.-s, -en b. narrative case c.-es c. inversion 2 a.preverb a. intranor esitive b.-a, -an b. nominative case c.-nen c. no inversion 3 a.i--(eb) a. intransitive b.-s, -en b. narrative case c.-es c. inversion

Semantic a. active b. telic a. inactive b. telic or stative a. active b. atelic

the classes, these three are the most consistent. Syntactic trait (a) shows the transitiv ity of v erb forms of this class, where a form is considered transitiv e if it has subject and direct object, and intransitiv e otherwise. Syntactic (b) shows the case used to mark the subject of v erb forms of this class when they are in Series II TAMs. Line (c) refers to v erbs that hav e dativ e subjects in Series III (“inv ersion”) and those that do not (“no inv ersion”); see also section 1.2.3. below. Semantic trait (a) shows whether v erbs of this class are v olitional and under the control of the subject (“activ e”) or not (“inactiv e”). (b) refers to a stativ e/non-stativ e distinction and within the latter to v erbs that express an action marked for hav ing an end point (“telic”) or without indication of such an end point (“atelic”). Many of these notions are discussed further below. Some v erb forms are exceptions to one or more of these trait correlations. Class 1 is exemplified in (1)—(3) abov e; ex-

amples of other classes are giv en below. Class 4 is omitted here but is described in section 1.2.3. Verb agreement also encodes information about subjects, direct objects, and indirect objects, as summarized in Table . The allomorphs of the third person indirect object marker and the sev eral v ariants of the third person plural subject marker are not included here. (Morphological and syntactic constraints on the cooccurrence of these in a single v erb form are giv en in Harris 1981, 30 f.) The prefixal portion of the third person indirect object marker, s-, is replaced by u- under certain circumstances (Harris 1981, 89 f). In Series I and II, the subject is marked by the set of subject markers in Table 77.3, the direct object by the direct object markers, and the indirect object by the indirect object markers. In Series III, on the other hand, the notional subject of Class 1 and 3 v erbs is marked instead by the indirect object markers, while the notional direct object of these v erbs is marked by subject markers, and their indirect objects are marked with the postposition -tvis. For Class 2 v erb forms in Series III, the subjects are marked by the subject markers, the indirect objects by indirect object markers. The facts stated abov e are described in all of the handbooks; v arious analyses of them are discussed in sections 1.2.2.—1.2.4. 1.2.2. Is Case Marking in Series II Ergative? One of the points of debate in the analysis of the facts described in 1.2.1. concerns the issue of whether case marking in Series II is ergativ e. Contrasts of the sort illustrated in (4)/ (5) lead to considering case marking in this series to be ergative.

Table 77.3: Three Sets of Agreement Markers.

1. person 2. person 3. person

Singular v-∅ --s/a/o

Subject Plural v--t --t --en, etc.

Direct Object Singular Plural gv-m-g-g--t ∅ ∅

Indirect Object Singular Plural gv-m-g-g--t s-s--t

1380

XXI. Syntaktische Skizzen

Some of the western dialects differ from literary Modern Georgian in case marking patterns (Fähnrich 1967; Harris 1985, 113— 115, 376—380). Although I hav e translated these two v erb forms differently into English, they are forms of a single root. (The a- prefix in ga-a-zard-a occurs in many, but not all, transitiv e forms; it may be considered a formant of Class 1 v erb forms, though not all contain it. The iin g a-i-zard-a is one of sev eral formants of Class 2. Traditionally, these Class 2 intransitiv es are considered to be deriv ed from the corresponding transitiv es.) The form in (4) is a Class 1 form, and that in (5) a Class 2 form. (Both classes also contain unpaired forms, which gov ern the case marking characteristic of the class.) On the basis of this distribution, case marking in Series II was traditionally considered ergativ e, for the subject of the intransitiv e in (5) and the direct object in (4) are marked with one case, while the subject of the transitiv e in (4) is marked with a distinct case (see also Boeder 1979, 459). In this analysis, Class 3, which contains mostly monov alent v erbs, has been ignored, or treated as irregular, or termed transitiv e; this class is illustrated in Series II in (6).

Holisky (1981a) showed that the v erbs of Class 3 are, in fact, quite regular with respect to morphology, syntax, and semantics and constitute a large and productiv e type. Giv en their ov erall regularity, we cannot exclude these v erbs when considering the issue of whether Series II case marking is ergativ e. Most of the v erbs of Classes 2 and 3 are intransitiv e; and, thus, in Series II some intransitiv es (Class 2) gov ern nominativ e case subjects, while others (Class 3) gov ern narrativ e. Since the div ision of v erbs gov erning nominativ e v s. narrativ e case subjects in Series II corresponds predominantly to those that hav e inactiv e v s. activ e semantics (see definition abov e following Table 77.2), the case marking pattern of this series in Georgian falls into the type often called activ e/inactiv e, or just activ e (Sapir 1917; Fillmore 1968, 54). Neither the case marking of major constituents nor other major rules (but see Harris 1981, 174) of Georgian morphosyntax are truly ergativ e (see article 64, “Ergativ ity” herein).

1.2.3. Is There Inversion in Series III? The problem of inv ersion in Series III is closely tied to the identification of basic grammatical relations throughout the language. Inv ersion is the name of the construction in which notional subjects hav e dativ e case marking (see Table 77.1) and indirect object agreement (see Table 77.3), while notional direct objects (if present) hav e nominativ e case marking and subject agreement. This occurs with Class 1 and 3 v erbs in Series III and with certain “affectiv e” v erbs in all series. The issue here is whether or not these facts are to be accounted for by means of a syntactic rule that changes grammatical relations. Most traditional descriptions hav e taken the notional grammatical relations in Table 77.1 as basic (eg. Čikobav a 1961, 1968a; Tschenkéli 1958), saying that Series I and II simply hav e different case marking rules; they hav e assumed a rule of inv ersion for Series III. By this approach, the three sentences (1)—(3) containing Class 1 v erbs and the three (7)—(9) containing Class 3 v erbs hav e the same “logical” (or “real”) subject, and (1)—(3) hav e the same logical direct object. (3) and (9) differ from the others in hav ing undergone inv ersion, so that their logical subjects are their “grammatical” (or “morphological”) indirect objects, and their logical direct objects, if they hav e them, are their grammatical subjects.

In the inv ersion example, (9), the notional subject, Merab, is in the dativ e case and conditions u-, a marker of third person singular indirect object agreement. The suffix -a in (3) marks third person singular subject agreement with γvino (‘wine’); in (9), where there is no ov ert final subject, the third person singular form appears by default.

77.  Georgian

Aronson (1970, 294), following Čikobav a (1968a), seeks a more formal approach that will accord with the analysis described abov e. He defines the subject as “that member of the predication which can be marked by the v erb for number in the third person”. This principle works well except in Series III, exactly where the grammatical relations are at issue. In Series III, dativ e nominals condition number agreement, except when a first or second person nominativ e nominal (notional object) is present.

Without further restrictions, the principle stated abov e has the unfortunate consequence of suggesting that ‘students’ is the subject of (10), but not of (11), since ‘students’ conditions agreement in the former, but not in the latter. While there is dialectal v ariation in agreement of the sort illustrated in (10) and (11), these represent the grammatical norm. Vogt (1971, 81) departs from the traditional v iew, defining the subject as that nominal that conditions one of the agreement affixes from the set called “subject markers” in Table 77.1. Consequently, in his v iew, ‘Merab’ would be subject in (1) and (2), but ‘wine’ would be subject in (3). Harris (1981) takes a different approach, identifying as subject those nominals that hav e a particular cluster of properties, including triggering Tav-Reflexiv ization and conditioning Subject Agreement (see also Boeder 1989, 168—172). In Series III, where this cluster of properties is div ided between two nominals, one is v iewed as the initial subject, while the other is a final subject resulting from the application of the rule of Unaccusativ e, applying with Inv ersion. According to this analysis, ‘Merab’ in (3) is the initial subject and final indirect object (henceforth “inv ersion nominal”). A similar approach is taken to the other relations. Thus, this analysis combines the insights of the

1381

traditional works and of Vogt (1971) by attributing different subject properties to different lev els of deriv ation. On this v iew, the grammatical relations that head the columns in Table 77.1 refer in v erbs of Classes 1 and 3 in Series III to initial, rather than final, status. Working within the framework of relational grammar, Harris (1981) v iews the initial indirect object marked with -tvis in Series III as a chomeur, as predicted by a univ ersal principle, The Chomeur Law (see article 27, “Relational Grammar”, in the first v olume, and section 1.5. below). The syntactic behav ior of this tvis-nominal had not prev iously been described. One adv antage of the analysis is that it accounts for this behav ior and permits generalization with nominals marked with -tvis in other constructions (see section 1.5. below). A problem with this analysis is that it makes reference both to initial subjecthood, in order to identify the trigger of Number Agreement, and to final indirect-objecthood, to identify the set of agreement markers used; this problem is taken up again in section 1.2.4. In addition to the three classes of v erb forms described in section 1.2.1., there is a Class 4, containing mostly affectiv e v erbs. These v erbs hav e a single case pattern in all three series, illustrated in (12) and (13) from Series I.

Most linguists hav e analyzed the Class 4 forms in all series in the same way they analyze the Class 1 and 3 forms in Series III. Class 4 forms meet the number agreement criterion of Chikobav a (1968a) and Aronson (1970), and they would see ‘Merab’ as the subject in (12). Class 4 v erbs hav e the same syntactic properties found by Harris for Class 1 and 3 inv ersion in Series III, and that analysis considers ‘Merab’ the initial subject and final indirect object, while ‘Natela’ is the initial direct object and final subject. Howev er, Šanize (1963) treats Class 4 differently from Series III; he sees ‘Natela’ as the subject in (12), objecting that its translation equiv alent in Russian is not sufficient reason for treating it as object. Merlan (1982) excludes from consideration the intransitiv es of the type illustrated in (13)

1382

and cites the existence of this type as grounds for rejecting analyses that inv olv e inv ersion, ev idently feeling that optional elements are inadmissible in syntactic rules. In this discussion, Merlan (1982, 298) does not address the fact that the Inv ersion rule in Harris (1981) affects only the initial subject and so applies in the same way to transitiv es and intransitiv es. She argues (Merlan 1982, 299) that the traditional analyses made by nativ e grammarians ov erlook the fact that there is a semantic difference between Class 4 and the other classes, and that because of this difference they must hav e a different structure. That article does not propose ways to account for the v arious syntactic traits that the inv ersion nominal shares with subjects, as established in Harris (1981). 1.2.4. How Complex is Agreement? Agreement may be regarded as a morphological process, but its relation to syntax has been one of the important theoretical issues discussed in recent years. As part of this debate, Anderson (1982; 1984a) supports the position that morphological operations depend upon syntactic information, and that morphology forms a distinct component that operates on the output of syntactic rules. He proposes to account for the fact that the first person subject marker v- does not cooccur with the second person object marker gwithout deletion rules of the sort proposed in Harris (1981, 31), which he finds implausible. He accomplishes this through the dev ice of so-called disjunctiv e ordering, which refers to the practice of arranging rules in blocks, where all the rules in a block apply disjunctiv ely. Jensen/Stong-Jensen (1984) hav e argued that this dev ice is too powerful and hav e demonstrated that these same Georgian facts can be accounted for through lexical morphology without the dev ice of disjunctiv e ordering and without deletion, but at the expense of accuracy. Observ ing that most of the subject properties of the inv ersion nominal are syntactic ones, while its indirect object properties are morphological ones, Anderson (1984b) proposes to account for the indirect object properties of this nominal directly through the morphology. On his analysis, ‘Merab’ in (9) is the subject in the syntax, and it is the indirect object in the morphology. At the center of the issue is the rule of Number Agreement for Series III as described in all of the handbooks of Georgian. In the modern language in

XXI. Syntaktische Skizzen

Series I and II, subjects are the only nominals that condition the rule; in Series III, the nominal marked in the dativ e case conditions Number Agreement (one of its subject properties), though the morpheme -t that marks it is otherwise used only for objects. This is the rule that Harris (1981) formulates as a global rule. Number Agreement is complicated still further by v ariation according to the person of the arguments inv olv ed, as illustrated in (10)/(11) in the preceding subsection. Anderson’s analysis circumv ents the problem that Number Agreement refers to two lev els of deriv ation by making reference instead to syntactic subjecthood, in order to identify the trigger of Number Agreement, and to morphological indirect-objecthood, to identify the set of agreement markers used. Aissen/Ladusaw (1988) address the question of whether certain complex types of agreement systems, for which multistratal analyses hav e been proposed, can be accounted for within monostratal theories. One of the complex agreement systems they consider is the Number Agreement described abov e for Modern Georgian. They propose a monostratal account of Number Agreement in Modern Georgian, based on a rule that refers to the inv ersion construction and on a conv ention that unmarked agreement (here Number Agreement conditioned by the subject) can be ov erridden by stipulating a marked agreement relation in the construction referred to. K’iziria (1985) has turned our understanding of Number Agreement in Modern Georgian upside down. Although it had long been known that indirect objects, and occasionally ev en direct objects, can condition number agreement in some non-standard dialects (Čikobav a 1968b; Harris 1985, 313 f), all nativ e grammarians and others had considered the system described in the preceding paragraphs to be the norm (eg. Normebi 1970, 182 f; Tschenkéli 1958, 484—490). K’iziria has now documented the existence of Number Agreement with objects outside the inv ersion construction in the literary language. Tuite (1987; 1988) surv eys Number Agreement in the Kartv elian languages with unprecedented completeness. He shows that agreement with indirect objects that are not initial subjects (that is, outside the inv ersion construction) is optional from a syntactic point of v iew, and he suggests that this sort of number agreement may depend upon pragmatic factors, such as topicality or empathy.

77.  Georgian

1383

1.3. Aspect

v olume). Both inv olv e rules that promote direct objects to subjects; they differ in that the passiv e applies to transitiv e structures, while the unaccusativ e applies to a direct object in the absence of a subject. Both constructions are found in Georgian; (14) and (15) illustrate the passiv e, formed analytically from a past passiv e participle and an auxiliary.

Mač’av ariani (1974) has studied morphological aspect in Georgian and has shown that the modern language distinguishes the following categories: perfectiv e/imperfectiv e, punctual/durati v e, and one-time/repeated. Holisky, in a series of works, has shown that in addition to these categories, lexical aspect plays an important role in the language. Holisky (1978) establishes diagnostics for stativ ity in Georgian and relates these to univ ersals of language. Georgian also distinguishes the category telic/atelic, where a v erb is telic if it is marked for hav ing an end point (Holisky 1979). Holisky (1981b) explores the relationship between morphological and lexical aspect. Following studies of English aspect by Vendler (1967) and more recent work by Dowty (1977), she shows that aspect in Georgian is better understood through the use of four categories, illustrated below. Accomplishments: Activities: States: Achievements:

gacmenda ‘he cleaned it’ itamaša ‘he played’ civa ‘it is cold’ ip’ova ‘he found it’

The illustrations prov ided here hav e the imputed property inherently, but deriv ation can alter these. Holisky (1981a) relates lexical aspect to morphological and syntactic categories, as briefly summarized in section 1.2.1. 1.4. Voice Some linguists, including Schuchardt (1895), hav e proposed that languages with ergativ e case marking — here they include Georgian Series II (see section 1.2.1. abov e) — are characterized by structures in which all transitiv e v erbs are passiv e. Čikobav a (1942; 1961) has argued effectiv ely that neither Georgian nor other languages with ergativ e case marking, including especially other languages of the Caucasus, are accurately characterized in this way. He observ es that a passiv e, a marked syntactic construction, cannot exist in the absence of a corresponding unmarked activ e. Although it is now known that ergativ e case marking in some languages does originate through a reinterpretation of passiv es, it is doubtful that the Kartv elian family, to which Georgian belongs, can ev er be reconstructed far enough back to determine whether the case marking of Series II originated in this way. Relational grammar distinguishes a passiv e construction from an unaccusativ e (see article 27, “Relational Grammar”, in the first

Syntactically, both examples inv olv e initial direct objects (varaudi ‘conjecture’ and ‘I’) that are promoted to subjecthood. The deletion of unemphatic pronouns, as in (15), is frequent in Georgian; in this example, the person and number of the nominal are indicated by agreement in the auxiliary. Ev idence for the initial relations are based on suppletion phenomena; while case marking, v erb agreement, and syntactic rules that are outside the scope of the present chapter support the claim that these are final subjects (Harris 1981, 104—109). Both examples also inv olv e initial subjects; in (14) this is čven (‘we, us’), and in (15) it is unspecified. When the initial direct object is promoted to subject, the initial subject becomes a chomeur. In Georgian this is marked with the postposition mier (‘by’); transitiv e subject chomeurs are also marked with mier in dev erbal nominal constructions (see section 1.5. below). Unaccusativ es, formed synthetically with the prefix i- or its v ariants, are illustrated by (5) v abo e and by (16).

1384

In (16), unlike (15), there is no implied agent, no unspecified subject. Ev idence that the final subject, in this instance orive porma (‘both forms’), is the initial direct object is drawn from a wide v ariety of phenomena, including case marking, suppletion, prev erb alternation, inv ersion, and — in Old Georgian — plural agreement (Harris 1981, Ch. 13; Harris 1982). Both the passiv e and the unaccusativ e are Class 2 v erb forms. Some v erbs hav e both analytic passiv e forms and synthetic unaccusativ e forms with distinct syntax, semantics, and functions, thus prov iding a clear basis for distinguishing between these (examples in Harris 1981, 192 f). On the other hand, there are synthetic forms with the i- formant which hav e the syntax and semantics of the passiv e; that is, they permit agentiv e subjects marked with the postposition mier. Thus, there is not a one-to-one correspondence of morphology and syntactic construction with regards to unaccusatives and passives. 1.5. Other Phenomena Involving Grammatical Relations in Simple Sentences Version in Georgian refers to syntactic processes that create possessiv e, benefactiv e, or other indirect objects and to the morphological marking of this on the finite v erb form. Version has been studied by Boeder (1968) within a traditional framework, and an analysis within the framework of relational grammar is proposed in Harris (1981, Ch. 6). It has been suggested that causativ es in Georgian are deriv ed from a biclausal underlying form, in which the matrix clause contains an abstract predicate CAUSE, which is realized as the causativ e suffix -ev/ -evin (Harris 1981, Ch. 5). Dav ies/Rosen (1988) hav e shown, howev er, that the same range of facts can be accounted for in a monoclausal analysis for Georgian and other languages. Work within the framework of relational grammar rev ealed for the first time that there exists a category of retired term in Georgian, the occurrence of which is predictable on the basis of univ ersal principles (see article 27, “Relational Grammar”, in the first v olume). In Georgian, retired terms occur in a number of disparate constructions, including inv ersion, analytic passiv es, causativ es, infinitiv es,

XXI. Syntaktische Skizzen

and other non-finite v erb forms. Syntactically, retired terms are characterized by the fact that they behav e like initial terms (initial subjects, initial direct objects, or initial indirect objects) with respect to those phenomena that are stated on initial termhood, while they behav e like non-terms with regard to those rules that are sensitiv e to final termhood. In Georgian, retired transitiv e subjects are marked with the postposition mier, retired intransitiv e subjects and direct objects with the genitiv e case alone, and retired indirect objects with the postposition tvis. The remarkably regular marking of retired terms is illustrated below by retired indirect objects from v arious constructions (see Harris 1981, Ch.11 et passim). The reader should bear in mind that final indirect objects are not marked with tvis, but with the dativ e case, in all series, as for example in (17a). (On the analysis presented here, dev eloped more fully in Harris 1981, in Series III it is the initial subject of v erbs of Classes 1 and 3 that is the final indirect object. In this Series the tvismarked nominal is the retired indirect object.) (17b) is an inv ersion construction, with a retired indirect object; the corresponding (17a) in Series II has a final indirect object and no retired terms.

77.  Georgian

1385

2.

Complex Sentences

Relativ e clauses are discussed in section 2.1., complement clauses in 2.2., and adv erbial clauses in 2.3. 2.1. Relative Clauses

One of the contributions of relational grammar has been the prediction of retired terms according to univ ersal principles and the study of their language-particular characteristics, such as this highly regular marking in Georgian. 1.6. Formation of Content Questions In Georgian, question words, such as vin (‘who’), ra (‘what’), rodis (‘when’), sad (‘where’), rog or (‘how’), and rat’om (‘why’), occupy focus position, immediately preceding the finite v erb. Since the v erb itself may occupy a v ariety of positions, the question word may be sentence-initial, as in (22b), or internal, as in (22a). (22c) is ungrammatical because the question word is separated from the v erb.

In Georgian, question words cannot mov e outside the clauses in which they originate, as illustrated in (23).

Thus, although we must assume a mov ement rule to account for the placement of question words within their clauses, this rule differs from familiar WH-Mov ement in that it does not mov e the word to initial position and in that it does not cross clause boundaries. Additional details and exceptions to the generalizations stated here are discussed in Harris (1984).

Early subsections here describe v arious types of finite relativ e clauses (2.1.1.), the possibility of extraposing them (2.1.2.), the relations relativ izable (2.1.3.), and structural differences between restrictiv e and non-restrictiv e relativ es (2.1.4.). Headless relativ es are described in subsection 2.1.5., and participles in 2.1.6. 2.1.1. Structural Types Modern Georgian employs a v ariety of strategies in the formation of relativ e clauses. In the written language there is a strong preference for relativ es formed with a relativ e pronoun (romelic ‘which’, vinc ‘who’, sadac ‘where’, etc.), as in (24) and (25).

The relativ e pronoun always occupies clauseinitial position, as in these examples. I assume a mov ement rule to account for this, but note that the rule for question words in Georgian is not comparable (section 1.6 above). In Modern Georgian the case of the relativ e pronoun is determined strictly by features of the relativ e clause: the grammatical relations of the relativ e pronoun in its clause, the class and series of the v erb. The case of the head is determined by the same features of its own clause. The number of the pronoun is generally the same as that of the head noun, as in (26).

1386

XXI. Syntaktische Skizzen

Masdars (dev erbal nouns) and abstract nouns are not ordinarily put in the plural; but when they hav e plural meaning in context, relativ e pronouns referring to them may be in the plural, as in (28).

particle is enclitic to the full NP. The particle may be attached to a postposition, as in romel-tana-c (which-with-Prt, ‘with whom’) (see Aronson 1972, 140; Ӡiʒiguri 1973, 252 for examples). It is entirely possible, too, for the particle to be omitted when the relativ e pronoun is only one constituent of an NP within its clause; this is illustrated in (33) in the next subsection. Since in structures of the type illustrated in (24)—(28) the relativ e clause follows the head noun and makes use of a relativ e pronoun, in what follows the structure is termed the (postnominal) relative pronoun strategy. All of the remaining types, seldom found in written Georgian but v ery common in the spoken language, make use of a particle roml ro (‘that’). The relativ izer rom (unlike the complementizer of the same form) generally occurs in the ‘floating’ position; that is, it occurs between the first constituent and the v erb of its clause. In some instances it can be clause-initial (for examples see Harris to appear). The first of the new types of relativ e clauses, referred to here as the postnominal gap strategy, is illustrated in (29). In this example and those that follow in this subsection, the relati v e clause is bracketed.

The head noun sakmianoba is in Georgian an abstract noun, formed with a suffix of abstract nouns, -oba, and hav ing a meaning more precisely equiv alent to “busy-ness”. In instances of this sort, the relativ e pronoun is thought to agree with the semantic plurality of the head noun, rather than with its grammatical number. The particle -c(a) marks relativ ized NPs or NPs containing relativ ized constituents, and it differentiates a relativ e pronoun (eg. romeli-c ‘which’) from the corresponding interrogativ e pronoun (eg. romeli ‘which?’). The particle is enclitic to the entire NP of which the relativ e pronoun is a constituent. In (24) and (26)—(28) the pronoun itself constitutes the entire NP; in (25), on the other hand, the pronoun is a constituent of the NP romlis šemasmenel-i-c (‘whose predicate’), and the

In a gapped relative clause, the relative nominal is not ov ertly represented in its clause; in the postnominal gap type, the relativ e clause follows its head. The head noun bears the case required by its function in the main clause, and it is this fact which permits us to see the structure of the clause clearly. For example, in (30), the NP erti matg ani (‘one of them’) would be in the dativ e case, which would be required by the v erb of its clause, akvs (‘has’), if it were a constituent of that clause. In addition to postnominal relati v e clauses, Modern Georgian has relativ e clause

Howev er, under a v ariety of circumstances a plural head noun may be followed by a relativ e pronoun that is singular in form; this happens especially when the relativ e word is inanimate as in (27) (K’v ant’aliani 1983, 71 ff).

77.  Georgian

1387

types that precede the head noun. We may consider first the prenominal gap strategy, illustrated in (31).

Like clauses produced with the postnominal gap strategy, this clause type lacks ov ert representation of relativ e nominal and contains the particle ro(m). The prenominal gap type generally precedes not only the head noun, but the entire main clause, as in this example. The final major relativ e clause type is the (prenominal) non-reduction strategy, in which the relativ e nominal is represented as a noun, rather than being ‘reduced’ to a pronoun or gap. The main clause contains a pronoun (or in (36) an adv erb) coreferential with the relativ e nominal; I assume that this pronoun acts as head.

2.1.3. Grammatical Relations Relativizable

2.1.2. Extraposition The relativ e clause may be separated from its head or pronominal representation in the matrix clause with all types of relativ ization, as illustrated in (33)—(36). In these examples the demonstrativ e ‘that’ is glossed ‘that(Dem)’ to distinguish it from the subordinating particle ‘that’.

With all relativ e clause strategies, a wide v ariety of grammatical relations are relativ izable. For the relativ e pronoun strategy, a range of possibilities is illustrated in the two preceding subsections, including subject, possessor, and object of a postposition; although not illustrated here, all other noun functions can be relativized with this strategy. With the postnominal and prenominal gap types, subjects and objects can clearly be relativ ized. Many informants also permit relativ ization of objects of postpositions, as long as the relation of the gapped nominal within the relativ e clause is clear from the context. More restrictiv e informants do not like such structures, though they do occur in spoken language.

XXI. Syntaktische Skizzen

1388

v stricti e

Other interpretations of sentence (38) are possible, as described below in connection with adverbial clauses. 2.1.4. Restrictive and Non-Restrictive Relatives Restrictiv e relativ e clauses may optionally be distinguished from non-restrictiv e ones by inclusion of the distal demonstrativ e pronoun before the head noun. (39) prov ides an illustration with the postnominal gap type.

A more literal translation of (39) would be ‘... not that fox ...’ This sort of construction is also used with the relativ e pronoun strategy (for examples, see Ӡiʒiguri 1973, 332 f). In the spoken language, some non-restrictiv e relativ es formed with a relativ e pronoun may sound artificial, as does (40).

one.

2.1.5. Headless Relatives The structural types discussed abov e hav e in common that they modify a head, at least in abstract structure. In Georgian there are also so-called headless relativ es that modify no head in the matrix clause. These are formed primarily with relativ e pronouns and hav e the characteristics noted abov e for this clause type; an example is (43).

Such clauses hav e general meanings, ‘a person who’, ‘someone who’, ‘a place where’, etc. It is possible for a headless relativ e to hav e a correlativ e pronoun, as shown in (44).

The structure in (44) then resembles a prenominal non-reduced structure, but differs from it (i) in hav ing a relativ e pronoun, and (ii) in lacking rom. A more natural counterpart of (40) is (41).

Howev er, in the written language the relativ e pronoun strategy is certainly used for both restrictiv e and non-restrictiv e v arieties. (27) is a clear example of a restrictiv e relativ e, and (42) prov ides an illustration of a non-re-

2.1.6. Participles In traditional Georgian linguistics, discussion of relativ e clauses is usually restricted to finite clauses, but recent works on other languages often include non-finite clauses as relativ es. Accordingly, participles are briefly described here. Future participles, present participles, and past passiv e participles can all be used as prenominal modifiers, but it is the last-named type that is most likely to inv olv e additional constituents, as in (45).

77.  Georgian

It would be possible to express the same meaning with a relativ e clause with the head masala-s; in that case, the relativ e clause could hav e either an activ e or a passiv e structure. With the participle, only the passiv e structure can occur. The participial clause may be compared with the passiv e finite clause (46) (see abov e for the structure of passi es).

In (46) the passiv e form of the v erb consists of an auxiliary and a past passiv e participle; when the participial clause occurs as a prenominal modifier, it is non-finite and accordingly lacks the auxiliary. In (46) the participle has the suffix -i of the nominativ e case; in (45) its form agrees with the dativ e case of its head, though this agreement is only minimally indicated (see Tschenkéli 1958, 38). The noun relativ ized, in this case masala (‘material’), is gapped from the participial clause. Other features of finite passiv es, including the optional agent, are present also in the reduced participial clause, as in (46).

1389

In (47) the clause beginning rom is the subject of cnobilia (‘it is known’). In (48) the dependent clause is the direct object of vpikrobt (‘we think’). 2.2.2. Complement Types Modern Georgian has the following types of complementation: indicativ e (Ind), subjunctiv e (Sbj), and dev erbal noun (masdar). Infinitiv es occur in Georgian, but their use is limited to purpose clauses and object raising constructions. Noonan (1985) has classified complement-taking predicates into semantic types and has made cross-language predictions about the sort(s) of complements they will permit. Vamling’s (1987) research on Georgian complementation on the whole supports Noonan’s predictions; the choice of indicativ e v s. subjunctiv e form, for example, is usually determined by the predicate of the main v erb, as shown by the following examples from Vamling (1987, 33 f).

2.2. Complementation 2.2.1. Grammatical Relations Sentential subjects and sentential direct objects are common in Georgian, as illustrated below. Complement clauses may also bear oblique grammatical relations; this is discussed in subsection 2.2.4.

The list of predicates taking subjunctive complements promises to be v ery long, for it includes not only the v erbs listed by Noonan (1985), but complex predicates like mizani-a (‘the goal is’), mtavari-a (‘the main [thing] is’), and the predicate in the main clause of (51), where the complement clause is gvepikra in the subjunctive.

1390

Masdars (Mas) are often possible where either indicativ e or subjunctiv e may occur, as in (52)—(53), likewise from Vamling (1987, 33 f).

Vamling notes that the semantic categories cannot fully predict the complement type; a particular problem is presented by the two sets of predicates of fearing and of negativ e propositional attitude; predicates of both of these groups permit complements of all three types. The sentences below (from Vamling 1987, 37 f) illustrate the possibility of taking either indicativ e (54) or subjunctiv e (55).

The use of the subjunctive in complements in Georgian was found to be related primarily to determined time reference and irrealis states and ev ents. Vamling (1987, 53—55) found that complement clauses could occur as masdars with those main clause v erbs that had noun indirect objects, while the complement must occur as a finite clause with main clause verbs that had noun direct objects. 2.2.3. Complementizers In terms of frequency of occurrence and flexibility of use, rom (‘that’) is the most important complementizer in Modern Georgian. As it occurs in almost ev ery kind of dependent clause in Georgian, rom itself does not indicate the type of dependent clause but does serv e to mark the dependent status of the clause (although it can occur in certain kinds of questions also). In noun clauses, rom occurs with both sentential subjects and sentential objects. With indirect questions, the complementizer tu (‘whether’) occurs, as in (56); if the embedded question is a content question,

XXI. Syntaktische Skizzen

tu may occur together with the question word or may be absent, as in (57). Additional conditions on the occurrence of tu are noted in Harris (1984).

Asyndetic constructions are also found, where a dependent clause is joined to a matrix clause without an ov ert conjunction. In some cases, the conjunction seems to hav e been deleted, as in (58b), which seems to be from (58a) (Basilaia 1974, 30).

In some cases, it would be stylistically awkward to hav e a conjunction, as in (51) abov e, where rom added to introduce the clause g vepikra would be infelicitous because of the rom which introduces the next clause; examples of this kind are numerous in my collection. Basilaia (1974, 30 ff) has argued that omission of the complementizer is not determined by the potential ambiguity of the resulting sentence. Since sev eral conjunctions, especially rom, are used in multiple functions, they do not necessarily disambiguate the relation between the main and dependent clauses. Therefore, ev en sentences in which a conjunction does occur may hav e an unclear relationship between the clauses, and this cannot be the determining factor in the omission of the complementizer. In some examples, v erb tense and mood help to disambiguate the relation between the two conjunctionless clauses; in other examples the correlativ e does this, as in (59) (Basilaia 1974, 38).

77.  Georgian

In (59) the use of the subjunctiv e mood in the predicate of the embedded clause (viq’o ‘I would be’) contributes to marking this as the dependent clause. The boundary between the two clauses is primarily marked by intonation, though v erb-final word order in the first clause may be an added clue. (Note that pitch of the v oice, in addition to the pause between the clauses, contributes to the intonational marking; Basilaia 1974, 39 ff.) Finally, the correlativ e, by the case it bears, makes it clear that the dependent clause is the predicate nominal. 2.2.4. Correlatives A v ariety of pronouns occur in main clauses as correlativ es, but the most important are the demonstrativ es — the proximate es, the distal igi, and the remote is. When a correlativ e is used with a complement clause, the full range of complementizers is av ailable: rom ‘that’, as in (60), an interrogativ e pronoun with or without tu (‘whether’), as in (61), or no complementizer, as in (59) abo v e.

In (60) the subject pronoun, es (‘this-Nom’), refers to what has preceded; the correlativ e pronoun, imas (‘that-Dat’) refers to the dependent clause that follows. It is the contrast between the proximate form (es) and the remote (is, here in the dativ e case, imas) that rev eals which pronoun refers forward and

1391

which backward; that is, in Georgian discourse proximate forms generally refer to what has come before, while remote forms refer to what is to come. (This generalization, pointed out to me by Nani Č’anišv ili, holds true of the modern written technical language. Nev ertheless, counterexamples exist in literature not far remov ed in time or place; see Ӡiʒiguri 1973, Ch. 6, for a sample.) The case forms of the pronouns rev eal the grammatical relations they, together with the propositions they refer to, bear in the main clause. In this they follow the general rules stated abov e in this chapter; as the v erb here is in a Series I form, the nominativ e pronoun refers to the logical subject of the clause, while the pronoun in the dativ e refers to the logical direct object. The v erb akvs (‘has’) in the main clause of (61) is an inv ersion v erb; the correlativ e imas and the complement clause represent the initial subject (final indirect object) of this v erb. The complement is an indirect question and, as such, has a question word, here with tu (‘whether’) as the complementizer. Most often the correlativ e precedes its dependent clause, as in (60)—(61), but the opposite order is also found, as in (62).

(62) illustrates a sentential subject, with a correlativ e in the proximate form, referring to what precedes it, and in the nominativ e case, representing its role as subject in the main clause. Although the translation makes use of a passiv e for greater naturalness in English, the v oice of the main clause in Georgian is active, not passive. It is possible for the complement clause to be separated from its correlativ e pronoun, as in (63).

XXI. Syntaktische Skizzen

1392

and

Again it is imas, the remote correlative in the dativ e case, that refers to the clause that follows and indicates that it bears the relation of direct object in the main clause. While the subject and direct object are the grammatical relations most often borne by complement clauses, with a correlativ e in the main clause it is possible for a dependent clause to bear an oblique relation. For example, in (64), mdg omareobs is an intransitiv e v erb which may occur with an oblique nominal marked with the postposition -ši (‘in’); here the complement clause bears that oblique relation, and the correlativ e is marked with -ši.

In (65) and other examples like it, the complement clause and its correlativ e pronoun represent the initial direct object which is, howev er, not a direct object in final structure.

correlati v es.

Two

examples

follow.

The demonstrativ e pronoun is in (66) is ev idently functioning as a correlativ e, ev en though here it modifies a noun pakt’i and in v pre ious examples did not.

Although the v erb of the complement clause in (67) is translated with an infinitiv e, it is a finite form in the indicativ e mood. In all examples in my collection, the complement of pakt’i (‘fact’) is in the indicativ e, as it is in (67); complements of sapuʒveli (‘basis’), varaudi (‘supposition’), and sabuti (‘ev idence’), on the other hand, are in the subjunctiv e. It should be noted that, although this construction is here considered a complement, it is structurally highly similar to the relativ e clause of the postnominal gap type. 2.3. Adverbial Clauses

It is entirely regular for the retired direct object of infinitiv es of purpose to be in the genitiv e case (Harris 1981, 155—157), as this one is. In addition to the types described abov e, Georgian has so-called appositiv e clauses or noun complements. These seem to occur in or with the full range of grammatical relations, complement types, complementizers,

Georgian has a wide v ariety of adv erbial clauses, including clauses of time, location, manner, purpose, reason, condition, and concession, not all of which can be illustrated in this brief sketch. They may be set off by one of three dev ices: (i) a subordinating morpheme, (ii) one of two non-finite v erb forms that are restricted primarily to adv erbial clauses, or (iii) intonation (punctuation) alone. 2.3.1. Subordinating Morphemes Most types can be constructed either (i) with a subordinating conjunction that expresses lexical content and is specific to the particu-

77.  Georgian

lar type, such as roca (‘when’), particular to the adv erbial clause of time, or tu (‘if’), particular to the conditional, or vinaidan (‘because’), particular to the clause of reason, or (ii) with the grammatical morpheme rom, which has no lexical meaning (see Thompson/ Longacre 1985, 172). Because of the large number of conjunctions av ailable, one can express subtleties of interclausal meaning-relations through the conjunction. An example of the expression of similar meanings with a lexical or grammatical subordinator is giv en in (68) and (69).

1393

2.3.2. Special Verb Forms At least two non-finite v erb forms can signal specific adv erbial clauses. The synchronic infinitiv e (Inf) is usually called by a term that describes its diachronic origin — the masdar in the adv erbial case — ev en in works where it is explicitly described as an infinitv e. The -isas form does not hav e a commonly accepted name at all. Their synchronic syntax has been little discussed (see 2.4. below on this), though they hav e often been treated from a diachronic point of view. Use of the infinitiv e in Modern Georgian has been described in only two constructions — in adv erbial clauses of purpose and in object raising constructions (Harris 1981, 154— 156, 53 f); it is only the first of these that concerns us in this subsection.

With either v ariant, the ‘when’ clause may optionally precede or follow the main clause. With this ‘when’ type, as with most others, a correlativ e may optionally be used, as in (70).

The correlativ e opens up the possibility of center-embedding the adv erbial clause, as in (71).

The infinitive generally lacks overt expression of a subject, but in the following example gamosaxat’avad (‘express/Inf’) does hav e a nominal (retired) subject, marked here by the postposition mier.

‘I will ask you for the bill for this when it [the bill] comes.’ (Gr. Orbel., cited by Ӡiʒiguri 1973, 338) The grammatical subordinator rom may also be used with a correlativ e, and with amit’om, (i)mit’om (‘because of this, because of that’) is nearly always used, as in (72). The dependent clause is bracketed. Note that all terms of an infinitiv e are retired terms

1394

XXI. Syntaktische Skizzen

(Harris 1981, 165—167), and retired transitiv e subjects are regularly marked by mier (see section 1.5. above and Harris 1981, 171). A second type of adv erbial clause uses a v erb form in -isas; this type expresses simultaneity. Examples are giv en in (75) and (76).

2.4. Shared Syntax of Dependent Clauses

In all the examples I hav e collected of the -isas construction, the subject is omitted and is understood as ‘one’ or as ‘we’ or ‘you’ in their general senses. Note that, as in these two examples, it is not necessary that the subject of the -isas form be the same as a constituent in the main clause. The English translation of (75), considered ungrammatical by many, in popular usage has this same property. 2.3.3. Intonation Alone Basilaia (1974, 46—54) has pointed out that adv erbial clauses may occur as true dependent clauses without any conjunction. Their dependent status is marked by intonation (or punctuation) and, in most cases, by use of a subjunctiv e (Sbj) v erb form. A wide v ariety of adv erbial clause types can be so marked, including temporal, causal, purpose, manner, conditional, and result clauses. An example is v gi en below in (77).

The syntax of finite dependent clauses of all three major v arieties is essentially like that of main clauses: there is no special word order, agreement, or case marking used for dependent clauses. Although the subjunctiv e might be considered a special morphology associated with dependency, the subjunctiv e occurs also in main clauses, while dependent clauses may hav e other forms, including the indicative. On the other hand, non-finite clauses, which occur only as dependent clauses, hav e certain syntactic characteristics that set them apart. First, unlike finite clauses, non-finite ones hav e no agreement to encode subject, direct object, or indirect object. Second, the range of syntactic constructions possible is reduced in non-finite clauses. For example, causativ es occur in non-finite forms with only a few specific v erbs; passiv e and inv ersion constructions do not occur at all in masdars or forms derived from them. Third, the marking assigned to subjects and objects with non-finite forms is different from that assigned with finite forms. In particular with non-finite forms, subjects of intransitiv es and direct objects are marked with the genitiv e case, subjects of transitiv es are marked with the postposition mier, and indirect objects are marked with the postposition -tvis (both postpositions gov erning the genitiv e case). It is noteworthy that the same marking is found with participles used as relativ e clauses (see (45)), with masdars as complement clauses (see (52)), with infinitiv es as purpose clauses (see (73)—(74)), and with the -isas forms based historically on masdars (see (75)—(76)). Examples of all possibilities can be found in Harris (1981, Chapters 10 & 11) and Lacabiʒe (1975). The unmarked word order of non-finite clauses is the same, mutatis mutandis, as that of finite clauses. For example in (74) the retired subject, ori c’inadadeba (‘two clauses’) is the first constituent, the retired direct object movlena-ta an mokmedeba-ta dap’irisp’irebis (‘an opposition of phenomena or of actions’) is second, and the v erb g amosaxat’avad (‘to

77.  Georgian

express’) is clause-final. We may observ e that, loosely speaking, this is the order of noun phrases as well, in the sense that genitiv e attributes and other modifiers precede the head. With respect to the presence v s. absence of arguments, different kinds of clauses present v arious pictures. In relativ e clauses of the postnominal relativ e pronoun type and prenominal non-reduction type, in complement clauses of all types, and in adv erbial clauses formed with subordinating morphemes, all arguments of the v erb may be present. In preand postnominal relativ e clauses of the gap type, the one constituent that is the same as the head must be gapped. In infinitiv es the subject is ordinarily deleted on coreference with a term argument of the matrix clause, but in (74) the infinitiv e has a subject. Forms in -isas ev idently do not permit ov ert expression of a subject. In the sections abov e, relativ e clauses, complement clauses, and adv erbial clauses are distinguished according to function in the usual way (cf. Keenan 1985; Noonan 1985; and Thompson/Longacre 1985). An exception is the so-called headless relativ e, which functions as a nominal. It is difficult, howev er, to find a formal criterion in Georgian which would group them in this way. We hav e noted already that the conjunction rom is used to signal dependent clauses of all three major v arieties. Correlativ es, too, occur with all three, as illustrated abov e. There are clauses that would be considered complements or adv erbials by functional criteria, yet which hav e the formal marking -c associated primarily with relativ es. (78) illustrates a noun complement with the “relativ e” pronoun ra-c; (79) illustrates a clause of location with the “relativ e” pronoun sada-c (see also Thompson/Longacre 1985, 183).

1395

The temporal clause in (68) also contains a conjunction rodesa-c (‘when’) that has the enclitic particle that characterizes relativ e words. In conclusion, there does not seem to be any formal marking or syntax that characterizes any one of the major types of dependent clause (relativ e, complement, adv erbial) to the exclusion of the others. The particle -c marks relativ e words; but the relativ e clauses in (29)—(32) (and others) lack this marking, while the adv erbial clauses in (68) and (78)— (79) hav e it. Some relativ e clauses are characterized by gaps, but others hav e relativ e pronouns or otherwise lack gaps. Masdars may be limited to complements, but the complement clauses in (47)—(51) are finite. Infinitiv es and -isas forms are limited to certain adv erbial clauses, but not all adv erbial clauses hav e them. All or most semantic types of adv erbial clause may be marked with contentful subordinating conjunctions, such as tu (‘if’); but adv erbial clauses can be marked by other means instead. In short, while we may distinguish these three major types by function, we cannot find formal correlates of these functions. (Some of the data adduced in the present paper were elicited in the USSR in 1975 on a research trip supported by the International Research and Exchanges Board.)

3.

References

Anderson, Stephen R. 1982. Where’s morphology? Linguistic Inquiry 13.571—612. —. 1984a. Rules as “morphemes” in a theory of inflection. 1983 Mid-America Linguistics Conference papers, 3—21. Boulder. —. 1984b. On representation in morphology: Case, agreement, and inv ersion in Georgian. Natural Language and Linguistic Theory 2.157—218. Apridoniʒe, Šukia. 1986. Sit’q’v atganlageba axal kartulši [Word order in modern Georgian]. Tbilisi. Aronson, Howard I. 1970. Towards a semantic analysis of case and subject in Georgian. Lingua 25. 291—301. —. 1972. Some notes on relativ e clauses in Georgian. The Chicago which hunt (Papers from the Relativ e Clause Festiv al), 136—143. Chicago Linguistic Society. Aissen, Judith L., and William A. Ladusaw. 1988. Agreement and multistratality. Papers from the 24th Annual Regional Meeting of the Chicago Linguistic Society, Part 2, 1—15. Chicago Linguistic Society. Basilaia, N. 1974. Uk’av širo rtuli c’inadadeba [The conjunctionless complex sentence]. Tbilisi.

1396

Boeder, Winfried. 1968. Über die Versionen des georgischen Verbs. Folia Linguistica 2.82—152. —. 1979. Ergativ e syntax and morphology in language change: The South Caucasian languages. Ergativity, ed. by Frans Plank, 435—80. New York. —. 1989. Verbal person marking, noun phrase and word order in Georgian. Configurationality, ed. by László Marácz & Pieter Muysken, 159—84. Dordrecht. Čikobava, Arnoldi. 1942. Ergat’iuli k’onst’rukciis p’roblemisatv is k’av k’asiur enebši: am k’onst’rukciis st’abiluri da labiluri v ariant’ebi [On the problem of the ergativ e construction in Caucasian languages: Stable and labile v ariants of this construction]. Enimk’is moambe 12.221—39. —. 1961. Ergat’iuli k’onst’rukciis p’roblema iberiul-k’av k’asiur enebši II [The problem of the ergativ e construction in Ibero-Caucasian languages]. Tbilisi. —. 1968a. Mart’iv i c’inadadebis p’roblema kartulši, I [The problem of the simple sentence in Georgian]. 2nd edn. Tbilisi. —. 1968b. Mart’iv i c’inadadebis ev oluciis ʒiritadi t’endenciebi kartulši [Basic tendencies of the ev olution of the simple sentence in Georgian]. Mart’iv i c’inadadebis p’roblema kartulši I, 269—80. Tbilisi. (First published, 1941.) Davies, William, and Carol Rosen. 1988. Unions as multi-predicate clauses. Language 64.52—88. Dowty, David. 1977. Toward a semantic analysis of v erb aspect and the English imperfectiv e progressive. Linguistics and Philosophy 1.45—77. Ӡiʒig uri, Šota. 1973. K’av širebi kartul enaši [Conjunctions in the Georgian language]. Tbilisi. Ertelišvili, P. 1963. Rtuli c’inadadebis ist’oriisatv is kartulši I: hip’ot’aksis sak’itxebi [On the history of complex sentences in Georgian I: Questions of hypotaxis]. Tbilisi. Fähnrich, Heinz. 1967. Georgischer Ergativ im intransitiv en Satz. Beiträge zur Linguistik und Informationsverarbeitung 10.34—42. Fillmore, Charles J. 1968. The case for case. Univ ersals in linguistic theory, ed. by Emmon Bach & Robert T. Harms, 1—88. New York. Harris, Alice C. 1981. Georgian syntax: A study in relational grammar. Cambridge. —. 1984. Georgian. Interrogativ ity (Typological studies in language 4), ed. by William S. Chisholm, Jr., Louis T. Milic & John A. C. Greppin, 63— 112. Amsterdam. —. 1985. Diachronic syntax: The Kartv elian case (Syntax and semantics 18). New York. —. To appear. On the history of relativ e clauses in Georgian. Proceedings of 1985 Conference on the Non-Slav ic Languages of the USSR (Folia Slav ica), ed. by Howard I. Aronson. Holisky, Dee Ann. 1978. Stativ e v erbs in Georgian and elsewhere. The classification of grammatical

XXI. Syntaktische Skizzen

categories, ed. by Bernard Comrie. International Review of Slavic Linguistics 3.139—62. —. 1979. On lexical aspect and v erb classes in Georgian. The elements (Papers from the Conference on Non-Slav ic Languages of the USSR), 390—401. Chicago Linguistic Society. —. 1981a. Aspect and Georgian medial v erbs. Delmar, NY. —. 1981b. Aspect theory and Georgian aspect. Tense and aspect (Syntax and semantics 14), ed. by Phillip J. Tedeschi & Annie Zaenen, 127—44. New York. Jensen, John T., and Marg aret Stong -Jensen. 1984. Morphology is in the lexicon. Linguistic Inquiry 15.474—98. Keenan, Edward L. 1985. Relativ e clauses. Language typology and syntactic description II: Complex constructions, ed. by Timothy Shopen, 141— 70. Cambridge. K’iziria, A. 1985. Obiekt’is mier zmnis šetanxmeba mrav lobit ricxv ši tanamedrov e kartulši [Agreement of the object with the v erb in plural number in Modern Georgian]. Iberiul-k’a v k’asiuri enatmecniereba 24.100—11. K’vant’aliani, L. 1983. Misamarti sit’q’v isa da mimartebiti nacv alsaxelis šetanxmeba kartulši [Agreement of relativ e words and relativ e pronouns in Georgian]. Tbilisi. Lacabiʒe, Lia. 1975. Natesaobiti brunv is pormata semant’ik’uri da sint’aksuri k’av širi zmnastan tanamedrov e kartulši [The syntactic and semantic relation of genitiv e case forms with the v erb in Modern Georgian]. Macne 4.150—57. Mač’avariani, Givi. 1974. Asp’ekt’is k’at’egoria kartv elur enebši [The category of aspect in the Kartv elian languages]. Kartv elur enata st’rukt’uris sak’itxebi 4.118—41. Merlan, Francesca. 1982. Another look at Georgian “inv ersion”. Papers from the Second Conference on the Non-Slav ic Languages of the USSR. Folia Slavica 5.294—312. Noonan, Michael. 1985. Complementation. Language typology and syntactic description II: Complex constructions, ed. by Timothy Shopen, 42— 140. Cambridge. (Normebi) 1970. Tanamedro v e kartuli salit’erat’uro enis normebi [The norms of the Modern Georgian literary language]. Tbilisi. Nozaʒe, L. 1974. Medioakt’iv zmnata c’armoebis zogi sak’itxi kartulši [Some questions of the formation of medioactiv e v erbs in Georgian]. Iberiulk’avkasiuri enatmecniereba 19.25—51. (Šaniʒe) Chanidzé, Akaki. 1963. Le sujet grammatical de quelques v erbes intransitifs en géorgien. Bulletin de la Société de Linguistique de Paris 58.1—27. Sapir, Edward. 1917. Rev iew of Het passiev e karakter v an het v erbum transitiv um of v an het v er-

78.  Swahili

bum actionis in talen v an Noord-Amerika, by C. C. Uhlenbeck, 1916. International Journal of American Linguistics 1.82—6. Schuchardt, Hug o. 1895. Über den passiv en Charakter des Transitiv s in den kaukasischen Sprachen (Sitzungsberichte der Akademie der Wissensch.) 133, 1. Wien. Thompson, Sandra A., and Robert E. Long acre. 1985. Adv erbial clauses. Language typology and syntactic description II. Complex constructions, ed. by Timothy Shopen, 171—234. Cambridge. Tschenkéli, Kita. 1958. Einführung in die georgische Sprache. Vol. I. Zürich.

1397

Tuite, Kevin. 1987. Indirect transitiv es in Georgian. Proceedings of the Thirtheenth Annual Meeting of the Berkeley Linguistic Society, 296—309. —. 1988. Number agreement and morphosyntactic orientation in the Kartv elian languages. Doctoral dissertation, University of Chicago. Vamling , Karina. 1987. Complementation in Georgian. Dissertation, University of Lund, Sweden. Vendler, Zeno. 1967. Verbs and times. Linguistics in Philosophy, 97—121. Ithaca. Vog t, Hans. 1971. Grammaire de la langue Géorgienne. Oslo.

Alice C. Harris, Nashville, Tennessee (USA)

78. Swahili 1. 2.

8. 9. 10.

Introduction Noun Classes and Noun Phrases: The Basic Facts Basic Sentence Types The Voice System Clause Subordination and Coordination A Closer Look at Agreement A Closer Look at Grammatical Function Assignment Evidence for Clausal Phrase Structure Conclusion References

1.

Introduction

3. 4. 5. 6. 7.

This cannot be a detailed account of the syntax of Swahili, for reasons of space, but also because the syntax of that language, although it surely belongs to the more familiar languages of the world, is by no means well studied in ev ery detail. In this article, I will outline the basic syntactic properties (section 2.—5.), and discuss some phenomena which might be interesting in the light of current syntactic theories (sections 6.—8.). Swahili, a Bantu language of Eastern Africa, has been in intensiv e contact with non-Bantu languages, most notably Arabic and in more recent times English. The influences of these languages, howev er, is largely confined to the lexicon; as far as syntax and morphology is concerned, Swahili can be considered as a fairly typical Bantu language. The number of speakers exceeds 30 millions, but its situation may be quite unique, insofar as by far the largest group speaks it only as a second language. This reflects its history as

the lingua franca of the East African trade became the official language of Kenya and Tanzania, and an important language in Uganda. Sev eral Swahili dialects can be identified, among them Amu on the Lamu island, Mvita in Mombasa, Mtang ata at the northern Tanzanian coast, Ung uja in central Zanzibar and the neighbouring coast, and King wana in the Katanga prov ince of Zaire. Comorian and Chimwini, the language of the Somalian town Brav a, are mostly considered as separate languages. Standard Swahili is based on Unguja. (See Polomé 1967, Whiteley 1969, Khalid 1977, Nurse/Spear 1985). There are quite a few good textbooks (e. g. Wilson 1970, Brauner/Herms 1979, Hinnebusch/Mirza 1979, Maw 1985), dictionaries (“Kamusi”, 1981, Höftmann 1979), a few grammatical descriptions (e. g. Ashton 1944, Polomé 1967, Miehe 1979 for older stages), which focus mainly on morphology, and some monographs and articles on specific problems. There is a journal, Kiswahili, dedicated to the study of the language and its literature which is published by the Institute for Kiswahili Research of the Univ ersity of Daressalaam. There is, howev er, no comprehensiv e reference grammar; the closest to that is still Ashton (1944).

2.

Noun Classes and Noun Phrases: The Basic Facts

2.1.  Bantu languages are famous for their elaborate noun class systems. Although the phenomenon is well-known from gender sys-

1398

XXI. Syntaktische Skizzen

tem of languages like Spanish or German, Bantu systems are impressiv e because of their complexity and the central role they play in syntax. They are morphologically and semantically perspicuous, as they are marked by prefixes and nouns denoting similar entities tend to occur in the same noun class. In Swahili, ev ery noun belongs to one of 14 classes, which is marked in most cases by the so-called nominal prefix, e. g. m- in m-toto ‘child’, ki- in ki-kombe ‘cup’, or ji- in ji-we ‘stone’. The class of a noun is, by and large, semantically predictable; e. g., nouns denoting human beings, plants, or artefacts tend to belong to specific noun classes. Nearly all classes come in singular-plural pairs, e. g. mtoto ‘child’ wa-toto ‘children’, ki-kombe ‘cup’ vi-kombe ‘cups’, ji-we ‘stone’ ma-we ‘stones’. A few noun stems occur with different nominal prefixes in different classes not related by the singular/plural distinction; e. g. m-tu ‘person’, ki-tu ‘thing’. Some nominal prefixes can be combined with a wide range of nouns and express such semantic notions as augmentativ es and diminutiv es (cf. e. g. m-ji ‘town’, jiji ‘city’, ki-ji-ji ‘v illage’) or abstract concepts (cf. m-toto ‘child’, u-toto ‘childhood’).

speaker’ v s. ‘near hearer’ (cf. also Leonard 1985). Note that in these two forms, the pronominal “prefix” occurs as suffix and influences the preceding v owel, and with the last one, the v owel of the prefix is changed to o (cf. also 8.1.1.). The pronominal prefix is also used with possessives, cf. ji-we l-ang u ‘my stone’, ji-we l-ako ‘your stone’. But note that with some relational nouns, such as baba ‘father’, the possessiv e will be a suffix, cf. babang u ‘my father’. The pronominal prefix occurs in a g enitival construction based on the preposition -a, cf. ji-we l-a Juma ‘Juma’s stone’, and in a possessiv e construction based on the preposition -enye, cf. ji-we l-enye uzito ‘stone with weight’, ‘heav y stone’. Furthermore, the pronominal prefix is used for a number of quantifier-like expressions, namely with -ote ‘all’ (cf. ma-we y-ote ‘all stones’), with -o -ote ‘any’ (cf. ji-we l-o l-ote ‘any stone’), with the interrogativ e -pi ‘which’ (cf. ji-we li-pi ‘which stone’), and with the intensifier -enyewe (cf. ji-we l-enyewe ‘the stone itself’). Note the contrast between -ote ‘all’ and number words or words like -ing i ‘many’, which agree using the nominal prefix (cf. mawe m-engi ‘many stones’).

2.2.  Adjectiv e stems (which include numerals) hav e nominal prefixes by which they agree with the head noun; e. g. m-toto m-dog o ‘small child’, ki-kombe ki-dog o ‘small cup’, or m-toto m-moja ‘one child’, wa-toto wa-tatu ‘three children’. So the basic difference between adjectiv e stems and noun stems is that adjectiv e stems combine freely with any noun class prefix, whereas noun stems do not. But only a subset of the adjectiv es and numerals actually agree with the head noun, namely those of Bantu origin; examples which lack agreement are kikombe safi ‘clean cup’, watoto saba ‘seven children’.

2.4.  As the examples giv en so far suggest, adnominal modifiers follow their head noun in Swahili. This also holds, of course, of relativ e clauses, which will be discussed in section (5.1.) below. One possible exception to that rule are demonstrativ es, which may precede the noun, especially in a more colloquial style; cf. li-le ji-we ‘the stone’. As the gloss suggests, preposed demonstrativ es can be considered as definite articles. Another exception is the quantifier kila ‘ev ery’, borrowed from Arabic, which has to precede the noun (cf. kila mtoto ‘ev ery child’). So it might be that a new category of prenominal determiners is dev eloping in Swahili. There is no indefinite article; howev er, the postponed number word -moja ‘one’ is sometimes used in this function, most notably when introducing an entity into discourse about which more information is given later.

2.3.  There is a second class of prefixes which plays a role in the agreement of pronominal modifiers, like demonstrativ es, and therefore is called pronominal prefix. For example, the pronominal prefix of the ji/ma classes are li and ya, and for the ki/vi classes they happen to be identical to the nominal prefixes, ki and vi. There are three demonstratives, a far-deictic, cf. ji-we li-le ‘that stone’, and two neardeictics, cf. ji-we hi-li and ji-we hi-lo. Traditionally, the latter ones are described as neardeictic ‘this stone’ and text-deictic ‘the aforementioned stone’; Barrett-Keach (1980) cites ev idence that, in fact, the distinction is ‘near

2.5.  There are three locative classes that are special insofar as there are no nouns which belong to them per se (with the exception of mahali ‘place’). Howev er, many nouns can be transformed to place nouns by a suffix -ni (cf. nyumba-ni ‘at, to, from the house’, kazi-ni ‘at, to, from work’). These locativ e nouns, together with place names which are inherently

78.  Swahili

locativ e, in turn, may agree with the help of one of the three locativ e prefixes (pa-, ku-, m-), which add further specification to the locativ e meaning of the noun (roughly ‘at’, ‘to’ or ‘near’, and ‘in’). 2.6.  Standard Swahili differs from many other Bantu languages that feature two different nominal prefixes, generally of the structure CV- or VCV- (where the first V also is called ‘preprefix’). Typically, using the CV prefix indicates that the noun is indefinite, nonspecific, or predicativ e, whereas the VCV prefix indicates that the noun is definite, specific, or referential. Greenberg (1978) assumes that the complex prefixes originate in the agglutination of demonstrativ e elements or articles with the noun.

3.

Basic Sentence Types

3.1.  Noun class agreement plays an important role in the syntax of Swahili sentences. In the following sentence, the v erb agrees with both the subject watoto wale ‘those children’ and the object machung wa haya ‘these oranges’ by means of their pronominal prefixes wa (class 2) and ya (class 6):

As it can be seen with this example, the socalled subject prefix is followed by a tense prefix, the object prefix, and the v erb stem. The subject prefix occurs with most tenses; exceptions being the general present tense marked by a, where it occurs with some allomorphic v ariations, and the habitual tense prefix hu, where it does not occur at all. The object prefix occurs only with transitiv e v erbs, and is not obligatory in this case: it occurs regularly with animate NPs, especially if they are definite, and sometimes with inanimate definite NPs. The conditions of object agreement will be further discussed in section (6.1.3.). 3.2.  There is no case distinction with NPs. The word order, though basically SVO, is relativ ely free — the two NPs and the v erb in (1) can be arranged in any order. Hence agreement, or cross-reference, bears a considerable functional load in determining sub-

1399

ject and object. Thus, Swahili is a headmarking language in the typology of Nichols (1986), as the syntactic functions of the arguments are marked on the head, viz. the verb. 3.3.  Example (1) has prov ided a first impression of the morphological structure of the Swahili v erb. It should be noted here that the v erb stem may be decomposed into a v erb root and suffixes which indicate argumentchanging processes, like passiv e (cf. section 4.). The last v owel, -a, can be considered as a marker of affirmativ ity and indicativ ity; there is a special subjunctiv e form with has -e as a final v owel and lacks the “tense” prefix (cf. wa-ya-on-e CL2-CL6-see-SUBJ), and there are special negativ e forms, some of which hav e the final v owel -i (cf. ha-wa-yaon-i NEG-CL2-CL6-see-NEG). With v erb stems of Arabic origin, howev er, the final v owel does not change, cf. wa-na-fahamu CL2-PRES-understand, with wa-fahamu as subjunctiv e and ha-wa-fahamu as negation. (See Contini-Morav a 1989 for a study of the semantics of tense, modus and negation). 3.4.  The copula behav es differently from other v erbs. There is a particle ni (negative si) which, combined with a following NP or adjectiv e, changes this to a predicate; cf. Juma ni mwalimu ‘Juma is a teacher’, nyumba si nzuri ‘the house isn’t nice’. It is, howev er, not obligatory to use an ov ert copula; cf. Juma mwalimu ‘Juma is a teacher’. There is a second copula which consists of a pronominal prefix; cf. Juma yu mwalimu ‘Juma is a teacher’, Kuchimba ku kazi ‘Cultiv ation is (hard) work’; it is related to a Bantu copula stem li, which is reduced to zero in Swahili. There are slight semantic differences between these two copulas; the latter one has a kind of ‘actual’ meaning, cf. Huyu yu mume wang u ‘This is my husband’ (implies that the speaker is still married), Huyu ni mume wang u (could be uttered by a widow pointing at a picture). To express tense distinctions, an auxiliary v erb kuwa ‘to be’ must be used; one example is Watu hawa wa-li-kuwa ni mwalimu ‘These people were teachers’. With the v erb kuwa, the copula nilsi, and especially the second type of copula which consists of a subject prefix only, typically are suppressed; so it looks as if the v erb kuwa were the copula (see Closs 1967 for a treatment of copula constructions, and 8.6. for auxiliary kuwa). 3.5. Possession and location are expressed with the help of copula constructions. The

XXI. Syntaktische Skizzen

1400

concept ‘to hav e’ is realized by a particle na ‘with’; this occurs either with a subject prefix (cf. watu hawa wa-na motokaa ‘these people hav e a car’), or with the v erb kuwa (cf. watu hawa wa-li-kuwa na motokaa ‘these people had a car’, lit. ‘were with a car’). The object prefix may be cliticized to the possession predicate, using the o-pronoun; cf. watu hawa wa-na-yo motokaa ‘these people hav e the car’. Location is expressed by the agreeing copula and an enclitic locativ e o-pronoun; Subject Object

1. SG ni ni

2. SG u ku

3. SG a m

e. g. watu hawa wa-po Nairobi ‘these people are in Nairobi’. In other tenses, the locativ e pronoun is attached to kuwa; cf. watu hawa walikuwapo Nairobi ‘these people were in Nairobi’. 3.6.  In addition to the pronominal prefixes of the noun classes, we also hav e a set of personal prefixes. Note the special form a for third person singular which replaces the pronominal prefix of class, 1, yu. 1. PL tu tu

2. PL m wa

3. PL wa wa

Reflexive -ji

We find some differences between subject and of course. Here is an example of a bitransiobject forms — the only trace in Swahili of a v ti e sentence: case system. The prefix ji acts as a reflexiv e pronoun which does not show any person category. In addition to the pronominal prefixes, there are also free forms; they do not distinguish between subject and object forms, and there are no free pronouns for non-animate NPs. In general, free pronouns are used The Swahili v erb can hav e only one cliticized only in special cases, such as emphasis. The object pronoun. If there are two objects, it is following example is a complete sentence: the “dativ e”-like object (denoting the recipi(2) Ni-na-ku-penda. ent or benefactiv e), and not the “patient”-like 1. SG-PRES-2. SG-love, ‘I love you.’ object, which shows agreement. This object typically follows the v erb directly, but other 3.7.  The agreement system is confronted with orders are possible. Furthermore, it is the a special problem with conjoined NPs. If the “dativ e” object which becomes the subject afconjoined NPs belong to the same singular ter passiv ization (see also 4.1.; 4.8.5. for diaclass, the complex NP typically will agree aclectal differences). cording to the corresponding plural class. Furthermore, if they denote human beings, the complex NP will agree according to class 2 (human plural). In other cases, there are two strategies: First, the plural prefix of class 8, vi-, might be used as a sort of neutral class (cf. Sabuni na maji vi-ta-ku-saidia ‘Soap (Cl. 9) and water (Cl. 6) will help you’); second, the last NP might trigger the agreement (cf. Vi-kombe na zawadi zi-li-tolewa kwa washindi ‘Cups (Cl. 8) and presents (Cl. 10) were giv en So the “dativ e” object is distinct from tradi(Cl. 10) to the v ictors’). (Cf. Giv ón 1970 and tional indirect objects. Following Hoekstra/ Roberts/Wolontis 1972 for Bantu in general). Dimmendaal (1983) and researchers on other Note that both strategies pose problems for Bantu languages, I will use the term direct a unification-based treatment of agreement, object instead of “dativ e object”, and call such as in GPSG, as the category of the conother objects nominal complements. Swahili, joined NP cannot be deriv ed from the catethen, exhibits a major difference to the argugories of the parts by simple unification of ment selection of better-known European features. languages: For v erbs with both a “dativ e” object and a “patient” object, it is the “da3.8.  Intransitiv e and simple transitiv e sententiv e” object which enjoys the syntactic status ces are not the only clause types in Swahili, of a direct object (cf. Faltz 1978).

78.  Swahili

In addition to ordinary transitiv e and bitransitiv e v erbs, we find v erbs with locativ e complements. For example, fika ‘arriv e’ has a locativ e complement, and weka ‘put’ has a direct object and a locative complement: (5a) Mtoto a-li-fika shule-ni ‘The child arrived at the school’ (5b) Mtoto a-li-ya-weka matunda sanduku-ni ‘The child put the oranges into the box’ There are some additional argument frames, and many v erbs may occur in more than one frame; this will be discussed in sections 4. and 7. Also, v erbs may gov ern sentential complements of various sorts (see section 5.2.). 3.9.  Speaking of objects, I should mention a frequent idiomatic construction which, at first sight, looks like a normal v erb-object construction, but actually has to be analyzed as a complex v erb (or a “phrasal predicate”, Harries 1970). Some examples: pig a pasi ‘beat’ + ‘iron’ — ‘(to) iron’, pig a mashine ‘beat’ + ‘machine’ — ‘typewrite’, pig a risasi ‘beat’ + ‘bullet’ — ‘shoot’, fanya kazi ‘do’ + ‘work’ — ‘work’, vunja jung u ‘break’ + ‘pot’ — ‘celebrate’, ona baridi ‘see’ + ‘cold’ — ‘be cold’, kata kiu ‘cut’ + ‘thirst’ — ‘quench thirst’. The noun in these constructions cannot be a direct object, since it does not agree with the v erb, cannot be promoted to subject by passiv e, and cannot be expanded by adnominals. Also, a complex v erb may gov ern a real object:

It is obv ious that the idiomatic ‘objects’ in fact aren’t objects. A construction like -piga pasi has to be analyzed as an incorporation of pasi into the v erb -pig a instead. As only words (that is, constituents of categories X°) can be incorporated, we can explain why pasi cannot be expanded. Also, we can explain why there is no object agreement in this case. There are actually two reasons for that: First, incorporated words typically are non-referential, and thus non-definite; second, they cannot play the role of normal syntactical arguments anymore, as they become part of the morphological component of the language. But note that the incorporated noun need not be adjacent to the v erb; as an alternativ e to

1401

(6), we might hav e Juma alizipig a ng uo hizi pasi, where -pig a and pasi arguably form a discontinuous constituent. 3.10.  There are v arious kinds of adverbials, for example simple adv erbs, prepositional phrases, and locativ es. Frequently they follow the v erb (or the v erb + object construction), but this is only a tendency; they also may precede the sentence, or come between the verb and the object: (7) (Leo) mwalimu (leo) aliwapa (leo) wanafunzi (leo) mpira (leo) Today, the teacher gave the pupils a football. (with alternative positions of leo ‘today’). Simple adverbials may be formed with adjectiv al roots by putting them into class 8, cf. Rehema ananyimba vi-zuri ‘Rehema is singing beautifully’. Furthermore, there is a set of typically onomatopoetic words which act mainly as adv erbials, so-called ideophones; examples are kuang uka tang ’ ‘fall down like a coin’ and kuang uka chubwi ‘fall down in water’, or kunukua mff ‘emit an unpleasant odor’. 3.11.  There are few simple prepositions in Swahili — basically only na ‘by’, kwa ‘with’, ‘for’ and kwenye ‘with’, ‘close to’, which are difficult to render in English because of their wide range of meanings. Howev er, prepositions may be deriv ed from nouns; some examples are kwa sababu ya ‘for reason of’, ‘because of’, juu ya ‘abov e’ (where juu is a noun meaning ‘height’). 3.12.  There are locative NPs, whose head noun is marked by a suffix -ni. As we hav e seen, they may be gov erned by the v erb. But they also occur as free adjuncts with any v erb, as in Mtoto a-li-imba nyumba-ni (m-le) ‘the child sang in the/(that) house’.

4.

The Voice System

As Bantu languages in general (cf. Guthrie 1962, Meeussen 1967), Swahili has an elaborate v oice system. In this section, I try to outline the facts in a theory-neutral way; see section (7.6.) for a discussion of recent theories of the v oice system. Voices are typically marked by suffixes of the v erb root, also called “extensions”. The suffix is followed by the final vowel (cf. 3.3.).

1402

4.1.  The passive can be deriv ed from v erbs which hav e a direct object (but see 6.2.1.). The direct object is promoted to subject, the former subject may be expressed by a prepositional phrase with the preposition na, and the v erb is marked with the suffix w or one of its allomorphic v ariants (l)ew, (l)iw, which are conditioned by rules of v owel harmony. One example: For the sentence Juma a-li-waona watoto ‘Juma saw the children’, we hav e the passiv e correspondent Watoto wa-li-onew-a (na Juma) ‘The children were seen (by Juma). As we hav e seen (cf. 4.), passiv e applies only to the direct object, not to a nominal complement (but see 4.8.5.). Also, it does not apply to idiomatic complex v erbs as discussed in (3.9.). 4.2.  Similar to the passiv e, the stative verb form adv ances the direct object to subject position. In this case, howev er, the former subject cannot be expressed anymore. The v erb is marked by the suffix k or its v ariants (l)ik or (l)ek. One example: Juma a-me-ki-vunja kikombe ‘Juma has broken the cup.’, kikombe ki-me-vunj-ik-a (*na Juma) ‘the cup is broken (*by Juma).’ 4.3.  There is another use of this suffix, namely to deriv e forms which express potentiality. One example: Juma a-na-fanya kazi hii ‘Juma does this work.’, kazi hii ya-fany-ik-a (*na Juma). ‘This work can be done (*by Juma).’ Some v erbs allow stativ e followed by potential marking at the same time, by k and by lik/lek, (e. g. ukuta w-a-bomo-k-a ‘the wall is crumbling down’, ukuta w-a-bomo-lek-a ‘the wall can be pulled down’). It is possible to apply the potential deriv ation to lexical intransitiv es, e. g. pona ‘heal’, ponek-a ‘be healable’. 4.4. Causatives are marked by the suffixes ish/ esh or iz/ez. They add an agent, or causer, to the argument frame of the v erb. If the original v erb was intransitiv e, its subject becomes the direct object of the deriv ed v erb, as seen in watoto wa-na-lala ‘the children are sleeping’ v s. Juma a-na-wa-lalisha watoto ‘Juma makes the children sleep’, ‘Juma puts the children to bed.’ If the original v erb was transitiv e, its subject becomes the direct object, and its object becomes a nominal complement of the deriv ed v erb: farasi a-na-kunywa maji ‘the horse is drinking water’, Juma a-nam-nywesha farasi maji ‘Juma makes the horse

XXI. Syntaktische Skizzen

drink water’. Although causativ es are quite frequent in running Swahili text, not ev ery v erb can be causativ ized. For example, there are no causativ e forms of ishi ‘liv e’ or tua ‘land’. There is a syntactic alternativ e to causativ e v erb deriv ation, the embedding of subjunctiv e clauses under the verb fanya ‘do, make’ (cf. 5.2.2.), e. g. Juma a-li-wa-fanya watoto wa-lale ‘Juma made the children sleep’. This alternativ e is less restricted and can be used, for example, with the v erb ishi: Kenyatta a-me-m-fanya Og ing a a-ishi Kisumu ‘Kenyatta has forced Oginga to liv e in Kisumu.’ There is a second, minor use of the causativ e suffix which expresses intensification, without changing the argument frame at all; witness chung a ‘search’ v s. chunguza ‘inv estigate’; nyama ‘be silent’ v s. nyamaza ‘be completely silent’; or fung a ‘close’ vs. fungisha ‘fasten tightly, imprison’. 4.5.  The applicative, which is marked by the suffix (l)e/(l)i, typically creates a non-subject argument. The applicativ e can occur with basically intransitiv e v erbs; cf. a-me-kufa ‘he died’ v s. a-me-wa-f-i-a watoto ‘he died for the children’. The new argument assumes object status. It often occurs with basically transitiv e v erbs; cf. a-li-u-pika wali ‘he cooked rice’ v s. a-li-wa-pik-i-a watoto wali ‘he cooked rice for the children’. Here, the new argument becomes an object, and the old object is demoted to the status of a nominal complement. If we start with a ditransitiv e v erb, we end up with a four-place v erb that has two nominal complements; e. g. a-li-m-pa mtoto wali ‘he gav e the child rice’ v s. a-li-wa-p-e-a wazee mtoto wali ‘he gav e the child rice for the parents’. Note that in the examples so far, the added argument has the role of a beneficiary. It also may hav e the role of an instrument; witness a-me-li-i-andika barua ‘he has written the letter’ v s. a-li-i/u-andik-i-a barua unyoya ‘he has written the/a letter with the/a quill’. In this case, the instrument may become object, or the former object may remain in object position (see the alternativ e agreement pattern). Furthermore, the additional argument may be a goal; e. g. a-li-kimbia ‘he ran (away)’ v s. a-li-wa-kimbi-li-a watoto ‘he ran after the children’. The applicativ e suffix may be used without any change in the argument structure with an intensifying meaning; some examples: nuka ‘smell bad’ v s. nuk-i-a ‘smell pleasant’, kosa ‘commit an error’ v s. kos-e-a ‘commit a triv ial error’, tuma ‘send’ and tum-i-a ‘use’.

78.  Swahili

4.6.  The reciprocal derivation remov es the direct object of a v erb. The v erb is marked by the suffix na, and the result is interpreted as a reciprocal form: Juma a-li-m-pig a Hassani ‘Juma beat Hassan’, Juma na Hassani wa-lipig -an-a ‘Juma and Hassan beat each other’. Howev er, unlike reciprocals in English, we can hav e “split subjects”: Juma a-li-pig ana na Hassani. — The reflexive, on the other hand, does not belong to the v oice system, as it is indicated by a special object prefix, not by a suffix: Juma a-li-u-kata m-kate ‘Juma cut the bread’, Juma a-li-ji-kata ‘Juma cut himself.’ 4.7.  Besides the v oice distinctions discussed so far, there are a few which are not productiv e anymore in Swahili, but whose traces still are v isible in the v ocabulary of the language. One of these deriv ations is the socalled contactive form, which expresses intensiv e contact with an object (cf. kama ‘squeeze’, kamata ‘take hold of, arrest’). Somewhat more transparent is the reversive form; examples are kunja ‘fold’ v s. kunjua ‘unfold’, fung a ‘close’ v s. fung ua ‘open’. Furthermore, there is a socalled static form, similar in its syntactic behav iour to the stative (cf. ficha ‘hide’ vs. fichama ‘be hidden’). 4.8.  Verbal deriv ations may interact with each other in v arious ways, that is, one deriv ation may feed another one. Verbs with up to four deriv ation suffixes are by no means unusual. In general, the meaning of these complex deriv ations can be deriv ed from the simple deriv ation and the order in which they apply. Howev er, not ev ery combination of deriv ations is possible. The following generalizations seem to hold, not only for Swahili, but for Bantu in general: 4.8.1.  There is no deriv ation after passiv e (but see 6.2.1., and Alsina 1990 for a case of an applicativ e with locativ e meaning after passive in Chichewa). 4.8.2.  There is no deriv ation after the suffix ek/ik marking the potential. Howev er, there are some v erbs, mainly used impersonally, where a reciprocal applies after the stativ e suffix ek/ik. One example is weza ‘manage’, ‘be able to’ v s. wez-ek-an-a ‘managable, possible’, as in i-na-wez-ek-an-a ‘it is possible’. 4.8.3.  Reciprocals can feed other deriv ations, e. g. applicativ e, causativ e, potential. Exam-

1403

ples are Sam na Juma wa-me-g omb-an-i-a mtoto ‘Sam and Juma quarreled ov er the child’; Hassani a-li-fung -an-ish-a Sam na Juma ‘Hassan made Sam and Juma tie each other’; Mpang o huu u-na-pat-an-ik-a ‘this plan is negotiable’. Occasionally, we also find passiv es of reciprocal v erbs. In these cases, it can be assumed that the reciprocal v erb form has been reanalyzed to a transitiv e v erb. An example is vita i-li-pig -an-w-a ‘the war was fought’, where pig a is ‘beat’ and pig ana is ‘beat each other’, ‘fight’. 4.8.4.  Causativ es may feed other deriv ations, for example reciprocal, potential, and passiv e (but not stativ es). Some examples: Sam na Juma wa-li-furah-ish-an-a ‘Sam and Juma pleased each other (lit. made each other happy)’, Juma a-na-nyw-esh-ek-a pombe ‘Juma can be made drink beer’. Furthermore, causativ e may feed applicativ e, as the following example shows: Zawadi i-li-rudi kwa mtoto ‘the gift went back to the child’, Yusuf a-li-i-rud-ish-a zawadi kwa mtoto ‘Yusuf gav e back the gift to the child’, and Yusuf ali-m-rud-ish-i-a mtoto zawadi ‘Yusuf gav e back the child the gift’. There are a few cases where a causativ e feeds another causativ e, for example Sam a-me-mw-o-a Jenny ‘Sam has married Jenny’, Jaji a-me-mw-o-z-a Sam Jenny ‘the judge married Jenny to Sam’, and Rais a-me-mw-o-z-esh-a jaji Sam Jenny ‘the president made the judge marry Jenny to Sam’. 4.8.5.  Applicativ es may feed some other deriv ations, for example, reciprocals, potentials, and passiv es. Examples are Sam na Rehema wa-li-andik-i-an-a ‘Sam and Rehema wrote to each other’, barua i-na-andik-i-k-a ‘the letter is writeable’, and watoto wa-li-andik-i-w-a barua (na Sam) ‘the children were written a letter by Sam’. There is some v ariation in the latter case, as some dialects (e. g. Mv ita, cf. Barrett-Keach 1980) allow also for the patient NP to become subject, e. g. barua i-liandik-i-w-a watoto (na Sam). There seem to be a few cases of a causativ e deriv ation from an applicativ e; one example: Juma a-li-mw-oa Mariamu ‘Juma married Mariamu’, Juma a-li-wa-o-le-a watu wale ‘Juma married from those people’, waziri a-li-mw-o-l-esh-a Juma watu wale ‘the Minister forced Juma to marry from those people’. Applicativ e deriv ations from applicativ es are equally rare and probably hav e to be analyzed as a special case, as they do not lead to additional arguments, and their meaning cannot be predicted in a

XXI. Syntaktische Skizzen

1404

regular way; for example, we hav e sema ‘speak’, semea ‘speak for’, and semelea ‘sermonize’; or og a ‘take a shower’, og ea ‘wash with’, and og elea ‘swim’. Howev er, we find regular cases with two applicativ es in other Bantu languages, for example in Haya (cf. Byarushengo e. a. 1977). 4.8.6.  Stativ es may feed other deriv ations, if they are compatible with intransiti v es, namely causativ es and applicativ es. For example, we hav e ondo-k-esh-a ‘cause to leav e’, ondo-k-e-a ‘leave for’. 4.8.7.  Although the reflexiv e is not part of the suffixal v erb deriv ation system, it may interact with it. For example, it can be argued that it may apply before or after applicativ e deriv ation. So the reflexiv e v erb form -ji-kati-a can mean (i) ‘cut oneself with’, which is an applicativ e deriv ation of ji-kat-a ‘cut oneself’, or (ii) ‘cut for oneself’, which is a reflexive form of kat-i-a ‘cut for’.

5.

Clause Subordination and Coordination

5.1.  There are three types of relative clauses, an analytic type, and two synthetic types with special relativ e forms of the v erb. All of these types are based on a special pronominal affix, which typically contains the vowel “o”. 5.1.1.  The analytic type contains a relativ e pronoun, which is based on the root amba, originally ‘say’, and a suffix containing o. One example, an object relativ e clause, is the following: watoto amba-o mwalimu a-li-waona ‘the children whom the teacher saw’. The relativ e pronoun heads the relativ e clause and agrees with the head noun (in the case at hand, with o); in the relativ e clause, we often find agreement with the relativ e pronoun (here, wa). This agreement is not obligatory in ev ery case, howev er; cf. the relativ ization on a nominal complement in kitabu ambacho mwalimu a-li-wa-onyesha watoto ‘the book which the teacher showed to the children’. 5.1.2.  The second type of relativ e clauses is based on a special relativ e v erb form, which contains an o-affix after the tense affix, and may be followed by an object prefix. This type is restricted to v erbs in a small number of tenses; namely the ones marked by na (pre-

sent), li (past) and ta (future; in this case, the future prefix has the allomorphic v ariant taka). Also, it occurs with the negation infix si. An example watoto a-li-o-wa-ona mwalimu ‘the children whom the teacher saw’; note that both o and wa refer to the children. Our second example turns out as kitabu mwalimu a-li-cho-wa-onyesha watoto ‘the book which the teacher showed to the children’; note that in this case, the relativ e prefix cho differs from both subject and object prefix. This last example also shows that the word containing the relativ e morpheme need not be adjacent to the head noun, although there is a tendency to that order. 5.1.3.  Finally, there is a second type of a synthetic relativ e form in which the v erb lacks a tense affix and the o-pronoun is enclitic; an example is watoto a-wa-ona-o mwalimu ‘the children the teacher sees’. The syntactic possibilities are more restricted here; for example, we cannot formulate a sentence corresponding to our second example. Otherwise, the three relativ e clause forms hav e a similar distribution. For example, ev ery one of them can function as an attributiv e or a restrictiv e relativ e clause. Also, ev ery type can occur as a free relativ e clause; witness wamba-o wanaimba wa-na-furahi, wa-nao-imba wanafurahi and wa-imba-o wanafurahi ‘(they) who sing are happy. (See 8.3. for further discussion of relative clauses). 5.1.4.  Relativ e forms occur quite frequently because a common way to mark focus are cleft sentences that inv olv e relativ e clauses. Cleft sentences consist of the copula ndi(negativ e si-), which agrees with the focused constituent by means of an o- pronoun, followed by the constituent in focus, and a lowpitched relative clause. Some examples: (8a) Ndi-lo dirisha a-li-lo-vunja mtoto yule. ‘It is the window that the child broke’ (8b) Ndi-ye mtoto yule a-li-ye-vunja dirisha. ‘It is the child who broke the window’ (8c) Ndi-po jana a-li-po-vunja mtoto yule dirisha. ‘It was yesterday that the child broke the window’ Swahili makes abundant use of this cleft construction. It may be simplified by using a non-relativ e v erb form and the copula ndior simply ni; for example, (8a) in this v ariant would become Ndilo/Ni dirisha a-li-(li)-vunja mtoto yule.

78.  Swahili

5.2.  Let us now turn to embedded clauses which are gov erned by v erbs. There are three types of sentential complements: tensed indicativ e clauses, subjunctiv e clauses, and infinitives. 5.2.1.  Indicativ e clauses are typically introduced by a complementizer kwamba (historically an infinitiv e of the v erb -amba ‘say’) or kuwa (the infinitiv e of ‘be’). Similar to English, the complementizer may be dropped if the embedded sentence immediately follows the governing verb. (9a) Mariamu a-li-ona kwamba/kuwa/ø watoto wa-li-kwenda shuleni. ‘Mariamu saw that the children went to school’ (9b) Mariamu a-li-mw-ambia Juma kwamba/kuwa watoto wa-ta-kwenda shuleni. ‘Mariamu told Juma that the children will go to school’ 5.2.2.  The following sentence is an example of an embedded subjunctive: (10) Juma a-li-taka Mariamu a-ki-som-e kitabu. ‘Juma wanted that Mariam should read the book’ Subjunctiv es behav e similar as in many other languages: They may occur under predicates denoting wishes (like taka ‘want’, omba ‘beg’, ambia ‘tell (to do something)’, under negation, and under modals like ilibidi ‘it is necessary’. They may occur as root clauses in an optativ e or adhortativ e meaning, cf. Tusome! ‘Let’s read!’. (There is, in addition, an imperativ e, consisting of the v erb root only in the singular and the root with suffix -ni in the plural). Subjunctiv e clauses typically hav e no ov ert subject, and the subject prefix agrees with some NP in the embedding clause, as in Juma a-li-wa-ambia watoto wa-some kitabu ‘John told the children that they should read the book’, but this is not a syntactic requirement, as example (10) shows. It should be noted that there are some noun-like modals borrowed from Arabic which gov ern the subjunctiv e and which are not v erbal; examples are afadhali (suggestion), tafadhali (request), lazima, sharti (necessity). 5.2.3.  The last type of embedded clauses, infinitiv es, has neither ov ert subjects nor sub-

1405

ject agreement (the infinitiv e marker, ku, is sometimes treated as a noun class prefix and is probably related to the locativ e ku-class; other Bantu languages may form the infinitiv e with different class markers, cf. Meeussen 1967). We find subject-controlled and object-controlled infinitive constructions: (11a) Juma a-li-jaribu ku-wa-angalia watoto. ‘Juma tried to take care of the children’ (11b) Juma a-li-mw-ambia Mariamu ku-waangalia watoto. ‘Juma told Mariamu to take care of the children’ As we have seen, some verbs, like ambia ‘tell’, may gov ern the subjunctiv e or the infinitiv e. Others, like jaribu ‘try’, are restricted to the infinitive. 5.2.4.  Sentential complements neither show object agreement nor undergo passiv ization. Furthermore, they cannot be topicalized; cf. *kwamba watoto walikwenda shuleni, Juma aliona ‘That the children went to school, John saw’. Infinitiv es, howev er, may be the subject of a copula sentence, as in [kupata chakula msituni] ni kig umu sana ‘to get food in the wood is very difficult’ 5.3.  In addition to embedded clauses which are gov erned by a v erb there is a wide v ariety of adverbial clauses, that is, embedded clauses in the function of an adjunct. 5.3.1.  One type of adv erbial clauses is introduced by a subjunction, followed by an indicativ e, subjunctiv e, or infinitiv e clause. Some examples with the subjunction ili ‘in order to’, which might gov ern a subjunctiv e or an infinitiv e, and kwa vile ‘because’, which gov erns the indicative: (12a) Tu-li-kutana ili tu-anzishe chama. (12b) Tu-li-kutana ili ku-anzisha chama. ‘We met in order to found a party’ (12c) Tu-li-kutana kwa vile tu-li-taka ku-anzisha chama. ‘We met because we wanted to found a party’ 5.3.2.  A second type consists of clauses which are formally relativ e clauses, headed by a subjunction. There are two subcases: Temporal clauses, which are headed by wakati ‘time’ and contain the relativ e prefix po, and manner clauses, which are headed by jinsi ‘man-

XXI. Syntaktische Skizzen

1406

ner’, ‘kind’ or kama ‘as’, and contain the relativ e prefix vyo. Note that the relativ e prefixes do not agree formally with their heads, analyzed as nouns; po is a locativ e prefix (and is applied to temporal location here), and vyo is the prefix of class 8, which is used to form adv erbials from adjectiv e stems (cf. vi-zuri ‘well’). Two examples: (13a) Wakati watoto wa-li-po-kwenda shuleni, Juma a-li-lala. ‘When the children went to school, Juma slept.’ (13b) Kama Juma a-li-vyo-eleza, watoto walikwenda shuleni. ‘As Juma explained, the children went to school.’ 5.3.3.  A third type of adv erbial clauses is marked by a special affix at the postition of the “tense” marker (possibly in addition with a subjunction, like kama). One example is the affix ki, which is used for clauses specifying the situational background of the main clause (see Hopper 1979 for that interpretation); it may specify an ev ent which is going on simultaneously, or it may specify the conditions in a “when”-clause: (14a) Juma a-li-pika chakula a-ki-wa-sikia watoto. ‘Juma cooked food when he heard the children.’ (14b) Watoto wa-ki-rudi nyumbani Juma ali-wa-pikeni chakula. ‘When the children return home, Juma cooks food for them.’ 5.3.4.  Other examples are the affixes ng a and japo, which mark factiv e and hypothetical concessiv e clauses, roughly to be translated using ‘although’, or ‘ev en if’. Furthermore, the two conditional affixes ng e and ng ali (hypothetical) belong to that type. Both protasis and apodosis hav e to be marked with these affixes: (15) Juma a-ngali-pika chakula, watoto wangali-rudi nyumbani. ‘If Juma had cooked food, the children would have returned home.’ 5.3.5.  Some of the subjunctions are based on the v erb -wa ‘be’ with a dummy subject prefix i-, such as i-ki-wa ‘when’, i-ng a-wa ‘although’, i-si-po-kuwa ‘except’. Others are of Arabic origin, such as ili ‘in order that’, (kwa) sababu ‘because’, ila ‘except if’. In general, it seems that the use of subjunctions is of rela-

tively recent origin in Bantu. 5.4.  Let us finally look at sentence conjunctions. There are conjunctions which are rather independent from the categories they conjoin, for example na ‘and’, au ‘or’ and wala ‘and not’ (after a negated first conjunct), and they can be used as NP conjunctions, v erb conjunctions, and sentence conjunctions. Howev er, the use of na as sentence conjunction is definitely not idiomatic. Also, note that au and wala are of Arabic origin. One v ery characteristic way of conjoining sentences in narrativ es, with the additional meaning of temporal succession, is the use of v erbs with the “tense” affix ka. A typical paragraph in a narrativ e starts with a sentence whose v erb has some definite time affix, such as li (past), followed by clauses marked by ka, which can be rendered by ‘and then’; cf. ni-li-kuja, ni-ka-ona, ni-ka-shinda ‘v eni, v idi, v ici’. The ka- marker is special, howev er, as it may occur with the subjunctiv e in purpositiv e clauses (cf. Nitakwenda sokoni ni-ka-nunue ndizi ‘I will go to the market and (I) buy some bananas’). Furthermore, it can occur without a subject prefix, to giv e emphasis; cf. Ka-sema nani ‘Who spoke?’. Hopper (1979) and Contini-Morav a (1989) analyze ka as a foreground marker, in opposition to the background marker ki mentioned in (5.3.3.). Another conjunction, tena, is used to conjoin adjectiv es (e. g. maembe haya mazuri tena mabivu ‘those mangos are fine and ripe’). When na is used, the second noun typically is an abstract noun; in our case, mazuri na ubivu ‘fine and (with) ripeness’. Tena is also a sentence-initial conjunction ‘then’, and an adv erb ‘again’. I will discuss a means to conjoin VPs in (8.1.6.).

6.

A Closer Look at Agreement

6.1.  Subject-agreement and object-agreement are not quite as simple as presented in section 3. I start by pointing out some differences between Bantu languages in their object agreement (and possibly also differences in dialects of Swahili): 6.1.1.  Some languages closely related to Swahili can hav e more than one object prefix (e. g. Chimwini, cf. Abasheik 1980, Kinyarwanda, cf. Kimeny 1980, and Haya, cf. Byarushengo e. a. 1977; see also Wald 1979). For

78.  Swahili

example, in Haya we find v erbs with up to three object clitics, as in Kat’ á-ka-ki-g a-músiig isa ‘Kato smeared it (mu) on him (ga) with it (ki)’. Interestingly, the order of the object clitics is the mirror-image of the preferred order of the corresponding full NPs; this might reflect an older stage with a basic ObjectVerb word order (cf. Krifka 1983). 6.1.2.  Not all object agreement in Bantu languages is the same. Bresnan/Mchombo (1987) show that object agreement in Chichewa is not syntactical agreement of the v erb with its object, but anaphoric agreement between an incorporated object pronoun and a topical noun phrase. Subject agreement, on the other hand, can be either syntactic agreement or anaphoric agreement. For example, they show, using ev idence from tone sandhi phenomena, that a string consisting of a nonagreeing v erb and an object NP forms a constituent, whereas a string consisting of an agreeing v erb and an (purported) object NP does not. Also, they show that the word order is Verb-Object with non-agreeing v erbs, but can be any order with agreeing v erbs (as the purported O is in fact a free NP). Swahili is different in that respect. First, there is no indication for the type of difference found in Chichewa. Second, as agreement with object NPs denoting human beings is v irtually obligatory, and there is no reason for claiming that these object NPs aren’t true objects, we should assume syntactic v erb-object agreement. 6.1.3.  A typology of object agreement in Bantu may look as follows: In general, object markers may express anaphoric agreement. Languages like Chichewa restrict their use of object markers to anaphoric agreement. Languages of this type v ary in one respect, insofar as some of them, like Chichewa, allow object pronouns in object relativ e clauses like ‘the man that I (him-)saw’, whereas others, like Chibemba, ban it (cf. Bresnan/Mchombo 1987, Wald 1979). Other languages show syntactic agreement in addition to anaphoric agreement. The use of syntactic agreement may differ between languages; in Zulu, for example, object agreement seems to be a sign of definiteness of the object. For Swahili, a good generalization is that objects denoting human beings show agreement far more frequently than others, and definite NPs show agreement more frequently than indefinite NPs, and indefinite

1407

referential NPs in turn more frequently than nonreferential NPs. According to a text surv ey by Wald (1979), object NPs denoting human beings agreed to 90% if they were definite, 42% if indefinite, and 10% if nonreferential; if they denoted non-human entities, the figures were 11%, 3% and 0%, respectiv ely. According to Alemán (1987), specificity may also play a role; she contrasts sentences like Na-m-tafuta mtu ku-ni-saidia ‘I’m looking for someone (specific) to help me’ v s. Na-tafuta mtu ku-ni-saidia ‘I’m looking for anyone that can help me’, or Si-ku-mw-ona mtu yo yote ‘I didn’t see anyone of them’ v s. Si-ku-ona mtu yo yote ‘I didn’t see anyone at all’. It seems plausible that object agreement originated in anaphoric agreement (cf. Giv ón 1976). A pattern like a-li-wa-ona, watoto ‘he saw them, the children’, with the topical NP watoto and anaphoric agreement, ev olv ed towards a-li-wa-ona watoto ‘he saw the children’, where anaphoric agreement is grammaticalized to syntactic agreement. As topical NPs typically are definite and denote human beings, this bias became grammaticalized as well. 6.1.4.  Wald (1979), following Ashton (1944), suggests that an additional function of syntactic object agreement, at least with non-human objects, might be to “draw attention” to the object. Hence object agreement would be a means to focus the object. Wald tries to explain that by saying that syntactic object agreement with non-human objects is so rare that its v ery occurrence MUST draw attention to the object. By that, object agreement would hav e dev eloped from a reference to topical entities to a focus marker(!). Maw (1976) also showed that object agreement is used in more circumstances than earlier descriptions had env isaged. She found that the v erb may agree with locativ e adjuncts, temporal adjuncts and manner adjuncts: (16a) A-li-mu-ulia busta-ni wageni. ‘He killed the stranger in the garden.’ (16b) A-li-mu-isha mwaka jana kazi yake. ‘He finished his work last year.’ (16c) A-li-vi-onelea kwa uchungu sana kifo cha dada yake. ‘She felt her sister’s death very bitterly.’ The locativ e adjunct (16a) and the temporal adjunct (16b) agree using one of the locativ e prefixes (here mu ‘inside’), and the manner adjunct agrees by the class prefix 8 (we hav e

1408

already seen with free relativ e clauses that the locativ e prefixes can hav e a temporal meaning and class 8 can hav e a manner-related meaning; cf. (13a/b)). In all cases, the agreement focuses the agreeing adjuncts; for example, (16a) would be appropriate as an response to ‘Where was it that he killed them?’, (b) to ‘When was it that he finished his work?’, and (c) to ‘how did she take her sister’s death?’. 6.1.5.  To sum up, “object” agreement in Swahili is a rather complex matter: We hav e to distinguish between anaphoric agreement and syntactic agreement, and with syntactic agreement we must in turn distinguish between two cases: object NPs denoting human beings nearly always agree, and other constituents (objects or adjuncts) may agree if the constituent is in focus. 6.2.  Now let’s look at subject ag reement. It turns out to be gov erned by rules at least as complicated, which becomes particularly clear when we look at so-called inversions. 6.2.1.  Ashton (1944) pointed out that locativ e adv erbials may trigger subject agreement, which she called locative inversion: (17a) Mwitu-ni m-me-lala wa-nyama. ‘In the wood animals are asleep’ (17b) Bonde-ni ku-na-lima wanawake. ‘There in the valley women cultivate’ (17c) Hapa pa-me-kufa simba. ‘On this spot there has died a lion’. According to Khamisi (1985), the v erb may be marked by the passiv e suffix w in these cases; cf. dukani pa-me-simam-(w)-a watu ‘in the shop there are people standing’. In contrast to real passiv es, these passiv es of intransitiv es do not hav e a normal agent phrase marked by na. Also, locativ e inv ersion may apply after passiv e, contrary to all deriv ations; cf. mtoto a-li-vu-li-w-a ng uo na Sam nyumba-ni ‘the child was undressed by Sam’ and nyumba-ni m-li-vu-li-w-a mtoto ng uo na Sam ‘in the house a child was undressed by Sam’. 6.2.2.  Maw (1976) showed that we can also hav e temporal subjects. Note that the subjects cannot occur with the locativ e ending -ni, although the subject prefixes are locative:

XXI. Syntaktische Skizzen

(18a) Asubuhi m-li-nyesha mvua nyingi. ‘In the morning it rained a lot’, ‘It rained all morning’ (18b) Mwaka jana m-li-fika wageni. ‘Last year foreigners arrived’ 6.2.3.  Perhaps most striking are the facts discussed by Whiteley/Mganga (1969). They show that many v erbs are quite unstable in their subject selection. In the following example, the v erb pika ‘cook’ occurs with the NP denoting the cooked object as a subject: (19) Chakula ki-me-pika mgeni wetu. lit. ‘The food has cooked our guest’, gloss: ‘the food (which we hav e been praising ...) has, as a matter of fact, been cooked by our guest’ For examples of this type, the semantic relationship of the NPs to the v erb is clear by the animateness of one NP; if both NPs are animate, only the normal interpretation is possible (for example, the meaning of yule kijana a-me-wa-pig a wa-toto wang u ‘that youngster has hit my kids’ cannot be rendered by watoto wang u wa-me-pig a yule kijana). Although animateness of one NP is an important condition (cf. Barrett-Keach 1980), in some cases we hav e inv ersion although both NPs are inanimate. In this case, other semantic properties of the arguments clarify the semantic role assignment; cf. nchi i-me-enea maji ‘the country is cov ered with water’ and maji ya-me-enea nchi ‘the water cov ers the country’, or bahari hii i-me-zama meli ‘this sea has sunk ships’. Another restriction is that the demoted object, though not an object, must be expressed explicitly; cf. *chakula ki-me-pika. 6.2.4.  It should be noted that there is a tendency, at least with adv erbial subjects, that the “logical” subject may not be a proper name or a pronoun (cf. Whiteley 1972); for example, nyumba-ni ku-li-fika mtoto/*Juma ‘to-the-house came some child/*Juma’. In most examples of inv ersions, the “logical” subject is, indeed, indefinite. 6.2.5.  These facts can be captured by assuming that what we hav e called subject agreement is in fact topic ag reement, that is, agreement with a constituent in topic position that need not be the logical subject. Often, the logical subject will be the topic; but if this is not the case, it may be demoted from subject position without any marking at the v erb. This, howev er, is possible only if the semantic

78.  Swahili

roles are recov erable by world-knowledge inference. 6.2.6.  There is one peculiar construction which shows that the v erb can agree either with the subject or with the topic (cf. Maw 1970): (20) Mtoto yule mambo yake a/ya-me-nichoka. CL1.Child this CL6.affairs his CL1/ CL6-TNS-1 SG-make-tired. Maw suggests quite similar glosses for each case. A plausible analysis is that mtoto yule in fact is the topic of the sentence, and that the best gloss is ‘As for the child, his affairs make me tired’. The v erb may either agree with the main topic mtoto yule or with the logical subject, mambo yake. Note that, when we make the reasonable assumption that the two NPs occupy different syntactic positions, syntactic agreement is not obv iously triggered by syntactic position. 6.2.7.  Cases of v ariable agreement can be treated in at least two different ways: They might be considered as a case of syntactic mov ement, similar to passiv e, with the difference that the v erb is not marked and the former subject does not end up in a PP, but as a nominal complement. Alternativ ely, we might assume that the mapping from semantic arguments to the syntactic arguments is not as rigid as in languages like English. Swahili might be closer to a topic-prominent language in the sense of Li/Thompson (1975). I do not intend to go into this problem here, but I will contribute some more observ ations on argument selection.

7.

A Closer Look at Grammatical Function Assignment

In the last section, we saw that “subject” and “object” agreement is related to topic and focus marking. But it is also related to the meaning of the v erb, or, to be more prećise, to the thematic roles of the v erb, which giv e us the “default” subject and object, as it were. 7.1.  First, let us hav e a look at the selection of the subject. In general, if a v erb has an agent argument, this will become the subject, cf. Juma a-li-fung a mlang o ‘Juma closed the door’. If no agent is present, an instrument

1409

may take the subject position, cf. Upepo u-lifung a mlang o ‘the wind closed the door’. In this case, the instrument probably has typical agent features so that it qualifies, in absence of a real agent, as the most agent-like participant. 7.1.1.  For some v erbs there is a middle construction where the agent is suppressed and the patient surfaces as subject; cf. Mlang o ume-fung a ‘The door closed’. Whiteley (1972) remarks that the middle is apparently restricted to perfectiv e and future, and that the potentiality of the door for closing properly is stressed. So the middle conv eys the meaning that it is some property of the non-agent subject which accounts for the fact expressed by the sentence. 7.1.2.  The same seems to be true for semantically triggered inversions, which are similar to middles, but for which the “original” subject is still present (cf. 6.2.3.). For example, Whiteley/Mganga (1969) note that in addition to the regular mtoto huyu a-me-li-ruka jiwe ‘the child jumped onto a stone’, we hav e jiwe lime-ruka mtoto huyu, ‘the stone is small enough for the child to hav e jumped on it’, were mtoto huyu ‘the child’ is an agent. Of course, inv ersions abide by additional restrictions, such as recov erability of theta-roles. Inv ersion, in general, may apply after passiv e; cf. watoto wa-li-som-e-w-a kitabu ‘the children were read a book’ v s. kitabu ki-lisom-e-w-a watoto ‘the book was read to children’. Whiteley (1972) sees as the common property of inv ersions that “one can infer from the sentence some quality of NP1 [the NP agreeing as subject] which is responsible for such an attribute”. In fact, this responsibility feature is present in the instrument and middle construction as well, were it is suggested that the instrument, or the theme, are responsible for the v erbal predicate to be applicable. Also, it cov ers the simple agent case; in Juma a-li-m-pig a mtoto ‘Juma beat the child’, Juma is responsible for the beating. Consequently, we can assume that the NP whose semantic role inv olv es responsibility becomes the subject. There are restrictions, howev er: First, as we hav e seen in the last section, there is a competing subject principle, related to topicality. Second, the “responsibility” property does not apply to the stimulus in experiencerstimulus-v erbs, such as in Juma a-li-wa-ona watoto ‘Juma saw the children’, as it is the

1410

experiencer (who is not responsible) that becomes subject. 7.2.  As for object selection, we hav e to account for the fact that in a transitiv e v erb with agent and theme, the theme will become the (direct) object, whereas in a ditransitiv e v erb with agent, theme and “dativ e”, the “dativ e” will be the object (cf. 3.8.). Note that this cannot simply be traced back to applicativ e v oice, as it holds ev en with simple, nonapplicativ e v erbs, such as pa ‘giv e’. With ditransitiv es, recipient and benefactiv es are not the only semantic role for objects; for example, in Juma a-na-m-la Ali pesa ‘Juma is scrounging (lit. eating) money from Ali’, the object Ali is neither recipient nor beneficiary. It is difficult to characterize the common core of the semantic role of these objects; maybe something like ‘indirectly participating in the ev ent’ is general enough. A few v erbs v ary in their object selection; for example, we hav e both a-li-u-paka ukuta rang i ‘he smeared the wall (with) paint’ and a-li-i-paka rang i ukuta ‘he smeared the paint (on) the wall’. 7.3.  With many transitiv e v erbs, the object may be dropped, yielding an indefinite interpretation; cf. Juma a-na-kula ‘Juma is eating’. With other v erbs, we get a definite interpretation; cf. Juma a-me-peleka ‘Juma has sent (it)’. Verbs with indefinite interpretation can be said to hav e an “immanent” object (cf. Whiteley 1972). According to Barrett-Keach (1980), only inanimate objects can be dropped, which shows the important role animacy has in the syntax of Swahili. 7.4.  Let us now turn to nominal complement selection. We hav e seen that NPs may end up as complements because they are demoted from subject or object status. Howev er, there are some cases which can be considered as free complement creation. 7.4.1.  According to Khamisi (1985), many v erbs with an agent role may be extended to include an instrument role as well, without the prepositional marking kwa. One example is Juma a-li-m-choma ng uruwe mkuki ‘Juma stabbed the pig (with a) spear’; another example, which requires a wider notion of instrument, a-li-kufa ndui ‘s/he died (of) smallpox’. This fits nicely to the assumption of Marantz (1984) and Baker (1988a, b) that instruments get their semantic role by the v erb directly, and not by a preposition.

XXI. Syntaktische Skizzen

7.4.2.  Locativ e arguments typically are marked by a locativ e noun (a place name or a noun with suffix ni). The specific type of locativ ity is determined by the v erb; cf. a-likwenda shule-ni ‘he went TO school’, a-li-kaa mji-ni ‘he liv ed IN the town’, and a-li-hama kijiji-ni ‘he mov ed FROM the v illage’. Some v erbs are ambiguous; for example, ruka ‘jump’, may hav e a goal or a source interpretation for its locativ e complement. Sometimes a non-locativ e noun may be used, as in a-li-hama kijiji. According to Driev er (1976), this is possible with source-locativ es (in the Mv ita dialect); according to Whiteley (1972), this might be done if the NP refers to a previously mentioned entity. 7.5.  There is an interesting construction which affects both subject and object selection:

In some analyses (cf. Harries 1970/71, Hawkinson 1979, Hyman 1977 for Haya, see also Hinnebusch/Kirsner 1980), this construction was treated as a case of possessor raising applicable in the case of inalienable possession. (21a) is deriv ed from mkono wa mtoto u-mefura ‘the arm of the child is swollen’ by movement of mtoto into subject position, and (21b) from wa-dudu wa-li-uma mkono wa mtoto by mov ement of mtoto into object position. In both cases, the possessed mkono is demoted to complement status. According to Baker (1988a), possessor raising consists of the incorporation of the possessor into the v erb; howev er, this does not explain why the possessor clearly has syntactic object status (with agreement), and why it might be syntactically complex (note that incorporation, as X°-mov ement, should affect only lexical heads). Scotton (1981) proposed that “possessor raising” has nothing to do with either possession or raising. First, there are many cases where the semantic trigger of the construction is not inalienable possession. For example, in Juma a-li-m-shika Ali shati ‘Juma grasped Ali at the shirt’, Ali and shati do not stand in the relation of inalienable posses-

78.  Swahili

sion. Second, the complement can be analyzed similarly to an adv erbial, namely an adv erbial that specifies the extent to which the described ev ent or state holds; a similar treatment was indicated by Whiteley (1972) who called this complement the “NP of limitation”. So a sentence like Juma a-li-m-shika Ali shati says that Juma grasped Ali, and limits this action to the shirt (a good paraphrase might be: “Juma grasped Ali shirtwise”); similarly, (21a) says that the child is swollen, and limits that state to the arm. Analyzed in this fashion, sentences like the following ones fall under that pattern as well: Wa-li-jenga nyumba vyumba viwili ‘they built the house two rooms (implying that no more are to be built)’, or Juma a-li-m-pig a mtoto bakora tatu ‘Juma beat the child three canes (to the extent of three strokes of a cane)’, or the more familiar-looking Juma a-li-tembea maili tatu ‘Juma walked three miles’. Note that the limitation complement will express the limitation either by referring to parts of the theme object, or, as in the last examples, by restricting the v erbal predicate directly by some measure phrase. 7.6.  In this section, I hav e suggested general principles of grammatical function assignments for simple v erbs. An obv ious question is how these principles relate to v erbal deriv ations, as discussed in section 4. At the time of this writing, the nature of these v erbal deriv ations is perhaps the most debated problem in Bantu syntax. There are two quite different types of theories: Verbal deriv ations hav e been described as morphological consequences of basically syntactic rules in the GB framework or its predecessors (cf. Vitale 1981 for Swahili, Baker 1988a), or as lexical deriv ations with subsequent syntactic effects in frameworks like Case Grammar or Relational Grammar (cf. Driev er 1976, Khamisi 1985; see also Kimenyi 1980 for Kinyarwanda, Bresnan/Kanerv a 1989 and Alsina 1990 for Chichewa). Let us hav e a short look at current versions of these theory types. According to Baker (1988a), the v oice system is based on syntactic mov ement of lexical heads (X°) to other heads, leading to incorporation. For example, in causativ es the v erb of a clause embedded by a predicate CAUSE is raised to that predicate, leav ing a trace (example: CAUSE [shorse drink water] → drinkiCAUSE [shorse ti water]). This explains the origin of causativ e v erb forms, the fact that the original subject (horse) becomes the ob-

1411

ject (in Swahili, the causativ e v erb allows for case-marking the adjacent NP horse as an object), and certain locality phenomena. Applicativ es, on the other hand, consist in the mov ement and incorporation of a preposition (example: cook [food] [FOR children] → cook-FORi [tichildren] [food]). As children gets its case from the preposition, which is now part of the complex v erb, it is consequently promoted to object status. — In a framework like this one, general syntactic principles can be applied to explain phenomena of v erbal deriv ations and their interactions. Also, the fact that syntactical changes associated with morphological operations occur in the same sequence as these morphological operations (Baker’s “mirror principle”) gets a natural explanation. According to the work of Bresnan, Mchombo and Alsina, v erbal deriv ations can be described by morpholexical operations concerning the semantic arguments of a v erb, and some general principles of grammatical function assignment. In the case of causativ es, two new arguments are added to the arguments of the basic v erb, an agent and a patient, and the patient is referentially identified with that argument of the basic v erb that is highest on a scale of arguments (where agent > benefactiv e > goal > instrument > patient > locativ e). An example: fall(ag, pt) → fall-CAUS(ag, pt (ag, pt)), where underlining expresses referential identification (the patient of CAUSE is the agent of the embedded proposition). Independent rules of GF assignment apply to that structure. Basically, GF’s are traced back to features [±o] for “objectiv e”, and [±r], for “thematically restricted”, where subject is [—r, —o], direct object is [—r, +o], and oblique (complement) is [+r, — o]. These features are assigned to thematic roles by specific rules and restrictions like “assign [+o] or [—r] to patients”, “assign [—o] to all non-patients”, “do not assign [+o] to arguments higher than instrumental”, or “do not assign [—r] more than once”. In the example at hand, the arguments of fallCAUSE(ag, pt, (ag, pt)) would get the following features: ag gets [-o], as it is a nonpatient; pt-ag gets [—r], as it contains a patient but cannot get [+ o] because it also contains an agent; finally pt gets [+o], as it is a patient and [—r] has been assigned already. In a second step, the feature assignment is completed by rules such as: “assign [—r] to the highest argument if it has feature [— o], otherwise assign [—o] to the argument with

XXI. Syntaktische Skizzen

1412

feature [—r]”, and “assign [+o] and [+r] where possible”. In the case at hand, we end up with the three arguments hav ing the features [—r, —o], [—r, +o] and [+r, +o] respectiv ely, which is spelled out as subject, direct object, and complement. Similar rules allow for the deriv ation of applicativ es, passiv es, stativ es, and reciprocals. And the feature assignment rules also determine GF-selection with simple v erbs — for example, the two arguments of hit(ag, pat) will end up with the features [—r, — o] and [—r, + o]. Although I hav e my reserv ations about the specific lexical theory sketched abov e (it might simply show how much one can “explain”, giv en enough features, feature rules, and ingenuity), I think that there are good arguments for an analysis of this type. One is that the result of v erbal deriv ations often is so idiosyncratic that a syntactic deriv ation is quite unlikely (e. g. the intensiv e meaning of causativ es, as in nya ‘rain’ — ny-esh-a ‘rain hard’). Another argument (cf. Alsina 1990) is that truly morphological operations, like nominalizations, may occur after v erbal deriv ations (cf. pend-w-a ‘lov e-PASS’, m-pend-w-a ‘friend’; fa ‘die’, f-ish-a ‘kill’, m-f-ish-a ‘murderer’; kema ‘shout’, kem-e-a ‘to rebuke’, mkem-e-o ‘a rebuke’). Furthermore, the syntactic approach does not explain the similarity between the GF-assignment of simple v erbs and deriv ed v erbs. For example, applicativ es that introduce benefactiv es create v erbs in which the benefactiv e is the direct object; but the same is true with basic v erbs that come with a benefactiv e object, such as pa ‘give’. According to the syntactic approach, this is a pure coincidence. Baker (1988a) tries to motiv ate it by assuming that v erbs like pa actually are applicativ e as well; howev er, there is no independent motiv ation for that, and the fact that pa can hav e a regular applicativ e (cf. 4.5.) which is not possible with other applicativ e v erbs in Swahili, argues against this hypothesis. Also, some of the generalizations for which Baker gav e a theoretical explanation prov ed to be wrong, for example that there are no causativ es after applicativ es (cf. 4.8.5.), or that there are no applicativ es after passiv es (cf. Alsina 1990 on Chichewa). As for the mirror principle, it has been shown (Alsina 1990) that a lexical approach can explain the correspondence between morphological structure and syntactic structure equally well, as soon as the scope of morphological affixes can be expressed.

8.

Evidence for Clausal Phrase Structure

In this section, I will discuss some phenomena that argue for the assumption of specific syntactic structures, especially with respect to the IP and the CP (cf. Chomsky 1986). 8.1.  There is interesting morphological ev idence for an I (INFL) node as a part of the complex v erb that consists of the subject and the tense prefix, but excludes the object prefix and the finite v erb. This structure can be exemplified as follows (alternativ ely, one might assume that V is mo v ed to I):

8.1.1.  Drawing mainly on the distribution of referential affixes based on “o”, BarrettKeach (1986) arriv ed at an interesting argument for assuming an INFL node. First, she observ es that the o-affix typically occurs word-finally, for example in the relativ e pronoun (amba-o), in the simple synthetic relativ e form (wa-imba-o ‘who sing’), as an object prefix in possessiv e predicates (cf. mwalimu ana-o watoto ‘the teacher has children’), or in morphological contractions like na-o ‘and they’. Consequently, she argues that the o-affix is a constituent-final allomorph of the pronominal affix. But as we hav e seen in section (5.1.2.), the o-affix also seems to occur word-internally with synthetic relativ es like wa-li-o-mw-andikia ‘(they) who wrote him’, which seems to be an exception. Howev er, under the assumption that the subject prefix and the “tense” prefix form a constituent, INFL, the o-affix simply could be analyzed as a clitic to that constituent: [INFLwa-li-o] [VPmw-andika]. (Note: This analysis of o-affixes as constituent-final allomorphs faces a problem, as we hav e both forms with deictics, as in hi-li and hi-lo; cf. section 2.3. Howev er, both deictic forms arguably cannot be analyzed as morphological complex anymore, due to morphological idiosyncrasies like the

78.  Swahili

1413

choice of the v owel after h and the form huyo instead of the regular h-ye).

it is not used with the ki tense; cf. ni-kí-la ‘when I eat’.

8.1.2.  Barrett-Keach has a second argument for INFL, which relates facts about word accent to syntax and morphology. Swahili has stress on the penultimate syllable. In compositions, the stress of the first word becomes secondary, cf. wafànya kázi ‘workers’. Now, with finite v erbs containing a tense prefix, we also find this secondary penultimate stress if we assume that the subject prefix, the “tense” prefix, and, if present, the relativ e prefix form a constituent. Some examples: [nì-na] [kupénda ‘I lov e you’, [ni-me-kwìsha] [andíka] ‘I hav e already written’, [wa-nà-o] [m-pénda] ‘(they) who love him’.

8.1.4.  Myers (1987) showed that Shona has the same internal v erb structure as proposed for Swahili, using ev idence from tone sandhi phenomena (Meeussen’s Rule, which says that a H(igh) lowers an immediately following H, does apply between words and between word-internal INFL and VP, but not within INFL or VP). He also showed that Meeussen’s rule does not apply within v erbs with a subjunctiv e or participial INFL. This distinction between two kinds of INFL is reminiscent of the two kinds of temporal prefixes in Swahili that do or do not require kuwith monosyllabic v erb stems. Myers argues for the [INFL VP] analysis for both cases; howev er, he distinguishes between two different lev els of morphological combinations, where Meeussen’s Rule affects only the higher-lev el combination. Similarly, we can assume that Swahili ni-na-ku-la and ni-ki-la differ in the morphological lev el at which the INFL’s ni-na and ni-ki combine with the VP.

8.1.3.  An independent argument for an internal word boundary in finite v erbs is due to Brandon (1975). He observ es that words hav e typically two or more syllables, arguably due to the penultimate stress rule, and that morphological rules ensure that. For example, whereas adjectiv es (and nouns) in class 5 do not normally show agreement, monosyllabic adjectiv es (and nouns) hav e retained the prefix ji (cf. duka kubwa ‘big shop’, duka jipya ‘new shop’). Now, this “bisyllabic word constraint” can explain an interesting fact about Swahili morphology (cf. Ashton 1944, 142 f), namely that monosyllabic v erb roots such as la ‘eat’, nywa ‘drink’, ja ‘come’, fa ‘die’ occur with the infinitiv e prefix in many tenses. For example, instead of *ni-na-la ‘I’m eating’, we hav e ni-na-ku-la. We can assume that there is a word boundary between ni-na and la, that la does not satisfy the bisyllabic word constraint, and hence has to be expanded, which is done by the infinitiv e prefix as in [ni-na] [ku-la]. Now, the interesting thing is that in case we hav e an object prefix, the infinitiv e prefix is not used; v iz. ni-na-ya-la mayai ‘I’m eating the eggs’. This can be explained by the fact that the object prefix and the v erb stem form a word that already meets the bisyllabic word constrained, as in [ni-na] [ya-la]. Again, we hav e to assume that the major boundary is after the tense prefix. There are, howev er, two residual problems with this analysis: First, there are also bisyllabic v erb roots which need a dummy prefix ku, namely enda ‘go’, isha ‘finish’ and others in some dialects; note that all these v erbs hav e an initial v owel. Second, the dummy ku is used only with certain tenses. For example,

8.1.5.  Myers cites ev idence that the [INFL VP] structure is indeed v ery common in Bantu and may be inherited from the protolanguage. It might be a reflex of the syntactical origin of the morphological structure of the v erb, which arguable goes back to an auxiliary + VP — construction (cf. Krifka 1983). For example, many temporal prefixes are related to v erb stems with an original meaning that is apt for a grammaticalization as an auxiliary; cf. future ta(ka) ‘want’, past li ‘be’, perfect me, reconstructed as *mala, possibly related to maliza ‘finish’. Especially interesting is the v erb kwisha ‘end’; it can gov ern infinitiv es or v erb stems, as in watoto wa-mekwisha (ku)ondoka ‘the children hav e already left’. 8.1.6.  Another argument for an INFL node comes from VP conjunction. If two v erbal predicates hav e the same subject and tense, they can be conjoined by na, but only the first bears a subject and a tense prefix, whereas the second has only the infinitiv e prefix ku. One example: (23) Watoto wa-li-mw-andikia mwalimu na ku-ki-soma kitabu. ‘The children wrote to the teacher and read the book’

1414

This can be explained if we assume that subject and tense prefix form an INFL node which gov erns a VP node. In example (23), the INFL wa-li gov erns the conjoined VP [VP[VPmw-andikia mwalimu] na [VPku-ki-soma kitabu]]. The infinitiv e marker ku of the v erb in the second VP can be traced back to a rule which makes the incomplete ki-soma a surface word. This “dummy” analysis of ku- is motiv ated independently, as we find infinitiv e v erb forms instead of agreeing v erb forms in special registers, for example in the headlines of newspapers; an example of that is Algeria ku-tu-saidia ku-tafuta mafuta ‘Algeria to help us to search oil’. This analysis predicts that the subject and tense prefix of the first v erb is the head of the whole sentence. Giv en the plausible assumption that in synthetic relativ es the relativ e affix occurs at the head of a clause, we can test this hypothesis: In a synthetic relativ e clause with a conjoined VP, we should expect that the relativ e pronoun is suffixed to the subject and tense prefix of the first v erb. This is indeed true; v iz. watoto wa-li-o-mw-andika mwalimu na ku-ki-soma kitabu ‘the children who wrote to the teacher and read the book’. 8.1.7.  We hav e found four arguments for an INFL node: the o-Allomorph and synthetic relativ e clauses of type I, penultimate word stress, the bisyllabic word constraint and monosyllabic v erb roots, and VP conjunction. Now there are certain restrictions with these constructions, as they may occur only with some tenses. The analysis giv en so far would be conv incing if all four phenomena would pattern similarly, that is, would be acceptable with the same tenses. But this is not the case: Synthetic relativ e clauses are possible with the three tense markers li (past), na (present) and taka (future) only. Infinitiv e prefixes with monosyllabic v erbs show up with these tenses, but in addition with tense markers like me (perfect) and nge/ng ali (conditional). And VP conjunction as discussed is possible with every tense marker. 8.2.  It may be tempting to analyze INFL further, along the lines of Pollock (1989). More specifically, we might introduce phrases AgrS-P (subject agreement) and TP (tense), and arriv e at structures like the following one: (24) [AgrS-PJuma[AgrS’ [a-[TPli-[VPki-soma kitabu]]]]]

XXI. Syntaktische Skizzen

Note that this dev iates from Pollock’s assumptions as AgrS-P dominates TP, and not v ice v ersa. There is no simple way to include negation into this setup (which is, in Pollock’s theory, a phrasal node NegP occuring between TP and AgrS-P). This phrase should dominate AgrS’, as it is typically marked by an outermost prefix ha (cf. 3.3.); but we might also hav e changes of the tense prefix and the final v owel (cf. h-a-ku-ki-som-i ‘he didn’t read’). There is one point in fav our of a complex analysis of INFL, howev er. Assuming analysis (24), the relativ e affix should be suffixed to the tense affix (the specifier of TP). If there were no tense affix, we simply could not form this construction. This is borne out by the facts; in tenseless relativ e clauses (cf. 5.1.3.), the relativ e affix is suffixed to the whole v erb, instead of the subject prefix (cf. a-ki-somacho, not *a-cho-soma). 8.3.  Let us now hav e a closer look at relative clauses. We will see that their structure prov ides us with clues about the structure of complementizer phrases (CPs). 8.3.1.  Of the three types of relativ e clauses (cf. 5.1.), the one based on amba clearly is the most recent one. This relativ e marker originated with an now obsolete v erb amba ‘say’ that occurred in the second synthetic relativ e form and subcategorized for a sentence. That origin may be still reflected by the fact that amba-relativ es may hav e a complementizer kwamba; cf. watoto amba-o kwamba mwalimu a-li-wa-ona. Barrett-Keach (1980) takes this as an argument to analyze amba as a specialized v erb in current Swahili and proposes the structure (25a). Howev er, another possibility is that although this type of relativ e clause must be traced back to a v erbal construction, amba- now is an agreeing relativ e pronoun in Spec-CP position, and kwamba is an optional complementizer in C-position (cf. (25b)); similar structures occur in southern v arieties of German, as die Kinder die wo der Lehrer gesehen hat. (25a) watoto [VPamba-o [S’ kwamba [Smwalimu a-li-wa-ona]]] (25b) watoto [CPamba-o [C’ kwamba [IPmwalimu a-li-wa-ona]]] Synthetic relativ e clauses cannot occur with that complementizer; cf. *watoto (kwamba) ali-o-wa-ona (kwamba) mwalimu. Hence they should either be analyzed as being of the category IP, or alternativ ely the mov ement of

78.  Swahili

the v erb containing the relativ e affix, a-li-owa-ona, to the Spec-CP position is barred because of the intervening complementizer. 8.3.2.  One difference between the relativ e clause types is that with synthetic relativ es, there is a clear tendency for postposing the subject (cf. Giv ón 1972 for a similar rule in Chibemba). This could be taken as an argument that the relativ e v erb is mov ed to the initial Spec-C position, leav ing the subject and other constituents behind. Howev er, as we hav e seen in (5.1.2.), or with jinsi wazee wa-li-vyo-ishi ‘the way the forefathers liv ed’, this rule is a tendency at best. According to Barrett-Keach (1980), the subject may occur in front of the v erb if it is stressed; this might suggest that a stressed subject may mov e into Spec-CP position, thus prev enting the relative verb to do so. 8.3.3.  Another difference between ambaclauses and synthetic relativ e clauses is that the relativ e pronoun in amba-clauses may be mov ed from an embedded clause. BarrettKeach giv es examples like kitabu amba-cho ni-li-mw-ona watoto [u-li-o-wa-pa] ‘the book which I saw the children to whom you gav e it’ v s. *kitabu ni-li-cho-mw-ona watoto [u-li-owa-pa]. This can be explained under the assumption that INFL as the site for marking relativ e clauses is only accessible for constituents within the local IP, whereas Spec-C is accessible for relativ e markers that are mov ed cyclically from embedded constituents. 8.3.4.  A related difference is that amba-relativ es allow for resumptiv e pronouns, whereas synthetic relativ es don’t. Barrett-Keach giv es examples like watoto amba-o (wao) wa-likwenda ‘the children who left’ v s. *watoto (wao) wa-li-o-kwenda (wao). Note, howev er, that possessiv e resumptiv e pronouns are possible in both cases when they occur embedded in an NP; cf. watoto ni-li-o-ki-soma [NPkitabu ch-ake] ‘the children of whom I read their book’. Following Barrett-Keach, we can explain these facts in terms of a bindingtheoretic restriction, saying that a pronoun cannot be bound by an o-antecedent in its minimal gov erning category (where IP or S’, and NP are gov erning categories). But note that this applies only to free pronouns and not to affixes (cf. kitabu ni-li-CHO-KI-soma ‘the book which I read’); hence we could conclude that affixes are not part of the syntax

1415

proper. 8.4.  The data on relativ e clauses suggest that Swahili clauses might be either IP’s or CP’s, where the full-fledged CP structure is v isible in the more recent construction. Incidentally, we find similar v ariations in other cases of sentential complementation (5.2.) and adv erbial clauses (5.3.): Ov ert complementizers typically represent later dev elopments; they are either deriv ed from v erbs or are borrowed from Arabic. The morphological marking of subordination, on the other hand, clearly represents the more traditional pattern. This suggests that the language is in a stage of syntactic transition. 8.5.  It is interesting in this respect to look at WH-movement, as this is typically analyzed as a mov ement to the Spec-C position. In Standard Swahili, WH-words do not mov e at all. Howev er, Alemán (1987) reports that there is a southern dialect in which syntactic mo v ement of Wh-Words into complementizer position is possible as a v ariant. (A similar case is reported for Kinande by Authier 1988). For example, for the question ‘Who did you say you saw’ we hav e two alternativ es (Standard Swahili has only the first one). (26a) U-li-sema kwamba nani a-li-kuja? (26b) Nanii a-li-sema kwamba ti a-li-kuja? Aléman observ es that more than one Whword may be mov ed, as in (27a), and that they may move to the same position (27b): (27a) Nanii u-me-amini kwamba wapij ti ame-kwenda tj? ‘Who do you believe went where? (27b) Nanii wapij u-li-mw-ona ti tj? ‘Whom did you see where? Aléman notes certain restrictions with two mov ed Wh-words, such as the following: At least one of the Wh-words must agree with the v erb. If two Comp-nodes are present, the agreeing Wh-words always mov es to the higher node. If two Wh-words fill the same node, then one of them must be agreeing, and the other one must be non-agreeing. Although data are lacking, it seems probable that WH-mov ement is an innov ation; at least it is not mentioned at all in traditional grammars. This dev elopment of WH-mov ement might be seen in connection with the establishment of full-fledged CPs in dependent clauses.

XXI. Syntaktische Skizzen

1416

8.6.  Let us now hav e a look at a construction which I hav e suppressed until now, namely so-called “compound tenses” (Ashton 1944): (28) watoto wa-li-kuwa wa-me-ki-soma kitabu. ‘the children had read the book’ According to morphological criteria, this sentence has two finite v erbs, wa-li-kuwa ‘were’ and wa-me-ki-soma ‘hav e read it’, which is contrary to basic assumptions of X-bar theory. In general, compound tenses consist of the auxiliary kuwa ‘to be’ with a tense prefix (li, na, ta, me, ka, ng e) or in the subjunctiv e, and a main v erb with another tense prefix, typically with some aspectual meaning, like perfect (me), imperfectiv e (ki), or progressiv e (na). Both agree with the subject, and the main v erb may also show object agreement. The two v erbs follow each other in that order; typically, they are adjacent, but sometimes we find an adv erb between them, as in watoto walikuwa daima wakig ombana ‘the children were all the time quarreling’. At first sight, it looks as if these clauses contain two finite v erbs. Howev er, there is a reason for assuming that the two v erbs are not equally ranked syntactically, but that the auxiliary forms the head of the construction. This follows from the explanation of relativ e clause structures giv en abov e. If we assume that the o-affix is realized as a clitic to the head of the relativ e clause, then we hav e a means to determine the head of a sentence with a compound tense: We simply put it into the first synthetic relativ e form and look where the o-affix ends up: (29) watoto wa-li-o-kuwa a-me-ki-soma kitabu ‘the children who had read the book’ It is the auxiliary; so we should assume that the auxiliary is the head of the construction. With this result, a plausible syntactic analysis of a sentence with a double v erb construction would be as follows:

I assume that the auxiliary stem kuwa embedds an IP without an ov ert subject (alternativ ely, an I’). Thus, we hav e only one finite verb per IP. In concluding the discussion of the double finite v erb construction, I want to mention that either of the two v erbs can be negated, with characteristic meaning differences (cf. Brauner/ Herms 1979, 242). For example, we hav e ha-kuwa a-na-fanya kazi (negated auxiliary; ‘he was without work’) v s. a-li-kuwa hafanyi kazi negated main v erb, ‘he did not work (although there was the possibility to do so)’).

9.

Conclusion

This concludes this short treatise of syntactic phenomena in Swahili. With the v iv id interest of different linguistic frameworks in Bantu syntax, we certainly can expect that the next years will bring with them many bold theoretical generalizations, fascinating discov eries, and embarassing falsifications. In a certain respect the Bantu languages may prov e to be an excellent laboratory for research into syntactic principles and parameters, as there are so many of them (around 500) that differ from each other in slight and, hopefully, principled ways.

10. References Abdulaziz, M. H. 1979. Muyaka. Nairobi. Alemán, Iris Belinda. 1987. Agreement and the ECP. Cornell University Ph. D. Dissertation. Allan, Keith. 1983. Anaphora, Cataphora and Topic Focusing: Functions of the Object Prefix in Swahili. Current Approaches to African Linguistics, ed. by Ivan R. Dihoff, Dordrecht. 322—335. Alsina, Alex. 1990. Where is the Mirror Principle? Stanford. MS. —, and Sam Mchombo. 1990. The Syntax of Applicativ es in Chichewa. Natural Language and Linguistic Theory 8. Ashton, Ethel Osteli. 1944. Swahili Grammar (including intonation). 2nd. edn, 1947. London. Authier, J.-Marc P. Null Object Construction in Kinande. Natural Language and Linguistic Theory 6. 19—37. Baker, Mark. 1988a. Incorporation. A Theory of Grammatical Function Changing. Chicago. —. 1988b. Theta Theory and the Syntax of Applicativ es in Chichewa. Natural Language & Linguistic Theory 6. 353—89.

78.  Swahili

Barrett-Keach, Camilla. 1980. The Syntax and Interpretation of the Relativ e Clause Construction in Swahili. Amherst: Uni v ersity of Massachusetts Ph. D. Dissertation. —. 1986. Word-Internal Ev idence from Swahili for AUX/INFL. Linguistic Inquiry 17. Brandon, Frank Robert. 1975. A Constraint on Deletion in Swahili. Proceedings of the 6th Conference on African Linguistics, Ohio State Univ ersity Working Papers in Linguistics 20, 241—59. Brauner, Sieg mund, and Irmtraud Herms. 1979. Lehrbuch des modernen Swahili. Leipzig. Bresnan, Joan, and Jonni M. Kanerva. 1989. Locativ e Inv ersion in Chichewa: A Case Study of Factorization. Linguistic Inquiry 20. 1—50. —, and Sam A. Mchombo. 1987. Topic, Pronoun and Agreement in Chichewa. Language 63. 685— 739. Byarusheng o, E. R.; A. Duranti; and L. Hyman. (eds.) 1977. Haya Grammatical Structure. Southern California Occasional Papers in Linguistics 6. Chomsky, Noam. 1986. Barriers, Cambridge, MA. Closs, Elizabeth. 1970. Some Copula Constructions in Swahili. Journal of African Languages 6. 105— 30. Contini-Morava, Ellen. 1989. Discourse Pragmatics and Semantic Categorization. The Case of Negation and Tense-Aspect with Special Reference to Swahili. Berlin. Eastman, Carol M. 1969. A Descriptiv e Statement of Operativ e Verbal Extensions in Swahili. African Language Studies X. 31—46. Faltz, Leonard M. 1978. On Indirect Objects in Universal Syntax. Chicago Linguistic Society. Givón, Talmy. 1970. The resolution of Gender Conflicts in Bantu Conjunction: When Syntax and Semantics Clash. Chicago Linguistic Society 6, 250—261. —. 1972. Studies in ChiBemba and Bantu Grammar. Studies in African Linguistics, Supplement 3. —. 1976. Topic, Pronoun and Grammatical Agreement. Subject and Topic, ed. by Charles N. Li. New York. Greenberg , Joseph H. How Does a Language Acquire Gender Markers? Univ ersals of Human Language, Vol. 3, ed. by J. H. Greenberg. Stanford. Press. Guthrie, Malcolm. 1962. The Status of the Radical Extension in Bantu Languages. Journal of African Linguistics 1. 202—20. Harries, Lyndon. 1970/71. Inalienable Possession in Swahili. African Language Review 9. 140—47. —. 1970. The Phrasal Predicate in Swahili. African Language Studies IX. 191—210. Hawkinson, Annie. 1979. Homonymy v ersus Unity of Form: The Particle -A in Swahili. Studies in African Linguistics 10. 81—109.

1417

Heukstra, Teun A., and Gerrit J. Dimmendaal. 1983. An Alternativ e Approach to Swahili Grammar. Rev iew of A. J. Vitale: ‘Swahili Syntax.’ Lingua 60. 53—85. Hinnebusch, Thomas J., and Robert S. Kirsner. 1980. On the Inference of “Inalienable Possession” in Swahili. Journal of African Languages and Linguistics 2. 1—16. —, and Sarah M. Mirza. 1979. Kiswahili. Msingi wa kusema, kusoma na kuandika. Washington D. C. Höftmann, Hildeg ard. 1979. Wörterbuch SwahiliDeutsch. Leipzig. Hopper, Paul. 1979. Aspect and Foregrounding in Discourse. Discourse and Syntax (Syntax and Semantics, 12) ed. by Talmy Givon, 213—41. Kamusi ya Kiswahili Sanifu. 1981. Dar es Salaam. Nairobi. Khalid, Abdallah. 1977. The Liberation of Swahili from European Approbiation. Nairobi. Khamisi, Abdu Mtajuka. 1985. Swahili Verb Derivation. University of Hawaii Ph. D. dissertation. Kimenyi, Alexandre. 1980. A Relational Grammar of Kinyarwanda. Los Angeles. Krapf, Ludwig . 1850. An Outline of the Grammar of the Swahili Language. Tübingen. Krifka, Manfred. 1983. Zur semantischen und pragmatischen Motiv ation syntaktischer Regularitäten: Eine Studie zur Wortstellung und Wortstellungsveränderung im Swahili. München. Leonard, Robert A. 1985. Swahili Demonstrativ es: Ev aluating the v alidity of competing semantic hypotheses. Studies in African Linguistics 16. 281—93. Li, Charles, and Sandra A. Thompson. 1976. Subject and Topic. A New Typology of Language. Subject and Topic, ed. by Ch. Li, 459—89. New York. Maw, Joan. 1969. Sentences in Swahili: A study of their internal relationship. London. —. 1970. Some Problems in Swahili Clause Structure. African Language Studies XI. 257—71. —. 1976. Focus and the Morphology of the Swahili Verb. Bulletin of the School of Oriental and African Studies 39. 390—401. —. 1985. Twende! A Practical Swahili Course. Oxford. —, and John Kelly. 1975. Intonation in Swahili. London. Meeussen, A. E. 1967. Bantu Grammatical Reconstructions. Africana Linguistica 3. Miehe, Gudrun. 1979. Die Sprache der älteren Swahili-Dichtung. Phonologie und Morphologie. Berlin. Myers, Scott. 1987. Tone and the Structure of Words in Shona. Amherst: Univ ersity of Massachusetts Ph. D. dissertation.

XXI. Syntaktische Skizzen

1418

Nichols, J. 1986. Head-Marking and Dependentmarking Grammar. Language 62. 56—119. Nurse, Derek, and Thomas Spear. 1985. The Swahili: Reconstructing the History and Language of an African Society. Philadelphia. Pollock, Jean-Yves. 1989. Verbs Mov ement, Univ ersal Grammar, and the Structure of 18. Linguistic Inquiry 20. 365—424. Polomé, Edg ar C. 1967. Swahili Language Handbook. Washington D. C. Riddle, E. 1975. Relational Grammar and Some Aspects of Swahili Syntax. Proceedings of the 6th Conference on African Linguistics, Ohio State Uni v ersity Working Papers in Linguistics 20, 260—69. Roberts, Linda, and Marjorie Wolontis. 1972. Conjunction and Concord in Bantu. Third Annual Conference on African Linguistics, ed. by E. Voeltz. Bloomington. Sacleux, H. 1909. Grammaire des dialectes swahilis. Paris. Schicho, Walter. 1982. Syntax des Swahili v on Lubumbashi. Veröff. der Institute für Afrikanistik und Ägyptologie der Universität Wien. Scotton, Carol Myers. 1967. Semantic and Syntactic subcategorization in Swahili Causativ e Verb Shapes. Journal of African Languages 6. 249—67. —. 1981. Extending Inalienable Possession: The Argument for an Extensiv e Case in Swahili. Journal of African Languages and Linguistics 3. 159—

74. Shepardson, Kenneth N. 1982. An Integrated Analysis of Swahili Augmentati v e-Diminuiti v es. Studies in African Linguistics 13. 52—76. Steere, Edg ar. 1870. Handbook of the Swahili language as spoken at Zanzibar. Rev ised 2nd edn, A. C. Madan, 1894. Tyler, A. 1985. Swahili Relativ e Clauses: A Generalized Phrase Structure Analysis. Studies in African Linguistics Suppl. 9, 298—303. Vitale, Anthony J. 1981. Swahili Syntax. Dordrecht. Wald, Benji. 1979. The Dev elopment of the Swahili Object Marker: A Study of the Interaction of Syntax and Discourse. Syntax and Semantics 12. 505—23. Whiteley, Wilfred H. 1968. Some Problems of Transitivity in Swahili. London. —. 1969. Swahili: The Rise of a National Language. London. —. 1972. Case Complexes in Swahili. Studies in African Linguistics 3. 1—45. —. 1972. Notes on the Syntax of Passiv es in Swahili. African Language Review 11. 391—404. —, and J. D. Mg ang a. 1969. Focus and Entailment. Further Problems of Transitiv ity in Swahili. African Language Review 8. 18—125. Wilson, P. M. 1970. Simplified Swahili. Nairobi.

Manfred Krifka Austin, Texas (USA)

79. Tagalog 1. 2. 3. 4. 5. 6. 7.

1.

Introduction Basic Clause Structure Word Order Nominals and Other Arguments Non-Declarative Sentences and Negation Subject Properties References

Introduction

Tagalog, a member of the West Indonesian branch of the Austronesian language family, is nativ e to the southern part of the Philippine island of Luzon. Since its adoption in 1937 as the Philippine national language (under the name Pilipino), it has spread rapidly ov er the entire Philippine archipelago, and it is estimated that by the year 2000 ov er 98% of all Filipinos will speak Tagalog as a first or second language. More than 300 years of

contact with Spanish and a briefer period of contact with English hav e heav ily influenced the Tagalog lexicon, and hav e had some influence on the phonology as well (leading, for example, to a phonemicization of an originally allophonic distinction between high and mid v owels). But contact with Spanish and English appears to hav e had negligible influence on Tagalog syntax. The following selectiv e sketch presents an ov erv iew of the major syntactic structures of Tagalog, with emphasis on certain aspects of Tagalog syntax that are of particular theoretical interest. For a more complete account of the grammar as a whole, see Schachter and Otanes (1972).

2.

Basic Clause Structure

Tagalog is basically a predicate-initial language (but see section 3 for certain non-pred-

79.  Tagalog

icate-initial constructions), in which the dominant clause type contains a predicate followed by one or more argument expressions. (There is also a set of weather and time predicates that do not take following arguments: e. g., the v erb kumulog ‘(to) thunder’ and the adjectiv e maag a ‘early’.) Predicates may be v erbs, nouns, adjectiv es, or prepositional phrases. By far the greatest syntactic complexity and interest is to be found in clauses with v erbal predicates, the subject of sections 2.1 through 2.1.2. Clauses with nonv erbal predicates are discussed in section 2.2. 2.1. Clauses with Verbal Predicates A basic clause with a v erbal predicate contains a v erb followed by a string of arguments all but one of which are case-marked to indicate the argument’s semantic role. (The order of arguments after the initial v erb is in general not fixed — see section 3 for details.) The one argument per clause that is not casemarked has its semantic role indicated by an affix on the v erb. (There is also affixation for aspect, with finite v erbs occurring in one of three aspect-marked forms: irrealis, imperfectiv e, perfect. Since aspect does not in general affect syntactic structure, it is henceforth ignored.) The non-case-marked argument will be referred to here as the Trigger. (Some prev ious treatments hav e referred to the argument in question as the topic and some as the subject. Howev er, as will become clear below, each of these labels appears to carry some inappropriate connotations, making a neutral term like Trig g er seem preferable.) In the following examples (and subsequent examples through the end of section 2.1.2), the v erbal affix that indicates the semantic role of the Trigger is italicized, as is the Trigger argument itself. (The following abbrev iations are used in the glosses: AT = Actor-Trigger affix, T = Trigger marker, Th = Theme marker, D = Direction marker, ThT = Theme-Trigger affix, A = Actor marker, DT = Direction-Trigger affix. Note that the Trigger, although not case-marked, has its own distincti e marking.)

1419

As the translations indicate, the Trigger argument is regularly interpreted as definite, a non-Trigger Theme as indefinite, and other non-Trigger arguments as either definite or indefinite. It is the regular association of the Trigger with definiteness that has led some prev ious analysts (e. g., Schachter and Otanes 1972) to identify the Trigger as a topic, but the Trigger does not necessarily hav e the other pragmatic properties typically associated with topics: e. g., its referent is not necessarily what the sentence is about. The v ariation in v erb form illustrated in sets of sentences like (1)—(3) is reminiscent of v oice-marking, and one might be tempted to analyze (1) as activ e and (2) and (3) as different types of passiv es (in which case the Trigger would correspond to the activ e or passiv e subject). Howev er, there is good reason to distinguish the Tagalog Triggermarking system from the v oice systems of familiar languages. In the first place, non-Actor-Trigger clauses such as (2) and (3) generally outnumber Actor-Trigger clauses such as (1) in texts — a situation that is certain ly unexpected if (1) is activ e and (2) and (3) passiv e. Second, while v oice systems typically make just a two-way distinction (between activ e and passiv e), the Tagalog Triggermarking system is much richer, allowing a wide range of semantic roles to be registered on the v erb. For example, in addition to arguments with the semantic roles illustrated in (1)—(3), arguments with roles like beneficiary, instrument, location, and cause may also serv e as Triggers, with a distinct v erbal affix indicating the role in each case. There also seems to be good reason to reject the term focus — which is the traditional term in the Philippinist literature — for the v erb-form v ariation in question. Focus now has a well-established usage as a term of pragmatics, where it indicates new or salient information. This pragmatic function is irrelev ant to the registration of the role of the Trigger, which, as noted, is regularly definite

1420

— a non-focus-like property. It therefore seems adv isable to use the neutral term Trigg er system for the phenomenon in question, and to refer to the v erb forms that comprise this system as x-Trigger forms, where x ranges ov er the semantic roles that the system distinguishes. As for the functions of the Trigger system — i. e., the conditions that determine the choice of one or another Trigger form — an unpublished textual study by Fay Wouk suggests that the central function has to do with marking the referential status of the arguments. The Trigger is always referential. Its reference is usually definite, but may it also be generic, or indefinite but specific (‘a certain ...’). Non-Trigger arguments, on the other hand, are not necessarily referential, and one — the non-Trigger Theme — is almost always indefinite. The choice of Trigger is thus to a considerable extent conditioned by the relativ e referential status of the arguments in the clause. If, for example, the argument with the role of Theme has a definite referent, the clause will in the great majority of cases be a Theme-Trigger clause. Each Tagalog v erb is lexically specified for the semantic roles of the arguments for which it is subcategorized. The system of semantic roles assumed here is based on the thematicrole system of Jackendoff (1972) and much subsequent work in generativ e grammar, with certain modifications that are needed to account for the Tagalog facts. The central semantic role in Jackendoff’s system is the role of Theme, which is the role played by the argument whose position, possession, etc. are at issue. (“Position” and the like must often be interpreted abstractly: e. g., in John explained the situation to Mary, the situation is the Theme.) Other roles in the system are Agent (typically, the causer of the ev ent expressed), Location, Source, and Goal. One important characteristic of the system is that a single argument may in some cases be assigned two different roles. For example, in (1)—(3), abov e, babae ‘woman’ is both Agent and Source. The modifications in the thematic-role system that are assumed here inv olv e the recognition of two macroroles, Actor and Direction. (A macrorole is a generalized semantic role that subsumes certain sets of thematic roles.) The Actor is the argument whose role in any ev ent is presented as central, the argument from whose perspectiv e or point of v iew the ev ent is presented. (In English translations of Tagalog, the Actor regularly cor-

XXI. Syntaktische Skizzen

responds to the activ e subject.) If the v erb assigns the thematic role of Agent to an argument, then this argument is always the Actor. For non-agentiv e experiential v erbs, the Actor is the experiencer (which, being the locus of the experience, carries the thematic role of Location in Jackendoff’s system), as in:

Finally, if the v erb assigns neither Agent nor Location (experiencer), the Actor may hav e the semantic role of Goal, Source, or Theme, the choice being lexically determined. For example, the v erb AT tumanggap/ThT tang g apin ‘receiv e’ in (6)—(7) selects the Goal rather than the Theme as Actor, while the v erb AT dumating/DT dating nan ‘reach’ in (8)—(9) selects the Theme rather than the Goal. (The AT infix -um-, which is present in the infinitiv al forms tumang g ap ‘receiv e’ and dumating ‘reach’, does not occur in irrealis forms — hence the absence of ov ert AT marking in (6) and (8).)

As the glosses show, Actor assignment takes precedence ov er thematic role assignment, so Actors are all treated alike syntactically: i. e., as non-Triggers they are all marked in the same way (e. g., by the case-marking preposition ng), as Triggers they all trigger the selection of a member of the same set of affixes (e. g., mag - or -um-), and they all ev ince the other Actor-related properties discussed in section 6.

79.  Tagalog

Direction is a metarole that subsumes Goal and Source, which again manifest syntactic similarities hav ing to do with case marking and Trigger affixation. Compare (8)—(9), in which the Direction is a Goal, with (10)—(11) (inv olv ing the v erb AT umalis/ DT alisan ‘leav e’), in which the Direction is a Source:

As the examples show, a Direction argument, whether it is a Goal or a Source, takes the same case marking (e. g., the preposition sa) and triggers the same v erbal affixation (e. g., -an). 2.1.1. Case-Marking and Trigger-Marking of Arguments All non-Trigger arguments are case-marked for their semantic role or metarole. All Trigger arguments are marked as such, and their semantic (meta)role is marked on the v erb. The way in which arguments are case-marked or Trigger-marked v aries according to the syntactic class of the argument. Nouns are marked prepositionally, while personal and demonstrativ e pronouns are marked by morphological variation. For common nouns, the Trigger-marking preposition is ang, the Actor-marking and Theme-marking preposition is ng (the conv entional spelling for phonetic [naŋ]), and the Direction-marking and Location-marking preposition is sa. Personal names hav e their own distinctiv e prepositions: the personalname counterparts of ang , ng , and sa are, respectiv ely, si, ni, and kay. Personal and demonstrativ e pronouns each occur in three forms, which may be called, on the basis of their functional parallelism to phrases introduced by ang , ng , and sa, the ang -form, the ng-form, and the sa-form respectiv ely. Angform and ng-form pronouns are nev er preceded by prepositions, and neither are saform demonstrative pronouns. Howev er, saform personal pronouns functioning as Direction or Location arguments are preceded by the preposition sa. (Sa-form personal pronouns also occur as possessiv e predicates and possessiv e modifiers. As possessiv e predicates they are optionally preceded by sa, and as

1421

possessiv e modifiers — cf. section 4 — they occur without a preceding preposition.) The forms of the personal and demonstrativ e pronouns are summarized in Charts 1 and 2. Chart 1 Personal Pronouns ang-form ng-form sa-form Singular 1st person 2nd person 3rd person Plural 1st person exclusive 1st person inclusive 2nd person 3rd person

ako ka/ikaw siya

ko mo niya

akin iyo kaniya

kami

namin

amin

tayo

natin

atin

kayo sila

ninyo nila

inyo kanila

(The two ang-formsof the 2nd-person-singular pronoun, ka and ikaw, are respectiv ely clitic and nonclitic — cf. the discussion of clitics in section 3.) Chart 2 Demonstrative Pronouns ang-form ng-form sa-form ‘this’ ito nito dito iyan niyan diyan ‘that (near addressee)’ niyon/ doon ‘that (not near iyon addressee)’ noon 2.1.2. Trigger-Marking of Verbs As noted in 2.1, the semantic role or metarole of the Trigger argument is marked by an affix on the v erb. There is no one-to-one correspondence between roles and affixes. Some roles may be marked by two or more different affixes, and some affixes may mark two or more different roles. Nonetheless, there is in fact a good deal of predictability in the Trigger-marking system, on the basis of the semantics of the verb. Consider, for example, two major classes of three-argument v erbs, one of which takes the Trigger affixes AT mag-/ThT i-/DT -an, the other the Trigger affixes AT -um-/ThT -in/ DT -an. It turns out, as noted by Ramos (1974), that the occurrence of a particular v erb in one of these two classes is predictable from the v erbal semantics, with the distinction based on whether the v erb is “centrifugal” or “centripetal”. A centrifugal v erb is one in which the Actor is both Agent and Source (while the Direction is Goal); with such v erbs the mag -/i-/-an set of affixes is used. Thus it is this set that is used with the

1422

v erbs meaning ‘hand to’ in (1)—(3), abov e. (Other v erbs showing the same pattern include those meaning ‘giv e to’, ‘take up to’, ‘take into’, etc.) A centripetal v erb, on the other hand, is one in which the Actor is both Agent and Goal (while the Direction is Source); with these v erbs the -um-/-in/-an set of affixes is used, as in the following examples:

(Other v erbs showing the same pattern include those meaning ‘get from’, ‘buy from’, ‘ask for from’, etc.) The semantic roles and metaroles of Trigger arguments that may be marked by a Trigger affix on the v erb include: Actor, Theme, Direction, Beneficiary, Location, Instrument, and Cause. The marking of each of these roles is discussed in turn below, with emphasis on such generalizations as can be made concerning the choice of affix in each case. (The discussion will be restricted to “simple” Trigger-marking affixes, and will generally exclude those which, in addition to indicating the role of the Trigger, hav e some further function such as marking the action as intensiv e, moderativ e, accidental, etc. In general, these dual-function affixes are predictable from the simple affixes by rules of deriv ational morphology.) The two most common Actor Trigger (AT) affixes are mag - and -um-. Mag - (which for some purposes may conv eniently be analyzed as consisting of two morphemes — see discussion of Beneficiary-Trigger v erbs below, present section) most commonly occurs if the Actor is just an Agent (as in mag luto ‘cook’, mag sig arilyo ‘smoke’) or is both Agent and Source (as in mag -abot ‘hand to’, magpasok

XXI. Syntaktische Skizzen

‘take into’), while -um- most commonly occurs if the Actor is both Agent and Goal (as in humiram ‘borrow’), both Agent and Theme (as in pumasok ‘enter’), or not an Agent (as in pumuti ‘turn white’, tumang g ap ‘receive’). There are a number of v erb roots that form AT v erbs with both mag - and -um-, with magforming a three-argument v erb whose Actor is Agent and Source, and -um- forming a twoargument v erb whose Actor is Agent and Theme. Examples include AT mag-akyat ‘take up to’ v s. umakyat ‘climb’, maglabas ‘take outside’ v s. lumabas ‘go outside’ and mag pasok ‘take into’ v s. pumasok ‘enter’. (The mag - v erbs of this type, as noted abov e, all also form ThT v erbs with i- and DT v erbs with -an. The -um- v erbs of this type all also form DT verbs with -in.) In addition to mag - and -um-, there are a number of other AT affixes that occur idiosyncratically as simple affixes with certain v erb roots: e. g., ma-, maka-, and mang -, as in matulog ‘sleep’, makakita ‘see’, and mang isda ‘go fishing’ respectiv ely. Ma- and mangalso hav e certain more systematic uses as simple AT affixes. For example, ma- occurs regularly in experiential inchoativ e v erbs such as mabing i ‘become deaf’, mauhaw ‘become thirsty’, and mang - (which has mam- as an allomorph in certain cases) occurs regularly in transitory inchoativ es such as mamula ‘blush’, mamuti ‘blanch’ (cf. the nontransitory inchoativ es inv olv ing the same roots in combination with -um-, pumula ‘turn red’, pumuti ‘turn white’). The most common Theme-Trigger (ThT) affixes are i- and -in, which are the most common ThT counterparts of AT mag - and AT -um- respectiv ely. Thus ThT i- commonly occurs, inter alia, when the Actor is both Agent and Source, as in (2), while ThT -in commonly occurs, inter alia, when the Actor is both Agent and Goal, as in (13). There are also certain ThT v erbs formed with -an (which is much more common as a DirectionTrigger affix). These are often v erbs whose meaning is such that their Theme undergoes a superficial change: e. g., hug asan ‘wash’, walisan ‘sweep’. Idiosyncratically, ma- occurs as a simple affix in a few ThT v erbs that correspond to AT maka- v erbs: e. g., makita ‘see’. (AT maka- and ThT ma- are common as dual-purpose affixes that indicate, in addition to their Trigger-marking function, ability or accidental occurrence. It seems plausible that AT makakita/ThT makita ‘see’ were originally ability forms meaning ‘can see’, and

79.  Tagalog

1423

that ov er the course of time these ability the argument designating the cause of the forms supplanted the original simple forms.) ve ent, as in: The usual Direction-Trigger (DT) affix is -an. The one systematic exception is a set of two-argument DT v erbs formed with -in. The roots of these v erbs all also form three-argument DT v erbs with -an. These are the same CT v erbs corresponding to AT -um- (or ma-) roots that were mentioned abov e as forming v erbs are commonly formed with the prefix two-argument AT v erbs with -um- and threeika-: compare CT ikatakbo ‘make run’ and argument AT v erbs with mag -. The DT -in AT tumakbo ‘run’ (also CT ikabing i ‘cause to v erbs correspond to the AT -um- v erbs, while become deaf’ and AT mabing i ‘become the DT -an v erbs correspond to the AT magdeaf’). If the AT v erb is formed with mag-, v erbs (and to ThT i- v erbs). Examples include stem-forming pag - follows the CT prefix, and DT pasukin ‘enter’ v s. pasukan ‘take into’, lathe ka- of the prefix is optional: cf. CT i(ka)basin ‘go outside’ v s. labasan ‘take outside’, pag -away ‘cause to fig ht’, AT mag -away and akyatin ‘take up to’ vs. akyatan ‘climb’. ‘fight’. Beneficiary-Trigger (BT) v erbs select as Tagalog also has a set of causativ e v erbs Trigger the argument denoting the benefiformed with the causativ e-stem-forming preciary of the ve ent, as in: fix pa-, which occurs in addition to a Trigger affix, as in:

The BT affix is i-. If the corresponding AT v erb is formed with mag -, then the BT affix is followed by the stem-forming prefix pag-, as in BT ipag -abot ‘hand to for’ (cf. AT magabot ‘hand to’). Mag - itself may in fact be analyzed as including this same stem-forming prefix in combination with an AT prefix m-: cf. De Guzman (1980). (Similarly, AT mang-, as in mang isda ‘go fishing’, may be analyzed as consisting of the AT prefix m- plus a stemforming prefix pang -, which surfaces in such BT forms as ipangisda ‘go fishing for’.) Instrument-Trigger (IT) v erbs select as Trigger the argument denoting the instrument of the ve ent, as in:

IT verbs commonly involve the prefix ipang-. Ipang - may be analyzed as consisting of the Trigger-marking prefix i- and a stem-forming prefix pang - that also occurs in instrumental adjectives. (Compare with the v erb in (17) the instrumental adjectiv e pang g uhit ‘used in drawing’. The stem-forming prefix pang - that occurs in instrumental v erbs and adjectiv es is distinct from the prev iouly mentioned stemforming prefix pang -, showing a different pattern of morphophonemic alternations.) Cause-Trigger (CT) v erbs select as Trigger

With such causativ e v erbs there are, in a sense, two Actors, one causing the other to act. Howev er, morphologically and syntactically only the “Causer” is treated as an Actor, while the “Causee” is treated as a Theme or a Direction. Thus, when the Causer is the Trigger, as in (19), the AT affix mag - is used, and a non-Trigger Causer is case-marked as an Actor, as in (20):

(19) and (20) also illustrate the Causee being treated as a Theme: i. e., taking Theme case-marking in (19) and triggering the occurrence of the Theme-Trigger affix -in in (20). The Causee is treated as a Theme, howev er, only if the v erb underlying the causativ e formation is one that is not already subcategorized for a Theme. (Thus the v erb underlying the causativ es in (19) and (20), AT pumunta/DT puntahan ‘go’, is not subcategorized for a Theme.) If the v erb underlying the causativ e formation is subcategorized for a Theme, then the Causee takes Direction case-

1424

marking, and, though it still triggers the occurrence of the Trigger-marking affix -in, the affix under these circumstances may be considered Direction-Trigger. This pattern is illustrated in (21) and (22). (The underlying v erb here is AT humingi/ThT hing in/DT hinan ‘ask for’.)

(Since the ordering of the two Direction phrases in (21) is not fixed — cf. section 5 — this sentence is potentially ambiguous, and could also mean ‘The woman will hav e a/the neighbor ask a/the maid for some rice.’) As examples (20) and (22) illustrate, in causativ e v erbs the affix -in is used whenev er the causee is Trigger, whether the corresponding non-Trigger is marked as Theme or Direction. When the Theme of the underlying v erb is the Trigger of the causativ e v erb, the affix i- is consistently used, and when the Direction of the underlying v erb is the Trigger of the causativ e v erb, the affix -an is consistently used: cf. ThT ipahing i/DT pahing an ‘cause to ask for’.

XXI. Syntaktische Skizzen

noun phrases), adjecti v es (or adjecti v e phrases), and prepositional phrases hav e similar distributions in Tagalog, and all in fact hav e distributions similar to v erbs. (The distributional similarities are ev en greater than has thus far been illustrated, since nouns, adjectiv es, prepositional phrases, and v erbs all also occur in essentially identical contexts as arguments or as modifiers — cf. section 4. The distinctions between major parts of speech are thus based not on distribution but on morphology or internal structure. For example, only adjectiv es occur in certain intensiv e formations and constructions. And only v erbs take Trigger-marking affixes and are inflected for aspect. Verbs are not, howev er, inflected for person, and inflection for number, though possible with certain ActorTrigger verbs, is never obligatory.) In addition to predicate-plus-Trigger constructions, there are also certain basic clauses in which a nonv erbal predicate is not followed by a Trigger. These are existential clauses: e. g.:

The same prepositional phrases that occur in existential predicates without a following Trigger also occur as possessiv e predicates with a following Trigger: compare (26) and (27):

2.2. Clauses with Nonverbal Predicates Nouns, adjectiv es, and prepositional phrases may all occur as predicates. Tagalog has no copula, and the structure of a basic clause with a nonv erbal predicate is in most cases similar to that of a clause with an intransitiv e-v erb predicate. Thus the predicate is initial, and it is followed by a Trigger expression. Some examples are:

As these examples illustrate, nouns (or

One other type of predicate to be noted is the definite predicate, found in equational constructions such as (28):

As the example illustrates, equational constructions consist of two Trigger expressions. In such constructions it is not altogether clear which of the two Trigger expressions is the predicate (cf. the English translation of (28)). Howev er, it is the first expression that has the predicate-like property of representing newer information. Thus (28) is an appropriate answer to Sino ang titser? ‘Who is the teacher?’, not to Sino ang Amerikano? ‘Who is the American?’

79.  Tagalog

3.

Word Order

The order of constituents that follow a clause-initial predicate is in general not fixed, and, clitics excepted (see below, present section), arguments and adv erbs (such as time expressions) may generally occur in any order. In sentences with v erbal predicates, there is often a preference for placing the Actor immediately after the v erb, so that an ordering like that shown in (29) is a common one:

1425

When two clitic pronouns co-occur, monosyllabic pronouns (whether ang-form or ngform) always precede disyllabic pronouns. Two monosyllabic pronouns nev er co-occur. The only hypothetical possibility of such a co-occurrence would inv olv e the monosyllabic 2nd-person-singular ang-form pronoun ka and the monosyllabic 1st-person-singular ng-form pronoun ko, but in fact a portmanteau disyllabic clitic pronoun kita occurs instead, as in:

Two disyllabic clitic pronouns may occur in either order. But any other ordering of ni Juan, si Maria, Clitic particles always follow monosyllabic and ng ayon would result in a grammatical clitic pronouns and precede disyllabic clitic sentence cognitiv ely synonymous with (29) pronouns. The clitic particles themselv es fall (though perhaps pragmatically distinguished into a number of different classes that deterfrom it). mine their order in relation to one another. Tagalog does, howev er, hav e a set of seThis ordering is partly fixed, partly free. For cond-position clitics that hav e a fixed posidetails, see Schachter (1973). tion in relation to other clause elements, Clitics in general follow the clause-initial following a clause-initial predicate (or other word, which may be the predicate, as in exclause-initial word — see below) and obligaamples cited thus far, but may also be a negatorily preceding all other non-initial constitutiv e marker, as in (32), a question word, as in ents. When two or more of these clitics co(33), etc. (See section 5 for a discussion of occur, their order in relation to one another negati es and questions.) is also largely fixed. Thus no other ordering v of the constituents of (30), which contains the clitics ko, na, and sila, is possible:

The Tagalog clitics are the ang-formand ng-form personal pronouns (cf. Chart 1, section 2.1.1) and the rather heterogeneous set of particles shown in Chart 3 (some of which in fact hav e a wider semantic range than the glosses on the chart indicate): Chart 3 Clitic Particles ba (interrogative na ‘already’ marker) kasi ‘because’ naman ‘instead’ kaya (speculation nga ‘really’ marker) daw (reported speech pa ‘still’ marker) din ‘too’ pala (surprise marker) ho (politeness marker) po (politeness marker) lamang (only) sana (optative marker) tuloy ‘as a result’ man ‘even’ yata (uncertainty marker) muna ‘for a while’

(Some clause-initial words cannot serv e as clitic hosts, and others serv e optionally. There are also some differences between the potential hosts of clitic pronouns and those of clitic particles. For details, see Schachter and Otanes (1972), pp. 187—193, 433—435.) Although predicate-initial constructions are basic in Tagalog, there are sev eral different construction types in which an argument or adv erb precedes the predicate. In one such construction, the clause-initial constituent — which may be the Trigger, one of certain types of non-Trigger arguments, or an adv erb — is immediately followed by the inv ersion marker ay. Some examples are:

XXI. Syntaktische Skizzen

1426

(Ikaw in (33) and ka in (34) are corresponding nonclitic and clitic forms — cf. Chart 1, section 2.1.1. As (34) illustrates, constituents that precede ay do not host clitic pronouns, and neither does ay itself.) Ay constructions occur more often in writing and in formal speech than they do in ordinary conv ersation. Fox (1985) has proposed a discourse function for ay constructions in narrativ es, suggesting that the referent of the pre-ay constituent is one that had been referred to at some earlier point in the narrativ e and is being reintroduced in the ay construction itself. Other non-predicate-initial constructions with special discourse functions are the contrastiv e inv ersion construction, illustrated in (35), and the emphatic inv ersion construction, illustrated in (36). In the former, the referent of an initial Trigger or adv erb is contrasted, either explicitly or implicitly, with a structurally similar element in another sentence. In the latter, the referent of an initial adv erb or directional complement is giv en roughly the same kind of emphasis that is giv en by the cleft construction in English:

(Note that the emphatic initial constituent in (36) serves as host to the clitic pronoun.) The order of constituents of noun phrases is discussed, together with other aspects of noun-phrase structure, in section 4.

4.

Nominals and Other Arguments

This section first examines the structure of noun phrases, cov ering pluralization and modification, then goes on to constructions of other types that may function as arguments.

The most common expression of pluralization in noun phrases inv olv es the proclitic mg a [maŋa]. Explicit pluralization with mga is optional if the context makes the intended plural meaning clear (except that mg a is nev er used if the head noun is modified by a quantifier, as in (42) below). Thus when a predicate nominal and a Trigger are to be interpreted as plurals, the pluralization may be indicated on either or on both:

In addition to pluralizing common nouns and demonstrativ es as abov e, mg a may also be used to pluralize a personal name, as in:

Personal names may also be pluralized by replacing the usual personal-name markers si, ni, and kay (cf. section 2.1.1) by sina, nina, and kina respectiv ely, in which case the interpretation is ‘X and others’: e. g., sina Santos ‘Santos and the others’. Possessiv e modifiers other than personal pronouns always follow the nouns they modify. These modifiers hav e the form of ng phrases or their functional equiv alents (i. e., ni plus a personal name or a ng-form demonv strati e pronoun), as in:

Personal-pronoun possessiv e two forms: either a ng-form the head noun, as in (40), or noun before it, as in (41). If used, the linker -ng [ŋ] comes the head

modifiers hav e pronoun after a sa-form prothe sa-form is between it and noun.

79.  Tagalog

Some nonpossessiv e modifiers regularly precede the head, others regularly follow it, and still others may either precede or follow the head. In any case, a linker comes between the modifier and the head. (This linker has two phonologically conditioned alternants, -ng and na.) The modifiers that regularly precede the head are the quantifiers: e. g.,

Those that regularly follow are nouns: e. g.,

Other modifiers may in general either precede or follow the head, though in some cases there is a pragmatic difference. For example, when a demonstrativ e modifier precedes a noun, the noun is potentially contrastiv e. On the other hand, when a noun precedes a demonstrativ e, the demonstrative is potentially contrastiv e. Thus itong sombrero (this-L hat) ‘this hat’ can be used to mean ‘this hat’ (as opposed, say, to this coat) while sombrero na ito (hat L this) can be used to mean ‘this hat’ (as opposed to that one). A Tagalog nonpossessiv e modifier generally has the structure of a clause with a deleted Trigger (see below for exceptions), and the head of a modification construction is generally understood as corresponding to this Trigger. For example, the head in the examples of (44) corresponds to the Trigger in the examples of (45):

As these examples illustrate, in Tagalog the

1427

process of deriv ing modifiers from clauses — i. e., relativ ization — typically inv olv es the deletion of a Trigger, and one may propose that in general all and only Triggers may be relativ ized. The one exception is that certain constituents of Triggers may also be relativ ized. For example, the possessiv e modifier of the Trigger in (46) may be relativ ized, as illustrated in (47):

Clauses with deleted triggers function freely not only as modifiers but also as arguments, taking such typical argument roles as Trigger, Actor, etc. Some examples are:

Such argument structures hav e exactly the same form as relativ e modifiers (compare, for example, (44) and (48)), and may in fact be analyzed as headless relative clauses. There are also sentential arguments. Those corresponding to statements are introduced by the linker -ng /na, as in (50); those corresponding to questions are introduced by kung ‘if’, as in (51).

XXI. Syntaktische Skizzen

1428

Such sentential arguments hav e exactly the same internal structure as the corresponding independent sentences, except that embedded questions nev er contain the interrogativ e particle ba that is optional in their non-embedded counterparts. (For the structure of questions, see section 5.)

5.

Non-Declarative Sentences and Negation

Tagalog polar questions are distinguished from the corresponding statements by a rising intonation pattern and, optionally, by the interrogativ e clitic particle ba. This particle also occurs optionally in question-word questions, which, howev er, hav e an intonation pattern distinct from that of both polar questions and statements. The questioned constituent in a questionword question is normally clause-initial. If this constituent is a non-Trigger argument or an adv erb, any clitics in the clause attach to it, as in:

There are also hortativ es, which differ from imperativ es only in hav ing first-personinclusi v e Actors. Examples are:

Negation is commonly expressed by one of three clause-initial markers. In imperativ e and hortativ e clauses, the negativ e marker is huwag; in existential and possessiv e clauses, it is wala (which occurs in place of an affirmativ e existential/possessiv e marker); and in clauses of other types it is hindi. Huwag and wala take a following linker, which is preceded by any clitic pronouns or particles. Examples are:

‘Which corner will the bus leave from?’ If the questioned constituent is a Trigger, the rest of the clause is put into the form of a headless relativ e clause (cf. section 4) and preceded by the Trigger marker ang . Under these circumstances, the clitic particle ba still follows the questioned constituent, but any clitic Actor pronoun occurs inside the headless relativ e clause. Some examples are:

Imperativ es of the most frequent type are syntactically similar to statements with v erbal predicates and second-person Actors (which may or may not be Triggers), except that the v erb is in the infinitiv e form rather than one of the three finite forms that occur in statements. Some examples are:

6.

Subject Properties

It was suggested in section 2.1 that it is inappropriate to identify the Tagalog Trigger with the familiar grammatical category subject be-

79.  Tagalog

cause this would entail analyzing the Trigger system as a v oice system, in disregard of the differences between the two types of systems. Another reason for not marking such an identification is the fact that certain syntactic properties that are commonly associated with subjects in other languages (cf. Keenan 1976) are in fact associated not with the Trigger but with the Actor in Tagalog. For example, it is the Actor that is the addressee of an imperativ e sentence (cf. (56)—(57), abov e). And it is also the Actor that controls the reference of a reflexiv e (expressed by sarili ‘self’ and a possessi v e pronoun), as in:

(Since the v erb in (66) is AT, sila in this sentence is Actor as well as Trigger. In (67), howev er, nila is Actor but not Trigger. So the generalization is clearly that it is the Actor — whether or not it is also the Trigger — that controls reflexive reference.) On the other hand, there are certain subject-like properties that are associated with the Trigger. One such property, noted in section 4, is relativ izability: only Triggers (and certain constituents of Triggers) may be relativ ized. Another subject-like property associated with Triggers is the ability to launch socalled floating quantifiers — i. e., quantifiers that are separated from the nouns they quantify. Some rele v ant examples are:

In addition, there is at least one subject property that may be associated with either the Actor or the Trigger. This is the property of deletability under identity in so-called equi-NP constructions, as illustrated by the

1429

following (1981):

examples,

adapted

from

Dell

In (70) the complement construction contains an AT v erb, and the deletable ako is thus both Trigger and Actor. In (71), howev er, the deletable ko is Actor alone, while in (72) the deletable ako is Trigger alone. So the Actor and the Trigger in this case share the subject property in question. Schachter (1976) argues on the basis of ev idence like the abov e that Tagalog in fact does not hav e subjects, and that subject is therefore — contrary to a common assumption among grammarians — not a univ ersally attested grammatical category. (Gil (1984) presents similar arguments with regard to the category direct object, which is equally problematic in Tagalog.) In any ev ent, Tagalog in this respect poses an intriguing problem for syntactic theory.

7.

References

De Guzman, Videa P. 1980. A reanalysis of the structure of Tagalog v erbs. Philippine Journal of Linguistics 11. 21—31. Dell, François C. 1981. On certain sentential complements in Tagalog. Cahiers de linguistique Asie Orientale 10. 19—41. Fox, Barbara A. 1985. Word-order inv ersion and discourse continuity in Tagalog. Text 5. 39—54. Gil, David. 1984. On the notion of “direct object” in patient prominent languages. Objects: towards a theory of grammatical relations, ed. by Frans Plank. London. Jackendoff, Ray S. 1972. Semantic interpretation in generative grammar. Cambridge. Keenan, Edward L. 1976. Towards a univ ersal definition of “subject”. Subject and topic, ed. by Charles N. Li. New York.

XXI. Syntaktische Skizzen

1430

Ramos, Teresita V. 1974. The case system of Tagalog verbs (Pacific linguistics series B 27). Canberra. Schachter, Paul. 1973. Constraints on clitic order in Tagalog. Parangal kay Cecilio Lopez, ed. by Andrew B. Gonzales (Philippine Journal of Linguistics special monograph 4). Quezon City.

—. 1976. The subject in Philippine languages: topic, actor, actor-topic, or none of the abov e. Subject and topic, ed. by Charles N. Li. New York. —, and Fe. T. Otanes. 1972. Tagalog reference grammar. Berkeley.

Paul Schachter, Los Angeles (USA)

80. Warlpiri 1. 2.

6. 7. 8. 9. 10.

Introduction Some General Features of Warlpiri Clause Structure Constituent Structure and the Sentence The Auxiliary and Agreement Predicators, Argument Structures and Case Array Anaphora Complex Clauses Sentential Adjuncts Operators and Logical Form References

1.

Introduction

3. 4. 5.

Warlpiri is an Australian Aboriginal language spoken in Central Australia. It has receiv ed a good deal of attention recently, because of its ‘nonconfigurational’ structure. The properties of Warlpiri that make it a ‘non-configurational language’, (such as free word order, discontinuous constituents and null anaphora) hav e been examined in v arious theories. Hale (1981a) and Nash (1986) present accounts of Warlpiri as a W* language in rev ised extended standard transformational grammar, as does v an Riemsdijk (1981). Bouma (1985, 1986) presents an account of nonconfigurationality within categorial grammar. Lexicalist accounts are giv en in Andrews (1985) and Simpson (1983b, 1991). Gov ernment-Binding accounts are giv en in Hale (1983), Jelinek (1984) and Laughren (1985a, b, c, 1989). Swartz (1988, 1991) looks at some of these properties in terms of a functional approach to grammar, and discourse pragmatics. A general introduction to Warlpiri grammar is giv en in Nash (1986), and v arious aspects are discussed in papers in Swartz (1982a), as well as other references in the bibliography. Apart from grammar, other work on Warlpiri includes work on language acquisition by Edith Bav in and Tim Shopen, and on an auxiliary language, Warlpiri sign

language, by Adam Kendon. As well, Michael Kashket has prepared a parser for Warlpiri, based on the Gov ernment-Binding theory. A large body of material in Warlpiri is av ailable, including oral and written material in machine-readable files prepared by the Massachusetts Institute of Technology Lexicon Project, as well as many books written by Warlpiri people and published by the Yuendumu Bilingual Resources Dev elopment Unit, and translations of the Bible prepared with the help of Stephen Swartz. In this sketch, we outline some of the important features of Warlpiri grammar, referring wherev er possible to published works on the topics discussed.

2.

Some General Features of Warlpiri Clause Structure

2.1. Verbal and Nominal Sentences Warlpiri has both v erbal and nominal sentences. In the former, illustrated by (1) and (2) below, the predicator is a v erb. In the latter, illustrated by (3) and (4), the predicator belongs to the other major part of speech in Warlpiri, i. e. the noun:

(See Appendix for abbreviations and glosses).

80.  Warlpiri

In a finite v erbal sentence, an auxiliary (AUX) is obligatory — in (1) and (2) abov e, the base of the auxiliary is the present imperfectiv e element ka. This is construed with the nonpast endings of the v erb. In addition, person marking clitics — corresponding to the direct arguments (subject, object) of the v erb — are suffixed to the base of the auxiliary. In (1), an intransitiv e sentence, the auxiliary contains just the first person singular subject clitic (glossed ‘1sS’), corresponding to the subject argument. But in the transitiv e sentence (2), in addition to the subject clitic immediately following the base, the auxiliary contains a non-subject clitic corresponding to the second person singular object argument (glossed ‘2sNS’). Nominal sentences are stativ e and hav e no phonologically ov ert auxiliary base, though they can optionally hav e pronominal clitics, as is shown by the parentheses in (3) and (4). This essay will be concerned primarily with the syntax of verbal sentences. 2.2. Free Word Order A prominent feature of Warlpiri surface syntax is free word order. A transitiv e sentence, with ov ert subject and object, may exhibit any of the theoretically possible orderings of these arguments and the v erb:

1431

This may be thought of as a phonological constraint — the AUX must be contained in the first phonological phrase of the phonological clause, where the phonological clause is an intonational unit marked by a final pause and a distinctiv e tonal melody. (Preceding the phonological clause there may be a topicalised element, as in (18)). Althought the choice of different word order alternativ es is conditioned by stylistic and discourse factors, as yet only partially understood, it is also true to an extraordinary degree in Warlpiri that different orderings are considered to be repetitions of one another. When asked to repeat an utterance, speakers depart from the ordering of the original more often than not (cf. Hale 1981a, 1983, Laughren 1984, Swartz 1988, 1991). 2.3. The Ergative Case System The sentences of (1) through (5) serv e to illustrate the essential character of the Warlpiri ergativ e case system. The subject of a canonical intransitiv e sentence, like (1), or the nominal sentences (3) and (4), is assigned the socalled absolutiv e (phonologically unmarked) case, as is the object of a canonical transitiv e sentence, such as (2). The subject of a transitiv e sentence of this type appears in the socalled ergativ e case (glossed ERG and realized morphologically by the ending -ng ku after disyllabics and -rlu after polysyllabics and certain exceptional disyllabics). Canonical transitiv es are, howev er, not the only ones in which a dyadic predicator can appear. Thus, sentence (5) contains a dyadic v erb which differs from that of (2) in that its subject is assigned the absolutiv e case while its object is assigned the dativ e (glossed DAT and realized morphologically by the ending -ku). Sentence (4) contains a dyadic nominal predicator, with a similar case assignment; the subject is assigned the absolutiv e case, and the other argument is assigned the dative case. 2.4. Agreement with the AUXILIARY

Only the auxiliary complex is restricted in its placement. It normally appears in second position (the so-called Wackernagel position), if its initial element is not a complementizer. Otherwise, an auxiliary may appear either in second position or in initial position, as in (8), (with the exception that the negativ e complementizer kula cannot follow the v erb).

These sentences also illustrate another prominent feature of the grammar of Warlpiri tensed clauses — to wit, agreement between the auxiliary complex and the arguments of the v erb. As mentioned earlier, this is realized by means of pronominal clitics suffixed to the auxiliary base. These elements appear in the order subject-non-subject, with a partial exception to be noted later, and they embody the person and number categories of the corresponding argument. This agreement is suf-

XXI. Syntaktische Skizzen

1432

ficiently “rich” to permit free use of so-called “null anaphora” in Warlpiri tensed clauses — i. e. arguments need not be expressed in syntax as ov ert noun phrases (cf. Hale 1973a, 1983, Jelinek 1984, Swartz 1991, Simpson 1991). This also holds for nominal sentences containing pronominal clitics. Thus, beside (1—5) abov e, we also find the following:

As an aside, it should be pointed out that this sentence illustrates an additional detail of Warlpiri grammar as well, namely, the fact that a third person singular dativ e argument is ov ertly registered in the auxiliary (here by the clitic -rla). Otherwise, third singular arguments, subject or object, are not ov ertly represented by clitics in the auxiliary, and we do not represent them in our interlinear glossing. (It is v ery rare for dativ e arguments of nominal predicates to be cross-referenced by clitics — hence the lack of a clitic agreeing with nalijaku in (4)). Sentence (8) also exemplifies the fact that an auxiliary complex with a complementizer, being disyllabic, or longer, may remain in initial position. The alternativ e, with the auxiliary following the v erb, is also possible here, of course. 2.5. Complex Clauses

As these sentences show, the clitics correspond to grammatical functions, not to the case categories of arguments. Thus, subject clitics — i. e. -rna (1sS = first person singular subject) and -pala (3dS = third person dual subject) — are construed with absolutiv e subjects in (1), (3), and (5), but with an ergativ e subject in (2). And non-subject clitics (i. e. -ng ku (2sNS = second person singular nonsubject) and -jana (3pNS = third person plural non-subject) — are construed with an absolutiv e object in (2) and with a dativ e object in (5). In tensed sentences whose v erb is triadic, taking both absolutiv e and dativ e “objects”, only the dativ e exhibits true object properties (cf. Carrier 1976, Simpson/Bresnan 1983, Swartz 1982b) and is represented by a clitic in the auxiliary complex:

The absolutiv e argument is not represented by a clitic in such cases. Here again, the direct arguments of the v erb, including the abv soluti e, may be v non-o ert:

As we show in Section 7., Warlpiri complex sentences inv olv ing tensed dependent clauses are adjoined in protasis or apodosis — they are not, strictly speaking, embedded. Infinitiv al dependent clauses, howev er, can be embedded. In addition to the use of these dependent clause types in forming complex sentences, Warlpiri makes liberal use of the “secondary predication” of nominal expressions. Secondary predicates may attribute a state to some referent, or describe the circumstances of an ev ent, such as reason, or the resulting state. Secondary predication is important in Warlpiri, shouldering a large portion of the expressive burden in the language. 2.6. Syntactic Categories The Warlpiri categorial (or part-of-speech) system recognizes two large classes, nouns and v erbs. The v erbal category is circumscribed in its notional content, while the nominal category appears, from the v iew point of Indo-European languages, say, to express a rather wide range of notions, including some which are typically expressed by v erbs in Indo-European languages. Apart from v erbs of emotion (e. g. yulkami ‘like, lov e’) and perception (e. g. nya-nyi ‘see’), Warlpiri v erbs typically denote ev ent types in which an entity undergoes, or is caused to undergo, a change (in location,

80.  Warlpiri

stance, or condition, e. g. yuka-mi ‘enter’, yirrpi-rni ‘insert’, pali-mi ‘die, expire’, etc.), assumes an attitude or stance (in some location or state, e. g. nyina-mi ‘sit, be in location or state’), or produces some effect (in itself or in some other entity, e. g. yula-mi ‘cry’, pi-nyi ‘affect (harmfully)’). (Here and elsewhere, when v erbs are cited in isolation, they are giv en in their nonpast tense forms. This serv es not only to prov ide pronounceable v erbal words but also to identify the conjugation to which a cited v erb belongs). (Cf. Hale 1982b, 1983, Hale/Platero 1986b, Hale/ Laughren 1986, Nash 1986, Swartz 1982b). The notional range of Warlpiri nouns extends from definite and fully referential expressions to ones which, although morphologically nominal, are almost exclusiv ely predicativ e in use. Thus, the nominal category encompasses deictics (e. g. the pronouns and demonstrativ es, such as ng aju ‘I’, nyampu ‘this’), the indefinite determiners (and “quantifiers”, e. g. jinta ‘one’), names (e. g. the subsection terms), substantiv es (e. g. karnta ‘woman’, pirli ‘stone’), attributiv es (e. g. wiri ‘big’), mental and psychological stativ es (e. g. ng ampurrpa ‘wanting, desirous’), and locativ es and directionals (e. g. kulkurru ‘in the middle’). Items toward the end of this listing are more often than not predicativ e in function, while those toward the beginning are more often argumentai (cf. Hale 1983). While nouns and v erbs are the two major lexical categories in Warlpiri, heading the major syntactic phrasal categories (noun phrase and clause), there is a third lexical category which plays a significant role in Warlpiri morpho-syntax — namely, the prev erb. While Warlpiri probably has only some 120 monomorphemic v erbs, it boasts an impressiv e inv entory of prev erbs, and the category accounts for the bulk of the v erbal v ocabulary in the language (cf. Hale 1982b, Nash 1982, 1986, Swartz 1982b). The Warlpiri prev erb is ev idently nominal in origin, and it is often ambiguous in its classification — as noun or prev erb — in the synchronic grammar. In its basic position, it precedes the v erbal stem, forming with it a metrical unit. Howev er, many are rather loosely associated with their v erbal hosts, in that an auxiliary, in seeking Wackernagel’s position, may interv ene between the prev erb and the v erb stem — particularly, but not exclusiv ely, when the auxiliary has the null base. Thus, with the prev erb pina ‘back, returning’, both (9a) and (9b) are possible:

1433

In (9a) and (9b), the string (of prev erb, auxiliary, and v erb) forms a phonological phrase, with a single main stress — on the initial syllable. Warlpiri words, complex or simple, carry primary stress on the first syllable. The string in (9a) (of PV, V, AUX) forms a single phonological word which also constitutes a phonological phrase, whereas in (9b) the string (of PV AUX V) consists of two phonological words which form a single phonological phrase. The nature of the morphological binding between the prev erb and the v erb is such that, with the most “productiv e” prev erbs, at least, the relativ e order of v erb and prev erb may be rev ersed — maintaining the metrical unity of the combination, normally:

As hinted in the preceding paragraph, prev erbs are not all equally “productiv e”. They range from those (like pina ‘back, returning’) which are fully productiv e, forming predicates whose semantics is entirely compositional, to those (like wurru-, the first two syllables of the v erb wurru-ka-nyi ‘stalk’, see (5) abov e) which, though clearly prev erbs morphologically, are unique in their occurrence (according to our records, at least) and utterly obscure in their semantics.

3.

Constituent Structure and the Sentence

3.1. Discontinuous Constituents and AUX Placement The second-position placement of the auxiliary in Warlpiri offers certain clues to surface constituent structure in the language. If we can assume, as is usually done, that what precedes a second-position auxiliary (in the normal unmarked execution of a sentence) comprises a single constituent, then an argument expression consisting of a noun and modifier (e. g. determiner, genitiv e, attributiv e) forms a single constituent, though it may consist of more than one word:

XXI. Syntaktische Skizzen

1434

Complex locativ e expressions, consisting typically of a nominal in the locativ e case in apposition with an inherently locativ e nominal, may also appear in pre-auxiliary position and therefore, presumably, form syntactic constituents when they so occur, (cf. Hale 1981a, 1982a, Laughren 1989, Nash 1986).

3.2. Flat Structure The ev idence afforded by the position of the auxiliary here coincides with another type of ev idence. The case inflections in (11), being marked once only, at the right-hand margins of the nominal expressions, indicate that these are single constituents. Where a nominal expression is discontinuous — a possibility in Warlpiri — each of the separate subconstituents must be marked for case:

The same ev idence applies in the case of infinitiv al expressions, as exemplified by the first two words in (13). The infinitiv al v erb and its object, inflected by a complementizer (appearing just once, at the right margin), may precede the auxiliary:

Both the position of the infinitiv al expression and its inflection by a single complementizer indicate that it is a single constituent in the surface syntax of (13). Like nominal expressions, infinitiv als too may be discontinuous, in which case the complementizer will appear not only on the v erb but on its object as well:

If it is true that what precedes the auxiliary in the abov e examples is a syntactic constituent, and if it is true that a sequence which cannot precede the auxiliary is not a constituent, but rather more than one, then it is ev ident that there is no surface constituent corresponding to the v erb phrase, as normally understood. The v erb may precede the auxiliary, — ev en a complex v erb including a prev erb may do so, as in (9a) and (6e). But it may not be accompanied there by any of its complements. Thus, (16) is ungrammatical (unless the first word is topicalised):

Here the v erb and its object, in the order object-v erb, jointly precede the auxiliary. The order v erb-object is equally ungrammatical:

The ill-formedness is presumably due to the fact that the v erb and its object do not form a single subconstituent within the clause. There is an apparent exception to the principle that the auxiliary is in second position in surface structure. This is the common Warlpiri topicalization construction in which a constituent, a topic, is displaced to the left of the clause to which it relates, the comment:

80.  Warlpiri

Sentence (16) would be grammatical with wawirri as a topic, but it is unlikely that this could save (17). The dislocated phrase, or topic, does not count in determining the surface positioning of the auxiliary. Typically, the topic is marked by a characteristic falling-rising intonation on the final two syllables, and it is normally, but not inev itably, separated from the comment by a pause (cf. Laughren 1984, Swartz 1991). Sentences of the type represented by (18) are not true exceptions to the canonical positioning of the auxiliary, since the relev ant domain to which the second-position principle applies is just the comment portion of the topicalization construction. The ev idence just adduced for the constituent structure of Warlpiri sentences in their surface form indicates that this structure is “flat”, in the sense that the v erb does not form a separate constituent with any of its arguments. This is consistent not only with the facts concerning the placement of the auxiliary but also with the free word order, exemplified in (5). This freedom is quite ev idently not at all constrained by any hierarchical organization uniting the v erb with particular ones of its arguments. This is not to say, howev er, that there is no phrasal structure at all within the clause, since nominal, locativ e, and infinitiv al expressions do form constituents. The free constituent order observ ed within finite clauses is also observ ed in nominal and locativ e expressions; their component words may be freely ordered, although there are preferred orders (cf. Swartz 1988, 1991). Howev er, in nominal expressions a case inflection applying to the whole must appear on the final word. The internal word order of infinitiv als is more rigid — the object must directly precede the v erb (unless the object is itself marked with the complementizer). This probably follows from the fact that the complementizer must appear on the infinitiv al v erb and must be final in the phrase.

4.

The Auxiliary and Agreement

The AUX has a flat template-like structure, comprising morphemes which express mood, aspect and tense (which we call “v erbal morphemes”), and morphemes which express agreement (which we call “argumental morphemes”). The full AUX consists of an auxiliary base, comprising an obligatory ASPect element optionally preceded by a sentential

1435

COMPlementizer element; the auxiliary base is followed by obligatory pronominal AGReement clitics. AUX is obligatory in finite v erbal clauses (although it may be phonologically null if there is no COMP and ASP is PERFectiv e). AUX is restricted in its occurrence in nominal-headed clauses: ASP is neutralised, and only the COMP elements kula (NEGCOMP) and kuja (FACTCOMP) can occur, while the AGR pronominal clitics are optional. AUX cannot appear in nonfinite clauses. 4.1. Verbal AUX Morphemes The sentential categories of tense, mood and aspect are realized discontinuously in the surface string of a Warlpiri sentence. The meanings associated with the v erbal inflectional suffixes correspond to those traditionally included in the tense and mood categories. For any giv en clause, the choice of v erbal AUX morphemes and the choice of tense/mood suffix on the v erb are interdependent. Thus, in a Warlpiri finite clause, tense-mood-aspectual information is encoded through the interaction of co-occurring members of each category. There is a perfectiv e/imperfectiv e contrast expressed in Warlpiri by means of the T/ASP AUX morpheme. The perfectiv e aspect morpheme is phonologically null, while the imperfectiv e aspectual morpheme is realized as -lpa in conjunction with the past and irrealis v erb forms, and as ka with the nonpast v erb forms. The perfectiv e-imperfectiv e contrast with -lpa is shown in (19).

The categories of sentential complementizers, which we refer to as COMP, are listed in the Appendix. Some of these will be discussed further in Section 8. There are complex compatibility constraints on the coexistence of COMP and T/ASP morphemes and v erbal tense/mood suffixes, (cf. Hale 1973a, Laughren 1982, Nash 1986). 4.2. Agreement: Person-Number Clitics The third category of morphemes which constitute the AUX are the agreement (AGR) pronominal clitics. These contain morphemes

XXI. Syntaktische Skizzen

1436

belonging to two subcategories: PERSON and NUMBER. These categories are indirectly related to the main predicator (v erbal or nominal) of a finite clause, in the sense that they manifest features which identify the direct arguments of the predicator, in addition to certain more peripheral arguments. The person-number auxiliary clitics and their interaction with the transitiv ity features of the clause are of considerable complexity (cf. Hale 1973a, Laughren 1977, Nash 1986, Hale/Laughren 1986, Simpson/Withgott 1986, Simpson 1991, Swartz 1982b, 1991). There is one series of subject clitics, another of non-subject clitics. Warlpiri personnumber clitics can be classified according to whether, in their surface form, they are composed of a portmanteau person-number morpheme, or whether they consist of two distinct clitics — one being the person morpheme, the other, the number morpheme. The clitics belonging to the latter set may appear discontinuously in the AUX template. The presence of a subject clitic is obligatory in the AUX except under some v ery specific conditions, which we will leav e aside here. The presence or absence of a non-subject clitic depends on a number of complex factors (cf. Hale 1973a, Laughren 1977). Which argument agrees with a subject clitic, and which arguments agree with non-subject clitics will be discussed in Section 5. Normally, both subject and non-subject clitics agree with the arguments of the v erb in both person and number. Howev er, there are certain exceptions. These include uses of plural for singular referents and of third person for second person referents, in “special language” (auxiliary languages used in addressing or referring to certain sets of kin relations). Another exception is found in a comitativ e construction in which nominals with dual or plural marking appear with subject clitics of all persons: (20) Jakamarra-jarra-rlujarra ya-nu. Jakamarra-DUAL-1deS go-PAST ‘I went with Jakamarra.’ A third exception, observ ed by Stephen Swartz in Lajamanu Warlpiri, is the appearance of second person clitics in sentences with presentational v erbs and non-second person subjects. He says that, typically, this combination is used in narrativ es to announce a surprise v de elopment.

The number portion of the clitic agrees with the actual subject of the sentence:

5.

Predicators, Argument Structures, and Case Arrays

Predicates in Warlpiri may be headed by either of the two morphologically distinct categories, v erbs or nominals (cf. section 2.1.). In either case this head, or predicator, denotes an action, process, or state inv olv ing one or more participants, commonly referred to as its arguments. The lexical entry of a giv en predicator defines its argument structure which, in turn, determines the initial syntactic structure of core arguments which it projects. While the surface syntactic structure of a Warlpiri sentence is ev idently “flat”, in the sense that there is no ev idence for a v erb phrase at that lev el of syntactic representation (cf. 3.2.), it is nonetheless clear that the syntactic organization of a predicator’s arguments, as defined by its lexical argument structure, exhibits an asymmetry distinguishing its subject from its complements, if any. This asymmetry is rev ealed in the Warlpiri systems of anaphora, or binding. We first look at v erbs with ergativ e subjects, and then at verbs with absolutive arguments. 5.1. Verbs with Ergative Subjects Consider, for example, the v erb of (23) below:

This v erb takes two core arguments, one assigned ergativ e case (ERG), the other the phonologically null absolutiv e case (ABS). The argument with ergativ e case agrees with the subject clitic -pala, while the argument with absolutiv e case agrees with the non-subject clitic -jana. Where one of these arguments anaphorically binds the other, the binder is the erga-

80.  Warlpiri

1437

tiv e argument, and the anaphor is the absolutiv e argument (represented only by the anaphoric clitic -nyanu ‘reflexive-reciprocal’ (REFL), occupying non-subject position within the auxiliary, as in:

And, correspondingly, the v erb in (26) exemplifies the ERG-DAT array. The ergativ e argument agrees with the subject clitic, and the dativ e argument agrees with the non-subject clitic. This direction of binding, holding for all v erbs which exhibit the ERG-ABS case array, is nev er rev ersed. Hence an ergativ e argument can only be anaphorically bound “from outside”; that is to say, an ergativ e argument of a nonfinite clause can only be bound by an argument in a matrix clause in a “control structure”, as in (13). In that sentence, the ergativ e argument, the eater, of the non-finite v erb ng a-rni-nja ‘eat’ is bound — i. e. “controlled” — by the object wawirri of the matrix v erb nya-nyi. Thus, for ERG-ABS v erbs, the ergativ e argument is the “prominent” argument. It is the ergativ e which may bind another argument clause-internally in reflexiv ereciprocal constructions; and it is the ergativ e which may be bound from outside, by a matrix argument. Neither of these properties holds of the absolutiv e in these v erbs. Since these are both properties indicating subjecthood, the ergativ e, therefore, is the subject. The facts of anaphoric binding and control, of course, are in accord with the facts of agreement. It is the ergativ e argument that agrees with the subject clitics, as can be seen in (24) and (23), as well as in other illustrativ e sentences used here. Not only for ERG-ABS v erbs, but for all v erbs whose case arrays include an ergativ e, the ergativ e is the subject. (25) exemplifies the prototypical ERG-ABS-DAT v erb yi-nyi ‘giv e’. The ergativ e argument is represented by the ergativ e-marked noun karnta-jarra-rlu and agrees with the subject clitic -pala. The absolutiv e argument is represented by the noun miyi (not cross-referenced by a clitic), while the dativ e argument is represented by the non-subject REFL clitic -nyanu. The agreement structure indicates that the ergativ e argument is the subject, as does the binding relation, according to which the ergativ e subject binds the dati v e object:

Although it is not a prominent feature of Warlpiri, there exists a class of v erbs whose case array consists solely of an ERG argument. Typically, these are morphologically complex v erbs containing prev erbs of clearly nominal origin. The single argument of such v erbs, not surprisingly, exhibits the properties of a subject, as illustrated by the agreement pattern of (27) and by the binding (control) relation of (28):

In (28), of course, the subject of the nonfinite v erb ng ung kurru-pang i-rninja may not appear — it is bound, or controlled, and so must be non-ov ert. It is without question an ERG argument, howev er, as is clear from the case marking which appears on the ov ert subject kurdu-jarra-rlu in the corresponding finite clause (27).

1438

5.2. Verbs with Absolutive Subjects Not all v erbs hav e an ergativ e argument in their case array, as is ev ident from examples in Section 2. Where there is no ergativ e, the absolutiv e assumes the subject function. This fact is illustrated in (1) abov e, where the v erb wang ka-mi ‘speak’ appears in its monadic use, representing the simple ABS array. The ABS-DAT array is exemplified by (5) abov e, and also in (29), where wang ka-mi appears in a dyadic use. In both (5) and (29), the absolute argument agrees with the subject clitic and the dativ e argument agrees with the nonsubject clitic:

The dativ e argument may be bound by the absolutiv e subject in a reflexiv e/reciprocal construction:

Similarly, the absolutiv e subject of a nonfinite v erb may be controlled by an argument of the matrix v erb, as illustrated in (31) in which the dativ e argument of the higher v erb controls the absolutiv e subject of nyina-njakurra.

XXI. Syntaktische Skizzen

cal object in such structures can be shown, in part. Thus, in (7) abov e, the dativ e argument, not the absolutiv e, is cross-referenced by non-subject agreement morphology in the auxiliary. The objecthood of the dativ e argument is shown also by control of infinitiv als in -kurra ‘OBJCOMP’ clauses (cf. Carrier 1976, Hale 1982a, 1983, Simpson/Bresnan 1983). The subjects of these infinitiv als are controlled by a matrix object. In (31) and (32) it is the dativ e argument which controls the v infiniti al subject.

5.3. Alternations in Semantic Role and Case Some v erb classes show an alternation between a dativ e object and an absolutiv e object inv olv ing a change of meaning. These include perception v erbs, and v erbs of impact and concussion (cf. Guerssel/Hale/Laughren et al. 1985, Hale 1982b, Hale/Laughren 1986, Laughren 1988b, Swartz 1982b, Simpson 1991). That both the dativ e and the absolutiv e arguments act as objects is shown by the fact that they control the infinitiv al subject of an OBJCOMP clause:

The sentences so far exhaustiv ely exemplify the assignments of grammatical case categories to the subject and object functions. The subject is assigned the ergativ e case, if there is one (i. e. if the v erb has an ergativ e in its case array); otherwise, the subject is assigned the absolutiv e. The object is assigned the dativ e, if there is one, otherwise the absoA special case of the alternation with v erbs lutiv e. In a triadic case array, as in (25), of impact and concussion is the “conativ e” therefore, the absolutiv e is assigned to an aror “attempted action” alternation, in which gument which bears neither the subject nor a second dativ e clitic is used in the AUX in the object function (cf. Swartz 1982b). addition to the clitic that cross-references the That the dativ e argument is the grammativ dati e object:

80.  Warlpiri

Again, both the dativ e and the absolutiv e argument can control the infinitiv al subject of OBJCOMP clauses. 5.4. Adjunct Datives

1439

non-subject clitic in the AUX. This is a dativ e case use of the suffix -ku — to denote a participant which, in some unspecified way, is affected by the action or ev ent denoted by the v erb, or prov ides the cause or purpose of that action or ev ent. The dativ e is not a direct argument of the v erb, but rather an adjoined argument, the so-called ‘adjunct dativ e’. Agreement with a non-subject clitic in the AUX may be interpreted as foregrounding the participant concerned. The adjunct dativ e may be expressed as a reflexiv e bound by the subject, as in (36) in which the dativ e benefactiv e ‘themselv es’ is expressed by a reflexiv e pronominal clitic -nyanu bound to the subject wati-patu-rlu and the subject pronominal clitic -lu:

In preceding sections, we hav e seen the suffix -ku used as a dativ e marking a core argument of the v erb. We turn now to other uses of this suffix in which its status as a dativ e object argument marker is more questionable. (35) illustrates some such uses: Intransitiv e v erbs (and sometimes ev en transitiv e v erbs) may go a step further, and permit the dativ e participant to act as the object. A v erb such as ng arlarrimi ‘laugh’ is essentially an intransitiv e v erb. Howev er, in (37) it appears with a dativ e participant, which agrees with the dativ e non-subject clitic in the AUX. That this dativ e is an object is shown by the fact that it can control an OBJCOMP -kurra clause:

In neither (35a) nor (35b) does the adjunct agree with a non-subject clitic in the AUX. In (35a), the nominal marked with this suffix indicates the purpose of the action, and is glossed PURP, while (35b) shows the -ku suffix indicating the frequency of an action. (This use of ku, glossed as FREQ (Frequentativ e), differs from the dativ e, and from the bare purposiv e uses, in that it may be followed by further case-marking). In (35c) the suffix -ku indicates for whose benefit the action is performed. The person and number of this participant agree with the

Agreement with the AUX allows the argument marked with -ku to be foregrounded, and raised to the status of object. Howev er the dativ e participant may act as the controller of a -rlarni OBVCOMP or “pure obv iativ e complementizer” clause, as in (38), ev en though such clauses cannot be controlled by objects (absolutiv e or dativ e) in the matrix clause.

XXI. Syntaktische Skizzen

1440

as itself. The anaphor can nev er be bound by an element bearing a non-subject grammatical function.

Control of the OBVCOMP is independent of whether or not the participant marked with -ku is foregrounded and agrees with the AUX, as the optionality of the third person singular dativ e non-subject clitic -rla in (38) shows. Thus there are three main uses of the suffix -ku: as adjuncts with no AUX agreement (the FREQ or PURP uses), as adjunct dativ es with AUX agreement and control of OBVCOMP clauses, and as objects with AUX agreement and control of OBJCOMP clauses. The relation of the dativ e participant to the action or ev ent may be made more specific by combining the v erb with one of a set of prev erbs that add dativ e participants: jirrn g anja, yirrkirnpa ‘with (dependent)’, jurnta ‘away from, remov al from’, kaji, ng ayi ‘for, on behalf of’, marlaja, marlang ka ‘because of, associated with’, piki(piki) ‘in danger of’. The meanings giv en to the prev erbs further specify the relation between the dativ e participant and the action or ev ent denoted by the v erb. Thus, in (39) the action is “cutting the boomerang”, the dativ e participant is “the little child”, and the prev erb kajiindicates that the relation between the child and the boomerang-cutting is benefactiv e:

6.1. Disjoint Pronominal Reference and Pronominal Coreference While a free pronoun occurs in the same syntactic position as a common or proper noun as the argument of a v erb, bearing the same set of grammatical functions and case-marking, its distribution is restricted by relations of possible or impossible coreference between a pronoun and another argument in the same clause. A free pronoun such as the third person pronoun nyanung u cannot be coreferent with the reflexiv e anaphor -nyanu and hence with the subject of a sentence in which -nyanu expresses the object of a v erb that takes an ERG subject. This constraint is demonstrated in (40)—(42):

See Craig/Hale (1988), Hale (1982b), Nash (1982, 1986b), Swartz (1982b), Simpson (1991).

6.

Anaphora

In Warlpiri, the pronominal features of person and number are expressed by two distinct syntactic categories: bound pronominal clitics realized in AUX, and optional free pronouns. The reciprocal-reflexiv e non-subject clitic -nyanu is always coreferent with the subject of the finite v erb in the same clause

The constraint which prev ents the pronoun from bearing the absolutiv e or dativ e object function in (38)—(40) does not apply when it bears a dativ e adjunct function and is construed with the anaphor -nyanu. This is shown in (41):

80.  Warlpiri

A dativ e pronoun construed with the anaphor -nyanu bearing the object grammatical function may be realized in a sentence containing a v erb which takes an absolutiv e subject and a dativ e object, as shown in (42):

1441

7.

Complex Clauses

7.1. Sentential Complements In Warlpiri, v ery few v erbs select ev ents as arguments, and thus few allow sentential complements. The main semantic classes appear to be v erbs of ordering, telling, enlisting, and v erbs meaning “fail”, or “cause to fail”, as in (44). Selected sentential complements must be non-finite, and may generally be replaced by a nominal bearing the same complementizer, as in (c).

See Farmer/Hale/Tsujimura (1986), Laughren (1985b, 1988a), Simpson (1991). No non-pronominal nominal expression may be construed directly with the anaphor nyanu, which would imply binding of that nominal by the v erb’s subject. Howev er, indirect construal is possible — a nominal which is predicated of the argument or adjunct expressed by the anaphor -nyanu may be ov ertly expressed as shown in (43). In (a) the ABS nominal jurru ‘head’ is predicated of the object expressed by -nyanu (see Hale 1981b, Laughren 1992, for a detailed study of partwhole syntax in Warlpiri). In (b) the ABS nominal murrumurru ‘pain’ is predicated of the argument associated with the anaphor -nyanu. In (c) the ABS nominal wati ‘man’ is predicated of the non-subject argument associated with -nyanu.

What the subject of the non-finite clause will be is predictable — if the matrix v erb has one argument, as in (a), that argument will control the subject of the non-finite clause. But if the matrix v erb has two arguments, as in (b), the object argument will control the subject of the non-finite clause (cf. Hale 1982b, Nash 1986, Simpson 1983). 7.2. Infinitive and Nominal Secondary Predicates Secondary predicates, as opposed to the primary v erbal or nominal predicate of a finite clause, hav e no independent tense, mood, as-

1442

pect or ov ertly realized pronominal clitics associated with them. A secondary predicate consists of a nominal or infinitiv e v erb to which a complementizer case and/or nominal case is suffixed. Non-subject arguments of the nominal or infinitiv e, as well as v arious adjuncts and modifiers, may also be expressed. There are basically two classes of secondary predicate: ev entiv e and stativ e. Ev entiv e predicates are headed by a complementizer case which indicates the temporal relation of the ev ent denoted by the secondary predicate to the ev ent or process denoted by the primary (or finite) predicate. The ev entiv e interpretation is attributable to the complementizer case, since a referential nominal bearing one of these cases will be interpreted as referring to some ev ent inv olv ing the referent of the subject of the secondary predicate and the referent of the nominal expression. In (45), stativ e and ev entiv e secondary predicates are compared. In (a) the ABS case-marked nominal nyurnu ‘sick’ is a stativ e predicate. Its subject is karnta ‘woman’ which is the ABS case-marked object of the v erb. In (b) nyurnu is interpreted as part of an ev entiv e predicate headed by the object complementizer case -kurra, meaning something like ‘inv olv ed with the sick one’. The subject of this ev entiv e secondary predicate is coreferent with karnta, the ABS case-marked object of the finite v erb.

See Hale (1982b), Nash (1986), Simpson (1983), Laughren (1992). 7.3. Eventive Predicates Ev entiv e predicates may be further classified according to their time-reference; whether they denote an ev ent taking place at the same time as the ev ent denoted by the main clause, or preceding it, or subsequent to it. 7.3.1 Simultaneous Event Warlpiri has three complementizer cases which indicate that the ev ent denoted by the

XXI. Syntaktische Skizzen

secondary predicate occurs at the same time as the ev ent denoted by the main v erb. These are illustrated below. In addition to their temporal content, each of these complementizer cases specifies how the understood subject of the secondary predicate is to be interpreted.

In (a) the secondary predicate consists of the infinitiv e form of the v erb ng una-mi plus the complementizer case -kurra which indicates that the understood subject of the infinitiv e v erb is obligatorily coreferent with the object marlu of the finite v erb nya-ng u. In (b) the secondary predicate is made up of the infinitiv e form of the v erb parnka-mi plus the complementizer case -karra which specifies that the understood subject of the infinitiv e is coreferent with the subject, wati-ngki, of the finite v erb. This relationship is further marked by the presence of the ergativ e caseending -rlu on the secondary predicate. In (c), the secondary predicate jarda-nguna-njarlarni contains the complementizer case -rlarni which indicates that the understood subject of the compound v erb jarda-ng una-nja is coreferent with the dati v e case-marked karnta-ku. Karnta-ku is neither the subject nor the object of the finite v erb, but rather a dative adjunct (see Section 5.4.). Whereas the subject of a nominal or infinitiv al predicate headed by either OBJCOMP

80.  Warlpiri

or SUBJCOMP cannot be ov ertly expressed and is obligatorily interpreted as being coreferent with, or controlled by, the object or subject of the finite v erb, the subject of a predicate headed by the OBVCOMP -rlarni may be ov ertly expressed by a dativ e noun phrase as in (d). In this case we may consider the combination of dativ e noun phrase plus -rlarni headed predicate and its non-subject arguments to be adjoined to the clause headed by the main finite verb. 7.3.2. Preceding Events

(Other translations are possible). In (47) the secondary predicate consists of the infinitiv e form of the v erb purra-mi, its object kuyu and the complementizer case -rla, homophonous with the LOC[ativ e] case suffix. This suffix indicates that the ev ent of cooking precedes the ev ent of eating denoted by the finite forms of the v erb nga-rni ‘eat’ — in the past tense in (a) and the immediate future tense in (b). The understood subject of the PRECCOMP is typically coreferent with the subject of the finite verb.

1443

(The SEQCOMP, as in (c) is homophonous with the OBJCOMP, but differs from it in time-reference, and in what may be the controller of the subject of the non-finite clause). Each of the complementizer cases in (48) signals that the ev ent denoted by the infinitiv e v erb or nominal to which it is suffixed follows, or is dependent on, the ev ent denoted by the finite verb. Another complementizer case which refers to a subsequent ev ent is the ev itativ e or negav ti e purposi v e (NEGPURP) complementizer -kujaku illustrated in (49). This complementizer sometimes allows the object of the nonfinite clause to be construed with an argument of the main clause. There is no obligatory control of the subject of the nonfinite clause.

This NEGPURP complementizer may also occasionally be used on finite v erb forms, as in (50):

7.3.3. Subsequent Events

7.3.4. Objects of Infinitive Verbs Whereas a finite v erb combines with AUX in such a way that certain of its arguments are expressed by means of person-number clitics with which a case-marked nominal expres-

1444

sion may be construed, an infinitiv e v erb does not combine with an independent AUX, falling as it does under the scope of the tense, mood and aspect morphemes associated with AUX and the related finite v erb in the clause. The arguments of the infinitiv al v erb may be realised as nominal expressions, apart from the subject argument of an infinitiv e to which is suffixed one of the complementizer cases that requires obligatory control of the subject, as in (46). The dativ e case argument of a v erb, finite or nonfinite, may always be expressed by a dativ e case-marked nominal expression. The object argument of an infinitiv e which would be expressed by an absolutiv e case-marked nominal expression of the corresponding finite v erb, cannot be thus expressed, since absolutiv e case fails to be assigned to the object of an infinitiv e v erb. Rather it falls within the scope of the complementizer case which marks the infinitiv e. Howev er, when the object nominal immediately precedes the infinitiv e, ov ert complementizer case-marking on the nominal is not obligatory (giv ing the appearance of an absolutiv e object), and the nominal and infinitiv e v erb form a single phonological phrase, as in (48b). When the object nominal occupies another position in the clause, it too is ov ertly marked by the complementizer case ending (cf. Hale 1982b, Laughren 1989). Compare (51) with the sentences in (48). Not only does the object nominal fall within the scope of the complementizer case, but so does a modifying nominal which has semantic scope ov er the infiniv ti e and its arguments.

Arguments of an ev entiv e secondary predicate need not be ov ertly expressed in Warlpiri. Occasionally, as shown in (52), none of the arguments of the infinitiv e v erb yi-nja ‘give-INF’ are overtly expressed:

XXI. Syntaktische Skizzen

(52) Wangka-ja-rna-rla say-PAST-1sS-3sDAT yi-nja-ku. give-INF-PURPCOMP ‘I said to him to give (someone something).’ 7.3.5. Stative Secondary Predicates A stativ e secondary predicate typically attributes some quality to its “subject” (which may be a nominal expression or a pronominal clitic or both). The secondary predicate consists of a nominal or infinitiv al expression which agrees in case-marking with its “subject”.

In (a) the ABS nominal predicate mata is predicated of the absolutiv e case-marked karnta, the subject of the sentence. The infinitiv e expression wapa-nja-warnu is also predicated of the woman. The ASSOC[iativ e] case transforms an ev entiv e predicate into a stativ e one, i. e. the state which results from inv olv ement in the activ ity, process or ev ent denoted by the v erb. The PRIV[ativ e] suffix -wang u can also transform an ev entiv e predicate into a stativ e one as in (b). (This is an unusual example, showing ergativ e case on warlu-ng ku in the non-finite clause, agreeing with the understood ergativ e subject of purranja-wangu). In (c) we hav e two stativ e secondary predicates, purra-nja-warnu is predicated of the absolutiv e object kuyu while yarnunjuku-rlu is predicated of the ergativ e subject kurdu-ngku. The relationship between “subject” and secondary predicate is formally indicated in each case by the case-marking on the predicate. This case-marking is identical to that on the nominal understood to be the “subject” of this secondary predicate.

80.  Warlpiri

7.3.6 Secondary Predicates and Semantic Cases Warlpiri has a number of semantic cases which fall into two classes: deriv ational and non-deriv ational (see Appendix and also Hale 1982b, Nash 1986, Simpson 1983). These cases are suffixed to nominal expressions. The resulting compound expression is predicated of some element in the sentence, which acts as the “subject” of the expression, as in (54). A semantic case can be suffixed to a nominal plus deriv ational case compound. A grammatical case, dativ e or ergativ e, can be further suffixed to a semantic case-headed predicate, thus indicating the argument of which it is predicated.

1445

-kurra ‘RESULT[ativ e]’. They refer to a state which is achiev ed or arriv ed at, that is different from the implicit or explicit original state. While the predicate formed with -karda is freely used and may take any argument or adjunct in the same clause as its “subject”, the predicate formed with -kurra is v ery restricted in its application. It may only be predicated of the absolutiv e object of an impact v erb. Only a few predicativ e nominals hav e been found with the resultativ e suffix, namely nyurnu ‘sick, dead’, yalyu ‘blood’ and g tarnn a ‘long time, v fore er’.

8.

Sentential Adjuncts

Finite tensed clauses apparently nev er appear as complements in Warlpiri. Howev er, they are used in a v ery common construction, the adjoined relativ e clause, illustrated in (56):

As these examples show, semantic-caseheaded predicates too may receiv e case in agreement with the nominal of which they are predicated. See Laughren (1992). 7.3.7 Inchoative There are two inchoativ e suffixes, illustrated in (55), which are suffixed to predicativ e nominals to form a resultativ e secondary predicate. These are -karda ‘TRANS[lativ e]’ and

Morphologically, adjoined relativ e clauses are marked by the presence in the AUX of one of a set of complementizer (COMP) morphemes, such as kuja abov e. The others include the factiv e ng ula (FACTCOMP), the nonfactiv e kaji (NFACTCOMP), and the relational yung u and its allomorphs (RELCOMP). Syntactically, these clauses are dependent on the main clause, and peripheral to it. They are nev er embedded in the main clause, and

XXI. Syntaktische Skizzen

1446

they are usually separated from it by a pause. They may be preposed to the main clause, which may start with the anaphorical element ngula, as in (57). There may ev en be multiple subordinations, as in (58).

will disambiguate them. Howev er, the choice of complementizer, the time references of both the main clause and the adjoined clause, and the presence of coreferent noun phrases in both the main and the adjoined clause are factors in determining what reading is giv en. An example of a T-relativ e reading is giv en in (59): Factiv e, no shared argument, same time reference, v T-relati e

There appears to be no difference in behav iour between coreferential noun phrases in clauses with NP-relativ e interpretations, and those in clauses with T-relativ e interpretations. Both are represented by pronominal clitics, if they bear the right grammatical functions. In terms of pronominalisation, usually the second of the two (whether this happens to be in the main clause or the adjoined clause) undergoes pronominalisation, but pronominalisation is not essential. Semantically, as the translations show, adjoined relativ e clauses are open to different interpretations, which hav e been the subject of some inv estigation (cf. Hale 1976; Larson 1983). Roughly speaking, there are two main classes of interpretation. The first is the “NPRelativ e” interpretation, in which the adjoined clause modified an argument of the main clause, and is often translated as a relative clause in English. The second is the “T-Relativ e” interpretation, a term deriv ed from the use of adjoined relativ e clauses to specify the time of the main clause, or to describe an ev ent holding at the same time as the main clause. Howev er, it has been extended to cov er the use of adjoined relativ e clauses for comments, which may be linked by any reasonable connection to the main clause. As well as time, these include comments on place, cause, purpose, reason, “enabling cause”, contrastiv e parallels, conditionals and so on. They are often translated into English by clauses headed by conjunctions such as when, where, while, if, whereas, because and so on. Many sentences may hav e both NP-relativ e and T-relativ e readings, and only context

9.

Operators and Logical Form

In content questions the interrogativ e element normally appears in initial position, preceding the auxiliary; in that position, presumably, it has scope ov er the remainder of the sentence:

There is, howev er, no ev idence that syntactic mov ement is inv olv ed in the formation of content questions in Warlpiri. Initial position is av ailable to any constituent simply by v irtue of the free surface ordering characteristic of Warlpiri syntax generally. Diagnostics, such as the so-called “weak cross-ov er effect”

80.  Warlpiri

1447

(cf. Farmer/Hale/Tsujimura 1986), assuming that they are v alid, suggest that no mov ement is inv olv ed in the formation of content questions:

The Warlpiri sentence here, unlike the English giv en in translation, can receiv e an interpretation according to which the question word binds the possessiv e pronoun. The Warlpiri sentence can hav e the meanings associated with the English passiv e counterpart: ‘Who is being chased by his dog?’ This would follow if the Warlpiri question word were not inv olv ed in an operator-v ariable relationship with a trace in object position resulting from syntactic mov ement. The English of the translation, of course, inv olv es mov ement and the pronoun may not, therefore, function as a bound v ariable without v iolating the restriction against weak crossover. Relativ ization in Warlpiri also fails to giv e ev idence of syntactic mov ement. When the relativ e clause is in protasis position, the internal head of the relativ e may appear in iniAPPENDIX: Abbreviations, Glosses (1) AUXILIARY (AUX) ELEMENTS (a) Sentential Complementizers (COMP) FACTCOMPv facti e FUTCOMP future negati e NEGCOMP v NFACTCOMP v nonfacti e PASTCOMP remote POTCOMP potential RELCOMP relational

tial position within its clause, as in (62) and (63). But, again, this position is simply av ailable by v irtue of the word order characteristics of Warlpiri generally.

Although the ev idence cannot be considered conclusiv e at this point, it is doubtful that operator-v ariable binding relationships are formed, through mov ement, in the syntactic representations of Warlpiri sentences. The interpretations of questions and relati v es, therefore, are ve idently effected through mov ement in logical form, rather than in syntax (cf. Larson 1983).

complementizerkuja, gn ula kapu/kapi, gn arra complementizerkula complementizerkaji past, v usitati ekala (with aspect ka)kala (causal, reason)yungu/yinga/yingi/yi

(b) Aspect (ASP) IMPF (Perfect (c)

past (PERF) Glosses

Person l first person exclusive e inclusive i 2 second person 3 third person

imperfectlpa ∅ is for

pronominal

Number singular s dual d plural p

(3sS and 3sNS, ∅, are not usually glossed).

IMPF not

present imperfect ka usually glossed).

agreement

clitics

Grammatical relation subject S non-subject NS 3sDAT REFL

third singular Dative reflexive

(AGR)

XXI. Syntaktische Skizzen

1448

(d) Pronominal agreement clitics

1s

Singular S NS ju rna

2s

n(pa)

ngku

3s

∅

∅ rla

1di 1de 2d 3d

Dual S rli rlijarra n(pa)pala pala

NS ngali(ngki) jarrangku ngku-pala palangu

1pi 1pe 2p

Plural S rlipa rna-lu nku-lu

NS ngalpa nganpa nyarra

3p

lu

jana

(DAT)

(2) COMPLEMENTIZER SUFFIXES (a) Simultaneous event object-controlled complementizer obviative-controlled complementizer subject-controlled complementizer

OBJCOMP OBVCOMP SUBJCOMP

-kurra -rlarni -karra

(b) Preceding or purposive event preceding event purposive desiderative purposive negative purposive preparatory purposive directional purposive

PRECCOMP PURPCOMP DESIDCOMP NEGPURP PREPCOMP SEQCOMP

-r l a -ku -kupurda -kujaku -kungarnti -kurra

(c) Stative ASSOC PRIV

associative, resultative, perfective privative, negative

-warnu -wangu

(3) NOMINAL SUFFIXES (A) Grammatical case ABS DAT ERG

Absolutive Dative Ergative

∅ (not usually glossed) -ku -ngku, -rlu

(B) Semantic case (a) Non-derivational case ALL COMIT EL LOC RESULT TRANS

allative: ‘to, into’ comitative: ‘with’ elative: ‘from’ locative resultative translative

-kurra -ngkajinta, -rlajinta -ngurlu -ngka, -rla -kurra -karda

(b) Derivational case ASSOC DENIZ

associative, perfective: ‘being’ denizen of: ‘belonging to’

LIKE PERL POSS POSS POSS PRIV PROP SOURCE

simile-former: ‘as, like’ perlative: ‘along’ possessive possessive (on pronouns) possessive (on kinterms) privative, negative: ‘without’ proprietive: ‘having’ elative of source: ‘from, because of’

-warnu -malu, -ngarna, -ngawurrpa, -wardingki -piya -wana -kurlangu -nyangu -nyanu -wangu -kurlu, -parnta -jangka

80.  Warlpiri

1449

(C) Number DUAL PAUCAL PL PL

dual plural, paucal: ‘few’ plural plural

-jarra -patu ∅ -rra (on some pronouns)

(D) Nominal formatives and other clitics CS OTHER WARD

change of state: ‘now, then’ other, next towards

-lku -kari -purda

(4) VERBS (A) Verbal inflections, arranged by conjugation Conjugation class I (a) Tense, co-occurring with Aspect NPST nonpast -mi, ∅ PAST past -ja IRR irrealis -ya-rla (b) Tense/Mood, no Aspect distinction FUT future -jul-ji IMP imperative -ya PREST presentational -nya (c) Non-finite verb forms INF: infinitive -nja NOMIC: -ngu

II

III

IV

V

-rni, -ni -rnu -ka-rla

-nyi -ngu -ngka-rla

-rni, -ni -rnu -nja-rla

-ni -nu -nta-rla

-ku -ka -rni-nya

-ngku -ngka -nga-nya

-lku -nja -rni-nya

-nku -nta -na-nya

-rni-nja -rnu

-nja -ngu

-rni-nja -rnu

-ni-nja -nu

(B) Verb formatives CAUS INCH

causative (transitive) inchoative

N-ma-ni N-jarri-mi

(C) Directionals HITHER BY THITHER

hither, to here past, by, across thither, to there

Acknowledgments We would like to thank the Warlpiri people who hav e been teaching us their language. We would also like to thank Robert Hoogenraad, Stephen Swartz and Dav id Nash for v ery useful comments and discussion. This work has been supported by the Department of Education of the Northern Territory of Australia, the Australian Institute of Aboriginal Studies and the MIT Lexicon Project (System Development Foundation, U.S.A.)

10. References Andrews, Avery. 1985. The major functions of the noun phrase. Language typology and syntactic description. Vol. 1, Clause structure, ed. by Timothy Shopen, 62—154. London.

-rni -mpa -rra Bittner, Maria, and Ken Hale. 1991. Remarks on definiteness in Warlpiri. Cambridge: MS. Massachusetts Institute of Technology. Bouma, Gosse. 1985. A categorial grammar for Warlpiri. Groningen: Nederlands Instituut. MS. —. 1986. Grammatical functions and agreement in Warlpiri. Linguistics in the Netherlands 1986, ed. by Frits Beukema, 19—26. Dordrecht. Brunson, Barbara A. 1988. A Processing Model for Warlpiri Syntax and Implications for Linguistic Theory. Technical Report Computer Systems Research Institute-208. University of Toronto. Carrier, Jill. 1976. Grammatical relations in Warlpiri. Cambridge: Massachusetts of Institute of Technology. MS. Craig , Colette, and Kenneth L. Hale. 1988. Oblique relations and reanalysis in some languages of the Americas. Language 64.2. 312—344.

1450

Farmer, Ann K., Ken Hale, and Natsuko Tsujimura. 1986. A note on weak crossov er in Japanese. Natural Language and Linguistic Theory 4.1. 33—42. Granites, Robin Japanang ka, Kenneth L. Hale, and David Odling -Smee. 1976. Surv ey of Warlpiri grammar. Cambridge: Massachusetts Institute of Technolology. MS. Guerssel, Mohammed, Kenneth L. Hale, Mary Laug hren, Beth Levin, and Josie White Eag le. 1985. A cross-linguistic study of transitiv ity alternations. In Papers from the Parasession on causativ es and agentiv ity at the Twenty-first Regional Meeting, Chicago Linguistics Society, 48—63. Chicago. Hale, Kenneth L. 1973a. Person marking in Walbiri. A Festschrift for Morris Halle, ed. by Stephen Anderson & Paul Kiparsky, 308—344. New York. —. 1973b. Deep-surface canonical disparities in relation to analysis and change: an Australian example. Current Trends in Linguistics, Vol. II, 401— 458. The Hague. —. 1976. The adjoined relativ e clause in Australia. Grammatical Categories in Australian Languages, ed. by Dixon, 78—105. Canberra. —. 1981a. On the position of Walbiri in a typology of the base. Bloomington. —. 1981b. Preliminary remarks on the grammar of part-whole relations in Warlpiri. Studies in Pacific languages and cultures in honour of Bruce Biggs, ed. by Jim Hollyman & Andrew Pawley, 333— 344. Auckland. —. 1982a. Preliminary remarks on configurationality. In Proceedings of the North Eastern Linguistics Society 12, 86—96. —. 1982b. Some essential features of Warlpiri main clauses. In Stephen Swartz (ed.), 217—315. —. 1983. Warlpiri and the grammar of non-configurational languages. Natural Language and Linguistic Theory 1.1. 5—47. —. 1986. Notes on world v iew and semantic categories: some Warlpiri examples. Features and Projections, ed. by Pieter Muysken & Henk v an Riemsdijk, 233—254. Dordrecht. —. 1990. Core structures and adjunctions in Warlpiri syntax. For the proceedings of the Tilburg Scrambling Conference, October 1990. —, and Mary Laug hren. 1986. The structure of v erbal entries. Preface to the Warlpiri Dictionary Verb fascicle. Cambridge: Massachusetts Institute of Technology. MS. —, and Paul Platero. 1986. Parts of speech. Features and Projections, ed. by Pieter Muysken & Henk van Riemsdijk, 31—40. Dordrecht. Jelinek, Eloise. 1984. Empty categories, case, and configurationality. Natural Language and Linguistic Theory 2.1. 39—76. Kashket, Michael Brian. 1984. A Gov ernmentBinding Based Parser for Warlpiri. A Free-Word

XXI. Syntaktische Skizzen

Order Language. Massachusetts Institute of Technology MSc. dissertation. Cambridge, MA. Larson, Richard K. 1983. Restrictiv e modification: relativ e clauses and adv erbs. Univ ersity of Wisconsin doctoral dissertation. Laug hren, Mary. 1977. Pronouns in Warlpiri and the category of number. Paper presented to the conference of linguists, Department of Education, Darwin, Northern Territory. —. 1982. A preliminary description of propositional particles in Warlpiri. In Stephen Swartz (ed.), 129—163. —. 1984. Some aspects of focus in Warlpiri. Paper presented at the Australian Linguistics Society. —. 1985a. The split case hypothesis re-examined: the Warlpiri case. Paper presented at the Australian Linguistics Society. —. 1985b. Warlpiri reflexiv es, ‘inherent’ reflexiv es and grammatical relations. Paper presented at the Massachusetts Institute of Technology. —. 1985c. Case assignment across categories in Warlpiri case. Paper presented at the Linguistics Society of America Winter Meeting. MS. —. 1988a. Some data on pronominal disjoint reference and coreference in Warlpiri. MS. —. 1988b. Towards a lexical representation of Warlpiri v erbs. Thematic Relations (Syntax and Semantics, Vol. 21), ed. by Wendy Wilkins, 215—242. New York. —. 1989. The Configurationality Parameter and Warlpiri. Configurationality: the Typology of Asymmetries, ed. by Laszlo K. Maracz & Pieter Muysken, 319—353. Dordrecht. —. 1992. Secondary Predication as a Diagnostic of Underlying Structure in Pama-Nyungan Languages. Thematic Structure: Its role in Grammar (Linguistic Models Series), ed. by I. Roca. 199—246. Berlin. Nash, David. 1982. Warlpiri prev erbs and v erb roots. In Stephen Swartz (ed.), 165—216. —. 1986. Topics in Warlpiri Grammar. Outstanding Dissertations in Linguistics. Third Series. Copyright 1985. New York, London [Published v ersion of Massachusetts Institute of Technology doctoral dissertation with the same name, 1980]. Riemsdijk, Henk van. 1981. On ‘Adjacency’ in phonology and syntax. Proceedings of the North Eastern Linguistics Society, 11. 399—413. Simpson, Jane. 1983a. Discontinuous v erbs and the interaction of morphology and syntax in Warlpiri. In Proceedings of the Second Annual West Coast Conference on Formal Linguistics. February, 1983. —. 1983b. Aspects of Warlpiri morphology and syntax. Massachusetts Institute of Technology doctoral dissertation. Cambridge. —. 1988. Case and complementizer suffixes in Warlpiri. Complex sentence constructions in Aus-

81.  The Syntax of Creole Languages

tralian Aboriginal languages (Typological studies in language, Vol. 15), ed. by Peter Austin, 205— 218. Amsterdam. —. 1991. Warlpiri Morpho-syntax: a lexicalist approach. Studies in Natural Language and Linguistic Theory. Dordrecht. —, and Joan Bresnan. 1983. Control and obv iation in Warlpiri. Natural Language and Linguistic Theory 1.1. 49—64. [Rev ised v ersion of a paper with the same name in Proceedings of the First Annual West Coast Conference on Formal Linguistics, Stanford University, January 1982.] —, and Meg Withg ott. 1986. Pronominal clitic clusters and templates. The syntax of pronominal clitics. Syntax and Semantics, Vol. 19), 149—174. New York. Speas, Marg aret J. 1990. Phrase Structure in Natural Language. Studies in Natural Language and

81.

1451

Linguistic Theory 21. Dordrecht. Swartz, Stephen (ed.) 1982a. Papers in Warlpiri grammar: in memory of Lothar Jagst. Work-Papers of SIL-AAB. Series A, Volume 6. Berrimah, N. T. —. 1982b. Syntactic structure of Warlpiri clauses. In Stephen Swartz (ed.), 69—127. —. 1991. Constraints on zero anaphora and word order in Warlpiri narrativ e text. SIL-AATB Occasional Papers 1 SIL, Darwin, N. T. [Published v ersion of Pacific College of Graduate Studies MA thesis with the same name, 1988]. —. 1988b. Pragmatic structure and word order in Warlpiri. Pacific Linguistics: Papers in Australian Linguistics No. 17.

Kenneth L. Hale, Cambridge/MA (USA),Mary Laughren, Brisbane (Australia),Jane Simpson, Sydney (Australia)

The Syntax of Creole Languages

1. 2. 3. 4. 5. 6.

Introduction The Lexicon Phrase Structure Sentence Grammar Conclusion References

1.

Introduction

The term ‘creole’ has been applied to a wide v ariety of languages. For instance, languages as v aried as Proto-Germanic, Egyptian, Songay, Mbugu and Middle English hav e been hailed as creoles. Such indiscriminate use depriv es the label of all meaning, and has led to the belief that there is no distinct creole typology (see Muysken 1988 and comments thereon in Section 5. below). In what follows, the term will be restricted to what, in Bickerton (1988), were described as ‘plantation creoles’: languages that arose among peoples of div erse linguistic backgrounds who had been remov ed from their traditional homelands under the aegis of colonial power and who were thus obliged to ‘inv ent’ some means of mutual communication. Such languages would include, for example, Gullah, Haitian, Sranan, Saramaccan, Papiamentu, Seselwa, Morisyen and Hawaiian Creole, to take a random sample of the two or three dozen exemplars. Much of what can be said about these languages would apply also to ‘fort’

and ’maritime’ creoles, as defined in Bickerton (1988). Howev er, it is only about ‘plantation creoles’ that one can make typological statements which are both general and detailed, without requiring frequent disclaimers of the type ‘except in the case of X, where ...’ Although questions surrounding the origin of creoles are and always hav e been controv ersial, and are thus best left unbroached in a study such as this, it is difficult to describe these languages, except at a superficial lev el, without giv ing some idea of the problems they faced and the solutions that, accordingly, they were obliged to adopt. The essentials of creole syntax were created, for each of the languages in question, in a single generation. The input to this generation was a highly v ariable pidgin continuum, the greater part of which was degraded to a point at which it is probably fair to say that it was no longer rule-gov erned (in any sense of linguistic rule); certainly, a syntactic system could not hav e been induced from it. At the same time, the input lexicon was radically reduced, both in the number of av ailable items and the properties which those items possessed. Yet the creole-forming generation was equipped with the same set of syntactic principles as are all other members of our species. The interaction between inv ariant principles and a degenerate lexicon is what giv es creole languages their unmistakeable stamp.

XXI. Syntaktische Skizzen

1452

2.

The Lexicon

In general, creoles, at least in their early stages of dev elopment, hav e a lexicon considerably smaller than that of their source (superstrate) languages, ev en though in most cases the latter supply 90% or more of the basic morpheme stock. Missing from the lexicon are reflexes of the less frequent content words, all or almost all bound morphology, and a large percentage of free grammatical morphemes, including especially superstrate articles, auxiliary v erbs and complementizers, as well as some question words, prepositions and conjunctions. Content words that surv iv e tend to be disyllabic: superstrate monosyllables with final consonants usually add an epenthetic v owel, although this may disappear ov er time if creole and superstrate remain in contact; trisyllabic or longer words tend to be lost prior to creole formation. With regard to the semantics of content words, out of the sev eral members of a single semantic set (e. g. ‘talk’, ‘say’, ‘speak’, ‘report’ etc.) only one may surv iv e (e. g. taki ‘talk’ in the Suriname creoles). Deficits in the content side of the lexicon are made up either by broadening the semantic ranges of existing morphemes or by coinage of new items (see Hancock (1980) for some typical ‘incoining’ dev ices). A part of the v ocabulary which tends to reduce ov er time is drawn from substrate languages. In Saramaccan, this has been estimated at as high as 50% (Price 1976) and in Berbice Dutch at 27% (Smith/Robertson/Williamson 1987) although a range between 2% and 10% is more common for most synchronic creoles. In general, little core v ocabulary is drawn from substrate sources; retentions are most numerous in ‘priv ate’ domains such as body parts, diseases, food, religion and so on. The deficit due to loss of a substantial percentage of grammatical morphemes is made up by ‘semantically bleaching’ content words and (where these are disyllabic) reducing them to monosyllables usually by truncating the unstressed syllable: thus in Sranan sabi ‘know’ giv es sa ‘irrealis particle’, in Saramaccan taki ‘talk’ giv es taa ‘factiv e complementizer’, in Crioulo kaba ‘finish’ giv es ba ‘completiv e particle’, and so on. These monosyllabic forms will be reduced in stress and are perceiv ed by speakers as quite different in meaning from their etyma ev en where phonological truncation has not occurrred, cf. Hawaiian Creole he stay g o come ‘He will

be coming’. The loss of ov ert case-distinctions in creole pronouns, although common (e. g. Haitian), is by no means always complete. Normally most pronouns are inv ariant for case in nominativ e, accusativ e and possessiv e forms. Nominativ e and accusativ e may be distinguished in the first person singular (Palenquero, Seselwa) or in the third (Sranan, Guyanese), but seldom elsewhere. Especially in Portuguese creoles, pronouns may hav e emphatic forms.

3.

Phrase Structure

3.1. Order Within NP Creoles are strictly configurational SVO languages. Howev er, the order of constituents within NP does not always follow what is predicted by this ov erall schema. In some cases, determiners, such as possessiv es, may follow nouns:

Sometimes determiners, as well as adjectives, may either precede or follow the head:

Sometimes ev en in the same NP determiners can both precede and follow the head:

Compare Haitian NP-sa-a ‘NP-this-the’. But in general, articles hav e ev olv ed too recently from demonstrativ e adjectiv es for such a split to manifest itself. For instance, the di and (d)a articles found univ ersally in English creoles are probably not reflexes of English the but truncated forms of creole disi ‘this’ or dati ‘that’ respectiv ely (note that where a creole is referred to as ‘English’, ‘French’ etc., no claims of genetic affiliation are intended; the terms are used neutrally and entirely for con v enience, for making generalizations about sets of creoles whose lexicons are drawn predominantly from a particular language).

81.  The Syntax of Creole Languages

A charcteristic of creoles is the absence of any determiner where referents are generic, hypothetical or ev en indefinite (See Bickerton (1981, 23—25) for discussion). Where the determiner is absent the pluralizer also is obligatorily absent. Pluralizers are also absent where there is any other indicator of plurality, e. g. a quantifier or a numeral, within the phrase. In general, the position of adjectiv es v is-av is their noun heads seems to be determined by their order in the superstrate language: English creoles hav e only preceding adjectiv es, Spanish and Portuguese creoles only following ones, while French show a mixture, with the majority being postnominal. Relativ e clauses and other types of complement, such as prepositional phrases, follow their heads universally. In contrast, genitiv e phrases are highly v ariable, probably more so than any other construction in creoles. We find at least fiv e types of possessiv e NP: possessor-possessed, possessed-possessor, possessor-possessiv e adjecti v e-possessed, preposition-possessor-possessed and possessed-preposition-possessor:

1453

If the Theme/direct object comes first, the indirect object is, in many cases, preceded by a verb (see Section 4.1. below). Although v erbs in general generate their Themes in direct-object position, large numbers of transitiv e (but no ditransitiv e) v erbs are also ‘ergativ e’ in nature: that is to say that like English melt they may appear either with Agent subject and Theme object, or with Theme subject and no object:

The restrictions on what verbs may behave in this way hav e not yet been clarified for any creole language: it may be that causativ e v erbs like kill (= ‘cause to die’) are exluded, or that selectional restrictions are inv olv ed (inanimate Theme subjects seem more acceptable than human ones, for instance). More study is needed here. Prepositional phrases do occur in the v erb phrase, although as noted abov e the inv entory of prepositions is depleted as compared with the source language. It has been claimed that there are postpositions in Berbice Dutch (Smith et al. 1987):

As (4 b), (8) indicate, there may be more than one type of genitiv e in the same language. In general, creoles follow the order of their superstrates, but this is certainly not the case in e. g. Papiamentu. 3.2. Order Within VP Verbs inv ariably precede their complements, whether these are nominal or sentential. Most if not all creoles hav e ‘dativ e-shift’, that is, constructions in which an indirect-object Goal NP precedes a direct-object Theme NP:

Howev er, it is by no means clear that this is the case. Muysken (1987) discusses some v ery similar structures in Saramaccan that can perhaps better be analysed as NPs modelled along the lines of the genitive NP:

XXI. Syntaktische Skizzen

1454

It is possible that both Dutch and Portuguese prepositions hav e been reanalysed as nouns and that the phrases in (12) and (13) are in fact possessiv e NPs similar in structure to (4 a). Adv erbs are rare in creoles generally. Occasionally they may be formed by reduplication (e. g. Saramaccan hesi ‘quick’, hesihesi ‘quickly’). More frequently a serial v erb construction may be used: Although TMA particles are almost always free, their behav ior is quite distinct from that of the ‘true’ v erbs: the latter, for instance, will undergo the process of v erb copying (sometimes called ‘predicate fronting’) discussed in Section 4.6.3., whereas the former will not.

4. Negation is inv ariably predicate-external (that is, nev er falls inside the predicate as do English not, French pas); this is true ev en where pa is the chosen negator (as it is in French creoles generally). In a large majority of creoles the negator precedes the predicate; in a few (Palenquero, Sao Tomense, Berbice Dutch) it follows:

3.3. The Structure of INFL Creoles hav e a characteristic array of particles that indicate tense, modality and aspect (TMA). These are always free morphemes, usually monosyllabic, and always appear in the order indicated:

The semantics of these particles and their combinations are consistent ov er a wide range of creoles (see discussion in Bickerton 1981, 73—99). There are, howev er, minor v ariations; completiv e aspect is marked in some creoles by a VP-final v erb deriv ed from a superstrate word ‘finish’ (Anglo-Creole done, Franco-creole fini(r), Ibero-Creole (a)kaba(r)). This particle may be reduced (e. g. Crioulo ba, Seselwa (i)n) and may in addition be incorporated into INFL as in the French creoles of the Indian Ocean:

Sentence Grammar

4.1. Serial Verb Constructions Perhaps no area of creole grammar has giv en rise to more confusion and misunderstanding than serial v erb constructions (SVCs). In early work they are often treated as conjoined sentences. For the past couple of decades a large majority of scholars has treated them as conjoined (or more often, embedded) VPs, except for Byrne (1987), who treated them as CPs. Many writers, including Baker (1988), Sebba (1987) and Lefebv re (1986) hav e treated them as ‘closed-list’ items, although with little agreement as to whether these lists were drawn directly from the lexicon (Lefebv re) or depended on v erb-subclass: unaccusativ e, unergativ e, triadic etc. (Baker). Howev er, most agreed that there were sets of ‘first’ v erbs and ‘second’ v erbs that could not commute (ev en in the face of empirical ev idence that v erbs of motion, for instance, or — less often — triadic v erbs could appear in both positions). This approach arose in part because one function of SVCs was to ensure that arguments affected by the loss of prepositions were properly gov erned and assigned case. This meant that among the statistically most frequent SVCs are sentences like (20), where the v erb equiv alent to g ive gov erns benefactiv e arguments, or (21), where the v erb equiv alent to take gov erns instrumental arguments:

81.  The Syntax of Creole Languages

As has been pointed out, one cannot hav e sentences like *nu bay u pote liv-la or *a tei sindeki kii faka, but this is because of their semantic incoherence rather than any syntactic constraint. In fact, as shown in Bickerton (1989), Bickerton/Iatridou (1988), potentially any v erb may enter into SVCs in any position, prov ided that semantic or pragmatic anomalies do not arise. As for SVCs themselv es, far from being lexically or categorially circumscribed they are simply IPs adjoined at I’ or V’. They cannot be VPs because the binding and control facts (among other: see Bickerton/Iatridou (1988) for further discussion) rule out this possibility:

1455

subject in a subordinate clause corefers with an NP in the matrix clause that nonsubject will hav e zero phonetic representation, in many cases obligatorily:

The mechanisms inv olv ed in constructions of this type will be discussed in Section 4.2. below. 4.2. Empty Categories

They must be adjoined because they are not subcategorized arguments of V and therefore cannot be attached to a V-complement node. At least if they are tensed, they must be IP rather than CP since otherwise matrix I would not be able to gov ern and bind subordinate I and would therefore be able to differ from matrix I in tense features, contrary to fact. At one time it was believ ed that SVCs in creoles carried tense only on the matrix v erb. This was challenged by Byrne (1987) for Saramaccan, where apparently there is dialect v ariation in this respect; Bickerton (1989) showed that tensing of subordinate clauses was the norm in Seselwa and Morisyen. TMA features in lower clauses cannot v ary from the matrix clause:

In general, extraction may take place from SVCs (see Section 4.6. below). A further feature of SVCs is that usually wherev er a non-

If it is the case that creole languages owe their typology mainly to the relativ e paucity of the lexicon that they inherited from their antecedent pidgin, one would expect them to maximize their use of empty categories (ECs), and this is indeed the case, as we hav e already seen in SVCs. In particular, relativ e pronouns tend to be lost (see Section 4.4. below); although they may be reconstituted at a later stage, the use of zero relativ izers (i. e. null operators) may continue in at least some environments. The ‘missing pronouns’ in sentences such as (24) abov e are in fact the traces of null operators that hav e been mov ed to an A’-position where they can be gov erned by a coreferent NP (see Bickerton/Iatridou (1988) for discussion):

In (26), t is the trace of the null operator 0 that has been mov ed to a position in which Amba can gov ern it. The motiv ation for this procedure lies in the potential ambiguity of SVCs. Consider the consequences if (26) carried an ov ert pronoun in object position: (27) *Kofii naki Ambak [PROi/k kiri enk/i] Since SVCs are not subcategorized for, the matrix v erbs are not ‘control’ v erbs and do not specify whether matrix subject or matrix object shall control PRO; since both c-command PRO, the latter may corefer with either, prov ided that indexing is disjoint. In other words, it would be possible to interpret (27) not just as ‘Kofi struck Amba dead’, but also as ‘Kofi struck Amba (and) Amba killed

XXI. Syntaktische Skizzen

1456

Kofi’. This possibility is av oided if a mechanism exists by which the subordinate object must be coindexed with the matrix object; in that case the subordinate subject, to be disjoint from the object, must corefer with the matrix subject, and the desired reading is the only reading. In contrast, pronouns may freely be inserted in Seselwa or Morisyen SVCs; since these SVCs are tensed, they must be attached within the gov ernment domain of matrix I, and therefore PRO is no longer c-commanded by the matrix object and thus cannot corefer with it. In other words, when there is no undesired reading, null operator mov ement is optional. ECs in subject position may be either PRO or pro. PRO is found in tenseless clauses in all creoles, and is unproblematic: whatev er its status v is-a-v is gov ernment, it is not assigned case if there is no lexical INFL, but it can corefer with any c-commanding NP. Subject proprox, found so far only in tensed SVC clauses in Saramaccan (Byrne 1985), Seselwa and Morisyen (Bickerton 1989), is another matter. In most pro-drop languages pro may occur in main clauses, may corefer as freely as an ov ert pronominal and may be freely replaced by an ov ert pronominal. In creoles it may occur only in the subordinate clauses of SVCs, must be bound by the matrix subject NP and may not always be replaceable by an overt pronominal. There are howev er v alid reasons for all these differences of behav ior. Null elements hav e to be identified; in many pro-drop languages they can be identified by a set of rich agreement features in INFL. Naturally, giv en the prev ailing morphological pov erty, AGR in creoles is featureless, so identification has to be by coindexing with a unique antecedent. In main clauses and in subcategorized subordinate clauses (which are CP-protected) subjects can v ary unpredictably; howev er, if an adjunct clause is attached at I’, a chain of obligatory coindexing (matrix subject and matrix I, matrix I and subordinate I, subordinate I and subordinate subject) insures that matrix subject will bind subordinate subject. These facts account for the restricted env ironments and obligatory corereference of pro. As for its replaceability, this is uncertain in Saramaccan, perhaps due to ongoing language change: howev er, in Seselwa and Morisyen, where tensed SVCs are more firmly established, ov ert subject pronouns may occur in SVCs (which incidentally, and contrary to appearances, do not then become co-ordinate structures, but remain SVCs, as shown

in Section 4.6.2.):

Since pleonastic elements such as it, there or the il of French il y a are lost in pidginization, a zero proarb is a feature of many creoles:

As (29) and (30) indicate, a single creole verb expresses, in perhaps all affirmativ e cases, both existence and possession: Morisyen, Seselwa and Hawaiian Creole hav e in addition a distinct negativ e form for both (napa, nomoa). Note that in creoles where both are found, proarb and proprox are in complementary distribution, the former occurring only in those env ironments (main and subcategorized subordinate clauses) from which the latter are barred, and vice versa. 4.3. ‘Accomplished’ Versus ‘Unaccomplished’ Complements In English there is no morphosyntactic distinction between complements whose action is necessarily accomplished (I manag ed to leave), necessarily unaccomplished (I failed to leave) or of unspecified accomplishment (I decided to leave). In all creoles there exists some degree of distinction: in some creoles, all three conditions are distinguished, while in the remainder the second and third conditions are collapsed. In Saramaccan, Papiamentu and Haitian, for example, all three conditions can be realised, and in the same way. In all creoles there

81.  The Syntax of Creole Languages

is a complementizer deriv ed from the ‘for’ preposition (pu and pa in French and Iberian creoles, from pour and para respectiv ely): the absence of this, or its replacement by auxiliary g o or equiv alent, indicates the ‘accomplished’ condition, its presence alone indicates the ‘indeterminate’ condition, and its presence plus a marker of past or anterior tense in the following clause indicates the ‘unaccomplished’ condition:

In Jamaican, Morisyen and Hawaiian Creole, howev er, the subordinate clause either cannot be tensed or cannot be marked ov ertly for tense, so that the use of the ‘for’ complementizer cov ers both the b and c conditions. Note that the distinction between the a and b/c conditions can be clearly indicated by the use of an adversative tag:

1457

4.4 Anaphora The status of anaphora in creoles depends on the extent to which anaphors were retained or lost during pidginization and whether reconstitution has taken place or not. Since anaphora in Iberian and Dutch creoles has hardly been studied so far, present remarks must be largely limited to French and English creoles. In English creoles, reflexiv es hav e been either retained or reconstituted without significant change; in consequence their distribution accords fairly well with Principle A of binding theory (Chomsky 1981). Howev er, French ‘short-range’ reflexiv e clitics were lost (along with cliticised forms in general) in the process of pidginization. Only reflexes of the emphatic pronoun series moi ‘me’, lui ‘him’ etc. surv iv ed in the forms mwen, li etc. A reflex of meme surv iv ed, and can be conjoined with li, but the meaning is different from the French ‘long-range’ reflexiv e lui meme: for most speakers, li-mem must hav e a referent outside the sentence that contains it, that is to say it is a sort of ‘long-range pronoun’. At some time considerably later than creolization (see Carden/Stewart (1988) for further discussion) reflexiv es based on bodyparts (ko-li, literally ‘body-his’; tet-li, literally ‘head-his’) were reconstituted by Haitian speakers, but by this time assignment of the features [+ pronominal, + anaphoric] to li had been firmly established. The result in at least one contemporary dialect of Haitian is illustrated in (37):

The reason for claiming absence of ov ert tensing rather than nonfinite status is that in all creoles except Seselwa (including Morisyen) the ov ert subject of a ‘for’ clause takes nominative case:

Corne (1988) shows a similar situation for Morisyen, although he argues (incorrectly, in this writer’s opinion) for a historical scenario the exact rev erse of Carden and Stewart’s. In Seselwa there is massiv e v ariation among three forms, li-mem, li and son lekor (literally, ‘his body’). For some speakers, only the first and third may be used reflexiv ely; for others, all three may be used, but preferred usage varies according to the verb:

1458

This situation appears anomalous in the light of current binding theory, but nothing in that theory states that there shall be anaphors and pronouns; only that, if there are anaphors and pronouns, their binding domains will differ in a certain (principled) way. The lexicallearning approach proposed in the introductory section predicts that if only one set of nonreferential items surv vi es pidginization, then that set will discharge all nonreferential functions, i. e. function as both anaphor and pronoun. While reflexiv e forms tend to be reconstituted ev en if lost, reciprocal forms seem always to be lost and are relativ ely seldom reconstituted. In some entirely unrelated creoles, the word for ‘friend’ functions as a reciprocal:

Similar constructions are found in Cabo Verdiense (with kompanyer) and in Morisyen (with kamarad). In Saramaccan, there is no difference between reciprocals and reflexiv es, so that (38 a) is ambiguous and can only be disambiguated by (38 b):

4.5. Relativization One empirical generalization that can be made about creoles is that relativ izers and question-words will not be homophonous (unlike English who, French qui etc.). We find pairs like Saramaccan ambe ‘who?’, di ‘relativ izer’, or Guyanese (a)hu ‘who?’, we ‘relativ izer’, or Papiamentu quen ‘who?’, cu ‘relativ izer’. It is true that in French creoles gen-

XXI. Syntaktische Skizzen

erally, the relativ izer is ki, but ki alone does not function as a question-word; rather it occurs in compounds such as Haitian ki moun, literally ‘who-person?’, though it may also be replaced by another form such as Seselwa lekel. Historically, at least some of the synchronically opaque forms in other creoles were deriv ed in a similar manner, e. g. synchronic Sranan suma ‘who’ was originally usuma, from the English forms ‘who?’ and ‘someone’. It is hard to think of any language, howev er primitiv e, as lacking question-words, so that the bimorphemic question-words found frequently though not univ ersally in creoles were probably created in the pidgin phase. Howev er, there is no similar necessity for any kind of relativ e pronoun or relativ izing complementizer; indeed, as early stage pidgins lack any kind of embedding, they would hav e no place for such a morpheme. It is likely that in their early stages creoles lack any kind of relativ izer. In Hawaiian Creole, the most recently formed, there is no relativ izer at the basilectal lev el (although v arieties closer to English hav e that) so that ev en if the subject of the clause corefers with the head, there is no ov ert marker of relativ ization:

In other creoles, similar structures may occur even though a relativizer exists:

Absence of relativ izer is especially common in existential sentences, e. g. the following from Seselwa:

81.  The Syntax of Creole Languages

Comparison with the ECs in SVC structures discussed in Section 4.2. abov e indicates what is happening here: the empty subject position contains the trace of a null operator that is adjoined to the relativ ized NP. Subsequent acquisition of a relativ izer by creoles is largely to circumv ent the processing problems discussed by Bever/Langendoen (1971). One aspect of relativ ization which should be touched on here relates to a striking asymmetry between French and English creoles. In French creoles, extraction from subject position (whether for a questioned or a focused constituent) inv olv es use of the relativ izer ki; in English creoles, relativ izers are barred from this environment:

1459

struction discussed in Section 3.2. discharges the functions of agentless passiv es. A few creoles (Papiamentu, Seselwa) hav e a rather marginal passiv e, but it is questionable whether this really forms a part of either the original or the synchronic basilectal grammar; in all probability such forms are deriv ed through contact with European languages that do hav e passiv es (it is interesting to note that the Papiamentu passiv izing v erb wordu is derived from a Dutch loan). Recent work by de Graff (1993) shows that Haitian, at least, has raising v erbs such as g enlè ‘seem’. It also seems likely that NPmov ement is inv olv ed in ‘ergativ e’ examples like (48 a), on the assumption that their underlying structure is that of (48 b), where e represents an empty node:

4.6.2. WH-movement

The reasons for this are still not entirely clear, but there is reason to believ e that the surv iv al of the relativ izer ki is inv olv ed. Assuming that both question-words and focused constituents are base-generated in place and that a link between these constituents and some v acant A-position is established by means of an operator which mov es from that A-position to an A’-position adjoining the fronted constituent, then that operator may be either null or ov ert. In French creoles ki was directly inherited as an operator; in English creoles no ov ert operator was inherited, and there is reason to believ e that the later reconstituted forms that appear in relativ e-pronoun position in those languages (Guyanese we, Sranan/Saramaccan di, etc.) are complementizers rather than operators. 4.6. Movement 4.6.1. NP-movement NP-mov ement (mov ement to an A-position) is sev erely limited in creoles. Most creoles lack a passiv e altogether; the ‘ergativ e’ con-

As has already been noted, WH-mov ement in creoles (except perhaps French creoles) inv olv es null operator mov ement. Howev er, it is worth looking at some specific examples. In general focused NPs and PPs (arguments in other words) can be freely linked with empty argument positions, subject of course to subjacency:

In some languages, a particle usually equiv alent to the pre-NP copula precedes the focused constituent (a in Guyanses, se in French etc.). This particle may be optionally omitted in some French creoles where ki is present (see Section 4.5., example (46)). There is no focusing particle in synchronic Saramaccan.

1460

XXI. Syntaktische Skizzen

Linking of focused constituents with empty A-positions is possible ev en when the latter lie within the lower clauses of SVCs:

Further, only v erbs (including most modals) and predicate adjectiv es undergo this process; prev erbal TMA markers do not, either individually or with V: More surprisingly, similar linkage is possible ev en when the SVC clause containing the EC is tensed and has an overt subject:

This rules out the possibility that even tensed and subjectful SVCs are co-ordinate, and tends to confirm that SVCs are adjunct IPs; if they were CPs and the CPs were not Lmarked (Chomsky 1986), as would hav e to be the case for adjuncts, the additional barrier should bar any linkage between the coindexed elements. 4.6.3. Verb Copying There remains a feature found perhaps ev erywhere but in Hawaiian Creole which is sometimes referred to as ‘predicate fronting’, but is perhaps better described as v erb copying. The term ‘predicate fronting’ suggests that a whole predicate is fronted, which (except in Hawaiian creole sentences such as (52)) is never the case:

In creoles generally, the v erb alone (or a predicate adjectiv e, alone) may appear initially, but only if at the same time it remains in situ:

Variation exists between creoles and ev en within creoles (whether dialectally or idiolectally is not yet determined) with regard to whether the copying process applies to all verbs in SVCs or only to the matrix verb. The existence of v erb-copying raises at least two interesting questions. One is the question why v erbs (heads, rather than maximal projections) are av ailable for focusing. The answer may be that for the process ‘Mov e alpha’, ‘alpha’ is defined for creoles as ‘argument-lev el unit’, that is, all arguments (whether NP or PP) and argument-taking constituents (V or Adj; the question whether adjectiv es in creoles are a subclass of v erbs will not be addressed here). Another is why it is necessary for v erbs to be copied: ‘v erbsecond’ and ‘scrambling’ processes in many languages successfully mov e v erbs from their deep-structure positions. One explanation might be that creole v erbs lack trace and thus cannot be mov ed. But this is not only implausible in itself; it would suggest that v erbcopying should not be blocked by the presence of other mov ed elements in intermediate Spec-C positions. At least in Saramaccan, the only creole for which the possibility has so far been tested, this is not the case:

81.  The Syntax of Creole Languages

Note that there is nothing semantically anomalous in the translation of (54 b), e. g. ‘He doesn’t know what to eat’. For some speakers (54 b) is ungrammatical because they cannot extract v erbs from lower clauses; for those who can, howev er, the most likely reason is that the presence of andi in the lower Spec-C position blocks v erb copying. This is confirmed by (55), where the only difference between the (a) and (b) sentences is that in (55 a) ka is a complementizer that nev er occurs in questions whereas in (55 b) naase is the WH-question form. This would seem to mean that v erbs do ‘mov e’ and do leav e trace and that the process of v erb-copying is subject, like WH-mov ement, to subjacency.

5.

Conclusion

The foregoing, while it outlines the core areas of creole grammar, has for reasons of space had to omit mention of many other features with respect to which creoles show striking similarities. Such features include the following: negativ e indefinite NPs, whether in subject position or elsewhere, are accompanied by v erbal negation; there are distinct co-ordinate conjunctions for clauses (usually a reflex of superstrate and) and NPs (usually a reflex of superstrate with); co-ordinates of the type of John went home and had dinner normally require an ov ert subject for the second conjunct; there are usually different copula forms for NP and locativ e PP predicates; nonpunctual aspect markers with stativ es or adjectiv es indicate ongoing process; questions inv olv e no syntactic differences from affirmativ es; v erbs of perception take tensed clausal complements with TMA markers and overt nominative subjects. A recent article (Muysken 1988) claimed that ‘the v ery notion of a “creole” language from the linguistic point of v iew tends to disappear if one looks closely: what we hav e is

1461

just a language.’ The present article has considered a wider range of data than Muysken considered, and is based on more accurate information (for instance, his claim that the Indian Ocean creoles lack serialization is simply false, as sev eral examples giv en here attest). The degree of similarity in creole grammars that has been shown would be striking enough ev en if it related to a group of languages that were genetically affiliated. When one considers that there is neither a direct genetic link nor (in a large majority of cases) any significant historical contact between these languages, and that the ancestral languages of their first speakers were extremely div erse typologically, the degree of similarity that prev ails is unexpected and demands to be accounted for.

6.

References

Baker, Mark. 1988. Object sharing in serial v erb constructions. MS, McGill University. Bever, Thomas G., and D. Terence Lang endoen. 1971. A dynamic model of the ev olution of language. Linguistic Inquiry 2. 433—463. Bickerton, Derek. 1981. Roots of language. Ann Arbor. —. 1988. Creole languages and the bioprogram. Linguistics: the Cambridge surv ey, Vol. 2, ed. by Frederick J. Newmeyer, 267—284. Cambridge. —. 1989. Seselwa serialization and its significance. JPCL 4. —, and Sabine Iatridou. 1988. Verb serialization and empty categories. Unpublished MS. Byrne, Francis. 1985. proprox in Saramaccan. Linguistic Inquiry 16. 313—320. —. 1987. Grammatical relations in a radical creole. Amsterdam. Carden, Guy, and William A. Stewart. 1988. Binding theory, bioprogram and creolization: ev idence from Haitian Creole. Journal of Pidgin and Creole Languages 3. 1—68. Chomsky, Noam. 1981. Lectures on gov ernment and binding. Dordrecht. —. 1986. Barriers. Cambridge, MA. Corne, Chris. 1988. Mauritian Creole reflexiv es. Journal of Pidgin and Creole Languages 3. 69—102. de Graff, Michel. 1993. Is Haitian Creole a prodrop language? Atlantic meels Pacific, ed. by F. Byrne & J. Holm, 71—90. Amsterdam. Hancock, Ian. 1980. Lexical expansion in creole languages. Theoretical orientations in creole studies, ed. by A. Valdman & A. Highfield. New York.

1462

Lefebvre, Claire. 1986. Relexification in creole genesis rev isited: the case of Haitian Creole. Substrata v ersus univ ersals in creole genesis, ed. by Pieter Muysken & Norval Smith, 279—300. Amsterdam. Muysken, Peter. 1987. Prepositions and postpositions in Saramaccan. Studies in Saramaccan language structure, ed. by M. C. Alleyne. Caribbean Culture Studies 2, University of Amsterdam. Price, Richard. 1976. The Guiana Maroons: a his-

XXI. Syntaktische Skizzen

torical and bibliographical introduction. Baltimore. Sebba, Mark. 1987. The syntax of serial v erbs. Amsterdam. Smith, Normal S. H., Ian E. Robertson, and Kay Wilkinson. 1987. The Ijo element in Berbice Dutch. Language in Society 16. 49—90.

Derek Bickerton, Honolulu, Hawaii (USA)

1463

XXII. Syntax und Nachbardisziplinen Syntax and Neighbouring Disciplines

82. Syntax and Philosophy 1.1. 1.2. 1.3. 1.4. 1.5. 1.6. 2.1. 2.2. 2.3. 2.4. 2.5. 3.1. 3.2 3.3. 3.4. 3.5. 4.1. 4.2. 4.3. a 4.4. 4.5. 4.6. 5.1. 5.2. 5.3. 5.4. 5.5. 6.1.

Introductory Remarks General Historical Remarks Plato and Grammar Aristotle on Language Hellenistic Grammar Summary Syntax and the Tradition of Analytic Philosophy Motivations The Concepts of Artificial Language and Formal Language The Rise of “ Ordinary Language Philosophy” and its View of Grammar Summary Unification of Formal Semantics and Generative Grammar Relation of Syntax to Semantics in Natural Languages An Alternative View The Prescriptive-Normative Notion of Grammar Summary Philosophical Problems Concerning Generative Grammar Why View Grammar this Way? Relations between Syntax as Formal Object and the Language User Ontological Issues; Realism Syntactic Categories Epistemology and Syntax Syntax and Theories of Cognition What can Syntax tell about the Mind? Syntax and Computation Modularity The Innateness Hypothesis; Does Syntax Help? Epiloque References

1.1. Introductory Remarks Over the past twenty-five years there has been a lot of interaction between philosophers and grammarians. This is due to a large extent to the emergence of generative grammar. Generative grammar presented

philosophers with a threefold challenge. First, it presented the syntax of natural languages as a precise formal structure worthy of investigation in its own right. Second, it forced philosophers to face the problem of language acquisition and to consider hypotheses about learning formulated in precise ways. Third, it showed that there are many clear and interesting questions to be raised and answered about the grammatical structures of natural languages even in contexts in which neither the questions nor the structures are semantically motivated. It helps to place these matters into perspective if we look briefly at the beginnings of speculation about language in Western thought. As with so many other sciences, these speculations too had their roots in philosophy. Our survey will show that most of the key issues that concern linguists and philosophers with respect to grammar today were already formulated in the period between Plato’s activities and the second century A. D. in which Apollonius Dyscolus formulated his grammar. 1.2. General Historical Remarks Philosophers today are interested in language mainly because of its semantics or system of meaning. This was not true of early Greek philosophy. As the CRATYLUS testifies, one of Plato’s main interests was in the origin of language, and another was the sound system of a language (PHILEBUS 18 b—c). This was also a focus for other philosophers such as Democritus. This should not be surprising. That only certain combinations of sounds can occur in various languages, and that we can present the phonological structure of a language as resting on a small set of basic letter-sounds is more amazing than the fact that once a part of a language has been given syntactic and phonological interpretation, we can discern certain elements as standing for certain elements of reality. To be sure, these speculations about origin, sound systems, and grammar should not be interpreted as amounting to “ theories”. But at least a

1464

glance at these early speculations should place the modern preoccupation with semantics into proper perspective. 1.3. Plato and Grammar In one of Plato’s dialogues we find an early recognition of the notion of grammaticality. By that we mean that we find here an explicit mention of the fact that in addition to conditions that tell us when a sentence is true or false — a matter of semantics — there are antecedent conditions of combinability of words or phrases, distinct from questions of truth and meaning, that a well-formed sentence has to meet. These conditions are on combination of elements from different syntactic categories. They are not the same as generalizations about what will be acceptable to a hearer as something that can be given an interpretation (Chomsky 1957). 1.3.1. SOPHIST 261 d—262 d In this passage Plato makes the following points. He considers what we would call meaning-bearing elements and points out that the combinations of some yield sentences while the linking of others does not. He goes on to characterize the “ right” combinations in terms of syntactic categories. We have here, then, the first rough proposal of grammaticality in terms of combination of such categories as noun, verb, predicate phrase. This much can be stated with confidence, even though the matter of exactly what the categories (“ onoma”, “ rhema”) were for Plato is not beyond controversy. The key point is that Plato gives here a characterization of a sentence in purely grammatical terms. He goes on to say that once we have this characterization of sentences, further conditions — which turn out to be what we would call semantic conditions — determine what it is for a sentence to be true and how such a sentence differs from false ones. 1.3.2. Remarks on Plato’s Characterization of Grammar We must not overestimate the significance of Plato’s remarks. Plato does not give a grammar for Greek, or even for a reasonable fragment of Greek. He only gives examples of the categories and rules of combination that would go into the construction of a grammar. This can be explained by pointing out that the whole context for the discussion is the refutation of Parmenidean monism, i. e. the view according to which there can be only

XXII. Syntax und Nachbardisziplinen

one real entity; any attempt to interpret reality as a plurality of intelligible entities will end up, according to this view, as paradoxical. Plato’s concern is not with language in its own right, but only with features of language attention to which will help defeat the monist. The remarks on language are only instrumental to a metaphysical enterprise. 1.4. Aristotle on Language In contrast to Plato, Aristotle held views about language that show him interested in natural languages as natural phenomena and not merely as sources of support for metaphysical theses. In the first chapter of DE INTERPRETATIONE, 16 a3—8, Aristotle gives a general characterization of certain relations between spoken and written language, language and thought, thought and reality. As we shall see, the amazing thing about this statement is that it sets out most of the fundamental questions that occupy linguistics and the philosophy of language for the subsequent 2400 years. 1.4.1. The Statement in DE INTERPRETATIONE, Chpt. 1., Concerning Language In this passage Aristotle makes the following statement about language: “ Spoken sounds are symbols of affections in the soul and written marks are symbols of spoken sounds. And just as written marks are not the same for all humans, neither are spoken sounds. But the affections of the soul, of which the spoken sounds are the signs, remain the same for all, and so do the elements of reality of which the affections of the souls are likenesses.” We see here Aristotle distinguishing between two, or two kinds of, relations. The first kind of relation is the “ symbolizing” relation. This one, R′, holds between written and spoken elements of language, and also between spoken language and elements of the mind, or more literally, “ affections of the soul.” R″ is another, or another kind of, relation. It holds between certain elements of the mind and certain elements of reality. In connection with these relationships, Aristotle makes some claims concerning what is universal and what varies across languages. 1.4.1.1. Interpretation R″, holding between mind probably roughly equivalent tions of conceptualizing or world. Aristotle thinks that

and reality, is to modern nointerpreting the R″ involves the

82.  Syntax and Philosophy

notion of likeness and obtains when reality impinges on the mind. The two aspects of this relation are compatible within the Aristotelian framework. We know from other writings of his that the same relationship can include both likeness and causal effect. This interpretation is the same in all humans, and for Aristotle it turns out to be a basic irreducible notion. The other relation, R′, seems to be the relation of language to understanding, except for the curious proviso that it is analogous to the relation between written and spoken linguistic elements. The oddness can be ameliorated if one thinks of language understanding as a representation of some kind, for this would construe it as analogous to the representation that we find between spoken and written sounds. Apparently, for Aristotle, the different sound systems realized by different languages map onto a basically universal human conceptual scheme. Language understanding and the representational relation between written and spoken elements remain for Aristotle basic primitive notions. Language understanding results in the ability to express thoughts in language and to understand what is expressed in language. It might be misleading for some to talk about Aristotle’s suggestions as representational, for in modern parlance to say that x represented y, is to assume that x ≠ y. But, as the discussions in DE ANIMA Bk. III show, this assumption is not held by Aristotle. 1.4.1.2. Philosophical Comments As we reflect on the two relations and what Aristotle says about these, we see unfolding the deep questions that occupied the study of language since Aristotle into our own times. In some sciences the basic questions change, while in others they seem to remain the same. The text under consideration shows the study of language to belong to the second class. One could look at much of the study of language, with exceptions like 19th century historical work, as a series of attempts to get clearer on the relations R′ and R″ that Aristotle posited. It is also worth noting that both of these relations remained for us basic primitive notions. One can also construe some schools of thought on language as disagreements with some of Aristotle’s characterizations. As we see, Aristotle posits a reality independent of the human mind and takes this to be the realm that we interpret. Kantians would say that this is to confuse phenomenal with noumenal realms. Aristotle

1465

thinks that the structure of the mind remains basically the same, regardless of linguistic variations. Whorfians think otherwise. Aristotle thinks that the structure of languages does not remain fixed across language families. Chomsky and others in his tradition think otherwise. Aristotle does not separate in his account what one might take as conceptual as opposed to empirical claims. We can see today that several of the components of the view that unfolds in this text are subject to empirical verification or disconfirmation. 1.4.2. Aristotle’s Notion of Designation In CATEGORIES chapter IV. Aristotle introduces a relation (“ semainein”) between some parts of language and some parts of reality. This relation is best labeled in modern English as “ designating”. According to Aristotle certain elements of language designate, or “ single out”, a number of items from the categories of substance, quality, quantity, relational property, etc. This is, then, a notion additional to the ones surveyed above. In addition to the language-mind and mind-reality relations we have also the language-reality relation. The elements of language that Aristotle seems to have in mind are what we call “ general terms”. They are not sentential units; hence the designation relation cannot be that of truth or falsehood. Furthermore, the examples given in the text mentioned are not proper names; hence it would be a mistake to construe the relation as that of naming. The best we can do, it seems, is to say that this is the relation of applicability, or of a general term being true of a certain set of entities. It is important to put it this way, since Aristotle’s examples are not predicate phrases; hence it would be a mistake to describe Aristotle’s view here as a view about predication. It turns out to be the case that designating is a primitive for Aristotle; so we have this, and language understanding, and conceptualizing the world, as key primitives that we have not eliminated to this very day. 1.4.2.1. Remarks on the Above A modern philosopher of language might be tempted to force Aristotle’s notion of designation into the conceptual straightjacket of having to be either naming or predication. Such forcing would be quite anachronistic. It would suggest that there can be no other important sub-sentential semantic relations except that of being predicated of something or

1466

being a name of something. There are, however, many other alternatives. The seemingly necessary dichotomy of naming and predicating has its origin not in all-pervasive features of natural languages, but in the technical apparatus of modern symbolic logic with its logical constants and combinations of variables and predicate letters. Needless to say, there are other frameworks for the analysis of semantic structure besides modern symbolic logic. In any case, Aristotle does not accept naming/predicating as a fundamental dichotomy. His notion of designating is the relation of applicability between general terms and elements of reality, sorted into fundamental categories. One could, though Aristotle did not, take this as the most fundamental notion and explain both naming and predicating as notions derived from designation. In any case, the importance for syntax is that given Aristotle’s basic notion, we can talk of the range of application of words like runs, tiger, etc. and not just names like John and predicate expressions like is a human or was a student. 1.4.3. Aristotle on the “ Parts of Speech” We see, then, that nouns and verbs can be important semantic units for Aristotle. The same is true for his conception of grammar. We saw the notion of a grammatical category emerging in Plato. Aristotle develops this notion further. The key grammatical categories become the famous “ parts of speech”. Plato’s meager list is expanded considerably by Aristotle in his RHETORIC and other works, a fact discussed by historians (Steinthal 1863/ 1961, 261—271). It would be senseless to ask if Aristotle would define these “ parts” in terms of purely syntactic features or would also use semantic characterizations. The contrast between semantic and syntactic theories was not yet consciously formulated at this stage. It is important, however, to note that the list concerns primarily classifications of expressions and that the criterion is to come up with a list of “ parts” or categories that would explain sentence construction. In making this a topic of importance by itself, Aristotle paved the way for Hellenistic grammar. 1.5. Hellenistic Grammar Between the second century B. C. and the second century A. D. we see the emergence of serious work on Greek grammar. The details of this development have been ably chronicled by historians earlier as well as in recent

XXII. Syntax und Nachbardisziplinen

times (Steinthal, Blank). We shall concentrate on a few issues in connection with which we can see the interaction between philosophers and grammarians. We shall canvass three issues. First, the relationship between the two leading schools. Second, the data utilized. Third, epistemological implications as seen in this period. The aim of these sketchy remarks is to show that the key issues emerging in this period are identical with those that became the focus in the emergence of generative grammar. 1.5.1. Two Schools or Two Approaches? While in Aristotle we find lists of “ parts of speech”, in the period under discussion we find attempts at formulating rules to build syntactic complexes out of simpler elements. We shall not only ignore these details but also the issue whether we should see Hellenistic grammar as a fight between two schools or a development through two stages (Steinthal, Blank). One alleged school is the earlier Stoic school, and the other that which is composed primarily of Alexandrian grammarians. What is of interest to us is the way in which the two schools — or stages — were characterized in ancient as well as modern times. The Stoic grammarians are said to be practicing “ philosophical” grammar. This turns out to mean that their main interest was in characterizing what the meanings of words are, and in relations of logic to language. They thought that once these matters were settled, there was no more room for rule and theory. That is to say, they regarded the particular ways in which different languages developed apart from assignments of meaning and the embodiment of logic as matters of accident. Thus these thinkers differ from Aristotle. For while he thought that linguistic form varies across language, the Stoics seem to have speculated about certain general ways in which logic and the assignment of meanings to constituents are encoded in languages. We see here perhaps the beginnings of a prejudice that runs deep in philosophy; namely the hunch that once we have settled semantic and logical issues, the “ purely syntactic” matters in natural languages are not worthy of serious theorizing and probably would not provide suitable material for such in any case. We do find, however, either as a contrast or as a further development of this, the work of the Alexandrians, which involved the assumption that grammatical rules are not matters of accident; one can give general ac-

82.  Syntax and Philosophy

counts of these, even apart from reference to matters of meaning and logic. As we shall see, this contrast of a view of grammar according to which insofar as it is intelligible and subject to rules and explanations, it must be based on semantics, with the view according to which pure syntax admits of laws and generalizations in its own right emerges once more in the twentieth century, in particular with the rise of generative grammar. According to one view, syntax is partly a matter of accident and partly abased on semantic considerations. According to the other view, syntax has a life of its own. 1.5.2. Data for Grammar In contemporary work on grammar we typically see as evidence sentences or pseudo-sentences that are submitted to the judgments of current speakers. Thus it is worth mentioning that the ancient grammarians considered a much wider range of data. For example, much of the interest centered on preserving the structure of classical Greek found in texts such as the Homeric compositions (Steinthal 1863/1961, 68). The resulting grammar is, then, partially normative in character. It captures usage that is good and exemplary in an objective sense. For it preserves that which users of the language consider linguistic usage at its best. The normative nature of a good grammar has been misunderstood in recent times. It is confused with arbitrary legislation. We shall return to this point later in this survey. 1.5.3. Epistemology Two views on knowledge of rules of grammar can be distinguished in this period (Blank). Blank calls one of these the empiricist view and the other the rationalist view, with Apollonius Dyscolos represented as belonging to the latter. The “ empiricists” thought that in the case of the study of language all we can do is observe data and regularities, with no further explanations possible. The “ rationalists” thought that one could summarize the grammar of a language as a system of rules and that these rules could admit of further more general explanations. It is not clear from the evidence available whether the dichotomies between rationalists and empiricists on the one hand and that between philosophical grammarians and the Alexandrians were seen as distinct. It is important, however, for our purposes, to realize that the distinctions are orthogonal to each other. All

1467

four of the possible combinations of views offer coherent stances, even though there might not have been a historical figure to represent each of these possibilities. Once more we see both in terms of the issues emerging and in terms of the possibilities ignored, a conceptual situation at the dawn of the discipline that is mirrored strikingly during the past thirty years. 1.6. Summary This part of our survey sketched the rise of systematic concern with different aspects of language from philosophical speculations. It is surprising to find that both in the initial statements of the key structures of relations between language, mind, and reality, as well as with regard to the early differences on what were taken as key issues, the earliest period reflects the debates and concerns that characterize philosophers’ concern with theories of language over the past thirty years. We see Aristotle sketching certain fundamental relations between language and mind, mind and reality, and again language and reality; the elucidation of these relations is still today the main task of interactions between philosophers and grammarians. We see in the Hellenistic period the emergence of concern with meaning, and a contrast between conceptions of pure grammar, and the interpretation of linguistic structure as based on logic and semantics. We also see contrasts in views concerning the nature and scope of explanations in linguistics. We seen, then, that the key questions have not been fully answered, nor replaced by more fundamental ones. Furthermore, the key primitives remained the same. One can only speculate about the reasons for these facts. Perhaps any really fundamental breakthrough would require new investigative techniques and technology, or we might be running up here against one of the permanent limitations of the study of humans by humans. 2.1. Syntax and the Tradition of Analytic Philosophy We shall not define “ analytic philosophy”. It will be taken to be, roughly, the family of traditions that have their origin in the thoughts of Frege and Russell and still dominate most of the philosophy of language done in AngloAmerican philosophy. These traditions involve preoccupation with the rise of symbolic logic, and the use of logic in clarifying philosophical issues. They also involve using the

XXII. Syntax und Nachbardisziplinen

1468

new technical framework of logic to analyze parts of natural languages and the structure of scientific theories. We shall review briefly the history of the interest of analytic philosophers in syntax in order to place into perspective the interactions of the past 30 years. 2.2. Motivation We saw that Plato’s interest in language was primarily instrumental to his metaphysical interests while Aristotle reflected more on language as a natural phenomenon. As we survey analytical philosophy we see that its interest in language, at least in the early phases, was also primarily instrumental. The interest in language was motivated partly by an attempt to link the allegedly most fundamental structures of language to the basic structures of reality, and the attempts to investigate the ways in which the structures generated by modern symbolic logic are embedded in natural languages. 2.2.1. Language and Reality There are several examples of analytic metaphysical works in which interest in language is instrumental to attempts to link certain linguistics structures to metaphysical conceptions of reality (Russell 1918, Wittgenstein 1922). For example, Russell is looking for analogues to the notion of individual constant in symbolic logic within natural languages. These are the so-called logically proper names. He then links his candidates, e. g. demonstratives, with what he considers metaphysical atoms, and develops on the basis of this an ontology. Wittgenstein, too, looked for certain basic simple entities in his metaphysical construction, and his interest in language was merely instrumental to this ontological search. In these enterprises, then, we can detect no genuine interest in what one would ordinarily call the grammars of natural languages. In fact, everyday grammatical constructions were seen as hindrances in the attempt to detect in language ontologically basic structures. 2.2.2. Logic and Language The disdain for grammar in the ordinary sense and the claim that grammatical structure is misleading is seen as even more pronounced in efforts to link modern quantificational logic to structures of languages like English. To appreciate the motivation for these claims one should look at detailed com-

parisons between the syntax of natural languages and the syntax of first-order predicate calculus (Quine 1970, 22—30). Within the context of this survey we shall restrict ourselves to a few remarks. First, there is no straightforward correlation between the syntactic categories of symbolic logic (individual constant, variable, predicate letter, etc.) and the grammatical categories of natural language such as noun phrase, verb phrase, etc. Secondly, in symbolic logic the purely referential function and the descriptive functions are syntactically separated; the logical and individual constants, variables, and quantifiers bear the burden of the former while predicate letters and compounds constructed partly with these carry the onus of the latter. We can find no such linguistic division in the grammars of natural languages. The interpretation of a language of logic will assign to individual constants elements of the domain over which the language is interpreted. Thus both syntactically and semantically individual constants are parts of languages of logic. It has been thought that the proper names such as those for persons, streets, etc. that are associated with natural languages, are the analogues within natural language to the individual constants of logic. This claim is, however, questionable. For one thing, it is part of knowing a language of logic to know what the individual constants pick out, but it is certainly not part of knowing, e. g., English that one should know the bearers of all of the proper names assigned to speakers of English. We need to know the syntactic role of proper names, i. e. that they make up a species of noun phrase, and we need to know their semantic role, namely singling out entities. But we do not need to know the referents of all of the names. In that sense proper names for certain entities are not parts of natural languages. These differences show the difficulties of “ translating” sentences of English into formulae of symbolic logic. They also give us clues as to why some philosophers thought that the grammar of natural languages was an impediment to uncovering the “ real” logical structure of language. 2.2.3. Summary The above shows that the syntax of natural languages is at this stage not an intrinsic object of interest for philosophers. Furthermore, none of the philosophical work of this era suggested that one can get precise about the purely syntactic features of natural lan-

82.  Syntax and Philosophy

guages. Russell and others felt that one has to “ cut through” ordinary grammar in order to uncover the important structures such as those of quantifiers and definite descriptions in languages like English. 2.3. The Concepts of an Artificial Language and a Formal Language The felt discrepancy between logic and natural languages led to the enterprise of constructing artificial languages, primarily based on the model of logic, in order to illuminate the nature of scientific theories and in the hope that formulating matters in this way would help to solve philosophical problems. The enterprise was aided considerably with the invention of a technical notion of “ formal language” by Alfred Tarski. We shall review these developments solely from the point of view of their implications for philosophers’ concern with syntax. 2.3.1. Syntactic Definitions in Logic and Nominalism In logical theory to say that a notion is defined semantically is to say that the phrase capturing the notion is assigned an abstract object; i. e. a part of what is taken to be reality. Thus semantic definitions carry implications as to what there is in the world. To say that a notion or phrase is defined syntactically is to say that it can be reduced to other expressions or explained in terms of the use of other expressions, without any semantic commitment, i. e. posit of what is or is not real. Hence nominalists who wanted to eliminate any entity that is not a particular — nonrecurrent spatio-temporal or merely temporal — from their ontology attempted to make maximum use of syntactic definitions. This is characteristic of Hilbert’s program for the foundations of mathematics or Quine’s characterization of the notion of logical truth (Quine 1970, 58). 2.3.2. The Notion of a “ Formal Language” The phrase “ formal” is used by linguists and philosophers in a number of ways. Here we shall restrict ourselves to the consideration of a strictly defined technical notion of what a formal language is. This notion was formulated by Tarski (Tarski 1936). Tarski’s aim was to construct languages that would be able to deal with semantic paradoxes such as the liar paradox, and to be suitable as the languages within which one could formulate precisely the body of knowledge contained in

1469

a scientific theory. We shall single out those constituents in the account of a formal language that have special significance of the uses that philosophers and linguists subsequently made of this notion. The syntax of such a language must contain an enumeration of all of the minimal significant elements, and rigorous rules of composition that define the class of well-formed sequences. As we shall see, in this respect the syntax of Tarski’s formal languages resembles the syntax that emerged from the later “ generative” tradition. A key difference is, however, that in a formal language the sole motivation for introducing syntactic elements and structures is the facilitation of expressing semantic content. The semantics of a formal language is given by positing a certain domain of entities in relation to which the language is interpreted, and then assigning elements and functions ranging over various subsets of these elements to various meaning bearing complexes and simples (for a good introductory survey see Rogers 1963). Tarski himself thought that this notion of a formal language was not applicable to natural languages. For one thing, within a formal language thus specified one can formulate hierarchies of languages and with the help of such structures escape the paradoxes. For another, the interpretations of such languages are defined over certain domains of entities, a condition naturally satisfiable when one constructs a language to express a scientific theory, but not satisfiable when one considers the indefinite “ domain” over which expressions of a natural language range. 2.3.2.1. The Syntax of Formal Languages We have already seen two features of this syntax: the characterization over a fixed set of expressions, and the purely semantic motivation. The latter point can be spelled out further by noting that in a formal language there will be an isomorphism between syntactic and semantic structures. Furthermore, reflection on the grammar of logic suggests that in a formal language the syntax will be rather simple in structure, while the information content will be rich. It is far from clear that natural languages have similar structures. 2.4. The Rise of “ Ordinary Language Philosophy” and its View of Grammar The history of the rise of so-called ordinary language philosophy has been well chronicled (Urmson 1956). Our concern here will

XXII. Syntax und Nachbardisziplinen

1470

be with its view of the grammars of natural languages. Ordinary language philosophy emerged partly as a reaction against the claim of analytic philosophers that ordinary language is confused or unclear. In reply it was pointed out that ordinary language has a variety of uses other than encoding scientific theories. If we pay close attention to these uses and what they demand of language, then we will understand the structure of natural languages and will discern order where the myopic formal logician saw only chaos. For example, if we see reference not primarily as a relation defined between logical constants — or their natural language analogues — and a part of reality, but rather as a human activity with a variety of purposes, then we will be able to make much better sense of the referring uses and devices that a language like English make available (Strawson 1950). Within this view, one can attempt to see the grammars of natural languages as generating structures with multiple purposes and potentialities for employment. 2.4.1. Natural Language as an Object of Interest in its Own Right The shift from viewing language as containing possible clues for metaphysics or as a vehicle for science to a consideration of language use as a human activity incorporating a variety of functions and purposes, led to an appreciation of the complexity of natural languages and their potential as objects of study in themselves. We see in this period the study of many types of locutions that are not related to semantics as the formal logician conceives of this notion (e. g. Urmson 1952). This shift also brought with it a different view of syntax. Grammatical structures of natural languages are now viewed as giving clues for the reconstruction of the varieties of uses for which a language like English is suited. 2.4.2. Limitations of Interest in Syntax by the School Natural languages are used to do other things besides making assertions. They can be used to question, explain, warn, suggest, advise, etc. This fact led philosophers to search for syntactic devices with which different uses can be correlated. These attempts were only partly successful. One can ask questions in other ways beside the use of interrogatives, and one can issue a command in ways other than the imperative. It may or may not be the case that for each grammatical construc-

tion or peculiarity we can give an explanation in terms of the functions that languages serve, but it is certainly not the case that for each function and use we have a separate grammatical form. Once this was seen clearly, philosophers of this school had their interest in grammar dampened. This shows, then, that even the ordinary language school of philosophy did not have a direct interest in grammar as such. 2.4.3. Speech Act Theories One can look at the ordinary language school as the flourishing of great philosophical “ naturalists” of English, e. g. Austin, Urmson, Ryle; this label is meant to contrast with “ biologists”. We need both naturalists who are marvelous at fact finding, and biologists who shift out the facts and shape the relevant ones into material for a theory with explanatory power. We shall consider the rise of the “ biologists” shortly. But we need to mention first a transition stage, marked by strong interest in classifying all of the different ways in which the use of language can enter into acts of different types that humans perform. This effort led to so-called speech act theories (Alston 1964 , Searle 1969). These theories at the least present a systematic account of the variety of speech acts we can perform by the use of English; in their more ambitious form they present a theory of meaning which is meant either as an alternative to or a supplement of the kind of semantics the logicians provide. Even within this theory syntax is not viewed as “ having a life of its own.” 2.5. Summary Formal semantics developed a view of syntax that in terms of its rigorous abstract structure foreshadowed generative grammar. It was, however, insisting that syntax has to be “ the slave of semantics” and conceived of semantics in terms of rigorous formal interpretation. Ordinary language philosophy came to view natural language as a fascinating human artifact, providing for a variety of uses that correspond to a variety of essential human activities, needed for any viable rational human form of life. Grammar within this tradition meant not formal syntax but the traditional grammar taught in school. Though there was some interest in grammar, this interest was limited by the implicit functionalist view of grammar that philosophers in this tradition and in the tradition of the ensuing speech-act theories held.

82.  Syntax and Philosophy

3.1. The Unification of Formal Semantics and Generative Grammar By 1966 we find a split picture of philosophy of language in Anglo-American philosophy. On the one hand, the construction of artificial languages and the use of Tarski’s ideas about formal language goes on without much interest in the peculiarities of natural languages. On the other hand, there is much good work on describing a number of specific facts about English; facts that from the point of view of symbolic logic or formal semantics might not be significant, but that point to ways in which the use of language serves the demands of human functioning. This work proceeds without any interest in formal theories. At this point philosophers begin to take notice of the rise of generative grammar. This view held that the main task of syntax was to consider the class of all possible combinations of the minimal meaning-bearing elements and construct a function that would delimit a subclass of this collection, i. e. the class of well-formed complexes. This formal specification of syntax resembles that given by Tarski’s theory. The syntax of generative grammar, however, takes the traditional grammatical categories as its building blocks, and has as its aim the explanation of all of the facts of grammaticality, regardless of whether there is or is not a semantic or functional explanation of these (Chomsky 1965). Philosophers were amazed that one could become so precise about the ordinary grammar of natural languages. Generative grammar was seen by some as presenting a challenge. Could one take a natural language like English on its own terms, with its traditional syntactic categories, and formulate a precise syntax and semantics that would explain particular facts of meaning and grammaticality? Montague’s theory was designed to meet this challenge (Montague 1970). 3.1.1. Some Aspects of Montague’s Theory Both Chomsky and Montague conceive of generative grammar as a formally rigorous way of generating all and only the wellformed formulae of a language like English. They opt, however, for different formalisms. Montague opts for a categorial grammar because he wants a framework that is so general and abstract that it will accommodate the grammars of all languages, natural or artificial. Chomsky opts for a formalism that will,

1471

he hopes, bring out conspicuously what is unique about human languages. Chomsky’s syntactic categories are those of traditional grammar, while Montague has to find indirect ways to capture those categories and at the same time have a grammar in which, like in Tarski’s formal languages, he can assign a certain type of abstract object as the “ meaning” of a given syntactic category and all that falls under it. The meaning of an expression in Montague’s framework will be a function that will pick out all and only those actual and possible entities that are within the range of application of the term. In simplified form, for example, the meaning of wise would be a function that would pick out all actual and possible wise things such as humans, actions, or institutions, and the meaning of walk would single out all those who walk. The semantic rules of composition are isomorphic to the syntactic generative rules, and yield the right semantic assignments to all sentences and their sub-parts. In this framework two different syntactic categories can have semantic objects of the same type as their “ meanings”, but one cannot have elements with different types of semantic objects as their meaning be members of the same syntactic category. 3.1.2. What should Motivate the Positing of Syntactic Structures? Some philosophers, influenced by logic, think of grammar as tailor-made for the generation of semantic complexes. For example, the combination of the adjective red with the noun apple should yield the complex red apple which has the intersection of the red things and apples in its range, and the combination of walk with slowly should yield a function that will single out the slow walkers. Another view was that which emerged from ordinary language philosophy. It said that we should look for some account or facilitation of language use behind each grammatical peculiarity. Phrases like hereby, and so-called “ parenthetical” constructions as in Mr. Jones, I believe, serve specific needs of activities involving the use of language. One could, however, consider a third view. According to this conception some syntactic structures and generalizations cannot be justified by either reference to semantics or reference to varieties of language use. On this view syntax has a “ life of its own”; its primary purpose is to account for grammatical evidence and to exhibit what is unique about human languages from the syntactic point of view. Philo-

1472

sophers have been slow to come to understand this conception, and even today there is a widespread skeptical attitude toward it. 3.2. Relation of Semantics and Syntax in Natural and Formal Languages We saw that within the framework of formal languages one can posit a hierarchy that will resolve the semantic paradoxes. If someone is to apply the same technical framework to natural languages, the burden is on this person to show how the hierarchies can be built into the semantics and syntax of this collection of languages. Furthermore, formal languages are individuated, for technical reasons, partly in terms of their vocabulary, including all of the constants. But we could hardly say that English has become another language just because a new name has been added by the language users or a previously used name — like Miller — has gained new application. If someone were to propose that names for persons, countries, etc. are not parts of a language like English, then we have another important difference between natural languages and formal languages. We need also to consider other constraints that will govern the syntax of a natural language and which do not enter into Montague’s considerations. One of these is learnability. In the case of formal languages this problem does not arise since we invent these systems. But in the case of natural languages it is imperative that the syntax chosen should be suitable for presenting the language as learnable by a human under normal circumstances. 3.2.1. Replies In reply one can say a number of things in defense of those who wish to carry on with Montague’s project. As to the paradoxes, it is an open empirical question whether one could detect in the structures of natural languages hierarchies of the sort Tarski’s conception needs, or whether one could find another way out of the paradoxes. With regard to the vocabulary one could claim that Montague’s system applies to English under certain idealizations, e. g. “ being frozen at an instance.” And finally, one could suggest that conditions like learnability could be added as further constraints of the class of Montague grammars that one might generate as adequate systems for natural languages. 3.2.2. Other Constraints Learnability is not the only constraint on an adequate syntax for a language like English.

XXII. Syntax und Nachbardisziplinen

There is also the matter of being a suitable structure to be related to phonological structures, and the condition that the syntax must be such as to answer the needs of the human processing mechanisms, whatever these may be. Human language processing is a biological process, hence one would expect that human biology places additional constraints of the permissible structures. This is not, then, a case of the usual scientific procedure of a description under idealization with other variables added. The additional constraints might force us to choose a framework for syntax and semantics that violates the conditions defining a “ formal language.” 3.3. An Alternative View In what follows we shall sketch an alternative to the view developed by Montague and his co-workers. This view takes as its main goal the characterization of what is unique to human languages, both in terms of syntax and semantics, and it leaves the issue of the relation between natural languages and formal languages in the sense defined to be an open empirical matter. It also considers the syntax of natural languages to be an abstract object of interest in its own right, regardless of any connections between it and semantic or functional considerations. 3.3.1. Syntax and Complexity One partial characterization of the grammar of natural languages is given in terms of formal complexity measures. The way to reach such a characterization was shown by some of Chomsky’s early work (Chomsky 1963). In this work a hierarchy of types of grammars and automata processing these grammars was developed. The hierarchy was arranged according to formally definable levels of complexity. This framework shows that if the grammars of natural languages are on a level of complexity, say F, then no automata on a level of complexity less than F can process fully natural languages. What is philosophically interesting in this work is not necessarily the specific hierarchy posited; recent and ongoing research has posited others as well. The crucial matter for philosophers is the logical form of the argument. From precise levels of complexity of types of grammars to levels of complextiy for processing agents. 3.3.2. Syntactic Structures According to the view under consideration syntactic structures are posited primarily on the basis of pure syntactic evidence (Chom-

82.  Syntax and Philosophy

sky 1957). Semantic considerations and other constraints influence the choices only when the syntactic evidence is neutral between several alternatives. In explaining and justifying the posited syntactic structures it is assumed that syntax “ has a life of its own”; that is to say, just as in the case of explaining anatomical structures or the structures of atoms, we reach a level of explanation that is beyond anything functional, purposive, or instrumental to some other set of considerations, so we reach such a level in the case of syntax. 3.3.3. Limits on Functional Explanations We are used to supplementing causal and structural explanations with functional explanations in biology. When considering the human organs such as the eye or the brain, we not only want to know what their constituents are, but also how the constituency helps with the functional demands such as seeing, thinking, etc. But in each of these cases we reach limits to such accounts. Nobody claims that we can find functional reasons for humans having the structure they have down to the last muscle or bone. One could say that “ anatomical structures have a life of their own”, to parallel our slogan about syntax, and to show that this view of syntax is analogous to the way science looks at structure in any of the other well-developed domains. Furthermore, the “ purely syntactic” is not only a part of that set of structures that are language specific. Again, parallel to physics or biology, we can posit syntactic universals that will have this function-independent status. The alternative view, then, is not an idiosyncratic view of science, but rather a way to view language as one views any other natural phenomenon. 3.3.4. Autonomy of Syntax The view outlined above can be described as a version of the “ autonomy of syntax” thesis. Since slogans and labels tend to be misleading, we should distinguish different ways in which one might view syntax as autonomous. An extreme view would construe syntax as completely independent of semantic and other functional considerations. The view adopted above says only that syntactic structure has some degree of independence and autonomy. This is quite sufficient for interpreting natural languages as not being “ formal languages” in Tarski’s sense (Moravcsik 1983). This conception also has implications for language processing, for it proposes that

1473

the processing is either a matter of a number of parallel processes or a matter of one type of process building on the results of the other, with various interactions allowed (Moravcsik 1983). 3.3.5. Syntax and Conceptions of Processing In approaching the question of processing we should distinguish two questions: (a) what does one know about syntax insofar as one knows a language like English? (b) what knowledge do I make use of when I understand a part of English in everyday communication? There is no a priori reason why the answers to the two questions could not be identical, but analogies with other types of cognitive processing make that very unlikely. While our understanding of elementary arithmetic involves a grasp of commutativity, we do not use this principle in everyday activities such as the adding up of a grocery bill. It is one thing to characterize the understanding on which we can fall back, and another the “ short-cuts” we use in everyday applied arithmetic. Language processing is likely to be an analogous process, with a pool of material mastered underlying everyday routines and shortcuts used in person-to-person communication. The shortcuts are facilitated by other parameters such as knowledge of social and psychological context. It is an open empirical question how much of the underlying abstract structure we utilize in everyday communication, and what all of the intervening factors are. It is clear, however, that we do have a large body of knowledge and conceptualization to rely on, for it is only on the hypothesis of our possessing this pool that we can plausibly account for intuitions of grammaticality. We shall take up the role of intuitions in more detail later on. 3.3.6. Syntax within the “ Alternative View” The syntax within this conception is not tied to the conditions of “ formal” languages. Thus it might be that items in what is by any reasonable syntactic criterion a single grammatical category have in fact different kinds of semantic objects as their meanings. Furthermore, the syntax envisaged by the alternative view is compatible with a semantics that does not generate determinate extensions and corresponding functions as the semantic interpretation of meaning bearing elements, but only partial functions as explanatory schemata that are the meanings of vari-

1474

ous expressions (Moravcsik 1987). Also, semantic interpretations of natural languages may be possible at any given time for only a certain set of projections of the current states of affairs over time and possibility, and not over all possible “ worlds” or states of affairs. The syntax is the firm foundation that gives stability to a natural language even over periods of drastic semantic change. It also provides for places for proper names in sentences, regardless of whether we will or will not consider these elements as semantically parts of a natural language. 3.3.6.1. The Role of Formal Semantics within this Framework Given the differences that this view posits between natural and formal languages, one might think that Tarski’s methodology for semantics might not have a role in this framework. There is, however, room for formal semantics within the alternative view. For one could construct a formal semantics for a selected fragment of English in order to demonstrate an important property of the meaning system of this language. For example, one might want to investigate the complexity of various quantificational structures of English so that one could draw conclusions about the complexity of the semantics of this language (Barwise 1979). Within such a project one could make use of the techniques of formal semantics. One could also investigate sequences of temporal references in English sentences and their dependencies, or various constructions that might force us to conclude that the semantics of a language like English requires for its characterization higher order quantification. Projects of this sort could be carried out without running into conflict with the general view about syntax of a whole natural language that is delineated by the alternative view. 3.4. The Prescriptive-Normative Nature of Grammar All of the conceptions of syntax canvassed so far assumed that the syntax of a natural language is a system of rules. Some philosophers have raised general skeptical questions about rule-following. Such general skepticism is not a specific problem for the characterization of syntax, and hence will not be treated here. There is, however, a special worry about the posited syntactic rule systems that we shall consider now. This concerns the descriptive and prescriptive status of syntax.

XXII. Syntax und Nachbardisziplinen

3.4.1. A Dilemma and its Resolution As one considers whether syntax is prescriptive or descriptive, the following dilemma presents itself. If syntax is purely descriptive, then it represents how we actually speak. Likewise, if logic and arithmetic are purely descriptive, then these disciplines represent how we actually reason and calculate. But theories representing how we actually reason and calculate must be trivial. For we already know how we actually reason and calculate: badly. Likewise, a theory that represents how we actually speak must be trivial, for our actual speech contains many mistakes. On the other hand, if these disciplines are prescriptive, then we need to give an account of the source of their authority. Are all such prescriptions arbitrary? Are normative questions of grammar separable from questions of style and form? It does seem that we can distinguish acceptable English from stylistically outstanding English. We can show this reasoning to be presenting a pseudo-dilemma by explaining how logic, mathematics, and syntax are both descriptive and prescriptive. The facts that we correct ourselves and have intuitions about correctness and incorrectness in these domains shows already that we conceive of ourselves as creatures who try to follow rules and live up to certain standards. Thus what the linguist should want to know is not merely what we do but what we in fact try to do and realize. In this sense linguistics recreates what is both descriptive and normative. In the neighboring disciplines of logic and mathematics we find a more complex situation. These disciplines present us with a double aspect of normativity. To be sure, they also contain the normativity we saw in connection with linguistics. For again, what we want to know is not just how humans reasons and calculate but how they think they should reason and calculate. But in addition to this, in logic and mathematics we have factors independent of our intuitions, such as consistency and arithmetic laws, to justify what is correct. Apart, however, from one sense in which logic and mathematics are normative, we can uncover a normativity that they share with linguistics. Hence we can say that what needs capturing is the system of internalized rules that the competent speaker of a language possesses; this is a system of rules that humans at their best want to follow. Our descriptions are descriptive, but what they are

82.  Syntax and Philosophy

to describe is a human value system. The scriptions of linguistic competence prescriptive in the sense that they are mere generalizations of how we in fact have.

1475

deare not be-

3.4.2. Syntax and Two Kinds of Rules One might object to the above by suggesting that we have drawn too close an analogy between grammar and logic. There are, after all, purely conventional rules such as those of etiquette or those governing on which side of the road we should drive. These are to a large extent arbitrary and thus do not resemble the rules of logic and mathematics. Why should not the rules of grammar be more like the rules of etiquette and less like those of mathematics? To be sure, in order to manage well, we must have rules about driving and about grammar, but perhaps all of these are arbitrary in their actual structure. One can reply to this point in a number of ways, depending on the view that one has of grammar. If one has a thoroughgoing functionalist and instrumentalist view of grammar, then one can justify the close analogy with mathematics that we posited. One could say then that the mathematics we use best serves our needs to calculate, theorize about the environment, etc., and the grammars of natural languages best serve the requirements of facilitating semantic interpretation, learnability, the needs of the processing mechanisms, etc. We saw, however, that a fully functionalist view is not tenable. We can, however, defend the analogy even within the “ alternative view”. The analogy with rules about driving and etiquette is bad in any case, since these rules develop within explicit institutional contexts, while this is not true of most syntactic distinctions. The claim that “ syntax has a life of its own” can be coupled with the view that the grammars of all natural languages have a certain common core, and that this core is represented in the minds of all humans. Thus the core underlying the syntax of all natural languages represents a unique human capacity to process linguistic units in certain ways. The task of the linguist is to come close to whatever the right system of rules is that reflects the biological foundation of human language use and understanding. Thus the choices are ultimately not arbitrary. Needless to say, here as everywhere else, the principle of underdetermination applies. That is to say, compatible with any set of empirical observations we find more than one theory with

equal explanatory power. Thus the task of the linguist is better represented by saying that he wants a partial order among possible grammars rather than that he is aiming at one unique set of grammatical rules that would capture the common core underlying the grammars of natural languages. Furthermore, in this whole discussion we left the issue of the ontology of the disciplines under discussion an open question. One might want to be Platonist with regard to the foundation of mathematics, or that of linguistics, or neither. Those are important choices, but they do not affect the issue of arbitrariness and that of the dual nature of these disciplines as being both descriptive and prescriptive. 3.4.3. Grammar and Freedom The notion of a natural language defined in terms of a system of rules with prescriptive force might lead one to think — as schoolchildren often do — that grammar is a constraint and restriction on the expression of human thought. There are, however, philosophical considerations that show exactly the opposite. Syntax, interpreted the way it has been in this essay, is not a restriction but one of the enabling conditions of intellectual human freedom. Humans are thinking and feeling beings. The unique complexity of their thoughts and feelings is reflected in the rules of natural language and in particular in grammar. Mastery of syntax is an enabling condition; it frees us to realize our human potential as syntactically thinking beings (Moravcsik 1977). There can be no thinking of any reasonable complexity and no language without rules. Logic and grammar conspire to make it possible for us to articulate and express thoughts. Rules do not restrict thinking; rather, by creating coherence, consistency, and regularity, they make thinking and language possible. The rules of grammar are in this respect similar to the rules of chess. We could hardly complain that the rules of chess restrict chess playing. The rules of chess are constitutive rules; they define the game of chess. The rules generate the class of well-formed or acceptable moves in chess. No rules, no chess. In the same way, the rules of syntax define a language. Without the rules of syntax, how could we generate novel utterances with indefinite length? Mere habits and regularities do not have the projective power and character that is necessary for natural languages. Needless to say, not all rules are of this sort. To show that a rule is an enabling

1476

condition for human freedom, one must show it to be a constitutive rule for an activity that can be viewed as realizing or at least facilitating the realization of human potential, rather than repressing that. Creativity is one of the human potentials, and thus we need to look at the relation between it and the rules of syntax. Creativity in language use manifests itself on three levels. First, as Chomsky stressed, we continually process sentences that we never encountered before (Chomsky 1957, 30). This is one of the strengths of syntax. For the novel utterances gain their freshness in semantic content rather than in grammatical structure. Syntax is the skeleton on which we hang varieties of semantic meat, i. e. combinations of meaning-bearing elements. Another level of creativity is the phenomenon of language change, either in everyday or in technical and artistic contexts. Here, too, the change is mostly in the meaning of expressions, or in shifting lexical items from one lexical category into another. This happens when someone invents the verb parenting, or people start talking about gifting. Similar considerations apply to metaphor. In putting together complex noun phrases with metaphorical meaning — e. g. the hawk on fire — the syntactic category remains fixed, and it provides the frame for the juxtaposition of lexical items that spark the new interpretation. Third, a poem or other literary composition can occasionally introduce systematic violations of grammar. But even in these cases, the new impact is derived from the contrast between the new sequence and the background of standard grammar. Without that contrast the new sequence would not have the effect on us that it has. This happens, for example, when a poet does not put together a wellformed sentence, but substitutes for that a string of adjectives. What is conveyed by the new structure is partly a function of the fact that the elements are adjectives and that what is expressed is meant not to have the logical form of an assertion. We see, then, that creativity on all three levels thrives on syntax. Once more we can see syntax as an enabling condition rather than as a hindrance to originality. 3.5. Summary We surveyed two conceptions of the relation of syntax to semantics. Both start out with the view that the syntax of a natural language can be represented as a rigorous formal system. We saw, then, how one conception

XXII. Syntax und Nachbardisziplinen

emerges out of applications of Tarski’s work on formal semantics. Even though this work was not meant to study the structure of natural languages, and its philosophical motivation was not the concern with a language like English for its own sake, other schools of philosophy did begin to be interested in language as a natural phenomenon. Montague’s work and the tradition that it started can be viewed as an application of Tarski’s methodology with interest shifted to natural languages as abstract objects to be studied for their own sake. We saw also, however, that there are problems confronting the view that a natural language is a formal language, and thus we turned to an alternative view. According to this view, syntax, even at the deepest level, has its own autonomous rules, not derivable from either semantic or other functional considerations. This view of grammar is compatible with a number of views of semantics, including ones according to which sentences of a natural language do not have “ truth-conditions” or conditions of determinate applicability. We also defended the conception of syntax as simultaneously descriptive and prescriptive. This defense is compatible with both of the conceptions of syntax surveyed in the preceding sections. The argument presented was designed to show that syntax has this dual role in the same way in which logic and mathematics share it. These disciplines show us what a rational human agent thinks we should be doing when we reason, calculate, or process language. We then applied this result to the notion of creativity, and showed that the prescriptive nature of syntax does not make it a restriction on human freedom and creativity. On the contrary, we showed that it is an enabling condition for freedom and its various aspects. 4.1. Philosophical Problems Concerning Generative Syntax The basic conception of generative syntax represents the grammar of a natural language as an algorithm that computes a function of that language into the set of well-formed ones and the complement class. Philosophers might wonder why one should view grammar in this way, what problems such a view encounters, and what the alternatives are. In the sections to follow various problems related to generative grammar will be reviewed, and suggestions made how these problems might be treated.

82.  Syntax and Philosophy

4.2. Why View Grammar in this Way? One might be skeptical of this way of viewing grammar on the ground that empirical evidence open to public inspection would support the minimalist hypothesis according to which a language is just a set of human speech-patterns and regularities. It is a disposition to respond to stimuli in certain ways, and any construction of rule systems is just an abstraction, possibly a convenient one, to present the data and the underlying disposition in an insightful way. Grammar is on this view purely instrumental; its main function is to predict verbal behavior and response. 4.2.1. Defense of the Generative View Every scientific theory wants to explain empirical facts, thus this can hardly be the bone of contention. But from this trivial observation it does not follow that one wants to explain only the publicly observable. What is at issue is partly which facts grammar is to explain and how those facts are to be described. Verbal behavior is a function of many factors; our mastery of a language is only one of these. One might, then, want to sort out the different factors influencing verbal behavior, and concentrate on one of these, under idealization. Given this view, the mastery of a language is the basic fact to be explained; and this, in turn, is divided into the mastery of phonological, grammatical, and semantic aspects of language. The mastery includes the internalization of rules and the development of intuitions of rule violations. Generative grammar takes the having of intuition to be a fact indirectly verifiable, and posits the internalization of a rule system to explain the having of intuitions of grammaticality. Thus the methodology is analogous to the one we would adopt to discuss such competencies as that of logic and mathematics. To take the regularities or dispositions to respond as the basic facts would be to go against standard scientific procedure. Regularities are not the basic facts in physics, chemistry, etc., but the observed data to be related to something more fundamental. Likewise, we always assume that no ultimate difference between two entities is purely dispositional. If X and Y differ in their dispositions, then this must be explained either in terms of some structural difference between X and Y, or in terms of constituency, or differences in key environmental conditions. Thus the methodology of

1477

generative grammar is a straightforward application of the methodology that has proved itself to be successful in the natural sciences. 4.3. Relations between Syntax as a Formal Object and the Language User There are many ways in which one can construe the relation holding between the rules constructed in a generative grammar and the mind of a competent language user. It is quite implausible to suppose that there is a very close connection between the application of these rules under ideal conditions and what we employ in our daily lives when we process a sentence. To suggest that there is a close link would be to say something analogous to the — implausible — claim that when we do simple arithmetic we use all of the laws, axioms, etc., of that discipline. It is more plausible to suppose that the mastery of the grammar of a language includes having internalized a system of grammatical rules, and that this is something we use partly some of the time and can always fall back on, in case no other clues enable us to take “ short cuts”. The intrinsic merits of a good syntax that describes the language adequately and is in line with syntactic universals does not depend on how closely it reflects the step-by-step processing of actual use. A good description of the structure of a Beethoven quartet does not depend for merit on how closely it mirrors what the competent music lover actually perceives in the process of listening. One could claim in both the case of music and of language that the structures posited represent what we would use under the idealized circumstances in which all other factors that normally enter into music appreciation and language understanding respectively have been taken out. To be sure, this way of looking at the matter depends on our ability to fill in the list of other factors even if this cannot be done completely, and thus have some predictions about language use and actual language understanding. The view would gain in power also if one could show that the factor of linguistic competence is the only constant factor in language processing. There seem to be practical contexts in which other factors play a minimal role. One such context is the correction of oneself or of others in cases in which no semantic or pragmatic reason for wanting to change the message, such as politeness, style, etc., seems relevant. We can see, then, from these reflections, that there may be many different relations be-

1478

tween the rules of syntax and the mind of the language user. The indirect relation between mastery and actual processing that we posited in this section is sufficient to make the study of syntax also an interesting study of the properties of the human mind. 4.4. Ontological Issues; Realism The syntax of a natural language is a complex abstract object. Philosophers with different metaphysical stances will want to classify it in different ways within their ontologies. In order to understand these ontological debates we must distinguish two different senses of the word ‘realism’. In one sense we use this term to describe the view according to which the abstract entities posited by a science as underlying the observables are parts of reality; in fact the more fundamental parts of reality. The contrasting view, instrumentalism, claims that the theoretical constructs have no independent reality, and serve only as instruments for predicting and explaining the observable (for a good discussion of realism in this sense see Cartwright 1983). In the other sense of ‘realism’ it denotes a Platonistic ontology; i. e. one in which abstract entities are posited as existing independently of the mind or of space and time. Mathematical entities, for example, are obvious candidates for being accounted for most adequately within a Platonistic ontology. The two senses of ‘realism’ are quite independent. One can be a realist in one of the senses and not the other. One could attempt to reduce the ontology of syntax to a non-realist ontology by claiming that syntactic structures are purely mental entities or that they can be shown to be equivalent to complexes of spatio-temporal entities, “ particulars,” in some way. AntiPlatonism is linked at times to empiricist views according to which what is fundamentally real is the realm of entities that can be observed by the senses. When we consider language, and syntax in particular, the view that observables are what is fundamentally real seems quite implausible. The essence of a natural language is the set of its constituent rules; these relate to observables only in very indirect ways. The observable may be, at some level, the most directly available evidence, but neither the elements of the explanatory structures nor the basic facts, such as language understanding, are on the level of the publicly observable. These considerations count also against instrumentalism. It would be very difficult, if at all possible, to define a

XXII. Syntax und Nachbardisziplinen

natural language in terms of observable entities. It could not be the sum total of all of the verbal output of past, present, and future speakers, for that corpus will include many errors, corrections, etc. Soon we become involved in counterfactuals such as “ the corpus that competent speakers would acknowledge as genuine language use,” etc., and thus we would be already on the road away from pure instrumentalism. There is another issue here that we will only touch. For the instrumentalist the main problem is how to predict observables, while for some of the non-instrumentalists the key notion is that of explanation and understanding. These alternatives represent different philosophies of science; grammar will be viewed by the two schools in different ways. Returning to Platonism, some detailed considerations should show the enormous task that a reductionist of any sort would face. Let us take simple notions like that of a word or a sentence. How could we identify words or sentences with spatio-temporal non-recurrent particular entities, or even classes of such entities? Using a distinction invented by C. S. Peirce, we can distinguish between a word-type, and all of the written and spoken manifestations of that word, which we will call tokens. Philosophers have tried, without much success, to say in a non-circular way what the tokens of the same type have in common. They represent in written form different shapes and only some normative considerations will draw the line and sort out the tokens of a letter from all other written marks. The same applies to spoken sounds. How many of the large variety of sounds associated by the many speakers of English with the letter-sound ‘w’ are we to count as tokens of that type? Again, it seems that we end up relying on normative criteria. It seems then that a word or letter-sound is not an entity that empiricists or nominalists would countenance. These are — so it seems — Platonic entities, attributes that have many spoken and written instances. Different languages will contain different such attributes. Laws among these will define the sound system of a language. The same consideration holds for sentences, and words. The most natural analysis, then, of letters, words, and sentences, as well as of the rules of syntax and phonology that establish certain relations among these, is that of the realist. The burden of proof lies on the anti-realist and his reductionistic programs. So far none of these has been successful. Indeed, the fact

82.  Syntax and Philosophy

that such programs have been unsuccessful in mathematics has a bearing on this issue. If we need Platonistic entities in mathematics, the ones in linguistics do not call for a new type of entity to be posited. 4.5. Syntactic Categories Syntactic categories, as we said earlier, are classes of expressions into which words and phrases are sorted on the basis of grammatical evidence, and they constitute the domain over which the rules of grammar range, separating the acceptable configurations from the unacceptable ones. Examples familiar from Indo-European languages are: noun, verb, noun phrase, verb phrase, adjective, etc. Since generative grammar aims at formulating linguistic universals, and these are defined partly on syntactic categories, it is essential that we should have a characterization of grammatical categories that is applicable across languages. Let us ask, then, a question like: “ What is a noun?” One might try to give examples, but this will hardly suffice for theoretical purposes. If we were to be restricted to a specific language, then one might try to answer this question by enumerating all of the nouns in that language. But, of course, languages have also a diachronic dimension; thus one would have to enumerate the nouns at each point of time, and even then one would want to know what enables one to go on from one enumeration to another. Furthermore, in a good theory we want to know not only what a noun happens to be, but what an element must have in order to be a noun. In short, we would also want to know what a “ possible” noun is. It would be unsatisfying if we could tell what gold is, but only in certain regions of the world. Such “ local” definitions would not do for scientific purposes. The same applies to syntax. Without a universal vocabulary within which the syntactic categories can be articulated, generative grammar and the related claims about linguistic universals rest on shaky foundations. From a philosophical point of view, then, this is one of the key requirements of this way of viewing grammar. 4.6. Epistemology and Syntax The following four epistemological questions present the philosopher’s key concerns with syntax from the point of view of the assessment of evidence for the claims of the discipline. First, how does a speaker-hearer know what is an acceptable sequence in English?

1479

Second, how does the speaker-hearer know what is a rule of English? Third, how is the knowledge referred to in the first two questions represented in the speaker’s mind? Fourth, how does the linguist know what a rule of English is, and what is a syntactically well-formed part of English? 4.6.1. How do we Know what is a Well-Formed Part of English? Modern analytic epistemology focuses on what it is to know or believe a sentence, or a proposition expressed by a sentence, to be true. Roughly, to know a sentence p to be true is to believe p and to have complete evidence for it. Belief becomes a similar epistemic state on this account, but with less evidental backing. An epistemology concentrating on propositional knowledge and belief, and on the motion of empirical or theoretical evidence, has difficulties dealing with the epistemology of syntax, at least when syntax is presented the way it has been in this essay. For key to syntactic competence is the claim that a competent speaker should know a sentence to be a genuine part of English, and that this knowledge is derived from his being able to apply the internalized rule system that he had learned. But the competent speaker does not base this knowledge on empirical evidence. He does not say that x must be a part of English because it is used, other say so, etc. He might say that while still learning but not once he possesses the full competence. Nor can such knowledge claims be based on introspective evidence. We do not have separate feelings for every well-formed sentence of a language we speak. Our knowledge about the grammaticality of sentences is based, then, on our understanding of the language. This, in turn, is best characterized as having the correct representations of sentences derivable from the internalized rules. Thus this phenomenon, a legitimate case of knowledge, does not fit into the framework of the kind of epistemology that regards as the paradigm case of empirical knowledge knowing a sentence like this is water to be true; that is to say cases in which the evidence for the knowledge claim is a set of observations via the senses that anyone can perform. Returning to our case of grammatical knowledge, we need to stress that the account given above does not entail that every claim of grammaticality made by a competent speaker is infallible. These claims are correlated to intuitions, and these — as we shall see — are

1480

not known directly via introspection. Hence we can be wrong about our intuitions and about our internal representations. 4.6.2. Intuitions and Rules We have intuitions of grammaticality about strings or sequences to which we are exposed. Since the sentences of a language like English are of arbitrary length, and we have intuitions also about sequences that we have not encountered before, we need to seek a source for these intuitions. We cannot suppose that each intuition has arisen as a result of a belief formed independently about a specific sentence. The hypothesis mentioned above, namely that we have internalized a rule system that captures the syntax of English, would provide a fine candidate for the common source of these intuitions. There are good reasons why the rule system as well as its applications should remain on the subconscious level. If processing a sentence or for that matter performing a simple operation of addition in our checkbook, would require conscious acts of rule applications that would fully justify every step we take, it would take a great deal of time to perform the most mundane of activities; human life would be very difficult under those circumstances. Thus we have a way of accounting for our syntactic intuitions. We need to account also for the state of having internalized a rule system. This cannot be a matter of propositional knowledge, but of a mental state involving also non-propositional insight, that one could call understanding (Moravcsik 1979). The intuitions that are derived from this understanding should be distinguished from our introspections that provide indirect evidence for mental states and events. We know our intuitions only indirectly. Introspection, verbal output, correcting others and ourselves, provide evidence of equal worth, and we need to reconstruct our intuitions from this broad basis. To identify the intuitions with the introspective data would bring with it conceptual and empirical problems. We would have to say that we cannot be wrong about our intuitions; an obviously implausible claim. Alternatively, we might have to say that the introspectively formed intuitions are infallible; once more an implausible claim. We need to posit, then, three levels. We have in the mind an internalized rule system. This gives rise to intuitions about particular sentences or sentence structures. Having these intuitions are mental states, not directly observ-

XXII. Syntax und Nachbardisziplinen

able. On the third level we have the observable evidence such as verbal output, introspective judgments, etc. The elements on the third level provide evidence for reconstructing the mental states posited on the second level. These, in turn, are to be explained that there is underlying them an understanding of the language in terms of an internalized rule system. 4.6.3. How are Rules Represented in the Mind? That very question might sound tendentious to some philosophers. For, they will say, why assume that the rules are represented instead of being in the mind in some other way? We have to remember, however, that syntactic competence is a state through time. The continued manifestation of syntactic competence depends on memory. To have the rules of syntax in one’s memory is to have some sort of a representation of these. Problems about how such a representation can take place, what the relation is between the representations and the abstract objects represented, what the elements of the representations might be, etc., occupy a large part of what is called today philosophical literature. Yet the problem is an empirical, not a conceptual, one. This, however, does not ensure that we can find solutions to it. It may be that there are limits to how well humans can understand themselves. But the problem arises in connection with all three of the components of language, the syntactic, semantic, and phonological components. At present all we have is a number of conceptual proposals and research programs (for a good review of these as well as an interesting original analysis, see Fodor 1975). 4.6.4. How does the Linguist Know what is a Well-Formed Formula of English? The task of the linguist is to reconstruct the rule system that constitutes the grammar of, e. g., English. According to the conception articulated in this essay, this rule system is internalized in the mind of the competent speaker-hearer, and as a result this agent has intuitions of grammaticality, phonological wellformedness, etc. Philosophers would be concerned what the data of the linguist are for this project. If you take intuitions to be identical with introspective reports, then the data consist simply of these reports. This leads us to difficulties mentioned before as well as all of the well-known problems of in-

82.  Syntax and Philosophy

trospective psychology. If we take the view of syntactic intuitions sketched above, then there is a variety of evidence from which the linguist is to reconstruct the intuitions which then form the basis of the rule system construction. The data will include introspective reports, but also verbal output, behavior associated with language processing, etc. Thus there is nothing mysterious in this conception about the epistemology of the linguist. 5.1. Syntax and Theories of Cognition Several of the issues raised above are really matters shared between epistemology and theories of mind, or philosophical psychology. One of the important impacts of the rise of generative grammar has been to stimulate philosophical psychology to rethink empiricist views about behavioral and dispositional theories of mind and look for other alternatives as well as to raise new questions about mental representation. As we shall see, one could also interpret philosophers’ preoccupation with syntactic competence and its relation to theories of mind or cognition as helping us to see that some of what might look initially as conceptual philosophic questions can be rephrased to be empirical issues that require empirical evidence for their settlement. 5.2. What can Syntax Tell us about the Mind? As our survey shows, in classical times there was no strong interest in natural languages as such, and analytic philosophers in the early part of this century did not think that the syntactic structure of natural languages showed anything interesting about the human mind. Yet the brief quotation from Aristotle that we examined above shows that there must be some relationship between the structure of language and the “ pathemata” (“ affections”?) of mind. It is only in the past few decades that philosophers became exposed to suggestions according to which descriptions of syntactic structure can tell us important things about the mind, and that we should expand our conceptions of the variety of ways in which one can give characterizations of the human mind. 5.2.1. Characterizing the Human Mind The proposal that certain characterizations of syntax of natural languages give us also a partial characterization of the mind origi-

1481

nates with Chomsky (Chomsky 1965). In order to place this suggestion into proper perspective we shall distinguish a number of ways in which we can give descriptions of a natural substance in general, and of the mind in particular. From the classical period onwards there has been a view stated and restated often, according to which humans should be described as primarily productive and creative creatures. Thus one way of characterizing humans is to state what is unique about the products of their creations and the conditions under which the creation of these products takes place. This is hardly a surprising claim, since we apply the same view to other animal species. Thus everyone would agree that one of the most important characteristics of spiders is that they weave webs, of beavers that they build huts and dams, of ants that they build ant hills, etc. By parity of reasoning we should ask what is distinctive about humans in terms of their products. To say that humans communicate is not to say anything interesting since this is also true of many other animal species. We can make use, however, of syntactic universals and specify what it is that is distinctive about human communication. We need a description of syntax from this point of view. This point can be generalized. It is distinctive of humans that they can compose and appreciate abstract objects like a Beethoven quartet or a Dylan Thomas poem. What we need in these cases too is a description of these products from the point of view of what makes these distinctive human products. If we had descriptions of this sort, they would give us a partial characterization of the human mind, in the same way in which descriptions of spiders as web-weavers or beavers as dam-builders give us important descriptions of the distinctive capacities of those creatures. A beginning in the direction of describing syntax from this point of view was made by the work on the complexity of grammars as related to the complexity of processing agents (see 3.3.1. above). These characterizations of the complexity of mental processing are theoretically at least as interesting as attempts to give a working model for how processing actually takes place. It is a prejudice of a technologically oriented culture to think that we have a theoretically interesting description only if we have an account that tells us how something “ works.” In fact, there is no correlation between how well we know the “ workings” of a mechanism and how well we understand its fundamental principles. Me-

XXII. Syntax und Nachbardisziplinen

1482

chanics can tell us how a car works, but only physicists can tell us the basic principles on which the structure of these contraptions is based. These reflections on syntax and other products suggest the following scheme: we should characterize a number of these distinctive products in such a way as to make the question, “ How is it possible for humans to produce and understand such entities?” theoretically fruitful. It is an open question as to how detailed answers we will be able to give to such questions. We may be able to learn a great deal from answers, even if they do not yield mechanistic models for understanding and production. An abstract characterization of the products has priority over such questions as “ How do we actually produce it or enjoy it?” For one cannot begin to answer the latter question without knowing what it is that we produce, enjoy, or understand, and the description of the object in a theoretically interesting way will be determining whether the latter question will be answered in a fruitful way. 5.3. Syntax and Computation The characterization of levels of complexity of grammars and corresponding automata already made certain assumptions about language processing or the computation of syntactic structure from a grasp of simple elements. In recent years efforts have been made outside of philosophy to sketch computer implementations of various types of grammars. A survey of this literature is clearly beyond the scope of this essay. We shall limit ourselves to a discussion of two questions that show the interaction between syntax and philosophy. First, if we achieve an implementation, what does this show about the human mind? Second, is it not the case that the issue of whether machines can process the grammar of a natural language is unmanageable because of the openness of the concept of a machine? The first question evokes two very different attitudes. On the one hand, it can be said that even if we do have a computer implementation, this does not show that humans process syntax that way. There are epistemological limitations on the value of simulation. The other attitude is expressed by the following: a computer implementation shows at least one way in which processing can take place. Why not use this as a starting hypothesis and go on from there, exploring ways in which the model has to be altered in

view of what we find out about humans? These two attitudes are not incompatible in theory, though they do lead at times to different research strategies. The matter of computer implementation is taken to raise the issue of whether such success would help answer the question of whether minds are really machines of a very complex type. This question is, however, not a well formulated philosophical question. The term ‘machine’, like the term ‘matter’, is not well defined. Every time physics makes progress we enlarge the concept of matter instead of saying that ultimate constituents are partly other than material ones. The same thing can happen to the notion of a machine. Thus unless we have a definite and theoretically interesting general notion of a “ machine” apart from specific proposals generated within automata theory, the question does not lead to fruitful speculation. 5.4. Modularity There is another issue that has been discussed in the philosophy of mind. It can be shown, however to be an empirical issue that cannot be resolved by philosophic arguments. Initially we see the issue involving two concepts of the mind. According to one, the mind is an all-purpose reasoning device, using basically the same procedures with different applications. According to the other conception the mind consists of separate faculties, each with its own objects, principles, and procedures. These are connected by a few allpurpose operations, but the more “ rich” content of the mind lies in what the different faculties handle. This is the modular conception of the mind, and its thesis the thesis of modularity (Fodor 1983). From a philosophical point of view, what we have here are two extreme views, with lots of room for proposals “ in between.” Furthermore, the matter is empirical, since we can answer the key question only by discovering how much, if anything, logical reasoning, mathematical calculation, language understanding, etc., have in common. It is true that one can use one or the other of the positions as a starting point for research. But other than empirical plausibility arguments, there is not much to choose between them initially. This is a case, then, in which further work on syntactic structure and syntactic processing can transform what seems initially to be a philosophic question into a series of viable empirical research programs.

82.  Syntax and Philosophy

5.5. The Innateness Hypothesis; does Syntax Help? The above helps us in getting a grip on another traditional philosophic debate, namely the one concerning whether the important ideas that shape our minds are innate or acquired by experience. The contrast between the innate and the acquired must be made conceptually clear in order to have a viable contrast. In a rough way one can say that a structure or capacity is acquired if the organism cannot develop it or contain it without environmental influences that seem to be required uniquely just for this capacity. A structure of capacity is innate if the organism can develop it on its own, in spite of variations in environmental stimuli. Thus, for example, certain bacteria or viruses are acquired, and so are the skills of playing chess or riding a bicycle. In theory something could be innate to an individual only, but for our purposes the interesting cases are those structures or capacities that are innate to the whole human species. Philosophers have debated at length as to which capacities are innate, and which structures can be assigned the same status (for a fine survey, see Stich 1975). Once we state the issue in the terms sketched it becomes clear that whether language processing is innate or not is an empirical rather than a conceptual question. Furthermore, it seems plausible to suppose that while some aspects of syntactic or phonological capacities and structures are innate, there are likely to be many language-specific features of various languages that need to be acquired by the speakers. Thus here too, we have a case in which traditional philosophical issues can be restated so as to admit of empirical answer. For our purposes we shall stay with a primarily conceptual issue, namely the link, if any, between what is innate and what is a priori. To say that a claim is a priori is to say that its verification or disconfirmation does not rest on evidence furnished by the senses. From this rough characterization we can see that there is no conceptual link between the innate and the a priori. There are many a priori truths, for example, much of mathematics, which are not given to us innately but need to be learned. Again, there may be universal linguistic principles that are given innately to the human mind, but this does not mean that these principles are not

1483

empirical. It may be that we bring these principles to bear on linguistic input right from the start. We have no evidence and need no evidence for them in order for language learning to take place. What this shows is not that these principles are a priori, but that we can do very well in the development of some of our capacities while relying on unsupported empirical principles. 6.1. Epilogue This survey of the interaction between philosophy and syntax was designed to leave us with a number of general impressions. First, it is amazing how the questions raised by Aristotle at the dawn of the subject remain with us as the key questions, and that though we have learned much about empirical applications, the basic primitives of the Aristotelian triangle between language, mind, and reality remain our basic primitives. Second, we have noted how misleading it is to characterize most of the history of analytic philosophy as having linguistic interests, or a genuine interest in natural languages. Such interest was slow in developing, and in recent times philosophers have taken technical tools designed for quite different purposes and started applying these to the study of natural language in its own right. It is still not clear whether these moves will prove in the long run to be fruitful, and whether we have the right tools for the analysis of natural language. We saw also that generative grammar did not grow out of philosophical tools. Rather, it gave plenty of food for thought for philosophers. It would be safe to say that it revitalized the philosophy of mind and raised epistemological questions that epistemology centering on public evidence and propositional knowledge has trouble handling. We saw also how certain philosophical questions can be restated into empirical ones without any assurance that thereby they will not remain “ mysteries” (Chomsky 1980). Finally, we saw also that many fundamental conceptual questions still haunt us, and there is a good chance that some of these, too, will remain permanently unsolvable by the human mind. Alston, William. 1964 . Philosophy of language. Englewood Cliffs. Aristotle. Categories. —. De Interpretatione. —. Rhetorik.

XXII. Syntax und Nachbardisziplinen

1484

B arwise, Jon. 1979. On branching quantifiers in English. Journal of Philosophical Logic 8. 47—80. B lack, David. 1982. Ancient philosophy and grammar. (American Classical Studies 10). Chico. Cartwright, Nancy. 1983. How the laws of physics lie. Oxford. Chomsky, Noam. 1957. Syntactic structures. Gravenhage. —. 1963. Formal properties of grammars. Handbook of mathematical psychology, vol. 2, ed. by P. Luce, R. Bush & E. Gallanter, 323—4 18. New York. —. 1965. Aspects of the theory of syntax. Cambridge. —. 1980. Rules and Representations. New York. Fodor, Jerry. 1975. The language of thought. New York. —. 1983. Modularity of mind. Cambridge. Montague, Richard. 1970. English as a formal language. Linguiaggi: nella societa e nella tecnica, ed. by B. Visentini et al., 189—222. Milan. Moravcsik, Julius. 1977. Grammar and freedom. Linguistic Studies offered to Joseph Greenberg ed. by A. Juilland, 103—14. Saratoga. —. 1979. Understanding. Dialectica 33. 201—15. —. 1983. Natural languages and formal languages: A tenable dualism. Language, logic, and method ed. by R. S. Cohen & M. Wartofsky, 225—39. Dordrecht.

—. 1987. Explanation and meaning. CSLI Vol. on Philosophy of Language, forthcoming. Plato. Cratylus. —. Philebus. —. Sophist. Quine, Willard V. OF. 1970. Philosophy of logic. Englewood Cliffs. Rogers, Robert. 1963. A survey of formal semantics. Synthese 15. 17—56. Russell, B ertrand. 1918. The philosophy of logical atomism. Department of Philosophy, University of Minnesota, Minneapolis. Searle, John. 1969. Speech acts. Cambridge. Steinthal, H. 1863. (1961). Geschichte der Sprachwissenschaften bei den Griechen und Römern. Hildesheim. Stich, Stephen.(ed.) 1975. Innate ideas. Berkeley. Strawson, Peter. 1950. On referring. Mind 59. 320—44. Tarski, Alfred. 1936. (1956). The concept of truth in formalized languages. Logic, semantics, metamathematics 152—278. Oxford. Urmson, James. 1952. Parenthetical verbs. Mind 61. 480—96. —. 1956. Philosophical analysis. Oxford. Wittgenstein, Ludwig. 1922. Tractatus logico-philosophicus. London.

Julius M. Moravcsik, Stanford (USA)

83. Syntax and Psychology 1. 2. 3. 4. 5. 6.

1.

Introduction Psychological Evidence for Syntax Grammars and Parsers On the Nature of the Syntactic Processor Conclusion References

Introduction

Syntax in the traditional sense has little to do with psychology. Although traditional grammar contains scattered remarks about the representation or transmission of ideas by means of word order and focus, nothing like a psychology of syntax has emerged. On the other hand, a comprehensive theory of language must ultimately connect the properties of syntax with the ways language is used and

acquired. In a broader sense syntactic knowledge can be and must indeed be seen as a psychological issue. Anybody who is involved in the present day research on syntax must be aware of the issues of explanatory adequacy that were brought into focus by Chomskyan linguistics. The purpose of this article is not to restate these questions. The emphasis will rather be laid on the questions which immediately concern the time-bound nature of language, where time-bound refers to both bound to real time perception/production and ontogenetic development. Two major questions have to be asked: (a) Is there evidence for syntax as a circumscribed mental ability in humans? If (a) can be answered positively,

83.  Syntax and Psychology

(b) What is the nature of the syntactic processing system and of syntactic development? At first glance, it may appear strange to ask question (a), because after all how could syntax be studied independently from phonology, semantics etc. if it were not a distinct system of knowledge and thus ultimately a product of the mind. Apart from such very general considerations, however, the involvement and the role of syntax in linguistic processing and development has always been a much debated issue in psycholinguistics as well as in neighbouring disciplines such as Artificial Intelligence. It was, for example, not clear for classical associationism that the comprehension of speech required more than the elicitation of the various concepts which are represented by the words contained in the stream of speech. According to associationism, a hearing sensation triggers a sensory word image while a similar image can trigger a motor image for production. Learning was seen as an initial pairing of sensory and motor images with subsequent establishment of links to an “ idea”. It was forerunners of modern psycholinguistics like Wilhelm Wundt and Karl Bühler in general psychology and Karl Kleist and Arnold Pick in neuropsychology who in recognizing the sentence as a central unit of language broke with this tradition (see de Bleser 1987). In present times, all researchers, of course, acknowledge the existence of syntax as a set of rules and/ or principles, and appreciate the work of generations of linguists who have unravelled formal properties that are specific to syntax. This, however, has not prevented many of them to believe and argue that syntax actually plays a rather minor role in language use. Schank (1975) and Riesbeck and Schank (1978), for example, argued for a primacy of conceptual information in language processing. Syntax is taken as a second resort which can be consulted when ambiguities or other uncertainties arise. (For a critique of this view see Dresher and Hornstein 1976). A similar conclusion had also been reached in Bever (1970). Confronted with the intricacies of hypothesized syntactic representations, many psychologists tend to believe that what humans manipulate in language use is unlikely to involve these kinds of representations. This is especially true for branches of psychology that deny mental representation altogether, like radical versions of connec-

1485

tionism; (for a critical evaluation see the contributions in Pinker and Mehler (1988), especially the one by Fodor and Pylyshyn). As will be shown below, however, research in psycholinguistics has succeeded in demonstrating a profound role of syntax in human language processing. What is less clear is the nature of the relation that holds between the syntactic component of a given grammar and the processing system that implements it. This is the general question of how the performance system is linked to the competence grammar. The minimalist but nevertheless extreme view would be that the competence grammar is directly couched into the performance system i. e., the processor is something like grammar supplied with the constraint of finite time, finite memory, and sequentiality. An alternative would be that the competence grammar is orthogonal to autonomous processing constraints which are specific to linguistic processing. A useful tripartite distinction has been proposed by David Marr. Marr argues for a separation of computational (Type 1) theories, which are suitable to decompose a problem, and algorithmic (Type 2) theories, which show how the computational theory can be implemented e. g., by specifying how certain steps in the generation or analysis process are carried out, and the level of implementation itself which deals with circuit analysis or the functioning of neurons and synapses in the brain without making reference to the entities that are manipulated by the process (see Marr 1977, Marr and Poggio 1977). Let us assume that question (a) — is there evidence for syntax as a circumscribed mental ability in humans? — can be answered positively. A much harder question is (b): What is the nature of the syntactic processing system and of syntactic development? An answer to (b) requires considerations of all sorts. Some of the most important ones are these: How autonomous or how interactive (e. g. with pragmatics and knowledge of the world) is the syntactic system? Is there a reason to isolate it in behavior? Does it implement a competence grammar in the sense of a Chomskyan view of language? If this is the case, which kind of grammars are most appropriate? Should the competence grammar be identical or close to identical with an idealized subsystem of the processing system? If autonomy in syntactic processing can be claimed, is there reason to argue in turn in favor of autonomous subsystems of the syn-

XXII. Syntax und Nachbardisziplinen

1486

tactic processor, e. g. units which reflect some of the modules that were proposed for Universal Grammar in Chomsky (1981), like e. g. thematic theory or binding theory? In which sense does syntax contribute to the fact that linguistic processing is impressively fast and efficient, but still fails at very specific tasks? And finally, how is it possible that children almost effortlessly acquire syntactic competence in any natural language in a remarkably short period of time and without experiencing even a fraction of the possible patterns they will produce and understand at puberty? The latter question crucially involves what Chomsky has termed the logical problem of language acquisition. It is a genuinely psychological question, and it touches on one of the deepest problems in our understanding of the nature of language. The academic enterprise that covers among other areas the relation between syntax and psychology, namely psycholinguistics, is divided into three areas of interest: production, comprehension and acquisition. Orthogonal to this is the study of syntactic knowledge and processing under conditions of focal brain damage. The present article will address these areas. It is organized as follows: In section 2. evidence in favor of syntactic involvement in language use is given. I will provide short overviews of work in the areas of comprehension, production, acquisition and neuropsychological breakdown. The presentation can, of course, only focus on certain issues, and it should be clear that a comprehensive review of the writings on these topics is out of the question. In section 3. I will turn to the relation that can be assumed to hold between grammars and parsers, and discuss some of the major models that have responded to this question in one way or the other. In section 4 . I will outline some leading ideas about the nature of the syntactic processor as envisaged in recent theoretical and experimental work on human sentence processing. I will neither try to proceed chronologically nor give a history of the relation between syntax and psychology, but there will be occasional excursions into previous stages of research such as the early times of generative grammar.

2.

Psychological Evidence for Syntax

In the following section we will turn to some of the results which clearly demonstrate the

role of syntactic knowledge in language use, acquisition and breakdown. We will also take a look at some of the earlier work in psycholinguistics. A useful overview is also given in Bock and Kroch (1988). These authors focus on the challenge from a functionalist point of view. Functionalism sees syntax as the result of a more general system of human performance. According to Bates and MacWhinney (1982), there is a mixture of syntactic cues (basically “ word order”), phonological cues (prosody, intonation), and semantic cues (animacy, potency) etc. which together yield certain interpretive preferences; at the same time, there is nothing in their scenario that could be identified as a designated syntax module that drives human linguistic performance. The driving force in linguistic performance is more or less seen as the need for efficient communication. Kroch and Bock distinguish different positions of functionalism with respect to explanations in syntactic change, syntactic categorization, acquisition and adult processing, and show that for each of them there are serious problems if one tries to argue that there is a direct relation between communicative function and form. 2.1. Comprehension It is common practice among linguists to ask people for grammaticality judgements. This is a task that normally nobody is confronted with. Nevertheless most people are able to judge sentences they have never heard or considered before for their grammaticality. If we agree that the assignment of structure to a string of words is a non-trivial matter that taps the tacit knowledge of the hearer about his or her language, the fact that such metalinguistic judgements can be made in a sense proves the psychological reality of syntax. If one considers a sentence like (1) The child drank the milk there is certainly a way to unravel the meaning of the sentence by using elementary semantic categories like AGENT, PATIENT or AFFECTED OBJECT, ACTION, and knowledge about entities, properties and events in general: a special order or hierarchization seems to be unnecessary for the detection of an agent/patient relation that holds for the relation ‘drink’. It is not surprising that severely disturbed aphasic patients with demonstrably very low comprehension above the lexical level understand such sentences

83.  Syntax and Psychology

more or less most of the time. But what about a sentence like (2)? (2) The child hit the girl Here, events in which the child hit the girl and events in which the girl hit the child are equally plausible. There is no way but a formal way to detect the relation that is expressed in this clause. Many years of research both in linguistics and psychology have shown that for humans the key to grasp the meaning of such sentences must be syntax. For anyone who is willing to consider the situation seriously, this is an easy way of showing the psychological relevance of syntax in comprehension. Forster (1979, 55) puts it in this way: “ The whole point of a language having a syntax is to provide a clear and unmistakable indication of the correct interpretation of the sentence. [...] Any move to allow the syntactic processor to be influenced by pragmatic factors works against the fundamental purpose of syntax.” For those who remain sceptical, however, there is more than this. In the early days of experimental psycholinguistics, George Miller and colleagues ran recall experiments in which subjects were presented with serially ordered material (letters, numbers, words) which they had to recall subsequently. The result of these investigations was that random collections of symbols are much harder to recall than symbols which are ordered in some way. This ordering may be superimposed by the processing device itself, as shown by the grouping (chunking) of numbers into subunits, or inherent in the material (Miller 1956). Miller and Isard (1963) demonstrated that their subjects could recall semantically and syntactically wellformed sentences better than semantically anomalous sentences, and semantically anomalous sentences better than random collections of words. What is interesting in this result is that the syntactic organization seems to suffice to improve performance significantly. One could critically argue that semantic abnormality does not prevent listeners to create conceptual representations, and that, as a consequence, the results do not necessitate the involvement of syntax proper in comprehension (see Schank 1975). This kind of argument, however, is incompatible with the other half of the finding, namely that semantically abnormal sentences are significantly harder to recall than normal sentences. In another series of experiments, Fodor, Be-

1487

ver and Garrett, following pioneering work of Ladefoged and Broadbent (1960), presented subjects with sentences of different constituent structure which were interspersed with clicks. The task of the subjects was to report where they heard a click. The overall result of this research was that subjects tended to perceive the click at constituent boundaries even when it was not there physically. (Fodor, Bever and Garret 1974 ; see also the literature cited therein). There were many other demonstrations of the psychological reality of constituent structure e. g., the scaling experiments by Levelt (1970), in which subjects have to rate the degree of relatedness between pairs of words. The outcome of these ratings is strikingly close to the bracketing of a syntactic surface structure. A problem that obfuscates results from recall experiments and other post-hoc methodologies like scaling, sentence-picture verification, sentence completion etc. lies in the nature of the experimental technique. Subjects have time to reprocess the input before they come up with their response. Therefore, socalled on-line techniques appear to be more suitable, if the immediate effect of syntax on comprehension is under investigation. Influenced by the derivational theory of complexity (see section 3.1. below) earlier studies of online processing were very much concerned with the question of syntactic autonomy as opposed to processing under the influence of semantic expectation or even general knowledge. The earlier conception of a serial processing machine was that full interpretation in the sense of a discourse representation or mental model (Johnson-Laird 1983) must be temporally preceded by a formal syntactic analysis. The hope was that this process could be made transparent by methods which allow the researcher to somehow look into the process instead of simply observing the final result. One of the first and certainly most interesting studies used the technique of speech shadowing (Marslen-Wilson 1973, 1985). In these experiments, subjects were asked to repeat back speech input word by word as soon as possible. Certain subjects (close shadowers) are able to do so at an average latency of about 270 msec. MarslenWilson could show that here again random word collections (as well as syntactically structured but nonsensical jabberwocky type of input) lead to an increase of shadowing latencies as compared with the latencies for semantically anomalous but syntactically

1488

well-formed speech. Thus, the off-line findings are replicable by the one-line methodology. One more recently developed on-line technique is the recording of gaze durations on letters or words and regressions in the text while subjects read. With such a method, Flores d’Arcais (1982) could for instance show that subjects have prolonged gaze durations when they read syntactically deviant sentences such as The old lady sat the white chair, even if they tend to not detect such violations consciously. The recording of eye movement for the investigation of syntactic processing was especially developed by Keith Rayner and colleagues (cf. Frazier and Rayner 1982). With the rise of experimental methods that allow the researcher to interfere with the subject’s incremental building of a mental representation many of the older answers concerning the role of syntax in comprehension had to be reconsidered. What had been shown in the earlier studies was that syntactic constituent structure clearly plays a role in language comprehension, while the question that remained open was whether syntactic parsing has primacy over the assignment of an overall meaning, or whether syntax and semantics go hand in hand. The radically non-autonomist view is that semantic and pragmatic expectation can bias the parser toward a certain analysis, while the radical autonomist view would maintain that syntactic decisions must be made before semantic let alone pragmatic factors become effective. Various experiments by Marslen-Wilson and Tyler tried to demonstrate that on-line comprehension is a bottom-up process that makes use of all possible information, including non-linguistic information, from the earliest moment on. This amounts to a view of language comprehension as an interactive parallel process according to which there is no primacy of purely structural processing whose time course could be distinguished from inferential processes that may apply on the basis of pragmatic expectation or knowledge of the previous discourse. In one of their experiments, subjects had to name as quickly as possible the auxiliary is or are that follows planes in (1a) and (1b) (3a) If you’ve been trained as a pilot, landing planes IS/ARE ... (3b) If you walk too near the runway, landing planes IS/ARE ... It turned out that subjects name is faster in

XXII. Syntax und Nachbardisziplinen

(3a) than are, while the opposite is true in (3b), although both continuations would be grammatically legitimate. This means that the subject’s anticipated internal representation of the critical clause containing the auxiliary has been influenced by the pragmatically preferred reading. More experiments to this extent are summarized in Marslen-Wilson and Tyler (1987), and it is argued that they present counter-evidence to Fodor’s (1983) thesis that linguistic processing proper is an encapsulated process that crucially excludes non-domain-specific information (such as knowledge of the world). Various positions between the radically interactive and the radically autonomist view are imaginable. As we shall see in section 4 ., the question has still not been resolved yet. Recent work on syntactic comprehension shows a level of linguistic sophistication that goes way beyond the question of constituency or the occasionally misleading distinction between “ syntax” and “ semantics”. Some of these experiments would not have been possible at earlier stages of syntactic theory (in fact, in many cases the Standard Theory as reflected in Chomsky 1965), because the theory did not specify the appropriate representational format. A lively area of research has, for instance, been inspired by trace theory, the theory of empty categories in general, and binding theory. Notice that the generative syntactic theory of the early stage did not know about the concept of an empty category (trace) that is bound by a lexically overt antecedent, or about the semantic autonomy of the pronominal inventory of the language. One question in recent research has been whether the relation that is postulated between an antecedent and its trace or a dependent pronominal is reflected in the behavior of the listener in the sense that the antecedent is mentally reactivated at the position of the trace or the pronominal. In a sentence like (4) The police saw the boy that the crowd at the party (*1) accused (*2) of the (*3) crime from Swinney, Ford, Frauenfelder and Bresnan (1988) it could be shown that the concept CROWD is still active at the probe point (*1), but dies away at (*2) and (*3), while the concept BOY is inactive at (*1), but becomes activated at (*2). In general, Wh-traces and overt anaphors seem to immediately trigger reactivation of the elements that bind them (see the overview of Nicol and Swinney

83.  Syntax and Psychology

1989). Weaker but still significant effects to this extent have also been demonstrated for NP-traces and PRO-subjects as compared with constructions in which syntactic theory does not postulate an empty category; the examples in (5) are from McElree and Bever (1989): (5a) The shrewd lawyer who argued for the defense was suspected by the judge (*) (5b) The shrewd lawyer who argued for the defense had spoken to the judge (*) At the position marked by (*), the probe word shrewd yields faster recognition in (5a) than in (5b) (see also MacDonald 1989). The reason is argued to lie in the structural difference between (5a) and (5b): The verb suspected is followed by a trace that is bound by the subject-NP which contains the probe word in (5a), but there is no trace after the non-passive verb spoken in (5b). Results like these are taken as direct demonstrations of the psychological reality of syntactic constructions, and are sometimes used in arguments in favor of or against a particular syntactic theory. See, however, Pickering and Barry (1991), Gorrell (1993), Gibson and Hickok (1993) for controversial discussion. We will return to on-line studies of syntactic comprehension in section 4 ., and then discuss some psychological issues of parser design. The points brought up in the present section should have shown that there is various evidence for syntactic structure in sentence comprehension. Which particular representation is used is, of course, a matter of constant debate. If the syntactic processor is a robust and highly automatized device, however, it seems unlikely that listeners or readers engage in syntactic processing only as a last resort as had been suggested earlier. Clearly, fast readers who rapidly scan a text are unlikely to compute every detail syntactically, but they also have a good chance of getting things wrong. It is also clear that one can to a considerable extent comprehend telepraphic and agrammatic speach or foreigner talk. But it seems equally clear that in these cases comprehension is severely limited. Admittedly comprehension will improve if listeners substitute what they miss using their own linguistic knowledge; and there is no doubt that their substitutions will involve syntax. 2.2. Production When one asks about the involvement of syntax in language production there is immedi-

1489

ately a feeling of triviality. Even though it is debatable whether humans always perform a full syntactic parse of the input in comprehension, it is impossible to doubt the involvement of syntax in production for the simple reason that speakers produce — in the majority of speech events — well-formed sentences. Notice that even when speech errors occur, these errors are not “ wild”, but follow patterns which show a relation transparent to syntax. As Garrett (1980, 216) puts it, “ the production system must get the details of form ‘right’ in every instance, whether those details are germane to sentence meaning or not”. So, in this sense it would be futile to have any doubts about a syntactic level in sentence production. What is an important question though is how the encoding of a message should take place. The task of the speaker is to encode a message which may be (vaguely) organized in propositional form, say in the sense of an idea. As long as we do not know what is to be encoded, descriptions of this process run the risk of being quite arbitrary. (See Vygotsky 1962, Fodor, Bever and Garrett 1974 and Fodor 1975) for different views concerning inner speech or a universal language of thought, mentalese). We shall not be concerned with this issue here. The questions which in production pertain to syntax are more mundane. The most prominent questions are: How far ahead do we plan a syntactically encoded message when we speak or start to speak? What are the syntactic domains for this planning? Early studies investigated hesitation phenomena such as uhs and pauses (Goldman-Eisler 1956) as well as breath cycles (e. g., Fodor, Forster, Garrett and Hakes 1974 ). GoldmanEisler was mostly concerned with decision processes preceding lexical choice i. e., with the proper encoding of a message. Naturally, hesitation effects occur within syntactic constituents e. g., these ... uh traveller cheques. Effects such as these are syntactically interesting because it is often the case that while the (semantic) head of the construction is still not available decisions have already been made about its syntactic nature in terms of agreement features. The studies on breath groups provided evidence that in contrast to hesitation, breathing falls into syntactic boundaries. At the time when experimental psycholinguistics was heavily influenced by the Standard Theory of transformational grammar there were theoretical reasons to assume that the entire surface clause constitutes the plan-

1490

ning unit for an utterance: If deep-structure is the level that determines semantic interpretation, and the derivation is seen as psychologically real sentence generation, it is plausible to assume that sentence planning proceeds clause-wise. The initial, grammarinspired idea was that the whole surface clause is planned and then generated from deep structure. Ford and Holmes (1978) studied responses to tones that were randomly presented while subjects were talking. Their finding was that response latencies occurred at surface as well as “ deep clause” boundaries. Ford (1982) points out problems which arise from the adoption of old style transformational grammar for production models. Speech errors (see Meringer and Mayer 1896 and a lot of subsequent work) have been a major source of evidence for the investigation of planning units in speech. Boomer and Laver (1968) in a study of unsolicited speech, found that sound exchanges (“ spoonerisms” such as You hissed my mystery lesson instead of You missed my history lesson) almost never take place across clause boundaries. On the other hand, word and phrase exchanges can cross clause boundaries, as in the example provided by Garrett and Shattuck (1) I saw a lot of cars down there towing trucks away in which trucks and cars have been reversed. Another finding is that such exchanges are category true i. e., it is unlikely that nouns and verbs exchange. Fodor, Bever and Garrett (1974 ) interpreted these results by arguing that word/phrase exchanges occur at a level of processing at which the clauses are still under construction. Thus, the premature selection of a lexical item can lead to an inverse insertion into the structure. Since syntactic structures are only later transformed into phonemic/phonetic representations, sound exchanges can only take place in intraclausal local environments. On the basis of a large corpus of speech error data, Garrett (1975, 1980) has developed a serial model of language production in which one level of processing serves as the input to another level. The transition is from an idea (message) via distinct levels to a form of representation of articulatory plans which are executed by the articulators. One major syntactic division in this scenario holds between a functional level and a positional level.

XXII. Syntax und Nachbardisziplinen

Roughly, at the functional level grammatical relations are specified, while the positional level holds surface syntactic templates including word order specifications and grammatical formatives. Lexical insertion into the preplanned positional level yields anticipation and exchange errors of various sorts. Fromkin (1971) observed that when two words or phrases exchange, the stress pattern of the intended utterance remains unaffected. This suggests that the prosodic structure of an utterance is autonomously organized, as already suggested by Arnold Pick and, for entirely different reasons, by Selkirk (1984 ). If it is true that exchanges tend to not cross clause boundaries, one can interpret this as evidence that syntactic planning proceeds clause-byclause, and that in a complex sentence the message is formulated ahead while one clause is still articulated. Rather than encoding a complete thought unit, the production system seems to work in the fashion of a cascade. Kempen and Hoenkamp (1987) propose an incremental procedural grammar which mimics this property. In their account, the message is identified with something like thematic roles. These roles are successively mapped onto syntactic categories which are subsequently “ grammaticized” i. e., supplied with function words and inflections. The computer implemented model is intended to capture the functional and the positional level proposed by Fromkin and Garrett as well as production phenomena such as selfrepairs of planning errors (Levelt 1983) and the omission of function words (Kean 1977). Levelt (1983) presents an interesting account of speech repairs in showing that rules of coordination and rules of mapping an answer onto a question share essentially the same constraints. A possible repair is, for instance, (2a), but hardly (2b) (2a) From the purple node you go left, uh right (2b) From the purple node you go left, uh purple node you go righ while the latter example is well possible when the preposition and the article remain in the repair. Consider a coordination such as (3) *From the purple node you go left and green node you go right. Obviously, the substring NP + you + go + X is not a constituent and can thus not be coordinated. Speech repairs thus seems to reflect the speaker’s syntactic intuitions very closely.

83.  Syntax and Psychology

Bock and Kroch (1988) report a number of studies which demonstrated that speakers tend to follow syntactic plans from previous discourse, say by using a passive sentence rather than an active one if there has been a passive sentence in the vicinity, and that this adaptation is unconscious and uninfluenced by extrasyntactic factors. There are further investigations which provide striking and non-trivial support for syntactic constituency in production. Woolford (1983), Joshi (1985) and others found that in bilingual settings code switching is heavily constrained by an asymmetry that must be assumed to hold between source and goal language and by the syntax of the source (or ‘embedding’) language. One major result of this kind of research which fits into Garrett’s model is that grammatical morphemes cannot be affected by the goal language. Joshi provides examples of switching from Marathi to English. The Marathi PP in (4 a) can be altered as in (4b), but not as in (4c,d): (4a) kaahi khurcyaa-war some chairs -on ‘on some chairs’ (4b) kaahi chairs-war (4c) *kaahi khurcyaa-on (4d) *on kaahi khurcyaa (4e) *war some chairs These examples do not only show that the postposition war (which is a quasi function word) cannot change to English, but also — as shown in (4 e) — that the position of war must be fixed at the level where the English vocabulary is inserted. In Garrett’s model this would be the positional level. Garrett’s model is up to now the most serious attempt of coming to grips with syntactic production. It has been criticized for mixing up function words and inflectional affixes and for not being explicit about the processes that are assumed to take place in the transition from one level to the next. Lapointe and Dell (1988) make an attempt at spelling out underlying assumptions in Garrett’s model in more detail, drawing on a difference between the establishment of syntactic fragments and lexical access which they see mainly as described in interactive activation models. 2.3. Acquisition Given the complexity of syntax, the question arises how little children can master it at an early age and without noticeable difficulty, special instruction etc. The acquisition of

1491

syntax is a central part of what has become known as Plato’s Problem or the Logical Problem of Language Acquisition in the Chomskyan research program of the study of natural language. (See Chomsky 1986b, chapter 3). In the USA, earlier understanding of learning was strongly under the influence of behaviorist psychology. Language was thought to be acquired by imitation and reinforcement. It is safe to say that this view has been completely abandoned, and that this was primarily the consequence of Chomsky’s criticism of behaviorist learning theory (Chomsky 1959) and the development of an alternative research hypothesis which ascribes to the child rich innate mental structures which unfold in the course of linguistic experience across time. As Brown (1973) first has shown for children acquiring English, there seems to be a specific time course for the development of syntactic structures. There may be a certain degree of individual variation, and there may be some variation across superficially similar constructions (e. g. passive) in different languages, but apart from this it seems to be true that the acquisition of syntax always unfolds in a number of distinguishable steps: (i) An initial step where functional elements like determiners and inflectional affixes are basically missing (mommy sock, no bed, other shoe etc.; see Radford 1990). (ii) Overt functional elements become productive. Here one finds a lot of variation depending on the role morphology plays in the respective language; as one can expect, morphology is used by the child earlier in richly inflecting languages than in morphologically impoverished ones such as English. (iii) Complex sentences emerge with overt marking of subordination by complementizers. Under normal conditions, most of the syntactic knowledge is acquired by the age of six. A robust, uniform and time-bound process like this is, according to Lenneberg (1967), characteristic of a genetic base. Chomsky compares this process with biological maturation as seen in fetal development or puberty. Evidence for this view comes from the observation that there must be a critical period during which the language processing system has to be neurally fixed. Exposure to language input after that period prevents normal development of the adult target grammar, as studies of Kaspar Hauser type cases and studies of wolf children or badly neglected children have shown (e. g. Curtiss 1977).

1492

Another important point speaking directly against behaviorist learning is that the child obviously develops the grammar more or less by himself or herself. The input is far away from the competence system that it ultimately yields; and it has been shown that (a) corrections normally do not concern the formal syntactic side of the errors the child produces and (b) even if that happens the child does not pay attention to the correction (McNeill 1970). Thus, there is a stage at which the child would insist on regular forms of inflection even if that particular form fails to appear in the adult model and can therefore not have been experienced, like *go-ed instead of went. Snow (1972) and Snow and Ferguson (1977) have argued that the child’s language may be a function of the particular way their mothers talk to them. This has become known as the Motherese Hypothesis. It has been criticized for different reason: First, children are obviously not conservative learners i. e., they do not exclusively produce forms that they have witnessed in the input, as we just saw (Newport, Gleitman and Gleitman 1977). Secondly it is unclear how grammar should unfold better with the help of motherese when the latter offers an even more impoverished input (Chomsky 1975, Wexler 1982). There is evidence from language contact and language change that supports the view that syntax does not emerge on the basis of the input alone: Pidgins may be considered ‘unnatural’ languages in the sense that they are naively invented by adults and used in the absence of a common language between different speakers. They are usually very simple, lack grammatical morphology, (case, gender, tense, subordinators etc.), and do not in general compare with what is normally found in a language with a longer history. Interestingly it could be shown that a pidgin is changed by the generation that learns it as the first language. The resulting system — a so-called creole — shows most of the complexities of a language that has existed for hundreds of years. Bickerton (1981, 1984 ) takes this as the “ ab ovo creation of language” (1981, 4 5), i. e. the emergence of a fully developed linguistic system from an impoverished adult model that could not, in any serious sense, have been the model for the target. Much recent research has shown that central syntactic properties of the adult

XXII. Syntax und Nachbardisziplinen

grammar, like for example the verb-second (V2) property or German main clauses (Weissenborn 1990, 1992, 1994 ; Roeper 1992; Meisel and Müller 1992; Penner 1992), can already be found in the linguistic knowledge of fairly young children. The assessment of children’s syntactic knowledge is less clear concerning interpretive processes as they have been studied over the years — mainly in the framework of GB-theory — such as coindexation with an empty pronominal, or reflexive versus pronominal binding (C. Chomsky 1969, Solan 1983, Jakubovicz 1984 ; Deutsch, Koster and Koster 1986; Lust 1986; Schaner-Wolles and Haider 1987; Grimshaw and Rosen 1990; Grodzinsky and Reinhart 1993 and others), long distance dependencies like WH-movement (de Villiers, Roeper and Vainikka 1990, Roeper 1986, Thornton and Crain 1994 ). Here we find much controversy about how much of the adult knowledge would have to be attributed to the young child. Given that interpretive abilities of small children cannot easily be addressed on the basis of their spontaneous productions, most of the above mentioned research uses various kinds of comprehension tests such as acting-out tasks, sentence-picture verification tasks, judgement tasks, etc., some of which have also successfully been used in neurolinguistic research. Despite our still fragmentary knowledge about the child’s early syntactic capacities, the research of the last few years has provided clear evidence that children acquire the essential syntactic properties of the target language between age two and three. This raises again the question how learning is possible given the complexity and abstractness of the regularities involved. The number of syntactic types of errors that occur in child language development is astonishingly small compared to the number of error types that would occur as logically possible combinatory options. This has lead to the overarching question of modern acquisition research, namely which assumptions should be made about the initial linguistic knowledge of the child. The actual course of syntactic development as characterized so far makes it plausible that there must be strong and possibly innate constraints. One of the most influential proposals in recent years has been Chomsky’s hypothesis that UG is part of the child’s initial biological endowment. But this can obviously not mean that the child is born with a particular grammar. The language

83.  Syntax and Psychology

acquisition device must be designed to cover all existing or even virtual human languages; it must be something like the blueprint of grammar. Its main effect would be that the child could only hypothesize properties about the target language which are compatible with some parametric option as made available by UG. This assumption constitutes one of the elements for a possible solution of a central issue in learnability, the so-called No Negative Evidence problem which stems from the classical observation by Brown and Hanlon (1970) that there are almost no explicit corrections of the child’s syntactic errors by the caretakers. This means that children have to derive their final syntactic competence and processing capacities from UG and the input data. To illustrate this point, the child will, for example, be confronted with sentences like (1) What do you think [we bought]? where the WH-filler what is one clause further up from the clause in which its argument status is licensed, namely to the right of bought. Learning this as a fact would not do because in the final state the child will also both accept and comprehend sentences such as (2) What do you think [Paul believes [we bought]]? and reject sentences such as those in (3): (3a) *What do you believe [the claim [that we bought]] (3b) *What do you wonder [who believes we bought]? The child must be able to project the knowledge that the sentences in (3) are ungrammatical from experience with positive data i. e., what appears in (3) cannot be data. Because of questions like these, research in learnability theory has proposed that learnable grammars must have certain properties which enable the child to project syntactic knowledge (in this case, constraints on long-distance dependencies) from input that appears to observe locality conditions (Culicover and Wexler 1980). For the examples in question one must assume the Subjacency Condition (Chomsky 1973) which roughly says that movement of a phrase may not cross more than a single bounding node (basically NP and CP) (see van Riemsdijk and Williams 1986, ch. 4 ; see also Berwick and Weinberg 1984 and Pritchett 1993 for discussion with

1493

respect to parsing). Chomsky (1986a, ch. 6) has moved to a notion of subjacency that tolerates no crossing of a phrasal node which is not selected and has some “ escape” position. Assuming a structural specifier position at the left edge of each of the dependent clauses in (2) through which the WH-phrase can move, there will be no crossing in the technical sense, whereas there will be a crossing of NP in (3a) and a crossing of the medial WHclause in (3b) because who blocks the position through which what could move otherwise. Constraints like the Subjacency Condition are supposed to be part of the universal endowment that determines the structure any kind of human language can take. The variability of languages is accounted for by assuming that apart from lexical idiosyncracy, languages can only differ with respect to a limited number of parametrical options such as the so-called Pro-Drop Parameter (Jaeggli and Safir 1989), or a parameter that determines whether the head in some language precedes or follows its complement. But in order for the principles of UG to apply to the input, the child has to find a way into concrete syntactic representations. How can this be done? One can hardly assume that the child is born with syntactic notions like “ noun”, “ verb”, “ preposition”, “ complementizer“ etc. Given that there are language which do not have prepositions and PPs or complementizers this could be an odd assumption. The view of a direct reflection of syntactic categories in the mind of the child has to be modified in the sense that there can at most be an innate schema for syntactic categories; nothing might be specified in this program as to how many such categories there are, or what their precise nature is. But given that a projection schema in the sense of X-bar theory is innate, it must be fed with real syntactic categories in order to become effective in linguistic performance. How should the child pave his or her way to syntactic categories then? It has been suggested (Pinker 1984 , 31 ff) that this can be done by a process of semantic bootstrapping. The idea is that the child acquires the relevant notions from the parental language by being exposed to objects, properties, actions etc. of the physical world in connection with units of speech that refer to or accompany them. Various hypotheses have been proposed to account for the course of language acquisition within the framework of a parameterized UG approach. Most of them constitute some

1494

version of the Continuity Hypothesis. The strong continuity hypothesis claims that the child’s grammar is constrained by UG throughout, and that it is always a true partial grammar of the adult grammar i. e., it stays within the parametric system of the target language the child is exposed to. The weak continuity hypothesis, on the other hand, allows that on his way to the mature stage the child may develop grammars which are inconsistent with the parametric system of the target grammar. The Maturation Hypothesis (Felix 1987, Borer and Wexler 1987) is opposed to both strong and weak continuity in allowing for the possibility that the child’s grammar may violate principles of UG. Notice though that it still remains in the range of Chomsky’s claim that the child is born with UG: The apparent discontinuity implied by the maturation hypothesis would be due to the fact that principles of UG may mature over time i. e., some of them may simply not be available at an early period of the acquisition process. A general problem concerns the question of how much of the difference between the child’s language and the adult language may be attributed to differences in the general cognitive and processing capacities of children and adults. An obvious point would be that adult-like syntactic processing requires an amount of short term memory that exceeds that of little children. Considerations of learnability have lead to the postulation of additional constraints by which the child’s grammar may differ from the target. Manzini and Wexler (1987) have proposed the Subset Principle as a universal constraint on learning that enables the child to preserve continuity in his or her hypotheses about the parameter of the target language. Since different languages apply different parameters, for example to define binding domains, the child must start with the least permissive system i. e., with a subset of the parametric options; otherwise the parameter could be set the wrong way, and unlearning would be required. Given the absence of negative evidence, it would be unclear on which basis a more liberal parameter could be reset to a more conservative one. While the subset principle seems to guarantee weak continuity, strong continuity may need further devices like a constraint on local wellformedness as proposed by Weissenborn (1994). Due to the centrality of Plato’s problem

XXII. Syntax und Nachbardisziplinen

for Chomskyan linguistic theory, theoretically inspired work in language acquisition has become a central part of syntactic research. Consequently, considerations of learnability these days serve as an evaluation metric for any linguistic hypothesis. Research in generative linguistics also had an impact on much work in second language acquisition and the theory of syntactic change. Of course, there is also research in acquisition that is more or less opposed to the assumptions of the Chomskyan program, but it is often not really possible to compare these lines of research because they occasionally show radically different opinions about the role they would attribute to syntax in human language processing and acquisition. 2.4. Neuropsychological Disorders It would have been natural to subsume a discussion of syntactic breakdown in pathology under the sections on comprehension and production, but the very nature of neuropsychological disorders detaches the phenomenon considerably from mainstream psycholinguistics. The study of language in a disrupted linguistic or language processing system is incomparably more demanding because it involves our understanding of the functioning of the brain. It is quite clear that in studying the undisrupted linguistic system one can abstract away from its neural hardwiring. In studying the disrupted system, however, e. g. in cases of acquired neurological disorders, the functioning of the cortex comes in as a complicating variable. It is difficult to argue that such malfunctioning is nothing more than a global reduction of the normal system and that — as a consequence — it enables one to explore the normal system by looking at pathology, although this possibility cannot be excluded on logical grounds. Take two well-known aphasic syndromes: Broca’s aphasia with agrammatism and Wernicke’s aphasia with paragrammatism. Both are known as classical syntactic disorders. The first manifests itself in the sparse output the speaker produces which is characterized by a tentative reduction towards only retaining lexical head categories such as N, V, A and canonical word order. The latter abounds in long rows of syntactically elaborate phrases which, however, contain numerous morphological and lexical misselections, unwarranted repetitions and blends of constructions. The syntactic disorders associated

83.  Syntax and Psychology

with both syndromes are quite different, and it is natural that researchers have tried to relate these functional disorders to the different lesion sites they are in the core cases accompanied with. Several other neuropsychological symptoms with modality specific functional asymmetries have been observed which shed light on the syntactic processing system of patients with neurological disorders. Many people are unaware of the fact that the history of neurolinguistics is considerably longer than the history of normal experimental psycholinguistics. The driving force behind the European research in the late 19th and early 20th century was not yet the interest in language, but the interest in the functioning of the brain. With researchers interested in neurological and neuropsychological information processing like Carl Wernicke, Ludwig Lichtheim, Arnold Pick, Wilhelm Wundt, Karl Bühler, Karl Kleist, Kurt Goldstein, especially in and from the German academic world of this time, a level of theorizing and sophistication of investigation had been reached that was only resumed much later in post-war America. (see de Bleser 1987 and Caplan 1987, ch. 2). Research of this time concentrated very much on sensory versus motor aspects of linguistic processing as dictated by the mapping to specific parts of the left hemisphere of the brain. A serious problem was, of course, that there was no understanding of the nature of language yet that would have been comparable with the insights that were possible from the 60ies onwards. Right from the beginning of modern linguistics there was a discussion about the characterization of different types of aphasia as disorders of competence or linguistic performance. For peripheral disturbances such as dysarthria, dyslexia etc. the answer could easily be found, however, in cases of central language processing deficits there has been controversy until the present day (cf. Whitaker 1970, Weigl and Bierwisch 1970, Linebarger, Schwartz and Saffran 1983). A central deficit of syntactic knowledge or syntactic processing appears independently of the modality in which language is used (comprehension (auditory/reading), production (speech/writing), naming, repetition etc.). Most of the clear cases of agrammatism and paragrammatism show this kind of multimodal disturbance, but there are exception. Miceli, Mazzucchi, Menn and Goodglass (1983), for example, found agrammatic Italian patients without a demonstrable compre-

1495

hension disorder. Agrammatic production is usually unsyntactic in the sense that primarily major categories (mostly nouns, but also verbs and adjectives) are used in telegraphic style to convey a communicative intention. A German man wich severe agrammatism describes a picture in which a boy cries because he has broken a glass as

A German woman with a milder form of agrammatism describes a picture in which a boy plays with a dog as

Agrammatism has often been characterized as a loss of access to function words or morphemes (Kean 1977, Bradley, Garrett and Zurif 1980; Friederici 1985), but in both of the examples in (1) and (2) there are function words/morphemes. The problem seems rather that functional morphemes that encode case, tense, agreement and relate it to word order or grammatical function are either avoided altogether or mis-selected as indicated by bold-face in the examples above (cf. Grodzinsky 1984 for Russian and Hebrew; Bayer, de Bleser and Dronsek 1987 for German). Many agrammatics are well capable of forming short syntactically correct phrases, but this needs their special attention and can only be performed to the detriment of phrasal complexity and communicative success. The phenomenon is compatible with the hypothesis that agrammatism is a central syntactic processing deficit (Berndt and Caramazza 1980) or that the patient adapts to a general motor deficit (Kolk and van Grunsven 1985). Contrary to earlier assumptions of a pure motor deficit, agrammatics normally also show syntactically disturbed comprehension. By far the most detailed investigation of aphasic comprehension in English is Caplan and Hildebrandt (1988). The theoretical background is

1496

GB-theory and the parsing assumptions of Berwick and Weinberg (1984 ). Caplan and Hildebrandt asked selected patients to act out with toy figures the meaning of sentences involving empty categories and referential dependencies (binding relations) of various sorts. Since it is known that comprehension improves when semantic cues can be used (as shown by the non-reversibility of the thematic roles in The child drank the milk), such cues must be excluded if one wants to get a glimpse of syntactic parsing pure. Relative clauses like (3a) are for certain patients easier to comprehend than those of type (3b, c): (3a) The frog hit the goat that kissed the cow (3b) The frog that the goat hit kissed the cow (3c) The frog hit the goat that the cow kissed The reason seems to be that a non-syntactic strategy of predication may help in the first case: “ frog-hit-goat” and “ goat-kiss-cow”; but such a strategy fails in the latter two cases. Thematic relations can only be detected when the relative operator is associated with an empty category (trace) in argument position: “ [The frogi [Øi that the goat hit ti]] kissed the cow” which leads to the propositions “ goat-hit-frog” and “ frog-kisscow”. See also Grodzinsky (1990, ch. 4 ) who finds that agentive active sentences and adjectival passives are correctly understood in roughly 90% of the cases, whereas syntactic passives are (with 60% correct) hardly above chance level. Grodzinsky hypothesizes that agrammatic patients have syntactic representations which lack traces, and are thus unable to detect thematic relations that are induced by antecedent-trace dependencies. As the details of Caplan and Hildebrandt’s work show, however, it is unlikely that such exclusively grammar based generalizations will stand up to the evidence. The idea that neuropsychological misfunction maps directly into nonaccess to certain syntactic categories or distinctions has been with us for a long time, but the empirical results it has borne have always been fragile. The problem is fundamental, because such purely structure based accounts ignore the functional plasticity that remains in the damaged brain. Caplan and Hildebrandt (1988, 73 ff) give a review of the role of short term memory (STM) in syntactic comprehension. Most of the studies on comprehension in patients with an STM deficit could not demonstrate a clear deficit for sentence comprehension. The reason seems to be that even in patients with

XXII. Syntax und Nachbardisziplinen

a memory span as small as two digits (as shown by their bad performance in the recall of unstructured collections of items) the syntactic parser creates structures which are immediately interpreted and sent to a more long-term semantic store. These results indicate that most if not all decisions by the syntactic parser must be extremely local. Cases of acquired grammar/parser disturbance can occasionally shed light on the thesis that syntax is an autonomous subsystem in linguistic performance. An interesting piece of evidence for formal linguistic autonomy in general and syntactic autonomy in particular comes from patients who can hardly comprehend or communicate but retain exceptional abilities for repetition. This rare syndrome is called sensory or mixed transcortical aphasia. Geschwind, Quadfasel and Segarra (1968) argue that it emerges as the result of a functionally as well as anatomically isolated Wernicke-Broca speech area in which the pathways to the rest of the neocortex are blocked. Most of the studies of such patients report that the patients do not repeat the input in a parrot-like fashion, but rather tend to correct syntactically deviant input spontaneously. *The cats drinks milk or *The boy gave she a present are spontaneously changed to the correct form while semantically anomalous sentences like I had a building for breakfast are not because the patients do not notice semantic anomaly (see Sasanuma and Monoi 1975, H. Whitaker 1976, Davis, Foldi, Gardner and Zurif 1978, de Bleser and Bayer 1986 and Dogil, Haider, Schaner-Wolles and Husmann 1993). Whitaker and de Bleser and Bayer could also demonstrate syntactic abilities in completion tasks. Given the noun Stiefel (‘boot’) a German transcortical aphasic coult turn it to ein verdammter Stiefel (‘a damned boot’) in which NP-internal word order and agreement is fully grammatical. This is striking evidence for the independence of local syntactic processes from access to the semantic system. On the other hand, it is well possible that syntactic decisions in normal processing can be influenced by non-syntactic information (see section 2.1.). Modern neurolinguistic research teaches us that syntax serves a highly automatized language processing system which is not found in other areas of higher cognitive functions. This system is specific to man, and specific to certain areas in the left hemisphere of his brain. Damage to the neurological corre-

83.  Syntax and Psychology

late of the syntactic processing system yields a situation in which normal communication is drastically impeded, mostly in all modalities of language use. The complexity factor that appears in impeded processing is structural in nature i. e., it can hardly be characterized in terms of phrase length or digit span of short term memory, but must be described in terms of the structures assigned by the processor on the basis of the competence grammar.

3.

Grammars and Parsers

3.1. The Derivational Theory of Complexity The beginning of experimental psycholinguistic research on syntax in North America was to a large extent provoked by the advent of generative transformational grammar. Fodor, Bever and Garrett (1974 ) give a comprehensive overview and evaluation of the work that has been done until the early seventies. The assumption underlying a lot of this research was that a transformational grammar as proposed in Chomsky (1957) or Chomsky (1965) could serve as a model for natural language processing. The idea was that in production the speaker starts with a deep structure which is semantically interpreted (approximately a Garrett-type “ message level”). This deep structure was then converted by the application of transformations into a surface structure. For comprehension, the application of rules had to be reversed. After the phonological decoding of a phonetic input stream a surface structure with labelled bracketing is reached. Transformations have to be reversed in order to lead from surface to deep structure, which then can be interpreted semantically (where “ semantics” equals thematic structure). Apart from the aspect of phrase structural complexity and sentence length a major expectation was that it costs time to apply a transformation, and the more transformations apply to a sequence the more processing time will be consumed. This has become known as the Derivational Theory of Complexity (DTC). The expectations connected with the DTC were confirmed by a first series of experiments, as reported in Miller (1962). These experiments were designed after the model of Chomsky (1957) in which transformations could introduce morpho-semantic elements such as negators and

1497

question morphemes into so-called kernel sentences. One problem was that transformations which do not have a noticeable semantic effect such as affix hopping or dosupport failed to exhibit a delay in recognition. The theoretical shift in transformational grammar that culminated in the ‘Aspects’ framework of Chomsky (1965) left part of the early studies without theoretical grounding. Later studies could also not confirm the DTC. A classic up to present days is the passive construction in English. The DTC prediction is that passives are “ harder” than actives. In a picture verification study with children, Slobin (1966) found out that this was only the case when subject and object were semantically reversible as in (1a), but not in (1b): (1a) The girl was hit by the boy (1b) The kite was flown by the child This was taken as evidence that — other things being equal — passives deviate from the canonical sentence pattern of the English clause as expected by the listener who follows certain heuristics. It was assumed that these elementary heuristics could be cancelled or weakened by the semantic content of the structure to be processed (see also section 3.3.). Fodor, Bever and Garrett (1974 , 327 f) point out cases in which the DTC clearly makes counterintuitive predictions. One of their examples is that according to the DTC the sentence in (2a) should be easier to process than the one in (2b): (2a) The first shot the tired soldier the mosquito bit fired missed (2b) The first shot fired by the tired soldier bitten by the mosquito missed Although the second sentence contains two passives, it is more readily comprehensible. Considerations like these and the rapid development of syntactic theory, which threw many assumptions about transformations over board, led to a quick abandonment of the DTC. On the other hand, the negative effects the DTC had on psycholinguistics was taken as a challenge by some linguists to develop grammars which are more suitable to these results and which are as a result better candidates for parser implementations. I want to briefly review four developments: Augmented Transitional Networks (ATN), Lexical Functional Grammar (LFG), Generalized Phrase Structure Grammar (GPSG) and Government and Binding (GB) Theory.

1498

3.2. Augmented Transition Networks As Chomsky (1957) has argued, English cannot be described as a finite state language, because it allows for recursion that can set mutually dependent elements such as if ... then or either ... or apart in an (in principle) unlimited fashion. It is much more difficult to argue that natural languages are not describable by recursive context-free rewrite rules. In fact, it was only recently that formal linguistics has found enough reason to say that they cannot. (See Shieber 1985 and much earlier Evers 1975 who speculated that Dutch cannot be parsed by a pushdown storage automaton; see also Bach, Brown and Marslen-Wilson 1987 for processing results that support this speculation). Chomsky was originally led to the transformational enrichment of phrase structure grammars by conceptual considerations. (See Chomsky 1975, chapter 5). As we have seen, the DTC was directly linked to the presence of transformations in a grammar that was considered to serve as a parser (though not by Chomsky himself). Thus, the negative results concerning the DTC invited the abandonment of transformations. On the other hand, syntactic complexity per se was always found to go hand in hand with processing complexity. A formal model which seemed to capture the complexity effect while avoiding the problems the DTC faced, was provided by Augmented Transition Networks (ATNs) as developed in Woods (1970) and applied to psycholinguistic questions in Kaplan (1972) and Wanner and Maratsos (1978). An ATN incorporates a finite state grammar with a finite number of nodes which are connected by directed arcs. An ATN is enriched in various ways in order to escape the limitations of ordinary finite state grammars. One such addition is a push-and-pop device (push-down store). Arcs which carry state names cause the name of the constituent to be pushed into a store from where it can only be removed (popped) when the final state of the constituent is reached. With this device context-free languages can be parsed. In order to handle cross-serial dependencies as originally diagnosed in natural language by Postal and later by various linguists working on Dutch and other Germanic languages, Kaplan (1972) proposed so-called registers. These are devices by which the name of a constituent can be withheld until more information to the right is accessible, thus allowing for changes

XXII. Syntax und Nachbardisziplinen

in the content of the register. By this device, limitation of context-freeness can be circumvented. Kaplan shows how a number of the processing constraints that Bever (1970) postulated could be captured by his ATN-model. Processing complexity is measured in the number of transitions that have to be performed or attempted in order to arrive at an analysis. For this to work, the arcs have to be ranked according to some metric that does not follow from the system itself. A problem that has been noticed a long time ago is that there is no theoretical limit to the information that can be stored and withheld in registers, and that storing does not hamper the parser’s further processing. Wanner and Maratsos (1978) show, however, that the mental capacities necessary for storing can be linked to those necessary for phrasal analysis. But even if this can be done, it remains unclear how the processing of unbounded dependencies could be made sensitive to the architecture of the processor. Why for example, is it possible to hold a WHphrase across a syntactic boundary of some kind but not another (see the discussion of subjacency in section 2.3.). In the ATNframework all of this has to be directly written into the net in the same way as rules like PASSIVE are written down directly. So while overcoming certain problems with the DTC and with the limitations of finite state grammars, the ATN-approach remains a technical device that does not by itself respond to the grammar which it implements. In the case of Kaplan (1972) this is a grammar with a large set of rules. 3.3. Lexical Functional Grammar Clearly inspired by the failure of the DTC is the development of Lexical Functional Grammar (LFG) (cf. also article 26 in volume I). On the basis of psycholinguistic evidence which could not demonstrate a syntactically motivated complexity difference between active and passive sentences (see Slobin 1966, and section 3.1.), Bresnan (1978) concluded that function changing transformations such as PASSIVE, RAISING etc. have to be eliminated and substituted by lexical rules. In this approach, the active form of a verb like eat is related to its passive form eaten by functional structures that are stored in the lexicon. Sentences in the active and passive voice are not related by transformations. Transformation are, however, retained for the description of unbounded dependencies as in WH-movement.

83.  Syntax and Psychology

The processing motivation behind Bresnan’s proposal is that it is easier to look up information in the lexicon than to compute a structure in the sense of ‘undoing’ a transformation. Slobin’s (1966) result was that English speaking children between 6 and 7 years of age understand active sentences better than passive sentences when agent and patient are semantically reversible, (as in The boy hit the girl, but not in The boy ate the apple). The DTC seems to wrongly predict a complexity effect also for non-reversible sentences such as The apple was eaten (by the boy). Bresnan argues that in the comprehension of passive sentences the syntactic complexity is always the same as in active sentences, but that in semantically reversible passives such as The girl was hit (by the boy) the mapping of syntactic structure onto a functional template in the lexicon may be more difficult. Thus, a complexity effect is predicted only for reversible passives, while DTC would predict such an effect throughout. With the same logic, though, a processing difficulty could likewise be predicted for reversible active sentences. Such a difficulty seems to be absent, however. Other aspects also indicate that the relation between grammar and parser must be seen with more caution. Forster (1979, 4 7) argues that the Slobin results may not be relevant for on-line processes at all, because the matching of sentences with pictures is likely to be confounded with general cognitive processes that cannot be equated with the decisions of the syntactic processor. Forster and Olbrei (1973), by using an on-line method found indeed evidence that passive sentences are more complex than active sentences, and that the reversibility effect is largely irrelevant. In its further development. LFG has emphasized the importance of the use of lexical information in syntactical processing. Ford, Bresnan and Kaplan (1982) and Ford (1988) show how attachment options are guided by the lexical force of a head to license an argument. In the following sentences (1) The woman wanted the dress on the rack (2) The woman positioned the dress on the rack it is the choice of the verb that determines that the PP is preferentially attached into the NP in (1) i. e., [NP ... [PP ...]], and into the VP in (2) i. e., [VP ... [PP ...]]. A number of processing principles are proposed which together with the grammar design of LFG

1499

make predictions about what the preferred attachment site of a phrase is. The applicability of these principles rests to a high degree on the choice of a particular lexical item, although certain default rules may apply once the lexical structure leaves the choice in limbo. It is an open question how a lexically driven processor would deal with SOV-languages like German (in embedded sentences), Hindi, Japanese, Turkish etc. in which the lexical head follows its complements and adjuncts rather than preceding them (see section 4. for more discussion). For LFG, the grammar/parser relation seems rather unproblematic, because the burden of computationally problematic transformation has been removed in favor of a more elaborate system of lexical structure and lexical look-up. In this way it is possible that the parser directly incorporates the grammar. The parser can in principle use all the information that becomes available online as lexical items are received, and their lexical templates can be consulted. Preferences are captured by parsing principles as indicated above. 3.4. Generalized Phrase Structure Grammar One of the initial motivations for the development of Generalized Phrase Structure Grammar (GPSG (cf. also article 25 in volume I) also rested on the problems that have been observed in connection with the DTC. To remedy the disagreement between the standard model of grammar, namely classical transformational grammar, and certain processing results, Gazdar (1981) suggested a theory of phrase structure that is enriched in such a way that it can handle movement — up to the treatment of non-local dependencies — without transformations. The parsing argument in favor of GPSG runs as follows: We know that humans can parse language rapidly. Context-free languages can be parsed in a time which is at least proportional to the cube of the length of the sentence. For grammars with transformations such a result does not exist, and efficient parsing may be impeded. Therefore natural language must be context-free. How are phenomena captured by GPSG which have classically been taken as chief arguments in favor of the transformational account such as trans-clausal WH-movement? The trick which allows abandoning transformations lies in the use of complex symbols. Instead of expanding a basic or underlying structure such as [S NP [VP

1500

V WH-NP]] which is subsequently manipulated by a transformation that yields [WHNPi [S NP [VP V ei]]], where ei is an empty placeholder symbol, GPSG employs a more elaborate rule system that allows direct generation of “ transformed” sentences. The transformation is encoded in the phrase structure rules themselves: S/NP → NP VP/ NP is derived by meta-rules from simplex phrase structure rules such as S → NP VP; this derived rule directly generates sentences in which an NP is missing, and in which it is encoded that this NP bears the grammatical relation of an object of the verb. Abstracting away from certain details, an embedded WHclause as in (I wonder) who Mary likes is then directly generated by the rule S’ → NPWH S/ NP. The category annotated with the feature for WH satisfies the slashed category S/NP and every slashed category dominated by it. As Torris (1987) points out, at least the earlier model of GPSG emphasized the fact that there is no distinction between a productive component of the grammar, namely the phrase structure rules, and a restrictive component, namely filters (see Chomsky and Lasnik 1977) or abstract principles as encoded in the various modules of GB-theory such as the Case-filter, the Theta criterion etc. (see Chomsky 1981). The question in the present context is whether there is psycholinguistic motivation behind GPSG that goes beyond the mathematical questions concerning the Chomsky hierarchy. Advocates of GPSG have usually been silent about the psychological aspects of their model. The underlying idea, however, must be almost the opposite of the LFG idea, namely that computation by means of phrase structure rules must be rather economical. Gazdar (1981) mentions, however, that psycholinguistic results suggest that the use of slashed rules bears a higher processing load as compared to the use of unslashed rules. Subject extractions are less costly than object extractions. In GPSG this follows because subject extractions do not involve slashed categories at all. Apart from this, the predictions for processing are not easily distinguishable from those of transformational grammar. Crain and Fodor (1985) take the uniformity aspect of GPSG as an advantage over Government and Binding (GB) Theory (see 3.5. below). They do not present direct evidence in favor of GPSG, but rather show that certain psycholinguistic evidence that

XXII. Syntax und Nachbardisziplinen

has been adduced against a GPSG-type parser (and grammar) cannot be upheld. Freedman (1982) as well as Frazier, Clifton and Randall (1983) argue on the basis of experimental data that the parser’s operations in structure building differ from its operations in checking grammatical principles, and that as a consequence a grammar/parser that draws a line between rules and principles has to be favored over GPSG. Freedman gave her subjects a sentence matching task in which first a sentence appeared and stayed on a screen which was then followed either by the same sentence or by a different sentence. The subjects had to give a timed same/ different response. One class of sentences involved what were then taken to be rule violations such as agreement violations, the other class involved violations of syntactic principles (such as the Specified Subject Constraint). The finding was that it took longer to match ungrammatical sentences like (1 a) that violate agreement rules than ungrammatical sentences that violate bounding principles such as (1 b): (1a) *Mary were writing a letter to her husband (1b) *What did you buy Picasso’s painting of? In fact, matching times for sentences like (1b) were about the same as those for fully grammatical sentences. The result was taken as a reflex of the different status of rules and principles in the human sentence processing system. Crain and Fodor (1987) argue, however, that the Freedman results do not permit such far-reaching conclusions. According to them, the examples of supposed rule violations can be spontaneously corrected by the subject while the other examples cannot. Shifting were to was in (1a) is obviously a simpler process that is accessible to almost anybody than trying to find the defect in (1b), a task that requires some training in linguistics. Frazier, Clifton and Randall (1983) provided experiments which suggest that structure building is procedurally different from gap filling (or more generally indexing of empty categories such as WH-traces and PRO in GB-Theory). According to them, the parser is likely to index empty categories in the first pass according to a Most Recent Filler Strategy (MRFS). In the following examples, the first empty node (indicated by []) in sentence a. is compatible with MRFS, but not in b.:

83.  Syntax and Psychology

(2a) Everyone liked the woman whoi the little childj started []j to sing those stupid French songs for []i last Christmas (RECENT FILLER) (2b) Everyone liked the woman whoi the little child forced []i []i to sing those stupid French songs last Christmas (DISTANT FILLER) The task was to read these sentences word by word and give a timed response which signals that the sentence has been comprehended. The result was that the RECENT FILLER sentences were responded to faster than the DISTANT FILLER sentences, and that the latter also yielded lower comprehension ranks. (Similar results were obtained with sentences that involve control verbs such as beg, which are ambiguous between a transitive and an intransitive reading.) The parser seems to temporally dissociate a first pass analysis (using the MRFS as a kind of default case) from the subsequent correct indexing of empty categories. This temporal dissociation is taken to reflect the modular structure of the grammar that is implemented. If this conclusion is justified it would speak against GPSG, where according to Crain and Fodor gap-filling goes hand in hand with structure building, while the control of an understood PRO-subject is entirely shifted to the semantics. A point of critique could be that the extensive use of rules does not lead to a theory that would adequately characterize the acquisition and processing of language. For a child that acquires language it seems to be cumbersone to learn idiosyncratic rules. Defending GPSG as a plausible model, Fodor (1990) proposes that learning and parsing rest on a process that strips away features from individual rules such that what is used in the end is a universal rule much like the expansion schema suggested by X-bar theory. In order for this reduction to work it must be assumed, however, that the input gets an initial structural description. How this should be possible without the use of rules remains unclear. But assuming that there are rules that can structure the input, why should there be the additional luxury of stripping away certain features? We will come back to the issue of rule systems versus the principle-and-parameters approach of GB-theory. For the time being it can be stated that GPSG has so far not born a psycholinguistically motivated theory of

1501

processing or acquisition. The psychological issue rests entirely on the familiar argument about context-freeness. 3.5. Government and Binding Theory The most influential framework in modern psycholinguistic research is Government and Binding Theory (also known as Principles and Parameters Theory; cf. also article 24 in volume I) as developed in Chomsky (1981) and its successors (see Chomsky’s (1986a) theory of barriers and the economy of derivation approach of Chomsky (1991) and his minimalist program (1993). This is so despite the fact that psycholinguistics had been misled by the DTC that was assumed for earlier stages of generative transformational grammar. GB-theory has changed considerably in comparison with the standard theory of Chomsky (1965). Most importantly, it has turned from a system of rules to a system of principles and parameters. Classical GB consists of X’-syntax as a projection/generation system that expands structure according to the Projection Principle (which requires that lexical/thematic information is preserved on each level of syntactic representation). In addition to that, several subcomponents filter the derivation of a sentence on the way from D-structure to LF. These components are thematic theory, case theory, binding theory and a theory of control. Variation in the order of constituents (in a single language) is subject to a generalized transformation known as move-α where α is a variable for any constituent. This movement leaves empty categories (traces; in psycholinguistics often referred to as gaps) which keep a record of the transformational history. Move-a is under the control of structure preservation and locality. The former says that landing sites of movement must match the categorial status of the item to be moved i. e., an NP may only move to an NP-position, a head only to a head position etc. The second says that movement can only proceed in local domains that guarantee the determination of the content of empty categories. The conceptual advantage of GB-theory and its successors is that the system has changed from a theory of construction specific rules like PASSIVE, MOVE-WH, SUBJECT-AUX-INVERSION etc. to a theory that captures the respective phenomena through the interaction of universal principles with a number of parameters which classes of languages follow (headinitial versus head-final order, null versus

1502

overt subjects etc.) and language particular properties such as richness of inflectional morphology etc. GB has offered a new perspective on comparative grammar. It is not obvious how universals or parametric choices could be expressed in a rule based grammar given that rule-based grammars will always be language specific. The GB-approach has been quite successful in comparison because it is now possible to strip variation as given by lexical idiosyncracy and parametric choice away from the universal base. GB-theory is a characterization of UG. As pointed out already in section 2.3., its goal is to approach an abstract theory of innate linguistic knowledge that permits the child to construct the grammar of an individual language L when exposed to data from L. The question for language use is in which sense GB as a highly abstract competence system is connected to the processor. If a parser runs something like a linguistically defined grammar, then of course, it will run a particular grammar in the sense that at least a certain lexicon and certain parameters must have been chosen. Advocates of rule based grammars seem to have the advantage of a transparent relation between grammar and parser. Bresnan and Kaplan (1982), for example, have such a tight link in mind when they suggest their strong competence hypothesis according to which the representational basis of the processor is isomorphic to the competence grammar. According to Stabler (1983), however, it is not necessary that grammars are used directly; it might be more plausible that processors use programs that are “ hardwired” rather than explicitly represented. This is more in line with the way researchers on parsing in the GB-tradition usually approach the grammar-parser relation. A famous pre-GB version of a parser for the Extended Standard Theory (EST) is PARSIFAL, also known as the Marcus Parser (Marcus 1980), which implements the grammar in such a way that the output will show annotations of strings that are equivalent to the annotations provided by EST. The parser does not, however, use rules of EST directly, but uses so-called ‘pattern-action rules’ which yield a comparable structural result. NPmovement, for example, has as its parsing equivalent the direct insertion of an NP into the phrase marker and the creation of a coindexed trace after it is realized that the following verb is of the seem class or shows passive morphology. Following Mar-

XXII. Syntax und Nachbardisziplinen

cus (1980), Berwick and Weinberg (1984 ) argue for an indirect mapping between grammar and parser. According to them, the parser may use a covering grammar G2 which generates a language that is weakly equivalent to the language generated by the competence grammar Gl. If this assumption is made, however, there is no necessary opposition between GB and its more or less rule based rivals. They could be seen as different stages in the same enterprise whereby the latter show a transparent relation to the actions performed by the parser. The predicate psychological reality of a linguistic theory like GB may then be an attribute which is quite different from the attribute psychologically real that one may use for a parser that naturally mimics human performance in sentence comprehension. With respect to grammar per se, Matthews (1991, 197) suggests that what is psychologically real is “ not the linguistic constructs that figure in a grammar, but the function that these constructs serve to specify”. A linguistic construct that figures in the grammar would, for instance, be Case theory; but Case theory would, according to this view, never be active in linguistic performance; what could be active are specific rules that are licensed among other things by Case theory. Not all GB-oriented research on parsing presupposes this to be true or desirable though. Pritchett (1992, 155) makes the strong claim that “ approaches to parsing not formulated directly in terms of grammatical principles are cognitively insufficient”, and “ human natural language processing may be characterizable wholly in terms of a grammar, where the grammar is viewed not as a system of rules, but as a set of conditions on representation, and where the parser applies the principles of grammar at every point during the processing of the input string.” Since Pritchett does not have an implemented parsing model, one has to more or less believe that this program can be worked out. Notice, however, that all parsers that can be subsumed under the so-called Principle-Based approach need devices of various sorts to prevent computational intractability. Kolb and Thiersch (1991, 265), for instance, note that X’-theory with binary branching and Chomsky-adjunction to any X’-level would allow for more than 35 000 different structures for the unspectacular embedded German sentence daß Karl den Hund schlug (“ that Karl beat the dog”) in case only prece-

83.  Syntax and Psychology

dence and dominance are checked and no principles have applied yet to inspect the licensing of possible empty categories. One can then reason about ways to remedy the problem, but certainly no trivial application of GB-type modules will solve the problem. GB-theory is certainly the most attractive framework of grammar if the perspective is a deeper understanding of UG and language learning, but it is far from clear that in its derivational form with various levels of representation it is simultaneously a competence grammar that could be directly used by a syntactic parser. To achieve the goal of efficient parsing, various changes seem to be necessary that make it less obvious that the design of the grammar carries over to the parser.

4.

On the Nature of the Syntactic Processor

According to the results reported in Fodor, Bever and Garrett (1974 ), the DTC has made it clear that it is very difficult to maintain a direct map between grammars and parsers. In order to understand sentence parsing it is not enough to know what the grammar looks like, but it is likewise necessary to find out what the parsing specific principles are which guide the on-line analysis of sentences. Since then the study of parsing principles has become one of the liveliest areas of syntactic research on human language processing. The question is how sentences are perceived word by word in such a way that efficiency is guaranteed i. e., semantic intepretation is not delayed and revisions are kept at a minimum. While research in this area had usually been built on intuitive judgements or off-line i. e. post-hoc methodologies, the last twenty years of research have developed techniques which allow more or less reliable snapshots of the process of sentence comprehension. One of these techniques is rapid serial visual presentation (RSVP) where subjets see a sentence “ develop” word by word at a speed that is determined by the examiner, and where they can respond to something in the input or something accompanying the input. Closely related to this is self-paced reading, a technique by which it is the subject who determines the point of time when a new fragment (word or phrase) of the input appears on a screen. In this way, exact records can be taken of the time required for reading parts of a sentence.

1503

A more elaborate method is the recording of eye movements while a subject reads a text (see also section 2.1.). In the auditory domain it is possible to use monitoring techniques of various sorts; subjects hear a text and have to react to certain input upon instruction (like “ press a key when the text becomes anomalous”) or have to deal with superimposed tasks like word recognition at certain choice points in the parse. In recent years non-intrusive techniques became popular by which records of event related brain potentials (ERP) are taken during processing. The latter technique suffered from the problem that it was uncertain what the electrophysiological manifestation of processes specific to syntax are, but there is now evidence for what is called the syntactic positive shift (SPS) (Hagoort, Brown and Groothusen, 1993) i. e., a wave form that is specific to agreement and subcategorization violations and can be distinguished from a semantic negative shift, the N4 00, that is observed when lexical semantic normality is violated (see Kutas and van Petten, 1988). Techniques of this kind have allowed psycholinguists observations which give rise to a fairly clear picture of human sentence perception, although many aspects of these results still leave room for theoretical alternatives. The basic insight is that decision making about phrase structure seems to be fast and automatized, while semantic decisions, say about pronominal reference, control relations and lexical desambiguation can be postponed and revised if necessary; see Bierwisch (1983) for a wealth of relevant observations. As the earliest experiments on recall and STM have shown, it is extremely important that the input is immediately structured in one way or the other. Take a row of numbers as in (1). (1) 7 5 3 4 1 3 8 2 5 ... This row is easier to store in memory if it appears in chunks as in (2) or (3): (2) 75 34 13 82 5 ... (3) 753 413 825 ... This is elementary syntax, and it may in this form not be unique to language, but seems to occur also in visual and auditory (e. g. musical) gestalt perception. Humans hate to store something in their mind that is unstructured and thus cannot be interpreted (in the broadest sense). The reason must be that there is little room for storing. Linguistic in-

1504

put wants to be structured and interpreted exhaustively as soon as possible such that the STM can be cleared again. The German examples (4 ) through (6) are syntactically wellformed but almost totally incomprehensible due to an overload of center embedding:

As the English translations indicate, such phrases are easy to comprehend once they are dissolved into chunks that can be cleared out of STM before the parsing process continues. The length of sentences per se seems to be no problem. A sentence with multiple right-branching embeddings causes no problem at all:

According to Kimball (1973) such a sentence is roughly parsed like three simplex clauses. There seems to be no doubt that the mind has at least two stores, one which builds syntactic structures, and another one which preserves them as interpreted (semantically recoded) units. The latter are hardly or not at all accessible for further syntactic (phrase structural) manipulations. Important evidence to this extent is the phenomenon of closure (cf. Kimball 1973, Frazier and Fodor 1978). Closure (also called ‘right association’) means that

XXII. Syntax und Nachbardisziplinen

new input is preferably attached into the last constituent under construction, which is normally the one on the “ low right” side of a phrase structure tree. Due to the principle of Closure, the last phrase in (8a) through (10a) cannot be attached as in (8b) through (10b):

Another property that has been proposed as a principle of human sentence processing is minimal attachment (MA) (Frazier 1978, Frazier and Fodor 1978, Frazier and Rayner 1982 etc.). It roughly says: New input has to be attached into existing structure using the smallest possible number of non-terminal nodes. MA must be distinguished from Closure, although the two may occasionally overlap; Closure is obviously related to STM while MA is not. MA, however, is also a principle of parsimony. Under conservative assumptions about phrase structure, MA predicts that the initial parse will favor (11a) over (11b) and that (12) is at first sight an odd sentence: (11a) John [VP bought [NP the book] [PP for Susan]] (11b) John [VP bought [[NP the book] [PP for Susan]]] (12) John [VP sent [[NP the letter] [PP to Susan]] [PP to the dean]] Obviously, the parser attaches the PP directly into the tree under construction, and it seems to do this in such a way that the PP ends up as part of the (now ternary branching) VP. Adjunction to NP would lead to one extra non-terminal, and is consequently a less fa-

83.  Syntax and Psychology

vored option. At least three theoretical positions have been discussed in connection with MA: (I) Gardenpath Theory: GP-Theory as mainly developed by Frazier and colleagues claims that the human syntactic processor starts out with minimal assumptions about the structure and constructs a tree top-down as compatible with the input available. In those cases where the initial assignment has to be revised as in (13), where to Susan has to be removed from the VP and attached into NP, there is a short-lived GP. Long-lived GPs arise only in certain cases like the famous examples from Bever (1970): (13a) The horse raced past the barn fell (13b) The boat floated down the river sank where the parser analyzes the input in the sense of a simple NP + VP clause and then crashes when the verb (fell or sank) appears because it has gone “ too far” in a very obvious analysis to be able to retract to the proper analysis NP + rel. clause + VP in which the last element would constitute the VP. (II) Context Theory: An alternative explanation of GP is provided by researchers who doubt that initial parsing is purely syntactic. Crain and Steedman (1985), Altman and Steedman (1988), Steedman and Altman (1989) argue that even when no context is provided, sentences are interpreted in a context (in the sense of a mental model set up by the hearer). In order to assume reference to a complex NP like the book for Susan it would be necessary to have a context which motivates this reference. This would, for instance, be a context in which several books are under discussion. Since such a context is normally not provided, the listener will not expect such a structure. Crain and Steedman, thus, reduce the GP-effect to a failure of what they call referential success. They argue that under appropriate context conditions even severe GPs like those in (13) would become parsable. (III) Thematic GP Theory: Another purely syntax based explanation for GP-phenomena is proposed in Pritchett (1988; 1992) which, however, tries to rely on syntactic principles of GB theory alone. The idea is that the appearance of a head dictates that its selection properties (θ-grid) should be respected as soon as possible. Since in (11) and (12) there is a verb that is compatible with or even requires a PP, the preferred option would be

1505

attachment into VP. The reason here is different from the one in (I). It is the presence of the head that drives attachment, while it is node-parsimony that does so according to MA. Pritchett’s theory also takes care of a difference in strength of GP-effects: Reattachment is easy when it remains in the original (θ-)licensing domain, but is impossible when the target position lies outside. (14 ) involves at worst a mild GP, while the GP-effect in (15) is strong; [] signal the true phrasal boundaries, and {} the GP-induces boundaries: (14) John {knew [the girl} was nice] (15) While Mary {sang [a song} played on the radio] In (14 ), the girl has to be “ lowered” from its direct object position (i. e. from the {}-domain) to become the subject of the embedded clause (the []-domain), but it is only reattached into the object of knew;in (15), however, a song would have to be reattached from an adjunct clause into the main clause. Various more fine-grained distinctions and mixtures of these approaches are under discussion, but the common core seems to be clear enough. It is undeniable that the human parser (i) builds structure which is compatible with the rules and principles of grammar (and possible discourse organization) and that (ii) in the absence of evidence to the contrary, not more structure than absolutely necessary is built. The human sentence processor can, thus, be characterized as intelligent but lazy. As Frazier and Fodor (1978) have pointed out, there is no logical alternative to the principle of laziness simply because of the recursive nature of syntax. Take a substring like (16) John saw Mary’s ... here the parser will hypothesize that what follows is some noun, but the sentence can, of course, continue as (17) brother’s ... little fat ... dog that always tries to bite him ... There is no sense in making more than minimal guesses about the input as dictated by the grammar because the expansion that may follow is logically infinite. Parsing research that is based on primarily head-initial languages like English suffers from the problem that it is hard to distinguish between strong top-down processes as

1506

assumed in (I) and strong bottom-up processes as assumed in (III) or other theories that rely heavily on projection from the lexicon. In English the lexical head is always available when its minimal maximal phrase is built. This is not the case in head-final language though. In languages like German, Hindi, Japanese or Turkish it is impossible to decide with absolute certainty about the thematic function a certain XP will ultimately end up with before the head appears which licenses it. The examples in (18) are from Inoue and Fodor (1993):

Up to the appearance of the verb the structure is locally ambiguous between a simplex clause as in (18b) or some other more complex structure as in (18c). According to Inoue and Fodor, Japanese listeners are surprised at the appearance of tabeta when (18a) continues as in (18c). Obviously certain commitments have already been made before the head has appeared. This is incompatible with strict forms of head-drivenness or delayed structure building. (see Inoue and Fodor 1993 for discussion of various models in connection with processing Japanese). Experimental results to this effect have been obtained for Dutch (Frazier 1987) and German (Bader 1990, Bader and Lasser 1994 , Bayer and Marslen-Wilson 1992) where the verb appears clause-finally in embedded sentences. With respect to the processing of head-final phrases, Frazier (1989) strongly argues against lexically driven parsing. She notices that there is a moment of difficulty in the Dutch sentence

XXII. Syntax und Nachbardisziplinen

because the parser has taken van Holland as part of VP at a time where the head of VP is not available yet. In spite of the delayed appearance of the verb, listeners build structure on the way to the verb, and there is a measurable effect at the point where the initial minimalist assignment turns out to be discouraged, and structural revision becomes unavoidable. Assume counterfactually that the parser simply remains idle until the head is available. This would not only be incompatible with what is known about STM, but would in addition be quite uneconomical. The information that the head provides is certainly of central importance, but it is clearly not the only reliable information. Notice the case morphemes in the Japanese examples in (18), or in German where the case paradigm offers various cues for the on-line hypotheses about the grammatical functions NPs may bear. This information is not waterproof as the following examples from the German show:

The form der Königin can be genitive or dative. Thus, the parser has a good chance to go wrong and have to revise the initial analysis after a longer stretch of input. Nevertheless, it would be surprising if the information available would not be made use of. For those languages where the head is immediately available when a phrase is being constructed, it is plausible that the processor will immediately exploit the subcategorization information provided by it. Frazier (1989) suggests, however, that even then it is possible that phrase structure is expanded before the head’s subcategorization properties are used. Mitchell (1987) found evidence that in self-paced reading subjects initially ana-

83.  Syntax and Psychology

1507

lyzed a postverbal NP as a direct object even if the verb is obligatorily intransitive as in: (21) After the audience had {departed the actor} sat down for a well deserved drink Note, however, that the first frame in Mitchell’s presentation contained the phrase ... departed the actor which in all likelihood will have introduced a bias in favor of the result obtained. Apart from this, the assumption of strong autonomy of structure building leads to intricate questions about the speed of lexical access and how it interacts with decisions in phrase structure expansion. A lot of evidence in this area suggests that lexical information is used immediately (see for instance Tanenhaus and Carlson 1989 and Gorrell 1991). This indicates that the parser may be opportunistic, using the available information that is consistent with the internalized grammar as parsimoniously as possible. For head-initial languages it is often difficult to decide what is going on exactly. Therefore, the real challenge for refining the existing parsing models comes from head-final languages. The little evidence that is available at present suggests that the processor makes immediate use of structural information irrespective of whether it is provided by the head or by other constituents. It seems to be universally true that the syntactic processor is not only intelligent and lazy, but in addition engages in guesses at the risk of getting things wrong. The preceding discussion was about phrase structure. What about all the other aspects syntactic theory has to deal with, — aspects that have a clear relation to semantic interpretation and discourse like scope and pronominal reference? There is no doubt that these are under the control of tree geometry. And there is no doubt that decisions very often have to be delayed. Consider pronominal reference in the following German examples from Bierwisch (1983):

The reference of sie remains ambiguous almost until the end of the entire clause, but there is not the slightest impression of a processing difficulty that would make these cases comparable with ambiguities or revisions in phrase structure parsing. Consider also the delay in the interpretation of the empty subject PRO in the examples in (23):

Until the control verb appears, the structure remains ambiguous, and full semantic interpretation (e. g. for truth conditions) is impossible. Again there is no trace of a conscious processing difficulty. The processor seems to be quite comfortable with this situation. Further examples to this extent proliferate (see Bierwisch 1983). There is a remarkable difference between this situation of semantic assignment where delay is no problem, and the situation of phrase structure building in which the parser is at pains to immediately decide on structure. How can this difference be explained? The obvious answer rests in the initial classical observation that the SMT puts pressure on the linguistic system to immediately assign structure to the input. In this situation, the primacy of phrase structure is unavoidable because it is simply impossible to exploit unstructured words for phrase level semantic interpretation. This still leaves the question why semantic assignment can wait, (and it is of course not clear that it always can). The properties one has ascribed to the syntactic parser can be seen as a compromise between efficiency and the pressure under which the system works. Efficiency is gained

1508

when the initial parse is not contradicted by later information, and the structure can be used for exhaustive interpretation. The pressure deriving from STM requires structural decisions without delay, and this, of course, introduces the risk of going wrong. Semantic processes of assigning pronominal reference etc. enjoy some more luxury as they can obviously be delayed if necessary. Since the input to these processes is syntactically structured, it is an interesting question how it is stored and in which time course it decays. After phrase structure has been built there is obviously a stage in processing at which initial (partial) semantic assignment can to some extent be manipulated. This state cannot be equated with long term memory because it clearly requires a verbatim representation of the input. Take the following example of an interrogative in Chinese from Huang (1982): (24) Zhangsan xiangxin [shei mai-le shu] Zhangsan believes who bought books ‘Who does Zhangsan believe bought books?’ It is standardly assumed that the word that corresponds to the Wh-phrase in English must have wide scope at the level of Logical Form (LF) because otherwise the lexical requirement that verbs like “ believe’ select a [-WH]-clause (which of course holds in Chinese as well) would not be met, and the sentence would be ungrammatical. But consider on-line processing now: In the absence of evidence to the contrary, the Chinese listener will most likely not anticipate an interrogative interpretation before the word shei is met. The stretch Zhangsan xiangxin ... must therefore be stored in such a way that its declarative mood remains open for change. If it were in a permanent non-linguistic memory store, the continuation shown in (24 ) could presumably not be integrated. The problem is not trivial as the Wh-element can in principle appear infinitely far to the right. LF has become an important issue in syntactic research because it has turned out that many of the processes of variable binding and scoping of operators are restricted by principles known from the syntax of movement (or similar implementations). It is for this reason that syntacticians use the notion movement when talking about the determination of the scope of a quantifier (also a Whquantifier that remains phonetically in situ). But there is one important difference: Overt movement requires the on-line processing of

XXII. Syntax und Nachbardisziplinen

filler-gap dependencies. There is some evidence that the parser — working on minimalist assumptions about the imput — associates the filler with the first position in which the grammar would permit a gap, even if it subsequently turns out that this gap was only spurious. For LF-scope assignment the picture is not like that, because — apart from the syntactician’s façon de parler — there is no filler and there is no gap which could be relevant for on-line processing. There is no displacement that would have a repercussion at the phonetic side of the grammar. For online processing this means that constraints on LF-“ movement” have to be respected after phrase structure parsing or at least after the onset of the phrase structure parsing process. It is thus plausible to assume a state of representation in which the verbatim syntactic form of the sentence is still available such that the scope of operators can be manipulated according to the known syntactic constraints on movement. Linguists have noticed, however, that there are preferences. May (1977), Fodor (1982), Reinhart (1983) and others observe that there is a tendency to interpret sentences with two quantified NPs in roughly such a way that the first has scope over the second. An on-line processing study of Kurtzman and MacDonald (1993) with the determiners every and a revealed, however, that this is only true for active sentences like (25a), but not for passives like (25b). In complex NPs like (25c), they found a strong preference for wide scope of the lower NP: (25a) Every kid climbed a tree ∀∃ (25b) Every tree was climbed by a kid ∀∃/∃∀ (25c) The nurse called every friend of a patient ∃∀ Parsing seems to be accompanied by the online creation of a mental model (JohnsonLaird 1983, Fodor 1982), but this cannot mean that the model cannot be altered as new information becomes available. Otherwise, linear order would dictate relative quantifier scope more strongly than it actually does. It needs far more research than presently available to determine with more certainty how phrase structure parsing lines up with syntax-based semantic decisions as characterized at the level of LF; at the same time it

83.  Syntax and Psychology

is almost certain that a simple identification of all these processes is wrong, and that the temporal domains in which information is used may differ accordingly. If this turns out to be true, the on-line creation of phrase structure has priority over delayed (or delayable) processes, and it may therefore be the only syntactic process that has the properties Fodor (1983) has claimed to hold for modularity.

5.

Conclusion

We have asked in the beginning of this article whether there is evidence for syntax as a circumscribed mental ability in humans. It seems to be fair to answer this question affirmatively, not only in the general sense that anything in language has to be somehow mental, but in the specific sense that there are very detailed patterns of spontaneous as well as experimentally elicited behavior that reflect syntactic structure. The second question asked for the nature of the syntactic processing system and the process of syntactic development. With respect to the latter it seems to be accepted by almost everyone in the field that the child is born with rich internal preconditions to acquire language. The process of syntax acquisition shows regularities which are easiest explained with increasing access to the internal preconditions of UG through general cognitive maturation and/or maturation of principles of grammar. Learning syntax in the literal sense seems highly impossible. With respect to the mature syntactic processing system, many aspects have found a satisfactory answer in the last twenty years. It is more or less agreed upon today that this system must be highly automatized, hardly open to conscious inspection, and linked to a system of internal tacit knowledge; this makes the processing system highly efficient and robust. Research on parsing preferences has yielded many insights into the architecture of this system that were unthinkable before. Anyone should be aware that this progress could not have been made without the contribution of syntactic theory and the sophistication that has meanwhile been reached in this domain of research.

6.

References

Altman, G., and M. Steedman. 1988. Interaction with context during human sentence processing. Cognition 30. 191—238. B ach, E., C. B rown and W. D. Marslen-Wilson.

1509

1986. Crossed and nested dependencies in German and Dutch: A psycholinguistic study. Language and Cognitive Processes 1. 249—262. B ader, M. 1990. Syntaktische Prozesse beim Sprachverstehen: Theoretische Überlegungen und experimentelle Untersuchungen. MA-thesis. Universität Freiburg i. Br. —, and I. Lasser. 1994 . German verb-final clauses and sentence processing: Evidence for immediate attachment. Perspectives on Sentence Processing, ed. by Ch. Clifton, L. Frazier & K. Rayer, Hillsdale, NJ. B ates, E., and B . MacWhinney. 1982. Functionalist approaches to grammar. Language acquisition: The State of the Art, ed. by E. Wanner & L. Gleitman, Cambridge. B ayer, J., R. de B leser and C. Dronsek. 1987. Form und Funktion von Kasus bei Agrammatismus. Grammatik und Kognition (Sonderheft 1 der Linguistischen Berichte), ed. by J. Bayer, Opladen. —, and W. Marslen-Wilson. 1992. Configurationality in the light of language comprehension. MS. Universität Wien and Birkbeck College London. B erndt, R., and A. Caramazza. 1980. A redefinition of the syndrome of Broca’s aphasia: Implications for a neuropsychological model of language. Applied Psycholinguistics 1. 225—278. B erwick, R., and A. Weinberg. 1984 . The grammatical basis of linguistic performance. Cambridge, MA. B ever, T. 1970. The cognitive basis for linguistic structures. Cognition and the Development of language, ed. by J. Hayes, New York. Bickerton, D. 1981. Roots of language. Ann Arbor. —. 1984 . The language bioprogram hypothesis. The Behavioral and Brain Sciences 7. 173—221. B ierwisch, M. 1983. How on-line is language processing? The Process of Language Understanding, ed. by G. Flores d’Arcais, G. Jarvella & R. Jarvella. New York. de B leser, R. 1987. From agrammatism to paragrammatism: German aphasiological traditions and grammatical disturbances. Cognitive Neuropsychology 4. 187—256. —, and J. B ayer. 1986. German word formation and aphasia. The Linguistic Review 5. 1—40. B ock, K., and A. Kroch. 1988. The isolability of syntactic processing. Linguistic structure in language processing, ed. by G. Carlson & M. Tanenhaus, Dordrecht. B oomer, D., and J. Laver. 1968. Slips of the tongue. The British Journal of Disorders of Communication 3. 2—12. B orer, H., and K. Wexler. 1987. The maturation of syntax. Parameter setting, ed. by T. Roeper & E. Williams, Dordrecht.

1510

B radley, D., M. Garrett and E. B . Zurif. 1980. Syntactic deficits in Broca’s aphasia. Biological studies of mental processes, ed. by D. Caplan, Cambridge, MA. B resnan, J. 1978. A realistic transformational grammar. Linguistic theory and psychological reality, ed. by J. Bresnan, M. Halle & G. Miller, Cambridge, MA. —, and R. Kaplan. 1982. Introduction: Grammars as mental representations of language. The mental representation of language, ed. by J. Bresnan, Cambridge, MA. Brown, R. 1973. A first language. Cambridge, MA. —, and C. Hanlon. 1970. Derivational complexity and order of acquisition in child speech. Cognition and the development of language, ed. by J. Hayes, New York. Caplan, D. 1987. Neurolinguistics and linguistic aphasiology. Cambridge. —, and N. Hildebrandt. 1988. Disorders of syntactic comprehension. Cambridge, MA. Chomsky, C. 1969. The acquisition of syntax in children from 5 to 10. Cambridge, MA. Chomsky, N. 1957. Syntactic structures. The Hague. —. 1959. A review of Skinner’s Verbal B ehavior. Language 35. 26—58. —. 1965. Aspects of the theory of syntax. Cambridge, MA. —. 1973. Conditions on transformation. A Festschrift for Morris Halle, ed. by S. Anderson & P. Kiparsky. New York. —. 1975. Reflections on language. New York. —. 1986a. Barriers. Cambridge, MA. —. 1986b. Knowledge of language: It’s nature, origin and use. New York. —. 1991. Some notes on economy of derivation and representation. Principles and parameters in comparative grammar, ed. by R. Freidin. Cambridge, MA. —. 1993. A minimalist program for linguistic theory. View from building 20: Essays in linguistics in honor of Sylvain Bromberger, ed. by K. Hale & J. Keyser. Cambridge, MA. —, and H. Lasnik. 1977. Filters and control. Linguistic Inquiry 8. 425—504. Crain, S. 1991. Language in the absence of experience. The Behavioral and Brain Sciences 14 . 597— 640. —, and J. D. Fodor. 1985. How can grammars help parsers? Natural language parsing, ed. by D. Dowty, L. Karttunen & A. Zwicky. Cambridge. —. 1987. Sentence matching and overgeneration. Cognition 26. 123—170. —, and M. Steedman. 1985. On not being led up the garden path: the use of context by the psycho-

XXII. Syntax und Nachbardisziplinen

logical syntax processor. Natural language parsing, ed. by D. Dowty, L. Karttunen & A. Zwicky. Cambridge. Curtiss, S. 1977. Genie: A psycholinguistic study of a modern-day “ wild child”. New York. Clahsen, H. 1982. Spracherwerb in der Kindheit. Tübingen. Davis, L., N. S. Foldi, H. Garner and E. B . Zurif. 1978. Repetition in the transcortical aphasias. Brain and Language 6. 226—238. Deutsch, W., C. Koster and J. Koster. 1986. What can we learn from children’s errors in understanding anaphora? Linguistics 24. 203—225. Dogil, G., H. Haider, Ch. Schaner-Wolles and R. Husmann. 1993. Radical autonomy of syntax: Evidence from transcortical aphasia. submitted to Aphasiology. Dresher, B . E., and N. Hornstein. 1976. On some supposed contributions of artificial intelligence to the scientific study of language. Cognition 4 . 321—398. Evers, A. 1975. The transformational cycle in Dutch and German. Doctoral dissertation, Rijksuniversiteit Utrecht. Felix, S. 1987. Cognition and language growth. Dordrecht. Flores d’Arcais, G. 1982. Automatic syntactic computation of semantic information during sentence comprehension. Psychological Research 4 4 . 231— 242. Fodor, J. A. 1975. The language of thought. New York. —. 1983. The modularity of mind. Cambridge, MA. Fodor, J. D. 1982. The mental representation of quantifiers. Processes, beliefs and questions, ed. by S. Peters & E. Saarinen. Dordrecht. —. 1990. Parameters and parameter setting in a phrase structure grammar. Language processing and language acquisition, ed. by L. Frazier & J. de Villiers. Dordrecht. Fodor, J. A., T. B ever and M. Garrett. 1974 . The psychology of language. New York. —, K. Forster, M. Garrett and D. Hakes. 1974 . A relation between syntax and respiration. Quarterly Progress Report 113, Research Laboratory of Electronics. MIT. Ford, M. 1982. Sentence planning units: implications for the speaker’s representation of meaningful relations underlying sentences. The mental representation of language, ed. by J. Bresnan. Cambridge, MA. —. 1988. Parsing complexity and a theory of parsing. Linguistic structure in language processing, ed. by G. Carlson & M. Tanenhaus. Dordrecht. —, J. B resnan and R. Kaplan. 1982. A competencebased theory of syntactic closure. The mental rep-

83.  Syntax and Psychology

resentation of language, ed. by J. Bresnan. Cambridge, MA. —, and V. Holmes. 1978. Planning units and syntax in sentence production. Cognition 6. 35—53. Forster, K. 1979. Levels of processing and the structure of the language processor. Sentence processing: Psycholinguistic studies presented to Merrill Garrett, ed. by W. E. Cooper & E. C. T. Walker. Hillsdale, NJ. —, and I. Olbrei. 1973. Semantic heuristics and syntactic analysis. Cognition 2. 319—347. Frazier, L. 1978. On comprehending sentences: Syntactic parsing strategies. Ph. D. dissertation, University of Connecticut. Storrs, CN. —. 1987. Syntactic processing: Evidence from Dutch. Natural Language and Linguistic Theory 5. 519—559. —. 1989. Against lexical generation of syntax. Lexical representation and process, ed. by W. MarslenWilson. Cambridge, MA. —, Ch. Clifton and J. Randall. 1983. Filling Gaps: Decision principles and structure in sentence comprehension. Cognition 13. 187—222. —, and J. D. Fodor. 1978. The sausage machine: A new two-stage parsing model. Cognition 6. 291— 325. —, and K. Rayner. 1982. Making and correcting errors during sentence comprehension: Eye movement in the analysis of structurally ambiguous sentences. Cognitive Psychology 14. 178—210. Friederici, A. 1985. Levels of processing and vocabulary types: Evidence from on-line comprehension in normals and agrammatics. Cognition 19. 133— 162. Freedman, S. 1982. Behavioral reflexes of constraints on transformations. Doctoral dissertation, Monash University. Fromkin, V. 1971. The non-anomalous nature of anomalous utterances. Language 47. 27—52. Garrett, M. 1975. The analysis of sentence production. The psychology of learning and motivation, ed. by G. H. Bower. New York. —. 1980. Levels of processing in sentence production. Language production, ed. by G. Butterworth. Vol. I. London. Gazdar, G. 1981. Unbounded dependencies and coordinate structure. Linguistic Inquiry 12. 155—184. Geschwind, N., F. A. Quadfasel and J. M. Segarra. 1968. Isolation of the speech area. Neuropsychologia 6. 327—340. Gibson, E., and G. Hickok. 1993. Sentence processing with empty categories. Language and Cognitice Processes 8. 147—162. Goldman-Eisler, F. 1956. The determinants of the rate of speech output and their mutual relations. Journal of Psychosomatic Research 2. 137—143. Gorrell, P. 1991. Subcategorization and sentence

1511

processing. Principle-based parsing, ed. by R. Berwick, S. Abney & C. Tenny. Dordrecht. —. 1993. Evaluating the direct association hypothesis: A reply to Pickering & Barry (1991). Language and Cognitive Processes 8. 129—146. Grimshaw, J., and S. T. Rosen. 1990. Obeying the binding theory. Language processing and language acquisition, ed. by L. Frazier & J. de Villiers. Dordrecht. Grodzinsky, Y 1984 . Language deficits and the theory of syntax. Brain and Language 27. 135—159. —. 1990. Theoretical perspectives on language deficits. Cambridge, MA. —, and T. Reinhart. 1993. The innateness of binding and coreference. Linguistic Inquiry 24. 69—101. Hagoort, P., C. B rown and J. Groothusen. 1993. The syntactic positive shift (SPS) as an ERP-measure on syntactic processing. Language and Cognitive Processes. 8. 439—483. Inoue, A., and J. D. Fodor. 1993. Informationpaced parsing of Japanese. to appear in: Japanese syntactic processing, ed. by R. Mazuka and N. Nagai. Hillsdale, NJ. Jaeggli, O., and K. Safir (eds.). 1989. The null subject parameter. Dordrecht. Jakubovicz, C. 1984 . On markedness and binding principles. Proceedings of the 14 th North Eastern Linguistics Association Meeting, ed. by C. Jones & P. Sells. Amherst, MA. Johnson-Laird, P. 1983. Mental models. Cambridge, MA. Joshi, A. 1985. Processing of sentences and intrasentential code switching. Natural language parsing, ed. by D. Dowty, L. Karttunen & A. Zwicky. Cambridge. Kaplan, R. 1972. Augmented Transition Networks as psychological models of sentence comprehension. Artificial Intelligence 3. 77—100. Kean, M. L. 1977. The linguistic interpretation of aphasic syndromes. Cognition 5. 9—46. Kempen, G., and E. Hoenkamp. 1987. An incremental procedural grammar for sentence formulation. Cognitive Science 11. 201—258. Kimball, J. 1973. Seven principles of surface structure parsing in natural language. Cognition 2. 15—47. Kolb, H.-P., and C. Thiersch. 1991. Levels and empty categories in a principles and parameters based approach to parsing. Representation and derivation in the theory of grammar, ed. by H. Haider & K. Netter. Dordrecht. Kolk, H., and M. van Grunsven. 1985. On parallelism in agrammatism. Agrammatism, ed. by M.L. Kean. New York. Kurtzman, H., and M. MacDonald. 1993. Resolution of quantifier scope ambiguities. Cognition 4 8. 243—279.

1512

Kutas, M., and C. van Petten. 1988. Event-related brain potential studies of language. Advances in Psychophysiology, 3, ed. by P. K. Ackles, J. R. Jennings and M. G. H. Coles. Connecticut. Ladefoged, P., and D. B roadbent. 1960. Perception of sequence in auditory events. Quarterly Journal of Experimental Psychology 13. 162—170. Lapointe, S., and G. Dell. 1988. A synthesis of some recent work in sentence production. Linguistic structure in language processing, ed. by G. Carlson and M. Tanenhaus. Dordrecht. Lebeaux, D. 1988. Language acquisition and the form of the grammar. Ph. D. Dissertation, University of Massachusetts, Amherst, MA. Lenneberg, E. 1967. Biological foundations of language. New York. Levelt, W. J. M. 1970. A scaling approach to the study of syntactic relations. Advances in psycholinguistics, ed. by G. B. Flores d’Arcais & W. J. M. Levelt. Amsterdam. —. 1983. Monitoring and self-repair in speech. Cognition 14. 41—104. Linebarger, M., M. Schwartz and E. Saffran. 1983. Sensitivity to grammatical structure in so-called agrammatic aphasics. Cognition 13. 361—392. Lust, B . (ed.). 1986. Studies in the acquisition of anaphora. Vol. 1. Dordrecht. Manzini, R., and K. Wexler. 1987. Parameters, binding theory, and learnability. Linguistic Inquiry 18. 413—444. Marcus, M. 1980. A theory of syntactic recognition. Cambridge, MA. Marslen-Wilson, W. D. 1973. Linguistic structure and speech shadowing at very short latencies. Nature 244. 522—523. —. 1985. Speech shadowing and speech comprehension. Speech Communication 4. 55—73. —, and L. K. Tyler. 1987. Against modularity. Modularity in knowledge representation and natural language understanding, ed. by J. Garfield. Cambridge, MA. Marr, D. 1977. Artificial intelligence — a personal view. Artificial Intelligence 9. 37—48. —, and T. Poggio. 1977. From understanding computation to understanding neural circuitry. Neuroscience Research Program Bulletin 15. 470—488. Matthews, R. J. 1991. Psychological reality of grammars. The Chomskyan turn, ed. by A. Kasher. Oxford. MacDonald, M. 1989. Priming effects from gaps to antecedents. Language and Cognitive Processes 4 . 35—56. McElree, B ., and T. B ever. 1989. The psychological reality of linguistically defined gaps. Journal of Psycholinguistic Research 18. 21—35. McNeill, D. 1970. The acquisition of language: The study of developmental psycholinguistics. New

XXII. Syntax und Nachbardisziplinen

York. Meisel, J., and N. Müller. 1992. Finiteness and verb placement in early child grammar: Evidence from simultaneous acquisition of French and German bilinguals. The acquisition of verb placement, ed. by J. Meisel. Dordrecht. Meringer, R., and K. Mayer. 1896. Versprechen und Verlesen. Eine psychologisch-linguistische Studie. Stuttgart. [reprinted 1978, Amsterdam]. Miceli, G., A. Mazzucchini, L. Menn and H. Goodglass. 1983. Contrasting cases of Italian agrammatic aphasia without comprehension disorder. Brain and Language 19. 65—97. Miller, G. A. 1967. The magical number seven, plus or minus two. Psychological Review 63. 81—97. —, and S. Isard. 1963. Some perceptual consequences of linguistic rules. Journal of Verbal Learning and Verbal Behavior 2. 217—228. Mitchell, D. 1987. Lexical guidance in human parsing: Locus and processing characteristics. Attention and performance XII, ed. by M. Coltheart. Hillsdale, NJ. Newport, E., H. Gleitman and L. Gleitman. 1977. Mother, I’d rather do it myself: Some effects and non-effects of maternal style. Talking to children, ed. by Snow & Ferguson. Cambridge. Nicol, J., and D. Swinney. 1989. The role of structure in coreference assignment during sentence comprehension. Journal of Psycholinguistic Research 18. 5—19. Penner, Z. 1992. The ban on parameter resetting, default mechanisms, and the acquisition of V2 in Bernese Swiss German. The acquisition of verb placement, ed. by J. Meisel. Dordrecht. Pickering, M., and G. B arry. 1991. Sentence processing without empty categories. Language and Cognitive Processes 6. 229—259. Pinker, S. 1984 . Language, learnability and language development. Cambridge, MA. —, and J. Mehler. 1988. Connections and symbols. Cambridge, MA. Pritchett, B . 1988. Garden path phenomena and the grammatical basis of language processing. Language 64. 539—576. —. 1991. Subjacency in a principle-based parser. Principle-based Parsing: Computation and psycholinguistics, ed. by R. Berwick, S. Abney & C. Tenny. Dordrecht. —. 1992. Grammatical competence and parsing performance. Chicago. Radford, A. 1990. Syntactic theory and the acquisition of English syntax. Cambridge, MA. Riesbeck, C. K., and R. C. Schank. 1978. Comprehension by computer: Expectation based analysis of sentences in context. Studies in the perception of language, ed. by W. J. M. Levelt & G. B. Flores d’Arcais. Chichester.

83.  Syntax and Psychology

Roeper, T. 1986. How children acquire bound variables. Studies in the acquisition of anaphora. Vol. 1, ed. by B. Lust. Dordrecht. —. 1992. From the initial state to V2. Acquisition principles in action. The acquisition of verb placement, ed. by J. Meisel. Dordrecht. Sasanuma, S. and H. Monoi. 1975. The syndrome of gogi (word-meaning) aphasia. Selective impairment of kanji processing. Neurology 25. 627—632. Schaner-Wolles, C., and H. Haider. 1987. Spracherwerb und Kognition — Eine Studie über interpretative Relationen. Grammatik und Kognition (Sonderheft 1 der Linguistischen Berichte), ed. by J. Bayer. Opladen. Schank, R. C. 1975. Conceptual information processing. Amsterdam. Selkirk, E. 1984 . Phonology and syntax. Cambridge, MA. Shieber, S. 1985. Evidence against the context-freeness of natural language. Linguistics and Philosophy 8. 333—343. Snow, C. 1972. Mother’s speech to children learning language. Child Development 43. 549—565. Slobin, D. L. 1966. Grammatical transformations and sentence comprehension in child- and adulthood. Journal of Verbal Learning and Verbal Behavior 5. 219—227. Solan. L. 1983. Pronominal reference. Child language and the theory of grammar. Dordrecht. Stabler, E. 1983. How are grammars represented? The Behavioral and Brain Sciences 6. 319—402. Steedman, M., and G. Altman. 1989. Ambiguity in context: A reply. Language and Cognitive Processes 4. 105—122. Swinney, D., M. Ford, U. Frauenfelder and J. B resnan. 1988. On the temporal course of gap-filling and antecedent assignment during sentence comprehension. Language structure and processing, ed. by B. Grosz, R. Kaplan, M. Macken & I. Sag. Stanford, CA. Tanenhaus, M., and G. Carlson. 1989. Lexical structure and language comprehension. Lexical representation and process, ed. by W. Marslen-Wilson. Cambridge, MA. Thornton, R., and S. Crain. 1994 . Successful cyclic movement. Language acquisition studies in generative grammar, ed. by T. Hoekstra & B. Schwartz. Amsterdam.

1513

Torris, T. 1987. Grammars and parsers: the case of GPSG. MS. Universität Köln. Vygotsky, L. 1962. Thought and language. Cambridge, MA. de Villiers, J., T. Roeper and A. Vainikka. 1990. The acquisition of long-distance rules. Language processing and language acquisition, ed. by L. Frazier & J. de Villiers. Dordrecht. Wanner, E., and M. Maratsos. 1978. An ATN approach to comprehension. Linguistic theory and psychological reality, ed. by J. Bresnan, M. Halle and G. A. Miller. Cambridge, MA. Weigl, E., and M. B ierwisch. 1970. Neuropsychology and linguistics: Topics of common research. Foundations of Language 6. 1—18. Weissenborn, J. 1990. Functional categories and verb movement in early German. The acquisition of German syntax reconsidered. Spracherwerb und Grammatik. Linguistische Untersuchungen zum Erwerb von Syntax und Morphologie (Sonderheft 3 der Linguistischen Berichte), ed. by M. Rothweiler. Opladen. —. 1992. Null subjects in early grammars: Implications for parameter setting theories. Theoretical issues in language acquisition, ed. by J. Weissenborn, H. Goodluck & T. Roeper. Hillsdale, NJ. —. 1994 . Constraining the child’s grammar: Local wellformedness in the development of verb movement in German and French. To appear: Syntactic theory and language acquisition: Crosslinguistic perspectives. Vol. 1: Phrase structure, ed. by B. Lust, J. Whitman & J. Kornfilt. Hillsdale, NJ. Wexler, K. 1982. A principled theory for language acquisition. Language acquisition: The state of the art, ed. by E. Wanner & L. Gleitman. Cambridge. —, and P. Culicover. 1980. Formal principles of language acquisition. Cambridge, MA. Whitaker, H. A. 1970. Linguistic competence: Evidence from aphasia. Glossa 4. 46—54. Whitaker, H. 1976. A case of isolation of the language function. Studies in neurolinguistics 2, ed. by H. Whitaker & H. A. Whitaker. New York. Woods, W. 1970. Transition network grammars for natural language analysis. Comm. ACM 13. 591—602. Woolford, E. 1983. Bilingual code-switching. Linguistic Inquiry 14. 520—536.

Josef Bayer, Jena (Germany)

XXII. Syntax und Nachbardisziplinen

1514

84. Syntax and Sociolinguistics 1. 2. 3. 4. 5. 6. 7. 8.

10. 11.

Introduction Defining ‘Sociolinguistics’ Syntactic Constraints on Code-Mixing Syntax in Multilingual Communities Syntactic Variation in a Creole Continuum Syntactic Peculiarities of Dialects or Registers Social Influences on Syntax Quantitative Approaches to Syntactic Analysis Quantitative Methods and Grammatical Theory The Locus of Grammars and of Variation References

1.

Introduction

9.

Sociolinguistics is the study of linguistic structures in relation to social structures. Syntactic structures have so far received less attention than phonological structures, but sufficient work has been done on them to show that the sociolinguistic approach promises to offer some valuable insights. At the same time it is fair to admit that at present the promises are hardly more than that (except in a few rather specialised areas listed below): there have been few sociolinguistic studies to match in sophistication the early syntactic work of William Labov on copula deletion and negative concord (1972). The main contribution of sociolinguists has been to study varieties of language which theoretical and descriptive linguists tend to ignore, rather than to explore the consequences of sociolinguistic findings for theories of syntax. Not surprisingly, perhaps, this research focus has allowed sociolinguistic work on syntax to be ignored by most syntacticians. It would be fair to say that this work has had no influence at all on the most popular theories of syntax such as the three described in Sells (1985) and Horrocks (1987), namely Government-Binding Theory, Lexical Functional Grammar and Generalised Phrase Structure Grammar. It is true that some theories attempt to take account of one particular strand of sociolinguistics, the study of discourse — two such theories are Systemic Grammar (Halliday 1985) and Functional Grammar (Dik 1978) — but these theories in turn are generally ignored by followers of other theories, and in any case it is debatable to what extent their analyses achieve their stated goals.

The aim of this article is to outline the part that syntax has played so far in sociolinguistics, focussing where possible on methods or findings which are likely to be of interest to general syntacticians. We start with a brief survey of the kinds of activity that are commonly classified as ‘sociolinguistics’.

2.

Defining ‘Sociolinguistics’

On the one hand, sociolinguists identify themselves by their METHODS — they typically study tape-recorded texts, which have been collected with some attention to the possible influences of social variables, and they typically analyse these texts in a quantitative way. One reason for analysing texts in this way is in order to find statistical evidence that social variables influence linguistic variables, but the texts are typically subjected to other, more purely linguistic, processing as well, in which the aim is to find ways in which the linguistic context influences a linguistic variable. For example, Labov (1972) studied the use of sentences like Clarence my friend (meaning ‘Clarence is my friend’) in New York, and investigated the influence not only of social variables (such as the speaker’s race) but also of the linguistic environment (e. g. whether the predicate was a nounphrase or an adjective-phrase). It is clear that there is nothing especially ‘social’ about the latter work, but it counts as ‘sociolinguistic’ because of the methods used. Another characteristic of sociolinguistic work is its SUBJECT MATTER. Sociolinguists typically study varieties of language whose social prestige is low — non-standard dialects dominated by a standard dialect (notably English), subordinate languages spoken by immigrant or other minorities, and pidgins or creoles. Once again this focus follows from the interest in social influences on language behaviour — speakers of such varieties often have some knowledge of the dominant variety, and may combine forms of the two varieties in interesting ways — but it is of course possible to study any such variety with regard to nothing but its own internal structure. Moreover, it could be studied without the use of quantitative methods. In either case, the work would still be considered by many to be an example of ‘sociolinguistics’,

84.  Syntax and Sociolinguistics

in contrast with ‘ordinary’ grammatical research (whose focus is still, regrettably, very much on standard languages). By this interpretation, then, a study of (say) relative clauses in a non-standard dialect would count as sociolinguistics, whereas a similar study relating to a standard dialect would not. It is clear that the concept ‘sociolinguistics’ has all the characteristics of a set of ‘family resemblances’ — all its instances share some properties with other instances, but there are no properties which are shared by all instances (which would not be shared equally by other branches of linguistics). Whatever the merits of interpreting the term in this way, the fact is that this is how it is often interpreted, so I too shall take it to include all these things. However, I shall ignore one important area of work which is often included in sociolinguistics, the study of discourse.

3.

Syntactic Constraints on Code-Mixing

We start with one of the most dramatic challenges to familiar assumptions about grammmars. In many multilingual communities code-mixing is normal — that is, speakers often change languages in mid-sentence. For example, Pfaff (1979) quotes attested utterances like (1). (1) No van a bring it up in the meeting. Here the first three words are in Spanish, meaning ‘They are not going to’, but the rest of the sentence, in English, has a syntactic structure which fits these Spanish words. Examples like these are sufficiently interesting in their own right, since they raise important issues about the cross-linguistic identification of categories. For example, in Spanish, ir a, ‘be going to’, is followed by an infinitive (a morphologically marked form of the verb); but in (1) it is followed by the uninflected English word bring, so it seems reasonable to assume that this too counts as an infinitive — i. e. that there is some psychologically real sense in which bring and a Spanish verb like com-er, ‘(to) eat’, belong to the same category. Even more interestingly, however, it turns out that people who use code-mixed utterances like (1) can also pass judgment on their relative acceptability, in just the same way

1515

that native speakers can judge the acceptability of an unmixed utterance (e. g. Sobin 1984 , Aguirre 1985). Moreover, when these judgments are taken in conjunction with sample texts, there seems to be a good deal of agreement among speakers about the kinds of mixing which are permitted. It is possible to generalise the judgments in terms of syntactic structures and to derive general constraints on code-mixing — i. e. rules about the positions in sentence structure where it is permissible to switch from one language to the other. This finding has given rise to a lot of research (amply documented in Bokamba 1987), covering a wide range of languagepairs. In each case the research has produced rules which control mixing in some particular community, though for some communities different researchers have found different rules by using different methods — for instance, there is disagreement as to whether Spanish-English bilinguals allow mixing within a noun-phrase, so that an attributive adjective and its noun head are in different languages. Pfaff (1979), using recordings of spontaneous speech, concluded that this was possible only if the position of the adjective was permitted by both Spanish and English; which means that it must precede the noun, and must belong to the small list of adjectivetypes which can occur in that position in Spanish, as in (2). (2) mi unico pleasure ‘my only pleasure’. In contrast, Sobin (1984 ) asked bilingual speakers to judge prepared examples, and found that they accepted examples which infringed Pfaff’s generalisation — e. g. examples in which an English colour adjective followed the noun. At the same time, his judges did not indiscriminately accept all examples, and his research confirmed the general conclusion that there are systematic limits to code-mixing. Reasssuringly, Sobin’s findings seem to support those of Poplack (1980), based on methods similar to Pfaff’s. Of course, it is always possible that different Spanish-English bilingual communities apply different rules, and that these researchers are all right, but with respect to different communities. One of the interesting questions that arises in this area is whether there is any syntactic evidence for the asymmetrical relation between the two languages — syntactically definable boundaries where it is permissible to

XXII. Syntax und Nachbardisziplinen

1516

switch languages in one direction but not in the other. There is good evidence for such asymmetries in some communities; for instance, Bautista (1980) reports that TagalogEnglish mixing does not allow an English complementiser to be followed by a Tagalog word, though the reverse switch is possible. Similarly, the data reported by Keddad (1986) can be interpreted as showing that in Algeria French-Arabic mixing is controlled by the general principle that an Arabic noun cannot be used as a dependent of a French word, though the reverse is permitted; for example, switching from Arabic to French is not permitted between a subject and verb, though it is between verb and object. However it is also worth noting that some of the most detailed and thorough analyses have found no clear evidence for asymmetries, so they may exist in some communities but not in others. It would be wrong to assume in advance that in every code-mixing community one language is recognised as the ‘matrix’ within which mixing takes place (Schmid 1986). Code-mixing may provide an extra source of evidence for deciding among competing analyses for a construction, to which the analyst can turn when analysing one or the other of the languages concerned. We have already noted that mixing suggests strongly that there is a psychologically real identity between some grammatical categories in different languages. The detailed patterns of mixing can be used further to defend particular analyses. For example, Blake (1987) argues that English infinitival to is an instance of the (Government-Binding) category INFL, whereas the corresponding Spanish word a is a preposition, on the basis of observed patterns of mixing between Spanish and English (to may be followed by a Spanish infinitive, and a by an English infinitive — as in our example (1) — but a Spanish verb cannot be followed by to). Whatever the merits of these arguments, it is clear that code-mixing data provides a potentially rich source of evidence for linguistic analysis (Woolford 1983). The most general, and presumably the most important, factual question is whether all or any of the constraints on code-mixing are universal. Opinion is sharply divided at present but according to Bokamba (1987), the literature to date does not contain any universal constraints: different constraints apply to different language-pairs. Various

rules and principles have been claimed to be universal, but the data-base has generally consisted of only a very small sample of language-pairs and the claims are easily refuted on the basis of other pairs. For instance, according to Sciullo et al. (1986), switching between subject and verb should always be easy, but Muthwi (1986) reports that switching between English and Kalenjin is not permitted at all in this position. These differences suggest, then, that the constraints are learned. If this is right, then the familiar question arises as to how such things can be learned, given that the only available evidence is negative. It is also important to note that at least one study (Lederberg et al. 1985) claims to show that judgment patterns concerning code-mixing cannot be learned because the same patterns can be produced by different judges whether or not they are members of a community in which code-mixing is permitted. The facts about code-mixing raise important questions for grammatical theory. It seems to be generally agreed that code-mixers can avoid mixing (e. g. when talking to a monolingual), so they clearly know which constructions and items belong to which language. Does this mean that they have two separate grammars from which they draw, or a single grammar in which each element is tagged for its language? Indeed, is this a meaningful question? In either case, what status do the constraints on code-mixing have: part of some grammar or part of some kind of data-base for pragmatics or processing? Various suggestions have been made, both formal (e. g. Sankoff/Mainville 1986) and psycholinguistic (e. g. Sridhar/Sridhar 1980), but we are clearly a long way from working through the implications of these facts for syntactic theory.

4.

Syntax in Multilingual Communities

Code-mixing is not the only interesting phenomenon which is found in multilingual communities. Other possibilities include important syntactic change in one or the other of the languages concerned, and the creation of altogether new languages. The first of these possibilities falls into the domain of historical linguistics, which is covered elsewhere in this encyclopedia, but it is worth mentioning that sociolinguistics has an important

84.  Syntax and Sociolinguistics

contribution to make through the study of contemporary multilingual communities and their social structures. This kind of work promises to throw light on the mechanisms by which syntactic change takes place. A particularly interesting example of this kind of work is the study of a small village in India called Kupwar by Gumperz/Wilson (1971). The inhabitants of Kupwar are divided into castes, each of which has one of three languages (Marathi, Urdu, Kannada) as its language. However, the castes need to communicate with one another and bilingualism or trilingualism is common. What is interesting is that although these languages have coexisted in Kupwar for centuries, their vocabularies are still completely separate. One assumes that this can be explained functionally, in terms of the need to symbolise the differences between castes. As far as syntax is concerned, on the other hand, the differences between the languages have been grossly reduced. For example, Kannada normally has a zero copula (i. e. no verb at all in sentences like ‘John (is) my friend’), whereas Urdu and Marathi have an overt copula; but the Kannada of Kupwar too has an overt copula. It is interesting to speculate on the causal connections between these changes and codemixing, which is presumably less constrained the more similar the languages concerned are in their syntax. It is also interesting to notice how commonly syntactic patterns are borrowed from one language into another, provided the communication patterns between their speakers are sufficiently intensive (Hudson 1980, 59 ff). An even more dramatic consequence of the coexistence of two languages in one community is the creation of a third language, combining the lexicon of one with the syntax of the other. This is reported (Muysken 1981) to have happened in Ecuador, where a newly created language called ‘Media Lingua’ has the syntax of Quechua, but a vocabulary which is 90% Spanish. More precisely, the voacabulary is 90% Spanish with respect to forms only, but the meanings and morphology (and, one assumes, syntactic valency information) remain Quechua. Once again, most interesting questions arise in connection with the formal and psychological relations among the grammars concerned — for example, is Media Lingua a completely separate grammar, or is it just a set of general principles for exploiting the grammars of Quechua and of Spanish?

1517

The case of Media Lingua is somewhat unusual. A much more common example of a new language being created in a multilingual situation is the birth of a PIDGIN language. These differ from the case of Media Lingua in being, at least in their origins, much simpler and smaller than any of the languages on which they are based. Rather interestingly, one of the main simplifications is the total abandonment of inflection, even when this is used by all the source languages. Less surprisingly, perhaps, early pidgins make no syntactic provision for relatively complex structures such as subordinate clauses. If a pidgin is used only for very simple transactional purposes it is possible to dispense with such things. Pidgin languages have been widely studied (Holm 1988), and this work has generated some interesting questions for grammatical theory. Perhaps the most general question is, once again, whether there are any universals of pidgin syntax, and if there are, how they can be explained. For example, it is very common for pidgin languages to have a zero copula (Ferguson 1971). Why should this be so? Another important set of questions arises about the ways in which pidgins develop if the functional demands on them are expanded: is there a universal sequence in which communicative needs are filled, or a universal set of ways in which they are filled? For instance, the need for relative clauses can be met in various ways, but do pidgins all tend to follow the same route as they change into creoles, by acquiring native speakers (Sankoff/Brown 1976, Romaine 1984)?

5.

Syntactic Variation in a Creole Continuum

When a pidgin is created, there is typically one language which is dominant as the model for its vocabulary — often a European language, because of colonialism — but its syntax may be radically different from that of this language, showing heavy influence from the indigenous languages. When it is learned as first language by the offspring of pidginspeaking parents these syntactic characteristics naturally persist, though the grammar may be expanded in the ways just alluded to. Because of the historical circumstances under which pidgins and creoles arise, it is often the case that the creole coexists with the dominant language on which the original pidgin

XXII. Syntax und Nachbardisziplinen

1518

was based. The creole is spoken by poor people, and the dominant model by rich people in the same country. Because of this social difference between them sociolinguists call the creole the ‘basilect’ and the dominant model the ‘acrolect’. For example, in many Caribbean countries the acrolect is (more or less) standard English, while the basilect is an English-based creole. What makes this sociolinguistic situation interesting for syntacticians is that the relative similarity between the basilect and acrolect allows intermediate varieties to develop — which would not be possible if the coexisting languages were completely different languages. The result is a continuous spectrum of grammar between the basilect and the acrolect, known as a ‘creole continuum’. As we have seen, the syntactic differences between acrolect and basilect may be considerable, but the semantic systems of the two varieties can also differ in fundamental ways — e. g. according to Bickerton (1975) the basilect in Guyana shows anteriority (temporal relations to some reference-time), but not deictic tense, whereas the converse is true of the acrolect (standard English). Where grammars differ in such fundamental respects, but are linked by a chain of intermediate grammars, we are likely to learn a great deal about the nature of grammar by studying the links between them. The distance between acrolect and basilect can be illustrated by the roughly synonymous pair of sentences in (3), taken from Winford (1985). (3a) Who do I want to go? (3b) A hu mi waan fi go? Some of the differences are quite superficial — e. g. the lack of t on waan — but others are much more significant: — the initial particle a, often used to mark the focus; — the lack of surface ‘case’ even in pronoun forms; — the lack of subject-auxiliary inversion in questions. As for the last two words, fi go, Winford argues persuasively that these do not correspond to the standard to go, but constitute a finite clause (whose subject position is empty), fi being a modal auxiliary. In standard English, of course, want does not allow a finite clause as complement. The existence of intermediate varieties — ‘mesolects’ — provides a natural test-bed for

a number of issues in grammatical theory. What is (or should be) the status of the notion ‘language’, given that by any standard definition of ‘same language’ every adjacent pair of mesolects must belong to the same language, and therefore the same must be true of the basilect and acrolect too? In particular, is it possible to believe that all dialects of (say) English share a ‘common core’ of syntax? Secondly, what do the mesolects tell us about the relations among the elements of the grammar? Does the order in which acrolectal features give way to basilectal ones relate in any way to their interrelations in the grammar? The discussion so far has accepted the claim (Bickerton 1975) that there is basically a single ‘route’ from basilect to acrolect. However, this has been denied by a number of creolists (e. g. Washabaugh 1977, Edwards 1985), who find that this view oversimplifies the sociolinguistic facts considerably. According to these scholars there are many independent dimensions of variation, and not just one. This claim adds to the interest of creole continuums by reducing their uniqueness. Other societies show multidimensional variation, as we shall see in section 7, so it should be easier to generalise from creole continuums. Somewhat similar questions arise in connection with ‘pidgin continuums’, such as those found in Western Europe among migrant workers. Thus the pidgin German of Turkish ‘Gastarbeiter’ ranges from a minimal command of a few words to something approaching German (e. g. Klein/Dittmar 1979). However this situation seems likely to throw less light on normal syntax than the much richer and more stable grammatical systems found in creole continuums.

6.

Syntactic Peculiarities of Dialects or Registers

According to our definition of sociolinguistics, the study of dialects, and especially of low-prestige dialects, is part of it. Factual information about such dialects is an important corrective to the tendency for syntacticians to take account only of standard dialects. It is all too easy to draw general and fundamental conclusions about the basic structure of a language on the basis of facts which turn out not to apply generally to all dialects of the same language. For example, the analysis of

84.  Syntax and Sociolinguistics

the English auxiliary verb system in Chomsky (1957) rested, inter alia, on the assumption that modal verbs could not be combined with one another. This analysis was generally accepted in Transformational Grammar for more than two decades, and is still the source of much of current theorizing, via the abstract INFL node of Government-Binding Theory (Sells 1985, Horrocks 1987). However, the original analysis would never have been proposed if Chomsky had spoken Scots English, since double (or even treble) modals are permitted there (Miller/Brown 1982), as in (4). (4) He will can come. The reason why differences between dialects are of special interest to grammarians is that the dialects concerned are likely to be similar in other respects. Thus the facts about Scots are highly relevant even to a grammar which claims to describe only standard English because it would be hard to explain the differences in terms of general typological differences — unless of course there is clear evidence that the dialects concerned are indeed different in some fundamental respects. In the absence of such evidence we ought to be able to write grammars for the two dialects which are different in only very specific ways — in terms of the valency requirements of particular modal verbs, for example. Otherwise we have no explanation for all the similarities between them. We should be especially suspicious of any analysis of one dialect which presents some fact about it as the automatic consequence of a basic theoretical assumption, when the same fact does not apply to a closely related dialect. A noteworthy example of this danger is the discussion of so-called ‘that — trace’ effects — the impossibility, in standard English, of extracting a subject across the complementiser that as in (5). (5) %Who do you think that ___ came? Various attempts have been made to show how this fact follows from general typological properties of English plus very general assumptions about the nature of grammatical structure; but these attempts must all be wrong because there are dialects of English, spoken in the Ozarks, which do allow structures like (5) (Sobin 1987). Until there is evidence for major typological differences between these dialects and others, it is fair to

1519

assume that the differences are rather specific (Hudson 1986). The examples quoted so far could easily be multiplied. For example, recent discussions of English dialects have turned up the following facts, all of which are surprising if one’s main experience has been in standard English: — Some dialects (spoken by Irish people and black Americans, but probably originating in the South of England) show explicitly whether an event happened once or habitually, by using do before the verb in the latter case; and be’s can be used, as the inflected present of be, to indicate habituality (Fasold 1969, Harris 1984 , 1986, Bailey/Maynor, 1985). — In Scots English (Romaine 1980) relative clauses are introduced almost exclusively by that, rather than by relative pronouns that begin with wh- (who, which, etc.). The word that can even be used with ’s, as a possessive, as in (6). (6) The man that’s hat was blown off ran after it. Structures like this suggest strongly that that is a pronoun in the dialects concerned, and not just a complementiser. The interest of examples like these is not just philological, but structural: they show that quite major differences between varieties can still be limited to quite specific parts of the language system. They are relevant to any discussion of how best to analyse the structures concerned, in either of the dialects concerned. Points similar to those just made in relation to dialects can also be made about REGISTERS — varieties of language defined in relation to the ways in which they are used, such as ‘high-level academic writing about linguistics’ or ‘popular journalistic radio commentary on sport’. Differences in syntax seem especially important for distinguishing registers (more so, for example, than phonological differences, whose main domain is in distinguishing between dialects). To take just one recent example, Bell (1985) discusses phrases like (7). (7) controversial cancer therapist Milan Brych This example has a structure which is not permitted in ordinary speech or writing, because of the absence of an article. Stylistically restricted registers like this cannot be dis-

XXII. Syntax und Nachbardisziplinen

1520

missed as performance errors, because they are obviously learned and systematic. Nor can they be ignored as marginal to ordinary language, when linguists devote so much attention to other constructions which are presumably much more peripheral to most people’s ordinary language experience, such as ‘gapping’ (8). (8) John invited Mary, and Bill, Sue. Sociolinguists subsume the concepts ‘dialect’ and ‘register’ under the general term ‘variety’. Any variety differences of the kind discussed here are a challenge to the syntactician, whether descriptive or theoretical, because they force a decision about what should be included in a grammar, which in turn requires a decision about the ontological status of the grammar. In particular, if the grammar is intended to be a psychologically plausible model of a speaker’s knowledge, then it clearly must cover the full range of registers known to a speaker (especially if the latter is idealised as knowing the language perfectly). But if that decision is taken, some means has to be found to distinguish the registers from one another, because these distinctions are also known to native speakers.

7.

Social Influences on Syntax

The notion ‘variety’ is rather a crude and unsatisfactory basis for characterising the ways in which syntactic constructions can differ socially. It is unsatisfactory because of the considerable overlap between varieties (Hudson 1980). Quantitative analysis of syntactic variation (described in the next section) has shown that there is at least a strong tendency for constructions to be unique in their social relations — some constructions are closely associated with formality of style, others with social status of speaker, others with sex of speaker, and so on through a wide variety of quantitative combinations and permutations of different social influences. A good example of the research which shows this kind of patterning is reported in Cheshire (1982a, 1982b). The speakers studied were a group of teenagers in Reading, in the south of England. One of the constructions studied was the use of ‘negative concord’ — sentences like (9). (9) I’m not going nowhere. This turned out to be associated closely with what Cheshire calls ‘vernacular loyalty’ — i. e. roughly speaking, its use was associated

with delinquency. This was equally true both for the girls and for the boys. Another construction was the use of what (rather than that) to introduce relative clauses, as in (10). (10) There’s a knob what you turn. This was also used to mark vernacular loyalty, but only for boys — that is, if a particular girl and boy both used negative concord about as often, the girl would probably use relative what significantly less often than the boy. The reverse was true for a third construction, ain’t (instead of present-tense, negative auxiliary have or be): among the ‘delinquents’, it was used more often by girls than by boys. If the only concept we have for describing facts such as these is ‘dialect’, then we have two options: either we ignore the differences among the three constructions and assign them all to the same dialect (‘delinquent adolescents’), or we recognise the differences and distinguish two different dialects (‘delinquent girls’, ‘delinquent boys’). In the first case the description fails to show important distinctions; in the second it misses important similarities (namely, that negative concord has the same status in both dialects). Worse still, in either case the description entirely lacks the quantitative dimension. All three of these non-standard constructions were in fact used by all the teenagers studied, including those who were not classified as ‘delinquent’, and the groups differed only in how often they used them. An alternative to the use of ‘dialect’ is to include every construction in a single, undivided grammar, but to link each one directly to any social factors which influence its use. The linkages needed for the Reading data are shown in (11). (11) negative concord — — — — — delinquents relative what — — — — — delinquent boys ain’t — — — — — — — — — delinquent girls Although the data from which these linkages are deduced is quantitative, it is debatable whether the linkages themselves need to be quantitative (Hudson 1986), because the quantitative differences could derive from aspects of performance such as the degree to which a speaker alligned themself with the social group concerned. Two main points arise from this rather simple example. One is that different constructions are favoured by different social cir-

84.  Syntax and Sociolinguistics

cumstances, which makes the concept ‘dialect’ much too crude a tool for showing all the necessary distinctions. The other is that the linkages between constructions and social circumstances must be part of the speaker’s knowledge about the constructions concerned — in other words, must be part of ‘competence’. There is no other way to explain the subtle differences found between different kinds of speaker — to say nothing of the ability that all speakers have both to draw conclusions about other speakers’ social characteristics on the basis of their use of socially sensitive constructions, and to ‘play’ with the social linkages in order to produce different social impressions. It is worth mentioning that these general conclusions about constructions are based on a fairly small number of sociolinguistic research projects, but that a very much larger number of projects in which the focus was on pronunciation lead to precisely comparable conclusions, with the difference that the social linkages are with pronunciation features (either general or in particular words).

8.

Quantitative Approaches to Syntactic Analysis

The evidence for claims about links between constructions and the social context is mostly quantitative. A variety of methods are used, but they all derive from the work of William Labov, who has been the dominant figure in quantitative sociolinguistics since his work in the 1960’s on phonological variation (see the collections in Labov 1972a, 1972b). Two particularly influential papers dealt specifically with syntactic variation in New York: one (1969) about the use of the zero copula (characteristic of black speakers), the other (1972c) about negative concord. These papers are both included in the 1972a collection. Labov’s work has inspired a spate of research projects, some of which included syntactic variation among the linguistic features studied. These works are reported at the annual ‘NWAV’ (‘New Ways of Analyzing Variation’) conferences (e. g. Denning et al. 1987), but the following is a representative sample of the syntactic studies: Sankoff (1973), Thelander (1976), Sankoff/Thibault (1977), Sankoff/Vincent (1977), Sankoff/Laberge (1978), Kroch/Small (1978), Sankoff/Thibault (1978), Klein/Dittmar (1979), Dines (1980), Naro (1981), Cheshire (1982a, 1982b), Weiner/La-

1521

bov (1983), Kikai et al. (1987), Lemieux (1987), Pooley (1988). These studies all use the same basic methodology (Hudson 1980, 138 ff). Assuming that the researcher has already identified a population of speakers to be studied, then a sample of these speakers is identified. A reasonable amount of speech by each speaker is recorded (under more or less carefully controlled conditions), producing a set of texts which can then be compared quantitatively, with respect to the use of a (pre-selected) set of constructions. For each construction, the researcher counts the number of times it is used in each text, as a proportion of the number of times it might have occurred (a notion that we shall discuss below). This gives a figure for each text which can be compared with figures for other texts, and also with the figures for other constructions in the same text. During the analysis various distinctions can be made between different kinds of use for a particular construction — e. g. according to the kind of linguistic context in which it occurs — and these distinctions can also be reflected in the final text figures. Statistical comparisons can be made among the various texts to discover significant trends, and on the basis of these statistics generalisations can be made. These methods clearly contrast with those used in descriptive linguistics, where the main method is non-quantitative: the elicitation of judgements from a native informant (who may be the linguist themself). However, a complementary method is also used in sociolinguistics which is somewhat like this elicitation method, namely judgements on tape-recorded speech as to the social characteristics of the speaker. This method is sometimes used as a check on the results of the textanalysis method. One of the details of the quantitative method as applied to syntax deserves some discussion. The aim of the method is to compare texts with respect to the frequency of occurrence of some construction, but this raises the question of what ‘frequency’ means in this context — what is the unit X such that the construction occurs, on average, so many times per X? X clearly cannot be a unit of time (e. g. ten minutes), because the figures are likely to be influenced, in a quite irrelevant way, by factors such as the speaker’s speed of delivery. In the most straightforward cases, it is possible to give quite a satisfactory answer:

XXII. Syntax und Nachbardisziplinen

1522

X is a fixed number of occasions on which the intended meaning was compatible with the construction. In other words, one counts the number of times the construction occurs, and presents it as a proportion of the times when it could have occurred without changing the meaning. If meaning is kept constant in this way, then we can probably assume that the choice between the alternative constructions is motivated by social influences, so whatever figures we find can be used as evidence for a sociolinguistic analysis of the data. In these straightforward cases, the construction has a meaning which is expressed by one or more other constructions that are all easily identified and mutually substitutable. For example, it would be quite easy to deal with any of the three constructions found in Reading English mentioned above (negative concord, relative what and ain’t), because in each case it is relatively easy to identify constructions whose meanings are the same. Thus, a negative-concord sentence like (12a) is an alternative to a standard one like (12b), (12a) I didn’t do nothing. (12b) I didn’t do anything. The researcher would therefore find out what proportion of sentences of either type had negative concord. If this figure had been calculated for each text, then it would be possible to make meaningful comparisons between the texts. This method of analysis is the one which was developed for handling variations in pronunciation, and when analysed in this way each construction is called a variant on some SOCIOLINGUISTIC VARIABLE. However, it has often been pointed out that there are some constructions where it is hard, or even impossible, to treat syntactic variation in this way (e. g. Lavandera 1978, Dines 1980, Cheshire 1987). For example, in some cases there is no alternative, synonymous construction because the variation involves semantic structures as well as syntactic ones — e. g. this may be true of some of the variation in creole continuums mentioned above (Winford 1984 ). It also arises in urban dialects such as Reading — for example, according to Cheshire (1982), Reading teenagers use tag questions consisting of in (a form of ain’t) plus subject in a way that is not parallelled in standard English, to show aggression after conveying new information. Thus when (13)

was uttered the speaker knew that the addressee did not know that he was going. (13) No, I’m going, mate, in I. Whether the correct account of this tag question should be in grammatical or pragmatic terms, the fact remains that in standard English, used according to standard pragmatics, no tag would have been used in these circumstances. In cases like this the construction under observation has no alternatives other than saying nothing, so there is no sociolinguistic variable, according to the normal definition of this term. Alternative methods of quantification have to be found for making texts comparable, all of which are likely to be somewhat less sensitive, because other variables are not controlled to the same extent. Many possibilities suggest themselves; for example, Sankoff/Thibault (1978) suggest taking some fixed number of running words — say, 1.000 words — as the unit in relation to which the occurrences of a construction are counted. Another possibility worth considering as the basic unit is a fixed number of words of the class on which the construction concerned depends — e. g. main verbs, in the case of tag questions. It is an empirical matter which of these alternatives offers the most revealing results, but it is certainly not necessary to conclude that syntactic variation cannot be studied quantitatively because some constructions have no alternatives.

9.

Quantitative Methods and Grammatical Theory

One of the facts about quantitative studies of syntax of which syntacticians ought to be aware is that it is possible to relate a construction quantitatively to its purely linguistic environment. Indeed, many ‘sociolinguistic’ studies are primarily concerned with precisely such relations, rather than with social influences, a classic example being Labov’s 1969 paper on zero copulas in the speech of Black Americans. Although this describes and compares the speech of different social groups, a greal deal of the discussion is concerned with the structure of the (transformational) rules responsible for the construction, and with their relations within the grammar. The level of argumentation about the pros and cons of the alternative analyses is as high as any in the transformational literature of

84.  Syntax and Sociolinguistics

the 60’s, and the conclusions are such as to interest even the most a-social grammarian. The first sentence of the paper is worth quoting: “ The study of the Black English Vernacular [...] provides a strategic research site for the analysis of English structure in general.” (Labov 1969). One thing that makes this paper (and most others which apply the quantitative method) important for the practising grammarian is that they provide a new source of evidence on which grammatical analyses can be built. The evidence consists of quantitative data on language use, showing how probable some construction is under various circumstances. (Figures for probabilities are different from those for actual occurrences in particular texts; the former can be derived from the latter, given a sufficient body of data (Cedergren/Sankoff 1974 ), and are clearly more relevant to grammar.) The probability typically varies from one linguistic environment to another; for example, Labov found a major difference in his study of zero copulas according to whether the subject was a pronoun or a full NP, illustrated respectively by the sentences in (14 ), in which the zero copula is marked by a gap. (14a) He just feel like he ___ getting cripple up from arthritis. (14b) Boot ___ always comin’ over my house to eat, to ax for food. Labov found that all the black speakers in his sample, under all circumstances, were more likely to use a zero copula after a pronoun than after a full NP, though all of them on occasion used the full (uncontracted) copula instead, and some speakers were much more likely to use it than others. How is this fact to be explained? It is hard to avoid the conclusion that the differences in performance between the two environments must relate to some difference in the grammar, though, as we shall see in the next section, it is a matter of considerable debate as to precisely what that difference might be. The alternatives to a grammatical account are an explanation in pragmatic terms, or one in psycholinguistic terms, but it is hard to see how either of these accounts could predict the differences concerned. Failing an alternative, then, we must assume that the grammar must distinguish the two environments in which the zero copula is used (i. e. in which

1523

the copula is not used). This in spite of the fact that, in terms of what is simply possible or impossible, there is no difference between them — the zero copula is possible, for all relevant (i. e. black) speakers, in both environments. Quantitative data allow one to investigate the interactions between rules in a way that is not otherwise possible. In Labov’s analysis of zero copulas, this involved considering the rule for reduced verb-forms (such as ’s for is). Interestingly he found precisely the same pattern of environmental influence here too: reduced forms were much more likely after a pronoun subject than after a full NP. This was equally true for white speakers, who did not use the zero copula at all. The conclusion Labov drew from this fact was that the zero copula is in fact a sub-case of the reduced copula, a conclusion which was strongly supported by other data — e. g. the fact that the zero copula seems to be ruled out in precisely the same environments where overt reduced forms are excluded, notably when stranded without a following complement. Thus (15a) is impossible, according to Labov’s analysis, for just the same reason as (15b). (15a) *He is as nice as he says he ____. (15b) *He is as nice as he says he’s. It could of course be argued that Labov could have arrived at precisely the same conclusions on the basis of nonquantitative data like (15) alone, but the quantitative data provide independent confirmation which makes the conclusions that much firmer. Moreover there are some cases where quantitative data are probably more reliable than judgements, influenced as the latter often are by prescriptive attitudes. The value of quantitative data in linguistic analysis has been widely recognised by sociolinguists (e. g. Wolfram 1975). Quantitative differences between constructions are also relevant to one of the most general issues in current syntactic theory, the status of ‘constructions’. The notion of ‘a construction’ is admittedly vague (e. g. is ‘agentless passive’ a different construction from ‘passive’?), but it has played an important part in most theories of grammar. However it has now become a matter of controversy because according to GovernmentBinding theory there are no ‘constructionspecific’ rules — that is, grammars do not refer to constructions, in the traditional sense. Linguists may refer to constructions such as ‘agentless passives’ or ‘zero-copula sentences’

1524

when they are discussing their analyses, but such notions play no part in the grammar itself. Instead, the grammar contains rules and principles which all apply to a variety of constructions; so a construction reflects the intersection of a set of rules and principles none of which is uniquely responsible for it. This view has been contested by other grammarians, most clearly by those working in the tradition of Construction Grammar (e. g. Fillmore et al. 1988). The relevance of quantitative data to this debate is that constructions have distinctive probabilities of occurrence, which are unlikely to be derivable from the intersecting probabilities of the rules and principles responsible for them (though it has to be admitted that this possibility has not in fact been investigated seriously). We have already seen examples of this — e. g. negative concord and zero-copula constructions. There are different ways in which a probability could be linked to a construction, but all of which require a direct linkage; one has already been hinted at, namely that the probability of some construction C could relate more directly to the social categories with which C is associated. But whatever kind of connection one assumes between constructions and probabilities, the fact remains that some probabilities seem to be unique properties of particular constructions. If this is so, then it probably follows that the construction must have independent status in the grammar, because otherwise there would be nothing to which the probability could relate. We have considered two ways in which probabilistic data may be relevant to descriptive and theoretical grammar: — as an extra source of evidence for the relatedness of constructions; — as evidence for the reality of constructions. Both of these suggestions raise general questions, notably about the psychological or social status of grammars, and the serious debate has obviously hardly begun.

10. The Locus of Grammars and of Variation One of the most general questions that sociolinguistic work raises is about the ‘locus’ of grammars: is a grammar a property of an individual, or of a community? Putting the same question in a different form, is a gram-

XXII. Syntax und Nachbardisziplinen

mar a model of a psychological or of a social reality? A second question, which is in fact closely related to the first, is about the locus of variation: what is the linguistic unit of variation? Is it a variety, or a linguistic item (e. g. a construction)? The evidence from sociolinguistics throws light on both of these questions. The traditional view on the locus of grammars is that they are properties of communities — e. g. ‘English grammar’ is defined in relation to the community of English speakers. This is a reasonable view if one thinks in terms of standard languages, supported by prescriptive grammars; and it is not unreasonable when one views language from the point of view of the language-learner, who aims to identify themself as far as possible as a member of some community (or set of intersecting communities) by learning its language perfectly. This view seems to be accepted by Chomsky, when he talks of an ideal speaker knowing the language of their community perfectly (Chomsky 1965). It is also accepted by Labov, who argues (1972b, 24 7) that many of the regularities discovered by sociolinguists are unlikely to be generally known to members of the community concerned, but they should nevertheless be included in a grammar — from which it follows that the grammar must be a property of the community, not of the individual (because it helps to predict the behaviour of the community, and not that of any given individual). Another major contributor to sociolinguistic theory, Bickerton, suggests that a community grammar generates the range of possible sub-grammars (‘lects’) that are available to individual speakers (Bickerton 1975). In contrast, most theoretically-oriented linguists these days seem to accept Chomsky’s other view of grammar as a part of the individual’s mind (a view which is quite separate from the much more contentious claim that language is a separate ‘module’ or ‘organ’ of the mind). The same is also true of many people working within the paradigm of sociolinguistics. For example, Kay (1978) and Kay/McDaniel (1979) argue that Labov’s own quantitative data show that there are regularities within some communities that could not be shown within a single grammar (because the influences of particular environments are ordered differently for different sections of the community). Similar points can be made in relation to the deep, underly-

84.  Syntax and Sociolinguistics

ing differences in structure between the alternants found, inter alia, in creole continuums (e. g. Harris 1984, Winford 1988). It is hard to be sure, in this dispute, whether there is any more at stake than terminology. It is beyond dispute that individuals have mental grammars, and that these may vary, at least in minute details, from person to person even within a small, close-knit community. It is also agreed that there are regularities in the linguistic differences between speakers in particular communities, and also regularities between the speech used under socially different circumstances, which cut across differences between individuals. The question at issue seems to be whether it is appropriate to call a desription of these regularities a ‘grammar’. Of course, there is more at stake than this, because those who believe that a grammar can be used for these purposes also believe that the regularities are such that they can be described successfully in terms of some grammar such as a transformational grammar. This is a theoretical matter, and probably one that can be resolved empirically. However there has been very little discussion in these terms, and there is no real substance to the debate at present. Let us assume, then, that ‘grammar’ refers to the knowledge of an individual speaker (leaving open the possibility of a cross-community description which may or may not also be called ‘grammar’). What relation is there between individual grammars and variation of the kind described in this article? It certainly is not the case that by narrowing the scope down to an individual speaker’s grammar we have identified a homogeneous ‘idiolect’, in which there is no variation — or more precisely, in which all variation can be described satisfactorily in terms of the notions ‘optional’ and ‘free variation’. Let us briefly review the phenomena that we have already surveyed which have involved variation within the speech of one speaker which could not be treated in this way. — Most recently we have seen that even if two constructions are both possible in some linguistic environment, they may not be equally probable. We assume that such probability differences need to be explained, and in at least some cases is seems that any explanation must refer to the specific construction concerned, rather than just to general pragmatic or processing principles. — We have mentioned ‘register’ differences, which involve variation within the speech of

1525

one person (e. g. between technical and nontechnical, or formal and informal). These differences are closely related to differences in the social circumstances, and such connections are known to the speaker — indeed, they could not occur if they were not known. — We have shown how code-mixing can involve the change from one language to another within a single sentence, and how this process is controlled by constraints on the points at which switches are permitted — constraints which are presumably at least partly learned, since they vary between communities. We have also mentioned that speakers who code-mix can in fact distinguish their languages when this is necessary, for instance when speaking to a monolingual, so the code-mixing is done on the basis of what speakers know about their languages. — It is agreed by all sociolinguists that speakers can, and do, have an active as well as a passive command of more than one variety. The multilingual speaker is just a particularly glaring case of this phenomenon. Where some construction is typically used by some kind of speaker, or under some kind of circumstances, this fact will be known by many speakers, and will be exploited by them in understanding speakers of different dialects, in deducing social information about strangers, and even in trying to pass themselves off as speakers of other varieties. The most general point that emerges from all of this is that people know a wide range of things about constructions (as well as about sounds and words). Some of these things are internal to language — e. g. about the linguistic environments where the construction can occur, or about the language to which the construction belongs. But some of the known facts concern relations between the construction and non-linguistic categories, such as the type of person who typically uses it, or the circumstances under which it is used. Whether we include these known facts under the term ‘competence’ seems to be nothing but a matter of terminology. We can at least call them ‘knowledge of language’. What seems to be needed, if the findings of sociolinguistics are ever to interact with theories of syntax, is a general theory about knowledge of language which can accommodate all these kinds of knowledge. (One candidate for such a theory is Word Grammar — Hudson 1980, 1984 , 1985, 1986, 1987, 1988, 1990.) Alongside this theory we need a model of performance which takes account of the

1526

fact that speakers identify themselves as members of particular social groups, acting under particular social circumstances. The findings of sociolinguistics will provide a rich source of data against which the predictions of this model can be tested. Sociolinguists and specialists in language structure need to collaborate in this enterprise. At present sociolinguists probably deserve the criticism levelled by Rickford (1988), one of their leading members, for “ a tendency to be satisfied with observation and description, and [for being] insufficiently imbued with the thirst for theoretical explanation and prediction which drives science onward.” At the same time, most syntacticians pay far too little attention to the findings of sociolinguists in their theory-building activities.

11. References Aguirre, Adalberto. 1985. An experimental study of code alternation. International Journal of the Sociology of Language 53. 59—82. B ailey, Guy, and Natalie Maynor. 1985. The present tense of be in white folk speech of the Southern United States. English World Wide 6. 199—216. B autisita, Maria L. S. 1980. The Filipino bilingual’s competence. A model based on an analysis of Tagalog-English code-switching. Canberra. B ell, Allan. 1985. One rule of news English: geographical, social and historical spread. Te Reo 28. 95—117. B ickerton, Derek. 1973. The nature of a creole continuum. Language 49. 640—669. —. 1975. Dynamics of a Creole continuum. Cambridge. B lake, Robert. 1987. Is tola the head of S? I don’t want to decidir la cuestion, but I’m going to. Variation in Language, ed. by Denning et al., 22—34. B okamba, Eyamba. 1987. Are there syntactic constraints on code-mixing? Variation in Language, ed. by Denning et al., 35—51. Cedergren, Henrietta, and David Sankoff. 1974 . Variable rules: performance as statistical reflection of competence. Language 50. 333—355. Cheshire, Jenny. 1982a. Variation in an English Dialect. A sociolinguistic study. London. —. 1982b. Linguistic variation and social function. Sociolinguistic Variation in Speech Communities, 153—166. London. —. 1987. Syntactic variation, the linguistic variable and sociolinguistic theory. Linguistics 25. 257— 282. Chomsky, Noam. 1965. Aspects of the Theory of Syntax. Cambridge.

XXII. Syntax und Nachbardisziplinen

Denning, Keith, Sharon Inkelas, Faye McNairKnox, and John Rickford. 1987. Variation in Langauge. NWAV-XV at Stanford. Dik, Simon C. 1978. Functional Grammar. Amsterdam. Dines, E. 1980. Variation in discourse — and stuff like that. Language in Society 9. 13—31. Edwards, Walter. 1985. Intra-style shifting and linguistic variation in Guyanese speech. Anthropological Linguistics 27. 86—93. Fasold, Ralph. 1969. Tense and the form be in Black English. Language 45. 763—777. Ferguson, Charles. 1971. Absence of copula and the notion of simplicity: a study of normal speech, baby talk, foreigner talk and pidgins. Pidginization and Creolization of Languages, ed. by Dell Hymes, 141—150. London. Fillmore, Charles, Paul Kay, and Mary I. O’Connor. 1988. Regularity and idiomaticity in grammatical constructions. Language 64. 501—538. Gumperz, John, and Robert Wilson. 1971. Convergence and creolization. A case from the IndoAryan/Dravidian border in India. Pidginization and Creolization of Languages, ed. by Dell Hymes, 151—167. London. Halliday, Michael. 1985. An Introduction to Functional Grammar. London. Harris, John. 1984 . Syntactic variation and dialect convergence. Journal of Linguistics 20. 303—327. —. 1986. Expanding the superstrate: habitual aspect markers in Atlantic Englishes. English World Wide 7. 171—199. Holm, John. 1988. Pidgins and Creoles I. Theory and structure. London. Horrocks, Geoffrey. 1987. Generative Grammar. London. Hudson, Richard. 1980. Sociolinguistics. London. —. 1984. Word Grammar. Oxford. —. 1985. A psychologically and socially plausible theory of language structure. Meaning, Form and Use in Context, ed. by Deborrah Schiffrin, 150— 159. Washington, D. C. —. 1986. Sociolinguistics and the theory of grammar. Linguistics 24. 1053—1078. —. 1987. Grammar, society and the pronoun. Language Topics. Essays in Honour of Michael Halliday, ed. by Ross Steele & Terry Treadgold, 4 93— 505. Amsterdam. —. 1988. Identifying the linguistic foundations for lexcial research and dictionary design. International Journal of Lexicography 1. 287—312. —. 1990. A Word Grammar of English. Oxford. Kay, Paul. 1978. Variable rules, community grammar and linguistic change. Linguistic Variation: Models and methods, ed. by David Sankoff, 71— 83. London.

84.  Syntax and Sociolinguistics

—, and Chad McDaniel. 1979. On the logic of variable rules. Language in Society 8. 151—187. Keddad, Sadika. 1986. An analysis of French-Arabic code-switching in Algiers. London University PhD Dissertation. Kikai, Akio, Mary Schleppegrell, and Sali Tagliamonte. 1987. The influence of syntactic position on relativization strategies. Variation in Language, ed. by Denning et al., 266—277. Klein, Wolfgang, and Norbert Dittmar. 1979. The Acquisition of German Syntax by Foreign Workers. Berlin. Kroch, Anthony, and Cathy Small. 1978. Grammatical ideology and its effect on speech. Linguistic Variation. Models and methods, ed. by David Sankoff, 45—55. London. Labov, William. 1969. Contraction, deletion and inherent variability of the English copula. Language 45. 715—762. [Reprinted in Labov 1972a.] —. 1972a. Language in the Inner City. Philadelphia. —. 1972b. Sociolinguistic Patterns. Philadelphia, Oxford. —. 1972c. Negative attraction and negative concord in English grammar. Language 4 8. 773—818. [Reprinted in Labov 1972a.] Lavandera, B eatrice. 1978. Where does the sociolinguistic variable stop? Language in Society 7. 171—183. Lederberg, Amy, and Cesareo Morales. 1985. Code switching by bilinguals: evidence against a third grammar. Journal of Psycholinguistic Research 14 . 113—136. Lemieux, Monique. 1987. Clitic placement in the history of French. Variation in Language, ed. by Denning et al., 278—299. Miller, James, and Keith B rown. 1982. Aspects of Scottish English syntax. English World Wide 3. 3—17. Muthwi, Margaret. 1986. Language use in pluri-lingual societies: the significance of code-switching. Norwich: University of East Anglia MA Dissertation. Muysken, Pieter. 1981. Halfway between Quechua and Spanish. The case for relexification. Historicity and Variation in Creole Studies, ed. by Arnold Highfield & Albert Valdman, 52—78. Ann Arbor. Naro, Anthony. 1981. The social and structural dimensions of a syntactic change. Language 57. 63—98. Pfaff, Carol. 1979. Constraints on langauge mixing: intrasentential code-switching and borrowing in Spanish/English. Language 55. 291—318. Pooley, Timothy. 1988. A Sociolinguistic Study of Roubaix (France). London PhD Dissertation. Poplack, Shana. 1980. Sometimes I’ll start a sen-

1527

tence in Spanich Y TERMINO EN ESPANOL: toward a typology of code-switching. Linguistics 18. 581—618. Rickford, John. 1988. Connections between sociolinguistics and pidgin-creole studies. International Journal of the Sociology of Language 71. 51—57. Romaine, Suzanne. 1980. The relative clause marker in Scots English: Diffusion, complexity and style as dimensions of syntactic change. Language in Society 9. 221—247. —. 1984 . Relative clauses in child language, pidgins and creoles. Australian Journal of Linguistics 4 . 257—281. Sankoff, David and Suzanne Laberge. 1978. Statistical dependence among successive occurrences of a variable in discourse. Linguistic Variation. Models and methods, ed. by David Sankoff, 119—126. London. —, and Sylvie Mainville. 1986. Code-switching of context-free grammars. Theoretical Linguistics 13. 75—90. —, and Pierrette Thibault. 1978. Weak complementarity: tense and aspect in Montreal French. Paper presented at conference on syntactic change. Michigan. Sankoff, Gillian. 1973. Above and beyond phonology in variable rules. New Ways of Analyzing Variation in English, ed. by Charles-James Beiley & Roger Shuy, 44—61. Washington, DC. —, and Penelope B rown. 1976. The origins of syntax in discourse. Language 52. 631—666. —, and Pierrette Thibault. 1977. L’alternance entre les auxiliaires avoir et être en français, parlé à Montréal. Langue Française 34. 81—108. —, and David Vincent. 1977. L’emploi productif du ne dans le français parlé à Montréal. Le Français Moderne 45. 243—256. Schmid, B eata. 1986. Constraints on code-switching: evidence from Swedish and English. Nordic Journal of Linguistics 9. 55—82. Sciullo, Anne-Mariedi, Pieter Muysken, and Rajendra Singh. 1986. Government and code-switching. Journal of Linguistics 22. 1—24. Sells, Peter. 1985. Lectures on Contemporary Syntactic Theories. Stanford. Sobin, Nicholas. 1984 . On code-switching inside NP. Applied Psycholinguistics 5. 293—303. —. 1987. The variable status of Comp-trace phenomena. Natural Language and Linguistic Theory 5. 33—60. Sridhar, S. N., and Kamal Sridhar. 1980. The syntax and psycholinguistics of bilingual code-mixing. Studies in the Linguistic Sciences 10. 203—215. Thelander, Mats. 1976. Code-switching or codemixing. International Journal of the Sociology of Language 10. 103—123.

XXII. Syntax und Nachbardisziplinen

1528

Washabaugh, William. 1977. Constraining variation in decreolization. Language 53. 329—352. Weiner, Judith, and William Labov. 1983. Constraints on the agentless passive. Journal of Linguistics 19. 29—58. Winford, Donald. 1984 . The linguistic variable and syntactic variation in creole continua. Lingua 62. 267—288. —. 1985. The syntax of fi complements in Caribbean English Creole. Language 61. 588—624. —. 1988. The creole continuum and the notion of

the community as locus of language. International Journal of the Sociology of Language 71. 91—105. Wolfram, Walt. 1975. Variable constraints and rule relations. Analyzing variation in language, ed. by Ralph Fasold & Roger Shuy, 70—88. Washington, DC. Woolford, Ellen. 1983. Bilingual code-switching and syntactic theory. Linguistic Inquiry 14 . 520— 536.

Richard Hudson, London (Great Britain)

85. Syntax und Stilistik 1. 2. 3. 4. 5. 6.

1.

Das Verhältnis von Syntax und Stilistik Stilistische Vorgaben Wirkungsbereiche des Stils Stilwerte und Stileffekte syntaktischer Phänomene Stilfiguren Literatur

Das Verhältnis von Syntax und Stilistik

1.1.  Die Fülle von Arbeiten zur Syntax in den letzten Jahrzehnten hat kaum eine linguistische Teildisziplin unberührt gelassen. Im Vergleich damit nimmt sich der Beitrag, den die Syntax zur Entwicklung der Stilistik geleistet hat, eher bescheiden aus. Dies gilt zunächst in quantitativer Hinsicht: neuere Arbeiten zur Stilistik ziehen weniger syntaktische Kriterien heran als etwa solche der Lexik, der Sprechakttheorie und der Texttheorie. Es gilt aber auch in inhaltlicher Hinsicht: Die Stiltheorie gründet ihre Fortschritte auf andere Überlegungen als die in der Syntax in jüngster Zeit entwickelten. Dabei haben sowohl die Syntax als auch die Stilistik eine sehr lange Forschungstradition aufzuweisen. (Vgl. für den Stilbegriff Gumbrecht 1986). Doch je stärker sich die strukturellen und formalen Methoden in der Syntax durchsetzten, um so deutlicher wurde der ganz andere Zugang der Stilistik zum sprachlichen Phänomenbereich: Während die Syntax als eine primär deskriptive Teildisziplin der Linguistik mehr und mehr auch Erklärungsinteressen entwickelte und diese auf die Erfassung universaler oder einzelsprachlicher Erscheinungen konzentrierte, ist die Stilistik von

alters her eine evaluative Disziplin, die heute stärker auch eine beschreibende Komponente ausgebildet hat. Allerdings gab es auch herkömmlich durchaus eine Schnittstelle von Syntax und Stilistik: die jeweilige linguistische Norm. Doch während Normbezüglichkeit in der Syntax wiederum ein deskriptives Kriterium ist, das den Systemcharakter beachtet und auf der Ebene der langue operiert, stellt die stilistische Norm eine Bewertungskategorie bereit und zielt in der Beschreibungspraxis fast ausschließlich auf die Charakterisierung individueller Texte, ist also an der parole festzumachen. Während die Syntax als Teilgebiet der Grammatik Aufschluß über Systemregularitäten geben will, kontrolliert die Stilistik den Anwendungsfall. Das gilt für die ältere wie für die neuere Stilforschung; es trifft besonders zu für die Individualstilistik, die vor allem den Sprachgebrauch der Dichter beschreibt, aber auch für die Allgemeinstilistik, die Regularitäten für das Gelingen der Textproduktion zusammenträgt, denn die (individuelle) Stilanalyse und die (globale) Stillehre orientieren sich daran, daß ein Text als gelungen oder zu gelingender in Rede steht oder aber daß nachgewiesen werden soll, warum er mißlungen ist. In jedem Fall sind die stilistischen Aussagen über syntaktische Phänomene dabei evaluative Prozeduren, die immer auch andere sprachliche Elemente einbeziehen. 1.2.  Einerseits ist mithin die Syntax nur ein Anwendungsbereich von Stilistik neben anderen, denn die linguistischen Kategorien sind nicht Beschreibungsselbstzweck, sondern Mittel bei der Bewertung von Texten. Von daher sind auch die Kontroversen um die Stilistik als wissenschaftliche Disziplin zu ver-

85.  Syntax und Stilistik

stehen. Sie läßt sich als Teilgebiet der Rhetorik, als literaturwissenschaftliche Sparte, als eine eigenständige Disziplin oder aber als eine linguistische Teildisziplin auffassen. Für alle Zuordnungen lassen sich Gründe anführen: In der Rhetorik liegt eine Wissenschaftsdisziplin vor, die theoretische und praktische Zielsetzungen verbindet: einen Text so zu gestalten und zu präsentieren, daß er möglichst gut gelingt. Die Stilistik kann dabei aufgefaßt werden als eine Hilfswissenschaft, die die zu erwartenden linguistischen Normbezüge bereitstellt, etwa im Bereich der Satzlänge angibt, welches die optimale Zahl von Wörtern pro Satz ist, mit denen ein Text am überzeugendsten wird. Als literaturwissenschaftliche Teildisziplin wird die Stilistik als Wissenschaft aufgefaßt, die den Rang von Dichtung dadurch nachweist, daß sie gerade die sprachliche Normabweichung eines Dichters in Einzelzügen heraushebt. Als eigenständige Disziplin wird die Stilistik etwa von Ullmann (1972) angesehen, der zwei Haupttypen von Stiluntersuchungen unterscheidet: „diejenigen, die den Stil einer Sprache erforschen, und diejenigen, die auf den Stil eines Schriftstellers gerichtet sind“ (Ullmann 1972, 113). Er faßt die Stilistik als „Schwesterdisziplin“ der Sprachwissenschaft auf, „der es nicht um die Elemente der Sprache als solche zu tun ist, sondern um deren Ausdruckspotential. Die Stilistik hat demnach die gleichen Untergruppen wie die Sprachwissenschaft“ (124 ). Das sind für ihn Phonologie, Lexikologie und Syntax. Damit aber wird die linguistische Organisation des Phänomenbereichs vor- und zugegeben. Weil nun auch die Linguistik zunehmend wieder erkennt, daß nicht nur das exakte algorithmische Erkenntnisinteresse eine Rolle spielt, sondern daß dem Performanzbereich in seiner Weite nicht auszuweichen ist, werden auch Bewertungsverfahren als linguistische Aufgaben begriffen, zumal sich zeigt, daß auch diese regelgeleitet sind (vgl. besonders Heringer 1989). Daher darf die Stilistik als linguistische Teildisziplin angesehen werden. Dies ist etwa auch der ausdrückliche Ansatz von Sandig (1986). Die Stilistik läßt sich somit auffassen als dasjenige linguistische Teilgebiet, das dem paradoxen Phänomen Rechnung trägt, daß Äußerungen einerseits normgerecht, andererseits in Maßen normverletzend sein sollen, wenn sie gelingen sollen. Bevor bestimmt werden kann, welcher Platz der Syntax dabei zukommt, und welche syntaktischen Phänomene es sind, die hier

1529

eine Rolle spielen, muß die dazu nötige Auffassung von Stilistik, wie sie sich in den letzten Jahren herausgebildet hat, skizziert werden.

2.

Stilistische Vorgaben

2.1.  Für die Erfassung stilistischer Phänomene ist davon auszugehen, daß sprachliche Varianten vorliegen müssen, aus denen der Sprecher oder Schreiber zur Erzielung eines bestimmten kommunikativen Zwecks auswählen kann. Die stilistische Prozedur ist ein Wahlakt (vgl. Guiraud/Kuentz 1975, 121 ff). Stil bezieht sich auf die „Art der Handlungsdurchführung“ (Sandig 1986, 23). Teilweise wird auch die Auffassung vertreten, daß von abstrakten sprachlichen Ebenen ausgegangen werden kann, auf denen stilistische Differenzierungen erst operieren (Ohmann 1964 , Jacobs/Rosenbaum 1971, Rochemont 1978). Mit der stilistisch bedingten Auswahl aus den Möglichkeiten des Sprachsystems werden Entscheidungen getroffen, die geplante kommunikative Handlungen über ihr bloßes Angemessensein hinaus in besonderem Maße zum Gelingen führen sollen. Dies gilt nicht nur für sprachliche Stile, sondern für sozial relevante Verhaltensweisen schlechthin. Es wird von Wahlkampfstil, Stil in der Werbung, der Mode, weiter von Baustilen, von Epochenstilen, Fahrstilen und Behandlungsstilen, Gesprächsstilen (Sandig 1983) und vielen anderen -stilen gesprochen. In diesen Bezeichnungen tritt, wenn man sie insgesamt betrachtet, nicht nur das soziale Element hervor, sondern gleicherweise auch das individuelle. Stil ist ein paradoxes Phänomen, das sich in seiner kollektiv-normeinhaltenden und in seiner individuell-normdurchbrechenden Perspektive auf die spezifische Art einer Handlungsdurchführung bezieht. Was die Sprache betrifft, so ist das daraus resultierende Kriterium der verantworteten Wahl so zu sehen, daß ein konventionelles Raster von verbindlichen Vorgaben der Entscheidung, welche aus dem System zur Verfügung stehende Variante zu wählen ist, vorgeschaltet ist. Stil bezieht sich nie auf ein einzelnes sprachliches Phänomen, sondern immer auf größere Komplexe, die in ihrem syntagmatischen Aspekt nach einheitlichen Stilprinzipien gerade erst ausgewählt sind (vgl. Trabant 1986). D. h. Stil ist ein konstitutiver Faktor für die Zusammenbindung von Elementen zu kohärenten größeren Komplexen

1530

oder aber zu isolierten oder isolierbaren Entitäten, die rekurrent innerhalb größerer Einheiten als Konstanten auftreten, was allerdings häufig zu stereotypen Erwartungen führt. So wird andererseits auch die Einhaltung eines Gebotes gefordert, das innerhalb rekurrenter Elemente Variationen vorsieht, identische Wiederholungen sind — abgesehen von gewissen Stilfiguren — verpönt. Jeder Text enthält stilistisches Potential. Beim Rezipienten ruft es bestimmte Wirkungen hervor. Sandig (1986) spricht in bezug auf diese Dichotomie von „stilistischem Sinn“ und „Stilwirkung“. Dafür einige Beispiele: Ein Text, der Sachverhalte überwiegend nominal faßt an Stellen, an denen eine verbale Ausdrucksweise möglich ist, wird gern als ‘typischer Behördenstil’ eingestuft: Im Falle der Zusammenveranlagung von Ehegatten zur Einkommensteuer sowie bei Anwendung des Splittingverfahrens ergibt sich eine teilweise Verdoppelung der Abzugsbeträge (Broschüre des Finanzministeriums). Ein Textstück, das in einer syntaktischen Parallele eine Steigerung enthält wie in der folgenden Passage: Mein Vater las die Zeitung, er las die Saarbrücker Zeitung, er las nur die Saarbrücker Zeitung (Ludwig Harig, Ordnung ist das ganze Leben) verwendet eine Stilfigur und führt zu einer individuell stilistisch gelungenen Formulierung. Ein Text wie Was machen Sie denn da? Das ist doch — Ich werd Ihnen ...! gibt sich mit seinen Satzbrüchen als dem Stil der Umgangssprache zugehörig zu erkennen. 2.2.  Die Quellen für die stilistische Einordnung sind noch vielfältiger. In der Forschung werden verschiedene Möglichkeiten diskutiert (vgl. z. B. Spitzer 1961, Bally 1970, Turner 1979, Sanders 1973, Sanders 1977). Anzusetzen ist zunächst bei textuell oder diskursiv bedingten universalen Restriktionen für die Anwendung des Systems in der aktuellen sprachlichen Produktion. Dafür hat sich das Raster der sogenannten Darstellungsarten (Heinemann 1975, 268 ff), Texttypen (Werlich 1979) oder Vertextungsstrategien (Eroms 1983) bewährt. Die Vertextungsstrategien geben aus den in einer Sprache möglichen Ausdrucksweisen die nach dem jeweiligen Vertextungszweck möglichen vor. Es werden damit gleichermaßen sprechakttheoretisch relevante Gliederungen, als auch die kohärente Textbildung erfaßt, und zwar über systematisch gleichartige Bedingungen. Dabei läßt sich von einer vierfachen Gliederung ausgehen:

XXII. Syntax und Nachbardisziplinen

Erzählen, Beschreiben, Argumentieren und Anweisen. Dieses Gliederungsraster steht allerdings teilweise mit einer Kategorisierung, die nach mündlichem und schriftlichem Sprachgebrauch trennt, in Kreuzklassifikation. Statistische Untersuchungen haben ergeben, daß primär geschriebenes und primär gesprochenes Deutsch sich syntaktisch u. a. durch eine unterschiedliche Frequenz bei satzeröffnenden Adverbialia (35%: 17,5%) und Subjekten (60% : 75%) differenzieren lassen (Winter 1961). Die mit den Strategien verbundenen Handlungsziele und die sprachlichen Ausprägungen sind u. a. die folgenden: Mit dem Erzählen wird eine chronologisch ausgerichtete Geschehens- oder Handlungswiedergabe beabsichtigt. Es werden Aussagesätze gewählt. Vergangenes soll evoziert werden. Es tritt dafür die Fülle des verbalen Wortschatzes ein, die Satzbaupläne sind ebenfalls kaum restringiert. Die Sätze können in ihrer Aufeinanderfolge die Geschehensabfolge simulieren, sie gehorchen dann dem Ikonizitätsprinzip. Sie können temporale Staffelungen aber auch durch spezifischen Subjunktorengebrauch signalisieren. In der geschriebenen Sprache ist das Präteritum obligatorisches Erzähltempus; temporale Staffelungen werden gewöhnlich durch das Plusquamperfekt bezeichnet. Mein Vater schlug die Augen auf, sein Mittagsschlaf war beendet. Er nahm die Zeitung vom Radio, wenn er am Küchentisch geschlafen und die Zeitung hinter sich auf das Radio gelegt hatte; er nahm sie vom Fußboden, wenn er auf der Chaiselongue geschlafen und sie neben sich auf den Fußboden gelegt hatte, kniffte noch einmal über den Rand, daß es zischte, und Mutter sagte: „Wenn du die Zeitung zusammenmachst, dann läuft es mir eiskalt über den Rücken“ (Ludwig Harig, Ordnung ist das ganze Leben). Der Beginn dieses originalen Erzähltextes ließe sich z. B. in eine temporale Staffelung überführen: Nachdem mein Vater seinen Mittagsschlaf beendet hatte, schlug er seine Augen auf und nahm die Zeitung entweder vom Radio auf ... oder vom Boden ... Als Stilideal bei Erzähltexten gilt die Sparsamkeit bei der Verwendung von Subjunktoren, denn die sich ergebende Schachtelung der Sätze verhindert einen gleichmäßigen Erzählfluß. Auch Satzkonjunktionen und Konnektoren sind unwillkommen. Entweder werden sie als dem mündlichen Sprachgebrauch zugehörig angesehen, vor allem Anknüpfungen mit und, da oder gar und da, oder aber sie stellen Mittel anderer Vertextungsstrate-

85.  Syntax und Stilistik

gien dar, insbesondere des Argumentierens, wie Satzanfänge mit aber, denn, allerdings, nur und anderen Konnektoren, vor allem solchen, die in das Vorvorfeld des Satzes treten. Das Erzählen ist, genau wie die anderen Vertextungsstrategien eine idealtypische Strategie. Die Textsorten, in denen sie dominiert, wie Romane, Zeitungsberichte oder Lebensläufe enthalten immer auch andere Vertextungsstrategien. Jede stilistische Bewertung hat dies in Rechnung zu stellen. Ist das Erzählen im Prinzip chronologisch orientiert, ist das Beschreiben archetypisch räumlich ausgerichtet: Es zielt auf die topographische Abbildung und erfaßt nicht nur Räume im konkreten Sinn, sondern auch alle abstrakten Topologien, wie physikalische Modelle, kunstgeschichtliche Bestandsaufnahmen (Bildbeschreibungen) und Katalogisierungen. Die zeitliche Dimension spielt bei Beschreibungen nur eine sekundäre Rolle, etwa bei Modellvergleichen. Daher ist das archetypische Tempus hier das neutrale Präsens. Die beim Beschreiben eingesetzten Verben und ihre Satzbaupläne sind außerordentlich restringiert. Es handelt sich in erster Linie um Befindlichkeits- oder einfache Ordnungsprädikate wie sein, liegen, stehen, angrenzen, bestehen aus. Die bei reinen Beschreibungen sich leicht einstellende Monotonie erzwingt geradezu stilistische Maßnahmen, die von der Verbindung des Beschreibens mit oder der Einlagerung in andere Vertextungsstrategien, vor allem dem Erzählen, bis zum massiven Einsatz von Stilfiguren, etwa Personifizierungen, reichen. Bezeichnend sind feuilletonistische Länderbeschreibungen nach folgendem Muster: Der Kühler unseres Chevrolets brüllt vor Anstrengung. Die Klimaanlage keucht. Der Feuerball am Himmel zeigt kein Erbarmen mit Maschine und Mensch ... Der Wüstenwind ... saugt gierig die Schweißperlen von der Stirn ... in der heißesten Gegend der Welt: im Death Valley (Süddeutsche Zeitung vom 10. 4 . 1990). Syntaktisch relevante Stilzüge des Beschreibens sind außer den genannten spezifischen Verben mit ihren Satzbauplänen weitgehend konnektorlose Satzverknüpfungen. Die räumliche Orientierung wird durch räumliche Adverbien wie dort, daneben, oberhalb, links geleistet. Kennzeichnend ist weiterhin der starke Gebrauch von Adjektiven. Gerade in ihnen, die nicht nur als ‘schmückende Beiwörter’ aufzufassen sind, manifestieren sich die Stellungnahmen des Beschreibers, wie

1531

etwa im folgenden Satz: Mit ihren gebrochenen Kurven und eigenwilligen Verkürzungen gehört die ungewöhnliche Darstellung zu den originellsten Schöpfungen der römischen B arockarchitektur (Kunstführer Rom). Sie lassen sich darüber hinaus als Reflexe der Verschränkung des Beschreibens mit anderen Vertextungsstrategien deuten, so die Zeitadjektive als Indikatoren für Erzählstrategien (alte B auernhöfe, barocker Turmhelm) und viele bewertende Adjektive als argumentierende Einsprengsel: gelungener Umbau, unvermittelte Begrenzung. Das Argumentieren ist die Vertextungsstrategie, die bereits in reiner Form über die reichsten Möglichkeiten verfügt. Es ist die diskursive und handlungsbegründende Strategie schlechthin. Das Argumentieren zielt darauf ab, den Hörer oder Leser von der Richtigkeit von Thesen zu überzeugen und ihn zu damit verbundenen Handlungskonsequenzen zu führen. Im Gegensatz zu den Strategien des Erzählens und Beschreibens, die sich auch im Einzelsatz manifestieren können und demzufolge ein hohes Maß an stilneutralem Potential aufweisen, ist das Argumentieren auf satzübergreifende Schemata abgestellt. Prototypisch sind die mathematischen oder logischen Schlußschemata und ihre informellen gebrauchssprachlichen Varianten. Sie begegnen sowohl schriftsprachlich als auch im mündlichen Gebrauch. Textsorten, in denen das Argumentieren dominiert, sind die Rede, insbesondere die politische Rede, der Leitartikel, der Essay, die Erörterung und die literarische oder künstlerische Kritik. In die eigentlichen argumentierenden Passagen können längere erzählende oder beschreibende eingelagert sein. Dennoch sind argumentierende Texte leicht gerade an syntaktisch zu erfassenden Eigenschaften zu erkennen: Hier liegt die Domäne der Konnektoren, Konjunktionen und begründenden Partikeln wie aber, allerdings, denn, dennoch, doch, einerseits—andererseits, ja, jedoch, meiner Meinung nach, nicht doch, schließlich und ähnlicher Wörter und Wortgruppen. Die auf Satzform gebrachten Schlußschemata nach dem Muster zunächst/zuerst ..., sodann ..., nun aber ..., folglich oder die Konditionalsätze (wenn ... dann) führen zu überdurchschnittlich langen Sätzen. Jedoch sind lange Sätze durchaus nicht ausschließlich verbindlich für die Argumentation. Gerade besonders kurze Formen sind effektiv: Das wär’s. B asta, punktum. Quod erat demonstrandum (q e d).

XXII. Syntax und Nachbardisziplinen

1532

Mit dem Anweisen als letzter Vertextungsstrategie soll ein unmittelbarer und möglichst simultaner Handlungsvollzug erreicht werden. Hier sind wieder Formen möglich, die nur aus einem Satz bestehen und im ‘Satzmodus’ der Aufforderung begegnen wie Geben Sie mir bitte noch eine Tasse Kaffee! Andere Ausdrucksweisen, mit denen die Aufforderungsillokution zum Ausdruck gebracht werden kann, sind z. B. Fragesätze (Könnten Sie mir wohl bitte noch eine Tasse Kaffee geben?) oder Aussagesätze (Ich habe keinen Kaffee mehr), aus denen die Aufforderung nach konventionalisierten Diskursregeln zu extrahieren ist. Die zwischen den drei angeführten Beispielen zu bemerkenden kommunikativ relevanten Unterschiede, die sich zunächst in einer unterschiedlichen Höflichkeitsform bemerkbar machen, werden vielfach als Stilunterschiede erfaßt. Sie sind es jedoch nur, wenn sie als rekurrente Merkmale eines individuellen oder institutionalisierten Verhaltens begriffen werden. Deutlicher tritt ihr stilistisches Potential zu Tage, wenn Aufforderungen zu Anweisungsketten verbunden werden. Dann ist das Anweisen mit seinen sprachlichen Formen zunächst vom Argumentieren und sodann auch von den anderen Vertextungsstrategien abzuheben. Im Gegensatz zum Argumentieren verzichtet das Anweisen auf Strategien, mit denen Handlungsvoraussetzungen und -konsequenzen einsehbar gemacht werden. Textsorten, in denen dies gut deutlich wird, sind Ketten militärischer Befehle (stillgestanden, die Augen links, Augen geradeaus), Kochrezepte (Die Kartoffeln waschen und in der Schale gar kochen lassen, dann sofort abpellen und durch die Presse geben, 1 kg davon abwiegen und kochendheiße Milch darunter rühren) oder Bedienungsanleitungen (Für den Empfang des 2. Programms drehen Sie die Taste a in die obere Stellung. Drehen Sie nun den Senderwahlknopf, bis der gewünschte Sender oben steht. Anschließend wird mit demselben Knopf die B ildschärfe abgestimmt). In allen Fällen wird der sofortige Vollzug der Handlungsanweisungen gefordert, wobei die Sanktionen, die bei der Nichtbefolgung eintreten, ganz unterschiedlich sind. Die hier verwendeten sprachlichen Formen, insbesondere die syntaktischen, zeichnen sich dadurch aus, daß sie sich durch Kürze von solchen abheben, die bei der Verwendung in anderen Vertextungsstrategien expliziter sind, z. B. Imperative gegenüber sämtlichen anderen finiten Formen, damit im Zusammenhang die Tempuslosigkeit, die so-

fortigen Handlungsvollzug signalisieren soll. Werden tempusenthaltende finite Formen gewählt, dann stehen sie im Präsens oder im Futur I. Die beim Anweisen verwendbaren Prädikate sind so gut wie ausschließlich Handlungsprädikate. Daher kommen hier die mit Handlungsverben verbindbaren Satzmuster zur Geltung, allerdings stets unter starker Aktantenreduktion. — Die Vertextungsstrategien sind ein erster Filter für die stilistische Wahl und ihre Bewertung. Sie müssen weitgehend eingehalten werden und stellen verbindliche kollektive Auswahlen aus dem sprachlichen System dar. Wenn von ihnen abgewichen wird, ergeben sich auffällige Stileffekte. Auf der Ebene des Satzes manifestieren sich diese für das Erzählen etwa in der Verwendung von Frage- und Ausrufesätzen anstelle von Aussagesätzen: Und die Trommel? Und das Pfeifchen? Wie dumpf, aber tapfer war sie geschlagen, wie hell, aber lockend war sie geblasen worden! (Ludwig Harig, Ordnung ist das ganze Leben). Auch beim Beschreiben können Ausrufesätze das Beschriebene für den Leser eindringlicher machen: Temperaturen von mehr als 50 Grad Celsius sind hier keine Seltenheit — im Schatten wohlgemerkt! (Süddeutsche Zeitung, 10. 4 . 1990). Das Argumentieren wird hin und wieder mit Anweisungssätzen durchbrochen, wie Aufgepaßt! oder mit Fragesätzen, die den Adressaten in die Argumentation einbeziehen: Ein paar Wochen früher als geplant, werden sich George B ush und Michael Gorbatschow wieder gipfelwärts bewegen ... Weshalb die neue Eile? (Süddeutsche Zeitung, 10. 4 . 1990). Und das Anweisen wiederum läßt die ganze Fülle von kaschierenden Satzmodi zu, mit dem sich der Verbindlichkeitscharakter dieser Strategie herunterspielen läßt.

3.

Wirkungsbereiche des Stils

3.1.  Den bedeutendsten Systemfilter für die normativ-soziale Seite des Stilphänomens stellen die sogenannten Funktionalstile dar, die besonders in der osteuropäischen Linguistik bearbeitet worden sind (vgl. Fleischer/ Michel 1975, Riesel/Schendels 1975, Porsch 1981, außerdem Halliday 1985 a). Sie werden auch als Funktionalsprachen bezeichnet (Havránek 1976), doch hat der Begriff Funktionalstile für den Stil der öffentlichen Sprache, der Sprache der Wissenschaft, der Presse, der Belletristik und der Alltagskom-

85.  Syntax und Stilistik

munikation seine Berechtigung. Alle diese ‘Stile’ lassen sich auffassen als je spezifische Systemausschnitte zur Erzielung unterschiedlicher primärer kommunikativer Effekte. Die öffentliche Sprache hat zwei sehr unterschiedliche Ausprägungen, einmal die Sprache der Direktive, d. h. der Exekutive und des Rechtswesens, insbesondere die Verwaltungssprache, und sodann die Sprache der politischen Kommunikation mit mannigfaltigen Formen. „Die Verwaltungssprache wird im Dienst von Beamten und Verwaltungsangestellten zur Erfüllung staatlicher Aufgaben gebraucht. Sie hat deshalb einen offiziellen, unpersönlichen und funktionalen Charakter“ (Wagner 1972, 97). Wagner weist ihr einen eigenen „Sprachstil“ zu und registriert als syntaktische Besonderheiten im wesentlichen die folgenden: Gehäufte Verwendung von Funktionsverbfügungen („Vorgangsglieder“), ein überdurchschnittlich hoher Prozentsatz von Passivsätzen (26% gegenüber 15,3% in allgemeinsprachlichen Texten. Wagner 1972, 17), dazu eine hohe Zahl von Passivumschreibungen. Weitere Merkmale sind: Fehlen des direkten Imperativs, hohe Frequenz nominaler Umschreibungen, niedrigere Frequenz bei personalen Subjekten (35,5% gegenüber 4 6,2% bei allgemeinsprachlichen Texten), hohe Zahl von substantivischen Umstandsangaben, Attribute, insbesondere partizipiale, Attributschachtelungen „bis zu einer Abhängigkeit fünften Grades“: dem dort vorliegenden Antrag auf Durchführung des Verfahrens auf Untersagung der Ausübung des selbständigen Gewerbes (Wagner 1972, 4 1). Die Sätze sind entgegen landläufiger Erwartung kürzer als in anderen Textgenera, weisen aber dafür häufig lange Wörter auf. Kennzeichnend sind ferner einleitende präpositionale Fügungen wie im Nachgang zu unserem Schreiben vom ..., während Nebensätze nicht sehr häufig sind (Wagner 1972, 66). — Im Gegensatz zur relativ einheitlichen Behördensprache lassen sich für die eigentliche politische Sprache keine generellen Stilmerkmale angeben. Während die Behördensprache durch die Dominanz der Vertextungsstrategie des Anweisens in Kombination mit dem Beschreiben gekennzeichnet ist, herrscht in der politischen Sprache die Argumentation vor. In der Sprechweise der Politiker schlagen sich vor allem individuelle Eigenheiten nieder, was weithin auch erwartet wird. Parteigebundene Redeweise ist syntaktisch kaum nachzuweisen, abgesehen von den aus der DDR bis 1989 bekannten stereotypen

1533

syntaktischen Versatzstücken des folgenden Typs: Die amtierenden Generalsekretäre versicherten, alles zu tun, um vorhandene Möglichkeiten zur kontinuierlichen Entwicklung und Vertiefung der allseitigen B eziehungen zwischen beiden Bruderländern zu nutzen. Die wissenschaftliche Sprache bezweckt Vermittlung von Erkenntnissen. Die eingesetzten Vertextungsstrategien sind vor allem die Beschreibung und die Argumentation. Als Stilzüge werden bei Riesel/Schendels (1975) angegeben: Logik, Abstraktion, Genauigkeit, Klarheit, vollständige Ausformulierungen. Kennzeichnend für den Stil der Fachsprachen — die in den einzelnen Wissenschaften sehr unterschiedlich sein können — ist die hohe Frequenz des Substantivs, die Verwendung von Attributketten, insbesondere des erweiterten Partizipialattributs. Die verbale Ausdrucksweise tritt zurück (vgl. Beier 1979), das Passiv oder passivähnliche Konstruktionen sind dagegen relativ häufig (vgl. Admoni 1979). Die Wissenschaftssprache zeigt im Deutschen auch stilistische Gemeinsamkeiten mit der Behördensprache, was vielfach zur Kritik an der deutschen Wissenschaftssprache geführt hat (vgl. Teubert 1988). Braun (1987, 39) nennt als syntaktische Auffälligkeiten des wissenschaftlichen Stils: ‘valenzspezifische Verbformen’, wie z. B. ansprechen (‘ein Gedicht spricht an’: unbelebtes Subjekt), Nominalisierung von Verben und Adjektiven, Satzglieder anstelle von Gliedsätzen, Attribute anstelle von Attributsätzen sowie die besonders häufige Verwendung der 3. Pers. Indik. Präsens im Passiv oder mit Passivsynonymen. Die Sprache der Presse und Publizistik, die in der Funktionalstilistik teilweise nicht als eigener Stil beschrieben wird, weist trotz einer sicher einheitlichen Grundintention, nämlich Unterrichtung der Öffentlichkeit, außer dem sehr pauschalen Stilzug der allgemeinen Verständlichkeit keine Besonderheiten auf. Vielmehr kommen die generellen sprachlichen Systemtendenzen in der Pressesprache am klarsten zur Geltung. So ließe sich dieser Funktionalstilbereich mit gewissen Abstrichen für die geschriebenen Sprachformen als stilistisch neutraler beschreiben. Die Untersuchungen, die sich mit Entwicklungstendenzen in der deutschen Gegenwartssprache befassen, stützen sich deshalb auch sehr stark auf die Pressesprache (Braun 1987, Sommerfeldt 1988, Glück/Sauer 1990). Die systemnatürlichsten Tendenzen jedoch weist die meist mündliche Alltagskommuni-

1534

kation auf, denn bei ihr sind die Normeingriffe, die sich bei der Schriftsprache mit ihrer jahrhundertelangen Tradition in den einzelnen Funktionalstilen bemerkbar machen, am wenigsten zu spüren. Allerdings sind gerade in der Alltagskommunikation, die, abgesehen von den Dialekten, als überhaupt untersuchenswerter Bereich für standardsprachnahe Register erst seit verhältnismäßig kurzer Zeit eine Rolle spielt (bahnbrechend Riesel 1964 ), die Probleme der Abgrenzung von der geschriebenen Sprache oder von regional geprägter ‘Umgangssprache’ noch nicht endgültig geklärt (vgl. insbesondere Bichel 1973). Als Stilzüge der Alltagskommunikation werden bei Riesel/Schendels (1975, 167) u. a. Situationsbezug, Ungezwungenheit, Emotionalität, Bildhaftigkeit und Individualität genannt. Es treten alle Vertextungsstrategien auf. Solange die Syntax auf dem Hintergrund der geschriebenen Sprache erfaßt wurde, mußte der Satzbau der gesprochenen Sprache anstößig oder defektiv erscheinen. Allerdings hat eine solche Sichtweise in der Dialektologie nie gegolten (vgl. etwa Schiepek 1899/ 1908 und Henn-Memmesheimer 1986). Erst durch Arbeiten wie die von Leska (1965), Betten (1976) oder Rath (1979) konnten wesentliche funktionalstilistische Eigenheiten der gesprochenen Sprache geklärt werden. Während Leska ihre Untersuchung der Syntax gesprochener Sprache noch auf die relative Nähe zur Schrift abstellt, rechtfertigt Betten (1976) z. B. Ellipsen, Anakoluthe und Parenthesen als normale, d. h. nicht abweichende und auch nicht stilistisch markierte Formen der Syntax gesprochener Sprache, was bei Rath (1979) weiter ausgearbeitet und um den Gesichtspunkt erweitert wird, daß die Ellipse wie die Ausklammerung funktional, d. h. dem mündlichen diskursiven Sprachgebrauch mit seinen spezifischen Bedingungen angemessen ist. Damit sind derartige syntaktische Stilzüge im Funktionalstil der mündlichen Alltagskommunikation unmarkiert. Doch ergeben sich hier auch Unterschiede bei den einzelnen Textsorten, die im mündlichen Sprachgebrauch begegnen (vgl. Rath 1979, 164 ). Erst im Vergleich mit anderen Funktionalstilen und den dort begegnenden äquivalenten Formen sind Ellipse und Ausklammerung im Hinblick auf den Systemplatz zu bestimmen. Wenn als Funktion der Ausklammerung von Rath u. a. Gedächtnisentlastung, Hervorhebung der ausgeklammerten Glieder und leichtere Rezipierbarkeit genannt werden (Rath 1979, 165—184 ), wird

XXII. Syntax und Nachbardisziplinen

deutlich, daß dies Faktoren sind, die teilweise auch für die geschriebene Sprache gelten. Dennoch ist damit nicht ausgemacht, ob die Ausklammerung aufs Gesamtsystem des Deutschen gesehen in der Gegenwartssprache die unmarkierte Form geworden ist. Funktionalstilistisch ist die Ausklammerung daher ein nicht eindeutiger Fall für stilwertgebende Faktoren. Das gleiche trifft für die Ellipsen zu. Andere funktionalstilistische Charakteristika der gesprochenen Sprache wie die Anakoluthe und Prolepsen erzielen dagegen beim Vorkommen in den übrigen Funktionalstilen Effekte, in ihrem Ausgangsstilbereich sind sie stilistisch nicht markiert. Als letzter Funktionalstil ist die Belletristik zu nennen. Für die Literatur lassen sich überhaupt keine funktionalstilistischen Vorgaben ansetzen. Dagegen gelten (literarische) Textsortenregeln und die Vorgaben der Vertextungsstrategien, allerdings in einer spezifischen Ausprägung, die sich nicht eindeutig auf die nichtliterarischen Vertextungsstrategien abbilden läßt: In der Epik dominiert zweifellos das Erzählen, das Drama simuliert diskursive Kommunikationsformen. Für die Lyrik gelten die stärksten individuellen Eigengesetzlichkeiten, während sich politisch engagierte Dichtung mit Appelltexten in Verbindung bringen lassen kann. An der Literatur hat sich daher insbesondere die individualstilistische Konzeption der Stilistik ausgebildet. Das gilt auch für die Syntax. Bei der starken Vorbildwirkung von Literatur werden andererseits individualstilistische Stilprinzipien als normativ-verbindliche oder zumindest auch für die nichtliterarische Sprache anzustrebende ausgegeben. Für das Deutsche wird gern das Stilideal der deutschen Klassik, vor allem die Schreibweise Goethes und Schillers und der älteren Wissenschaftssprache herangezogen (vgl. Meyer 1906, Engel 1914 ). Doch gilt dies nicht ohne Einschränkungen. Bezeichnend ist die Kritik von Reiners (1976, 24 ): „Selbst Männer, deren Namen wir mit tiefster Ehrfurcht nennen ... Kant und Hegel, Klopstock und Herder, Fichte und Schelling, Wilhelm und Alexander von Humboldt, Ranke und Lamprecht sind nicht Meister, sondern ‘Büßer der deutschen Sprache’ gewesen“. Reiners kritisiert im folgenden vor allem den Satzbau. Seine ‘Stilkunst’ ist zudem überhaupt eine stark normativ angelegte Stilistik, in der die Funktionalstile allenfalls pauschaliert werden. Für die Syntax in der Dichtkunst lassen sich keine verbindlichen Regeln angeben. Denn es ist gerade die extensive Nutzung ei-

85.  Syntax und Stilistik

ner syntaktischen Eigenheit oder die Normdurchbrechung, die den Stilzug der Unverwechselbarkeit und Einheitlichkeit oder aber der Abweichung schlechthin hervorruft. U. a. sind an dichterischer Syntax der jüngeren deutschen Sprachgeschichte folgende Phänomene herausgearbeitet worden (vgl. vor allem Langen 1957, Betten 1987 und Admoni 1987): Die Satzantithese ist kennzeichnend für den barocken Alexandriner (der Schultern warmer Schnee wird werden kalter Sand), der kunstvolle Periodenbau gilt in der barocken Prosa überhaupt, während in den eher volkstümlichen Textgenres die assoziative Reihung vorherrscht. Beides steht im Gegensatz zum „Taciteischen Kurzstil“ (Langen 1957, Sp. 986), der auch gepflegt wird. Mit dem Ende des 18. Jahrhunderts dringt das Natürlichkeitsideal, aus Frankreich kommend, in die Dichtersprache ein. Damit werden Formen der gesprochenen Sprache aufgenommen. Im Pietismus finden sich häufig dynamisierende verbale Präfixformen, überhaupt herrscht der Verbalstil vor. Bei Gottsched wird endgültig der Kanzleistil als Ideal verworfen. Ungezwungen sollen kürzere und längere Sätze miteinander abwechseln (Langen 1957, Sp. 1036). Auch die Wortfolge soll dem natürlichen Fluß folgen. An Lessings Stil rühmen schon die Zeitgenossen die entwickelnde, zum gedanklichen Mitvollzug auffordernde Periodenbildung, Klopstock gilt in stilistischer Hinsicht als Muster für die Verlebendigung von Beschreibungen und Erzählungen durch den ausgreifenden Gebrauch von Bewegungsverben. Im Sturm und Drang ebenso wie bei Herder dringen in starkem Maße Eigenheiten der gesprochenen Sprache in die schriftsprachigen Register, verstärkt und gesteigert beim jungen Goethe. In der deutschen Klassik wird das Natürlichkeitsideal wieder verlassen. Der Satzbau wird formell und distanziert, er verliert diskursive Elemente weitgehend. Da das syntaktische Stilideal der deutschen Klassik jedoch außerordentlich einflußreich war, unterliegen in den Stilistiken seit der Zeit Partikeln, Klammerformen, aber auch der lange Periodenbau weitgehend einem Verdikt. (Vgl. etwa Engel 1914 , 289 ff, der Schachtelsätze, Treppensätze (daß-daß-daß-Sätze) als „Stopfstil“ kritisiert). Mit der Romantik werden archaische Satzmuster, aber auch volkstümliche und mehr noch vermeintlich volkstümliche Formen für die gehobenen Stilschichten erschlossen, etwa die Ellipsen und der Partikel- und Konnektorengebrauch der Alltagssprache, was zu

1535

vielfältigen Formtypen führt (vgl. Simmler 1986). Bei Heine finden sich dann u. a. parataktische Reihungen, anaphorische Satzbindungen und die Ironisierung volkstümlicher syntaktischer Formen. Dem späten 19. Jahrhundert wird, auch in der Syntax, das Etikett „pathetischer Stil“ beigelegt (Langen 1957, Sp. 1325), gipfelnd etwa in den Reden Wilhelms II, dessen Satzbau durch den Wechsel von markiger militärischer Kommandosprache und ausuferndem Periodenbau gekennzeichnet ist. Aus dem Stilideal des Realismus, dem jedes Pathos fremd ist und der bewußt wieder alltagssprachliche Formen aufnimmt und sie mit solchen anderer Funktionalstile mischt, sind die heute gültigen Stilideale entstanden. Dies führt einerseits zu sehr individuellen Stilformen, andererseits ist Funktionalität, Angemessenheit, Sachlichkeit, kurz „kommunikative Adäquatheit“ (Sanders 1990, 66) das wichtigste gegenwärtige gebrauchssprachliche Stilideal. — Für das Französische sind u. a. von der literarischen Stilistik an syntaktischen Phänomenen die Stilwerte der Inversion untersucht werden (Ullmann 1972), für das Englische die Zuordnung von Verbgebrauch und Personencharakteristik (Kennedy 1982) oder der Gebrauch von syntaktischen Schablonen (Ringbom 1975). Die dichterische Sprache steht nicht nur im Wechselspiel mit der Allgemeinsprache, sie erschließt vor allem auch neue Ausdruckssysteme. Für sie gilt ganz besonders das Wort von Spitzer (1961, II, 517), Grammatik sei gefrorene Stilistik. In der Syntax ist allerdings weniger die Entwicklung gänzlich neuer Ausdrucksweisen zu verzeichnen, sondern vielmehr die Verwendung von markierten Konstruktionen, die dadurch ihre Markierung, z. B. als funktionalstilistisch der Umgangssprache zugehörend, verlieren können. Während im Bereich des Wortschatzes und der Metaphernverwendung das Innovationspotential der Literatur unerschöpflich ist, lassen sich in der Syntax eher konsequente und überzeugende Nutzungen seltener oder eben markierter Konstruktionen nachweisen. Ein Beispiel dafür ist im Deutschen die Wiederdurchsetzung archaischer Verbstellungsformen, z. B. die Spitzenstellung des Verbs in Aussagesätzen. Sie wurde in der Zeit der deutschen Klassik in der Lyrik (Sah ein Knab ein Röslein stehn) aufgenommen, fand sodann Verwendung in dichterischer Prosa und läßt sich in der Gegenwart bis in die Sachprosa hinein nachweisen. Dort gilt sie aller-

1536

dings eher als emphatische, d. h. markierte Variante der Zweitstellung (vgl. Erben 1980, 119 f, 270). Auch für die stärkere Akzeptierung von Charakteristika der Syntax gesprochener Sprache sowie für den Gebrauch von Partikeln läßt sich deren ungezwungene Verwendung in der Dichtung ansehen. 3.2.  Die Funktionalstile mit ihren z. T. recht global formulierten Anforderungen sind ein Verwendungsraster für sämtliche sprachlichen Mittel. Sehr viel genauere Anforderungen an die stilistischen Vorgaben lassen sich über die Charakterisierung von Textsorten und ihren sprachlichen Bedingungen erzielen. So ist, was die Syntax betrifft, etwa für die Textsorte ‘Wetterbericht’ festgestellt worden, daß hier ein besonderer Typ von Kurzsätzen dominiert (vgl. Sandig 1970). Die Sätze in informellen Briefen zeichnen sich durch Kürze, häufige Kontaktsignale und umgangssprachliche Erscheinungen aus. Märchen weisen ebenfalls vor allem kurze Sätze mit wenig Einschachtelungen auf. Es treten gehäuft Geschehens- und Handlungsprädikate auf, dazu kommen feste Anfangs- und Schlußformeln, die zu den bei den Erzähltexten gängigen Stilzügen dazu treten (Gobyn 1984 , 224 f, vgl. auch Gobyn 1984 a). Wichtig aber ist auch die Tatsache, daß nicht nur beim Märchen, sondern bei so gut wie allen Textsorten die textsortenspezifischen Charakteristika nur eine kleine, wenn auch differenzierende Menge darstellen. Diese ist innerhalb größerer Ordnungskategorien mit erheblich weiter gefaßten, aber dennoch gegeneinander abgehobenen Formen zu betrachten, also im Rahmen der Funktionalstile, der Vertextungsstrategien und der kommunikativen Bereiche ‘mündlicher gegenüber schriftlicher Sprachgebrauch’. Und das gibt auch die Berechtigung ab, von diesen Kategorien auszugehen. Andernfalls ergäbe sich in Hinblick auf eine komplette textsortenstilistische Darstellung eine nicht zu bewältigende Redundanz. Stilbewertende Aussagen beziehen sich stets auf sämtliche Vorgabenbereiche, die systematisch oder normativ geregelt sind. Sie beziehen sich jedoch immer auch auf die Bewertung des individuellen Gelingens, das, von standardisierten Textsorten abgesehen, niemals ausschließlich in der Einhaltung der Normen, sondern in ihrer Transzendierung besteht. Die Fülle der Textsorten ließe sich unter diesem Aspekt skalieren nach solchen, die wenig bis gar keine individuellen Stilzüge aufweisen, solchen, bei denen die individuellen Stilanforde-

XXII. Syntax und Nachbardisziplinen

rungen sich mit normativ geforderten die Waage halten und solchen, bei denen stilistische Idiosynkrasien dominieren. In die erste Gruppe gehören Textsorten wie liturgische Texte, Behördenbriefe, Wetterberichte, Verkehrsmeldungen und Zeitungsnachrichten. Doch finden sich auch hier zunehmend Erwartungen auf individualstilistische Gestaltungen, etwa bei den Wettermeldungen der regionalen Fernsehprogramme. Der zweiten Gruppe lassen sich Textsorten wie Lebensläufe, Bewerbungsschreiben, Essays, Festreden und Regierungserklärungen zuordnen, bei denen zumindest ein individuelles Stilmerkmal erwartet wird. In die dritte Gruppe gehören alle literarischen Texte, allerdings durchaus auch mit unterschiedlichen Anforderungen an den Grad der individuellen Gestaltung. So sind Gedichte des Barock und Kunstmärchen nach relativ festen Schemata gebaut, während Romane, besonders aber die moderne Lyrik sich zunehmend als individuelle Textsorten schlechthin begreifen lassen. Bahnbrechend für die stilistische Interpretation, bei der literarische und linguistische Kategorien gleichermaßen herangezogen werden, ist Spitzer (1961/1926). Für die moderne englische Lyrik zeigt Ohmann (1967) an Beispielen wie the river wended vales (Dylan Thomas), daß der Interpretationsprozeß das linguistische Wissen des Lesers aktiviert und immer mehr als eine Deutung erlaubt. Vgl. ferner Baker (1967). Auch nichtliterarische Texte lassen sich hierher rechnen, etwa Freundesbriefe, Musikkritiken und Plädoyers. Die individuellen Faktoren sind jedoch in bezug auf die Syntax wiederum insgesamt sehr ungleichmäßig verteilt. So wird, von der Lyrik abgesehen, die individuelle syntaktische Gestaltung sehr häufig durch den Gebrauch von Stilfigur erzielt (vgl. 5.), die ja gleichsam ‘zugelassene Verstöße’ gegen die Norm darstellen. So läßt sich auch von daher sagen, daß Stileigentümlichkeiten der Syntax seltener sind als solche des Wortschatzes. 3.3.  Alle stilbewertenden Aussagen haben den Bedingungen zu genügen, die sich aus den normativen Erwartungen und den individuellen Abweichungen ergeben. Da sich Stil auf die Wirkung von Texten bezieht und sich stilistischer Sinn als ein Moment der Verstärkung der Sprecherintention begreifen läßt, sind in dieser Hinsicht alle Einhaltungen der Norm ohne Information, sie sind stilistisch unmarkiert. Sie sind aber nur dann neutral, wenn sie sich in allen Funktionalstilen ver-

85.  Syntax und Stilistik

wenden lassen. Informativ in bezug auf die Verstärkung des stilistischen Sinnes sind alle die sprachlichen Elemente, die entweder zu Paradigmen eines bestimmten Funktionalstils oder einer Textsorte gehören, d. h. bereits Auswahlen aus dem System der Sprache darstellen, die also kollektiv vorgenommen worden sind und damit konventionell bestehen. Diese sprachlichen Mittel weisen von Haus aus einen bestimmten Stilwert auf. Alle Abweichungen von diesen Vorgaben führen zu Stileffekten. Sie sind als sprachliche Phänomene markiert. Dies läßt sich leicht an Beispielen aus dem Wortschatz beleuchten: Wörter, die an allen Stellen der Kommunikation begegnen können, sind stilneutral. Stilneutralität läßt sich jedoch auf zweierlei Weise erreichen: Entweder existiert nur ein einziges Wort zur Bezeichnung eines Gegenstandes oder eines Sachverhaltes, oder aber aus einer Synonymenreihe wird der unmarkierte Ausdruck gewählt. Das ist der Ausdruck, der in allen Funktionalstilen verwendbar ist. Autosemantika und Synsemantika, z. B Pronomina und Funktionswörter, verhalten sich im Prinzip gleich. So sind Ausdrücke wie Schreibmaschine, Wolke, er, sie neutral, weil zu ihnen keine Synonyme existieren. Damit ist nicht ausgeschlossen, daß sich zu ihnen ad-hoc-Synonyme bilden lassen, die dann starke Stileffekte ergeben. Stilneutral sind auch Wörter wie Auto, Kind, Kleidung oder dennoch, aber aus einem anderen Grunde: Sie neutralisieren Paradigmen, in denen Ausdrücke wie Wagen (‘gehoben’), Kiste (‘umgangssprachlich’), Gör (‘umgangssprachlich, regional’), Klamotten (‘umgangssprachlich’), Bekleidung (‘fachsprachlich’), gleichwohl (‘schriftsprachlich’) und dessenungeachtet (‘behördensprachlich’) bestehen. Diese Ausdrücke ergeben in ihren Herkunftsbereichen Stilwerte. Für den Satzbau gilt im Prinzip die gleiche Markiertheitsskala, nur ist die Menge der neutralen Elemente erheblich größer als beim Wortschatz, bzw. die der Synonymierelation hier parallele Paraphraserelation ist in viel stärkerem Maße text- oder diskursfunktional zu erklären. So ist etwa das Passiv, das sich anstelle des Aktivs wählen läßt, nicht stilistisch, sondern zunächst morphologisch markiert. Gleiches gilt für die Funktionsverbfügungen, die an die Stelle von einfachen Verbkonstruktionen treten können. An einer beliebigen einzelnen Textstelle sind alle diese Konstruktionen stilistisch unbewertbar. Erst über die Gesamttexte, die im Hinblick auf

1537

Funktionalstil und Textsorte zu sehen sind, lassen sich in solchen Fällen quantitative Aussagen treffen, nämlich ob die zu erwartende Frequenz eintritt; dann ergeben sich Stilwerte, oder aber ob sich Abweichungen davon nachweisen lassen. In diesen Fällen läßt sich von Stileffekten sprechen. Die Masse der durch Mittel des Satzbaus erzielten Stileffekte aber wird durch unerwartete, seltene oder veraltende Konstruktionen, oder aber durch die Stilfiguren erzielt, die unmittelbare Wirkung hervorrufen. Syntaktische Stileffekte aller Art gehören mit allen sonstigen stilistisch markierten Elementen eines Textes zu seinem stilistischen Potential, das in einer evaluativen Prozedur bewertet wird, d. h. je nach Rezipient oder Analysator zu unterschiedlichen Ergebnissen führen wird.

4.

Stilwerte und Stileffekte syntaktischer Phänomene

Im folgenden werden die wichtigsten syntaktischen Phänomene, die in den Stilistiken behandelt werden, aufgeführt und über die in 2. und 3. vorgenommenen Einordnungen hinaus kommentiert. 4.1.  Die Satzarten weisen stilistisches Potential in dem Sinne auf, daß der gleiche Sachverhalt durch den Wechsel von Aussage-, Frage-, Aufforderungs- und Ausrufesätzen zum Ausdruck gebracht werden kann. Dies gilt im Prinzip für jeden möglichen Wechsel zwischen den Satzarten, doch läuft es im allgemeinen darauf hinaus, anstelle von Aussage- oder Aufforderungssätzen andere Satzmodi zu wählen. Damit wird zunächst dem Stilerfordernis der Variation Genüge getan. Darüber hinaus gelten die meisten Formen, die anstelle von Aussagesätzen gewählt werden, als expressiv markiert. Der informationsvermittelnde feststellende Aussagesatz Er sieht alt aus läßt sich in einen Fragesatz Sieht er nicht alt aus? oder einen Ausrufesatz Wie alt sieht er aus! umformen. Besonders dieser Satzmodus erlaubt die zusätzliche Signalisierung von emotionalen und situationellen Konnotationen. Sein Einsatz ist stark textsortengebunden. In der Wissenschaftssprache ist er kaum anzutreffen, während er in der Alltagskommunikation und in der öffentlichen Sprache, sowie in der Dichtung, vor allem in der Lyrik, häufig begegnet. Er gilt genau wie der Fragesatz in solchen Funktionen auch als Stilfigur. Fragesätze können

1538

auch implizite Negierungen verhüllt zum Ausdruck bringen: Wer hätte das gedacht? (= ‘Niemand’) oder umgekehrt Bejahungen beinhalten: Wer würde hier nicht helfen wollen? (= ‘Jedermann’). Anstelle von Aufforderungssätzen wie Machen Sie bitte das Fenster zu finden sich ebenfalls stark konventionalisierte Formen mit Fragesätzen: Könnten Sie bitte das Fenster zumachen? und Aussagesätzen Hier zieht es, bei denen indirekte Sprechakte vorliegen. Fragesätze und Ausrufesätze anstelle von möglichen Aussagesätzen bringen diskursive Formen in die Texte hinein; sie simulieren direkten Sprecher-Hörer-Kontakt. Es finden sich aber auch Aussagesatzformen anstelle von Aufforderungen: B ei dem Umfang dieser B estellung erwarte ich von Ihnen einen Sonderrabatt (Ermert 1979, 101). 4.2.  Die Satzbaupläne (Satzmodelle) sind als solche bis auf wenige Ausnahmen stilistisch neutral. Doch ist ihre Verteilung nach Vertextungsstrategien und Funktionalstilen sehr unterschiedlich geregelt. Zu den von vornherein stilistisch markierten Satzmodellen gehören im Deutschen vielfach solche, die den Genitiv enthalten: sich einer Sache entledigen, sich eines Vorkommnisses erinnern, jemanden einer Sache gemahnen gegenüber stilistisch neutralen Formen: sich von einer Sache freimachen, sich an ein Vorkommnis erinnern, jemanden an eine Sache erinnern. Die Genitivfügungen stellen Archaismen dar und sind auf gehobene Textsorten beschränkt. — Das Vorkommen von Satzmodellen in unterschiedlichen Textsorten regelt sich hauptsächlich danach, ob erzählt, beschrieben, argumentiert oder angewiesen wird. Nur sekundär ist „die Thematik, der Inhalt der Mitteilung“ (Riesel/ Schendels 1975, 163) dafür verantwortlich. Denn ob Handlungsprädikate, die die am reichsten distribuierten Satzbaupläne eröffnen, oder die zahlenmäßig sehr eingeschränkten Befindlichkeitsprädikate oder Prädikatsnomina (B onn liegt am Rhein. Der Ottomotor ist eine Verbrennungskraftmaschine) begegnen, richtet sich nach der Darstellungsart. Die Möglichkeit zu individueller stilistischer Wahl besteht in der Mischung der Vertextungsstrategien, indem z. B. in erzählenden Partien argumentierende Passagen oder in beschreibende Partien erzählende mit jeweils anderen Satzbauplänen eingelagert werden. — Verblose Sätze, die leicht als stilistisch markiert angesehen werden können, sind eigentlich nur konsequente Anwendungsformen des Beschreibens, bei dem die referen-

XXII. Syntax und Nachbardisziplinen

tielle Information dominiert und die an Verben gebundene prädikative auf ein Minimum beschränkt ist, z. B. Träume, Träume — Flakkerndes und Flammen, B ildung, ewig dem Vergängnis nah, Räume, Räume — Suchen und Verdammen, Schatten, Schreie der Apostata (Gottfried Benn). Verblosigkeit in Erzähltexten signalisiert dagegen Spannung und Fülle, ist also ein Mittel, über das Erzählte hinaus Expressivität durch den raschen Wechsel der erzählerischen ‘Momentaufnahmen’ auszudrücken, z. B. Heute nur noch Tee, noch einmal die ganze Untersucherei, nachher ist man erledigt. Morgen endlich Operation (Max Frisch, Homo Faber). Derartige Verkürzungen, die den Satzmodellen ihr strukturelles Zentrum entziehen, sind Fälle von Ellipsen. Auch bei der Ellipse sind neutrale Vorkommen und stilistisch markierte zu unterscheiden. Alle Formen, die Verkürzungen im Diskurs betreffen, sind natürliche Ausdrucksformen, allerdings mit diskursspezifischen Eigentümlichkeiten, etwa bei der Behördenbefragung: Ihr Name? — Meier. — Geboren in? — München oder im ungezwungenen Alltagsgespräch: Wo kommst du denn her? — Aus der Stadt. — Jetzt schon? Als „usuelle Ellipsen“ (Fleischer/Michel 1975, 178), die in Grußformeln und Floskeln wie Guten Morgen, Tschüß, Mahlzeit begegnen, sind sie ebenfalls unmarkierte Formen. Hier wären expandierte stilistisch markiert: Ich wünsche Ihnen einen guten Morgen! Als Stileffekte erzielendes Mittel wird die Ellipse in sehr vielen Textsorten eingesetzt. Von der Verwendung als Stilfigur abgesehen (vgl. 5.), lassen sich Ellipsen danach klassifizieren, welche Syntagmen ausgelassen sind. Es läßt sich sagen, daß im allgemeinen nur restituierbare Glieder ausgelassen werden können, also thematische. Das Verb, insbesondere das finite Verb, kann außer in Fällen wie den obigen nur in standardisierten Formeln wie Eintritt verboten, Da capo al fine, Schluß jetzt! weggelassen werden. Dies gilt analog auch für Textsorten wie Wetterberichte oder Verkehrsmeldungen, wie z. B. Auf den Fernstraßen lebhafter Verkehr. An den Grenzen Wartezeiten von etwa einer Stunde. Die stilistisch markierte Auslassung des Verbs findet sich in expressiver Alltagssprache, sowie in der Pressesprache und in der Dichtung (s. o.). — Das Subjekt kann zur Erzielung redundanzfreier kommunikativer Strukturen ausgelassen werden, dies wird in bestimmten Textsorten benötigt, etwa im Telegramm: ankomme um drei Uhr, aber auch in der Alltagssprache, dort ganz allgemein: komme gleich,

85.  Syntax und Stilistik

macht nichts. Ähnliches gilt für Auslassung bei passivischen Formen: wird sofort erledigt. In derartigen Fällen werden allenfalls funktionalstil- oder textsortenspezifische Stilwerte transportiert, jedoch keine Stileffekte erzielt, weil Ellipsen erwartet werden können. Davon zu trennen sind Weglassungen fakultativer Ergänzungen, wie er raucht, er trinkt oder sie liest, sie schauen zu. Entweder handelt es sich um generische Aussagen (‘er ist Raucher’) oder um grammatisch geregelte Nichtsetzungen von Aktanten bei situationeller oder kontextueller Eindeutigkeit. 4.3.  Der Satzumfang und die Satzkomplexität sind sowohl funktional- als auch individualstilistisch wichtige Gestaltungsmittel. Eine Informationsmenge kann auf mehrere Sätze verteilt oder in wenige oder gar einen einzigen Satz verdichtet werden. Diese grundsätzlich unterschiedlichen Möglichkeiten lassen sodann die Ausgestaltung nach hypotaktischer gegenüber parataktischer Konstruktionsweise, jeweils mit oder ohne kompakte Attribute zu. Zudem ergeben sich vielfältige Kombinationsmöglichkeiten. Gerade hierin und im Wechsel von kurzen und langen Sätzen schlagen sich eher individuelle stilistische Absichten nieder, während die reine Satzlänge (= Anzahl der Wörter pro Satz) stärker Textsortenregeln gehorcht und zeitlichen Veränderungen unterliegt. Mehrere Untersuchungen haben nachgewiesen, daß im Deutschen die Sätze seit dem 17. Jahrhundert laufend kürzer werden und zwar in allen (untersuchbaren) Funktionalstilen: Für die Literatur ist ermittelt worden: im 17. Jh. 36,3 Wörter; im 18. Jh. 26,2, im 19. Jh. 1. Hälfte: 30,3 Wörter, 2. Hälfte 23,4 , im 20. Jh. 19,3 Wörter. Ähnliche Zahlen gelten für die Wissenschaftssprache, die didaktische Literatur und die Publizistik (Sommerfeldt 1988, 216; Braun 1987, 37 ff). Kennzeichnend für den kurzen Satz ist das rasche Aufeinandertreffen von Thema und Rhema, eine Folge von kurzen Sätzen hebt daher die Rhemata hervor. Dieses stilistische Potential wird besonders in der Dichtung, aber auch in der Werbungssprache genutzt: Es ging um Schreibweise, Klangfarbe, Verteutschung, um Neuwörter. B ald hatte man sich theologisch verstrickt. Denn fromm waren sie alle. Jede protestantische B esserwisserei wurde verfochten. Jedermann glaubte sich näher bei Gott (Günter Grass, Das Treffen in Telgte). Millionen Sparer haben einen. Sie haben keinen. Warum? B undesschatzbriefe (Werbetext). Auffällige Stileffekte ergeben sich bei

1539

der künstlichen Auflösung des Satzes, auch dies begegnet in der Dichtung und in der Werbungssprache: Ivy summte. Wie zum Hohn (Max Frisch, Homo Faber). Der Canon CLC 500. Seiner Konkurrenz um Jahre voraus. Durch Laser-Technologie. Und digitale Auflösung von Abbildungen. In mehr als 250 Halbtonstufen (Werbetext, Spiegel 39/1988). Da in diesen künstlich verkürzten Sätzen die Themata unterdrückt sind, ergibt sich eine große Nachdrücklichkeit in der Aussage. Derartige Formen und extrem kurze Sätze sind stilistisch markiert. Unmarkiert ist der „Satz mittlerer Länge“ (Sowinski 1988, 80). Die langen Sätze können entweder durch Erweiterung der Satzglieder gebildet werden oder durch hypotaktischen Bau oder aber durch Kombination von beiden. Erweiterung der Satzglieder betrifft vor allem die Attribute. Generell sind attributive Regeln rekursiv. So finden sich durchaus unmarkierte Verwendungen größerer Attributkomplexe mit gleichartigem Bau: der archäologische Nachweis bedeutender B ausubstanz frühchristlicher Kirchen (Ausstellungskatalog ‘Die Bajuwaren’). Stilistisch markiert sind Fälle, die mit der Homogenität des Baues sachliche Rekursivität abbilden: dies ist der Schlüssel des Schlosses der Tür des Hauses des hölzernen Mannes. Normalerweise sind derartige Häufungen unter stilistischem Aspekt verpönt. — Bei den Satzgefügen wird maßvoller Gebrauch der ebenfalls grammatisch möglichen Häufungen als Stilideal angesehen. Die Stilkritik richtet sich im Deutschen einerseits gegen übermäßige Länge, andererseits gegen die Verwendung der Satzklammer bei Häufungen. Ein Satz wie der folgende gilt als stilistisch mißlungen: Nachdem der Kaiser sich dann noch die Mitglieder der B otschaft, unter welchen sich auch die Attachés, welche aus ... eingetroffen waren, befanden, hatte vorstellen lassen, hatte auch der B otschafter die Ehre ... (Tägliche Rundschau, zitiert bei Engel 1914 , 295). 4.4.  Die Wortstellung läßt sich für die stilistische Gestaltung in besonderem Maße nutzen. Allerdings müssen die dafür verfügbaren Möglichkeiten zunächst von den grammatischen Vorgaben abgezogen werden: Diese sind vor allem die festen Stellungen des finiten Verbs. Wenn davon abgewichen wird, ergeben sich auffällige Stileffekte, die so gut wie ausschließlich nur in literarischen Textsorten, insbesondere in der Lyrik genutzt werden: Klingt im Wind ein Wiegenlied, Sonne warm

1540

herniedersieht ... (Theodor Storm). Die Topologie des Satzes regelt sich sodann nach der sogenannten Grundreihenfolge, die teils universal, teils sprachspezifisch ist. Die Grundreihenfolge wird aber nur in textuell unmarkierten Sätzen eingehalten, vor allem in Texteröffnungssätzen. Ab dem zweiten Satz eines Textes wird der Grundreihenfolge die auf die Vorgängersätze Bezug nehmende Thema-Rhema-Gliederung aufgeprägt. Auch dafür gibt es — statistisch zu erwartende — Regularitäten. Im Deutschen tendiert das Subjekt an die Satzspitze. Zählungen haben ergeben, daß es im Texteröffnungssatz im 19. Jahrhundert in 58,7% und im 20. Jahrhundert in 57,3% der Fälle im Vorfeld steht. Vom 2. Satz des Textes an steht es im 19. Jahrhundert in 4 9,8% und im 20. Jahrhundert in 50,5% der Fälle im Vorfeld (Sommerfeldt 1988, 24 0). Die Besetzung des Vorfeldes, von Drach (1963, 17) als „Ausdrucksstelle“ des Satzes benannt, ist für stilistische Absichten merklich herausgehoben (Beneš 1971). Da das Subjekt der neutrale Thematisierungsfall ist, aber textuell gesehen durchaus auch Objekte oder Adverbialangaben unmarkierte thematische Glieder sein können, ist die Besetzung des Vorfeldes erst dann als stilistisch markiert anzusehen, wenn rhematische Glieder, Verbteile oder unerwartete kontrastierte Glieder im Vorfeld stehen: Bewundernswert sind diese sozialen Einrichtungen (Robert Musil, Der Mann ohne Eigenschaften). In diesem Satz steht das Rhema an der Spitze, während das nach hinten verschobene Thema durch den Verweisausdruck diese sogar deutlich gekennzeichnet ist. Abweichungen von der üblichen Besetzung des Mittelfeldes sind auffälliger stilistisch markiert, aber seltener, z. B.: Nur noch vereinzelt bewohnt sei die Stadt (Günter Grass, Das Treffen in Telgte) (unmarkiert: ... sei die Stadt bewohnt). — Auf die bemerkenswerteste Eigentümlichkeit des Deutschen, die Klammerkonstruktion, ist oben schon hingewiesen worden. Die Variation von Rahmenkonstruktion und Ausklammerung ist im Bereich der Wortstellung wohl das wichtigste stilistische Mittel. Allerdings sind in vielen Textsorten auch hier zunächst allgemeine kommunikative Gründe zu bedenken, etwa Verständlichkeit oder Eindeutigkeit der Beziehungen, ehe von stilistischer Nutzung gesprochen werden kann. So begegnen Beispiele des folgenden Typs überwiegend nur in der Dichtung: In Schütz’ Gesicht hatte sich, während man Albert lobreden, des Meisters schwierigen Tonsatz beklagen und

XXII. Syntax und Nachbardisziplinen

zum Schluß dessen Wünsche hersagen hörte, nichts ereignet (Günter Grass, Das Treffen in Telgte). Manchmal sind Klammerkonstruktionen nötig, um uneindeutige Bezüge zu vermeiden: Es wurden nur wenige Vorschläge, die jeden überzeugen konnten, für notwendige Verbesserungen gemacht. Der Satz Es wurden nur wenige Vorschläge für notwendige Verbesserungen, die jeden überzeugen konnten, gemacht ist nicht eindeutig und wird daher vermieden. 4.5.  Die Diathesen werden in den Stilistiken vielfach als die syntaktischen Wahlmöglichkeiten schlechthin beschrieben. Grundsätzlich kann im Deutschen ein Geschehen im morphologischen Aktiv, im Vorgangspassiv oder im Zustandspassiv ausgedrückt werden. Die Zahl der Verben, bei denen diese Möglichkeiten nicht bestehen, ist gering (Brinker 1971, 64 ). Auch beim Diathesengebrauch ist die angenommene Wahlfreiheit in starkem Maße grammatisch und textuell vorgegeben: Hier sind es vor allem Thema-Rhema-Bedingungen, die die Wahl der Diathesen steuern: Vielfach wird eine Passivform gewählt, um einen neutralen Anknüpfungsfall mit thematischem Subjekt im Vorfeld zu erhalten: Gestern habe ich im B us meinen Schirm liegen lassen. Er wurde später auf der Fundstelle abgegeben. — Im Einzelfall ist bei der Fülle der alternativen Formulierungen, die sich wählen lassen, kaum zu entscheiden, ob eine bestimmte Diathese, vor allem eine morphologisch stärker markierte, aus stilistischen Gründen, d. h., um eine Ausdrucksabsicht zu verstärken, gewählt wurde. Es lassen sich aber sowohl funktionalstilistische und textsortenspezifische Gesetzmäßigkeiten für das zahlenmäßig höhere oder niedrigere Vorkommen einer Diathese, vor allem des Passivs, angeben, als auch über den individuellen Gebrauch, wenn er rekurrent ist, Aussagen treffen. Zunächst sind die Ausdrucksvarianten an sich zu benennen. Ihr genauer grammatiktheoretischer Status ist unter Stilgesichtspunkten weniger von Belang. Im Deutschen können anstelle des Akkusativpassivs mit den beiden Typen des werden- und sein-Passivs vielfältige andere Formtypen eintreten. In erster Linie ist es das Dativpassiv, das bei höherwertigen Verben, die den Dativ erfordern oder als fakultatives Glied zulassen, bildbar ist. Mit seinen drei Auxiliarverben gehorcht es unterschiedlichen funktionalstilistischen Bedingungen: Unmarkiert sind die Verwen-

85.  Syntax und Stilistik

dungen mit bekommen, alltagssprachlich ist kriegen, während erhalten stilistisch gehoben ist: er kriegt die Meinung gesagt; wir bekommen den Antrag genehmigt; er erhält die Zusage ausgesprochen. Wie das Akkusativpassiv erlaubt das Dativpassiv eine Zustandsvariante, die allerdings selten zu belegen ist: Die Neuauflage hat alle Nachträge eingeordnet. Weitere Diathesemöglichkeiten sind u. a. das regional beschränkte, also ebenfalls stilistisch markierte gehören-Passiv: Eine solche Maßnahme gehört verboten, die sein zu-Konstruktionen (Der Auftrag ist zu erledigen), sowie die vielfältigen Funktionsverbfügungen. Die häufigsten Typen sind dabei solche, bei denen ein aktivisches oder passivisches Syntagma als Vergleichsform besteht: Die Angelegenheit gerät in Vergessenheit; das Gesetz kommt zur Entscheidung. Für die Wahl solcher Fügungen anstelle von einfachen aktivischen oder passivischen Formen (man vergißt die Angelegenheit; über das Gesetz wird entschieden) sind überwiegend aktionsartliche Gründe maßgeblich. Doch da aktionale und aspektuelle Kategorien im Deutschen morphologisch nur in Ansätzen ausgebildet sind, lassen sich die mit den Funktionsverbfügungen vorliegenden Ausdrucksalternativen durchaus zu stilistischen Zwecken nutzen. Deutlich wird dies daran, daß bestimmte Funktionalstile, vor allem die Literatursprache, aber auch die Alltagssprache Funktionsverbfügungen kaum verwenden (vgl. Sowinski 1988, 228 f). Derartige stilistische Markierungen sind als Grund für das generelle Stilverdikt in der älteren grammatischen und stilistischen Literatur anzunehmen (vgl. dazu Eisenberg 1989, 307 ff, Reiners 1976, 14 0). Auch der Gebrauch des Passivs ist stark textsortengebunden. So weist die Gegenwartsliteratur 1,5%, die Publizistik dagegen 9% werden-Passivsätze auf (Brinker 1971, 68). Durch diese funktionalstilistischen Vorgaben ist der Gebrauch der Diathesen und ihrer weiteren Ausdrucksalternativen für die individuelle stilistische Wahl sehr bedeutsam. So wird der anonyme Zugriff, der sich die Hauptperson in Kafkas Roman ‘Der Prozeß’ ausgesetzt sieht, bereits im Texteröffnungssatz durch einen Passivsatz mit ‘Täterverschweigung’ deutlich: Jemand mußte Josef K. verleumdet haben, denn ohne daß er etwas B öses getan hätte, wurde er eines Morgens verhaftet. Oder eine Passivkette wie die folgende, bei der ein Parallelismus in der Wortstellung die ungewöhnlich besetzte Ausdrucksstelle noch hervorhebt, gibt ein starkes Ironiesignal ab:

1541

Hochgelobt ist die B uche, schätzenswert ist die Fichte, vielbesungen ist die B irke, aber unantastbar ist allein die Eiche, wenn nicht gerade ihr Holz zum Schreinern eines deutschen Tisches oder zum Auslegen eines deutschen Parkettfußbodens gebraucht wird (Ludwig Harig, Heilige Kühe der Deutschen). 4.6.  Weitere „konkurrierende syntaktische Konstruktionen“ (vgl. Starke 1969/70) in der deutschen Gegenwartssprache begegnen an zahlreichen Stellen des Systems und lassen sich auf ganz unterschiedliche Weise stilistisch nutzen: Die Reichhaltigkeit des deutschen Kasussystems führt zu mannigfaltigen Konkurrenzen. Sie reichen von der Zulässigkeit verschiedener Kasus bei einzelnen Verben z. B. rufen mit Dat oder Akk oder entbehren mit Gen oder Akk bis zu subjektenthaltender gegenüber subjektloser Ausdrucksweise (ich friere gegenüber mich friert). In die Unterschiedlichkeiten werden bisweilen grammatisch schwer zu beweisende Differenzierungen hineininterpretiert, indem etwa dativische und akkusativische Bindungen als grundsätzlich unterschiedliche Möglichkeiten ausgegeben werden (vgl. dazu kritisch Starke 1969, 4 7). Hier schlagen sich entweder regionale Differenzierungen oder diachron zu deutende Gegebenheiten nieder, die stilistisch nutzbar sind zur Zeichnung von Lokalkolorit oder wie im Falle des Genitivs zur Markierung gehobenen Stils. — Die Konkurrenz der be-Verben gegenüber einfachen Verben mit Präpositionalfügungen ist den Diathesen zu vergleichen: Der Gärtner bepflanzt das B eet mit B lumen gegenüber Der Gärtner pflanzt B lumen auf das B eet eröffnet vor allem andere Nutzungen der Thema-Rhema-Gliederung (vgl. Eroms 1980). Die stilkritischen Aussagen zu den be-Verben gründen sich darauf, daß sie in der öffentlichen Sprache und in der Publizistik häufiger als in anderen Textsorten verwendet werden. Konkurrenzen zwischen Dat bzw. Akk und Präpositionalphrase (jemandem schreiben — an jemanden schreiben; ein B uch schreiben — an einem B uch schreiben) sind vornehmlich grammatisch bedingt (vgl. Matzel 1976). An integrativen Behandlungen syntaktischer Phänomene liegen mit Admoni (1987) und Kreye (1989) zwei Untersuchungen mit unterschiedlicher Zielsetzung vor. Admoni bewertet größere Stichproben ausgewählter deutscher Prosa des 19. Jahrhunderts nach der Verwendung der wichtigsten der im Vorstehenden genannten syntaktischen Parame-

XXII. Syntax und Nachbardisziplinen

1542

ter. Dabei gelangt er zu knappen Charakterisierungen der untersuchten Autoren und kann Entwicklungstendenzen des deutschen Satzbaus belegen. Kreye stützt sich bei seinen Analysen auf vollständige Passagen von Erzähltexten, die er bis zu den kleinsten syntaktischen Einheiten untergliedert. Eine wichtige Rolle spielen dabei die Kategorisierungen nach ‘Rede’ und ‘Redeaufnahme’. Die genaue frequenzanalytische Bewertung syntaktischer Phänomene führt zu ausdruckssyntaktisch konstituierten Textprofilen, die sich auf den Textgehalt beziehen lassen. Dabei werden textuelle Auffälligkeiten ebenso wie statistisch zu erfassende Phänomene beachtet. Ziel ist die exakte Erfassung des „individuell Charakteristischen eines Textes“ (Kreye 1989, 1).

5.

Stilfiguren

Seit alters werden gewisse, immer wieder verwendete, auffällige Stileffekte erzielende Mittel klassifiziert und in rhetorischen Stillehren behandelt. In den sogenannten Stilfiguren (colores rhetorici) schlägt sich das stilistische Paradoxon am sinnfälligsten nieder, als mit ihnen „zugelassene Verstöße“, „akzeptierte Abweichungen“ vorliegen. Die Stilfiguren lassen sich in Figuren des Ersatzes, der Hinzufügung, der Auslassung und der Anordnungsveränderung einteilen. Sie entsprechen damit den vier gängigsten Transformationstypen der ersten Phase der generativen Transformationsgrammatik (vgl. Jacobs/Rosenbaum 1971). Während die Ersatzfiguren hauptsächlich lexikalische Mittel darstellen, sind die anderen zum großen Teil syntaktische. Die jeweiligen Verstöße werden als ausdruckssteigernd aufgefaßt. Die wichtigsten syntaktischen sind die folgenden: Das Zeugma, bei dem zwei nicht zueinander passende Glieder an ein Verb angeschlossen werden: Ich dachte ... der lieben Westfalen, womit ich so oft in Göttingen getrunken, bis wir gerührt einander ans Herz und unter die Tische gesunken (Heinrich Heine). Das Zeugma wird so gut wie ausschließlich nur in ironisierenden Kontexten verwendet. Beim Chiasmus wird eine spiegelbildliche Anordnung der Satzglieder gewählt. Sie kommt in gehobenen Stilschichten vor: Das Spiel ist Leben, das Leben Spiel (Ernst Jünger, Eine gefährliche Begegnung). Der Anakoluth, die Unterbrechung der Satzkonstruktion, ist dagegen eine der häufigsten Stilfiguren. Sie begegnet

in der Alltagskommunikation, aber auch in der schönen Literatur: Das muß ich Hertha nachher noch aufmutzen; die sich moquiert, wenn ich etwas mit Worten präludiere: und sie spricht von so-was! (Arno Schmidt, Kaff auch mare crisium). Die Ellipse ist ebenfalls in vielen Funktionalstilen vertreten. Sie wird in der Dichtung gern verwendet, um das Erzählte zu verlebendigen: „Hyazinth, geh jetzt zu —“ „Ich weiß schon —“ „Du weißt nicht, sage ich dir, Hyazinth —“ (Heinrich Heine, Reisebilder). Eine Sonderform ist die Aposiopese, das plötzliche Verstummen, mit dem starke Betroffenheit zum Ausdruck gebracht werden soll: „... ist es mir doch lieb, wenn ich mal wieder gutes Deutsch sprechen kann mit einem Gesichte, das ich schon einmal in Hamburg gesehen, und denke ich an Hamburg —“ Hier ... wurden des Mannes Äuglein flimmernd feucht ... (Heinrich Heine, Reisebilder). Mit der Parenthese, die in allen Funktionalstilen und Textsorten häufig vorkommt, werden sukzessive Konstruktionen umgangen. „Die Aussage wirkt ungezwungen, aus dem Stegreif gesprochen, lebendig“ (Schneider 1969, 4 94 ). Im folgenden Beispiel wird sie mit einem Vergleich, der fast auf ein Zeugma verkürzt ist, kombiniert: So belehrt, wolle er in die Kunst, wie grad durchs Fenster, den Einstieg finden und — falls die Musen geneigt — zum Dichter werden (Günter Grass, Das Treffen in Telgte). — Nachträge der folgenden Art, mit denen sich Verknappungen erzielen lassen, werden ebenfalls als Stilfiguren aufgefaßt: Dann setzte er sich: erleichtert (Günter Grass, Das Treffen in Telgte). Werden erzählte Geschehnisse in umgekehrter Reihenfolge wiedergegeben, dem Hysteron proteron, wird eine paradoxe Situation simuliert: Ihr Mann ist tot und läßt Sie grüßen (Goethe, Faust). Prolepsen ziehen das Referenzobjekt nach vorn: Erinnerst du dich an den Strand von Alassio, wie ruhig es da noch war? Parallelismen werden zu unterschiedlichen Zwecken eingesetzt, u. a. zur Kennzeichnung von Gleichförmigkeit, zuweilen in ironischer Absicht: Mit seinen Reitern und Musketieren habe er zeitgemäß gehandelt, wie die hier versammelten Herren gezwungen seien, zeitgemäß zu handeln, indem sie mit ihren Huldigungspoemen Fürsten zu loben hätten, denen die Mordbrennerei geläufig wie das tägliche Ave sei, deren größerer Raub als sein Mundraub mit Pfaffensegen bedacht werde, denen Untreue praktisch wie ein Hemdwechsel sei und deren Reue kein Vaterunserlang halte (Günter Grass, Das Treffen in Telgte). Parallelismen enthalten häufig eine

85.  Syntax und Stilistik

Klimax: Diese Lippen brauchten nur zu pfeifen, — et la Prusse n’existait plus — diese Lippen brauchten nur zu pfeifen — und die ganze Klerisei hatte ausgeklingelt — diese Lippen brauchten nur zu pfeifen — und das ganze heilige römische Reich tanzte (H. Heine, Reisebilder). — Die Stilfiguren zeigen gut die generellen Wirkungsmechanismen des stilistischen Sinnes bei der Syntax. Erst im Rahmen größerer Komplexe erschließt sich im Normalfall das stilistische Potential des Einzelsatzes. Es verstärkt die Aussageabsicht und wird unter Bezug auf sehr unterschiedliche Ebenen gedeutet.

6.

Literatur

Admoni, Wladimir. 1979. Die Verwendung der grammatischen Formen in den Fachsprachen. Fachsprachen und Gemeinsprachen, Jahrbuch 1978 des Instituts für deutsche Sprache, hrsg. von Wolfgang Mentrup, 218—28. Düsseldorf. —. 1987. Die Entwicklung des Satzbaus der deutschen Literatursprache im 19. und 20. Jahrhundert. Berlin. B ailey, Richard W., und Dolores M. B urton. 1968. English Stylistics: A Bibliography. Cambridge, England. B aker, William E. 1967. Syntax in English Poetry 1870—1930. Berkeley, Los Angeles. B ally, Charles. 1970 (1963). Traité de stylistique française. 2 Bde. Genf. B eier, Rudolf. 1979. Zur Syntax in Fachtexten. In: Fachsprachen und Gemeinsprachen. Jahrbuch 1978 des Instituts für deutsche Sprache, hrsg. von Wolfgang Mentrup, 276—301. Düsseldorf. B eneš, Eduard. 1971. Die Besetzung der ersten Position im deutschen Aussagesatz. Fragen der strukturellen Syntax und der kontrastiven Grammatik, 160—82. (Sprache der Gegenwart 17). Düsseldorf. B etten, Anne. 1976. Ellipsen, Anakoluthe und Parenthesen. Fälle für Grammatik, Stilistik, Sprechakttheorie oder Konversationsanalyse? Deutsche Sprache 4. 207—30. —. 1987. Grundzüge der Prosasyntax. Stilprägende Entwicklungen vom Althochdeutschen zum Neuhochdeutschen. (Reihe Germanistische Linguistik 82). Tübingen. B ichel, Ulf. 1973. Problem und Begriff der Umgangssprache in der germanistischen Forschung (Hermaea Germanistische Forschungen Neue Folge 32). Tübingen. B raun, Peter. 1987. Tendenzen in der deutschen Gegenwartssprache. Sprachvarietäten. 2., veränderte und erweiterte Aufl. Stuttgart, Berlin, Köln, Mainz.

1543

B rinker, Klaus. 1971. Das Passiv im heutigen Deutsch. Form und Funktion (Heutiges Deutsch I/ 2). München, Düsseldorf. Chatman, Seymour, und Samuel R. Levin. 1967. Essays on the language of literature. Boston. Drach, Erich. 1963 (1937). Grundgedanken der deutschen Satzlehre. Darmstadt. Eisenberg, Peter. 1989. Grundriß der deutschen Grammatik. 2. Aufl. Stuttgart. Engel, Eduard. 1914 . Deutsche Stilkunst. 22. Aufl. Wien, Leipzig. Enkvist, Nils Erik. 1973. Linguistic Stylistics. The Hague, Paris. Erben, Johannes. 1980. Deutsche Grammatik. Ein Abriß. 12. Aufl. München. Ermert, Karl. 1979. Briefsorten. Untersuchungen zu Theorie und Empirie der Textklassifikation (Reihe germanistische Linguistik 20). Tübingen. Eroms, Hans-Werner. 1980. Be-Verb und Präpositionalphrase (Monographien zur Sprachwissenschaft 9). Heidelberg. —. 1983. Stilistik. Kritische Stichwörter zur Sprachdidaktik, hrsg. v. Margareta Gorschenek & Annamaria Rucktäschel, 235—246. München. —. 1986. Textlinguistik und Stiltheorie. Akten des VII. Internationalen Germanisten-Kongresses Göttingen 1985, Bd. 3, 10—21. Tübingen. Fleischer, Wolfgang, und Georg Michel (Hg.) 1975. Stilistik der deutschen Gegenwartssprache. Leipzig. Francis, W. Nelson. 1967. Syntax and Literary Interpretation. In Seymour Chatman & Samuel R. Levin, 209—216. Glück, Helmut, und Wolfgang Werner Sauer. 1990. Gegenwartsdeutsch (Sammlung Metzler. Bd. 252). Stuttgart. Gobyn, Luc. 1984 . Das Märchen und seine Spielformen. Pragmatics and Stylistics, hrsg. v. Willi van Peer & Jan Renkema, 219—54 . Leuven, Amersfoort. —. 1984a. Textsorten. Brüssel. Guiraud, Pierre, und Pierre Kuentz. 1975 (1970). La stylistique. Paris. Gumbrecht, Hans Ulrich. 1986. Schwindende Stabilität der Wirklichkeit. Eine Geschichte des Stilbegriffs. In: Gumbrecht & Pfeiffer 1986, 726—88. —, und K Ludwig Pfeiffer. (Hg.) 1986. Stil. Geschichten und Funktionen eines kulturwissenschaftlichen Diskurselements. Frankfurt am Main. Halliday, M. A. K. 1985. An Introduction to Functional Grammar. London. —. 1985a. Spoken and Written Language (Language and Learning 5). Geelong, Vic. Havránek, B ohuslav. 1976. Die funktionale Schichtung der Literatursprache. Grundlagen der Sprach-

1544

kultur. Beiträge der Prager Linguistik zur Sprachtheorie und Sprachpflege, Teil 1 (Reihe Sprache und Gesellschaft Bd. 8/1), hrsg. v. Jürgen Scharnhorst u. a., 150—61. Berlin. Heinemann, Wolfgang. 1975. Das Problem der Darstellungsarten. In Fleischer & Michel 1975, 268— 300. Henn-Memmesheimer, B eate. 1986. Nonstandardmuster. Ihre Beschreibung in der Syntax und das Problem ihrer Arealität (Reihe Germanistische Linguistik. 66) Tübingen. Heringer, Hans-Jürgen. 1989. Grammatik und Stil. Praktische Grammatik des Deutschen. Frankfurt a. M. Jacobs, Roderick A, und Peter S. Rosenbaum. 1971. Transformations, Style, and Meaning. Waltham, MA, Toronto. Kennedy, Chris. 1982. Systemic Grammar and its Use in Literary Analysis. Language and Literature. An Introductory Reader in Stylistics, hrsg. v. Ronald Carter, 83—99. London. Kreye, Horst. 1989. Satzform und Stil. Heidelberg. Langen, August. 1957. Deutsche Sprachgeschichte vom Barock bis zur Gegenwart. Deutsche Philologie im Aufriß. 2., überarbeitete Aufl. Berlin. Lausberg, Heinrich. 1973. Handbuch der literarischen Rhetorik. 2 Bde. München. Leska, Christel. 1965. Vergleichende Untersuchungen zur Syntax gesprochener und geschriebener deutscher Gegenwartssprache. In Beiträge zur Geschichte der deutschen Sprache und Literatur 87, 427—464. Matzel, Klaus. 1976. Dativ und Präpositionalphrase. Sprachwissenschaft 1. 144—86. Meyer, Richard M. 1906. Deutsche Stilistik. München. Ohmann, Richard. 1964 . Generative Grammars and the Concept of Literary Style. Word 20. 423—39. —. 1967. Literature as Sentences. In Chatman & Levin 1967. 231—38. Porsch, Antje. 1981. Die funktionalstilistische Theorie und ihr Verhältnis zur Differenziertheit der Sprache. Kommunikation und Sprachvariation (Reihe Sprache und Gesellschaft Band 17), hrsg. v. Wolfdietrich Hartung u. a., 280—307. Berlin. Püschel, Ulrich. 1982. Die Bedeutung von Textsortenstilen. Zeitschrift für germanistische Linguistik 10. 28—37. Quirk, Randolph. 1982. Style and Communication in the English Language. London. Rath, Rainer. 1979. Kommunikationspraxis. Analysen zur Textbildung und Textgliederung im gesprochenen Deutsch. Göttingen. Reiners, Ludwig. 1976 (194 3). Stilkunst. München. Neubearbeitung 1991. Riesel, Elise. 1964 . Der Stil der deutschen Alltags-

XXII. Syntax und Nachbardisziplinen

rede. Moskau. —, und E. Schendels. 1975. Deutsche Stilistik. Moskau. Ringbom, Håkan. 1975. The style of Orwell’s preface to ‘Animal Farm’. Style and Text. Studies presented to Nils Erik Enkvist. 243—49. Stockholm. Rochemont, Michael S. 1985. A Theory of Stylistic Rules in English. New York, London. Sanders, Willy. 1973. Linguistische Stiltheorie. —. 1977. Linguistische Stilistik. Göttingen. —. 1990. Gutes Deutsch — besseres Deutsch. Praktische Stillehre der deutschen Gegenwartssprache. 2. Aufl. Darmstadt. Sandig, B arbara. 1970. Probleme einer linguistischen Stilistik. Linguistik und Didaktik 1. 177—94. —. (Hrsg.) 1983. Stilistik. Bd. 1. Probleme der Stilistik. Bd. 2 Gesprächsstile. (Germanistische Linguistik 3—6). Hildesheim, Zürich, New York. —. 1986. Stilistik der deutschen Sprache. Berlin, New York. Schiepek, Josef. 1899/1908. Der Satzbau der Egerländer Mundart. Teil I und II. Prag (Beiträge zur Kenntnis deutsch-böhmischer Mundarten 1). Schneider, Wilhelm. 1969. Stilistische deutsche Grammatik. 5. Aufl. Freiburg, Basel, Wien. Simmler, Franz. 1986. Syntaktische Strukturen in Kunstmärchen der Romantik. Akten des VII. Internationalen Germanisten-Kongresses Göttingen 1985. Bd. 3, 66—96. Sommerfeldt, Karl Ernst (Hg.) 1988. Entwicklungstendenzen in der deutschen Gegenwartssprache. Leipzig. Sowinski, B ernhard. 1988. Deutsche Stilistik. 6. Aufl. Frankfurt am Main. Spillner, B ernd. 1974 . Linguistik und Literaturwissenschaft. Stilforschung, Rhetorik, Textlinguistik. Stuttgart, Berlin, Köln, Mainz. Spitzer, Leo. 1961 (1926). Stilstudien. 2 Bde. München. Sumpf, Joseph. 1971. Introduction à la stylistique du français. Paris. Starke, Günther. 1969/70. Konkurrierende syntaktische Konstruktionen in der deutschen Sprache der Gegenwart. Untersuchungen im Funktionsbereich des Objekts. Teil 1—5. Zeitschrift für Phonetik 22 (1969). 25—65/154 —95. 23 (1970). 53—84 / 232—60/573—89. Taylor, Talbot J. 1980. Linguistic Theory and Structural Stylistics. Oxford, New York. Teubert, Wolfgang. 1988. Der gelehrte Jargon — Ein deutsches Phänomen? Sprachreport 1, 9—12. Mannheim. Trabant, Jürgen. 1986. Der Totaleindruck. Stil der Texte und Charakter der Sprachen In Gumbrecht/ Pfeiffer 1986, 169—88.

86.  Syntax und Sprachdidaktik

Turner, G. W. 1979. Stylistics. 4 . Aufl. Harmondsworth, Middlesex. Ullmann, Stephen. 1972. Sprache und Stil. Tübingen. Wagner, Hildegard. 1972. Die deutsche Verwaltungssprache der Gegenwart. Düsseldorf.

1545

Werlich, Egon. 1979. Typologie der Texte. 2. Aufl. Heidelberg. Winter, Werner. 1961. Relative Häufigkeit syntaktischer Erscheinungen als Mittel zur Abgrenzung von Stilarten. Phonetica 7. 193—216.

Hans-Werner Eroms, Passau (Deutschland)

86. Syntax und Sprachdidaktik 1. 2. 3. 4. 5. 6. 7.

Geschichtlicher Aspekt Prädikat Subjekt Objekte Adverbial- und Attributskonstruktionen Kategorialgrammatik und Sprachdidaktik Literatur

Ein Unterricht, der Sprache lehrt, muß das Problem, das die Vermittlung syntaktischer Strukturen mit sich bringt, lösen. Das setzt Kenntnisse der Syntax einer Sprache voraus, macht Wissen über mögliche Methoden der Vermittlung notwendig und erfordert Erfahrungen im Umgang mit dem Personenkreis, dem diese Vermittlung gilt (Hoppenkamps 1978; Butzkamp 1985). Aus den drei genannten Komponenten sprachwissenschaftliche Daten, sprachdidaktische Modellvorschläge und sprachentwicklungsbedingte Resultate, eine umfassende Theorie der Syntaxvermittlung zu entwickeln, war bisher nicht möglich (Wode 1977, 17; Kerkhoff 1983, 74 ; Braunroth 1984 , 266—267). Die mit ihnen verbundenen Fachdisziplinen verfolgten ihre eigenen Erkenntnisziele, und diese lassen sich mit den Bedürfnissen einer umfassenden Theorie nicht in Einklang bringen. Das hat eine lange Tradition.

1.

Geschichtlicher Aspekt

1.1. Die Anfänge einer Sprachvermittlung Das Interesse an grammatischem Wissen läßt sich seit der Antike belegen. Grammatik zählte dort zu den ‘septem artes liberales’. Mit Dionysios Thrax (1. Jh. v. Chr.) beginnt eine Jahrhunderte umfassende Entwicklung, die das Wissen über Grammatik überliefert und ein Wissen, wie mit Sprache umzugehen ist, zu sammeln begann (Schmidt 1852—53).

In der Antike war die Sprachbetrachtung eng mit philologischen Problemen verbunden. Die Scholastik interessierte der Zusammenhang zwischen der vom Wort bezeichneten Wirklichkeit und der Bedeutung desselben (Arens 1969, 35—50). Seit der Zeit des Humanismus ist erstmals Interesse am Bau der Sprache nachzuweisen. Sprachwissen war nötig, um antike Texte lesen und verstehen zu können. Durch das Einsetzen intensiver Reisetätigkeiten in dieser Zeit entsteht die Notwendigkeit, fremde Sprachen lernen zu müssen (Arens 1969, 62—80). Parallel dazu entwickelt sich ein völlig neues Verhältnis zur Volkssprache. Sie wird im Aufbau und Ausdruck als gleichwertig mit Latein, Hebräisch und Griechisch bewertet. Das führt zu den ersten umfassenden Versuchen einer Beschreibung der deutschen Sprache (Ickelsamer 1531; Albertus 1573; Oelinger 1573). Die Reformation veränderte das Verhältnis zur Sprache, weil sie das Ziel verfolgte, jeder solle die Heilige Schrift in seiner Muttersprache lesen können. Ein breites Interesse am Leseund Schreibunterricht setzt ein (Frank 1976, 33—38). Das zwang zum Herausarbeiten grammatischer Strukturelemente des Deutschen, denn grammatisches Wissen war nötig, um zu einer einheitlichen Orthographie im Deutschen zu gelangen (Garbe 1978). Mit der Schrift ‘Aristarchus sive de contemptu linguae Teutonicae’ hatte Opitz (1617) eine intensive Auseinandersetzung mit dem Deutschen bewirkt. Die ‘Fruchtbringende Gesellschaft’ veranlaßte, eine Terminologie der deutschen Sprachlehre zu entwickeln (Harsdörffer 1651; Schottel 164 1; 1663; Gueintz 1663). Diese Entwicklungen stehen im Zusammenhang mit den Zielen einer Bildungsreform, wie sie in Ratkes Reformplänen von 1612 formuliert worden sind (Ratke 1612). In ihnen wird das Deutsche als Ausgangs-

1546

sprache für das Erlernen der drei klassischen Sprachen Hebräisch, Griechisch und Lateinisch gefordert. ‘Deutsche Sprachlehre’ ist als Unterrichtsgegenstand erstmals in der Weimarer Schulordnung (1619) belegt. Sie wird als ‘Propädeutikum für den anschließenden Lateinunterricht’ vorgesehen. Die Kinder erwerben Kenntnisse der Wortarten und Flexion des Deutschen (Frank 1976, 4 8—50). Die ‘Köthener Lehrordnungen’ unterteilen die deutsche Sprachlehre nach den Sachbereichen ‘Wortsprechung’, ‘Wortschreibung’, ‘Wortforschung’ und ‘Wortfügung’. Letzteres berührt erstmals den Bereich der Syntax in der Schule (Ratke 1612, 81—106). Ein erster Versuch der Sprachvermittlung, dem so etwas wie eine methodisch didaktische Idee zugrundelag (Geißler 1959, 117), war der ‘Janua linguarum’ von Comenius (1631). Für ihn sollte Schule ein Tor zur Welt der Dinge und Sprachen sein. Worte erschließen diese Welt, und Sätze die in ihr existierenden Sachverhalte. Er wählte achttausend Wörter aus dem Latein und verband sie zu tausend Sätzen, die in über hundert Sachgebiete geordnet wurden. Mit seinem ‘Orbis pictus’ (Comenius 1658) tat er dasselbe für das Deutsche. Die ‘Wissenschaft’ belächelte die ‘Narretei’. Weise (1672) kritisierte den unzureichenden Wortschatz der Verben und das Fehlen der Konjunktionen. Damit geschah zum ersten Mal das, was die Beziehung von ‘Sprachwissenschaft’ und den didaktisch motivierten Versuchen einer Sprachlehre auch in Zukunft kennzeichnen wird: Das Sprachverständnis der Didaktiker wirkt auf die Wissenschaftler naiv, ist im Ansatz verkürzt und im Detail nicht selten unzutreffend. 1.2. Der erste Unterricht in der Muttersprache Die ‘Deutsche Sprachlehre’ von Adelung (1781) brachte für die Schulgrammatik neue Aspekte. Seit Bödiker (1690) und spätestens seit Gottsched (174 8) war die deutsche Grammatik regulärer Unterrichtsgegenstand. Sie wurde wie die lateinische Grammatik als Regelwissen vermittelt (Frank 1976, 94 —96). Das wird von Adelung (1781) als ‘langweiliges Gedächtniswerk’ kritisiert. Der Schüler muß Verständnis für sprachliche Zusammenhänge gewinnen, dann werde er ein besseres Sprachgefühl erlangen. Über das grammatische Wissen müsse die Erziehung zu einem guten Stil geleistet werden. Moritz (1793) verfolgt diesen Gedanken noch weiter und for-

XXII. Syntax und Nachbardisziplinen

dert zur eigenständigen Regelfindung durch den Schüler auf. Ein anderer Denkanstoß ging von Herder (1766) mit seiner Schrift ‘Über die neuere deutsche Literatur’ aus. Grammatik wird als ‘erste Lehrerin der Logik’ bezeichnet (Herder 184 7, 14 7). Die Rezeption des ‘Hermes, or a philosophical inquiry concerning universal grammar’ von Harris (1751) durch den Gymnasiallehrer Roth (1795) im ‘Antihermes oder philosophische Untersuchung über den reinen Begriff der menschlichen Sprache und die allgemeine Sprachlehre’ führte zum ersten systematischen Versuch, die deutsche Grammatik in eine Denklehre zu überführen. Ausgangspunkt im Unterricht war der einfache Satz. Die Bestandteile eines Satzes mußten als logische Elemente erkannt werden. Auf dieser Basis waren wieder eigene Sätze zu bilden (Roth 1799). Pestalozzi (1807) hatte empfohlen, den Sprachunterricht syntaktisch anzugehen. Den Kindern wurden Satzfragmente zum Einüben vorgegeben. Die Gedanken wurden von Krause (1813) aufgegriffen und im ‘Versuch eines Lehrbuchs der deutschen Sprache’ (Krause 1817/18) methodisch aufgearbeitet, so daß zum ersten Mal zwischen der Vermittlung von Wissen über die Sprache und dem Einüben in die Sprache systematisch unterschieden wurde. Vom Satz her eine Gedankenschulung zu entwickeln, das hatte auch Becker (1827) mit seinem Buch ‘Organism der Sprache als Einleitung zur deutschen Grammatik’ versucht. Er bewirkte mit seinen Arbeiten ein grammatisches Paradigma, das die Schulgrammatik über das 19. Jh. hinaus bestimmen sollte (Becker 1829; 1831; 1833). Seine Syntax führt er auf ein einfaches Bauprinzip zurück. Die organische Gestaltung des ganzen Satzes bestehe darin, in ihm ein Allgemeines mit einem Besondern im Gegensatze von Tätigkeit und Sein zu einer organischen Einheit verbunden zu sehen. Bei der Entwicklung des Satzes wiederhole sich in jedem besonderen Verhältnis dieser Art der zu einer Einheit verbundene Gegensatz auf das Neue. Ausgehend von der Vorstellung, im Prädikat den ‘Hauptbegriff’ und im Subjekt den ‘Beziehungsbegriff’ vorzufinden, leitet er daraus das prädikative Verhältnis ab. Becker überträgt den Gedanken auf die einzelnen Satzglieder und kommt so zu einer Beschreibung des ‘attributiven’ und ‘objektiven Verhältnisses’. Aus diesen leitet er dann alle weiteren syntaktischen Strukturen ab. Er klassifiziert die verschiedenen Objekte und führt

86.  Syntax und Sprachdidaktik

das präpositionale Objekt ein. Die verschiedenen adverbialen Verhältnisse werden als adverbiale Objekte in der z. T. noch heute gebräuchlichen Form festgelegt, und die Nebensätze auf dieselbe Weise geordnet. Becker glaubte, daß die Vermittlung syntaktischen Wissens eine Schulung des logischen Denkvermögens der Kinder bedinge. Daher wollte er beginnend mit der ersten Klasse, daß die Kinder schrittweise zum Erkennen der besonderen Satzverhältnisse geführt werden sollten, um so ihre Sprachfertigkeiten verbessern und ihre Vorstellungskraft in besonderer Weise ordnen zu können. Er hätte seine Wirkung auf den Unterricht und die Schulgrammatik nicht gewonnen, wenn nicht im Streit um die Brauchbarkeit des Ansatzes ein Schweizer Lehrer namens Wurst (1836) eine Sprachlehre in Beckers Sinn geschrieben hätte, in der klar formulierte Regeln, zahlreiche Beispiele und Übungsaufgaben gegliedert in einzelne Lektionen für den Unterricht angeboten worden wären. Diese ‘Sprachdenklehre’ wurde zum meist verwendeten Buch für den Sprachunterricht. Trotz dieser Wirkung fand eine sprachwissenschaftliche Auseinandersetzung nicht statt (Raumer 1857, 217), so daß mit Ausnahme von Steinthal (1850) Becker entweder überhaupt nicht wahrgenommen wurde oder nur pauschale Ablehnung fand. Eine ernsthafte sprachwissenschaftliche Auseinandersetzung leistete erst Glinz (194 7). Schulgrammatik und wissenschaftliche Grammatik gingen bis zu diesem Zeitpunkt getrennte Wege. 1.3. Unterschiedliche Entwicklungen am Gymnasium und an der Volksschule Verschiedene Wege hatten sich zwischen dem Gymnasium und der Volksschule entwickelt. Deutsch war seit 184 0 im Gymnasium als Lehrfach anerkannt, besaß aber eine untergeordnete Funktion. Im Schnitt wurden nicht mehr als zwei Stunden Deutsch gegeben. Mit Hiecke (184 2) ‘Der deutsche Unterricht auf deutschen Gymnasien’ wurde der syntaktischen Betrachtung ein fester Ort im Unterricht zugewiesen. Die Grundlage des Unterrichts ist das Lesestück; um einen Text deuten zu können, ist eine syntaktische Analyse nötig. Das geschieht durch Zergliedern der Konstruktion in einzelne Satzglieder und eine Paraphrasierung derselben in einfache Konstruktionen. In diesen Übungen wird eine Förderung des Denkvermögens und eine

1547

Schulung der sprachlichen Fertigkeiten gesehen. Sprache wird geschichtlich gesehen. Im vergleichenden Grammatikunterricht sieht Hiecke ein Mittel, Schüler für Sprachveränderungen sensibel zu machen. Er tritt deshalb für einen komparativen Unterricht in lateinischer und deutscher Grammatik ein, um so ein Gefühl für die Verschiedenheit des Satzbaus der jeweiligen Sprachen zu vermitteln. Einer sprachwissenschaftlich fundierbaren Sehweise waren allerdings Grenzen gesetzt. Im Gymnasium wurde zwischen der ‘Sprachlehre’ und der ‘Sprachkunde’ unterschieden. Erstere diente lediglich dazu, praktische Fertigkeiten zu erwerben. Letztere war ausschließlich sprachhistorisch begründet (Frank 1976, 4 51—4 70). Mit Junggrammatikern wie Brugmann (1885) oder Paul (1880) trat ein Wandel in der Sprachwissenschaft ein, und dieser blieb nicht ohne Wirkung auf das Satzverständnis in der Schulgrammatik. Die Relativierung der Bedeutung des Historischen für die Sprachbeschreibung, die Bezugnahme auf die gesprochene Gegenwartssprache und das Interesse am einzelnen Sprecher schufen Berührungspunkte mit den Inhalten der Schule (Arens 1969, 337—362). Mensing (1903) gab das erste Sprachbuch für das Gymnasium heraus, das zum Standardwerk bis weit in die 30er Jahre hinein wurde. Es sollte sprachgeschichtliches Verständnis wecken, zur Sprachrichtigkeit erziehen, mit der Wortbildung bekannt machen und in die Satzlehre einführen. Durch Erkennen der Wortklassen sollte die Binnenstruktur des Satzes durchschaubar gemacht werden. Der Satz wird nach dem Schema von Satzgegenstand und -aussage interpretiert, indem nach Subjekt und Prädikat als den notwendigen Bestandteilen eines Satzes gefragt wird, anschließend ist nach den Erweiterungen zu suchen. Die Analyse wird durch eine graphische Darstellung veranschaulicht. Das Augenmerk gilt den formalen Besonderheiten. An ausgewählten Musterbeispielen wird die grammatische Regel, die einer Konstruktion zugrundegelegt wird, erarbeitet und eingeübt. Sprache reduziert sich bei einer solchen Vorgehensweise auf eine Sammlung von Mustersätzen zur Darstellung grammatischer Regeln und Formen. Demgegenüber entwikkelt die Volksschule einen Ansatz, der sich auf die Reformpädagogik bezieht und die Hinwendung zum Kind betont sowie am induktiven Vorgehen orientiert ist. Das eigene

1548

Sprachverhalten und -erleben wird in den Mittelpunkt der Betrachtungen gestellt. Für die Volksschule ist Sprache die gesprochene Sprache des kindlichen Alltags. Der Sprachunterricht der Volksschule betont die Abkehr von der formalisierten Satzanalyse. Das besondere pädagogische Interesse gilt der synchronen Betrachtungsweise von Sprache. Der Sprachunterricht müsse „vor das System Sprache“ treten, fordert Hartnacke (1918) und entwickelt die Idee, Satzglieder seien nicht Einheiten, die durch Zerlegen erkannt werden können, sondern sie seien Elemente, die aus ihrer Erweiterbarkeit heraus betrachtet werden müßten. Er konnte so den Unterschied zwischen Valenzabhängigkeit und Attribution, Satzteile erster und Satzteile zweiter Ordnung verdeutlichen. Dependentielle und hierarchische Strukturen werden als Organisationsprinzipien der Sprache von Müller (1921) erkannt, sie versucht, dem Kind interdependente Relationen bei der Satzkonstruktion zu erklären. Statt Sprachgesetze will sie Beziehungsverhältnisse im Satz beschreiben und besprechen. Neumann (1929) betont in diesem Sinn die besondere Rolle des Verbs, welches den Satz als Ganzheit zu beschreiben erlaubt. Tests, wie den der Ersetzbarkeit von Termen, werden bei FlorstedtStieber (1935) vorgeschlagen, als dieser Methoden entwickelt, den Kindern das Erkennen von Subjekt und Prädikat zu ermöglichen. Zur selben Zeit versuchte Lyon (1890), sprachwissenschaftliches Wissen in den Unterricht einzubringen. Die Laut- und Wortbildungslehre werden zum Ausgangspunkt der Betrachtung syntaktischer Phänomene. Kern (1883) arbeitet Beziehungen zwischen der Wortartenlehre und der Satzanalyse heraus. Eine völlig neue Perspektive führt Sütterlin (1900) ein, als er den Satz nicht von der Form her definiert, sondern mit der Grundabsicht eines Sprechers oder des Redeschritts eines Autors in Verbindung setzt. Grundsätzlich blieb Grammatik aber etwas Formales (Braun/Krallmann 1983). Das Wissen über die Grammatik der Muttersprache war oberflächlich und nach nur kurzer Zeit wieder vergessen (Schwarz 1937; Augst 1976). Der Sinn von Grammatikunterricht wurde in zunehmender Weise in Frage gestellt. Gaiser (1950) fragte offen: „Wieviel Grammatik braucht der Mensch?“ Er gab zur Antwort, daß die Art und Weise, wie Satzanalyse betrieben worden sei, wenig zum Verständnis einer Sprache beigetragen hätte

XXII. Syntax und Nachbardisziplinen

(Weisgerber 1950; Pfleiderer 1954 ), und auch das Argument, der Fremdsprachunterricht bedürfe dieser Grundlagen, lasse sich nicht plausibel machen (Essen 1962, 87). 1.4. Veränderungen in der Sprachdidaktik durch die Linguistik Das Scheitern einer für den Unterricht erkenntnisleitenden Satzbeschreibung hatte verschiedene Gründe. Es gab eine Vielzahl von einzelnen Ansätzen, Syntax synchron zu beschreiben, Methoden für die Analyse zu entwickeln und sich von den hergebrachten Satzglieddefinitionen zu lösen, doch keiner wurde konsequent genutzt. Drach (1937) hatte zwar ein topologisches Konzept für die Syntax entwickelt, es fand indes keinen Eingang in den Sprachunterricht. Der Satz war als zentrales Thema des Sprachunterrichts seit Becker unbestritten. Die Entwicklung einer Syntax aber scheiterte am Fehlen eines theoretischen Konzepts, aus dem der Satz als Ganzes zur Erklärung seiner Funktion hätte abgeleitet werden können. Satz blieb somit linguistisch ein weitgehend morphologisch definiertes Konstrukt mit einer Reihe inhaltlicher Implikationen, die sich aus der Lexikalisierung einzelner Wortarten herleiteten (Weisgerber 1950; 1951; Menzel 1975, 31—35), auf Intonationsmuster und Klangbögen (Essen 1962; Graucob 1970) intuitiv bezogen oder mit der allgemeinen Sprachentwicklung in Zusammenhang gebracht wurden (Ingendahl 1975). Das änderte sich erst, als Glinz (1952) systematisch strukturalistische Verfahren zur Segmentierung von Äußerungsstrukturen für die Satzanalyse adaptierte. Die Satzbetrachtung wird so zu einem Experimentierfeld, das sich nicht nur dem Wissenschaftler eröffnet, sondern von einem Schüler auf ganz ähnliche Weise genutzt werden kann (Menzel 1975, 136—138). Ebenso bedeutsam war für die Sprachdidaktik die Auswertung der Dependenzrelationen im Rahmen deutscher Grammatiken (Erben 1960; Brinkmann 1962; Fourquet 1970; Grebe 1973; Engelen 1975), die Systematisierung durch Heringer (1970; 1972; 1978) und die für den Ausländerunterricht konzipierten Arbeiten (Helbig/Schenkel 1969; Helbig 1971; Helbig/Buscha 1972). Zwar ist die Idee der Wertigkeit eines Verbs nicht neu (Meiner 1781), sie zur Basis der Grammatik zu machen, war allerdings erst von Tesniere (1959) versucht worden (Fink 1977; Zöfgen 1982). Die Erörterung syntaktischer Themen erhält in den von Glinz (1952; 1970—71) vorge-

86.  Syntax und Sprachdidaktik

schlagenen Operationen ein methodisch zuverlässiges und wissenschaftlich begründetes Fundament. Sie gewinnt durch die Valenz des Verbs eine Perspektive, die ein Wesensmerkmal des Satzaufbaus erfaßt (Seyfert 1976; Engel 1977; Tarvainen 1979; Eroms 1981). Es handelt sich um eine linguistische Sichtweise. Methoden zur Erkennung der sprachlichen Segmente und Namen zur Kennzeichnung der Abhängigkeiten erlauben auch Schülern, theoriegeleitet Sprache zu beschreiben (Aßheuer/Hartig 1976). Eine Öffnung für linguistische Seh- und Vorgehensweisen wurde durch die Meinung unterstützt, das Ziel der Sprachdidaktik müsse die Vermittlung der Einsicht in den Bau der Sprache sein (Edelstein/Schäfer 1969, 4 9; Strehle 1962). Die Entdeckung der generativen Grammatik weckte in der Sprachdidaktik große Erwartungen (Hartmann 1969; Frey 1966; Eichler 1972; Bünting/Kochan 1973; Henrici 1979). Eine kindgemäße bzw. didaktisch aufbereitete Hinführung zu dem Modell würde mit einem Verstehen des Sprachsystems an sich Hand in Hand gehen (Roberts 1964 ; Thomas 1965; Hartmann 1969; Hundsnurscher 1970; Roth 1971; Menzel 1975; Ader 1976; Ader/ Kress 1978). Es entstand die Vorstellung, eine Basis für eine pädagogische Grammatik gefunden zu haben (Bünting 1970, 73; Menzel 1975, 170). Zur Krise kam es, als sich zeigte, daß der Anspruch dieser Theorie den sprachlichen Gegebenheiten bisher nicht genügen konnte (Weisgerber 1974 ; Tegge 1975, 399; Wode 1977, 21; Oomen-Welke 1982, 22). Damit war die sprachdidaktische Relevanz in Frage gestellt (Hamburger Autorenkollektiv 1975). Eine zunehmende Kritik an der Rezeption von Wissenschaft allgemein und der Linguistik insbesondere sowie Zweifel am universalgrammatischen Ansatz des Modells der generativen Grammatik bewirkten eine generelle Distanzierung von der Syntaxtheorie (Hamburger Autorenkollektiv 1975, 5; Strassner 1977; Müller-Michaels 1980; Reuschling 1981, 166—168). Die Sprachdidaktik wandte sich von der sprachtheoretischen Diskussion ab und suchte Wege, die sprachwissenschaftlich weniger belastet erschienen. Ein „anderer Grammatikunterricht“ wurde von Boettcher/Sitta (1978) gefordert. Syntaktische Strukturen einer Sprache seien aus situativen Handlungszusammenhängen herauszuarbeiten (Henrici 1979, 81; Böttcher/Sitta 1983, 378—398; Klein 1984 , 279). Es gehe darum, den

1549

Sprachgebrauch bewußtzumachen und nicht die Theorie über eine Sprache zu vermitteln oder Regeln eines abstrakten Sprachsystems (Doderer 1965, 117; Menzel 1975, 85; Zander 1978, 61; Klein 1984 , 275—283). Wichtig sei, Intuition für den Aufbau des Satzes und die Funktion des Wortes zu wecken (Glinz 1961, 292; Helmers 1966, 223; Spitta 1990, 4 ). Pragmatik und Textlinguistik riefen mehr Aufmerksamkeit hervor (Bremer Kollektiv 1974 ; Rank 1977; Henrici 1979, 85; Braunroth 1984 , 267). Sie lenkten zugleich von der Tatsache ab, daß auch ein freies Sprechen über syntaktische Strukturen theoriegeleitet erfolgen muß, wenn ein solches nicht willkürlich werden soll. Daran ändert sich auch nichts, wenn zwischen einer produktions-, rezeptions- und analyseorientierten Grammatik unterschieden wird (Kleineidam 1989, 24 6— 24 9) und diesen spezifische Arbeitsformen zugewiesen werden. Ein Scheitern sprachwissenschaftlicher Erklärungsversuche für syntaktische Kategorien sieht Primus (1987, 66—67) in den heterogenen Merkmalen syntaktischer Phänomene, die durch homogen definierte grammatische Relationen zu erfassen versucht worden sind (Reis 1982). Sprachdidaktisch sind die Versuche, eine linguistisch formal ausdefinierte Syntax dem Unterricht zugänglich zu machen, mißlungen, weil sie die natürliche Sprache nur extrem rudimentär darzustellen erlaubten und bereits für einfache Konstruktionen einen sehr umfangreichen theoretischen Rahmen im Vorfeld vermitteln mußten. Ein Grund, warum sich die Sprachdidaktik im Umgang mit syntaktischen Fakten schwer getan hat, lag in der „ontologischen“ Bewertung der Theoriefragmente bzw. einer Psychologisierung linguistischer Theorien (Hamburger Autorenkollektiv 1975, 6). Es wurde so getan, als seien syntaktische Daten von einer Theorie vollständig beschreibbar. Diese Fehleinschätzung hatte auch dazu verleitet, linguistische Theorien zu didaktisieren (Bildungsplan 1969, 10; Menzel 1975, 95; Kerkhoff 1983, 14 ). Zu bedenken sind Beobachtungen zur progressiven Entwicklung syntaktischer Konstruktionstypen, auch wenn die Forschungen darüber noch unzureichend sind (Felix 1977). Unbestritten ist, daß sprachliche Phänomene im Unterricht besprochen werden sollen (Böttcher/Sitta 1978, 157; Wode 1981; Butzkamp 1981, 55; Balhorn 1981, 4 4 ; Felix 1982; Oomen-Welke 1982, 14 ), und daß syntaktische Themen dazu gehören (Reuschling 1982). Offen ist hinge-

1550

gen, wie das, was über Sprache gewußt wird, über das „naive Sprechen“ hinaus für den Sprachgebrauch nutzbar gemacht werden kann (Herrlitz 1975; Nündel 1976; Schwenk 1976; Zander 1978; Spitta 1990). Eine grundlegende Schwäche der gegenwärtigen Satzbetrachtungen in der Schule ist der Selektizismus und der willkürlich anmutende linguistische Eklektizismus (Sitta 1974 , 4 32; Menzel 1975, 80; Kerkhoff 1983). Dieser wird sogar als sprachdidaktisch notwendig angesehen (Gleason 1965, 4 94 ; Nickel 1968, 13; Aßheuer/Hartig 1976, 25; Ader 1983). Unter dem Hinweis auf eine schülernahe Betrachtungsweise werden unterschiedliche Beschreibungsebenen, Kategorien mit verschiedenen theoretischen Ansätzen und pseudomathematische Veranschaulichungstechniken gewählt (Sowinski 1969, 171; Hamburger Autorenkollektiv 1975; Menzel 1975; Henrici 1979; Reuschling 1981; Ader 1983). Auf diese Weise entsteht Inhomogenität im methodischen Vorgehen und bei der Darstellung syntaktischer Eigenschaften, die bereits in sich heterogene Züge tragen. Es überrascht daher nicht, wenn in der Sprachdidaktik die Methodikfrage völlig offen ist, eine verbindliche Methode für den Sprachunterricht nicht existiert (OomenWelke 1982; Neuner 1983; 1989) und in zunehmendem Maße integrativ verfolgt wird (Homberger 1977, 36; Boettcher/Sitta 1978; Wunderlich/Conrady 1978, 301; Klein 1984 , 279; Hohmann 1990), wobei ein systematischer Grammatikkurs übereinstimmend abgelehnt wird (Boettcher/Sitta 1978, 182; Gewehr 1979, 40; Balhorn u. a. 1981, 42). Eine neue Sichtweise von Syntax ist nicht zu erkennen. Die zugrundeliegenden grammatischen Konzepte sagen zu wenig über die semantische Leistung von Konstruktionen aus und finden keinen Zugang zu pragmatischen Funktionen. Syntax endet schließlich wieder bei der Bestimmung von Satzbauplänen, Morphologie und Besonderheiten der Satzgliedfolge sowie dem Einüben in die Formen derselben (Menzel 1975; Reuschling 1982). 1.5. Syntax als „Operation“ Syntax im Unterricht muß als „Weg“ definiert werden. Nicht das Wissen über die Syntax ist das Lernziel des Unterrichts, sondern Syntax als Methode, wie ein Bezug zum Handeln (Klein 1979) und Denken durch das Verknüpfen sprachlicher Mittel erreicht wird (Weisgerber 1983). Der operationale Umgang

XXII. Syntax und Nachbardisziplinen

mit sprachlichen Mitteln steht im Mittelpunkt. Dies geschieht aber nicht, um ein abstraktes System zu ermitteln oder die Hypothesenbildung von Regeln zu ermöglichen. Im Vordergrund steht die Grundfrage, inwieweit sprachlich das erreicht wird, was vom Sprecher intendiert und vom Hörer erkannt werden soll (Heringer 1988, 1). Ein einzelnes Wort leistet das nicht, dazu bedarf es syntaktisch geleiteter, semantischer Funktionen. Den Anspruch, satzsemantische Aspekte zu berücksichtigen, erheben verschiedene sprachtheoretische Ansätze. Erben (1984 , 30) betont für eine moderne Grammatik die Priorität der Semantik gegenüber morphosyntaktischen Formvarianten, Heringer (1988, 35) hebt die Notwendigkeit, Sinnzusammenhänge und grammatische Strukturen aufeinander zu beziehen, hervor, und Polenz (1988, 51) erkennt ganz allgemein in der deutschen Grammatik die Tendenz, das Inhaltliche wieder in die Beschreibungen einzubeziehen. Was mit der Suche nach der Tiefenstruktur eines Satzes in der Generativen Transformationsgrammatik begonnen hatte (Chomsky 1965), führte über die Kritik am Transformationsbegriff zur Entwicklung der Generativen Semantik (Lakoff 1970), der Kasusgrammatik (Fillmore 1968; 1977) und zu grundsätzlichen Neuansätzen in der Lexikalischen Funktionalen Grammatik (Bresnan 1978; 1982). Die Syntax der Kategorialgrammatik leitet sich aus der Darstellung semantischer Funktionen her (Bartsch u. a. 1977, 205). Im Gegensatz zur Syntaxdarstellung strukturalistischer und transformationsgrammatischer Ansätze läßt sich das kategorialgrammatische Konzept auf elementare Kategorien zurückführen. Psycholinguistisch kann keinem der bekannten Modelle eine durch empirische Ergebnisse eindeutig belegbare Präferenz zugewiesen werden. Sprachdidaktisch sind solche Konzepte zu favorisieren, bei denen semantische und pragmatische Sprachfunktionen integriert werden können. Der kategorialgrammatische Ansatz bietet ein Konzept, das in die 30er Jahre zurückreichend (Ajdukiewicz 1935) einer Idee folgt, die eine Vielfalt von Lernzielen zu realisieren im und für den Sprachunterricht eröffnet. Der Grundgedanke dieses Konzeptes besteht aus der Annahme, aus zwei oder nicht wesentlich mehr Basiskategorien zusammengesetzte Kategorien zu bilden, die semantische Effekte und das syntaktische Konstruktionsprinzip anzeigen (Bartsch 1977).

86.  Syntax und Sprachdidaktik

Viele kategorialgrammatische Arbeiten beschränken sich auf die Kategorien „Individuen-“ , „Gattungsname“ und „Satz“ . Mit Hilfe dieser Menge elementarer Kategorien lassen sich komplexe bilden, indem angegeben wird, wie und zu welchen Resultaten Verbindungen dieser elementaren Kategorien führen. Ein Satz stellt sich als Resultat einer Folge von Verknüpfungsoperationen dar. Im einfachsten Fall ist er das Ergebnis einer Verschmelzung aus einem Individuennamen und einem finiten Verb. „N“ sei der Kategorienname für Individuennamen und „S“ stehe für Satz. Die Äußerung (1) Peter schläft erhielte die Interpretation: Peter gehört zur Kategorie „N“ , schläft zu einer Kategorie des Typs „S/N“ , d. h. nimm einen Ausdruck der Kategorie „N“ und mache einen Ausdruck der Kategorie „S“ .

Die Verknüpfungsoperationen sind semantisch motiviert, so daß sie zugleich eine Interpretation im Hinblick auf Aussagen über die Wirklichkeit beinhalten. Ein Individuenname nimmt Bezug auf ein spezielles Objekt. Ein einstelliges Verb kann einem logischen Prädikat entsprechen. Eine Verknüpfung bedeutet prädikatenlogisch, eine Aussage über einen speziellen Wirklichkeitsausschnitt zu machen. Pragmatisch wirksam ist der Bezug zur Wirklichkeit, weil die Aussage auf einen Ausschnitt bezogen werden muß, der von Sprecher und Hörer als möglicher Objektbereich anzuerkennen und als wahr oder falsch zu bewerten ist. Für jemanden, der eine Sprache lernt oder sich mit ihren Funktionen beschäftigt wie im muttersprachlichen Unterricht, kommt es immer darauf an, herauszuarbeiten, mit welchen Mitteln erwünschte Effekte auf der semantischen und pragmatischen Seite erreicht werden können. Es beginnt bei der normgerechten Wahl der morphologischen Formen und reicht bis zur Bewertung der Situationsangemessenheit lexikalischer, textsortenspezifischer und allgemein stilistischer Elemente. Für das Grundschulkind stellt sich das besondere Problem, Sachverhalte erstmals primär sprachlich vermitteln zu müssen, wobei zusätzlich der Schriftlichkeit eine besondere Rolle zukommt. Eine solche Aufgabe ist nur lösbar, wenn Bezüge zur gedachten Welt kor-

1551

rekt hergestellt werden (Ingendahl 1975, 79; Weisgerber 1983, 354 ). Das setzt die Beherrschung von Satzkonstruktionen voraus, was allgemein angenommen wird (Joos 1966; Pregel 1969; 1970; Menzel 1975, 35; Spies 1989), und es macht die Fähigkeit erforderlich, die Bezeichnungsleistung einer Konstruktion abschätzen zu können (Kaper 1959, 192). Das Kind muß also Testverfahren erwerben, die nicht nur prüfen, ob ein Wort passend gewählt worden ist, sondern wie aus Worten der intendierte Gegenstand erkannt werden kann (Selinker 1972; Aßheuer/Hartig 1976; Wode 1988, 93—105). Eine kategorialgrammatisch angelegte Betrachtungsweise erlaubt nun dem Kind, seinen Sprachgebrauch semantisch zu überprüfen, indem Äußerungssegmente in ihrer Abfolge als Auf-, Ab- und Umbau semantisch testbarer Aussagen erfahren werden können. Etwas anders stellt sich die Situation für den Personenkreis dar, der bereits über eine diskursgrammatische Kompetenz verfügt, sich aber der Normerwartung des gewählten sprachlichen Inventars unsicher ist, weil er die Sprache nicht oder noch nicht ausreichend beherrscht. Wichtig ist, dem Lernenden zeigen zu können, worin sich die syntaktischen Wege einzelner Sprachen unterscheiden. Grundsätzliche Zweifel am pädagogischen Sinn solcher Erklärungen werden nicht geäußert (Engelen 1979; Butzkamp 1981, 55—57). Sie sind aber nicht mit dem Erwerb oder der Unterstützung des Erwerbs solcher Strukturen gleichzusetzen. Das kategorialgrammatische Konzept bietet mit seiner Technik, Operationen im Hinblick auf syntaktische Resultate zu definieren, eine gute Bezugsbasis für den Sprachlerner, der nicht abstrakte Merkmale wie Kasus oder morphologische Markierungen lernt, sondern beobachten und erkennen muß, wie die Sprache bestimmte Mittel wählt, um Äußerungssegmente zusammenzufügen, so daß dadurch Bezeichungseinheiten entstehen (Wode 1977, 20—21). Unter diesem Aspekt kommt den formalen Mitteln Aufmerksamkeit zu. Es wäre jedoch falsch, darin den einzigen Schwerpunkt zu sehen. Konstruktionsprinzipien bedingen sich gegenseitig. In der Sprachentwicklung lassen sich vergleichbare Interdependenzen beobachten. Das Wissen um solche Konstruktionsvarianz ist daher nicht nur stilistisch von Bedeutung (Wode 1977). Darüberhinaus ist zu demonstrieren, welche Wege zu identischen semantischen Resultaten

XXII. Syntax und Nachbardisziplinen

1552

führen, und worin sie sich dennoch unterscheiden lassen und deshalb nicht austauschbar sind. Als Problem eigener Art stellt sich der Unterricht dar, wenn Sprache einem Personenkreis vermittelt werden muß, der sich eine neue Sprache natürlich erworben hat (Burstall 1974 ; Asher 1977; Meyer-Ingwersen 1977; Keim 1978). Hier ist die analytische Arbeit des Lehrers gefordert. Er muß aus den Fehlern auf die syntaktische Kompetenz des Sprechers schließen, um daraus Aufgaben und Übungen zu entwickeln. Dafür bedarf es eines Modells, das pragmatische, semantische und syntaktische Funktionen deutlich voneinander abhebt, ohne den funktionalen Zusammenhang außerachtzulassen (Nold 1981, 153—154 ). Die semantische Komponente im kategorialgrammatischen Beschreibungsansatz zwingt zu einer solchen Differenzierung und klammert dabei die Wirksamkeit syntaktischer und pragmatischer Faktoren nicht aus, so daß für die Entwicklung von Thesen über den Stand des Sprachlerners gerade dieses Modell eine gute Basis bietet. Damit wird allerdings vorerst nur ein Zugang zur „reproduktiven“ (Felix 1977, 25—33) bzw. „rezeptiven Kompetenz“ (Heringer 1988, 2—4 ) geschaffen. Der Lehrer kann Fehler beschreiben, so daß ihm Einblick in das sprachliche Verhalten des Sprechers, wie dieser Bezüge zur Wirklichkeit arrangiert und welche sprachlichen Wege er dabei wählt, möglich wird. Bedeutsam für die Wahl eines Modells als pädagogische Grundlage ist die Möglichkeit einer flexiblen Kategorienbildung. Hier bietet die Kategorialgrammatik im Gegensatz zu den alternativen Modellen intutiv leicht vermittelbare Zugänge. Der Auf- und Ausbau komplexer Strukturen kann in Abhängigkeit zum Entwicklungsstand des Lerners gestaltet werden (Sucharowski 1977). Das Grundinventar elementarer Kategorien läßt sich um eine begrenzte und leicht zu handhabende Zahl erweitern (Primus 1987, 68), ohne an Übersichtlichkeit zu verlieren. Die Verbindung zu den bekannten syntaktischen Funktionen bleibt erhalten, und die komplexen Kategorien lassen sich wie Umschreibungen derselben lesen, so daß weder für den Lehrer noch für den Schüler Probleme entstehen, wenn er Kenntnisse in der herkömmlichen grammatischen Terminologie und Begrifflichkeit nachweisen muß. Die Forderung nach der Einbettung einer pädagogisch begründeten Auseinandersetzung mit Sprache in die

Sprachwissenschaft ist gewährleistet, ohne daß die Fachwissenschaft didaktischen Problemstellungen im Wege steht (Henrici 1979, 85).

2.

Prädikat

2.1. Prädikat als Operationen zum Satz Für die Sprachdidaktik war die syntaktische Relevanz des Prädikats stets unbestritten (Glinz 1952). Weglaßtest und Umstellprobe belegen die hervorgehobene Stellung (Aßheuer/Hartig 1976, 69—72; Oomen-Welke 1982, 84 —104 ). Das Prädikat ist nicht weglaßbar, ohne daß eine Äußerung ungrammatisch wird oder ihre pragmatisch-semantische Funktion signifikant verändert. Umstellungen sind nur bedingt möglich und bewirken bei Erststellung eine pragmatische Veränderung. Die Aussage- wird zur Frageform. Das Prädikat ist Träger des grammatischen Tempus, Modus, Numerus und der Person. Syntaktisch bedeutsam ist die Übereinstimmung des Numerus und der Person mit dem Subjekt, wodurch eine besondere Nähe zu diesem Satzglied entsteht. Das Prädikat legt darüber hinaus die Kasus der von ihm abhängigen (pro)nominalen Terme fest. Aus der Sicht des Satzes kann das Prädikat wie eine Folge von „Operationsanweisungen“ zur Interpretation oder Konstruktion eines Satzganzen gedeutet werden, indem es Hinweise darauf gibt, wie aus Einzelteilen ein Ganzes abgeleitet oder gebildet werden kann. Satzbetrachtung wird so zur Bestimmung des Wegs, wie aus Wörtern ein Satz wird. Äußerung (1) Peter hat geschlafen Syntaktische Interpretation (i—iv) (i) SUCHE NACH DER FINITEN FORM DES VERBS, (1a)Peter hat geschlafen (ii) SUCHE NACH DEM INFINITEN VERBTEIL UND VERKNÜPFE BEIDES ZUR VERBALPHRASE. (lb) Peter hat geschlafen (iii) SUCHE DEN NÄCHSTSTEHENDEN, NOCH NICHT VERKNÜPFTEN NOMINALEN TERM (1c)Peter hat geschlafen (iv) UND MACHE AUS DIESEM UND DER VERBALPHRASE EINEN SATZ. (1d)Peter hat geschlafen. Der zu analysierende Satz wird in den syntaktisch bereits interpretierten Teilen unterstrichen. Zur Dokumentation der Analyse

86.  Syntax und Sprachdidaktik

könnte ein „Protokoll“ erstellt werden, in dem die zusammengehörenden Teile notiert und grammatisch klassifiziert werden. (1a) (hat) V-FIN (1b) ((geschlafen)V-INFIN (hat)V-FIN) VP (1c) (Peter)N ((geschlafen)V-INF (hat)V-FIN) VP (1d) ((Peter)N ((geschlafen)V-INF (hat)V-FIN) VP)S So wie sich Operationsanweisungen zur Interpretation formulieren lassen, sind auch Anweisungen zur Konstruktion möglich. Äußerung (1) Peter hat geschlafen Konstruktionen (i—iv) (i) WÄHLE EIN LEXEM DER KATEGORIE NOMEN, DAS AUF EIN INDIVIDUUM MIT NAMEN PETER VERWEIST.

(1 a) Peter = Nomen (ii) WÄHLE EIN LEXEM DER KATEGORIE VERB, DAS AUF DIE EIGENSCHAFT SCHLAFEN HINWEIST.

(1b) schlaf- = Verb (iii) WÄHLE DIE MARKIERUNG FINITER UND IN(1c) (iv)

FINITER VERBTEIL DES PERFEKTS, DIE MIT DER 3. SINGULAR ÜBEREINSTIMMT. _N hat geVen FÜGE DIE LEXEME EIN, SO DASS EIN AUSSAGESATZ ÜBER EIN INDIVIDUUM GEBILDET WIRD.

(1d) Peter hat geschlafen. Ziel sprachdidaktischer Darstellung muß die Beschreibbarkeit natürlicher Äußerungen sein. Es wird bewußt auf Formalisierungen im Sinn baumgraphischer Darstellungen oder quasimathematischer Notationen verzichtet, obwohl solche für das zugrundeliegende Modell existieren (Vennemann/Jacobs 1982). Diese Darstellungstechniken stellen nämlich für sich ein didaktisches Problem dar und lenken, wenn ihr heuristischer Wert nicht erkannt wird, von der Sprachbeobachtung ab (Wunderlich 1978, 305). Diese ist aber das vorrangige Ziel. 2.2. Syntaktische Interpretation der Äußerungsstruktur Die vorgeschlagene „Protokollierung“ der grammatischen Interpretation löst die originale Stellung der einzelnen Ausdrücke in bestimmten Fällen auf. Das geschieht nicht ohne Absicht. Die syntaktische Operation soll bewußt gemacht und gegen die „syntaktische Regel“ hervorgehoben werden, die die Klammerstellung oder die Zweitstellung des

1553

Prädikats bewirkt (Clahsen 1981). Regularitäten der Abfolge einzelner Terme lassen sich wie die syntaktischen Operationen als Anweisungen formulieren, die ein „syntaktisches Protokoll“ im Hinblick auf eine Äußerung interpretieren. Von „Äußerung“ und nicht von „Satz“ wird gesprochen, um anzuzeigen, daß es im Unterricht nicht darum geht, ein spezielles grammatisches Paradigma zu belegen, sondern Ziel ist es, ein sprachliches Ereignis zu erklären, das ein Satz sein kann, aber auch von Fall zu Fall eine Nominal-, Verbal- oder eine andere syntaktische Phrase beinhalten könnte, wie das Beispiel (2) zeigt. (2) A: Wo steckt er? B: Peter? A: Kommt gleich! Eine Eigentümlichkeit der deutschen Verbalphrase ist die Konstruktion der Vergangenheitstempora mit haben und sein, die seit Anfang des 9. Jahrhunderts beobachtet wird und vom 16. Jahrhundert in der heutigen Form der Grammatikalisierung üblich geworden ist (Ebert 1976, 57—61). Sie beinhaltet eine Zweiteilung der Verbalphrase in einen finiten und infiniten Teil der Konstruktion. Die Gliedfolge beider Teile hat eine eigenständige Entwicklung durchlaufen (Lenerz 1977; 1984 ), so daß heute zwischen der Folge im Haupt- und der im Nebensatz unterschieden werden muß. (3) Peter hat Inge eine Rose mitgebracht. (3a) Es gab Ärger, weil Peter eine Rose mitgebracht hat. (4) Peter hätte Inge eine Tulpe mitbringen können. (4a) Isabell meinte, daß Peter Inge eine Tulpe hätte mitbringen können. Das Phänomen, das als „Satzrahmen“ oder „Satzklammer“ bezeichnet wird, ist auch im Zusammenwirken mit einigen adverbialen und nominalen Ergänzungen sowie einer Reihe von Partikeln bzw. Präpositionen zu beobachten. (5) Peter bringt über den Vorfall alles in Erfahrung. (5a) Peter hat über den Vorfall alles in Erfahrung gebracht. (5b) Die Sache ist aufgeklärt, weil Peter alles über den Vorfall in Erfahrung gebrachthat. (6) Peter schenkt Blumen her

XXII. Syntax und Nachbardisziplinen

1554

(6a) Stefan ist immer überrascht, wenn Peter Blumen herschenkt. Entsprechend sind für einzelne Verben Operationen festzulegen, die die Suche nach einer möglichen Verbergänzung beschreiben und zu prüfen erlauben, wie in einer Äußerung eine solche zu finden ist. Viele Verben besitzen die Eigenschaft, zur Konstruktion eines Satzes mehr als nur ein nominales Glied an sich zu binden, was von altersher zur Unterscheidung von transitiven und intransitiven Verben geführt hat. (7) Inge küßt Peter. (8) Inge überreicht Peter Blumen. (9) Inge beschuldigt Peter der Untreue. Operationen müssen daher verbspezifisch angeben, ob ein oder mehrere nominale Glieder zur Realisierung eines Satzes nötig sind. Für das Verb „besuchen“ , das zwei nominale Terme zu sich nehmen kann, wären dann die folgenden Operationen vorstellbar: Äußerung (10) Peter hat Uli besucht Syntaktische Interpretation (i—iii) (i) SUCHE NACH DER FINITEN VERBFORM, (10a) (ii)

(10b) (iii)

(10c)

DEM INFINITEN VERBTEIL UND VERBINDE BEIDE ZUR VERBALPHRASE: (besucht hat)VP SUCHE NACH DEM DER VERBALPHRASE AM NÄCHSTEN STEHENDEN (PRO)-NOMINALEN TERM UND BILDE DANN EINE ERWEITERTE VERBALPHRASE, WENN DAMIT KEINE MORPHO-SYNTAKTISCHEN BEDINGUNGEN VERLETZT WERDEN: ((Uli)N (besucht hat)VP)VP SUCHE NACH DEM DER VERBALPHRASE AM NÄCHSTEN STEHENDEN (PRO)-NOMINALEN TERM UND BILDE DANN EINEN SATZ, WENN DAMIT KEINE MORPHO-SYNTAKTISCHEN BEDINGUNGEN VERLETZT WERDEN: ((Peter)N ((Uli)N (besucht hat)VP)VP)S

Auf dieselbe Weise würde die Äußerung (11) Peter hat Uli Blumen geschenkt. als Satz erkannt werden, wenn das Verb schenken den Hinweis erhält, drei nominale Terme zur Satzkonstruktion zu benötigen. Die Suche nach Termen, die zur Konstruktion eines Satzes nötig sind und miteinander verbunden werden müssen, folgt im Deutschen nur bedingt topologischen Regeln. Die Satzgliedfolge wird durch die Wahl der Satzperspektive beeinflußt, ist abhängig von dis-

kursgrammatischen Bedingungen und stützt sich auf den morpho-syntaktischen Rahmen bzw. das Wissen um die Sorte von Dingen, die es denotiert. Das hat zur Folge, daß im Deutschen die Verknüpfungsoperation nicht nur in eine bestimmte Richtung erfolgt. Während in der Nebensatzkonstruktion vom Verbalkomplex ausgehend die Glieder linksseitig „aufgespürt“ werden, sind sie im Aussagesatz links und rechts verteilt. Der durch ein Verb gesetzte nominale Rahmen bedingt daher nicht nur eine bestimmte Anzahl von (pro)nominalen Termen, sondern legt sie auch durch spezielle Kasus fest. Eine Suchoperation wird somit mehrfach gelenkt. Sie sucht den noch unverknüpften Term in eine bestimmte Richtung und prüft, ob dieser mit der morphologischen Bedingung des geforderten Kasus kollidiert. Äußerung (12) Den Jungen tadelt der Lehrer. Syntaktische Interpretation (i) SUCHE RECHTSSEITIG NACH EINEM (PRO)-

(12a)

NOMINALEN TERM, WENN ES DIE KASUSMORPHOLOGIE ZULÄSST, INTERPRETIERE IHN ALS AKKUSATIVPHRASE, ((derLehrer)*NP.ACC (tadelt))VP

Die Interpretation scheitert an der Unverträglichkeit der morpho-syntaktischen Bedingung. der kann keinen Akkusativ anzeigen. (ii) ANSONSTEN SUCHE LINKSSEITIG NACH EI(12) (iii) (12c)

NEM TERM, DER IM SINN DES ERFORDERLICHEN KASUS GEDEUTET WERDEN KANN, ((denJungen)NP.ACC (tadelt))VP SUCHE DEN (PRO)NOMINALEN TERM, DER ALS NOMINATIV INTERPRETIERT WERDEN KANN, UND BILDE EINEN SATZ. ((derLehrer)NP.NOM (denJungen)NP.ACC (tadelt)VP)S

2.3. Besonderheiten der Wertigkeit Da sich die Wertigkeit des Verbs als ein dominierendes Konstruktionsprinzip darstellt, muß der „Interpreter“ wissen, welche Operationen vom Verb ausgehen (Vennemann 1977; Strube/Hemforth 1990), um zu einer angemessenen grammatischen Interpretation zu gelangen. Die Suche nach dem Verb ist daher eine ganz wesentliche Operation. Der durch ein Verb bedingte Satzrahmen ist im Deutschen nicht nur nominal besetzt. Witterungsverben werden mit dem unpersönlichen Personalpronomen es gebraucht, (13) Es regnet.

86.  Syntax und Sprachdidaktik

Reflexivverben verlangen ein Reflexivpronomen, (14) Peter schämt sich. und viele Verben verbinden sich mit speziellen Präpositionen. (15) Peter spottet über Anne. Eine besondere Rolle kommt sein, werden oder bleiben in der Rolle der Kopula zu, weil sich damit spezielle semantische Regeln verbinden. (16) Peter ist verliebt. (17) Inge wird Lehrerin. (18) Eva bleibt Ministerin. Die syntaktische Operation zur Beschreibung der Kopulaausdrücke muß anzeigen, wie der Adjektivterm bzw. der entsprechende Nominalterm mit dem Subjekt verknüpft wird und auf den Denotatsbereich desselben referiert. Von den üblichen Testverfahren können nur die sprachdidaktischen Ansätze des Phänomen Kopula erklären, die Umformungsübungen an dieser Stelle einführen und so den attributiven Charakter herausstellen. Das ist pädagogisch bedeutsam, weil Kinder die Besonderheit des Konstruktionsprinzips intuitiv wahrnehmen und irritiert reagieren, wenn die Tests einen solchen Bezug nicht durchsichtig machen. Bei einer Reihe von Verben, die als Verben des Sagens, Wahrnehmens und Denkens zusammengefaßt werden, tritt an die Stelle des nominalen Terms nicht selten ein Satz (19) Isabell sagt Andrea zu, daß sie Peter morgen besucht. (20) Sylvie sieht, daß Hans eine andere hat. (21) Inge hofft, daß sie Peter gleich antrifft. oder eine Infinitivkonstruktion. (22) Peter verspricht, Isabell zu küssen. (23) Hans bittet, gleich zu ihm zu kommen. (24) Andreas erkennt, bei Andrea keine Chancen mehr zu haben. Eine mögliche syntaktische Interpretation muß daher nach einem Term suchen, der u. U. nicht von einem Nomen dominiert wird, sondern ein eigenständiger Satz ist oder eine Infinitivkonstruktion. Syntaktische Interpretation (i) SUCHE (RECHTSSEITIG) NACH EINEM TERM, DER NOMINAL, ALS (ERWEITERTER) INFINITIV MITzuODER (daß-)SATZ REALISIERT SEIN KANN, UND INTERPRETIERE IHN ALS AKKUSATIVPHRASE, WENN KEINE MORPHOLOGISCHEN BEDINGUNGEN ODER GRÜNDE DES DENOTATSBEREICHS DAGEGEN SPRECHEN.

1555

Wenn der Term als Satz erkannt wird — das Auffinden eines daß kann diese Identifikation ermöglichen —, finden die bereits vorgestellten Operationen ihre Gültigkeit, indem das Prädikat gebildet wird und seine „Mitspieler“ identifiziert werden. Wird der Infinitiv vorgefunden — das zu ist ein mögliches Indiz —, dann gelten die Verknüpfungsoperationen zum Auffinden (pro)nominaler Terme mit Ausnahme des Subjektterms. 2.4. Funktion der „Semantischen Rollen“ In der Valenzdiskussion wurden neben den morphologischen Rahmenbedingungen auch sog. semantische Rollen in die grammatische Beschreibung einbezogen. Einzelne (pro)nominale Terme werden in Abhängigkeit zur morphologischen und strukturellen Kennzeichnung dann als Agens, Patiens, Experiencer, Thema, Instrument, Ziel, Herkunft oder Ort gedeutet (Fillmore 1977; Koch 1978; Jakkendorf 1983; Wunderlich 1985). (25)Inge schenktPeter (i) AGENS

Blumen

PATIENSOBJEKT

Es ist vorstellbar, diesen Deutungsvorgang in das allgemeine Konzept zu integrieren, so daß nach einer Zuweisung geprüft wird, ob sie vom lexikalischen Befund her einen Widerspruch auslöst. (26) Inge heiratet den Stein. Der als Akkusativobjekt markierte Term dieser Äußerung setzt einen menschlichen Adressaten voraus. Es kann sich daher bei den Stein nicht um einen Gegenstand handeln. Wenn das dennoch der Fall ist, dann sind nichtgrammatische Interpretationsregeln nötig, die den Stein als Namen einer Person, metaphorisch oder eine spezielle Redeweise „erkennen“ und interpretieren. 2.5. Semantische Aussage Grundbaustein ist das einzelne Wort und jeweilige Eigenschaften, wie sich aus Wörtern umfassendere Einheiten entwickeln. Vom Wort her erschließt sich für das Kind ein differenziertes Verständnis der Welt. Eine inhaltliche Auseinandersetzung mit Problemen der Syntax findet hier eine sinnvolle Basis. Wenn das Wort ein Prädikat ist, dann kann ein Weg angegeben werden, der als

XXII. Syntax und Nachbardisziplinen

1556

mögliche Einheit am Ende einen Satz darstellt, der etwas über ein Ereignis, Zustände oder allgemeine Sachverhalte aussagt. Somit wird nicht nach einem abstrakten System gesucht, sondern nach „Anweisungen“ , wie aus einzelnen Wörtern ein Gebilde entsteht, das sich zu einer Einheit zusammenfindet und das wir als Satz verstehen und auf semantische bzw. pragmatische Weise deuten. Ein Satzglied erkannt zu haben, bedeutet dann, etwas darüber zu wissen, wie aufgrund sprachlicher Gegebenheiten aus der Sicht des Sprechenden Dinge aufeinander bezogen werden sollen. Die kategorialgrammatische Beschreibungsweise erlaubt nämlich, syntaktisch interpretierte Formen mit analogen Operationen semantisch zu deuten. Ein grammatisches Prädikat läßt sich semantisch als Funktor deuten. Bezogen auf das Beispiel (27) besitzt dieser drei Argumentstellen (Bartsch 1977, 219—226). Äußerung (27) Peter hat Uli Blumen geschenkt Grammatische Struktur (27a) ((Peter)N ((Blumen)N ((Uli)N (geschenkthat)VP)VP)VP)S Semantische Interpretation (i—iv) (i) SUCHE DAS GRAMMATISCHE PRÄDIKAT UND ERSETZE ES UNTER VERNACHLÄSSIGUNG VON TEMPUS UND MODUS DURCH DEN GLEICHLAUTENDEN FUNKTOR.

(27b) SCHENK-_(_, _, _) (ii) SUCHE DIE DEM FUNKTOR „SCHENK-“ LINKSSEITIGE, ERSTE ENTITÄT, MIT EINER DEM DATIV NICHT WIDERSPRECHENDEN MARKIERUNG UND IN DER SEMANTISCHEN ROLLE EINES PATIENS:

(27c) SCHENK- (_, _, ULI) (iii) SUCHE DIE DEM FUNKTOR „SCHENK-“ LINKSSEITIGE, ZWEITE ENTITÄT MIT EINER DEM AKKUSATIV NICHT WIDERSPRECHENDEN MARKIERUNG UND IN DER SEMANTISCHEN ROLLE EINES OBJEKTES:

(27d) SCHENK- (_, BLUMEN, ULI,) (iv) SUCHE DIE DEM FUNKTOR „SCHENK-“ LINKSSEITIGE ENTITÄT MIT EINER NICHT DEM NOMINATIV WIDERSPRECHENDEN MARKIERUNG UND IN DER SEMANTISCHEN ROLLE EINES AGENS:

(27e) SCHENK- (PETER, BLUMEN, ULI) Die Argumente werden aufgrund von Kriterien identifiziert, die morphologisch, strukturell oder semantisch motiviert sind. Wird die beschriebene Darstellungsform benutzt — was an sich nicht nötig ist —, dann liegt nach

dem Auffüllen der Leerstellen eine semantische Interpretation in Form einer prädikatenlogischen Notation vor. Sie beschreibt die Bedeutung des Funktors, wobei die Bedeutungen von Peter, Uli und B lumen die äußerste, mittlere bzw. innerste Argumentposition erfüllen. Die bisherige Interpretation hatte das Tempus- und Modusmorphem hat ge- -en ausgeklammert. Tempus und Modus können in die Betrachtung einbezogen werden, wenn sie als semantische Operationen gesehen werden, die eine Aussage als semantische Einheit „nehmen“ und einem temporal und pragmatisch näher zu beschreibenden System zuweisen, aus dem hervorgeht, wie der Sprecher seine Aussage unter dem Gesichtspunkt der Zeitlichkeit und in welchem Sprechhandlungszusammenhang verstanden wissen will. Wie bei der syntaktischen Beschreibung soll auch hier nicht ein Formalismus im Vordergrund stehen, sondern es soll nur sichergestellt sein, daß es theoretische Grundlagen gibt, die diese Sichtweise absichern. Pädagogisch kommt es darauf an, den Akt bewußt zu machen, daß einzelne Wörter und Formen dazu beitragen, einen Bezug zu den Dingen herzustellen, den Kommunikationsteilnehmer einzubeziehen und zu einem Handlungsumfeld in Raum und Zeit eine Verbindung zu erzeugen. Theoretische Grundlagen bzw. Anknüpfungspunkte bietet die Sprechakttheorie (Gewehr 1979; Pommerin 1983) und Ansätze, die institutionelle Kontexte berücksichtigen (Rehbein 1977). Sie zusammen können dazu beitragen, einen Bezug zu einer gedachten oder vorgefundenen Wirklichkeit auch theoretisch abgesichert zu ermöglichen.

3.

Subjekt

3.1. Widersprüchlichkeiten der traditionellen Subjektdefinition Die Sprachdidaktik sieht im Erkennen des Subjekts keine Probleme. Subjekt wird als Thematräger (Grebe 1973, 4 90; Vennemann 1974 , 266; Weigand 1979, 266), als Bezugsgröße, die auf Personen oder Gegenstände referiert (Keenan 1976, 318; Sasse 1978, Weigand 1979, 266), oder als Träger der Handlung erklärt (Keenan 1976; Li/Thompson 1976). Das ist zwar aufgrund einer hohen diskursgrammatischen Wahrscheinlichkeit oft nicht unzutreffend (Engel 1972, 212—213), es ist aber nicht aus der Syntax des Deutschen

86.  Syntax und Sprachdidaktik

herzuleiten und durch Gegenbeispiele widerlegbar (Reis 1982, 176). Ähnliche Irritationen werden durch die Vorstellung ausgelöst, das Subjekt sei mit der semantischen Rolle Agens verbunden (Bever 1970; Reis 1982; 1986, 68). Faktisch trifft dies nur bei den Tätigkeitsverben zu. Der in der deutschen Sprachdidaktik verwendete Fragetest „wer tut etwas?“ gelangt daher nur dann zum Erfolg, wenn ein Äußerungssegment erfragt wird, das semantisch „Agens“ ist und morphologisch mit Nominativ markiert wird (Menzel 1975, 13 4 —136; Aßheuer/Hartig 1976, 80—84 ). Kinder orientieren sich an unterschiedlichen Merkmalen (Brünner 1982; Brekle 1971, 487). Ein typischer Orientierungspunkt ist die eingangs zitierte Erstposition, die im Deutschen in 60% mit dem Subjekt besetzt ist (Engel 1972, 4 3—4 4 ; Reis 1982, 175) — noch deutlicher fallen die Zahlen für das Niederländische aus (Abraham 1982, 8). Die Erstposition ist diskursgrammatisch Träger der Topik und des Themas und stimmt mit der strukturellen Eigenschaft überein (Keenan 1976). Die traditionelle Methodik setzt mit dem Fragetest beim morphologischen Kriterium an und betont das strukturelle Erfassen des Subjekts durch die Weglaßprobe. Das Weglassen verhindert einen Aussagesatz und erleichtert das Erfragen mithilfe der Ersetzung durch das Fragepronomen. Diese Technik widerspricht den syntaktischen Gegebenheiten des Deutschen nicht. Das ist im Italienischen, Rumänischen und Ungarischen anders (Benkö 1972, 96). Probleme macht eine Äußerung wie (28) Ihm fehlt hier die Erfahrung Grundschulkinder interpretieren ihm als Subjekt. Sie erkennen damit ein strukturelles Phänomen korrekt und haben mit der satzinitialen Funktion des ihm eine wichtige Subjekteigenschaft erfaßt (Primus 1987, 24, 100). Bei Inversionsverben — den statischen wie frieren, grauen, fehlen oder den dynamischen wie zerbrechen, mißlingen, passieren — muß die Satzkonstituente aus dem Dativ hergeleitet werden (Endres 1976; Primus 1987, 100— 104 ). Darin ist ein Konflikt mit der traditionellen Grammatik enthalten. Die herkömmliche Begrifflichkeit grammatischer Konventionen bezieht sich auf das morphologische Kriterium und betrachtet das Subjekt als Nominativ markiert (Glinz 1961, 82; Schmidt 1967, 126; Grebe 1973, 531). Der Wider-

1557

spruch wird nun nicht dadurch behoben, daß das morphologische Kriterium zum Konstituenz erklärt wird (Reis 1986, 71). Zwar hat im Deutschen und Englischen, anders als im Japanischen beispielsweise, die Kongruenz auch in prädikativen Konstruktionen einen großen Geltungsbereich (Harweg 1971). Dennoch kann es zu Widersprüchen führen, wenn ein zweites nominativmarkiertes Glied vorliegt (Abraham 1982, 9; Primus 1987, 103), (29) Das sind Blumen. (29a) *Das ist Blumen. (30) Eine Reihe von Kindern sind hier. (30a) Eine Reihe von Kindern ist hier. 3.2. Subjekt als offene Kategorie Eine Lösung kann erreicht werden, wenn sprachdidaktisch akzeptiert wird, daß Satzglieder nicht durch ein spezifisches Kennzeichen eindeutig bestimmbar sind (Aßheuer/ Hartig 1976, 83; Rall u. a. 1977; OomenWelke 1982, 81). Linguistisch ist der Standpunkt nicht neu (Helbig 1978, 5; Kaneko 1978, 126; Hiersche 1979, 24 9). Damit wird nicht behauptet, daß Satzglieder nicht für sie typische Eigenschaften besitzen und in denselben beschreibbar wären. Allerdings variiert die Menge dieser Eigenschaften im Einzelfall. Für das Deutsche spielen strukturelle und morphologische Eigenschaften eine wichtige Rolle, während im Englischen und Französischen strukturelle Relationen die Satzglieder ausreichend zu definieren erlauben (Abraham 1982, 16; Primus 1987, 103). Es sollte daher nicht vom Subjekt schlechthin, sondern von Subjekteigenschaften eines Terms gesprochen werden. Ein Satzglied, das satzinitial, nominativ-, kongruenzmarkiert, passivierbar und durch ein Reflexivpronomen erreicht wird, ist dann ein gut erkennbarer Kandidat für ein Subjekt. Man könnte von einem Prototypen sprechen. Die Einbeziehung der semantischen Rollen und diskursgrammatischer Besonderheiten erlaubt darüberhinaus Satzgliedcharakterisierungen, die die Beobachtungsgabe für sprachliche Eigenschaften im Sprachunterricht schärfen (Keenan 1976, 303—334). (31) Anne begrüßt den neuen Gast. In dieser Äußerung wird das vorfindliche Subjekt Anne durch folgende Eigenschaften gekennzeichnet: Es ist nicht weglaßbar, d. h. ein Weglassen macht die Äußerung unverständlich oder setzt ein besonderes Kontext- bzw. Diskurs-

XXII. Syntax und Nachbardisziplinen

1558

management voraus. Es steht in satzinitialer Position, d. h. es ist mit der letzten Operation vor der Satzerkennung verbunden. Die WerFrage ist erfolgreich anwendbar, da es im Nominativ steht. Es besteht Kongruenz zwischen Prädikat und Subjekt im Numerus und in der Person. Die Passivbildung ist möglich. Als semantische Rolle ist das Agens wirksam. Es präsupponiert in der Erstposition die Existenz einer Person. Ein Sprecher kann so direkt Bezug auf diese nehmen. In der Funktion des Thematrägers ist von der Person Anne die Rede. Demgegenüber ist der potentielle Subjektkandidat in (32) Dem Kind ist schlecht geworden. wesentlich undeutlicher als „Subjekt“ zu erkennen, weil auf ihn nur eine Teilmenge der üblichen Beschreibungseinheiten zutrifft: Dem Kind befindet sich in Erstposition und besitzt satzinitiale Funktion, es ist das, worauf der Sprecher Bezug nimmt und was vom Zuhörer als bekannt erwartet werden kann. Andererseits gibt es keine Nominativmarkierung, es fehlt die Rolle als Agens, und der Subjektterm ist nicht im üblichen Sinn passivierbar, da die Konstruktion bereits passivähnliche Züge trägt. Für die Kongruenz fehlt ein Bezugsterm. Eine solche Betrachtungs- und Vorgehensweise regt didaktisch dazu an, weitere subjekttypische Eigenschaften zu berücksichtigen, wobei dies unabhängig von der Frage geschieht, mit welcher Systemhaftigkeit sie jeweils verwendet werden (Keenan 1976; Reis 1982). Subjekte sind nur schwer metaphorisierbar, sie besitzen einen weiteren Skopus und speziellere Quantoreneigenschaften als andere Satzglieder (Keenan 1976). Das Subjekt gehört zu den Satzgliedern, die bei koordinierenden Konstruktionen im Fall der Subjektgleichheit wegfallen, im Imperativ und bei Infinitivkonstruktionen nicht realisiert werden (Primus 1987, 65). Das Subjekt kann in einer Reihe von Fällen als daß-Satz formuliert werden und ist selbstverständlich durch einen Relativsatz erweiterbar. Für eine pädagogisch gelenkte und didaktisch reflektierte Beschäftigung mit diesem Satzglied ergeben sich nicht nur zahlreiche Anknüpfungsmöglichkeiten, sondern das breite Spektrum der satzgliedspezifischen Gebrauchsweisen fordert zum reflektierenden Umgang dieser Funktion und seiner für die jeweilige Sprache gültigen Gebrauchsbedingungen heraus. Sprachdidaktisch wird damit von der Illusion Abschied genommen, Satz-

glieder seien durch eine Handvoll Tests hinreichend erklärt. Die Sichtweise ist komplexer, sie ist deswegen aber nicht komplizierter. Da die Tests Eigenschaften ermitteln und nicht das Satzglied schlechthin, weiß der Benutzer, worauf seine Beobachtung gerichtet ist. Erkannte und beobachtete Eigenschaften lassen sich zusammentragen, im Rahmen einer Theorie bewerten, ordnen und erlauben dann grammatische Aussagen. Es muß allerdings einschränkend erwähnt werden, daß eine konsistente Theorie zur Beschreibung fehlt, so daß die Analysen auf bekannte Verfahren beschränkt bleiben. Trotz dieser Einschränkung bedeutet eine klare Trennung zwischen Sprachbeobachtung und grammatischer Aussage über Sprache für den Sprachunterricht einen Fortschritt. Für den Sprachlerner bleibt durchschaubar, wie Grammatik mit dem Beobachteten umgeht. Sprecher einer anderen Sprache können kontrastiv Übereinstimmungen und Abweichungen feststellen, ohne von einer vagen Satzgliedbegrifflichkeit irritiert werden zu müssen. Das Kind findet einen Weg, seiner Intuition Ausdruck zu verleihen und sie an der Sprache selbst zu erproben. Es gibt keinen Formalismus oder ein sog. Modell, das die Wahrnehmung einengt bzw. irreleiten müßte.

4.

Objekte

4.1. Akkusativobjekt Mit der Bestimmung von Objekten hat sich die Sprachdidaktik schon immer schwer getan (Aßheuer/Hartig 1976, 91). Die Gründe dafür sind im Sprachsystem des Deutschen selbst zu suchen, das für die Funktion des Objekts keine einheitliche Kodierung vorsieht. Kasus markieren nicht nur Objekte (Fourquet 1970, 79; Abraham 1982, 10), sondern treten auch zur Kennzeichnung von Zeit- und Mengenangaben auf. (33) Peter wartete den ganzen Tag. (34) Inge kostete das alles eine Menge Geld. Das Erkennen eines Kasus sichert nicht den Zugang zum Objekt (Nikula 1982, 295). Dieses Dilemma wird sprachdidaktisch durch die Anwendung des Weglaßtests zu umgehen versucht (Aßheuer/Hartig 1976, 72—74 ). Da die Objektfunktion eng mit der Semantik einzelner Verben verbunden ist, und die Verben als Operatoren eine bestimmte Anzahl von Operanden fordern, ist dieser Test erfolgreich.

86.  Syntax und Sprachdidaktik

Sprachdidaktisch ungelöst bleibt das Problem, daß einzelne Verben nicht mit einem einzigen Operator identisch sind, weil sie unterschiedliche Bedeutungen realisieren können (Helbig/Schenkel 1969; Engel/Schuhmacher 1978). (35) Peter setzt beim zweiten Takt ein. (36) Inge setzt alle weiblichen Tricks ein. (37) Der Trainer setzt heute Peter ein. (38) Peter wird von Inge zum neuen Intendanten eingesetzt. Während in (35) und (38) einsetzen mit einem Präpositionalausdruck bei/zu etwas gebraucht wird, muß in (36) und (37) ein Akkusativterm verwendet werden. Dieser variiert allerdings hinsichtlich der semantischen Rolle. Die Bedeutungen von einsetzen unterscheiden sich voneinander. Der Weglaßtest setzt voraus, daß die möglichen Bedeutungen eines Verbs erkannt und hinsichtlich ihrer Wertigkeit beurteilt werden können. Das erfordert im Einzelfall mehr linguistische Kompetenz als sprachdidaktisch erwartet werden kann. Die Weglaßprobe ist keine pädagogische Operation schlechthin, sondern ein linguistisches Instrument. Vereinzelt wird in der Sprachdidaktik versucht, strukturelle Merkmale durch Transformationen zu ermitteln (Aßheuer/Hartig 1976, 175—180; Oomen-Welke 1982, 4 0—4 2). Als Umformungstest wird die Passivkonstruktion bevorzugt. Das Umformen einer Konstruktion fällt Grundschulkindern leichter als der Weglaßtest. Allerdings ist nicht jede Satzkonstruktion ins Passiv transformierbar, und das Passiv kennt aufgrund des Vorgangs- und Zustandspassivs unterschiedliche Realisationsformen (Primus 1987, 183—190; Johansen 1988, 66—68). Mit dem Erwerb eines Umformungstyps wird ebenfalls nur eine Teilmenge möglicher Objekte erkannt. Wenn aber das Objekt wie das Subjekt durch Objekteigenschaften beschrieben wird, dann lassen sich die unterschiedlichen Realisationen aufgrund gemeinsamer Eigenschaften als zusammengehörig erkennen, ohne daß sie jeweils alle Eigenschaften miteinander teilen müssen. Im Fall des Akkusativobjekts in (39) Peter jagt den Hund. wird das Segment den Hund unmittelbar vom Verb gebunden. Es hat eine eindeutige Markierung als Akkusativ. Es kann Subjekt im Satz werden, (39a) Der Hund wird von Peter gejagt.

1559

ist durch einen Relativsatz erweiterbar, (39b) Peter hat den Hund, den er entdeckt hat, gejagt. bleibt beim Infinitiv erhalten (39c) Inge mißfällt es, den Hund zu jagen. und ist nicht ohne einen Wandel der semantischen Aussage des Verbs weglaßbar, (39d) Peter jagt. (= ist auf der Jagd oder ist ein Jäger) sofern nicht von einer definiten Ellipse auszugehen ist. (39e) A: Eine Motte! B: Ich jag’ schon. Bei adverbialen Erweiterungen schiebt es sich um die erweiterte Position hinter dem finiten Verbteil nach rechts. (39f) Peter jagt den ganzen Nachmittag den Hund. Mit einer Äußerung wie (39g) Peter jagt, was er kriegen kann. wird die Realisation als Objektsatz erkennbar. Das Objekt verbindet sich mit der semantischen Rolle eines Patiens. Eine Umstellung in die Erstposition, (39h) Den Hund hat Peter gejagt. kommt einer Hervorhebung des zugrundeliegenden Sachverhalts gleich. Traditionell wäre zwischen dem affizierten (4 0) und dem effizierten Objekt (4 1) zu unterscheiden (Sommer 1931, 14). (40) Peter schlägt den Hund. (41) Inge bekommt Peter. Sprachdidaktisch bedeutet die Komplexität dieser Strukturen, differenzieren zu müssen, ob Programme zur Erweiterung der syntaktischen Kompetenz entwickelt werden oder ob Formen des Bewußtmachens dazu beitragen sollen, sich mit Struktureigentümlichkeiten beobachtend und beschreibend auseinanderzusetzen. Grundsätzlich sollte dabei gelten, Aufgaben zu entwickeln, die viele der strukturellen Beziehungen herzustellen erlauben und nicht zu einer reduzierten Verständnisweise verleiten. 4.2. Dativobjekte (42) Inge gibt dem kleinen Peter den Liebesbrief zurück.

1560

Peter ist durch Dativ morphologisch markiert. Um dem kleinen Peter zum Subjekt des Satzes machen zu können, bedarf es einer Umformung durch die bekommen-Konstruktion, da die übliche Passivbildung nicht möglich ist: (43) Der kleine Peter bekommt von Inge den Liebesbrief zurück. Eine Umstellung in die erste Position ist nur möglich, wenn der Kasus morphologisch eindeutig identifizierbar ist oder der vorausgegangene Kontext eine referentielle Zuweisung sicherstellt. Topologisch gilt, Peter folgt dem finiten Verbteil. Der Unterschied zum vorausgegangenen Konstruktionstyp ist eindeutig, so daß in der herkömmlichen Grammatik und in der Sprachlehre vom Unterschied zwischen dem direkten und indirekten Objekt die Rede ist (Eichler/Bünting 1976, 41). Während Schülern der morphologisch korrekte Gebrauch Schwierigkeiten bereitet, macht der strukturelle keine Schwierigkeiten. Der Unterschied wird erkannt. Für den Umgang mit dem Dativobjekt ist daher das ganze Spektrum der zu beschreibenden Merkmale von besonderer Bedeutung, um zur Sicherheit bei der Verwendung der morphologischen Markierung zu gelangen. 4.3. Genitivobjekte Zu den eher problemlosen Fällen zählt das seit dem Frühneuhochdeutschen auf eine überschaubare Anzahl von Fällen beschränkte Genitivobjekt (Rausch 1897; Ebert 1976; 1986). (44) Inge bezichtigt den Jungen einer schrecklichen Eifersucht. Das Genitivobjekt einer schrecklichen Eifersucht ist nicht durch Umformung ins Passiv transformierbar. Es kann aber durch einen Infinitiv ersetzt werden: (45) Inge bezichtigt den Jungen, schrecklich eifersüchtig zu sein. In vielen Fällen kann die Genitvkonstruktion durch eine Präpositionalkonstruktion ersetzt werden. (46) Inge schämt sich ihrer Schwäche für den Jungen. (47) Inge schämt sich wegen ihrer Schwäche für den Jungen. Zu klären bleibt, was der Eifersucht oder ihrer Schwäche vom Genitiv im folgenden unterscheidet und als Objekt auszeichnet:

XXII. Syntax und Nachbardisziplinen

(48) Inge war des Nachts immer zuhause. Topologisch befindet sich des Nachts nach dem finiten Verb, der Eifersucht am Ende des Satzes. Wird der Eifersucht weggelassen, wird im vorausgegangenen Kontext nach einem möglichen inhaltlichen Bezug gesucht. Kann dort keiner gefunden werden, bleibt die Äußerung unverständlich. Das ist bei des Nachts nicht der Fall. Die Äußerung verliert zwar an Aussagegenauigkeit, aber das Fehlen löst keine „Verständniskrise“ aus. Vergleichbares ist bei dem Akkusativ den ganzen Tag zu beobachten. Die andere Wirkungsweise von den ganzen Tag gegenüber den Peter macht zusätzlich eine Umformung deutlich: (49) Inge war zu Hause, das gilt für den ganzen Tag. (49a) *Inge bezichtigt der Eifersucht, das gilt für den Jungen. Die Äußerung (4 9a) ist grammatisch nicht akzeptabel. Diesem Umstand wird mit der Unterscheidung in oblique Objekte und freie Formen Rechnung getragen. In der Sprachdidaktik haben sich dafür die Kategorien „Ergänzungen“ und „Angaben“ durchgesetzt (Reuschling 1981, 184 ; OomenWelke 1982, 34 —4 0; 65—67). „Ergänzungen“ beziehen sich auf alle durch das Verb selegierten (pro)nominalen Terme. Als „Angaben“ gelten die restlichen Satzglieder. Darin wird eine Vereinfachung der Kategorisierung von Satzgliedern gesehen (Heringer 1972, 4 7; 1978; 50—51). 4.4. Der sog. freie Dativ Schwierig bleibt der Umgang mit einigen Realisationen von Dativtermen. Die morphologische Markierung des Dativs ist im Schwinden (Primus 1987, 181). In einzelnen Dialekten ist er nicht mehr markiert und wird durch Präpositionalkonstruktionen ersetzt (Ebert 1982, 4 9—51). Die Situation wird zusätzlich erschwert, weil in der linguistischen Diskussion keine einheitliche Einschätzung darüber existiert, wie die Objektfunktion bei dativmarkierten Äußerungssegmenten einzustufen ist (Johansen 1988, 7—27). (50) Inge macht ihrem Chef ein neues Gutachten. Einerseits kann in der genannten Äußerung ihrem Chef weggelassen werden, ohne daß ein „Kontext-Management“ ausgelöst oder eine „Verständniskrise“ einsetzen würde, und der Ausdruck steht in einer Adverbial-Position,

86.  Syntax und Sprachdidaktik

(50a) Inge macht ein neues Gutachten. (50b) Der Chef bekommt ein neues Gutachten von Inge gemacht. (50c) Der Chef läßt sich ein Gutachten von Inge machen. andererseits kann ihrem Chef durch ein bekommen-Passivin die Subjektposition gebracht werden (Eroms 1978), und eine Reflexiv-Konstruktion ist denkbar. Ein zusätzliches Problem stellen Konstruktionen dar, in denen der dativmarkierte Term Träger von Subjekteigenschaften ist, wie dies bei den Inversionsverben bereits gezeigt worden ist (Primus 1987, 99—124 ). Sie lassen sich auf mehrwertige Verben zurückführen. Der Dativ ist nicht weglaßbar, und die nominalen Terme basieren auf semantisch motivierten Beziehungen innerhalb einer Satzkonstruktion. Die dativmarkierte Konstruktion unterscheidet sich durch die Art der Passivbildung. Weitere Dativkonstruktionen sind weglaßbar und zeigen einzelne Objekteigenschaften. Der Dativus Commodi und Incommodi (51) Inge lebt nur ihren Kindern. (51a) Nur ihren Kindern lebt Inge. (51b) Inge lebt nur für ihre Kinder. kann an erster Stelle stehen, tritt aber nur bei Verben auf, die kein Dativobjekt zu sich nehmen. Er findet sich daher nur bei einer bestimmten Verbgruppe. Üblicher als der Dativ ist die Umschreibung mit für. Der Dativus Ethicus zur Charakterisierung einer persönlichen Stellungnahme (52) Und daß du mir nicht wieder zur Inge gehst. (52a) Und daß du mir nicht wieder der Inge den Hof machst. (52b) *Und daß du nicht wieder der Inge den Hof machst, und das geschah mir. ist nicht erststellenfähig, wird als 1. bzw. 2. Person realisiert und ist neben einem Dativobjekt auffindbar. Er ist weglaßbar, aber der Geschehenstest ist nicht möglich. Der Dativus Judicantis zur Bezeichnung des Standpunktes einer Person gegenüber der geäußerten Sache wird üblich in der Für-Umschreibung gebraucht. Er ist in der ersten Stelle möglich. Darin unterscheidet er sich vom Dativus Ethicus, ansonsten stimmt er mit diesem überein. Eine Umformung in passivähnliche Konstruktionen ist in beiden Fällen nicht möglich. (53) Ferien oder Urlaub, das war uns zu einem Fremdwort geworden.

1561

Der Pertinenzdativ unterscheidet sich von den bisherigen Dativtermen. Zwar ist er erststellenfähig, doch eine Umschreibung mit für ist nicht möglich. Syntaktisch bezieht er sich nicht auf das Verb, sondern auf ein Nomen der Satzkonstruktion. (54) Peter blickt seiner Freundin verliebt in die Augen. Er kann daher in eine Genitivkonstruktion umgeformt werden und zeigt attributive Funktion. Die Dativkonstruktion besitzt eine höhere Stellungsflexibilität. Erschwert wird die allgemeine Situation zusätzlich durch den Umstand, daß eindeutig durch das Verb bedingte Dativterme im allgemeinen Sprachgebrauch weggelassen werden. (55) Inge hat ihm immer wieder geschrieben. Er hat nicht geantwortet. (55a) Inge hat immer wieder geschrieben, und er hat nicht geantwortet. Sprachdidaktisch wird diese Situation nicht gelöst, indem der zu beobachtenden Vielfalt durch eine grobrastigere Klassifikation begegnet wird. Entscheidend ist, wenn eine solche didaktisch erforderlich ist, daß sie nach nachvollziehbaren, linguistisch gesicherten und für den Betroffenen selber prüfbaren Kriterien erfolgt. Es müssen spezifische Eigenschaften sein, wie sie sich aus strukturellen, topologischen, semantischen und morphologischen Bedingungen herleiten lassen und durch entsprechende Testverfahren aufgedeckt werden können. Wenn der korrekte Gebrauch zu lernen ist, dann muß mithilfe eines entsprechenden Sprachkorpus ein solcher Ausschnitt aus der Sprachwirklichkeit angeboten werden. 4.5. Präpositionalobjekte Ein Bereich, der linguistisch sehr unterschiedliche Bewertung erfährt und sprachdidaktisch nicht darüber hinaus thematisiert worden ist, stellt die Zuordnung von Äußerungssegmenten zu den Präpositionalobjekten dar (Glinz 1961; Aßheuer/Hartig 1976, 85). (56) Inge hat an Peter gedacht. Ein Weglassen ist nicht möglich, ohne daß eine als ungrammatisch empfundene Form entsteht. Die Präposition ist nicht durch eine andere ersetzbar (Heringer 1968, 4 34 ; Steinitz 1971, 4 0—4 6; Engelen 1975, 110; Lerot 1982, 261). Das Präpositionalobjekt ist nicht subjektfähig und kann hier durch eine Genitivkonstruktion ersetzt werden. Adverbiale Er-

XXII. Syntax und Nachbardisziplinen

1562

weiterungen schieben die Nominalphrase um die jeweilige Position nach rechts weiter. (56) Inge hat wegen zunehmender Klagen mit Peter über Petra gesprochen. Werden die einzelnen Präpositionalphrasen interpretiert, dann erhält über Petra andere Eigenschaften. Die Phrase kann nicht in eine andere Position geschoben werden, ohne daß die Gefahr eines Mißverständnisses entsteht, d. h. die Phrase würde eine attributive Lesart erhalten. (56a) Inge hat wegen zunehmender Klagen über Petra mit Peter gesprochen. Eine Verschiebung hinter den finiten Verbteil bedeutet eine Hervorhebung. (56b) Inge hat über Petra wegen zunehmender Klagen mit Peter gesprochen. Wenn die Präpositionalphrase über Petra weggelassen wird, (56c) Inge hat wegen zunehmender Klagen mit Peter gesprochen. dann bleibt zwar die Äußerung semantisch interpretierbar, die Bedeutung des Verbs sprechen bezieht sich nur auf den Umstand des Sprechens an sich. Davon zu unterscheiden ist das inhaltlich festgelegte Sprechen, das mit der Phrase sprechen über lexikalisiert wird. Ein Weglassen von wegen zunehmender Klagen schränkt die Aussage zwar ein, löst aber kein Kontext-Management aus oder bedingt eine semantische Neuorientierung für eine Deutung. (56d) Inge hat mit Peter über Petra gesprochen. Die Phrase wegen zunehmender Klagen ist als Kausalsatz anfügbar. (56e) Inge hat mit Peter über Petra gesprochen, weil Klagen zunehmen. Etwas Vergleichbares ist mit über Petra nicht möglich. Eine Umformung im Sinne des Geschehens- und Tuntests ist nur mit Peter erfolgreich. (56f) Inge hat über Petra wegen der Klagen gesprochen, und zwar tat sie das mit Peter. (56e) *Inge hat wegen der Klagen mit Peter gesprochen, und zwar tat sie das über Petra. Sprachdidaktisch lag der Schwerpunkt für die Behandlung der Objekte auf dem Gesichtspunkt der grammatischen Vollständig-

keit eines Satzes (Engelen 1975; Heuer 1977). Eine solche Sicht kollidierte mit der Erfahrung von Äußerungen in konkreten Alltagssituationen. Sie erklärt nicht, warum die Verständigung dennoch möglich ist, und daß keine Schwierigkeiten auftreten, wenn einzelne grammatisch erwartete Terme fehlen. Mit dem hier skizzierten Konzept kommt es zu solchen Problemen nicht. Es zeigt die Vielfalt syntaktischer Zusammenhänge auf und erlaubt, über syntaktische Strukturbeobachtungen und -beschreibungen hinauszugehen, ohne sogleich von der Ungrammatikalität einer Äußerung reden zu müssen. Das Interesse gilt nicht einer abstrakten Darstellung von Satzbauplänen oder der Fixierung einer Norm, sondern es zielt auf das Verstehen von Konstruktionsweisen im Hinblick auf eine unter Kommunikationspartnern geteilte Diskurswelt ab.

5.

Adverbial- und Attributskonstruktionen

5.1. Spezifikation Adverbiale und Attribute unterscheiden sich von den Objekten, weil sie nicht satzkonstitutiv sind. Der in der Sprachdidaktik übliche Umstell- und Weglaßtest erfaßt daher diese Konstruktionen (Oomen-Welke 1982, 34 — 4 5). Allerdings sind nicht alle weglaßbaren Terme automatisch Adverbiale oder Attribute. Der Test reicht insofern nicht aus. Ähnlich ungenau bleibt die Anwendung der herkömmlichen Wie-Frage. Adverbiale und Attribute zeigen gegenüber den Operationen mit Objekten klare Unterschiede. Objekte sind integraler Bestandteil der Prädikatsoperationen und führen schrittweise zur Identifikation eines Satzes als Konstruktionseinheit. Adverbial- bzw. Attributskonstruktionen beziehen sich auf Äußerungssegmente einer Kategorie, um diese mit anderen Äußerungssegmenten zur selben Kategorie zu verknüpfen (Bartsch 1975; Primus 1987, 77—82). (57) Inge bedankte sich am Freitag für die Blumen. Auf der Ebene der bisher besprochenen Operationen würde eine Konstruktion erkannt werden, die den Satz Inge bedankte sich für die B lumen erfaßt. Erst die zusätzliche Operation ermöglicht, die nicht interpretierte Phrase am Freitag aufzugreifen und als Er-

86.  Syntax und Sprachdidaktik

weiterung einer Satzkonstruktion in die bisher geleistete Interpretation zu integrieren. Syntaktische Interpretation (i—ii) (i) NIMM DIE ALS SATZ ERKANNTE KONSTRUKTION

(57a) Inge bedankte sich am Freitag für dieBlumen. (ii) UND MACHE DURCH EINBINDEN DES UNVERBUNDENEN TERMS WIEDER EINEN SATZ.

(57) Inge bedankte sich am Freitag für die Blumen. Entsprechend würde eine Konstruktionsregel vorgehen, indem sie den Satz in Abhängigkeit von der Wertigkeit des Prädikats aufbaut und dann morphologische und topologische Bedingungen angibt, die adverbiale Temporal-, Lokal-, Modal- und Kausalerweiterungen erlauben. Die Operation ist von der Art, daß sie den Ausdruck einer vorfindlichen Kategorie nimmt und neuerlich erzeugt. Der bereits erkannte Satz bleibt nach der Erweiterung ein Satz. Sprachdidaktisch steht dieses Verständnis von Syntax dem Gedanken vom Entfalten der Satzglieder nahe (Hartnacke 1918). Der Charakter einer Konstruktion wird nicht durch das Zergliedern erkannt, sondern durch die Formen seiner Erweiterung. Der Auf- und Ausbau von Konstruktionen bewirkt zugleich deutliche Veränderungen der Bezeichnungsleistung. Anders als bei „reduktionistischen“ Methodiken, die Bedeutung vernachlässigen, dient sie als Bezugsbasis und kommt dem Bedürfnis des Sprachbenutzers entgegen, nicht abstrakte Konstruktionen kennenzulernen, sondern sein Ausdrucksvermögen durch bewußteres Konstruieren zu verbessern. Das „Experimentieren“ mit Konstruktionen, um syntaktische Zusammenhänge zu ermitteln, ist zugleich auch ein Erproben der semantischen, diskursgrammatischen und pragmatischen Leistungsfähigkeit von Ausdrucksformen. 5.2. Satz-Adverbial Warum Weglaßbarkeit sprachdidaktisch ein problematisches Kriterium ist, verdeutlicht das Äußerungsbeispiel: (58) Inge ist wahrscheinlich bei Peter. Inge ist bei Peter ist eine grammatisch interpretierbare Satzkonstruktion und eine semantisch überprüfbare Aussage. Mit dem Zusatz wahrscheinlich wird der Wahrheitsge-

1563

halt der Aussage in Frage gestellt. Aus der dem Satz entnommenen Aussage wird ein Satz mit einer Aussage, die semantisch ganz andere Schlüsse nach sich zieht. Wörter wie wahrscheinlich, vielleicht u. ä. müssen daher syntaktisch als Satzadverbiale ausgewiesen werden, so daß die entsprechende Operation damit initiiert werden kann. Es sind nicht nur Präpositionalphrasen und Adverbien, welche die beschriebene Operation bewirken können, sondern es können ebenso Konjunktionen zu einer Operation führen, die einen Satz mit einem vorhandenen Satz erweitert. (59) Inge schläft, weil sie die ganze Nacht nicht schlafen konnte. Beide Sätze sind für sich interpretierbar. Erst die Konjunktion weil leistet die Integration des weil-Satzes für den Ausgangssatz (Butzkamp 1982, 77—78). 5.3. Attribute Präpositionalphrasen sind nicht automatisch Adverbiale. Bevor sie so gedeutet werden, muß geprüft worden sein, inwieweit die vorhandene Konstruktion nicht als Attribut zu verstehen ist. Denn so, wie die Operation einen Satz in einen erweiterten Satz überführen kann, lassen sich analog Operationen für die Erweiterung der Nominal- oder Verbalphrase formulieren. Das Prinzip, Äußerungssegmente einer Kategorie durch weitere Äußerungsteile in dieselbe Kategorie zu überführen, gilt ebenso für nominale Terme (Bartsch 1977, 214). (60) Inge hat die Geburtstadt ihrer Mutter besucht. Die Phrase ihrer Mutter bliebe uninterpretiert, wenn nicht eine Operation vorhanden ist, die erkennt, daß eine genitivmarkierte und uninterpretierte Phrase neben einem nominalen Term zu einer Konstruktionseinheit verknüpft werden kann. Das gilt auch für die Äußerung (61) Inge hat die Wohnung von ihrer Mutter verlassen. Die durch das Prädikat gesteuerten Operationen interpretieren nicht die Präpositionalphrase von ihrer Mutter. Sie kann nur syntaktisch berücksichtigt werden, wenn es eine Operation gibt, die erkennt, daß von neben einem nominalen Term steht und einen erweiterten nominalen Term im Sinne eines Attributs bilden kann.

XXII. Syntax und Nachbardisziplinen

1564

An dieser Stelle ist auf den Pertinenzdativ hinzuweisen. Auch hier wird ein nominaler Term auf einen weiteren bezogen und schafft dadurch eine neue semantische Deutung des Gesamtausdrucks, wie das am Beispiel (54 ) gezeigt worden ist. Syntaktische Interpretation (i) NIMM DEN NACH DER SATZANALYSE NICHT INTERPRETIERTEN NOMINALEN TERM UND MACHE DARAUS EINEN ERWEITERTEN NOMINALEN TERM.

Diese Operation bedarf in Abhängigkeit von der jeweiligen morphologischen Realisation, wenn der uninterpretierte Term genitivmarkiert, eine von-Phrase oder dativmarkiert ist, entsprechender topologischer oder struktureller Hinweise, damit die Konstruktion angemessen erfaßt werden kann. Ein (pro)nominaler Term ist nicht nur durch nominale Phrasen in eine erweiterte Nominalphrase zu überführen, sondern kann mithilfe des Relativsatzes durch einen Satz spezifiziert werden. Das Relativpronomen kann durch eine entsprechende Operation in seiner Funktion erklärt werden. Für die folgende Äußerung gilt dann, (62) Inge, die immer verärgert ist, kommt heute. die Analyse erkennt den Satz Inge kommt heute und die immer verärgert ist. Wenn das die einen Eintrag besitzt, der es als Relativpronomen kennzeichnet, bewirkt eine entsprechende Operation das Erkennen des zweiten Satzes als Attribut zu Inge und interpretiert ihn als einen erweiterten nominalen Term. Eine solche Sichtweise des Attributs ist weniger abstrakt als der in der Sprachdidaktik übliche Umstell-und Weglaßtest. Die Umstellprobe bedeutet einen Eingriff in die Satzperspektive. Er stört die Thema-Rhema Beziehung in einer Aussage. Der Weglaßtest manipuliert die semantische Aussage eines Satzes. Ein Laie läßt nicht einfach einen Term in einer Konstruktion weg oder stellt diesen um, sondern er sucht vorrangig nach einer damit verbundenen Bedeutung oder Absicht. Das zugrundeliegende Verfahren interpretiert schrittweise alle Teile einer Äußerung, ohne die Satzperspektive zu verändern, die Illokution zu beeinflussen und die Semantik nach dem „Alles oder Nichts Prinzip“ zu manipulieren. Man könnte von einem „sanften

Testverfahren“ sprechen. 5.4. Ad-Adverb Das „attributive Konstruktionsprinzip“ gilt auch für die Modifikation von Verbalphrasen. (63) Inge ist schnell zu Peter hinübergelaufen. Das Adjektiv schnell kann nur in die Konstruktion einbezogen werden, wenn eine Operation sicherstellt, daß ein unflektiertes Adjektiv hinter dem finiten Verbteil auf besondere Weise mit der vorhandenen Konstruktion zu verknüpfen ist. Es kann sich mit dem Prädikat verbinden, wenn die Operation angewandt wird, Syntaktische Interpretation (i) MACHE AUS EINER ERKANNTEN VERBALPHRASE EINE ERWEITERTE VERBALPHRASE.

so daß gilt: Grammatische Struktur (63a) ((schnell) ((hinübergelaufen)V-INF (ist)V-FIN) VP) VP 5.5. Ad-Attribute Was für die Verbalphrasen gilt, ist auch bei Konstruktionen mit Adjektivphrasen zu beobachten. (64) Die schrecklich verliebte Inge besuchte Peter. Die Prädikatsoperationen erkennen das Äußerungsfragment Die Inge besuchte Peter. Eine Operation zur Identifikation adjektivischer Attribute würde verliebte als ein Äußerungssegment erkennen, das aus Inge ein verliebte Inge zu bilden erlaubt. Bevor jedoch diese Operation angewendet wird, ist zu prüfen, ob nicht ein Ausdruck links von einem Adjektiv steht, und ein nicht kaususmarkiertes Adjektiv oder eine Partikel ist. (65) Peter küßt die äußerst verliebte Inge. In diesem Fall muß eine Operation sicherstellen, daß weitere links stehende unflektierte Adjektive oder Partikeln den rechts stehenden Ausdruck spezifizieren können, so daß aus äußerst und verliebt ein äußerst verliebt gebildet werden kann. Eine Komplikation kann entstehen, wenn noch weitere Partikeln links aufzuspüren sind, die einer Interpretation bisher unzugänglich waren. (66) Peter küßt die stets äußerst verliebte Inge.

86.  Syntax und Sprachdidaktik

In diesem Fall zeigt sich, daß stets nicht im Sinne stets äußerst, sondern als stets verliebt bzw. stets äußerst verliebt gedeutet werden muß. Wenn daher linksseitig mehrere nicht flektierte und uninterpretierte Terme vorhanden sind, ist eine Operation nötig, die inhaltlich prüft, ob eine einfache Verknüpfung des links- mit dem rechtsseitigen Ausdruck möglich ist oder nicht. Im Gegensatz zur herkömmlichen sprachdidaktischen Sichtweise, die attributive Relationen durch das Wegstreichen zu ermitteln versucht, wird das Attribut hier als etwas erkannt, das durch Hinzufügen die Bezeichnungsleistung einer Kategorie modifiziert. Ein solches Vorgehen bewahrt im Sinne eines „sanften Testverfahrens“ die vorfindliche Struktur. 5.6. Negation Abschließend sei noch die Negation erwähnt, weil sie attributiv, adverbial und adsentential verwendet werden kann (Jacobs 1982) und bei der Sprachentwicklung eine besondere Rolle spielt (Wode 1988, 84—90; 233—237). (67) Inge liebt Peter nicht. (68) Inge liebt nicht, sondern haßt Peter. (69) Inge liebt nicht Peter, sondern hat sich in Hans verliebt. Einzelne Operationen, wie sie aus den verschiedenen attributiven Funktionen abgeleitet werden konnten, müssen auch hier angewendet werden und dadurch erklären, welche Reichweite die jeweilige Negation besitzt, und wie die Verknüpfung im einzelnen zu leisten ist, damit der semantisch erwünschte Effekt tatsächlich erreicht wird. 5.7. Natürlichsprachliche Fragmente und grammatische Deutungen Nachdem betont worden ist, daß mit diesem Ansatz keine formalgrammatischen Aussagen verfolgt werden, bleibt zu klären, wie mit natürlichsprachlichen Äußerungen umzugehen ist, die auf den ersten Blick nicht im Sinne der Grammatik beschreibbar erscheinen.

1565

soll. Es setzt daher ein „Situations- bzw. Kontext-Management“ ein, das die Äußerung als handlungsbegleitend klassifiziert und damit im Handlungskontext nach dem nicht verbalisierten Prädikat sucht, welches im „zum Abschreiben gereichten Blatt“ zu erkennen ist. Die Analyse kann die Operationen nur auf einem erschlossenen Prädikat „abschreiben“ aufbauen. Das Abschreiben bezieht sich auf ein Objekt, das ebenfalls aus dem Handlungskontext als „das gereichte Blatt“ erkannt und sprachlich aufgrund des das kodiert vorfindbar ist. Der Kontext läßt einen Imperativ erwarten, daher muß nicht nach einem Subjektterm in der sprachlichen Realisation gesucht werden. Mit dem Ausdruck ja wird ein vorhandener Sachverhalt bestätigt. Syntaktisch ist von einer Operation auszugehen, die etwas aus dem Geäußerten bzw. vorfindlichen Handlungskontext als eine Kategorie nimmt und als erweiterte Äußerung derselben Kategorie zuordnet. Der durch die Äußerung das Angesprochene gibt durch das ja kund, daß „er das Blatt abschreiben wird“ . Das Einbeziehen der Diskurswelt ist eigentlich immer notwendig, da mit der Bestimmung der Verbformen Aussagen über den Modus und das Tempus verbunden sind, so daß die Aussagen hinsichtlich einer gedachten oder vorliegenden Welt und mit Rekurs auf eine zeitlich und im Verhältnis zum Sprecher bestimmbare Dimension gedeutet werden. Die Diskurswelt wird immer mitvollzogen, wenn bei der semantischen Analyse Personalpronomen zu berücksichtigen sind. Pro-Formen deuten auf Aussagen der Äußerungsumgebung hin. Sie bedingen eigenständige semantische Operationen, die erst die Möglichkeit für eine Interpretation der im Text bzw. in der Situation befindlichen Entitäten schaffen (de Beaugrande 1981, 50—87).

6.

Kategorialgrammatik und Sprachdidaktik

6.1. Spezifische Probleme für die Didaktik der deutschen Syntax

Die Suche nach dem Prädikat und damit verbundenen Operationen scheitert, wenn die Äußerung das syntaktisch analysiert werden

Für die Sprachtheorie des Deutschen stellt sich eine Reihe von Problemen, die eine einfache Darstellungstechnik der Konstruktionsprinzipien nicht erlauben und somit für die Vermittlung derselben keine einheitliche Lösung erwarten lassen. Typisch für das Deutsche ist die teils morphologisch-katego-

1566

riale und teils strukturell-determinierte Syntax (Primus 1987, 254 —255). Vielfach zu beobachten ist ein topologischer und semantischer Einfluß, der Verunsicherung beim Erkennen von Struktureigentümlichkeiten bedingen kann. Die Syntax des Deutschen wird daher auch als eine Relations- und Positionssyntax charakterisiert (Brinker 1972, 10; Johansen 1988, 4 2). Sprachdidaktisch hat dies zur Folge, daß ein Lernangebot gemacht werden muß, das die genannten Aspekte beinhaltet und zur Auseinandersetzung mit ihnen hinführt. 6.2. Ungelöste Probleme in der Sprachdidaktik Grundsätzlich muß die Sprachdidaktik dazu anleiten, die Syntax als ein Bauprinzip anzuerkennen, das erklärt, warum Teile zueinander in Beziehung treten, und wie aus ihnen ein Ganzes entsteht. Die Behandlung der Grammatik im Unterricht und besonders der syntaktischen Eigenschaften ist gekennzeichnet durch einen sich rhythmisch wiederholenden Perspektivenwandel. Der Wechsel zwischen formalen und freien Formen grammatischer Betrachtungsweisen bestimmt die Geschichte des Grammatikunterrichts (Frank 1973; Erlinger 1969; Reuschling 1981). Dieses Hin und Her kann mit einer sprecher- oder hörerseitig ausgerichteten Sprachbetrachtung in Verbindung gebracht werden oder muß unter dem Aspekt der Förderung der produktiven, reproduktiven und analytischen Fähigkeit eines Sprachlerners gesehen werden (Felix 1977). Eine Äußerung, die vom Sprecher her gedeutet wird, zielt auf die Konstruktion von Sätzen. Die zentrale Frage ist dann das „Wie“ der Konstruktion. Steht dagegen der Hörer bzw. Sprachrezipient im Vordergrund, wird nach dem „Warum“ der Form gefragt. Lernt der Sprecher und/oder Hörer eine Sprache als Zweitsprache, kommt neben dem Wie und Warum die Frage „Was ist anders?“ hinzu. Die Perspektive des Konstruierens läßt die formalen Gesichtspunkte in den Vordergrund treten. Wenn die Hörerseite eingenommen wird, interessieren die mit der Äußerung jeweils verbundenen semantischen Aussagen und ihre pragmatischen Intentionen. Wird eine Zweitsprache erworben, dann tritt neben dem Problem von Kodierung und Dekodierung das Problem der Beherrschbarkeit auf. Es ist daher nicht zufällig, daß der „freie Grammatikunterricht“ stets mit Textbetrachtungen verbunden wird, während sich der

XXII. Syntax und Nachbardisziplinen

formalgrammatische Unterricht an Satzmodellen orientiert und mit graphischen Illustrationen zum Satz auseinandersetzt (Sowinski 1969; Reuschling 1982; Sucharowski 1979). Der Zweitsprachenerwerb hingegen betont die Notwendigkeit einer Methodenvielfalt. Der analytische Umgang mit Sprache wird sowohl von den produktiven als auch von den reproduktiven Spracherfahrungen geleitet. Es liegt nahe, daß muttersprachlicher Unterricht eher hörerseitig ausgerichtet ist, und daß der Zweitsprachenerwerb die Sprecherseite mehr im Auge hat. Dabei muß mitbedacht werden, daß dem analytischen Umgang mit Sprache wenig Einfluß auf eine Verbesserung der Sprachproduktion beigemessen wird (Siebenbrodt 1966; Maas 1975; Spies 1989). 6.3. Verhältnisbestimmung zwischen Kategorialgrammatik und Sprachdidaktik Mit den Beschreibungsansätzen der Kategorialsyntax ist ein Bezugssystem vorhanden, das theoretisch zuverlässig, praktisch nachvollziehbar und intuitiv zugänglich ist (Buszkowski 1988). Es bietet eine sinnvolle Basis für die im Sprachunterricht notwendige produktive und reproduktive Perspektive. Die schrittweise Entwicklung der Operationsanweisungen von einzelnen Wörtern bzw. Wortformen her erlaubt dem Sprachbenutzer, sich zugrundeliegende Regelhaftigkeiten bewußt zu machen und in einer ihm verständlichen Weise zu formulieren. Durch den semantischen Hintergrund syntaktischer Beschreibungen wird ein Zugang zu inhaltlich definierten Aufbauprinzipien gewonnen. Die Praktiken und Erfahrungen aus dem kommunikationsgeleiteten Sprachunterricht (Boettcher 1976) stehen zu einem kategorialgrammatischen Ansatz nicht im Widerspruch. Doch ist dieser naturgemäß enger gefaßt, weil er sich auf die syntaktischen Gegebenheiten einer Sprache bezieht. Das bedeutet aber nicht die Ausklammerung semantischer und pragmatischer Faktoren. Sie besitzen einen klar erkennbaren Platz bei der Sprachanalyse und -beschreibung. Sprachdidaktisch wird mit dem vorgeschlagenen Konzept ein Weg gewonnen, der die morphologische, strukturelle, semantische, topologische und pragmatische Ebene nicht voneinander isoliert. Der sprachwissenschaftliche Beschreibungsansatz erlaubt, die sprachlichen Faktoren einerseits frei beobachtend und zugleich theoriegeleitet zu be-

86.  Syntax und Sprachdidaktik

schreiben, eines einengenden Formalismus bedarf es nicht. Er ist im Grundsätzlichen einfach zu handhaben, ohne dabei die linguistische Komplexität zu ignorieren. Mit der zugrundeliegenden Theorie kann Intuition über Sprache freigesetzt werden (OomenWelke 1982, 7). Die beschriebene Vorgehensweise eröffnet kontrastiven Sichtweisen den Weg (Sroka 1988) und zwingt zur Explikation impliziten Sprachwissens. Alle nachfolgend genannten Sprachen beispielsweise zeigen einen vergleichbaren Umgang mit dem Subjekt (Fanselow/Felix 1987, 72—74). (71) Er liebt Inge. (Deutsch) (72) He loves Inge. (Englisch) (73) Il aime Inge. (Französisch) (74) On ljubit Inge. (Russisch) Problematisch wäre der Schluß, damit sei das Subjekt in seinen Eigenschaften schlechthin erkannt. Das Lateinische (75a) Gallos frequenter bellum gerere Krieg führen Die Gallier häufig putat. glaubt-er (75b) Galli frequenter bellum gerunt. zeigt das Subjekt eines infiniten Satzes im Akkusativ, obwohl es auch das Subjekt im Nominativ besitzt. Das Georgische nutzt zur Realisation von Subjekt drei Kasus: den Nominativ (76a), den Ergativ (76b) und den Dativ (77a). (76) (76a) (76b) (77) (77a)

Mutter malt mir ein Bild. deda-c mi-xat’av-s (me) surat-s. deda-m da-mi-xat’-a me surat-i Mutter malte mir kein Bild. deda-s ar da-u-xat’av-s der Mutter nicht sie gemalt hat cem-tvis surat-id mich für Bild

Die vorgeschlagene Betrachtungsweise bewirkt Sensibilität für semantische Relationen und öffnet den Blick auf Diskurseigenschaften einzelner Äußerungsteile (Henrici 1979, 78). (78) Inge war die ganze Nacht bei mir, weil sie sich nur so Einsicht in alles bisher Gewesene erhofft hatte. Doch sie wurde enttäuscht. Um eine solche Äußerung zu verstehen, bedarf es nicht nur syntaktisch motivierter Operationen, sondern darüber hinaus sind Angaben nötig, wie einzelne Wörter auf vorausge-

1567

gangene bezogen werden, um semantisch interpretiert werden zu können, bzw. wo außerhalb des Geäußerten liegende Referenzpunkte für die Kommunikationspartner überhaupt zu finden sind. Regeln werden durch das Beobachten der Wirkweisen von Operationen zur Verknüpfung entwickelt. Diese legen dar, wie aus einzelnen Wörtern zusammenhängende Phrasen mit Funktionen entstehen, die sich syntaktisch, semantisch oder pragmatisch bewerten lassen. vielleicht gehört zu den Wörtern, die syntaktisch „einen Satz nehmen und einen Satz machen“ , die semantisch eine Aussage als nicht tatsächlich, sondern nur möglich klassifizieren und die pragmatisch Routinen aktivieren können, die etwas über das Verhältnis des Sprechers zum geäußerten Gegenstand aussagen (Wunderlich 1976, 73). Mit der Unterscheidung zwischen „syntaktischer Operation“ (79a) und „syntaktischer Regel“ (79b) (79a) NIMM KATEGORIEN DER EINGABE UND

(79b)

MACHE EINE KATEGORIE FÜR DIE AUSGABE, DIE DAS VERKNÜPFUNGSERGEBNIS ANZEIGT. WENN A AUSDRUCK EINER BESTIMMTEN KATEGORIE, UND WENN B AUSDRUCK EINER BESTIMMTEN KATEGORIE, DANN GELTEN BEI DER VERKNÜPFUNG FÜR A UND B DIE MORPHOLOGISCHEN UND TOPOLOGISCHEN BEDINGUNGEN, WIE SIE IN (I)—(I + N) DEFINIERT SIND.

wird über die Angabe von Verknüpfungen hinaus auf Besonderheiten der Serialisierung hingewiesen, wie sie in den jeweils zu beobachtenden Sprachen gelten. Der Unterschied zwischen sprachspezifischer und universeller Information kann so auf besondere Weise deutlich gemacht werden (Dowty 1982, 117). 6.4. Sprachdidaktisch offene Fragen und kategorialgrammatische Antworten Die Spezifikationsrelation ist das Kernstück für eine sprachdidaktische Rezeption. Wenn nämlich unabhängig von der Äußerungsabfolge die Teile in ihrer spezifizierenden Wirkung aufeinander betrachtet werden, entsteht eine Operator-Operand-Beziehung, womit eine uralte sprachdidaktische Idee aufgegriffen wird. Die Satzbeschreibung sollte durch die Zweierteilung sog. logische Beziehungen im Satz aufdecken (Becker 1827; 1829; 1831). Ein weiterer, wichtiger Anknüpfungspunkt an die bisherige Sprachdidaktik ist die Explikation der Dependenzrelationen. Doch an-

1568

ders als in der Dependenzgrammatik (Heringer 1970; 1972; 1978), wo Satzglieder gleichgewichtig nebeneinanderstehen und allein aus dem Verb abgeleitet werden, verweist die kategorialgrammatische Darstellung auf die syntaktischen Besonderheiten der einzelnen Satzglieder, wie sie natürlichsprachlich empfunden werden, ohne die Wertigkeit der Verben außer acht zu lassen. Darüber hinaus gerät die Konstituentenstrukturdarstellung nicht in Widersprüchlichkeiten bei der Behandlung attributiver und adverbialer Relationen, wie das bei der Dependenzgrammatik der Fall ist (Vennemann/Jacobs 1982, 99). Unbestritten war für die Sprachdidaktik die Möglichkeit des operationalen Zugriffs auf die Sprache ein Zugang zu syntaktischen Strukturen (Aßheuer/Hartig 1976; Henrici 1979; Oomen-Welke 1982). Dieser Weg beschränkte sich indes auf das Identifizieren einzelner Satzglieder, ohne Konstruktionen in ihrer Funktion erklären zu können. Kategorialgrammatisch werden über die Definition von Operationen Regeln entwickelt, die Konstruktionen zu klassifizieren ermöglichen und zugleich ihr Aufbaukonzept interpretieren können. Dabei wird einem Denkansatz gefolgt, der konstruktiv-konventionell ist, weil mit den Operationen „Regeln“ für einzelne Wörter bzw. Wortklassen definiert werden, die Aussagen darüber machen, wie sich Wörter in syntaktischen Konstruktionen verhalten. Damit bewahrt der Ansatz den Anschluß an die Tradition des formalen Grammatikunterrichts. Zugleich ist er offener Sprachunterricht, weil er interpretativ angelegt ist. Es gibt kein geschlossenes System von Definitionen grammatischer Kategorien, Regeln werden heuristisch formuliert und eröffnen einen von der Theorie vorgegebenen Weg für die semantische und pragmatische Dimension. Wichtig ist, daß der linguistische Bezug zu keiner Zeit verwischt oder gar aufgegeben wird. Es ist ein Konzept, das sich auf andere Sprachen anwenden läßt und Hinweise bieten kann, wo und wie Relationen in der fremden Sprache anders gesehen werden. Der Betrachtungs- und Beschreibungsansatz fordert kontrastiv zu Schlüssen über Besonderheiten von Sprachen heraus. Da er Bauprinzipien vom einzelnen Wort her definiert, entzieht er sich der Gefahr des formalen Grammatikunterrichts, abstrakt Konstruktionsprinzipien im Sinne von Mustern lernen zu lassen. Syntaktische Kom-

XXII. Syntax und Nachbardisziplinen

petenz entwickelt sich immer mit und im Bezug zum Wortschatzwissen. Die im Ansatz implizit enthaltene semantische Perspektive bewirkt eine erhöhte Sensibilität für diskursgrammatische und pragmatische Phänomene, weil Kontext und Situation die Bezeichnungsleistung einer Konstruktion im Hinblick auf das, was gesagt werden soll, und das, was gesagt wird, beeinflussen. 6.5. Lernerspezifische Aspekte Die Personengruppen, die sich mit den Inhalten der Syntax im Unterricht auseinanderzusetzen haben, sind Schüler im muttersprachlichen Unterricht der Primar- und Sekundarstufe sowie die Personen, die Deutsch als Zweitsprache im Unterricht lernen. Der Fremdsprachenunterricht in der Sekundarstufe erwartet Kenntnisse in der grammatischen Terminologie (Raasch 1983). Unabhängig von der Sinnfrage, ob und inwieweit grammatisches Wissen ein Indiz für Sprachkompetenz bedeuten kann, verlangen die Lehrpläne Wissen über Satzglieder und Satzbaupläne des Deutschen sowie die grundlegenden Wortarten und die Fertigkeit ihrer Bestimmung (Reuschling 1982; Erlinger 1986; Knobloch 1986). Vorschriften hinsichtlich der Wahl eines bestimmten theoretischen Ansatzes gibt es nicht. Empfohlen werden Ansätze, die die Darstellung der Dependenzrelation gewährleisten. Für den Primarstufenunterricht wird die Elementarisierbarkeit grammatischer Aussagen erwartet. Im Unterricht der Sekundarstufe geht es vornehmlich um Erweiterungen des grammatischen Wissens zur Unterstützung des Fremdsprachenunterrichts. Die Gruppe derer, die Deutsch als Zweitsprache lernt, ist auf eine Grammatik angewiesen, die Funktionsweisen im Vergleich zur eigenen Sprache erklärt, ohne jeweils ein Theoriekonzept vorauszusetzen. Die kategorialgrammatischen Beschreibungen sind von sich aus elementar. Sie basieren auf den aus der Wortartenlehre vertrauten Kategorien. Strukturelle Komplexität entsteht dann, wenn aus den Elementaraussagen weiterreichendere Konstruktionszusammenhänge benannt werden. Sie bleiben erfahrungs- und einzelfallbezogen, wie dies zum Verständnis einer Konstruktion nötig ist. Verallgemeinerungen, die zugleich den Einstieg in die Sprachtheorie bedeuten können, erfolgen in Abhängigkeit zur Gruppe der Lernenden, zu den die Sprachlernsituation konstituierenden Faktoren wie dem Sprach-

86.  Syntax und Sprachdidaktik

erwerbstyp oder der Altersgruppe und mit Rücksicht auf jeweils angestrebte Lernziele. Mit dem Grundsatz, Sprache in ihrer konkret existierenden Form zu beobachten, über ihre Eigentümlichkeiten hypothetisch zu sprechen, Kategorien offen zu definieren und flexibel anzuwenden und das alles in enger Bindung an die Semantik sowie unter Einbeziehung diskursgrammatischer und pragmatischer Faktoren zu realisieren, gewinnt der Sprachunterricht eine pädagogisch bedeutsame Perspektive, die durch einen Aphororismus im „Der Baum der Erkenntnis“ von Maturana/Varela (1984 ) wie folgt charakterisiert wird: „Jedes Tun ist Erkennen, und jedes Erkennen ist Tun“ .

7.

Literatur

Abels, Kurt et al. (Hrsg.) 1978. Sprachunterricht. Bad Heilbrunn. Abraham, Werner. (Hrsg.) 1982. Satzglieder. Tübingen. Adelung, Johann Chr. 1781. Deutsche Sprachlehre. Berlin. Ader, Dorothea. 1983. Grammatik als Sprachförderung, Handbuch Deutschunterricht, Bd. 1, hrsg. v. Peter Braun & Dieter Krallmann. Düsseldorf. —, und Axel Kress. 1978. Zum gegenwärtigen Stand der Sprachdidaktik. Sprachunterricht, hrsg. v. Kurt Abel u. a., 237—62. Bad Heilbrunn. Ajdukiewicz, Kazimiers. 1935. Die syntaktische Konnexität. Studia philosophica I. 1—27. Admoni, Wladimir. 1970. Der deutsche Sprachbau. München. Albertus, Laurentius. 1573. Teutsch Grammatick oder Sprach Kunst. Augusta Vindelicorum. Arens, Hans. 1969. Sprachwissenschaft. Der Gang ihrer Entwicklung von der Antike bis zur Gegenwart, Bd. 1 und Bd. 2. Frankfurt a. M. Asher, J. J. 1977. Learning another language through actions. Los Gatos, Cal. Aßheuer, Johannes, und Matthias Hartig. 1976. Aufbau einer Schulgrammatik auf der Primar- und Sekundarstufe. Düsseldorf. Äugst, Gerhard. 1976. Welchen Sinn hat der Grammatikunterricht in der Schule? Diskussion Deutsch. 7. 211—27. B ach, Emmon. 1983. On the relationship between word-grammar and phrase-grammar. Natural language and linguistic theory 1. 65—89. B alhorn, Heiko. 1981. Didaktische Aspekte zum Lernbereich „Sprache“ . Sprachunterricht, hrsg. v. Heiko Balhorn et al., 11—52. München. B artsch, Renate. 1977. Kategorialgrammatik. Einführung in die Syntax, hrsg. v. Renate Bartsch. 205—71. Kronberg.

1569

—. 1975. Subcategorization of adnominal and adverbial modifiers. Formal semantics of natural language, ed. by Edward L. Keenan, 175—87. London. B ar-Hillel, Yehosua. 1964 . Language and information. Reading MA. B ausch, Karl-Richard. (Hrsg.) 1979. Beiträge zur didaktischen Grammatik. Königstein. —, et al. (Hrsg.) 1989. Handbuch Fremdsprachenunterricht. Tübingen. —, und Frank G. Königs.(eds.) 1986. Sprachlehrforschung in der Diskussion. Tübingen. B ecker, Karl F. 1827. Organism der Sprache als Einleitung zur deutschen Grammatik. Frankfurt a. M. —. 1829. Deutsche Grammatik. Frankfurt a. M. —. 1831. Schulgrammatik der deutschen Sprache. Frankfurt a. M. —. 1833. Über die Methode des Unterrichts in der deutschen Sprache. Frankfurt a. M. B enkö, L., and S. Imre. (eds.) 1972. The Hungarian language. The Hague. B ever, T. G. 1970. The cognitive basis for linguistic structures. Cognition and the development of language, ed. by J. R. Hayes, 279—362. New York. Bildungsplan Fach Deutsch an den Gymnasien Hessens vom 19. 2. 1969. 1969. Hrsg. v. Kultusministerium Hessen. Wiesbaden. B ödiker, Johann. 1690. Grund-Sätze Der Deutschen Sprachen im Reden und Schreiben. Berlin. B oettcher, Wolfgang, und Horst Sitta. 1983. Grammatik in Situationen. Handbuch Deutschunterricht. Bd. 1. Sprachdidaktik, hrsg. v. Peter Braun & Dieter Krallmann, 378—98. Düsseldorf. —. 1978. Der andere Grammatikunterricht. München. B raun, Peter, und Dieter Krallmann. 1983. Sprachunterricht nach 194 5 — Geschichte, Veränderung, Bestand. Handbuch Deutschunterricht. Bd. 1. Sprachdidaktik, hrsg. v. Peter Braun & Dieter Krallmann, 3—19. Düsseldorf. —, und —. (Hrsg.) 1983. Handbuch Deutschunterricht. Bd. 1. Sprachdidaktik. Düsseldorf. B raunroth, Manfred. 1984 . Zur Geschichte und Situation der Sprachdidaktik. Diskussion Deutsch 15. 255—70. B rekle, Wolfgang. 1971. Ist das Erfragen von Satzgliedern im Grammatikunterricht zu empfehlen? Deutschunterricht, 484—89. Berlin. Bremer Kollektiv. 1974 . Grundriß einer Didaktik und Methodik des Deutschunterrichts in der Sekundarstufe I und II. Stuttgart. B resnan, Joan A. 1978. Realistic transformational grammar. Linguistic theory and psychological reality, ed. by Morris Halle & G. A. Miller, 1—59. Cambridge, MA.

1570

—. (ed.) 1982. The mental representation of grammatical relations. Cambridge, MA. B rinker, Klaus. 1972. Konstituentenstrukturgrammatik und operationale Satzgliedanalyse, Frankfurt a. M. B rinkmann, Hennig. 1962. Die deutsche Sprache. Gestalt und Leistung. Zweite, revidierte Auflage. Düsseldorf. B rugmann, Karl, und Hermann Osthoff. 1878. Morphologische Untersuchungen auf dem Gebiet der indogermanischen Sprachen. Leipzig. B rünner, Gisela. 1982. „Wer oder was kennst du?“ . Probleme des Grammatikunterrichts in der Grundschule. Sprache beschreiben und erklären, hrsg. v. Klaus Detering et al. 136—46. Tübingen. B unting, Karl-Dieter. 1970. Wissenschaftliche und pädagogische Grammatik. Linguistische Berichte 5. 73—82. —, und Detlef C. Kochan. 1973. Linguistik und Deutschunterricht. Kronberg Ts. B urstall, C. u. a. Primary French in the balance. Slough: NFER 1974. B uszkowski, Wojciech et al. (eds.) 1988. Categorial grammar. Amsterdam. B utzkamp, Wolfgang. 1980. Praxis und Theorie der bilingualen Methode. Heidelberg. —. 1985. Natürliche Erwerbssituationen als Bezugspunkte für die Sprachlehrmethodik. Neusprachliche Mitteilungen aus Wissenschaft und Praxis 1. 5—11. —. 1981. Rationaler und ratiomorpher Grammatikerwerb. Linguistik und Didaktik 12.4 5/4 6. 4 9— 64. —. 1982. Zur Vermittlung von Weil-Sätzen: Problemskizze anhand einer Unterrichtsanalyse. Deutsch lernen 4. 71—81. Chonsky, Noam. 1965. Aspects of the theory of Syntax. Cambridge Mass. Clahsen, H. 1981. Der Erwerb des Syntax in der frühen Kindheit. Wuppertal: Dissertation 1981. Comenius, Johann Amos. Janua Linguarum. Große Didaktik. In neuer Übersetzung hrsg. v. Andreas Flintner. 1954 . Pädagogische Texte, ed. by Wilhelm Flintner. Düsseldorf. —. 1658. Orbis sensualium pictus. Nachdruck der Erstausgabe. Dortmund 1978. de B eaugrande, Robert A. und Wolfgang U. Dressler. 1981. Einführung in die Textlinguistik. Tübingen. Donatus. Ars grammatica. Grammatici Latini, ed. by H. Keil. Bd. 1—4. Leipzig, 1864. Doderer, Klaus. 1965. Wege in die Welt der Sprache. Stuttgart. Dowty, D. R. 1981. More on the categorial analysis of grammatical relations. Subjects and other subjects. Proceedings of the Harvard Conference on

XXII. Syntax und Nachbardisziplinen

the Representation of Grammatical Relations, ed. by A. Zaenen, 115—53. Indiana University Linguistics Club. Bloomington. Drach, Erich. 1937. Grundgedanken der deutschen Satzlehre. Frankfurt a. M. Ebert, Robert, P. 1976. Historische Syntax des Deutschen. Stuttgart. —. 1986. Historische Syntax des Deutschen II: 1300—1759. Bern. Edelstein, W., and W. Schäfer. 1969. Unterrichtsziele im Sprachunterricht in der differenzierten Gesamtschule. Deutscher Bildungsrat: Gutachten und Studien der Bildungskommission 12. Stuttgart. Eichler, Wolfgang. 1972. Zum Aufbau eines für die Schulpraxis geeigneten Grammatikmodells. Der Deutschunterricht 24.3. 69—97. —, und Karl-Dieter B unting. 1978. Deutsche Grammatik. Form, Leistung und Gebrauch der Gegenwartssprache, zweite revid. Auflage. Kronberg/Ts. Endres, Rolf. 1976. Subjekt und Subjektlosigkeit als Kategorien der Inhaltsseite der Sprache. Sprachwissenschaft 1. 292—327. Engel, Ulrich. 1972. Regeln zur „Satzgliedfolge“ . Zur Stellung der Elemente im einfachen Verbalsatz. Sprache der Gegenwart. 17—75. Düsseldorf. —. 1972. Syntaktische Besonderheiten der deutschen Alltagssprache. Sprache der Gegenwart. 199—228. Düsseldorf. —. 1977. Syntax der deutschen Gegenwartssprache. Berlin. —, und Helmut Schumacher. 1978. Kleines Valenzlexikon deutscher Verben. Zweite Auflage. Tübingen. Engelen, B ernhard. 1979. Aufbauprinzipien wissenschaftlicher Grammatiken und der Aufbau des Grammatikunterrichts. Sprachdidaktik. Neue Perspektiven und Unterrichtsvorschläge, hrsg. v. Gewehr & Wolf. Düsseldorf. —. 1975. Überlegungen zum Syntaxunterricht auf der Grundlage eines Dependenzmodells. Deutschunterricht 27.6. 19—32. —. 1975. Untersuchungen zu Satzbauplan und Wortfeld in der geschriebenen deutschen Sprache der Gegenwart. München. Erben, Johannes. 1984 . Deutsche Syntax. Eine Einführung. Bern. —. 1960. Abriß der deutschen Grammatik. München. Erlinger, Hans D. 1969. Sprachwissenschaft und Schulgrammatik. Strukturen und Ergebnisse von 1900 bis zur Gegenwart. Düsseldorf. —. 1986. Begründungen für Grammatikunterricht: einst und heute. Der Deutschunterricht 38.2. 104—16. —. 1988. Studienbuch: Grammatikunterricht. Paderborn.

86.  Syntax und Sprachdidaktik

Eroms, Hans W. 1981. Valenz, Kasus und Präpositionen. Heidelberg. —. 1978. Zur Konversion der Dativphrasen. Sprachwissenschaft 3. 357—405. Essen, Erika. 1962. Methodik des Deutschunterrichts. Heidelberg. Fanselow, Gisbert, und Sascha W. Felix. 1990. Sprachtheorie 1. Grundlagen und Zielsetzungen. Zweite Auflage. Tübingen. —, und —. 1987. Sprachtheorie 2. Die Rektionsund Bindungstheorie. Tübingen. Felix, Sascha. 1977. Kreative und reproduktive Kompetenz im Zweitspracherwerb. Neuere Perspektiven der Fremdsprachendidaktik, hrsg. v. Hans Hunfeld, 25—33. Kronberg Ts. —. 1982. Psycholinguistische Aspekte des Zweitsprachenerwerbs. Tübingen. Fillmore, Charles J. 1968. The case for case. Universals in linguistic theory, ed. by Bach, Emmon & J. Harris, 1—90. New York. —. 1977. The case for case reopened. Grammatical relations (Syntax and Semantics 8), ed. by Peter Cole & J. M. Sadock, 59—81. New York. Fink, S. R. 1977. Aspects of pedagogical grammar based on case grammar and valenc theory. Tübingen. Florstedt, F., und W. Stieber. 1935. Neue deutsche Sprachlehre. Frankfurt. Fourquet, Jean. 1970. Prolegomena zu einer deutschen Grammatik. Düsseldorf. Frank, Horst J. 1976. Dichtung, Sprache, Menschenbildung. Geschichte des Deutschunterrichts von den Anfängen bis 1945. Bd. 1—2. München. Frey, Emmy. 1966. Lage und Möglichkeiten der Schul- und Volksgrammatik. Der Deutschunterricht 18. 5—46. Garbe, B urckhard. (Hrsg.) 1978. Die deutsche rechtschreibung und ihre reform 1722—1974 . Tübingen. Gaiser, Konrad. 1950. Wieviel Grammatik braucht der Mensch? Die Pädagogische Provinz. Monatsschrift für Erziehung und Unterricht 4. 590—99. Geißler, Heinrich. 1959. Comenius und die Sprache. Pädagogische Forschungen. Heidelberg. Gewehr, Wolf. 1979. Brauchen wir eine neue Grammatikkonzeption? Überlegungen zur Neugestaltung des Grammatikunterrichts. Sprachdidaktik. Neue Perspektiven und Unterrichtsvorschläge, hrsg. v. Wolf Gewehr, 40—59. Düsseldorf. Gleason, Henry A. 1965. Linguistics and English grammar. New York. Glinz, Hans. 194 7. Geschichte und Kritik der Lehre von den Satzgliedern. Bern. —. 1952. Die innere Form des Deutschen. Bern. —. 1956—57. Die Begründung der abendländi-

1571

schen Grammatik durch die Griechen und ihr Verhältnis zur modernen Sprachwissenschaft. Wirkendes Wort 7. 129—35. —. 1961. Der Deutschunterricht im engeren Sinne oder Sprachlehre und Sprachkunde. Handbuch des Deutschunterrichts. Teil 4 , hrsg. v. August Beinlich. Emsdetten, Westf. —. 1961. Der deutsche Satz. Zweite Auflage. Düsseldorf. —. 1970—71. Deutsche Grammatik. 2 Bd. Bad Homburg. Gottsched, Johann C. 174 8. Grundlegung der Deutschen Sprachkunst. Nach den Mustern der besten Schriftsteller des vorigen und jetzigen Jahrhunderts. Leipzig. —. 1777. Kern der größeren Deutschen Sprachkunst. Leipzig. Graucob, Karl. 1970. Der muttersprachliche Unterricht in der Grundschule. Siebente Auflage. Kiel. Grebe, Paul. 1973. Duden. Grammatik der deutschen Gegenwartssprache. Dritte neubearb. und erweiterte Auflage. Mannheim. Gueintz, Christian. 1663. Deutsche Sprachlehre Entwurf. Hildesheim 1978. Heinimann, S. 1963. Zur Geschichte der grammatischen Terminologie im Mittelalter. Zeitschrift für romanische Philologie 79. 23—37. Hamburger Autorenkollektiv. 1975. Sprachunterricht gleich Linguistik? Zur Kritik des linguistisierten Sprachunterrichts. Zur Praxis des Deutschunterrichts 4 , hrsg. v. Susanne Glauber et al. Stuttgart. Harris, James. 1751. Hermes, or a philosophical inquiry concerning universal grammar (= übersetzt von Chr. G. Ewerbeck, Halle 1788). London. Harsdörffer, Georg Ph. 1651. Die Teutsche Sprache. Heinz Zirnbauer. Philobiblon 5, 1961. 12—49. Hartmann, Peter. 1959. Die Sprachbetrachtung bei Leo Weisgerber. System und Kritik. Der Deutschunterricht 11.1. 104—24. Hartmann, Wilfried. 1969. Unterrichtsbeispiel zur Arbeit mit der generativen Grammatik im Deutschunterricht. Wirkendes Wort 19. 289—310. Hartnacke, Wilhelm. 1918. Deutsche Sprachlehre im Sinne der Selbsttätigkeit — und im Dienst der Sprachsicherheit. Wegweiser und Stoffsammlung für den deutschen Sprachunterricht aller Schulgattungen. Leipzig. Harweg, Roland. 1971. Subjekt und Prädikat. Folia Linguistica V. 253—76. Hayes, John R. (ed.) 1970. Cognition and the development of language. New York. Helbig, Gerhard (Hrsg.) 1978. Beiträge zu Problemen der Satzglieder. The Hague, Paris. —. (ed.) 1971. Beiträge zur Valenztheorie. The Hague.

1572

—, und Joachim B uscha. 1972. Deutsche Grammatik. Ein Handbuch für den Ausländerunterricht. Leipzig. —, und Wolfgang Schenkel. 1969. Wörterbuch zur Valenz und Distribution deutscher Verben. Leipzig. Helmers, Hermann. 1966. Didaktik der Deutschen Sprache. Stuttgart. Henrici, Gert. 1979. Sprachunterricht in der Lehrerausbildung. Theorie und Praxis einer PädagoLinguistik. Weinheim. Herder, Johann G. 1766—67. Über die neuere Deutsche Literatur. Erste Sammlung von Fragmenten, Sämtliche Werke, hrsg. v. B. Suphan. Bd. 1. Berlin 1877. Heringer, Hans-Jürgen. 1968. Präpositionale Ergänzungsbestimmungen im Deutschen. Zeitschrift für deutsche Philologie 87. 426—57. —. 1970. Zur Analyse von Sätzen des Deutschen auf der Unterstufe. Linguistik und Didaktik 1. 2—28. —. 1970. Theorie der deutschen Syntax. München. —. 1972. Deutsche Syntax. Zweite völlig neubear. Auflage. Berlin. —. 1978. Wort für Wort. Interpretation und Grammatik. Stuttgart. —. 1988. Lesen lehren lernen: eine rezeptive Grammatik des Deutschen. Tübingen. Herrlitz, Wolfgang. 1975. Grammatikunterricht in der Schule? Diskussion der Probleme, hrsg. v. Bernd Sowinski, 167—82. Fachdidaktik Deutsch. Köln. Heuer, Knut. 1977. Das Verb im Grammatikunterricht der Orientierungsstufe. Deutschunterricht 28.3. 64—73. Hiecke, Robert H. 184 2. Der deutsche Unterricht auf deutschen Gymnasien. Ein pädagogischer Versuch. Leipzig. Hiersche, Rolf. 1979. Zur deutschen Satzgliedlehre. Sprachwissenschaft 4. 233—53. Hohmann, Heinz-Otto. 1990. Fremdsprachenunterricht und Methodik. Neusprachliche Mitteilungen 1990, 17—24. Homberger, Dietrich. 1977. Sprachbuch. Sekundarstufe I. Mitteilungen 4, 34—8. Hoppenkamps, Hermann et al. 1978. Sprachdidaktiken in der Diskussion — Kriterien ihrer Beurteilung. Linguistik und Didaktik 33. 1—42. Hundsnurscher, Franz et al. 1970. TSG — Transformationelle Schulgrammatik. Erster Versuch. Zweite durchgesehene Auflage. Göppingen. Ickelsamer, Valentinus. 1531. Ein Teutsche Grammatica. Vier seltene Schriften des 16. Jahrhunderts, hrsg. v. Heinrich Fechner. Hildesheim. Ingendahl, Werner. 1975. Sprechen und Schreiben. Heidelberg.

XXII. Syntax und Nachbardisziplinen

Jackendorf, Ray S. 1983. Syntax and cognition. Cambridge, MA. Jacobs, Joachim. 1982. Syntax und Semantik der Negation im Deutschen. München. Johansen, Ingeborg. 1988. Der heterogene deutsche Dativ. 1988 Zur Syntax, Semantik und Sprachgebrauchsbedeutung. Heidelberg. Johnson, David E. 1977. On Keenan’s definition of „subject of“ . Linguistic Inquiry 8. 673—92. Joos, M. 1966. Language and the school child. Language, and learning, ed. by J. A. Emig et al. New York. Kaneko, Thhru. 1978. Wozu Satzglieder? Beiträge zur Valenztheorie, hrsg. v. Gerhard Helbig, 105— 28. The Hague. Kaper, W. 1959. Kindersprachforschung mit Hilfe des Kindes. Groningen. Keenan, Edward L. (ed) 1975. Formal semantics of natural language. London. —. 1976. Towards a universal definition of „subject“ . Subject and topic, ed. by Charles N. Li, 303— 34. New York. Keim, Inken. 1978. Gastarbeiterdeutsch. Untersuchungen zum sprachlichen Verhalten türkischer Gastarbeiter. Tübingen. Kerkhoff, Lothar van den. 1983. Zum Verhältnis von Sprachtheorie und Sprachdidaktik. Eine kritische Untersuchung neuerer sprachdidaktischer Konzepte und ihrer theoretischen Grundlagen. Frankfurt a. M. Kern, F. 1883. Die deutsche Satzlehre. Berlin. Kleineidam, Hartmut 1989. Grammatiken. Bausch u. a. (Hrsg.): Handbuch. 1989, 246—49. Klein, Klaus-Peter. 1979. Didaktische Aspekte einer Theorie des sprachlichen Handelns. Sprachdidaktik. Neue Perspektiven und Unterrichtsvorschläge, hrsg. v. Wolf Gewehr, 13—26. Düsseldorf. Klein, Wolfgang. 1984 . Sprachdidaktische Konzeptionen und ihre Auswirkungen auf die Gliederung von Sprachbüchern für die Grundschule. Diskussion Deutsch 15.75. 271—85. Knapp-Potthoff, A. und K. Knapp. 1982. Fremdsprachenlernen und -lehren. Syntax und Reihenfolge beim Spracherwerb. Stuttgart. Knobloch, Clemens. 1986. Wortarten in der Schulgrammatik? Probleme und Vorschläge. Der Deutschunterricht 38.2. 37—49. Koch, W. 1978. Kasus-Kognition-Kausalität. Zur semantischen Analyse der instrumentalen „mit“ -Phrase. Lund. Krause, Karl H. 1817/18. Versuch eines Lehrbuchs der deutschen Sprache. Halle. —. 1813. Versuch planmäßiger und naturgemäßer Denkübungen für Elementarschulen. Halle. Lakoff, George. 1970. Global rules. Language 4 6, 3. 627—39.

86.  Syntax und Sprachdidaktik

Lenerz, Jürgen. 1977. Zur Abfolge nominaler Satzglieder im Deutschen. Tübingen. —. 1984 . Syntaktischer Wandel und Grammatiktheorie: eine Untersuchung aus der Sprachgeschichte des Deutschen. Tübingen. Leopold, W. 194 7—4 9. Speech development of a bilingual child. Vol. I—IV. Evanston. Lerot, Jaques. 1982. Die verbregierten Präpositionen in Präpositionalobjekten. Satzglieder, hrsg. v. Werner Abraham, 261—91. Tübingen. Li, Charles N. (ed.) 1976. Subject and topic. New York. —, and S. A. Thompson. 1976. Subject and topic: A new typology of language. Subject and topic, ed. by Charles Li, 457—90. New York. Lyon, Otto. 1890. Historische und gesetzgebende Grammatik. Programm der Ammerschule zu Dresden-Altstadt. Dresden. Maas, Utz. 1976. Kann man Sprache lehren? Für einen anderen Sprachunterricht. Frankfurt a. M. Margarjan, A. A. 1978. Zum Problem des grammatischen Subjekts in der deutschen Gegenwartssprache. Deutsch als Fremdsprache 5. 291—97. Matzel, Klaus. 1976. Dativ und Präpositionalphrase. Sprachwissenschaft 1. 144—86. Meiner, Johann W. 1781. Versuch einer an der menschlichen Sprache abgebildeten Vernunftlehre oder Philosophie und allgemeine Sprachlehre. Leipzig. Mensing, Otto. 1903. Deutsche Sprachlehre für höhere Schulen. Dresden. Menzel, Wolfgang. 1975. Die deutsche Schulgrammatik. Kritik und Ansätze zur Neukonzeption. Paderborn. Meyer-Ingwersen, Johannes et al. 1977. Zur Sprachentwicklung türkischer Schüler in der Bundesrepublik. Zwei Bd. Kronberg Ts. Moritz, Karl Ph. 1793. Vorlesungen über den Styl oder praktische Anweisung zu einer guten Schreibart. Berlin. Müller, Lotte. 1921. Vom Deutschunterricht in der Arbeitsschule. Leipzig. Müller-Michaels, Harro. 1980. Positionen der Deutschdidaktik seit 1949. Königstein Ts. Neumann, Friedrich. 1929. Die Sinneinheit des Satzes und das indogermanische Verbum. Festschrift für E. Husserl zum 70. Geburtstag gewidmet, 297— 314. Halle. Neuner, Gerhard. 1989. Methodik und Methoden. Handbuch, hrsg. v. Bausch u. a., 145—52. —. 1983. Zum Wandel der Übungsformen in der Methodik des Fremdsprachunterrichts. Handlungsorientierter Fremdsprachunterricht, hrsg. v. Albert Raasch, 67—74. Tübingen. Nickel, Gerhard. 1968. Der moderne Strukturalismus und seine Weiterführung bis zur generativen

1573

Transformationsgrammatik. Neusprachliche Mitteilungen 21.1. 5—15. Nikula, Henrik. 1982. Satzglieder und Verbvalenz. Satzglieder, hrsg. v. Werner Abraham, 293—307. Tübingen. Nold, Günter. 1981. Grammatik im Fremdsprachunterricht angesichts der kommunikativen Wende. Kommunikation in Europa, hrsg. v. F. J. Zapp et al., 144—55. Frankfurt a. M. Nündel, Ernst. 1976. Zur Grundlegung einer Didaktik des sprachlichen Handelns. Kronberg Ts. Oelinger, Albert. 1573. Underricht der Hoch Teutschen Spraach. Argentina. Oomen-Welke, Ingelore. 1982. Didaktik der Grammatik. Eine Einführung an Beispielen für die Klassen 5—10. Tübingen. Opitz, Martin. 1617, Aristarchus sive de contemptu linguae, ed. by G. Witkowski. Leipzig 1888. Otto, Ernst. 1925. Methodik und Didaktik des neusprachlichen Unterrichts. Bielefeld. Priscianus. Institutionum grammaticarum libri XVIII. Hrsg. v. H. M. Hertz. Berlin. 1855—59. Paul, Hermann. 1880. Prinzipien der Sprachgeschichte. Halle. [Achte Auflage 1970. Tübingen.] Pestalozzi, Johann H. 1807. Wochenschrift für Menschenbildung von Heinrich Pestalozzi und seinen Freunden, Bd. 1. Aarau. Pfleiderer, Wolfgang. 195 4 . Grammatik und Sprachunterricht. Der Deutschunterricht, 4 6. . 21—38. Polenz, Peter von. 1988. Deutsche Satzsemantik. Grundbegriffe des Zwischen-den-Zeilen-Lesens. Berlin. Pommerin, Gabriele. 1983. Pragmatik und Sprachunterricht. Handbuch zum Sprachunterricht, hrsg. v. Bernhard Weisgerber et al. 110—29. Weinheim. Pregel, Dietrich. 1970. Zum Sprachstil des Grundschulkindes. Düsseldorf. Primus, B eatrice. 1987. Grammatische Hierarchien. Eine Beschreibung und Erklärung von Regularitäten des Deutschen ohne grammatische Relationen. München. Raasch, Albert (ed.) 1983. Grammatische Terminologie: Vorschläge für den Sprachunterricht. Handlungsorientierter Fremdsprachunterricht, hrsg. v. Albert Raasch. Tübingen. Rall, Marlene et al. 1977. Dependenz-Verb-Grammatik für Deutsch als Fremdsprache. Heidelberg. Rank, B ernhard. 1977. Der Sprachunterricht und die Theorie der Sprechtätigkeit — Thesen zur „Aufhebung“ des Grammatikunterrichts. Linguistik und Didaktik 31. 171—256. Ratke, Wolfgang. 1612. Memorial, welches zu Frankfurt auf dem Wahltag Anno 1612, den 7. Mai dem deutschen Reich übergeben. Die neue Lehrart. Pädagogische Schriften Wolfgang Ratkes, hrsg. v. Gerd von Hohendorf (Erziehung und Gesellschaft.

1574

Materialien zur Geschichte der Erziehung, hrsg. v. Robert Alt. 1957). Berlin. Raumer, Rudolf von. 1857. Der Unterricht im Deutschen. Geschichte der Pädagogik. Teil 3. Dritte Auflage, hrsg. v. Rudolf von Raumer. Stuttgart. Rausch, Georg. 1897. Zur Geschichte des deutschen Genitivs seit der mittelhochdeutschen Zeit. Dissertation, Gießen. Rehbein, Jochen. 1977. Komplexes Handeln. Elemente zur Handlungstheorie der Sprache. Stuttgart. Reis, Marga. 1982. Zum Subjektbegriff im Deutschen, ed. by Werner Abraham. Satzglieder, 173— 211. Tübingen. —. 1986. Subjekt-Fragen in der Schulgrammatik? Deutschunterricht 38.2. 64—84. Reuschling, Gisela. 1982. Grammatikunterricht in den Lehrplänen für die Grundschule. Diskussion Deutsch 13. 242—56. —. 1981. Sprachbücher für die Grundschule 1920—1978. Ein Beitrag zur Geschichte des Sprachbuchs. München. Roberts, P. 1964 . English syntax: A book of programed lessons. New York. Robins, R. H. 1957. Dionysios Thrax and the western grammar tradition. Transact. of the Philol. Soc. Oxford, 67—106. Rötzer, Hans Gerd (ed.) 1973. Zur Didaktik der deutschen Grammatik. Darmstadt. Roth, Georg M. 1795. Antihermes oder philosophische Untersuchungen über den reinen Begriff der menschlichen Sprache und die allgemeine Sprachlehre. Gießen. —. 1799. Systematische deutsche Sprachlehre. Gießen. Roth, H. L. 1971. Transformationsgrammatik in der englischen Unterrichtspraxis. Frankfurt a. M. Sandmann, Manfred. 1954 . Subject and predicate. A contribution to the theory of syntax. Edinburgh. Sasse, H.-J. 1978. Subjekt und Ergativ: Zur pragmatischen Grundlage primärer grammatischer Relationen. Folia Linguistica 12.3—4. 219—52. Schmidt, M. 1852. Über Dionysios Thrax Philologus VII, 360—382. —. 1853. Über Dionysios Thrax. Philologus VIII, 231—53; 510—20. Schmidt, Wilhelm. 1967. Grundfragen der deutschen Grammatik. Berlin. Schottel, Justus Georg. 1663. Ausführliche Arbeit von der Teutschen Haupt Sprach, hrsg. v. F. E. Koldewey. Leipzig 1900. —. 164 1. Lobrede von der Uhralten Hauptsprache der Teutschen, hrsg. v. F. E. Koldewey. Leipzig 1900. Schwarz, Sebald. 1933. Systematische Grammatik des Deutschen in der höheren Schule. Zeitschrift für Deutschkunde 47. 732—42.

XXII. Syntax und Nachbardisziplinen

Schwenk, Helga. 1972. Zur Theorie des Sprachunterrichts im Fach Deutsch. Linguistik und Didaktik 11. 178—84. Selinker, L. 1972. Interlanguage, International Review of Applied Linguistics in Language Teaching 10. 209—31. Seyfert, Gernot. 1976. Zur Theorie der Verbgrammatik. Tübingen. Sitta, Horst. 1974 . Didaktik und Linguistik. Diskussion Deutsch 5. 431—45. Siebenbrodt, Magdalene. 1966. Ein Beitrag zur Einwirkung des Grammatikunterrichts auf schriftliche Schüleräußerungen. Habil.schrift Halle, Saale. Sommer, Ferdinand. 1931. Vergleichende Syntax der Schulsprachen. Dritte Auflage. Stuttgart. Sowinski, B ernhard. 1969. Möglichkeiten und Grenzen strukturalistischer Sprachbetrachtung in der Schule. Wirkendes Wort 19. 163—75. Spies, B arbara. 1989. Zur Aneignung von grammatischem Wissen bei Primarstufenschülern. Untersuchungen zum Verständnis der Wortarten. Sprachbewußtsein und Schulgrammatik, hrsg. v. Eduard Haueis. Osnabrücker Beiträge zur Sprach-Theorie 40. Osnabrück. Spitta, Gudrun. 1990. Grammatik — nein danke? Die Grundschulzeitschrift 32. 2—4. Sroka, Kazimierz A. 1988. Grammatical categories: their nature and interlanguage variation. Categorial grammar, ed. by Wojciech Buszkowski et al. 333—54. Amsterdam. Steinitz, Renate. 1971. Adverbial-Syntax. Berlin. Steinthal, Heymann. 1850. Klassifikation der Sprachen, dargestellt als die Entwickelung der Sprachidee. Berlin. Strassner, Erich. 1977. Aufgabenfeld Sprache im Deutschunterricht. Zur Wandlung sprachdidaktischer Konzepte zwischen 1945 und 1975. Tübingen. Strehle, Heinrich. 1962. Einige grundsätzliche Bemerkungen zum funktionalen Grammatikunterricht. Deutschunterricht 15.12. 693—704. Strube, Gerhard, Hemforth, B . und H. Wrobel. 1990. Auf dem Weg zu psychologisch fundierten Modellen menschlicher Sprachverarbeitung: Echtzeitanalysen des Sprachverstehens. Sprache und Wissen. hrsg. v. Felix, s., Kanngießer, G. und G. Rickheit. Opladen. 115—134. Sucharowski, Wolfgang. 1977. Semantische Motivation syntaktischer Kategorien für den Schulunterricht. Papiere zur Linguistik 16. 131—58. —. 1978. Kategorialgrammatik und Grammatikunterricht. Linguistik und Didaktik 34 —35. 109— 28; 303—24. —. 1979. Syntax. Kritische Stichwörter zur Sprachdidaktik, hrsg. v. M. Goroschenk u. a. Rucktäschel. München. Sütterlin, Ludwig. 1900. Die deutsche Sprache der Gegenwart. Ihre Laute, Wörter, Wortformen und Sätze. Leipzig.

86.  Syntax und Sprachdidaktik

Tarvainen, Kalevi. 1979. Dependentielle Satzgliedsyntax des Deutschen. Oulu. —. 1981. Einführung in die Dependenzgrammatik. Tübingen. Tegge, Egon. 1975. Zur linguistischen Legitimation von Sprachbüchern. Diskussion Deutsch 6. 399— 417. Tesniere, Lucien. 1959. Elements de syntaxe structurale. Paris 1959 [= übersetzt v. Ulrich Engel, Stuttgart 1980] Thomas, O. 1965. Transformational grammar and the teacher of English. New York. Vennemann, Theo. 1974 . Topics, subjects and word order: from SXV to SVX via TVX. Historical Linguistics, ed. by John M. Anderson & Ch. Jones, 259—301. Amsterdam. —. 1977. Konstituenz und Dependenz in einigen neueren Grammatiktheorien. Sprachwissenschaft 2. 259—301. —, und R. Harlow. 1977. Categorial grammar and consistent basic VX serialization. Theoretical Linguistics 4. 227—54. —, und Joachim Jacobs. 1982. Sprache und Grammatik. Darmstadt. Wahlen, Hartmut. 1982. Sprachlehre und Sprachkunde in der Hauptschule. 7. bis 9. Jahrgangsstufe. (Exempla 34) Donauwörth. Weigand, Edda. 1979. Zum Zusammenhang von Thema/Rhema und Subjekt/Prädikat. Zeitschrift für Germanistische Linguistik 7. 167—89. Weise, Christian. 1672. Die drei ärgsten Erznarren. Comenius und die Sprache. (Pädagogische Forschungen. Veröffentlichungen des Comenius-Instituts. Band 10), hrsg. v. Heinrich Geißler, 1959 Heidelberg. Weisgerber, B ernhard. 1974 . Theorie der Sprachdidaktik. Heidelberg. Weisgerber, B ernhard et al. 1983. Handbuch zum Sprachunterricht. Weinheim. Weisgerber, Leo. 1950. Das Tor zur Muttersprache. Düsseldorf. —. 1950—51. Grammatik im Kreuzfeuer. Wirkendes Wort 1. III. 129—39. Wittgenstein, Ludwig. 1953. Philosophische Unter-

1575

suchungen. Oxford. Wode, Henning. 1988. Einführung in die Psycholinguistik. Theorien, Methoden, Ergebnisse. München. —. 1977. Lernerorientiertheit im Fremdsprachenunterricht: FU als Spracherwerb. Neue Perspektiven der Fremdsprachdidaktik, hrsg. v. Hans Hunfeld, 17—23. Kronberg. —. 1981. Learning a second language. I: An integrated view of language acquisation. Tübingen. Wolfrum, Erich. 1976. Taschenbuch des Deutschunterrichts. Zweite erweiterte Auflage. Baltmansweiler. Wunderlich, Dieter. 1976. Studien zur Sprachakttheorie. Frankfurt a. M. —. 1985. Über die Argumente des Verbs. Linguistische Berichte 97. 183—227. —, und Peter Conrady. 1978. Von der Schulgrammatik zur Grammatik der Sprechhandlung. Didaktische Trends, hrsg. v. W. Born & G. Otto, 298— 312. München. Wurst, Raimund Jakob. 1836. Praktische Sprachdenklehre für Volksschule und die Elementarklassen der Gymnasial- und Real-Anstalten. Nach Dr. K. F. Beckers Ansichten über die Behandlung des Unterrichts in der Muttersprache bearbeitet. (73. Auflage) Reutlingen. —. 1836. Theoretisch-praktische Anleitung zum Gebrauche der Sprachdenklehre. Ein Handbuch für Elementarlehrer, welche vorwärts schreiten und sich vorbereitend mit dem neuesten Standpunkte der deutschen Sprachwissenschaft bekannt machen wollen. Reutlingen. Zander, Sönke. 1978. Psychologie und Sprachdidaktik. Bochum. Zimmermann, Günther. 198 4 . Erkundungen zur Praxis des Grammatikunterrichts. (Schule und Forschung). Frankfurt a. M. Zöfgen, Ekkehard. 1982. Verbwörterbücher und Verbvalenz im Französischunterricht. Linguistik und Didaktik 49—50. 18—61.

Wolfgang Sucharowski, München (Deutschland)

1576

XXIII. Service-Artikel Service-Article

87. 1. 2. 3. 4. 5. 6.

1.

Formale Sprachen Geschichtlicher Überblick Notationenverzeichnis Grundlagen der Theorie der formalen Sprachen Syntaxen Erweiterungen generativer Syntaxen in der Linguistik Literatur

Geschichtlicher Überblick

Unter einer formalen Sprache vers tehen wir im folgenden eine endliche oder unendliche Menge von Zeichenketten. Die Theorie der formalen Sprachen, wie in dies em Artikel entwickelt, geht auf den norwegis chen Mathematiker Axel Thue zurück, der in Probleme über die Veränderungen von Zeichenreihen nach gegebenen Regeln (1914) Produktionens ys teme, mit denen aus Zeichenketten neue Zeichenketten erzeugt werden, einführt und Eigens chaften dies er Sys teme unters ucht, — s o die Frage, ob die Äquivalenz zweier beliebig gegebener Produktionens ys teme immer ents cheidbar is t. Dies e Sys teme s ind unter der Bezeichnung Thue- bzw. Semi-Thue-Systeme bekannt. Die Arbeit Thues s cheint zunächs t wenig beachtet worden zu s ein. Die Theorie der formalen Sprachen bekommt ers t um 1936 aus der Richtung der mathemati s chen Logik neue Impuls e, wobei vornehmlich Ents cheidbarkeits fragen im Vordergrund s tehen. In dies em Zus ammenhang s tehen die Arbeiten von Kleene (1936) und Church (1936) über rekurs ive Funktionen und bes onders Turing (1937), der das Konzept der Turingmas chine entwickelt. Wichtige Beiträge zu Ents cheidbarkeits fragen bzgl. formaler Sys teme liefert s päter Pos t (1944; 1946 Pos ts Ents prechungs problem). Die genannten Arbeiten s ind rein mathematis cher Natur ohne Bezug zu natürlichen Sprachen. Die Verbindung zur Linguis tik

wird von Choms ky (vgl. Choms ky 1956; 1957; 1959) in der von ihm entwickelten „generativen Grammatik“ herges tellt. In der Linguis tik geht Choms ky von Pos itionen des amerikanis chen Strukturalis mus aus , die er in neuer Weis e interpretiert und weiterentwikkelt. Dazu gehören die Kons tituentenanalys e (Bloomfield 1933, Harris 1951) und der Begriff der s yntaktis chen Trans formation (Harris 1952). Bas ierend auf den früheren mathematis chen Arbeiten ents teht nun die Theorie der formalen Sprachen als mathematis che Metatheorie der generativen Syntaxen und der von ihnen erzeugten Sprachen. Aus s chlaggebend für die linguis tis che Modellbildung erweis t s ich die Übernahme des Rekurs ivitäts begriffs , der bereits von Bar-Hillel (1953) auf natürlich-s prachliche Strukturen angewendet wurde. Um 1960 wird die Theorie der formalen Sprachen in der im we s entlichen von Choms ky ges chaffenen Form für die Informatik interes s ant, nachdem ers tmals eine höhere Programmier s prache formal definiert wurde, nämlich ALGOL-60 in der Backus Naur-Form (Backus 1959). Dies er Formalis mus erweis t s ich als Notations variante für kontextfreie Syntaxen, wodurch die Verbindung zur generativen Syntax herges tellt is t. Von nun an wird die Theorie der formalen Sprachen grundlegend für das Gebiet der Programmier s prachen und Compiler und entwickelt s ich dadurch in der Folgezeit zu einem Kerngebiet der theoretis chen Informatik. Viele der hier gewonnenen Ergebnis s e haben wiederum Eingang in die linguis tis che Syntax und bes onders in die Computerlinguistik gefunden.

2.

Notationenverzeichnis

Hier eine Auflis tung der Termini und Notationen, die im folgenden verwendet werden. Es bedeutet

87.  Formale Sprachen

⋀ ∨ gdw := x∈M x∉M {a1, ..., an} {x B(x)} N card(M) ∅ N⊆M N⊂M N=M M—N M⋃N

Mi

M⋂N

Mi P(M) (a1, ..., an) M×N M1 × ... × Mn

logisches „und“ logisches „oder“ logische Äquivalenz („genau dann wenn“ ) ist definitionsgemäß gleich x ist Element der Menge M x ist nicht Element von M die aus den Elementen a1 und ... und an gebildete Menge die Menge aller x, auf die das Prädikat B zutrifft die Menge der natürlichen Zahlen beginnend bei 0 die Kardinalzahl von M die leere Menge N ist eine Teilmenge von M (M ist eine Obermenge von N, auch M ⊇ N) N ist eine echte Teilmenge von M (M ist eine echte Obermenge von N, M ⊃ N) die Mengen N und M sind gleich (d. h. N ⊆ M und M ⊆ N) das Komplement von N bzgl. M, definiert durch M — N := {x  x ∈ M ⋀ x ∉ N} die Vereinigungsmenge von M und N, definiert durch M ⋃ N:={x  x ∈ M ∨ x ∈ N} die Vereinigungsmenge M1 ⋃ ... ⋃ Mn (abgek. Mi) i die Durchschnittsmenge von M und N, definiert durch M ⋂ N := {x  x ∈ M ⋀ x ∈ N} die Durchschnittsmenge M1 ⋂ ... ⋂ Mn die Potenzmenge von M, definiert durch P(M) := {N  N ⊆ M} das n-Tupel bestehend aus den Komponenten a1 und ... und an das kartesische Produkt von M und N, definiert durch M × N := {(x, y)  x ∈ M ⋀ y ∈ N} das n-fache kartesische Produkt der Mengen M1 und ... und Mn, definiert durch M1 × ... × Mn := (M1 × ... × Mn—1) × Mn

Zwei Mengen M und N heißen disjunkt gdw M ⋂ N = ∅. Jede Teilmenge eines n-fachen kartes is chen Produkts von Mengen M1 und ... und Mn is t eine n-stellige Relation zwischen M1 und . . . und Mn. Is t M1 = ... = Mn = M, s o s pricht man von einer n-s telligen Relation

1577

auf M. 1-s tellige Relationen auf einer Menge M s ind Teilmengen von M. Eine Relation R ⊆ M1 × ... × Mn is t eine Funktion gdw für alle (x1, ..., xn—1, xn) gilt: s ind (x1, ..., xn—1, y1) und (x1, ..., xn—1, y2) ∈ R, dann y1 = y2. Die Menge der (n — 1)-Tupel (x1, ..., xn—1), für die es ein y gibt, s o daß (x1, ..., xn—1, y) ∈ R, heißt der Definitionsbereich der Funktion, Mn ihr Wertebereich. Die Menge der y ∈ Mn, für die es ein (x1, ..., xn—1) gibt, dem s ie zugeordnet s ind, heißt die Wertmenge (oder Bildmenge, das Bild) der Funktion. Is t f eine Funktion, Df ihr Definitions bereich und W ihr Wertbereich, s o notieren wir dies f: Df → W (zu les en „Df wird durch f in W abgebildet“ ). Für (x1, ..., xn—1, y) ∈ f s chreibt man auch f(x1, ..., xn—1) = y.

3.

Grundlagen der Theorie der formalen Sprachen

Die Theorie der formalen Sprachen baut auf dem Grundbegriff Zeichen (Symbol) auf. Was ein Zeichen is t, wird nicht mathematis ch definiert s ondern kann nur informell bes chrieben werden. Unter einem Zeichen wird ein Ding vers tanden, das eine äußere Form hat, als o eine Figur is t, und Zeichen werden voneinander durch die äußere Form unters chieden. Der Zeichenbegriff der Theorie der formalen Sprachen is t von dem der s trukturalis tis chen Linguis tik dadurch vers chieden, daß von jeglichem Inhalt als einem Übers ichhinaus weis en des Zeichens abs trahiert wird. Buchs taben, Ziffern und Interpunktions zeichen s ind in dies em Sinn Zeichen, aber auch ges chriebene Wörter einer natürlichen Sprache können als formals prachliche Zeichen betrachtet werden, s ieht man von ihrer Inhalts seite ab. Eine endliche, nicht-leere Menge von Zeichen heißt ein Alphabet (oder Vokabular). Eine Folge a1 ... an (n  0), worin jedes ai (1 ≤ i ≤ n) Zeichen eines gegebenen Alphabets V is t, heißt eine Zeichenkette (kurz Kette oder Wort) über V. n legt dabei die Länge der Kette fes t. Ins bes ondere gibt es Ketten der Länge 1 und die Kette der Länge 0, die als leeres Wort bezeichnet und ε notiert wird. Mit dem Begriff der Kette is t die Operation des Aneinanderreihens , der Verkettung, verbunden. Sei V ein Alphabet. Für je zwei Ketten x, y über V, x = x1 ... xn und y = y1 ... ym is t die Verkettung von x und y, notiert xᵔy oder xy, durch xy := x1 ... xny1 ... ym gegeben. Für das leere Wort gilt xᵔε = εᵔx = x. Aus dies en Fes tlegungen folgt die As s oziativität der Verkettung, d. h. (xy)z = x(yz) = xyz.

XXIII. Service-Artikel

1578

Die Verkettung is t eine Operation, die in der klas s is chen Mengenlehre nicht vorges ehen is t, s o daß zu fragen is t, wie s ie s ich mengentheoretis ch begründen läßt. Ketten können als geordnete Mengen mit endlich vielen Komponenten aufgefaßt werden, wodurch s owohl die Reihenfolge als auch die mögliche Wiederholung von Zeichen abgedeckt si t, d. h. a1 ... an is t das n-Tupel (a1, ..., an). Die Verkettung is t dann eine bes ondere Operation über n-Tupeln (eins chl. 1-Tupeln und 0-Tupeln) derges talt, daß (a1, ..., an)  (b1, ..., bm) = (a1, ..., an, b1, ..., bm). Al s grundlegende Unter s cheidung ergibt s ich weiter die zwis chen Zeichen(typus) und Zeichenvorkommen. Zeichen als Typen s ind nur die Elemente von Alphabeten. In Ketten haben wir hingegen Zeichenvorkommen. So bes teht die Kette aabb über dem Alphabet {a, b} aus 4 Zeichenvorkommen, die 2 Zeichentypen zugeordnet s ind. Allerdings s agt man in freierer Sprechweis e (der wir uns , wenn Verwechs lungen aus ges chlos s en s ind, ebenfalls bedienen) auch, daß eine Kette a1 ... an aus n Zeichen bes teht, wobei Zeichenvorkommen gemeint sind. Die Zeichen eines Alphabets müs s en identifizierbar und voneinander unter s cheidbar s ein, m. a. W. man muß Gleichheit bzw. Ungleichheit von Zeichen fes st tellen können. Dies gewährleis tet aber noch nicht, daß die Ketten über einem Alphabet eindeutig in ihre Zeichenvorkommen zerlegbar s ind. Sei z. B. V = {a, aa, b, ab} ein Alphabet und aabab eine darüber gebildete Kette. Zerlegen wir s ie in ihre Zeichenvorkommen, s o erhalten wir mehrere Lös ungen, nämlich aabab, aaba b, aabab, aabab, aabab. Jede einzelne dies er Zerlegungen is t eine andere Kette, bes onders auch da jeder Kette ihre Länge (d. h. die Anzahl ihrer Zeichenvorkommen) eindeutig zugeordnet wird. Um in jedem Fall eine eindeutige Zerlegung zu erhalten, genügt es , jedem Zeichenvorkommen ein s ons t nicht benutztes Begrenzungs zeichen vorangehen (oder nachfolgen) zu las s en, z. B. ′, s o daß die ers te der obigen Lö s ungen al s ′a′a′b′a′b er s cheint. Man kann das Begrenzungs zeichen auch als ers ten (bzw. letzten) Bes tandteil der Zeichen(typen) ans ehen, s o daß die Definition der Verkettung nicht abgeändert werden muß. Das Begrenzungs zeichen wird man bes onders bei Alphabeten, deren Elemente längere Einheiten s ind, verwenden. Dazu gehören auch die Vokabulare natürlicher Sprachen, wo der Wortzwis chenraum, das Blank,

als Begrenzungs zeichen dient. Im folgenden wird ebenfalls das Blank benutzt, außer bei nur aus einzelnen Buchs taben bes tehenden Alphabeten, wo wir es weglassen können. Einige Definitionen: Sternmenge eines Alphabets: Die Menge aller Ketten über einem Alphabet V heißt die Sternmenge (der Stern) von V, notiert V*. Da Ketten aus beliebig vielen Zeichenvorkommen bes tehen, is t V* unendlich. Ins bes ondere gehört das leere Wort ε zu V*. Wir bezeichnen mit V+ die Menge V* — {ε}. Teilstück einer Kette: Sei x eine Kette über einem Alphabet V. y is t eine Teilkette von x gdw es Ketten u und v aus V* gibt, für die x = uyv gilt. y is t ein Anfangsstück (Endstück) von x gdw es ein z aus V* gibt, für das x = yz (x = zy) gilt. Formale Sprache: Sei V ein Alphabet. Jede Teilmenge von V* is t eine formale Sprache über V. Speziell s ind dann V, V*, ∅ und {ε} formale Sprachen über V. Statt ‘formale Sprache’ s agen wir hinfort auch einfach ‘Sprache’. Verkettungsprodukt: Sei V ein Alphabet, A ⊆ V* und B ⊆ V*. Das Verkettungs produkt AB is t definiert durch AB := {xy  x ∈ A ⋀ y ∈ B}. Ents prechend gilt für n formale Sprachen A1, ..., An (n  2) A1 ... An := {x1 ... xn  x1 ∈ A1 ⋀ ... ⋀ xn ∈ An}. So is t das Verkettungs produkt von A = {an n  1} mit B = {bn  n  1}, worin A, B ⊆ {a, b}*, AB = {anbm  n, m  1}. n-fache Potenz einer formalen Sprache: Sei L eine formale Sprache. Die n-fache Potenz von L, notiert Ln, wird rekurs iv, wie folgt, definiert. 1. L0 = {ε}. 2. Ln+1 = LnL. Beispiel: Sei V = {a, b} und L = {ambm  m 1}. Dann ist L2 = {ambmakbk  m, k  1}. Stern einer formalen Sprache: Der Stern einer formalen Sprache L, notiert L*, is t die Vereinigungs menge aller Potenzen von L: L* :=

Ln.

Man benutzt folgende Schreibkonventionen: Is t a ein beliebiges Zeichen eines Alphabets , s o notiert man die Kette, die genau aus n a-Vorkommen bes teht, durch an. Definition (rekursiv): 1. a0 = ε. 2. an+1 = ana. Für {an  n  0} s chreibt man a*. Dies e Notationen werden auf Ketten erweitert. Sei x

87.  Formale Sprachen

eine Kette über einem Vokabular V. Dann soll gelten: 1. (x)0 = ε. 2. (x)n+1 = (x)nx. So gilt z. B. ababababababab = (ab)7 für a, b ∈ V. Analog zu a* s chreibt man (x)* für {(x)n  n  0}.

4.

Syntaxen

4.1. Zum Begriff der formalen Syntax Die Theorie der formalen Sprachen in ihrer heutigen Aus prägung, in der s ie zugleich eine Theorie formaler Syntaxen si t, entwickelte s ich ungefähr s eit 1955 in enger Verbindung mit demjenigen Zweig der Linguis tik, der unter dem Namen „generative Grammatik“ bekannt is t. Beide wurden von Noam Choms ky begründet (vgl. Kap. 1.). Neu ist (a) das Konzept der generativen Syntax als Beschreibungsmodell für natürlichsprachliche Satzmengen, welches sehr bald zu dem der Transformationsgrammatik erweitert wird (Chomsky 1957; 1965), und (b) die mathematische Theorie der generativen Syntaxen und der von ihnen erzeugten Sprachen, auch Typ-i-Sprachen (0 ≤ i ≤ 3) genannt, die den Kern der Theorie der formalen Sprachen bildet. Die Theorie geht von der in Kap. 3. gegebenen Definition der formalen Sprachen aus. Sätze einer natürlichen Sprache können demnach als Zeichenketten angesehen werden, deren Grundzeichen je nach Beschreibungsebene Phoneme, Grapheme, Morpheme oder Wörter sind, und Mengen natürlich-sprachlicher Sätze sind so betrachtet formale Sprachen. Ganz allgemein legt eine Syntax für eine formale Sprache eine berechenbare Funktion fes t, deren Bildmenge die formale Sprache is t. Mit G der Syntax, fG der Funktion und L der Sprache gilt fG: N → L, worin N die Menge der natürlichen Zahlen is t. Gibt es für ein gegebenes L ein s olches G, s o s agt man, daß L mittels G aufgezählt wird bzw. daß L eine aufzählbare Menge is t. Alle endlichen Sprachen (darunter die leere Menge) s ind aufzählbar. Unter den unendlichen gibt es aufzählbare, aber auch nicht-aufzählbare, d. h. s olche, für die kein G exis tiert, wie in 4.2.2. gezeigt wird.

1579

Prinzipiell las s en s ich für Syntaxen G vers chiedene Typen von Formalis men finden, s o wie Funktionen auf mannigfache Weis e bes chreibbar s ind. In der linguis tis chen Syntax und der Informatik s ind zumindes t zwei Typen zu nennen, nämlich die generativen (oder Typ-i-)Syntaxen und die Automaten, die wir hier informell einander gegenübers tellen. Die generativen Syntaxen beruhen auf dem Konzept der Wiederschreibregel (rewriting rule), das auf Thue (1914) zurückgeht. Eine Wieders chreibregel is t eine Anweis ung, eine Zeichenkette z1 durch eine Zeichenkette z2 zu ers etzen (= wiederzus chreiben), notiert z1 → z2. Eine Kette uz1v wird dadurch in uz2v umgewandelt. Eine generative Syntax G is t durch zwei dis junkte Alphabete, die Ends ymbole und die Nicht-Ends ymbole, und durch eine darüber gebildete Menge von Wieders chreibregeln fes tgelegt, durch die, beginnend bei einem aus gezeichneten Nicht-Ends ymbol (wie z. B. SATZ), Zeichenketten abgeleitet werden. Die Menge der s o erhaltenen Ketten, die nur aus Ends ymbolen bes tehen, is t die von G erzeugte Sprache (notiert L(G)). Da der Erzeugungs prozeß s o ges taltet werden kann, daß s ukzes s ive die Elemente von L(G) aufgezählt werden (vgl. 4.2.1.), is t G eine berechenbare Funktion und damit eine Syntax gemäß obiger Definition. Ein Automat is t hingegen eine abs trakte Mas chine, die Zeichen von einem Rechenband lies t oder auch darauf s chreibt und damit Zeichenketten verarbeitet. Die zentrale Operation is t das Durchlaufen von Zuständen. In Abhängigkeit von dem aktuellen Zus tand des Automaten und dem gerade geles enen Zeichen wird ein Nachfolgezus tand erreicht und können vers chiedene Operationen, wie auf das Band s chreiben u. a., durchgeführt werden. Es werden Anfangs - und Endzus tände unters chieden. Diejenigen Zeichenketten, die (je nach Automatentyp in einem näher zu definierenden Sinn) verarbeitet werden, wenn der Automat von einem Anfangs zus tand aus einen Endzus tand erreicht, bilden die Sprache des Automaten. Die allgemeins ten Automaten s ind die Turingmas chinen. Deren Sprachen s ind aufzählbar, s o daß Turingmas chinen Syntaxen im oben gegebenen Sinn s ind (vgl. Hermes 1971, Maurer 1977). Die aus der Computerlinguis tik bekannten Übergang s netzwerk-Grammatiken gehören zu den Automaten, da ihr Charakteris tikum das Durchlaufen von Zus tänden unter Verarbeitung von Zeichen ist.

1580

4.2. Die Syntaxen und Sprachen der Typ-i-Hierarchie (0 ≤ i ≤ 3) 4.2.1. Typ-0-Syntaxen In 4.1. wurde das Konzept der generativen Syntax informell vorges tellt. Hier nun die übliche Definition dies er Syntaxen, die auch vom Typ 0 genannt werden. Definition . . 4. 2. 11 (Produktionssystem): Ein Produktionssystem (auch Semi-Thue-System genannt) is t ein Paar (V, R), worin V ein Alphabet und R eine endliche Menge von Paaren (u, v) mit u, v ∈ V* is t. (u,v) heißt Wiederschreibregel, Produktion oder einfach Regel. Statt (u, v) is t die Notation u → v verbreitet. Definition 4. 2. 1. 2. (Ableitung einer Kette): Sei P ein Produktions s ys tem (V, R). Für Ketten x, y ∈ V* führt x direkt zu y gdw es z1, z2 ∈ V* gibt, für die x = z1uz2 und y = z1vz2 und (u, v) ∈ R. Die Relation ‘führt direkt zu’ wird → ges chrieben, als o x → y. Um auf das Produktions s ys tem oder eine Regel r daraus Bezug zu nehmen, notiert man x y, x y bzw. x y, was ‘x führt in P (bzw. ‘in R’ bzw. ‘mittels r’) direkt zu y’ bes agt. Für Ketten x, y ∈ V* führt x zu y gdw es w0, ..., wn ∈ V* (n  0) gibt, s o daß x = w0, y = wn und w0 → ... → wn. Die Folge w0, ..., wn heißt eine Ableitung der Länge n. Eine Ableitung der Länge 1 heißt eine Regelanwendung. Die Relation ‘führt zu’ wird notiert, als o x y bzw. x y. Beispiele: (a) Sei P ein Produktions s ys tem (V, R) mit V = {a, b} und R = {ab → aab, ba → bba}. abbab → abbaab → abbbaab → aabbbaab → aabbbaaab is t eine Ableitung in P der Länge 4. (b) Sei V die Menge der Phoneme der deuts chen Standards prache vereinigt mit {#}, worin # ein Begrenzungs zeichen is t. Stimmhafte deuts che Kons onanten werden des onoris iert, wenn s ie in den Aus laut geraten, vgl. Tage /tage/ vers us Tag /tak/. Dies e Ers cheinung läßt s ich durch Produktions regeln wie g# → k# u. a. bes chreiben (# zeigt den Wortaus laut an). So wird die zugrundeliegende Form tag# in tak# überführt. Gemäß der Definition s ind auch Ableitungen der Länge 0 möglich, es gilt als o für jede Kette x x x. Die Relation ‘führt zu’ is t s omit reflexiv und trans itiv. Beides gilt nicht für ‘führt direkt zu’. Statt ‘x führt (direkt) zu y’ s agt man auch ‘y wird aus x (direkt) abgeleitet’ (s tatt ‘abgeleitet’ sagt man auch ‘erzeugt’, ‘generiert’).

XXIII. Service-Artikel

Definition .4.2.1.3 (Typ-0-Syntax): Eine Syntax vom Typ 0 G is t ein 4-Tupel (VN, VT, R, S). Darin s ind VN und VT dis junkte Alphabete. Die Elemente von VN heißen NichtEndsymbole (Nicht-Terminalsymbole), die von VT Terminal- oder Endsymbole. R is t eine Menge von Regeln u → v, worin u, v ∈ (VN ⋃ VT)* und u ∉ VT*. S is t ein aus gezeichnetes Element aus VN, das Startsymbol. Statt ‘Typ-0-Syntax’ s agt man auch ‘Regelgrammatik’, ‘Erzeugung ss yntax’, ‘Phra s enstrukturgrammatik’. Wir s prechen bezüglich u von der linken, bezüglich v von der rechten Seite einer Regel. Laut Definition können Ketten, die nur aus Ends ymbolen bes tehen, s owie das leere Wort nicht linke Seiten von Regeln s ein. Dies hängt mit dem Begriff der von einer Typ-0-Syntax erzeugten Sprache zus ammen. Eine Typ-0Syntax is t ein s pezielles Produktions s ys tem, nämlich (VN ⋃ VT, R), und die von ihr erzeugte Sprache die Menge aller nur aus Ends ymbolen bes tehenden Ketten, die in ihr aus gehend vom Starts ymbol ableitbar s ind. Man möchte aber nicht, daß aus dies en Ketten noch weitere Ketten ableitbar s ind. Die Unters cheidung in nicht-terminale und terminale Symbole ges chieht analog der linguis tis chen Unters cheidung in Symbole für s yntaktis che Kategorien und s olche für lexikalis che Formative. Ketten, die nur aus Ends ymbolen bestehen oder leer sind, heißen Endketten. Definition 4. 2. 1. 4. (Typ-0-Sprache): Sei G eine Typ-0-Syntax. Die von G erzeugte Sprache, L(G), is t definiert durch L(G) := {x  x ∈ VT* ⋀ S x}. L(G) heißt eine Typ-0Sprache. Beispiel: Sei G = ({S, A, B}, {a, b}, {S → ASB, S → ε, AB → BA, A → a, B → b}, S). L(G) is t die Menge aller Ketten über {a, b}*, die aus genaus o vielen a-Vorkommen wie bVorkommen s be tehen, ein s chließlich ε. Wir zeigen nun Satz 4. 2. 1. 1. : Jede Typ-0-Sprache is t aufzählbar. Beweis: Für jede Typ-0-Sprache L gibt es eine Typ-0-Syntax G, die L erzeugt. Es is t zu zeigen, daß G eine berechenbare Funktion f fes tlegt, s o daß L die Bildmenge von f is t. Wir können nun alle Ketten k aus L finden, die durch n Regelanwendungen beginnend bei S erzeugt werden. Für n = 1 s eien dies e k1, ..., km1, für n = 2 km1 + 1, ..., km1 + m2, allgemein für n = i (i > 1) km1 + ... + mi—1 + 1 ..., km1 + ... + mi (mi  0 für i  1). f is t nun gegeben durch f(j) = kj mit j einer natürlichen Zahl. Da jede Kette aus L

87.  Formale Sprachen

durch eine Anzahl n von Regelanwendungen erzeugt wird, muß s ie s ich in dies er Lis te befinden und bekommt einen Index j (evtl. auch mehrere Indizes ) zugeordnet. Wir haben damit eine berechenbare Funktion gefunden, deren Bildmenge L ist. L ist somit aufzählbar. Umgekehrt gilt auch Satz 4. 2. 1. 2. : Jede aufzählbare Sprache is t vom Typ 0. Jede aufzählbare Sprache is t s omit durch eine Typ-0-Syntax erzeugbar. Dies er Satz is t eine Folgerung aus der Thes e Turings , daß jeder Algorithmus , und d. h. jede berechenbare Funktion, s ich durch eine Turingmas chine dars tellen läßt, und aus dem Haupts atz für Turingmas chinen. Dies er bes agt, daß eine formale Sprache genau dann vom Typ 0 is t, wenn s ie die Sprache einer Turingmas chine is t. So ergibt s ich ins ofern eine Äquivalenz zwis chen Typ-0-Syntaxen und Turingmas chinen, als s ie dies elben Sprachen definieren. Ein aus führlicher Beweis findet s ich in Maurer 1977. 4.2.2. Typ-1-Syntaxen Die Typ-0-Syntaxen bilden für mathematis che Betrachtungen in der Theorie der formalen Sprachen ein grundlegendes Konzept. Als lingui ss ti che s Be s chreibung s mittel kommen s ie in unbes chränkter Form aber kaum in Frage, da die Ges amtheit ihrer formalen Möglichkeiten undurchs chaubar is t, wie im folgenden durch Satz 4.2.2.3. erhellt. Man betrachtet daher einges chränkte Klas s en von Typ-0-Syntaxen. Definition .4.2.2.1 (Typ-1-Syntax): Eine Typ-0-Syntax G = (VN, VT, R, S) heißt vom Typ 1 gdw alle Regeln aus R folgende Form haben: (1) S → ε oder (2) u → v mit Länge (u) ≤ Länge (v), wobei v ∈ ((VN ⋃ VT) — {S})*, falls S → ε aus R, und v ∈ (VN ⋃ VT)* sonst. Die Bedingung in (2) gewährleis tet, daß eine Anwendung dies er Regeln keine Kette erzeugt, die kürzer is t als die Aus gangs kette. Durch 1) wird das leere Wort fakultativ als Element der erzeugten Sprache zugelas s en. Die von einer Typ-1-Syntax erzeugte Sprache heißt Typ-1-Sprache. Beispiel: Sei G = ({S, A, B, C}, {a, b}, {S → ε, S → AB, S → ACB, C → AB, C → ACB, AB → BA, A → a, B → b}, S). G is t eine Typ-1-Syntax. G erzeugt dies elbe Sprache wie die Typ-0-Syntax im vorangehenden Beis piel (vgl. bei Definition 4.1.2.4.), als o is t

1581

diese Sprache auch vom Typ 1. Da jede Typ-1-Syntax vom Typ 0 is t, folgt, daß auch jede Typ-1-Sprache vom Typ 0 is t. Zu fragen bleibt, ob jede Typ-0-Sprache auch vom Typ 1 is t. Die Antwort is t ‘nein’. Für den Beweis führen wir den Begriff der entscheidbaren Sprache ein. Definition 4. 2. 2. 2. (ents cheidbare Sprache): Seien L1 und L2 formale Sprachen über einem Alphabet V und L1 ⊆L2. L1 heißt entscheidbar relativ zu L2 gdw es ein mathematis ches Verfahren gibt, das für jedes x ∈ L2 findet, ob x ∈ L1 oder nicht. Für ‘L1 is t ents cheidbar relativ zu V*’ s agt man einfach ‘L1 ist entscheidbar’. Definition 4.2.2.2. is t nicht s treng im mathematis chen Sinn, da der Begriff mathematisches Verfahren unpräzis iert is t. Genau genommen is t eine Funktion gemeint, deren Definitions bereich L2 und deren Wertebereich {0, 1} is t, s o daß jedes Element aus L1 effektiv auf 1 und jedes aus L2—L1 auf 0 abgebildet wird. Effektiv bedeutet, daß die Rechnung tats ächlich durchführbar si t und nach endlich vielen Rechens chritten abbricht und ein Ergebnis liefert. Ein mathematis ches Verfahren nennt man auch einen Algorithmus. Es gilt Satz 4. 2. 2. 1. : Jede Typ-1-Sprache is t entscheidbar. Beweis: Sei G = (VN, VT, R, S) eine Typ1-Syntax und x ∈ V*. Es is t heraus zufinden, ob S x gilt oder nicht. Für x = ε kann nur die Regel S → ε angewendet werden. Für x ≠ ε is t x ∈ L(G), wenn S = w1 → ... → wn = x, wobei Länge (w1) ≤ Länge (wi+1) (1 ≤ i < n). Für zwei beliebige Ketten y und z über VN ⋃ VT läßt s ich ents cheiden, ob y z. Sei F die Menge aller Folgen von Ketten über VN ⋃ VT der Form w1, ..., wm mit wm = x, worin Länge (wi) ≤ Länge (wi+1) (1 ≤ i < m) und wi ≠ wk für i ≠ k. Dann is t F endlich und für jede Folge darin läßt s ich ents cheiden, ob S = w1 → ... → wm = x gilt. Gibt es eine s olche Folge, s o gilt x ∈ L(G), andernfalls x ∉ L(G). Zwis chen ents cheidbaren und aufzählbaren Sprachen gelten folgende Beziehungen: Satz 4. 2. 2. 2. : Eine Sprache L über einem Alphabet V is t ents cheidbar gdw. L und V* — L aufzählbar sind. Beweis: Jede ents cheidbare Sprache L is t aufzählbar. Zunächs t is t V* aufzählbar, denn s eine Elemente können gemäß einer bere-

1582

chenbaren Funktion f: N → V* (z. B. in alphabetis cher Reihenfolge) aufgezählt werden. Die Elemente aus V* werden nun in der Reihenfolge ihrer Aufzählung daraufhin überprüft, ob s ie aus L s ind oder nicht, und damit wird, wenn L ents cheidbar is t, eine Aufzählung für L gefunden. Is t ein Element aus V* nicht aus L, s o is t es aus V* — L, s omit wird zugleich eine Aufzählung für V* — L gefunden. Sind umgekehrt L und V* — L aufzählbar, s o wird x ∈ V* als i.tes Element von L oder V* — L durch abwechs elndes Aufzählen dieser Sprachen gefunden. Es gilt aber auch Satz .4 .2 .2 .3 : Es gibt eine aufzählbare Sprache L, deren Komplement V* — L nicht aufzählbar ist. Gemäß Satz 4.2.2.2. gibt es dann eine aufzählbare Sprache, die nicht ents cheidbar is t. Damit s ind nun folgende Inklus ions beziehungen zwi s chen Sprachenmengen herges tellt. Sei F die Menge aller formalen Sprachen, L0 die Menge der aufzählbaren oder Typ-0-Sprachen, E die Menge der ents cheidbaren Sprachen und L1 die Menge der Typ1-Sprachen (wobei wir immer von dems elben Alphabet V ausgehen können). Es gilt F ⊃ L0 ⊃ E ⊇ L1. Es gilt sogar Satz 4. 2. 2. 4. : Es gibt eine ents cheidbare Sprache, die nicht vom Typ 1 is t, als o E D L1. Beweis von Satz 4. 2. 2. 3. : Wir betrachten eine Typ-0-Syntax G = (VN, VT, R, S0), worin VT = {a} und VN aus einigen Zeichen Si (i  0) bes teht. Jede Typ-0-Sprache über VT wird durch eine s olche Syntax erzeugt (da die Nicht-Terminale in Si’s umbenannt werden können). Is t R = {u1 → v1, ..., un → vn}, s o is t die Syntax durch die Kette u1 → v1# . . . #un → vn s pezifiziert. Für jede Kette aus (VN ⋃ VT ⋃ {→, #})* is t auf Grund von Definition 4.2.1.3. ents cheidbar, ob s ie eine Typ-0-Syntax is t. Falls ja, legt s ie eine formale Sprache über {a} fes t (die auch ∅ s ein kann). Somit is t die Menge dies er Syntaxen ents cheidbar und damit aufzählbar. Es gibt als o eine Aufzählung G1, G2, G3, ... dies er Syntaxen. Sei L eine Sprache über VT, wie folgt: für jedes n  0 gehört an zu L gdw an von Gn erzeugt wird. Wir zeigen, daß L aufzählbar is t. Da jedes L(Gi) aufzählbar is t, is t auch L(Gi) aufzählbar. Eine Aufzählung für L(Gi) läßt s ich auch als Folge der Regelanwendungen dars tellen, die die Elemente aus L(Gi) erzeugt.

XXIII. Service-Artikel

Stehe (Gi, j) für die j.te Regelanwendung in Gi. Wir ordnen alle Regelanwendungen in der Folge der Pfeile:

Jedes mal wenn (Gi, j) eine Endkette ergibt, wird diese als Element von L(Gi) gelistet, womit wir die gewüns chte Aufzählung für dies e Menge erhalten. Jedes mal wenn (Gi, j) ai ergibt, wird ai als Element von L gelis tet, wodurch L aufgezählt wird. Wir zeigen nun: VT* — L is t nicht aufzählbar. Angenommen, es is t aufzählbar. Dann gibt es ein k, für das L(Gk) = VT* — L. Da aber ak zu VT* — L gehört gdw. ak nicht von Gk erzeugt wird, ergibt s ich ein Widers pruch. VT* — L kann daher nicht aufzählbar s ein. Da L aufzählbar is t, is t damit der Satz bewiesen. Beweis von Satz 4. 2. 2. 4. : Der Beweis läuft analog dem von Satz 4.2.2.3. Wir betrachten jetzt alle Typ-1-Syntaxen G = (VN, VT, R, S0), worin VT = {a} und VN aus einigen Zeichen Sj (i  0) bes teht. Wie oben gilt, daß die Menge dies er Syntaxen ents cheidbar und damit aufzählbar is t. Sei G1, G2, G3, ... eine s olche Aufzählung und x1, x2, x3, ... eine Aufzählung von {a}*. Sei D = {xi  xi ∉ L(Gi)}. D is t ents cheidbar, da L(Gi) ents cheidbar is t. Angenommen, D is t vom Typ 1. Dann gibt es ein m, s o daß D = L(Gm), und es gilt xm ∈ L(Gm) gdw xm ∉ L(Gm), ein Widers pruch. D is t als o ents cheidbar, aber nicht vom Typ 1. Da alle Typ-1-Sprachen ents cheidbar s ind, gilt E ⊃ L1. Für lingui ss ti che Sprachbe s chreibungen wird eine s peziellere Form der Typ-1-Syntaxen betrachtet, die man als kontexts ens itiv bezeichnet (vgl. Chomsky 1957). Definition .4 .2 .2 .3 (kontexts ens itive Syntax): Eine Typ-1-Syntax G = (VN, VT, R, S) heißt kontextsensitiv gdw alle Regeln aus R folgende Form haben:

87.  Formale Sprachen

(1) S → ∈ oder (2) uAv → uxv mit u, v ∈ (VN ⋃ VT)*, x ∈ (VN ⋃ VT)+ und A ∈ VN. Regeln der Form (2) heißen kontextsensitive Regeln. Durch eine s olche Regel wird ein Element aus VN in Abhängigkeit von einem gegebenen Kontext, in dem es s teht, in eine Zeichenkette überführt. Zu beachten is t, daß der Kontext auch leer s ein kann. In der generativen Grammatik werden kontexts ens itive Regeln auch in der Form (2a) A → x / u___v mit A, x, u, v wie in Definition 4.2.2.3. notiert. A → x heißt Kern der Regel. Die beiden Schreibweis en (2) und (2a) s ind aber nicht ganz gleichwertig. Mit Notation (2) is t der Kern einer Regel nicht immer eindeutig gegeben, vgl. z. B. die Regel AB → ACB mit A, B, C ∈ VN, die als A → AC /___B oder B → CB / A___ interpretiert werden kann. Auf die erzeugten Sprachen (als Mengen von Endketten) hat dies keine Aus wirkungen, doch auf die Interpretation von Endketten hins ichtlich ihrer internen s yntaktis chen Gliederung. Gemäß der Definition s ind die von kontexts ens itiven Syntaxen erzeugten Sprachen eine Teilmenge der Typ-1-Sprachen. Es gilt aber auch Satz 4. 2. 2. 5. Für jede Typ-1-Sprache gibt es eine kontexts ens itive Syntax, die s ie erzeugt. Die Ein s chränkung auf kontext s en s itive Syntaxen hat als o keine Aus wirkungen auf die erzeugbare Sprachenmenge. Zum Beweis dies es Satzes benutzen wir den Begriff der s eparierten Syntax. Definition 4. 2. 2. 4. (Äquivalenz zweier Typ0-Syntaxen): Zwei Typ-0-Syntaxen G1 und G2 sind äquivalent gdw L(G1) = L(G2). Definition 4. 2. 2. 5. (s eparierte Syntax): Eine Typ-0-Syntax G = (VN, VT, R, S) heißt separiert gdw jede Regel aus R folgende Form hat: (1) u → v mit u, v ∈ VN+ oder (2) u → v mit u ∈ VN und v ∈ VT ⋃ {ε}. Satz 4. 2. 2. 6. : Zu jeder Typ-0-Syntax G gibt es eine äquivalente separierte Syntax Gs. Beweis: Sei G = (VN, VT, R, S) mit VT = {a1, ..., an}. Sei G′ eine Typ-0-Syntax, wie folgt: G′ = (VN′, VT, R′, S) mit VN′ = VN ⋃ {A1, ..., An}, wobei {A1, ..., An} ⋂ VN = ∅, und R′ = R″ ⋃ {A1 → a1, ..., An → an}, wobei R″ aus R folgendermaßen ents teht: Alle Regeln aus R der Form u → v mit u ∈ VN und v ∈ VT s owie alle, in denen kein ai vorkommt, s ind auch in R″. In allen übrigen Regeln werden die ai durch Ai ers etzt. Dies e

1583

neuen Regeln gehören ebenfalls zu R″. Da dies nur eine Umbenennung von ai’s in R is t und die Aj mittels R′ wieder durch ai ers etzt werden können, gilt L(G) = L(G′). Seien u1 → v1, ..., um → vm alle Regeln aus R′ mit vi = ε und ui ∈ VN′* (1 ≤ i ≤ m). Sei VNs = VN′ ⋃ {B1, ..., Bm}, wobei {B1, ..., Bm} ⋂ VN′ = ∅, und Rs = (R′ — {ui → vi  1 ≤ i ≤ m}) ⋃ {ui → Bi  1 ≤ i ≤ m} ⋃ {Bi → ε  1 ≤ i ≤ m}. Dann is t Gs = (VNs, VT, Rs, S) eine s eparierte Syntax, und es gilt offens ichtlich L(G′) = L(Gs) und, da L(G′) = L(G), auch L(Gs) = L(G). Konstruktion eines Gs zu einem G, s o daß L(G) = L(Gs): G = ({S, B, C}, {a, b, c}, {S → aSB, aS → aBC, BC → b, aB → ε, BC → bc, B → ab}, S). Gs = ({S, B, C, A, B1, C1, D}, {a, b, c}, {S → ASB, AS → ABC, BC → B1, AB → D, BC → B1C1, B → AB1, A → a, B1 → b, C1 → c, D → ε}, S). Beweis von Satz 4. 2. 2. 5. : Die Beweis idee is t, zu einer beliebigen Typ-1-Syntax eine äquivalente kontexts ens itive Syntax zu kons truieren. Sei eine beliebige Typ-1-Syntax gegeben. Nach Satz 4.2.2.6. gibt es dazu eine äquivalente s eparierte Syntax. Dies e is t nach dem Kons truktions verfahren auch eine Typ1-Syntax, da keine Längenveränderung der Regels eiten erfolgt. Wir können daher von dies er s eparierten Syntax aus gehen. Sei dies e G = (VN, VT, R, S). R enthält Regeln der Form (1) S → ε oder (2) A → a mit A ∈ VN und a ∈ VT oder (3) A1 ... Am → B1 ... Bn mit n  m und A1, ..., Am, B1, ..., Bn ∈ VN. Regeln der Form (1) und (2) können in einer kontext s en s itiven Syntax auftreten. Jede nicht kontexts ens itive Regel der Form (3) ri wird durch eine Menge kontexts ens itiver Regeln ers etzt. Es werden dabei m neue NichtTerminale C1, ..., Cm, die in keiner weiteren Regel auftreten, eingeführt. Die Regeln sind

1584

Sei dies e Regelmenge Ri′, R′ = Ri′ und R″ die Menge der Regeln aus R, die bereits zu einer kontexts ens itiven Syntax gehören können. Sei Ni die Menge der in Ri′ neu eingeführten Nicht-Terminale Cj und V′ = Ni. Behauptung: Mit G1 = (VN1, VT, R1, S), worin VN1 = VN ⋃ V′ und R1 = R′ ⋃ R″, is t eine zu G äquivalente kontexts ens itive Syntax gegeben. Beweis der Behauptung: (i) Wir zeigen, daß L(G) ⊆ L(G1). Für jede nicht kontexts ens itive Regel ri aus R der Form A1 ... Am → B1 ... Bn gilt xA1 ... Amy , xB1 ... Bny mit x, y, ∈ (VN1 ⋃ VT)*. Die Ketten, die ri erzeugt, können als o auch durch die Regelmenge Ri′ erzeugt werden. Da Ri′ ⊆ R1 und R″ ⊆ R1, erzeugt G1 alle Endketten, die auch G erzeugt, als o gilt L(G) ⊆L(G1). (ii) Wir zeigen, daß L(G1) ⊆ L(G). Da jedes Cj ∈ V′, das in Regeln aus Ri′ vorkommt, nicht in den Regeln aus R1 — Ri′ auftritt, kann es nur mittels Regeln aus Ri′ weiter ers etzt werden. Für jede Ableitung S xA1 ... Amy z mit z ∈ L(G1), x, y ∈ (VN1 ⋃ VT)* und A1 ... Am der linken Seite einer Regel aus R1, worin Ak ∉ V′ (1 ≤ k ≤ m), gibt es eine Ableitung S xA1 ... Amy xB1 ... Bny z mit Br ∉ V′ (1 ≤ r ≤ n) für ein Ri′ ⊆ R1 oder eine Ableitung S A1 ... Amy xB1 ... Bny z. Dafür gibt es aber auch eine Ableitung S xA1 ... Amy xB1 ... Bny z, und, da dies auf jeden Ableitungs s chritt zutrifft, gilt auch S xA1 ... Amy xB1 ... Bny z. Daher gilt L(G1) ⊆ L(G). Zus ammen mit (i) gilt die Behauptung. 4.2.3. Kontextfreie Syntaxen Die in der Linguis tik verwendeten Phras ens trukturs yntaxen haben zumeis t die Form einer kontextfreien Syntax oder lehnen s ich stark an diesen Syntaxtypus an. Definition .4 .2 .3.1 (kontextfreie Syntax): Eine Typ-0-Syntax G = (VN, VT, R, S) heißt kontextfrei oder vom Typ 2 gdw jede Regel aus R die Form A → x mit A ∈ VN und x ∈ (VN ⋃ VT)* hat. Eine Regel dies er Form heißt kontextfreie Regel. Die von einer kontextfreien Syntax erzeugte Sprache heißt ebenfalls kontextfrei oder vom Typ 2. Regeln der Form A → ε heißen Tilgungsregeln. Kontexts ens itive Regeln mit leerem Kon-

XXIII. Service-Artikel

text s ind demnach kontextfrei. Man beachte jedoch, daß die kontextfreien Syntaxen keine Teilmenge der kontexts ens itiven s ind, da die linke Seite von Tilgungs regeln nicht das Starts ymbol zu s ein braucht und dies e Regeln unbe s chränkt angewendet werden können. Für die Sprachen gilt aber: Die Menge der Typ-2-Sprachen (L2) is t eine echte Teilmenge der Menge der Typ-1-Sprachen (L1). Um dies zu zeigen, s ind einige weitere Sätze und Festlegungen vonnöten. Zunächst zeigen wir Satz 4. 2. 3. 1. : Jede kontextfreie Sprache is t vom Typ 1. Beweis: Sei L die Sprache und G = (VN, VT, R, S) eine kontextfreie Syntax, s o daß L = L(G). Enthält R keine Regeln der Form A → ε, s o is t G bereits kontexts ens itiv und damit auch L. Andernfalls wird eine zu G äquivalente Typ-1-Syntax G1 kons truiert, indem Mengen Xi rekurs iv, wie folgt, gebildet werden: (i) X1 = {A  A ∈ VN ⋀ A → ε}, (ii) Xi+1 = Xi ⋃ {A  A ∈ VN und es gibt ein x ∈ Xi* mit A → x} für i  1. X1 enthält alle Nicht-Terminale, die direkt zu ε führen, und jedes Xi+1 außer den Elementen von Xi alle Nicht-Terminale, die direkt zu Ketten aus Xi* führen. Da VN endlich is t, gibt es eine natürliche Zahl i, für die Xi = Xj für jedes j > i. Sei dies es i gleich n für uns er G. Xn enthält alle A aus VN, für die A ε. Is t S ∈ Xn, s o gilt S ε, als o ε ∈ L(G). Sei nun G1 = (VN1, VT, R1, S1), worin R1 folgendermaßen aus R entsteht: (1) Alle Regeln aus R der Form A → x mit x ≠ ε sind aus R1. (2) Zu jeder Regel rk aus R (1 ≤ k ≤ card (R)) der Form A →y, worin y = x1A1 ... xmAmxm+1 mit Ai ∈ Xn und xi ∈ ((VN ⋃ VT) — Xn)*, wird die Regelmenge Mk gebildet, die genau die Regeln enthält, die ents tehen, wenn man in y ein oder mehrere Ai tilgt, dabei aber die rechte Seite nicht zu ε wird. Alle diese Regeln gehören zu R1. (3) Is t ε ∉ L(G), s o wird noch die Regel S1 → S, bei ε ∈ L(G) die Regeln S1 → S und S1 → ε zu R1 gefügt, worin S1 ein neues Starts ymbol nicht aus VN is t. Dadurch wird gewährleis tet, daß in G1 das Starts ymbol nicht auf einer rechten Regelseite auftritt. Mit VN1 = VN ⋃ {S1} is t G1 kons truiert. G1 is t kontextfrei und zugleich kontexts ens itiv. Behauptung: Es gilt L(G1) = L(G). Beweis der Behauptung: Da die Regeln nach (2) aus denen von R durch Fortlas s en

87.  Formale Sprachen

1585

von Zeichen ents tehen, die auf das leere Wort führen, ohne daß dies e Regeln s elbs t das leere Wort erzeugen, und durch (3) gewährleis tet is t, daß ε ∈ L(G) gdw ε ∈ L(G1), gilt L(G1) ⊆ L(G). Anderers eits gibt es für jede Ableitung A z mit z ≠ ε auf Grund von (1) und

tax G gibt es eine äquivalente kontextfreie Syntax G1, deren Regeln folgende Form haben: (1) A → x mit Länge (x) > 1 oder (2) S → x mit S dem Startsymbol von G1 und x ∈ VT oder x = ε

(2) auch eine Ableitung A z, da R1alle Regeln enthält, in denen kein, eins oder mehrere Zeichen, die in R auf ε führen, getilgt s ind. Es bleiben damit alle Möglichkeiten, ein z ε L(G) mit z ≠ ε zu erzeugen, in G1 erhalten. Zus ammen mit (3) gilt daher auch L(G) ⊆ L(G1). Damit is t die Behauptung bewies en, woraus der Satz folgt. Es gilt als o L(G) = L(G1). Beispiel: Kons truktion einer Syntax G1: Sei G = ({S, A, B}, {a, b, c}, {S → aAbSa, S → AB, A → aca, A → ε, B → ε}, S). Wir kons truieren Xn: X1 = {A, B}, X2 = {S, A, B} = X3 = Xn. Da S ∈ Xn, gilt ε ∈ L(G). Wir erhalten G1 = ({S1, S, A, B}, {a, b, c}, R1, S1) mit R1 = R2 ⋃ R3 ⋃ R4, worin R2 = {S → aAbSa, S → AB, A → aca} nach (1), R3 = {S → abSa, S → aAba, S → aba, S → A, S → B} nach (2) und R4 = {S1 → ε, S1 → S} nach (3). In G führt B nur zu ε. Des halb kann eine in G1 abgeleitete Kette, in der B vorkommt, zu keiner Endkette führen. Sie führt in eine Sackgasse. Sackgas s en s ind für die erzeugte Sprache ohne Belang. Daher können aus R1 alle Regeln fortgelas s en werden, in denen B vorkommt, ohne daß s ich die erzeugte Sprache verändert. Führt in einer Typ-0-Syntax G keine Ableitung zu einer Endkette, s o is t L(G) = ∅. Dies is t zu unters cheiden von dem Fall, wo für ein G L(G) = {ε}. Beispiele: (a) G1 = ({S, A}, {a, b, c}, {S → aA, A → Sb, A → aAc}, S). L(G1) = ∅. (b) G2 = ({S, A}, {a, b, c}, {S → ε, S → aA, A → aAc}, S). L(G2) = {ε}. In Kapitel 4.2.2. haben wir ges ehen, daß zwei Syntaxtypen, von denen der eine ein Spezialfall des anderen is t, nicht unbedingt vers chiedene Sprachenmengen fes tlegen. So is t die Menge der Typ-1-Sprachen gleich der Menge der von kontexts ens itiven Syntaxen erzeugten Sprachen. Ebens o is t zu fragen, ob vielleicht alle Typ-1-Sprachen durch Typ-2Syntaxen erzeugt werden. Die Antwort is t ‘nein’: es gibt eine Typ-1-Sprache, die nicht vom Typ 2 is t. Als o L1 ⊃L2 (L2 die Menge der Typ-2-Sprachen). Für den Beweis behandeln wir zunächs t zwei weitere Sätze über kontextfreie Syntaxen. Satz 4. 2. 3. 2. : Zu jeder kontextfreien Syn-

Endketten der Länge 1 werden als o nur direkt vom Starts ymbol aus erzeugt. Aus ges chlos s en s ind in G1 Regeln der Form A → B mit B ∈ VN und A → x mit x ∈ VT und A ≠S. Beweis des Satzes: Sei G = (VN, VT, R, S). Ohne Bes chränkung der Allgemeinheit können wir davon aus gehen (vgl. vorigen Satz), daß G auch kontexts ens itiv is t und s omit ε nur direkt von S aus erzeugt wird. Sei VN ⋃ VT = V. Für jedes x ∈ V werden Mengen Ui(x) rekursiv, wie folgt, gebildet: (1) U1(x) = {x}, (2) Ui+1(x) = Ui(x) ⋃ {y  y ∈ V und es gibt ein z ∈ Ui(x) mit z → y } für i  1. Für jedes x aus dem Gesamtvokabular V werden s o alle diejenigen Zeichen ges ammelt, die von x aus mittels R ableitbar s ind. Da V endlich is t, gibt es eine natürliche Zahl i, für die Ui(x) = Uj(x) für jedes j > i. Sei für uns er G i = n. Ein a ∈ VT gehört zu L(G) gdw a ∈ Un(S). Es wird eine neue Regelmenge R1 konstruiert: (1) Alle Regeln W → w1 ... wm mit wi ∈ V (1 ≤ i ≤ m) und m  2, für die es eine Regel X → x1 ... xm aus R mit X ∈ Un(W) und ein wi mit wi ∈ Un(xi) gibt, und (2) alle Regeln der Form S → a mit a ∈ (VT ⋂ Un(S)) und (3) S → ε, falls diese Regel in R ist, gehören zu R1. Behauptung: Die Syntax G1 = (VN, VT, R1, S) ist G äquivalent. Beweis der Behauptung: Auf Grund der Regeln (1) gibt es für jede Ableitung W X x1 ... xm w1 ... wm mit m  2 eine Ableitung W w1 ... wm und umgekehrt für jedes W w1 ... wm eine Ableitung W X x1 ... xm w1 ... wm (man beachte die Reflexivität von , s o daß W = X und x1 ... xm = w1 ... wm möglich s ind). Durch die Regeln (2) enthalten L(G) und L(G1) dies elben Ketten der Länge 1. ε ∈ L(G) gdw ε ∈ L(G1) wegen (3). Daher gilt L(G) = L(G1) und damit der Satz.

1586

Beispiel: Kons truktion eines G1 wie in Satz 4.2.3.2.: Sei G = ({S, A, B, C}, {a, b, c}, {S → A, S → BS, B → C, B → aBc, B → b, A → a, C → c}, S). Dann is t G1 = ({S, A, B, C}, {a, b, c}, R1, S), und es gilt U1(S) = {S}, U2(S) = {S, A}, U3(S) = {S, A, a} = U4(S) = U7(S), U1(A) = {A}, U2(A) = {A, a} = U3(A) = U7(A), U1(B) = {B}, U2(B) = {B, C, b}, U3(B) = {B, C, b, c} = U4(B)= U7(B), U1(C) = {C}, U2(C) = {C, c} = U3(C) = U7(C), U1(a) = {a} = U2(a) = U7(a), U1(b) = {b} = U2(b) = U7(b), U1(c) = {c} = U2(c) = U7(c). R1 = R′ ⋃ R″ mit R′ = {S → a} nach (2) des Beweis es . R″ ents teht nach (1) wie folgt: Für S → BS treten die Regeln S → BS, S → BA, S → Ba, S → cS, S → cA, S → ca, S → bS, S → bA, S → ba, S → CS, S → CA und S → Ca ein, für B → aBc die Regeln B → aBc, B → aCc, B → abc und B → acc. Das s ind alle Regeln von R″. Damit ist G1 konstruiert. Definition . . 4. 2. 32 (Chomsky-Normalform): Eine kontextfreie Syntax G = (VN, VT, R, S) s teht in Chomsky-Normalform gdw jede Regel aus R folgende Form hat: (1) A → BC oder (2) A → a mit A, B, C ∈ VN und a ∈ VT. Satz 4. 2. 3. 3. : Zu jeder kontextfreien Syntax G, für die ε ∉ L(G), gibt es eine äquivalente Syntax in Chomsky-Normalform. Beweis: Auf Grund der vorherigen Sätze können wir ohne Bes chränkung der Allgemeinheit davon aus gehen, daß G s epariert is t und nur Regeln der Form A → x mit Länge (x)  2 oder x ∈ VT enthält. Für jede Regel ri aus R der Form A → B1 ... Bn mit B1, ..., Bn ∈ VN und n > 2 werden neue NichtTerminale C1, ..., Cn—2 und folgende Regeln eingeführt: A → B1C1, Cj → Bj+1Cj+1 für jedes j mit 1 ≤ j ≤ n—3, Cn—2 → Bn—1Bn. Wir nennen dies e Regelmenge Ri und s etzen {C1, ..., Cn—2} = Ni. Sei R′ die Menge der Regeln aus R der Form A → x mit Länge (x) = 2 oder x ∈ VT. Die Syntax GCH = (VNCH, VT, RCH, S) mit VNCH = VN ⋃ Ni und RCH = R′ ⋃ Ri s teht in der Choms ky-Normalform. Es gilt L(GCH) = L(G). Der Beweis wird analog dem von Satz 4.2.2.5. geführt. Beispiel: Kons truktion eines GCH: Sei G =

XXIII. Service-Artikel

({S, A, B}, {a, b}, {S → ABSA, S → ABAB, A → a, B → b}, S). G is t bereits s epariert und enthält nur Regeln der Form A → x mit Länge (x)  2 oder x ∈ VT. Wir erhalten GCH = ({S, A, B C1, C2, D1, D2}, {a, b}, {S → AC1, C1 → BC2, C2 → SA, S → AD1, D1 → BD2, D2 → AB, A → a, B → b}, S). Wir kommen nun zu einer wichtigen allgemeinen Charakteri s ierung kontextfreier Sprachen, nämlich Satz 4. 2. 3. 4. : Für jede kontextfreie Sprache L gibt es natürliche Zahlen p und q, s o daß für jedes z ∈ L mit Länge (z) > p gilt: Es gibt w1, w2, w3, w4, w5, s o daß z = w1w2w3w4w5, worin (i) Länge (w2w3w4) ≤ q, (ii) w2w4 ≠ ε und (iii) w1(w2)iw3(w4)iw5 ∈ L für jedes i  0. Beweis: Is t ε ∈ L, s o werden p und q für L — {ε} bes timmt. Sie erfüllen (i) bis (iii) dann ebenfalls für L. Durch Satz 4.2.3.3. wis s en wir, daß es eine kontextfreie Syntax G in Choms ky-Normalform mit L(G) = L — {ε} gibt. Sei G = (VN, VT, R, S) und z1 → ... → zn eine beliebige Ableitung in G mit Länge (z1) = 1 und zn ∈ VT*. Sei (A1, ..., Ak) ein k-Tupel Tk von Zeichen aus VN, wobei, falls k > 1, für jedes Ai (1 ≤ i ≤ k—1) und irgendwelche s , t, x, y ∈ (VN ⋃ VT)* gilt: Ai → s Ai+1t, und es gibt ein zj = s Aiy (1 ≤ j > n—1), s o daß zj+1 = xs Ai+1ty. D. h. Tk is t eine Folge von Nicht-Terminalen, die s ukzes s ive voneinander abgeleitet werden, bildlich ges prochen, ein „Pfad“ durch die Ableitung, wie folgendes Beispiel zeigt: X → YZ → VWZ → vWZ → vwZ → vwUV → vwuV → vwuv mit U, V, W, X, Y, Z ∈ VN, u, v, w ∈ VT und den Regeln X → YZ, Y → VW, Z → UV, U → u, V → v, W → w. Die möglichen Pfade s ind hier (X, Y, V), (X, Y, W), (X, Z, U), (X, Z, V). Sei Tk s o, daß es für z1 → ... → zn kein Th mit h > k gibt (d. h. Tk is t ein längs ter Pfad). Dann gilt Länge (zn) < 2k. Dies wird durch volls tändige Induktion bewies en. Die Behauptung gilt für k = 1, da dann n = 2 und Länge (zn) = 1 < 21. Angenommen die Behauptung gilt für ein (k—1)  1. Für jede Ableitung z1 → ... → zn mit Tk dem längs ten Pfad von Nicht-Terminalen gilt: z1 = Z, z2 = XY mit Z, X, Y beliebigen Symbolen aus VN und es gibt höchs tens zwei (k—1)-Tupel Tk—1 für z2 → ... → zn (wobei an deren ers ter Stelle das X- bzw. das Y-Vorkommen von z2 s teht). Daher gilt zn = xy mit Länge (x) < 2k—1 und

87.  Formale Sprachen

1587

Länge (y) < 2k—1. Als o gilt Länge (zn) < 2k—1 + 2k—1 = 2k. Wir s ind nun in der Lage zu zeigen, daß es natürliche Zahlen p und q gibt, s o daß die Eigens chaften (i) bis (iii) erfüllt s ind. Sei u = card (VN). Wir s etzen p = 2u und q = 2u+1 und betrachten ein beliebiges z ∈ L mit Länge (z) > p. Dann gibt es für jede Ableitung S z ein Tk mit k  u+1. Als o gibt es ein X ∈ VN, welches in Tk mindes tens zweimal vorkommt. Seien X1 und X2 dies e beiden Vorkommen von X. Dabei gibt es X1 und X2, die s o verteilt s ind, daß für Tk = (A1, ..., Ak) gilt: X1 = Ai, X2 = Aj für zwei i, j, wobei (k—u) ≤ i < j ≤ k. X s ei im folgenden s o gewählt. Sei nun z - v1v2v3v4v5 mit v1, ..., v5 ∈

VT*

und

gelte

, wobei s da X-Vorkommen in (a) X1 und das in (b) X2 is t. Dann gilt X v2v3v4 und Länge (v2v3v4) ≤ 2u+1 = q. Ferner, da G in Choms ky-Normalform s teht, haben wir v2v4 ≠ ε. Setzen wir wi für vi für jedes i ∈ {1, 2, 3, 4, 5}, bekommen wir (i) und (ii) des Satzes . Da es in G eine Ableitung S w1Xw5 w1w2Xw4w5 w1w2w3w4w5 gibt, gibt es in G auch S w1Xw5 w1w2Xw4w5 w1(w2)iX(w4)iw5 w1(w2)iw3(w4)iw5 für jedes i 0. Damit gilt die Eigens chaft (iii). Damit si t der Satz bewie s en. ' Auf Grund dies es Satzes läßt s ich nun zeigen, daß es eine Typ-1-Sprache gibt, die nicht vom Typ 2 is t. Man braucht nur eine Typ-1Sprache zu finden, die eine der Bedingungen (i) bis (iii) nicht erfüllt. Eine s olche Sprache ist L = {anbncn  n  1}. Daher Satz 4. 2. 3. 5. : Die Sprache L = {anbncn  n 1} ist vom Typ 1, aber nicht kontextfrei. Beweis: Wir zeigen, daß L nicht kontextfrei is t. Angenommen, L is t kontextfrei und p und q s eien natürliche Zahlen derart, daß (i) bis (iii) von Satz 4.2.3.4. erfüllt s ind. Sei z1 ∈ L, z1 = akbkck mit k > p/3. Dann gibt es Teilketten w1, ..., w5 von z1 derart, daß z1 = w1w2w3w4w5, wobei w2w4 ≠ ε und alle Ketten zi = w1(w2)iw3(w4)iw5 mit i 0 aus L s ind. Dann können w2 und w4 nur Ketten der Form ag, bh oder cj mit g, h, i  0 s ein, da s ons t in (w2)2 und (w4)2 a-, b- oder c-Vorkommen in gemis chter Reihenfolge auftreten, was L nicht zuläßt. Da w2w4 ≠ ε und mit wachs endem i die Länge von (w2)i(w4)i zunimmt, anderers eits w1, w3 und w5 unverändert bleiben, gehören auch zi zu L mit unglei-

cher Anzahl der a-, b- und c-Vorkommen. Da es in L keine s olchen Ketten gibt, is t die Annahme, L s ei kontextfrei, zu einem Widerspruch geführt. L ist also nicht kontextfrei. Wir zeigen, daß L vom Typ 1 is t. Dazu kons truieren wir eine Typ-1-Syntax, deren Sprache L is t. Die Syntax is t G = ({S, A, B}, {a, b, c}, R, S) mit

G is t vom Typ 1. Die Regeln (1) und (2) erzeugen alle Ketten der Form an+1BAn mit n  0. Hierauf können nur die Regeln (3) und (5) angewendet werden. Falls n  1, führt (5), worauf nur noch n-mal (4) anwendbar is t, in eine Sackga ss e. Mit (3) erhält man an+1bBcAn—1, durch (n—1)-malige Anwendung von (4) an+1bBAn—1c. Daraufhin kann wieder nur (3) benutzt werden, ohne daß man in eine Sackgas s e gerät, dann (n—2)mal (4) us w., s o daß s ich s chließlich an+1bnBcn und durch (5) an+1bn+1cn+1 ergibt. Für n = 0 erhalten wir abc durch (5). Dies s ind die einzigen Ableitungs möglichkeiten für Ketten aus L(G), und damit gilt L(G) = L. Damit is t der Satz bewiesen. Ableitungen, Bäume, Ambiguität: Is t G = (VN, VT, R, S) eine kontextfreie Syntax und x ∈ L(G), s o gibt es normalerweis e vers chiedene Ableitungen S x. Befinden s ich nämlich in einer Ableitungs zeile mehrere Vorkommen von Nicht-Terminals ymbolen, können dies e in vers chiedener Reihenfolge abgearbeitet werden. Beispiel: Sei G = ({S, A}, {a, b}, {S → AS, S → AA, S → b, A → a}, S). Es gilt S ab mit den Möglichkeiten S → AS → aS → ab und S → AS → Ab → ab. Über die Ableitungen wird Endketten eine Strukturbes chreibung in Form eines Generationsbaumes (kurz: Baum) zugeordnet. Definition 4. 2. 3. 3. (Baum): Sei ein Paar (K, E) gegeben mit K und E endlichen nicht-leeren Mengen. Die Elemente von K heißen Knoten. Die Elemente von E s ind Paare (k1, k2) von Knoten. Sie heißen Kanten. Die Kante e = (k1, k2) geht von k1 aus und führt nach k2. Eine endliche Folge von Kanten e1, ..., en heißt ein Pfad gdw die zweite Komponente von ei gleich der ers ten von ei+1 für jedes i mit 1 ≤ i < n is t. Ein Paar (K, E) heißt ein Baum gdw (i) es genau einen Knoten gibt, zu dem keine Kante führt (dieser Knoten heißt die Wurzel des Baumes), und

1588

(ii) es für jeden Knoten k ungleich der Wurzel genau einen Pfad von der Wurzel nach k gibt. Die Knoten, von denen keine Kante aus geht, heißen Blätter. Sei nun G = (VN, VT, R, S) eine kontextfreie Syntax und x1 → ... → xr mit x1 ∈ VN einer Ableitung in G. Wir kons truieren einen Baum dafür in folgender Weis e. Die Knoten des Baumes werden durch n-Tupel natürlicher Zahlen darges tellt und bekommen als Etikette terminale oder nicht-terminale Symbole zugewies en. Die Wurzel is t das 1-Tupel (1) und erhält x1 als Etikett. Für jedes i (1 ≤ i ≤ r—1) s ei Bi → yi eine Regel aus R, für die xi = uiBivi und xi+1 = uiyivi gilt. Sei dies es Bi-Vorkommen bereits Etikett eines Knotens und s ei dies er durch das m-Tupel (z1, ..., zm) bezeichnet. Für yi = a1 ... ak, worin a1, ..., ak ∈ VN ⋃ VT, wird jedes aj (1 ≤ j ≤ k) Etikett eines neuen Knotens (z1, ..., zm, j). Außerdem wird die Kante (z1, ..., zm), (z1, ..., zm, j)) gebildet. Auf dies e Weis e erhalten wir für die Ableitungen in obigem Beispiel den Baum

Vers chiedenen Ableitungen kann als o ders elbe Baum zugeordnet s ein. Umgekehrt gibt es aber auch den Fall, daß einer Ableitung mehrere Bäume zugeordnet s ind, wie folgendes Beispiel zeigt. Beispiel: G = ({S}, {a}, {S → SS, S → a}, S). Die Ableitung S → SS → SSS → aSS → aaS → aaa hat die zugeordneten Bäume

und

XXIII. Service-Artikel

(wir erwähnen nur die Etikette und gehen davon aus , daß benachbarte Knoten von links nach rechts geordnet sind). Um zu erreichen, daß jeder Ableitung genau ein Baum zugeordnet wird, führen wir eine Normierung des Ableitungs prozes s es in Form der Linksableitung ein. Definition 4. 2. 3. 4. (Links ableitung): Sei G = (VN, VT, R, S) eine kontextfreie Syntax. Eine Ableitung x1 → ... → xn in G heißt eine Linksableitung gdw jedes xi und xi+1 (1 ≤ i ≤ n—1) die Form xi = yAz und xi+1 = yuz mit A ∈ VN, u, z ∈ (VN ⋃ VT)* und y ∈ VT* haben und A → u aus R ist. In einer Links ableitung wird als o immer das am weites ten links s tehende Nicht-Terminals ymbol wiederges chrieben. Mit x1 = B für ein B ∈ VN erhalten wir für jede Links ableitung genau einen Baum und umgekehrt is t jedem Baum genau eine Links ableitung zugeordnet. Außerdem gibt es offens ichtlich für jedes x ∈ L(G) eine Links ableitung S x, s o daß die gesamte Sprache erhalten wird. Auf dies er Bas is kann s innvoll der Ambiguitätsbegriff eingeführt werden. Definition 4. 2. 3. 5. (ambige Syntax): Eine kontextfreie Syntax G = (VN, VT, R, S) heißt ambig gdw es für ein x ∈ L(G) mehr als eine Links ableitung in G gibt, s o daß S x. Andernfalls heißt G nicht ambig (oder eindeutig). Beispiel einer ambigen Syntax: G = ({S}, {a}, {S → SS, S → a}, S). Für aaa gibt es die beiden Links ableitungen S → SS → aS → aSS → aaS → aaa und S → SS → SSS → aSS → aaS → aaa. Definition 4. 2. 3. 6. (ambige Sprache): Eine kontextfreie Sprache L heißt ambig (auch inhärent ambig) gdw jede kontextfreie Syntax G mit L(G) = L ambig is t. Andernfalls heißt L nicht ambig (oder eindeutig). Das Vorhandens ein einer ambigen Syntax bes agt nicht, daß auch die davon erzeugte Sprache ambig is t. Man kann s ogar auf einfache Weis e zu einer beliebig gegebenen kon-

87.  Formale Sprachen

textfreien Syntax eine ambige, die dies elbe Sprache erzeugt, kons truieren. Doch gibt es auch ambige kontextfreie Sprachen. Eine s olche is t L = {ambncn  m, n  1} ⋃ {ambmcn  m, n  1}. Zum Beweis vgl. Salomaa (1973, 205). 4.2.4. Einseitig-lineare Syntaxen Die eins eitig-linearen oder Typ-3-Syntaxen s ind s pezielle kontextfreie Syntaxen einfacher Bauart. Sie s tehen in einer Äquivalenz-Beziehung zu den endlichen Automaten: beide Formalis men legen dies elben Sprachen fes t (vgl. Kap. 4.2.6.). Endliche Automaten wurden u. a. zur Bes chreibung morphologis cher Sys teme (Kos kenniemi 1983) verwendet, bes onders aber für die Modellierung neurolinguis tis cher Prozes s e vorges chlagen, wobei die Gehirns truktur als endlicher Automat ges ehen wird (Kleene 1956). Äquivalent las s en s ich hierfür auch eins eitig-lineare Syntaxen nehmen. Die Frage, ob dies e Formalis men für die Bes chreibung natürlicher Sprachen aus reichen, wurde von Choms ky (1957) aufgeworfen und verneint, doch is t die Dis kus s ion anges ichts neuer Ergebnisse wieder aufgelebt. Definition 4. 2. 4. 1. (lineare eins eitig-lineare, abs chließende Regel): Sei G = (VN, VT, R, S) eine kontextfreie Syntax. Eine Regel aus R heißt linear gdw s ie die Form A → xBy mit A, B ∈ VN und x, y ∈ VT* hat. Sie heißt rechts-linear gdw y = ε, links-linear gdw x = ε. Sie heißt abschließend gdw s ie von der Form A → x mit x ∈ VT* ist. Definition 4. 2. 4. 2. (lineare Syntax, Typ-3Syntax): Eine kontextfreie Syntax G = (VN, VT, R, S) heißt linear gdw jede Regel aus R linear oder abs chließend is t. Sie heißt rechtslinear (links-linear) gdw jede Regel aus R rechts -linear (links -linear) oder abs chließend is t. Rechts -lineare und links -lineare Syntaxen zus ammen heißen einseitig-linear oder vom Typ 3. Die von einer Typ-3-Syntax erzeugte Sprache heißt Typ-3-Sprache (oder einseitig-lineare Sprache). Beispiel: (a) Sei G = ({S, A, B, C}, {a, b, c}, {S → aaS, S → bA, A → aaA, A → cB, B → bbC, C → c}, S). G is t rechts linear. L(G) = {(aa)mb(aa)ncbbc  m, n  0}. (b) Sei G = ({S, A, B, C, D}, {Katze, Schlange, Maus , frißt, beißt, s ieht, die}, R, S} mit R = {S → die A, A →Katze B, A → Maus B, A → Schlange B, B → beißt C, B → frißt C, B → s ieht C, C → die D, D → Maus , D → Katze, D → Schlange}. G is t rechts -li-

1589

near und erzeugt L(G) = {die} {Katze, Maus , Schlange} {beißt, frißt, s ieht} {die} {Katze, Maus, Schlange}. Einen Sub-Typu s der ein s eitig-linearen Syntaxen bilden die regulären Syntaxen. Definition 4. 2. 4. 3. (reguläre Syntax): Eine rechts -lineare (links -lineare) Syntax G = (VN, VT, R, S) heißt rechts-(links-)regulär gdw jede Regel aus R folgende Form hat: (1) S → ε oder (2) A → aB (A → Ba für links-regulär) oder (3) A → a mit A, B ∈ VN und a ∈ VT, wobei, falls S → ε aus R, B ≠ S gilt. Die Vereinigungs menge der rechts - und links regulären Syntaxen is t die Menge der regulären Syntaxen. Es gelten folgende Sätze: Satz 4. 2. 4. 1. : Die rechts -linearen und die link s -linearen Syntaxen erzeugen die s elbe Sprachenmenge. M. a. W.: zu jeder rechts linearen Syntax G1 gibt es eine links -lineare Syntax G2 derart, daß L(G1) = L(G2). Satz 4. 2. 4. 2. : Zu jeder eins eitig-linearen Syntax G1 gibt es eine reguläre Syntax G2 derart, daß L(G1) = L(G2). Satz 4. 2. 4. 3. : Die Menge der Typ-3-Sprachen (L3) is t eine echte Teilmenge der Menge der Typ-2-Sprachen (L2 ⊃ L3). Beweisskizze von Satz 4. 2. 4. 1. : Man beweis t, daß es zu jeder rechts -linearen Syntax G1 eine äquivalente links -lineare G2 gibt und umgekehrt. Sei G1 = (VN1, VT, R1, S1). Nach Satz 4.2.3.2. können wir ohne Bes chränkung der Allgemeinheit davon aus gehen, daß R1 keine Regeln der Form A → B mit A, B ∈ VN1 und als Tilgungs regel höchs tens S1 → ε enthält (andernfalls können wir eine s olche Syntax kons truieren und in G1 umbenennen). Wir kons truieren G2 = (VN2, VT, R2, S2) folgendermaßen: S2 is t ein neues Zeichen, S2 ∉ VN1 und VN2 = VN1 ⋃ {S2}. (1) Zu jeder Regel der Form A → xB aus R1 mit B ∈ VN, x ∈ VT* gibt es in R2 die Regel B → Ax. (2) Zu jeder Regel der Form S1 → xB aus R1 mit x ∈ VT* gibt es in R2 B → x. (3) Zu jeder Regel der Form A → x aus R1 mit x ∈ VT+ gibt es in R2 S2 → Ax. (4) Für jede Regel der Form S1 → x aus R1 mit x ∈ VT* ist S2 → x in R2. Damit is t R2 kons truiert. Man zeigt nun leicht, daß es für jedes x aus L(G1) eine Ablei-

XXIII. Service-Artikel

1590

tung in G2 gibt derart, daß S2

x, und um-

gekehrt, daß es für jedes x aus L(G2) S1 x gibt. Sei nun G2 = (VN2, VT, R2, S2) eine links lineare Syntax (wieder ohne Regeln der Form A → B, worin höchs tens S2 → ε als Tilgungs regel auftritt). Sei G1 = (VN1, VT, R1, S1) mit S1 ∉ VN2, VN1 = VN2 ⋃ {S1}, wobei R2 s piegelbildlich zu (1) bis (4) oben kons truiert wird: (1) Zu jedem A → Bx ∈ R2 gibt es B → xA ∈ R1. (2) Zu jedem S2 → Bx ∈ R2 gibt es B → x ∈ R1. (3) Zu jedem B → x ∈ R2 mit x ∈ VT+ gibt es S1 → xB ∈ R1. (4) Für jedes S2 → x ∈ R2 mit x ∈ VT* ist S1 → x ∈ R1. Man zeigt nun analog zum obigen Fall, daß L(G1) = L(G2). Beweisskizze von Satz 4. 2. 4. 2. : Sei G1 = (VN1, VT, R1, S1). Auf Grund der Sätze 4.2.4.1., 4.2.3.1. und 4.2.3.2. können wir ohne Bes chränkung der Allgemeinheit davon aus gehen, daß G1 rechts -linear is t und keine Regeln der Form A →B mit B ∈ VN1 und A → ε mit A ≠ S1 enthält. Sei G2 = (VN2, VT, R2, S) mit R2, wie folgt: Die Regeln aus R1 der Form A → xB und A → x mit x ∈ VT gehören auch zu R2. Für jede Regel r aus R1 der Form A → xB oder A → x mit x = a1 ... an, (n > 1) und a1, ..., an ∈ VT werden n—1 neue Nicht-Terminals ymbole A1, ..., An—1 eingeführt und die Regeln A → a1A1, ..., An—2 → an—1An—1 s owie im Fall A → xB die Regel An—1 → anB und im Fall A → x die Regel An—1 → an zu R2 gefügt. Dies e Regeln liefern das s elbe Ableitungs ergebnis wie r. Damit is t R2 kons truiert. VN2 is t die Vereinigungs menge aus VN1 und der Menge aller neu eingeführten Nicht-Terminalen. Man s ieht leicht, daß es für jedes x aus L(G1) eine Ableitung S x in G2 gibt und umgekehrt für jedes x aus L(G2) eine Ableitung S1 x in G1, somit gilt L(G1) = L(G2). Beweis von Satz 4. 2. 4. 3. : Der Satz bes agt, daß es eine kontextfreie Sprache gibt, die nicht eins eitig-linear is t. Behauptung: L = {anban  n  0} is t eine s olche Sprache. L wird, wie man unmittelbar s ieht, durch die kontextfreie Syntax G = ({S}, {a, b}, {S → aSa, S → b}, S) erzeugt. Angenommen, L is t eins eitig-linear. Sei G′ = (VN′, VT, R′, S) eine rechts -lineare Syntax derart, daß L(G′)

= L. Sei R″ = {A → x  A → x ∈ R′ ⋀ Länge (x)  2 }. Sei 1max die maximale Länge der rechten Seiten der Regeln aus R″, u = card (R″) und k eine beliebige natürliche Zahl derart, daß 2k > 1max · u. Dann werden in jeder Ableitung S → akbak mindes tens (u+1)-mal Regeln aus R″ angewendet, d. h. mindes tens eine Regel zweimal. Es gibt als o in G′ eine Ableitung S xA xyA xyz mit xyz = akbak, wobei A ∈ VN, x, z ∈ VT* und y ∈ VT+. Dann is t auch xyiz für jedes i > 1 aus L. In y kann kein b vorkommen, da dann in yi i-mal b vorkommen würde. Als o gilt y = aj mit j > 0. Da xyz = akbak, gilt für jedes i > 1 entweder xyiz = akaj(i—1)bak oder xyiz = akbakaj(i—1). Solche Ketten s ind aber nicht aus L, als o is t die Annahme, L s ei eins eitiglinear, zu einem Widerspruch geführt. 4.2.5. Abgeschlossenheitseigenschaften formaler Sprachen Formale Sprachen s ind Mengen, damit s ind über ihnen die mengentheoretis chen Operationen wie Vereinigung, Durch s chnitt s bildung, Komplementbildung anwendbar. Dazu kommen, da s ie Mengen von Zeichenketten s ind, die Operationen der Verkettung, Sternbildung (vgl. Kap. 3.) und Eins etzung. Das Res ultat der Anwendung einer s olchen Operation is t wieder eine formale Sprache. Hier interes s iert nun, welche Eigens chaften die res ultierende Sprache hat, und bes onders , welche Eigens chaften der Aus gangs s prachen beibehalten werden. Die Unters uchung dies es Fragenkreis es hat zur Entdeckung mathematis cher Strukturen geführt, deren Kenntnis für die Kons truktion von Syntaxen hilfreich s ein kann. Als neues Gebiet hat s ich hieraus die Theorie der abs trakten Sprachenfamilien (AFLs , abstract families of languages) entwikkelt. Eine gute Einführung gibt Salomaa (1973). Sei op eine Operation und L eine Menge von formalen Sprachen, auf die op anwendbar is t. L heißt abgeschlossen bzgl. op gdw eine Anwendung von op auf Sprachen aus L als Resultat wieder eine Sprache aus L ergibt. Die drei Operationen Vereinigung, Verkettung und Sternbildung werden unter der Bezeichnung reguläre Operationen zusammengefaßt. Es gilt Satz 4. 2. 5. 1. : Jede der Sprachenmengen Li (0 ≤ i ≤ 3) is t abges chlos s en bzgl. der regulären Operationen. Wie der nachs tehende Beweis zeigt, is t es damit möglich, Syntaxen in Teilen zu kons truieren, wie es im Bereich der natürlichen Sprachen gehandhabt wird. Liegt zum Bei-

87.  Formale Sprachen

s piel eine kontextfreie Syntax für Deklarativs ätze und eine kontextfreie Syntax für Frages ätze vor, s o läßt s ich daraus eine kontextfreie Syntax für die Ges amtmenge dies er Sätze kons truieren. Liegen Syntaxen des s elben Typs i für vers chiedene s yntaktis che Kategorien wie Nominalphras e, Verbalphras e u. a. vor, s o kann daraus eine Syntax des s elben Typs i für Verkettungs produkte dies er Kategorien gewonnen werden, mithin auch für Sätze. Beweis von Satz 4. 2. 5. 1. : Seien zwei Sprachen L1 und L2 von dems elben Typ i. Der Satz bes agt, daß dann auch L1 ⋃ L2, L1L2 und L1* vom Typ i s ind. Sei G1 = (VN1, VT1, R1, S1) und G2 = (VN2, VT2, R2, S2) mit L(G1) = L1 und L(G2) = L2. Wir nehmen an, daß VN1 und VN2 dis junkt s ind, was durch Umbenennung von Zeichen immer erreicht werden kann. Teil 1: Abges chlos s enheit bzgl. der Vereinigung: Wir kons truieren eine Syntax G3 = (VN3, VT3, R3, S3) derart, daß L(G3) = L1 ⋃ L2, wie folgt: VN3 = VN1 ⋃ VN2 ⋃ {S3}, wobei S3 ∉ VN1 ⋃ VN2. VT3 = VT1 ⋃ VT2. R3 = R1 ⋃ R2 ⋃ {S3 → S1, S3 → S2}. G3 is t von dems elben Typ i wie G1 und G2. Aus S3 werden nur S1 und S2 direkt abgeleitet und dies e erzeugen wegen der Dis junktheit der nicht-terminalen Alphabete L(G1) und L(G2). Daher L(G3) = L1 ⋃ L2, und L(G3) is t vom Typ i. Teil 2: Abges chlos s enheit bzgl. der Verkettung: (a) L1 und L2 s ind von dems elben Typ i, 0 ≤ i ≤ 2: Auf Grund der Sätze der vorigen Kapitel kann ohne Bes chränkung der Allgemeinheit davon aus gegangen werden, daß L1 und L2 durch s eparierte Syntaxen G1 und G2 des Typs i erzeugt werden. Sei G3 = (VN3, VT3, R3, S3) mit VN3 = VN1 ⋃ VN2 ⋃ {S3}, S3 ∉ VN1 ⋃ VN2, VT3 = VT1 ⋃ VT2 und R3 = R1 ⋃ R2 ⋃ {S3 → S1S2}. G3 is t vom Typ i. Wegen S3 → S1S2, der Dis junktheit von VN1 und VN2 s owie der Separiertheit von G1 und G2 is t ein (möglicherweis e leeres ) Anfangs s tück jeder von G3 erzeugten Kette k aus L1 und das nach Abzug dies es Anfangs s tücks übrigbleibende Ends tück von k aus L2. Ferner werden alle Ketten aus L1L2 erzeugt. Somit gilt L(G3) = L1L2, und L(G3) ist vom Typ i. (b) L1 und L2 s ind vom Typ 3: Auf Grund von Satz 4.2.4.1. werden L1 und L2 durch rechts -lineare Syntaxen G1 und G2 erzeugt. Sei G3 = (VN3, VT3, R3, S1) mit VN3 = VN1 ⋃ VN2, VT3 = VT1 ⋃ VT2 und R3 = (R1 —

1591

{A → x  A → x ∈ R1 ⋀ x ∈ VT1*}) ⋃ R′ ⋃ R2, wobei R′ = {A → xS2  A → x ∈ R1 ⋀ x ∈VT1*}. Jede abs chließende Regel A → x aus R1 wird als o durch A → xS2 ers etzt. G3 is t vom Typ 3. Sei S1 y eine Ableitung in G3 mit y ∈ L(G3). Dann is t ein Anfangs s tück von y durch Regeln aus R1 erzeugt und das nach Abzug dies es Anfangs s tücks übrigbleibende Ends tück durch Regeln aus R2. R2 kommt ers t zur Anwendung, wenn S2 erzeugt is t. Wir erhalten s omit in G3 eine Ableitung S uS2 uv = y mit u ∈ L1 und v ∈ L2. Als o gilt L(G3) ⊆ L1L2. Umgekehrt gilt für jedes y ∈ L1L2 y = uv mit S1 u, als o S1 uS2, und S2 v, als o S1 uS2 uv. Daher L1L2 ⊆ L(G3). Als o gilt L(G3) = L1L2, und L1L2 ist vom Typ 3. Teil 3: Abges chlos s enheit bzgl. der Sternbildung: (a) L1 vom Typ 3: Sei G1 rechts -linear und G = (VN1, VT1, R, S1) mit R = R1 ⋃ R′ ⋃ {S1 → ε}, worin R′ = {A → xS1  A → x ∈ R1 ⋀ x ∈ VT1*}. G is t vom Typ 3. Durch R1 wird L1 und durch R′ werden alle möglichen Verkettungen von Elementen aus L1 erzeugt. Da auch ε ∈ L(G), is t offens ichtlich L(G) = L1*. L1* ist also vom Typ 3. (b) L1 vom Typ 2: Sei G1 kontextfrei und G = (VN1 ⋃ {S}, VT1, R, S) mit S ∉ VN1 und R = R1 ⋃ {S → SS1, S → ε}. Da durch S → SS1 und S → ε beliebig lange Ketten S1 ... S1 erzeugt werden und da G auch ε erzeugt, gilt L(G) = L1*. G is t vom Typ 2, als o ist L1* vom Typ 2. (c) L1 vom Typ 1: Sei G1 vom Typ 1 und L1 = L(G1). Um ein G mit L(G) = L1* zu kons truieren, läßt s ich nicht wie in (b) verfahren, da es dann eine Ableitung der Form S1S1 xuvy → xwy mit x, u, v, w, y Ketten und u, v ungleich ε geben kann, die durch S1 → xu und S1 → vy ents teht. Durch die Anwendung der Regel uv → w kann die Ableitung zu einer Endkette führen, die nicht aus L1* is t. Sei G1 wieder s epariert, und s ei G1′ = (VN1′, VT1, R1′, S1′). Dabei bes teht VN1′ aus allen Zeichen A′, für die A ∈ VN1. VN1 und VN1′ s ind s omit dis junkt. Für S1 ∈ VN1 is t S1′ ∈ VN1′. Die Regeln aus R1′ ents tehen dadurch, daß für jedes A ∈ VN1 alle A-Vorkommen in den Regeln aus R1 durch A′ ers etzt werden und die übrigen Zeichen an ihrer jeweiligen Stelle erhalten bleiben. Sei G = (VN, VT1, R, S) mit VN = VN1 ⋃ VN1′ ⋃ {S}, S ∉ (VN1 ⋃ VN1′) und R = (R1 ⋃ R1′ ⋃ R′) — {S1 → ε, S1′ → ε}, worin

1592

R′ = {S → S1, S → S1S1′, S → S1S1′S, S → ε}. G is t vom Typ 1. Durch R′ werden S1 und beliebig lange Ketten S1S1′ ... S1S1′, S1S1′ ... S1S1′S1 erzeugt. Wegen der Dis junktheit von VN1 und VN1′ und da auch ε ∈ L(G), gilt offens ichtlich L(G) = L1*. L1* is t als o vom Typ 1. (d) L1 vom Typ 0: Sei G1 vom Typ 0 und L1 = L(G1). Um ein G mit L(G) = L1* zu kons truieren, können wir nicht wie in (c) verfahren, da es eine Ableitung S1 ε geben kann, worin ε über mehrere Ableitungs s chritte erzeugt wird. Sei G = (VN, VT1, R, S) mit VN = VN1 ⋃ {S, X, Y, Z, U}, {S, X, Y, Z, U} ⋂ VN1 = ∅, R = R1 ⋃ R′, worin R′ = {S → XYZ, Y → S1UY, Y → ε, UZ → Z, XZ → ε} ⋃ {aZ → Za  a ∈ VT1}. In jeder Ableitung S x mit x ∈ L(G) werden zunächs t die Regeln S → XYZ, Y → S1UY und Y → ε angewendet. Sie erzeugen alle Ketten X(S1U)nZ mit n  0. Falls n = 0, kommt XZ → ε zur Anwendung und ε ∈ L(G) wird geliefert. Falls n ≠0, können nur UZ → Z und Regeln aus R1 benutzt werden. Die Regeln der Form aZ → Za las s en Z nach links wandern, bis ein Zeichen aus VN direkt vor Z s teht. Das nächs te U links wird daher nur erreicht, wenn aus dem ihm in X(S1U)nZ rechts direkt folgenden S1 eine Endkette, als o ein Element aus L(G1), abgeleitet wurde. Ers t dann kann wieder UZ → Z angewendet werden. Die U-Vorkommen wirken s omit als Trenner zwis chen den aus den S1 abgeleiteten Ketten. Is t Z bis vor X gelangt, erhält man durch XZ → ε ein Element aus L(G). Jede s olche Endkette is t aus L1* und, da von S1 aus jedes x ∈ L(G1) ableitbar is t, gilt L(G) = L1*. L1* ist also vom Typ 0. Die Einsetzungssätze für Typ-0 und Typ-2Sprachen: Definition 4. 2. 5. 1. (Eins etzung): Sei V ein Alphabet. Für jedes a ∈ V s ei Va ein Alphabet, und die Vereinigungs menge aller Va s ei VG. Wir definieren eine Funktion s : V* → VG*, wie folgt: (1) s(ε) = {ε}, (2) s(a) ⊆ Va* für jedes a ∈ V, (3) s(ax) = s(a)s(x) für jedes a ∈ V und x ∈ V*. Die Funktion s heißt Einsetzung (Substitution). Für jedes M ⊆ V* sei s(M) := s(x). Definition .4.2.5.2 (Homomorphismus): Seien V, Va und VG wie in Definition 4.2.5.1. Eine Eins etzung s : V* → VG* heißt Homomorphismus gdw für jedes a ∈ V s (a) genau

XXIII. Service-Artikel

ein Element enthält. Als o können Sprachen durch Eins etzungs operationen in abs trakte Mus ter, die ihrers eits formale Sprachen s ind, gewonnen werden. Im linguis tis chen Bereich s ind hier Satzmus ter, wie die präterminalen Ketten in generativen Grammatiken, zu nennen. Es gilt Satz 4. 2. 5. 2. (Eins etzungs s atz für Typ-2und Typ-0-Sprachen): Sei L eine Typ-i-Sprache über einem Alphabet VT, i ∈ {0,2}, und s ei für jedes a ∈ VT s (a) von dems elben Typ i wie L. Dann is t s (L) ebenfalls vom Typ i. Folglich s ind die Sprachenmengen L0 und L2 auch abgeschlossen bzgl. Homomorphismus. Für den Beweis gehen wir von L = L(G) mit G = (VN, VT, R, S), worin VT = {a1, ..., an}, aus. (a) Beweisskizze für Typ-2-Sprachen: Sei G vom Typ 2. Für jedes ai (1 ≤ i ≤ n) werde s (ai) durch die Typ-2-Syntax Gi = (VNi, VTi, Ri, Si) erzeugt. Wir können davon aus gehen, daß VN und die VNi paarweis e dis junkt s ind. Sei Rs = R1 ⋃ ... ⋃ Rn ⋃ R′, worin R′ aus R ents teht, indem für jedes ai alle ai-Vorkommen in den Regeln von R durch Si ers etzt werden und die übrigen Zeichen an ihren Stellen erhalten bleiben. Wir s etzen Gs = (VNs, VTs, Rs, S) mit VNs = VN ⋃ VN1 ⋃ ... ⋃ VNn und VTs = VT1 ⋃ ... ⋃ VTn. Gs is t vom Typ 2. Man zeigt leicht, daß s (L) = L(Gs). Also ist s(L) vom Typ 2. b) Beweis für Typ-0-Sprachen: Sei G vom Typ 0. Für jedes ai (1 ≤ i ≤ n) werde s (ai) durch die Typ-0-Syntax Gi = (VNi, VTi, Ri, Si) erzeugt. G und die Gi s eien s epariert und VN und die VNi paarweis e dis junkt. Sei Gs = (VNs, VTs, Rs, Ss) mit VNs = VN ⋃ VN1 ⋃ ... ⋃ VNn ⋃ {Ss, Y, A1, ..., An}, {Ss, Y, A1, ..., An} ⋂ (VN ⋃ VN1 ⋃ ... ⋃ VNn) = ∅, VTs = VT1 ⋃ ... ⋃ VTn, Rs = R1 ⋃ ... ⋃ Rn ⋃ R′ ⋃ R″ ⋃ R‴ ⋃ {Ss → YS, Y → ε}. Dabei ents teht R′ aus R, indem für jedes ai ∈ VT alle ai-Vorkommen in den Regeln aus R durch Ai ers etzt werden. Es wird R″ = {YAi → YSi  1 ≤ i ≤ n} und R‴ = {Ya → aY  a ∈ VTs} ges etzt. Wir zeigen, daß L(Gs) = s(L). Für jedes x ∈ L(G), x = a1 ... an gibt es eine Ableitung Ss YA1 ... An in Gs. YA1 muß durch YS1 ers etzt werden, worauf nur S1 durch Regeln aus R1 weiter ers etzt werden kann. Y kann nur nach rechts vers choben werden, und zwar genau dann, wenn ihm ein Ends ymbol unmittelbar folgt. Y gelangt daher unmittelbar vor A2 (falls vorhanden) gdw aus S1 ein Element von L(G1) erzeugt is t (das

87.  Formale Sprachen

auch das leere Wort s ein kann). Dies geht s o weiter mit allen nachfolgenden Aj. Steht Y am rechten Ende der Kette, s o wird es durch Y → ε getilgt, als o eine Endkette erzeugt. Aus gehend von YA1 ... An erhält man s omit das Verkettungs produkt L(G1) ... L(Gn), als o für jedes a1 ... an ∈ L das Eins etzungs ergebnis s (a1 ... an). Für ε gilt, wie man leicht s ieht, ε ∈ L(G) gdw s (ε) ∈ L(Gs). Als o gilt s (L) ⊆ L(Gs). Anderers eits werden aber durch Gs nur Elemente aus s (L) erzeugt, als o L(Gs) ⊆ s(L), also gilt der Satz für Typ-0-Sprachen. Auch die eins eitig-linearen Sprachen s ind abges chlos s en bzgl. der Eins etzung, wie in Kap. 4.2.6. gezeigt wird. Für Typ-1-Sprachen gilt der Satz einges chränkt auf ε-freie Eins etzung, d. h. falls s (a) ⊆Va+ für jedes a ∈ VT. Der Beweis für Typ-1-Sprachen erfolgt mit Hilfe des Platzbedarfssatzes. Definition 4. 2. 5. 3. (Platzbedarf): Sei G = (VN, VT, R, S) vom Typ 0. Für eine Ableitung A = (S = x1 → ... → xn = x) bzgl. G is t der Platzbedarffür xn bzgl. A definiert durch PLG(xn, A) = max {Länge (xi)  1 ≤ i ≤ n}. Der Platzbedarf für xn bzgl. G is t definiert durch PLG(xn) = min {PLG(xn, A)  A is t eine Ableitung S xn}. Es folgt unmittelbar, daß für jedes G und x PLG(x)  Länge (x). Satz .4 .2 .5 .3 (Platzbedarfss atz): Falls G eine Typ-0-Syntax und p eine natürliche Zahl is t derart, daß PLG (x) ≤ p · Länge (x) für alle x ∈ L(G) — {ε}, dann is t L(G) vom Typ 1. Der Beweis erfolgt durch volls tändige Induktion über p mittels Kons truktion geeigneter Typ-1-Syntaxen (vgl. Salomaa 1973, Maurer 1977). Mit Hilfe dies es Satzes läßt s ich die Abges chlos s enheit der Typ-1-Sprachen unter ε-freier Eins etzung beweis en. Wir brauchen nur im Beweis für Typ-0-Sprachen (Satz 4.2.5.2.) für G und die Gi Typ-1-Syntaxen zu wählen. Die Gi haben dann keine Regel Si → ε. Die einzige Regel nicht vom Typ 1 in Rs is t dann Y → ε. Es tritt aber in jeder abgeleiteten Kette höchs tens ein Y auf. Daher erfüllt Gs den Platzbedarfs s atz für p = 2, als o is t L(Gs) vom Typ 1. Die Durchschnittssätze für Typ-i-Sprachen: Satz 4. 2. 5. 4. : Jede der Sprachenmengen L0, L1 und L3 is t abges chlos s en bzgl. der Durchschnittsbildung. Satz 4. 2. 5. 5. : Die Sprachenmenge L2 ist nicht abges chlos s en bzgl. der Durchs chnitts bildung. Es gilt aber Satz 4. 2. 5. 6. : Is t L1 eine Typ-2-Sprache

1593

und is t L2 eins eitig-linear, s o is t L1 ⋂ L2 vom Typ 2. Dies e Sätze s ind wichtige Hilfs mittel, um die Zugehörigkeit von Sprachen zu Typen zu zeigen. So is t eine Sprache L nicht vom Typ 2, wenn eine nicht-kontextfreie Teilmenge von L als Durchs chnitt von L und einer eins eitig-linearen Sprache darge s tellt werden kann. Beweis von Satz 4. 2. 5. 4. für L0 und L1. Seien L1 = L(G1) und L2 = L(G2) Typ-0Sprachen mit G1 = (VN1, VT1, R1, S1) und G2 = (VN2, VT2, R2, S2) Typ-0-Syntaxen. Wir können wieder davon aus gehen, daß VN1 ⋂ VN2 = ∅. Sei Gd = (VN1 ⋃ VN2 ⋃ VN′, VT1 ⋃ VT2, R, ⋃ R2 ⋃ R′, Sd), worin VN′ = {Ai  ai ∈ VT1 ⋃ VT2} ⋃ {Sd, B1, B2}, Sd, B1, B2 ∉ VN1 ⋃ VN2 und R′ die folgenden Regeln enthält: Sd → B1S1B2S2B1, B2ai → AiB2 für jedes ai ∈ VT1 ⋃ VT2, bAi → Aib für jedes b ∈ VT1 ⋃ VT2, B1Aiai → aiB1 für jedes ai ∈ VT1 ⋃ VT 2. B1B2B1 → ε. Durch R1 und R2 werden in Gd alle Ketten der Form B1x1B2x2B1 mit x1 ∈ L(G1) und x2 ∈ L(G2) aus gehend von Sd erzeugt. Durch R′ erhält man die Kette x1B1B2B3 gdw x1 = x2, und durch B1B2B3 → ε wird x1 erzeugt. Da dies die einzig möglichen Ableitungen s ind, gilt als o: L(Gd) = L1 ⋂ L2, und L(Gd) ist vom Typ 0. Für Typ-1-Sprachen beweis t man den Satz in ders elben Weis e, indem für G1 und G2 Typ-1-Syntaxen angenommen werden. Gd enthält dann als einzige Regel mit längerer linker als rechter Seite B1B2B3 → ε. Gd erfüllt den Platzbedarfs s atz für p = 5, als o is t L(Gd) vom Typ 1. Zum Beweis für L3 vgl. Kap. 4.2.6. Beweis von Satz 4. 2. 5. 5. : Wir brauchen nur Typ-2-Sprachen L1 und L2 zu finden derart, daß L1 ⋂ L2 nicht vom Typ 2 is t. Wir wählen L1 = {anbncm  n, m 1} und L2 = {ambncn  n, m  1}. L1 = L(G1) mit G1 = ({S1, A}, {a, b, c}, R1, S1), worin R1 = {S1 → S1c, S1 → Ac, A → ab, A → aAb}, und L2 = L(G2) mit G2 = ({S2, A}, {a, b, c}, R2, S2), worin R2 = {S2 → aS2, S2 → aA, A → bc, A → bAc}. Als o s ind L1 und L2 vom Typ 2. L1 ⋂ L2 = {anbncn n  1}, nach Satz 4.2.3.5. nicht vom Typ 2. Aus Satz 4.2.5.5. folgt weiter, daß L2 nicht abges chlos s en is t bzgl. der Komplementbildung zu VT*: Nicht für jede über VT gebil-

1594

dete Sprache L vom Typ 2 is t VT* — L auch vom Typ 2. Dies folgt aus der Gleichung L1 ⋂ L2 = VT* — ((VT* — L1) ⋃ (VT* — L2)). Wäre L2 abges chlos s en bzgl. der Komplementbildung, müßte, falls L1 und L2 vom Typ 2, wegen der Abges chlos s enheit von L2 bzgl. der Vereinigung auch L1 ⋂ L2 vom Typ 2 sein. Dies widerspricht aber Satz 4.2.5.5. Satz 4.2.5.6. wird in Kap. 4.2.6. bewiesen. 4.2.6. Äquivalente Darstellungen einseitiglinearer Sprachen Die eins eitig-linearen Sprachen werden außer durch Typ-3-Syntaxen durch endliche Automaten und reguläre Aus drücke bes chrieben. Dies e Formalis men s ind einander äquivalent: Jede dies er Sprachen wird durch eine Typ-3Syntax, einen endlichen Automaten und einen regulären Aus druck darges tellt, wohingegen andere Sprachen nicht davon erfaßt werden. Die vers chiedenen Charakteris ierungen eins eitig-linearer Sprachen zeigen wichtige mathematis che Eigens chaften, die für die Einordnung von Sprachen in die Typ-i-Hierarchie ausgenutzt werden können. Einseitig-lineare Sprachen und endliche Automaten: Eine allgemeine Charakteris ierung von Automaten wurde bereit s in Kap. 4.1. gegeben. Automaten legen formale Sprachen fes t. Die Sprache eines Automaten A is t die Menge aller Zeichenketten x über einem gegebenen Vokabular, die aus gehend von einem Anfangs zus tand von A volls tändig verarbeitet werden, wobei die Prozedur in einem Endzus tand von A abbricht. Man s agt bzgl. eines s olchen x „x wird von A akzeptiert“ . Wichtige Automatentypen s ind die Turingmas chinen, die linear bes chränkten Automaten, die Pushdown-Automaten(auch Kellerautomaten genannt) und die endlichen Automaten, die in ders elben Beziehung zueinander s tehen wie die Typ-i-Syntaxen (0 ≤ i ≤ 3). Es gilt nämlich: Satz 4.2.6.1.: (a) Eine Sprache L ist vom Typ 0 gdw es eine Turingmaschine gibt, von der L akzeptiert wird. (b) L ist vom Typ 1 gdw es einen linear beschränkten Automaten gibt, der L akzeptiert. (c) L ist vom Typ 2 gdw es einen PushdownAutomaten gibt, der L akzeptiert. (d) L ist vom Typ 3 gdw es einen endlichen Automaten gibt, der L akzeptiert. Die Beweis e von (a) (bis d) finden s ich in den eins chlägigen Lehrwerken, z. B. Maurer

XXIII. Service-Artikel

1977. Endliche Automaten erfreuen s ich einer großen Beliebtheit als Dars tellungs mittel für eins eitig-lineare Sprachen. In der Linguis tik vers ucht man, s ie wegen ihrer einfachen Struktur und Handhabung in vers chiedenen Bereichen (z. B. Morphologie) anzuwenden. Speziell bilden s ie die Bas is der Übergangs netzwerkgrammatiken (vgl. Chri s taller/Metzing 1979). Ein endlicher Automat bes teht, ans chaulich bes chrieben, aus einem Kas ten, der Steuereinheit, die endlich viele vers chiedene Zus tände annehmen kann. An ihr befindet s ich ein Les ekopf, unter dies em ein in nebeneinanderliegende Felder aufgeteiltes Band, das Eingabeband, auf dess en Feldern Zeichen eingetragen s ind. Der Les ekopf lies t das Zeichen in dem Feld, über dem er s teht. In Abhängigkeit von dem Zeichen und dem aktuellen Zus tand der Steuereinheit rückt das Eingabeband um ein Feld nach rechts oder links , wobei die Steuereinheit in einen Übergangs zustand übergeht, oder es bleibt stehen. Bewegt s ich das Eingabeband jedes mal in dies elbe Richtung, s o liegt ein einseitiger endlicher Automat vor. Bei zweiseitigen endlichen Automaten kann s ich das Band s owohl nach links als auch nach rechts bewegen. Da die zweis eitigen endlichen Automaten dies elben Sprachen akzeptieren wie die eins eitigen, können wir uns darauf bes chränken, die einseitigen zu behandeln. Man unters cheidet zwis chen deterministischen und nicht-deterministischen endlichen Automaten, abgekürzt FDA (finite deterministic automaton) und FNA (finite non-deterministic automaton). Bei den FDA is t der Übergangs zus tand in Abhängigkeit vom geles enen Zeichen und dem aktuellen Zus tand der Steuereinheit eindeutig fes tgelegt, bei den FNA kann se mehrere mögliche Übergangs zustände geben. Unter den Zus tänden der Steuereinheit werden Startzus tände und Endzus tände aus gezeichnet. Eine Folge von Zeichen auf dem Eingabeband heißt Bandinschrift. Eine Bandins chrift x = a1 ... an wird von einem endlichen Automaten A akzeptiert gdw s ukzes s ive a1, ..., an geles en werden und dabei die Steuereinheit, die s ich bei a1 in einem Startzus tand befindet, nach Les en von an einen Endzustand erreicht. Definition 4. 2. 6. 1. (endlicher determinis tis cher Automat, FDA): Ein FDA A is t ein 5-Tupel (K, V, d, F, S). Darin s ind K und V Alphabete. Die Elemente von K heißen Zustände, die von V Eingabezeichen. Es gilt F ⊆

87.  Formale Sprachen

K, S ∈ K. F is t die Menge der Endzustände, S is t der Startzustand.d: K × V → K ⋃ {∅} is t die Übergangsfunktion.d wird auf Ketten aus V*, wie folgt, erweitert: (i) d(Z, ε) = Z, (ii) d(Z, a1 ... an) = d(d(Z, a1), a2 ... an) für jedes Z ∈ K und a1, ..., an ∈ V. Definition 4. 2. 6. 2. (Sprache eines FDA): Sei A ein FDA. L(A) := {x  x ∈ V* ⋀ d(S, x) ∈ F} heißt die Sprache von A. Für jedes x ∈ L(A) sagt man: x wird von A akzeptiert. Beispiel: Sei A = (K, V, d F, Z) ein FDA mit K = {Z, X, Y}, V = {a, b, c}, F = {Z, X} und d, wie folgt: d(Z, a) = Z, d(Z, b) = X, d(Z, c) = X, d(X, a) = Y, d(X, b) = ∅, d(X, c) = X, d(Y, a) = ∅, d(Y, b) = Z, d(Y, c) = Z. L(A) = (a* ⋃ a* {b, c} c* ⋃ a* {b, c} c* {a} {b, c})*. L(A) enthält also auch ε. Definition 4. 2. 6. 3. (endlicher nichtdeterminis tis cher Automat, FNA): Ein FNA A is t ein 5-Tupel (K, V, d F, X) mit K, V, F wie in Definition 4.2.6.1. und X ⊆ K, der Menge der Startzus tände. Für jedes Paar (Z, a) aus K × V gilt d(Z, a) ⊆ K. Definition 4. 2. 6. 4. (Sprache eines FNA): Eine Kette a1 ... an mit a1, ..., an ∈ V wird von einem FNA A akzeptiert gdw es Zus tände Z1, ..., Zn+1 ∈ K gibt derart, daß Z1 ∈ X, Zi+1 ∈ d(Zi, ai) für 1 ≤ i ≤ n und Zn+1 ∈ F. Die Menge aller von A akzeptierten Ketten heißt die Sprache von A, notiert L(A). Definitions gemäß is t ε ∈ L(A) gdw ein Startzustand zugleich ein Endzustand ist. Beispiel: Sei A = (K, V, d, F, X) ein FNA mit K = {S, T, C, B}, V = {a, b, c}, F = {B, T}, X = {S} und d, wie folgt: d(S, a) = {C, T}, d(S, b) = ∅, d(S, c) = ∅, d(T, a) = ∅, d(T, b) = ∅, d(T, c) = ∅, d(C, a) = ∅, d(C, b) = {C, T}, d(C, c) = {B, T}, d(B, a) = {T}, d(B, b) = {B}, d(B, c) = ∅. L(A) = {a} ⋃ {a} b* {b, c} ⋃ {a} b* {c} b* ⋃ {a} b* {c} b* {a}. Es folgt unmittelbar, daß jeder FDA zugleich ein FNA is t. Bemerkens wert is t, daß beide Automatentypen die s elbe Sprachenmenge festlegen, es gilt Satz 4. 2. 6. 2. : Für jeden FNA A1 gibt es einen FDA A2, derart daß L(A1) = L(A2), und vice versa. Beweisskizze: Wir brauchen nur noch zu zeigen, daß es für jeden FNA A1 einen FDA A2 gibt. Sei A1 = (K, V, d, F, S), K′ die Menge aller nicht-leeren Teilmengen von K,

1595

F′ = {Ki′  Ki′ ∈ K′ ⋀ Ki′ ⋂ F ≠ ∅}, und s ei für jedes Ki′ ∈ K′ d′ (Ki′, a) die Vereinigungs menge aller d(Z, a), worin Z ∈ Ki′ und a ∈ V. Mit A2 = (K′, V, d′, F′, {S}) erhalten wir einen FDA, der dies elbe Sprache wie A1 akzeptiert. Die Zus tände s ind hier durch Teilmengen von K bezeichnet, die durch neue Zustandsnamen ersetzt werden können. Beispiel (Kon s truktion eine s zu einem FNA äquivalenten FDA): Sei der FNA A1 = ({B, C}, {a, b}, d, {B, C}, {B}) mit d(B, a) = {B, C}, d(B, b) = {C}, d(C, a) = ∅, d(C, b) = {B, C}. Dann is t der ges uchte FDA A2 = ({{B}, {C}, {B, C}}, {a, b}, d′, {{B}, {C}, {B, C}},{ {B}}) mit d′, wie folgt: d′({B}, a) = {B, C}, d′({B}, b) = {C}, d′({C}, a) = ∅, d′({C}, b) = {B, C}, d′({B, C}, a) = {B, C}, d′({B, C}, b) = {B, C}. Als neue Zus tands namen führen wir noch E1 für {B}, E2 für {C} und E3 für {B, C} ein. Wir s ind nun in der Lage, Satz 4.2.6.1.(d), den s og. Haupts atz für endliche Automaten, zu beweisen. Beweis: Sei L eine eins eitig-lineare Sprache. Nach Satz 4.2.4.2. können wir davon aus gehen, daß L = L(G), wobei G rechts regulär is t. Sei G = (VN, VT, R, S). Wir nehmen zunächs t an, daß ε ∉ L. Wir kons truieren den FNA A = (K, V, d, F, X) mit K = VN ⋃ {T}, worin T ∉ VN, V = VT, F = {T} und X = {S}. Für jede Regel r ∈ R der Form B → aC gilt C ∈ d(B, a). Is t r von der Form B → a, so T ∈ d(B, a). Damit ist d gegeben. Behauptung: L(G) = L(A). Beweis der Behauptung: Sei a1 ... an ein beliebiges Element aus L(G) mit a1, ..., an ∈ VT. Dann ents teht a1 ... an durch s ukzes s ive Anwendung von Regeln S → a1C1, C1 → a2C2, ..., Cn—1 → an mit C1, ..., Cn—1 ∈ VN. Ents prechend gilt für den FNA A C1 ∈ d(S, a1), C2 ∈ d(C1, a2), ..., T ∈ d(Cn—1, a1). Als o wird a1 ... an von A akzeptiert. Es gilt als o L(G) ⊆L(A). Sei nun a1 ... an ein beliebiges Element aus L(A). Dann gilt C1 ∈ d(S, a1), C2 ∈ d(C1, a2), ..., T ∈ d(Cn—1, an) für irgendwelche C1, ..., Cn—1 ∈ K. Dies e Übergänge ents tehen auf Grund von Regeln S → a1C1, C1 → a2C2, ..., Cn—1 → an aus R. Als o ist a1 ... an aus L(G). Daher L(A) ⊆ L(G). Sei nun ε ∈ L(G). Es wird X = {S, T} ges etzt und der Übergang T ∈ d(T, ε) eingeführt. Dann is t ε auch aus L(A). Anderers eits is t dies die einzige Möglichkeit für A, ε zu

1596

akzeptieren, s o daß ε ∈ L(G) gdw ε ∈ L(A). Damit gilt L(G) = L(A). Sei nun L = L(A) mit A = (K, V, d, F, X) einem FNA. Ohne Bes chränkung der Allgemeinheit können wir davon aus gehen, daß K ⋂ V = ∅ (was durch Umbenennung immer erreichbar is t). Wir kons truieren die rechts reguläre Syntax G = (VN, VT, R, S1) mit VN = K ⋃ {S1}, wobei S1 ∉ K, und VT = V und d, wie folgt: Für jedes d(B, a) = {B1, ..., Bm} für irgendwelche Zus tände B, B1, ..., Bm ∈ K und a ∈ VT gilt: (1) Die Regeln B → aB1, ..., B → aBm sind aus R. (2) Für Bi ∈ F (1 ≤ i ≤ m) ist auch B → a aus R. (3) Für B ∈ X sind dazu auch S1 → aB1, ..., S1 → aBm aus R. (4) Für B ∈ X und Bi ∈ F (1 ≤ i ≤ m) ist dazu S1 → a aus R. Is t ε ∈ L(A), s o is t S1 → ε aus R. Behauptung: L(G) = L(A). Beweis der Behauptung: Sei a1 ... an ∈ L(G) mit n > 1 und a1, ... an ∈ VT. Dann wird a1 ... an durch s ukzes s ive Anwendung von Regeln S1 → a1C1, C1 → a2C2, ..., Cn—1 → an aus R erzeugt. Die Regel S1 → a1C1 kann nur gemäß (3) ents tanden s ein, es gibt als o ein B ∈ X, für das C1 ∈ d(B, a1). Die Regel Cn—1 → an geht auf (2) zurück, d. h., es gibt ein B ∈ F, s o daß B ∈ d(Cn—1, an). Die übrigen Regeln s ind nach (1) ents tanden, für jedes Ci → ai+1Ci+1 gilt Ci+1 ∈ d(Ci, ai+1. Daher res ultiert a1 ... an ∈ L(A). a ∈ L(G) mit a ∈ VT wird durch S1 → a erzeugt. Dies e Regel ents teht gemäß (4), d. h., es gibt ein B ∈ X und ein C ∈ F, s o daß C ∈ d(B, a). Daher a ∈ L(A). ε ∈ L(G) ents teht durch S1 → ε. Dies e Regel tritt nur auf, wenn ε ∈ L(A). Für alle x ∈ L(G) gilt s omit x ∈ L(A) und also L(G) ≤ L(A). Sei nun a1 ... an ∈ L(A) mit n 1. Dann gibt es Zus tände C1, ..., Cn ∈ K, Cn ∈ F und C ∈ X, s o daß C1 ∈ d(C, a1), C2 ∈ d(C1, a2), ..., Cn ∈ d(Cn—1, an). Nach (1) s ind dann die Regeln C → a1C1, C1 → a2C2, ..., Cn—1 → anCn aus R, nach (2) auch Cn—1 → an, nach (3) auch S1 → a1C1. Für n = 1 is t nach (4) S1 → a1 aus R. G erlaubt s omit die Ableitung S1 → a1C1 → ... → a1 ... an—1Cn—1 → a1 ... an, als o a1 ... an ∈ L(G). Sei nun ε ∈ L(A). Dann is t S1 → ε aus R und als o ε ∈ L(G). Als o L(A) ⊆ L(G). Damit is t die Äquivalenz von G und A gezeigt. Damit gilt der Satz. Beispiel (Kon s truktion einer zu einem FNA äquivalenten regulären Syntax): Sei der FNA A = ({S, B, C, D}, {a, b, c}, d, {S, B},

XXIII. Service-Artikel

{S, D}) mit d wie folgt: d(S, a) = {S}, d(B, a) = {C}, d(S, b) = {B, D}, d(C, b) = {C}, d(D, c) = {C}, d(C, a) = {D} und d(X, y) = ∅ für alle anderen (X, y). Dann is t die reguläre Syntax G = ({S, B, C, D, S1}, {a, b, c}, R, S1) mit R, wie folgt: S → aS, B → aC, S → bB, S → bD, C → bC, D → cC, C → aD nach (1), S → a, S → b nach (2), S1 → aS, S1 → bB, S1 → bD, S1 → cC nach (3) und S1 → a, S1 → b nach (4). Auf dies er Grundlage läßt s ich die Abges chlos s enheit der eins eitig-linearen Sprachen bzgl. des Durchs chnitts und des Komplements beweisen. Beweis von Satz 4. 2. 5. 4. für L3: Sei L eine Sprache aus L3. Dann gibt es auf Grund von Satz 4.2.6.2. einen FDA, der L akzeptiert. Sei dies er A = (K, V, d, F, S). Wir kons truieren dazu den FDA A1 = (K ⋃ {T}, V, d′, (K — F) ⋃ {T}, S) mit K ⋂ {T} = ∅ und d′, wie folgt: (1) Für jeden Übergang d(Z1, a) = Z2 mit Z1, Z2 ∈ K und a ∈ V gilt d′ (Z1, a) = Z2. (2) Für jeden Übergang d(Z1, a) = ∅ gilt d′(Z1, a) = T. (3) Es gilt d′(T, a) = T für jedes a ∈ V. Es gilt L(A1) = V* — L(A). Als o is t L(A1), die Komplements prache von L, eins eitiglinear, als o in L3. Als o is t L3 abgeschlossen bzgl. des Komplements. Seien L1 und L2 aus L3. Wegen der Beziehung L1 ⋂ L2 = V* — ((V* — L1) ⋃ (V* — L2)) gilt auch die Abges chlos s enheit bzgl. des Durchschnitts. Ferner können wir jetzt zeigen, daß L1 ⋂ L2 vom Typ 2 is t, falls L1 ∈L2 und L2 ∈ L3 (Satz 4.2.5.6.). Beweis von Satz 4. 2. 5. 6. : Ohne Bes chränkung der Allgemeinheit können wir davon aus gehen, daß L1 und L2 Sprachen über dems elben Alphabet VT s ind. Sei zunächs t ε ∉ L1 und ε ∉ L2. L1 werde von der kontextfreien Syntax G1 = (VN, VT, R, S) in Choms kyNormalform erzeugt und L2 von dem FDA A = (K, V, d, F, z) akzeptiert. Sei F = {f1, ..., fk}. A (fi) is t der FDA, den man aus A erhält, indem F durch {fi} ers etzt wird. Dann gilt L1 ⋂ L2 = (L1 ⋂ L(A(f1))) ⋃ ... ⋃ (L1 ⋂ L(A(fk))). Für ein fes tes i wird L1 ⋂ L(A(fi)) durch die kontextfreie Syntax Gi = ((VN ⋃ VT) × K × K, VT, R′, (S, z, fi)) erzeugt, worin R′ aus allen Regeln folgender Form besteht: (1) (A, k, k1) → (B, k, k2) (C, k2, k1), wobei k, k1, k2 ∈ K, A, B, C ∈ VN, A → BC ∈ R.

87.  Formale Sprachen

(2) (A, k, k1) → (a, k, k1), wobei k, k1, k2 ∈ K, a ∈ VT und A → a ∈ R. (3) (a, k, k1) → a, worin a ∈ VT und d(k, a) = k1. Jede Ableitung s x in G wird durch die ers te Komponente der in den Regeln (1) und (2) auftretenden Tripel s imuliert. Durch (3) werden alle Tripel eliminiert gdw ein Element aus L(G) durch A(fi) akzeptiert wird. Als o is t L1 ⋂ L(A(fi)) die von Gi erzeugte Sprache. Wegen Satz 4.2.5.1. is t als o L1 ⋂ L2 vom Typ 2. Is t ε s owohl aus L1 als auch aus L2, s o wird ein neues Starts ymbol X für G1, nicht aus VN, gewählt und die Regeln X → S und X → ε zu R gefügt. Unter den Endzus tänden von A befindet s ich dann z. Die Syntax für L1 ⋂ L(A(z)) hat (X, z, z) als Starts ymbol, und es werden alle Regeln der Form (X, z, k1) → (S, z, k1) s owie (X, z, z) → (ε, z, z) und (ε, z, z) → ε zu R′ hinzugenommen. Als o ε ∈ L1 ⋂ L2. Damit ist der Satz bewiesen. Einseitig-lineare Sprachen und reguläre Ausdrücke: Wir kommen zu einer weiteren Charakterisierung einseitig-linearer Sprachen. Definition (rekursiv) .4.2.6.5 (regulärer Aus druck): Seien V und W = {⋃, *, ∅, (,)} dis junkte Alphabete. Eine Kette x über V ⋃ W heißt regulärer Ausdruck über V gdw gilt: (1) x ∈ V oder x ist das Zeichen ∅, oder (2) x ist von der Form (y ⋃ z), (yz) oder y*, worin y und z reguläre Ausdrücke über V sind. Jeder reguläre Aus druck x über V s tellt eine formale Sprache in folgender Weis e dar: Der reguläre Aus druck ∅ s teht für die leere Sprache. Jeder reguläre Aus druck a ∈ V s teht für die Sprache, deren einziges Element a is t. Für alle regulären Aus drücke x und y über V s teht (x ⋃ y) für die Sprache x ⋃ y, (xy) für die Sprache xy und y* für die Sprache y*. Beispiel: Sei V = {a, b}. ((ab)* ⋃ (ba)*) is t ein regulärer Aus druck über V und bezeichnet die Sprache, die aus allen Ketten gerader Länge, in denen a und b abwechs elnd aufeinander folgen, sowie aus ε besteht. Es gilt Satz 4. 2. 6. 3. : Eine Sprache wird von einem regulären Aus druck darges tellt gdw s ie eins eitig-linear ist. Beweis: (vgl. Salomaa 1973, 43 f.) Sei RA ein regulärer Aus druck über einem Alphabet V ⋃ W mit W wie in obiger Definition. Durch RA = a mit a ∈ V wird die Sprache {a} darges tellt. Alle einelementigen Sprachen s ind eins eitig-linear, da s ie durch die einzige

1597

Regel S → a erzeugbar s ind. Ebenfalls is t die leere Menge eins eitig-linear, da s ie von jeder Typ-3-Syntax ohne abs chließende Regeln erzeugt wird. Die nach (2) gebildeten regulären Aus drücke s tellen den Abs chluß der Vereinigung, Verkettung und Sternbildung über den einelementigen Sprachen der Form {a} mit a ∈ V dar. Auf Grund von Satz 4.2.5.1. Teil 3 s ind dies e Sprachen auch eins eitig-linear. Als o bezeichnet jedes RA eine eins eitig-lineare Sprache. Sei nun L eine eins eitig-lineare Sprache. Dann wird L von einem endlichen determinis tis chen Automaten A = (K, V, d, F, S) akzeptiert. Sei F = {f1, ..., fk} und A(fi) der Automat, den man aus A erhält, wenn F durch {fi} (1 ≤ i ≤ k) ers etzt wird. L(A) is t dann

L(A(fi)). Falls F = ∅, s o L(A) = ∅,

wobei L(A) durch den regulären Aus druck ∅ bezeichnet wird. Betrachten wir nun A(f), worin f irgendein fi is t. Sei K = {s 1, ..., s n} und bezeichne L(i, j, k) mit 0 ≤ k ≤ n und 1 ≤ i, j ≤ n die Sprache, die aus allen Ketten x bes teht, für die gilt: d(s i, x) = d(s i1, x1), = ... = d(s ir, xr) = s j, worin s i, s i1, ... s ir, s j bei der Abarbeitung von x s ukzes s ive durchlaufen werden und iu ≤ k für u ∈ {1, ..., r}. Für i ≠ j is t dann L(i, j, 0) entweder leer oder bes teht aus Zeichen aus V. Falls i = j, bes teht L(i, j, 0) aus ε und möglicherweis e aus Zeichen aus V. ε wird durch den regulären Aus druck ∅* bezeichnet. Für jedes L(i, j, 0) gibt es somit einen regulären Ausdruck. Angenommen, für ein bes timmtes k mit 0 ≤ k ≤ n—1 werde jede Sprache L(i, j, k) durch einen regulären Aus druck bezeichnet. Da L(i, j, k+1) = L(i, j, k) ⋃ L(i, k+1, k) L(k+1, k+1, k)* L(k+1, j, k), wird jedes L(i, j, k+1) durch einen regulären Aus druck bezeichnet, als o alle L(i, j, k) mit 0 ≤ k ≤ n. Das s elbe gilt auch für L = L(A(f)) und wegen der Abges chlos s enheit bzgl. der Vereinigung auch für L(A). Damit is t der Satz bewiesen. Aus Definition 4.2.6.5. ergibt sich sofort Satz 4. 2. 6. 4. : Die Menge der eins eitig-linearen Sprachen is t abges chlos s en bzgl. der Einsetzung. 4.2.7. Unlösbare Probleme Mit Satz 4.2.2.3. wurde gezeigt, daß nicht alle Typ-0-Sprachen ents cheidbar s ind. Dies bedeutet, daß es nicht für alle Typ-0-Syntaxen G und Ketten x einen Algorithmus gibt, mit dem s ich effektiv fes ts tellen läßt, ob x ∈ L(G)

1598

oder x ∉ L(G). Wir haben hiermit ein Beispiel für ein unlösbares Problem. Bevor wir auf weitere unlös bare Probleme in der Theorie der formalen Sprachen eingehen, s oll der Begriff des unlös baren Problems kurz beleuchtet werden. Lösbar / unlösbar bes agt, daß es einen / keinen Algorithmus zur Lös ung gibt. Für eine Klas s e K von formalen Sprachen gilt ein Problem P als lös bar gdw P für jedes Element von K lös bar is t, andernfalls is t P für K unlös bar. D. h.: is t P für K unlös bar, s o is t P für wenigs tens 1 Element aus K unlös bar, kann aber für andere Elemente aus K lös bar s ein. Der oben erwähnte Satz, daß das Element-Problem für Typ-0Sprachen nicht lös bar is t, bes agt nur, daß es eine s olche Typ-0-Sprache gibt, für die es unlös bar is t. Doch gibt es , wie man weiß, andere Typ-0-Sprachen, für die es lös bar is t (z. B. alle Typ-1-Sprachen). Unlö s barkeit s s ätze s tecken theoreti s che Grenzen der Anwendbarkeit mathematis cher Modelle ab. So wird man es vermeiden, in der Linguis tik von Syntaxtypen aus zugehen, deren Sprachenmenge die aller Typ-0-Sprachen is t, da es kein Analys everfahren gibt, das für eine beliebig gegebene Syntax G dies es Typs und einen beliebigen Satz s ents cheidet, ob s von G erzeugt wird. Die Lös barkeit von Problemen s pielt naturgemäß in der Computerlinguis tik eine zentrale Rolle, da s ie Voraus s etzung für die Kons truktion eins chlägiger Algorithmen ist. Im Rahmen der Theorie der formalen Sprachen werden vers chiedene Klas s en von Problemen unters chieden, die hins ichtlich ihrer Lös barkeit zu betrachten s ind. Sei D irgendeine Klas s e von Formalis men, die formale Sprachen definieren, z. B. ein Typ generativer Syntaxen, ein Automatentyp o. a., und L(d) mit d ∈ D die durch d definierte Sprache. Wichtige Probleme für D s ind folgende: (1) Das Element-Problem: Für ein beliebiges d ∈ D und eine beliebige Zeichenkette k über einem gegebenen Alphabet V soll entschieden werden, ob k ∈ L(d) ist. (2) Das Unendlichkeitsproblem: Für ein beliebiges d ∈ D soll entschieden werden, ob L(d) unendlich ist. (3) Das Leere-Menge-Problem: Für ein beliebiges d ∈ D soll entschieden werden, ob L(d) leer ist. (4) Das Teilmenge-Problem: Für beliebige d1, d2 ∈ D soll entschieden werden, ob L(d1) ⊆ L(d2).

XXIII. Service-Artikel

(5) Das Äquivalenzproblem: Für beliebige d1, d2 ∈ D soll entschieden werden, ob L(d1) = L(d2). Es ergibt s ich unmittelbar, daß aus der Lös barkeit des Teilmengen-Problems für D die Lös barkeit des Äquivalenzproblems für D folgt. Aus L0 ⊃ L1 ⊃ L2 ⊃L3 folgt, daß, wenn eins der Probleme (1) bis (5) für Li lös bar is t, es auch für Lj mit j > i lös bar is t und, wenn eins dies er Probleme für Li unlös bar is t, es auch für Lj mit j < i unlös bar is t. Die Ergebnis s e zeigt folgende Tabelle (L s teht für ‘lös bar’, U für ‘unlösbar’): Problem L3 L2 L1 L0 Element-Problem L L L U Unendlichkeitsproblem L L U U Leere-Menge-Problem L L U U Teilmenge-Problem L U U U L U U U Äquivalenzproblem Die Unlös barkeit des Element-Problems für Typ-0-Sprachen erweis t s ich als fundamental, da die Unlös barkeit vieler weiterer Probleme darauf zurückgeführt werden kann. So Satz .4 .2.7.1 : Da s Leere-Menge-Problem für L0 ist unlösbar. Beweis: Angenommen, es is t lös bar. Sei G = (VN, VT, R, S) eine Typ-0-Syntax und k eine beliebige Kette aus VT*. Sei G1 = (VN ⋃ {X, #}, VT, R ⋃ {X → #S#, #k# → ε}, X), wobei X, # ∉ VN. L(G1) is t entweder leer oder hat als einziges Element ε. Welche der beiden Möglichkeiten zutrifft, si t auf Grund der Annahme ents cheidbar. Dann is t aber auch ents cheidbar, ob k aus L(G) is t oder nicht, und damit is t das Element-Problem für Typ-0-Sprachen lös bar, ein Widers pruch zu Satz 4.2.2.3. Als o is t das LeereMenge-Problem für Typ-0-Sprachen unlösbar. Dies er Satz dient als Aus gangs punkt weiterer Sätze, so Satz .4 .2 .7 .2 : Das Leere-Menge-Problem für L1 ist unlösbar. Beweis: Eine Lös ung für das Leere-MengeProblem für L1 würde eine Lös ung dies es Problems für L0 bedingen, wie aus folgendem Satz hervorgeht. Satz 4. 2. 7. 3. : Sei L eine Typ-0-Sprache über dem Alphabet VT und a, b ∉ VT. Dann gibt es eine Typ-1-Sprache L1 folgender Ges talt: L1 bes teht aus Ketten der Form aibx mit i  0 und x ∈ L, und für jedes x ∈ L gibt es ein aibx ∈ L1 mit i  0.

87.  Formale Sprachen

Beweis: Sei L = L(G) mit G = (VN, VT, R, S), und s eien X1 und X2 neue Nicht-Terminale nicht aus VN. Wir kons truieren die Syntax G1 = (VN ⋃ {X1, X2}, VT ⋃ {a, b}, R1, X1) mit R1, wie folgt: R1 = {x → y  x → y ∈ R ⋀ Länge(x) ≤ Länge(y)} ⋃ {x → X2Länge(x)—Länge(y)y  x → y ∈ R ⋀ Länge(x) > Länge(y)} ⋃ {X1 → bS, X2b → ab} ⋃ {cX2 → X2c  c ∈ VN ⋃ VT ⋃ {b}}. Dies e Syntax is t vom Typ 1, und die davon erzeugte Sprache erfüllt die Bedingungen für L1. Wäre das Leere-Menge-Problem für jede Sprache L1 lös bar, s o auch für jede Typ-0Sprache, ein Widers pruch zu Satz 4.2.7.1. Als o gibt es eine Typ-1-Sprache, für die es unlösbar ist. Man zeigt nun leicht, daß auch das Unendlichkeits problem für L0 und L1 unlösbar is t. Wir brauchen G1 im Beweis von Satz 4.2.7.1. nur s o abzuwandeln, daß für ein beliebiges k ∈ VT* die Regelmenge von G1 R ⋃ {X → #S#, #k# → +k+, +k + → +k+k+, + → ε} mit + einem neuen NichtTerminal. Dann is t L(G1) unendlich gdw k ∈ L(G), andernfalls is t L(G1) leer. Aus der Unlös barkeit des Element-Problems für L0 folgt die Unlös barkeit des Unendlichkeits problems für L0. Zus ammen mit Satz 4.2.7.2. ergibt s ich die Unlös barkeit des Unendlichkeitsproblems für L1. Weitere Unlö s barkeit s s ätze gehen auf Posts Entsprechungssatz (Post 1946) zurück. Definition . . 4. 2. 71 (Entsprechungsproblem): Ein Ents prechungs problem si t ein Quadrupel (V, A, B, n) mit V einem Alphabet, n  1 und A = (x1, ..., xn), B = (y1, ..., yn), worin xi, yi ∈ V+ (1 ≤ i ≤ n). Eine Lös ung für ein Ents prechungs problem is t eine nicht-leere Folge von Indizes i1, ..., ik, für die xi1 ... xik = yi1 ... yik gilt. Beispiele: a) Das Ents prechungs problem ({a, b}, A, B, 4) mit A = (aa, b, a, b) und B = (a, a, ba, b) bes itzt die Lös ung 1, 2, 3, 4, da aabab = aabab. b) Das Ents prechungs problem ({a, b}, A, B, 2) mit A = (ab, a2b2) und B = (a3b3, a4b4) bes itzt keine Lös ung, da die mit B gebildeten Ketten Teilketten aaa oder aaaa enthalten und s olche Ketten mit A nicht gebildet werden können. Satz 4. 2. 7. 4. (Pos ts Ents prechungs s atz): Es gibt keinen Algorithmus , der für jedes Ents prechungs problem ents cheidet, ob es eine Lösung besitzt oder nicht.

1599

Dies er Satz wird auf das Element-Problem für Typ-0-Sprachen zurückgeführt, und man bewei s t, daß jede s Ent s prechung s problem lös bar is t gdw das Element-Problem lös bar is t. Aus des s en Unlös barkeit folgt die Unlös barkeit des Ents prechungs problems (zum Beweis vgl. Salomaa 1973, Maurer 1977). Aus gehend von der Unlös barkeit des Ents prechungs problems beweis t man weiter einige Unlös barkeits s ätze für kontextfreie Syntaxen. So gilt Satz 4. 2. 7. 5. : Seien G, G1 und G2 beliebige kontextfreie Syntaxen. Folgende Probleme s ind unlös bar (d. h., es gibt G, G1 und G2, für die sie unlösbar sind): a) Ist L(G1) ⋂ L(G2) leer? b) Ist L(G1) ⋂ L(G2) unendlich? c) Ist L(G1) ⋂ L(G2) einseitig-linear? d) Ist L(G1) ⋂ L(G2) kontextfrei? e) Ist VT* — L((G) leer? f) Ist VT* — L((G) unendlich? g) Ist VT* — L((G) kontextfrei? h) Ist L((G) einseitig-linear? i) Ist L(G1) ⊆ L(G2)? k) Ist L(G1) = L(G2)? Eine weitere Klas s e von Problemen vzgl. kontextfreier Sprachen betrifft Ambiguität (vgl. Kap. 4.2.3.). Auf Grund von Pos ts Entsprechungssatz läßt sich zeigen: Satz 4. 2. 7. 6. : 1) Es gibt kein Verfahren, das für jede kontextfreie Syntax G ents cheidet, ob G ambig is t. 2) Es gibt kein Verfahren, daß für jede kontextfreie Sprache L ents cheidet, ob L ambig is t. (Zum Beweis vgl. ebenfalls Salomaa 1973.)

5.

Erweiterungen generativer Syntaxen in der Linguistik

5.1. Allgemeines Für die s yntaktis che Bes chreibung natürlicher Sprachen is t unter generativer Sicht bes onders das Konzept der kontextfreien Syntax herangezogen worden. Dies is t unter dem As pekt adäquat, daß natürliche Sprachen — s ieht man von wenigen exotis chen Strukturen ab — im wes entlichen kontextfrei bes chreibbar zu s ein s cheinen (vgl. Klenk 1989). Dies e Syntaxen erweis en s ich aber als unhandlich und unübers ichtlich, da bereits für kleinere Aus s chnitte einer natürlichen Sprache viele Regeln benötigt werden — wobei die Ebene der lexikalis chen Formative s ogar noch unberücks ichtigt is t. Durch die Verquickung kontextfreier Regeln mit anderen Formalis men

1600

oder durch Abänderungen des kontextfreien Formalis mus s elbs t wird vers ucht, größere Trans parenz zu erreichen s owie genuin linguis tis chen Erfordernis s en wie z. B. der Dars tellung genereller Eigens chaften einer Sprache nachzukommen. Im folgenden s oll zwis chen Wieders chreibFormalis men und anderen Formalis men unters chieden werden. Wieders chreib-Formalis men s ind oder erzeugen Produktions regeln vom Typ 0 (in den hier behandelten Fällen vom Typ 2), die ihrers eits die Elemente der Sprachen erzeugen. Hierzu gehören u. a. Syntaxen mit Metaregeln und 2-Schichten-Grammatiken. Zu den anderen Formalis men zählen z. B. s olche, die über Baums trukturen operieren, s ei es in Form von Trans formations regeln, die einer Kette mit ihrer Baums truktur eine andere Kette mit deren Baums truktur zuordnen, s ei es in Form von Unifikations formalis men, die mit komplexen, aus Merkmal s mengen be s tehenden Nicht-Terminalen und Prinzipien der Merkmals übertragung arbeiten. Die Unifikations grammatiken unters cheiden s ich von den übrigen genannten Syntaxen darin, daß s ie Bedingungen der Wohlgeformtheit von Bäumen einführen: ein Element der Sprache is t eine Endkette, die durch die Produktions regeln abgeleitet werden kann und deren zugeordneter Baum den Wohlgeformtheit s bedingungen genügt. Die Bes chreibung der für Wohlgeformtheits bedingungen vorge s chlagenen Formali s men würde einen s elbs tändigen Artikel füllen, s o daß hier darauf verzichtet werden muß. Man vergleiche dazu die grundlegenden Arbeiten von Bres nan/Kaplan (1982) zur lexikalis chfunktionalen Grammatik, Gazdar/Klein/Pullum/Sag (1985) zur Generalized Phras e Structure Grammar, Kay (1985) zur funktionalen Unifikations grammatik s owie Artikel 25 in Bd. I dieses Handbuchs. 5.2. Die Formalismen 5.2.1. Syntaxen mit Metaregeln Metaregeln s ind aus der Generalized Phras e Structure Grammar (GPSG) bekannt. Sie bes chreiben Beziehungen zwi s chen Mengen kontextfreier Regeln. Definition 5. 2. 1. 1. (Metaregel): Seien VN, VT und VA dis junkte Vokabulare. Eine Metaregel m hat die Form r ⇒ s , worin r und s Schemata kontextfreier Regeln der Form A → x mit A ∈ VN und x ∈ (VN ⋃ VT ⋃ VA)* s ind, wobei jedes Element aus VA in s auftritt gdw es in r auftritt, und höchs tens ein

XXIII. Service-Artikel

Vorkommen in r und s hat. Die Zeichen aus VA heißen Kettenvariable. Definition 5. 2. 1. 2. (die aus einer Metaregel erzeugte Regelmenge): Sei m = r ⇒ s eine Metaregel und G = (VN, VT, R, S) eine kontextfreie Syntax. m wird auf G angewendet, indem die Regelmenge Rm gebildet wird, die aus R bes teht und aus allen das Schema s erfüllenden Regeln, für die es eine Regel aus Rm gibt, die r erfüllt (wie man s ieht, wird Rm rekurs iv gebildet). Dabei erfüllt eine Regel A → x das Schema p = B → X0a1X1 ... anXn, worin ai ∈ VN ⋃ VT (1 ≤ i ≤ n) und Xi ∈ VA (0 ≤ i ≤ n), gdw A = B und x = x0a1x1 ... anxn mit xi ∈ (VN ⋃ VT)* (0 ≤ i ≤ n). Rm is t die durch m bzgl. G erzeugte Regelmenge. Für M = {m1, ..., mn} mit mi = ri ⇒ s i (1 ≤ i ≤ n) s ei RM die Regelmenge, die aus R bes teht und aus allen ein Schema s i erfüllenden Regeln, für die es eine Regel aus RM gibt, die ri erfüllt. Definition .5 .2 .1 .3 (Syntax mit Metaregeln): Eine Syntax mit Metaregeln G is t ein 6-Tupel (VN, VT, VA, R, M, S), worin VN, VT und VA wie in Definition 5.2.1.1. s ind, R eine endliche, nicht-leere Menge kontextfreier Regeln der Form A → x mit A ∈ VN und x ∈ (VN ⋃ VT)*, M eine endliche, nicht-leere Menge von Metaregeln und S ein aus gezeichnetes Symbol aus VN ist. Definition 5. 2. 1. 4. (Sprache einer Syntax mit Metaregeln): Sei G eine Syntax mit Metaregeln (VN, VT, VA, R, M, S), G′ die kontextfreie Syntax (VN, VT, R, S), RM die gemäß Definition 5.2.1.2. bzgl. G′ erzeugte Regelmenge. Die Syntax Gobj = (VN, VT, RM, S) is t die aus G erzeugte Objektsyntax und deren Sprache L(Gobj) = {x S x ⋀ x ∈ VT*} zugleich die Sprache der Metaregel-Syntax G. Hier is t zu beachten, daß ein Gobj nicht kontextfrei zu s ein braucht. RM bes teht zwar aus kontextfreien Regeln, doch können es durch rekurs ive Anwendung von M unendlich viele s ein, was kontextfreie Syntaxen nicht erlauben. Mit unendlich vielen kontextfreien Regeln las s en s ich auch nicht-kontextfreie Sprachen erzeugen. Beispiele: (a) Sei G = ({S}, {a, b}, {X}, {S → aSb, S → ab}, {S → aX ⇒ S → Xa, S → Xb ⇒ S → bX}, S) eine Syntax mit Metaregeln. Dann is t Gobj = ({S}, {a, b}, {S → aSb, S → ab, S → Sba, S → ba, S → baS}, S). (b) Für die Bildung des englis chen Pas s ivs wird die Metaregel VP → X NP ⇒ VP[PAS] → X (PP[by]) angegeben, worin X eine Kettenvariable und VP, VP[PAS], NP und PP[by]

87.  Formale Sprachen

Nicht-Terminale s ind. Die runden Klammern bei PP[by] s tehen dafür, daß dies es Symbol auch fehlen kann. X vertritt Symbole für Klas s en von Verben, die ein Pas s iv bilden können. Angewendet etwa auf die kontextfreie Regel VP → V[2] NP, worin V[2] für die Klas s e der Verben s teht, die s ich bzgl. ihres kategorialen Kontexts wie see verhalten, ergeben s ich VP[PAS] → V[2] und VP[PAS] → V[2] PP[by], wodurch aktivis chen Kons truktionen wie sees the dog pas s ivis che wie is seen und is seen by the man gegenüberges tellt werden. Die res ultierende Objekts yntax kann mit geeigneten weiteren Regeln die drei Strukturen erzeugen. (c) Sei G = ({S}, {a, b, c}, {S → abc}, {S → aX1bX2cX3 ⇒ S → aaX1bbX2ccX3}, S) eine Syntax mit Metaregeln. Die Objekts yntax Gobj enthält alle Regeln der Form S → anbncn (n  1). Dies e erzeugen die Sprache {anbncn  n  1}, welche kontexts ens itiv, aber nicht kontextfrei ist. Die Menge der Sprachen, die durch Syntaxen mit unendlich vielen kontextfreien Regeln erzeugt werden, is t L0. Auch die Syntaxen mit Metaregeln erzeugen ganz L0, und dies gilt s ogar, wenn man dies e Syntaxen auf einelementiges VA eins chränkt (Peters /Us zkoreit 1982). Da daher mit dies en Syntaxen die in 4.2.7. behandelten Probleme auftreten, werden Form und Anwendung der Metaregeln in linguis tis chen Bes chreibungen weiter res tringiert. Daß die Objekts yntax kontextfrei is t, wird z. B. durch den endlichen Abschluß einer Menge von kontextfreien Regeln unter einer Menge von Metaregeln garantiert. Darunter vers teht man, daß eine Metaregel m ∈ M nicht auf eine Regel angewendet werden darf, die bereits aus einer Anwendung von m res ultiert. Somit kann die erzeugte Regelmenge RM nur endlich s ein. Weitere Res triktions möglichkeiten werden im Rahmen der GPSG (Gazdar/Klein/Pullum/Sag 1985) eingeführt. 5.2.2. 2-Schichten-Grammatiken Die 2-Schichten-Grammatik wurde von A. van Wijngaarden (danach auch Van-Wijngaarden-Grammatik genannt) zur Bes chreibung der Programmier s prache ALGOL68 entwickelt. Obwohl für natürliche Sprachen wenig verwendet (vgl. aber Chas tellier/Colmerauer 1969), behandeln wir dies en Typus , weil hier ähnlich wie in Syntaxen mit Metaregeln aus Regeln andere Regeln abgeleitet werden, die dann die Sprache erzeugen. Doch is t der Formalismus ein anderer.

1601

Eine 2-Schichten-Grammatik bes teht aus einem Produktions s ys tem P = (VN1, VT1, R1) mit VN1 einer Menge von Nicht-Terminalen, VT1 einer Menge von Ends ymbolen, wobei VN1 ⋂ VT1 = ∅, und R1 einer endlichen Menge von kontextfreien Regeln, den Metaregeln (nicht zu verwechs eln mit den Metaregeln in 5.2.1.), und einem Sys tem H = (VH, c, R2), worin R2 eine endliche Menge von Regels chemata, Hyperregeln genannt, is t. Sie haben die Form kontextfreier Regeln A → x mit A ∈ VH und x ∈ VH {{c} VH}* ⋃ {ε}. VH is t ein Alphabet, des s en Symbole nicht-leere Ketten über VN1 ⋃ VT1 ⋃ V2 s ind, worin V2 eine Menge neuer Symbole nicht aus VN1 ⋃ VT1 is t. Als weiteres neues Zeichen wird c, das Trennzeichen, eingeführt. Aus P und H wird eine neue Schicht von kontextfreien Regeln R, wie folgt, erzeugt. Für jedes X ∈ VN1 s ei LX = {x  X x ⋀ x ∈ VT1*}. Is t h eine Hyperregel und s ind X1, ..., Xn die Symbole aus VN1, die in den in h vorkommenden Symbolen aus VH auftreten, und gilt LXi ≠ ∅ (1 ≤ i ≤ n), s o erhält man eine Regel aus R, indem jedes Xi durch ein x ∈LXi wiederges chrieben wird und dabei alle Vorkommen des s elben Xi durch das s elbe x ers etzt werden. Alle s o bildbaren Regeln gehören zu R s owie diejenigen Hyperregeln, in denen kein Symbol aus VN1 vorkommt. Sei ein einfaches Beis piel gegeben: P = ({NUM, GEN, PER}, {s g, pl, mas k, fem, 1per, 3per}, {NUM → s g, NUM → pl, GEN → mas k, GEN → fem, PER → 1per, PER → 3per}). Es is t LNUM = {s g, pl}, LGEN = {mas k, fem}, LPER = {1per, 3per}. Sei V2 = {s , np, vp, v, pron, art, n, ich, s ie, wir, der, die, Mann, Frau, Männer, laufe, läuft, laufen} und R2 = {s → PER NUM np c PER NUM vp, PER NUM vp → PER NUM v, PER NUM np → PER NUM pron, 3per sg np → sg GEN art c sg GEN n, 3per pl np → pl art c pl n, 1per sg pron → ich, 3per sg pron → sie, 1per pl pron → wir, 3per pl pron → sie, 1per sg v → laufe, 3per sg v → läuft, 1per pl v → laufen, 3per pl v → laufen, sg mask art → der, sg fern art →die, pl art → die, sg mask n → Mann,

1602

sg fem n → Frau, pl n → Männer}. Alle linken Seiten der Regeln sowie alle rechts s tehenden maximalen Ketten, in denen das Trennzeichen c nicht vorkommt, s ind Elemente von VH. Mit LNUM, LGEN und LPER werden aus den ers ten 4 Hyperregeln 14 Regeln aus R erzeugt, wie s → 1per s g np c 1per s g vp, während die übrigen Hyperregeln s chon zu R gehören. Wie man s ieht, bes chreibt das Sys tem Konkordanzen von Person, Numerus und Genus. R is t die Regelmenge einer Syntax G = (VN, VT, R, S), worin VN ⊆ (VT1 ⋃ V2)* — {ε}, VT ⊆ (VT1 ⋃ V2)* — {ε} ⋃ {c}, VN ⋂ VT = ∅ und S ∈ VN. Die linken Seiten der Regeln aus R s ind aus VN. Damit s ind die Regeln kontextfrei. Die Sprache von G is t L(G) = {x  x ∈ VT* ⋀ S x}. Setzt man im obigen Beis piel für VT die Menge der deuts chen Wörter aus V2 vereinigt mit {c} und für VN die übrigen Elemente aus VH an und is t das Starts ymbol s , s o is t L(G) eine Menge deuts cher Sätze, deren Wörter noch durch c getrennt s ind. Auf L(G) wird ein Homomorphis mus h angewendet (vgl. Definition 4.2.5.2.), der c durch ε s ubs tituiert. Das Res ultat si t die Sprache der 2-SchichtenGrammatik. Eine 2-Schichten-Grammatik 2G is t s omit ein 5-Tupel (P, H, VN, VT, S) mit P, H, VN, VT und S wie oben definiert. Die Sprache von 2G is t L(2G) := h(L(G)) mit h(a) = a für a ∈ VT — {c} und h(c) = ε. Beispiel einer Ableitung mit der oben erzeugten Grammatik G: s→ 3per pl np c 3per pl vp → pl art c pl n c 3per pl vp → die c pl n c 3per pl vp → die c Männer c 3per pl vp → die c Männer c 3per pl v → die c Männer c laufen Da die Metaregeln unendliche Sprachen erzeugen können, können s owohl R als auch VN unendlich s ein. In dies en Fällen is t G keine kontextfreie Syntax. Es gilt s ogar, daß die durch 2-Schichten-Grammatiken erzeugten Sprachen genau die Typ-0-Sprachen s ind (Beweis von Sintzoff 1967). Beispiel einer 2-Schichten-Grammatik für die nicht-kontextfreie Sprache L = {anbndn  n  1}: 2G = (P, H, VN, VT, s tart) mit P = ({NUM, ABD}, {a, b, d, i}, {ABD → a, ABD →b, ABD →d, NUM →i, NUM →NUM i}), V2 = {s tart}, H = ({s tart, NUM a, NUM

XXIII. Service-Artikel

b, NUM d, i NUM ABD, ABD, NUM ABD, i ABD}, c, R2) mit R2 = {start → NUM a c NUM b c NUM d, i NUM ABD → ABD c NUM ABD, i ABD → ABD}, VT = {a, b, d, c} und VN = V2 ⋃ {in{a, b, d}  n  1}. Es gilt LABD = {a, b, d}, LNUM = {in  n  1}, R = {start → in a c in b c in d  n  1} ⋃ {i in a → a c in a  n  1} ⋃ {i in b → b c in b  n  1} ⋃ {i in d → d c in d  n  1} ⋃ {i a → a, i b → b, i d → d}. Beginnend bei s tart werden direkt vor a, b und d gleich viele i-Vorkommen erzeugt, die durch die anderen Regeln durch ebens oviele a-, b- und d-Vorkommen wiederges chrieben werden. Daher L = L(2G). 5.2.3. Transformationssyntaxen Im Rahmen dies es Artikels wird nur auf die formalen Eigen s chaften der Tran s formations s yntaxen eingegangen. Ihre Anwendung als linguis tis ches Bes chreibungs mittel is t s eit Choms ky (1957) hinlänglich bekannt, s o daß s ich eine nochmalige Dars tellung hier erübrigt. Da vers chiedene Subtypen von Trans formations s yntaxen entwickelt wurden, s ollen deren grunds ätzliche Gemeins amkeiten betrachtet werden. (Nicht behandelt wird die Governmentand Binding-Syntax (vgl. Choms ky 1981), da dies e keine Trans formations s yntax im eigentlichen Sinn is t.) Generell gilt: Eine Transformationssyntax is t ein 4-Tupel TG = (B, T, A, VT), worin B eine nicht-leere Menge von Bäumen gemäß Definition 4.2.3.3., T eine endliche Menge von BaumUmwandlungs regeln, den Transformationsregeln (abgekürzt T-Regeln), A eine Anwendungs vors chrift für T und VT ein Alphabet is t. B kann unendlich s ein. Für B gibt es zwei dis junkte Vokabulare VNB und VTB, derart daß jedes Blatt eines Baumes b ∈ B ein Etikett aus VTB und jeder andere Knoten ein Etikett aus VNB trägt. Die Wurzeln aller b ∈ B haben das s elbe Etikett. Man bezeichnet B als die Basis von TG. Eine T-Regel t is t ein Paar (SB, SV) mit SB einer Strukturbeschreibung und SV einer Strukturveränderung. SB is t ein Baums chema (mit etikettierten Knoten). Ein Baums chema unters cheidet s ich von einem Baum dadurch, daß es variable Teils tücke enthalten kann. Es legt s omit eine

87.  Formale Sprachen

Klas s e von Bäumen fes t. Liegt ein gegebener Baum b in der von SB definierten Baumklas s e, s o kann t auf b angewendet werden. SV gibt an, wie b in einen neuen Baum b′ umzuwandeln is t. SV bes teht aus einer Folge von Anweis ungen, die auf die Operationen Tilgung und Adjunktion, die elementare Operationen heißen, zurückgeführt werden. Die Tilgung: Is t k ein Knoten in b, s o wird mit tilge k der größte Teilbaum in b mit Wurzel k′ gelös cht, wobei k = k′ oder jeder Pfad von k′ zu einem Blatt über k führt. Die Adjunktion: Sind b und b′ Bäume und is t k ein Knoten in b, s o wird mit adjungiere b′ an k in Form f b′ beim Knoten k in einer Form eingehängt, s o daß ein neuer Baum ents teht. Bzgl. f unters cheidet man Tochter- und Schwesteradjunktion: b′ wird — von links nach rechts betrachtet — an ers ter oder letzter Stelle direkt unter k eingehängt (als ers te bzw. letzte Tochter) oder direkt links oder rechts neben k unter dem Knoten, von dem eine Ecke zu k führt (als linke bzw. rechte Schwes ter). Als weitere Operationen nennt man die Substitution, die Ers etzung des größten Teilbaums mit Wurzel k durch einen anderen Baum, und die Permutation, die Ums tellung zweier Teilbäume in der Links -Rechts -Anordnung. Subs titution und Permutation las s en s ich auf Folgen von Adjunktionen und Tilgungen zurückführen. Bei allen Operationen is t niemals die Wurzel des Eingabebaums , auf dem t operiert, betroffen, s o daß der Ergebnis baum das s elbe Wurzeletikett hat wie der Eingabebaum. Ferner gibt es auch für jeden Ergebnis baum zwei dis junkte Vokabulare terminaler und nicht-terminaler Etiketten für die Blätter bzw. die übrigen Knoten. Für T-Regeln pflegt man eine Reihenfolge fes tzulegen, gemäß der s ie zur Anwendung kommen. Solche und weitere Res triktionen werden in der Vors chrift A angegeben. Eine Ableitung mit einer Syntax TG verläuft in der Weis e, daß beginnend bei einem beliebigen b0 ∈ B eine Folge von T-Regeln t1, ..., tm durchs chritten wird, s o daß t1(b0) = b1, ..., tm(bm—1) = bm. Dabei bezeichnet ti(bj), worin bj ein Baum is t, den Ergebnis baum der Anwendung von ti auf bj. Erfüllt bj die Strukturbes chreibung von ti nicht (d. h. liegt bj nicht in der von der Strukturbes chreibung von tj definierten Baumklas s e), s o gilt ti(bj) = bj. Der Ableitungs prozeß richtet s ich nach der Anwendungs vors chrift A. Bei Vorgabe einer Reihenfolge für T bes teht A aus einer Folge F = (t1, ..., tn), worin alle Regeln aus T wenigs tens einmal vorkommen. Für je-

1603

des b ∈ B verläuft die Ableitung gemäß F. Alternative Ableitungen für ein b s ind möglich durch Aus zeichnung mancher Regeln als fakultativ. Dies e können in der Reihenfolge übers chlagen werden. ti(bj) muß eindeutig s ein. Da es aber vorkommt, daß eine SV auf vers chiedene Knoten eines Eingabebaums anwendbar is t, s o daß vers chiedene Ergebnis bäume möglich s ind, muß A auch Angaben darüber enthalten, in welcher Reihenfolge vers chiedene Teilbäume eines Baums abzuarbeiten s ind. Dies is t bes onders bei s elbs teinbettenden Strukturen, d. h., wenn in einem Baum wenigs tens zwei Knoten mit dems elben Etikett auf einem Pfad liegen, zu berücks ichtigen. Aus der Linguis tik kennt man hier die zyklische Anwendung der Folge F bei Satzeinbettungen: F wird zuers t nur auf die tiefs t eingebetteten Sätze angewendet und ganz durchlaufen, dann wiederholt s ich dies er Vorgang s ukzes s ive bei den nächs t höheren Sätzen bis hin zum obersten Satz. Nun bleibt noch, die Sprache einer Trans formations s yntax TG fes tzulegen. Dazu gehen wir von der Menge der Bäume aus , die aus allen b ∈ B durch den bes chriebenen Ableitungs prozeß erhalten werden und auf die keine nicht fakultative T-Regel mehr anwendbar is t. Sei dies die Menge der Endbäume (EB), und s ei EK die Menge der Endketten dies er Bäume, d. h. die Folge der Etiketten ihrer Blätter von links nach rechts . Die Sprache von TG is t durch L(TG) := EK ⋂ VT* gegeben. VT braucht dabei mit VTB nicht identisch zu sein. Beispiel: Sei TG = (B, T, A, VT) mit VT = {a, d, c}. Sei G = ({S, A, D}, {a, d, c}, {S → ASD, S → c, A → a, D → d}, S) eine kontextfreie Syntax und B die Menge der Bäume, die den Links ableitungen aller k ∈ L(G) bzgl. G gemäß dem Verfahren in 4.2.3. zugeordnet werden. Bes tehe T aus der einzigen T-Regel t = (s b, s v), worin s b das Baumschema

mit X einer beliebigen evtl. leeren und Y einer nicht leeren Folge von Teilbäumen is t. Die geklammerten Zahlen zeichnen dabei die Teilbäume aus , auf die s v Bezug nimmt. s v bes teht aus der Folge von Anweis ungen: ad-

XXIII. Service-Artikel

1604

jungiere (1) als rechte Schwes ter an (2), tilge (1) an s einer urs prünglichen Stelle. A enthält als einzige Res triktion, daß t zuers t auf einem am tiefs ten eingebetteten Teilbaum, auf den es angewendet werden kann, operiert und dann s ukzes s ive auf den nächs t höheren. Somit wird der s s Ba i baum

tionellen Ableitung auch unter Wahrung der Wiederauffindbarkeit beträchtliche Tilgungen vorkommen können und dabei, s chon um kurze Sprachelemente zu gewinnen, eine große Anzahl von Trans formations zyklen zu durchlaufen is t. Dabei s pielt die Anzahl der eingebetteten Teilbäume mit Wurzeletikett S (S-Bäume, S das Starts ymbol der Bas is s yntax) eine Rolle. Peters /Ritchie konnten zeigen, daß Ent s cheidbarkeit herbeigeführt wird, wenn eine maximale Anzahl der SBäume im Bas is baum angenommen wird: für jede Kette x über dem Endvokabular VT, jede Trans formations s yntax TG (in der in Peters / Ritchie einges chränkten Form) und jede natürliche Zahl n is t es ents cheidbar, ob x durch TG aus gehend von einem Bas is baum mit höchs tens n S-Bäumen abgeleitet wird oder nicht. Der kleins tmögliche Wert für n bzgl. eines gegebenen x is t der Wert der s og. Zyklus funktion für TG an der Stelle x. Mittels der Zyklus funktion wird eine Klas s e von Tran s formation s s yntaxen, deren Sprachen ents cheidbar s ind, gefunden (vgl. Peters /Ritchie 1973).

6. überführt. Die Sprache L(G) = {ancdn  n  0} wird in L(TG) = {(ad)nc  n  0} übersetzt. Generative Kapazität: Man s pricht von Transformationssyntaxen mit kontextfreier (kontextsensitiver) Basis, wenn B die Menge von Bäumen is t, die nach dem in 4.2.3. bes chriebenen Verfahren den Sprachelementen einer kontextfreien (kontexts ens itiven) Syntax G zugeordnet werden (im Fall einer kontexts ens itiven Syntax mit Notation der Regeln in der Form A → x / u___v, vgl. Definition 4.2.2.3.). Für Trans formationss yntaxen mit kontextfreier Bas is wurde von Peters /Ritchie (1973) bewies en, daß die Menge der von ihnen erzeugten Sprachen gleich der Menge der Typ-0-Sprachen is t. Zugrundegelegt wird hier s da Aspects-Modellvon Chom s ky (1965), das bezüglich Tilgungen Res triktionen einführt, die im principle of recoverability fes tgelegt s ind. So muß ein zu tilgender Teilbaum „wiederauffindbar“ s ein, s ei es , daß er im Eingabebaum mehrfach auftritt, s ei es , daß er nur aus einem Knoten mit terminalem Etikett aus einer fes t vorgegebenen Symbollis te bes teht. Die generative Kapazität umfaßt auch unter dies en Bes chränkungen ganz L0. Dies liegt daran, daß in einer trans forma-

Literatur

Backus, J. W. 1959. The s yntax and s emantics of the propos ed international algebraic language of the Zurich ACM-GAMM conference. ICIP Paris. Bar-Hillel, Y. 1953. On recurs ive definitions in empirical s cience. Proceedings of the 11th International Congress of Philosophy, Brüssel, Bd. 5. Bloomfield, L. 1933. Language. New York. Bresnan, J. , and R. Kaplan. 1982. Lexical functional grammar. A formal s ys tem for grammatical repres entation. The Mental Repres entation of Grammatical Relations , ed. by J. Bres nan. Cambridge, MA. Chastellier, G. de, and A. Colmerauer. 1969. WGrammar. Proceedings of the 24th National Conference of the ACM. New York. Chomsky, N. A. 1956. Three models for the des cription of language. I. R. E. Trans actions on Information Theory, IT-2, 113—24. —. 1957. Syntactic Structures. Den Haag. —. 1959. On certain formal properties of grammars. Information and Control 2. 137—67. —. 1965. Aspects of the Theory of Syntax. —. 1981. Lectures on Government and Binding. Dordrecht. Christaller, T. , and D. Metzing. 1979/80. Augmented Tran s ition Network Grammatiken. 2 Bde. Berlin.

87.  Formale Sprachen

Church, A. 1936. An uns olvable problem of elementary number theory. American Journal of Mathematics 58. 345—63. Gazdar, G. ; Klein, E. ; Pullum, G. ; and I. Sag. 1985. Generalized Phrase Structure Grammar. Oxford. Harris, Z. 1951. Structural Linguistics. —. 1952. Discourse analysis. Language 28. 18—23. Hermes, H . 1971. Aufzählbarkeit, Ent s cheidbarkeit, Berechenbarkeit. Berlin. Kay, M. 1985. Pars ing in functional unification grammar. Natural Language Pars ing, ed. by D. R. Dowty, L. Karttunen & A. M. Zwicky. 251—78. Kleene, S. C. 1936. General recurs ive functions of natural number s. Mathematical Annual 112. 727—42. —. 1956. Repres entation of events in nerve nets and finite automata. Automata Studies. Klenk, U. 1989. Computerlinguis tik und die Theorie der formalen Sprachen. Computational Linguistics, ed. by I. Bátori, W. Lenders & W. Putschke. Koskenniemi, K. 1983. Two-level model for morphological analys si . International Joint Conference of Artificial Intelligence (IJCAI) 1983, Karls ruhe, 683—85. Maurer, H . 1977. Grundlagen der Programmier-

1605

sprachen. Theorie der Syntax. Mannheim. Peters, P. S. , and R. W. Ritchie. 1973. On the generative power of trans formational grammars . Information Sciences 6. 49—83. Post, E. 1944 Recurs ive enumerable s ets of pos itive integers and their decis ion problems . Bull. Am. Math. Soc. 50. 284—316. —. 1946. A variant of recurs ively uns olvable problem. Bull. Am. Math. Soc. 52. 264—8. Salomaa, A. 1973. Formal Languages . New York, London. Sintzoff, M. 1967. Exis tence of a Van Wijngaarden Syntax for Every Recurs ive Enumerable Set. Annales de la Société Scientifiques de Bruxelles 81. Thue, A. 1914. Probleme über Veränderungen von Zeichenreihen nach gegebenen Regeln. Skrifter utgit av Videnskapsselskapet i Kristiania I 10. Turing, A. 1937. On computable numbers , with an application to the Ent s cheidung s problem. Proc. London Math. Soc. 42. 230—65. Uszkoreit, H. , and P. S. Peters. 1982. Es s ential variables in metarules . 1982 Annual Meeting of the Linguistic Society of America, San Diego.

Ursula Klenk, Göttingen (Deutschland)

1607

Namenregister / Name Index A Aarsleff 1142, 1145, 1146, 1149 Abailard 1139 Abasheik 1406 Abbott 1087 Abeillé 406 Abhyankar 206 Abney 530, 1355 Abraham 637, 796, 801, 806, 820, 841, 1047, 1072, 1073, 1557, 1558 Abraham ibn Esra 163 Abramov 1277 Abu Aswad al-Du’ali 208 Adamec 356 Adelung 133, 144, 713, 714, 717, 718, 1044, 1141, 1144, 1146, 1151, 1546 Ader 1549, 1550 Ades 401 Admoni 143, 232—240, 1533, 1535, 1541 Åfarli 801 Aguirre 1515 Ahn 871 Aissen 602, 605, 608, 1382 Aitchison 1173 Ajdukiewicz 18, 395, 412, 1550 Akinlabi 1177 Akmajian 500, 917, 959 Aksakov 141 Al-|Ghazalī 1138 Alambert d’ 717 Alarcos Llorach 259 Albertus 1545 Alemán 1407, 1415 Alexander de Villa-Dei 130, 184 Algeo 217 Allan 830, 836, 837, 839 Allen 1219 Allerton 112, 304, 312, 334, 1001, 1220 Allwood 643 Alsina 598, 746, 1403, 1411, 1412 Alston 1470 Altman 1505 Altmann 884, 982, 1293, 1294 Amborn 663 Ammann 270, 355 Ammon 122, 123 Ananthanarayana 206 Anbari 211 Andersen 902, 903, 1084, 1086,

1090, 1156, 1218 Anderson 15, 127, 304, 331, 332, 334, 335, 642, 675, 676, 702, 979, 1080, 1156, 1169, 1186, 1382 Andersson 643 Andrea Guarna von Cremona 164 Andrews 697, 698, 1111, 1277 Antinucci 1055, 1221 Antipater von Tarsos 162 Anttila 1170 Anward 801 Aoun 518, 523, 535, 540, 709, 711, 712, 715, 718, 802, 870, 871, 873 Apollonios (Apollonius) 89, 91, 101, 132, 133, 135—138, 141, 146, 159, 161, 166—168, 175, 176, 186, 187, 189, 190, 1463, 1467 Apridoniʒe 1378 Arens 131, 1141, 1545, 1547 Aristoteles 132, 137, 147, 161, 162, 174 Aristotle 1137, 1148, 1464—1466, 1468, 1481, 1483 Arnauld 147, 162, 163, 177, 184, 1140, 1149, 1153 Arnold 998 Aronson 1381 Arrivé 266 Asher 1552 Ashton 1397, 1407, 1408, 1413, 1416 Askedal 305 Aßheuer 1549, 1550—1552, 1557—1559, 1561, 1568 Astarabadhi 209, 210, 212 Augst 1548 Austin 388, 1341, 1470 Authier 1415 Auwera 1213 Avis d’ 878 Awóbùlúyì 801, 805, 807, 815 Awóyalé 800, 801, 803, 805

B Baalbaki 209, 210 Babby 556, 1055 Bach 18, 80, 313, 398, 399, 405, 407, 410, 480, 482, 484, 713, 734, 935, 946, 1325, 1498 Bacher 213

Backus 1576 Bacon 967, 971, 1139 Bader 1506 Bailey 122, 1519 Bak 172 Bakalla 1045 Baker 20, 23—25, 27—30, 32, 33, 41, 44, 45, 510, 524, 531, 559, 561, 731, 734, 735, 737, 739, 741, 742, 744, 745, 801, 802, 805, 808, 813, 820, 854, 920, 921, 939, 1293, 1295, 1297, 1299, 1304, 1305, 1307, 1311, 1314, 1410—1412, 1454, 1536 Balhorn 1549 Bally 138, 157, 168, 233, 235, 271, 273, 356, 718, 1530 Baltin 413, 527 Bágbóṣé 800, 801, 803—806, 815 Bańczerowski 137, 138 Bani 1250 Bar-Hillel 18, 261, 396, 397, 407, 412, 475, 618, 1576 Barlow 722, 1236 Barrett-Keach 1398, 1408, 1410, 1412—1415 Barry 1489 Barss 538, 958 Bartsch 398, 619, 1044, 1045, 1047, 1217, 1270, 1310, 1550, 1556, 1562 Barwise 1474 Basilaia 1390, 1391, 1394 Baskervill 217 Bates 649, 829, 1486 Batistella 538 Batliner 1009, 1017, 1018, 1022 Batteux 137 Baudouin de Courtenay 132, 156, 179, 233, 667 Bauer 183, 185, 187, 226, 1112, 1152 Bäuerle 60, 80, 732 Baum 299 Baumgärtner 853 Bautista 1516 Bavin 1430 Bayer 73, 984, 1495, 1496, 1506 Bazell 259, 263, 268, 275, 277 Beaman 829 Bean 1268, 1271, 1278 Beaugrande, de 830, 1565

1608

Beauzée 184, 717, 1144, 1154 Bech 10, 44, 69, 73, 276, 277, 591, 781, 850, 931, 1289, 1292, 1300 Bechert 1103, 1271 Becker 133, 147, 148, 155, 160, 164—167, 174, 176, 177, 181, 182, 185, 225, 226, 232, 1546—1548, 1567 Beckman 52, 136, 144, 152, 158, 169, 172, 183 Bednarczuk 1152 Behaghel 190, 221, 227, 233, 710, 852, 1044, 1151, 1153, 1155, 1158, 1172, 1267— 1270, 1281, 1282 Behrens 689, 883, 885, 1072 Beier 1533 Beleckij 136 Belić 175 Běličová 351 Bell 1110, 1176, 1519 Belletti 530, 531, 541, 597, 1315, 1349—1353 Beneš 356, 358, 1540 Benešová-Buráňová 361 Benfey 1142, 1149 Benincà 1348, 1353, 1359, 1360 Benkö 1557 Bennett 981 Bennis 1239 Benson 341 van Benthem 64, 408, 411, 413 Benveniste 649, 670, 831, 1045, 1155, 1214 Berezin 131 Bergsträsser 1129 Berman 55, 64, 877, 1000, 1277 Berndt 1495 Berneker 1047 Bernhardi 137, 141 Bernhardt 1283 Berwick 1493, 1496, 1502 Besch 122, 123 Betten 767, 1534, 1535 Beutel 1009, 1011 Bever 485, 1171, 1222, 1459, 1485, 1487, 1489, 1497, 1498, 1503, 1557 Bhartṛhari 199—206 Bhat 1103 Bhatia 1095, 1096, 1098, 1112 Biber 829, 830 Bibliander 1139, 1140, 1158 Bichel 1534 Bickerton 636, 801, 1451, 1453, 1455, 1456, 1492, 1518, 1524 Biener 1152, 1153, 1158, 1267, 1269 Biere 305 Bierwisch 9, 26, 49, 121, 122, 124, 254, 484, 618, 733, 738, 746, 782, 848, 850, 1007,

1008, 1495, 1503, 1507 Biljarskij 168 Billroth 137, 141, 143, 144, 152, 168, 176, 181 Biloa 962 Birkenmaier 994—996 Blake 604, 605, 607, 835, 838, 840, 911, 1032, 1077, 1078, 1084, 1085, 1088, 1091, 1094, 1098, 1104, 1181, 1516 Blank 1466, 1467 Blansitt 835 Bleser, de 1485, 1495, 1496 Bloch 283, 284, 290, 296 Bloomfield 54, 56, 90, 91, 94, 103, 108, 134, 217, 225, 233, 281—283, 286—290, 296, 330, 717, 722, 1117, 1128, 1576 Blümel 1169 Boas 167, 1078 Bock 842, 1486, 1491 Bödiker 1546 Boeder 1277, 1380, 1381, 1384 Boehner 177 Boettcher 1549, 1550 Bogusławski 362 Bohas 208 Bohnhoff 677 Boisson 917 Bok-Bennema 1087 Bokamba 1515, 1516 Bolinger 482, 670, 897, 998, 1000, 1001 Bolkestein 346, 347, 837 Bolli 1268, 1272 Bondarko 233, 234 Boole 489 Boomer 1490 Boost 846, 847 Bopp 1141, 1142, 1144, 1145, 1149, 1154, 1156 Borer 513, 514, 518, 670, 1228, 1289, 1494 Borgstrøm 1040 Borsley 329, 573—575 Borst 1140 Bosch 670 Bossong 680, 1098 Bottari 1355 Bouchard 531, 556 Bouma 404, 410, 411, 1430 Bowerman 1178 Bradley 1495 Brailovskij 175 Brame 530, 575, 1355 Brandi 1353 Brandon 1413 Brandstetter 767 Brandt 852, 1006, 1008, 1016, 1020 Brandt Corstius 682 Braun 1533, 1539, 1548 Brauner 1397, 1416

Braunmüller 680, 1269, 1273 Braunroth 1545, 1549 Brauße 1252 Bréal 164 Breeze 670 Breivik 1069 Brekle 484, 1557 Bresnan 72, 133, 329, 395, 498, 528, 565, 581, 583, 584, 588, 590—592, 594—600, 696, 699, 731, 734, 736, 737, 739, 741, 782, 802, 827, 1000, 1230, 1235, 1237, 1365, 1407, 1411, 1432, 1438, 1488, 1498, 1499, 1502, 1550, 1600 Brettschneider 679 Breva-Claramonte 213, 1137—1140, 1148, 1157, 1158 Breznik 141, 154 Brinker 306, 1540, 1541, 1566 Brinkmann 233, 234, 238, 248, 251, 252, 1548 Broadbent 1487 Broberg 140, 189 Brody 950 Brömser 357 Brøndal 163, 258 Broschart 651, 653—658 Brough 201 Brown 217, 313, 832, 900, 1491, 1493, 1498, 1503, 1517, 1519 Browning 498, 539, 546 Brugmann 1147, 1150—1153, 1157, 1277, 1281, 1282, 1547 Bruhns 124 Brünner 1557 Brunot 233 Bugarski 431 Bühler 161, 766, 767, 1485, 1495 Bühler, Ch. 190 Bunting 1549, 1560 Burgess 1089 Burnet 135, 162, 183 Burquest 672 Bursill-Hall 131, 146, 1139 Burstall 1552 Burton-Roberts 972 Burzio 28, 520, 562, 807, 963, 1080, 1081, 1305, 1348— 1350, 1352—1354 Buscha 158, 1548 Buslaev 135, 140, 154, 160, 167, 174 Busse 312 Buszkowski 399, 407, 412, 1566 Butler 340, 344, 1280 Buttmann 137 Butzkamp 1545, 1549, 1551 Byarushengo 1406 Bybee 897, 901, 917, 933, 1060, 1258 Bynon 1147, 1155, 1168 Byrne 185, 257, 801, 1454—1456

1609

C Campbell 803, 1032, 1127, 1158, 1223 Capell 653, 917 Caplan 1495, 1496 Caramazza 1495 Carden 801, 1337, 1457 Cardinaletti 1293, 1297, 1312, 1356, 1359 Cardona 199, 202, 206 Carlson 967—969, 972, 974, 975, 1507 Carnap 288, 395, 434 Carrier 744, 746, 1432, 1438 Carrier-Duncan 732 Carroll 378 Carter 209, 210, 212, 213 Cartwright 1478 Casad 666 Cassiodorus 1137 Castelvetro 1148 Cattell 502 Cedergren 1523 Červenka 356 Chafe 341, 357, 829, 832, 926, 967, 974, 1067, 1206, 1228, 1238, 1323 Chakravarti 1148 Channon 1338 Chao 291, 295, 777, 787—789, 1248 Charachidzé 1081, 1082 Charudeva Shastri 207 Chastellier 1601 Chatman 296 Cheng 535 Cherubim 216, 1034 Cheshire 118, 1520—1522 Chevalier 131 Chiang 1045 Chien 514 Chierchia 410 Chisholm 1114 Chomsky 1, 2, 5—7, 11, 23—26, 37—39, 41, 47, 49, 51, 56, 61, 89, 90, 93—96, 98, 109, 112, 113, 121—123, 216, 217, 242, 270, 283, 329, 330, 333, 356, 358, 359, 362, 397, 398, 405, 407, 414, 431, 436, 437, 440, 465, 469, 470—485, 487—489, 492—494, 498, 500—503, 511, 512, 514, 515, 518, 521—523, 525, 526, 528, 531—533, 536—540, 545—547, 554, 555, 557—560, 562—564, 570, 571, 575, 581, 596, 615, 619—621, 631, 635, 636, 639, 642, 644, 695, 696, 707, 709, 711—716, 718, 735, 737, 738, 748—751, 769—771, 776, 802, 815, 818, 827, 848, 850, 853, 854, 861,

863, 885, 888, 901, 906, 913, 915, 931, 944—946, 950, 953, 956, 959—962, 998, 999, 1032, 1034, 1035, 1045, 1072, 1083, 1167, 1170, 1172, 1177, 1218, 1219, 1273, 1289, 1294, 1297, 1299, 1303, 1304, 1307, 1309, 1310, 1355, 1356, 1364, 1412, 1457, 1460, 1464, 1465, 1471, 1472, 1476, 1481, 1483, 1486, 1488, 1491—1494, 1497, 1498, 1500, 1501, 1519, 1524, 1550, 1576, 1579, 1582, 1589, 1602, 1604 Chrakovskij 1045 Christaller 1594 Christensen 962 Christie 1169 Christophersen 989 Christy 1146, 1147 Chrysippos 162 Chung 715, 717, 718, 1103 Church 397, 408, 1576 Čikobava 1380—1383 Cinque 523, 539, 1296, 1297, 1302, 1309, 1348, 1352— 1355, 1357—1360 Clahsen 1273, 1553 Clark 173, 313, 681, 800 Claudi 1157, 1235 Clausen 848, 855, 1267 Clements 564, 677 Clifton 1500 Closs 1169, 1399 Closs-Traugott 1170 Cole 909, 1277, 1280, 1284 Collart 131 Collberg 1284 Collinge 1154, 1155 Collis 259, 261, 262 Colmerauer 1601 Comenius 1141, 1546 Comrie 116, 602, 677, 687, 688, 700, 733, 734, 740, 741, 746, 839, 840, 883, 890, 891, 910—912, 925, 933, 1032, 1033, 1037, 1043, 1072, 1077, 1078, 1081, 1085—1087, 1092, 1097, 1101, 1104, 1109, 1110, 1114, 1115, 1128, 1156, 1176, 1178, 1179, 1186, 1203, 1237, 1239, 1241, 1244, 1246 Condillac 152, 170, 174, 1148, 1149, 1151 Connell 1240 Connolly 341, 387, 393, 841 Conrady 1550 Contini-Morava 1399, 1406 Contreras 883 Cooper 359, 834 Corbett 728, 1236, 1241 Cordemoy 1151, 1156

Cordin 1353 Cormack 405, 749, 871 Corne 1457 Coseriu 123, 649, 650, 1043— 1145, 1156, 1168, 1171 Court 800 Covington 136, 146, 147, 159, 213, 1139 Cowan 213 Coward 207 Craig 400, 412, 800, 837, 1151, 1157, 1440 Crain 514, 580, 1492, 1500, 1501, 1505 Creider 801 Cresswell 60, 64, 267, 398, 411, 412 Croft 647, 649, 651, 976, 1032, 1033, 1035, 1036, 1041, 1077, 1078, 1081, 1087, 1101, 1115, 1128, 1131, 1132 Crook 1227 Crowley 800 Cruttenden 1001 Crystal 829, 1000 Culicover 959, 998, 1072, 1493 Culy 475 Curme 218, 220 Curry 398—402, 405, 412 Curtiss 1491 Curtius 1145, 1147, 1149 Cutler 1060

D Dahl 298, 327, 399, 915—918, 967, 972, 975, 976, 1052, 1063, 1092, 1247, 1311 Dalgarno 163 Dalrymple 597, 943 Damoiseau 801, 818 Damourette 922 Daneš 343, 350—352, 355—358, 362, 886, 1047 Dante 1140, 1141, 1153 Das 202 Davidse 340, 343, 358 Davidson 56, 449, 732 Davies 602, 604, 605, 1112, 1147, 1384 Davis 711, 1496 De Guzman 1423 DeCarrico 953 Déchaine 820 Dederding 311 DeGraff 801, 819, 1459 DeLancey 831, 1098, 1102, 1147, 1155 Delbrück 149, 281, 680, 1116, 1123, 1150, 1151, 1153, 1154, 1156, 1269, 1271, 1281 Delfitto 870 Dell 564, 1429

1610

Demers 654, 836 Demirdache 536 Democritus 1463 Demuth 1240 den Besten 528, 1080, 1081, 1271, 1298, 1300 den Dikken 801—803, 811 Denkel 352 Denning 1521 Derbyshire 687, 693, 842, 1050, 1089, 1090, 1112 Descartes 1140 Desclés 405, 411 Destutt de Tracy 168, 170, 175, 182 Deutsch 1492 Dezső 1043 Di Benedetto 132 Diderichsen 134, 258, 262, 266, 269, 270, 272, 277, 1141— 1144, 1147, 1267 Diderot 717, 1145 Diem 672 Diesing 73 Dik 101, 336, 340, 341, 344— 347, 368—370, 386, 387, 391, 393, 681, 690, 749, 750, 771, 827, 832, 833, 838, 1083, 1093, 1186, 1278, 1514 Dimmendaal 837, 1400 Dines 1521, 1522 Dinser 1170 Diogenes Laertios 145, 161 Diokles 145, 159 Dionysios 163 Dionysios Thrax 132, 1545 Dionysius Thrax 646 Dirven 339 DiSciullo 16, 20, 25, 28, 30, 31, 33, 510, 737, 939, 1516 Dittmann 243, 245, 250—252 Dittmar 123, 1518, 1521 Dittrich 150, 180 Dixon 649, 652, 662, 666, 912, 1070, 1077—1079, 1081, 1084, 1087, 1089, 1091, 1092, 1098, 1101, 1186, 1239, 1298 Djakonov 1089 Dobiaš 134, 138, 141, 146, 155 Dobrovie-Sorin 1354 Doderer 951, 1549 Dogil 1496 Doke 667 Dokulil 349, 351 Donatus 1137 Donhauser 1023 Donnet 131—133, 136, 137, 144 Dönnges 312 Dougherty 748, 751, 771, 946, 955 Dowty 55, 56, 60, 69, 396, 403— 405, 410, 427, 695, 733, 734,

738, 739, 742, 932, 935, 936, 1080, 1081, 1092, 1383, 1567 Drach 235, 847, 1044, 1267, 1540, 1548 Dresher 1219, 1485 Dressler 633, 830, 1044 Driever 1410, 1411 Dronsek 1495 Drösser 1133 Dryer 711, 734, 835, 837, 838, 918, 1051—1055, 1057, 1059, 1060, 1220, 1311 DuBois 833, 903, 1094 Dubost 312 Ducrot 915 Duden 226, 311 Dumarsais 163, 182, 184 Dumézil 691, 1095, 1103 Dunnigan 678 Duponceau 1146 Duran 163 Durand 331, 332 Duranti 834, 836, 1055, 1236, 1237 Durie 800 Dutra 833 Dyscolos (Dyscolus, Dyskolos)

cf. Apollonios E Eastman 842 Ebert 484, 970, 1156, 1269, 1272, 1560 Edelman 1098 Edelstein 1549 Edmondson 980, 1271, 1289 Edwards 1518 Egger 131, 146, 186 Egli 80, 131, 161 Eichler 1549, 1560 Einarsson 1284 Eisenberg 249, 435, 466, 774, 848, 1541 Eisenhardt 1134 Ellendt 152 Elmedlaoui 564 Elmer 1280, 1281 Éménanjọ 801, 810, 812, 816 Emeneau 1124 Emonds 483, 519, 522, 540, 1170, 1346, 1351 Emons 306, 311, 321 Endres 1557 Engdahl 75, 76, 539, 643, 877, 944, 945, 963 Engel 134, 311, 312, 325, 848, 855, 865, 1267, 1534, 1535, 1539, 1549, 1556, 1557, 1559 Engelen 1548, 1551, 1561, 1562 Engelkamp 313 England 1091 Enkvist 830

Erben 233, 234, 1267, 1536, 1548, 1550 Erdmann 186, 892, 893, 1151, 1267, 1269 Erguvanli 883, 884, 890, 891 Erlinger 1566, 1568 Ermert 1538 Eroms 306, 311, 325, 1530, 1541, 1549 Ertel 831, 832 Erteschik-Shir 998, 1001 Esau 484 Essen 1548 Esser 357 Estival 1186 Etzler 142, 173 Evans 80, 951, 960 Everaert 674 Evers 14, 484, 1293, 1498 Ézikèójìakụ 810, 816

F Fabb 510, 814 Faber 1001 Fabricius-Hansen 60, 71 Fágbrun 808, 809 Fähnrich 1380 Falkenberg 1008 Faltz 399, 405, 674, 700, 1400 Fanselow 1, 38, 695, 946, 955, 1172, 1293, 1295, 1298, 1299, 1302, 1307, 1309, 1567 Faraclas 1240 Farghaly 1004 Farkas 971, 974 Farmer 1071, 1365, 1441, 1447 Farra 209 Farrell 607 Fasold 1519 Fassi Fehri 528 Fauchois 801, 812, 819 Fauconnier 725 Fearn 142, 163, 177 Felix 1, 1172, 1294, 1494, 1549, 1552, 1566, 1567 Fenk-Oczlon 631 Ferguson 722, 1492, 1517 Féry 50, 51, 53, 54 Feys 398, 400, 401, 405, 412 Fichte 1149 Fick 1149, 1150, 1155, 1156 Fiengo 535, 774, 784, 955, 962 Filipec 356 Filipovich 812 Fillenbaum 924 Filliozat 206 Fillmore 106, 304, 313, 440, 465, 482, 490, 681, 682, 690, 831, 909, 910, 1066, 1093, 1380, 1524, 1550, 1555 Finck 134, 1044 Fink 1548

1611

Finney 803 Firbas 270, 343, 355, 358, 360— 362, 837, 886 Fischer 435 Fitialov 316, 323 Fleischer 1538 Fleischman 1222, 1248 Fleischmann 1268, 1269 Fleury 130 Flores d’Arcais 1488 Florstedt-Stieber 1548 Flynn 397 Fodor 64, 480, 484, 485, 580, 1218, 1480, 1482, 1487, 1489, 1497, 1500, 1503—1506, 1508, 1509 Foldi 1496 Foley 339, 651, 688, 700, 834, 910, 1068—1071, 1082, 1083, 1111, 1114, 1181 Fong 514 Forchheimer 670 Ford 923, 1488, 1490, 1499 Forsgren 226 Forster 1487, 1489, 1499 Forsyth 1247 Fortescue 1087, 1112 Fortunatov 175 Fought 280 Foulet 1047 Fourquet 258, 263, 267, 269, 270, 275, 296, 1269, 1270, 1272, 1548, 1558 Fox 342, 1426 Frachtenberg 1087, 1089, 1090 Frampton 546 Franchetto 1086 Francis 122 Frank 1545—1547, 1566 Frantz 836 Fraser 331 Frauenfelder 1488 Frazier 1057, 1176, 1488, 1500, 1504—1506 Freedman 1500 Frege 57, 66, 84, 395, 618, 867, 1467 Frei 1222 Freidin 538, 556, 715 Frey 871, 976, 1549 Fried 339 Friederici 1495 Friedin 1219 Friedman 404 Friedrich 1151, 1179, 1223 Fries 291, 295, 648, 652, 1011, 1013, 1026, 1278, 1281, 1312 Fromkin 1490 Fromm-Sadeniemi 1277 Fuchs 1001 Fujii 1068 Fukui 513, 527, 850, 1365 Fulda 1151, 1152, 1155

G Gaaf 1279, 1280, 1281 von der Gabelentz 133, 147, 148, 162, 164, 165, 179, 180, 182, 185, 188, 270, 355, 356, 1044, 1045, 1147 Gaifman 136, 283, 316, 318, 319, 327, 331, 353, 355 Gair 1096, 1097 Gaiser 1548 Gallmann 226 Gamal-Eldin 1112 Gamillscheg 627 Gamkrelidze 1124, 1129 Garbe 1545 Gardner 1496 Garrett 485, 1487, 1489—1491, 1495, 1497, 1503 Gary 734, 1112 Gaurinath Sastri 206, 207 Gazdar 69, 96, 133, 270, 329, 395, 475, 525, 570—572, 575, 578, 580, 716, 728, 748, 751, 771, 788, 789, 827, 840, 950, 959, 1364, 1499, 1500, 1600, 1601 Geach 68, 399, 407, 408, 410— 412, 940, 946, 947 Gebauer 150 Gebert 1055 Geckeler 244 Geertz 900 Geis 923 Geißler 1546 Gelenius 1141 Geluykens 340, 1009 Geniušienė 674 Gentzen 398 Georgopoulos 501 Gerritsen 1268, 1273 Gerstmann 897 Gerstner(-Link) 967, 968 Gerzymisch-Arbogast 356 Geschwind 1496 Gessner 1140, 1141 Gewehr 1556 Gibson 1489 Gil 870, 874, 1429 Gilii 1141 Gilligan 1060 Gilman 900 Ginsberg 973 Giorgi 1315, 1350, 1352, 1355, 1357 Gipper 242, 244, 245, 251 Girard 133, 147, 155, 1144 Giusti 1356—1358 Givón 339, 345, 652, 657, 663, 689, 691, 702, 829, 837, 842, 886, 894, 895, 897, 902, 903, 916, 1032, 1043, 1057—1059, 1062, 1063, 1083, 1121, 1153,

1155—1157, 1220, 1228, 1229, 1235—1241, 1248, 1407, 1415 Gleason 218, 290, 291, 652, 1550 Gleitman 169, 516, 748, 751, 760, 777, 1289, 1492 Glinz 134, 147, 149, 186, 187, 225, 226, 242, 252, 253, 258, 264, 1547—1549, 1557, 1561 Glück 1533 Gobyn 1536 Goethe 1145, 1146 Goetz 134 Goldman-Eisler 1489 Goldsmith 19, 32, 52, 564, 801 Goldstein 1495 Goodall 755, 774 Goodglass 1495 Goodman 923 Goodwin 1249 Goossens 385, 1219 Gorayska 331 Goropius 1141 Gorrell 1489, 1507 Gottsched 1546 Götze 312 Graffi 1355 Grassmann 1116 Graucob 1548 Greaves 341 Grebe 1556, 1557 Greč 147, 167—169, 172 Green 801, 924 Greenberg 211, 282, 295, 637, 676, 681, 682, 687—689, 692, 839, 842, 843, 902, 903, 912, 918, 926, 1032—1036, 1039, 1040, 1043, 1046, 1047, 1050, 1051, 1053—1055, 1057, 1059—1061, 1063, 1084, 1088, 1094, 1109, 1110, 1113, 1133, 1146, 1147, 1176—1178, 1180, 1181, 1216, 1230, 1238, 1270, 1399 Gregor 1036 Gregor von Korinth 133, 137, 144 Gregores 1098, 1099 Gregory the Great 1137 Greule 312 Grewendorf 18, 615, 616, 863, 1006—1008, 1080, 1081, 1271, 1289, 1293, 1295, 1299—1307, 1310—1312, 1314, 1316 Grice 352 Grierson 1078, 1086, 1098 Griesbach 134, 848, 1267 Grigorjan 136 Grimm 221, 225, 1116, 1142, 1146, 1153, 1281 Grimshaw 410, 502, 510, 516, 565, 746, 802, 1355, 1492

1612

Grinder 779 Grodzinsky 1492, 1495, 1496 Groenendijk 78, 877, 951 Groot, de 99, 258, 259, 263, 264, 268, 272—274, 351, 375 Groothusen 1503 Gross 953 Gruber 55, 565, 733 Grunig 131, 133, 137, 153, 266, 270 van Grunsven 1495 Gruntfest 211 Gruskij 131 Guarna 1148 Guasch 1099 Guasti 1355 Gueintz 1545 Guerssel 1438 Guillaume 208 Guillaume de Conques 34 Guiraud 1210, 1529 Gulya 1142 Gulyga 233, 234 Gumbrecht 1528 Gumperz 1517 Gundel 881, 883, 885, 886, 895 Gunji 1364, 1365, 1372 Güntert 224 Günther 743 Gussenhoven 998, 1001 Guthrie 1401 Gyarmathi 1143, 1158

H Haarman 209, 211 Haas 1238 Haase 148 Habel 122 Haberbeck 435 Habrum 1169 Haegeman 30, 74, 1239 Haftka 848, 850—852, 863—865, 1023 Hagège 649, 650, 677, 681, 1218 Hagoort 1503 Haider 692, 696, 746, 850, 1214, 1219, 1271, 1272, 1292, 1293, 1295, 1297, 1300, 1302, 1303, 1311, 1492, 1496 Haïk 871, 948, 952, 960 Haiman 339, 661, 680, 837, 897, 901, 903, 923—929, 1041, 1114, 1156, 1169, 1171, 1222—1224, 1277 Hajičová 331, 332, 334, 335, 342, 343, 353—355, 357, 360—364 Hakes 1489 Hale 565, 696, 702, 800, 914, 1071, 1151, 1157, 1237, 1298, 1300, 1365, 1430—1436, 1438, 1440—1442, 1444— 1447

Hall 260, 270, 1148, 1181 Halle 2, 49, 51, 480, 488, 564, 888, 999, 1035, 1168 Halliday 98, 99, 339, 340, 344, 356—358, 835, 837, 998— 1001, 1047, 1514, 1532 Hamblin 74 Hammond 1114 Hamp 1117 Hancock 1452 Hankamer 750, 757, 774, 775, 777, 779, 781, 784, 1335 Hanlon 1493 Hannay 341, 344, 346, 833 Hansen 150, 176 Hanzeli 1142 Happ 304, 306, 312 Harbert 693, 1277, 1280, 1284 Harbsmeier 1046 Harlow 1055, 1056, 1060 Harman 218, 477, 525, 571 Harper 323 Harries 1401, 1410 Harris 96, 98, 134, 137, 142, 163, 174, 175, 217, 277, 280, 282, 284, 288, 289, 291, 294—296, 469, 472, 477—479, 484, 512, 527, 607, 697, 770, 890, 928, 1102, 1136, 1140, 1151, 1152, 1156—1158, 1186, 1216, 1217, 1220—1224, 1237, 1241, 1249, 1278, 1378—1384, 1390, 1392— 1394, 1519, 1525, 1546, 1576 Harrison 283 Harsdörffer 1545 Hart 286 Hartig 1549—1552, 1557—1559, 1561, 1568 Hartmann 249, 990, 1277, 1549 Hartnacke 1548, 1563 Hartung 484 Harweg 1557 Hasegawa 1365 Haspelmath 674 Hassan 837 Haudry 1206 Hauenschild 997 Hausser 410 Havers 186, 224, 1116, 1117, 1166, 1171, 1281 Havránek 257, 259, 260, 349, 1532 Hawkins 637—640, 826, 838, 839, 880, 881, 886—889, 893—895, 988—990, 993, 994, 996, 1032, 1035, 1040, 1051, 1054, 1055, 1057, 1059— 1063, 1088, 1094, 1109, 1110, 1113, 1114, 1116, 1125, 1130, 1155, 1176, 1178—1180, 1217, 1218, 1220—1224, 1278 Hawkinson 1410

Hayes 49, 52—54 Hays 136, 283, 313, 316, 318— 320, 323, 327 Haywood 177 Heath 311, 842, 1032 Heeschen 243, 251, 767 Hegel 1149 Heger 118, 123, 990 Heidolph 248, 852, 883 van Heijnoort 412 Heim 58, 63, 73, 76, 77, 79, 80, 83, 554, 951, 952, 962, 971, 972, 989, 990, 994, 996 Heine 1132, 1155, 1157 Heinemann 1530 Heinrichs 212 Helbig 131, 158, 244, 245, 247, 252, 305, 306, 311, 312, 1128, 1548, 1557, 1559 Hellan 962 Hellwig 331, 333 Helmers 1549 Hemforth 1554 Hempel 470 Hemsterhuis 1141 Hendriks 411 Hengeveld 345, 346, 385 Henn(-Memmesheimer) 127, 1534 Henning 1282 Henrici 1549, 1550, 1552, 1567, 1568 Henry 655 Henzen 247 Hepple 404, 405, 411 Herbart 281 Herbst 311, 312 Herder 1146, 1151, 1157, 1546 Heringer 306, 311—313, 325, 330, 1529, 1548, 1550, 1552, 1560, 1561, 1568 Herling 152, 153, 159, 164, 168—173, 226, 1267 Hermann 648, 757, 1153 Hermes 1579 Hermon 1277, 1280, 1284 Herms 1397, 1416 Herrlitz 1550 Hervás y Panduro 1141, 1142 Herzog 1168, 1170 Hess 654 Hesse 325 Hestvik 55, 64 Hettrich 678 Hetzron 680 Heuer 305, 1562 Hewitt 1086, 1087, 1112 Heyse 137, 149, 165—167, 174, 226 Heyse, K. W. L. 226 Hickmann 339, 341 Hickok 1489 Hiecke 1547

1613

Hiersche 1557 Higginbotham 78, 81, 535, 537, 538, 548, 550, 565, 740, 744, 945, 946, 949 Hilbert 1469 Hildebrandt 1495, 1496 Hilgard 132, 137 Hill 282, 1227 Himmelmann 655, 1069 Hinch 917 Hindley 412 Hinds 1069, 1112 Hinnebusch 1397, 1410 Hinrichs 34 Hippel 189 Hirschfeld 213 Hirt 1150, 1152, 1153 Hjelmslev 132—134, 217, 258, 259, 264, 265, 268, 272, 288, 289, 296, 714, 715 Hoberg 1306 Hock 1147, 1152 Hockett 95, 134, 263, 280, 291— 295, 351, 474, 652, 718, 1047, 1065 Hoeksema 405, 411, 950, 951, 986 Hoekstra 802, 933, 936, 1400 Hoenigswald 1117, 1143, 1146, 1155, 1158 Hoenkamp 1490 Hoffmann 1127 Hoffmeister 149 Hofling 1237 Höftmann 1397 Höhle 9, 16, 871, 1017, 1072, 1267, 1305, 1312 Hoji 870—872, 874, 1365 Holenstein 628 Holes 1112 Holisky 1100, 1102, 1378, 1380, 1383 Holm 1517 Holmback 996 Holmes 1490 Homberger 1550 Homer 1137 Hoppenkamps 1545 Hopper 649—651, 657, 720, 836, 1038, 1039, 1133, 1146, 1153, 1157, 1406 Horacek 1272 Hořejši 351, 356, 357 Hörmann 313 Horn 738, 921, 922, 1331 Hornstein 55, 537, 718, 942, 943, 1219, 1485 Horrocks 1514, 1519 Horvath 1002, 1004, 1072 House 218 Householder 137, 146, 282 Hovdhaugen 1136—1139 Howard 1369

Høysgaard 139, 141, 155, 157, 158, 163, 183, 187, 188, 190, 257 Hualde 1112 Huang 48, 500, 502, 513, 518, 523, 535, 536, 545, 552, 781, 870—872, 962, 963, 1303, 1312, 1508 Huber 484 Huck 396, 405, 413 Hudson 298, 311, 313, 321, 325—327, 331—336, 404, 710, 712—715, 717, 718, 827, 1517, 1520, 1525 Hughes 60 Hukari 579, 580 Hull 877 Hülser 145, 161, 162 Hulst 51 Humboldt 134, 216, 225, 232, 242, 243, 485, 507, 1032, 1145—1147, 1153 Hundsnurscher 1549 Hunt 168 Hunyadi 871 Hurford 682 Husmann 1496 Hussein 800 Husserl 395 Hutchinson 801, 812 Hutchisson 671, 674, 675 Hyams 514 Hyman 677, 803, 836, 1121, 1123, 1124, 1221, 1237, 1250, 1271, 1410 Hymes 280 Hyvärinen 312

I Iatridou 801, 1455 Ibn Ḥajib 209 Ibn Barun 213 Ibn Fâris 1138 Ibn Ḥazm 1158 Ibn Hisham 209 Ibn Janâḥ 1138 Ibn Jinni 208, 209, 211, 1138 Ibn Khaldûn 1150, 1157 Ibn Maḍa 209 Ibn Qureish 213 Ibn Ya’ish 209 Ibrahim 1150 Ickelsamer 1545 Ígwè 801 Íhìn 801 Ihm 136, 181, 323 Ihre 1142 Ineichen 1043—1045, 1047 Ingendahl 1548, 1551 Ingram 670 Inkelas 51 Innes 1004

Inoue 1365, 1506 Ioup 872, 874, 875 Isačenko 768 Isard 1487 Ishii 536 Ishikawa 1365, 1366 Isodorus 1138 Isokrates 162 Itkonen 207, 1217, 1218 Ivanov 1124, 1129 Ivić 131

J Jackendoff 96, 112, 329, 356, 359, 484, 492, 493, 498, 511, 526, 527, 532, 565, 619, 620, 690, 691, 744, 758, 785, 802, 853, 871, 953, 955, 959, 967, 985, 998, 999, 1170, 1177, 1322, 1420, 1555 Jackson 1127 Jacobs 74, 355, 356, 359, 362, 448, 690, 694, 732, 796, 871, 873, 878, 881, 887, 888, 891, 979, 983—986, 1017, 1072, 1132, 1226—1228, 1309— 1311, 1529, 1542, 1553, 1568 Jacobsen 653, 654, 670, 836, 837, 1136 Jacobson 405, 410, 413, 575, 947, 949, 1092 Jaeggli 518, 540, 801, 1304, 1493 Jäger 1142 Jagić 131, 186, 190 Jaimini 199 Jakobson 257, 258, 349, 350, 623, 896, 903, 1035, 1045, 1179, 1253 Jakubovicz 1492 Jankowsky 1143, 1145, 1146, 1153 Jansen 800, 802 Jantsch 1134 Jäppinen 331 Jedlička 259, 260 Jelinek 654, 836, 906, 1230, 1430, 1432 Jensen 1382 Jespersen 105, 137, 141, 151, 154, 168, 180, 216, 218—220, 233, 284, 332, 355, 835, 903, 917, 923, 925, 927, 1119, 1147, 1152, 1155, 1158, 1218, 1280, 1281 Jodłowski 142 Johansen 1207, 1559, 1560, 1566 Johns 531 Johnson 363, 559, 602—604, 608, 687, 801, 902, 1077, 1085, 1304 Johnson-Laird 899, 928, 1487, 1508

1614

Jokinen 411 Jolly 1152, 1153 Jones 127, 935, 1142, 1143, 1169 Jong, de 828 Joo 871, 872, 874, 877 Joos 1551 Joseph 1112 Joseph ben Abba Mari ibn Kaspi 163 Joshi 206, 395, 406, 1491 Juda ben David Hayyoudj 163 Juilland 715 Julien 166, 182 Jung 713, 716 Jurjani 209—212 Justeson 1055, 1058

K K’iziria 1382 K’vant’aliani 1386 Kac 571, 933 Kachru 1095 Kadmon 952 Kahn 49 Kahr 680, 681 Kaisse 1228, 1229 Kakumasu 1089 Kamp 73, 79, 80, 951, 971, 976 Kaneko 1557 Kanerva 598—600, 699, 802, 1411 Kang 411, 946 Kanngießer 122, 1170 Kaper 1551 Kaplan 402, 588, 595—597, 1365, 1498, 1499, 1502, 1600 Karcevskij 153, 273, 352 Karjalainen 1112 Karttunen 74—77, 80, 410, 411, 877, 947, 948, 963 Kashket 1430 Kaspar 572 Kātyāyana 199, 200, 202, 203 Katz 480, 484 Kaufmann 742 Kavanagh 1271, 1272 Kay 970, 981, 1524, 1600 Kayne 403, 501, 502, 523, 530, 531, 539, 540, 638, 641, 1299, 1303, 1347, 1349, 1353— 1355, 1357, 1359 Kean 1490, 1495 Keddad 1516 Keenan 133, 398, 399, 405, 602, 637, 638, 675, 676, 687—689, 696, 699, 700, 719, 720, 734, 739, 829, 839, 842, 877, 885, 912—914, 949, 1037, 1040, 1043, 1045, 1047, 1050, 1056, 1060, 1062, 1076—1078, 1082, 1085, 1088, 1111, 1176, 1178, 1179, 1203, 1239, 1280, 1337, 1429, 1556—1558

Kefer 907 Keijsper 361 Keil 91 Keim 1552 Keizer 342, 345 Keller 1259 Kellogg 142, 218 Kemmer 1093, 1210 Kempen 1490 Kempson 749, 871 Kendon 1430 Kennedy 1535 Kenstowicz 1004 Kerkhoff 1550 Kern 140, 158, 172, 189, 227, 1548 Keyser 528, 565 Khalid 1397 Khamisi 1408, 1410, 1411 Khan 1097, 1098 Khanmagomedov 917 Kibrik 1082, 1104, 1184, 1239 Kiefer 361 Kikai 1521 Kilwardby 174, 179, 1139 Kim 536, 871, 875, 1060 Kimball 1504 Kimenyi 734, 742, 1406, 1411 Kindt 770 King 122, 1168, 1169 Kinkade 654 Kintsch 313 Kinyarwanda 1411 Kiparsky 206, 732, 733, 738, 746, 963, 1145, 1154, 1168 Kirsner 1410 Kiss 592, 693, 696, 871, 880, 890, 1072, 1073 Klaiman 837, 1086, 1093, 1098, 1277 Klavans 55, 1228—1230 Kleene 398, 1576, 1589 Klein 26, 27, 60, 122, 124, 270, 316, 356, 570, 575, 580, 716, 766, 768, 770, 772, 774, 777, 784, 788, 792, 795, 796, 827, 935, 1270, 1271, 1364, 1518, 1521, 1549, 1550 Kleineidam 1549 Kleinschmidt 142 Kleist 1485, 1495 Klemensiewicz 138, 151, 173 Klemm 159 Klenk 1599 Klima 480, 483, 489, 518, 526, 899, 915, 916, 920, 921, 941, 953, 1168 Klimov 1046, 1076, 1098, 1099, 1103, 1186, 1278 Klokeid 836, 1250 Knobloch 1568 Koch 308, 962, 1278, 1555 Kochan 1549

Kodzasov 1239 Koefoed 273 Koehn 668, 1089 Koenig 928 Koenitz 355, 357, 360, 362 Koenraads 1171 Koerner 127, 1153 Kohrt 484, 1271 Koktová 343, 359 Kolb 247, 1502 Kolk 767, 1495 Koller 161, 305 Kölver 1044 König 980, 981, 984, 1258 Koopman 531, 800, 801, 813, 945, 1303 Kopchevskaya 918 Korhonen 112, 311, 312 Kornai 1589 Körner 257, 1046 Kornfilt 530, 1113 Koschmieder 175, 1118 Koskenniemi 1589 Kossuth 1284 Koster 27, 641, 644, 816, 946, 958, 1061, 1271, 1303, 1308, 1311, 1492 Kovatcheva 1281 Krallmann 1548 Krámský 990 Kratzer 972, 1300, 1311 Kraus 1143, 1158 Krause 1546 Kreiman 670 Kress 1041, 1549 Kreye 1541, 1542 Krifka 56, 977, 1407, 1413 Křižková 355, 920 Kroch 413, 527, 869, 872, 1127, 1486, 1491, 1521 Kroskrity 836 Krüger 220 Kruisinga 220 Krzeszowski 118, 122 Kubińsky 154, 608 Kudrjavskij 130, 174, 175, 178 Kuentz 1529 Kuhn 469 Kühner 142, 149, 164, 165, 182 Kuipers 654, 918, 1034 Kukenheim 131, 213 Kulagina 259, 317 Kummer 484 Kunjunni Raja 199, 200, 207 Kuno 357, 360, 770, 775, 785, 786, 788, 831, 834, 836, 870, 871, 955—959, 961, 963, 1057, 1171, 1220, 1277, 1337, 1343, 1369, 1374 Kunze 310, 317, 325, 330, 331, 334, 335 Kuroda 360, 531, 536, 870, 1369 Kürschner 1007, 1565

1615

Kurth 1134 Kurtzman 1508 Kuryłowicz 266, 350, 351, 1117, 1155, 1186, 1278 Kurzová 1199, 1278 Küstner 325 Kutas 1503

L Laberge 1250, 1521 Labov 118, 121, 1127, 1168, 1170, 1521—1524 Lacabiʒe 1394 Ladd 998, 1001 Ladefoged 1000, 1487 Ladusaw 1382 Laet, de 1141 Lahiri 49, 51—54, 78 Laka 531 Lakoff 61, 363, 487—491, 526, 538, 749, 783, 801, 871, 902, 921, 932, 946, 953—956, 958—960, 1092, 1332, 1335, 1550 Lamb 296, 351 Lambek 261, 396—399, 404— 408, 412 Lambrecht 342, 726, 1220, 1221 Lamy 1151, 1157 Lancelot 162, 163, 177, 184, 1140, 1149, 1153 Lane 141, 163, 167—170 Lang 782, 795, 869, 1006—1008 Langacker 505, 518, 899, 953, 955, 1147, 1154—1156, 1227, 1228, 1262 Langdon 1223 Langen 1535 Langendoen 538, 1171, 1459 van Langendonck 331, 334, 335 Láníran 801, 804—807 LaPointe 1228 Lappin 998, 1001 Larcher 211, 212 Larive 130 Larson 35, 532, 802, 875, 1316, 1446, 1447 Lascaratou 883 Lasnik 1, 23, 39, 48, 83, 505, 511, 513, 514, 516, 518, 521—523, 535, 537—540, 545, 546, 550, 551, 554, 557, 559, 560, 564, 707, 711, 715, 955, 959, 960, 962, 1356, 1500 Lass 1178, 1218, 1222—1224 Lasser 1506 László 312 Latour 311 Laughren 1430, 1431, 1433— 1436, 1438, 1441, 1442, 1444, 1445 Lavandera 1522 Laver 1490

Law 1139, 1141, 1142 Lawler 967, 968, 972, 974, 975 Lawrence 1040 Lazard 1099, 1103 Lazaresku 312 Lazarus 313 Le Laboureur 1150 Lebeaux 538, 554, 962 Lecerf 136, 181, 323 Lederberg 1516 Leech 313 Lees 469, 480, 483, 484, 487, 526, 907, 953 Lefebvre 802, 812, 820, 1454 Lehmann 654, 678, 681, 707, 708, 712, 713, 715—717, 720, 725, 727, 827, 1040, 1046, 1047, 1051, 1053—1056, 1059, 1060, 1063, 1114— 1120, 1124, 1129, 1132, 1151—1153, 1155, 1166, 1167, 1176, 1179, 1199, 1203, 1208, 1217, 1218, 1223, 1236— 1240, 1270, 1278, 1288, 1311 Lehr-Spławiński 154 Lehtinen 719 Lehtola 331 Leibniz 1141, 1142, 1145 Lemieux 1521 Lenerz 694, 841, 848, 878, 992, 993, 1006, 1167, 1169, 1171, 1173, 1219, 1267, 1269, 1271—1273, 1277, 1282, 1288, 1289, 1306, 1553 Lenneberg 1491 Leonard 1398 Leopold 1146 Lerch 1045 Lerot 1561 Leska 1534 Leśnewski 395 Levelt 312, 767, 1487, 1490 Levin 410, 597, 598, 1079 Levine 579 Levinson 963 Lewis 58, 70, 79, 141, 397, 411, 719, 928, 932, 951, 971, 972 Lewontin 509 van Leynseele 801 Li 708, 746, 800, 803, 804, 870, 871, 873, 1040, 1047, 1048, 1066—1069, 1071—1074, 1121, 1169, 1220, 1221, 1409, 1556 Liberman 50 Lichtheim 1495 Liddy 902 Lieb 121, 123, 124, 352, 431— 436, 443—445, 447, 448, 461, 463, 465, 1128, 1170 Lieber 16 Lightfoot 122, 514, 711, 718, 1128, 1134, 1136, 1153, 1154,

1156—1158, 1170, 1171, 1173, 1218, 1219, 1223, 1224, 1229, 1273, 1280, 1281 Lima 312 Linacre 257 Lindroth 258, 259 Linebarger 73, 1495 Link 422 Liu 942 Löbner 989, 990, 993—997 Locke 1140, 1144, 1148 Lockwood 1283 Loeweke 1089 Loflin 1169 Löfstedt 1212 Longacre 98, 680, 830, 923, 1111, 1393 Longobardi 502, 530, 539, 870, 872, 873, 1348, 1355, 1357 Lonzi 1352 Lord 680, 801, 811, 812 Łoś 146, 154, 160 Losee-Furbee 753 Lötscher 361, 1073, 1074 Ludolf 1142 Lugebil 146, 148, 155, 176, 181 Lumsden 812 Lust 1492 Lutz 357 Lutzeier 435, 465 Luukko-Vinchenzo 1011, 1019, 1020, 1027 Lyon 1548 Lyons 107, 111, 396, 397, 412, 646, 648, 649, 675, 676, 718, 831, 1077, 1081

M Maas 1566 Mač’avariani 1383 Macaulay 1169 MacDonald 926, 1489, 1508 Mackenzie 342, 345 MacLeish 1278 MacWhinney 649, 829, 1486 Madvig 164 Magloire-Holly 818 Magometov 1239 Mahajan 531 Mainville 1516 Malachowskij 236 Maling 578, 750, 944 Mallinson 835, 838, 840, 1032, 1077, 1078, 1084, 1088, 1094, 1098, 1104, 1112, 1181 Manandise 1230 Manaster-Ramer 571 Manfredi 801, 805, 806, 810, 812, 819 Mangal Deva Shastri 205 Manzini 556, 938, 962, 963, 1315, 1353, 1354, 1494

1616

Marácz 707 Marantz 510, 531, 695, 737, 741, 811, 1079, 1095, 1410 Maratsos 1498 March 142 Marchand 650, 1281 Marchese 678, 923, 926, 1240, 1250 Marcus 136, 143, 317, 323, 1502 Marlett 607 Marouzeau 1119—1121 Marr 1485 Marshman 182 Marslen-Wilson 313, 1487, 1488, 1498, 1506 Martin von Dakien 159 Martinet 95, 132, 134, 233, 259, 1045, 1077, 1078, 1081 Martinus de Dacia 136 Marty 176, 355, 1252 Masica 1053, 1095, 1223, 1277 Massam 814 Mater 692 Materna 363 Mathesius 270, 349, 351, 355, 1047 Matilal 200, 202, 204 Matisoff 800 Matthews 90, 98, 102, 107, 167, 305, 306, 311, 330, 332—334, 526, 718, 720, 953, 1502 Matthiessen 342 Matzel 1014, 1541 Mätzner 218 Maurer 1267—1269, 1282, 1579, 1581, 1593, 1594, 1599 Maw 1397, 1407—1409 May 55, 61, 63—65, 78, 81, 83, 522, 533, 535, 739, 871, 877, 878, 941, 942, 944, 945, 948, 949, 962, 963, 1089, 1508 Mayer 1490 Mayerthaler 122, 622, 623, 633, 1170 Maynor 1519 Maytinskaya 670 Mazzocco 1153 Mazzucchi 1495 McCall 412 MacCarthy 19, 564 McCawley 61, 487, 488, 570, 932, 946, 955, 1280, 1281, 1319, 1328, 1336, 1369 McClelland 432 McCloskey 501, 715, 717, 718 McConvell 834 McDaniel 1524 McElree 1489 Mchombo 588, 598, 1230, 1235, 1237, 1407, 1411 McLendon 1102 McLeod 1089 McNeill 1492

van Megen 321 Mehler 1485 Meibauer 1009, 1020, 1023 Meigret 184 Meij, de 796, 1073 Meillet 175, 680, 1078, 1150, 1155—1157, 1222, 1254 Meiner 139—141, 157, 163, 168, 188, 189, 257, 304, 1548 Meisel 1492 Mel’ḉuk 304, 311, 321, 324, 351, 715, 718, 1078 Melanchthon 130 Menn 1495 Mensing 142, 1547 Menzel 1548—1551, 1557 Meringer 1490 Merlan 1090, 1098, 1112, 1113, 1381, 1382 Mervis 1092 Metcalf 1141—1143 Metzing 1594 Mey 339, 342 Meyer 1534 Meyer-Ingwersen 1552 Mganga 1408, 1409 Miceli 1495 Michael 131, 141, 155, 163 Michael der Sýnkellos 132, 133, 137 Michel 1532, 1538 Miehe 1397 Mieszek 1004 Mihaila-Cova 312 Mikkelsen 137, 154, 167, 187 Miklosich 131, 176, 185, 186, 227, 1136, 1154, 1277 Mikuš 157, 259, 263, 264, 267, 268, 270, 272 Mill 291 Miller 471, 475, 478, 485, 673, 911, 1223, 1472, 1487, 1497, 1519 Minami 1375 Mirza 1397 Mistrík 356 Mitchell 680, 1506, 1507 Mithun 343, 655, 680, 841, 842, 1062, 1098, 1100, 1205, 1237, 1238 Miyagawa 1366 Mohanan 589, 1103 Molnár 713 Moltmann 83 Monboddo 1149 Monoi 1496 Monreal-Wickert 131 Montague 61, 63, 66, 397, 411, 412, 414—420, 422, 423, 425—427, 431, 440, 941, 963, 1471, 1472, 1476 Moortgat 408, 409, 411, 413 Moravcsik 723, 1236, 1239,

1278, 1473, 1474, 1475, 1480 Mordechay 752 Morgan 899, 932, 1322, 1335 Moritz 1546 Moro 1348 Morrill 408 Mosel 1086 Moshi 599, 731, 736, 737, 739, 741 Motsch 254, 484 Mounin 131 Moutaouakil 212, 384, 385 Mrazović 312, 1277 Mubarrad 209, 211 Mufwene 1169 Mukařovský 349 Mukherjee 1277 Mulder 802 Müller 35, 38, 40, 43, 45, 48, 946, 1295—1297, 1307, 1492, 1548 Müller-Michaels 1549 Müllerová 356 Munro 1069, 1114, 1227, 1231, 1234 Murray 170 Mustanoja 1280, 1281 Muthwi 1516 Mutschmann 174, 180 Muysken 11, 12, 800, 801, 1156, 1451, 1453, 1461, 1517 Myers 1413 Myhill 833, 1068, 1186

N Nadkarni 1158 Näf 1026 Nanni 1321 Napoli 782, 950, 951 Naro 1241, 1521 Nash 1430, 1433—1436, 1440— 1442, 1445 Neale 81 Nebeský 136, 260, 317 Neidle 556 Neijt 775, 777, 786, 789 Nelimarkka 331 Nespor 49, 54 Neumann 189, 1548 Neuner 1550 Neurath 288 Newman 1000 Newmeyer 208, 470, 472, 478, 955 Newport 1492 Nichols 717, 814, 1076, 1080, 1082, 1087, 1089, 1090, 1095, 1099, 1103, 1104, 1110, 1181, 1277, 1399 Nickel 1550 Nicol 1488 Nicole 147, 163, 177 Nida 283

1617

Nieminen 138 Nikula 311, 312, 767, 1558 Nishida 411, 1230 Nishigauchi 76, 77, 535, 536, 1365 Niyekawa-Howard 1369 Nold 1552 Noonan 680, 1111, 1389 Nordmeyer 26 Noreen 134, 135, 150, 173, 178— 180, 182, 183, 257, 296 Notkers 1272 Nunberg 968 Nündel 1550 Nurse 1397 Nwáchukwu 801, 810—812 Nygaard 190

O O’Connor 998 O’Grady 398 Oates 834 Ochs 1236 Oehrle 404, 412 Oelinger 1545 Ogihara 60 |Ogwuéléka 810 Ohijenko 142, 143, 154, 167 Ohmann 1529, 1536 van Oirsouw 750, 751, 753—759, 761, 770 Ojeda 413, 575, 670 Olátúnyí 808 Olbrei 1499 Olovjannikova 1239 Olsen 743, 767, 781, 1267, 1271, 1298 Oomen-Welke 1549, 1550, 1552, 1557, 1559, 1560, 1562, 1567, 1568 Opitz 1545 Oppenrieder 1009, 1011, 1014, 1017—1022 Ostler 732, 733 Otanes 1418, 1419, 1425 Otto 175, 224 Ouhalla 1289, 1304, 1316 Owens 208, 210, 211, 1138, 1150, 1157 Oyèlárǎn 802, 815

P Paardekooper 275 Pafel 64, 872, 874, 875, 878 Pagliuca 1060 Pallas 1141, 1143 Pan 968 Panevová 331—335, 353 Pāṇini 199, 202—206, 1136 Pape 312 Parker 677, 1156

Parsons 56, 63 Partee 59, 67, 399, 410, 411, 821 Pasch 358, 931, 1006 Patañjali 100—203, 205, 206 Paul 133, 153, 160, 164, 176— 178, 186, 216, 222—225, 227, 233, 281, 1045, 1118, 1146, 1147, 1150, 1152—1154, 1156—1158, 1282, 1547 Pauliny 260, 352 Pawley 800 Payne 676, 678, 679, 915—920, 1062, 1086, 1102, 1111, 1239, 1241 Pecorella 132, 163 Peirce 896, 1478 Pelletier 967, 971, 975 Penner 1492 Percival 330, 1138, 1139 Pereira 943 Perel’muter 1248 Perkins 1060 Perl 311, 312 Perlmutter 28, 329, 528, 562, 601—607, 696, 697, 731, 735, 781, 827, 833, 909, 949, 1080, 1081, 1348, 1352 Perrott 719 Pesetsky 76, 516, 517, 531, 801, 1302, 1311 Peškovskij 130, 141, 234 Pestalozzi 1546 Peters 76, 427, 749, 877, 1141, 1601, 1604 Peterson 170 Petkevič 363 Petrov 171, 189 Petrus Helias 139, 167, 1139 van Petten 1503 Petyt 122 Pfaff 1515 Pfleiderer 1548 Philippaki-Warburton 883, 1112 Phillips 580, 800 Pianesi 1352 Pica 554 Picallo 1357 Pichon 922 Pick 1485, 1490, 1495 Pickering 1489 Pickett 286 Pierce 514 Pierrehumbert 52, 1018 Pike 286, 296, 330 Pinborg 131, 137, 174, 1139 Pinker 1485, 1493 Pinkster 346 Piou 812 Pirejko 1086 Pittaro 924 Planck 980 Plank 1046, 1076, 1098, 1103, 1152, 1158, 1186

Plátek 354 Platero 1433 Plato 91, 646, 688, 1137, 1463, 1464, 1468, 1478, 1494 Platon 132, 147, 161 Platzack 1173, 1219, 1267, 1271, 1272, 1311 Poggio 1485 Pokorny 1033, 1132 von Polenz 250, 1550 Poletto 1353 Polinskaja 1089 Pollard 289, 406, 410, 570, 573, 574, 576, 579, 580, 592, 1364, 1372 Pollock 61, 501, 519, 530, 531, 1310, 1351, 1414 Polomé 1397 Pommerin 1556 Pooley 1521 Popa 312 Popjes 1086 Poplack 122, 1515 Porphyrios 162 Porsch 1532 Poser 735 Posner 362 Post 1599 Postal 475, 477, 487, 505, 530, 537, 550, 570, 601—608, 696, 709, 731, 735, 750, 773, 779, 827, 833, 909, 944, 953, 958, 1498 Pott 1149 Poutsma 219, 220 Powell 831 Pregel 1551 Price 1452 Prideaux 340 Priestly 1240 Primus 692, 693, 700, 701, 890, 980, 981, 983, 985, 986, 1040, 1077, 1083—1096, 1088— 1090, 1092—1094, 1097, 1098, 1100—1102, 1104, 1549, 1552, 1557—1562, 1566 Prince 50, 832, 892, 893, 1003 Prinzhorn 1271, 1272 Priscian 91, 94, 134, 135, 137, 161, 163, 167, 175, 184, 189, 1137 Pritchett 1493, 1502, 1505 Protagoras 1137 Proudian 580 Pulleyblank 1177 Pullum 270, 413, 475, 498, 570, 571, 575, 580, 687, 716, 827, 842, 1034, 1040, 1050, 1089, 1090, 1129, 1176, 1228, 1364, 1600, 1601 Pulte 757 Pustejovsky 565 Pustet 1069

1618

Q Qimḥi 1138 Quadfasel 1496 Quine 928, 946, 969, 1468, 1469 Quintilian 1148 Quirk 92, 104, 889, 979, 998, 1000, 1002

R Raasch 1568 Radford 713, 1491 Radtke 1347 Rall 312, 1557 Ramat 1040, 1047, 1130, 1262 Ramos 1421 Ramsey 924, 927 Ramus 1140 Randall 744, 746, 1500 von Randow 312 Rank 1549 Ranke 1127 Ransom 837 Rappaport 897 Rask 1143, 1146 Rath 767, 1534 Ratke 1545, 1546 Rau 206 Raue 313 Rauh 207, 731 Raumer 1547 Rausch 1560 Rayner 1488, 1504 Reckendorf 1045 Reed 142, 218, 676, 678 Reesink 926 Reh 1132, 1155 Rehbein 1556 Rehling 155 Reichling 130, 184 Reid 1239 Reiman 325 Reiners 1541 Reinhart 27, 38, 518, 538, 549, 689, 715, 802, 871, 872, 881, 941, 942, 945, 958, 960, 963, 1077, 1303, 1492, 1508 Reis 690—692, 858, 859, 861, 878, 993, 1006, 1016, 1020, 1023, 1267, 1277, 1283, 1289, 1549, 1557, 1558 Reis, H. 227 Reiter 973 Renzi 1347, 1353 Reuchlin 1139, 1141 Reuland 549, 957, 958, 997 Reuschling 1549, 1550, 1560, 1566, 1568 Revzin 259, 317 Rhodes 842, 884 Richter 435, 1152, 1158 Ricken 131 Rickford 1526

Rickmeyer 312 van Riemsdijk 30, 538, 540, 558, 707, 1295, 1298—1300, 1430, 1493 Ries 91, 166, 167, 185, 227, 228, 1151 Riesbeck 1485 Riesel 1532, 1534, 1538 Rijkhoff 387 Ringbom 1535 Risselada 346, 347, 837 Ristad 564 Ritchie 1604 Rivarol 1151 Rivero 1229 Rizzi 47, 500, 518, 523, 525, 531, 539, 540—542, 544, 546, 597, 642, 1293, 1313, 1315, 1350—1354 Roberts 559, 811, 814, 819, 1112, 1304 Robertson 1452 Robins 147, 647, 652, 653, 1032, 1127, 1128, 1140—1142, 1146 Robinson 210, 327, 331 Rochemont 978, 998, 1002, 1072, 1529 Rocher 201, 207 Rodríguez 213 Roeper 514, 1492 Rogers 1469 Romaine 121, 1156, 1210, 1517, 1519 Roodbergen 206 Rooth 53, 59, 65, 67, 399, 411, 986 Rosch 1092 Rosen 28, 602, 604—606, 612, 827, 913, 1384, 1492 Rosenbaum 487, 756, 784, 935, 1529, 1542 Rosengren 305, 852, 858, 1006, 1008, 1016, 1020, 1023, 1025, 1026 Rosiello 1149 Ross 43, 136, 484, 487, 491, 499, 505, 512, 524, 545, 596, 657, 728, 750, 752, 758, 770, 772, 774, 776, 779—781, 783—786, 788, 801, 834, 934, 953—955, 960, 1312 Roth 161, 181, 182, 312, 1546, 1549 Rothkegel 313, 321 Rothstein 528 Rounds 572 Rousseau 1144 Rouveret 715, 718 Rowlands 1246 Rozwadowski 137, 154 Ruhlen 1109, 1110 Rumelhart 432 Rumsey 1090

Russell 71, 291, 1128, 1237, 1341, 1467—1469 Růžička 254, 260 Ryle 291, 1470

S Sabel 1293 Sacy, de 177 Sadock 31—33, 491, 1006, 1009, 1011, 1012, 1019, 1020, 1023, 1026, 1027, 1111, 1114 Sadzinski 312 Saeed 1068, 1069 Sætningsskema 1267 Saffran 1495 Safir 518, 991, 992, 997, 1493 Sag 64, 65, 270, 289, 336, 406, 410, 413, 564, 570, 573—576, 579, 580, 592, 716, 746, 752, 761, 774, 775, 777, 779, 785, 786, 827, 935, 1335, 1364, 1372, 1600, 1601 Saito 48, 513, 514, 522, 523, 535, 538—540, 545, 546, 711, 1365 Sajnovics 1142, 1143 Sakkaki 209, 211, 212 Saksena 1098 Salmasius 1141 Salomaa 122, 1590, 1593, 1597, 1599 Saltarelli 883, 890, 893, 1087, 1112 Salvi 1249, 1348 Samarin 667, 669 Samuel ben Jakob 163 Sanctius 144, 163, 168, 1140, 1153 Sanders 339, 750, 784, 1530, 1535 Sandig 767, 1529, 1530, 1536 Sandkühler 1031 Sandmann 170 Šanize 1381 Sankoff 121, 122, 1250, 1516, 1517, 1521—1523 Sapir 216, 282, 653, 654, 723, 1098, 1158, 1166, 1217, 1380 Sarraj 209—211 Sarup 201 Sasanuma 1496 Sasse 651, 678, 688, 691, 693, 703, 833, 883, 884, 886, 1034, 1040, 1068, 1071, 1077, 1078, 1081, 1094, 1113, 1114, 1131, 1214, 1270, 1271, 1278, 1556 Sauer 1533 Šaumjan 133, 259, 260, 261, 263, 270 Saussure, de 94, 157, 217, 233, 258, 264, 268, 270, 349, 350, 1044, 1116, 1128, 1151, 1217

1619

Sauvel 1227, 1231, 1234 Savin 312 Scaglione 131, 183, 1150—1153, 1155, 1156, 1158, 1268, 1269 Scaliger 177, 1139, 1141, 1142, 1148, 1157 Ščerba 233 Scha 81 Schabes 406 Schachter 648, 652, 653, 663, 664, 667, 679, 682, 688, 700, 752, 801, 901, 1111, 1113; 1418, 1419, 1425, 1429 Schäfer 1549 Schaner-Wolles 1492, 1496 Schank 1485, 1487 Scharfe 207 Schaub 1112 Scheerer 653 Scheidius 1141 Schendels 1532, 1534, 1538 Schenkel 306, 312, 1548, 1559 Scherer 186, 1149 Scherpenisse 637, 1047, 1267 Schiefer 765 Schieffelin 830, 885 Schiepek 1534 Schildt 1272 Schiller 800—802, 808, 817, 821 Schlegel, A. 232, 1032, 1144— 1146, 1158 Schlegel, F. 232, 1032, 1142— 1144, 1146, 1152 Schleicher 1044, 1124, 1128, 1142, 1145, 1147, 1154, 1155 Schlobinski 123 Schlözer 1141 Schmerling 1001 Schmidt 164, 179, 233, 314, 1278, 1516, 1545, 1557 Schmitter 242 Schmitthenner 164, 165 Schneider 89, 1542 Schnelle 352, 354, 431, 432 Schoell 134 Schoenthal 248, 249 Scholz 1010, 1014, 1015, 1019, 1025, 1277 Schönfeldt 306 Schönfinkel 57, 396, 402, 407, 412 Schooneveld 351 Schötensack 174, 177 Schottel(ius) 1141, 1148, 1154, 1545 Schreiber 312 Schubert 967, 971, 975 Schuchardt 1078, 1150, 1383 Schuhmacher 1559 Schultz-Lorentzen 167 Schulz 134 Schulze 1086 Schumacher 311, 312, 321 Schütze 875

Schwartz 173, 678, 971, 1495 Schwarz 1090, 1548 Schwenk 1550 Scotton 1410 Scrieckius 1141 Searle 356, 372, 388, 432, 1470 Sebba 802, 803, 808, 1454 Sechehaye 258 Sedlak 835, 838 Seefranz-Montag 1271, 1277, 1278, 1284, 1285 Segarra 1496 Seidel 167, 229 Seiler 118, 123, 621, 674, 1214 Seldin 412 Selinker 1551 Selkirk 15—18, 20, 49, 54, 1490 Sells 536, 588, 1034, 1514, 1519 Šendels 233, 234 Sereecharoensatit 800 Setälä 138 Sewell 217 Sextus Empiricus 174, 176, 178, 180 Seyfert 1549 Sgall 331, 332, 334, 335, 339, 350—356, 360—362, 364, 629, 920, 1068 Shamir 475 Sharma 1096 Shattuck 1490 Shaumyan 405, 714, 715, 717, 718, 1077 Sheintuch 835 Shepard 1169 Sherry 1155 Shibatani 696, 734, 1092, 1277, 1365 Shieber 410, 475, 580, 943, 1498 Shopen 1034, 1109—1111, 1113, 1114, 1430 Showalter 678 Shukla 206 Sibawayh 208, 210, 212 Sidarus 213 Siebenbrodt 1566 Siegel 663 Siewierska 339—341, 344—346, 837, 843, 864, 880, 884, 1114 Sigwart 176 Silman 237 Silverstein 834, 1038, 1039, 1070, 1079 Simmler 1535 Simons 671, 679 Simpson 1430, 1432, 1436, 1438, 1440—1442, 1445 Singh 206 Sintzoff 1602 Sirotinina 830 Sitta 226, 1549, 1550 Sjoblom 755 Skalička 258, 262, 349—351, 1176

Skinner 484 Slobin 1498, 1499 Smaby 951 Small 1521 Smári 183 Šmilauer 143, 176, 350 Smith 289, 291, 999, 1109, 1144, 1145, 1158, 1179, 1223, 1452, 1453 Smotrickij 156 Smullyan 400, 402, 412 Sneed 434 Snow 1492 Soames 949 Sobin 1515, 1519 Soboleva 259—261, 270 Sohn 1112 Solan 958, 1492 Solnit 800 Somers 305, 306, 314, 334 Sommer 186 Sommerfeldt 312, 1533, 1539, 1540 Ṣónaiya 804, 807 Sova 266 Sowinski 1539, 1541, 1550, 1566 Spear 1397 Speas 513, 527 Spencer 32, 55, 71 Sperber 1006, 1008, 1020, 1023 Spies 1280, 1551, 1566 Spillner 122 Spitta 1549, 1550 Spitzer 1530, 1535, 1536 Sportiche 518, 525, 531, 540, 550, 709, 712, 715, 718, 802, 945, 1303 Spranger 1282 Sridhar 697, 1112, 1277, 1516 Sroka 1567 St. Anselm 1139 St. Augustine 1137, 1139 Staal 206, 1136 Stabler 1502 Stahlke 801 Stalnaker 928, 932, 1330 Stampe 682 Stankiewicz 1137, 1141, 1148— 1150, 1155 Starke 1541 Starosta 304, 321, 327, 331— 335, 681 Stassen 1032 Staudacher 483 von Stechow 8, 13, 37, 44, 59, 61, 63, 68—70, 190, 356, 398, 411, 621, 714, 795, 796, 963, 964, 1072, 1130, 1289—1291, 1304, 1306, 1307, 1309—1311 Steedman 413, 1505 Steele 405, 411, 682, 1228, 1231, 1237

1620

Steever 1169 Steger 246 Stegmüller 434 Steiner 321 Steinitz 484, 624, 853, 1561 Steinthal 131, 148, 225, 1146, 1153, 1466, 1467, 1547 Stephens 1055, 1058 Stern 150, 165 Sternefeld 8, 13, 25, 29, 37, 40— 42, 45—48, 69, 82, 190, 863, 1130, 1289—1291, 1293, 1302, 1306, 1307, 1309 Stewart 801, 808, 819, 1457 Stich 1483 Stiehl 1134 Stiernheim 1142 Stimm 898, 899 Stipa 1141—1143 Stockwell 916, 1169, 1180, 1267—1269, 1273 Stokhof 78, 877, 951 Štolc 260 Stolt 1269 Stong-Jensen 1382 Stowell 11, 21, 70, 513, 518, 527, 528, 530, 537, 713, 718, 934, 945, 946 Strahlenberg 1141 Strassner 1549 Strawson 924, 1470 Strecker 311 Strehle 1549 Streitberg 1116 Strigin 972 Strömsdörfer 1033 Strube 1554 Stump 60, 771, 821 von Stutterheim 792 Suárez 1098, 1099 Subba Rao 200 Subramania Iyer 206 Sucharowski 1552, 1566 Sugioka 971, 974 Sulkala 1112 Suppes 122 Sütterlin 156, 1548 Suyuṭi 209 Svoboda 355 Swadesh 652, 653, 654 Swartz 1430—1433, 1435, 1436, 1438, 1440 Sweet 219, 717, 718, 1153 Sweetser 1181 Swinney 1488 Sylvain 801 Szabolcsi 63, 402, 405, 408, 410, 413, 530, 746, 890, 1072 Szamosi 1000 Szanyi 312 Szemerényi 252

T Taglicht 979, 998, 1002 Tai 750, 771 Takubo 1375 Tallerman 880, 893 Talmy 932 Tanenhaus 1507 Tappe 1300 Taraldsen 539, 643, 1352 Tarski 395, 414, 1469, 1471, 1473 Tarvainen 311, 312, 1549 Taylor 1112 Tchekhoff 1092 Tegge 1549 Teigeler 312 ten Kate 1141 Tendeloo 166, 167 Téné 213, 1138 Teng 1067 ter Meulen 997 Tesnière 99, 101, 111, 112, 137, 141, 169, 212, 258, 259, 261, 264, 265, 268, 269, 299, 301—305, 309—311, 316, 323, 324, 332, 350, 352, 618, 1269, 1548 Testelec 880, 891 Teubert 312, 1533 Thelander 1521 Thepkanjana 800 Thibault 1521, 1522 Thiersch 863, 1271, 1289, 1300, 1502 Thomas 1549 Thomas of Erfurt 1139 Thomas von Erfurt 146, 159, 168, 190 Thomason 412, 1330 Thompson 339, 342, 649—651, 654, 657, 663, 680, 708, 720, 800, 803, 804, 836, 923, 1038—1040, 1048, 1066— 1069, 1071—1074, 1111, 1114, 1121, 1153, 1157, 1176, 1220, 1221, 1393, 1409, 1556 Thornton 1492 Thráinsson 1284 Thue 1576, 1579 Thümmel 147, 848, 1267 Thurmair 1009, 1012, 1013, 1019, 1021, 1026 Thurot 139 Tieffenbach 189 Tiktin 140, 151, 177, 178 Timberlake 1156, 1233 Tobin 339, 341 Togeby 258, 275 Toman 46 Tomanetz 1151, 1153, 1269 Tomlin 339, 341, 342, 840, 842, 884, 1034, 1039, 1050, 1088—

1090, 1094, 1095, 1220 Torrego 501, 544 Torris 1500 Tottie 920 Touratier 1199 Trabalza 131, 164 Trabant 1529 Trager 289, 291, 999 Trask 1186 Traugott 923, 925, 1155, 1169, 1173, 1176, 1181, 1224, 1246, 1258 Trautz 311 Travis 43, 513 Trendelenburg 147 Tripp 1281 Trnka 349, 351, 357, 1281 Trommelen 51 Trostnikov 175 Trubetzkoy 257—259, 275, 349, 1035 Tryon 1238 Tschenkéli 1247, 1277, 1380, 1382, 1389 Tschirch 245 Tsujimura 1441, 1447 Tsunoda 1091, 1103 Tuite 1382 Tura 891, 893 Turgot 1143, 1149, 1158 Turing 1576 Turner 177, 1530 Tyler 1488

U Uhlig 101 Uhlířová 355, 364 Uhmann 888 Ullmann 1529, 1535 Ulrich 361, 689, 833, 882 Ultan 1248 Ulvestad 273, 1014 Urbina 1072 Uriagereka 539, 707 Urmson 1469, 1470 Uspenskij 261 Uszkoreit 410, 690, 694, 840, 878, 1601 |Uwaláàka 801, 810, 812, 816

V Vachek 122, 349 Vaillant 1278 Vainikka 1492 Valdman 801 van Valin 339, 651, 688, 700, 703, 717, 834, 910, 1068— 1071, 1077, 1078, 1081— 1083, 1094, 1098, 1099, 1111 Valkonen 331 Vamling 1389, 1390 Van der Auwera 928, 984

1621

Van Eijk 654 Vanderslice 1000 Vanelli 1353 VanLehn 872, 875 Varga 881, 888, 890 Varro 134, 1137, 1148 Vater 305, 312, 321, 990, 997 Vattuone 883 Vaugelas 184 Veith 122 Velten 1278 Vendler 968, 976, 1383 Vendryes 175, 1158 Vennemann 50, 53, 330, 398, 604, 618, 619, 637, 711, 717, 827, 1031, 1032, 1040, 1044, 1045, 1047, 1051—1057, 1059, 1060, 1077, 1118, 1119, 1123, 1125, 1129, 1131—1134, 1147, 1151, 1176, 1177, 1180, 1216, 1217, 1221, 1270, 1271, 1278, 1279, 1281, 1310, 1553, 1554, 1556, 1568 Vergnaud 557, 564, 715, 718 Verhaar 718 Verma 1103 Vermeer 1096 Vernaleken 144, 154 Verner 1116 Versteegh 209—211, 213, 1138 Vesper 222 Vester 346, 375 Vet 346, 833 Vico 1144, 1148, 1149, 1154, 1156 Viehweger 254 Vijay-Shankar 406 Vikner 528 Vilela 312 Vilkuna 689, 880, 881, 883, 890, 1103 Villiers, de 1492 Vincent 1220, 1222, 1521 Vinogradov 130, 141, 180, 273 Virgilius Maro 1137 Visser 1279, 1280 Vitale 757, 1411 Vlk 363 Vodička 349 Vogel 49, 54 Vogt 1381 Vondrák 1277 Voorhoeve 677 Vostokov 168 Vrbová 357 Vries, de 385 Vygotsky 1489

W Wackernagel 186, 680, 848, 1116, 1145, 1151—1154, 1156—1158, 1229, 1269

Wagner 1277, 1278, 1281, 1533 Wald 127, 1236, 1406, 1407 Wali 1096, 1097 Walker 359, 1169 Wall 427 Walter 650, 653, 657 Wandruszka 248 Wanner 1229, 1289, 1498 Warmington 1120 Washabaugh 1518 Wasow 484, 504, 505, 537, 551, 580, 711, 787, 945, 955, 957 Watanabe 535, 536 Watkins 1117, 1151, 1152, 1154, 1179, 1223, 1278 Watters 385, 1004 Webelhuth 863, 946, 1295, 1298, 1300, 1306, 1307, 1309—1311 Weber 677, 679 Wechter 213 Weerman 1173, 1267, 1272, 1273 Wegener 247, 250, 355, 1282 Wehr 833 Weigand 1556 Weigl 1495 Weil 185, 268, 270, 355, 1047, 1120, 1144 Weinberg 55, 537, 718, 942, 943, 1493, 1496, 1502 Weiner 1521 Weinreich 1147, 1168, 1170 Weir 406, 1089 Weise 1546 Weisgerber 234, 242—251, 253, 1548, 1549—1551 Weisler 580 Weiss 288, 990, 995, 997 Weissenborn 1492, 1494 Weiß 213 Wells 284, 285, 286, 290, 291, 296, 1167 Welmers 810, 816, 1247 Wendt 220 Werlich 1530 Werner 723 Wernicke 1495 Westphal 1149, 1154 Wexler 962, 963, 1315, 1492— 1494 Whalen 1277 Whistler 837 Whitaker 1495, 1496 Whiteley 910, 1397, 1408—1411 Whitney 216 Whorf 1124 Wierzbicka 649, 901, 1078, 1094 Wiese 944 Wiesemann 670 Wiesinger 127 Wijngaarden, van 1601 Wilder 1292 Wildgen 610, 614, 618 Wilhelm von Ockham 177

Wilkins 565, 690 Wilkinson 605 William of Conches 1139, 1157 Williams 16, 20, 25, 28, 30, 31, 33, 510, 514, 523, 528, 531, 536, 538, 549, 565, 695, 702, 707, 735, 737, 738, 745, 755, 762, 779, 786, 801, 803, 808, 814, 871, 877, 878, 906, 936, 939, 942, 943, 1493 Williamson 819, 821, 1452 Willim 556 Wilmans 154 Wilson 1006, 1008, 1020, 1023, 1397, 1517 Wimmer 311 Windisch 1152, 1154 Winford 1518, 1522, 1525 Wingerd 801, 818, 819 Winkler 1011, 1013—1015, 1019, 1023 Winston 801, 810 Winter 1154 Wirth 339 Witherspoon 836 Withgott 1436 Wittenburg 410 Wittgenstein 291, 1468 Wiwel 139, 141, 182, 187 Wode 1545, 1549, 1551, 1565 Wolfart 378 Wolfram 1523 Wood 404 Woods 1498 Woolford 1491 Wotjak 311 Wright 1246 Wunderlich 186, 733, 742—744, 746, 771, 781, 789, 1006, 1007, 1017, 1018, 1023, 1550, 1553, 1555, 1567 Wunderlich, H. 227 Wundt 94, 133, 141, 147, 149, 150, 154, 157, 167, 171, 172, 177, 178, 182, 185, 186, 222, 257, 258, 281, 283, 1044, 1150, 1153, 1485, 1495 Wurst 149, 150, 177, 226, 1547 Wurzel 51, 633

X Xafaji 211

Y Yang 554, 962 Yule 832

Z Zaefferer 1006—1008, 1020, 1026, 1114 Zaenen 578, 596—599, 944

1622

Zander 1549, 1550 Zander-Lüllwitz 306 Zandvoort 220 Zanuttini 1310, 1352, 1355 Zaręba 1242 Zaslavskij 136 Zeevat 410 Zemb 360, 1267 Zenon 162 Zgusta 1124 Zielonka 407

Zifonun 312 Zimmermann 63, 619—621, 848, 850—852, 854, 863, 942, 1006, 1008, 1016, 1020 Zipf 902 Žiugžda 155, 172 Ziv 835 Zöfgen 1548 Zograph 1098 Zonneveld 51 Zorilla 312

Zribi-Hertz 771 Zubaydî 1138 Zubin 831 Zubizarreta 938 Zurif 1495, 1496 Zwanziger 1214 Zwarts 399, 405, 986 Zwicky 334, 577, 716, 718, 923, 1006, 1009, 1011, 1012, 1019, 1020, 1023, 1026, 1027, 1111, 1114, 1228, 1229

1623

Sprachenregister / Language Index English und Deutsch sind nicht berücksichtigt / English and German are disregarded A Abaza 1087 Abxaz 1087, 1095 Acehnese 1099, 1100, 1103 Acooli 662 Adygh 1087 Afrikaans 917 Agaw 662, 680 Aghem 346, 1004 Agul 1086 Ainu 1040 Akkadian 1187 Albanian (cf. Albanisch) 660, 663, 666, 670, 674, 675, 683, 833, 883, 1073 Albanisch (cf. Albanian) 1033 Alemannisch 621 Aleut 1196 Algonquian 673 Altaic 918 Altakkadisch 1208 Altenglisch (cf. Altenglisch) 1268, 1278—1280 Altfranzösisch (cf. Old French) 627 Altgriechisch (cf. Ancient Greek) 723, 1207, 1208, 1211, 1260, 1264 Althebräisch 1129 Althochdeutsch 1167, 1171, 1172, 1207, 1208, 1210, 1263, 1267—1269, 1281, 1283

(cf. Old HighGerman) Altindisch 1211, 1269

(cf. Old Indian) Altisländisch 1169, 1210 Altsächsisch 1269 Altsemitisch 1033 Amahuaca 1089 Amele 1114 Amharic 673, 678, 683

(cf. Amharisch) Amharisch 1033, 1129

(cf. Amharic) Ami 1238 Anatolian 1120 Ancient Greek 666, 675, 707, 1093, 1248, 1249

(cf. Altgriechisch) Anglo-Creole 1454 Apalai 667, 668, 1089 Apurina 842, 1089

Arabic 210, 666, 669, 671, 675, 676, 683, 838, 1004, 1087, 1125, 1138, 1415, 1516

(cf. Arabisch) Arabic, classical 682, 1246 Arabic, Egyptian 679, 1196 Arabisch 19, 725, 1045, 1129

(cf. Arabian) Aramaic 660, 1138 Arecuna (Arekuna) 842, 1089 Armenian 676, 890, 1072, 1073, 1196 Aryan 1150 Asheninca 676, 678 Asurini 1089 Austronesian 673, 801 Avar 1077—1079, 1081, 1084, 1086, 1088 Avestisch 1214, 1215

B Bacairi 1089 Bairisch 73, 621, 630, 633

(cf. Bavarian) Balti 1194 Baluchi 1086 Bandjalang 1091 Bantu 673, 677, 734, 736, 742, 833, 836, 1113, 1236 Barai 817 Barasanos 1089 Bartang 1189 Baskisch 723

(cf. Basque) Basque 673, 757, 883, 890, 891, 893, 1072, 1073, 1084, 1087, 1095, 1196

(cf. Baskisch) Bats 1099, 1100, 1102, 1103 Bavarian 970

(cf. Bairisch) Bengali 52, 53, 981 Berber 671 Berbice Dutch 1452, 1453 Bhili 1098 Bimoba 838 Birmanisch 1044 Blackfoot 838 Bolante 670 Boni 677, 682, 1068 Breton 1073 Bulgarian 883,

1249

(cf. Bulgarisch) Bulgarisch 1033

(cf. Bulgarian) Burarra 1197 Burushaski 1095, 1187, 1189

C Cabo Verdiense 1458 Cahuilla 1227, 1228, 1230—1233 Cambodian 900 Canela-Kraho 1086, 1095 Cantonese 1124 Cashinahua 1095 Catalan 1073, 1357 Cayuga 652, 655—657, 669, 841, 842, 1062 Celtic 917, 1116

(cf. Keltisch) Chadic 670, 675 Chavante 1089 Chechen 675, 1095 Cherokee 755 Chibemba 1407, 1415 Chichewa 29, 1237, 1403, 1411, 1412 Chimwini 30, 1397, 1406 Chinese 513, 535, 536, 545, 556, 679, 725, 804, 809, 870, 871, 903, 962, 969, 970, 1066, 1068, 1069, 1104, 1121, 1125, 1247, 1248, 1508

(Mandarin; cf. Chine-sisch, Putonghua) Chinesisch 147, 166, 182, 188, 264, 287, 292, 294, 1036, 1044, 1045

(Mandarin; cf. Chi-nese, Putonghua) Chinook 1195 Choctaw 1070, 1099 Chol 693, 1188, 1189 Chontal 693 Chuckchee (Chukchee, Chukchi)28, 740, 917 CiBemba 676 Clallam 836 Comorian 1397 Coos 693, 842, 1084, 1087, 1090 Coptic 1123 Cora 666 Cree 346, 378 Creole 1492, 1517 Crioulo 1452, 1454

1624

Cupeño 1227, 1228, 1230—1233 Czech (cf. Tschechisch)343, 676, 916, 917, 1073, 1249, 1250

D Dagur 682 Dakota 1099, 1103, 1188 Danesfani 1190 Dänisch (cf. Danish)134, 137, 140, 150, 155, 157, 162, 168, 182, 187, 269, 272, 1267 Danish (cf. Dänisch) 1143 Dasenech 670, 1084 Dhalandji 1188, 1193 Diegueño 1213 Djingili 1095 Dogri 1098 Dullay 662, 666, 674, 683 Dutch (cf. Holländisch, Niederländisch) 378, 411, 753, 755—758, 833, 874, 980, 1061, 1062, 1080, 1299, 1300, 1303, 1454, 1498, 1506 Dutch Creole 1146, 1147 Dyirbal 346, 347, 660, 666, 688, 693, 912, 1070, 1077—1079, 1081, 1085, 1089, 1091, 1239

E East Cushitic 674 Eastern Pomo 1095, 1099, 1102 Enga 1084, 1196 Eskimo 653, 688, 693, 1079, 1084, 1087, 1146, 1147, 1190 Eskimo, Greenlandic 658 Eskimo, westgrönländisches 31 Estonian 918, 1158, 1193

F Faröisch 642 Fasu 1084, 1089 Fijian 918 Finnisch (cf. Finnish) 138, 1011, 1013, 1039 Finnish (cf. Finnisch) 411, 689, 719, 839, 881, 883, 890, 918, 981, 982, 986, 1054, 1073, 1103, 1142, 1143, 1158, 1193, 1233 Flemish 1239 Franco-Creole 1454 Französisch (cf. French) 13, 31, 147, 155, 157, 163, 175, 177, 184, 235, 263, 264, 268, 271, 621, 641, 726, 777, 991, 1047, 1173, 1211—1214, 1256, 1257, 1259, 1260, 1262, 1263, 1279, 1285, 1557 French (cf. Französisch) 303, 411, 519, 535, 539, 556, 666, 670, 672, 676, 694, 821, 833,

917, 919—922, 931, 937, 938, 974, 980, 982, 983, 1080, 1085, 1121, 1122, 1157, 1190, 1221—1223, 1240, 1247, 1249, 1299, 1311, 1352, 1354, 1359, 1457 Friesisch (cf. Frisian) 637, 1267 Frisian (cf. Friesisch) 970 Frühmittelenglisch 1279 Frühneuhochdeutsch 1261

G Ga 917 Gälisch (cf. Gaulish) 1039, 1040, 1127, 1131 Garawa 1091 Gaulish (cf. Gälisch) 1123 Gbeya 667 Ge’ez 1033 Genuese 883 Georgian (cf. Georgisch) 697, 702, 719, 890, 1102, 1191, 1247, 1377 Georgisch (cf. Georgian) 1131 Ghomala 670 Godié 678 Gokana 677 Gollango 683 Gothic (cf. Gotisch) 693, 1142 Gotisch (cf. Gothic) 1277 Grebo 838, 840 Greek (cf. Griechisch) 91, 513, 650, 660, 662, 666, 672, 673, 675, 683, 693, 833, 883, 901, 1073, 1120, 1125, 1143, 1145—1147, 1152, 1154, 1248 Griechisch (cf. Greek) 136, 146, 148, 159, 161—163, 176, 177, 186, 187, 190, 1033, 1202, 1203, 1545 Grönländisch 134, 167, 261, 262, 287 Guarani 1098, 1100, 1102, 1103, 1105 Guaymi 1102 Guyarati 1084, 1195 Gullah 1451 Gunwinggu 834 Gurung 1095 Guyanese 1452, 1458, 1459

H Haida 842, 1040, 1099 Haitian 800, 801, 803, 812—815, 818, 819, 1451, 1452, 1456— 1459 Halkomelem 836 Harare 1033 Haruai 908, 913 Hausa 670, 672, 676, 677 Hawaiian Creole 1451, 1452,

1456—1458, 1460 Haya 834, 1244, 1250, 1406, 1410 Hebräisch (cf. Hebrew) 20, 163, 184, 282, 295, 830, 1545 Hebrew (cf. Hebräisch) 676, 835, 874, 1138, 1140, 1239, 1495 Hethitisch 1201, 1202, 1206, 1207, 1208 Hianacoto 1089 Hindi 839, 1082, 1084, 1088, 1097, 1098, 1129, 1194, 1247, 1499, 1506 Hixkaryana (Hishkaryana) 693, 842, 1040, 1089 Hmong-Mien 800, 801 Holländisch (cf. Niederländisch, Dutch) 45, 991 Hua 661, 903, 925, 926, 1084, 1095 Huichol 911 Hungarian (cf. Ungarisch) 342, 658, 689, 693, 696, 705, 707, 710, 716, 719, 833, 881, 883, 885, 888, 900, 902, 933, 982, 1004, 1072—1074, 1076, 1086, 1142, 1143, 1158 Hupda 1089 Huronisch 135 Hurrian 1089

I Iacaltec 702 Ibero-Creole 1454 Icelandic (cf. Isländisch) 596, 688, 697, 698, 833, 1247, 1312 Ìgbo 800, 801, 803, 809—812, 814, 816—818, 917 Ila 838 Indonesian 659 Ingush 1084 Iranian 1187, 1195 Irisch (cf. Irish) 1033 Irish (cf. Irisch) 1196, 1245— 1247 Irokesisch (cf. Iroquoian) 1034 Iroquoian (cf. Irokesisch) 652, 655, 657, 666, 670, 673, 676, 680, 682, 1228, 1238 Ishkashim 1189 Isländisch (cf. Icelandic) 642, 1041, 1127, 1169, 1284 Italian (cf. Italienisch) 513, 539, 541, 598, 666, 672, 697, 833, 870, 885, 900, 913, 937, 938, 1073, 1080, 1122, 1190, 1236, 1249, 1295, 1296, 1299, 1305, 1312, 1313, 1347 Italienisch (cf. Italian) 40, 49, 160, 606, 607, 617, 629, 642, 1209, 1255, 1258, 1557

1625

J Ịz|n 821 Jacaltec 837, 1188, 1189 Jamaican 1457 Jamamadi 1089 Japanese (cf. Japanisch) 513, 518, 535, 536, 582, 673, 676, 681, 696, 715, 750, 756, 758, 870, 871, 874, 875, 877, 881, 909, 980, 1067, 1125, 1220, 1320, 1337, 1363, 1499, 1506 Japanisch (cf. Japanese) 76, 79, 283, 284, 740, 784, 1044, 1045, 1557 Javanese 900 Jur Luo 1086, 1090

K Kabardian (cf. Kabardisch) 1087, 1089 Kabardisch (cf. Karbadian) 1034 Kala Lagaw Ya 1082, 1088, 1091, 1250 Kalkatungu 1091 Kaluli 830 Kambata 1084 Kamchadal 1189, 1192 Kannada 697, 1103, 1517 Kanuri 680 Kapingamarangi 1192 Karelian 1193 Karen 838 Kashmiri 1098 Kâte 1084 Keltisch (cf. Celtic) 1033, 1129 Kewa 1084, 1095 Khanty 1192, 1193 Khinalug 1239 Kinande 1415 Kinyarwanda 742, 838, 1087, 1406 Korean (cf. Koreanisch) 411, 870, 872, 874, 875, 881, 917, 981 Koreanisch (cf. Korean) 13, 20, 21, 45, 74 Kuikuro 1086 Kumauni 1098 Kung 1104 Kungkari 1193 Kurdish 1086, 1192 Kwa 800, 815 Kwakwa’la 676

L Lahnda 1098 Lahu 1066 Lakhota 700 Lapp 1142, 1143, 1158 Lardil 1189—1191, 1194, 1196

Latein (cf. Latin) 26, 50, 136, 137, 141, 146, 148, 159, 163, 169, 175, 184, 261, 287, 631, 722, 724, 1011, 1044, 1207— 1209, 1212, 1213, 1256— 1260, 1264, 1545 Latin (cf. Latein) 303, 304, 342, 450, 556, 646, 650, 666, 669, 672—674, 688, 691, 706—709, 717, 828, 829, 838, 920, 980, 1119—1123, 1125, 1137, 1140, 1142, 1143, 1146, 1150, 1154, 1193, 1195, 1249, 1300 Laz 691, 697, 702, 1084, 1102— 1104 Lezgian 1086 Lhasa Tibetan 1099, 1102, 1103 Lillooet 654 Lisu 688 Litauisch 155, 170, 172 Livonian 1193 Logbara 1104 Lovari 1073 Luiseño 411, 1227, 1228, 1230— 1233 Lummi 836, 1187 Lushootseed 654 Lyélé 678

M Maasai 607 Mabuiag (cf. Kala Lagaw Ya) Macedonian 1249 Macushi (cf. Makusi) 1087 Maithili 1192 Makah 836, 837 Makusi (Makushi; cf. Macushi) 693, 842, 1196 Malagassisch (cf. Malagasy) 1040 Malagasy (cf. Malagassisch) 676, 829, 838, 884 Malaiisch (cf. Malay) 166 Malay (cf. Malaiisch) 1114 Malayalam 30, 980, 981 Malayo-Polynesian 918 Maltese 1087, 1246 Mangarayi 1090, 1113 Mansi 1192 Marathi 1084, 1098, 1491, 1517 Maung 917 Maxakalí 677 Mayan 673, 833, 917, 1196 Media Lingua 1517 Mende 1004 Miao 1104 Mingrelian 1187 Mirityabin 1238 Misumalpan languages 800 Mittelenglisch 1280 Mittelhochdeutsch 1167, 1172, 1207, 1268, 1281, 1283

Mohave 1084 Mohawk 475, 655, 656 Mon-Khmer 800, 801 Mongolisch 1042 Mordva 1086 Morisyen 1451, 1455—1458 Motu 1104 Mundani 675 Munji 1189 Mura Pirahã 670

N Nadëb 1089 Nahuatl 1132 Navaho (cf. Navajo) 1202, 1213 Navajo (cf. Navaho) 700, 836, 1099 Nepali 1194 Neyo 1250 Ngandi 842 Ngangikurrunggurr 1239 Ngizim 670 Nguni 667 Niederländisch (cf. Holländisch,Dutch) 51, 263, 264, 274, 637, 644, 1267, 1557 Niger-Congo 673 Nitinaht 836 Niuean 835, 1084 Nivkh 733 Nokuoro 1084 Nootka 652—654 Norwegian (cf. Norwegisch) 578, 694, 801, 930, 931, 1125 Norwegisch (cf. Norwegian) 641, 643, 1267 Nyawaygi 1091

O Oceanic 800 Oiampi 1089 Ojibwa (Ojibwe) 678, 842, 884 Old English (cf. Altenglisch) 1218 Old French (cf. Altfranzösisch) 1220 Old High German (cf. Althochdeutsch) 1219, 1220 Old Indian (cf. Altindisch) 672 Old Indic 1154, 1157 Old Irish 1139 Old Spanish 1229 Old-Quiché 693 Oneida 655 Onondaga 655 Ormuri 1086 Oromo 674, 1084 Oroshor 1189 Òweré 816, 817 Ozark 1196

1626

P

S

T

Pahari 1195 Palauan 676, 835 Palenquero 1452 Pama-Nyungan 1084, 1191 Pamir 1189 Panare 1086, 1089 Papiamentu 1451, 1456, 1458, 1459 Papuan 800 Pari 1084, 1086, 1090 Päri 1196 Pashto 1086 Persian (cf. Persisch) 667, 838, 1190 Persisch (cf. Persian) 1214, 1215 Philippine languages 105, 682, 688 Pitta-Pitta 1091 Pocomchi 693 Polish (cf. Polnisch) 342, 535, 556, 571, 836, 917, 921, 1004, 1073, 1242, 1294 Polnisch (cf. Polish) 138, 146, 151, 154, 160, 172, 173, 738 Portugiesisch (cf. Portuguese) 1212, 1260 Portuguese (cf. Portugiesisch) 1241, 1248, 1454 Provençal 1139 Punjabi 1084, 1095, 1098 Putonghua (cf. Chinese, Chine-sisch) 1121, 1124

Sahaptin 1087 Salishan 653—656, 666 Samoan 1984, 1086, 1187 Sanskrit 204, 556, 901, 1136, 1143, 1145—1147 Saramaccan 1451—1456, 1458— 1460 Sardinian 1248 Schottisch 1127 Schottisch-Gälisch 1132 Schwedisch (cf. Swedish) 134, 136, 144, 150, 152, 158, 169, 170, 173, 178, 183, 643, 733, 1172, 1267 Schweizerdeutsch 30 Scots English 1519 Scots Gaelic 1245—1247 Semitic 671, 800 Seneca 655 Serbo-Croat (cf. Serbokratisch) 883, 1241 Serbokratisch (cf. Serbo-Croat) 54, 1033, 1213 Seri 607 Serrano 1227, 1228, 1230—1233 Seselwa 1451, 1452, 1454—1459 Shambala 834 Sherpa 838, 1187 Shina 1194 Shona 1247, 1413 Shonghai 1104 Shugn 1189, 1190 Sindhi 1098 Sinhalese 1086 Sino-Tibetan 800 Siouan 670 Siuslaw 693, 1087, 1089 Slovak (cf. Slowakisch) 1073 Slowakisch (cf. Slovak) 260 Slowenisch 141 Somali 670, 674, 679, 682, 683, 1068, 1069 Sorbisch 32 Southern Paiute 679 Spanisch (cf. Spanish) 617, 623, 727, 1258, 1259, 1264 Spanish (cf. Spanisch) 411, 544, 561, 672, 676, 926, 927, 1100; 1193, 1248, 1398, 1515, 1517 Spätmittelhochdeutsch 1282 Squamish 836, 918, 1196 Sranan 1451, 1452, 1459 Suaheli 741 Sumerian 1095 Surselvan 898, 899 Sursurunga 671, 674, 675 Swahili 659, 666, 674, 679, 719, 724—726, 757, 907, 911, 974, 1087, 1236, 1248, 1397 Swedish (cf. Schwedisch) 801, 835, 919, 921, 944 Syriac 660

Tabasaran 917, 1086, 1192 Tabassaran 1237 Tagalog 652, 653, 655, 657, 660, 661, 688, 700, 839, 1069, 1070, 1111, 1418, 1516 Tai-Kadai 800, 801 Tamil 1337 Tangale 1004 Tasmania 682 Tatar 1086 Tauya 926 Teribe 1090 Tewa 836 Thai 1104, 1114 Tibeto-Burman 1187 Tigre 1033 Tigrinya 669 Tiwa 836 Tlingit 1099, 1100, 1103 To’abaita 679 Toba Batak 842, 1037 Tocharian 1124 Tojalabal 753 Tok Pisin 1210, 1248, 1250 Tongan 918, 1103 Tongic 1196 Tschechisch (cf. Czech) 46, 143, 150, 259, 260, 276, 277 Tucanoan 1125 Tuki 962 Tungus 918 Turkana 837, 1084 Turkic 673 Türkisch (cf. Turkish) 235, 725, 740, 1200, 1201, 1203, 1207, 1259 Turkish (cf. Türkisch) 530, 658, 664, 666, 670, 674, 675, 681—683, 719, 754, 757, 881, 883, 884, 890, 891, 902, 919, 927, 981, 1073, 1113, 1499, 1506 Turoyo 661 Tuscarora 655, 838 Twi 838 Tyaapukay 1091 Tzeltal 693

Q Quechua 31, 673, 677, 679, 757, 981, 1069, 1113, 1123, 1517 Quilente 676

R Rajasthani 1195 Romansch 1223 Rumanian (Romanian; cf. Rumänisch) 673, 689, 882—884, 1073, 1083, 1087, 1090, 1103, 1248, 1354 Rumänisch (cf. Rumanian) 140, 143, 151, 723, 1033, 1213, 1557 Rushan 1189, 1190 Russian (cf. Russisch) 556, 673, 674, 707—710, 713, 753, 757, 758, 830, 883, 891, 899, 902, 910, 911, 919—921, 1073, 1103, 1241, 1244, 1247, 1250, 1308, 1495 Russisch (cf. Russian) 45, 140, 144, 147, 155, 160, 168, 170, 171, 189, 270, 273, 607, 724, 733, 734, 784, 852, 994—996, 997, 1011

U Ubyx 1087 Udi 607, 702, 1084, 1086, 1187, 1237 Ukrainisch 142 Ungarinjin 1090 Ungarisch (cf. Hungarian) 159, 733, 734, 1557 Upper Chehalis 654 Uradhi 1095 Uralic 673, 918, 933 Urarina 1089

1627

Urdu 1098, 1517 Urubu 1089 Usan 926 Uto-Aztecan 673, 1237 Uzbek 838

Vatai 682 Vedic Sanskrit 1120 Veps 1193 Vietnamese 756, 900, 927 Vote 1193 Vulgar Latin (cf. Vulgärlatein) 1248 Vulgärlatein (cf. Vulgar Latin) 49, 1209, 1254, 1257, 1258

Walisisch (cf. Welsh) 1036 Wambon 346 Wangkumara 1104, 1188 Wappo 1084, 1187 Wargamay 1191 Waris 1194 Warlbiri (cf. Warlpiri) 411, 696, 702 Warlpiri (cf. Warlbiri) 1072, 1237, 1300, 1365, 1413 Warluwarra 1193 Warrungu 1091 Wayana 1089 Welsh (cf. Walisisch) 893, 1073, 1246, 1247 Western Desert 1188, 1193

W

Y

Waigali 1095 Wakhi 1188

Yg Dii 677 Yaghnobi 1086

V

Yalarnnga 1091 Yana 653, 679 Yao 909, 911 Yaqui 838 Yazgulyam 1189 Yemenite 672 Yidin 733 Yindiny 1085, 1091 Yorùbá 667, 799—803, 807, 808, 812, 814—816, 819, 1246, 1247 Yuculta (cf. Yukulta) 1095 Yukatekisch 726 Yukulta (cf. Yuculta) 834, 1091, 1190

Z Zapotec 756 Zulu 981 Zürichdeutsch 26, 78

1629

Sachregister / Subject Index A 1 Advancement Exclusiveness Law (1AEX) 607 A’-chain 528 A’-position 528 A-Bewegung 38 A-binding 945 A-chain 528 A-position 528 A-prefix 1099 A-quer-Bewegung 38 a-structure 598 (A)-element 962 Abduktion 1170 Abgeschlossenheit 1590 Abhihitānvaya-Theorie 200 Ableitung (cf. derivation) 1587 absolutive 1078, 1436 abstraction operator 397 abstractness 371 Abstraktion 65, 83 accent, syntactic 437, 459 Accessibility Hierarchy 1176 accomodation 989 accusative language 1076 accusative rule 1077 AcI 935, 1213, 1290 acquisition 1491 acrolect 1518 active argument 1184 active type 1098 Ad-Adverb 1564 Ad-Attribut 1564 additive particle 979 adjacency 115, 335 Adjazenzpaar 777 adjective 111, 661 adjunct (cf. Adjunkt) 527, 906, 1369 adjunct dative 1439 Adjunct Island Condition 500 Adjunkt (cf. adjunct) 8, 15 Adjunktion 621, 1603 Adnexion 173 adnominal 1369 adposition 680, 1113, 1261 adposition type 1058 adsential 1310 adverb 111, 663, 1454 adverbial 305, 1323, 1329, 1370, 1401, 1562 Adverbialsatz 853 adversity passive 1369

affect-α 514 affirmative-polarity context 982 affix 1144, 1228 affix hopping 476, 478 agent 104 agglutinating language 1144 agglutination theory 1145 agrammatism 1495 agreement (AGR) 117, 457, 530, 1235, 1379, 1382, 1431, 1435 Agreement Hierarchy 1241 agreement prefix 1099 agreement rule 1096 Akkusativierung 246 Akkusativobjekt 1558 Aktionsart 1245 Akzent, fokussierender 1017 Akzent, nichtfokussierender 1017 Akzeptabilität 471 Algebra 416 alienation 897 alignment pattern 1184 alignment typology 1076 Allomorphism Principle 1084 allophone 118 Alltagskommunikation 1534 Alphabet 1577 Ambigierungsrelation 422 Ambiguitätsbegriff 1588 analogic-semantic chance 1117 Analogie 1260 Analysebaum / analysis tree 426 analytic philosophy 1467 Anapher (cf. anaphor) 39, 1203 anaphor (cf. Anapher) 521, 959, 1335, 1440, 1457 anaphor, bound 940 anaphoric domain 959 anaphoric reference 940 Ancestral Active Hypothesis 1187, 1193 Ancestral Ergative Hypothesis 1187 Angabe 1560 antecedent-government 540 Antezedens 40 Antezedens, striktes 41 Antipassiv (cf. antipassive) 28 antipassive (cf. Antipassiv) 1090 antitopic 1323 Anvitābhidhāna-Theorie 200 Anweisen 1532 aorist 1249 aphasia, Broca’s 1494

aphasia, Wernicke’s 1494 apodosis 923 applicative 1402 Applikativierung 29 apposition 107 Äquivalenzproblem 1598 arbitrary constituent 456 Arc Pair Grammar 608 Argument 731, 905, 1056 Argumentieren 1531 argument position 941 argument structure 515, 905 Argument, implizites 58 argument, syntactic 908 Aristotelian subject 1069 Artikel 1260 aspect 1244, 1383 aspectual marker 1375 Atomizitätsthese 20 attention flow 834 Attraktor 613 Attribut 1562, 1563 Attribution 66, 651, 1202 Attributsatz 1201 Aufstieg / ascension 604 Augmented Transition Network (ATN) 1498 Ausdruck, bedeutungsvoller / meaningful expression 417 Ausdruck, wohlgeformter / proper expression 417 Ausklammerung 1271 Auslautverhärtung, deutsche 50 Aussagesatz 1019 Aussagesatzplan 847 Äußerung (cf. utterance) 1553 Automat 1579, 1594 Automat, endlicher 1594 Automat, linear beschränkter 1594 autonomy of syntax 1473 autosemanticum 652 Autosemantikum 1252 Auxiliarkomplex 475 auxiliary 1349, 1431, 1435 auxiliary verb 1328

B B-prefix 1099 Bach-Peters-Paradox 80, 946 Bach-Peters-sentence 1336 background 998 Bare-Remnant-Ellipsis 65

1630

barrier (cf. Barriere) 540 Barriere (cf. barrier) 41, 616 basic order 1061, 1089, 1094 basic pattern 1051, 1055 basilect 1518 Bathysmatik 237 Baum 1587 Baummetaphorik 140 be-Präfigierung 743 Bedeutung (cf. meaning) 289 Behaghel’s Law 1155 behavior 1036 Behördensprache 1533 Beiordnung 152, 170 (B)-element 962 Belletristik (Stil) 1534 Benefaktiv 734 Beobachtungsadäquatheit 1 Beschreiben 1531 Beschreibungsadäquatheit 1 Bestimmende 1151 Bestimmte 1151 Betrachtung, gestaltbezogene 243 Betrachtung, inhaltbezogene 244 Betrachtung, leistungbezogene 244 Betrachtung, wirkungbezogene 244 Bewertungsprozedur / evaluation procedure 470 Bezugsnomen 1201 bilateral correlation 1052 Binaritätsdogma 286 binding (cf. Bindung) 518 Binding Inheritance Principle 578 Binding Theory (cf. Bindungstheorie) 547 Bindung (cf. binding) 38 Bindungsprinzipien 38 Bindungstheorie (cf. Binding Theory) 37 bivalent VP 802 bleeding order 1168 Bogen / arc 603 borrowing 1117 bounding convention 596 bounding node 500 bridge verb 1295 bunch 325

C c-command (cf. C-Kommando) 518, 540, 701, 715, 871, 941, 958 c-command domain 548 c-command, maximal 1097 C-Kommando (cf. c-command) 27, 38 c-selection (cf. C-Selektion) 516 C-Selektion (cf. c-selection) 23 case (cf. Kasus) 517 case assigner 707

case filter 557 case grammar 490 case hierarchy (cf. Kasushierarchie) 692, 1083 case marker, ergative 1187 case marking 706, 1379, 1421 case position 523 case relation 691 case system, ergative 1431 case theory (cf. Kasustheorie) 556 categorial grammar (cf. Kategorialgrammatik) 330, 395 categorial grammar, combina- tory 399 categorization 298 category (cf. Kategorie) 108, 395, 646 category function 437, 459 category, complex 571 category, empty (cf. Kategorie, leere) 518, 944, 1455 category, governing 552 category, static 301 category, syntactic 647 causality 930 causative (cf. Kausative) 933, 1185, 1384, 1402, 1423 causativization 1366 cause-consequence relationship 924 causee 740 causer 740 centre-embedding 94 chain (cf. Kette) 522 Chain Condition 524, 562 chain of correlation 1055, 1058 Chaos, deterministisches 1133 Chaostheorie 1133 Chômeur 602, 736 Chômeur Advancement Ban 607 Chomsky-Normalform 1586 class concept 719 class system 659 classification (cf. Klassifizierung) 441 classification of phrases 113 classification of sentences 109 classification of words 110 clause 96 clause, nonrestrictive 1345 cleft construction (cf. Spaltsatz) 1326, 1404 clitic 1226, 1425, 1431, 1436 clitic climbing 1354 clitic left dislocation 1359 closed class 652 closure 1504 cluster concept 719 co-constituents 983 code-mixing 1515 cognitive principle 886

cognitivism 431 Coherence Condition 588 Coherence Condition, extended 595 coindexation 548, 942, 959 collective predicate 968 colloquial speech 829 combinator 399 command 536, 955 comment 104, 832, 1065 communicative dynamism 343, 355 COMP 1271 comparative stripping 1341 comparison 1340 competence 507, 1156 complement 305, 497, 527, 1326, 1364, 1389 complement order 464 complement supplanting 462 complement suppression 462 complementation 103 complementizer 528, 1322, 1327, 1390 complete functional complex 552 Completeness Condition 588 Complex Noun Phrase Con- straint 499 complex pattern 1061 composition 351, 400, 1144 comprehension 1486 computational system 510, 517 concept (cf. Konzept) 446 Condition A 549 Condition B 548 Condition C 549 conditional 923 configuration 316 configurationality 1071, 1300 conjunction 679, 930, 1334, 1406 conjunction reduction 1332 connexion 299 consistent language type 1056 constituency analysis 96 constituency hierarchy 95 constituent analysis 451 constituent category, basic 444 constituent category, derived 444 constituent function 437 constituent negation 804 constituent order 387 constituent structure (cf. Konstituentenstruktur) 329, 453, 582 constraint 496 construction (cf. Konstruktion) 114 construction marker 116 construction type 1176 constructional rule of functional control (CRFC) 591 contactive form 1403

1631

content word 652 contentive typology 1076 Context Theory 1505 Continuity Hypothesis 1494 contradiction 869 contrastive focus 890 control (cf. Kontrolle) 1327 control construction 800, 935 control infinitive 1290 control sentence 575 control, anaphoric 589 converse 307 conversion 650 coordinate structure 336 coordinating conjunction 450 coordination (cf. Koordination) 100, 399, 679, 748, 869, 934, 980 copula 107, 1399 coreference 549 correlative 1391 corrigibility 92 covert movement 535 CP-IP-System 26 Cross-Category Harmony 1054, 1217 cross-hierarchy correlation 703 cross-reference 722 crossover 944 crossover, strong 39, 505, 945 crossover, weak 945, 1303, 1446 CV-Skelett 19

D D-genericity 969 D-structure (cf. D-Struktur, Tiefenstruktur) 510, 526 D-Struktur (cf. D-structure) 22 Darstellungsart 1530 dative 1437 dative shift 734, 742, 1453 Dativobjekt 1559 Dativterm 1560 decoding perspective 1113 decomposition 698 definiteness effect / Definitheitseffekt 991 Definitheit 988 Degrammatikalisierung 1256 Deklarativsatz 855, 862 deletion (cf. Tilgung) 750, 784 deletion constraint 784 deletion rule (cf. Tilgungsregel) 754 deletion target site 756 demonstrative pronoun 675 demonstrative system 676 Demorphologisierung 1256 dependency 101, 316, 350, 706 dependency graph 320 dependency relation (cf. Dependenzrelation) 715

dependency syntax 298, 316, 352 dependency theory 329 dependency, unbounded 335 dependent 334, 1055 Dependenzgrammatik (cf. grammar, dependency) 1568 Dependenzsyntax (cf. dependency syntax) 298 Dependenzrelation (cf. dependency relation) 1548 Derivation (cf. Ableitung) 15, 319 Derivational Theory of Comple- xity (DTC) 1497 descriptive adequacy 509 Desemantisierung 1257 Desententialisierung 1201 designation 1465 Deskriptivismus 233 Determinans (Determinierende) 273, 1152 Determinant 990 Determinatum (Determinierte) 273, 1152 determiner 1321, 1355, 1452 Determinierte 1152 diagram 897 dialect 1518 diathese 307, 730, 1540 Diptychon, korrelatives 1206 direct case 1097 direct passive 1369 directionality 396 Directionality Constraint 750 discourse reference 869 discourse-configurationality 1073 disjoint reference 503, 550 Diskordanz 727 dislocation 872, 884, 892 dissimilation principle (cf. Dissimilationsprinzip) 1056 Dissimilationsprinzip (cf. dissimilation principle) 638 Distinktheit 41 Distribution 277, 283 distributivity 875 divergent reanalysis 1156 dominance (cf. Dominanz) 834 dominance hierarchy 830 Dominanz (cf. dominance) 14, 1130 Dominanz, appliktive 260 Dominanz, kommunikative 260 Dominanz, unmittelbare 15 domination 517 Doppeldependenz 158 Doppelkategorisierung 850 double object construction 835 double subject construction 1067 Doubling Acquisition Hypothe- sis 1178 DP-split 1298

Dualität 174 Dualität, doppelte 263, 288 Dualität, einfache 288 Dualitäts-Hypothese 295 dummy-Symbol 481 Dynamisierung 1130

E E-language 508 E-Rekonstruktion 787 E-Sprache 7 E-type analysis 951 each-movement 962 early immediate constituent (EIC) 886 echo 1344 Echo-w-Frage 859 ECP (cf. Empty Category Principle) Einordnung 152 Einsetzung (substitution) 1592 Element-Problem 1598 Ellipse (cf. ellipsis) 65, 177, 763 Ellipse, kontextkontrollierte 766 Ellipsenarten 767 Ellipseregel 768 ellipsis (cf. Ellipse) 92 embedded predication 390 Emporstufung / advancement 604 Empty Category Principle (ECP) 41, 711, 1296, 1302 empty word 301 encoding perspective 1114 Entdeckungsprozedur / disco- very procedure 470 Entscheidungsfragesatz 855, 857 Entsprechungsparadigma 240 Entsprechungssatz 1599 Equi-NP 776 equi-NP-deletion 1327 Erase-Relation 608 Ergänzung 1560 Ergänzungsfragesatz 853, 855, 859 Ergänzungsverfahren 769, 788 ergative 1239, 1379, 1436 ergative alignment 1192 ergative language (cf. Ergativsprache) 1076 ergative rule 1078 Ergativität (cf. ergativity) 1046 ergativity (cf. Ergativität) 1184, 1305, 1348 Ergativsprache (cf. ergative language) 632 Erklärungsadäquatheit 1 ERP (event related brain potenti- als) 1503 Ersatzinfinitiv 1289 Erzählen 1530 Eselssatz 79

1632

etymology 1141, 1149 Eurotra 331 Evaluation Grammar / ableitungsbewertende Grammatik 122 Eventum 1206 exceptional case marking (ECM) 557 exclamative 1345 Exklamativsatz 1026 Exklusion 41 Experiencer-Konstruktion 1277 explanatory adequacy 509 expletive 523, 533 explicit quantification 975 Exposition 1206 expression class 648 expression rules 385 Extended Standard Theory (EST) 492, 1171 external agreement 1240 external motivation 1123 extraction (cf. Extraktion) 404, 1358 Extraktion (cf. extraction) 643 extraposition 872, 1324

F factitive 1082 familiarity 831 familiarity hierarchy 830 FDA (finite deterministic automaton) 1594 feature (cf. Merkmal) 333 feature assignment 585 feeding order 1168 Fehlerkorrektur 217 Feld 272, 276 Feld, grammatisches 233 Feld, topologisches 846, 855, 865 Feldbegriff 237 Final 1 Law 605 finality 930 Finite State Grammar 473 Finitum 3, 9 Flexion 15, 1144 flexional language 1145 Flexionsmorphem 618 Flexionsverfall 1280 Flexivphrasen 20 FNA (finite non-deterministic automaton) 1594 focus (cf. Fokus) 343, 350, 383, 493, 595, 832, 880, 909, 978, 998, 1072, 1419 focus particle 978 focus position 889 focus prominence 1072 focus theory 887 focus, secundary 1000 focus, unmarked 999

focus, wide 888 focus-configurationality 1073 focus-prominence 1069 focusing device 1002 focusing element 1324 Fokus (cf. focus) 791 Fokusausdruck 795 Fokuskonstruktion 1211 Foot Feature Principle 578 Form 290 form class 648 form word 110 formal hierarchy 830 formal language 1469 formal semantics 1471 formal universal 496 Formalsubjekt 1282 Formel 415 Formmerkmal 1010 Formtyp 1007, 1019 Fragesatz 1020 frame 353 frequency 1036 full interpretation 514 full word 110, 301, 652 function (cf. Funktion) 108, 119 function word 652 function, dynamic 301 function, pragmatic 371 function, semantic 371 function, syntactic 371, 437, 458, 708 function-argument biuniqueness 583 functional anaphor 590 functional control 589 functional generative description (FGD) 353 Functional Grammar (FG) 339, 368, 1514 Functional Paradigm 369 functional schema 584 Functional Sentence Perspective (FSP) 339 functional structure 493, 582 functionalism 339, 350 functor 1056 Funktion (cf. function) 164, 232 Funktionale Grammatik, gehaltbasiert 233 Funktionale Grammatik, gestaltbasiert 234 Funktionalstil 1532 Funktionskomposition 737 Funktionstyp 1007 Funktionstypologie 1047 Fuß 50 Futur (cf. future) 1258 future (cf. Futur) 1244 fuzzy grammar 491 Fuzzy-Logik 1133

G Gapping 750, 756, 772, 777, 783, 1333 Gardenpath Theory 1505 Geach Rule 407 Gedächtnisbeschränkung 611 GEN operator 972 Generalisierung 1133 Generalized Hierarchy Condi- tion 1088 Generalized Hierarchy Rule 1085 Generalized Phrase Structure Grammar (GSPG) 96, 570, 1499 generative semantics 487 generic anaphor 976 generic NP 967 generic tense 967 genericity 966 Genitivobjekt 1560 Gitter, metrisches 50 givenness 832 Glied 272 Gliederung des Satzes 133 global rule 489 Glossem 134 glottonic interpretation 1144 governee 712 government 117, 330, 448, 518, 540, 705 government-binding theory (GB) 515, 1501 government configuration 714 government paradigm 713 government structure 712 governor 712 GPSG (cf. Generalized Phrase Structure Grammar) grammar (cf. Grammatik) grammar, Arabic 208 grammar, comparative 1143, 1145 grammar, constitutive 298 grammar, context free (cf. Grammatik, kontextfreie) 397 grammar, daughter-dependency 325 grammar, dependency (cf. Dependenzgrammatik) 210 grammar, dependential 298 grammar, generative (cf. Grammatik, generative) 93 grammar, Hebrew 213 grammar, hellenistic 1466 grammar, indexed 406 grammatical function 438, 460, 587, 872 grammatical relation 334, 827 grammaticality 93, 491, 1464, 1479 grammaticalization 1155

1633

Grammatik (cf. grammar) Grammatik, 2-Schichten- 1601 Grammatik, disziplinbezogen-definitorische 227 Grammatik, emergente 1133 Grammatik, funktionale 232 Grammatik, generative (cf. grammar, generative) 469 Grammatik, historisch deskripti- ver Ansatz 221 Grammatik, historische 218 Grammatik, inhaltbezogene 242 Grammatik, kontextfreie (cf. grammar, context free) 414 Grammatik, pädagogische 1549 Grammatik, praktische (Schul-) 219 Grammatik, psychologisch-erklärender Ansatz 222 Grammatik, satzbezogen-logi- scher Ansatz 225 Grammatik, satzbezogene 218 Grammatik, traditionelle 216 Grammatik, wissenschaftliche 219 Grammatik, wortbezogene 217 Grammatikalisierung 1132, 1254 Grammatikalität 471 Grammatikschreibung 217 Grammatikschreibung, deutsche 220 Grammatikschreibung, englische 219 Grammatiktradition, indische 199 Grammatikvereinfachung 1169 Grammatische Funktion (GF) 56 Grammatizität 1253 Greek tradition 213 grouping relation 826 growth principle 1154 Grundausdruck / basic expres- sion 418

H habitual 974, 1245 Handlungsfaktor des Agens 202 Handlungsfaktor des Ausgangspunktes 204 Handlungsfaktor des Empfän- gers 204 Handlungsfaktor des Instru- ments 204 Handlungsfaktor des Objekts 203 Handlungsfaktor des Ortes 204 Handlungstyp 1008 hanging topic 1294 Harmonie 1130 Häufigkeit 1130 Hauptsatz 171 head (cf. Kopf) 98, 333, 497, 1055 head constraint 774

Head-driven Phrase Structure Grammar (HPSG) 579 Head Future Convention 572 Head Future Principle 572 Head Movement Constraint (HMC) 525 head parameter 518 head-final language 1363 head-government 540 head-movement 524 headedness, asymmetric 803, 809 Heavy Constituent Shift 1321 Hebrew hypothesis 1138 Herabstufung / demotion 604 Heterogenität, sprachliche 1170 Hierarchie (cf. hierarchy) Hierarchie von Komplexen 149 Hierarchie, grammatische 1037 Hierarchie, thematische 55 hierarchy (cf. Hierarchy) 1117 hierarchy constraint 688 hierarchy universal 701 Hilfssymbol 473 Hineinquantifizieren / quanti- fying in 61 Homogenitätsannahme 1170 Homomorphismus 1592 homonymy 1138 HPSG (cf. Head-driven Phrase Structure Grammar) human mind 1481 Hyperregel 1601 hypotaxis 101, 1152

I I-genericity 971 I-language 507 I-Sprache 7 i-within-i filter 950, 961 icon 896 iconicity (cf. Ikonizität) 896 iconicity principle 650 ideal type 1051 ideophone 667 idiolect 432 if-clause 924 if-cleft 892 Ikonizität (cf. iconicity) 1041 illucationary act 1341 illucationary conversion 388 illucationary force 388 illucationary operator 388 ilm al-ma’aaniy 212 image 896 immediate constituent (IC) 95, 886 immediate dominance (ID) rule 572 Imperativ (cf. imperative) 1014 imperative (cf. Imperativ) 1185,

1342, 1428 Imperativsatz 857, 1023 imperfect 1249 imperfective 1244 Impersonalie 1277 implicational universal 1178 inactive argument 1184 inchoative 1445 inconsequential 926 indefinite pronoun 678, 1322 Indefinitheit 988 indexation 940 individual psychology 1153 individual-level predicate 967 Infinitiv, nominalisierter 627 infinitive 1289, 1393, 1441 infinitive, coherent 1292 infinitive, incoherent 1292 Infinitumphrase 26 INFL 615, 814, 1412 inflection phrase (IP) 26 information structure 342 Inhaltsfeld 1267 inherence 872 inherent case 558 Inklusion 41 Inkorporation 27 Inkorporation (Kopfbewegung) 23 Inkorporationstheorie 20 Inkorporierung 737 innateness hypothesis 1483 Innovation 1168 instrumentalization 584 integrational grammar 433 Integrational Linguistics (IL) 430 Integrational Syntax (IS) 430 intentionalist mentalism 432 interjection 683 intermediate dominance princi- ple 785 internal motivation 1123 internal reconstruction 1117 interpretive semantics 492 Interrogativator 75 interrogative 1343, 1376 interrogative pronoun 678, 1322 Interrogativität 74 Intonation 52, 115, 1122 intonation focus 998 intonation structure 457 intonation unit 829 Intonationsphrase 53 intransitive realignment 1186, 1192 Intransitivierung 733 intuition 1480 inversion 1320, 1380, 1409 Inversionsverb 1557 island constraint 405, 499, 774 island effect 1296

1634

island phenomenon 579 isolating language 1144 Isomorphie 266 item, functional 528 item, lexical 527

J junction 309 juxtaposition 107

K Kāraka 202 Karmamīmāṃsā (Ritualwissenschaft) 199 Kasus (cf. case) Kasushierarchie (cf. case hierarchy) 623 Kasusmorphem 204 Kasusrektion 624 Kasustheorie (cf. case theory) 34 kategoriale Füllung 1011 Kategorialgrammatik (cf. categorial grammar) 18, 1550 Kategorialsyntax 1566 Kategorialtypologie 1046 Kategorie (cf. category) 166, 421 Kategorie, bindende 38 Kategorie, funktionale 22 Kategorie, kommunikativ-grammatische 234 Kategorie, leere (cf. category, empty) 39 Kategorie, logisch-grammatische 234 Kategorie, strukturelle 234 Kategorien, lexikalische 16 Kategorienindex 421 Kategorienmerkmal 641 Katz-Postal hypothesis 487 Kausativ (cf. causative) 734, 740 Kausativierung 30 Kerngrammatik 635 Kernsatz / kernel sentence 479 Kette (cf. chain) 415 kind predicate 967 Klammerkonstruktion 1266 Klassifizierung (cf. classification) 1031 Kohäsion 73 Koindizierung 38 kommand (k-command) 955 Komparativ 71 Komparativellipse 782 Kompetenz, syntaktische 6 Komplementsatz 1213 Komplex 145 Komposition 17 Komposition, synthetische 71 Kongruent 722 Kongruenz / agreement 722 Kongruenz, externe 725

Kongruenz, grammatische 727 Kongruenz, interne 725 Kongruenz, semantische 727 Kongruenzkategorie 722 Konkordanz / concord 722 Konnektivität 1129 Konnexion 135, 260 Konsonantenmelodie 19 Konstituente, prosodische 49 Konstituente, unmittelbare 285, 296 Konstituentenstruktur (cf. constituent structure) 284 Konstruktion (cf. construction) 287 Konstruktionstyp, markiert 641 Kontakt- und Klammerprinzip 269 Kontextabhängigkeit, globale 766 Kontextabhängigkeit, struktu- relle 766 Kontextinformation 765 Kontrolle (cf. control) 69 Kontrolleur 722 Konzept (cf. concept) Konzept, funktionales 989 Konzept, relationales 989 Koordination (cf. coordination) 66, 168 Koordinationsreduktion 770, 783 Kopf (cf. head) Kopf (einer Phrase) 8 Kopf (einer Projektion) 13 Kopf, morphologischer 17 Kopf-Bewegungs-Beschränkung 43 Kopfbegriff, relativierter 16 Kopfbewegung 38 Kopfprinzip 8 Kopula 167 Korrelation 1045 Kreolisierung 636 Kuspe 614

L Lambda-Operator 58 λ-Tiefenstruktur 267 Lambek calculus 407 language acquisition 1156, 1491 language acquisition device (LAD) 1170 language contact 1158 language faculty 506 language family 1141 language index 1110 language of logic 1468 language processing 1473, 1485 language type 1176 language variability 431 langue 94 lassen-passive 1291 Lautgesetz 1128

learnability 1472 Leere-Menge-Problem 1598 Leerstellenbildung 1203 left dislocation (cf. Linksextraposition) 884, 1220, 1293, 1359 left-right asymmetry 885 lektón 133, 174 level of complexity 1472 lexical category 526 lexical decomposition 488 lexical entry 321 Lexical-Functional Grammar (LFG) 581, 1498 lexical mapping 597 lexical rule for instrumentaliza- tion 592 lexical rule of functional control (LRFC) 589 lexical subcategory 709 lexicalism 333, 395 lexicase theory 327, 331 lexicon 510, 515, 583 Lexikalisierung 1263 Lexikon 23, 622 Lexikoneintrag 731 LF-component 533 LF-scope 1508 LFG (cf. Lexical Functional Grammar) linear precedence rule 572 linear relation 99 Linearität (cf. linearity) 268 linearity (cf. Linearität) 517, 715 linearization hierarchy 830 linguistic prototype 651 linguistic universal 1050 linguistic variation, component approach 120 linguistic variation, grammar approach 121 linguistic variation, holistic approach 120 linguistic variation, language approach 121 linguistic variation, unified theory 124 linguistic variation, variety approach 120 linguistics, comparative 1142 Linguistik, areale 1033 Linguistik, diachrone 1128 Linguistik, genealogische 1033 Linguistik, synchrone 1128 linking 732 Linksableitung 1588 Linksadjunktion 864 Linksextraposition (cf. left dislocation) 865 Linksversetzung 1206 LIPOC (Language Independent Preferred Order of Consti- tuents) 838

1635

local domain 551 locality 571, 715 logical form (cf. LF, Logische Form) 510 Logische Form (cf. LF, logical form) 47, 55 Logische Form, transparente 55, 83 logophoric Function 677 logos 132, 161 long distance binding 1315

M m-command (cf. M-Kommando) 530 M-Kommando (cf. m-command) 41 M-Selektion 23 macrorole 1420 major constituent constraint 774, 784 mapping principle 599 Marcus Parser 1502 marked focus 999 markedness (cf. Markiertheit) 902 Markiertheit (cf. markedness) 629, 635, 1035 Markiertheit, externe 636 Markiertheit, interne 635 Markiertheitsskala 622 marking function 437, 459 marking strukture 454 masdar 1393 mass term 967 material implication 923 Maturation Hypothesis 1494 meaning (cf. Bedeutung) 290 Mehrdeutigkeit, semantische 5 Mehrdeutigkeit, strukturelle 5 Mehrebenenansatz 602 Mehrfachverknüpfung 157 mentalese 1489 Mentalismus 7 Merkmal (cf. feature) Merkmal [+/- proj(ection)] 12 Merkmal [+/- max(imal)] 11 Merkmal, grammatisch-semantisches 1251 Merkmal, grammatisches 1251 Merkmal, intonatorisches 1016 Merkmal, lexikalisch-semanti- sches 1251 Merkmal, morphologisches 1013 Merkmal, semantisches 1251 Merkmalsanalyse 639 Merkmalsebene 1010 Merkmalsvererbung, morphologische 16 mesolect 1518 metalanguage 496

Metaregel 1600, 1601 Meussen’s Rule 1413 middle construction 1314, 1409 minimal attachment 1504 Minimaleinheit 134, 151, 267 Minimalitätsbarriere 43 Mittelfeld 3, 9, 848, 992, 1267, 1289 MNLU 391 modal marker 1375 modal strukture 493 Modalität 60 Modalpartikel 1009 Modalverb 1171 model of word-combination 93 modification 102 modifier 1321 Modistic approach 1139 Modularität (cf. modularity) 34, 242, 610 modularity (cf. Modularität) 1482 module 341 Moduspartikel 1009 Monem 134 monostratality 571 Montague-Grammatik 413 Morphem (cf. morpheme) 134, 291 morpheme (cf. Morphem) 95 morpheme, argumental 1435 morphological case 707 morphological negation 917 Morphologie (cf. morphology) 15 Morphologie, autosegmentale 19 Morphologisierung 1254 morphology, category-changing 650 morphology, category- establishing 650 morphosyntactic case 1083 Morphosyntax 244 Most Recent Filler Strategy (MRFS) 1500 Motherese Hypothesis 1492 Motivated Chômage Law 606 move-α- 398, 522, 863 movement 498 movement, unbounded 499 multi-event 808 multi-factor agreement 1098 Mutation 1168

N N-Determinationsstruktur 137, 259, 282 N-Ellipse 781 N-Inkorporation 27 N-movement 1357 N-V-Ring 141

N-V-Ringstruktur 261 Nachfeld 4, 847, 1267, 1289 name 942 narrow focus 888 national psychology 1153 natural kind 971 Natürlichkeitsrelation 630 Natürlichkeitstheorie (NT) 610 ne-extraction 1349, 1351 Nebensatz 171 NEG-raising 921, 1310 Negation 73, 914, 1260, 1309, 1330, 1428, 1454, 1565 negative cohesion 1311 negative morpheme 918 negative particle 917 Negative Polarity Item (NPI) 73, 1331 negative raising 1332 negative verb 918 negative-polarity context 982 Netzwerk, relationales / relatio- nal network 603 neurolinguistics 1495 New Topic 346 Nexion 135 Nexusfeld 1267 Nirukta (Etymologie) 199 no negative evidence problem 1493 no-ambiguity constraint 775, 784 Nomen (cf. noun) 137 nominal prefix 1398 Nominalgruppe 988 Nominalisierung 1200 nominalization 935 nominative 1076 non-configurational language 1430 non-configurationality 1072 Non-local Feature Principle nonreduction strategy 1387 Norm, stilistische 1528 noun (cf. Nomen) 111, 648, 653, 1433 noun class system 1397 noun phrase (NP) 1321, 1426 nouniness 657 NP-movement 1304, 1459 NP-relative 1446 NP-splitting 1297 nuclear tone 978 nucleus (cf. Nukleus) 106 Nukleus (cf. nucleus) 53, 1201 Nukleus, interner 1202 Nukleusbildung 1202 null Infl 818 Null-Complement Anaphor 778 null-subject 1352 number agreement 1382 numeral 682, 1321 Nyāya (Logik) 199

1636

O Oberflächenanapher / surface anaphor 779 Oberflächenstruktur 267, 482

(S-Struktur;cf. S-structure) object (cf. Objekt) 1380 object agreement 1406 object assignment 381 object drop 1313 object patterner 1052 object prefix 1399 object selection 1410 Objekt (cf. object) 1558 Oblique Law 606 occurrence 710 of-insertion 558 Ökonomie-Prinzip 1041 only-float 1339 ónoma 161, 689, 1464 Onset 49 Opacity Condition 502 open class 652 Operation, polynomische 419 operator 867 ordinary language philosophy 1469 Ordnung, natürliche 184 origin of language 1137 overt movement 535

P P-Marker (cf. phrase-marker) 473 P-Reduktion 789 Pada-vāda (Wortlehre) 200 paradigm 445 Paradigmatik 236 Paradigmatik, syntaktische 238 paragrammatism 1495 parallelism condition (PARR) 564 parameter 123, 506, 513, 1172, 1273, 1501 parameter, typological 657 Parameterwert 635 parasitic gap 538, 643 parataxis 101, 108, 1152 Parenthesis 383 parenthetical 1324, 1346 parole 94 parser 509, 1497 Parsing 217 parsing principle 1503 part of speech (cf. Wortart) 646, 1148, 1319, 1466 participle 111, 1388 particle (cf. Partikel) 652 Partikel (cf. particle) 1252 Passiv (cf. passive) 29, 248, 601, 735, 738, 1045 passive (cf. Passiv) 462, 558,

576, 695, 1090, 1325, 1383, 1402 passivization 1304, 1367 Passivmorphem 739 past tense 1244 past particle 1349 Patial Deletion Constraint (PDC) 761 Peanoaxiome 4 perfect 1245, 1248 perfective 1244, 1247 performance 507, 1156 performance principle 886 Performanz 6 Peripherie 635 Permutation 1603 personal hierarchy 834 personal prefix 1400 personal pronoun 669, 1187 PF-component 533 Φ-feature 518 philosophy 1463 phonetic change 1117 phonetic erosion 1221 phonetic form (PF) 510 phoné 132 phoric function 459, 677 Phrase 8, 97, 332, 354 phrase structure (PS) 570 phrase structure grammar 570

(PSG;cf. Phrasenstrukturgram-matik) phrase structure grammar, mo- dern 571 phrase structur grammar, classi- cal 570 phrase structure rule 511, 751

(PS-rule;cf. PS-Regel) Phrase, phonologische 51, 54 phrase-marker (cf. P-Marker) 512 Phrasenstruktur 7, 149 Phrasenstrukturgrammatik (cf. phrase structure grammar) 5, 473, 1167 Physikalisierung 286 Physikalismus 291 Pied-Piping 76, 524, 1292 pied-piping, infinite 1299 Platzbedarfssatz 1593 Plural 81, 1262 polarity-sensitivity 920 Polygrammatikalisierung 1258 Polymorphismus 60 Polysemie 1257 polysynthesis 1146 PoNMH (Postpositional Noun Modifier Hierarchy) 839 Port Royal Grammar 1140 Port Royal Grammatiker 147, 163, 177, 190 Position 272

possessive 1356 possessive pronoun 673 Possessivization Principle 1358 possessor raising (cf. Possessoranhebung) 1410 Possessoranhebung (cf. possessor raising) 28 postnominal gap strategy 1386 postposition 1453 Prädikat (cf. predicate) 137, 154, 165, 1552 Prädikatsrelation 602 Prager Schule (cf. Prague School) 233 pragmatic function assignment 382 pragmatic pivot 1070 pragmatic principle 880 pragmatics 340 Prague School (cf. Prager Schule) 349 Präpositionalobjekt 1561 Präpositionsinkorporierung 742 predicate (cf. Prädikat) 374, 689, 905, 1436 predicate formation 376 predicate frame 374 predicate operator 385 predicate, basic 376 predicate, derived 376 predicate, eventive secondary 1442 predicate, nonverbal 1424 predicate, stative secondary 1442, 1444 predicate-initial language 1418 predication 651 predication operator 385 prenominal gap strategy 1387 preposition 112, 1401, 1453 present tense 1244 presentative 833 Pressesprache 1533 pressure of position 1281 presupposition 493 preverb 1433 primal word 1148 primary focus 1000 primitive constituent 454 primitive relation 696 principle 506, 513 principle (C) 945, 959 principle of recoverability 1604 Prinzip der eindeutigen Bindung 45 PrNMH (Prepositional Noun Modifier Hierarchy) 838 PRO 40, 518, 624 pro, A’-bound 1296 pro-adjective 679 pro-adverb 679 pro-drop parameter 642, 781, 1348, 1352, 1493

1637

pro-form 669 pro-numeral 679 pro-verb 679 production 1489 progressive 1245, 1246 Projection Principle 527 projection, maximal (cf. Projektion, maximale) 517 projectivity 321 Projektion 283 Projektion, maximale (cf. projection, maximal) 8, 15 Projektionshöhe 620 prominence typology (cf. Prominenztypologie) 1065 Prominenztypologie (cf. prominence typology) 1040 Pronominal 40, 959 pronominal prefix 1398 pronominalization 953, 1236 pronoun 521, 940, 1358, 1440 proper government 540, 711 proper name 969 Proportionsproblem 80 protasis 923 protasis clause 899, 926 proto-agent 1092 proto-patient 1092 proto-role 1102 Prototyp 628, 1038 PS-Regel (cf. phrase structure rule) 478 PSG (cf. phrase structure grammar) pseudo-cleft construction (cf. Pseudospaltsatz) 1326 Pseudospaltsatz (cf. pseudo cleft construction) 1211 psycholinguistics 1485 psychology 1484 PTQ 423 purposive infinitival 931 Pushdown-Automat 1594

Q Q’-lowering 1339 quantification 918, 966 quantifier (cf. Quantor) 867, 942 quantifier binding 868 quantifier raising (cf. Quantorenanhebung) 871, 941 quantifier scope 397, 1338 quantifier-float 1326 Quantor (cf. quantifier) 990 Quantorenanhebung (cf. quantifier raising) 61 Quantorenskopus 64 question 1428 question phrase 534 question word 1385 quotation 897

R R′ (symbolizing) 1464 R″ (conceptualizing) 1464 r-expression 521, 959 raising 498, 519, 575, 1327 raising infinitive 1290 raising verb 1348 rank-shifting 98 Raster, thematisches 55 realignment 1197 realism 1478 realization 351 Reanalyse (cf. reanalysis) 1261, 1272 reanalysis (cf. Reanalyse) 1145, 1156 Rechtsadjunktion 864 Rechtsextraposition (cf. right dislocation) 865 reciprocal pronoun 675, 1337, 1403 reconstruction (cf. Rekonstruktion) 537, 1124, 1229 reconstruction, historical 1179 recursive category 98 Redeteil 161 Reduktionsverfahren 768, 788 reference 651 referential dependency 1094 Referenz, exklusive 988 Referenz, inklusive 988 refl feature 1373 reflexive 1337, 1350, 1457 reflexive pronoun 674 reflexivization 953, 1095, 1313, 1373 Regel E 791 Regel R 797 Regelumkehrung 1134 register 1520 Reihenfolgebeziehung 180 Reihenfolgemerkmal 1015 Rekonstruktion (cf. reconstruction) 83 Rekonstruktion, innere 1264 Rektion 139, 615, 638 Rekursivität 4, 1576 relation 119 Relation, finale 604 Relation, grammatische 601, 1040 Relation, initiale 604 Relation, oblique 602 relational category 329 Relational Succession Law 607 relational typology 1076 Relationale Grammatik / relatio- nal grammar (RG) 601 relationization 298 relative clause (cf. Relativsatz) 536, 1299, 1385, 1404, 1414, 1445

relative clause, restrictive 1322, 1388 relative pronoun (cf. Relativpronomen) 677 relative pronoun strategy 1386 relative scope 867 Relative Time Hypothesis 1180 relativisation operator 389 Relativitätsprinzip 243 relativization 1372, 1458 relativized minimality 542 Relativkonstruktion 1211 Relativprinzip 1200 Relativpronomen (cf. relative pronoun) 1204 Relativsatz (cf. relative clause) 79, 853, 1199 Relativsatz, appositiver 1204 Relativsatz, prädikativer 1213 Relativsatz, restriktiver 1204 relevant category 1062 Repellor 613 restriction 353 restrictive particle 979 restrictor 379 Resultativ (cf. resultative) 744 resultative (cf. Resultativ) 807 Resumptivum 1203 retired term 1384 Revised Extended Standard Theory (REST) 1171 Reziprozität 82 Rhema (cf. rheme) 161, 355, 689, 1047, 1464 rheme (cf. Rhema) 342, 973 right coordination reduction (RCR) 784 right dislocation (cf. Rechtsextraposition) 885, 1221, 1360 Right-Node-Raising (RNR) 68, 750, 773, 783, 1333 rigid minimality 542 role relation 465 Rolle, semantische 1555 Rolle, thematische 55 root-theory 1154 Rückwärtsellipse 770 rule system 1477, 1480 rule-type 318 rule-weight 122 -rV suffix 809

S S-modifier 1330 s-selection (cf. S-Selektion) 516 S-Selektion (cf. S-selection) 57, 62 S-structure (cf. S-Struktur, Oberflächenstruktur) 511, 534 S-Struktur (cf. S-structure) 22 sandhi effect 1122 sarcasm 900

1638

satellite 376 Satz (cf. sentence) 167, 221, 281, 1546 Satz-Adverbial 1563 Satzakzent 1017 Satzanalyse 1548 Satzart 1007, 1537 Satzaspekt 239 Satzbauplan 249, 1538 Satzdefinition 199, 228 Satzglied 3, 134, 153 Satzklammer 3, 9, 992, 1266, 1289 Satzmodell 251 Satzmodus 851, 1006 Satzperspektive 270 Satzperspektive, funktionale 1047 Satzsandhi 205 Satzstruktur 848 Satztyp 851 Satztypselektion 852 Satzumfang 1539 Satzverhältnis 149, 164 Schulgrammatik 1546 scope (cf. Skopus) 867, 963, 979, 986 scope function 438, 459 scope value 876 scope, wide 868 Scrambling 45, 863, 871, 1306 segment 522 Seichtstruktur 267 Selbstorganisation 1134 Selektionsbeschränkung 481 Selektionsmerkmal 481 self-insertion 953 Semantem 274 semantic argument 908 semantic bootstrapping 1493 semantic case 1445 semantic dependency 1094 semantic participant role 708 semantic pivot 1070 semantic role 909 semantic role hierarchy 834 sense anaphor 1335 sentence (cf. Satz) 91 sentence grammar 91 sentence stress 999 sentence, complex 1152 sentence, declarative 109 sentence, imperative 109 sentence, interrogative 109 sentential complement 1405 sentential negation 915 sequence 115, 441 serial construction, covert 809 serial construction, overt 809 serial verb construction 799, 1454 Serialisierung (cf. serialization) 637, 1040, 1270

serialiation (cf. Serialisierung) 1289 si-construction 1353 Śikṣā (Phonetik) 199 Silbe 49 Skopusambiguität (cf. scope) 62 slash feature 1370 sloppy identity 947 sloppy reading 942 Sluicing 779 Small Clause / Koprädikativ 69, 530 Small Clause, resultative 70 social distance 900 sociolinguistics 1514 Solidarität 265 sṓmata 174 Sortenkonzept 989 sound system 1463 Soziolekt 1168 Spaltsatz (cf. cleft construction) 1211 specificity 875 Specified Subject Condition 503, 960 specifier (cf. Spezifikator) 497, 527 speech act 212 speech act theory 1470 speech error 1489 speech-shadowing 1487 Spezifikator (cf. specifier) 8 Spezifikator-Kopf-Kongruenz 36 Spezifiziererposition 851, 857 Sphoṭa 205 Spiegelprinzip 24 Spiralenprinzip 271 split ergative-accusative 1096 split intransitivity 1103 Sponsor 608 Sprachbund 1223 Sprachdidaktik 1545 Sprache, aufzählbare 1581 Sprache, aussagenlogische 415 Sprache, dichterische 1535 Sprache, disambiguierte 417, 421 Sprache, einseitig-lineare 1589 Sprache, entscheidbare 1581 Sprache, formale 1576 Sprache, politische 1533 Sprachfamilie, abstrakte 1590 Sprachforschung, ganzheitliche 243 Sprachkunde 1547 Sprachlehre 1547 Sprachliste 1140 Sprachtypologie 616 Sprachunterricht, kommunikationsgeleiteter 1566 Sprachveränderung 1130 Sprachwandel 1127, 1166 Sprachwandel, analogischer 1260 Spur (cf. trace) 39

Spurentheorie 25 Stabilität 1130 Stabilitätstheorie 611 stage-level predicate 967 Stamm 16 standard negation 915 Standard Theory 1169

(ST; cf. Stan-dardtheorie) Standardtheorie (cf. Standard Theory) 480 Standardwert 641 state of affairs 374 static form 1403 stemma 300 stemma, virtual 302 Stil 1529 Stileffekt 1537 Stilfigur 1542 Stilideal 1530, 1534 Stilistik 1528 Stratal Uniqueness Law 605 Stratifikationsgrammatik 296 Stratum 603 Stray-Affix-Filter 23 stress 999 stripping 1333 strong competence hypothesis 1502 structural case 558 structural description 507 structural relation 99, 694 structure 872 Struktur 257 Struktur, hybride 151 Struktur, morphologische 15 Struktur, prosodische 49 Struktur, semantische 732 Struktur, syntaktische 2 Strukturalismus 1128 Strukturalismus, amerikanischer 280, 470 Strukturalismus, europäischer 257 Strukturanzeiger 292 Strukturbedeutung 291 Strukturdiagramm 218 Strukturdomäne 619 Strukturerhaltungsprinzip 483 Strukturkasten 293 Strukturtyp 1008 subcategorization 1366 Subcategorization Principle 573, 1085 subjacency (cf. Subjazenz) 500 Subjacency Condition 1493 Subjazenz (cf. subjacency) 41 subject (cf. Subjekt) 104, 689, 1068, 1380, 1428 subject agreement 1408 subject assignment 381 Subject Condition 499 subject control 934

1639

subject drop 1311 subject prefix 1399 subject selection 1409 subject-inversion 833 subject-prominence 1066 subject-topic cycle 1220 Subjekt (cf. subject) 137, 154, 165, 1556 Subjekt, psychologisches 1045 Subjekt-Lücke 781 Subjekteigenschaft 1557 Subjektextraktion 643 subjunction 930 subordinate 332 Subordination 170, 318, 680, 934 Subordinator 1201 Subset Principle 1494 substitution 402, 947, 1603 superiority effect 1302 suprasegmental 114 surface syntax 436 Symbol 415 symmetry 901 synonymy 1138 synsemanticum 652 Synsemantikum 1252 syntactic base form 443 syntactic change (cf. Syntaxwandel) 1116, 1136 syntactic constellation 872 syntactic level 334 syntactic pivot 1069, 1078 syntactic position 210 syntactic relation 686, 912 syntactic structure 436 syntactic theory, Arabic 208 syntactic typology (cf. Typologie, syntaktische) 1109 syntactic weight 880 syntagm (cf. Syntagma) 96 Syntagma (cf. syntagm) 156, 267 Syntagmatik 236 Syntax, ambige 1588 Syntax, autolexikalische 30 Syntax, generative 1579 Syntax, kontextfreie 1584 Syntax, kontextsensitive 1582 Syntax, negative 185 Syntax, natürlichkeitstheoreti- sche (NTS) 610 Syntax, reguläre 1589 Syntax, strukturalistische 257 Syntax, traditionelle 216 Syntaxforschung 130 sýntaxis 132, 267 Syntaxwandel (cf. syntactic change) 1126, 1166 sýnthesis 132, 267 Systemic Functional Grammar 339 Systemic Grammar 1514

T T-Marker 478 T-relative 1446 table of coreference 493 Tail 383 TAM paradigm 1378 taxonomic NP 967 Téchnē-Grammatik 132, 135 Teilmenge-Problem 1598 temporality 385 Tempus (cf. tense) 58 tense (cf. Tempus) 1244 tense marker 1375 tense prefix 1399 tense, absolute 1244 tense, compound 1416 tense, relative 1244 Tensed S Condition 503, 960 term 379 term formation 379 term insertion 380 term operator 379 Terminalsymbol 473 Termrelation 602 test of completeness 91 test of correctness 92 text grammar 91 text type (cf. Texttyp) 830 Textsorte 1536 Texttyp (cf. text type) 1530 TF-Kompositionalität 796 TF-Markiertheit 796 that-trace effect (cf. that-traceEffekt) 544, 1302, 1519 that-trace-Effekt (cf. that-trace effect) 42 Thema (cf. theme) 355, 1047 thematic dependence 1093 Thematic GP Theory 1505 thematic hierarchy 690, 1094 thematic independence 1093 thematic relation 690, 827 thematic relation, basic 1092 thematic role 515, 690, 1092 thematic role system 1420 theme (cf. Thema) 342, 383, 973 Theme First Principle 1220 theory integration 434 there-insertion 1325, 1339 theta-chain, unbalanced 991 Theta-Kriterium 47 theta-marking 523 theta-position 523 Theta-Rolle 733 Thetarektion 624 Tiefenanapher / deep anaphor 779 Tiefenstruktur (cf. D-structure) 481 tier 20 Tilgung (cf. deletion) 1603 Tilgungsregel (cf. deletion rule)

770, 1584 title 898 TMA particle 1454 Tonakzent / pitch-accent 52 Tonhöhenverlauf 1018 Tonmelodie 19 topic (cf. Topik) 104, 343, 350, 383, 595, 689, 832, 880, 909, 1065, 1236, 1323, 1419 topic agreement 1408 topic continuity 837 topic drop 1312 topic marker 1375 topic-configurationality 882 topic-focus articulation (TFA) 355 topic-prominence 1066 topic-shifting 1237 topicalization 498, 872, 1297, 1360, 1370 Topik (cf. topic) 180, 791 Topik-Wegfall 781 Topikausdruck 795 tough-movement 1325 trace (cf. Spur) 518 trace theory 504 transference 310 transformation 97, 472 Transformation, generalisierte 5, 479 Transformation, grammatische 1167 Transformation, obligatorische 483 Transformation, optionale 483 Transformation, singuläre 479 transformational rule 511 transformationeller Zyklus 482 Transformationsgrammatik 469, 473 Transformationsgrammatik, generative 1166 Transformationsregel 1602 Transformationssyntax 1602 transitive realignment 1189, 1194 Transivität 136, 1039 Transparenzprinzip 1171 transposition 111 trigger 1419 trigger system 1420 trigger-marking 1421 Turingmaschine 1579, 1594 Typ, logischer 57 Typ, morphologischer 1264 Typ-0-Syntax 1580 Typ-1-Syntax 1581 Typ-2-Syntax 1584 Typ-3-Syntax 1589 type (cf. Typus) 1145 type-raising 402 Typenlogik (TL) 426 typological adequacy 371 Typologie (cf. typology) 1128 Typologie, evolutive 1264

1640

Typologie, holistische 1040 Typologie, morphologisch motivierte 1032 Typologie, syntaktische 1031, 1043 typology (cf. Typologie) 1176 Typus (cf. type) 1031 Typus, struktureller 1031

U Übergangsnetzwerk-Grammatik 1579 Übersetzungsbasis / translation base 420 Übersetzungsfunktion 418 Übersetzungshomomorphismus 418 UG (cf. universal grammar, Universalgrammatik) Umfang 235 Umformungstest 1559 Umhierarchisierung / revalua- tion 604 Umlaut 1167 unbounded dependency 578, 595 unconscious rationality 1217 Unendlichkeitsproblem 1598 Uniformitarianismus 1129 uniformity 547 unilateral correlation 1052 Uniqueness Condition 587 Unity of Indexing Hypothesis 991 Universal 637, 687, 1034, 1047 universal grammar 485, 507, 1139, 1219, 1492

(UG; cf. Universalgrammatik) universal, diachronic 1181 Universal, implikatives 1034 Universalgrammatik 1, 612, 635

(UG; cf.universal grammar) Unlösbarkeitssatz 1598 Unterordnung 152, 170 ’uṣuwl 209, 211 utterance (cf. Äußerung) 94

V V-Dependenzstruktur 138, 282 V-Determinationsstruktur 259 V-Inkorporation 30 V-modifier 1329 V-V compound 811 V-visibility 814 V-visibility, extended 816 V/2 order (cf. Verbzweitposition) 1219 Vākya-vāda (Satzlehre) 200 valency (cf. Valenz) 112, 302, 353, 448, 905 valency alternation 576 valency dictionary 312

valency raising 308 valency reduction (cf. Valenzreduktion) 308 valency, qualitative 303, 374 valency, quantitative 303, 374 valency, selectional 304 Valenz (cf. valency) 3, 55, 732, 1549 Valenzerweiterung 734 Valenzreduktion (cf. valency reduction) 733 Valenzumordnung 734 Variable 39 variable pronoun, bound 940 variable, bound 940 variable, component 119 variable, holistic 120 variable, linguistic 118 Variable, syntaktische 415 Variation 611 Variation, sprachliche 1255 variety 1520 Vedānga 199 Velarisierungsregel 54 verb 111, 138, 648, 653, 1432 verb agreement 1086 verb copying 1460 verb patterner 1052 verb phrase deletion 785 verb, complex 1401 verb, ergative 1348 verb, intransitive 1348 verb, meteorological 450 verb, psychological 1350 verb, transitive 1350 verb-focus 805 verbal morpheme 1435 verbal root 1149 Verbalphase 863 Verbendposition (V/E) 846 Verberstposition (V/1) 846 verbiness 657 Verbklassifikation 201 Verbzweitposition/-stellung (cf. V/2 order) 846, 1169 Verbzweitsatz 25 Verifikationismus 470 Verkettung 1577 Verschiebeprobe 187 Verschmelzung / union 605 Vertextungsstrategie 1530 view function 459 Visibility Condition 560 Visser’s generalization 594 vocative 1324 voice marker 1375 voice system 1401 Vokalmelodie 19 Völkerpsychologie 1044 Vorfeld 3, 9, 847, 1267, 1289 Vorwärtsellipse 770

VP-constituent 1301 VP-Ellipse 778 VP-Hülle 35 Vyākaraṇa (Grammatikwissenschaft) 199

W w-Frage 859 w-Imperativsatz 858 Wackernagel-Position 848 Wandel, syntaktischer (cf. Syntaxwandel) weather verb 1348 Weglaßprobe 187, 1557 Weltansicht 242 wh-extraction 1294 wh-focus 1004 WH-Island Condition 499 wh-movement 1294, 1415, 1459 wh/Q-interaction 877 what-cleft 892 Wiederschreibregel 1579 Wissenschaftssprache 1533 Wohlgeformtheitsbedingung 1172, 1600 word (cf. Wort) 95, 110 word class 646 word grammar 327, 331, 1525 word order (cf. Wortstellung) 466, 826, 868, 1050, 1216, 1359, 1371, 1378, 1425, 1431 word order change 1150, 1216 word order typology 1050 word order, consistent 1217 word-formation 510 word-sentence 1150 Wort (cf. word) Wort, leeres 175 Wort, phonologisches 50, 54 Wort, volles 175 Wortart (cf. part of speech) 166, 201 Wortfolgetypologie 1039, 1046 Wortgruppe 239 Wortstellung (cf. word order) 268, 637, 992, 1539 Wortstellungstypologie 1032 written language 829 Wunschsatz 1025 Wurzel 16

X X-bar model 96 X-bar theory 329, 497, 513, 527 X-bar-Schema 2, 9 X-bar-Schema, enge Version 11 X-bar-Schema, erweiterte Form 13 X-bar-Schema, morphologisches 18 X-bar-Theorie 1172 X-Projektion 10

1641

Z Zeichen 1577 Zeichen, grammatisches 1252 Zeichen, kategorematisches 1252

Zeichen, lexikalisches 1252 Zeichen, synkategorematisches 1252 Zeichenvorkommen 1578 zero copula 1522

Zweigliederung 281 Zweiseitigkeit 178, 291 Zweiteiligkeit 178, 291 Zweitsprachenerwerb 1566 Zyklizität 1131