Models of Inflection (Linguistische Arbeiten) 3484303883, 9783484303881

Citation preview

Linguistische Arbeiten

388

Herausgegeben von Hans Altmann, Peter Blumenthal, Herbert E. Brekle, Gerhard Heibig, Hans Jürgen Heringer, Heinz Vater und Richard Wiese

Models of Inflection Edited by Ray Fabri, Albert Ortmann and Teresa Parodi

Max Niemeyer Verlag Tübingen 1998

Die Deutsche Bibliothek - CIP-Einheitsaufnahme Models of Inflection / ed. by Ray Fabri ... - Tübingen : Niemeyer, 1998 (Linguistische Arbeiten ; 388) ISBN 3-484-30388-3

ISSN 0344-6727

© Max Niemeyer Verlag GmbH, Tübingen 1998 Das 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. Gedruckt auf alterungsbeständigem Papier. Druck: Weihert-Druck GmbH, Darmstadt Einband: Industriebuchbinderei Nädele, Nehren

Table of Contents

Preface Ray F abri Models of Inflection

vii

1

Geert Booij The Demarcation of Inflection: a Synoptical Survey

11

Philippe Ségéral and Tobias Scheer A Generalized Theory of Ablaut: the Case of Modern German Strong Verbs

28

Albert Ortmann The Role of [±animate] in Inflection

60

Teresa Parodi Aspects of Clitic Doubling and Clitic Clusters in Spanish

85

Birgit Gerlach Mediopassive in Greek and the Hierarchy of Functional Categories

103

Ilse Zimmermann Das dativische pronominale Klitikum in der DP-Struktur des Bulgarischen

119

Hilke Elsen The Acquisition of Past Participles: One or Two Mechanisms?

134

Katrin Lindner Overgeneralization Revisited: the Case of German Past Participles

152

Conxita Lieo Proto-articles in the Acquisition of Spanish: Interface between Phonology and Morphology

175

Dunstan Brown Stem Indexing and Morphonological Selection in the Russian Verb: a Network Morphology Account

196

vi

Table of Contents

Wolfgang Ullrich Wurzel Drei Ebenen der Struktur von Flexionsparadigmen

225

Martin Neef The Reduced Syllable Plural in German

244

Damaris Niibling Wann werden die deutschen Präpositionen flektieren? Grammatisierungswege zur Flexion

266

Preface

The aim of the volume is twofold: to contribute towards a better understanding of the mechanisms of inflection, and to encourage the reader to reflect upon the differences and the similarities in the approaches adopted by the various contributors. The book provides a platform for researchers with different views on inflectional morphology to discuss the results of their research in the light of the particular framework they have chosen to work in. The main thrust of the volume is to encourage the reader to discover what the common features of the different models are. In particular, it should stimulate readers to think about whether, after careful investigation, apparent insurmountable differences might turn out to be purely superficial in nature, or even non-existent. On the other hand, they might, of course, come to the conclusion that the analyses discussed here indeed do reflect fundamentally different and possibly incompatible views on inflectional morphology. The contributions in this volume have been ordered according to how general or specific the subject matter they deal with is, starting off with the more general topics and moving on to the more specific. The first paper provides an overview of the main controversies that rage within the area of inflectional morphology, and places the various contributions within a more general theoretical perspective. A number of papers follow dealing with general theoretical aspects such as the difference between derivation and inflection, irregular verb inflection, animacy in inflection, and clitics. These are followed by studies of functional categories, articles on the acquisition of aspects of inflection, the description of a formal implementation of a fragment of Russian verb inflection, and finally articles that deal specifically with some aspects of German inflectional morphology. Unsurprisingly, since the area we are looking into is so vast, there are thematic issues and also specific of approaches that are not included in this volume. In particular, work within two currently influential frameworks, namely Distributed Morphology and Optimality Theory, is missing. This is so for the very mundane reason that we received no contributions from researchers working specifically within these frameworks. It would perhaps be apt here for us to describe briefly how this volume came into being. The three editors worked together on a project on inflection for a number of years. The project is one of a number of sub-projects in a multi-project research programme (Sonderforschungsbereich) funded by the Deutsche Forschungs-Gemeinschaft (German Research Foundation), involving researchers from the three universities of Düsseldorf, Cologne and Wuppertal. The underlying theme of the multi-project research programme is "Theorie des Lexikons" (Theory of the Lexicon) and the project in which the editors were involved started out from the morphology of agreement phenomena and later moved on to investigate inflectional categories in general.

viii

Preface

As colleagues at the Heinrich-Heine University in Dusseldorf, the three editors spent hours of deliberation on various areas involving different aspects of inflection, and in particular on the various models proposed to account for inflectional phenomena and the interaction between morphology and syntax. The conception of this volume came about as a result of long hours of discussion with colleagues and students, usually starting off in the General Linguistics department, and most often ending in different kinds of a restaurants and pubs in Düsseldorf, accompanied by good food and one or two glasses of wine. The outcome of these intellectualcum-culinary activities was twofold: the organisation of a workshop at the annual general meeting of the Deutsche Gesellschaft für Sprachwissenschaft (German Linguistics Society) at the university of Freiburg in 1996, and publication of this volume, in which some of the contributions are based on talks delivered at the workshop. The work in preparing this volume was hampered to some extent by the physical distance between the three editors, who went separate ways a few months before the Freiburg workshop. However, the strong personal and professional links that were forged between the three, and the wonders of modern technology (which are turning the great wide world - at least the richer part of it - into a global village) have made this co-operative effort between Düsseldorf, Cambridge and Malta possible. For the publication of a volume of this nature, editors do not work in a vacuum but draw help and inspiration from those around them, namely colleagues, friends, and family, all of whom provide different kinds of support. We would therefore like to thank the following persons for their help and advice: Dieter Wunderlich, for both intellectual and material support, in providing us with the knowledge and the analytical tools we need to be able to understand and become familiar with the basics of linguistics in general and inflection in particular. Dieter has also been very generous in providing funds and actual physical space for us to be able to bring the enterprise to a satisfactory conclusion; Richard Wiese, for his support in making publication of this volume in the "Linguistische Arbeiten" series possible; Thomas Gamerschlag, without whose dedication and thoroughness in formatting we would not have been able to send off the final version in time for publication; all the participants and speakers at the workshop in Freiburg for their encouraging and inspiring contributions; finally, our partners and friends for supporting us and enduring our occasional uncommunicativeness, or absence, with good grace. We hope to have compiled a volume that contains articles that are stimulating and challenging, and that provide the reader with at least a taste of current thought on inflection. We also hope that this volume serves to stimulate researchers working within the field of inflection to read each others' ideas, to be more tolerant of other people's work, to adopt a more critical attitude towards their own work, and finally to look for possible areas of co-operation with others working in related fields. Düsseldorf/Malta/Cambridge, April 1998

Ray Fabri (Universität Düsseldorf/University

of Malta)

Models of Inflection

1. Introduction The last few years have witnessed a 'Renaissance' within the field of morphology, in particular in the area of inflection. A number of models that attempt to account for both old and new problems have been (and are still being) proposed and discussed. This Renaissance has been particularly evident within the framework of generative grammar, in which, for a long time, inflection was treated as some sort of Cinderella. The situation changed after the publication of an influential paper by Pollock (1989), who attributed significant syntactic variations in the syntax of French and English (in particular, differences in word order) to differences in what can be considered to be morphological properties of these languages. In order to explain the observed differences, Pollock split up the INFL(ection) position in syntax into T(ense) and AGR(eement). These syntactic positions encode functional information that is typically encoded by morphological entities like affixes and morphological processes like ablaut in inflectional languages. Overnight inflection was propelled into the foreground in discussions on models of grammar within the generative framework, and promoted to the regal position of 'Queen of Linguistic Diversity'. It should be noted, however, that outside of the generative framework, and in particular within the European linguistic tradition, the study of inflection has enjoyed uninterrupted attention at least since the 19th century. What follows is a brief overview of what I consider to be some of the most controversial questions that currently occupy researchers working on inflection. For reasons of space, this overview is of necessity very schematic and rather simplified, and it is, of course, biased by my own theoretical interests and work. It should, however, serve its function as a general introduction to the topic of inflection and help to set the general framework within which the individual contributions to this volume can be located and the connections between them made explicit.

2. Models and questions 2.1 Connectionism and Symbolism Two approaches to morphology that differ quite radically in the assumptions they make about the nature of the language processing mechanism are connectionism and symbolism. The basic controversy circles round the notion of a rule. While proponents of the connectionist model

2

Ray Fabri

claim that generalisations made by the linguistic system are based solely on pattern recognition, symbolists believe that it is symbol manipulation, i.e. the manipulation of specific language rules, that lies at the basis of the system. Connectionists claim that the rules assumed by linguists in their analyses are simply an artefact of the analytical stance they adopt in analysing the data. With respect to inflection, the connectionist model essentially claims that it is only phonological patterns that are learnt and generalised by the language acquisition device and not specific inflectional rules. In contrast, symbolists claim that word formation processes are ruleor constraint-based, and that the morphology is separate from (though, of course, related to) the phonology. To take an example, one of the questions that have been hotly debated by the two factions is how irregular forms should be dealt with in a model of inflection. The term irregular forms is being used here to cover both subregular and suppletive forms. Subregular forms are those forms that do not fit within the general pattern, and therefore cannot be accounted for by the 'regular' rule (or default case) which accounts for the largest number of cases. However, sub-regular forms also display a more or less regular pattern of their own and can, therefore, be accounted for by a sub-rule. Often, a particular pattern is restricted to base forms of a particular type and, moreover, there are a number of "exceptions", i.e. word forms having the required base form but a different derivation. Thus, in English the present/infinitive, past and participle forms of certain verbs display a common ablaut alternation, namely 111, /ae/ and IAI, respectively. Examples of such verbs are ring {rang, rung), sing {sang, sung), sink {sank, sunk) and drink {drank, drunk). This suggests that verbs with a final sequence consisting of the front high vowel III and velar nasal /η/ (e.g. /ir)/ and /drirjk/) lower the vowel to /ae/ in the past, and to IAI in the participle. This fact can be expressed by a rule or constraint that applies only to a restricted class of verbs, namely those with the /... ig ... / sequence in the base, and is therefore a kind of 'sub-rule'. Note that there are, however, also 'exceptions' to this particular alternation, since a number of verbs which display the /... irj ... / sequence in the base form, do not form the past and the participle with /œ/ and IAI. Thus, the past and participle forms of bring, namely brought and brought, change /ig/ to /o:t/, rather than to /aeq/ and /m)/, while fling (flung, flung) has both forms in /Λη/. Suppletion refers to cases in which the form of the verb changes in an unpredictable, idiosyncratic way. This is the case of verbs like go {went) and be (am, are, is, was, were) in English. Of course, such phenomena are not restricted to English. For example, in Maltese (an Arabic language spoken on the Maltese Islands) feminine forms are built by the addition of the suffix -a to the root form; thus tifia 'girl' is derived from the base form tifi. (The stem tifi is equivalent in meaning to 'child' and itself never occurs as a full form, the masculine form being tifel 'boy', in which the /e/ is inserted for reasons of syllabification.) In the case of the stem ragl, however, from which ragel 'man' is derived, the feminine form is not *ragla, as would be expected, but the suppletive mara 'woman'.

Models of Inflection

3

In connectionist models such as that proposed in the very influential article by Rummelhart and McLellan (1986), it is assumed that irregular forms are subject to the same patterning process as that involved in the formation of regular forms. In contrast, symbolists such as Pinker and Prince (1991) claim that the linguistic evidence points to the fact that regular and irregular formation are subject to two qualitatively different processes. They thus assume a two-level model of inflection in which the formation of regular forms is rule-based while that of irregular forms is associative, i.e. these forms are not generated by rules but stored as patterns in memory. However, what at first appeared to be an unbridgeable chasm separating these two mutually exclusive approaches now seems to be narrowing down to some extent. Thus, connectionists have developed models that are subtler than the original, rather naive model proposed by McLellan and Rummelhart (1986), allowing some kind of (internal) symbol manipulation by making use of hidden levels. Moreover, it was clear from the beginning that at least one part of the two-level model of Pinker and Prince (1991), namely that part dealing with irregular forms, is in a sense "connectionist", or at least very close in spirit to that model. This is so because the two-level model assumes that irregular forms are stored in memory and not derived by general formation rules. Although there seems to be some degree of convergence, the debate about the properties of the underlying model is still in full swing and promises new insights into the workings of inflectional systems. For two studies within the connectionist framework in this volume, see the contributions by Elsen and Lindner. 2.2 Generativism and non-generativism Another dichotomy is that between those who take a functionalist and/or typological approach, and those who work within some formal structural framework, as is the case with studies within the generative grammar. The former faction includes approaches taken within, for instance, the Natural Morphology model in Dressier (1985), Wurzel (1989), Mayerthaler (1988), and the work of typologists such as Greenberg (1963), Corbett (1987) and Plank (1991), among others. The latter includes works such as Anderson (1992), Lieber (1992), Halle and Marantz (1993), Wunderlich and Fabri (1996) and Gazdar and Cahill (1997). Unfortunately, the relation between these factions, in particular that between "natural morphologists" and "generativists", often seems to be one of antagonism and mutual dislike, with an often complete rejection by one side of the ideas of the other. It is difficult to justify the extent of this mutual dislike. The roots of the dichotomy lie in the different sorts of explanations that are sought for linguistic phenomena, which also leads to different methodological approaches. The "generativists" tend to focus on very specific areas of language and come up with primarily system-internal structural explanations based on abstract notions such as ccommand (see Chomsky 1995:35, and references therein), head movement (see, for example,

4

Ray Fabri

Culicover 1997:chap. 10), feature checking and the distinction between strong and weak features (see Chomsky 1995, and Marantz 1996 as well as Pollock 1989 on strong and weak features, among others). The search is for a minimalist (i.e. economical) model of Universal Grammar consisting of a small set of principles, and parameters whose setting determines the form of a particular language. In contrast to the generativists, functionalists tend to lay more stress on intuitive explanations based on extra-linguistic factors (rather than abstract, specifically linguistic principles), such as the properties of the articulatory system, and those of the perceptive system (iconoclasm), as well as the exigencies of communication. Descriptive typologists criticise the generativists because the latter tend to base their UG principles on a restricted database, often looking (in-depth) into one language. The universal principles that generativists postulate are often challenged by apparent counter-evidence from other languages. Typologists prefer to examine data from a large number of languages and look for cross-linguistic correlations. For a few typical examples, see Greenberg (1963), Corbett (1991), Nichols (1986) and Plank and Schellinger (1997). Typologists also criticise the generativists because they believe that the formalisms that the latter develop are too inflexible and restrictive, and because they have a tendency to ignore facts that do not behave the way they should according to the rules of the system they have devised. In his criticism of "transformational-generative grammar", Comrie (1981:5/8) states that it "is characterised by a number of questionable assumptions that are crucial to the argumentation, these assumptions being for the most part untestable, at least at present, so that acceptance of the paradigm becomes simply a matter of faith." Moreover, "what is even more disturbing is that attempts to bring such data [i.e. counter examples] to the attention of practitioners of the one-language approach to universale meet with lack of interest, rather than any attempt to work out the implications of the data for the over-all theory." Although, these two factions often display a very belligerent attitude, and tend to be mutually intolerant and arrogant, in actual fact the positions they represent are not as far apart as some of their proponents would have us believe. There are, fortunately, researchers working within these fields who hold the very reasonable view that the different approaches actually complement one another, and that they can have a positive, stimulating effect upon each other. Thus, for example, Kirby (1997:29) concludes that "functional explanations for language universale are compatible with generative approaches to language acquisition. (...) Indeed, in some circumstances the functional approach and the 'formal' approach to universale may both be required in order to explain cross-linguistic patterns." In fact, one can generally observe a certain degree of convergence between the two approaches. For example, in the last few years the generativists have been taking an interest in and exploring more intensively languages other than just English and German or perhaps a couple of Romance languages, thus widening their data-base and often radically changing their models as a result. Thus, to cite one example, Baker (1988) bases his theory of incorporation on a wide variety of languages, including, among others, Chichewa, Groenlandie Eskimo, Iro-

Models of Inflection

5

quoian, Kinyarwanda, Romance languages and Tzotzil. For a study based on cross-linguistic data in this volume, see Ortmann's contribution, which deals with the role that the gender category animacy/human plays in determining a number of cross-linguistic asymmetries. On their part, typologists often also try to develop formally explicit models. A particularly successful attempt in this direction is Network Morphology, which was proposed by Corbett and Fraser (1993). In this model the Default Inheritance Trees of DATR, a formal knowledgerepresentation language, are used to implement and make explicit ideas about morphological processes and relations. Brown's contribution to this volume is an example of a study within this framework. Brown analyses the nominal system of Russian and gives specific examples of lexical entries for the word forms he discusses. 2.3 Lexicalism and syntacticism Within the generative fortress a battle has been raging for some time between two camps: those who claim that inflection forms a component within the grammar that is independent of the syntax, and those who consider inflection to be indistinguishable from the syntax. Of course, such controversies have a long tradition and did not originate within the generative framework. What is different within this approach is that there has been an attempt to formulate the relevant questions within a more explicit theoretical framework. The question is whether there is enough evidence to support the claim that inflection is subject to its own set of principles and constraints. Such evidence would justify the assumption that inflection must be formally distinguished from other subsystems within the grammar: either from word formation processes such as derivation, or from processes involving other structural levels such as syntax. Authors like Ouhalla (1991) and Fassi Fehri (1993) propose models in which inflection is (partly or wholly) localised within syntax. In her book Deconstructing Morphology Lieber (1992) also takes this view. In contrast, in Morphology By Itself AronofF (1994) comes out very strongly in favour of a morphological component (including inflection) that is independent of other components of the grammar, although he does not model his ideas within any specific formal framework. In this volume, Booij discusses the problem of inflection versus derivation, and Zimmermann presents a specific lexicalist model of the morphologysyntax interface based on a discussion of data from Bulgarian. (Like Brown, Zimmermann proposes explicit lexical entries, as well as derivations, for the data she discusses.) Finally, of particular interest with regard to the question about the interface between inflection and syntax is the status of clitics, which are "hybrid" elements that typically behave both like affixes and like words, i.e. they display morphological as well as syntactic properties. Clitics are therefore equally a challenge to those who strictly separate the morphology and the syntax and to those who treat them within one and the same component. In the present volume

6

Ray Fabri

Parodi presents a study that deals with the status of object agreement clitics in Romance languages, in particular with the issue of clitic doubling and the role of the animacy hierarchy in inflection. Theorists who assume that inflection is a well-defined autonomous subsystem do not only have to ask themselves where exactly this system is to be localised within the grammar (see section 2 .4), but they also need to be able to describe the properties of this component. That is, they have to specify which principles or constraints, rules or hierarchies must be assumed to account for, or at least adequately describe, the observed phenomena. One question that arises in this connection, for example, is whether the postulated principles should be violable, for instance in the sense of Optimality Theory (see Prince and Smolensky 1993), or whether they are inviolable as is the case in Halle and Marantz's (1993) Distributed Morphology. A third possibility is to postulate both types of principles, i.e. to assume that a model may operate with both hard (inviolable) and soft (violable) principles. Wunderlich and Fabri's (1996) Minimalist Morphology model is an example of such a model of inflection The principles that regulate paradigm construction are violable, while those that determine the combination of stems and affixes are inviolable. A study within the Minimalist Morphology framework in this volume is that by Gerlach, who discusses whether a particular linear order of inflectional categories corresponds to an inviolable hierarchy of functional categories. 2.4 Early and late insertionism Theorists who assume that inflection (or morphology) forms an independent component are divided among themselves by another controversial issue: the locus of the inflectional component within the grammar. I am referring specifically to the so-called "early" versus "late insertionist" positions. Essentially, the question here is whether morpho-phonological forms are built up in a particular component first (call it "the lexicon"), projecting their properties in the syntax at a later stage, or whether purely abstract syntactic categories are first checked in the syntax and then assigned a morpho-phonological form, i.e. "spelled out" at later stage in the grammar. Halle and Marantz (1993) in Distributed Morphology and Stephen Anderson (1992) in A-Morphous Morphology argue for a late insertionist position, while Wunderlich and Fabri (1996) in Minimalist Morphology, and other proponents of lexicalist models, such as DiSciullo and Williams (1987), argue for an early insertion approach. The drift between early and late insertionist, however, can also be narrowed. The terms early and late insertion only make sense in a serial, and therefore procedural, model. Since procedural models can often be "translated" into declarative ones, it is worthwhile asking whether the differences between early and late insertion are neutralised when the model is "declarativised". Looked at from the point of view of a declarative model, the dichotomy between early and late insertion seems to be superficial and not substantive. This matter,

Models of Inflection

7

however, and in particular the concept of 're-adjustment rules' in late-insertionist models needs to be looked into in detail before any specific conclusions are reached. In my opinion, the difference between early and late insertion models does not result from a fundamentally different view of the structure of the morphological component itself, but rather from a different conception of the syntax. Thus, late insertion models are based on a fully specified syntax, while early insertion models are usually compatible with an underspecified syntax. 2.5 Item-and-Arrangement, Item-and-Process, Word-and-Paradigm An important and basic question that arises in connection with the way in which a particular model of inflection is conceived of concerns the representation of morphemes, i.e. stems and affixes. For example, affixes can be treated as lexical entities, as in an Item and Arrangement model, "in which morphemes are the basic units of meaning and in which they are arranged linearly" (Matthews 1991:22). Alternatively, they can be conceived of as processes in the sense of an Item and Process model, "in which the structure of a word is specified by a series of operations" (ibid.) (See Hockett 1958 and Spencer 1991 for a detailed discussion). These two different conceptions of affixation are mirrored in the debate surrounding Halle and Marantz's Distributed Morphology model, on the one hand, and Stephen Anderson's A-Morphous Morphology, on the other. Several other questions arise in connection with the choice of a particular form of representation. For example, a model based on the notion of Item and Arrangement works well when it comes to accounting for phenomena involving affixation, however, it does not lend itself well to the representation of typical non-affixal phenomena such as ablaut and conversion (i.e. zero derivation). Different proposals about how to deal with such non-affixal phenomena in a model based on affixes as units of representation have been made in the literature. For example, in the Minimalist Morphology model, if ablaut forms are not derivable on some general phonological basis, they are explicitly listed in the base lexicon as templates in the form of default inheritance trees. With regard to the status of ablaut in the grammar, the study by Ségéral and Scheer in this volume provides an analysis of ablaut as governed by a universal vowel sequence. A third model mentioned in Hockett (1958) and discussed by Robins (1959) and Matthews (1972) is the Word and Paradigm model. The question is whether the notion of a paradigm has any relevance in the analysis of inflectional forms, i.e. whether paradigms can be regarded as objects which possess specific properties that can be defined formally, and which pertain to the inflectional component of the grammar. This is in fact the position taken in work by Wurzel (see, for example, Wurzel 1987, 1989, and this volume), Carstairs-McCarthy (see, for example, Carstairs-McCarthy 1987, 1994), and in the Minimalist Morphology model by Wunderlich and Fabri. An alternative view considers the paradigm to be simply an epiphenomenon, or even just a practical visual aid that serves to illustrate abstract relations between inflectional word

8

Ray Fabri

forms. According to this view, which is held, for example, by Halle and Marantz in their work on Distributed Morphology, paradigms lack an independent existence and play no role whatsoever in the formation of inflectional forms.

3. Conclusion The above is a very brief overview of what I consider to be the most salient theoretical questions that occupy researchers working on inflection. The list of open questions and controversies can be extended further, for example by the inclusion of historical factors that are responsible for changes within inflectional systems, or by a discussion of the implications of dialectal and stylistic variety for models of inflection. In this volume, Niibling discusses the phenomenon of language variation and change, in particular the concept of grammaticalisation. Another important area of interest is that of language acquisition, since the results of studies on the acquisition of inflectional systems are often used to provide evidence and arguments for or against particular theoretical claims. Studies within this area in this volume are those by Elsen and Lindner, which have already been mentioned in connection with the connectionist model, and that by Lieo, who presents a study of the acquisition of the article in Spanish. More specifically, Lieo studies the elements that can occupy the pre-nominal position in child language. Finally, markedness, both in the synchronic and the diachronic dimensions, is a notion that also plays a significant role in discussions on inflectional models. (See Battistella 1996 for a recent, detailed discussion of the concept of markedness.) As has been said above, this article is simply meant to provide a basis of reference for the diverse, detailed studies that follow and that can be located in one or more of the niches described here. Thus, the above overview mentions only a few of the many models of inflection that have been proposed to account for the observed phenomena. A model which has not been mentioned is Wortdesign, which is described by Neef in this volume. Neef uses this model to give an exhaustive description of German noun plurals. The main concern behind the enterprise that resulted in the present collection of papers has been that of providing a platform upon which workers doing research within very different frameworks can present their work on inflection for discussion. I believe that, with a small amount of good will, researchers can learn considerably from each other, even though their theoretical frames of reference might actually be, or at least appear to be, very different. Competition in doses can be very healthy, but all kinds of fundamentalism, as it is often bitterly advocated by some researchers, must be avoided at all costs. Even researchers who assume what might seem to be radically different approaches to the various complex problems involved should be able to discuss their work with each other and to show understanding and respect for each other's point of view. I am in fact convinced that in most cases a close scrutiny of these apparent differences will reveal that they do not have a basis in reality, and that the differences

Models of Inflection

9

in approach and underlying assumptions will turn out to be merely superficial or even complementary. After all, it is the same set of phenomena that are the focus of different analyses over and over again, so that most differences are simply the result of looking at this same set of phenomena from different angles.

References Anderson, Stephen R. (1992): A-Morphous Morphology. Cambridge: Cambridge University Press. Aronoff, Mark (1994): Morphology by Itself. Stems and Inflectional Classes. Cambridge, Mass.: MIT Press. Baker, Mark (1988): Incorporation: A Theory of Grammatical Function Changing. Chicago: University of Chicago Press Battistella, Edwin (1996): The Logic of Markedness. Oxford: Oxford University Press. Carstairs-McCarthy, Andrew (1987): Allomorphy in Inflection. Beckenham: Croom Helm. - (1994): Inflection classes gender and the principle of contrast. Language 70, 737-788. Chomsky, Noam (1995): The Minimalist Program. Cambridge, Mass.: MIT Press. Comrie, Bernard (1981): Language Universals and Language Typology. Oxford: Blackwell. Corbett, Greville (1987): The morphology-syntax interface. Language 63, 299-345. - (1991): Gender. Cambridge: Cambridge University Press. Corbett, Greville and Norman Fraser (1993): Network Morphology: A DATR account of Russian inflectional morphology. Journal of linguistics 29, 113-142. DiSciullo, Anna-Maria and Edwin Williams (1987): On the Definition of Word. Cambridge, Mass.: MIT Press. Dressier, Wolfgang U. (1985): Morphonology. Ann Arbor: Karoma. Fassi Fehri, Abdelkader (1993): Issues in the Structure of Arabic Clauses and Words. Dordrecht: Kluwer. Gazdar, Gerald and Lynne Cahill (1997): The inflectional phonology of German adjectives, determiners and pronouns. Linguistics 35, 211-245 Greenberg, Joseph H. (1963): Some universale of grammar with particular reference to the order of meaningful elements. In: J. H Greenberg (ed.) Universals of Language. Cambridge, Mass.: MIT Press, 73-113. Halle, Morris and Alec Marantz (1993): Distributed Morphology and the pieces of inflection. In K. Hale and S.J. Bromberger (eds.) The View from Building 20 Linguistics in Honor of Sylvain Bromberger. Cambridge: MIT Press, 111-176. Hockett, Charles F. (1958): Two models of grammatical description. In: M. Joos (ed.) Readings in Linguistics (end edition). Chicago: University of Chicago Press. Kirby, Simon (1997): Competing motivations and emergence: Explaining implicational hierarchies. Linguistic Typology 1, 1,5-31. Lieber, Rochelle (1992): Deconstructing Morphology: Word Formation in Syntactic Theory. Chicago: University of Chicago Press. Marantz, Alec (1996): The Minimalist Program. In Gert Webelhuth (ed.) Government and Binding Theory and the Minimalist Program. Oxford: Blackwell, 351 - 382 Matthews, P.H. (1972): Inflectional Morphology. Cambridge: Cambridge University Press. - (1991): Morphology. Cambridge: Cambridge University Press. Mayerthaler, Willi (1988): Naturalness in Morphology. Ann Arbor: Karoma. Nichols, Johanna (1986): Head-Marking and Dependent-Marking Grammar. In Language 62, 56-119. Ouhalla, Jamal (1991): Functional Categories and Parametric Variation. Pinker, Steven and Alan Prince (1991) Regular and irregular morphology and the psychological status of rules of grammar. In L.A. Sutton, C, Johnson and R. Shields (eds.) Proceedings of the 17,hAnnual Meeting of the Berkley Linguistics Society, Berkley 230-251. Plank, Frans (ed.) (1991): Paradigms: The Economy of Inflection. Berlin: Mouton de Gruyter. Plank, Frans and Wolfgang Schellinger (1997): The uneven distribution of genders over number: Greenberg Nos. 37 and 45. Linguistic Typology 1-1, 53 - 101.

10

Ray Fabri

Pollock, Jean-Ives (1989): Verb Movement, Universal Grammar, and the Structure of IP. Linguistic Inquiry 20, 365-424 Prince, Alan S. and Paul Smolensky (1993): Optimality Theory: Constraint Interaction in Generative Grammar (= Technical Report #2). Rutgers University Center for Cognitive Science. Robins (1959): In defense of WP. Transactions of the Philological Society, 116-144. Rummelhait, D. E. and J. C. McClellan (1986): Parallel Distributed Processing. Cambridge: MIT-Press. Spencer, Andrew (1991): Morphological Theory. Blackwell: Oxford. Wunderlich, Dieter and Ray Fabri (1996): Minimalist Morphology: an approach to inflection. Zeitschrift für Sprachwissenschaft 14, 2, 1995, 236-294. Wurzel, Wolfgang Ulrich (1989): Inflectional Morphology and Naturalness. Dordrecht: Reidel. - (1987): Paradigmenstrukturbedingungen: Aufbau und Veränderung von Flexionsparadigmen. In: A.G. Ramat, O. Carruba and G. Bernini (eds.) Papers from the 7,h International Conference on Historical Linguistics. Benjamins: Amsterdam, 629-644.

GeertBooij

(Vrije Universiteit

Amsterdam)

The Demarcation of Inflection: a Synoptical Survey

1. Introduction: the demarcation of inflection A volume on inflection and inflectional models like the present one raises the question whether inflection constitutes a set of phenomena or a module of the grammar that can be demarcated somehow from other morphological phenomena or other modules of the grammar. Such a demarcation of inflection from other kinds of morphology is presupposed in certain formal models of inflection such as Anderson's A-morphous Morphology (Anderson 1992), and Wunderlich's minimalist model of inflection (Wunderlich 1996). It is the aim of this contribution to present a survey of demarcation criteria, to evaluate them, and to discuss how the results of this evaluation bear on the position of inflection in the grammar. The primary distinction between inflection and derivation is a functional one: derivation (i.e. word formation except compounding) is that kind of morphology that serves to create new lexemes, whereas inflection serves to create different forms of the same lexeme. Therefore, it is also said that derivation, unlike inflection, creates words for new concepts. However, one should realize that derivation has a secondary function in that it is also used to make stylistic variation possible. For instance, of the following two alternative phrasings of a referring expression, the second makes use of derivation (of reader from read). (1)

He who reads this book /The reader of this book

The formal means by which inflection and derivation are expressed are often the same. In both, the processes of affixation, vowel change, reduplication etc. may be used. For instance, in many IndoEuropean languages inflection is expressed primarily by suffixation, which is also a kind of morphological operation used in derivation. Whether a sharp demarcation of inflection with respect to derivation is possible, is a classical problem in morphological theory. Whereas some linguists claim that there is no sharp demarcation between the two, and that there is a cline from prototypical derivation to prototypical inflection (Bybee 1985, Dressier 1989, Plank 1994), a number of generative morphologists make a sharp distinction which is reflected by their organizational model of the grammar, the split morphology hypothesis (Perlmutter 1988, Anderson 1982; 1992). In such models, derivation is considered as pre-syntactic morphology taking place in the lexical component, whereas inflection is equated with post-syntactic morphology. However, split morphology is not a necessary component of generative morphology: it is also possible to assume one morphological module that deals with both inflection and derivation.

12

Geert Booij

In order to come to grips with this demarcation problem, I will review the different arguments and criteria proposed in the literature for distinguishing between the two (see also Scalise 1986 and Dressier 1989). This will lead to the conclusion that inflection and derivation behave differently in a number of ways, and that they have to be distinguished in order to make the right generalizations about the formal make up of morphologically complex words. Yet, as will be shown in Section 3 in a discussion of the split morphology hypothesis, distinguishing inflection and derivation should not lead to separation of the two because there is some interaction between inflection and derivation that cannot be accounted for in split morphology theories. My findings are then summarized in Section 4.

2. Criteria of demarcation In this section I will review nine different criteria for distinguishing inflectionfromderivation. 2.1 Change of word class Thefirstcriterion is that derivation, unlike inflection, may change the word class of the input word. That is, derivation may cause transposition of word class. This is of course a consequence of the lexical enrichment and stylistic variation functions of derivation, which do not apply to inflection. However, since derivation does not necessarily change word class, the fact that a morphological process does not change word class, is no proof of its inflectional nature. First, a category-determining affix may happen to attach to a word of the same category. An example is the Dutch nominalizing suffix -er that may be affixed to nominal bases, e.g. wetenschap 'science' - wetenschapper 'scientist1. Second, languages may have so-called evaluative morphology that is transparent for the syntactic category and the gender of the base, but that is nevertheless felt as derivation as far as the semantic change involved is concerned. For instance, the Italian diminutive suffix -in can be attached to nouns to form nouns, and to adjectives to form adjectives: ragazzo "boy' - ragazzino 'little boy', ragazza 'girl' - ragazzina 'little girl', giallo 'yellow' - giallino 'yellowish'. This shows that the Italian diminutive suffix is transparent for the syntactic category and gender of its stem. On the other hand, Dutch diminutive suffixes are category-determining, and always create nouns, e.g. blond'id.' - blondje 'girl with blond hair'. Moreover, the Dutch diminutives are always neuter, unlike their base words: de stoel 'the chair, non-neuter1 versus het stoeltje the little chair, neut'. Thus, Dutch diminutives are a clearer case of derivation than the Italian ones. A problem for the demarcation criterion discussed here is that it is not true that inflection never changes the syntactic category of its inputs (Haspelmath 1996). For instance, infinitives do not only exhibit verbal properties, but also nominal ones. This is illustrated by the Dutch infinitival phrase het boeken kopen 'the buying of books'. The external syntax of the infinitive shows nominal behaviour,

The Demarcation of Inflection

13

since it occurs with the determiner hel 'the'. On the other hand, its internal syntax is that of a verb, since it allows for a preverbal preposition-less noun phrase-complement, boeken. Other examples of Dutch infinitives, preceded by a determiner and a preposition are: (2)

Ik ben

aan

het

fietsen

I

at

the

cycle-LNF

am

Ί am cycling' Ik zette het

op een

lopen

I

on a

walk-INF

put

it

Ί started running' In Romance languages, infinitives also function as nouns, as in French le diner 'the dinner'and le manger 'the food'. Infinitives also feed nominal word formation. In Dutch, as in many Germanic languages, verbal compounding is unproductive, whereas nominal compounding is productive. Infinitives behave like nouns in this respect: Dutch has many compounds of the type school-zwemmen

'school-swim-INF,

1

school-swimming which do not have finite forms, and thus cannot be interpreted as the infinitival forms of verbal compounds (Booij 1989). In many languages, participles behave like adjectives in that they can be used attributively and as predicates, and agree in gender, number and case with the noun that they modify. On the other hand, participles still have verbal potential in that they case-mark noun phrase arguments, as in the following example from German (Haspelmath 1996): (3)

ein den Richter überraschendes (NOM.SG.NEUT.) Faktum 'a fact that surprises the judge'

Here, the participle überraschendes 'surprising' agrees in number, case and gender with its head

Faktum 'fact'; yet is has an accusative-marked verbal complement den Richter 'the judge-ACC'. Participles also feed deadjectival word formation, as in English spoi'/eafoess: and its Dutch equivalent bedorvenheid. They lexicalize quite often as adjectives with an idiosyncratic meaning, e.g. Dutch gesloten (past participle) 'closed, close-mouthed', and woedend (present participle) 'raging, angry'. In Biblical Hebrew, participles may have the distribution of nouns. For instance, they can be preceded by a determiner, and they can be inflected for number, gender, and state (construct state when followed by a specifier or complement, absolute state if there is no specifier or complement). Yet, they are still verbal in that they allow for verbal complements marked with the accusative particle et (Dyk 1994). Gerunds are another case of transpositional inflection: they are verbal forms with nominal properties. For instance, in John's reading the papers the gerund reading behaves externally as a noun since it assigns genitive case to John, whereas it behaves as a verb with respect to its nominal, prepositionless complement the papers.

Geert Booij

14

Thus, we see that forms that are traditionally considered as inflection may affect the external syntactic valency of a word, and are therefore category-changing. The reason why nevertheless they are usually considered inflection is that these forms are felt to belong to the paradigm of forms of a word, and can be made for each word of the relevant word class (cf. section 2.3). As far as these data are concerned, inflection and derivation still seems to differ in this respect because in the case of class-changing inflection the relevant word forms exhibit properties of both classes simultaneously, whereas in derivation it is only the new word class that determines the combinatory possibilities of the derived word. An even more complicated case is that of the possessive adjectives in Sorbían (Corbett 1987: 303). These words have the external distribution of adjectives, and are derived by means of the adjectivizing suffix -owa, nevertheless, the nominal stem is still accessible for agreement phenomena, which we would only expect if the adjectival suffix were inflectional: (4)

mojeho (GEN. SG. MASC.) mu_ ]Now]A-a (NOM. my husband's

SG. FEM.) sotra (NOM.SG.FEM.)

sister

In this example, the possessive pronoun mojeho agrees with the masculine nominal stem mu 'husband' of the adjective muowa with respect to gender and person, whereas the adjective mu owa itself agrees in gender, person and case with the head noun sotra. Corbett (1987: 305-06) therefore made the following comments on the nature of possessive adjectives such as mu owa. "Clearly, given the control possibilities of P[ossessive] A[djectíves] [...], their formation is relevant to syntax. Thus they meet the sufficient condition for being derivational, and they fully meet the condition of what is inflectional. They are therefore difficult to accommodate to the claim that morphology is divisible into two parts [i.e. inflection and derivation, GEB]"

What may be at stake here is that the nominal stem is still accessible because the derivational suffix -owa creates relational adjectives, and has no specific meaning contribution of its own except for a general 'relational' meaning. That is, it should not be seen as unambiguous evidence in favour of giving up the difference between derivation and inflection. 2.2 Obligatoriness The second criterion found in the literature is that derivation is optional, whereas inflection is obligatory. For instance, given that Latin nouns are inflected for number and case, each Latin noun must be inflected for these two categories, and has an ending indicating number and case. Whether this applies to all words and/or all languages, depends on one's analysis. For instance, the English noun book may be claimed to lack a specification for number, which is an inflectional category for English nouns. This would imply that not all inflection is obligatory. Alternatively, one may claim that each word of the relevant category must be inflected for all inflectional properties of that category. This implies that a word like book is specified as singular by means of a zero-morpheme,

The Demarcation of Inflection

15

or will receive the feature [singular] on the basis of a paradigmatic opposition, i.e. its position in a paradigm of related word forms, as in Wunderlich's (1996) minimalist morphology. Since the obligatory nature of inflection is theoiy-dependent, the criterion of obligatoriness is not always helpful as a demarcation criterion. 2.3 Paradigms A characteristic difference between inflection and derivation is that inflection is organised in terms of paradigms. Each cell in the paradigm specifies the form of a word for a particular value (property) of the relevant inflectional categories, such as number, person, tense, and case. One possible consequence of this view is the assumption of zero-markers in case there is no explicit marking for a particular inflectional property; thus a singular noun as book is given the morphological analysis book-e because bookfillsthe cell for SG. The same applies to the expression of present tense in works which can be analyzed as work-0-s (HWA-PRES-3RD PERS.

SG.).

This difference between inflection and derivation appears to be relativized by morphologists who assume zero-morphemes in derivation. Given data such as the following from Dutch: (5)

val]v 'fall' vang]v 'catch' beloof]v 'promise'

val]N 'fall' vang]v-st]N 'catch' belof]v-te]N 'promise'

We may reason that each verb has a corresponding deverbal event noun with a nominalizing marker that is expressed as -st in the case of vang, -te in the case of beloof, and as zero in the case of val. This reasoning seems to presuppose that each verb has a paradigmatic cell for a deverbal event noun. However, there is a difference with the use of zero-morphemes in inflection, because zeromorphemes are only assumed for derivational morphology if there are also non-zero morphemes for the relevant morphological category. Whereas we may assume a zero-morpheme for the English singular nouns without there being an overt counterpart, in derivational morphology at least one overt marker for the morphological category involved is usually required. This is known as the overt analogue criterion, which distinguishes derivation from inflection (cf. Sanders 1988). Related to the paradigmatic structure of inflection, we often find that there is no one-to-one correspondence between inflectional properties and their formal expression: two or more properties may be expressed by the same form, or vice versa (Matthews 1991). An inflectional property will be expressed in more than one way if the language involved has inflection classes (declensions for nouns, and conjugations for verbs); each class may have its own formal expression for a particular array of inflectional properties. Whereas in Latin mens-is 'table, DAT. PL.' the properties 'dative' and 'plural' are expressed by the suffix -is, the same properties are expressed by -ibus in the noun partibus 'part'. Inversely, in inflectional systems there might be one formal expression for two different

16

Geert Booij

arrays of properties, i.e. syncretism. For instance, the ending -ibus is both the dative pi. and ablative pi. form. In this respect, inflection clearly differsfromderivation. A further characteristic of inflectional paradigms in many languages is that more than one stem form has to be used for the formation of the set of inflectional forms. Latin uses three stem forms for each verb, one for the present tense, one for the perfect, and one for the past participle. Thus, the verb armare 'to arm' has the stem forms arma, armav, and armai, as in arma Ί arm', armav-i Ί have armed', armat-us 'armed, past participle'). The stem arma-t is also used in word formation, as in armatura 'armament'. This type of stem allomorphy is, however, not an exclusive characteristic of inflection: we also find cases where different stem forms of a base word have to be used in derivation. Latin itself is an example, as shown in the preceding paragraph. In Germanic languages many non-native words have two stem forms, ome for native derivational morphology, and another one for non-native derivation. A English word like drama has two stem forms, drama as in the plural form dramas, and dramat, as in dramat-ic (Booij 1997). Since the words of an inflectional paradigm are more closely connected to each other than derivationally related words, analogy applies more frequently within inflection. For instance, whereas Latin honos 'honor1 changed to honor because of the genitive form honor-is (from honos-is), the derived adjective honestus 'honest' kept its /s/. This is why it is called 'paradigmatic leveling'. Clearly, this term presupposes that inflectional forms of a word have stronger paradigmatic relations than words derived from the same base. Therefore, although we cannot exclude that derivationally related words may also form a sort of paradigmatic network, the forms of an inflectional paradigm clearly show a stronger relation between each other. 2.4 Generality and productivity A number of properties of inflection reflect the basic generalizations concerning the differences between inflection and derivation discussed above. First, if inflection is obligatory in the sense that each word has a paradigm the cells of which have to be filled, we expect that all words of the relevant category undergo the pertinent inflectional rules. That is, inflectional rules tend to be general (apply to all relevant words) and are productive (that is, new word-forms can be made in accordance with the rule). This is the main reason for considering certain types of class-changing morphology discussed in Section 2.1 as inflection. Productivity of inflectional patterns is certainly a universal tendency, but not without exceptions: we do find paradigmatic gaps, i.e. words for which certain inflectional forms are not available. Dutch has a number of complex verbs that only exist in the infinitive, and do not havefiniteforms, for instance bloemlezen 'to make an anthology1. French has a number of defective verbs for which not all tense forms can be formed. The verb frire 'tofry",for example, has no plural forms for the present indicative (Morin 1995).

The Demarcation of Inflection

17

Moreover, the property of generality does not always hold. In English many nouns do not have a plural form at all (courage, food, grace, March, assuredness, etc.). However, this is of course because the relevant nouns are uncountable, and therefore one might say that such words do not belong to the domain of plural inflection. A similar case is that many English adjectives do not have comparative or superlative forms (instead, one has to use more/most A). Here, we might say that the relevant domain of adjectives is defined prosodically, and that comparative formation applies without exception within that domain. Languages may also have pluralia tantum, i.e. nouns that only occur in the plural, such as Dutch Alpen 'Alps', notulen 'minutes' and hurken 'haunches'. In the latter case, we do seem to have arbitrary gaps, since, for instance, nothing would be wrong, semantically or otherwise, with the word Alp 'mountain of the Alps'. 2.5 Semantic transparency Another corollary of the more general and productive nature of inflection is that it is semantically more transparent than derivation. Whereas derived words often have a meaning that is not purely a compositional function of the meaning of its morphological constituents, this is very rarely the case with inflection. Exceptions are again formed by cases of inherent inflection such as plural nouns: the Dutch plural noun letteren 'letters' has the special meaning 'arts and humanities', and whereas English cloth means 'woven material', the plural clothes has the meaning 'garments'. Such inflectional forms exhibit the phenomenon of lexical split: the semantic relation between two formally related words is no longer transparent. It is a pervasive phenomenon in derivation, and relatively rare in inflection. The criterion of semantic regularity is also involved in the issue whether the system of conjugational classes in Hebrew (the binyanim) and other Semitic languages is a matter of inflection or of derivation. Since the different binyanim of a verbal root often have unpredictable meaning aspects, one is inclined to consider this system as derivation. On the other hand, the fact that the binyanim of a verbal root such as qatal 'to kill' form a kind of paradigm reminds us of inflection. The best interpretation appears to be that binyanim are inflectional classes, and that Hebrew derives new verbs by changing the inflectional class (binyan) of a verb. That is, transposition of conjugational class is a form of derivation (Aronoff 1994). 2.6 Psycholinguistic differences The differences between derivation and inflection outlined in the preceding sections may also have a psycholinguistic reflex in that products of derivation will more readily be stored in the mental lexicon, whereas inflectional forms, being mostly regular and formed according to productive rules, will often be made 'on the spot'. This will in particular be the case for languages with rich inflectional systems, for which it is simply impossible to store all the possible inflectional forms of a lexeme. The

18

Geert Booij

distinction between storage and rule does not completely coincide, however, with that between inflection and derivation. Irregular inflectional forms, and regular forms with a high token frequency appear to be stored, whereas regular inflectional forms with a low frequency are produced by rule (Stemberger & MacWhinney 1988, Baayen et al. 1997). On the other hand, there are very productive and regular derivational categories that can easily be extended by rule, and for which it is therefore implausible that all its members are stored in the mental lexicon. This is in particular the case for languages with agglutinating morphology like Turkish where with one root we may have millions of different word forms which cannot possibly be stored (Hankamer 1989). In another recent paper, Baayen et al. (ms.) provided another type of evidence for the difference between inflection and derivation: it is the token frequency of plural nouns that contributes to the cumulative frequency effect of the relevant stem, whereas it is only type frequency that affects cumulative frequency effects in the case of words that are derivationally related to that stem. This underscores the point made before that inflectional forms are more strongly related to each other than derivational forms. The distinction between inflection and derivation has also been investigated in studies of aphasia, with unclear conclusions. Badecker & Caramazza (1989) investigated the speech of an Italian aphatic who made many inflectional errors, but almost no derivational ones. They therefore concluded that the grammar must distinguish inflection and derivation, although, as they point out, this does not imply that inflection and derivation belong to two different components of the grammar (as in the split morphology hypothesis, cf. Section 3). On the other hand, there are also speakers with agrammatism (Broca aphatics with poor syntax and almost no function words) whose inflectional morphology is not affected, and preserved to the same extent as their derivational morphology (De Bleser & Bayer 1988). 2.7 Recursivity A consequence of the functional differences between derivation and inflection is that, whereas an inflectional process is applied only once to a word, to create a word form that fills a cell of the paradigm, derivational morphology may apply recursively, because each derivational step may add some additional meaning. For instance, in the Dutch adjective werke-loos-heids-loos 'without unemployment', the suffix -loos Svithout' occurs twice. Recursive application of derivational morphology is also found for a number of languages in the domain of evaluative morphology. For instance, we find two consecutive diminutive suffixes in Polish kote ek /kot-ek-ek/ 'dear little cat', and in Afrikaans huis-ie-tjie 'dear little house' (-/'e and -tjie are allomorphs of the diminutive suffix). Note that the two occurrences of the diminutive suffix in these examples do not have the same meaning, the more peripheral one does not express smallness, but endearment.

The Demarcation of Inflection

19

2.8 Syntactic relevance An important demarcation criterion often proposed in the literature is that inflection is that part of morphology that is relevant to syntax (e.g. Anderson 1982: 587). Particular inflectional forms of words may be required by the syntactic context, i.e. they are determined by agreement or government (rection). This is what I call 'contextual inflection' (Booij 1994). Typical examples are agreement in number and person between subject and finite verb, and the selection of particular case forms of nouns by verbs and prepositions. Note, however, that not all inflection is dependent on syntax. For instance, the number of a noun in subject position is not determined by syntactic context, but depends on whether the speaker wants to refer to a singular or to a plural entity. That is, there is also inherent inflection (e.g. number of nouns, tense, aspect, comparatives, and superlatives), which is closer to derivation than contextual inflection. This is also reflected by the fact that inherent inflection tends to be more idiosyncratic than contextual inflection (lexical split, lacking forms, forms without base words, etc., cf. Booij 1994). This difference between inherent and contextual inflection has also been observed by Kurylowicz who distinguished between inflectional categories with a primarily syntactic function such as case, and inflectional categories with a primarily semantic or autonomous function. He pointed out that number is "a semantic trait of the noun" (Kurylowicz 1964:31), and that "degrees of comparison [...] represent the autonomous inflection of the adjective. This inflection is intrinsically semantic and never assumes a special syntactic function" (Kurytowicz 1964:34). The criterion that syntactically relevant morphology is inflection is not so easy to apply in all cases. Note that derivation is also relevant to syntax in that if often determines the syntactic category and the syntactic valency of the words it creates. For instance, the Dutch prefix be- creates transitive verbs from verbs and nouns. The transitivity effect shows that Ae-prefixation is syntactically relevant. Yet, we may consider òe-prefixation derivation, because of its potential for word class transposition, and the often unpredictable meaning of the èe-verb. We meet a similar problem when we want to determine whether the formation of adverbs in -ly in English is inflection or derivation. The use of the adverb(ial form) happily in They sang happily is required by the syntactic context. This does not necessarily imply that -ly suffixation is a matter of inflection: one might also say that the syntactic context requires an adverb, and that suffixation with -ly is the morphological answer to this need, i.e. morphology creates adverbs. Similarly, the use of a than NP phrase requires the use of an adjective, as in John is bigger than Peter, but we can also use the comparative form without a //jaw-phrase. Therefore, when one uses 'relevance to syntax' as a criterion for inflection, this does not help us that much: clear cases of derivation can also be relevant syntactically, and on the other hand not all cases of what is traditionally considered inflection have a specific syntactic impact, as is the case for the plural inflection of object nouns in Germanic languages that have no object agreement, and for the semantic use of cases.

Geert Booij

20 2.9 Order of morphemes

In a complex word with both derivation and inflection, inflection is usually peripheral with respect to derivation. For instance, in the Dutch diminutive moeder-tje-s 'little mothers', the diminutive suffix -tje precedes the plural suffix -s, and a form like *moeder-s-tje is ill formed. This is one of the most important formal reasons for distinguishing between inflection and derivation: derivational suffixes are not attached to words in the concrete sense, but to stems, i.e. words minus their inflectional endings. In the Italian example given in section 2.1., the diminutive suffix -in is not attached to ragazzo "boy1, but to the stem ragazz-. The peripherality of inflection has been stated as a universal by Greenberg: (6)

Universal 28. If both the derivation and the inflection follow the root, or they both precede the root, the derivation is always between the root and the inflection (Greenberg 1963: 93)

It has been claimed that German diminutives such as Kind-er-chen 'small children' are counterexamples to the claim that inflection is always peripheral with respect to derivation, because the plural morpheme -er precedes the diminutive suffix -chen. However, this example is very idiosyncratic, and a form like Männerchen 'little man' is ill-formed. The sequence -er in this word can probably be seen as an extension of the stem of the lexeme Kind 'child'; this implies that the plurality is expressed by zero, just as is the case for all other words in -chen such as Mädchen 'girl'. As we saw in section 2.8, inherent inflection appears to share a lot of properties with derivation; this is in line with the generalization that contextual inflection tends to be peripheral with respect to inherent inflection. For instance, in Dutch finite verbs, the (contextually determined) number agreement suffix is peripheral with respect to the (inherent) tense-suffix, e.g. werk-te-n 'work-PASTPL'.

Morphologists who do not accept arigiddistinction between inflection and derivation have tried to establish principles for the ordering of affixes within a complex word. The best known proposal is that of Bybee (1985). According to her, the order of affixes is determined by the degree of relevance of an affix for the meaning of the word. Since derivational affixes such as the causative suffix have a considerable and specific effect on the meaning of the word, and thus have a higher semantic relevance, they occur close to the stem, whereas affixes for aspect, tense and the like are more peripheral: they have more general, hence vaguer meanings. Moreover, inflectional markers often do not pertain to the meaning of the complex word itself, but express the relation of a word to situation and context. Tense, for example, expresses the time relation between the event or situation expressed by the verb and the moment of speaking, and case expresses the relation of a noun to other parts of the sentence. Bybee (1985: 35) established the following tendencies in the ordering of verbal inflectional markers with respect to the stem: (7)

stem-aspect-tense-mood-number/person

The Demarcation of Inflection

21

As already mentioned, contextual inflection tends to be peripheral with respect to inherent inflection. To put it differently, syntactically relevant morphemes tend to occur at the periphery, in order to be visible for the syntax (Williams 1981). For instance, as Greenberg pointed out, there is a strong universal tendency for case affixes to be peripheral with respect to number affixes. This is in line with the observation that inherent inflection is more like derivation than contextual inflection: (8)

Universal 39. Where morphemes of both number and case are present and both follow or precede the noun base, the expression of number almost always comes between the noun base and the expression of case (Greenberg 1963: 95)

In sum, the following universal tendency appears to occur: contextual inflection is peripheral with respect to inherent inflection, and inherent inflection is peripheral with respect to derivation. This generalization presupposes, and thus supports, the inflection-derivation distinction. 2.10 Conclusions In a number of respects, the differences between inflection and derivation appeared to be of a gradual nature. In particular, inherent inflection appeared to occupy an intermediate position, in between derivation and contextual inflection. Nevertheless, there is a number of phenomena that support a distinction between inflection and derivation: (i) (ii) (iii)

the role of the paradigm: the network of paradigmatic relations between related words is much tighter in the case of inflection; generalizations about affix order require the inflection-derivation distinction; inflection typically exhibits a complex relation between form and meaning (inflectional classes and paradigms).

Therefore, it is justified to assume that inflection forms a special subcomponent of the morphological part of the grammar, with specific formal properties.

3. Split morphology? The differences between inflection and derivation discussed above have led some linguists to assume an organizational model of the grammar in which there is a strict separation of derivation and inflection. Derivation is located in a pre-syntactic morphological component, and functions to enrich the lexicon. Inflection, on the other hand, is located in a post-syntactic component of morphological spell out rules, since the correct inflectional form of a word depends on its position in syntactic structure. This model is called the model of 'split morphology' (Perlmutter 1988), and is also advocated in Anderson (1982; 1992). An additional reason for this separation is that, whereas in

22

Geert Booij

derivational morphology there is usually a one-to-one relation between form and meaning, this is different for inflection, as emplified above in Section 2.3. Therefore, inflectional rules are seen as realizational rules or spell out rules that specify the formal expression of each array of inflectional properties. A variant of this organizational model is proposed in Beard (1994): derivation is pre-syntactic as far as its semantic and syntactic aspect is concerned, inflection is post-syntactical. Both derivational properties (e.g. Agent, Action) and inflectional ones are spelled out by the same realizational component. The reason for this conflation of the formal expression of derivational and inflectional categories is that derivation and inflection often make use of the same afBxes. For instance, the Dutch suffix -s expresses both '3d pers. sg. present tense' for verbs, 'plural' for nouns, and deadjectival nominalization as in goed-s 'the good', and English -er is both the comparative and the deverbal agentive suffix. It should be realized, however, that the fact that the choice of a particular inflectional form is determined by syntax does not necessarily imply that inflection is post-syntactic. One can also assume that inflection applies pre-syntactically, and that rules such as subject-verb agreement only have a checking function: they check whether the relevant morphosyntactic properties of two words in a specific syntactic construction are compatible. For instance, since the English nouns people and books are marked as plural, the second due to an inflectional process, they both require a plural finite verb if they are the head of a subject noun phrase. That is, the presence of a singular finite verb will qualify such a sentence as ungrammatical. An additional argument adduced in favour of the split morphology hypothesis is that it predicts that inflection does not feed derivation, i.e. that we should never find inflectional morphemes inside derivational morphemes. Thus, this model directly accounts for the peripherality of inflection with respect to derivation. However, as we will see below, inflection is not always peripheral to derivation, and therefore this argument is not valid. Another organizational variant in which derivation and inflection are not completely separated, but distinguished within the lexical component, is the hypothesis of level-ordered morphology (Kiparsky 1985). In this model, a variant of strong lexicalism, morphological processes are assigned to different, ordered strata or levels in the lexicon. The idea then is that derivation is located at an earlier level (or earlier levels, if more than one derivation level is assumed) than (regular) inflection. This ordering predicts that inflection cannot feed derivation. On the other hand, such an organizational model maintains the possibility that derivational and inflectional processes induce the same phonological processes, which is often, but not always, the case. This variant faces the same problems as the split morphology hypothesis, unless we allow for loops between the different levels. The basic problem for the split morphology hypothesis is that inflection sometimes does feed word formation. This is particularly clear in the case of compounding: plural nouns may function as the first or second constituent of compounds:

The Demarcation ofInflection

(9)

23

Italian: lava-piatti 'lit. wash-dishes, dish washer" portalettere 'lit. carry letters, postman' Dutch: steden-raad 'cities council' huizen-rij 'row of houses'

Plural nouns also occur in Dutch derived word with the collective suffix -dom such as scholier-endom 'set of pupils'. Nouns with case endings also occur inside compounds; the generalization appears to be that only inherent case, i.e. case selected on the basis of semantics, and not on the basis of syntactic structure is involved. For instance, we find the following compounds with word-internal case suffixes in Finnish (Booij 1994): (10) maa-lta-pako 'country-ABLATTVE-flight, rural depopulation'

As already shown, inflection does not only feed compounding, but also derivation. In most European languages past participles feed de-adjectival word formation, as in Dutch gevreesd-heid 'feared-ness'. In Breton, the diminutive suffix is not only attached to singular nouns, but also to plural nouns such as bag "boat' andpaotr "boy1 (Stump 1990): (11) sg bag paotr

dim. bag-ig paotrig

plu bag-où paotr-ed

plu dim bag-où-ig-où paotr-ed-ig-où

Breton plural nouns also feed two other derivational processes, the formation of denominal verbs and of denominal adjectives: (12) aval'apple' delienn 'leaf

aval-où Ή.' delioù "PL'

aval-où-a'to look for apples' deliaoù-ek 'full of leaves'

As already pointed out above, infinitives often behave as nouns with respect to compounding. In Dutch verbal compounding is unproductive, unlike nominal compounding. Yet, compounding with infinitives as head or modifier is productive: (13) school-zwemm-en 'school-swim-lNF, school swimming'; boek-bind-en "book-bind-iNF, book binding'

lijd-en-s-verhaal 'suffer-iNF-story, Passion'; et-en-s-tijd 'eat-iNF-time, eating time' In sum, both the split morphology hypothesis and the level ordering hypothesis have problems with the types of interaction of inflection and word formation presented above. Therefore, one should assume one morphological module in which both derivation and inflection are accounted for. In this

Geert Booij

24

module, fully inflected forms are generated by the inflectional subsystem of that module. These inflectional forms are then selected by the syntax. Some types of inflected forms, instances of inherent inflection may feed the word formational part of the morphological module. This conception of the position of morphology in the grammar is corroborated by some observations about allomorphy, which will now be dealt with. 3.1 Stem allomorphy Derivational processes may make use of stem allomorphs that are formally identical to specific inflectional forms. This implies that such inflectional forms have to be available for derivation, another problem for the separation of inflection and derivation. An example of stem allomorphy in which inflected forms play a role is the role of the feminine form of French adjectives in derivation: the adverbial suffix -ment is always added to the feminine form of the adjective without a feminine meaning being implied (with four exceptions: brillament •brilliantly', savamment 'learnedly', èloquement 'eloquently' and apparemment 'apparently1): (14) masc. faux 'false' lent 'slow' heureux 'happy' certain 'id.'

fem. fausse lente heureuse certaine

adverb faussement lentement heureusement certainement

In these cases one could still defend an analysis that does not refer to the notion 'fem. inflected adjective' by assuming a latent stem-final consonant which then surfaces before the suffix -ement, just as it surfaces before the suffixes -e and -esse. This analysis, however, fails to explain why, if an adjective has a suppletive or irregular feminine form, it is this suppletive form that shows up in the adverb in -ment: masc. beau "beautiful' blanc Svhite' fou 'stupid' sec 'dry' vieux 'old'

fem. belle blanche folle sèche vieille

adverb bellement blanchement follement sèchement veillement

These 'feminine' allomorphs also show up in other types of deadjectival words. Therefore, the only generalizing analysis is that in which a stem allomorph that is formally identical to the feminine form of the adjective is the formal basis for derivation. In other words, we use a 'rule of referral' that specifies the form of the stem of adverbs by referring to a specific inflected form of the adjective. The same allomorphy pattern is found in related languages such as Spanish, as in clar-a-mente 'clearly', with the feminine suffix -a. As Rainer (1993: 267) pointed out, this feminine suffix reflects

The Demarcation of Inflection

25

the historical origin of the suffix -mente which is a case form of Latin mens, mentis 'mind' a feminine noun that requires the modifying adjective to agree in gender, as in clara mente 'with a clear mind'. These other Romance languages therefore also require such a rule of referral that predicts the right allomorph of an adjective on the basis of the feminine form of that adjective. Another relevant case is that of the five verbs of Dutch which are special in that they have an infinitive form in -n instead of the regular -en. (16) doen'todo' gaan 'to go' slaan 'to hit'

staan'to stand' zien 'to see'

These verbs have a second stem which is formally identical to the (irregular) infinitival form, and functions as the verbal base for the formation of the present participles (e.g. doen-d 'doing') and for derivation. (17) be-doen-ing'to-do' vol-doen-ing 'satisfaction' aan-doen-ing 'affliction'

boete-doen-ing 'penance' voor-zien-ing 'provision' her-zien-ing 'revision'

Crucially, we cannot say that the relevant derived words are derived from the infinitive, because the Dutch suffix -ing is a deverbal suffix, whereas the infinitive behaves as a noun in word formation (Booij 1989). That is, we have to say that these verbs have stem allomorphs to be used in derivation that is phonologically identical to the infinitival form of these verbs. Thus, a proper model of the morphological module of a grammar has to allow for derivation to have access to specific inflectional forms because the correct stem allomorphs for certain derivational processes are to be computed on the basis of inflectional forms.

4. Conclusions This article did not present a formal model of inflection. Instead, it focused on the demarcation of inflection and derivation, which is often presupposed in such formal inflectional models. We saw that inflection and derivation do differ in a number of ways, which justifies the assumption of a specific inflectional submodule with particular formal properties. However, we also saw that inflection interacts with derivation in a number of ways, and therefore, inflection and derivation cannot be separated in two different components of the grammar, as the split morphologists would have it.

26

Geert Booij

References Anderson Stephen R. (1982) Where's Morphology?. Linguistic Inquiry 13, 571-612 - (1992) A-Morphous Morphology. Cambridge: Cambridge University Press. Aronoff, Marie (1994) Morphology by Itself. Cambridge Mass. : ΜΓΓ Press Baayen, Harald, Cristina Burani & Robert Schreuder (1997) Effects of Semantic Markedness in the Processing of Regular Nominal Singulars and Plurals in Italian. In Geert Booij & Jaap van Marie (eds.) Yearbook of Morphology 1996. Dordrecht: Kluwer, 13-41. Baayen, Harald, Rochelle Lieber and Robert Schreuder (ms.) The Morphology of Simplex Nouns. Max Plank Institut fur Psycholinguistik, Nijmegen. Badecker, William & Alfonso Caramazzo (1989) A Lexical Distinction between Inflection and Derivation. Linguistic Inquiry 20, 108-116. Beard, Robert (1994) Lexeme-Morpheme-Base-Morphology. Albany: State of New Yoik University Press. Bleser, Ria de & Josef Bayer (1988 On the Role of Inflectional Morphology in Agrammatism. In: Hammond & Noonan (eds.), 45-70. Booij, Geert (1989) Complex Verbs and the Theory of Level Ordering. In: Booij, Geert & Jaap Van Marie (eds.) Yearbook of Morphology 1989. Dordrecht: Foris, 21-30. - (1994) Against Split Morphology. In: Booij, Geert & Jaap Van Marie (eds.) Yearbook of Morphology 1993. Dordrecht: Kluwer, 27-49. - (1997) Autonomous Morphology and Paradigmatic Relations. In: Booij, Geert & Jaap Van Marie (eds.) Yearbook of Morphology 1996. Dordrecht: Kluwer. Bybee, Joan (1985) Morphology. The Relation between Meaning and Form. Amsterdan: John Benjamins. Corbett, Greville (1987) The Morphology/Syntax Interface: Evidence from Possessive Adjectives in Slavonic. Language 63, 299-345. Dressier, Wolfgang U. (1989) Prototypical Differences between Inflection and Derivation. Zeitschrift für Phonetik, Sprachwissenschaft und Kommunikationsforschung 42, 3-10. Dyk, Janet (1994 Participles in Biblical Hebrew. A Computer-assisted Study of Old Testament Hebrew. Amsterdam: Free University Press. Greenberg, Joseph H. (1963) Some Universale of Grammar, with Particular Reference to the Order of Meaningful Elements. In: Id. (ed.) Universals of Language. Cambridge Mass.: ΜΓΓ Press, 73-113. Hammond, Michael & Michael Noonan (eds.) (1988 Theoretical Morphology. San Diego etc.: Academic Press. Hankamer, Jorge (1989) Morphological Parsing and the Lexicon. In: Marslen-Wilson, William (ed.) Lexical Representation and Process. Cambridge Mass/London, U.K.: ΜΓΓ Press, 392-408. Haspelmath, Martin (1996), "Category-changing Inflection". In: Booij, Geert & Jaap Van Marie (eds.) Yearbook of Morphology 1995. Dordrecht: Kluwer, 43-66. Kiparsky, Paul (1985) Some Consequences of Lexical Phonology. Phonology Yearbook 2, 85-138. Kuiylowicz, Jerzy (1964) The Inflectional Categories of Indo-European. Heidelberg: Carl Winter Universitätsverlag. Matthews, Peter H. (19912 )Morphology. Cambridge: Cambridge University Press [1974i], Morin, Yves-Charles (1995) De l'acquisition de la morphologie: le cas des verbes morphologiquement défectifs du français. In: Bat-Zeev Shyldkrot, Hava & Lucien Kupferman (eds.) Tendences Récentes en Linguistique Française et Générale. Amsterdam: John Benjamins, 295-310. Perlmutter, David (1988) The Split Morphology Hypothesis: Evidence from Yiddish. In: Hammond & Noonan (eds.), 79-100. Plank, Frans (1994) Inflection and Derivation. In: Asher, R. E. (ed.) The Encyclopedia of Language and Linguistics, Vol 3. Oxford: Pergamon Press, 1671-1678 Rainer, Franz (1993) Spanische Wortbildung. Tübingen: Niemeyer. Sanders, Gerald (1988) Zero Derivation and the Overt Analogue Criterion. In: Hammond & Noonan (eds.), 155-175. Scalise, Sergio (1986) Inflection and Derivation. Linguistics 22, 561-581 Stemberger, Joseph Paul & Brian MacWhinney (1988 Are Inflected Forms Stored in the Lexicon?. In: Hammond & Noonan (eds.), 101-116. Stump, Gregory T. (1990) Breton Inflection and the Split Morphology Hypothesis, In: Hendrick, Randall (ed.), The Syntax of the Modern Celtic Languages. San Diego: Academic Press (=Syntax and Semantics, Vol. 23), 97-119.

The Demarcation of Inflection

27

Williams, Edwin (1981) On the Notions Lexically Related' and Mead of a Word'. Linguistic Inquiry 12,245-274. Wunderlich, Dieter (1996) Minimalist morphology: the role of paradigms. In: Geert Booij & Jaap van Marie (eds.) Yearbook of Morphology 1995. Dordrecht: Kluwer, 93-114.

Philippe Ségéral

(Université Paris 7) and Tobias Scheer (Université Paris 8)

A Generalized Theory of Ablaut: the Case of Modern German Strong Verbs 1. Introduction1 In this article, we discuss non-arbitrary vowel colour alternations known as Ablaut or Apophony that lack any contextual conditioning and are exploited for the purpose of grammatical opposition. Some examples of such context-free alternations in various languages are given below: (1)

Berber Somali Hebrew Ge'ez CLArab. German English

fa6 (imperat.2S) yiqiin (past 3MS) äaamar (qal-pf 3MS) xägiir (adj SM) katab-(pf I act) helf-a (près IS) sir) (près IS)

i-fu0 (prêt 3MS) yaqaanfn-aal (près 3MS) yi-Smor (qal-ipf 3 MS) xägaar (adj SF) ya-ktub- (ipf I act) half (prêt IS) ssi) (prêt IS)

má oqoón (neg.past)

ga-holf-sn (past part) SAI) (past part)

The question raised by these data concerns the status of the alternating vowel: are its different manifestations lexically determined, or are they predictable? In other words, are there as many lexical entries as there are words, or are the different grammatical forms of the cognate words related by a derivation? In the remainder of the paper, we shall argue for the latter option. The existence of a derivational link between two apophonically connected forms was most succinctly formalized by Kurylowicz (1956, 1968). In his approach, any pair of apophonically related forms A and Β is of derivational nature to the extent that Β (forme fondée) is the nonarbitrary output of a derivational operation (le fondement A~>B) performed on A (forme de fondation). This process obeys two principles: 1) proportionality, that is, for any pair displaying the same apophonic operation, the grammatical relation is constant; 2) polarization, that is, the distance between A and Β is maximal. Even though this perspective does establish a derivational relation between the alternating vowels, it does not state any more general feature of Apophony. Particularly, any two vowels under these provisions may contract an apophonic relation. Moreover, the vowel colour of the derived form is not a priori predictable from the colour of the base-form. From such a perspective, the pairs of alternating vowels found in different languages are lexically determined. They have nothing in common but the derivational relation they contract. Proposing a much more constrained approach to the derivational character of apophonic morphology, we claim that apophonic systems obey a precise regularity defining the substantial 1

This paper is a revised and augmented version of Ségéral and Scheer (1994). We would like to thank Ray Fabri, Jean Lowenstamm, Utz Maas, Damaris Niibling and Albert Ortmann for helpful comments and criticism.

29

A Generalized Theory of Ablaut: the Case of Modern German Strong Verbs

(vowel colour) and the structural (directionality) properties of the alternations.2 The regularity we find evidence for is: 1) applicative, that is, the vowel of the base-form being known, the vowel of the derived form is predictable; only the vowel of the base-form is lexically recorded; 2) crosslinguistically unique. Under these provisions, only a very small subset of logically possible combinations of two vowels is predicted to be able to contract an apophonic relation. In section two, we present this subset as evidenced in Guerssel and Lowenstamm's (1994) analysis of the apophonic system of Classical Arabic. Section three adduces more evidence for the cross-linguistic generalization of Guerssel and Lowenstamm (1994). Section four then discusses the question of how the basic character of a given grammatical category is established. From section five on, we show that the analysis of apophony in Classical Arabic also accounts for one of the most intriguing apophonic systems, i.e. the Ablaut found in German strong verbs. We demonstrate that one form of the verb (present tense), together with the universal apophonic mechanism, contains all the information necessary in order to predict the other verbal forms (preterite, past participle). The analysis of German strong verbs is carried out on a strictly synchronic level.

2. The Apophonic Path Let us consider one of the major cases of context-free vocalic alternations that carry grammatical information, i.e. the verbal system of Classical Arabic (CA). The vocalic pattern of the subsystem of Measure I Active verbal forms is shown below (for other Measures, cf. Guerssel and Lowenstamm 1994):

ktb Drb lbs kbr

perfective 3MS katob-a Darab-a labis-a kabur-a

imperfective 3MS ya-ktob-u ya-Dr/b-u ya-lbas-u ya-kbur-u

write beat clothe grow

alternation pf - ipf a -u a-i i-a u-u

alternations ipf u-a i-a a-i u-u

As can be seen, the second vowel of the template alternates in perfective and imperfective forms of the same verb. These alternations do not seem to be predictable because they are ambiguous: 1) one vowel is always the input of several outputs (a phenomenon referred to as opacity by Guerssel and Lowenstamm 1994), 2) [i] and [a] are always the input and output of each other (polarity). Two derivational directions are conceivable: pf —> ipf or ipf —> pf (3) shows that opacity and polarity obtain in any event.

One consequence of this approach is that, unlike in Kurylowicz's system, apophony exclusively concerns vowel-colours, never vowel-length. Contrasts in length are not apophonic.

Philippe Ségéral and Tobias Scheer

30 (3)

hypothesis 1 pf I > a

u

hypothesis 2 ipf -» pf

ipf

> u

Now if there is any point in claiming that a derivational relation holds between both verbal forms, the minimal request is for the derivation to be applicative: that is, for a given input, there is one and only one output. Assuming the perfective forms to be basic,3 a problem arises with the vowel [a] in the perfective of kataba and Daraba. in contrast to pf [i], which gives ipf [a] (and only ipf [a]), and pf [u], which results in ipf [u] (and only in ipf [u]), pf [a] sometimes surfaces as ipf [i], sometimes as ipf M Hence, if the relation linking the two sets of vowels (pf and ipf) is to be applicative, the phonetic object [a] appearing on the surface must have a twofold underlying identity, i.e. /a/ and something else. Guerssel and Lowenstamm propose that the second origin offa] is Issi, the null set. Indeed, this option seems to be the only possible since any choice among the vowels appearing on the surface would lead into the same dead end of opacity and polarity. Thus, the postulated fourth object must be different from the three observable vowels [i,a,u]. Moreover, it does not ever appear on the surface, lai is the only object displaying these characteristics. (4) shows that assuming four underlying vowels (/0,i,a,u/) instead of three (/i,a,u/) for Classical Arabic, the ablauting system turns out to be applicative and learnable since one input has one and only one output. Thus, a learner of CA posits underlying lai for a verb whose ipf form is [i], whereas /a/ is fixed for verbs with imperfectives in [u]. The [a] appearing on the surface in perfective forms at the place of lai is a propagation from the active-marker /A/ in the template's first vocalic position. That is, the [a] of DArab- is a copy of the active /A/, whereas the [a] in kAtab- is the true realization of the lexically present /a/: (5) shows the different underlying identities of [Darab] and [katab]. (4)

pf 0

ipf

ΐΓ^^ί

X

as^^a

pf ipf 0 ==> i

pf ipf Daraba ya-Dribu

I

labisa

a

==> a

==> u

"

ya-lbasu

kataba ya-ktubu

(5)

A I

A

1

CV CV C V

D

ιι r

b

a I ι cv cv

cv

I

I

I

k

t

b

/kAtab/

u > u representation u ==> u of thekabura ID A of r 0 the b/ apophonic derivation A comprehensive lexical ya-kbuni inputs and the outputs can be stated in terms of the so-called Apophonic Path.

3

Cf. Guerssel and Lowenstamm (1994) for detailed discussion. This assumption is implicit in traditional grammars.

A Generalized Theory of Ablaut: the Case of Modern German Strong Verbs

(6)

31

Apophonic Path 0 = > i= > a= > u = > u

The Apophonic Path is strictly applicative and, moreover, it restricts the possible pairs of vowels that can be said to contract an apophonic relation in two ways: 1) substantially, that is, in terms of vowel colour (e.g., i and u can never be apophonically related), and 2) structurally, that is, with regard to the directionality of the derivation (if, say, / and a can contract an apophonic relation as to their colour, only input /' = > output a is possible, a = > i not being a possible apophonic derivation).

3. Universality of the Apophonic Path Given the above analysis of CA, the question arises whether the Apophonic Path merely encodes a language specific property or whether it also applies to other languages. In this section, we show that the pattern evidenced by Guerssel and Lowenstamm (1994) for CA controls context-free vocalic alternations in a much wider range of languages. Let us consider a few examples from other languages that display the relevant vocalic alternations.'' KABYLE BERBER - The table below gives characteristic patterns from Kabyle Berber (henceforth KB), an Afro-Asiatic but non-Semitic language (data and analysis from Bendjaballah 1995).5 (7)

alternation β - i6 i-a a-u u -u

imperfective xöam χθίτ nnaK öuß

perfective ya-xöim (neg) ys-xGar ya-nnuK i-öuß

'work' 'choose' 'fight' Tje tired'

It is clear that KB instantiates exactly the alternations found in CA. Although not every verb shows alternating vowels in KB, it is striking that all substantial (no correspondences such as i-u, 0-a and so forth) and structural (no directionalities such as a=>i,

u=>a)

properties of the alternations

found in KB are strictly identical with those of CA. Space restrictions prevent us from presenting the apophonic analysis of Ge'ez and Akkadian developed in Ségéral (1995,1996) and Ségéral (forth), respectively. Other Semitic languages like Hebrew and Syriac also exhibit alternations that possibly fall under the scope of apophony (e.g. Heb pf3MS kaabed, kaatab vs. ipßMS yi-xbad, yi-xtob "be heavy, write", Synac pf3MS lobc§, qsTal vs. ipf 3MS nc-lbâS, ncqTol "cloths, kill". Imperfective forms (represented by the imperative here) are termed Aorist in traditional Berber grammar, perfective ones Preterite. Cf. Bendjaballah (1995) for discussion. The underlying identity of [a] is lai. The perfective [i] can be observed in negative forms only for templatic reasons (cf. Bendjaballah 1995).

Philippe Ségéral and Tobias Scheer

32

Moreover, whenever two stem-vowels exhibit alternations between imperfective and perfective forms, both of them match precisely the ones observed in CA: (8)

alternation ai-ua ui-ua

apoph. derivation a=>u, i=>a u=>u, i=>a

imperf. naöi 6run

perf. i-nuöa i-ônira

'search' 'need'

The complex vocalic alternations of KB turn out to be systematic rather than lexical accidents on the assumption that they are determined by the Apophonic Path. Thus, the apophonic instrument developed when dealing with CA makes the KB alternations transparent, and guarantees their legibility and learnability. SOMALI - The existence of Ablaut phenomena has long been recognized in Cushitic languages (Hetzron 1969, Cohen 1972, 1974, Andrzejewski 1975, Appleyard 1992, Voigt 1985). Such phenomena are observable mainly in the verbal morphology: in Awngi, Beja, Afar, Rendille, Somali, among others, there is a small set of verbs (less than ten), seemingly the remnants of a larger set, which display 1) prefixed personal morphemes, 2) radical vowel alternations related to aspect or tense. Let us consider in more detail evidence from the verbal morphology of standard Somali, a main representative of Cushitic. In Somali, four verbs belong to the class defined above:7 yiil "be in a place [inanimate subj.]', yimi(d) 'come', yidhi [h] 'say', yiqiin 'know, recognize'. The most salient alternation concerns the so-called "general past" and "general present" (affirmative) forms:8 (9)

PAST 3mS PRES 3mS

-i-a-

yiil yaal [-laa]

yimi [d] yamaad-daa

yidhi yidhaah-daa

yiqiin yaqaan [-naa]

The Ablaut in (9) is a true instantiation of the ζ = > a derivation predicted by the Apophonic Path. Beside general past and present, there is a third major grammatical form in Somali usually referred to as "infinitive", on which several other "tenses" are built.9 In the case of yiqiin for example, the infinitive is aqoón (-in) or oqoón (-in). The same vowel colour o also appears in the infinitives ofyiil (inf oól) and yidhi (inf. odháríf0 (but not in yimi, inf. imán). A two-step derivation can thus be

inferred: past = > present = > infinitive, corresponding to i => a => o. At the present stage of our demonstration, we cannot account for the observed o occurring instead of the expected u. The following sections show that this o is /u/ underlyingly, which is lowered for independent reasons. Thus, the apophonic system underlying the small set of Somali 7 8

9

10

We leave aside a fifth verb belonging to this class, yahay "be", which is highly irregular. We use the terms "present" and "past" that stem from traditional Somali grammars. In fact, the categories referred to as temporal also have strong aspectual features. The "infinitive" of infixed verbs is used in compound forms such as the future, progressives (past and pres.), negatives (past, past progr., pres. progr., imperative) and conjunctives (pres. and pres. progr.). The case of the imperative is complex. For a detailed analysis see Ségéral and Scheer (1997). We have no explanation for the unexpected radical vowel -á- in odhán. Nevertheless, the crucial fact here is the presence of the o.

A Generalized Theory ofAblaut: the Case ofModern German Strong Verbs

33

prefixed verbs can be read through the same Apophonic Path 0=>i=>a=>u=>u

that proved

successful in Semitic and Berber. Moreover, the Apophonic Path can shed new light on all the remaining verbs in Somali, which can be referred to as "weak". The regular verbs do not display any radical Ablaut. Grammatical morphemes are suffixed to their non-Ablauting stem as shown below for the verb keen 'carry1: (10) 2S 3MS

1 - PAST 2 - PRES 3 - PRES Progr 4 - PRES NEG 5 - PRES Progr NEG keen-t-ay keen-t-aa keén-ay-s-aa mákeen-t-ó má keén-ay-s-ó keen-ay keen-aa keén-ay-aa má keen-ó má keén-ay-ó

6 - CONJ PRES ínkeen-t-ó ín keen-ó

As can be seen, the personal morphemes are 3MS -0- vs. 2S -t- (note that t~>s /y

); the infixed

morpheme -ay- denotes progressive aspect. In word-final position, three different morphemes occur: -ay (1 in (10)), -aa (2,3 in (10)) and -o (4,5,6 in (10)). It is generally admitted (e.g. Cohen 1972:57ff, Appleyard 1992:1270) that these morphemes are reduced forms of a former auxiliary be exhibiting an ablauting stem. The apophonic path suggests an unexpected interpretation: all three forms can be analyzed as ablauting variants of a unique morpheme of the shape [-a + apophonic vowel-]: (11)

1-a-y I I

2,3-a-a I ==>

a

4,5,6-0 I ==>

u

In the third form (short -o), the lowering of the /u/ is due to merging with the preceding -a-. Unlike in the other forms, this -a- therefore is absent from the surface (cf Voigt 1985:102, Ségéral and Scheer 1997).11 The apophonic interpretation of the three allomorphs predicts the following derivational ranking of the three associated categories: [PAST] = > [PRES] = > [NEG/PROGR/CONJ], This is a striking result since the same organization of grammatical categories comes out of the apophonic analysis of the previously discussed prefixed verbs yiil, yimi, yidhi, yiqiin. The concordance of prefixed and non-prefixed verbs as to their formal and derivational features can be demonstrated only if the Apophonic Path is assumed. By these means, apophonic theory allows for a cohesive and comprehensive interpretation of the whole verbal morphology of standard Somali. Moreover, the analysis of the verbal systems of other Cushitic languages (e.g. Oromo, Afar, Beja, Gawwada) carried out in Ségéral and Scheer (1997) shows that apophonic theory is a valuable formal guide through morphology: based on purely formal and phonological grounds, it is able to reveal a coherent picture of the exact derivational relations holding within the plethoric forms of the Cushitic verb.

11

In view of the other suffixes, a long [-00] could be expected. The observed short [-0] resulting from the merging of -a- and the apophonically delivered -« is to be related to the absence of word-final long [00] (and [uu]) in Somali.

Philippe Ségéral and Tobias Scheer

34

If it is trae that all of these non-arbitrary vocalic alternations can be accounted for by means of one principle, i.e. the Apophonic Path, then this principle must be deeply anchored in the grammar. We would like to make the maximal claim here that the Apophonic Path is part of Universal Grammar (UG). Hence, any speaker of any language is equipped to interpret grammatical relations expressed by apophony correctly. In the remaining sections, we extensively discuss the vocalic alternations of Modern German for two reasons. First, this system is reputed for its extreme complexity and generally viewed as being entirely lexicalized. Second, German does not belong to the Afro-Asiatic family. Apart from the inherent challenge it poses, the German system provides an excellent test for our claim that the apophonic regularity evidenced in CA has universal value. However, before turning to the German case, let us briefly discuss some issues related to the directionality of the derivation.

4. Which way do apophonic derivations go? The Apophonic Path is a mere formal instrument that can possibly serve to express any functional link between two or more related categories. It does not say anything about the purpose grammar exploits it for. The examples from the discussed languages, as well as other work that has been carried out so far, point to an apophonic link between the following grammatical categories: (12)

language a. Somali b. Ge'ez c. German d. Cl.Arabic e. Kab. Berber

basic category past pf,sg,active,mase present nucleus perfective imperfective

derived category(ies) près,neg,conj,opt etc. ipf.pl, passive, fern prêt, past part, onset imperfective perfective

Ségéral and Scheer (1997) Ségéral (1995, 1996) see below Chekayri and Scheer (1996) Guerssel and Lowenstamm (1994) Bendjaballah (1995)

The above grammatical categories are related, as we claim, by an apophonic derivation. The notion of derivation implies that one category is basic and serves as input for some operation. Hence, the basic and derived status of the categories involved should be shown independently of any apophonic consideration. Otherwise, the argumentation meets the risk of circularity: a given category cannot be said to be basic because the Apophonic Path requires it to be basic. Rather, the Apophonic Path predicts which category in a given language is basic and which is derived. The above table (12) displays the predictions for the languages which have been discussed above. Independent evidence for the correctness of the claims entailed by apophony should be available from particular languages or from general considerations. The claims that singular is basic while plural is derived, that past participle is derived from the preterite and the preterite from the present, and that the Onset is derived from the Nucleus seem

A Generalized Theory of Ablaut: the Case of Modern German Strong Verbs

35

quite natural and intuitive. Within the languages discussed here, they are traditionally assumed to hold anyway.12 However, a contradiction seems to arise when comparing Classical Arabic (CA) and Kabyle Berber (KB): apophonically, the perfective forms are basic in the former, whereas imperfective forms are claimed to be the derivational input in the latter. The key to this apparent paradox is provided by morphology. Comparing CA and KB, it appears that if both languages are opposed as to the aspectual derivation, they perfectly agree with respect to their morphological pattern. Indeed, in both CA and KB the morphologically non-complex form, i.e. the one without prefix (the perfective in CA, the imperfective in KB), is the input of the derivation. Hence, morphological complexity is an independent indicator of the basic character of a given grammatical form. This morphological criterion also holds for Ge'ez (prefixed plurals), and partly for German (prefixed past participles). The irrelevance of conceptual issues in the apophonic derivation confirms the status of the apophonic path as a formal instrument making no claim as to the way it may be used. The question concerning the basic character of perfectiveness vs. imperfectiveness across languages certainly remains open. Singular seems to be more basic than plural, but there could well be a language where plural forms are basic and singulars derived. Such a language could use apophony to run this derivation without any problems. Still, though, there would be independent criteria suggesting that the plural forms are basic. As we have seen, one possible way of showing the basic character of a form is by looking for evidence from morphology. In all of the languages discussed here, grammarians regard the forms that apophony predicts to be the input of the derivation as basic, be it implicitly or explicitly. Yet, the strongest argument for the correctness of what apophony tells us about the basic character of certain forms comes from its cross-linguistic validity. If context-free vocalic alternations in various genetically unrelated languages are not controlled by a single mechanism, they must be random. Nobody would expect them to display a cross-linguistic regularity. However, if a general theory can show that there is such a regularity, and if it can formalize what is common to all of these alternations, then its implications must be taken seriously. In the case of the directionality of derivation, this means that the apophonic analysis points to a certain form, identifying it as basic. This is the only way in which the system of vocalic alternations of a particular language can be connected with all the other apophonic systems. Hence, if there is any cross-linguistic regularity, the directionality of the derivation must be the one predicted by apophony. On the basis of the maximal hypothesis that there is only one apophonic regularity and that it controls all systems of context-free vocalic alternations, we shall explore the situation of one of the systems that represents the highest challenge for our theory. The three forms of German strong verbs (present, preterite, past participle) are reputed to be entirely lexicalized and by no means derivable one from each other. In the remainder of the paper, we shall carry out a strictly synchronic

12

For a specific argument from syntactic Binding theory that regards the directionality present ==> preterite ==> past participle, cf. Postma (1996).

Philippe Ségéral and Tobias Scheer

36

analysis showing that the German system is controlled by a derivational mechanism that supposes lexical information for one form only, and that this mechanism is identical to the one operating in CA, KB and the other languages mentioned.

5. Apophony in German strong verbs 5.1 German strong verbs

The stem-vowel o f so-called strong verbs in New-High-German (NHG) alternates in different morphological categories. Strong verbs are thus opposed to weak verbs, which display no alternation.

(13) strong verb 'sing' weak verb 'say'

present lsg sing-e sag-e

preterite lsg sang sag-t-e

past participle ge-sung-en ge-sag-t

i => a => u a—a—a

In (14) w e give an exhaustive list o f strong verbs. 13 Oil

6

2 1 I a 1

3

4

ai

1

5 I

b

2

ai

ii

ii

II a l 3 2 4 3 5

ii yy au

oo oo oo

bleiben, gedeihen, leihen, meiden, preisen, reiben, scheiden, scheinen, schreiben, schreien, schweigen, speien, steigen, treiben, weisen, zeihen oo biegen, bieten, erkiesen, fliegen, fliehen, frieren, schieben, stieben, verlieren, wiegen, ziehen 00 logen, küren, trügen oo saugen, schnauben

sich befleißen, beißen, bleichen, gleichen, gleiten, greifen, kneifen, kreischen, leiden, pfeifen, reißen, reiten, scheißen, schleichen, schleifen, schleißen, schmeißen, schneiden, schreiten, spleißen, streichen, streiten, weichen

7 23

16 11 3 2

4 6

au

3

3

saufen

1

b

7

ii

3

3

fließen, genießen, gießen, kriechen, riechen, schießen, schließen, sieden, sprießen, triefen, verdrießen

11

Illal

8

I

a

18

2 9

I

a

U binden, dingen, dringen, finden, gelingen, klingen, ringen, schlingen, schwinden, schwingen, singen, sinken, springen, stinken, trinken, winden, wringen, zwingen 3 beginnen, gewinnen, rinnen, schwimmen, sinnen, spinnen

3 10

I

3

3

glimmen, klimmen

2

4 11

I

a

a

bringen

1

5 12

I

υ

υ

schinden

1

b 1 13

ε

a

3

bergen, bersten, gehen, helfen, schelten, sterben, verderben, werben, werfen

9

2 14

ε

3

3

dreschen, fechten, flechten, melken, quellen, schmelzen, schwellen

7

3 15

ε

a, υ

3

werden

1

6

13

For the three athematic verbs tun, gehen, stehen, see section 11.1 Cases of Ablaut-independent radical vowel-alternations aie not considered: 1) "Präterito-Präsentien", e.g. modals such as dilrfen-durfte-gedurft, mögen-mochte-gemocht, whose actual present forms are former preterites, the initial present having been lost (cf. Paul et al. 1989:262f). 2) "Rückumlaut", i.e. verbs identifying themselves as weak through their dental suffix, e.g. brennen-brannte-gebrannt, but exhibiting a root-vowel alternation due to the failure of Umlaut in the preterite (cf. Paul et al. 1989:257): brennen, denken, kennen, nennen, rennen, senden, wenden.

A Generalized Theory of Ablaut: the Case of Modern German Strong Verbs IV 1 16

ε

αα

3

2 17

ee

αα

00 befehlen, stehlen

3 18

ee

αα

3

4 19

3

αα

D kommen

5 20 flete αα

OO

brechen, erschrecken, sprechen, stechen, treffen

5 2

nehmen

1 1

gebären

1

6 21

œ

D

3

erlöschen

1

V 1 22

ε

αα

ε

essen, fressen, messen, vergessen, stecken

5

2 23 3 24

ee ee

00 αα

00 bewegen, heben, pflegen, scheren, weben ee geben, genesen, geschehen, lesen, sehen, treten

5 6

4 25

ii

αα

ee liegen

1

26

I

αα

ee bitten

1

6 27

I

αα

ε

5

7 28 asae 00 VI 1 29 a UU 2 30 αα UU

sitzen

1

00 erwägen, gären, schwären a backen, schaffen, wachsen, waschen αα fahren, graben, laden, schlagen, tragen

3 4 5

3 31 4 32

œ

00

a

0

00 schwören 3 schallen

1 1

Vila 1 33

αα

ii

αα blasen, braten, raten, schlafen

4

a a

ii

a a

3

2 34 3 35

b l 36 a υ

I

37

fallen, halten, lassen fangen, hangen

2

ii

a ι hauen, laufen

2

2 37

a ι

ii

a ι heißen

1

3 38 4 39

uu 00

ii ii

UU

rufen 00 stoßen

1 1 Total 1

170

The first column in (14) shows the traditional classification in seven major classes (J. Grimm), with subdivisions a, b, c, ..., and 1, 2, 3, .... Column two shows our numbering of the different Ablaut-series; columns three to five display the Ablaut-series, i.e. the three root-vowels of the present, the preterite and the past participle; column six lists the members of series in question; column seven gives the total number of verbs belonging to the series. The German system represents a challenge for anybody supporting a derivational, non-lexical hypothesis for two reasons. First, there are no fewer than 39 different Ablaut-series on the surface (as opposed to four in CA). That is, a strong verb may belong to 39 different vowel-patterns. Second, the vocalic system of German is much more complex than that of CA: the latter has only three

vowels,

the

former

has

no

fewer

than

19 vocalic

segments

(16

vowels

[Y,yy,00,œ,i,ii,e,ee,aîae,a,aa,û,uu,o,oo,] and three diphthongs [au,ai,i]), 16 of which participate in the 39 Ablaut-series. This last point is to be taken as the main obstacle for an analysis that claims that the regularity controlling such a system knows only four vowels (β,ί,α,ιί). 5.2 Irrelevant features Since Ablaut is a phonological process, a phonetically complex vocalic system must be reduced to its phonologically relevant properties. First, apophonic regularities as understood above exclusively concern the quality of the vowels involved. Any contrast in length is thus irrelevant for ablauting systems. Second, tenseness/ ATRness is not contrastive in German: long vowels are +ATR, short

Philippe Ségéral and Tobias Scheer

38

vowels are -ATR (except for low [αα] of course).14 Third, following current representational systems (cf. Lowenstamm forthcoming), we assume the two diphthongs [ai], [au] occurring in strong verbs to be vowels followed by off-glides. (15)

[al]

Ο Ν Ο Ν

[au]

II a j

Ο Ν Ο Ν II a w

As apophony involves vocalic alternations only, the off-glide residing in an Onset position is not affected. We shall assume that the object over which apophony operates is the vowel hosted in the Nucleus, i.e. [a] in the above cases. The phonologically relevant inventory of German vowels can thus be reduced to the 7-vowel system /1, E, Y, Œ, U, O, A /. Thus, table (14) is to be recast as shown in (16), reducing 39 phonetic to 18 (A to R) phonological Ablaut-series. Pres Prêt Part A Β C D E F G H I J Κ L M Ν 0 Ρ

Q R

I I I I I I E E E A A A A Y ΠU 0 0

A A A A U

E A U

0

0

A A 0 I I υ 0 0 0 I I A

E 0 0 I A A 0 0 0 u o 0

0

U

table) examples 25, 26, 27 liegen, bitten,sitzen 11 bringen 8 binden, dingen,... 9 beginnen,... 12 schinden 3,7,10 biegen, bieten,... 22,24 essen, fressen,... 13,15-18,20 bergen, bersten,... 14,23,28 dreschen,fechten,... 1,2 beißen, bleiben,... 33-37 blasen, braten,... 29, 30 backen, schaffen,... 5, 6, 32 saugen,... 4 lügen,küren,trügen 21,31 schwören,... 38 rufen 39 stoßen 19 kommen

no. 3 1 18

6 1 24

11 19 15 39 12 9 4 3 2

1 1 1

5.3 The apophonic mechanism Under the derivational hypothesis outlined above, only the present form of a strong verb is recorded

14

The distribution of [ee], [ses] and [ε] in strong verbs is slightly more complicated, but still complementary. Both the right and left consonantal environment determine whether the long allophone of ÍE/ is [ee] or [xx], That is, [ass] occurs iff /E/ stands after a velar or a labial consonant and before an [r] or a velar. Elsewhere, [ee] appears. Compare e.g. [aeae] geBäRen, schWäRen, GäRen, erWäGen vs. [ee] stehlen, scheren, lesen.

A Generalized Theory of Ablaut: the Case of Modern German Strong Verbs

39

in the lexicon. Both the vowels of preterite and past participle forms are derived from the lexically recorded vowel by means of apophony. The apophonic path restricts the possible derivations in three ways: first, only the three cardinal vowels plus zero may participate in the alternation; second, derivations such as u==>a or a==>i that go against its directionality are prohibited; third, only adjacent vowels in the path enter an apophonic relation: "jumps" ignoring a path's member like i==>u are ruled out.15 Given these restrictions, the following four Ablaut-series, and only these, are predicted to occur in a system of three apophonic degrees: (17)

degree

1 0 1 a u

==> ==> ==> ==>

2 i a u u

==> ==> ==> ==>

3 a u u u

The apophonic derivation starts out with a lexically given vowel (those listed under "degree 1 " in (17)), which we shall call the entering vcwel. 5.4 Partial apophony In the case of three grammatical forms X (lexical), Y (derived from X) and Ζ (derived from Y), related by two derivations, it might be the case that only one derivation is apophonic. We shall call such cases partial apophony. Two different variants of partial apophony are logically possible, i.e. X = > Υ — Ζ and X - Y = > Ζ.16 However, in the light of what was claimed in the preceding section, the latter case is predicted not to occur. Indeed, the entering vowel would be Y, which itself is the output of a nonapophonic derivation performed on X. Since Y is not lexical and is apophonically unpredictable from X, speakers would be unable to access it. Under our approach, Y is necessarily absent from the lexicon because only X has a lexical status. Hence, the apophonic frame we have discussed so far makes the first prediction: if it is true that the three grammatical forms are related by an apophonic derivation, then a pattern like *X--Y=>Z should not occur. Scanning the data reveals that two out of 170 verbs actually do display this property: kommen, schinden. Grammars (e.g. Paul et al. 1989:248, Paul 1921:440) point out that these two verbs are notoriously irregular and isolated. We will come back to them below. Note that the prediction *X—Y=>Z illustrates a more general observation holding for the whole system of NHG strong verbs: the present and the preterite contract a close relationship, their forms

15

16

In southern German varieties such as Alemanic that lack preterite forms, the remaining alternations are lexicalized. The symbol "-" stands for "non-apophonic relation", as opposed to "==>" which refers to an apophonic relation.

Philippe Ségéral and Tobias Scheer

40

differ solely by means of the stem-vowel alternations (près sing- vs. preterite sang-). In contrast, the formation of the past participle involves additional morphological gestures, viz. the affixes ge- and en (ge-sung-en). Moreover, genetically, the past participle was an ¡dependent nominal form in IndoEuropean, lacking any derivational or apophonic relation with the other two degrees. Given partial apophony, we have to distinguish three kinds of strong verbs: 1) those displaying a fully apophonic series (i.e. two apophonic derivations respecting the path as listed in (17)), 2) halfapophonic verbs showing a pattern of partial apophony X=>Y—Z, and 3) non-apophonic verbs (i.e. violating the path or of the shape *X--Y=>Z). 5.5 Predictions made by apophony Consider the 64=26 logically possible three-member series when four vowels are assumsed ([0,1,a,u]) and no order is required, as shown below: 0 0 0 aao a a i U0Í i 0 u iau U U 0 0 i i

i00 uai a0u 0ui

ia0 00a aau iii

a00 0aa u0u iui

aae i0a uau aii

ua0 iaa aea 0 i 0 0U0 aui u i i U 0 0

00i aaa ii0 uui

0a1 u0a iu0 0ia

101 uaa ai0 0ua

1a1 00u au0 iia

a0i eau Ul 0 iua

aia aua ui a 1111a

0iu 0uu iiu iuu

According to the definitions of apophony and partial apophony arrived at in the previous section, the frame we have developed so far predicts certain series to be a pattern of a possible NHG strong verb. Other series will be predicted not to occur in NHG. Out of the set shown in (18), the apophonic system recognizes 16 series as possible instances of apophony (including the four fully apophonic series mentioned in (17)): (19)

3 different vowels A, Β, C

apophonic X==>Y==>Z

1 0==>i==> a 2 i==>a ==>u

half-apophonic X==>Y—Ζ

5 0==>i — u 6 i==>a — 0 7 a==>u~ 0 8 a==>u~ i

2 different vowels A, A, Β

A, Β, Β

1 vowel A, B, A

3 a==>u==>u 9 u==>u~0 10 u==>u-i 11 u==>u-a

12 0==>i—I 13 i==>a—a

A, A, A 4 u==>u==>u

14 0==>i—0 15 i==>a~i 16 a==>u~a

The 16 series above thus constitute the frame in which NHG strong verbs are predicted to fit: any of the verbs is predicted to be an instantiation of one of them. Having established the predicted set of possible NHG verbs, we proceed by mapping these predictions onto the 19 existing series shown in (16). Our analysis will be carried out in two steps. Step one concerns the verbs that display only the three cardinal vowels [i,a,u], i.e. where no complex (i.e. mid or front rounded) vowel occurs in any of the three forms (series B, C, E, J, K, L, P). In step 2, we shall face the main challenge of NHG, i.e. to show that it is possible to deal with mid and front rounded vowels by means of a formal instrument that ignores these articulations.

A Generalized

Theory of Ablaut:

the Case of Modern

German Strong

Verbs

41

6. Series lacking mid andfrontrounded vowels (B, C, E, J, K, L, P) According to the apophonic mechanism that we hold responsible for the grammatical derivations in NHG strong verbs, the following verbs without complex vowels are non-apophonic: 1

E

I

U

U

12

schinden

Ρ

υ

I

υ

38

rufen

J

A

I

I

1,2

sich befleißen beißen,...

Κ

A

I

Α

1

3 3 - 3 7 blasen, braten, raten

39 12

None of these series is predicted to exist according to (19). Series E and Ρ contain one single verb each, rufen (Ρ) violates the predicted directionality, schinden (E) (see the above discussion) has the prohibited shape *X--Y=>Z (/—«==>«). These two verbs are lost and they are unrecoverable under the apophonic hypothesis. The case of the series J and Κ is much worse for our hypothesis: 39+12, i.e. 51, truly ablauting verbs seem to be rejected by the apophonic mechanism. For the time being, we have to record this situation to be a serious setback to our ambitions. However, further development of the analysis will show that these two series are simply unanalysable at this stage of the discussion. We shall return to these verbs after having resolved the problem of complex vowels. Two series, L and B, are declared half-apophonic by the apophonic mechanism: L

A U A

Β

I

A

29, 30 backen, schaffen,... A

11

9 1

bringen

The series represented by L corresponds to number 16 of (19), a==> u-a. bringen, the only verb ofB, illustrates number 13 of the predicted series, i==>a—a. Finally, only one series, C, representing 18 verbs appears to be M y apophonic: C

I

A

U

8

binden, dingen, dringen, finden, gelingen, klingen,

ringen,...

18

In sum, one very encouraging fact emerges from the preceding considerations: there is no NHG series of three different vowels A,B,C that would have to be declared non-apophonic by our frame. The only NHG series with three different vowels involving [0,1,a,u], series C, is precisely one of the two predicted series of this kind according to (19). This fact appears to be striking confirmation of the postulated derivational directionality. Indeed, no less than 24 different three-membered series involving [0,1,a,u] are logically possible: (23)

l e a i 20au -> 3 e i a

4 0i u 5eua 6 oui

7 aol 8aou 9 aie

lOaiu 11 a u e 12 a u i

13 1 0 a 14 i e u 15 i a a

-»16 i au 17iu0 18 i u a

19 u o a 2Ou0i 21 n a n

22 u a i 23 u i e 24 u i a

42

Philippe Ségéral and Tobias Scheer

The apophonic mechanism eliminates 22 of them, identifying o==>i==>a and i==>a==>u as the only possible patterns for a NHG verb with a series of three different vowels. It can hardly be accidental that the only NHG series involving three different cardinal vowels, C, precisely matches one of the two predicted possibilities (i==>a==>u). Nevertheless, we have to admit that the numeric result is not very favourable for our purpose: apart from the 51 verbs of series J and Κ which apparently are excluded, only 28 verbs out of 170 are half-apophonic (series L and B) or fully apophonic (series C). However, we must keep in mind that only verbs involving exclusively cardinal vowels have been considered so far. Verbs with at least one complex vowel prevail (89 out of 170) and we shall now turn to an analysis of such verbs.

7. Series showing mid orfrontrounded vowels (A, D, F, G, Η, I, M, N, O, Q, R) 7.1 Infra-segmental representations The relevant series exhibiting a mid or front rounded vowel in at least one of their forms are A, D, F, G, Η, I, M, N, O, Q and R. The vowels were referred to as "complex vowels" because they are marked with respect to [i,a,u] and they do not occur in the minimal vocalic inventory a language can deal with, i.e. [i,a,u]. Consequently, all the current models of infra-segmental representation confer a special status to [i,a,u], making them more basic than mid or front rounded vowels. This treatment is most prominent in models that deal with phonological primitives whose size exceeds a single distinctive feature, such as Particle Phonology (Schane 1984), Dependency Phonology (Anderson and Ewen 1987) or Government Phonology (Kaye et al. 1985). In order to approach the complexvowel series, we use the infra-segmental representations of the latter frame. However, we should like to emphasize that any other representational model would yield the same result for our purpose. Indeed, all that is needed for a discussion of the series with complex vowels is the decomposition of the segments into smaller units. The internal structure of the relevant vowels according to Kaye et al. (1985) is shown in (24b). The phonological primitives are called Elements. They are noted in capital letters and defined by a feature-matrix. The underscored feature ("hot feature") defines the Element's salient property, cf (24a). Elements combine into segmental expressions that define a vowel. Elemental expressions are headed, i.e. one Element, the head, is prominent with respect to the other Elements) that are called operators). Conventionally, heads are underscored. The head of an expression confers more of its properties to the expression's phonetic value than the operators) (cf. Kaye et al. 1985 for a more detailed discussion of the fusion-process of different Elements). For example, I-A represents a front mid vowel that is closer to I than to A, i.e. [e,e]. The inversely headed expression A-I still stands for a vowel in-between I=[i] and A=[a], but closer to A than to I, i.e. [ae]. Consider the following elemental combinations where Elements reside on autosegmental lines, one for each salient

A Generalized Theory of Ablaut: the Case of Modern German Strong Verbs

43

property. Note that v, the neutral or "cold vowel", represents the maximally unmarked vocalic articulation [i]. (24) I

a. phonological Elements U A

b. vocalic expressions

1

back-round -back +high -ATR -low

+round +back +high -ATR -low

-round +back +high -ATR -low

-round +back -high -ATR +low

I round

ν

Ï

high

A

Ia

U—ν

I ν

_v—I—-I I I I

ν

U—U— U

I

I I I

I I I

ν

I

1

I ν

I

A—A—A—ν—A

y 0]

Tense-contrasts are dealt with by the ATR-Elementl [ι] and [u] are the direct phonetic translations of I and U. All mid vowels are a combination of FU head and A operator. Tensed mid and high vowels in addition bear-I: [i,u,e,o]=[i,u,e,o]+I. Since ATR is phonologically irrelevant in NHG (cf. above), we will omit it in the representations below. 7.2 Apophony and complex vowels Given the infra-segmental structures introduced in 7.1, let us consider an example of a series involving mid vowels. Below, the vowels of series H (bErgen - bArg - gebOrgen) are given in elemental notation. (25) present

I = I A [ε]

preterite A

[a]

past participle ==> U

A M

Obviously, series Η can be read as a fully apophonic series i==>a==>u, provided the Element A is neglected in the present and in the past participle. Our general position is that precisely this is the way apophony deals with non-cardinal vowels. Of course, the conditions under which Elements are neglected must be predictable. We come back to this issue in section 9. Let us first state the condition under which apophony can deal with non-cardinal vowels: (26) apophony applies to phonological Elements, not to segments. The apophonic path, thus, must be recast as follows: (27)

Apophonic Path e => I=> A=> U=> U

44

Philippe Ségéral and Tobias Scheer

(27) is a condition on the universality of the apophonic mechanism. If there is any point in claiming that there is a generalizable cross-linguistic apophonic regularity, two requirements must be fulfilled: 1) the number of vowels the mechanism functions with may not exceed the number of vowels found in maximally simple vocalic systems, [i,a,u], such as the one of CA; 2) the objects the mechanism deals with must be able to drive apophonic derivations in languages provided with a more complex vocalic system. (27) satisfies both requirements. Hence it is not to be assigned the status of an additional stipulation, since it follows straightforwardly from the universality of apophony. Thus, within a "complex" vowel, two different kinds of Elements are to be distinguished: (28) a. one Element that undergoes apophony (the I in the present of the above example, the U of its past participle). b. one Element that does not participate in the apophonic derivation (the A in the present and the past participle). We shall call (28a) apophonic Element, (28b) parasitic Element. 7.3 The distribution of parasitic Elements in series with complex vowels This distinction of course raises the question of how speakers know which Element is apophonic and which is parasitic in a complex vowel. In the case of bergen, for example, the lexically recorded [e]=I-A offers two possible Elements for entry into the apophonic path. I is the correct one yielding the A of the preterite form barg. In contrast, entering the path with the second Element offered by [ε], A, would lead to the preterite *burg. Hence, the correct distinction between apophonic and parasitic Elements is crucial for the apophonic mechanism. Our derivational hypothesis can be maintained only if there is a way for the speaker to identify the apophonic Element of lexically recorded complex vowels. In order to provide an answer to this question, let us examine in detail which Elements the apophonic frame predicts to be apophonic, and which parasitic. Consider a verb like biegen - bog - gebogen, belonging to series F. Its pattern i-o-o could be read half-apophonically as I=>A—A with a parasitic U in both the preterite and the past participle. However, the optimal apophonic reading of biegen is I = > A = > U , provided a parasitic U is inferred in the preterite, and a parasitic A assumed in the past participle, biegen is fully apophonic. In a case like e-a-e (series G, e.g. essen-aß-gegessen), the optimal apophonic reading is I = > A ~ E (inferring a parasitic A in the present), thus only half-apophonic. In the table below, the procedure of optimal apophonic reading is followed for each of the series involving complex vowels, i.e. A, D, F, G, Η, I, Μ, N, O, Q, R. Column 1 of the table below shows the series. Column 2 specifies its optimal apophonic reading and gives the result ("APOPH" for fully apophonic, "half-ap" for half-apophonic, and "lost" for non-apophonic). In column 3, the parasitic Elements that must be inferred according to the optimal apophonic

A Generalized Theory of Ablaut: the Case of Modern German Strong Verbs

45

reading are given for the present, the preterite and the past participle ("..." indicates that no parasitic Element must be assumed, "/" that the vowel is apophonically uninterpretable). 1

2

3

series

optimal apophonic

parasitic Elf :ments to be infA~e I=>A=>U I=>A=>U I=>A~e I=>A=>U I=>A=>U A=>U=>U A=>U=>U I=>A=>U 0=>I=>A

4 PRET

PART

U

A A

U A U u

A A A A A

/

/

/ A A A U A+U A /

/

17

nb 3 6 24 11 19 15 4 3 2 1 1

The result of this evaluation is striking for two reasons: (30) a. I is never parasitic b. the distribution of parasitic A and U is complementary with respect to the preterite and the past participle: only A is parasitic in the latter, only U in the former. Both can be parasitic in the present. (The irregular parasitic A in the preterite of series M will receive an explanation in section 8.) We would like to stress that it is the apophonic hypothesis alone that allows one to discriminate Elements in this way. Such a result is not possible outside of the apophonic frame. The two complementary distributions show that the apparent disorder prevailing in these 89 verbs in fact responds to a precise regularity. This regularity then, if any, must be the frame defined by apophony. Apart from suiting the NHG system on formal grounds, apophony also proves successful numerically. Only one verb, kommen, out of 89 must be declared lost. And still the nonapophonicity of kommen does not originate in the configuration of parasitic Elements, since it would be lost anyway (cf. the above discussion of this verb). All other verbs are half-apophonic (14 in all) or fully apophonic (74). Thus, both the formal and the quantitative result strongly support the apophonic hypothesis. 7.4 The Apophonic Theorem Parasitic Elements must be detected in the present in order to run the apophonic derivation. The stem vowel of the present being lexically recorded, speakers have no indication as to which of the

17

The attentive reader will have noticed that there should be a parasitic U inferred in the present as well as in the past participle. In fact, the Us in question are not parasitic but lexically present. The discussion at this stage does not allow us to justify this statement, but we shall return to the case of stoßen extensively below.

46

Philippe Ségéral and Tobias Scheer

Elements offered by the lexicon is to be affected by apophony. Hence, there must be a way for them to unambiguously identify the apophonic Element. A completely different situation obtains in the other two degrees: 1) unlike in the present, the apophonic Elements of the preterite and past participle vowels are a derivational output. Speakers thus have direct access to the Element to which apophony is to be applied, and no independent identificational mechanism is needed. 2) the situation of parasitic Elements in degrees two and three is the reverse of the one prevailing in degree 1 : in the present, parasitic Elements are given by the lexicon, whereas no lexical information whatsoever concerning parasitic Elements is available in the preterite and the past participle. It follows that parasitic Elements in degrees 2 and 3 must be predictable independently from apophony (cf section 9). The result achieved in the preceding section allows us to answer the crucial question as to how speakers identify the Element to which they apply apophony within the lexically recorded vowel. (30) shows that entering and parasitic Elements are complementarity distributed in NHG. Hence, (31) Apophonic Theorem a. if an Element is parasitic, it is not a possible entering Element b . if an Element is not parasitic, it is an entering Element18 The addition of the null-set being without consequence for the result, it does not make sense to talk about 0 being parasitic. Since it is initially always present, a is the default-entering Element under any circumstances and in any language. It follows that in NHG, 0 and I are the only possible entering Elements. Hence, once learners have figured out that 0 and I are possible entrances into the apophonic path, but A and U are not, they are equipped to face any NHG verb they do not know yet. They will be sure to apply apophony to the correct Element simply by ignoring any occurrence of A and U in what the lexicon provides. Note that the above Apophonic Theorem is one possible mechanism a language can recur to in order to assure the correct functioning of apophony when facing a very complex vocalic system. Other languages might develop palliatives of a different kind for this purpose (cf. the functioning of Old English described in Ségéral and Scheer 1996). Yet, whichever language-particular mechanism guarantees correct apophonic interpretations, the apophonic path remains universally valid. Given the Apophonic Theorem for NHG, the theory makes much more precise predictions about possible NHG apophonic series. Indeed, the set of predictions in (19) allowed for no fewer than 16 series that could be declared apophonic or half-apophonic. In the light of the present section, there are only four possible apophonic series left:

18

This statement entails a prediction. In a language with the minimal vocalic inventory [i,a,u], there are no complex segmental expressions. Hence, no Element can ever be parasitic. If (3 lb) is correct, any Element must be a possible entering Element in such a language. As we have seen above, this prediction is born out for Classical Arabic as well as for Kabyle Berber where any vowel is a possible input into the Apophonic Path.

A Generalized Theory of Ablaut: the Case of Modern German Strong Verbs (32)

fully apophonic 0==>I==>A I==>A==>U

47

half-apophonic 0==>I~X I=>A—X

8. The real nature of the "serious setback" Let us return now to what appeared to be a serious setback for the apophonic theory in section 6. Recall that no less than 51 verbs (series J a-i-i and Κ a-i-a) had to be declared non-apophonic according to the stage of investigation we had reached then. However, since we have have learned more about parasitic Elements and their distribution, the series J and Κ now appear in a completely different light. The observation "A and U are always parasitic in the present" forces us to reconsider their apophonic derivation: the [a]=A they display in the present is parasitic. A minus A being a, the real entering Element of these series is 0. Both exhibit an I in the preterite, which fits perfectly with the predicted apophonic output of 0. Series J has I in the participle, too, making it half-apophonic. Series K, however, shows an A and can thus be regarded as fully apophonic. (33)

series J a-i-i Κ a-i-a

apophonicity half-apophonic APOPHONIC

legibility 0 ==> i 0==>i==>a

number 39 (e.g. beißen) 12 (e.g. raten)

As a matter of fact, the "serious setback" turns out to be a simple illustration of a series that possesses the pecularity of entering the path with 0. These facts could only emerge after the distribution of parasitic Elements was understood. Although this analysis has the benefit of adding 51 verbs to the apophonic family, it also has the disadvantage of "losing" 13 verbs belonging to series L a-u-a backen (previously half-apophonic, cf. (21)) and M a-0-0 saugen (previously fully apophonic, cf. (29)). The surface A in the preterite being an apophonic 0, there is no way of deriving a U (series L a-u-a, series M a-0-0, reading 0-UU) or an A (series M a-0-0, reading 0-A-U) in the preterite by means of the apophonic path. Series L and M are thus predicted not to represent apophonic verbs. When considering diachronic data in section 11.4, we shall see that the exclusion of precisely these verbs is a powerful argument in favour of the apophonic mechanism.

9. Predictability of parasitic Elements Let us now turn to another crucial point of the demonstration. Under our analysis, the root-vowels of degree 2 (preterite) and 3 (past participle) are absent from the lexicon. They combine from 1) an apophonic Element that is delivered by the apophonic path and 2), should the occasion arise, a parasitic Element that must be predictable independently of apophony. As we have seen in (29), the parasitic Elements U and A are in complementary distribution, that is the former occurs in preterite forms only, the latter exclusively in the participle.

48

Philippe Ségéral and Tobias Scheer

9.1 Parasitic U in the preterite Parasitic U in the preterite occurs in series F, I, Μ, Ν and Ο (M (saugen) is not to be considered anymore since it was identified as non-apophonic in the preceding section). We shall show below that all of these parasitic Us originate in the consonantal neighbourhood of the root-vowel. It is therefore useful to have an idea of which consonants are likely to provide a U colouring influence. Recall that U is the phonological vector of velarity (cf. the tongue-body position defined by the matrix in (24)) and labiality (cf its hot feature). And indeed, it comes as no surprise that most models of consonantal representation, whichever kinds of primitives they recur to, establish a relationship between [u,w], labial and velar/uvular consonants (cf., among others, Clements 1993, Harris and Lindsey 1995, Scheer 1996, Anderson and Ewen 1987). Further discussion of consonantal identities would go far beyond the scope of this article. All we need to know when discussing the consonantal context of U-influenced root-vowels is that the same primitive, U in the frame used here, contributes to the definition of [u,w] as well as of any labial and velar consonant. Hence, [w], labials and velars/uvulars will be suspected of being possible sources of U. By contrast, a parasitic U could never originate, say, in a coronal or a palatal consonant. The detection of parasitic U will proceed in two steps. First, we shall discuss the case of series F and O, which are remarkably homogeneous. Second, the remaining verbs will be submitted to a distributional analysis of the consonantal context of the root-vowel. 9.1.1 Series F and N: abnormal vowel length in the present Consider first that U is present in all degrees of series F ( biegen-bog-gebogen) and Ν (lügen-loggelogen) except for the present of the former. Its presence cannot be related to any consonantal context: it occurs in any verb, even in those with dental consonants (e.g. fließen, sieden). The apophonic hypothesis that postulates the parasitic character of U is thus forced to assume its lexical presence in the underlying representation of series F and N. Moreover, U must reside outside the apophonically variable root-vowel. This follows from the fact that the root-vowels of the preterite and participle forms are defined by the sole means of apophony. As U is not a possible output for the preterite in NHG (only a and I are possible inputs), it can never ever be delivered by apophony. Still, however, it is there. The theory thus predicts a lexical structure of the type /fliUßen/ fließen, /biUgen/ biegen for the relevant verbs. Under this analysis, the present I delivers an A in the preterite via apophony, hence /flAUß/, /bAUg/. The sequence /AU/ combines along the described elemental fusion process, yielding 101. According to its length, the result is [fbs], [boog]. But what happens in the present? Why does the U not leave a trace in the phonetic output /fliUßen/ [fliisan], /biUgen/ [biiganj?

A Generalized Theory of Ablaut: the Case of Modern German Strong Verbs

49

The answer comes from a more general observation: in NHG, there are no sequences such as *[iw]. If *[iw], for a reason we ignore, is illicit in NHG, a conflict arises in the present.19 (34) /ONON/ I I ι υ

[ONON]

conflict/IU/, result [ii]

V I (U)

We face here a typical case of compensatory lengthening where the loss of an object to the right of a vowel entails lengthening. This view is confirmed by another fact that shows how learners can correctly lexicalize parasitic U. That is, the length of the root-vowel within series F, Ν and J (bleichen-blich-geblichen) is not random. It is driven by the voice-value of the consonant on its righthand side. E.g. reiten (J, prêt. ritt) has a short [i] in degrees two and three, whereas scheiden (J, prêt, schied) shows long [ii] in these forms. The same holds for fliegen (F, prêt, [floog]) vs. gießen (F, prêt. [gos]). The parallel case to the present /IU/ in series F, Ν is the sequence /AI/ in series J. No conflict arises here, the result is the diphthong [ai], observable throughout the present of series J. Thus, no compensatory lengthening occurs. The consonantal context voiced vs. voiceless does not interfere since there is no opposition such as short vs. long diphthongs in NHG. The case of the present of series F, Ν is different since no diphthong occurs. Hence, the vowel in the present is expected to obey the length contrast controlled by the voicing of its right-hand consonantal neighbour. However, it does not: where gießen [gìsan] is expected, [giissn] occurs. As we have seen above, long vowels in this context are the result of compensatory lengthening. That is, "something" was lost here so that the root-vowel could expand. This "something" is precisely the trace of the predicted parasitic U. Note that when the parasitic U combines with the root-vowel, as in the preterite, no compensatory lengthening occurs because nothing has been lost. Unless the loss of a lexically present /U/ is assumed, there is no way in which to explain the abnormal length of fließen, sieden etc.20 Moreover, the prediction of the presence of a lexical /U/ in series F (except for klimmen, glimmen) and Ν based on their homogeneous behaviour, which contrasts with any other strong verb, is a piece of historical evidence anyway. All of the verbs mentioned and only these verbs constitute Grimm's class Π, whose identifying property is a labial context: verbs belong to class Π precisely iff they exhibit a historical [ w] on the right-hand side of the root-vowel.21

19

20

21

One way out of this situation is simply to neglect the U as in the above examples. Another is to combine the I and the U of /IU/. The result then is [yy], that is, series Ν lägen, küren, trügen. glimmen and klimmen, members of series F, exhibit a short root-vowel in the present and thus do not participate in the process under discussion. Their parasitic U in the preterite will be discussed in the next section. The traditional classification is based on the former right context of the root-vowel: I=[ j], II=[ w], III=[ NC/NN/LC], IV=[_N/L], V=[_C],

Philippe Ségéral and Tobias Scheer

50 9.1.2 Labial and velar environments

We are thus left with the following verbs carrying a parasitic U in the preterite that has to be predicted synchronically. The table below shows that any of the verbs which apophony predicts to bear a parasitic U has a labial or a velar consonant in the right or left context of the root-vowel (for [ ç, J] see note 22). (35)

LAB

context lab

lab_

VEL

vel

veibs

series

glimmen, klimmen

F

heben, weben

J

quellen, schmelzen, schwellen, melken bewegen, pflegen, erwägen

J

r scheren, gären, schwären schwören

_J

dreschen _Ç

Ρ

erlöschen J

fechten, flechten

Recall that precisely these two groups of segments are potential U providers. Note that our prediction entails not only that the above verbs must have a labial or a velar/uvular root vowel context, but also that all the strong verbs with this kind of context must bear parasitic U. As a matter of fact, not every adjacent velar/ labial has a U-like influence on its environment (e.g. werben-warb, schwinden-schwand). Nevertheless, the contamination of the root-vowel by a Ucarrying neighbour is predictable to a large extent. We have identified the following two parameters that drive the spreading of U. First, potential U-providers on the righthand side of the root-vowel liberate parasitic U only if they are voiced (compare heben-hob, pflegen-pflog with treffen-traf, stecken-stak). Second, the root-structure is a conditioning factor. In a potentially U-providing preceding environment [lab ], parasitic U does occur only if 1) it is preceded by another consonant (compare [(Χω, ] schmelzen-schmolz with [Qab ] werben-warb) and 2) followed by an LC cluster (compare [ LC] schmelzen-schmolz with [ NC] schwinden-schwand). By contrast, in the mirror-situation of a potential following U-provider, parasitic U occurs only if its source is not followed by another consonant (compare f CyeiVI scheren-schor with [ CveiCl bergen-barg). We do not know why unvoiced labials/velars do not contaminate the root-vowel, neither is the causal relation between spreading and the mentioned structures clear. Moreover, there are three disobeying verbs: liegen-lag, geben-gab, gebären-gebar. This situation results from the fact that we have not understood the syntagmatic relations holding within German clusters yet. Second, actually understanding of internal structure of consonants does not allow us to make more precise predictions. Be this as it may, there can be no doubt that the predictions apophony makes about

A Generalized Theory of Ablaut: the Case of Modern German Strong Verbs

51

parasiticism in the preterite are correct: it is obvious that the presence of U in the preterite is not a lexical feature but depends on the environment of the root-vowel.22 9.2 Parasitic A in the past participle The apophonic frame predicts a parasitic A in the participle of series D, F, H, I, Μ, Ν, O and Q, that is the series of M y apophonic verbs. Only one series, C, is fully apophonic but lacks parasitic A in the preterite. The peculiarity of series C (binden-band-gebunden, traditional class Dial) is its context [ (NC)homorç]. Hence, synchronically, it is very simple to predict parasitic A in participles: it occurs everywhere except in verbs with a [ (NQhomorg] context. Of course, there should be a reason why this particular context does not provide A or, rather, why any context but Γ (NQhomorel provides A. As a matter of fact, what we face here is a process referred to as Brechung in German diachronics. Brechung is vowel harmony from right to left where a low or mid suffix-vowel lowers high root-vowels to mid articulations.23 However, Brechung does not take place if there is a sequence [ N C ] ^ or [NN] between the suffix vowel and the root vowel (inf bind-an "bind1, svimm-an 'swim'). The parasitic A predicted by apophonic theory is obviously identical with the one originating in Brechung non-high suffix-vowels. Indeed, apophony provides an opportunity to observe the synchronic effect of Brechung. It shows that this process, which is considered to have been inoperative for centuries, is still active in the grammar.24 22

23

24

A good indicator for the accurateness of the predictions made by apophony is the apparently absurd claim that [f, ç] are velars. Indeed, both segments turn out to have velar affinities in German. First, regarding [f] as a velar is historically obvious and synchronically recoverable by learners: any NHG [J] (but in [#JC]contexts as e.g. schlafen) dates back to Old-High-German (OHG) [sk]. Thus, erlöschen, dreschen < OHG lesean, drescan. This move has set free the velar component U of OHG [sk] that has remained floating since then. In the case of erlöschen, it has even contaminated the present (OHG [e]+U=NHG [ce]). Second, the classification of [ç] as a velar is based on its underlying identity, which is /χ/, [ç] and [χ] are in complementary distribution in German, [ç] occurring after front vowels only, [χ] after non-front vowels. The underlying identity of [ç, χ] is revealed for example by Umlaut in Köchin "cook fem" [kœçin] vs. Koch "cook mase" [Ιοχ] where the suffixal [i]=I triggers Umlaut /o/-->[ce] and /χ/—>[ç] simultaneously. More concomitant evidence for the crucial role played by the environment of the root-vowel comes from the evolution [r] > [κ]. In Middle-High-German (MHG), the preterite (singular) of classes III-V was uniformly [a]: e.g. stehlen-stahl, scheren-schar (NHG prêt, stahl, schor). At that time, the consonant transcribed as "r" was the alveolar [r]. Strikingly, the move prêt, [a] > [o] in the mentioned veibs chronologically coincides with the shift alveolar [τ] > uvular [κ]. This situation can hardly be regarded as an accident. More generally, the prediction of a parasitic U for the defined set of verbs sheds light on the disparate evolution of the preterite of strong verbs from MHG to NHG, which has not been widely reflected upon in historical grammars. OHG inf e33-an vs. lsg près ijj-w "eat", 3sg preterite scalt-a vs. 3pl preterite scul-un "shall", lpl près zeohemes vs. 2sg près ziuh-is "pull", cf. Scheer (1995) for extensive discussion. Brechung does not seem to pattern with NHG [ NN|-verbs such as rinnen - rann - geronnen (IIIa2,3): as Brechung was blocked by [ NN]-contexts, these NHG verbs should not bear the parasitic A in the participle. And indeed, in these cases, the parasitic A is due to an independent process of vowel-lowering before Nasals that occurred in the evolution from MHG to NHG. Compare MHG sunne, sumer, miinech, kilnec, gewunnen with NHG Sonne, Sommer, Mönch, König, gewonnen (cf. Paul et al. 1989:79). MHG

Philippe Ségéral and Tobias Scheer

52 10. The apophonic recipe / numeric result

Having gone through the entire apophonic mechanism, we can now state the key to the NHG system of strong verbs under the form of a "recipe". It sums up the different operations that are performed from the lexicon to the derived forms. (36)

Apophonic recipe for learners of N H G a. take the vowel in the present b. get rid of A and U. The remaining Element (I or 0) is your input into the apophonic path c. run apophony d. according to context, eventually add U in the preterite, A in the past participle

The simplicity of the mechanism is striking. It is all that a learner needs in order to break up the apparently inscrutable opacity of the German Ablaut-system. Below we give the explicit result of the preceding analysis. It splits the 170 NHG strong verbs into three groups: fully apophonic, half-apophonic and non-apophonic. 0l>II>A

II>AI>U

sum

percentage

apophonic

K,Q: 13

C,D,F,H,I,N,0: 87

100

59%

half-apophonic

J,P: 40

A,B,G: 15

55

32%

sum

53

102

155

91%

15

9%

170

100 %

non-apophonic

11. Concomitant diachronic evidence GEHEN - When establishing the list of NHG strong verbs in section 5 .1, we neglected the three athematic verbs gehen (ging-gegangen), stehen (stand-gestanden), tun (tat-getan). These so-called Wurzelverben are peculiar because of the direct contact of the root-vowels with the personal suffixes in MHG (l.sg. tuo-n, stâ-n, gâ-ri) with no mediation of a thematic vowel (cf. participles of class Ilia prove the independence of Brechung and lowering before Nasals: their root-vowel is invariably [u], never [0], e.g. MHG glimmen - glamm - geglummen (cf. Paul et al. 1989:247f). The move [u] > [o] thus cannot be due to Brechung (the predicted MHG form would be *geglommen) but affected all and only the verbs with an [_NN]-context from MHG to NHG.

A Generalized Theory of Ablaut: the Case of Modern German Strong Verbs

53

for example Paul et al. 1989:269ff, Braune and Eggersl987:303ñ). What does apophony have to say about their case? While tun is apophonically unrecoverable, stehen allows a halfapophonic reading. The most interesting case however is gehen. This verb could receive a fully apophonic interpretation if both I and A were neglected in the present [ee]=I-A The derivation 0==>I==>A would ensue. Gehen would be the only NHG strong verb with a parasitic I in the present. Recall that the absence of parasitic I otherwise is absolutely regular. Hence, apophony has to predict that something is wrong with this "parasitic I". And indeed, one more time, diachronic facts show the correctness of this claim. In OHG and MHG, forms with [aa] and [ee] cohabited: OHG l.sg. gam besides gän, MHG gân besides gên etc. (cf. Paul et al. 1989:270fF, Braune and Eggersl987:306). In any event, the forms in [aa] are diachronically primitive, as witnessed by the older OHG gangan, stantan that show only [a] (cf. Braune and Eggersop op.cit). The development of forms in e, i.e. the projection of the "parasitic I" in question onto [a], seems to be due to the influence of suffix-vowels under non-elucidated circumstances (Paul et al., Braune and Eggers op.cit.). Whatever its origin might be, it is striking that the prediction made by apophonic theory, i.e. I's status as an alien within the rootvowel, turns out to be a diachronic fact. PARASITIC U IN THE PRETERITE - The synchronic prediction "there is no lexical U in the root-vowel of preterite forms" is a piece of diachronic evidence. Historically, all Us occurring in NHG preterite forms are a NHG innovation. Still in MHG, [a] was the only rootvowel in the preterite singular (i.e. the degree that extended to the plural in NHG when the ¡dependent fourth Ablaut-degree occurring in the preterite plural disappeared): MHG glimmenglamm, malken-malk, leschen-lasch, pflegenpflak [

(vs. o in NHG). In class II where the lexical

w] merged much earlier with the root-vowel, the non-lexicality of U appears only in the

oldest sources of OHG (Braune and Eggers 1987:47,275), in Gothic (Braune and Ebbinghaus 1981:113) and Common Germanic: e.g. got IIa pliuhan - plauh - pluhans 'flee'. Once more, historical reality validates our predictions, which have been established on purely synchronic grounds. STOSSEN - Let us now turn to the announced discussion of stoßen - stieß - gestoßen [ooii-oo]. In section 7.3, we booked this verb as a fully apophonic one, offering the apophonic reading 0 = > I = > A. Consequently, we must assume parasitic A and U in the present, and a parasitic U in the participle. Nevertheless, we did not note any parasitic U, either in the present or in the participle, promising to justify this omission later on. Two reasons make us suspicious about stoßen. First the impossibility to predict a parasitic U from the consonantal context [st

ß] in the participle. Second, stoßen is the only NHG strong verb that does not observe the

strict complementary distribution of parasitic Elements: recall from section 5.3 that only U was parasitic in the preterite, as opposed to only A in the participle. If apophonic theory really controls the NHG system of strong verbs, there is no reason to assume that one single verb should be able to violate a tough regularity of absolute generality. Apophony thus forces us to

54

Philippe Ségéral and Tobias Scheer

consider the parasitic U in the participle to be alien to the root-vowel, but does not originate in the consonantal environment either. Is there any other possible source? Consider the case of class II ( fließen - flöß - geflossen) discussed in section 9.1.1 It turned out that synchronically as well as historically an underlying structure /iw/ must be assumed, which is interpreted as [ii] since *[iw] is illegal. Hence, class II is a precedent for a lexically present U that is neither part of the apophonically relevant vowel nor of the consonantal environment, and that never appears as such on the surface (dropped in the present because of */iw/, merged with A yielding [o] in the other degrees). Thus, our theoretical framework forces us to predict a lexical structure for stoßen where / w/ is the second element of a diphthong. Accordingly, the long [oo] in the present and the participle must be the merged result of this / w/=/ U/ and the lexical vowel over which apophony operates in reality, i.e. A=[a], The synchronic structure /stAUsen/ is thus predicted for stoßen. In the present and the participle, /aw/ freely combine to [oo]. In order to derive the preterite, apophony operates over the Nucleus containing A. As A is always parasitic in the present, the input into the apophonic path is 0. Therefore, the output defining the preterite-vowel is I, yielding the sequence */iw/. Just as in the present of class II, ""/iw/ being illegal, the [w] is dropped, and compensatory lengthening ensues, resulting in long [ii]. To sum up, the prediction made by apophony is that stoßen offers an underlying diphthongal structure /au/. Now, a look at the etymology of this verb reveals exactly this diphthong for Common Germanic: stauym (cf. for example Braune and Eggers 1987:23). In this case, apophonic theory not only accounts for the Ablautvariations of NHG, but is also an instrument that correctly predicts etymologies. CLASS VI - At the end of our analysis in (37), 15 verbs that cannot be accounted for by apophony remained. These belong to three different groups. First, kommen, schinden: these verbs are notoriously irregular (cf. the above discussion). Second, saugen, saufen, schnauben (class IIa3,4). These verbs are apophonically uninterpretable. Finally, backen, schaffen, wachsen, waschen, fahren, graben, laden, schlagen, tragen, schallen. Apophonic theory points them out as aliens in the system of strong verbs. This prediction is arrived at on strictly synchronic grounds. Now it happens that these verbs are all (except schwören) and only the members of the traditional class VI. They do not represent isolated cases. Now it is a striking fact that originally, class VI never recurred to apophony. Recall that apophony is concerned with qualitative vocalic alternations. Indeed, class VI operated with a quantitative opposition (so-called "quantitative Ablaut") in the Indo-European period that later evolved into qualitative contrasts. However, as there has never been apophony in class VI, these "alien" qualitative alternations are detected by apophonic theory as such with surgical precision. Once more, apophonic theory makes a synchronic prediction that turns out to be a piece of historical evidence. This synergetic pattern of synchronic and diachronic data strongly supports all elements of the analysis that have contributed to this result.

A Generalized Theory of Ablaut: the Case of Modem German Strong Verbs

55

Given the alien character of class VI, there is no reason for apophony to account for the 10 verbs it contains. These alternations are lexicalized since their support, vowel-quantity, turned into a qualitative opposition. They are therefore not to be booked as apophonically relevant. Their absence from the set of challenging verbs brings the predictive power of apophonic theory to 153 out of 160 items, i.e. 96%. THE MYSTERY OF CLASS VII - It has long been noticed that the formerly reduplicating verbs grouped in class VII exhibit a very special Ablaut-pattern. Indeed, they present a disparate range of vowels in present and participle forms but are absolutely uniform as to the presence of a palatal element [i] in the preterite in OHG (cf. Braune and Eggers 1987:185ff, Scheer 1997 for a synopsis). This [i]-uniformity in the preterite is even more striking in NHG (cf. table (14)). The Ablaut-pattern of class VII is an OHG innovation. The morphological marker of reduplication, guaranteeing the opposition of present and preterite forms in Common Germanic, broke down in OHG. Hence, verbs that formerly relied on reduplication in order to express the present-preterite opposition had to disappear or to conform to the dominant tense-marking of that period, i.e. Ablaut. This process of integration of the formerly reduplicating verbs into the class of ablauting verbs has been described in detail and made the object of different theories (e.g. Bech 1969, 1971:61-73, van Coetsem 1990, Fulk 1987, Lüdtke 1957). Yet, "this classical mystery of Germanic morphology" (Bech 1969:3) has never received a satisfactory treatment, i.e. one that does not resort to analogical processes. In particular, the presence of the [i] in the preterite has remained unexplained. However, approaching the problem in the light of apophonic theory provides an immediate solution: apophony predicts that there must be an I contained in the preterite. This follows straightforwardly from the parasitic character of A and U in the present. As a matter of fact, any of the surviving 40 formerly reduplicating OHG verbs have "dark" vowels in their present (OHG [a,ä,ai,ou,ö,uo], e g fallan, rätan, skaidan, houwan, scrötan, wuofari). Thus, with only A and U intervening in their production, the apophonic recipe predicts that such verbs must enter the path with what remains after removing A and U, i.e. 0. The next step on the path leading to the preterite vowel is [i]. Apophonic theory based on a strictly synchronic analysis offers a very simple and straightforward account for a classical problem of Germanic diachronics. Moreover, the fact that OHG and NHG verbs, although separated by more than one millenary, still obey the same apophonic pattern shows the unbroken continuation of the Ablaut-mechanism. In particular, it shows its presence in the grammar of NHG.

12. The status of the Apophonic Path Reaching the end of our demonstration, we would like to address the question of the status to be assigned to the core component of apophonic theory, that is the Apophonic Path. All

Philippe Ségéral and Tobias Scheer

56

models of segmental representation nowadays confer a basic status to the fundamental vocalic triangle going back to Jakobson and Halle (1956) whose space is defined by the three extreme vocalic articulations i - a - u. We are not aware of a phonological model dealing with primitives such as [mid] or E. The Apophonic Path can be viewed as a development of the vocalic triangle in the sense that 1) it makes explicit the most unmarked "vowel", i.e. zero, which is more or less explicitly stated in the various models (e.g. the "cold vowel" v° in Government Phonology), 2) it relates these four fundamental units by a directionality, which in fact merely encodes universal markedness (cf. the chapter 9 of Chomsky and Halle 1968): zero for sure is the most unmarked element, [u] the most marked vowel, and [i] seems less marked than [a]. In sum, the shape of the Apophonic Path in its entirety is very natural. It would be surprising if the four objects in question were connected in some other way. It is to be expected that the naturalness of the Apophonic Path is best visible when primary linguistic gestures are performed. Let us briefly consider two such fundamental operations, that is, epenthesis and onomatopoetic formations. Cross-linguistically, the epenthetic vowel is very frequently [i] (e.g. Old Spanish, Old French, Soninké, Dida, Arabic, Serbo-Croatian). There is no reason why [i] and not, say, [u], should assume this function. Now, since "epenthesis" means "replacing nothing by something", the first step of the Apophonic Path 0=>i could provide an explanation for epenthetic [i]. Moreover, if apophony is assumed, the phenomenon called "epenthesis" loses its exceptional status: it is to be considered an ordinary derivational process. In the same way, no unorganic [i]s "fall from heaven" anymore since they are the regular output of an apophonic derivation performed on /0Λ25 Recall that lai is never parasitic, thus a possible entering Element in any language (cf. section 7.4). Onomatopoeia, another kind of synchronically active formation, makes use of fundamental phonological operations such as reduplication and vocalic alternation. Consider some onomatopoetic expressions from German and French given in (38), which are far from being exhaustive. The vocalic alternations they are built on are overt instantiations of the Apophonic Path.26 We take this fact for another significant piece of evidence confirming the presence of the apophonic path in Universal Grammar. (38) GERMANI ritsch plitsch der BiLirum ri-

zick tick

25 26

A ratsch platsch BaLarum razack tack

U

FRENCH

Butzemann (Löffelstiel) ratsch

I trie patati cahin [ε] mie

pif ding

A frac patata caha mac zacapaf ding [ε]

U

zou [u] cou [u] pouf [u] dong [5]

For further discussion about the status of linguistic objects that "fall from heaven", cf. Scheer (1996:127fl). The regularity underlying onomatopoetic expressions was already noticed by Grammont (1933).

A Generalized Theory of Ablaut: the Case of Modern German Strong Verbs

57

13. Conclusion The German case discussed in this paper contributes a major piece of evidence to apophonic theory. We have seen by which means the apophonic legibility can be guaranteed even under extremely hostile external conditions such as the complex vocalic system of German, which displays 16 vowels and 3 diphthongs on the surface. The way German maintains the apophonic legibility of its verbal system as well as its learnability is summed up in the Apophonic Theorem: entering and parasitic Elements are complementarily distributed. Under our analysis, only the vowel of the present tense form is lexically recorded, the rootvowels of the subsequent degrees being the output of apophonic derivations. This system correctly predicts the vocalization of over 90% of the 170 NHG strong verbs. Moreover, many secondary predictions made on pure synchronic grounds turn out to be well-known historical realities. In sum, the underlying representations we arrive at when following the implications of apophonic theory are identical with the recorded forms of previous stages of the language. In particular, apophony detects all and only the verbs of a whole class that was never subject to qualitative Ablaut (traditional class VI). Finally, the apophonic analysis sheds light on an old problem of German historical linguistics, that is, the integration of the formerly reduplicating verbs into Ablaut. Nevertheless, the most important result of this paper is the cross-linguistic stability of apophonic alternations. We have shown that the regularity that underlies the systems of context-free vocalic alternations exploited for grammatical purpose in languages such as Classical Arabic, Berber and Somali Ablaut also drives the alternations found in NHG strong verbs. Henceforth, the hypothesis that this apophonic regularity a = > I ==> A ==> U = > U has universal character is legitimate.

References Anderson, J.M and C.J. Ewen (1987): Principles of Dependency Phonology. Cambridge: Cambridge UniversityPress. Andrzejewski, Bogumil W. (1975): Verbs with vocalic mutation in Somali and their significance for HamitoSemitic comparative studies. In: Bynon, Theodora and James (eds) Hamito-Semitica, 361-374. The Hague: Mouton. Appleyard, David L. (1992): Vocalic Ablaut and Aspect Marking in the Verb in Agaw. In: J AAL 3.2, 126-150. Bech, Gunnar (1969): Das germanische reduplizierte Präteritum. In: Historisk-filosofiske Meddelelser 44.1, 154. - (1971): Reduplikation und Innovation. In: Bech, Gunnar: Beiträge zur genetischen idg. Verbalmorphologie, 61-73. Historisk-filosofiske Meddelelser 44.5, Kopenhagen. Bendjaballah, Sabrina (1995): Aspects du système verbal du berbère (kabyle). DEA thesis, Université Paris 7. Braune, Wilhelm and Ernst Α. Ebbinghaus (1981): Gotische Grammatik. Tübingen: Niemeyer 1981. Braune, Wilhelm and Hans Eggers (1987): Althochdeutsche Grammatik. Tübingen: Niemeyer 1987. Chekayri, Abdeliah and Tobias Scheer (1996): The apophonic origin of Glides in the verbal system of Classical Arabic. In: Lecarme and al. (1996), 62-76.

58

Philippe Ségéral and Tobias Scheer

Chomsky, Noam and Morris Halle (1968): The Sound Pattern of English. New York: Harper and Row. Clements, George Ν. (1993): Lieu d'articulation des consonnes et des voyelles: une théorie unifiée. In: Laks, Bernard and Annie Rialland (eds) L'architecture des représentations phonologiques, 101-145. Paris. van Coetsem, Frans (1990): Ablaut and Reduplication in the Germanic Verb. Heidelberg: Carl Winter Universitätsverlag. Cohen, David (1972): La mutation aspectivo-temporelle dans quelques langues couchitiques et le système verbal chamito-sémitique. In: Thomas, Jacqueline M. C., Lucien Bernot (eds) Langues et techniques; nature et société. En hommage à A. Haudricourt à l'occasion de son soixantième anniversaire, vol.I (= Approche linguistique), 57-63. Paris: Klincksieck. - (1974): Alternances vocaliques dans le système verbal couchitique et chamito-sémitique. In: Caquot, André, David Cohen (eds) Actes du Premier Congrès International de Linguistique Sémitique, 40-48. La Haye: Mouton. Dillmann, August (1907): Ethiopie Grammar. Amsterdam: Philopress. Fulk, R.D. (1987): Reduplicating verbs and their development in Northwest Germanic. In: Beiträge zur Geschichte der deutschen Sprache und Literatur 109, 159-178. Grammont, Maurice (1933): Traité de Phonétique. Paris: Delagrave. Guerssel, Mohand and Jean Lowenstamm (1994): Ablaut in Classical Arabic measure I active verbal forms. Paper presented at the second conference on Afro-Asiatic Languages, Nizza 1994. Also in: Lecarme et al. (1996):, 123-134. Harris, John and Geoff Lindsey (1995): The elements of phonological representation. In: Durand, Jacques, Francis Katamba (eds) Frontiers of Phonology. Harlow, Essex, 34-79. Hetzron Robert (1969): The verbal system of Southern Agaw. In: Near Eastern Studies 12, Univerity of California Publications. Jakobson, Roman and Morris Halle (1956): Fundamentals of Language. The Hague: Mouton. Kaye, J.D., J. Lowenstamm and J.-R. Vergnaud (1985): The internal structure of phonological representations: a theory of Charm and Government. In: Phonology Yearbook 2, 305-328. Kuiylowicz, Jerzy (1956): L'apophonie en Indo-Européen. Wroclaw. - (1968): Indogermanische Grammatik. Band II Akzent, Ablaut. Heidelberg: Carl Winter. Lecarme et al. (1996) = Lecarme, Jacqueline, Jean Lowenstamm, Ur Shlonsky (eds.) Studies in Afroasiatic Grammar. The Hague: Holland Academic Graphics. Lowenstamm, Jean (1990): Vocalic length and syllabic structure in Semitic. In: Kaye, A.S (ed) Semitic Studies in Honor ofW. Leslau on the occasion of his 85 th birthday. Wiesbaden: Harrassowitz. - (forthcoming): The beginning of the word. In: Renisson, John (ed) (forthcoming) Phonologica 1996 Liidtke, Helmut (1957): Der Ursprung des germanischen ë2 und die Reduplikationspräterita. In: Phonetica 1, 157-183. Orwin, Martin (1995): Colloquial Somali. London: Routledge. Paul, Hermann (1921): Deutsches Wörterbuch. Halle (Saale): Niemeyer 1921. Paul, Hermann, Peter Wiehl and Siegfried Grosse (1989): Mittelhochdeutsche Grammatik. Tübingen: Niemeyer 1989. Postma, Gertjan (1996): Ablaut, duplicative structures and the syntax-phonology interface. Paper presented at the Morfologiedagen, Meertensinstituut, September 1996. Schane, Sanford S. (1984): The fundamentals of Particle Phonology. In: Phonology Yearbook 1, 129-155. Scheer, Tobias (1995): Halbechte Rektion in germanischem Wandel und althochdeutscher Brechung. In: Linguistische Berichte 160, 470-511. - (1996): Une théorie de l'interaction directe entre consonnes. Doctoral dissertation, University Paris 7. - (1997): Des Ablauts gemeiner Gegner: Laryngale. In: Dhumbadji! 3.1. Ségéral, Philippe (1995): Une théorie généralisée de l'apophonie. Doctoral dissertation, University Paris 7. - (1996): L'apophonie en ge'ez. In: Lecarme et al. (1996), 360-391. Ségéral, Philippe and Tobias Scheer (1994): L'apophonie dans les verbes forts de l'allemand moderne. In: Nash, Léa, Georges Tsoulas (eds): Actes du premier colloque Langues et Grammaire, 307-322. Paris: Université Paris 8.

A Generalized Theory ofAblaut: the Case ofModern German Strong Verbs

59

-,- (1996): Modern German and Old English strong verbs: two ways of running Ablaut; Paper presented at the GGS-conference, Berlin 1996. (1997): Ablaut and the organization of Cushitic verbal systems. Paper presented at GLOW, Rabat/ Marocco, March 1997. Voigt, Rainer M. (1985): Die beiden Präfixkonjugationen des Ostkuschitischen. In: Afrika und Übersee 68, 87-104. Wurzel, Wolfgang Ullrich (1970): Studien zur deutschen Lautstruktur. Berlin: Akademie-Verlag.

Albert Ortmann (Heinrich-Heine-Universität

Düsseldorf)

The Role of [±animate] in Inflection

1. Introduction1 This article is concerned with the issue of how natural languages morphologically classify the individual referents of nouns. More specifically, the following questions are raised: which distinctions inherent to individual objects are more fundamental than, or 'superior' to, other distinctions, and which are the different ways in which languages reflect these distinctions morphologically. Due to the relevance at the interface of conceptually based semantic form and morphosyntactic structure, it is expected that a fundamental conceptual property of individuals has influence on various areas of the grammar. Therefore, the method used will be to start out from the distinctions found in gender systems and look at the role they play 'outside' of gender. The goal is to provide evidence for the following generalisations: (1) The primary distinction in grouping nouns by virtue of conceptual properties of their referents is animacy, or humanness, respectively. Other distinctions, for example those based on sex or physical shape, are secondary. (2) The fundamental role of animacy is attested not only in gender systems, but also in numerous cross-linguistic asymmetries in the grammar. The linguistic category par excellence of classifying individuals is gender, which is in fact the only nominal category that systematically reflects inherent conceptual properties of individuals (Seiler 1986, Wunderlich 1993). As far as the role of gender in linguistic research is concerned, a good deal of work has been done, most notably in Corbett (1991), where many aspects are addressed in much detail. Also, issues like the number of gender classes, or the extent to which the assignment of nouns is semantic, or the diachronic development of gender have been documented for many languages. A look at the recent literature shows that most authors agree that the study of gender is particularly fascinating, and moreover there also seems to be agreement that the complexity of gender is difficult to grasp. This view is expressed in the introductory sentence of Corbett's monograph: "Gender is the most puzzling of the grammatical cateFor comments and criticism on parts or earlier versions of this paper I am grateful to Grev Corbett, Ray Fabri, Ingrid Kaufmann, Maitin Krämer, and Dieter Wunderlich. This work was carried out in the SFB (Research Programme) 282 "Theorie des Lexikons", supported by the German Science Foundation (DFG). I also gratefully acknowledge an Academic Research Collaboration Programme, supported by the Deutsche Akademische Austauschdienst (DAAD), which enabled cooperation with the Network Morphology Group at the University of Surrey (UK).

The Role of [¿animate] in Inflection

61

gories" (I.e.: 1). This indicates that a lot of issues, including some very basic ones, are as yet far from being understood. The questions raised above are related to these issues. There are hardly any statements in the literature as to the most fundamental dimensions in grouping individuals by morphological means such as gender. Corbett is no exception; for example, he does not touch upon the question of the conceptual background of gender. Other studies, such as Seiler (1986:16ñ), do raise the issue, but make no attempt to look for a hierarchical relationship among the various distinctions expressed by gender. The present article neither aims at developing a formal account of nominal features nor at providing an ontology of object properties. Also, the points made will be fairly neutral with respect to a particular theory of functional categories. The aim of the paper is to take a crosslinguistic perspective and search for general patterns found with some, but not with other, gender features, and investigate their influence on various grammatical areas. In addition, some external motivation for the privileged morphological status of animacy will be discussed. In the following, I first illustrate the role of animacy in the organisation of gender systems (section 2). In section 3, the bulk of the paper, I show in which areas of grammar animacy plays a systematic role that is unknown for any other category of nominal classification. Some language-external factors, based on psychological and neurological findings, that lie behind these asymmetries are discussed in section 4, followed by a brief conclusion in section 5.

2. The role of animacy and humanness in gender systems In this section I will argue that animacy rather than sex is the most fundamental opposition in the organisation of gender systems, on the basis of morphological evidence. Before discussing the data, a few notes of clarification are in order. 2.1 Prerequisites I understand the term 'gender' to cover both what is meant by gender in the narrower sense, as is generally the case in the Indo-European languages, as well as what is often referred to as noun (or nominal) classes. This distinction widely correlates with that of 'overt' vs. 'covert' gender. Overt gender implies the morphological realisation of the gender on the noun itself (as in the Bantu languages), whereas covert gender refers to the absence of such a marker on the noun, the specification being evident from agreement. Similarly, there are two major systems of classification, one based on the conceptual property of animacy, or humanness, respectively, the other on the sex of individual referents (nature based' vs. 'sex-based'; see Heine 1982). Often, but by no means exclusively, the covert type is sex-based, while the overt type is animacybased. While the term gender is sometimes reserved for the former type (e.g. Seiler 1986,

62

Albert Ortmann

Sands 1995), I consider both types to be one and the same phenomenon 'gender', due to the following criteria (cf. also Corbett 1991:4f): First, the defining property of gender is agreement. This is explicitly stated in the following generalisation: (3) If a language distinguishes gender, there is always agreement with respect to gender According to (3), for a language with gender we expect some sort of covariation between a property of a noun and at least one other category. Whether such agreement is found on, for example, adjectives, demonstratives or verbs, or only between pronoun and antecedent, is a matter of variation. Second, gender is the most idiosyncratic of the functional categories of the noun. This is so because gender is either specified by virtue of inherent, as opposed to instantiated, properties (natural kinds, or 'sorts'), or it is arbitrarily specified and hence must be learnt for the noun in question. The gender specification of a given noun is invariant.2 Hence the major difference between the two semantically based nominal agreement categories gender and number is that the former is inherently fixed for a noun, whereas the latter is usually instantiated and gives rise to different word forms in the paradigm of a noun. I will refer to the more general concept of grouping nouns morphologically into kinds as 'nominal classification' (following Sands 1995:247f), or 'classification of nouns'. In addition to gender, this concept also comprises classifier systems: noun classifiers, numeral classifiers, possessive classifiers, or demonstrative classifiers, all of which share a common basis with gender classes, namely a conceptual classification of individuals. The latter, in addition, are a means for codifying grammatical relations in discourse (antecedent-pronoun), constituent structure (concord) or argument linking (see section 3). Another prerequisite concerns the correspondence between conceptual properties and the morphological category gender. On the one hand, it is very common for the category gender to classify nouns according to inherent properties of their referents in the above sense, rather than to classify them merely according to inherent grammatical properties (such as ending in a consonant or high tone, being disyllabic, or what else may constitute 'formal' systems of gender assignment; see Corbett 1991:33ñ). In every language with gender, at least for a core of the noun inventory, gender is assigned according to semantic criteria. On the other hand, it is almost equally common for languages to assign nouns to gender classes in an arbitrary manner. For example, for a noun denoting the concept of 'table', such as German Tisch, masculine gender does not reflect any semantic or conceptual property. However, I will not be concerned 2

This does not exclude the possibility for noun roots to be unspecified for gender class. This seems to be the case, for example, with many Swahili nouns, where certain conceptual differentiations may correlate with the specification of a corresponding noun class, signalled by its special prefix. However, if such a particular surface instantiation of these noun roots is chosen, the noun class foi this instantiation is fixed.

The Role of [¿animate] in Inflection

63

with the difference between 'natural' and 'grammatical' gender - for the present purpose it is not crucial to what extent the assignment of the entire inventory of nouns in a language corresponds to the semantic basis. Rather, what I am dealing with is the question of which gender assigning rules are operative in a language, in other words, which semantically interpretable categories are involved. As to the representation of the interface between semantics and morphology, following Wunderlich (1993), one can formulate gender assignment rules like the ones in (4) in order to state the predictable correspondence between semantic properties and morphosyntactic features. (4)

Default rules for gender assignment: FEMALE (x) x ANIMATE(x)

->

x

MALE (x)

->

x

HUMAN(x)

-»

x

The reading of these correspondences is 'if a noun in its semantic decomposition contains the predicate to the left of the arrow (for example, 'FEMALE' in German Stute 'mare', or 'HUMAN' in Swahili -toto 'child'), its morphosyntactic gender will be the one specified to the right'. Semantic predicates of noun referents are thus conceived as generative properties for gender. Note crucially that the rules only have default status, thus allowing for 'formal' assignment in the above sense, or for idiosyncratic specifications in lexical entries that have to be learned for the item and block the application of the otherwise semantically justified default rule.3 This can easily be extended to other properties, such as dimensional or shape properties, which are relevant, for example, for class 11/10 in Southern Bantu, containing long, thin objects. Concerning the general issue of the paper, there are hardly any statements in the literature as to the most fundamental dimensions in qualifying individuals by morphological gender. If at all, such statements have been made with regard to classifier systems (see section 3.4). Now asking which gender distinctions are primary and which are secondary, I should provide some understanding of the notion of'primary' or 'fundamental' in gender systems, i.e. in what way should one gender distinction be dominant over another? As a possible answer, one could think of implicative universale of the form "If a language has separate gender classes for feminine and masculine, it will also have classes for animate and inanimate", or "If a language has morphological gender, it will distinguish animacy (or humanness)". Unfortunately, it is quite obvious that universale of this kind would not hold, as is immediately evident from languages of the Indo-European or Afro-Asiatic families with predominantly sex-based gender systems. How3

A somewhat different conception of this correspondence is found in Pollard & Sag (1994). Here, the correspondence is conceived of as interpretation rules: if a noun has, e.g., the feature [female +], its semantics will contain an attribute FEMALE. Obviously, this format is not appropriate for languages with grammatical gender, as one would not want to assign the semantic property of being female to a noun like German Lampe 'lamp' (fem).

54

Albert Ortmann

ever, that the issue of primary and secondary gender distinctions is indeed a matter of substance is clear from languages with "rich" differentiation, that is four or more classes as the result of more than one dimension of classification. In such a system, the question arises about which dimension possibly constitutes a subdimension of which other dimension. In order to find this out, the most reliable criterion is the neutralisation of gender classes in the plural, where many languages exhibit syncretism. This is the method pursued in the following subsection. 2.2 The 'vitalist' and the 'hominist' classification In the vitalist classification, humans and animals are morphologically contrasted to non-living objects, and consequently [ianimate] is the primary distinction here. An example of 'human' being a subcase of the primary category 'animate' is Lak, a language from the Lak subgroup of Dagestan (North-East Caucasian). In the singular, [+animate] nouns are further divided into male and female humans (classes I and II, respectively) and animals (class III), while there is only one marker for the plural. (5) gives the agreement affixes, found on targets such as adjectives and verbs, for the four gender classes (after Drossard 1982:159, Corbett 1991:207f):4 (5) Lak agreement prefixes and suffixes: Singular Plural I 0/-W ^ " m a II

d-/-r

III

b-/-w

IV

d-/-r

b-/-w

l

e

humans"

"some female humans - older" "most remaining anim.s, some inan.s"

d-/-r

"residue - largely inanimates"

As the syncretism in the plural clearly shows, animals are in one class with humans in Lak. This renders [ianimate] as the primary distinction, and humans are treated as a subclass of [+animate]. The vitalist classification is also the most common one in the Amerindian languages. It is found, for example, throughout the languages of both the Algonquian (see Corbett 1991:20ff) and of the Siouan familiy. In the hominist (or 'rationalist') classification, the feature [±human] is the primary distinction, and [Ianimate] is a subdivision of [-human]. This is the case in languages where nouns for humans are further distinguished, for example into [+fem] and [-fem] (or, more hypotheti4

Here and in the following examples, numeric class labels can be redefined into semantically motivated features, along the line of the correspondence rule given above. Thus a less arbitrary specification for Cl II would be , and for Cl I. Likewise, Cl III would be , and the rest , as morphological features are per default specified as '-' in the absence of an underlying value. However, what is of relevance here is not the names of features but rather the existence of some semantic basis behind (at least some of) the classes.

The Role of [¿animate] in Inflection

65

cally, into 'adult' and 'child', which hardly seems to exist systematically). In such a system, nouns denoting non-human entities are also often subdivided, and usually grouped into animals and things. Therefore, contrary to the vitalist classification, the latter two sorts form a natural class that excludes humans. A typical representative is the gender system of Proto-East Caucasian (as reconstructed by Deeters, cited after Schmidt 1994:187). (6) Proto-East Caucasian agreement prefixes: singular plural I w-b II j

"Masculine rational" "Feminine rational"

III

"Other individuals (animals, plants, material things" "Material nouns, collectives"

IV

This overall syncretism is still found in most contemporary North-East Caucasian languages (such as Tsova-Tush, Khinalug, Archi, Godoberi; see for example Smeets 1994), no matter if the number of (and assignment to) classes that are further distinguished is maintained, as in Bezhta, or has changed, as in Chamalal (both after Drossard 1982:158): (7) Bezhta agreement prefixes: singular plural I "male persons" II

b

III

b

>b

"female persons"

II

"animals, things"

III IV V

J IV

j

Chamalal agreement prefixes: singular plural I v^ "male persons"

"things"

j --—

j/v/d—

"female persons" "other nouns distributed"

The point to be noted is that the conceptual class of humans has a semantic base in all of these languages, and often this is the only class that is semantically based. The hominist classification is also predominant among the Bantu (see 2.3) and other NigerCongo languages. It is also found in the Dravidian languages like in Tamil and Kannada. These languages are representative of languages that divide only [+human] any further. Kannada, for example, treats the masculine/feminine-distinction as a subcase of [+human], while animals and infants are assigned to what is traditionally called 'neuter', or [-human] in our terms. This is illustrated here for the third person verbal agreement forms of ma.Du 'do' and for remote personal pronouns (after Sidhar 1990:198, 221f).

Albert Ortmann

66 (8) a. Kannada 3rd person agreement: present 3.SG.MASC ma:Dutta:ne 3.SG.FEM ma:Dutta:Le 3.PL.HUMAN ma:Dutta:re 3.SG.NON-HUMAN ma:Duttade 3.PL.NON-HUMAN ma:Duttave

past ma:Dida(nu) ma:DidaLu ma:Didaru ma:Ditu ma:Didavu

b. Kannada 3rd person (remote) pronouns: mase fem human non-human neut

SR

plural

avanu avaLu avu

avaru avu

What the examples from Kannada reveal is that while the plural distinguishes only the dimension of 'rational' and 'neuter' - [ihuman] in terms of a formal feature - the singular in addition has the distinction of masculine and feminine for rationale, which constitutes a subdivision of [+human]. Turning now to the relevance of these classification types for the general issue of this paper, a striking asymmetry can be observed. One basic type of gender system features [ihuman] as the major distinction, hence the term 'hominist' classification. Often [+human] is further divided, namely according to the feature [±feminine], and in addition [-human] can be devided, and here [± animate] is the relevant feature. This hierarchical order is evident from syncretisms in the plural. By contrast, a classification type that from a logical point of view is equally well conceivable does not exist: we do not find a system where only the specification [+fem] (or [+masc]) are subdivided into [±human], that is where only females (or males) are further classified into animals vs. humans. Such a classification would be just as conceivable, since both sex-based properties and humanness logically imply animacy, and hence are conceptual subclasses of animates, while only one hierarchical order among the two is found, viz. [±feminine] below [Ihuman].5 5

An apparent exception to this generalisation is the well-known split in noun declension between animate and inanimate in the Slavic languages (Coibett 1991:1610). In the masculine gender of Russian, there is no overt accusative marking in the singular, the latter case thus being non-distinct from the unmarked nominative, e.g. dub 'oak (NOM/ACC)' vs. duba (GEN). By contrast, animate masculines are overtly marked in accusative contexts with the same suffix used for the genitive, thus sosed 'neighbour (NOM)' vs. soseda (GEN/ACC)'. The same opposition is found in the plural of feminines and neuters. However, there are good reasons against assuming animacy as a further gender distinction in Russian. It does not sufficiently fulfil the defining criterion of gender, in that its marginal role in agreement would be unexplained: the targets for animates are the same as for inanimates except for just the accusative-genitive syncretism. Hence there is no independent form in any paradigm for animate nouns. For these reasons, Corbett introduces the notions of 'subgender' and 'dependent target gender', instead of identifying six genders for Russian. On the other hand, the fact that Slavic has introduced an asymmetry based on animacy in its grammar confirms the fundamental role of this distinction also in languages whose gender systems follow other dimensions (see also 3.3).

The Role of [ianimate]

67

in Inflection

We have thus arrived at the following generalisations: (1) The concept of animacy in its wider sense, comprising both humanness and 'animacy' in the narrow sense, constitutes the primary dimension in grouping nouns into gender classes. This generalisation is, of course, not meant to marginalise the male-female opposition in gender systems: languages with gender systems based exclusively on this distinction (such as most members of Indo-European, as well as many Afro-Asiatic and Australian languages) are not deviant, or probably not even 'marked' in any sense. Rather, morphological findings such as the asymmetry in subdivisions suggest that animacy reflects a more highly ordered category than [ifeminine] (or [imasculine]). (2) Both the features [± animate] and [±human] are indispensable, as in many languages both are instantiated. However, the similarity of the two concepts must be stressed over their difference. Languages tend to separate the concepts of humans and things as extreme poles of individuation. As animals are somewhere in between these poles, they may be grouped together with either pole. Similarly, in the assignment of further referential entities such as plants, body parts or spirits, languages draw their specific borderlines rather than making a major ontological choice between one of the two dimensions. For these reasons, in the remainder of this paper we can focus on the common nature of these features and leave the more finely graded difference between vitalist and hominist classifications aside. Consequently, where no confusion can occur, I will simply use the term 'animacy' to capture the overall distinction, even if in many cases humanness may be the more precise notion. 2.3 The role of animacy in Bantu nominal classes The Bantu languages exhibit a number of peculiarities in their noun class systems that indicate the superior role of humanness as opposed to other properties. These peculiarities concern, among others, (i) the extent to which gender assignment is driven semantically, (ii) the different diachronic developments of the phonological form of the prefixes of the noun classes involved, (iii) the treatment of loan words, and (iv) psycholinguistic findings. These types of evidence will be briefly illustrated in turn. (i) Consider the reconstruction of the nominal classes of Proto-Bantu and their semantic basis (quoted from Herbert 1985:175, 178): (9) Nominal classes of Proto-Bantu: Singular Plural ba1/2 mumumi3/4 5/6 limama6 kibi7/8 lini9/10 nilini11/10 lu-

"humans" "plants" "fruits" "liquids" "inanimates" "animals" "elongated object set"

Albert Ortmann

68

12/13 14 15 15/6

kabukuku-

tu-

small objects",11 'masses" ti; 'infinitives"» », 'paired body parts',ιι "

II

ma-

Note that it is commonly assumed that in the proto language nouns were assigned to nominal classes largely according to semantic criteria. The contemporary languages differ from the proto language in that the assignment to the classes has widely shifted from semantic to phonological criteria (see also Corbett 1991:46ff and references there). This can be seen from the treatment of loan words, for which see (iii) below, as well as in diachronic changes. For example, fruits or long objects are no longer systematically treated as a noun class of their own. Moreover, the number of classes has been reduced for most contemporary languages, a condition for that change being, amongst others, phonological similarities between prefixes. For example, class 11 (lu-) merges with class 5 (li-) in many of the modern languages, the common prefix being li- (Herbert 1985:179, 1991:107). Thus the proto class 11 no longer exists in languages such as Chichewa, where nouns formerly prefixed by lu- in the singular now show li-, and ma- in the plural, hence a shift to class 5/6. By contrast, what has remained stable throughout all changes in a system otherwise prone to the loss of differentiation is the semantic basis of class 1/2, i.e. the class for humans. As Herbert (1985:179) states, "class 1/2, the 'human class', exhibits the greatest semantic integrity of any of the Bantu gender classes".6 (ii) The morphological marking of class 1/2 is kept distinct from class 3/4 in spite of the same phonological affinity in the proto language as in the case of classes 5/6 and 11/10: although the singular markers of classes 1 and 3 are identical (/MM-), for these classes there seem to be no indications of collapsing in any variety: "I am unaware of any language in which gender classes 1/2 and 3/4 collapse. [...] However, somewhat comparable mergers of non-identical prefixes are attested" (I.e.: 179). The point is even clearer from the fact that in many Bantu languages the human class is not only resistant to fusion with other classes, but in addition the prefixes for class 1 and 3 have split into phonologically distinct markers: "class 1 and 3 (*mu) prefixes develop differently in a number of languages, possibly as a functional mechanism for distinguishing human and non-human markers" (Herbert 1985:179). (iii) There are several ways of assigning borrowings to nominal classes in the Bantu languages. One common strategy is to accommodate nouns according to affinities of initial syllables and existing class prefixes. Another strategy is a designated class for loans, usually 5/6, as in Zulu but also other default classes are found in other languages. However, class 1/2 is nowhere found as the default class for borrowed nouns. But not only is this formal class reserved for humans, semantic properties are also respected for no other group of individuals than for It should be mentioned that the predominantly semantic nature of noun class assignment for Proto-Bantu is not quite uncontroversial; see Herbert (1991). Note, however, that this does not make the matter less irrelevant: even if Proto-Bantu had a large number of arbitrarily assigned nouns, the crucial point is for which classes syncretisms can arise in the course of language change and which remain stable.

The Role of [¿animate] in Inflection

69

humans, which are assigned class 1/2 rather than being assigned a class by one of the above criteria (I.e.: 181). (iv) According to Herbert, there is exactly one type of over-regularisation in first-language acquisition of classes, namely "a tendency for human-denoting nouns [arbitrarily assigned to other classes in the adult language, A O.] to migrate to class 1/2 in children's speech" (I.e. : 185; 1991:110). Likewise, "one point of similarity between the aphasie data, child data, and historical change relates to the prominence of the CI. 1/2 human class. Many errors for human referents involve their transfer to this class" (I.e. : 126). Taken together, these observations provide evidence that class 1/2 is not only maintained in all contemporary Bantu languages as the noun class of humans, but in addition is the only class with consistent semantic assignment. Although in general the assignment based on semantics changed into a system of idiosyncrasies and phonological criteria, the [±human] differentiation has the highest priority: (10) HUMAN(x)

1/2

»

phonological assignment

Note finally that in addition to the role of [+human], for many Bantu languages there is a development towards a semantic dichotomy distinguishing animate and inanimate nouns, most notably in plural syncretisms and in the behaviour of agreement targets, i.e. the concordial prefixes found with verbs and other categories (for details cf. Herbert 1985:185ff, 1991:109). It is thus clear that the situation in Bantu by and large is the same as in the Caucasian languages: regardless of many differences among related languages in their class systems (for example in the number of classes, in the division between 'humans' and 'animals', or in the presence or absence of a further differentiation based upon sex), the role of animacy/humanness as the overall morphological distinction is the only common denominator of all contemporary Bantu and Caucasian languages. What we can conclude from this subsection, then, is that the morphological discrimination of animacy, which is argued to be the most fundamental gender distinction, is perfectly stable also in gender systems that are otherwise subject to changes, most notably to reductions of the number of classes and internal reassignments. The object of the next section is to point out the role of animacy in grammatical areas other than the gender system.

3. The relevance of animacy and humanness outside of gender systems According to the above definition, gender is always an agreement feature. As a consequence, gender is found only in languages which exhibit agreement. However, the conceptual dimensions that lie behind the morphological category of gender may have systematic effects in the

70

Albert Ortmann

grammar other than featuring as an agreement category. In this section, I will highlight the role that these conceptual dimensions play in other grammatical areas. It turns out that the distinction of animate and inanimate (or human and non-human, respectively) affects areas of morphology and syntax like no other property inherent of individuals. The role of animacy in morphological splits is also illustrated at some length in Comrie (1981:178ff). However, the present study is intended to be more systematic in the following respects: in the presentation of the grammatical fields where such splits are found; the startingpoint of this paper is the role of animacy in gender systems (cf. section 2), on which Comrie does not comment at all; finally, Comrie a priori relates animacy to other dimensions of linguistic prominence, such as pronominality or grammatical person, thus arriving at "a natural human interaction among several parameters, which include animacy in the strict sense, but also [...] various means of making an entity more individuated" (I.e. :192). While it is obvious that any description of the complex field of morphosyntactic splits as such must indeed refer to an interaction of several dimensions (often somewhat misleadingly named 'animacy hierarchy', cf. subsection 3.6), I consider it essential to learn more about the status of each single one of these dimensions. Hence what follows is an in-depth study of animacy in its narrow sense, which represents the only of these dimensions that reflects properties inherent of referential individuals, as opposed to properties of linguistic individuation (such as pronominality, topicality and definiteness). One effect (often observed even in languages where gender is based upon the sex opposition) is the animacy-based dichotomy of wh-pronouns such as who - what, qui - que, wer - was etc. The wrong choice gives rise to a violation of sortal restrictions, if not to ungrammaticality. Similarly, the English 'Saxon genitive', the phrasal suffix's, is restricted to humans. Examples like these already indicate that animacy is so fundamental that it is manifest even where it plays no role in the original gender system. Let us have a look at more systematic cases now. 3.1 Linking: case marking and agreement dependent on the animacy of arguments One area where animacy plays a crucial and systematic role in numerous languages is argument linking. By linking, I understand the specific morphosyntactic means of licensing and realising the arguments of a predicate, basically the verb. Three linking devices are available in Universal Grammar, namely positional linking (i.e. configurationality), structural morphological case, and morphological agreement (for a theoretical approach see Wunderlich 1997). In the following I illustrate some cross-linguistically typical asymmetries in object linking that are based on the animacy distinction.

The Role of [¿animate] in Inflection

71

3.1.1 Systematic marking of object agreement restricted to animates As has often been pointed out, most notably by Givon (1976:159), it is typical of languages with object agreement that the object markers are restricted to object noun phrases with either human (or animate) referents or with a definite (or specific) interpretation. One such language is Swahili: (11) a ni- li -soma ki-tabu lSG-PAST-read 7-book Subject Object Ί read a book'

b. ni-

li -hi -soma

1 SG-PAST-CL7-read

Subject Ί read the book1

c. ni - li -mw -ona m- tu m-moja 1SG-PAST-CL1 -see 1-person 1-one Subject Object Ί saw one person'

ki-tabu 7-book

Object

d. ni - li -mw-ona yula m-tu lSG-PAST-CLl-see DEF 1-person Subject Object Ί saw the person'

While the presence of an agreement marker for a non-human object correlates with the interpretation of the object as [+definite], (1 la) vs. (1 lb), for human objects agreement is specified on the verb regardless of the value for definiteness, as in (1 lc,d). Thus we face a grammatical asymmetry, viz. the obligatoriness of object agreement based on the animacy of a noun phrase. To this rather well-known example numerous others, and much clearer ones, can be added. In Palauan, an Austronesian language of Micronesia, the conditions for agreement with the direct object are a combination of [+human] and [+specific] (Woolford 1995:658fï). (12a) is an example of the perfective aspect without object agreement, the object being [-human,-spec]. By contrast, there is object agreement with specific human objects, as (12b) and (12c) show: (12) a. Te-'illebed

a bilis a rengalek

3. SUB J-PERF. hit

dog

children

'The kids hit a dog / the dog / some dog(s)' c. Ak mils-terir

b. Mchelebede-terir a rengalek! hit-3PL.OBJ

children

'Hit the children!'

a retede el sensei.

I saw-3PL.OBJ three Ί saw three teachers.'

teacher

Thus objects specified as [-human] are 'ignored' in object agreement, in the sense that the verb displays no morphological specification for them.7 In Plains Cree, an Algonquian language, the direct object only triggers verbal agreement and inverse morphology if it is animate (13a,b). With inanimate objects, the paradigm for intransi-

7

Except for objects with specific singular reference, which similarly to definite objects in Swahili are respected in agreement even if they are [-human].

72

Albert Ortmann

tive verbs is used, hence the lack of object agreement in (13c,d) (see Wolfart & Carroll 1981:67ff).8 (13) a. ni-wap-am-aw 1-see-ANIM-DIR Ί see him' c. peyak waskahikan ni-wap-aht-en one house l-see-INAN-non3rd Ί see one house'

b.

ni-wap-am-aw-ak 1-see-ANIM-DIR-PL Ί see them'

d. niso waskahikan-a ni-wap-aht-en two house -PL l-see-INAN-non3rd Ί see two houses'

In this language, then, only [+animate] is relevant, and [-animate] is ignored in object linking. Let us now turn to languages where objects are linked by case. 3.1.2 Case marking only with human objects In Palauan, the same difference that governs the presence of object agreement is responsible for the presence of the case particle er, an object linker found in the imperfective aspect (Josephs 1975, Woolford 1995:660). (14) a. Ng-milengelebed a bilis. 3SG-IMPF.hit dog 'S/he hit a dog / the dogs / some dogs.'

b. A sensei a mengelebed er a rengalek. teacher hit CASE children 'The teacher is hitting the children.'

While an object with [-human,-specific] reference is not marked for case, [+human] objects always receive the case particle. The same holds for the Arabic language Maltese, where direct objects are marked for case with lil only if the object refers to [+human] individuals (Fabri 1993:117f): (15) a. Raj -t Iii Pawlu. see-lSG CASE Paul Ί saw Paul.' c. Xtraj-t il-ktieb. buy-lSG DF-book Ί bought the book.'

b. *Raj-t Pawlu.. see-lSG Paul

d. *Xtraj-t lil-il-ktieb buy-lSG CASE-DF-book

Note that the gender system of Maltese is built entirely on masculine and feminine, not on [±human]. However, it is the latter feature that governs the asymmetry in object case marking. Similarly, in Spanish the distribution of the case marker a is governed by [lanimate], but in addition the feature [±specific] is involved (data due to Teresa Parodi, p.c.): g The agreement morphology of Cree is highly complex, mainly due to its inverse system; see Fabri (1995) for a theoretical account. These complexities, however, aie of no relevance for the point to be made here.

The Role of [¿animate]

73

in Inflection

(16) a. busco una secretaria que sepa inglés search-lSG a secretary that know-SBJ.3SG -» animate, but non-specific [English c. busco un lápiz search-lSG a pencil - » non-specific, inanimate

b. busco el lápiz search-lSG the pencil -» specific, but inanimate d. busco a la vaca con una mancha search-lSG CASE the cow with a spot -> specific, animate

(16a) and (16b) show that neither [+animate] nor [+specific] alone trigger object marking by a. Consequently, all [-specific] or [-animate] NPs occur as objects without the marker, see (16c). What is required for the use of a is the conjunction of [+animate] and [+specific] as in (16d). The distribution of the case marker shows that object linking can be determined by languagespecific combinations of the feature pairs [animate] / [human] and [def] / [specific]. Although, of the four features, [±def] is the only feature that is relevant in the morphology of the noun phrase, [ispecific] and [ianimate] are necessary for an adequate description of object linking, hence they are clearly established in Spanish. 3.1.3 Agreement in ditransitive verbs Not only the conditions under which an object of a verb is specified, but also which object is specified on the verb in the case of ditransitive constructions, is often governed by animacy. In many languages, agreement in ditransitive verbs is not generally with the direct object (as one would expect if agreement were to be analysed in purely structural terms) but rather with the animate object. Typically, it is the indirect object that is animate. For example, in Palauan it is the human indirect object rather than the non-human direct object that is specified on the verb: (17) a. Ng-mils-terir a buu' a rengalek a Sabino. 3SG-gave-3PL.OBJ betel nut children Sabino 'Sabino gave the kids some betel nuts.'

(Woolford 1995 .660)

b. Ak-mils-terir a buk. lSG-gave-3PL.OBJ book Ί gave them a/the book.' Recall from 3 .1.1 that the suffix -ferir specifies the direct object of transitive verbs. Being human, however, the indirect object has priority here. In sum, when looking at object agreement and object case marking, we generally find that the more 'specific' or 'animate' a referent is, the more likely is morphological specification. This behaviour can be functionally explained by the requirement for an economic and efficient linking system. According to Plank (1980:293f), it is a common strategy "to distinctively mark direct objects only if the risk of actual ambiguity is relatively high, i.e. if they have some of the semantic or pragmatic poperties (such as animateness, topicality, definiteness) normally characteristic of subjects". Thus the need for efficient argument linking accounts of the interaction

Albert Ortmann

74

between the several grammatical dimensions the object asymmetries are based upon. Of all these dimensions (to which thematic role or pronominality could be added; see Culy 1995 for an intricate case study), animacy is the only one to reflect an inherent property of referential individuals, rather than their role in the linguistic context. 3.1.4 Subject agreement While asymmetries in object linking like the previous ones have often been mentioned in the literature, much less has been said about the different behaviour of subjects with respect to the animacy distinction. This is so because, unlike object agreement, subject agreement tends to be obligatory when it is established in a language. However, subject-verb agreement is also often restricted by animacy. This is the case, for example, in Persian. In this language, number agreement on the verb is only obligatory for [+human] subjects, whereas plural agreement with [-human] subjects particularly emphasises the concept of plurality (Windfuhr 1979:31). Similarly, in Turkish this asymmetry is based on [±animate], where plural verb agreement with inanimates is only found when the verb is widely separated from the subject in terms of linear order (Lewis 1967:246). Another case in question is Georgian, where for many speakers number agreement is also restricted to animate subjects, whereas inanimates are found with the singular, which thus serves as the grammatical default (Harris 1981:21). (18) knut-eb-i kitten-PL-NOM

gorav-en.

burt-eb-i

goravs.

roll-3PL

ball-PL-NOM

roll-3SG

'The kittens are rolling.'

'The balls are rolling.'

Another aspect of subject-verb-agreement where animacy is the crucial trigger is the application of semantically based resolution rules with conjoined noun phrases. As Corbett (1983a&b) points out, number resolution - that is a specification in which the semantics of the conjoined NP as a whole is respected - in subject-verb-agreement is basically dependent on two factors. One is word order: a conjoined subject preceding the verb is more likely to exhibit plural agreement than it would if it followed the verb. The other factor is animacy: resolution is significantly more likely with animate conjuncts. In Corbett (1983a:181), this asymmetry is documented with statistics for Medieval Spanish, German, Russian and Serbo-Croat. Corbetts findings are supported by the following lyrical examples from German: singular on the verb is possible only for the inanimate subject in (19a). (19) a. Marmor, Stein und Eisen bricht / brech-en marble, stone and iron break-3SG.PRES / break-3PL.PRES 'marble, stone and iron break'

b. Hänsel und Gretel *verirr-te / verirr-te-n sich im Wald. Jack and Jill *get.lost-PAST(3SG) / get.lost-PAST-3PL in-the forest 'Jack and Jill got lost in the forest.'

The Role of [±animate] in Inflection

75

In contrast to animacy-based asymmetries in the resolution of conjuncts, no parallel asymmetries referring to masculine and feminine are known, such that number resolution would be triggered by, say, masculine rather than by feminine, or vice versa. To sum up, we have seen a number of striking asymmetries in verbal argument linking, where the animacy of an argument governs the way it is licensed or the choice of a particular morphological specification. By contrast, there do not seem to be any linking asymmetries referring to, for example, masculine and feminine, which confirms our view of animacy as the primary dimension in grouping individuals. 3.2 Word order Animacy and humanness cross-linguistically also play an important role in a grammatical field outside of inflection, viz. the serialisation of noun phrases, or simply word order. This is done in interaction with a bunch of other properties, such as definiteness, given/new, or grammatical relations, which will be kept invariant in the following examples. Particularly suggestive is the order of objects in Sesotho, a Bantu language. Although allowing for either order between two objects as long as both are animate or both are inanimate, Sesotho generally requires an animate object phrase to precede an inanimate object. In particular, it does not matter if the animate phrase is the indirect object, as in (20a), or the dircet object, as in (20b) (cf. Siewierska 1988:56): (20) a. Ke-phehét-sé ngoná lijó. 1 sg-cooked-ASP child food Ί cooked the child food.' b. Ke-bítsliísé banâ mokéte. lsg-called children feast Ί called the children for the feast.'

* Ke-phehét-sé ngoná lijó.

* Ke-bltslítsé mokéte band

Another case of word order governed by animacy is the co-ordination of noun phrases. The linear order of the conjuncts usually is, all other things being equal, human preceding nonhuman. In Shona, for example, the reverse order gives rise to ungrammatically (Siewierska 1988:59): (21) mìirume né ìmbwà man and dog 'man and dog'

* imbwá né munirne dog and man

As already mentioned, animacy is one of several factors, grouped within a larger hierarchy (see section 3.5.), that are relevant, among other things, for the serialisation of noun phrases.9 Of all g

See Müller (1998) for an Optimality-Theoretic account of the interplay of these factors in deriving the unmarked word order in the German Mittelfeld. Under Miiller's analysis, of all linearization constraints, the

Albert Ortmann

76

properties inherent of individuals, animacy seems to be the only one that cross-linguistically shows an effect on word order. 3.3 Nominal inflection: declensional classes and pronouns An example of animacy playing a role in noun inflection is the assignment to inflectional classes. Declensional classes, as opposed to gender classes, are only relevant for the choice of inflectional allomorphs rather than being reflected in agreement. An example for the role of animacy is Cree, an Algonquian language, which uses different plural suffixes, depending on animacy (Wolfart & Carroll 1981:20f, 39). (22) ikwesis-ak girl-PL

sisip-ak

vs.

duck-PL

minis-α

asiotin-a

berry-PL

cap-PL

The suffix -ak can be characterised as [+pl,+animate], and -a as the 'elsewhere' suffix [+pl]. Similarly, in Breton the suffix -ed is by and large reserved for animate nouns (Press 1986). (23) bag

bag-où

boat

boat-PL

vs.

paotr

paotr-ed

boy

boys-PL

This contrast is remarkable since Breton, like most other Indo-European languages, has grammatical gender with masculine and feminine. The dimension of animacy hence reflects a second noun-inherent distinction, viz. an inflectional class, since in addition all nouns are specified for the feature [lanimate]. Another case in point is the so-called weak declension in German noun inflection, which constitutes a subclass of masculine nouns. This declension, illustrated with the paradigm for Junge 'boy' in (24), is characterised by schwa or zero in the nominative singular and -en in all other case/number combinations, as opposed to the 'elsewhere' inflection for [-feminine] nouns, exemplified with Tisch 'table': (24) NOM GEN DAT ACC

SG der Tisch des Tisches dem Tisch(e) den Tisch

PL die Tische der Tische den Tischen die Tische

SG der Junge des Jungen dem Jungen den Jungen

PL die Jungen der Jungen den Jungen die Jungen

In contemporary German, the weak declensional class comprises many nouns denoting animals like Bär 'bear', Löwe 'lion', Hase 'hare', Affe 'monkey', Schimpanse 'chimpanzee', nouns denoting persons like Junge 'boy', and especially those derived from country names (der) Russe '(the) Russian', (der) Franzose '(der) French', and nouns denoting persons which are produc'Animacy constraint' ([+animate] precedes [-animate]) is dominated by only two others, viz. the 'Nominative constraint' (l+nom) precedes l-nom]) and the 'Definiteness constraint1 (l+def precedes [-def]).

The Role of fiammate J in Inflection

77

tively derived by suffixes such as -at, -ist as in Demokrat 'democrat' or Sozialist 'socialist'.10 Historically, the weak declension used to comprise many more masculine nouns, especially nouns with abstract meaning. Meanwhile those nouns which do not fall into the semantic class of animates have widely been reassigned within the system, in that the -en suffix is now reanalysed as part of the root. This is illustrated by the change of Haufe 'stack', Friede 'peace', Glaube 'faith' into Haufen, Frieden, Glauben. The semantic force behind this ongoing change can be clearly seen from contrastive pairs like 'weak' Rappe 'black horse' versus 'elsewhere' Rappen '(Swiss coin)', or Franke 'Franconian' versus Franken (another coin); see Wurzel's contribution to the present volume. Therefore, [+animate] clearly constitutes an inflectional class also in German. Similar innovations establishing an animacy distinction in inflection are not hard to find. For example, in Russian noun declension there is an overt marking of the accusative case which depends on the animacy of the noun referent (see footnote 5). The same trend as in noun inflection is observed for pronominal systems. In addition to the languages with animacy-based gender systems, which always reflect this distinction in their pronominal systems, separate pronouns for humans or animates are also wide-spread in other languages. Often, such forms are clearly innovative, which is remarkable, given the extremely conservative nature of pronominal systems. For example, in some Germanic languages there is a tendency towards a form for [+human] in the pronominal system, often derived from a former plural form. Thus in Swedish, which has maintained the tripartite sex-based gender in a reduced fashion, there is an innovative distinction in [-neuter]. For nouns that denote nonhuman entities, and that were formerly masculine or feminine, the innovative form den is used for many animal-denoting nouns such as hunden 'dog' or katten 'cat, as opposed to [+human], which are pronominally referred to as han and hun, respectively. (25) han (male) hun (female) [+human,-neuter,-fem] [+human,-neuter,+fem]

den (common) [-human, -neuter]

det (neut) [-human,+neuter]

Similarly, Persian is an example of a language without gender that nevertheless has established a [±human] differentiation for third person pronouns (Amin-Madani & Lutz 1972:150). The former general personal pronoun, singular u and plural ifan, is now restricted to [+human], while for [-human] the form an (plural anha) has emerged. The same holds for Finnish, viz.

hän vs .se. The implication would be that animacy is so fundamental that it tends to be grammaticalised even in languages where it is not manifest in the 'original' gender system.11

10 11

This is not to say that every masculine noun denoting animates falls into this declension class. Two further interesting case studies are found in Pasch (1986) and Aikhenvald (1994), respectively. The former involves [±animate] verb agreement next to adjective-noun agreement according to synchronically arbitrary noun classes in Dongo (Mba < Ubangi < Niger-Congo); the latter concerns an animacy-based

Albert Ortmann

78 3.4 Classifiers

It was pointed out in section 2 that for the issue of fundamental distinctions in the grouping of nouns, classifier systems are as interesting as gender systems, since both are instances of nominal classification (cf. also the introduction and contributions in Craig 1986). In a crosslinguistic survey, Allan (1977) identifies seven conceptual dimensions which are found crosslinguistically in classifier systems. These dimensions are named the 'categories of classification'. The first of these categories is called 'material' (other categories being 'shape1, 'consistency', or 'size') and comprises "all classifiers which typically refer to the essence of the entities referred to by nouns" (l.c.:299). The superiority of animacy as opposed to distinctions based on sex or age, for example, is evidenced by the finding that "probably all languages which have material classifiers distinguish between animacy and inanimacy" (l.c.:299).12 In other words, some languages further distinguish between 'male' and 'female' in some way or other, but from what is known about classifier systems, no language has classifiers for 'male/female' or any other dimension without having classifiers for 'animate/inanimate'. This generalisation also entails that the animacy distinction is the most stable one if a classifier system is reduced in diachronic change. This is indeed what happens, for example, in Yucatec Maya. Out of originally more than eighty classifiers, only four are still in use today (Bevington 1995:44f): (26) -p'el. -tul·.

'non-living things'

-ts'it.

'long, slender objects'

'animals and human beings'

-kul:

'trees'

Just like with noun class systems as illustrated from Bantu in 2.3., in classifier languages [±animate] is still observed in a system otherwise subject to the loss of differentiation. 3.5 Typological salience hierarchies As is well-known, hierarchies that specify degrees of relevance or salience have proved to be indispensable both for stating cross-linguistic generalisations and for theoretical accounts of language-specific phenomena.13 As a matter of fact, the term 'animacy hierarchy', which is often used in the literature, calls for a note about terminology: typically, animacy in the literal sense of an inherent property of individuals, as it is understood here, is only one dimension in these hierarchies, next to others such as specificity/definiteness, person, pronominality, given vs. new information, or grammatical or thematic roles (cf. Comrie 1981:187ff, Siewierska

12 13

classifier system in addition to [tfeminine] verb prefixes and pronouns in Tatiana (Arawakan < AndeanEquatorial). Seiler ( 1986:103) comes to the same result As an example of the latter see Fabri's (1995) analysis of the inverse system of Plains Cree.

The Role of fiammate]

in Inflection

79

1988:29ff). This renders the name 'animacy hierarchy' somewhat inaccurate, at least unless there is evidence for its predominant status. Therefore, I would rather speak of 'relevance' or 'salience' hierarchies. In the typological literature, various hierarchies have been proposed. Among them are the 'personal hierarchy' (27a) (Allan 1987; see also Siewierska 1988:30), the 'semantic hierarchy' (27b) (Siverstein 1976), and the 'plurality splits' hierarchy (27c) by Smith-Stark (1974): (27) a. 1st p. > 2nd p. > 3rd p. human > higher animals > other organisms > inorganic matters > abstracts b. 1st > 2nd > 3rd human > 3rd animal > 3rd inanimate c. speaker > addressee > kin > rational > human > animate > inanimate What all of these hierarchies have in common is that elements to the left are more likely to undergo or trigger particular grammatical phenomena than elements on the right, in the sense of implications. For example, the 'plurality splits' hierarchy is to predict, for those languages in which number marking is not obligatory for all nouns, implications like the following: if a noun on some position of the scale is marked for plural, so will be all nouns to the left. The point of interest is that all of the hierarchies proposed contain some notion of animacy, often further distinguished as to include 'human' in addition. Crucially, however, an ordering such as masculine > feminine, or vice versa, is not found as a further step on such a hierarchy. This indicates most clearly that the sex-opposition is not relevant in governing grammatical asymmetries of the kind discussd throughout this section. 3.6 Summary To the numerous examples of animacy-based morphological splits others could be added ad libitum. Most of these splits would probably match one of the types introduced in the preceding subsections. I will now conclude by discussing two other minor types of splits: (i) Predicative adjective agreement: Apart from the agreement asymmetries in argument linking discussed above in 3.1, there are other domains where agreement is governed by [+human]. An example is predicative adjective agreement. Thus in Georgian only adjectives referring to humans exhibit number agreement (Tschenkéli 1958:43). (ii) Verbal markers: In Plains Cree (Algonquian), a suffix immediately following the verbal root marks verb stems for the animacy of their argument (Wolfart & Carroll 1981). (28) a. ohpik-i b. ohpik-an grow-ANIM grow-INAN 'someone grow' 'something grow'

c. waap-am see-ANlM 'see someone'

d. waap-aht see-INAN 'see something'

80

Albert Ortmann

The suffix can be seen as reflecting a change in argument structure. In addition to introducing an argument, it imposes a sortal restriction concerning the animacy of that argument. To conclude, the presentation of morphological asymmetries based on animacy in this section was to provide evidence for the general claim that animacy is the most fundamental distinction in classifying individuals in natural language. The features [tanimate] and [±human] cross-linguistically show effects on crucial aspects of the grammatical system, in particular in the field of argument linking. This is also true of many languages which do not have this feature in their gender system, like Maltese or Spanish, or of languages that do not have the category of gender at all (Palauan). I have stressed at several places that similar asymmetries based on sex are not found. Note that this is not to say that there are no sex-based asymmetries in grammar at all. In Romanian there is an interesting split in object clitics: in the perfect tense, the weak masculine pronoun is procliticised to the auxiliary, while the corresponding feminine form is encliticised to the main verb participle: l-arn vazut vs. am vazut-o Ί have seen him/her' (Alexandra Popescu, p.c.). Similarly, in Alawa, an Australian Aboriginal language, according to Sands (1995:260) "only the feminine class takes adjective agreement while nouns belonging to the masculine class take no class affixes on modifiers". However, these are the most drastic examples I am aware of, and they can hardly be seen as challenging the point made in this section. In particular, asymmetries in the core areas of grammar, such as argument linking, are only based on animacy. This can be regarded as evidence for the claim that the animacy distinction is the primary one in grouping individuals according to inherent semantic properties.

4. External motivation for the privileged status of [±human] and [±animate] In the preceding sections I have presented a good deal of data from different languages to give a systematic impression of the extent to which inflection is governed cross-linguistically by the concept of animacy. I now turn to the issue of whether the asymmetries based on animacy found in language reflect a status of this concept 'in the world', i.e. outside language, that exceeds the status of other concepts. For this purpose, I start out from ontological considerations and then discuss psychological evidence for the importance of the concept of animacy. To start with, from an ontological perspective it is clear, if not trivial, that the animacy distinction has a higher status than any based on sex. Just as colours, for example, are properties only of physical objects, the range of objects for which the concepts male and female can be distinguished is necessarily that of animate entities.14

14

This does not imply, of course, that for any language a noun morphologically treated as masculine or feminine always refers to an animate object. Rather, as I have pointed out in 2.1, the focus has to be set on those parts of the noun inventory to which gender assignment rules apply, in other words, where a systematic correspondence between linguistic and real-world properties is observed.

The Role of [¿animate] in Inflection

81

The reason why animacy in general and humanness in particular play such a fundamental role can furthermore be found in what is sometimes called the 'anthropocentric' character of language. After all, it is humans, and only humans, that use speech. Far from being trivial, this 'observation' accounts, for example, for the fact that the use of first and second person is restricted to human referents. It provides a simple rationale for the hominist classification and for phenomena where [±human] marks the split in the dimension of animacy (cf. subsections 2.2 and 3.1). There is ample evidence that both the hominist and the vitalist classifications reflect equally fundamental distinctions in the cognitive system. Of particular interest is the study of the development of cognition in infancy and early childhood. Next to the beginning of the separation between self and non-self, the differentiation between human and non-human objects is one of the first distinctions of social cognition to be gradually acquired. A stronger interest in people as opposed to inanimate objects is found as early as in the first three months. Though not being completed, the differentiation is further developed by the end of infancy, that is at the age of two; for example, persons are assigned abilities such as emotions and the perception of objects by that age (Flavell 1977:49-60). As for the stage of early childhood, recent experiments indicate striking asymmetries in the cognitive representation of animate and inanimate objects. Gelman & Markman (1987) find that three- and four-year-old children are capable of inductive inferences from one object to another, on the basis of either category membership or perceptual similarities. They observe the following asymmetry: with animates, children tend to draw inferences to objects of the same category, regardless of differences in outer appearance. For example, after learning that a small brown snake lays eggs, they are likely to infer this fact to a picture of a cobra. By contrast, for inanimates, more inferences are drawn to objects of similar visual appearance (i.e. colour, position) rather than to those of the same category; for example, from brown sugar to sand instead of sugar cubes. A different experiment, reported in Keil (1989: chap. 9), involves transformations on (pictures of) animals and artefacts in such a way that an object is turned into a different object. After these operations, subjects where asked to which kind the result of the operation belonged. Children of different age were ready to accept that, for example, a former coffee-pot is now a birdfeeder, or a garbage can is now a chair. However, even kindergarten children clearly refused to accept that one biological kind is turned into another, for example a raccoon into a skunk, or a chicken into a turkey, regardless of its new outer appearence. Further evidence for the importance of the animacy concept in our cognitive system comes from recent research in neurology. It has often been pointed out that brain damage can cause selective impairments of cognitive functions. One such selective impairment concerns the semantic memory, where concepts, in particular words and their meanings, are stored. Sometimes even more selective disturbances affect only some categories of semantic knowledge but not others. These highly specific disturbances generally seem to involve animacy. Warrington

82

Albert Ortmann

& Shallice (1984) have investigated patients with cerebral lesions who were found to have specific difficulties in identifying living things, while the knowledge of inanimate objects is preserved; another case study is offered in McCarthy & Warrington (1988). Hence the existence of category-specific memory loss and the categories involved indicates that our semantic knowledge is divided into animates and inanimates, on the top of further subdivisions (see also Hodges & Greene 1995:153, 163ff and references there). Evidence of the kind discussed here indicates that the privileged status of the concept of animacy in natural language, as opposed to other properties of individual objects, has a cognitive basis. In particular, the linguistic primacy of animacy reflects the neurological organisation of semantic knowledge.

5. Conclusion In this paper I have provided evidence for the fact that the concept of animacy constitutes the most fundamental distinction for natural languages in grouping nouns according to inherent properties of their referents. In particular, the dimension of animacy is superior to distinctions based on sex, or physical properties such as shape. The first piece of evidence I have given comes from cross-linguistic patterns in the organisation of gender systems. I have shown that, in noun class systems with more class distinctions in the singular than in the plural, it is always the dimension of animacy, if any, that is respected in the plural, while other dimensions may be collapsed resulting in morphological syncretism. Moreover, the case of Bantu indicates that morphological classes for humans are much more stable than other classes with respect to diachronic change and other (e.g. psycholinguistic) criteria. Even languages that lack gender altogether tend towards a differentiation based on animacy. This is most notably manifest in languages that use classifiers for grammatical areas such as counting or possession. Moreover, countless asymmetries in the grammar show that the concept of animacy plays a crucial role also in languages with gender systems based on sex. The major types of asymmetry are: (1) overt morphological specification in argument linking, such as case marking or agreement; (2) word order phenomena, with the universal tendency of animate noun phrases preceding inanimate ones; (3) specific forms for pronouns or inflectional affixes, yielding declensional classes; (4) classifier systems; (5) phenomena governed by salience hierarchies. By contrast, corresponding asymmetries based on shape or sex distinctions are hardly found cross-linguistically; features like [±feminine] play virtually no role in morphology and syntax outside of the gender system, indicating that no gender feature is of similar importance for natural language grammar as [±animate].

The Role of [¿animate] in Inflection

83

The ubiquitous occurrence of morphological splits based on animacy and humanness reflects the ontological status of this dimension, which follows from the anthropocentric nature of language. Moreover, it corresponds to the importance of animacy in cognition as well as in the structure of semantic knowledge in the brain, as is clearly indicated by recent findings in the fields of cognitive development and clinical neurology.

References Aikhenvald, Alexandra Y. (1994) Classifiers in Tariana. Anthropological Linguistics 36, 407-465 Allan, Keith (1977) Classifiers. Language 53, 285-311 - (1987) Hierarchies and the choice of left conjuncts. Linguistics 23, 51-77 Bevington, Gary (1995) Maya for Travelers and Students. A Guide to Language and Culture in Yucatan. Austin: The University of Texas Press Comrie, Bernard (1981) Language Universals and Linguistic Typology. Oxford: Basil Blackwell Coibett, Greville G. (1983a) Hierarchies, Targets and Controllers. Agreement Patterns in Slavic. London: Croom Helm - (1983b) Resolution Rules: agreement in person, number and gender. In: Gazdar, Gerald, Ewan Klein & Geoffrey Pullum (eds.) Order, Concord and Constituency, 175-206. Dordrecht: Foris - (1991) Gender. Cambridge: Cambridge University Press Craig, Colette (1986, ed.) Noun Classes and Categorization. Amsterdam: Benjamins Culy, Christopher (1995) Ambiguity and Case Marking in Donno So (Dogon). In: Akinlabi, Akinbiyi (ed.) Theoretical Approaches to African Linguistics, 47-58. Trenton, N.J. : Africa World Press Drossard, Werner (1982) Nominalklassifikation in ostkaukasischen Sprachen. In: Seiler, Hansjakob & Franz Josef Stachowiak (eds.) APPREHENSION. Das sprachliche Erfassen von Gegenständen. Teil II: Die Techniken und ihr Zusammenhang in Einzelsprachen, 155-178. Tübingen: Narr Fabri, Ray (1993) Kongruenz und die Grammatik des Maltesischen. Tübingen: Niemeyer. - (1995) The inverse morphology of Plains Cree (Algonquian). In: Booij, Geert & Jaap van Marie (eds) Yearbook of Morphology, 17-42. Dordrecht: Kluwer Flavell, John H. (1977): Cognitive Development. Englewood Cliffs, N.J.: Prentice Hall Gelman, Susan A. & Ellen M. Markman (1987) Young Children's Inductions from Natural Kinds: The Role of Categories and Appearances. Child Development 58, 1532-1541 Givón, Talmy (1976) Topic, Pronoun, and Grammatical Agreement. In: Li, Cahrles N. (ed.) Subject and Topic. New York: Academis Press Heine, Bernd (1982) African Noun Class Systems. In: Seiler, Hansjakob & Christian Lehmann (eds.) APPREHENSION. Das sprachliche Erfassen von Gegenständen. Teil I: Bereich und Ordnung der Phänomene, 189-216. Tübingen: NanHerbert, Robert K. (1985) Gender systems and semanticity: two case histories from Bantu. In: Fisiak, Jacek (ed.) Historical Semantics - Historical Word-Formation, 171-197. Amsterdam: Mouton - (1991) Patterns in Language Change, Acquisition and Dissolution: Noun Prefixes and Concords in Bantu. Anthropological Linguistics 33, 103-134 Hodges, J.R. & J.D.W. Greene (1995) Disorders in Memory. In: Kennard, Christopher (ed.) Recent Advances in Clinical Neurology No. 8, 151-169. Edinburgh: Churchill Livingstone Josephs, Lewis S. (1975) Palauan Reference Grammar. Honululu: The University Press of Hawaii (= PALI Language Texts: Micronesia) Keil, F. (1989): Concepts, Kinds, and Cognitive Development. Cambridge, MA: MIT Press Lewis, G.L. (1967) Turkish Grammar. Oxford: Clarendon McCarthy, Rosaleen A. & E.K. Warrington (1988): Evidence for modality-specific meaning systems in the brain. Nature 334, 428-430

84

Albert Ortmann

Müller, Gereon (1998) German Word Order and Optimality Theory. Ms., University of Stuttgart Pasch, Helma (1986) Die Mba-Sprachen. Die Nomionalklassensysteme und die genetische Gliederung einer Gruppe von Ubangi-Sprachen. Hamburg: Buske Plank, Frans (1980) Encoding grammatical relations: acceptable and unacceptable non-distinctions. In: Fisiak, Jacob (ed.) Historical Morphology, 289-325. The Hague: Mouton Pollard, Carl & Ivan A. Sag (1994) Head-driven Phrase Structure Grammar. Chicago & London: The University of Chicago Press Press, Ian (1986) A Grammar of Modern Breton. Berlin: Mouton de Gruyter Sands, Kathrin (1995) Noun Classification in Australian Languages. Anthropological Linguistics 37, 247 346 Schmidt, Karl Horst (1994) Class Inflection and Related Categories in the Caucasus. In: Aronson, Howard I. (ed.) Non-Slavic Languages of the USSR. Papers From The Fourth Conference, 185-192. Columbus (Ohio): Slavica Publishers Seiler, Hansjakob (1986) APPREHENSION. Language, Object, and Order. Part III: The Universal Dimension of Apprehension. Tübingen: Narr Siewierska, Anna (1988) Word Order Rules. London: CroomHelm Silverstein, M. (1976) Hierarchy of Features and ergativity. In: Dixon, R.M.W (ed.) Grammatical Categories in Australian Languages, 112-171. Canberra: Australian Institute of Aboriginal Studies Smeets, Rieks (1994, ed.): The indigenous languages of the Caucasus, Vol. 4: North East Caucasian languages, Part 2, The three Nalch languages and six minor Lezgian languages. Delmar, N.Y.: Caravan Smith-Stark, T. Cedric (1974) The plurality split. Chicago Linguistics Society 10, 657-671 Sridhar, S.D. (1990) Kannada. London: Routledge (= Descriptive grammars series) Tschenkéli, Kita (1958) Einführung in die Georgische Sprache. Band I: Theoretischer Teil. Zürich: Amirani Walter, Heribert (1982) Genus- und Nominalklassensysteme und die Dimension der Apprehension. In: Seiler, Hansjakob & Christian Lehmann (eds.) APPREHENSION. Das sprachliche Erfassen von Gegenständen. Teil I: Bereich und Ordnung der Phänomene, 217-228. Tübingen: Narr. Warrington, E.K. & T. Shallice (1984): Category specific semantic impairment. Brain 107, 829-854 Windfuhr, Gemot L. (1979) Persian Grammar. History and State of its Study. The Hague: Mouton. (= Trends in Linguistics) Wolfart, Christoph H. & Janet F. Carroll (1981) Meet Cree: A Guide to the Cree Language. Lincoln: University of Nebraska Press Woolford, Ellen (1995) Object Agreement in Paluan: Specificity, Humanness, Economy and Optimality. In: Beckman, Jill, Suzanne Urbanczyk & Laura Walsh (eds.): Papers in Optimality Theory, 655-700. Amherst: GSLA. (= University of Massachusetts Occasional Papers in Linguistics 18) Wunderlich, Dieter (1993) Funktionale Kategorien im Lexikon. In: Beckmann, Frank & Gerhard Heyer (eds.) Theorie und Praxis des Lexikons, 54-73. Berlin: de Gruyter - (1997) Cause and the structure of verbs. Linguistic Inquiry 28, 27-68 Wurzel, Wolfgang Ullrich (1998) Drei Ebenen der Struktur von Flexionsparadigmen, (this volume)

Teresa Parodi (University of Cambridge)

Aspects of Clitic Doubling and Clitic Clusters in Spanish1

Introduction The paper provides evidence for the analysis of Romance clitics as agreement markers, in accordance with analyses such as Rizzi 1986, Kaiser 1992, Kaiser & Meisel 1992, Suñer 1988, Franco 1993, Cummins & Roberge 1994, Bonnet 1995, Uriagereka 1995, Torrego 1995, Sportiche 1996. The evidence considered relates two areas: clitic doubling and clitic clusters. As regards the former, it is seen as a manifestation of agreement, conditioned by a case hierarchy on the one hand and a specificity hierarchy on the other. The same hierarchies appear to be ruling the possible combinations of clitics in clusters. The focus will be set on two varieties of Spanish, Standard as well as Río de la Plata Spanish. Clitic pronouns have a special role. In older generative analyses (for example Kayne 1975, Strozer 1976, among others) they have been treated as arguments, even though their behaviour differs from that of "normal" arguments in several ways. One of these differences concerns the position of clitics in the sentence: in Spanish (and Italian) they occur pre- or postverbally, depending on whether the verb is finite or non-finite, in French they occur preverbally, as oposed to arguments which occur postverbally. As opposed to full NPs, which can occur at different positions in the sentence, the position of clitics is fixed. Arguments can combine freely with one another, provided the argument structure of the verb is observed; clitics, however, cannot. Apart from observing the argument structure of the verb, clitic clusters are subject to further conditions, as will be shown below. Early analyses dealt with the question whether clitics were moved from an argument position to the one where they appear on the surface (Kayne 1975) or were base-generated in this position, to the left of the finite verb (Strozer 1976). More recent analyses address the question whether clitics are better accounted for as a category of their own, possibly an agreement element (Suñer 1988, Franco 1993, Cummins & Roberge 1994, Bonet 1995a & b, Uriagereka 1995a & b, Torrego 1995, Sportiche 1996). The sort of projection and its characteristics varies according to the different authors. Of special interest for the understanding of clitics are the phenomena of clitic doubling and clitic clustering. It is mainly clitic doubling, the co-occurrence of a clitic and a coreferent full

1

This paper originated during my work at the project "Lexical Foundation of Agreement" at the University of Dusseldorf. I am especially indebted to Alexandra Popescu for the discussion which brought about many of the ideas in the paper. I would also like to thank the members of the project, namely Emanuela Canclini, Ray Fabri, Birgit Gerlach, Albert Ortmann, Dieter Wunderlich, for helpful comments and criticism.

Teresa Parodi

86

NP, which has led to the analysis of clitics as instances of agreement. Clitic clusters, i.e. the cooccurrence of more than one clitic in a sequence, have been considered a template with an idiosyncratic order ever since Perlmutter's (1970) analysis. In this paper I will analyse clitic doubling in two varieties of Spanish as an instance of agreement, exploring especially the conditions under which it comes about. In a further step the discussion will extend to the consequences of the agreement assumption on clitic clusters, i.e. on the ordering and the restrictions on the possible combinations. Clitic doubling is going to be dealt with in the following section, clitic clusters in section 2.

1. Conditions on clitic doubling 1.1 The facts In this section I will explore the conditions under which clitic doubling, i.e. the co-occurrence of a clitic plus a coreferent lexical or pronominal NP, is found. To be examined are case (accusative and dative), the syntactic category of the double (full NP or pronoun) and its feature values for animacy, definiteness and specificity. I consider here definiteness as grammatical information with a morphological correlate, while specificity is a semantic issue related to reference (Enç 1991). Clitics can always occur on their own (cf. la, 2a for accusative and dative respectively), which is expected if they are pronouns. In fact what cannot occur on its own is a pronominal object (lb, 2b). In other words, if the accusative or dative object of a verb is pronominal it must be doubled by a clitic (3c, 4c).The presence of the full pronoun double is the preferred option. This holds for both varieties under consideration here. (1) accus, a. la veo 3accfs see. Is Ί see her'

(2) dative

2

b. * veo see. Is

a ella K2she

c. la 3accfs

veo see. Is

a ella Κ she

a. le doy la carta 3dat.s give.Is the letter Ί give him the letter1

Κ = case marker

Clitic Doubling and Clitic Clusters in Spanish

87

b. *doy la carta a él give. 1 s the letter Κ he Notice, though, that in all of these examples the object, i.e. the double, is specific. In the case of a non specific object we can observe a difference between accusative and dative objects: accusative does not allow doubling by a clitic, dative does (3a-b). (3)

a. (*lo) 3acc.s

veo

a alguien

see. Is

Κ someone

Ί see someone' b. le 3dat.s

doy

la carta

a alguien

give. Is

the letter

Κ someone

Ί give the letter to someone' This asymmetry between dative and accusative objects can also be observed when the object is a full NP. While clitic doubling with dative objects seems to be unrestricted, it is restricted by further conditions in the case of accusative objects as I will show shortly. In the examples under (4) the dative object varies in definiteness and specificity and the sentence is always grammatical. By definite I mean the morphological marking by means of an article; by specific I mean referring. Although it is possible to have just the full NP as in (4a), doubling by a clitic is the strongly preferred option in both varieties under consideration. (4)

a. doy la carta al vecino/a un vecino give. Is the letter K-the neighbour/a neighbour Ί give the letter to the/a neighbour' b. +definite, + specific le doy la carta al vecino 3dat.s give. Is the letter K-the neighbour Ί give the letter to the neighbour1 c. +definit, -specific

le doy

la carta al primer vecino que aparezca K-the first neighbour who might turn up

d. -définit, +specific

le doy

la carta a un vecino que lleva gafas Κ a neighbour who wears glasses

e. -définit, -specific

le doy

la carta a un vecino Κ a neighbour Ί give the letter to a neighbour'

Teresa Parodi

88

In the case of accusative objects the realisation of a full NP together with a clitic double depends on whether the NP is definite or indefinite, doubling being disallowed with the latter in both varieties (cf. 5). As we can see from the examples, the specifity of the object NP does not make any difference. Animacy does not make any difference either.

(5) +specific, +animate a. veo a una mujer see. Is Κ a woman Ί see a woman' *la veo 3acc.fs see. Is Ί see a woman' -animate

a una mujer Κ a woman

(Spain, RPlata)

c. veo una mesa see. Is atable Ί see a table' d. *la veo 3accfs see. Is Ί see a table'

una mesa atable

(Spain, RPlata)

-specific, +animate e. busco una mujer look for. 1 s a woman Ί am looking for a woman' f. *la busco a una mujer que sepa inglés 3accfs look for. Is a woman who know.3s.subj English Ί am looking for a woman who would know English' -animate

(Spain, RPlata)

g. *la busco una mesa que tenga tres patas 3accfs look for. Is atable which have. 3s. subj three legs Ί am looking for a table which would have three legs' (Spain, RPlata)

As regards definite NPs, clitic doubling is excluded in Standard Spanish and allowed in the Rio de la Plata variety, provided, however, that the double be specific. Clitic doubling with a definite but non-specific NP is disallowed in both varieties. Examples (6g) and (6h) would only be possible in a contrastive, marked utterance.

Clitic Doubling and Clitic Clusters in Spanish

89

(6) +specific, +animate a. veo a la mujer see. Is Κ the woman Ί see the woman' b. la veo 3accfs see. Is Ί see her' c. la veo a la mujer 3accfs see. Is the woman Ί see the woman' -animate

(Spain *)

d. compro la mesa buy. Is the table Ί buy the table' e. la compro 3accfs buy. Is Ί buy it' f. la compro la mesa her buy. Is the table Ί buy the table'

-specific, +animate

-animate

(Spain *)

*la busco a la mujer que sepa inglés 3accfs look for. Is the woman who know.3s.subj. English Ί am looking for the woman who would know English'

(Spain, RPlata)

h. *la compro la mesa que tenga tres patas 3accfs buy.ls the table which have.3s.subj. three legs Ί (will) buy the table which would have three legs'

(Spain, RPlata)

The following table summarises the observations: (7) Clitic doubling in the dialects under consideration Standard Spanish obligatory

Rio de la Plata Spanish

acc, pronoun

acc, pronoun

dat, pronoun

dat, pronoun dat, full NP

optional not allowed

dat, full NP

acc, +definite, +specific NP

acc NP

acc, +definite, -specific NP acc, -definite NP

90

Teresa Parodi

In this table both varieties are compared with respect to the obligatoriness, optionality or impossibility of the occurrence of a clitic in addition to a coreferential NP. An accusative or a dative pronoun obligatorily require the additional presence of the clitic in both varieties. The same holds for the dative realised as a full NP in Río de la Plata Spanish: the additional presence of a clitic is obligatory. Standard Spanish, on the other hand, allows for clitic doubling with a full NP, although not obligatorily. Río de la Plata Spanish allows for a definite NP to occur doubled by an (optional) clitic, which is disallowed in Standard Spanish. Indefinite NPs as accusative objects cannot be doubled by clitics. The conditions which play a role in clitic doubling can be summarised as follows: a. case and argument structure: clitic doubling is realised with dative rather than with accusative objects (cf. 2c-d; 3c, 2f; 2j, 21, 2m vs 4a-c, 5a-d). Jackendoff (1990), Dowty (1991), Joppen and Wunderlich (1995), Wunderlich (1996), Kiparsky (1997) assume a hierarchy of semantic roles, which maps as a whole onto the hierarchy of syntactic functions. For the hierarchy itself I follow Larson (1988), whose tests show that in terms of argument structure the recipient role ranges over the theme role. This results in the following hierarchy of the argument structure and the respective mapping onto syntax: (8)

actor/agent nominative

> goal/recipient > patient/theme > dative object > accusative object

b. definiteness and specificity: (i) clitic with an indefinite accusative object is not possible either in Standard or in Rio de la Plata Spanish; (ii) clitic doubling with the accusative object is possible in Río de la Plata Spanish as opposed to Standard Spanish, only, however, if the object is specific (cf. 6a-g). Standard Spanish does not allow for clitic doubling with accusative objects if these are realised as a full NP. c. pronoun or full NP: clitic doubling with accusative objects is possible in both varieties under consideration if the object is a pronoun, and it is disallowed in Standard Spanish with a full NP. The conditions (b) and (c) can be related to an animacy hierarchy of the sort discussed by Comrie (1981:178-193) and Silverstein (1976). (9)

1 + 2 pronoun > 3 pronoun > full NP/ animate > full NP/ inanimate

In addition to animacy Comrie includes in this scale other factors such as definiteness and specificity as a means of individuating an entity. Definiteness has a morphological correlation: in the languages under consideration there is a distinction between definite and indefinite articles. Specificity is not marked morphologically, it is related to reference. It might be a further

Clitic Doubling and Clitic Clusters in Spanish

91

differentiation which cross-cuts the scale at this stage. It is not a subcategory of definiteness, because definite and indefinite NPs can be both specific and non-specific. A further distinction which cross-cuts the human/non-human or animate/inanimate opposition according to Comrie (1981:188) is the pronoun vs non- pronoun distinction. It is difficult to relate the distinction pronoun/non pronoun to an animacy hierarchy in its literal sense, since pronouns do not necessarily have animate reference and a pronoun with inanimate reference ranks higher in the hierarchy than a noun whose referent is high in animacy. For all of these reasons I think the relevant hierarchy should be called more appropriately the specificity hierarchy and stand as follows: (10) 1 + 2 pronoun > 3 pronoun > full NP/ animate > full NP/ inanimate full NP/ definite > full NP/ indefinite / specific > / non specific The claim I want to make is that the hierarchy is implicational, with its left end allowing for clitic doubling and its right end disallowing it. The varieties of Spanish analysed in the present paper represent different segments of this hierarchy: Standard Spanish only allows for doubling with pronouns, whereas Río de la Plata Spanish goes one step further in this hierarchy and also allows doubling with NPs which are definite and specific. 1.2 The analysis Different analyses of clitic doubling have been offered. Most recent analyses (Suñer 1988, Franco 1993, Uriagereka 1995a, Sportiche 1996, Torrego 1995) consider doubling to be an instance of (object) agreement. I understand agreement here as feature-matching between a category X and a category Y in a linking relation (Franco 1993:285). In the preceding section we have seen that the relevant features in clitic doubling are case (dative or accusative) and argument structure on the one hand (hierarchy of thematic roles), and a specificity hierarchy on the other. This hierarchy has been seen to include subcategories such as the pronoun/non-pronoun distinction, definiteness and specificity. The central condition for doubling to be grammatical is in Suñer's (1988) and Franco's (1993) analyses the feature matching between clitic and double, which Suñer formulates as the Matching Principle. This Principle stipulates that the clitic should match the features of the doubled NP. Observing that there is not a one-to-one correspondence in all cases, Franco modifies the feature matching to feature "harmony" (Feature Harmony Principle, Franco 1993:291ff). The problems Franco (1993) finds are basically related to the feature specificity. The fact that dative objects can be doubled even if the NP is non-specific violates in Franco's eyes the Feature Matching Principle. But Suñer (1988) states that the features are irrelevant for the da-

Teresa Parodi

92

tive: doubling is universal in this case. Suñer's statement makes sense if one sees the whole issue in the perspective of a case and argument hierarchy. In this perspective, and considering furthermore doubling as an instance of agreement, we end up with unrestricted agreement for the two higher ranking cases, nominative and dative, whereas agreement in the accusative is subject to certain conditions. In Standard Spanish accusative object clitics can only be doubled by pronouns. In Río de la Plata Spanish the double can be a full NP, but this has to be specific; the specificity of the NP is required for the doubling to be grammatical, clitics being inherently definite and specific. Franco (1993) claims that in Río de la Plata Spanish clitic doubling is disallowed if the nominal object is inanimate and indefinite; based on this he tries to establish a relationship between the animacy hierarchy and agreement. However, as the examples discussed in the previous section show, plain animacy is not an issue in Río de la Plata Spanish, but rather specificity: non-specific objects disallow clitic doubling. This can be explained by the Feature Matching Principle: the doubling of non specific objects is disallowed because the features of clitic and double do not match. A further issue which needs an explanation is why clitic doubling with dative objects is unrestricted while clitic doubling with accusative is not. Sportiche (1996) interprets dative clitics as instances of agreement, while accusative clitics are licensers of specificity. The interpretation for the dative and accusative is certainly correct. It is, however, possible to interpret the doubling with dative and accusative not as different phenomena, but rather as different points on a scale: an agreement scale which goes from nominative over dative to accusative. A nominative NP agrees with verbal suffixes in the features of person, case and number. A dative object agrees with the corresponding agreement marker, the dative clitic, in person, case and number. An accusative object requires additionally a specificity feature on the argument of the verb. Differences in the properties of 1st and 2nd person clitics, on the one hand, and 3rd person clitics, on the other, are claimed by Uriagereka (1995) to result in differences in doubling. Strong clitics, i.e. 1st and 2nd person clitics, are considered to be "neutralized maximal projections that can behave either as heads or as fully projected phrases - but do not have the internal structure of an intermediate level necessary for a specifier" (1995a: 112); as a consequence they are not allowed to double. In this respect they differ from 3rd person clitics, which are the only ones which make the doubling of a full pronoun necessary. He illustrates this with examples such as (1 la) vs. (1 lb) (his examples 3la & d): (11) a. (*te) me entregaron a ti 2dat.s lacc.s gave.3pl to you '(they) gave me to you' b. te lo entregaron a ti 2dat.s 3acc.ms gave.3pl to you '(they) gave it to you'

93

Clitic Doubling and Clitic Clusters in Spanish

These examples, though, are related, not to doubling in the sense discussed up to this point - in Uriagereka's as well as in this paper -, but to the co-occurrence of two clitics together. I have shown above that the presence of a clitic in a dative context like (11a) is obligatory: the sentence is ungrammatical if only the full pronoun is realised. In other words, the doubling of a 1st or a 2nd person clitic plus a 1st or a 2nd person strong pronoun is not a problem at all, as (1 lb), one of Uriagereka's examples, as well as the following under (12) show: (12) accusative

me visita lacc.s visit.3s 's/he visits me'

a mi Κ me

The difficulty is not the doubling itself, but the simultaneous presence of 1st and 2nd person clitics, i.e. there is actually a restriction on clustering, an issue which will be discussed in section 2. 1.3 Other Romance languages: Romanian, Colloquial French, Portuguese The analysis of clitics as agreement markers and the conditions regulating clitic doubling are supported by evidence from other Romance languages, such as Romanian, Colloquial French and Portuguese. According to Popescu (1997) clitic doubling in Romanian is conditioned by the features [+/-topic], [+/-human] and [+/-specific], as well as by case and the syntactic category of the double. In the case of the accusative object, the clitic co-occurs with a pronominal if this carries the features [+specific] or [+human]. When a full NP is involved, the feature [+/topic] counts as well: is [-topic], doubling with a non topicalised full NP takes place if this one is [+specific], [+human]. For a topicalised full NP [+specific] is enough for doubling to take place. In the case of the dative object, the doubling of a non topicalised full NP or pronoun requires these to be [+specific]; topicalised dative objects, whether full NP or pronoun, are doubled without further requirements. As in Spanish, the doubling of definite and indefinite NPs is subject to the specificity feature: a [+specific] NP can be doubled, a [-specific] one cannot. (13) Clitic doubling in Romanian according to Popescu (1997:52 and 57) +topic ( 0 cl-V S)

accusative

NP

+specific

-specific

+

-

NP/ pronoun

+specific

-specific

+human

-human

+

-

-

+

pronoun dative

-topic (S cl-V 0 )

+

+

-

Teresa Parodi

94

Popescu (1997:60f) observes that Romanian clitics display the morphological behaviour of inflection markers insofar as they co-occur with NPs and have agreement information (gender, number, person). They even allow pro-drop as strong inflection affixes do. A further interesting fact about Romanian is that doubling started with the dative, as can be seen in texts from the 17th and 18th century (Popescu 1997): this coincides with the claim made in the present paper that the case hierarchy represents an agreement scale as well and that Standard and Rio de la Plata Spanish rank on different slots on it. Kaiser (1992) analyses clitic pronouns in French and Portuguese. With respect to French he shows that the doubling of an object clitic is obligatory if the coreferent element is a pronominal. An exception is indefinite pronouns, which do not allow doubling. Full NPs pattern in a similar way: clitic doubling is disallowed with indefinite NPs but is possible with definite ones. In Kaiser's (1992) corpus 18,4% of definite NPs appear in doubling contexts. He does not carry out a distinction between accusative and dative objects. Kaiser (1992) analyses subject clitics in Colloquial French as inflection affixes and observes that object clitics behave in exactly the same way: they both select verbs as a host, they are syntactically and phonologically bound to that host and form with it a unit for syntactic operations. As subject clitics, object clitics in Colloquial French behave as agreement markers, and thus license and identify pro in the same way. Clitic doubling in Portuguese is more restricted than in Spanish, Romanian or French. It is limited to pronouns, and it is obligatory in pronominal contexts such as (14) below. (14) a. vejote see. Is 2accs Ί see you1 b. *vejo a ti

a ti Κ you

(Kaiser 1992:132)

According to Kaiser (1992) Portuguese clitics have clearly affixal status. They not only behave like typical bound morphemes but they also show other properties of affixes, such as the ability of bringing about morphophonological changes in the host and the possibility of being mesoclitic, i.e. they appear between the stem and the tense affix (cf. 15). Mesoclitic elements are not very common but they provide important evidence for the affixal status of clitics: otherwise it is not possible to explain how the finiteness and tense morphemes are affixed to the verb. (15) dir- seá que oAlentejo também é urna regiäo muito agrària say- impers. fut.3s that Alentejo also is a region very agricultural 'it will be said that Alentejo is a very agricultural region' (Kaiser 1992:134) A further characteristic which speaks for the interpretation of clitics as affixes is that, except in the case of mesoclisis, they are always enclitic, i.e. they only attach to the verb in one direction.

Clitic Doubling and Clitic Clusters in Spanish

95

Kaiser (1992) considers Portuguese clitics to be agreement markers carrying information on person, number, gender and case. In addition they have the feature [+pronominal], as they only agree with pronominal objects. The facts in these Romance languages show similarities with the phenomenon of clitic doubling in Spanish. These languages allow doubling to a different extent, but the conditions regulating doubling in all of them involve case and the specificity scale. 1.4 Summary To summarise, clitic doubling is seen as an instance of agreement, agreement being defined as feature matching. Case and specificity have been identified as the relevant features to be matched and they are ordered on a scale. The case hierarchy ranks nominative > dative > accusative. It is considered to be an implicational scale for agreement decreasing from left to right. In addition to this, the specificity scale (see 10) also conditions clitic doubling (cf. 16) This scale includes categories which also play a role in languages known to show subject and object agreement; this is the case, for example, with the role of specificity in the organisation of Bantu paradigms. (16) 1 + 2 > 3 > NP +definite, + specific > NP +definite,-specific > NP,-definite The dialects under consideration here, Standard and Rio de la Plata Spanish occupy different positions on the agreement scale: both of them show obligatory agreement with nominative, obligatory or strongly preferred agreement with dative for Rio de la Plata and Standard Spanish respectively, and a range of possibilities with accusative depending on the dialect, the syntactic category of the double and the fulfilment of the feature matching for specificity. According to this analysis clitics have affixal character: I would consider dative clitics as affixes and accusative clitics as quasi affixal. The fact that accusative clitics are subject to additional conditions indicates that they still keep some of their pronominal character. As affixes clitics build paradigms: evidence for this are clitic sequences and the rules for their combination. In the next section I will examine the closely related area of clitic clusters and show that they are also conditioned by the case and specificity hierarchies. The fact that the sequences cannot be combined freely and that they are conditioned by the same hierarchies as clitic doubling is further evidence for their status as agreement markers.

Teresa Parodi

96

2. Clitic clusters: Possible combinations of two objects This section deals with the rules for the co-occurrence of two clitics. In order to keep the parallel to the preceding section the observations will refer only to clitics in argument positions, disregarding ethical datives. As opposed to the realisation of two lexical arguments, the combination of two (or more) clitics is subject to special constraints, as one actually might expect from affixes organised in a paradigm.

2.1 Clitic clusters in Spanish In the following tables all logically possible combinations for both dialects of Spanish discussed here are displayed. The combinations of 1+1 and of 2+2 are excluded in this form; they are realised as reflexive constructions. Combinations marked with an asterisk are logically possible but do not occur. (17) a. Standard Spanish dative acc sg

plural

1

3

I

2

3

*te-me

*le-me

*

*os-me

*les-me

*

*le-te

* nos-te

*

*les-te

2

I

*

2

*me-te

3m

me-lo

te-lo

se-lo

nos-lo

os-lo

se-lo

3/ 1

me-la

te-la

se-la

nos-la

os-la

se-la

*te-nos

*le-nos

*

*os-nos

*les-nos

2

*me-os

*

*le-os

*nos-os

*

*les-os

3m

me-los

te-los

se-los

nos-los

os-los

se-los

3f

me-las

te-las

se-las

nos-las

os-las

se-las

*

b. Río de la Plata Spanish dative acc

I

sg

1

#

2

*me te

3m

me-lo

3/ 1 2

plural

plural

singular

singular 2

plural 3

1

*te-me

*le-me

*

*les-me

*

*le-te

*nos-te

*les-te

te-lo

se-lo

nos-lo

se-lo

me-la

te-la

se-la

nos-la

se-la

*

*te-nos

*le-nos

*

*me-los/as

*

3m

me-los

3/

me-las

te-los te-las

*se-los/as *nos-los/as se-los se-las

nos-los nos-las

2

3

*les-nos *se-los/as se-los se-las

97

Clitic Doubling and Clitic Clusters in Spanish

1. As can be seen from the table the combination of 1st and 2nd person is disallowed in whichever order (18a); if 1st and 2nd person have to be combined, then one of them, usually the dative, has to be realised by a strong pronoun (18b).3 (18) a. *te me presento 2dats laccs introduce.Is Ί introduce myself to you' b. me presento a ti 1 accs introduce. Is Κ you Ί introduce myself to you' 2. 1st and 2nd person clitics only combine with 3rd person. In this case the 3rd person must be accusative. 3. Two 3rd person clitics can combine with one another. In other words, in a clitic cluster there is always a 3rd person. 4. Case also plays a role in the ordering: dative always precedes accusative. Perlmutter (1970) claimed that clitics form a template, which amounts to describing the order and stating that there is no further explanation for it. This claim is still accepted, among others, by Bonet (1995b:75f) and Torrego (1995). In what follows I will try to give an explanation for the observed sequence. Perlmutter (1970) observes that Spanish clitics are ordered by person, while French clitics are ordered by case. A similar claim is found in Bonet (1995b:75f). Perlmutter's claims are based on the analysis of clusters including, not only arguments, as in the present paper, but also reflexives, impersonale, "spurious se" and ethical datives. Both Perlmutter (1970) and Bonet (1995b) need to postulate a template with a more or less ad hoc ordering in order to account for non-argument clitics, e.g. ethical datives as in (19) (my gloss, Perlmutter's translation). (19) a. s e m e le perdio el pasaporte 3s ldats 3dats lose.3s.past the passport 'My child's passport got lost on me' b. *se le me perdió el pasaporte al niño

al niño Κ the child

(Perlmutter 1970:197)

Both see, however, the category person as an important ordering criterion. According to Perlmutter (1970) case is irrelevant for Spanish, while Bonet (1995b) does see some relation3

These limitations only hold if both clitics realise arguments. They do not hold in the same way if one of them is an ethical dative (dative of interest).

Teresa Parodi

98

ship between person and case, stating that "if there is a dative clitic related to some argument, the clitic related to the direct object has to be third person" (1995b:66). This is in a different formulation one of the observations also made in the present paper. The observations here, however, are part of a wider generalisation showing the interplay of different conditions. As Perlmutter correctly observes, in Spanish se takes the first place in the cluster, whether it is a "spurious se", a reflexive or an impersonal. It is followed by 2nd, 1st and 3rd person. This is interpreted as a template, in which elements from the different slots cannot be repeated: that is how co-occurrence restrictions are accounted for. Again a similar claim is made by Bonet (1995b:77), based on an analysis of Barceloni dialects. According to her, "clitics seem to be ordered in an ad hoc but fixed fashion in each dialect" and "clitic order is established through the mapping to a template with specific positions for each clitic, which varies from dialect to dialect". In the following I will provide an analysis which does not require templates. Person certainly plays a role in the ordering, but it is not the only criterion. Furthermore, at least as far as Standard and Río de la Plata Spanish (and other Romance) languages are concerned, the ordering seems to follow some principle, i.e. it is not merely ad hoc. The ordering by person can be seen as part of the specificity hierarchy mentioned above, which follows the animacy hierarchy postulated in Silverstein (1976) and Comrie (1981) (cf. (17) above). Relevant at this stage is the person feature. Items which are higher in the hierarchy have to precede in the linear order lower ranking items. Assuming the animacy hierarchy in (20), it follows that the linear orders in (21a) are allowed and those in (21b) disallowed. (20)

1+2 > 3

(21) a. 2/1 3

>3 3

b. *3 *2 *1

< 2/1 < 1 goal/recipient > patient/theme subject > dative object > accusative object (23) case/argument structure animacy

A > Β a > b

Clitic Doubling and Clitic Clusters in Spanish

99

This means that in order for a sequence of two clitics to be allowed the argument which is higher in the specificity hierarchy must have a higher position in terms of case. Otherwise a clitic cannot be used; a full (tonic) pronoun occurs instead in the regular position for the corresponding argument. Under (24) the interaction of these rules is made explicit: dative lsg: me lsg: me lpl: nos lpl: nos

> > > > >

accus. 3sg: lo/a 3pl: los/as 3sg: lo/a 3pl: los/as

> dative b. * 1 sg: me *lsg: me * lpl: nos * lpl: nos

dative 2sg: te 2sg: te 2pl: os 2pl: os

> > > > >

accus. 3sg:lo/a 3pl: los/as 3sg: lo/a 3pl: los/as

dative d. *2sg: te *2sg: te *2pl: os *2pl: os

>

accus. lsg: me lpl: nos lsg: me lpl: nos

dative 3sg: le sg: le pi: les pi: les

>

accus. 3sg:lo/a pi: los/as 3sg: lo/a 3pl: los/as

dative f. *3sg: le *3sg: le *3pl: les *3pl: les

> < < <
< <
[[ pli'v ]- θ ] ν0,«ρ i

Analyzing the examples in (4), Rivero regards /Θ/ as the mediopassive affix and /ik/ as the perfective affix. 7 Besides the shortcoming that the ending /i/ remains unanalyzed, the identification of the relevant morphemes is somewhat problematic. Consider the examples in (6). Based on these examples, Rivero correctly assumes that /s/ marks perfective in the active. In the analytical forms, however, she assumes that the auxiliary is always generated under Asp 0 . Thus, /s/ is a bound allomorph to exo. (6)

a. perfective active 9a ayapiso FUT love.PFV.lSG 'G) will love' b. mperfective active 6a fut

ayapó love.lSG

'(I) will be loving'

(Rivero, 1990)

The main shortcoming of this approach is that it cannot account for the combination of perfect and perfective in examples like (7). 5

6

I do not adopt Rivero's orthographic transcription for the Modern Greek examples but prefer the transcription proposed by Joseph & Philippaki-Warburton (1987), because it is much nearer to the phonology of Modem Greek. The forms I refer to as mediopassive are 'passive' in Rivero's terminology. Like other linguists concerned with the mediopassive in Modern Greek (for example Campos 1987; Lascaratou & Philippaki-Warburton 1983), Rivero does not differentiate between the syntactical function passive and its morphological realization. Note that passive is neither the only nor the main function of mediopassive morphology. (See also section 4.1.) Although Rivero does not explicitly identify /ik/ as perfective affix, this is the only reasonable option. Joseph & Smimiotopoulos (1993) come to this conclusion, too.

Birgit Gerlach

106 (7)

éxo

Sési

have.lSG

tie.PFV.3SG

Ί have

tied'

(example from Mackridge, 1985:118)

It is not possible to generate both the auxiliary and the affix /s/ in one single Asp0 node. Assuming a lexical approach like the one suggested by Wunderlich and Fabri (1995), however, the hierarchy of categories is observed both in the morphology and in the syntax, so it is possible that analytical forms evolve which repeat morphological information more distinctively (or, as in the case of perfect versus perfective, more specifically). In the following, I will not argue against the hierarchy of categories expressed in (3a.ii), which, I think, is confirmed by the data. Unlike Rivero, I assume that /Θ/ fuses mediopassive voice and perfective aspect, and that /ik/8 expresses past. Consider the array of data in (8). The examples are ordered according to present tense, past tense, future tense and perfect forms. In (8a) the subjunctive marker na is added, because perfective forms cannot refer absolutely to present tense. (8) Forms with perfective aspect Active a. Present (na) (SBJ)

'you would loosen' b. Past é-li-s-es (PST)-l00Sen-PFV-PST.2SG

'you loosened1 c. Future 6a

Mediopassive lí-s-is (na) li-8-ís loosen-PFV-2SG (SBJ) loosen-MP.PFV-2SG

'you would be loosened' or 'you would loosen yourself Ιί-θ-ik-es loosen-MP.PFV-PST-PST.2SG

'you were loosened' or 'you loosened yourself li-s-is 9a Π-θ-ís

FUT

loosen-PFV-2SG FUT

'you will loosen'

'you will be loosened' or 'you will loosen yourself

d. Perfect éx-is

lí-s-i éx-is

loosen-MP.PFV-2SG

li-θ-ί

have-2sG

loosen-PFV-3SG have-2sG loosen-MP.PFV-3SG

'you have loosened'

'you have been loosened'

The perfective affix is invariantly /s/ in the active (active as default interpretation of voice remains unmarked) and /Θ/ in the mediopassive. Thus, voice and aspect are fused; this is compatible with either order (VOICE < ASPECT or aspect < VOICE).9 /ik/ expresses past tense in g The affix -ik poses several problems which are not directly connected with the topic of this paper. Whether the affix is /ik/ or /k/ depends on the analysis of the mediopassive affix as /Θ/ or /θί/ respectively, but since this is irrelevant here, I don't diverge from Rivero's segmentation. Further, no matter what morphological information /ik/ contributes to the verb form, it is always redundant. Wunderlich and Fabri (1995) explicitly restrict the fusion of information about more than one category within one affix to adjacent functional categories, as the hierarchy in (1) shows. This is the case in Modern Greek.

Mediopassive in Greek

107

the context of perfective mediopassive, and θα (specified as [+max] for syntactically visible particles) expresses future tense. The personal ending /es/ fuses past and 2nd person (so it can specify 2nd person in the context of past) 10 . In all other contexts, 2nd person is expressed by /is/. This analysis of the morphological forms clearly suggests the affix order VOICE/ASPECT < TENSE < NUMBER/PERSON, which is in line with the hierarchy claimed in (1). Moreover, given that the person affix of the aparemphato does not contribute checking information (see, FN 4) the forms in (8d) are compatible with one single hierarchy, allowing that Aspect is split into a lexical (morphological) and a syntactic part. Fusion of categories also plays an important role in the analysis of imperfective forms, which are discussed in the following section.

3. Mediopassive in Modern Greek imperfective verb forms As has already been mentioned above (see 3b), there are analyses of the imperfective mediopassive forms in Modern Greek which seem to constitute a counterexample to the hierarchy in (1), assuming that mediopassive voice is separated from the verbal root, namely that it is outside of aspect and other categories. Consider the set of personal forms of lino 'loosen' listed in (9) (following Triandafillidis' [1941]=1993 grammar). Here, the imperfective mediopassive forms are doubly framed. (9)

Personal forms of lino - Ί loosen' a. imperfective forms active present past 1 2 3 1 2 3

medio passive present past

lin-o

é-lin-a

lín-ome

lin-ómun

lin-is

é-lin-es

lín-ese

lin-ósun

lin-i

é-lin-e

lín-ete

lin-ótan

lin-ume

lín-ame

lin-ómaste

lin-ómastan

lin-ete

lín-ate

lin-ésaste

lin-ósastan

lin-un

é-lin-an

lín-onde

lin-óndan

b. perfective forms active present past 1 2 3 1 2 3

10

mediopassive present past

lí-s-o

é-li-s-a

li-θ-ό

Ιί-θ-ik-a

lí-s-is

é-li-s-es

Ιΐ-θ-ís

Ιί-θ-ik-es

lí-s-i

é-li-s-e

li-θ-ί

Ιί-θ-ik-e

lí-s-ume

li-s-ame

Ιί-θ-úme

Ιί-θ-ik-ame

lí-s-ete

li-s-ate

Ιί-θ-üe

Ιί-θ-ík-ate

lí-s-un

é-li-s-an

li-θ-ύη

Ιί-θ-ik-an

In the active, /es/ allows the prefix lèi for prosodie reasons, see the analysis of the past tense marking in section 3.

Birgit

108

Gerlach

The imperfective paradigm is given in (9a) and the perfective paradigm is given in (9b). As can be seen, the personal endings differ with respect to tense and with respect to voice: In the perfective mediopassive forms (see 9b) the personal endings are the same as for the active voice in both verbal aspects (apart from different stress patterns). Therefore I regard them as unmarked for voice. Remember from the previous section that mediopassive voice is marked together with perfective aspect and does not interact with the personal endings. In imperfective aspect, however, both categories, tense and person, are involved in the marking of the mediopassive voice (see 9a). The argument that voice is separated from the verbal root is exclusively based on the particular ways in which the verb endings in (9a) are segmented. Several such ways have been proposed in the literature, and I will show that they are not convincing. In particular the relation of person and voice (Warburton 1970) and tense and voice (Hamp 1961) is somewhat problematic. If one compares lin-ome Ί am loosened' with Ιίη-ο Ί am loosening', for instance, one may conclude that the person marker precedes mediopassive, and if one compares lin-ósun 'you were loosening' with the other 2sg forms, one may conclude that the tense marker precedes mediopassive. Accordingly, Hamp (1961) assumes the order root < aspect < tense < voice, and Warburton (1970) assumes the order root < aspect < person < voice. Note that an aspect affix occurs only in perfective forms. According to the discussion in section 2 this affix fuses the aspect and voice information. In imperfective aspect verb forms, however, there is no aspect affix. Therefore neither perfective nor imperfective verb forms give any reason to assume the order of aspect before voice. Let us now consider whether tense may precede voice. As already pointed out in section 2, only past tense is realized morphologically, but in different ways (see also FNs 8 and 10). First, as far as the augmentation of the past tense forms is concerned, no interaction between voice and tense affixes takes place: Modern Greek past tense forms usually are stressed on the antepenultimate syllable. If any past tense verb form has only two syllables, the 'augment' /e/ is prefixed to this form to bear the stress. That means, as Kaisse (1982) has shown, that the augment is only the effect of a prosodie constraint on past tense verb forms. The mediopassive, however, can never co-occur with the augment, because all mediopassive verb forms have at least three syllables. Second, the stressed past tense affix /ús/ is restricted to verbs with stress on the ending in the imperfective active.11 This affix always bears the stress and, since it is restricted to the active, it does not interact with the mediopassive either. Third, the past tense allomorph /ik/ is restricted to perfective mediopassive forms. Although there is an interaction between mediopassive voice and past tense in this case, no contradicting affix order is justified by these forms, as has been shown by the detailed discussion in the previous section. Fourth, in

11

However, these forms are being replaced by newer forms without this affix already: a. older forms (Thessaloniki) b. newer forms (Athens) ayap-ús-es ayápa-j-es I0ve-PST-PST.2SG

love-PST.2sG

'you loved'

'you loved'

See also example (14) below.

Mediopassive

in Greek

109

the imperfective active forms past tense is also indicated by the theme vowels /a/ or /e/, but neither of them occurs in the imperfective mediopassive past. In all these cases either voice precedes tense or it is unmarked. Fifth, although the forms in the imperfective mediopassive consistently have /o/ as theme vowel, this vowel also occurs in three of the corresponding present forms. So there is no reason to identify this vowel as tense marker and nothing indicates that past precedes mediopassive. Let us now examine the suggestion that person precedes voice. Warburton (1970) analyses Imi, /s/ and /t/ as the respective person morphemes for first, second and third person in active and mediopassive voice, followed by a complex node Q in which voice, tense and number are fused. The overall structure she proposes is shown in (12) where brackets show optional elements. (10) The structure of the verb in Modem Greek: Warburton (1970, p. 148): V

V-»

( M o d a l ) + ( P r e f i x ) + V base + ASPECT + THEMATIC + PERSON + VOICE + TENSE + NUMBER VOWEL m-

s-

t-

1

2

3

This analysis puts not only voice, but also tense and number after person. Thus, it is maximally incoherent with the hierarchy in (1). Moreover, in order to make her analysis work, Warburton needs a great number of readjustment rules to derive the actual forms. Number, tense and voice are a combination of information appearing in the surface structure as the result of a lot of stipulated context-specific spell-out rules. This kind of analysis is very uneconomic, because of the abundance of rules and the complexity of the ordering mechanism, which has to be added in order to produce the right forms. In her 1973 paper, Warburton proposes a different analysis. Here she considers the vowel a- as the past tense marker. She can thus treat the mediopassive forms like the active forms, where the lowering of the theme vowel marks past tense: the back vowels /o/ and /u/ become /aI, and the front vowel /i/ becomes /e/ (see 9a). Furthermore, she assumes that the theme vowels loi or /e/ indicate mediopassive in the imperfective. As before, /m/, Is/ and Iti are considered to be person markers, so voice precedes person and no conflict with (1) arises. Based on imperfective mediopassive forms that differ from those given in (9a) in the past tense insofar as they are dialectal variants to the normative forms in (9a), she proposes the segmentation of personal endings as shown in (11).

Birgit

110

Gerlach

(11) Segmentation of personal endings: Warburton (1973) a. imperfective mediopassive, present tense o-m-e e-s-e e-t-e MP-1-?

MP-2-?

MP-3-?

b. imperfective mediopassive, past tense o-m-un-a o-s-un-a o-t-a-n MP-1-7-PST

MP-2-7-PST

MP-3-PST-?

o-m-a-st-e

e-s-t-e

o-n-t-e

MP-1-?-?-?

MP-2-?-?

MP-7-3-7

o-m-a-st-e

o-s-a-st-e

o-n-t-a-n

MP-l-PST-?-?

MP-l-PST-?-? MP-7-3-PST-?

This analysis is still in conflict with (1) because the person affix precedes the past tense affix. Note that there are a number of question marks in (11) for the possible information given by segments which Warburton does not provide with an analysis, /a/, claimed to be the past morpheme, either precedes or follows these segments so that the parallelism to the active forms rests only on vowel quality anywhere in the ending. Another point against analyzing /a/ as a tense marker is that it is not obligatory in first and second person singular, so that tense marking in these forms must be regarded as optional despite the fact that the forms are clearly past tense even without /a/. Most of the forms are spelled out by means of readjustment rules, which, by their very nature, can at most have descriptive adequacy. In fact, Warburton needs 34 additional rules to derive the actual forms, so this analysis is not very convincing. In the following, I will sketch an alternative analysis for the active forms in (9) which shows that the information indicating tense, person and number is distributed among two constituents in a way that conforms with (1). All personal endings in the active and mediopassive perfective forms can be decomposed into a vowel melody (i.e. a vowel feature that is not associated with a segment, with /e/ as unmarked vowel) and a string of segments whose first element is an empty vowel segment. For the active set of the personal endings the lexical entries for the affixes and the vowel melodies are shown in (12). Category features before the slash indicate the output information, and those after the slash indicate the input information (the selection condition). Missing information is filled in according to the paradigm construction described in Wunderlich & Fabri (1995), so the combination of two empty outputs leads to 3sg.12 (12) a. fully specified affix:

12

/ite/:

[+2,+pl] / [+mp]

b. vowel melodies:

/[+low]/: /f+round]/: /[+round,+high]/: /[+high]/: /[ ]/

[+1 ν +pl] [+1] / [-past] [+pl] / [-2,-past] t ] / [ -pl, -past] elsewhere

c. affixes:

/Vme/: /Vte/: /Vs/: /Vn/: /V/:

[+l,+pl] [+2,+pl] [+2] [+pl] [ ]

In the system of Modern Greek verbal inflection the final /i/ bears no information by itself, which is the reason why the aparemphato - actually identical with 3sg forms - can be used without information.

checking

Mediopassive in Greek

111

By putting the two constituents together, the empty vowel segment ('V in (12c)) is associated with a melody. Example (13) shows how the imperfective active verb forms result by applying the affixes and vowel melodies in (12) to a verb with consonantal stem ending representing the class of verbs which are stressed on the stem. (13) Imperfective active of verbs with stress on the stem present tense

past tense

lin-o lín-is lín-i lín-ume lín-ete lín-un

é-lin-a é-lin-es é-lin-e lín-ame lín-ate é-lin-an

No problems arise with these verbs, but if the verb stem already ends in a vowel (which is the case with verbs with stress on the ending), applying the affixes in (12) would cause hiatus. In the present tense forms this is avoided by deletion of the less specific vowel (see 14a). Here, specificity should not be regarded as phonologically determined (like for example in terms of sonority hierarchy as suggested by Warburton (1973) which makes the wrong predictions) but as morphologically determined specificity: A vowel that bears person or number specification in the output may never be deleted. This can be seen in the present tense paradigm in (14a) 13 . The past tense endings remain unchanged because of the past tense affix /-us/ between stem and personal ending in the older forms or because of the glide /-j/ in the newer forms (see (14b)). (14) Imperfective active of verbs with stress on the ending - older paradigm a. present tense ayap-ó ayapá-s ayapá ayap-úme ayapá-te ayap-ún

b.

past tense older forms ayapús-a ayapús-es ayapús-e ayapús-ame ayapús-ate ayapús-an

newer forms ayápaj-a ayápaj-es ayápaj-e ayapáj-ame ayapáj-ate ayápaj-an

Though the tense information is thus clearly borne by the vowel melody, there is no distinct position in the affix order for it. Note that no other restrictions concerning order or selection are made than those that are given in the input information. None of the minimal morphemes given in (12) survives in the mediopassive forms listed in the last two columns of (9). It is purely speculative to assume that, for instance, the vowel 13

As in the past tense, there are newer forms in the present tense, too. Here, the stem vowel is resistant to deletion and remains in all cases. The vowel of the ending is deleted except in 1st and 3rd person singular (αγιρά-ο and ayapá-i), where the ending consists of only the vowel. One has to assume then that reanalysis of the personal endings takes place.

112

Birgit Gerlach

melody [+round] or the string /Vs/ may play any role in these forms. So we are well advised if we assume that all these endings are monomorphemic, namely express a complex information that is fused into one single exponent: the four categories, voice, tense, number, and person, are fused into one affix as can be seen in the examples in (15), and no argument can be made concerning a violation of (1). (15) Forms with imperfective aspect active a. lín-is

mediopassive lin-ese

loosen-2sG

loosen-MP.2sG

'you loosen'

'you are loosened' or 'you loosened yourself

b. é-lin-es

lin-ósun

(PST)-loosen-2sG

loosen-MP.psT.2sG

'you loosened'

'you were loosened' or 'you loosened yourself

c. 0a

lin-is 0a

lin-ese

FUT

loosen-2sG

FUT

'you will loosen'

'you will be loosened' or 'you will loosen yourself

loosen-MP.2sG

Evidence for this assumption is given by the fact that the imperfective mediopassive in Modern Greek is not at all fixed. Instead, a great variation can be observed, the greatest one in the past forms. This is shown by a study of Ruge (1972), who compares the normative paradigm (Triandafillidis [1941=] 1993) with several dialects (Athens, Kavala and Istanbul). Though Ruge's method is rather informal simply questioning speakers in the respective regions and noting the first of the given forms (which were more than one in the most cases), his results are confirmed by Iordanidou's more recent study. Iordanidou (1994) compares the verb forms of a much larger corpus consisting of informal speech and questionnaires as well as literary, scientific and journalistic texts. She finds out that, generally, imperfective mediopassive forms are used very seldom in informal speech (Iordanidou gives 0,6% for the best case). No similar variation can be found in the imperfective active or perfective active and mediopassive forms. In my opinion this variation forces the linguist to abstain from segmenting the forms into single morphemes, because it is impossible to refer to uniform forms. The startling difference between the realization of the mediopassive in imperfective and perfective aspect becomes much clearer after a closer look at the development of the forms. Therefore in the next section mediopassive and passive forms in Ancient Greek will be considered as well as the change in function they have gone through.

4. The development of the mediopassive in Greek For Ancient Greek one has to assume the affix order in (16). The passive is marked directly after the verb root. The middle is marked together with person and number at the end. Between these voice categories tense and mood may be marked.

Mediopassive

in

113

Greek

(16) The instantiation of functional categories in Ancient Greek V < VOICEPas/ASPECT < TENSE < MOOD < V0ICEM®/PERS0N/NUMBER

Evidence for this order is given below in (17) and (18). The table in (17) shows the relation between aspect and voice in Ancient Greek. (17) Active, middle and passive voice in Ancient Greek a. imperfective active middle

passive

lú-oo

lú-omai

loosen-lSG

loosen-lSG.MiD

Ί loosen'

Ί loosen myself or Ί am loosened'

b. perfective active

middle

passive

é-lu-sa-n

e-lu-sá-meen

e-lu-thée-n

PST-Ioosen-PFV-1 SG(PST)

PST-lOOSen-PFV- lSG.MID(PST)

PST-lOOSen-PFV.PAS-lSG(PST)

Ί loosened'

Ί loosened myself

Ί was loosened'

(17a) shows that the imperfective aspect in Ancient Greek is the same as in Modern Greek: The middle marker which has middle as well as passive functions is fused with the personal ending. The situation is different in perfective aspect (see 17b). Here three voices are distinguished: active, middle and passive. In the active and the middle voice the affix for perfective aspect is -sa\ the middle, just as in the imperfective, is fused with the personal ending, resulting in -mai or -meen for 1st person singular present or past. In the passive there is a fused affix, too: -thee indicates the perfective aspect and the passive voice at the same time, while person/number is marked independently. The similarities of Ancient and Modern Greek affixes are easy to see: Ancient Greek -sa became -5 in Modern Greek, -thee became θ(ϊ)14, -mai became -me and -meen became -mun. The data (18) reveals that both positions for voice markers can be filled at the same time (voice markers are bold in (18)). In addition it is possible to separate both markers by marking tense and mood. (18) Future, optative and middle in Ancient Greek a. i. lú-so-mai

100Sen-PAS.PFV-FUT- ISG.MID

Ί will loosen myself 'or Ί will be loosened'

Ί will be loosened' (perfective)

b. i. lú-oi-meen

loosen-PAS.PFV-ΟΡΤ-1 SG

Ί want to loosen myself or Ί may be loosened'

Ί may be loosened' (perfective)

loosen-FUT-oPT- ISG.MID

S e e F N 8.

ii. lu-the-iee-n

loosen-oPT-1 SG.MID

c. i. lu-s-oi-meen

14

ii. lu-thée-so-mai

Íoosen-FUT-ISG.MID

ii. lu-thee-s-oi-meen l00sen-PAS.PFv-FUT-0PT- ISG.MID

114

Birgit Gerlach

In (18a) examples of future tense forms are given. Unlike in Modern Greek, future tense is marked synthetically in Ancient Greek. In imperfective forms (18a.i) the future affix -so directly follows the verb root, then middle is marked with the personal ending. In perfective forms (18a.ii) the aspect/voice affix -thee comes after the verb root, followed by the future affix -so. Additionally the mediopassive personal ending -mai is outmost. 15 That means that tense is marked between both voice positions. (18b) shows that not only tense affixes but also mood affixes can occur in this position. Here the optative affix (-/ in (18b.i) and -iee in (18b.ii)) occupies the position. The forms in (18c) finally show that between both voice positions tense and mood affixes can occur.16 These data show that it is reasonable to assume an affix order like the one in (16) for Ancient Greek. This order differs from the hierarchy in (1) in at least two respects: on the one hand, there are two different positions for one category, namely voice, and on the other hand, one would not expect the mediopassive voice marker as a derivational affix to be separated from the verb root by tense and mood affixes which are inflectional. On the contrary, one would predict it to be positioned inside inflectional morphology, presumably directly after the verb root just like the passive voice marker. Furthermore the possibility of filling both voice positions (compare (18a.ii) and (18c.ii)) leads either to unwanted redundancy (assuming that passive and mediopassive are the same) or to contradictions (assuming that passive and middle are different instantiations of the same category). An explanation for the possibility of affix orders like the one in (16) results from the comparison of passive and middle voice functions. As can be shown, they are basically different and therefore belong to different categories. In this respect I agree with Klaiman (1991), who makes a distinction between basic voice (depending on inherent verbal semantics i.e.: active/middle) and derivational voice (operations on argument structure i.e.: passive). The emergence of the two voice positions in Ancient Greek can be explained historically, if the development of the two voice markers is considered. 4.1 T h e Ancient Greek middle marker The active/middle distinction in Indo-European is definitely older than the active/passive distinction. Middle functions subsume reflexive as well as passive and several other functions. 17 However, even if the passive is expressed by middle morphology, it often differs from 'regular' passive (agent realization is rarely possible, subject has to be animate etc.) 18 . Middle affixes probably originate in clitic pronouns. Georgiev (1985) for example derives them from pronouns partially from Ancient Greek and partially from Proto-Indo-European:

15

Only future passive has mediopassive personal endings in Ancient Greek. All other tenses and moods of the passive voice have unmarked personal endings. Although these forms are used very seldom, Ancient Greek grammars (see Smyth, 1966, for example) give them as regular.

17

See Klaiman ( 1 9 8 8 , 1 9 9 1 ) for more information about the functions of middles as opposed to passives. See Smirniotopoulos (1991) for more details especially for Modern Greek.

Mediopassive in Greek

115

Primary endings19 moi ('me'), soi ('you'), toi ('this one') > -mai, -sai, -tai (Reflexive, personal or demonstrative pronouns, dative) Secondary endings *me'm ('me') *so ('you'), *to ('this ') > -meen, -so, -to (Reflexive, personal or demonstrative pronouns, accusative) Both the historical data and the function of the middle suggests that the middle markers originate in cliticized reflexive or personal pronouns. Similar developments can be seen in the Romance languages where reflexive pronouns come increasingly closer to the verb and achieve more and more affix status fulfilling middle functions at the same time.20 Regarding the similarities between clitics and middle markers on the one hand and the differences between middle and passive in function on the other, the position of the middle marker is no longer surprising: because of their derivation from pronouns they are peripheral. The question is then, why the Ancient Greek system additionally invented another voice category, namely passive, despite the fact that passive could already be expressed by middle morphology. 4.2 The Ancient Greek passive marker Bakker (1994) shows that the passive function of -thee has emerged from the different conditions for two aspect affixes in Ancient Greek (for convenience his observations are illustrated in the table in (19)). (19) Two aspect affixes in Ancient Greek categorial information structural effect subject properties causative/unaccusative alternants intransitive verbs of motion, emotion and cognition passive interpretation for transitive verbs inherent perfective verbs

-thee[+perfective] intransitive unaccusative + + +

-sa[+perfective] +volitional causative

As can be seen in (19), the aspect affixes are in complementary distribution. While -thee is intransitivizing, -sa is restricted to volitional subjects. The distribution results from these conditions. Verbs which have causative/unaccusative alternants mark perfective aspect with -thee on the unaccusative and with -sa on the causative alternant. Moreover -thee is chosen for the perfective aspect of intransitive verbs of motion, emotion and cognition. Passive interpretation can only be realized by -thee and never by -sa+middle personal ending. Only -thee and never -sa can be attached to inherent perfective verbs. This shows that in Ancient 19

20

The difference between primary and secondary endings is roughly one of tense and mood: in all past tenses and in the optative, secondary endings are chosen; primary endings are chosen elsewhere. See Barber (1976) for more comments on this point.

116

Birgit

Gerlach

Greek the categorial information of -thee contains more than aspect information (in addition it is at least intransitivizing). Probably the reinterpretation from the aspect affix to the aspect and voice affix is already over at this stage. The position of -thee is exactly as predicted: because it originally marked a derivational process on the argument structure of the verb it directly follows the verb root. 4.3 From Ancient Greek to Modern Greek The development from Ancient to Modern Greek verb forms can be characterized as a strategy of repair to rearrange the affix order in accordance with the hierarchy in (1). The relation of aspect and voice in Modern Greek is as shown in (20). (20) Active, middle and passive in Modern Greek a. imperfective active

passive

middle

lin-o

lin-ome

loosen-lSG

loosen-lSG.MP

Ί loosen'

Ί loosen myself or Ί am loosened'

b. perfective active

passive

middle

é-li-s-an

li-6i-k-an

(PST)-loosen-PFV- PST.ISG

loosen-PFV.MP-PST-PST. lSG

Ί loosened'

Ί loosened myself or Ί am loosened'

The three voice system with active, middle and passive distinction has been changed into a two voice system with active and mediopassive distinction by the following steps.21 First, perfective aspect has been assimilated to imperfective aspect by reinterpretation of -thee from the intransitivizing perfective aspect marker to passive voice and perfective aspect marker. At first glance it is surprising that the passive affix, and not the middle affix, has been chosen as an affix for the mediopassive in the perfective aspect, but this results from the above mentioned distribution of the aspect affixes in Ancient Greek: due to the restriction of the perfective marker -sa to volitional subjects, it was impossible for the middle affix to receive passive interpretation in combination with -sa. Only the semantically unmarked -thee could receive passive as well as middle interpretation. Therefore morphological parallel forms in imperfective and perfective aspect were not possible. Then, the identical interpretation of both voice categories in perfective aspect made double voice marking redundant. Middle personal endings were replaced by personal endings without any voice information. In Modern Greek there are no longer perfective verb forms with both voice markers. Furthermore the categories which were marked between verb root and middle voice affixes - tense and mood - have been

21

See Barber (1976) regarding the development of voice systems. Barber predicts that three voice systems are always unstable.

Mediopassive

in Greek

117

externalized and are now marked by clitics or particles (future and subjunctive in examples (8a/c) and (15c) here repeated as (21) for convenience). (21) Future and subjunctive forms in Modern Greek a. forms with imperfective aspect 9a lin-is

0a

lin-ese

FUT loosen-2sG

FUT

loosen-MP.2sG

'you will loosen'

'you will be loosened' or 'you will loosen yourself

b. forms with perfective aspect (na) lí-s-is

(na) Ιϊ-θ-ís

(SBJ) loosen-PFV-2sG

(SBJ)

'you would loosen'

'you would be loosened' or 'you would loosen yourself

6a

li-s-is

loosen-MP.PFV-2sG

9a

Η-θ-ís

FUT loosen-PFV-2sG

FUT

loosen-MP.PFV-2sG

'you will loosen'

'you will be loosened' or 'you will loosen yourself

Therefore, the different realization of mediopassive voice in Modern Greek is no longer puzzling, but can be seen as the result of the reorganization of verbal morphology due to the universal tendency to obey the hierarchy of functional categories in (1).

5. Conclusion The Modern Greek verb forms do not constitute a counter-example to the universal hierarchy of functional categories assumed in modern approaches to inflectional order. Rather, the affix orders in imperfective aspect (22a) and in perfective aspect (22b) are in agreement with the hierarchy. ( 2 2 )

a.

[

B·

[[[[

]

V

]

VOICE/TENSE/PERSON/NUMBER ] v

]VOICE/ASPECT]TENSE]PERSON/NUMBER

The different realizations of the mediopassive have been motivated by the development of the marker from Ancient Greek. Exceptions to the hierarchy, which in fact appeared in Ancient Greek, were only transitional and were repaired to conform the affix order to the predicted order of functional categories.

References Barber, E.J.W. (1975). Voice - beyond the passive. In: Proceedings of the first annual meeting of the Berkeley Linguistics Society. Berkeley: University of California. Bakker, Egbert (1994). Voice, Aspect and Aktionsart. Middle and Passive in Ancient Greek. In: Barbara Fox & Paul J. Hopper (eds.). Voice. Form and Function, 23-47. Amsterdam: Benjamins. Bakker, W.F. (1970). The aspectual differences between the present and aorist subjunctives in Modern Greek. In: Hellenika (Thessaloniki) 23,78-108.

118

Birgit

Gerlach

Bybee, Joan L. (1985). Morphology. A Study of the Relation between Meaning and Form. Typological Studies in Language, Vol. 9. Amsterdam/Philadelphia: Benjamins. Georgiev, Vladimir I. (1985). Das Medium: Funktion und Herkunft. In: Bernfried Schlerath (ed.). Grammatische Kategorien. Funktion und Geschichte. Wiesbaden: Reichert. Gerlach, Birgit (1995). Das griechische Verb und die Hierarchie der funktionalen Kategorien. MA thesis. Heinrich-Heine-University, Düsseldorf. Hamp, E. (1961). To Rhima en ti Simerini Omiloumeni Elliniki Glossi. Athena 65,101-128. Iordanidou, Anna (1994). Morphological variation of the passive imperfect in Modern Greek. In: Irene Philippaki-Warburton, Katerina Nicolaidis and Maria Sifanou: Themes in Greek Linguistics, 45-52. Amsterdam/Philadelphia: Benjamins. Janda, Richard D. & Brian D. Joseph (1992). Pseudo-Agglutinativity in Modern Greek Verb-Inflection and "Elsewhere". CLS 28,251-267. Joseph, Brian D. & Jane C. Smirniotopoulos (1993). The Morphosyntax of the Modern Greek Verb as Morphology and not as Syntax. Linguistic Inquiry 24,388-398. Kaisse, Ellen M. (1982). On the preservation of stress in Modern Greek. Linguistics 20, 59-82. Klaiman, M.H. (1988). Affectedness and control: a typology of voice systems. In: Masayoshi Shibatani (ed.). Passive and Voice. Typological Studies in Language 16. Amsterdam/Philadelphia: Benjamins. - (1991). Grammatical Voice. Cambridge: Cambridge University Press. Koutsoudas, A. (1962). Verb Morphology in Modern Greek: A Descriptive Analysis. Bloomington: Indiana University. Mackridge, Peter (1985). The Modern Greek Language. A Descriptive Analysis of Standard Modern Greek. Oxford: Oxford University Press. Muysken, Pieter (1986). Approaches to affix order. Linguistics 24, 629-643. Pollock, Jean-Yves (1989). Verb Movement, Universal Grammar, and the Structure of IP. Linguistic Inquiry 20, 365-424. Rivero, Maria-Luisa (1990). The Loccation of Nonactive Voice in Albanian and Modern Greek. Linguistic Inquiry 21,135-146. Ruge, Hans (1972). Imperfekt Passiv im Neugriechischen. Ein Vergleich zwischen normativer Grammatik und Alltagssprache in Athen, Kavala und Istanbul. Glotta 51,142-59. Smyth, Herbert Weir (1966). Greek Grammar. Revised by Gordon M. Messing. Cambridge: Harvard University Press. Triandafillidis, Manolis ([ 1941]=1993). Τριανδαφυλλιδις Μανόλης. ΝΕΟΕΛΛΗΝΙΚΗ ΓΡΑΜΜΑΤΙΚΗ (ΤΗ Σ ΔΗΜΟΤΙΚΗΣ). ΑΝΑΤΥΠΩΣΗ ΤΗΣ ΕΚΔΟΣΗΣ ΤΟΥ ΟΕΣΒ (1941) ΜΕ ΔΙΟΡΘΩΣΗΣ. Thessaloniki: Instituto Neoellinikon Spudon. Warburton, Irene P. (1970). On the Verb in Modern Greek. Language Science Monographs, Vol. 4. Indiana University Publication. - (1973). Modern Greek verb conjugation: Inflectional morphology in a transformational grammar. Lingua 32, 193-226. Wiese, Richard (1995). The Phonology of German. Oxford: Clarendon Press. Wunderlich, Dieter (1992). Funktionale Kategorien im Lexikon. In: Frank Beckmann und Gerhard Heyer (eds.). Theorie und Praxis des Lexikons, 54-73. Berlin: de Gruyter. - (1996). A Minimalist Model of Inflectional Morphology. In: Chris Wilder, Manfred Bierwisch, HansMartin Gärtner (eds.): The Role of Economy Principles in Linguistic Theory, 267-298. Berlin: Akademie Verlag. Wunderlich, Dieter & Ray Fabri (1995). Minimalist Morphology: An Approach to Inflection. Zeitschrift för Sprachwissenschaft 14,236-294.

Ilse Zimmermann

(Berlin/Potsdam)

Das dativische pronominale Klitikum in der DP-Struktur des Bulgarischen Im Rahmen neuerer Theorieentwürfe zur Rolle der Syntax und des Lexikons in der LautBedeutungs-Zuordnung (s. Chomsky 1995, Bierwisch 1996) wird die Integration der dativischen pronominalen Klitika des Bulgarischen in die Struktur der DP untersucht. Dabei geht es um ihre prosodischen, morphosyntaktischen, semantischen und informationsstrukturellen Eigenschaften.1

1. Aufgabenstellung Das Bulgarische weist neben den Vollformen der Personalpronomina auch klitische Formen im Dativ und im Akkusativ auf. In Sätzen und satzartigen Modifikatoren vertreten diese Klitika topikalische spezifisch referierende Argumente bzw. duplizieren diese (s. Rudin 1995, Dimitrova-Vulchanova & Hellan 1995). In Substantivgruppen (DPs) hat das dativische Klitikum diese Funktion. (1)

M.knig\a] [ta]][mu]\ Buch fem d- cl-dat 'das Buch von Ivo'

(na Ivo) präp Ivo

(2)

[[[[ nov]a ] [ ta ]] [ mu ]] kniga (na Ivo) neu fem dcl-dat Buch präp Ivo 'das neue Buch von Ivo'

Es stellen sich folgende Fragen. • Wie kommt es zu einer Struktur vom Typ (1) und (2) und welchen Charakter haben die Struktur und ihre Komponenten? • Welche Rolle spielen dabei Syntax, Morphologie und Phonologie? • Welche lexikalischen Informationen sind relevant? • Welche syntaktischen Bewegungen sind in der Strukturbildung beteiligt?

1

Die Untersuchung fußt auf Zusammenarbeit mit Ivanka Petkova Schick (s. Schick, Zimmermann 1995, 1996a, 1996b). Für hilfreiche Diskussion und einschlägige Hinweise danke ich Peter Gallmann, Birgit Gerlach, Ursula Kleinhenz, Albert Ortmann, Teresa Parodi, Alexandra Popescu, Christopher Piñón, Cristina Schmitt und Wolfgang Ullrich Wurzel. Besonderer Dank gebührt Brigitta Haftka. Eine Vorfassung dieser Arbeit ist in den ZAS Papers in Linguistics, vol. 6 (1996), 192-209 erschienen.

Ilse Zimmermann

120

• Wie wird Kongruenz zwischen Determinierer, Adjektiv und Substantiv bzw. zwischen dem dativischen pronominalen Klitikum und der durch es duplizierten «α-Phrase garantiert? • Inwiefern ist die Generalisierung zutreffend, daß es sich bei den Klitika und den durch sie duplizierten Phrasen um Argumentausdrücke handelt? • Was ist der Bedeutungsbeitrag der beteiligten Konstituenten? Ich vernachlässige die komplizierten Bedingungen fiir die Duplizierung einer topikalischen Dativ· oder Akkusativphrase durch ein entsprechendes Klitikum, darunter auch die Frage, inwieweit diese Duplizierung fakultativ ist. Vorrangig geht es in dieser Arbeit um die Integrierung des pronominalen Klitikums in die DP-Struktur.

2. Grundannahmen zur Laut-Bedeutungs-Zuordnung Das in (3) skizzierte Modell der Laut-Bedeutungs-Zuordnung rechnet mit dem Lexikon als fundamentalem Lieferanten der fur die Korrelierung der Phonologischen Form (PF) und der Semantischen Form (SF) relevanten Strukturbausteine. Die Operationen Merge und Move bewerkstelligen den morphosyntaktischen Strukturaufbau. (3)

LEXIKON MERGE MOVE

PF

Φ

=> SF

Die Theorie sieht vor, daß es Operationen gibt, die fur die semantische Interpretation relevant sind, für die phonologische Interpretation jedoch unsichtbar sind, und umgekehrt. Ich nenne die in die PF übergehende morphosyntaktische Struktur Oberflächenstruktur (OS) und die zur SF zu amalgamierende Struktur Logische Form (LF). Anders als Chomsky (1995) nehme ich mit Bierwisch (1996) an, daß die Laut-Bedeutungs-Zuordnung zwischen PF und SF stattfindet.2 Alle zwischen diesen Schnittstellen vermittelnden Derivationsschritte und Struktureinheiten sind verborgen.

2 Zur Rechtfertigung der SF als Repräsentation der grammatisch determinierten Bedeutung sprachlicher Äußerungen und zu ihrer Unterscheidimg von der konzeptuellen Struktur als Repräsentation der Welterfahrung von Sprechern und Hörern s. Bierwisch (1982, 1987, 1988, 1989, 1996), Lang (1987, 1990, 1994) und Stiebeis (1996).

Dativisches Klitikum in der DP-Struktur

121

3. Hypothesen der Analyse In der Behandlung der pronominalen Klitika verfolge ich wie in Schick, Zimmermann (1996a, 1996b) und wie Cavar & Wilder (1994), Wilder & Óavar (1994) und Óavar (1996) eine syntaktische Lösung. Dabei sind die offensichtlichen Parallelitäten in der Strukturierung von Sätzen, satzartigen Modifikatoren und Substantivgruppen ins Auge zu fassen. Kennzeichnenderweise figurieren die pronominalen Klitika in erweiterten Verb- und Substantivprojektionen relativ weit links im Verhältnis zu den übrigen Konstituenten der jeweiligen Konstruktion. Das Strukturschema (4) deutet das an. (4)

[... [fp (XP) [f [ cl F ]... [lp ... L ... ]... ] ] . . . ]

Wie in Schick, Zimmermann (1996a, 1996b) nehme ich an, daß das pronominale Klitikum Adjunkt einer funktionalen Kategorie F ist und somit jenseits der lexikalischen Projektion LP piaziert ist und in dieser Position basisgeneriert wird. Bewegung von pronominalen Klitika ist nicht vorgesehen. XP in der SpecF-Position ist eine topikalische, gegebenenfalls durch das pronominale Klitikum duplizierte Phrase bzw. deren Spur. Vgl. : (5)

a. tazi mu biiga dieses cl-dat Buch 'dieses Buch von Ivo'

na Ivo präp Ivo

b. na Ivo tazi mu kniga (6)

a. Ana mu pomaga na Ivo. Anna cl-dat hilft präp Ivo 'Anna hilft dem Ivo.' b. Na Ivo mu pomaga Ana.

In den Beispielen wird deutlich, daß die durch das Klitikum mu pronominal verdoppelte Phrase na Ivo in der sichtbaren OS entweder innerhalb der lexikalischen Projektion liegt wie in (5a) und (6a) oder in die SpecD- bzw. SpecC-Position wandert wie in (5b) und (6b). Die in (4) angegebene Konfiguration kommt spätestens durch LF-Bewegung der topikalischen Phrase zustande und liegt der SF zugrunde. Die Plazierung des dativischen pronominalen Klitikums in der Substantivgruppe verlangt in der OS Adjazenz zu einem definiten Determinierer (s. (1), (2), (5)). Um das zu garantieren, sind entsprechende Annahmen über die Struktur der DP und die Definitheitskennzeichnung von Konstituenten erforderlich. (7) gibt die Basisstruktur von DPs an.3 (7) 3

[dp D ([pp (XP) [ F [ f (cl) F ]) [np [n· ... Ν ... ] ([DP na DP ]) ] (]]) ] Vgl. die DP-Analyse in Zimmermann (1991a, 1991b, 1992, 1993), die mit Ausnahme der FP für die Plazierung des Klitikums der hier angenommenen DP-Strukturierung weitgehend entspricht.

122

Ilse Zimmermann

Die das Klitikum beherbergende FP und die «α-Phrase können abwesend sein. Bezüglich der Eingliederung der «α-Phrase in die DP-Struktur nehme ich an, daß sie als Tochterkonstituente von NP piaziert ist.4 FP in (7) entspricht weitgehend der funktionalen Strukturdomäne InflP, die Szabolcsi (1983, 1987) zwischen DP und NP für die Plazierung des prominentesten Arguments von Ν annimmt. Es erscheint nicht abwegig, Possessivpronomen in SpecF unterzubringen. Sie können im Bulgarischen unter bestimmten Voraussetzungen von einem Klitikum begleitet sein. F kann das Merkmal +top haben und damit eine topikalische XP in SpecF und das Auftreten des Klitikums als F-Adjunkt legitimieren. FP ist dann einer funktionalen TopP vergleichbar, die Rizzi (1995) fur die Plazierung von Topiks vorsieht Wenn Definitheit nicht in D gekennzeichnet ist, sondern an einem adjektivischen Kopf oder am Substantiv erfolgt (vgl. (5) vs. (2) vs. (1)), ist D phonologisch leer.5 Zwecks Lizensierung seiner Definitheitskennzeichnung wandert entweder die betreffende Adjektivphrase in die SpecD-Position (vgl. Gallmann 1996) oder Ν wird D adjungiert (vgl. Longobardi 1994, Cavar & Wilder 1994, Wilder & Cavar 1994). Beide Operationen sind fur SF nicht sichtbar. Die resultierenden OS-Konfigurationen in (8) sind typisch für die Abgleichung von Merkmalen, hier von +def. (8)

a.

[dp APj [d· [d 0 ] [fp [f cl F ][m> t¡... Ν ... ]]]]

b.

[dp [d Ni [d 0 ]] [fp [f cl F ][np ... t¡... ]]]

Diese aus der Bewegung von définit gekennzeichneten Konstituenten resultierenden Konstellationen erfüllen die für die Enklise des klitischen Pronomens erforderlichen Adjazenzverhältnisse. Das Klitikum steht adjazent zu dem definiten Determinierer und klitisiert in der prosodischen Struktur an AP¡ wie in (2), an N¡ wie in (1) oder an D wie in (5). Eine Ergänzung ist allerdings nötig. Adjektivgruppen oder Partizipialgruppen können komplex sein. Vgl. : (9)

mnogo gordijat mu ot uspexa prijatel na Ivo sehr stolzer-der cl-dat auf Erfolg-der Freund präp Ivo 'der auf den Erfolg sehr stolze Freund von Ivo'

(10) davno zabravenite mu ot vsicki stari pestìi na ovoarja längst vergessene-die cl-dat von alle alte Lieder präp Schäfer-der 'die längst von allen vergessenen alten Lieder des Schäfers' 4

Möglicherweise ist die wa-Phrase in der Basisstruktur von DP in SpecN, also links von N1, zu piazieren. Dann sind transformationelle Umstrukturierungen vorzunehmen, die alle nicht mit Ν kongruierenden Phrasen in postnominale Stellungen bringen. Ich diskutiere solche Strukturbildungen hier nicht. Zu leeren funktionalen Köpfen s. Zimmermann (1990).

Dativisches Klitikum in der DP-Struktur

123

Es fragt sich, wie es hier zur Adjazenz des Adjektivs bzw. Partizips und des Klitikums kommt. Zwei Möglichkeiten sind zu prüfen. Erstens: Das Klitikum bewegt sich aus FP an den Kopf der in SpecD figurierenden Phrase. Eine solche Bewegung ist jedoch fragwürdig und nicht vorgesehen. Die zweite Möglichkeit besteht darin, daß in der nach SpecD bewegten Phrase und ihrer Spur komplementäre Elidierung von Formativketten stattfindet (vgl. Wilder 1994, 1996). (11) skizziert das in verallgemeinerter Form, wobei Y und Ζ phonologische Repräsentationen von Endketten einer bewegten Konstituente XP und ihrer Spur (Kopie) sind. (11) [xpYZJi , [χρ ¥ Ζ ]i Auf diese Weise kommt es auch in Fällen wie (9) und (10) zu der erforderlichen Adjazenzstellung der an der Klitisierung beteiligten Köpfe.6 Daß es sich bei den pronominalen Klitika um Köpfe handelt, ist in den angegebenen syntaktischen Repräsentationen (4), (7) und (8) immer schon vorausgesetzt worden. Das Klitikum wird in der hier verfolgten Analyse als ein nicht projizierendes D betrachtet, das an einen phonologisch leeren Kopf F adjungiert ist. Es wird noch deutlich werden, welche Funktion diesem F in der Laut-Bedeutungs-Zuordnung zukommt. Ich fasse die in diesem Abschnitt skizzierten Annahmen zusammen, indem ich für das Beispiel (2) die OS und die LF angebe. ( 12) novata mu kniga na Ivo a. die OS [DP [AP novata ], [D· [D 0 ] [FP [F [D mu ] [ F 0 ]] [NP t¡ kniga [DP na Ivo ]]]]] b. die LF [DP [D 0 ] [FP [DP na Ivo], [F [F [D mu ] [F 0 ] ] [NP novata kniga U ]]]]

6

Die in (9) und (10) nach SpecD bewegte und am rechten Rand reduzierte AP und ihr in situ befindlicher Rest sind Konstellationen von gespaltenen DPs bzw. PPs vergleichbar. Óavar (1996) nimmt fur diese allerdings keine komplementäre Elidierung von Endketten an, sondern extrahiert den rechts verbleibenden Rest aus der aufzuspaltenden DP bzw. PP und bewegt die an ihrem rechten Rand entleerte Hülle nach SpecC. Ein aus Cavar (1996:60) entnommenes Beispiel aus dem Kroatischen illustriert das: (i) [ Na kakav t¡ ]j je Ivan [ krov ], bacio loptu tj ? auf was für ein Aux Ivan Dach geworfen Ball 'Auf was fur ein Dach hat Ivan einen Ball geworfen?' In den hier zur Debatte stehenden attributiven Konstruktionen sehe ich für eine solche Lösung keine unabhängige Evidenz.

124

Ilse Zimmermann

4. Die Bausteine der Analyse Das Bulgarische hat im Vergleich zu anderen Balkansprachen nichttautologische Definitheitskennzeichnung. Definitheit wird wie in (5) durch Demonstrativpronomen signalisiert. In DPs ohne Demonstrativpronomen bleibt die Definitheit in D stumm und zeigt sich in der hierarchisch höchsten Phrase mit einem adjektivisch flektierenden Kopf wie in (2), (9) und (10) in Gestalt des enklitischen Formativs t-, das teilweise flektiert und Genus- sowie Numerusunterscheidungen signalisiert.7 Fehlt eine solche Phrase, trägt der lexikalische Kopf Ν wie in (1) die Definitheitskennzeichnung.8 Semantisch kommt die Definitheit erst in D zum Tragen. DEinheiten binden das referentielle Argument von N. Das heißt, daß das Formativ t- als Definitheitsmarker von +N-Einheiten keine Bedeutung trägt.'Die von ihm eingebrachte Definitheitskennzeichnung wird in der OS in den in (8) angegebenen Konstellationen mit dem phonologisch leeren definiten D lizensiert. Mit Penœv (1993) nehme ich an, daß das der Definitheitskennzeichnung dienende Formativ t- zur morphologischen Struktur von +N-Einheiten gehört, und gliedere es in diese als enklitisches Annex ein.10 Für die in (1) und (2) angedeutete morphologische Struktur für knigata bzw. novata, fur den phonologisch leeren definiten Determinierer und für das dativische Klitikum werden die folgenden Lexikoneinträge wirksam.11 (13) Lexikoneintrag fur das Flexiv -a a. lai b. +fem+max c. aV+N (+fem -i-Flektion)^ -max d. λχ [ χ ] (a) repräsentiert die PF des jeweiligen Formativs. (b) gibt seinen kategoriellen Beitrag zur morphologischen Struktur der flektierten Einheit an. Im gegebenen Fall transportiert -a die Genusinformation +fem und die wortstrukturelle Information +max, die redundanterweise besagt, daß das so charakterisierte Wort als syntaktisches Atom mit dem syntaktischen Merkmal +MIN dienen kann (s. dazu Muysken 1982, Stiebeis 1996). In (c) gebe ich die morphologi•7

8 9

10 11

Ich klammere in dieser Untersuchung - vereinfachend - die Betrachtung von Kardinalia und Totalitätspronomen wie vsiäd 'alle' aus, die auch Definitheitskennzeichnungen tragen können. Es wäre ausführlich zu der diesbezüglichen Analyse von Dimitrova-Vulchanova, Giusti (1995) Stellung zu nehmen. Fälle mit impliziter Definitheit betrachte ich hier nicht. Vgl. auch die attributiven Formen der Adjektive des Russischen, die historisch aus der Kurzform des Adjektivs mit klitischem definiten D hervorgegangen sind (s. Bailyn 1994). Ganz analog schließe ich das russische Reflexivität signalisierende Formativ -sja enklitisch an die flektierte Wortform von Verben, Partizipien und Adveibialpartizipien an (s. Zimmermann 1995). Die einzelnen in (a)-(d) verzeichneten lexikalischen Informationen sind in ihrer Substanz den Einträgen in Jackendoff (1975) analog. Im Hinblick auf eine minimalistisch orientierte Morphologie-Konzeption folge ich im Prinzip Wunderlich & Fabri (1995).

Dativisches Klitikum in der DP-Struktur

125

sehen Charakteristika an, die das Formativ von seinem wortstrukturellen Partner verlangt, -a kombiniert sich mit Adjektiv- und Substantivstämmen, ausgenommen Stämme wie gordost 'Stolz', die i-Flektion haben, (d) beinhaltet die Bedeutungscharakterisierung. Im Fall des Flexivs -a ist es die identische Abbildung. Resultierende Einheiten sind die Beispiele in (14).12 (14) Resultierende Konfigurationen [[ nov ] a ]k mit k = +V+N+fem-neutr-pl-lps-2ps+max [[ kn'g ] a ]k mit k = -V+N+fem-neutr-pl-lps-2ps+max (15) Lexikoneintrag für den enklitischen Definitheitsmarker -ta a. Ital,

[[

]p _

]p

b. +def c. Ç +fem-pl ^ / •i lai] / l aV+N+max ^ ßneutr γρίJ d. λχ [ χ ] Das heißt: Das Formativ ist ein prosodisch anlehnungsbedürftiges und morphologisch gebundenes Morphem (s. Inkelas 1990). Semantisch ist es leer. Nur für linksadjazente Wörter im Singular mit femininem Genus signalisiert -ta hier Kongruenz. Sonst handelt es sich um eine Art Vokalharmonie mit dem Auslaut des linken wortstrukturellen Nachbarn. Vgl. die folgenden Beispiele, die alle prosodische Wirte von -ta sein können. (16)

kniga

fem 'Buch'

moja f e m 'meine' nova f e m 'neue' kupena zaminala gordost

mase 'Vater' mase 'Jüngling' dva mase 'zwei' (s. Anm. 7) roga mase pi 'Hörner' deca neutr pi 'Kinder' pisma neutr pi 'Briefe' basta

junosa

edna f e m 'eine'

fem 'gekaufte' fem 'verreiste' fem 'Stolz'

( 17) Resultierende Konfigurationen [ [ [ n o v ] a ] i t a ]]k mit k = +V+N+def+fem-neutr-pl-lps-2ps+max [[[ knig] a ] [ta

]]k mit k = -V+N+def+fem-neutr-pl-lps-2ps+max

Im Gegensatz zum Maskulinum -ät, zum Neutrum -to und zum Plural -te, die alle als flektierte Formen von t- mit entsprechender morphologischer Struktur gelten, wird -ta morphologisch 12

Für die folgenden aus morphologischer und prosodischer Strukturbildung resultierenden Repräsentationen wird nur die strukturelle Typisierung der jeweiligen Einheit angegeben. Dabei werden die morphosyntaktischen Merkmale mit vorhersagbaren Minuswerten ergänzt.

Ilse Zimmermann

126

nicht dekomponiert, wie der Lexikoneintrag (15) zeigt. Es ist hervorzuheben, daß die fur -ta charakteristische Vokalharmonie mit dem Auslaut des prosodischen Wirts fur die morphologische Integriertheit dieses Formativs in dessen Wortstruktur spricht.13 (18) Lexikoneintrag für den definiten Artikel a. / 0 / b. +D+def+spezatop+MIN c. -V+N d. λΡ [ιχ [Ρ χ ]] Das phonologisch leere D bringt in die erweiterte N-Projektion strukturelle Informationen bezüglich Definitheit, Spezifizität, Topikalität und syntaktischer Projektionsfáhigkeit ein. Genus-, Numerus-, Person-Merkmale und auch die kategorialen Charakterisierungen -V+N übernimmt D als funktionale Kategorie von seinem Komplement (s. dazu Grimshaw 1991). (d) beinhaltet die semantische Seite der Definitheitskennzeichnung und die Bindung des referentiellen Arguments von N. (19) Lexikoneintrag für das pronominale Klitikum mu a. Imul, [[ ] p _ ] p v [ _ [ ]p]p b. +D-V+N+def+spez+top+regiert+obliqueaneutr+max+MAX d. χ mit χ e Ν (a) kennzeichnet das klitische Pronomen als prosodisch anlehnungsbedürftige Einheit, (b) charakterisiert es als morphologisch nicht analysierbare und syntaktisch nicht projektionsfahige Einheit im Dativ Singular, 3. Person Maskulinum oder Neutrum, ferner als spezifisch referierendes topikalisches definites Pronomen. Seine in (d) angegebene Semantik ist schlicht eine Individuenvariable. Diese Lexikoneinträge und die in (8) angegebenen OS-Konstellationen führen zu folgender prosodischer Strukturbildung der betrachteten Beispiele: (20) Resultierende prosodische Konfigurationen [[[ nova ]p [[[ kniga

[ta]]p[/m/]]p

]p [ t a ] \ [ m u ] \

Es müssen nun noch einige kurze Betrachtungen zum Kasussystem des Bulgarischen angestellt werden (s. dazu ausführlicher Schick & Zimmermann 1996b). Personalpronomen weisen in ihrem Lexikoneintrag Spezifizierungen bezüglich der Kasusmerkmale aregiert ßoblique auf (zu solchen Merkmalsystemen s. Bierwisch 1967, 1996, Fries 1996). Nichtpronominale DPs haben

13

Auf eine Diskussion, wie -ta in einer Morphologiekomponente zu behandeln wäre, die - anders als hier mit später Integration von Formativen in die morphologische Struktur von Wörtern rechnet, muß ich verzichten.

Dativisches Klitikum in der DP-Struktur

127

keine Kasusmerkmale. Sie treten in Nominativ- und in Akkusativpositionen auf. Alle Präpositionen regieren den Akkusativ. Die Präposition na dient der Kennzeichnung von DPs als Dativphrasen, so daß sich gegebenenfalls Kongruenz mit einem dativischen pronominalen Klitikum wie in den Beispielen (1), (2), (5), (9), (10) feststellen läßt. An den Beispielen wird auch deutlich, daß die bulgarische «α-Phrase adnominalen Genitivphrasen bzw. Phrasen mit den Präpositionen von, of, de, di anderer europäischer Sprachen entspricht. Über einen Genitiv verfugt das Bulgarische nicht. Wie schon in (7) vorweggenommen, sind bulgarische Dativphrasen mit dem analytischen Dativmarker na folgendermaßen repräsentiert: (21)

[DP +regiert +oblique

na

[DP

IVO]],

[ dp -»regiert -(-oblique

na

[Dp

nego]]

1

regiert -oblique

Das bedeutungsentleerte na wird also als phrasales Affix angesehen, das der Dativkennzeichnung von DPs dient. Schließlich ist nun noch die fur die syntaktische Eingliederung des klitischen Pronomens in Anspruch genommene funktionale Kategorie F mit dem Merkmal +top näher zu betrachten. Sie hat folgenden Lexikoneintrag:14 (22) Lexikoneintrag für die funktionale Kategorie F +top a. / 0 / b. +top d. (λγ) α λΡ (λχ)ρ Xr [y = x ] : [ P y r ] k

k

+def

mit β = +, wenn α = und k = +top+regiertyobliqueôfemeneut^plTilpsG2ps Die Funktion dieser funktionalen Kategorie liegt vor allem in ihrer Semantik. Die Bedeutungsangabe in (d) besagt, daß das pronominale Klitikum und auch die durch es gegebenenfalls duplizierte Phrase Topikstatus haben (vgl. Rudin 1995), wobei beide Ausdrücke als semantisch miteinander identifizierte Entitäten gelten und die Phrase in SpecF das Klitikum gewissermaßen expliziert.15 Topiks sind im jeweiligen Diskurs gegebene, als existent vorausgesetzte Größen anzusehen (s. dazu Jäger 1995). Dem trägt in (22d) der unsymmetrische Konnektor ':' Rechnung. λΡ ist durch die NP-Bedeutung bzw. bei Sätzen und satzartigen Konstruktionen durch die VP-Bedeutung zu spezifizieren. Dabei wird angenommen, daß neben dem referentiellen 14

15

(22d) weicht minimal von der fìir F in Schick & Zimmermann (1996a, 1996b) angenommenen Bedeutungsrepräsentation ab. Bemerkenswerte Nähe hat die in (22d) in der Präsupposition repräsentierte Beziehung zwischen dem Klitikum und der es explizierenden Phrase mit Schmitts (1996: Kap. 3) Auffassung, daß in der Basisstruktur von Sätzen in der Objektposition eine Art small clause aP im Spiel sei mit dem Klitikum in Speca und der duplizierten Phrase als Komplement des Identität setzenden Prädikats α . Ich installiere diese Beziehung in der Semantik, ohne sie in einer entsprechenden syntaktischen Form zu repräsentieren. Es ist hier nicht möglich, die Einzelheiten von Schmitts Analyse zu diskutieren.

128

Ilse Zimmermann

Argument noch ein weiteres Argument von Ν bzw. V unspezifiziert ist und erst in der FP zur Geltung kommt. Ferner enthält (22d) die Forderung an die referentielle Argumentstelle λτ, daß sie durch eine als +def gekennzeichnete Einheit zu spezifizieren ist.16 Die Argumentadresse k beinhaltet die für die Topiks erforderliche Kongruenz bezüglich Topikstatus, Kasus, Genus, Numerus und Person.17 Die Boolesche Kondition für die Wertefestlegung von α und β besagt, daß das Klitikum in FP abwesend sein kann. Dann repräsentiert die Phrase in SpecF allein das Topik. Zwei weitere Bauelemente für die kompositionale Laut-Bedeutungs-Zuordnung müssen noch beleuchtet werden, und zwar zwei im System der semantischen Interpretation von sprachlichen Ausdrücken zur Verfügung stehende Templates. Mit Ortmann (1995) will ich - anders als in Zimmermann (1991b) - annehmen, daß die Bedeutung nichtrelationaler Substantive auf folgende Weise um eine Argumentstelle angereichert werden kann: (23) ARG(umentstellenerweiterung) XQ X x X r [ Q r ] A [ x R r ] -V+N

wobei R eine nicht näher spezifizierte Zugehörigkeitsrelation repräsentiert Angewendet auf die Bedeutung von kniga ergibt sich (24). (24)

λχλΓ[ΒυθΗΓ]Α[χΚΓ]

Auf diese Weise können alle Substantive neben der referentiellen Argumentstelle eine weitere Argumentstelle haben und dativische Klitika wie mu und die durch sie duplizierte Μα-Phrase erhalten in DPs mit einem nichtrelationalen Nomen wie kniga 'Buch', mit einem relationalen Nomen wie prijatel 'Freund', mit einem deadjektivischen Nomen wie gordost 'Stolz' bzw. mit einem deverbalen Nomen wie pristigane 'Ankunft' den gleichen Status, semantisch, informationsstrukturell und morphosyntaktisch. Vgl. : (1)

knigata mu na Ivo Buch-das cl-dat präp Ivo 'das Buch von Ivo'

(25) prijateljat mu na Ivo Freund-der cl-dat präp Ivo 'der Freund von Ivo'

16 17

Mindestens für DPs ist diese Bedingung wesentlich. Es sei hier erwähnt, daß es keinen Sinn macht, diese für Kongruenz relevanten Informationen in der erweiterten V- oder N-Projektion zu haben. Das käme zustande, wenn das Klitikum als F und nicht als F-Adjunkt angesehen würde.

Dativisches Klitikum in der DP-Struktur

129

(26) gordosttamu na Ivo ot uspexa Stolz-der cl-dat präp Ivo präp Erfolg-der 'der Stolz Ivos auf den Erfolg' (27) pristiganeto mu na Ivo Ankunft-die cl-dat präp Ivo 'die Ankunft Ivos' Ferner ist für die Integration von Modifikatoren das Template MOD vorzusehen (s. Zimmermann 1992). Es macht u.a. Adjektivphrasen zu Modifikatoren von NPs. (28) MOD(ifikationstemplate) XQ2 XQ1 Xr [ Qi r ] Λ [ Q2 r ] mit Qi, Q2 e S/N und mit Kongruenz der unifizierten Argumentstellen, wenn QI und Q2 durch +N-Einheiten ausgedrückt sind. Es ist wesentlich zu verstehen, daß dieses Template auf die Bedeutungsstruktur des Modifikandums via Funktionale Komposition angewendet werden kann, so daß es zur Argumentvererbung kommt. Ich illustriere das an novata kniga aus Beipiel (2). (29) MOD (novata') (ARG (kniga')) = XQ2 XQi λτ [ Qi r ] Λ [ Q2 r ] ( λχ [ NEU χ ]) (λ(2 λχ λΓ [ Q r ] Λ [ χ R r ] (λγ [ BUCH y ] )) = λχ λτ [[ BUCH r ] Λ [ χ R r ]] Λ [ NEU r ] Diese Repräsentation kann nun mit der Bedeutung der «α-Phrase auf zweierlei Art verknüpft werden, je nachdem ob diese Konstituente als nichttopikalische oder topikalische Einheit repräsentiert ist. Im ersteren Fall steht die «α-Phrase in NP (s. (7)), im letzteren Fall in der SpecFPosition (s. (12b)). Entsprechend ergeben sich folgende zu unterscheidende semantische Repräsentationen: (30) novata kniga na Ivo' = 0D' ((MOD (novata')(ARG (kniga')))(na Ivo')) = vr [[ BUCH r ] Λ [ IVO R r ]] Λ [ NEU r ] (31 ) novata mu kniga na Ivo' = 0D(0F (W)(MOD (novata')(ARG (biigd)))(na Ivo')) = ir [ y = IVO ] : [[[BUCHr] Λ [ y R r ] ] Λ [ N E U r ] ]

130

Ilse Zimmermann

5. Zusammenfassung Die hier vorgestellte Integration des dativischen pronominalen Klitikums in die Struktur der Substantivgruppe des Bulgarischen weicht in einigen Punkten von Grundannahmen der gegenwärtigen minimalistischen Syntaxtheorie ab. Das betrifft erstens den Mechanismus des feature checking. Ich rechne grundsätzlich mit dem Funktionieren der Bedingungen für die semantische Amalgamierung von Konstituenten, die in den mit Argumentstellen assoziierten Argumentadressen repräsentiert sind. Eine Zuordnung von Laut und Bedeutung, d.h. eine Derivation, bricht zusammen, wenn die betreffenden Voraussetzungen nicht erfüllt sind. Diese erforderliche Verträglichkeit zweier in der semantischen Amalgamierung zu kombinierenden Konstituenten ist in meinen Augen eine entscheidende Form des feature checking. Zweitens rechne ich mit Templates, die Bedeutungsstrukturen ohne entsprechende Formative anreichern. Anders als Siloni (1995) und Bailyn (1994) schreibe ich attributiv verwendeten Adjektivgruppen Modifikatorstatus nicht durch besondere syntaktische Konstituenten oberhalb von AP zu, sondern wende ein die Modifikatorfunktion lieferndes Template an. Mindestens für slavische Sprachen wie auch fürs Deutsche, Englische, Französische, Italienische halte ich diese Analyse für angemessen. Sie schließt nicht aus, daß es satzartige Modifikatoren gibt, die Sätzen vergleichbare funktionale Strukturdomänen aufweisen. Drittens benötige ich folglich weniger funktionale Strukturdomänen in erweiterten Projektionen, als üblicherweise angenommen wird. Insbesondere sehe ich keine Agr-Phrasen vor. Die betreffenden Verträglichkeiten sind in den Argumentadressen und entsprechenden Kombinationsvorschriften, z.B. MOD, verankert (s. auch Wunderlich 1992). Viertens mache ich massiv von syntaktischen Operationen Gebrauch, die für die PF bzw. für die SF nicht sichtbar sind. Dabei ist zu klären, welche Beschränkungen für die resultierendenRepräsentationen bestehen, insbesondere auch welche Rolle Spuren (oder Kopien) von bewegten Konstituenten spielen. Fünftens betrachte ich die hier analysierten Klitika nicht wie Rudin (1995) und Maaßen (1994) als funktionale Köpfe in der erweiterten lexikalischen Projektion von Verben oder hier von Substantiven, sondern als Adjunkte einer funktionalen Kategorie. Deren Funktion besteht darin, Topiks aus der lexikalischen Projektion VP bzw. hier NP herauszuheben und informationsstrukturell als im Diskurs gegebene Entitäten zu interpretieren. Sechstens sieht die Analyse der klitischen Pronomen des Bulgarischen keine Bewegung der Klitika vor. Sie werden in FP basisgeneriert und klitisieren in der prosodischen Struktur an einen adjazenten Wirt. Die Lexikoneinträge der Klitika enthalten in der phonologischen Charakterisierung entsprechende Bedingungen.

Dativisches Klitikum in der DP-Struktur

131

Im Ganzen versteht sich die vorliegende Untersuchung auch als ein Beitrag zur Beleuchtung der Rolle des Lexikons in der minimalistisch orientierten Theoriebildung zur Laut-BedeutungsZuordnung.

Literatur Bailyn, John F. (1994): The Syntax and Semantics of Russian Long and Short Adjectives: An X'-Theoretic Account. In: J. Toman (ed.), 1-30. Bierwisch, Manfred (1967): Syntactic Features in Morphology: General Problems of So-called Pronominal Inflection in German. In: To Honor Roman Jakobson, 239-270. The Hague, Paris: Mouton. - (1982): Formal and Lexical Semantics. Linguistische Berichte 80, 3-17. - (1987): Semantik der Graduienmg. In: M. Bierwisch & E. Lang (Hgg.), 91-286. - (1988): On the Grammar of Local Prepositions. In: M. Bierwisch, W. Mötsch & I. Zimmermann (Hgg.), 163. - (1989): Event Nominalizations: Proposals and Problems. In: W. Mötsch (Hg.), 1-73. - (1996): Lexical Information from a Minimalist Point of View. Manuskript. Bierwisch, Manfred & Ewald Lang (Hgg.)(1987): Grammatische und konzeptuelle Aspekte von Dimensionsadjektiven. Berlin: Akademie-Verlag (= Studia grammatica 26/27). Bierwisch, Manfred, Wolfgang Mötsch & Ilse Zimmermann (Hgg.)(1988): Syntax, Semantik und Lexikon. Berlin: Akademie-Verlag (= Studia grammatica 29). Óavar, Damir (1996): On Cliticizatian in Croatian: Syntax or Phonology? In: ZAS Papers in Linguistics, vol. 6, 51-65. Óavar, Damir & Chris Wilder (1994): Clitic Third in Croatian. In: Linguistics in Potsdam 1, 25-63. Chomsky, Noam (1995): The Minimalist Program. Cambridge, Mass., London, England: MIT Press. (= Current Studies in Linguistics Series, 28). Dimitrova-Vulchanova, Mila & Giuliana Giusti (1995): Quantified Noun Phrase Structure in Bulgarian. Erscheint in: D. Lightfoot & J. Toman (eds.). Dimitrova-Vulchanova, Mila & Lars Hellan (1995): Clitics and Bulgarian Clause Structure. Erscheint in: H. van Riemsdijk (ed.). Fanselow, Gisbert (ed.)(1993): The Parametrization of Universal Grammar. Amsterdam, Philadelphia: Benjamins (= Linguistik Aktuell 8). Fehrmann, Dorothee, Uwe Junghanns & Gerhild Zybatow (Hgg.) (erscheint): Formale Slavistik. Materialien der Konferenz Formale Beschreibung slavischer Sprachen, Universität Leipzig, 30.11.-2.12.1995. Fries, Norbert (1996): Flexionsmorphologie und Syntax des Imperativs im Deutschen und Neugriechischen. Mit einer Skizze der Flexionsmorphologie des Verbs für beide Sprachen. Lund (= Sprache und Pragmatik 42). Gallmann, Peter (1996): Die Steuerung der Flexion in der DP. In : Linguistische Berichte 164, 283-314. Gramatika na sävremennija bâlgarski kniäoven ezik. Tom 2. Morfologia. Sofìja: Izdatelstvo na Bâlgarskata akademija na naukite 1983. Grimshaw, Jane (1991): Extended Projections. Manuskript. Haftka, Brigitta (Hg.)(1994): Was determiniert Wortstellungsvariation? Studien zu einem Interaktionsfeld von Grammatik, Pragmatik und Sprachtypologie. Wiesbaden: Westdeutscher Verlag. Inkelas, Sharon (1990): Prosodie Constituency in the Lexicon. New York, London: Garland Publishing, Inc. Jackendoff, Ray S. (1975): Morphological and Semantic Regularities in the Lexicon. Language 51, 639-671. Jäger, Gerhard (1995): Topics in Dynamic Semantics. Dissertation, Humboldt-Universität Berlin. Erschienen als CIS-Bericht-96-92, Universität München, Centrum fur Informations- und Sprachverarbeitung Kenesei, Istvan (ed.)(1987): Approaches to Hungarian, vol. 2: Theories and Analyses. Szeged: Jate. Lang, Ewald (1987): Semantik der Dimensionsauszeichnung räumlicher Objekte. In: M. Bierwisch, E. Lang (Hgg.), 287-458.

132 -

Ilse Zimmermann

(1990): Sprachkenntnis, Objektwissen und räumliches Schließen. Zeitschrift für Linguistik und Literaturwissenschaft 78, 59-97. - (1994): Semanüsche vs. konzeptuelle Struktur: Unterscheidung und Überschneidung. In: M. Schwarz (Hg.), 25-40. Lightfoot, David & Jindíich Toman (eds.)(to appear): Formal Approaches to Slavic Linguistics, vol. 3. Ann Arbor: Michigan Slavic Publications. Longobardi, Giuseppe (1994): Reference and Proper Names: A Theory of N-Movement in Syntax and Logical Form. Linguistic Inquiry 25, 609-665. Lutz, Uli & Jürgen Pafel (eds.)(1996): On Extraction and Extraposition in German. Amsterdam, Philadelphia: Benjamins (= Linguistik Aktuell 11). Maaflen, Christine (1994): Spanische klitische Pronomina: "Es war einmal ein Wackernagelpronomen ..." In: Β. Haftka (Hg.), 345-361. Mötsch, Wolfgang (Hg.)(1989): Wortstruktur und Satzstruktur. Berlin (= Linguistische Studien des Zentralinstituts fur Sprachwissenschaft der Akademie der Wissenschaften der DDR, Reihe A, H. 194). Muysken, Pieter (1982): Parametrizing the Notion "Head". Journal of Linguistic Research 2.3, 57-75. Ortmann, Albert (1995): Possessive DPs und Grammatikschnittstellen. Vortrag auf der Konferenz "Zur Struktur von nominalen w-Phrasen und zum Status der D-Elemente". 5.-6.4.1995, Universität Tübingen, SFB 340. Pencev, Jordan (1993): Bâlgarski sintaksis. Upravlenie i svârzvane. Plovdiv: Plovdivsko universitetsko izdatelstvo. van Riemsdijk, Henk (ed.) (to appear): Empirical Approaches to Language Typology. Clitics in the Languages of Europe. - Berlin: Mouton, de Gruyter. Rizzi, Luigi (1995): The Finite Structure of the Left Periphery. Manuskript. Rudin, Catherine (1995): On Pronominal Clitics. - Paper presented to the First Conference on Formal Approaches to South Slavic Languages, Plovdiv, October 12-15, 1995. Schick, Ivanka P. & Ilse Zimmermann (1995): Flexive und klitische Annexe in der Definitheitskennzeichnung des Bulgarischen. Handout zu einem Vortrag im Wortbildungszirkel des Forschungszentrums Allgemeine Sprachwissenschaft, Berlin am 20.12.1995 - (1996a): Das dativische pronominale Klitikum in der Substantivgruppe des Bulgarischen. Erscheint in: D. Fehrmann, U. Junghanns & G. Zybatow (Hgg.) - (1996b): Das dativische klitische Pronomen in der Substantivgruppe des Bulgarischen. Manuskript Schmitt, Cristina. J. (1996): Aspect and the Syntax of Noun Phrases. Doctoral Dissertation, University of Maryland, College Park. Schwarz, Monika (Hg.)(1994): Kognitive Semantik. Ergebnisse, Probleme, Perspektiven. Tübingen: Narr (= Tübinger Beiträge zur Linguistik 395). Siloni, Tal (1995): On Participial Relatives and Complementizer Do : A Case Study in Hebrew and French. Natural Language and Linguistic Theory 13, 445-487. Steube, Anita (Hg.)(1990): Syntaktische Repräsentationen mit leeren Kategorien oder Proformen und ihre semantischen Interpretationen. Berlin (= Linguistische Studien des Zentralinstituts für Sprachwissenschaft der Akademie der Wissenschaften der DDR, Reihe A, H. 206). Stiebeis, Barbara (1996): Lexikalische Argumente und Adjunkte. Zum semantischen Beitrag von verbalen Präfixen und Partikeln. Berlin: Akademie Verlag GmbH (= Studia grammatica 39). Szabolcsi, Anna (1983): The Possessor Ran Away from Home. The Linguistic Review 3, 89-102. - (1987): Functional Categories in the Noun Phrase. In: I. Kenesei (ed.) 167-189. Toman, Jindñch (ed.)(1994): Formal Approaches to Slavic Linguistics. The Ann Arbor Meeting: Functional Categories in Slavic Syntax. Ann Arbor: Michigan Slavic Publications. Walter, Hilmar, & Georgieva Karvanbasieva (1987): Lehrbuch der bulgarischen Sprache. Leipzig: Enzyklopädie. Wilder, Chris (1994): Coordination, ATB and Ellipsis. In: C. J.-W. Zwart (ed.), 291-329. - (1996): Rightward Movement as Leftward Deletion. In: U. Lutz & J. Pafel (eds ), 273-309. Wilder, Chris & Damir Éavar (1994): X°-Bewegung und Ökonomie. In: B. Haftka (Hg.), 11-32.

Dativisches Klitikum in der DP-Struktur

133

Wunderlich, Dieter (1992): Towards a Lexicon-based Theory of Agreement. Theorie des Lexikons 20. Arbeiten des SFB 282. Düsseldorf: Heinrich-Heine-Universität. Überarbeitete Version in: Theoretical Linguistics 20 (1994), 1-35. Wunderlich, Dieter & Ray Fabri (1995): Minimalist Morphology: An Approach to Inflection. Zeitschrift ftlr Sprachwissenschaft 14, 236-294. Zimmermann, Ilse (1990): Zur Legitimierung leerer Köpfe. In: A. Steube (Hg.), 75-90. - (1991a): Die Syntax der Substantivgruppe: Weiterentwicklungen der X1-Theorie. In: I. Zimmermann (Hg.), 1-32. - (1991b): The "Subject" in Noun Phrases: Its Syntax and Semantics. In: I. Zimmermann (Hg.), 33-68. - (1992): Der Skopus von Modifikatoren. In: I. Zimmermann & A. Strigin (Hgg.), 251-279. - (1993): The Syntax of "Possessor" Phrases. In: G. Fanselow (ed.), 201-225. - (1995): 'Sich vertun' auf Deutsch und auf Russisch. Ein Sprachenvergleich zur Arbeitsteilung von Morphologie und Syntax. Handout zu einem Vortrag auf der Konferenz "Slavische und deutsche Sprachwelt. Typologische Spezifika der slawischen Sprachen im Vergleich mit dem Deutschen". 27.-31.3.1995, HumboldtUniversität Berlin, Institut für Slawistik. - (Hg.)(1991): Syntax und Semantik der Substantivgruppe. Berlin: Akademie-Verlag (= Studia grammatica 33). Zimmermann, Ilse & Anatoli Strigin (Hgg.)(1992): Fügungspotenzen. Zum 60. Geburtstag von Manfred Bierwisch. Berlin: Akademie Verlag GmbH (= Studia grammatica 34). Zwart, C. Jan-Wouter (ed.)(1994): Minimalism and Kayne's Symmetry Hypothesis. (= Groninger Arbeiten zur Germanistischen Linguistik 37).

Hilke Elsen (Universität

München)

The Acquisition of Past Participles: One or Two Mechanisms?

1. Introduction In the linguistic and psychological literature, the formation of the English past tense is the current topic for the debate between symbolists and connectionists concerning the number and kinds of mechanisms which guide the acquisition of regular and irregular morphology. Traditionally, two separate mechanisms are assumed to exist: rote, or more recently, associative and rule learning. Symbolists (e.g., Pinker & Prince, 1988; Pinker, 1991; Marcus, Pinker, Ullman, Hollander, Rosen & Xu, 1992; Marcus, 1995) argue that irrregular items are lexically represented and learned associatively or by rote. Regular inflection is achieved with the help of a rule. A U-shaped behaviour pattern results from the interaction between the two mechanisms. All early forms are produced correctly because each is stored as a whole in the lexicon. Then the child discovers the suffixation rule for regular inflection and applies it to irregulars as well. When an irregular form is used often, the application of the rule is blocked. Insufficient use of a form results in failure of the blocking device. Hence, overregularizations occur (e.g., goed, comed). Finally, the exceptions to the rule are recognized. Irregulars are stored in the lexicon, whereas regulars are formed using a rule. Properties of children's grammatical systems are responsible for the transition from rote to rule learning, independently of the lexical development. Recently, connectionists have offered an alternative account for the acquisition of inflection (e.g., Rumelhart & McClelland, 1986; Plunkett & Marchman, 1991, 1993). They have developed models both capable of memorizing patterns and generalizing regularities with the help of one single mechanism. When simulating the acquisition of the English past tense these models show many U-shaped patterns of error which add up to a U-shaped behaviour pattern comparable to children's output (see Plunkett & Marchman, 1991, 1993). Though only one associative mechanism is used, non-linear behaviour results. The onset of overregularizations is said to be triggered by a sufficient number of regulars in the lexicon which allows for the abstraction of general patterns. Thus, a relationship between lexical and morphosyntactic acquisition is assumed. It is suggested that the development in both domains is governed by the same learning mechanism (critical-mass hypothesis, see Plunkett & Marchman, 1993). The first overregularizations may be related to growth in vocabulary size. Incremental increases in new regular verbs correlate with qualitative shifts in the way forms are treated (Marchman & Bates, 1994). But first overregularizations are also related to the proportion of regular and irregular verbs in the lexicon. The level of generalization in simulations was gene-

Acquisition of Past Participles

135

rally low as long as the percentage of regulars remained below 50% (Plunkett & Marchman, 1993: 54). In other words, when a sufficient number of verbs was reached and the proportion of regulars exceeded the 50% mark, the network model was able to generalize regular patterns so that overregularizations were produced. Thus, in contrast to earlier models (e.g. Rumelhart & McClelland 1986), the regularizaron process does not depend on high frequency numbers of regulars. Furthermore, as Plunkett & Marchman (1993, in press) point out repeatedly, sensitivity to the phonological shape of the stem is an influencing factor on the acquisition of inflectional morphology. Interestingly, relations between lexical and morphological development were not found in Marcus et al.'s (1992) investigation of the spontaneous speech of 83 children. Evidence of a sudden non-linear increase of new verbs ('spurt') or changes in the proportion of regulars has not yet been found in naturalistic studies of individual children (Pinker & Prince, 1988; Marcus et al., 1992). One purpose of this paper is to present data from continuous observation of the acquisition of German verbs by A. which show a sudden acceleration in verb growth, a clear change in verb vocabulary composition and a relation between structural and quantitative changes in the verb vocabulary and the onset of overregularizing. The second aim is to compare results obtained from different data samplings: continuous longitudinal data from one child (Elsen, 1991), longitudinal data from periodic recordings of several children (e.g., Marcus et al., 1992, Clahsen & Rothweiler, 1993, Weyerts & Clahsen, 1994) and parental report information on a large population (Marchman & Bates, 1994). To what extent limited real-life data can be compared with network results is a question to be addressed. The discussion of learning mechanisms is based on the English past tense. In spoken German the past tense (Präteritum) is quite rare and is being replaced by the present perfect which is formed with the help of haben 'to have' or sein 'to be' and the past participle (Partizip Perfekt), e.g., ich bin gegangen Ί have gone', ich habe genommen Ί have taken'. This investigation will therefore focus on the acquisition of participles. In the following paragraphs, some basic facts about the formation of the German Partizip Perfekt are provided. German past participles may take the suffix -(e)t or -(e)n. Both may combine with a change in the stem vowel (Ablaut). Some verbs show a change in the final stem consonant (Grammatischer Wechsel). The use of the prefix ge- is phonologico-rhythmically conditioned. It is used independently of suffixation when the first syllable of the finite base form is stressed (note that in German some prefixes can be separated from the base for the construction of the verbal constituent in verb-second sentences), e.g., arbeiten 'to work' - sie arbeitet 'she works' - ge-arbeit-et 'worked', helfen 'to help'- sie hilft - ge-holf-en, mit arbeiten 'to collaborate'- sie arbeitet mit - mit-ge-arbeit-et, but vorbereiten 'to prepare'- sie bereitet vor - vorbereit-et, miß trauen 'to mistrust'- sie miß 'traut - mißtrau-t, ent gehen 'to escape'-

136

Hi Ike Elsen

sie ent geht - entgang-en. This study will disregard the use of the prefix because its acquisition is determined by phonologico-rhythmical factors alone. The use of the suffix is morphologically conditioned. German verbs are usually grouped into three classes: weak, mixed, strong. Weak verbs form past participles by adding -{e)t, e.g., kochen 'to cook'- ge-koch-t, hoffen 'to hope'- ge-hoff-t. Mixed verbs (about 8 verbs) form the participle with the help of -(e)t and change of stem vowel, e.g., rennen 'to run'- ge-rann-t, brennen 'to burn'- ge-brann-t. Strong verbs (about 170) form past participles with the help of -(e)n and Ablaut, e.g., gehen 'to go' - ge-gang-en, werfen 'to throw'- ge-woif-en. The vowel alternations of present, preterite and past participle stems follow various patterns. 88 verbs show an identical preterite and participle stem vowel, 36 verbs show an identical present and participle stem vowel and 49 verbs show three different stem vowels (numbers according to Heringer, 1989). However, there are further verbs which show non-regular participle formation, e.g. auxiliaries and modals (about 6 verbs, depending on definition, e.g. können 'can'- gekonn-t, wollen 'to want to'- ge-woll-t). In many syntactic constructions, the participles of modals appear as infinitives.1 For participle formation, modals may use vowel change and -(e)t. But they show different patterns of stem vowel alternation and inflection in the present singular forms2, among other things. A problem arises for some modals, e.g. wollen 'to want to', because they form their participles with the dental suffix, but show a different paradigm for the present tense. Auxiliaries like sein 'to be' and haben 'to have' have also to be treated separately, as they show suppletion (sein) or some mixed formations with Ablaut in the conjunctive and the dental suffix with assimilation, e.g., haben - hatte - gehab-t 'to have - had - had', du hast 'you have' (2nd sg present, instead of *du habst), sie hat 'she has' (instead of *sie habt), ich/sie hatte 'I/she had'(instead of *ich/sie habte). The expression 'regular verb' is used for weak verbs, 'irregular verb' for others, as the formation of their participles is not predictable. As yet it is not clear to what extent the knowledge of morphological paradigms may influence the formation of past participles. In this paper, however, for figures 2 and 3 numbers of regular verbs include the modals wollen 'to want to' and sollen 'shall', as both use the dental suffix for the formation of preterite and participle without vowel change, comparable to regulars (though they do not show the present tense patterns of regular verbs). The term 'overregularization' refers to the substitution of the suffix -(e)t for -(e)n, irrespective of vowel alternation, e.g., werfen 'to throw' - ge-werf-t (rather than geworfen), sein 'to be' - ge-wes-t (rather than gewesen). A list of the child's first verbs and overregularizations is given in appendix I and II.

2

Cf. Ersatzinfinitiv in most grammars of German, for example Er hat singen wollen vs. *Er hat singen gewollt 'he has wanted to sing' (cf. e.g. Engel 2 1991: 463f.). For example, German verbs show -e, -st, -t in the present tense singular. Modals do not form the first and third person singular with the help of a suffix. Compare ich sag-e — du sag-st — sie sag-t Ί say — you say — she says' and ich v/ill-0 - du will-st - sie wiIl-0 Ί want to - you want to - she wants to'.

Acquisition of Past Participles

137

In the following, several aspects of the girl's verb acquisition are presented and compared to reports on other children. A.'s data is discussed with regard to the dual-mechanism and singlemechanism account previously described. The predictions of the two approaches under discussion will be compared to the presented results: The two-mechanism approach states that the onset of overregularizations should not be related to measurable increases in the number of regular verb types and predicts a clear dissociation between the mechanisms which are responsible for lexical and morphological acquisition. In the single-mechanism account lexical and morphological acquisition is governed by the same basic learning mechanism so that interdependencies between both domains are expected. In particular, a non-linear critical mass relationship between lexical and morphological development should be found because the acquisition of a vocabulary of a certain size allows for the abstraction of general patterns. This leads to the production of the first overregularizations, that is, productive usage. Additionally, overregularizations should not appear in small vocabularies where irregular verbs outnumber regulars. One important question is whether a morphological rule is really needed to account for the formation of past tense and past participles. The results of the investigation of A.'s language acquisition are very similar to those predicted by the network models. That is, although the data presented here does not rule out a symbolic account which assumes two different mechanisms for the acquisition of verb morphology, it can easily be interpreted with the help of the more simple one-mechanism account. These findings shed new light on the relevance of network simulations. They illuminate possible limits of fragmentary data samples.

2. Method For this paper data is taken from a diary study of a German-speaking girl, Α., an only child. The mother was the only observer-recorder. Data on pronunciation and essential linguistic and non-linguistic situations was collected continuously during the child's whole waking-time: all new items (all new words, all new word-forms, all novel pronunciations of established items) were documented (IPA phonetic transcription). Imitations were distinguished from deferred imitations and spontaneous productions. Notes were transferred to filing cards twice a day with additional comments on situation, frequency of use and changes in articulation. The entire inventory of sounds, words and word forms was checked three times a month. Estimations of relative frequency such as 'often, rare, none' were made. On this basis insights into non-linear use of a lexical item or morpheme were obtained. Continuous notes ended when A. was 2;5 and had completely acquired the phonological system. That is, she was able to distinguish and produce phonemes and replace foreign sounds by German ones in non-native words (Elsen, 1991). As the mother and child were together permanently throughout the study, a complete

138

Hilke Elsen

recording of words, word forms, and phonological development during this period of time was obtained. Although the study concentrated on phonology, notes included word forms and word combinations. Comments on situation and referents and information about frequency and mortality of individual lexical items were also recorded. Additionally, audio-recordings were taken twice, later once a month (from 0;4 up to age 8). Some audiotaped data were verified by sonagrams and checked with a trained phonetician. Inteijudge agreements were over 80%. Audio-recordings as well as handwritten notes were used to describe the babbling-period and to allow checks on later development. After 2;5 data were collected first daily, then in greater intervals. The child's lexicon together with detailed data on phonological and lexical development up to 2;5 is given in Elsen (1991).

3. Results The present data collection presents several aspects already observed for other children (see Dromi, 1987; Marcus et al., 1992; Clahsen & Rothweiler, 1993; Clark, 1993). Figure 1 shows the number of new words that were acquired during each third of month. Total vocabulary in types does not increase at a constant rate. Nouns dominate the lexicon (Elsen, 1996b). In general, early verbs are correctly inflected e.g fallen 'to fall', machen 'to make', weinen 'to cry', kleckern 'to spill', vergessen 'to forget', schreiben 'to write', essen 'to eat'. Some, however, show incorrect participles right from the start, e.g. verloren 'lost', gewesen 'been'. Not all verbs are overregularized. Up to the beginning of 2;5, approximately 25% of the irregular verbs show overregularizations (sometimes one token) [see appendix II]. Less frequently, irregularizations of regular verbs are found {-(e)n added to a regular verb, use of a stem vowel which does not appear in the target present tense stems or both), for example hingelogen, correct hingelegt 'put down' (2;4,6), gespritzen, correct gespritzt 'syringed' (2;4,16). Invented verbs are inflected, e.g., gedongt 'has made 'dong" at 2; 1.6, bähacht, hat gebähacht 'has made 'bäh" at 2; 1.7. Sometimes the girl corrects herself, but the corrected form is not always closer to the target, e.g. fanaeogt] corrected to fansBogr)], correct angezogen 'dressed' (2;0,15), aufgestunden corrected to aufgestunken, correct aufgestanden 'stood up' (2;6,6). Double markings appear, for example [lönt], correct verloren 'lost' (1,6,2), fallent, correct gefallen 'fallen' (for several days at the beginning of 1 ;6, along with fallt), geziehten, correct gezogen 'pulled' (2;4,19). However, some new observations are made which are underestimated or which have not yet been reported in the literature. For the formation of a participle, the choice of stem may vary, e -g-> gegengt (geganSen 'gone') at 2;8.25, gegingt (preterite stem), gegangt at 2;8.28, gegeht (infinitive stem) at 2;9.7, gegangt at 2;9.13. ~(e)n and -(e)t may be added to irregular stems other than infinitive or participle, e.g. gerunterfällt correct runtergefallen 'fallen down'

Acquisition of Past Participles

Vi

re h o *

«E*

e

«2 SM O h 4) ε 3 Ζ.

ÖJD £

δ

140

Hilke Elsen

(2;1,19), weggelaufen correct weggelaufen 'run away' (2;4,2). Irregulars and regulars may have stem vowels which do not appear in the target stems at all, for example [fon], correct gefunden 'found' (1;5,24), fapazurjkt], correct abgezogen 'drawn off (2;1,4), geschmockt, correct geschmeckt 'tasted' (2;5,0) (for more examples see Elsen 1997).3 Several forms coexist, partly for months. A. applies regular and different irregular patterns to regular and irregular verbs. The examples show that inflection patterns are obviously less homogeneous than generally assumed and that the difference between regular and irregular formations are less clear cut than postulated by symbolists, which makes a rote/rule distinction less probable. There are several sudden non-linear increases ('spurts') in the rate of vocabulary growth (Fig. 1), when the acquisition of new words nearly doubles (at the end of 1;2, at the end of 1;5, at the beginning of 1;9, and at the beginning of 2;3) or at least clearly rises (at the beginning of 2;0). Looking at the rate of verb growth, we also see that there is a sudden non-linear increase in the acquisition of verbs. For few irregulars, the first usage is overregularized (see appendix II), e.g., Ioni 'losted', for which only overregularized forms are noted at the first day of appearance. When examining the spontaneously produced verbs we notice that the rate of acquisition of regulars differs from that of irregulars. In Figure 2 the accumulation of A.'s regular and irregular verbs is documented. At the end of 1;5, there is a point in time when the gap between regular and irregular verbs widens in favour of the regulars and an increase in new verbs is noted: after learning around one, two or three new verbs every ten days, the child acquires seven new verbs at the end of 1;5, 18 verbs at the beginning of 1;6, then 14 verbs at the middle of 1;6. Figure 3 presents the proportion of regular and irregular verb types in relation to the total verb vocabulary level.4 Irregulars initially predominate. After some time, the proportion of regulars increases continuously. The graphs are clear evidence that verb vocabulary composition is undergoing substantial change. When the rate of vocabulary rises at the end of 1;5, the rate of regular verbs rises, too. All verbs which were produced spontaneously up to this point of time are listed in appendix I. There are thirty verbs in the child's lexicon at 1;5,27, 15 regulars and 15 irregulars. If verbs with different prefixes are counted once and the irregular verb haben 'to have' is treated as weak according to the dental suffix of the past participle (haben - ge-hab-t), 13 irregulars and 15 regulars are recorded by the end of 1;5. These differences in ratings do not seem to be crucial. The assignment to the part of speech 'verb' might be questioned, especially for the first verbs. But their use as verbs soon becomes plain either because of their inflectional behaviour or use in correct situations. When the increase of new verbs is noted, the first overregularized form is found: [dafalt], correct gefallen 'fallen' (1;5.30). Further overregularizations appear at 1;6.2 [lönt], correct

3

4

Even errors with ge-placement were found, e.g. gerunterfällt, correct runtergefallen 'fallen down' (2;1,19), schubstgehat, correct hat geschubst 'has pushed' (2;8,8). For the proportion of A.'s regular and irregular veibs as a function of age see Elsen (1997).

o - a -a -fíw a -a -ζ'τνι a -a - iSw - a a -oíw -a -a iiUw h a -a OI-I w ha -a tfm h a a Htfiw - o -a -¿:iw a -a - 9¡I W -a a - SilW - a -a tfiw a -a - £iw - a a - z'i η - a -a - ÑM a -a o'iw - a -a - iium - a a -oittw a -a - (POW - a o

! » M (β

Hilke Elsen

142

M

α

Ρ S s -e α b

>

•e £ •s

Iκ ' · n O'o

S

8 seSeiueaMd

S 8'

Acquisition of Past Participles

143

verloren 'lost', at the end of 1;6 [mitanêmt], correct mitgenommen 'taken along with', and at 1;7.24 [vest], correct gewesen 'been'. According to Marcus et al. (1992), low-frequency irregular verbs are inherently prone to overregularization. Note that these are not low-frequency verbs, especially not in talking to children. To what extent frequency is the decisive factor for overregularizations is discussed in detail in Lindner (this volume). Furthermore, some show overregularizations from their very first usage. At 1;6.2 only overregularized tokens are recorded for Ioni. These examples mark the beginning of overregularizations at a point of compositional change and abrupt increase of new regular verbs at the end of 1;5. Interestingly, at this time further productive verb inflection is recorded such as 3rd sg. present tense for e.g., schlafen 'to sleep', essen 'to eat', sitzen 'to sit', wischen 'to wipe'. However, overregularizations do not appear at a constant rate. There are three new overregularization types at the end of 1;5/1;6. The first two are used several times for a few days. Then all three forms disappear. Only one new type with one token is produced at 1;7. The same holds for 1;8. Then there are three new overregularization types at 1 ;9, but nothing new during 1 ; 10/1 ; 11. At 2;0 seven new types are recorded. At 2;1 eight new types are noted. At 2;2 one new type appears. At 2;3, three new overregularization types are found. At 2;4 it's only one. Though most of them only appear once initially, for runterfallt 'fallen down' frequent use is noted at 1 ; 11/2;0, for getrinkt 'drunk' during 2;0/2;l, for geschmeißt 'thrown' during 2;0/2;l, for weggetunt 'put away' during 2;1, for reingetunt 'put in' during 2;1. After daily note taking ceased, observations show that times of high and low production of overregularized forms still alternate. That is, we can see waves of development. It is interesting to note that a further increase in the acquisition of regular verbs is found at 1 ;9 when we find an increase in new overregularization types. The same happens at 2;0, when types and tokens of overregularizations increase. Overregularizations and sudden accelerations in verb growth show a correlation, as overregularizations show the same pattern of development as verb vocabulary development itself. These waves are investigated in more detail by Lindner (this volume) who found systematicity in the way different patterns of erroneous participle formation were used over time. Taken together the data is in line with the findings of the acquisition of the English past tense as shown by Marchman & Bates (1994): The acquisition of verbs shows a sudden nonlinear increase. The composition of the verbal lexicon undergoes changes. When structural and quantitative changes in the verb vocabulary appear, that is, when regulars exceed the 50% mark, the first overregularizations are found. Furthermore, for the production of overregularizations a wave-like development is noted which correlates with non-linear increases in the verb lexicon. Moreover, the formation of incorrect past participles shows various patterns which may coexist. Overregularizations as well as irregularizations, though less frequently, are found. Regulars and irregulars have double-markings, stems other than infinitive or participle or nonexisting stems, both with -(e)I or -(e)η suffixes.

144

Hilke Elsen

4. Discussion The data shows that the child produces her first overregularizations when the composition of the verb vocabulary changes and the proportion of regulars exceeds 50%, as predicted by the network. If the surge of morphological errors was simply due to more opportunities for errors because of more new verbs, then the number of overregularizations should increase in the course of development as numbers of new verbs (and, thus, numbers of opportunities) rise, too. The numbers of verbs increase throughout the study, but the use of overregularizations sometimes rises (at the end of 1;5/1;6, at 1;9, at 2;0/2;l), then again drops (at 1;7, 1;10, 2;2). Another possible explanation is that the increase of overregularizations might be due to more talking in general. However, the investigation of noun plurals, which is currently under way, shows that overregularization rates of nouns do not correlate with those of verbs, which should be expected if such errors were simply the result of more talking. So it is highly probable that the onset of overregularizations was triggered by quantitative and structural changes in the verb vocabulary. These findings contrast with those of Pinker & Prince (1988) or Marcus et al. (1992). Marcus et al. (1992) report data on children that show no sudden non-linear increase in new verbs. Furthermore, the 'onset' of overregularizations does not correlate with compositional changes in the lexicon (cf. Marcus et al. 1992: 86-88, 99). The same holds for Weyerts & Clahsen's (1994) data on Simone. Again, there is no sudden non-linear increase of new verbs. Simone's first overregularizations are noted before changes in the composition of the verbal lexicon appear, which may be due to an underestimation of the number of regular verbs in the Weyerts & Clahsen data. In non-continuous samples only higher frequency items can be noted. Rare tokens will be missed. Consequently, the degree of variation as well as the actual amount of types will be underestimated. The present data shows that several verbs are initially produced with very low frequency, in part less than once a week (considering only spontaneous productions). Essen 'to eat' is said once at 0; 11.3, once at 1;3.4, at 1;3.7 and then daily. Lassen 'to let' is produced once at 1 ;6.10. The next productions are noted at 1;7.0. Singen 'to sing' is recorded once at 1;3.22, once at 1;6.11 and at 1;6.20. Then it appears more often. Some verbs are produced daily from the start, e.g., anziehen 'to dress', though most early verbs are produced only occasionally during several days. The same holds for irregularizations and overregularizations. Especially the first overregularized forms are quite infrequent. West (gewesen 'been'), for example was recorded only once (1;7.24). Low frequency items like those mentioned will only be noted in data with high recording density. Thus, differences in results might be due to the method of data collection. Token frequency of many regular verbs is initially extremely low, in German as well as in English, whereas frequencies for many irregulars tend to

Acquisition of Past Participles

145

be higher.5 Thus, numbers of types in a continuous sample will be very different from those in non-continuous samples. There the rise of new verbs depends upon an increase of token numbers. A related problem arises for sudden non-linearities. According to connectionists, and in contrast to symbolists's assumptions (e.g. Marcus 1995), non-linearities in vocabulary growth are a contributing factor to the onset of overregularizations, not a necessary or sufficient condition (Marchman & Bates 1994, Plunkett & Marchman in press). Nevertheless, although the relation to morphological development is not clear, it should be stressed that for A. a sudden spurt-like increase in verb vocabulary exists. Again, such non-linearities in Marcus et al.'s and Clahsen's data may have been missed. However, careful investigation of weekly recordings do show non-linear verb acquisition and changes in the composition of the verbal lexicon in relation to early overregularizations, as Champaud (1996) demonstrated with data on a French child. The same problem arises for early overregularizations, which might be, initially, also quite infrequent. We cannot rule out the possibility that the child's second or third wave of overregularizations, when type and token numbers increase, is equivalent to the first overregularizations found in other children. In other words, investigations such as that of Marcus et al. (1992) are very likely to miss early regulars, which tend to have low frequencies, as well as relationships between vocabulary and 'onset' of overregularizations. Furthermore, as pointed out by Plunkett & Marchman (1993: 63), Marcus et al.'s recordings probably occurred during a later period of verb acquisition when a regularization process had already set in. Thus they do not necessarily contradict network results. What is more, as most rare tokens will be missed in periodic recordings, the actual amount of variation will be underestimated. Lindner (this volume) analyzed A.'s overregularized forms from particular verbs and found systematic variation. Variety of error types and variability across children cannot be explained by symbolists, but is in line with a one-mechanism approach (cf. Plunkett & Marchman in press). To gain quantitative results on type numbers in a child's actual repertoire from periodic recordings is not possible. Comparability with network results is questionable. To compare reallife information on type numbers with network observations, we have to rely on continuous samples from single children. On the other hand, there is still a further method of data collection: the parental report technique (see Marchman & Bates, 1994). When parents are asked about their children's linguistic abilities with the help of checklists, data on the development of a great many children can be obtained, though only for those aspects which are explicitly requested in the list. Facts not mentioned will be missed. However, Figure 3 is quite similar to Marchman & Bates' (1994) results. They collected data on 1130 children between 1;4 and 2;6 using the parental report technique and found that early in acquisition irregulars predominated before a shift in favour of the regular verbs was noted. These then predominated. Marchman & 5

Interestingly, irregulars and regulars seem to differ in lexical frequency effects. Large effects on language processing are found for irregulars, whereas hardly any effects are found for regulars (cf. Stemberger 1994). This emphasizes the relevance of type numbers as opposed to token numbers.

146

Hilke Elsen

Bates stated that 'the shape of the relationship between verb learning and the productive use of past tense morphology [was] consistent with the non-linear 'critical mass' assumptions of the single-mechanism theory' (Marchman & Bates 1994: 360). Thus, results from continuous data collection match those from a different methodology.6 As all three sampling methods have their limits, their results should be considered together, though not without reservation. In two investigations on the acquisition of German past participles, the network model is assumed to be able to generalize the regular pattern only because of high numbers of regular verbs in the input instead of reaching a critical mass (Clahsen & Rothweiler, 1993, Weyerts & Clahsen, 1994). In English, regular forms outnumber irregulars. In German, regulars and irregular forms are said to be found in equal numbers. Thus, Clahsen and colleagues conclude that learning of the regular pattern must be due to a symbolic inflectional rule. But the model does not need high frequencies of regular verbs to generalize the regular pattern. As already pointed out, a change in the proportion of regulars and irregulars together with a growth in vocabulary size triggered the formation of regular patterns in the model. It is exactly this relation between changes of quantitative and structural properties of the verb vocabulary and first overregularizations that is found in the present data. Regular forms are not more frequent than irregular forms, and nevertheless the child overregularizes. Therefore, though the idea of a rule is not explicitly excluded, it is not necessary to assume a symbolic rule for the formation of regular inflection. Furthermore, Clahsen and colleagues found that -n was not overapplied in their corpora.7 „In all the data there are no participles in which an irregular stem pattern has been extended to a weak verb" (Clahsen & Rothweiler 1993: 1) and the only kind of stem error they found were regular stems replacing irregular ones (ibd.: 1 and cf. ibd. : 20f., 31). For their 70 children, Weyerts and Clahsen (1994) found only three cases of irregular stems replacing regular stems, three cases of wrong marked stems of irregular verbs and only one example of a double marking (ibd.: 449). Thus, they conclude that a child can qualitatively distinguish between regular and irregular inflection. However, A.'s data shows that inflection patterns are obviously less homogeneous than generally assumed and differences between regular and irregular patterns are less clear-cut than postulated by symbolists. This makes a rote/rule distinction less probable but is consistent with a one-mechanism approach. A. applies irregular patterns, vowel change or -n suffix or both, to regular and irregular verbs. Not only stem vowels from various verb forms are found (such as 2nd/3rd person singular present tense (ge-nimm-t, correct genommen 'taken'), past tense (ge6

7

In several previous investigations, A.'s phonological and lexico-semantic acquisition already yielded parallels to other children ( Elsen, 1991, 1994,1995, 1996a,b). Only the participle affix -t is overregularized by the children, whereas the -n affix is not overextended" (Clahsen & Rothweiler 1993:1). „it is only the regular -t suffix which is overregularized by children" (Weyerts & Clahsen 1994: 430) (Emphasis mine), though actually -n errors were found to be 12,5% of suffix-errors in Weyerts & Clahsen's data. Irregularization errors did not occur in their data (Clahsen & Rothweiler 1993: 31).

Acquisition of Past Participles

147

ging-t, correct gegangen 'gone '), conjunctive (ge-läg-en, correct gelegen 'lain')) but various examples of past participles with stems that do not appear in the target stems at all (e.g. getrogen, correct getragen 'carried', gewoppt, correct gewippt 'balanced'). Both -n and -t are applied to wrong stems. There was only one child in Clahsen & Rothweiler (1993) who had comparable results: Peter, an SLI child. His inflectional system was treated as a special case. Furthermore, some of A.'s erroneous forms were used often during several months and were more frequently applied than the correct participles (e.g. ausgezungen, correct ausgezogen 'taken o f f ) or showed both -t and -n interchangeably (e.g. trunken, trunkt, correct getrunken, getan, getanen, getant, correct getan). In contrast to Clahsen & Rothweiler's results (1993: 23), A.'s incorrect forms cannot be called 'occasional'. Her data calls into question the claim that the child qualitatively distinguishes between regular and irregular verb inflection because marked as well as unmarked stems and regular as well as irregular suffixes were overapplied, partly for several months. Thus, we suggest that both irregular and regular inflection are based on one associative learning mechanism. The present data is consistent with the current connectionists' position. As already noted by Marchman & Bates (1994), the beginning of overregularizations should be triggered by the composition of the verbal lexicon. This contradicts Marcus et al.'s (1992) claims of independent development of the lexicon and verb morphology. But it is in line with the 'critical mass' hypothesis. However, the present results are not entirely inconsistent with a symbolic approach as long as one accepts the existance of an input trigger for the appearance of overregularizations. Some further facts are consistent with a connectionist framework. There are irregularized regular verbs which, according to symbolists, should not be produced, at least not in higher numbers and certainly not several times per token. They cannot be explained by symbolic approaches. The same holds for the variation found for A.'s patterns of incorrect participle formation which should not appear according to symbolists. Thus, a qualitative distinction between regular and irregular verbs should be questioned. The production of the first overregularizations at a time of change in verb vocabulary composition is not predicted by a symbolic account, nor are further correlations between verb lexicon and overregularization patterns.

5. Concluding Remarks After presenting data of the acquisition of German verbs, the observations were compared to findings from different data samples. Several results were quite similar to those found in Marchman & Bates (1994) for the acquisition of the English past tense. The acquisition of verbs shows a sudden non-linear increase. The composition of the verbal lexicon undergoes changes. The first overregularizations are found when structural and quantitative changes in the verb vocabulary appear. These and further observations were found to contrast with recent

148

Hilke Elsen

symbolists' investigations of periodically recorded samples. It was argued that differences are due to the method of data collection. Though A.'s data does not rule out a symbolic account entirely, some findings are more in line with a single-mechanism approach. Changes in the rate of verb vocabulary acquisition coincide with overregularization patterns, resulting in a wave-like development The onset of overregularizations appears when the proportion of regulars exceeds the 50% mark. The formation of past participles shows various patterns: overregularizations, irregularizations, double-markings, stems other than infinitive or participle or non-existing stems, both with -(e)t or -(e)η suffixes. Different forms may coexist. Consequently, the assumption of a qualitative distinction between regular and irregular inflection and the existence of a symbolic rule is questioned. Instead, the data suggests a relationship between lexical and morphosyntactic development and the use of one associative learning mechanism for both irregular and regular inflection. Further detailed investigations are needed to see whether the present findings are due to idiosyncrasy or whether current theories must be refined so that, one day, more of the developmental facts can be explained. It is to be hoped that the dawning of the new millenium will see the divergent attitudes converge.

Acknowledgements Parts of this paper were presented at the 7th International Morphology Meeting, Workshop The Acquisition of Morphology in LI, Vienna, February 1996, and at the 18th Annual Meeting of the Deutsche Gesellschaft fur Sprachwissenschaft, Workshop Models of Inflection, Freiburg, February 1996, together with Katrin Lindner, who concentrated on blocking and further investigated the waves. A shorter version of this article was given at the 21st Boston University Conference on Language Development, Boston, November 1996. I am grateful to Katrin Lindner, Conxita Lieo, Teresa Parodi and Kim Plunkett for their comments on earlier drafts of this paper. Thanks to Hans J. Hanke for help with the plots. The preparation of this work was facilitated by grants from the Freistaat Bayern.

References Champaud, Christian (1996): Overgeneralization in early acquisition of verbal morphology: the case of French. Paper presented at the Workshop: The Acquisition of Morphology in LI, 7th International Morphology Meeting, February 1996, Vienna. Clahsen, Harald & Rothweiler, Monika (1993): Inflectional rules in children's grammars: evidence from German participles. Yearbook of Morphology 1992. 1-34. Clark, Eve (1993): The Lexicon in Acquisition. Cambridge: Cambridge University Press. Dromi, Esther (1987): Early Lexical Development. Cambridge: Cambridge University Press. Elsen, Hilke (1991): Erstspracherwerb. Der Erwerb des deutschen Lautsystems. Wiesbaden: DUV. - (1994): Phonological constraints and overextensions. First Language 14. 305-15.

Acquisition of Past Participles -

149

(1995): Der Aufbau von Wortfeldern. Lexicology 1. 219-242 (1996a) Two routes to language: stylistic variation in one child. First Language, 16. 141-158. (1996b): Linguistic team-work - the interaction of linguistic modules in first language acquisition. In: Clark, Eve (ed.) The Proceedings of the Twenty-seventh Annua! Child Language Research Forum. Stanford: CSLI Publications. Elsen, Hilke (1997):. Acquiring verb morphology: German past participles. Proceedings of the Boston University Conference on Language Development 21, 1996. Boston: Cascadilla Press. Engel, Ulrich (21991):. Deutsche Grammatik. Heidelberg: Groos. Heringer, Hans J. (1989): Lesen lehren lernen. Eine rezeptive Grammatik des Deutschen. Tübingen: Niemeyer. Lindner, Katrin (this volume) Overgeneralization revisited: The case of German past participles. Marchman, Virginia A. & Bates, Elizabeth (1994): Continuity in lexical and morphological development: a test of the critical mass hypothesis. Journal of Child Language 21. 339-366. Marcus, Gary F. (1995): The acquisition of the English past tense in children and multilayered connectionist networks. Cognition 56. 271-279. Marcus, Gary F., Pinker, Steven, Ullman, Michael, Hollander, Michelle, Rosen, T.John & Xu, Fei (1992): Overregularization in language acquisition. Monographs of the Society for Research in Child Development, 228.57.4. Pinker, Steven (1991): Rules of language. Science 253. 530-535. Pinker, Steven & Prince, Alan (1988): On language and connectionism: Analysis of a parallel distributed processing model of language acquisition. Cognition 28. 73-193. Plunkett, Kim (in press): Connectionism and developmental psychology. Journal of Child Psychology and Child Psychiatry. Plunkett, Kim & Marchman, Virginia (1991): U-shaped learning and frequency effects in a multi-layered perceptron: Implications for child language acquisition. Cognition 38. 43-102. - (1993): From rote learning to system building: acquiring verb morphology in children and connectionist nets. Cognition 48. 21-69. - (in press): Learning from a connectionist model of the acquisition of the English past tense. Cognition. Rumelhart, David & McClelland, James L. (1986): On learning the past tenses of English verbs. In: McClelland, James L., Rumelhart, David and the PDP Research Group (eds.): Parallel Distributed Processing. Vol. II. Psychological and Biological Models. Cambridge, MA: Bradford Books/MIT Press. 216-270. Stemberger, Joseph P. (1994): Rule-less morphology at the phonology-lexicon interface. In: Lima, Susan D., Corrigan, Roberta L. & Iverson, Gregory K. (eds.). The Reality of Linguistic Rules. Amsterdam/Philadelphia: John Benjamins. 147-169. Weyerts, Helga & Clahsen, Harald (1994): Netzwerke und symbolische Regeln im Spracherwerb: Experimentelle Ergebnisse der Flexionsmorphologie. Linguistische Berichte 154. 430-460.

150

Hilke Elsen

APPENDIX I A. 's first 30 verbs (spontaneously produced, regulars underlined) No. age

verb

gloss

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30

essen trinken kleckern anziehen stehen gucken wickeln wischen singen rutschen ausziehen schlafen aufmachen sitzen aufräumen putzen klingeln drücken gehen drehen tanzen sehen haben finden kaputtmachen regnen warten helfen fallen geben

to eat to drink to spill to dress to stand to look to swaddle (a baby) to wipe to sing to slide (e.g., on a slide) to take off to sleep to open to sit to tidy up to clean to ring to press, to excrete (euphemism) to go to turn to dance to see to have to find to destroy, to spoil, to ruin to rain to wait to help to fall to give

beg. 0; 11 0 11.18 1 2.30 1 3.5 1 3.8 1 3.13 1 3.15 1 3.16 1 3.22 1 3.25 1 3.28 1 4.4 1 4.8 1 4.8 1 4.9 1 4.11 1 4.28 1 5.2 1 5.3 1 5.11 1 5.16 1 5.16 1 5.18 1 5.24 1 5.25 1 5.26 1 5.26 1 5.26 1 5.26 1 5.27

Acquisition of Past Participles

151 APPENDIX II

A. 's first overregularized verbs age

child's form

target

gloss

further deviant participles

1;5.30

defallt [dafalt]

gefallen

fallen

fallt, fallnt

1,6.2

lont* [lönt]

verloren

lost

end 1;6 1;7.24 1;8.21

mitgenehmt* [mitanemt] mitgenommen gewesen west* [vest] runterfallt* [xuintafalt] runtergefallen

1;9.3 1;9.23 1;9.26

gehaut* trunkt wehgetant

gehauen+ getrunken wehgetan

beaten drunk hurt

2;0 2;0.10

getrinkt eingereiht* [ainairaipt]

getrunken eingerieben

drunk rubbed in

2;0.14

abgebeißt fapabais^t]"

abgebissen

biten off

taken along with been fallen down getrinkt wehtanen (1;7,3), wehd/getanen, wehgetunt, -getui trunkt

2;0.15

annesongt fanaGogt]

angezogen

dressed

annesoggen

2;0.16 2;0,17 2;0/2;l 2;i 2;1 2; 1,3 2;1.4

reinetunt genehmt geschmeißt wehgetunt weggetunt weggenimmt abbesunkt fapszugkt]

reingetan genommen geschmissen wehgetan weggetan weggenommen abgezogen

put in taken thrown hurt put away take away drawn off

reingetut

2;1.5 2;1.9 2; 1.15

hinneschmeißt* weggenehmt gelauft geschlaft gegeht ausesiehnt faus^as^ïnt]

-geschmissen weggenommen gelaufen geschlafen gegangen ausgezogen

chucked taken away weggenimmt run slept gone aus(e)d/s/zungen, taken off

gekommt weggetant gehelft geziehten gekannt

gekommen weggetan geholfen gezogen gekonnt

come put away help pull been able to

2; 1.22 2,2.14 2-3.21

2;3,25 2;3.28 2;4.19 2;5.0

(see above) weggetant weggenehmt

auseziehne weggetunt

* the first participle token was overregularized + for this verb regular participles exist although they are not used by the mother (prime caretaker) # the notation [s"'] refers to a slighly [J]-like pronunciation of [s]

Katrin Lindner (Universität

München)

Overgeneralization Revisited: the Case of German Past Participles1

1. Introduction Over the last decades the conceptualization of the acquisition process of inflection has undergone various changes. Behaviorists believed in pattern learning. In the late fifties, however, the findings by Berko about children's early application of the plural suffix -s to nonce nouns and the suffix -ed to nonce verbs led researchers to reconceptualize the learning process in terms of the acquisition of rules. In some frameworks, in particular that of transformational grammar and its later developments, such rules have the status of mental operations. Yet not all formations are rule-governed: irregular verb forms have been said to be learned by rote, i.e. the appropriate form is listed in the lexicon. This "rote & rule theory" (Pinker & Prince 1992) has seriously been called into doubt when Rumelhart and McClelland (1986) published their study about a simulated development of the past tense in a connectionist framework. They demonstrated that in training a pattern associator with pairs of present tense or base and past tense forms of sets of irregular and regular verbs they are able to mimick the developmental curves ("U-shaped behavior", see below) found in studies based on natural data from children (e.g. Brown 1973; Ervin 1964; Bybee & Slobin 1982; Slobin 1973). The model associates input and output nodes via an intricate network of nodes. When the model starts to learn it compares the stored configuration with the input. If both do not match the connections are adjusted to accomodate the new input. In the beginning all connections are set at 0. Following frequency counts for English verbs (by Kucera & Francis 1967) the model was trained on a representative sample of English regular and irregular verbs: it was first fed with the 10 most frequent verbs in English (8 of them irregular), then with 410 verbs (76/410 irregular) - simulating a vocabulary spurt -, at last with 86 verbs of low frequency (with 14/86 irregular). Rumelhart and McClelland showed that learning can be conceptualized as a process of internally adjusting patterns to new evidence, that only one multi-componential processing mechanism is necessary for learning the relevant patterns, and that there is a correlation of lexical and morphological development (cf. the second training phase as a vocabulary spurt). These findings started the 1

This paper developed out of a joint presentation with Hilke Elsen at the Worshop 'The Acquisition of Morphology in LI' at the 7th International Morphology Meeting in Wien and at the annual meeting of the Deutsche Gesellschaft fur Sprachwissenschaft in Freiburg in February 1996; Hilke Elsen concentrated on the "critical mass hypothesis" (cf. Elsen this volume). Thanks to her for discussions and for sharing some of her unpublished data with me. To the Max-Planck-Institut fur Psycholinguistik in Nijmegen I am indebted for a research stay during which this paper was written. For comments and helpful suggestions on earlier drafts I want to thank Heike Behrens, Monika Baumann, Martha Crago, Elisabeth Löbel and Teresa Parodi.

Overgeneralisation Revisited: the Case of German Past Participles

153

debate between connectionists (e.g. Rumelhart, McClelland, Bates, Marchman and Plunkett) and symbolists (e.g. Pinker, Marcus and their collaborators) on the number and kinds of mechanisms involved in learning inflection. Under the influence of associative networks (cf. also Bybee 1985; 1988) the "rote & rule theory" was modified since rote learning did not adequately capture facts about irregular inflection: for instance, that there are different degrees of similarity among strong verbs (cf. suppletive sein 'to be' on the one side and, on the other, "families" of strong verbs like singen, wringen, sinken, trinken, - 'to sing', 'to wring' to sink', 'to drink' - etc.) or that, in experimental settings, nonce verbs resembling strong verbs in phonological structure may be inflected analogously by adult participants (Bybee & Moder 1983; Prasada & Pinker 1993). Thus the "rote part" (Pinker & Prince 1992) was extended. Irregulars are now considered to be "memorized pairs of words, but the linkages between the pair members are stored in an associative memory structure with certain connectionist-like properties" (Pinker & Prince 1992:233). This new symbolist model has become known as the 'dual mechanism model': irregular forms are learned via associations and stored as wholes in the lexicon while regular verb forms are rulegoverned. However - following Plunkett (in press) - a crucial question has not been answered yet: how do children identify the (default) rule in their language?2 The answer is particularly important if a language provides more than one type of regular inflection as, for instance, Norwegian and Icelandic (cf. Ragnasdóttir et al. 1996). In the meantime other simulations have been run modifying some of the properties of the Rumelhart & McClelland model objected to (e.g. Plunkett & Marchman 1991; 1993; cf. also Plunkett 1995; in press). The relationship between morphological an d lexical development severely criticised by symbolists (cf. Pinker & Prince 1988; Marcus et al. 1992) - is maintained.3 Crucial is still the composition of the verb vocabulary. However, the matter is now considered in terms of a "critical mass" that has to be reached for generalizations - and consequently overgeneralizations - to become effective. The shift in vocabulary composition from a larger proportion of irregular verbs to a larger proportion of regular verbs occurs "when the verb vocabulary expands from 20 - 60 verbs (i.e. from 42% regulars to 67% regulars)." (Plunkett & Marchman 1993:62). Such findings have been confirmed with natural data from

3

For verb inflection Marcus et al. solve this question in the syntax. Tense marking is triggered by the respective head since every main clause has to have tense. Yet regular and irregular inflection also exists independent of syntactic wellformedness conditions, e.g. with past participles or plurals, and the theory should be able to account for regular and irregular inflection in general. This relationship has been one of the major issues in the debate. The correlation of an explosive vocabulary growth (as simulated by the second phase in Rumelhart & McClelland's model) with overgeneralizations was interpreted in two directions: (a) that children's input includes more regular than irregular verbs or that (b) children's data show a vocabulary spurt of regular verbs. According to Marcus et al. (1992:80ff.) neither (a) nor (b) is confirmed by the data from the Harvard children. - Related to the issue about high frequency of regular verbs is the claim that connectionist models depend on frequency (e.g. Weyerts & Clahsen 1994). That this is not the case is shown by Plunkett (in press).

154

Katrin Lindner

children learning English or German (Marchman & Bates 1994; Elsen this volume; cf. also Champaud 1996 for French data). Ever since Berko's study, children's regularization of forms - be they nonce words or irregular forms - have been a matter of great interest. 'Regularizations', 'overregularizations' (cf. Marcus et al. 1992) or - as they are called more often - 'overgeneralizations' were first found to occur at a particular phase in children's development. Various studies (e.g. Ervin 1964; Slobin 1973; Bowerman 1982a; 1982b; Bybee & Slobin 1982) have shown that in the beginning children produce correct forms, presumably learned by rote; in a subsequent period they produce both correct and overgeneralized forms until they start to use the correct forms predominantly." This stage-like development became known as a particular developmental curve called 'U-shaped behavior'. For some time, however, there have been questions about the frequency of overgeneralizations, the duration of each "stage" and the use of the concept "stage" in this matter (e.g. Maratsos 1979, Bowerman 1982b). Plunkett & Marchman (1991) found in their simulations of learning the English past tense that percentages of overgeneralizations are low. Moreover, they found that the over-application of the suffix occurs selectiveley with some irregular verbs while others are still inflected correctly, and that such errors may appear, disappear and reappear over a protracted period of time. This behavior they called the "micro-Ushaped behavior". Calculating the overgeneralization rate from data of 25 English-speaking children Marcus et al. (1992:35f.) stated as well that it is fairly low (median from 1% to 24% for the individual children). They also noted that overgeneralizations still exist at low level in school-age years and are extremely rare among adults suggesting quantitative differences between children and adults. Thus what is left of the 'U-shaped behavior' is a rather flat curve overall consisting of smaller and smaller curves. In the context of the symbolist tradition overgeneralizations have been interpreted as evidence for children's insights into rules and the obligatory marking of tense. More recently, however, they are explained as instances of "blocking-plus-retrieval-failure" of those forms stored in memory (cf. Marcus et al. 1992, Pinker 1995); ie. retrieval of the listed item fails so that the default rule is not blocked from applying. This hypothesis will be examined in this paper. In the current connectionist perspective (e.g. Plunkett & Marchman 1993) overgeneralizations appear shortly after or when generalization of patterns has started. Overgeneralizations as well as recovery from overgeneralization errors are due to internal reorganization processes, due to competing hypotheses with regard to similarities (or phonological subregularities) of stems and mapping strengths of a given stem and past tense pair (Plunkett & Marchman 1991). If, besides overgeneralizations, "irregularizations" (e.g. blends) occur they occur late in development. Competing hypotheses for the formation of past participles will also be of concern to this paper. 4

For such developmental curves in other domains cf. e.g. Bever (1982) and Strauss (1982).

Overgeneralisation Revisited: the Case of German Past Participles

155

Although most of the discussion about the two proposals on learning mechanisms have been carried out with English data there are two contributions in support of the symbolist position with German data. Clahsen & Rothweiler (1993) and Weyerts & Clahsen (1994) investigated children's use of past participles because the past tense or, rather, the preterite form is disappearing in spoken German. Despite lower frequency of regular verbs in German compared to English - following frequency counts by Ruoff (1981) and type-token frequencies of children's caretakers -, they found that children treat the suffix of regular verbs as the default case. Furthermore, "irregularization" (here the substitution of a stem vowel in regular verbs or substitution of the default suffix by the suffix for irregular verbs) is said to be practically non-existent and analogical formations are not found (cf. Clahsen & Rothweiler 1993:31; also Xu & Pinker 1995 for English data). This paper will first discuss the hypothesis by Marcus et al. (1992:1, 6ff.) that overgeneralization is due to blocking and retrieval failure and a finding reported by various authors (cf. MacWhinney 1978, Bybee & Slobin 1982, Marcus et al. 1992:117f.) that overgeneralization occurs more often with low-frequency verbs than with high frequency verbs. Then it will present observations which have to be accounted for in a comprehensive theory of learning inflection. In particular it will consider an issue which has been assumed not to exist: variety among overgeneralized forms in the acquisition of past participles. It will show that contrary to the findings of Clahsen and his colleagues children try out a number of ways to arrive at the appropriate form. As mentioned before the discussion about learning mechanisms has been based on the English past tense. It is well known that the preterite forms in spoken German are disappearing and past tense is being expressed by the 'Perfekt'.5 In her study of longitudinal data from seven children Behrens (1993) found that early use of preterite forms of main verbs is very rare around age 2; 1 to 2; 4. Past participles and the 'Perfekt' on the other hand are used from early on. This paper will focus on the acquisition of past participles. A short description of their formation, a prediction for children's learning as well a definition of 'overgeneralization' will help to understand and structure the task at hand.

2. The formation of German past participles and overgeneralization With regard to inflectional properties verbs in German may be grouped into weak, strong, mixed and irregular verbs. This verb categorization provides a more systematic approach than the categorization used most often in the English literature distinguishing only between regular and irregular verbs (for more details cf. Lindner 1995): the regular verbs correspond to the 5

I use the term 'Perfekt' since the German 'Perfekt' is not semantically equivalent to the English present perfect.

156

Katrin Lindner

weak verbs, while the irregular verbs comprise the other three classes. The formation of past participles involves affixation ((-)ge-, -t, -eri) and vowel change (e.g. Ablaut), ge- may be used as a prefix or infix or does not occur; its distribution is syntactically and prosodically determined.6 The four verb classes may be distinguished along the following lines (examples provide the infinitive, the 1. person singular in the present tense, in the past tense and in the 'Perfekt'): Weak verbs rely on affixation: they take the pre-/ infix ge- and the suffix -/. e.g. suchen ('to look for'), ich such-e - ich such-t-e - ich habe ge-such-t. Strong verbs take the pre-/ infix ge-, the suffix -en and Ablaut. e.g. fallen ('to fall') , ich fall-e - ich fiel - ich bin ge-fall-en. fliegen ('to fly'), ich flieg-e - ich flog - ich bin ge-flog-en. trinken ('to drink'), ich trink-e - ich trank - ich habe ge-trunk-en. Mixed verbs - as the name indicates - show a mixture: they take the pre-/ infix ge-, the suffix -t and vowel change. e.g. bringen ('to bring'), ich bring-e - ich brach-t-e - ich habe ge-brach-t. rennen ('to run'), ich renn-e - ich rann-t-e - ich bin ge-rann-t. Irregular verbs are irregular to a larger degree like the verb sein (' to be') or to a lesser degree like wissen ('to know') or haben ('to have'), cf. sein ('to be'), ich bin - ich war - ich bin ge-wes-en. wissen ('to know'), ich weiß - ich wuß-t-e - ich habe ge-wuß-t. haben ('to have'), ich hab-e - ich hat-t-e7 - ich habe ge-hab-t. Ablaut in strong verbs may be captured with the following three patterns (cf. also Clahsen & Rothweiler 1993). In the first pattern, ABA, the vowels of the present tense and the past participle stem are the same, as in fallen ('to fall'): ich fall-e vs. ich fiel vs. ich bin ge-fall-en. In the second pattern, ABB, the vowels of the preterite and the past participle stem are the same, as in fliegen ('to fly'): ich flieg- e vs. ich flog - ich bin ge-flog-en or steigen ('to climb'): ich steig-e vs. ich stieg - ich bin ge-stieg-en. In the third pattern, ABC, there is a different vowel in each verb stem, as in trinken ('to drink'): ich trink-e vs. ich trank vs. ich habe ge-trunk-en.

6

7

ge- occurs if the accent is on the first syllable of the finite base form; this condition holds also for complex verbs with a separable first constituent. Ge- is used as an infix with such separable verbs, as a prefix in all other cases. Cf. ich 'fahre - ich bin ge-'fahren (Ί drive' - Ί have driven') ich 'fahre ... durch - ich bin 'durchge- fahren. (Ί drive through...' - Ί have through-driven') in contrast to ich ent'laufe - ich bin entlaufen (Ί run away' -' I have run away7. Note the assimiliation of the final consonant of the verb stem to the dental leading to an interlude.

Overgeneralisation Revisited: the Case of German Past Participles

157

Since in spoken German children barely hear the preterite forms in all three patterns the second form - here labelled with the letter Β - can be put in brackets leading to A(B)A, A(B)B and A(B)C. For one of the issues to be discussed here the second and third pattern, A(B)B and A(B)C, will be summarized under A(B)X. This leads to the distinction of two basic patterns for the vowel change: A(B)A and A(B)X in strong verbs. The differentiations among the verb classes and the patterns of vowel change just considered along with type and token frequencies allow for predictions for children's learning of verb inflection (for details cf. Lindner 1995). One prediction to be taken up here concerns the patterns A(B)A vs. A(B)X among strong verbs. Given that with A(B)A verbs the verb stems for the present tense and the past participle do not differ there is no indication that these verbs should be treated as strong verbs and, consequently, that their past participles should be supplied with the suffix -en. Thus it can be predicted that children with less linguistic experience will overgeneralize the dental suffix in the beginning more often to past participles of these verbs than to those of A(B)X verbs. To summarize, children learning to produce German past participles have to learn the distribution of (a) two kinds of suffixes -t and -en, (b) ge- as a pre- or infix, (c) patterns of vowel change and different verb stems. Thus children have to master a complex task. Note also that both suffixes in (a) are homonymous; in particular the suffix -en is homonymous with the suffix for the infinitive in German. The task becomes even more complex when, in order to produce the 'Perfekt', on top of (a) - (c), children also have to attend to (d), the distribution of the auxiliary haben ('to have') or sein ('to be') both classified here as irregular verbs. As for the term 'overgeneralization', it is usually defined as "the application of the regular pattern to an irregular stem" (Marcus et al. 1992:1). Thus overgeneralization in this paper concerns the substitution of the suffix -t for the suffix -en irrespective of the verb stem chosen. Since the vowel in the verb stems may vary according to particular patterns (A(B)A, A(B)X) the dental suffix may be added to the present tense verb stem (a), to the preterite verb stem (b) or to the past participle stem (c). Cf. Table 1. Table 1 : Types of Overgeneralized Forms Patterns

Example

A(B)A (a) A(B)X (a) (b)

ge-fallt fallen ('to fall') steigen ('to climb') ge-steig-t, ge-stieg-t fliegen ('to fly')

(a)(b)(c)

trinken ('to drink')

Overgeneralized Form

ge-flieg-t, ge-flog-t ge-trink-t, ge-trank-t, ge-trunk-t

Target Form

ge-fall-en ge-stieg-en geflogen ge-trunk-en

However, it will shortly become clear that this definition does not capture all of the overgeneralized forms children use. Therefore I will distinguish two kinds of overgeneralizations: over-

158

Katrin Lindner

generalizations in a more narrow - traditional - sense will be called those forms with the dental suffix -t added to the verb stem (illustrated in Table 1) vs. overgeneralizations in a wider sense when some pattern is applied to a verb form. Examples for the second kind could be the dental suffix added to some verb form (cf. fallent in Table 4), a blend or an analogical formation. Substitutions of the dental suffix in weak verbs by the suffix -en of strong verbs will not be considered (for details cf. Lindner 1995).

3. Method The findings reported in this paper are based in part on diary data from a German-learning girl, A. (cf. Elsen 1991), collected continuously from the child's first word at age 8 months to 2 years 5 months. The method is described in detail in Elsen (1991). This data is phonetically transcribed.8 It is supplemented by some unpublished data (Elsen personal communication) presented here in transliterated form. The other data reported on - also transliterated - is crosssectional. It has been collected from 60 normal children age 2; 2 to 6; 11 in a research project in München.9 The data comes from two sources. The conversational sample collected for each child (with 600 utterances for the two and three year olds and 120 utterances for the 4 to 7 year olds) and from an experiment eliciting verb forms (present tense, preterite and past participles). Since children's productions of past participles may sometimes be ambiguous or lack some of the characteristics (e.g. ge-) the categorization by Elsen was followed with regard to the longitudinal data. For the cross-sectional data the following two criteria were used: a verb form is accepted as a past participle if the form is associated with some variant of an auxiliary and/ or if the verb form itself is modified in some systematic way (e.g. the child uses one of the verb stems with a suffix but omits the prefix - as in funden or fandet instead of gefunden ('found') - or s/he uses forms as illustrated in Table 1). All children in the cross-sectional data produced at least four past participles in spontaneous speech, each past participle from a different verb type.

g In citing this data the transcriptions by Elsen (1991) have been followed as closely as necessary. Markers for g interludes have been omitted. The research project "Funktionale Determinanten im Spracherwerb" has been supported by grants from the Deutsche Forschungsgemeinschaft to Prof. Dr. Rolf Castell (Ca 50/6-1 from 1990 - 1992) and to Prof. Dr. Elmar Seebold (Se 249/4-2 from 1992-1994). Its major aim has been to compare aspects of linguistic behavior of children with and without language impairment. Thus the data base also comprises data from 23 children with specific language impairment age 4; 3 to 6; 10. The data has been collected by Christina Schelletter (who participated in the project from 1990-1992), by Kerstin Täubner, Sabine Stoll and myself.

Overgeneralisation Revisited: the Case of German Past Participles

159

4. Frequency and 'blocking-and-retrieval-failure' In their monograph on children's overgeneralizations Marcus et al. (1992:6ff) set up the following hypothesis: Overgeneralization is due to blocking and retrieval failure. This hypothesis combines two principles or mechanisms: the "blocking principle" and the "principle of psychology of memory" (Pinker 1995). According to the first principle "an idiosyncratic form listed in the mental dictionary as corresponding to a particular grammatical modification of a word (past tense, in this case) blocks the application of a general rule that would effect the same grammatical modification" (Pinker 1995:112).

The second principle holds that the probability of successful retrieval of an item increases with repeated encounters. Thus "the cure for overregularization is living longer, hearing the irregulars more often and consolidating them in memory improving retrievability." (Marcus et al. 1992:18). 'Blocking' is a notion in accordance with claims and hypotheses in various linguistic theories (cf. Rainer 1988). Thus it may be considered relevant to explain the behavior of competent speakers. Pinker (1984) suggested to apply the term also to children's behavior (cf. the Unique Entry Principle p. 171, 181ff ). The blocking principle is conceptualized as part of Universal Grammar the child is said to be endowed with (also Pinker 1995). Marcus et al. (1992) as well as Pinker (1995) emphasize, however, that children's memory is more diffuse so that retrieval of an idiosyncractic form is more difficult for them than for adults. One finding in the literature which seems to create no difficulties for this hypothesis is that Overgeneralizations occur more often with low frequency than with high frequency verbs. I will start to discuss this finding before turning to the main hypothesis. The reasoning behind this finding is simple: with an irregular verb of high frequency children do not have difficulties accessing the proper entry so that the default rule is blocked. Thus overgeneralizations are expected to occur with irregular verbs of low frequency. Crucial for this issue, however, are overgeneralizations with verbs which are not of low frequency. In the longitudinal data of A. (Elsen 1991; this volume) the first overgeneralized past participles are those given in Table 2.

Katrin Lindner

160 Table 2: A.'s First Overgeneralizations Age

Overgeneralization

Target Form

Gloss

1; 5.30 1 ; 6.2 end of 1; 6 1;7.24

[dafált] [lönt] [mitenëmt] [vest]

gefallen verloren mitgenommen gewesen

'failed' 'losted' 'taked along' 'beed'

Since no token frequency count for verbs in A.'s data nor a type-token count for her input or for the output of the young children in the cross-linguistic study is currently available101 have to rely on observations of young children and their caretakers. In their interactions the verbs fallen ('to fall'), nehmen ('to take'), sein ('to be') and their various verb forms are, no doubt, used very often, likewise anziehen ('to dress') or ausziehen ('to take off) which will be focussed upon in later examples of this paper. The fact that a frequent verb like fallen ('to fall') is overgeneralized by A. is interesting. Recall that fallen is an A(B)A verb. With these verbs it was predicted that children with less linguistic experience will overgeneralize the dental suffix in the beginning more often than with those from A(B)X verbs. In the cross-sectional data overgeneralizations with A(B)A verbs are more frequent at first, but decrease rapidly with age and MLS (mean length of sentences in words),11 as shown in Table 3. Compare the percentages of suffix substitutions (S) of -en by the dental suffix -t in the second and third column with those in the fourth column as well as with the percentages in the fourth and fifth columns labelled VS (an abbreviation for the use of a verb stem other than the past participle stem along with the dental suffix, as in ge-beiss-t 'bited' instead of ge-biss-en 'bitten'). In the first MLS group only one of the two children (n = 1/2, age 2; 6 and 2; 10) produces overgeneralizations. This child uses all three of his A(B)A verbs (fallen, fangen, fahren) with the -t suffix but no prefix (er is fallt/fahrt - 'he has failed/ drived', er hat fangt - 'he has catched'). With regard to A(B)X verbs he overgeneralizes one form (aufgetrinket 'up-drinked') but not the other (reingangen - 'into-gone'). The data in this group is rather scarce. However, in the second MLS group (age 2; 2 to 3; 6) with all six children (n = 6/6) producing overgeneralizations six out of 32 instances with the prefix and 8 out of 9 verb forms without the prefix still have the dental suffix. From MLS group 3 onwards the percentage of suffix substitutions with A(B)A verbs decreases rapidly to 5% and then to 2%. 10

11

For a cumulative type count for A. cf. Elsen (this volume). There are also a number of recordings (Elsen 1991:16) which, so far, have not been analyzed. Type-token frequency counts for all the verbs in the crosssectional data is currently under way. The conventional measure for children's productions, in particular to account for syntactic complexity, is the mean length of utterances (MLU in words or morphemes). In the project in München the mean length of sentences (MLS) is examined in terms of being a better predictor than MLU for children's productions, in particular for those of children with specific language impairment who produce very long utterances with lots of grammatical errors.

Overgeneralisation Revisited: the Case of German Past

Participles

161

Now compare the percentages for A(B)X verbs. Suffix substitutions only are fairly rare. If children overgeneralize past participles of this verb group they tend to substitute the verb stem and the suffix (VS). Thus, besides frequency, the phonological structure of the various verb stems in the paradigm - here homophony of verb stems - may be an important factor for this issue as well. This implies that overgeneralizations may be due not only to low frequency but to different - sometimes competing - factors influencing production. However, before going into this issue in more detail let us return to the main hypothesis of 'blocking and retrieval failure' responsible for overgeneralizations. 'Blocking' has a history in linguistics going back to Hermann Paul in 1896. He mentions it in the context of word formation processes when the formation of a morphologically complex word is blocked by a synonymous word. Rainer (1988) shows that this concept has undergone some modification in linguistic theory. What remains is that the blocking words or tokens are stored. Storage clearly is assumed for irregular verb forms. But what about regular or weak verb forms? There is an ongoing discussion in the psycholinguistic literature about the storage of regularly inflected items as a whole with competent language users (e.g. Stemberger & MacWhinney 1988). Baayen et al. (1996) show that storage of complex words and morphological rules are compatible with one another. In an on-line comprehension task about plural and singular forms in Italian they found that certain plural forms have shorter response latencies than others, namely those where the plural form expresses the basic concept, has higher frequency in usage (e.g. denti 'teeth' vs. nasi 'noses') etc. These nouns are stored as wholes in the lexicon in contrast to other plural forms which are not. On the other hand response latencies for singulars (e.g. naso 'nose' or dente 'tooth') remain the same. Since their endings are semantically vacuous they are considered less likely to be stored as full forms. Given this complex situation for mature language users how is storage to be conceptualizedfor language users who just start learning their language? Are the forms stored in parts or as holes, are the forms stored with their meaning or separately? Marcus et al. assume that irregular items are stored as wholes; presumably children's early use of correct forms are taken as evidence. Why should not regular/ weak forms of high frequency be stored as wholes? There is one recent suggestion for the build up of children's lexical entries. Kim et al. (1991; 1994) have found in experiments that children aged 3; 2 to 5; 2 (mean 4; 3) differentiate between denominal and root verbs in using the weak inflection more often with the derived verb. This finding is particularly interesting since the stimuli are words which due to their phonological but not to their semantic properties might have also been inflected as strong verbs However, Kim et al. suggest that neither phonological nor semantic information is enough to ex-

Katrin Lindner

162

Table 3 : Percentages for children's substitutions o f vowels and suffixes with past participles in t w o subgroups o f strong verbs A ( B ) A and A(B)X.

MLS groups

A(B)X

A(B)A +prefix

MLS 1,8 - 2,49

-prefix

+prefix

S 100% (3/3)

V S 100% (1/1)

-prefix

n = 1/2 MLS 2,5 - 2,99

S 19% (6/32)

S 88% (8/9)

S 7% (3/46) V S 28% (13/46)

η = 6/6 M L S 3,0 - 3,49

S 2% (2/97)

S 80% (4/5)

S 5% (2/38)

S 100% (1/1)

V S 18% (17/94)

η = 10/14 MLS 3,5 - 3 , 9 9

S 1% (1/94) V S 34% (4/12)

S 2% (1/49) V S 14% (7/49)

η = 4/11 I 3% (1/38)

1 2 % (1/49)

MLS 4,0 - 4 , 4 9

S 1% (1/79)

η = 6/12

V S 8% (6/79) I 1% (1/79)

Legend: MLS groups = children grouped accordig to mean length o f sentences. η = no o f children with overgeneralizations by total no o f children in that MLS group A ( B ) A = subgroup o f strong verbs with the same stem v o w e l in the present tense and in the past participle stem; A ( B ) X = subgroup o f strong verbs with a different stem v o w e l in the present tense and in the past participle

plain children's behavior; rather, children have to have the relevant structural information right from the beginning. 12 However, even by 3; 2 - the youngest participant in the experiment -

12

"Children are born prepared to distinguish regular inflection from irregular inflection..." (Kim et al. 1991:45). Their results for the preschool children, however, are not quite convincing. First, the percentages provided for children's performance still leaves room for questions: the authors report that regular inflection was chosen for the denominal verbs in 64.1%, irregular inflection in 5.6%, no change was observed in 20%, and uncodeable were 9.8%. Thus the percentages of 'no change' and 'uncodeable' still account for about a third of the results. Moreover, the corresponding strong verbs were marked with the regular inflection in 46,6% of the cases. Second, the design of the experiment was such that children got enough information to treat these denominal verbs as different from the verbs they knew. Thirdly, although the group of children were preschoolers mean age 4; 3 it is not quite clear how many young three year olds were in this group. The children in the cross-sectional sample, for instance, rarely missed the target form once they were 4 years old. Nevertheless, the phenomena need further attention why despite phonological similarity to strong verbs such verbs are inflected following the pattern of fewer constraints.

Overgeneralisation Revisited: the Case of German Past Participles

163

these children have more experience with language and thus with verb usage than the children and, in particular, the child A. whose behavior is described here. Therefore, still, evidence is needed for storage of items and thus the form of lexical entries with very young children. The other big problem is how does retrieval work in young children if - as claimed - it fails in case of overgeneralized forms. Retrieval, as observed and analyzed with competent participants, does not seem to be a simple straightforward matter either (cf. e.g. Dell 1988; Levelt 1989; Bock & Levelt 1994). Here is an illustration from the free speech sample of the crosssectional data again involving a verb which, no doubt, belongs to the earlier and most frequent verbs for young children, namely anziehen 'to dress'.13 (1)

Child C (age 2; 6) and experimenter E are talking about a witch and two children. The witch is wearing a dress and a hat while the children are in bathing suits playing with water. C weil die [the witch] ä angezogen. 'because that one is dressed.' About 160 turns later C and E are talking about wearing T-shirts: C has du den weiss angezo* weiss angezogt? 'have you dressed that one in white in white dressed?' E mhm. E den hab ich weiss angezogen ja. 'that one I have dressed in white yes.' C und du? 'and you?' E ich hab mich auch weiss angezogen. Ί have dressed myself in white as well.' E und du? 'and you?' C dun weisses T-shirt an. 'you a white T-shirt on.' C XXX auch ein weisses T-shirt. 'xxx also a white T-shirt.' E du hast wieder das gelbe T-shirt an ne? 'you are wearing again the yellow T-shirt right?' E is des dein Schlafanzug? 'are these your pyjamas?' E hm hast du dich noch gar nicht angezogen heute? 'hm have'nt you dressed yet?' C shakes his head. E nein? 'no?' E warum das denn nicht? 'why not?' E warum hast du dich noch nicht angezogen? 'why have'nt you dressed yet?' C weil ich mich nich angezog hab. 'because I haven't dress.'

13

(1)

(2)

(3)

(4)

Anziehen is also produced very early by A. Her first imitated forms are [anda] [ândjï] [andi] at 1; 3.0; her first spontaneous form is [antci] at 1; 3.5 (Elsen 1991:143).

164

Katrin Lindner

At first the child produces the correct form angezogen. With the second occurrence the child stops in the middle of the verb stem (angezo*) and corrects himself in producing an overgeneralized form angezogt. If the child's overgeneralization were due to a retrieval problem and if that retrieval process were straightforward then one might suppose that the correct input given to a child a number of times would have an immediate effect on his output. Although E provides the proper form angezogen four times the child does not produce it in his last utterance, rather, he omits the suffix altogether. There may be a number of factors interfering in the child's production process; these may also include that, for some reason, the child evades the verb form altogether (cf. his utterance: du η weisses T-shirt an.). - In particular the research within the generative grammar framework has emphasized that children often are not affected by the immediate input. Examples for children's "insensitivity to input" abound in the literature (e.g. Brown & Hanlon 1970; Marcus et al. 1992; Pinker 1995). The example just presented may be interpreted in favor of their position. But it is not clear how the concept (and process) of retrieval failure is integrated into this position. More has to be known about storage and retrieval before the hypothesis by Marcus et al. can be tested. One way to attack the problem is to investigate children's comprehension and production and their relationship to one another. Thus it will also have to be decided upon whether one lexicon or two - one for production, one for comprehension - have to be assumed. There has been some discussion regarding a two-lexicon model in relation to phonological data from children who missed the target form but claimed to have said it (e.g. Menn & Matthei 1992). There is also some debate about a two lexicon-model regarding morphology in competent language user (cf. Monsell 1987). Marcus et al. are not concerned with this issue. Yet anecdotes show that children are aware of their productions and are able to distinguish between overgeneralized forms they use and forms other people use (e.g. Pinker 1995). Given this state of affairs it may be useful to look at children's overgeneralizations in more detail.

5. Various ways to form a past participle A number of times it has been mentioned in this paper that there is more variety in formations and that there are different factors influencing children's productions. It was demonstrated that 'high frequency' which should protect strong verb forms from overgeneralization can be overriden by 'homophony' or, more precisely, by low perceivability of characteristics of strong past participles due to homophonic verb stems (as with A(B)A verbs). It was also shown that the influence of this factor diminishes with increasing linguistic proficiency. Such factors tell us something about children's learning process.

Overgeneralisation Revisited: the Case of German Past Participles

165

Let us return to the German longitudinal data. Before looking at overgeneralizations from A. in more detail I would like to provide some background for the development of her past participles forms. A.'s first past participles from strong verbs are with finden ('to find1, target form gefunden; cf. Elsen 1991:189):" (2)

1; 5.24 1; 6.25

[zoka foñ]'sock found', later [dekl dafifl]'lid found', [fuman] possibly imiated

1; 6.27 1; 7.2 end of 1 ; 7 to 1 ; 8 1 ; 10.13

[lisal gaftuna] 'key found', repeatedly [fiiinñ] [fionda] as well as occasionally [ftunan] [finnan] daily ['aurto fundan] 'car found'.

The other verb is fallen (target form gefallen, for more details cf. Table 4): (3)

1; 5.26 and 1; 5.27 1; 5.28

[va'n] [valñ]

1; 5.30

[vain], later [dafalt] repeatedly that day and daily up to 1; 6.2.

Up to age 1; 6 A. has 15 weak, 14 strong verbs and one irregular verb {haben 'to have', cf. Elsen this volume). By the end of 1; 5 she acquires seven more verbs, three of them are weak verbs; by the beginning of 1; 6 she has 18 more verbs, 11 of them weak verbs. By 1; 6.10 her vocabulary contains 49 verbs, 43% strong and irregular verbs and 57% weak verbs. For Α., within a few days the following has happened: she produces the first strong past participles (at 1; 5.24, cf. (2) and (3)), then the first weak past participle along with the first overgeneralization (from 1; 5.30 onwards) while by that time the number of weak verbs exceeds 50% of her verb vocabulary. Thus A.'s development is compatible with the development reported in the connectionist literature. Clearly the development of A.'s verb lexicon is not a linear one but shows verb spurts (cf. Elsen this volume). These verb spurts seem to correlate with particular phases in the development of her past participles. At about 1 ; 6 A. starts to use not only the dental suffix but she occasionally uses the pre-/infix (e.g. [puitomat papa] 'Pappa has broken' by 1; 6.2). By the next spurt, at 1 ; 9, the child produces ge- more often and begins to use the auxiliary (e.g. Wurscht gegessen habe 'sausage eaten have'). Thus the child gradually assembles the necessary "ingredients" for her 'Perfekt' forms, starting out with the (dental) suffix, then attending to the pre-/ infix and then to the auxiliary. The fact that she begins with the suffix is not amazing when the 14

From 0; 11 - 1; 5.27 A. acquires the following strong verbs in this order: essen, trinken, anziehen, stehen, singen, ausziehen, schlafen, sitzen, gehen, sehen, finden, helfen , fallen, geben ('to eat1, 'to drink', 'to dress', 'to stand', 'to sing', 'to take off, 'to sleep', 'to sit1, 'to go' ,'to see', 'to find', 'to help1, 'to fall', 'to give'). By age 1; 5 - 1; 6.2 she uses past participles from three strong verbs: fallen (cf. Table A,),finden(cf. (2)) and verlieren (cf. Table 2).

Katrin Lindner

166

position of past participles in linear order is taken into account. Past participles form part of the sentence bracketing in German and are thus often in last position; therefore, due to recency, they may be recalled more easily.15 Note that with regard to Aktionsarten A. starts out with telic verbs like finden ('to find') and fallen ('to fall') and telic constructions with machen (cf. Behrens 1993; Lindner in prep.).16 Unique references to past events are observed around 1; 9 (cf. [filton vas das habazen] 'that was a sled (I) have seen'). In examining A.'s past particles of particular verbs it is easy to see the variety of forms she produces. As a first example take the past participles from the verb fallen and related verbs (Elsen 1991:186, 295) in Table 4. Table 4: Past Participle Forms from fallen, runterfallen and Related Verb Forms Other Forms of the Paradigm

Age

Past Participle Form

1; 5.26 1; 5.27 1; 5.28 1; 5.30 up to 1; 6.2 1; 6.3 1; 6.5 - 1; 6.8

[van] [va'n] [valñ ] repeatedly that day [vain] then repeatedly [dafalt] [dafált] daily [fait] [faint] [vald] [falñt] often also [vált] [wait] [faint]

1; 6.23

[vain] often [vaint]

1 ; 8 and 1;9 1; 9.11 1; 10.6 1; 10.7 1; 9 to 2; 1 1; 11 to 2; 0 2; 1.19

[Kuinto-fetan] [valan] also [ga-valan] [KUintafabn] [KUintofalan] [febn] [Kunto-falt] often waserunterfällt 'something failed down' 2; 3 to 2; 4 [feten] (infinitive)

1; 6.2

15

16

1 ; 6.14 [dui velst] imitated 1; 6.28 [felt munto-] (3.sg.pres.) 1; 7.9 [valt] (3.sg.pres.)

Moreover, experimental evidence is needed to decide whether the dental suffix may be perceived more easily than the suffix -en since the latter is quite often assimilated in rapid speech. Other reasons for the early use of -t may be morphological, e.g. the ending -t for the 3. person singular present tense learned about the same time. - With regard to the use of ge-/-ge- the rhythm of the utterance may play a role; children seem to produce it more often if it fits the trochaic rhythm of the language. A.'s first weak past participles are with machen [gamat] gemacht 'made' at 1; 5.30; [puitemat papa] at 1; 6.2 Pappa kaputtgemacht. By 1; 6.2 she also uses kleckern 'to spill' as in [gekat mama] 'Mama spilled' or kleben 'to stick to' at 1; 6.16 [gsget] 'stuck'.

Overgeneralisation Revisited: the Case of German Past Participles

167

Her first productions are fairly close to the target form (except for the missing prefix). However, on the fifth day she produces her first overgeneralizations. These forms continue for the next two days. In the following week she varies in using overgeneralized [fait], some variant of it and [faint], the latter winning out by age 1; 6.5 - 1; 6.8. A competition between forms with or without final -t is observed for the next two weeks. In her first overgeneralizations A. adds the dental suffix to the stem producing overgeneralizations in the narrow sense. With [faint] she produces an overgeneralization in the wider sense adding the -t to a constituent which could either be classified as a past participle without prefix or as the infinitive. A decision between the two alternatives is rather difficult to make: on the one hand, at 1; 6.2 (cf. Table 2), the beginning of this period, A. uses [lont] 'losted' which looks like the past participle with the dental suffix; three months later, at 1; 9.26, she uses [vegatänt] ('hurted', target form: wehgetari) having over-applied the -en suffix to the past participle for the last two months (cf. footnote 20). However, such formations - besides those with fallen - are extremely rare. On the other hand, from 1 ; 8 onwards present tense forms become more influential (see below) and by 2; 0 to 2; 2 she adds the -t to forms which look like infinitives (cf. Table 5). By 1; 8 A. no longer experiments with the -t in case offallen but uses verb forms which are influenced by present tense forms from the same paradigm: she first imitates du fällst 'you fall' but then produces spontaneously er fällt runler 'he falls down' or er fallt 'he falls'. A few weeks later the form [Kuinta-felan] of the related verb appears. By 1 ; 9 A. also produces the past participle offallen both without and with the prefix, the latter with an intermediate pause between prefix and verbstem. For the next two months the blend [felan] is the third form to be used; it is homonymous to the infinitive observed later at 2; 3 to 2; 4. By the end of the observation period at 2; 5 she still does not seem to use the appropriate form gefallen consistently.17 The items in Table 4 show that A.'s past participles include overgeneralizations in the narrow (or traditional) sense and those in the wider sense. The latter comprise forms with a final -t as well as blends. The data also show that different forms compete at the same day (as at 1; 6.2 and 1; 6.3), for a few weeks or even one or two months. Nevertheless, the variety of forms indicates some systematicity. A.'s first productions show the dental added to the verb stem (following the pattern for past participles of weak verbs), then to the form [fain] in [faint]. Later her productions are affected by the singular forms of the present tense of the paradigm she is acquiring at the same time.18 Thus the development of her past participles forms cannot be separated from the development of other forms in the paradigm; the influence of competing forms or patterns within the same paradigm is another factor to be taken into account in children's learning process. 17

18

In case of runterfallen ('to fall down1) Α., like many other children (cf. Lindner 1995), often does not use the prefix; cf. at 1; 10 [Kumtafalsn], at 1; 11 to 2; 0 [ Kumta-falt] (Elsen 1991:295). It is feasable that the second syllable is taken to be the prefix. Later there are also forms where the infix is pre-fixed; cf. waserunterfällt in Table 4. The relationship between past participles and present tense forms is also studied by Schindler (1997) with data from four children age 1; 8 to 3; 6.

Katrin Lindner

168

The examples given in Table 4 do not demonstrate the only way A. is "working" on her past participle forms: a different way is manifested with the past participles of ziehen - ausziehen/anziehen/abziehen ('to pull' - 'to take off, 'to put on/dress', 'to pull ofïïremove') with the correct form gezogen [gatsögan] / ausgezogen [ausgatsôgan] / angezogen [arjgatsögan] and abgezogen [apgatsögan]. As mentioned above ziehen and its related verbs seem to be rather frequent in interactions of young children with their caretakers. Cf. Table 5 (from Elsen 1991:138, 143, 149): Table 5: Past Participle Forms with ziehen-anziehen-ausziehen-abziehen Major Age Characteristics

Past Participle Form

1; 4.7

['aöogg] imitated

1; 8.27

['auisdunpn]

1; 10.15

['auissduigan] ['auiOaduigan]

1; 1 0 . 2 0

['auisatslna]

Related Forms

PHASE I Analogy

Infinitive + schwa

11.21 11.24 11.25 11.27 Analogy vowel change 2; 0.15 Analogy

[finnan] [duigkan] [zunpn]

['auisazuigan] ['austsugan] later ['ausatsögan] ['ausaöögan] ['auiOatsuigan] f'anaGoqt] corrected to ['anaBogn]

+-t

Analogy

2; 1.4

['apazurjkt]

Analogy

2; 1.14

['ausatsuqan]

PHASE II "Infinitive" + -t

"Infinitive"

reinetunt weggetunt wehgetunt 2; 2.14

['ausatsmt]*

2; 4.14

will ich auch mal anziehn werden (Ί want also once to be dressed')

Present tense 2; 4.19 stem + t + en

hab richtig geziehten ('have really pulled')

['ausale a nt]

Overgeneralisation Revisited: the Case of German Past Participles

"Infinitive" 2; 4.21

169

nich anzieht! worden ('not been dressed')

* for a more narrow transcription cf. Elsen (1991: 149). A first approximation of the target form is an imitated form of ausgezogen at 1; 4.7. No other attempt is noted during the next four months. Considering the child's productions in the subsequent months, the data can be assigned to two phases: the first phase is dominated by analogies, the second by forms which resemble infinitives. A. first produces forms of the past participle modelled on past participle forms of strong verbs of the A(B)C type she has acquired or is using about the same time:19 e.g. [finnan] 'found' (from age 1; 6.25 - 1; 7.29 see (2) above) or [dmqkan] 'drunk' (imitated in ['am dafa duiqkan] 'coffee drunk' at 1; 8.16, spontaneously produced trunken, trunkt 'drunk', 'drunked' by 1; 9.22 and getrunken at 1; 9.27). The past participle o f ' t o sing' [zuirpn] appears at age 1; 10.5 (around Christmas 1988). Phase I holds for about three months. It is briefly interrupted by a form which looks like an infinitive with schwa at 1; 10.20. Twice A. comes up with a form close or closer to target: while at 1; 11.24 her second form barely differs from the grammatical form, her correction at 2; 0.15 leads to the correct suffix yet not to the correct stem vowel in ['anaGogr)]. Her first attempt ['anaBogkt] shows the same vowel but the dental suffix is added to the verb stem. The pattern with the dental suffix is also used in the next example ['apazuqkt], again an analogical formation now with abziehen. In the following months this suffix will become more important. It even looks as if the two phases are "connected" via the dental suffix. Phase II, at 2; 2, begins with forms with the dental as the final suffix. As with some of the formations in Table 4 it is added to the whole constituent. This pattern seems to be quite pervasive now when forms like reinetunt (2; 0.16 and 2; 1.12), weggetunt (repeatedly at 2; 1.23 and 2; 1.24), wehgetunt (by 2; l), 20 ('put into', 'put away', 'hurt') are taken into account; it is even applied to the weak verb ausleeren 'to empty', cf. ['aussieht] at age 2; 3.4. The following forms in Table 5 are either infinitives or show both suffixes -t and -en added to the verb stem as in geziehten. Of particular interest are the constructions involving the future tense and the passive will auch mal anziehn werden ('want also once be dressed' - target: will auch mal angezogen werden) as well as nich anziehn werden ('not be dressed' - target: nich angezogen werden)2X They point to another factor influencing children's learning process of past partici19 20

21

These productions are reminiscent of'gang effects' (cf. e.g. Stemberger & MacWhinney 1988:107). For wehtun and its past participle wehgetan ('to hurt', 'hurt') Elsen (1991: 349) notes that from 1; 7.3 onwards up to 1; 9.3 A. uses [vetaran] or variants of it like [vetatänan], [vedatönan], [vegatönan], At 1; 9.5 she is reported to say [vagatóñ]; one day later she uses the grammatical form. However, [vegatönan] reappears later in the month. By 1; 9.24 she uses [kofegatün] ('head hurt') and one day later [kof vëgatônt], the past participle with the dental suffix. These examples also demonstrate that A. not only over-applies -t but also -en. Moreover, these forms too may disappear and reappear again. In the cross-sectional data children by 2; 8 show similar difficulties with the 'Ersatzinfinitiv' (cf. Lindner 1995). It is also possible that such forms are produced due to capacity limitations.

170

Katrin Lindner

pies, the influence of other complex constructions. In the cross-sectional data the 2 year olds as well as the language-impaired preschool children are found to substitute the past participle with the infinitive of strong and weak verbs once the auxiliary is used. It is suggested that there is a competition with other syntactic patterns, here the modal verb construction requiring a finite verb and an infinitive; and the infinitive wins (cf. Lindner 1995). The data in Table 5 suggest that the child explores the phonological structures first before relations to other forms in the paradigm or to other constructions come into play. Up to 2; 5 no correct past participle form with gezogen or related verbs is reported. Looking at the development of other strong past participles they appear to be fairly different. For instance, verbs with a diphthong in the infinitive and the present tense stem are predominantly overgeneralized with the present tense stem and the suffix -t (as in eingereiht, abgebeißt, reinsteigt hat, hineschmeißt/ hingeschmeißt, hatebeißt - 'rubbed-in', 'bited-off, 'intoclimbed has', 'down-throwed', 'has bited'). Yet only a few instances are observed before age 2; 5 (cf. Elsen this volume Appendix II); most of them appear later than the period under study in this paper. Overgeneralizations with the present tense stem and the dental suffix are also quite common with these verbs in the cross-sectional data (cf. Lindner 1995, cf. also Stemberger 1993 for English-learning children). There are also other verbs whose past participles show less variation in the course of development, cf. finden - gefunden ('to find, found') or trinken ('to drink') with its past participle form getrunken. By 1; 6.17 A. produces [dunjka]; by 1; 8.16 she imitates ['QUI dafa dunjkan] ('coffee drunken too'). By 1; 9.23 she uses the correct form next to the over-generalized form [tKUiqkt], At 2; 0 and 2; 1 she is reported to use repeatedly [gatKirjkt] (cf. Elsen 1991:333); around this time the influence of the present tense forms appear to be fairly strong (cf. the blends in Table 4 and the forms with tun in Table 5). By 2; 5 she uses the correct past participle, often in conjunction with another word as in [milçatKuqkan] ('milk drunken'). Reduced variation with these verbs can also be observed in the data from the cross-sectional study. Children tend to overgeneralize less past participle forms in the class A(B)C like trinken trank - getrunken than with A(B)B as in ausziehn - ausgezogen. Recall that it was this group of A(B)C verbs which provided the model for A.'s analogical formations in Table 5. Thus, no doubt, there is systematicity among the various kinds of overgeneralized forms. The data manifest different attempts to build up the appropriate form.22

6. Summary and conclusion In this paper the main hypothesis about children's overgeneralizations by Marcus et al. (1992) was discussed. It maintains that overgeneralizations are due to 'blocking-and-retrieval-failure'. 22

A more comprehensive study needs to take other overgeneralizations within one paradigm into account, e.g. children's regulaiizations of present tense forms of modal and other strong verbs. Cf. Lindner (1995).

Overgeneralisation Revisited: the Case of German Past Participles

171

Given the current knowledge about storage and retrieval in young children the hypothesis cannot be tested. Furthermore, one finding reported by various authors related to symbolists or connectionists was also examined that overgeneralization occurs more often with low frequency than with high frequency strong verbs. With regard to this finding it was shown that in case of a particular subgroup of strong verbs - those with homophonic forms for the present tense and the participle stem - even high frequency verbs are subject to overgeneralizations for children with little linguistic experience. In the remaining part of the paper children's overgeneralizations were explored in more detail. A first set of findings concerns the variety of forms, a variety which so far has not been noted in the literature. Overgeneralizations of two kinds were found: those in a wider sense (like the application of the dental suffix to forms resembling an infinitive as in [faint], analogical formations and blends) and those in a more narrow or traditional sense (where the dental suffix is applied to the verb stem be it the present tense, the past tense or the past participle stem). While both kinds were observed with a small number of verbs (mostly fallen and ziehen) the latter kind was dominant with past participles of A(B)B verbs with a diphthong in the present tense stem or verbs of the A(B)C group. Differences were also found with regard to the time span when overgeneralizations appear. While for fallen and ziehen the child did not produce the correct past participle up to age 2; 5 she did suceed with past participles of the A(B)C group. On the other hand there are past participles (like those with the diphthong in the present tense stem) which tend to be overgeneralized more often later than the period under study. The second set of findings concerns the evidence for factors influencing particular formations: - homophonic forms (with A(B)A verbs) - or low perceivability of characteristics of strong verbs - were observed to lead to suffix substitutions even with past participles of verbs of high frequency. - competing phonological structures of past participles of other strong verbs were models for analogical formations. - competing phonological - morphological structures within the same paradigm acquired at the same time lead to blends. - competing syntactic constructions, for instance modal verb constructions, influence the choice of formations once children produce auxiliaries. Thus, overall, there is systematicity in children's productions. One way to find out about this systematicity is to look at the development of participles of individual and related verbs and recurring processes demonstrated here with the verbs fallen and ziehen. Yet there are also processes recurring across paradigms, for instance the application of the dental suffix to infinitive-like forms or the preference for the present tense stem with the dental suffix. As for the relationship of overgeneralizations in the wider sense to those in the narrow/traditional sense, one type subsumed under the former seems to disappear with

172

Katrin Lindner

growing linguistic experience, since it has not been observed in the cross-sectional nor the longitudinal data at a later age: the application o f the dental suffix to the infinitive-like form. Overgen-eralizations o f the traditional kind have been found with some past participles right from the beginning. With regard to a longer time span (cf. Lindner 1995) they seem to be the dominant kind. This may manifest progress in the child's attempts to form past participles. Symbolists and connectionists both postulate that strong and irregular verbs are learned by association. Y e t there seems to be a difference between them in so far as symbolists have not found much variety among children's productions. For instance Clahsen and Rothweiler (1993:31) c o m e to the conclusion that children "stick very carefully to the particular form o f the entries." Connectionists, however, have pointed to a variety o f overgeneralization types. Selective over-application o f particular patterns to verbs as well as the kinds o f competition observed in this paper seem to be more in line with the assumption o f organizing or reorganizing factors in children's learning process proposed in the connectionist literature.

References Baayen, Harald, Christina Burani & Robert Schreuder (1996) Effects of semantic markedness in the processing of regular nominal singulars and plurals in Italian. In: Yearbook of Morphology. In press. Behrens, Heike (1993) Temporal reference in German child language. Form and function of early verb use. Doctoral Dissertation. Universiteit von Amsterdam. Berko, Jean (1958) The child's learning of English morphology. Word 14, 150-177. Bever, Thomas G. (1982) Regression in the service of development. In: Thomas G. Bever (ed.): Regressions in mental development. Hillsdale, NJ: Erlbaum. 153-188. Bock, Kathiyn & William Levelt (1994) Language production. In: Morton A. Gernsbacher (ed.): Handbook of Psycholinguistics. San Diego: Academic Press. 945-984. Bowerman, Melissa (1982a) Reorganizational processes in language development. In: Lila R. Gleitman & Eric Wanner (eds.): Language acquisition: the state of the art. London: Cambridge University press. 319-346. (1982b) Starting to talk worse: clues to language acqusition from children's late speech errors. In: Sidney Strauss with Ruth Stavy (eds.): U-shaped behavioral growth. New York: Academic Press. 101-145. Brown, Roger (1973) A first language. The early stages. Cambridge, MASS: Harvard. Brown, Roger & Camille Hanlon (1970) Derivational complexity and order of acquisition in child speech. In: John R. Hayes (ed.): Cognition and the development of language. New York: Wiley. 11-54. Bybee, Joan L. (1985) Morphology. A study of the relation between meaning and form. Philadelphia: Benjamins. - (1988) Morphology as lexical organization. In: Michael Hammond & Michael Noonan (eds.): Theoretical morphology. Approaches in modern linguistics. San Diego: Academic Press. 119-141. Bybee, Joan L. & Carol Lynn Moder (1983) Morphological classes as natural categories. Language 59, 251270. Bybee, Joan L. & Dan I. Slobin (1982) Rules and schemes in the development and use of the English past tense. Language 58, 265-289. Champaud, Christian (1996) Overgeneralisation in early acquisition of verbal morphology: the case of French. Paper presented at the Workshop 'The Acquisition of morphology in LI' at the 7th International Morphology Meeting in Wien. Clahsen, Harald & Monika Rothweiler (1993) Inflectional rules in children's grammars: evidence from German participles. Yearbook of Morphology 1992. 1-34.

Overgeneralisation Revisited: the Case of German Past Participles

173

Dell, Gary S. (1988) The retrieval of phonological forms in production: tests of predictions from a connectionist model. Journal of Memory and Language 27, 124-142. Elsen, Hilke (1991) Erstspracherwerb. Der Erwerb des deutschen Lautsystems. Wiesbaden: DUV. - (this volume) The acquisition of past participles: one or two mechanisms. Ervin, Susan (1964) Imitation and structural change in children's language. In: Eric H. Lenneberg (ed.): New directions in the study of language. Cambridge, MASS: MIT Press. 163-189. Kim, John J., Gary F. Marcus, Michelle Hollander & Steven Pinker (1991) Children's inflection is sensitivity to morphological structure. Papers and Reports in Child Language Development 30, 39-46. Kim, John J., Gary F. Marcus, Steven Pinker, Michelle Hollander & Marie Coppola (1994) Sensitivity of children's inflection to grammatical structure. Journal of Child Language 21, 173-209. Kucera, Henry & Nelson W. Francis (1967) Computational analysis of present-day American English. Providence, RI: Brown University Press. Levelt, William J.M. (1989) Speaking. From intention to articulation. Cambridge, MASS: MIT Press. Lindner, Katrin (1995) Regularization in verb inflection: a comparison of normal and language-impaired children learning German as their first language. MS München. - (in prep.) Normale und gestörte Sprachentwicklung bei Kindern. Studien zum Verstehen und zur Produktion. MacWhinney, Brian (1978) The acquisition of morphophonology. Monographs of the Society for Research in Child Development. 174.43.1 Maratsos, Michael (1979) How to get from words to sentences. In: Doris Aaronson & Robert W. Rieber (eds.): Perspectives in Psycholinguistics. Hillsdale, N.J.: Erlbaum. 285-353. Marchman, Virginia & Elizabeth Bates (1994) Continuity in lexical and morphological development: a test of the critical mass hypothesis. Journal of Child Language 21, 339-366. Marcus, Gary F., Steven Pinker, Michael Ullman, Michelle Hollander, T. John Rosen & Fei Xu (1992) Overregularization in language acquisition. Monographs of the Society for Research in Child Development. 228.57.4. Menn, Lise & Edward Matthei (1992) The "two-lexicon" account of child phonology: looking back, looking ahead. In: Charles A. Ferguson, Lise Menn & Carol Stoel-Gammon (eds.): Phonological Development. Models. Research, Implications. Timonium, Maryland: York Press. 211-247. Monsell, Stephen (1987). On the relation between lexical input and output pathways for speech. In: Alan Allport, Donald G. Mackay, Wolfgang Prinz & Eckart Scheerer (eds./. Language perception and production. London: Academic Press. 273-311. Paul, Hermann (1896) Über die Aufgaben der Wortbildungslehre. Sitzungsberichte der PhilosophischPhilologischen oder der Historischen Classe der k.b. Akademie der Wissenschaften zu München, 692-713. Repr. in: Leonhard Lipka & Hartmut Günther (eds.): Wortbildung. Darmstadt: Wissenschaftliche Buchgesellschaft. 17-35. Pinker, Steven (1984) Language learnability and language development. Cambridge, MASS: Harvard University Press. - (1991) Rules of language. Science 253. 530-535. - (1995) Why the child holded the baby rabbits: a case study in language acquisition. In: Lila R. Gleitman & Mark Liberman (eds.): Language. An invitation to cognitive science. Vol 1. Cambridge, MASS: MIT Press. 107-133. Pinker, Steven & Alan Prince (1988) On language and connectionism: analysis of a parallel-distributed processing model of language acquisition. Cognition 28, 73-193 - , - (1992) Regular and irregular morphology and the psychological status of rules of grammar. In: Proceedings of the 17th annual meeting of the Berkeley Linguistic Society. Berkeley, CA: Berkeley Linguistic Society. 230-251. Prasada, Sandeep & Steven Pinker (1993) Generalisation of regular and irregular morphological patterns. Language and cognitive processes 8, 1-56. Plunkett, Kim (1995) Connectionist Approaches to Language Acquisition. In: Paul Fletcher & Brian MacWhinney (eds.): The Handbook of Child Language. Oxford: Blackwell. 36-72.

174 -

Katrin Lindner

(in press) Connectionism and developmental psychology. Journal of Child Psychology and Child Psychiatry. Plunkett, Kim & Virginia Marchman (1991) U-shaped learning and frequency effects in a multi-layered perceptron: implications for child language acquisition. Cognition 38, 43-102. - , - (1993) From rote learning to system building: acquiring verb morphology in children and connectionist nets. Cognition 48, 21-69. Pregel, Dieter & Gerd Rickheit (1987). Der Wortschatz im Grundschulalter. Hildesheim: Olm. Ragnasdóttir, Hrafhhildur, Hanne G. Simonsen & Kim Plunkett (1996) The acquisition of past tense inflection in Icelandic and Norwegian children. Paper presented at the 28th annual meeting of the Child Language Research Forum in Stanford. Rainer, Franz (1988) Towards a theory of blocking: the case of Italian and German quality nouns. Yearbook of Morphology, 155-185. Ruoff, Andreas (1981) Häufigkeitswörterbuch gesprochener Sprache. Niemeyer: Tübingen. Rumelhart, David E. & James L. McClelland (1986) On learning the past tenses of English verbs. In: James L. McClelland, David E. Rumelhart and the PDP Research Group (eds.): Parallel Distributed Processing. Vol. II. Psychological and Biological Models. Cambridge, MASS: Bradford Books / MIT Press. 216-270. Schindler, Simone (1997) Heißt nicht gesungen ... GESINGEN! Untersuchungen zum Erstspracherwerb von Partizip-Perfekt-Formen im Deutschen. MA Thesis München. Slobin, Dan I. (1973) Cognitive prerequisites for the development of grammar. In: C. Ferguson & D. I. Slobin (eds.): Studies of child language development. New York: Holt, Rinehart & Winston. 175-208. Stemberger, Joseph Paul (1993) Vowel dominance in overregularizaton. Journal of Child Language 20, 503521. Stemberger, Joseph Paul & Brian MacWhinney (1988) Are inflected forms stored in the lexicon? In: Michael Hammond & Michael Noonan (eds): Theoretical morphology. Approaches in modern linguistics. San Diego: Academic press. 101-116. Strauss, Sidney & Ruth Stavy (1982) U-Shaped behavioral growth. New York: Academic Press. Weyerts, Helga & Harald Clahsen (1994) Netzwerke und symbolische Regeln im Spracherwerb: Experimentelle Ergebnisse zur Entwicklung der Flexionsmorphologie. Linguistische Berichte 154, 430-460. Xu, Fei & Steven Pinker (1995) Weird past tense forms. Journal of Child Language 22, 531- 557.

Conxita Lieo (Universität Hamburg)

Proto-articles in the Acquisition of Spanish: Interface between Phonology and Morphology

1. Introduction This paper deals with the early appearance of "proto-articles" in Spanish. Its purpose is twofold, having a theoretical and an empirical orientation. Empirically, it is well-established that the first words children produce are content words, that can be equated with nouns (Hawayek 1995: 150). On the other hand, the appearance of "fillers" or "filler syllables" has also been recognized for the early stages of language production (Peters & Menn 1993, Peters & Strömqvist 1996, among others). Fillers seem to be a sort of "place-holders" for functional words, which are argued to enter the language at a later age. In early monolingual Spanish data, that had been collected with the purpose to study phonological acquisition, it could not be ignored that such fillers made their appearance before the 25-word-point and during the one-word stage. A detailed inspection of their development leads to the establishment of a clear link between fillers and articles. This link helps to clarify the function of prosodie constraints in the early grammar, as they have been proposed, for instance, by Demuth (1996). At the same time, the early appearance of proto-articles has theoretical consequences, because of its bearing on the question of the availability of functional categories. That is, are functional categories available from the very beginning or do they have to develop after a stage consisting only of content words? The evidence presented in this paper argues in favor of the availability of the functional category "Det" at the one-word stage, i.e., at the onset of language production.

2. Data The data analyzed in the present study belong to a longitudinal research project involving four monolingual children acquiring Spanish in Madrid and five monolingual children, acquiring German in Hamburg.1 Their productions began to be recorded at about nine months of age, in the pre-word stage. The Spanish recording sessions took place on a monthly basis and the German recordings were conducted twice a month. The children were visited at their homes by

'

The study was supported by the DFG (projects BIDS and PAIDUS), which I want to thank for its support. I want to express my gratitude to the research assistants, Dr. El Mogharbel and Dr. Michael Prinz, and to the students who worked for the project, especially Cristina Trujillo and Fatima Marinho, as well as to my colleague, Prof. Antonio Maldonado, who conducted the recordings in Madrid.

176

Conxita Lieo

one or two investigators and were recorded while playing or interacting with the mother and one investigator. Toys, infant books and a few props for object naming were used. The Spanish data were video and audio-taped, the German data were audio-taped and supplemented by careful notation of the situational context. The recordings were made by means of a highfidelity Sony TCD-D10 PRO cassette recorder and a portable Beyerdynamic microphone concealed in a vest worn by the children. Both data sets were phonetically transcribed by two trained phoneticians, who made a narrow transcription using IPA, supplemented by additional symbols designed for child language. For the purpose of the present study the data of two Spanish children, Maria and Miguel, have been selected. The close inspection of their productions began at the 25-word point, i.e. the first recording in which the child produces a minimum of about 25 word types, which has been introduced in the literature on phonological acquisition research as a reference point, more reliable than age. Because several fillers were already produced at that point, the data were traced back to the occurrence of the first fillers. And they were inspected forwards, up to 1; 10 for Maria and 1;8 for Miguel. The analyzed sessions correspond to the following children at the following age points:2 Maria (I;4,21), (1;6,3), (1;7,24), (1;10,17); Miguel (1;4,5), (1;5,1), (7;6,7), (1;7,26), (1;8,23). As regards the German children, the 25-word point was also inspected for all children of the project, corresponding to the following sessions: Bernd (1;8,6), Britta (1;5,28), Johannes (1;8,1), Marion (1;5,3) and Thomas (1;6,16). A preliminary analysis lead to the conclusion that the German data contained almost no fillers at all at this point. A more detailed analysis of the German data as well as a comparison with the Spanish data is presented elsewhere (Lieo 1997). Here the German data will only be drawn upon as far as they illustrate a prosodie difference between the two sets of early lexicons, that might be crucial in trying to provide an explanation for the earlier appearance of the proto-article in Spanish than in German.

3. Results The focus of analysis is on the emergence and development of the so-called filler syllables preceding content words, especially nouns and only a few verbs. The production of these filler syllables will now be longitudinally described for each child. Since these syllables are viewed here as proto-articles, a summary of the different forms of the definite and indefinite articles in Spanish are summarized in table 1, in order to facilitate the presentation and discussion of the data.

2 As usually, the age of the child is indicated by means of (year;months,days). The session corresponding to the 25-word point has been underlined.

Proto-articles in the Acquisition of Spanish

177

DEFINITE ARTICLE

INDEFINITE ARTICLE

SINGULAR

SINGULAR

el la

MASC. FEM.

PLURAL

los las

un una

PLURAL

unos unas

Table 1. Definite and indefinite article in Standard Spanish María (1;4,21): A few of the spontaneously produced nouns are preceded by a syllable, generally an optional laryngeal followed by a low or middle central vowel, as shown in (la). Besides 'parí produced once as [pah], most of the nouns produced are disyllables, some of them being pronounced with three syllables because of the filler. Trisyllables are generally reduced to disyllables {Alberto, rastrilló)·, in some cases, they remain trisyllabic with the first syllable substituted by a default syllable, i.e., similar in form to the filler syllables (lb). The often produced words 'st (yes) and 'no' are never preceded by a filler, except one single occurrence of 'no' as [?anoh], which might correspond to an emphatic 'que no'. In a few cases, the filler has a different form: [hu], [?u] or even [1B]. (I) a. [apah] h

pan

'bread'

b. [?ajki:ja]

rastrillo

'hackle'

pala

'shovel'

[?waßebeh]

'shovel'

[?aJ3eJjoh]

chupete babero

'pacifier'

['ha'beta?] pala [?u'pailas h ] pala

'shovel'

[?abaijo]

babero

'bib'

[lB'baieh]

pala

'shovel'

[a'memch]

mamá cubo

'mum'

[?9'bAla ]

[hu'guwa]

'bib'

'bucket'

Maria (1;6,3): Most of the nouns produced are disyllabic, some with a preceding filler syllable, and a few trisyllables. Filler syllables appearing before monosyllables and disyllables are of three different forms, shown in (2): an optional laryngeal with a central generally low vowel (2a), with a front mid vowel (2b) and a syllabic nasal or a nasal preceded by a back high vowel (2c). The fillers given in (2a) constitute the most frequent type. Although the distribution of these syllables appears to be random, a certain correlation between feminine with (2a) and masculine with (2b) and (2c) might begin to take place, but still in a very inconsistent way. That is, it could be argued that the filler syllable preceding feminine nouns is reminiscent of la and the one preceding masculine nouns is reminiscent of el (2b) and un (2c). Trisyllables are either reduced to disyllables or have a reduced first syllable (2d).

Conxita Lleó

178

[?a'pa:le:h]

peine

'comb'

b. [hsiTie:?] tren 'train'

[WOwoh]

globo

'balloon'

[hclcuwo] cwèo'bucket'

'TV'

['h8'ci:joh]s;7/a

(2)8.

[h0 ? 'dajje h ] tele

'chair'

[aT)ubah] [a'wu:] [awa'wn:] c.

pupa 'pain' luz 'lamp' guau guau 'bow-wow'

['mpab]

palo

'stick'

d. [amaJ'jB]

caramelo 'candy'

papa

'porridge'

[am'mejnoh]

caramelo 'candy'

diente

'tooth'

[aJjoi'di]

calcetín' sock'

[hiceXa]

carretilla 'push cart'

[n'daJj ε ]

A few verb forms make their appearance in this session. The imperative form 'mira1 appears a few times, spontaneously and repeating an adult utterance; it is never preceded by a filler. The forms 'cae' and 'se cae' are produced as [gai'ja^1], [aí'gajjah] and [o:gaj]. The filler syllable appearing in the last two forms might correspond to 'se'. The monosyllables 'sf and 'no' are produced several times, and they are never preceded by a filler. Maria (1;7,24): In general, the prosodie structure of words does not differ from the previous session. Many nouns are produced with three syllables, either a target disyllable expanded by means of an initial filler syllable or a target trisyllable, generally produced with an initial reduced syllable, comparable to a filler in its form. Phonologically, most of the initial syllables of nouns are default syllables, formed by an optional laryngeal and a central open or mid central vowel; examples of these are given in (3a). In (3b) there appear examples of other initial syllables, [0η], [ej], [h0], [he], [hY], It is not the case that feminine nouns are in (3a) and masculine nouns in (3b); what seems rather to be the case is that the most frequent initial syllable has a central vowel, either low or mid, as shown in (3 a), and it is applied to feminine as well as to masculine nouns; and that all nouns receiving one of the other syllables, appearing in (3b), are masculine, with the exception of 'pupa'. Monosyllabic nouns are preceded by a filler, even when imitating an adult model.3 Examples of trisyllables with an initial reduced syllable are listed in (3c). Here, all initial syllables resemble fillers; the most common type has a central low or mid vowel, and two words (mordisco and calcetín), having a front vowel or a high back, are masculine.

3 It has been often reported that young children omit function words in repetition tasks. An explanation for the behavior of the Spanish children is offered in the discussion.

Proto-articles in the Acquisition of Spanish

(3) a. [a'fah]

'custard' 'chair'

b. [e'ce] [he'fah]

tren flan

'train'

[asija]

flan silla

[?ahsijah]

silla

'chair'

[ehupah]

pupa

'pain'

[a.waw.wo]

globo 'balloon'

gorro

'cap'

[azece]

leche

'milk'

[ejijojo] [annanah]

nene

'child'

[abado] [abado] [habadoh] [haebato] [abado] [habadoh]

plato plato plato plato pato pato papa

'dish' 'dish' 'dish' 'dish' 'duck' 'duck'

[hYt0jlo]

toro

'bull'

[waebabah]

'porridge'

'custard'

c. [?3EGa.Gah]

'doll'

[aßatcih]

muñeca bocadillo chupachup petit suisse

[haßha'ei]

petit suisse

'yoghourt'

[?9teteah]

chupete

'pacifier'

[?aebibih]

chupete calcetín mordisco

'sock'

[a?ka.i:joh] [a.pa'su:]

[ud?nedi] [h0cikoh]

'sandwich' 'lolly' 'yoghourt'

'pacifier' 'bite'

Many verb forms make their appearance in this session, for instance 'mird, 'a dormir', 'esperà, 'a ver', 'se acabó'. In general, they are not preceded by filler syllables, only the form 'a ver' is produced either [veh], [ha'vch], [a'vch], [a^vaeh], [?a'veh] and [?a'veh]. The imperative form mira 'look', in spite of being a trochaic disyllable, like the majority of nouns, is never produced with a filler, as if the child was differentiating between the two word classes, in spite of prosodically similar targets. Some other target forms that are longer than two syllables are truncated to disyllables: [Vaijah] for espera, [TcawtE] and [ga'w^oa] for 'se acabó', [ma'mi], ['maemih], ['mamaeh] for 'a dormir". The latter is also rendered as [hemeh], [?asmeh] and [hameh], in a way more similar to disyllabic nouns; and the same is true for 'a comer·' rendered as [ava'jeh] and [haeba'je]. Maria (1;10,17): There appear more initial filler syllables than in the previous sessions, but they are not more differentiated, consisting mainly of an optional laryngeal plus a central low vowel (4a,b); a few different syllables appear before some masculine nouns (4c), but the undifferentiated or default syllable has been extended to more nouns, feminine (4a) as well as masculine (4b). Trisyllables are generally produced with a reduced first syllable, as in previous sessions. Some examples are given in (4d).

Conxita Lieo

180 playa

'beach'

b. [hß'wowo]

globo

'balloon'

[?a'sijah] [ha'sije]

silla silla

'chair' 'chair'

[a'wowoh] [ha'vaço]

globo vaso

'balloon' 'glass'

[awKu'gB?]

vaca

'cow'

tren

'train'

[?a'bAkah]

vaca

'cow'

río

'river'

(4) a. [a'bajlah]

[?a'?ijoh] 11

'hands'

[a'Yaijo ]

gorro

'cap'

pelo

'hair'

a. [ha'bada h ]

pelota

[?a3l'xuwoh]

cubo

'bucket'

[?agagah]

muñeca

'ball· 'doll'

[bc9b h i]

lápiz

'pencil'

[?a:pBsanE^]

manzana

'apple'

[a:mano] c. [mbhelo11]

manos

[îaîga'dzjçB^jcas/to h

'house (dim.)'

[aga'dija ] [e'gadi neh] [?akB'çiireh]

gatito calcetines cocina

'cat (dim.)' 'socks' 'kitchen'

[aßata'jo11] [apa'cacoh]

flotador payaso

'life-belt' 'clown'

The imperative form 'mira' appears 21 times, never with a filler. On the other hand, several target infinitives preceded by the preposition 'a' or 'para' are rendered with an initial central low vowel optionally preceded by a laryngeal: 'a jugar', 'a pinta?'a guardar', 'a dormii·', 'para patinar'. The form 'a ver' always has an initial low central vowel preceded by an optional laryngeal. There are no monosyllables in these data besides the particles 'sì and 'no'. During this session there appear a few two-word utterances for the first time. A complete list of them is given in (5). It is interesting to notice that there are no filler syllables in these utterances. (5) [?a'ui: 'marnu]

abrir mamá

[ñame 'ot^ah]

dame otra

[mia 'bupha]

mira pupa

[mia 'kumeh]

mira cuna

['mi^ja 'xa*· vi*·]

mira Javier

[ma'ijß? VaTcih] María aquí

[?ala 'phapA]

hola papi

[?mÌB gabduh] mira (a) Kaput

Miguel (1;4,5): All productions are disyllables, corresponding to adult nouns, and the word 'adiós'. Some of them are preceded by a filler syllable formed by an optional laryngeal and the vowel [a] (6a); only the word 'lápiz' has a different initial syllable [l£d]; trisyllables are generally truncated to disyllables; in a few cases, they have a default initial syllable (6b).

Proto-articles in the Acquisition of Spanish

(6) a.

181

b.

['ha'tJilca]

tic tac

'clock'

[habe'ba]

oveja

'lamb'

[haveva:] [?a'vava] [avava] ['ldöa'vi]

vaca vaca boca lápiz

'cow' 'cow' 'mouth' 'pencil'

[m'geka]

muñeca boquita

'doll' 'mouth (dim.)'

p i 'giga]

Miguel (1;5,1): Most productions are again disyllables, preceded or not by a filler syllable, target trisyllables being generally reduced to disyllables. Filler syllables are more differentiated than in the previous session, as shown in the examples in (7): one type of filler consists of an optional laryngeal followed by a vowel, either mid front (7a) or high back (7b); another type contains a nasal nucleus (7c); and there are a few cases containing a liquid consonant (7d). As it can be seen from the data in (7), there is a certain correlation between the gender of the target noun and the type or subtype of initial syllable: all target nouns in (7a) are feminine, whereas most nouns in (7b) and (7c) are masculine. Exceptions to these generalizations are target feminine vaca and nariz, and masculine guau guau, appearing in both sets. Also target feminine pelota appears under (7c). Actually, the distinction exhibited in (7a) as opposed to (7b-c) might also be one of 'definiteness', because the fillers appearing in (7b-c) remind of the indefinite article W and those of (7a) remind of the definite 'et. (7) a. [?0nani] [enani] [?0nane] [?anani]

nariz nariz nariz

'nose' 'nose' 'nose'

b. [unna'ni] [?ug'lapi] [?u?uma:]

nariz lápiz camión

'nose' 'pencil' 'truck'

nariz

'nose'

[u:vaj]

barco

'ship'

[he'gak'.a]

vaca

'cow'

púlelo]

reloj

'clock'

[?e'kaka] [he'gaka]

vaca vaca

'cow 'cow'

[uinvasva] [?u:n'kik:9]

guau guau 'bow-wow' quiquiriquí 'cock-a-doodle-do'

'cow'c. 'cow' 'water'

c.

[?ej'kaka] vaca [he:gaga] vaca [he'a:va] agua

[nldka] [hn'gika]

tic tac tic tac

'clock' 'clock'

guau guau 'bow-wow'

[m'gyka] [n'gaga] [ni'kapi]

tic tac vaca lápiz

'clock' 'cow' 'pencil'

[lalapi]

lápiz

'pencil'

[hm'baßa]

guau guau

'bow-wow'

[lava]

agua

'water'

[m'bite]

[he:vava] d.

pelota

'ball'

There are a few verb forms, which do not seem to be differentiated from the nouns, because they once receive an initial filler, as in [he'mi:jae] interpreted as mira and [hœ'javi] interpreted

Conxita Lieo

182

as abrir. It has to be noticed that these forms appeared only once so that their reliability is doubtful. Nevertheless, Miguel's preferred productions, especially for nouns, are trisyllabic, with the stress generally on the second syllable and in a few cases on the last syllable. Miguel (1;6,7): The same prosodie tendencies of the previous session continue to develop for nouns: preference for disyllabic and trisyllabic productions consisting of a disyllable preceded by a filler syllable. But fillers are now reduced to two basic forms: an optional laryngeal followed by a central rather low vowel is the most frequent filler, applied to feminine as well as to masculine nouns (8a), and a syllable containing a nasal (8b); in a few cases, the filler is constituted by la (8c). The data support a tentative distinction between definite and indefinite article, without gender differentiation; this result might be carrying out a tendency already started in the previous month, as pointed out above. As regards target trisyllabic nouns, they are generally reduced to disyllables, many of them being headed by the kind of filler syllable that precedes target disyllables (8d). a. [hagAodc:e]

coches

'cars'

b. [m:ba]

pan

'bread'

[ha'gode]

coche

'car'

[unte]

tren

'train'

n

[habadas:e]

coche

'car'

[u bejje]

peine

'comb'

[ha'guna]

cuna

'eradle'

[hgflgodeej]

coches

'cars'

[?avevo]

huevo

'egg'

[n:'luna]

luna

'moon'

[a'veaßo]

huevo

'egg'

[huvsjvo:]

huevo

'egg1

[hajevo:]

huevo

'egg'

c.

[abada:]

moto

'motorcycle'

[hdagoodee]

coches

'cars'

[habata:]

moto

'motorcycle'

[lagoöc:j]

coches

'cars'

[?amuma:] [a'muma:] [had3i'ga:] [hagi'ga]

zumo zumo tic tac tic tac

'juice' 'juice' 'clock' 'clock'

[la gu:na] d. [hapata] [abada:]

cunas

'cradles'

zapatos pelota

'shoes' 'ball'

A relevant fact for the interpretation of proto-articles, which will be discussed below, is that several times in this session Miguel produces a noun without filler when looking for the object, and then repeats the same noun preceded by a filler when pointing to the object that he has just found. This is the case for cuna, produced as [ha'guna], and huevo, [hajevo:]. A few verb forms appear: [ma'me] for dame, [ka'ja] for a callar, [a'n:a] for anda, [mi'a] for mira. Interestingly, they all exhibit an iambic prosodie structure in the child's production. Certainly, they constitute a very reduced set to allow for strong generalizations, but these imperative forms are treated differently from nouns in that, besides being iambic and not trochaic, they are not preceded by a filler. For the first time, a few two-word utterances are produced; they are all given in (9):

Proto-articles in the Acquisition of Spanish

(9)

183

[ma'me e'mu:]

dame el muu ('cow')

[oda Vejvo]

otro huevo

[no'da 'doda]

no está (el) coche4

Miguel (1;7,26): At this point, Miguel's data exhibit a clear gender difference, expressed by the proto-articles adopting two main forms: [u] or [un] for masculine (10a) and [a] for feminine (10b). The masculine form alternates a few times with [ε] and once with [εΐ]; and the feminine form alternates twice with [la] (10c). This points to the existence of two different types of article, definite and indefinite, but the occurrences of what seems to be the definite article are too scarce to assume that it has been acquired; on the other hand, as regards the feminine proto-article, being mainly realized as the central low vowel, it is not possible to decide wheter it corresponds to the definite form la or to the indefinite una. The examples marked with an asterisk (*) correspond to cases in which the form of the proto-article is "wrong" given the gender of the target noun, i.e., biberón is masculine and moto feminine, in Standard Spanish.

'foot'

b. [aT)ici] [ekuja]

mariposa

'bicycle' 'butterfly'

tren

'train'

[a'pojB]

mariposa

'butterfly'

tren

'train'

[hcvoijc]

mariposa

'butterfly'

'brush' 'pencil'

[hamKamoija] [aWa:]

mariposa pelota

'comb'

[a'bota]

pelota

'butterfly' 'ball' 'ball'

a. [un'pe:]

pié

'foot'

[υ'ρε:]

pié

[u'd8n] [υη'άεη]

[?upm'0aei] pincel lápiz [u:a'pi0] peine [υ:δεηε]

bici

[Aka'jo]

camión

'truck'

[hamaθakεta]

bicicleta

'bicycle'

[un"ko39]

coche

'car'

[?a'v8vi](*)

biberón

'bicycle'

[ιιηιιε'ζε:]

jersey

'sweater'

[υη'ίεο]

Teo

c.

[uteo]

Teo

[ε'πιοΐο](*)

moto

'motorcycle'

[o'deo]

Teo

[elpapel] [la'toja] tuya [la'gaga] caca

papel 'yours' 'filth'

'paper'

The initial filler [a] , besides being the feminine form of the article, appears in contexts where prepositions like de, en or a plus infinitive would be expected, as for instance in [a:'baba:] 'de papá', [îa'gcnte] 'de dientet, [a*bate] 'en el wáter', [îaba'ja] and [agaja:] 'a callar'. Some of the occurring verb forms have an initial syllable [ε] or [?ε], which is interpretable as the verb es or as the reflexive se: [ε-.gajo] 'se cayó', [ενΛενο] '(está) durmiendo'. 4

This utterance is a repetition of the utterance produced by an adult immediately before.

Conxita Lleó

184

There are some two-word utterances, and a few three-word utterances. The following examples represent some spontaneously produced multiword utterances: (11)

[la moto albata:] [νενυη teo gaiöe] [ΐεο a'batc]

la moto de papà se va(?) Teo a casa Teo en el wáter

Miguel (1;8,23): The differentiation that began in the previous sessions, and that partly crystallized in the last one, is now clearly established, the form of the fillers preceding nouns are [ε], [e], [u], [a] with optional laryngeal, [εΐ], and [n] for masculine; [a] with optional laryngeal, [na], [la] and [lo] for the feminine. Two distinctions have been clearly established, one of gender and another one of definiteness. Examples are given in (12). The data in (12a) contain masculine nouns exclusively (the only exception being moto), the proto-article forms corresponding to the definite article el, (12b) also contains masculine nouns, the proto-articles corresponding to the indefinite article un, and the nouns in (12c) are all feminine, a few having the indefinite article una, and a few the definite article la, some of the fillers still retain the form of the proto-article, and it is not possible to decide whether they correspond to the definite or to the indefinite Spanish article (la or una). There is only one case of gender mismatch, again the word moto, which is feminine in Standard Spanish (shortened for motocicleta). The child might be attributing the masculine to moto, because of its ending in -o. Although in Standard Spanish no unambiguous correspondence between gender and ending of the noun can be established (see Harris 1991), there is a clear tendency for nouns ending in -a being feminine and nouns ending in -o being masculine. And this is certainly the case in Miguel's early lexicon.5 Another piece of evidence for the grammar of gender in Miguel's data is the attribution of the masculine to muu 'cow', which in Spanish is feminine, Ία vaca'. Apparently, the child takes the ending of the noun, the onomatopoetic [mu:], as evidence for treating it as a masculine. (12) a. [a'öol]

b. sol

'sun'

[η'άεΐ]

[c'uoßwch] coche

'car'

[în'ctem]

tren

'train'

[?ε'ςονε1ι]

coche

'car'

[hu:'YOue:]

coche

'car'

[e'godch]

coche

'car'

[?ggoôe]

coche

'car'

[ewweJ'jo]

camion

'truck'

[hmbexo]

perro

'dog'

[ε'διιπω.β]

zumo

'juice'

[ηΐβ'γοη]

león

'lion'

tren

'train'

^ The form elamoto, segmented as el amoto, instead of the standard la moto, is quite common in adult spoken varieties of Spanish. That is, speakers are "misled" by the ending of the noun, and classify it as masculine.

Proto-articles in the Acquisition of Spanish

[?euodo](*) moto 11

185

'motorcycle'

c.

[hejyiga ]

tic tac

'clock'

[îogeôe]

casa

'house'

[he'mu:]

muu

'cow'

[adei'ida]

estrellitas

'stars (dim.)'

h

[e'mo:]

muu

'cow'

[a'woÖB ]

mariposa

'butterfly'

[el'bes] [el'luBŒ]

pez Curro

'fish'

[ha'boza:] [a'bala]

mariposa pala

'butterfly'

[albeRo]

perro

'dog'

[a'vaga]

vaca

'cow'

leche

'milk'

[nai'nuna* ] [la:luna] [la'boza:]

luna luna mariposa

'moon' 'moon' 'butterfly'

[lo*baleh] [una'bah]

pala pala

'shovel' 'shovel'

['laleo] 1

'shovel'

Verb forms do not generally get a filler, as for instance ['kome] for corre, [ko'lae^] for correr or [veßa] bebe, the only initial syllable corresponds to the pronoun se and appears in [o'wode] 'se rompe' or [0n0*bobe] 'no se rompe'. There appear about twenty two-word utterances in this session. Some examples are given in (13): (13)

[ge'dabalo]

aquí está (Pata)Palo

paTwkt! 'ölußol] (la) pelota de fútbol [uep9 'leteh]

[me'a UWB]

mira huevo h

e'balE ]

ese/a (es) la pala

bebe leche

Some two-word utterances are just constituted by the word otro plus a noun; obviously, otro is treated as an invariable form, appearing always in the masculine, independently of the gender of the noun: (14) ['ογο Yi'yayo] [du γί'γυ:] [DUIO uii'gah]

otro tictac-reloj otro tic tac otro tic tac

[?odo 'lune] [?odu 'bado] [hudja bis]

otra luna otro pato otro pez

3.1. Form of the proto-aiticle Filler syllables begin to appear in the position preceding nouns at a very early age. They cannot yet be considered articles because they are not fully developed, whether formally or functionally. At first, they generally consist of a laryngeal consonant (either [h] or [?]) followed by a central low or mid vowel. This type of syllable appears before monosyllables and disyllables. On the other hand, a similar underspecified or default syllable is often found as a substitute for the initial syllable of a trisyllable or, in a few cases, a target quadrisyllable.

186

Conxita Lleó

These filler syllables, when considered longitudinally, approximate in a progressive way to the adult articles. What begins as an undifferentiated syllable, a sort of a "place-holder" for the article, evolves through several paths to forms that are progressively more and more differentiated, and that finally will be indistinguishable from full-fledged articles (for a description of subsequent stages see Lleó 1997). It is thus justified to consider them as proto-articles. On the one side, they appear very frequently with monosyllabic and disyllabic nouns. Especially the latter alternate in their realizations with and without the filler. They appear with the filler in contexts which generally require the article, as shown in §3.2. Verbs are generally not preceded by a filler, except monosyllabic forms like ven, and a few occurrences of what can be interpreted as a preposition plus an infinitive. 3.2. Contexts of appearance: meaning of the proto-article Both Spanish children begin producing nouns at about one year of age. Still at the one-word stage, at about the 25-word point and even earlier, the Spanish children begin to produce the first filler syllables that can be classified as proto-articles. Although at first they are not numerous, in the following sessions the number of proto-articles gradually grow. Even though the proto-article is not phonetically well developed, it is contextually consistent, in the sense that all proto-articles are produced in contexts which require them. A classification of different types of context will help to see the underlying systematicity of the occurrences. a) The default syllable is used in contexts which can be called "specific-presentational". They occur as answers to questions like 'what is this?', 'what are you giving to the doll?', when the child is at the same time pointing to an object or figure. The child may also take an object and "present" it while saying its name. This is a kind of context which occurred often, in recordings in which words were elicited from the child by means of picture books or objects that the child was expected to name. The use of the early proto-articles is not yet obligatory in these contexts, and they are often missing. Maria's and Miguel's productions were in general adjusted to the expectations, in the sense that they often produced nouns provided with the filler when asked to name objects. In a few cases, the adult asks "what's the name for this?", i.e.,"cómo se llama?" and sometimes provides the article herself, probably in order to "help" the child's lexical access. In these cases, the child does not produce the article. b) Other contexts, in which the article does not appear can be characterized as "acting": the child takes an object and acts on it, i.e., plays with it and accompanies the action with the word. In these cases, the child is often naming the object as in a "vocative", and the noun is never preceded by the article. c) There is a third context, in which the article is often but not always omitted: "echo" productions, i.e., repeating an utterance immediately after the adult has produced it. Generally, the adult produces the noun preceded by an article. In the literature it has generally been said that in such imitations, children omit functional words. Although this is the case in the present data up to a certain extent, in many cases of repetitions Spanish children produce the noun preceded by its article.

Proto-articles in the Acquisition of Spanish

187

As it was pointed out in relation to Miguel's data, at (1;6,7) he produced a few nouns without articles in what could be called a "volitional" context, in which he remembered about an object or was reminded about it by the adult. After having produced the name of an object without article, when he found the object, a few minutes later, he showed it in a "specific-presentational" way and then produced it preceded by the proto-article.

4. Discussion It has often been reported that young children concentrate on the production of content words and leave function words out of their productions, even in repetition tasks. On the other hand, it has been convincingly argued that young children are aware of function words in perception. Gerken and Mcintosh (1993) have demonstrated that children acquiring English are capable of decoding content words in their language better if the functional words that accompany them are grammatical in English than if nonsense or invented syllables appear in the position of functional words. Another phenomenon that has been often discussed in the literature are fillers or filler syllables, which have been attributed different functions. A rough summary of the interpretations that have been given to fillers would attribute them to prosodie reasons. They are often seen as empty syllables that contribute to the rhythm of the utterance, making thus their appearance during or after the two-word stage. Some authors (Peters & Menn 1993, among others) have emphasized their function as "place-holders" for functional words and thus as protomorphemes. But the fillers have always been reported at an age in which the child already has some syntax, i.e., is beyond the one-word stage. This view is not different in studies on the acquisition of Spanish. López Ornat (1994:108) finds that 'children acquiring Spanish, at the two-word stage, produce amalgams of article + noun (or unanalyzed "nominal" utterances), where there is a pre-posed sound stamming from the unstressed article: "enéne" (for 'el nene'), "aquéca" (for 'una muñeca').' But as we have seen above, fillers reminiscent of the article make their appearance before the two-word stage in our acquisition data of Spanish. A methodological observation should be made here. It is important to note that our data were collected at a very young age, with the purpose of investigating phonological acquisition. Some of the data which have inspired claims on fillers or on amalgamated forms were collected with the aim of investigating syntactic or morphosyntactic acquisition, thus focusing on the two-word period and thereafter. Given the early character of our data and the very careful narrow transcription to which they were submitted, it was possible to discover the presence of syllables, which otherwise might have gone unnoticed.

188

Conxita Lleó

4.1. Contextual conditioning Focusing on the analysis of our data, it is obviously not enough to characterize the sequences of default syllable plus (target) noun as "amalgamated" utterances. Certainly, at the very beginning of their appearance filler syllables do not yet exhibit distinctions of gender, number or definiteness. One reason for this lack of differentiation certainly is an undifferentiated phonological system, in which a lot of underspecification is at work. In the first sessions in which prenominai fillers appeared, laryngeals were the preferred consonants, plus some labials and a few coronals; as regards vowels, these were highly undifferentiated, with predominance of central low and mid vowels. Emerging from such a limited and undifferentiated phonological module, subtle morphological distinctions were not feasible. On the other hand, filler syllables are used in consistent contexts, especially the one characterized as "specific-presentational" above, whereas they are never used in the vocative, for instance. Certainly, they are not used in all "specific-presentational" situations, but when used, the context can often be characterized as "specific-presentational". Default syllables preceding disyllabic nouns are decisive to decide about the function of fillers. They are not obligatorily produced, as would be the case if they simply were part of a trisyllabic template. That is, monosyllabic nouns are generally produced with a filler, but disyllabic nouns are either produced with or without the filler. And when produced with the filler, it fits in a context, in which Standard Spanish would use an article. As far as trisyllables are concerned, they are either truncated to disyllables or produced as trisyllables with a reduced first syllable; and this reduced initial syllable resembles the proto-articles of disyllables. Certainly, the latter production of trisyllables could also be interpreted as undergoing a process of truncation of the initial syllable, which is then filled by the default syllable. This seems to underly the interpretation of López Ornat above, as regards 'muñeca'. And it is applicable to many cases in our data, especially when the child alternates the trisyllabic with a disyllabic production of target trisyllables. Leaving the problem of trisyllables as partly unsolvable for the moment, target disyllables, with their dual productions, with and without a filler, deliver the best evidence to pursue the analysis of fillers as proto-articles. Taking into consideration the contextual conditioning mentioned above, fillers used as proto-articles fulfill a quantificational function: they are used to designate a specific entity, in general a single noun. On the other hand, these proto-articles soon begin to agree with the following noun. Maria at 1;6 and Miguel at 1;5 begin to make a gender distinction, so that a "primitive" agreement between Det and Ν takes place for the first time at this early age. Clearly, the agreement posited here for the early Spanish data does not coincide yet with the agreement in Standard Spanish because the proto-classes of nouns overlap, but are not identical with noun classes in Standard Spanish. Nevertheless, our analysis gives support to Hyams (1986: 138-40), who finds agreement within NP at a very early age in the acquisition of Italian. And it suggests the adoption of the DP hypothesis, according to which the Noun functions as the complement of the article.

Proto-articles in the Acquisition of Spanish

189

Along these lines, it can be argued that the relational notion of complementation is already used by the child, as displayed by our Spanish data, from 1;6 on for Maria and 1;5 for Miguel. It is important to emphasize that the first two-word utterances appear at 1;10 in Maria's data and at 1;6 in Miguel's data. And that the first proto-articles make their appearance during the one-word stage, when no predicational relation has explicitly been used yet. Our data cast thus doubt on three-stage proposals, according to which there should be an initial stage of projection of a lexical element, followed by a predicational stage of two-word utterances and a third complementational stage of Det+N. Such a model of acquisition based on a three-stage progression has been argued for by Hawayek (1995:15 If). She postulates a first stage, in which the syntactic structure for noun phrases consists in the projection of a lexical element (fig. la); at this stage the child produces bare nouns. This is followed by a predicational stage of two-word utterances, represented as in (fig. lb), in which the child may say este papá or papá este\ este does not act as a determiner but as a 'pronominal deictic', because it has the same referent as the noun which precedes or follows it. Such sentences are better described equationally: X = Y. The representation in (fig. lb) encodes the fact that children have acquired a predication relation, established in terms of a mutual c-command relationship. It seems to be that this particular knowledge precedes, and must be acquired before children develop a notion of complementation. In the third stage the functional element is inserted in the position of the head (X); the noun is inserted in the position of the complement (Y) so that it may function as such (fig. lc). xmax

Xmax

γ max

Xo

Y°

γ max

X°

(N)

(Ν)

(Fig. la) Head

(N)

(Fig. lb) Predication

(Del)

(N)

(Fig. lc) Complementation

The precedence of two-word utterances over the acquisition of all inflectional morphology (including the acquisition of articles) has also been defended by Pizzuto & Caselli (1992) for Italian. Certainly, the criteria applied by them differ from the ones applied in the present study, since they were trying to establish the morphophonological mastery of inflectional morphemes, whereas here the child is credited with some functional category as soon as (s)he begins to have command of some morphophonological distinction. Besides, the present analysis refers to articles only, and not to the whole inflectional morphology. But as regards the acquisition of the functional category 'Det', evidence has been presented for the appearance of stage (lc) be-

190

Conxita Lleó

fore stage (lb). In fact, our data show that when two-word utterances are produced, the production of proto-articles diminish, in the sense that the initial two-word utterances contain almost no fillers. There are prosodie reasons for this reduction, which will be discussed below. Probably, the reduction of proto-articles in the production of early two-word utterances might have given the impression that the stages are: I. one-word stage: projection of a lexical unit II. two-word stage without functional words: predicational stage III. two-word stage plus functional words: complementational stage Our data support the following sequence of stages: I. one-word stage: projection of a lexical unit II. one-word stage plus functional words: complementational stage III. a. two-word stage without functional words: predicational stage, and b. single content words with functional words: complementational stage IV. two-word stage plus functional words: stage of integration of complementation within predication As our Spanish data have shown, Ilia and Illb take place simultaneously. But at stage III functional words are not found in two-word utterances, being only used with single nouns. The fact that they are integrated within two-word utterances at stage IV has erroneously suggested that functional words appear for the first time within this type of utterances. However, the presence of Det+N at stage II, before the two-word period, provides evidence that complementation is acquired before predication. It also provides evidence in favor of the availability of functional categories from the earliest stages of acquisition. It actually supports what Demuth (1992: 84) characterizes as the Functional Projection Hypothesis, by which "the building of syntactic structure may actually precede the phonetic (or morpho-lexical) realization of functional heads themselves". She proposes a PF Filter (1993: 101), by which "phonetic form (e.g. the surface realization of lexical items and morphosyntactic strings) will develop gradually over time, subject to maturational constraints on production". This proposal is confirmed by our data on Spanish, in the sense that the functional head is present from the earliest productions, but it is undifferentiated because the phonological component is still immature. And still, in spite of its immaturity, the Spanish phonological component allows for the presence of the functional category earlier than in other languages, due to its prosodie characteristics. 4.2. Prosodie conditioning What is it that makes the Spanish data somehow special, in that proto-articles appear much sooner than in all other sets of data analyzed so far? A cursory comparison with our own data on the acquisition of German revealed that prenominai filler syllables began to appear later in

Proto-articles in the Acquisition of Spanish

191

German: not before 1;8 or even 1;10 for some children. In any case, no German children produced fillers at 1;4 or 1;5, the time when Spanish children began to produce them.6 The appearance of the article in Spanish is phonologically favored by the tendency of this language to long words. Most Spanish target words are disyllables, but there are many target trisyllables from the very beginning. In order to see if this might constitute a decisive prosodie factor, sorting out Spanish from other often reported languages, including German, target trisyllables appearing at the 25-word point where listed and compared to equivalent words, also produced by the German children. The complete list of Spanish trisyllables and a few quadrisyllables, with the German equivalent, and English translation appears in (15): abuela Antonio amarillo babero bicicleta caballo camello cerdito chupete cocodrilo conejo galleta hermana mariposa pajarito patito pelota pollito rastrillo tijeras trompeta zapato

Orna (proper name) gelb Lätzchen Fahrrad Pferd Kamel Schwein(chen) Schnuller Krokodil Hase Kekse Schwester Schmetterling Vogel Ente Ball Küken Rechen Schere Trompete Schuh

'grandma' 'yellow' 'bib' 'bicycle' 'horse' 'camel' 'pig(gy)' 'pacifier' 'crocodile' 'rabbit' 'biscuit' 'sister' 'butterfly' 'bird' 'duck' 'ball' 'chicken' 'rake' 'scissors' 'trumpet' 'shoe'

With the exception of cocodrilo, mariposa and trompeta, which are trisyllabic in German, all Spanish multisyllables correspond to monosyllables and disyllables in German. In order to be able to assert the ratio of trisyllables in the early Spanish lexicon, and to quantify the difference in number of syllables between early Spanish and German words, all words produced by three Spanish children and by all five German children of the project, at the 25-word session, were counted for number of syllables in the target form and in the child's productions. Table 2 shows the percentages of monosyllables, disyllables and trisyllables (plus a few quadrisyllables) ac-

6

Peters and Strömqvist (1996:228) mention similar cases for German in the data of Stern & Stern, when the child, Günther, was already 2;5. My analysis of the German data is presented in Lieo (1996).

Conxita Lieo

192

cording to the target form and the child's productions (the numbers in parentheses refer to number of tokens):

José Maria Miguel Total

Target forms 1-syl 2-syl % Ν % Ν (49) 30 (36) 40 (28) 11 ( 5) 61 2 ( i) 71 (42) 18 (42) 53 (119)

Bernd Britta Marion Johannes Thomas Total

22 ( 9 ) 40 (27) 24 (26) 25 (23) 41 (31) 30 (116)

73 59 72 71 58 67

(30) (40) (78) (65) (44) (257)

multis. % Ν 30 (36) 28 (13) 27 (16) 29 (65)

Child productions 1-syl 2-syl Ν Ν % % 34 (41) 40 (48) 15 ( 7) 67 (31) 5 ( 3) 75 (44) 23 (51) 54 (123)

5 1 4 3 1 3

13 ( 6 ) 43 (29) 31 (34) 26 (24) 50 (38) 33 (131)

( 2) ( O ( 4) ( 3) ( 0 01)

81 (39) 57 (39) 68 (73) 73 (66) 47 (36) 65 (253)

3-syl % Ν 26 (32) 17 ( 8) 20 (12) 23 (52) 6 0 1 1 3 2

(3) (-) ( i) ( i) (2) (7)

Table 2. Percentages of monosyllables, disyllables, trisyllables and quadrisyllables produced by Spanish and German children at the 25-word point, according to target and produced forms. All data refer to tokens. Obviously, multisyllables are more numerous in Spanish than in German, representing 29% of all target forms, whereas in German they only amount to 3%. Multisyllables and especially trisyllables have probably contributed to the early development of a pretonic syllable, in the following sense. As has been proposed for the acquisition of English and Dutch (Fikkert 1994, Demuth & Fee 1995, Demuth 1996), the first prosodie units produced are binary, which in the acquisition of Spanish implies two-syllabic units. Monosyllables cannot be prosodically analyzed as binary, since Spanish is not quantity-sensitive,7 and closed syllables are acquired late in Spanish, anyhow. This means that monosyllables would be soon avoided in the acquisition of Spanish. Actually, besides particles like si and no, which constitute utterances in themselves and are thus excluded from the prosodie structure that will give rise to full sentences, monosyllabic nouns are soon provided with a filler. For some children it is also the case that if the monosyllable is a verb, as in ven, it is often produced as a disyllable. That is, filler syllables preceding monosyllables might exclusively respond to the prosodie constraint requiring bisyllabicity. Disyllabic nouns do not violate this constraint, and nevertheless they get fillers very soon. According to current metrical theories (see especially Hayes 1995), feet are binary, which leads the child to an initial truncation of the initial syllable of trisyllabic nouns. But the Spanish child is used to trisyllables and soon learns that there is a mismatch between the Foot and the Prosodie Word. Feet are disyllabic, often trochaic, but Prosodie Words are often trisyllabic, so that 7 Or it is so only marginally. See Harris (1983).

Proto-articles in the Acquisition of Spanish

193

parsing Prosodie Words implies adjoining one more syllable to the left of the metrical Foot, as in the following diagram: (16) Example of Foot and Word construction Word Layer Foot Layer Syllable

(

χ ) (x .) σ σ σ ?a ßej joh 'babero' (María 1;4,21)

The syllable to the left may be considered as an extrametrical syllable that is adjoined to the Word level by Stray Adjunction, or by some other mechanism.8 Independently of the formal description that one gives to this prosodie phenomenon, the crucial point is that the Spanish child becomes soon sensitive to adding a weak syllable to the left of a disyllabic Foot in order to construct words. This makes him/her sensitive to the determiner preceding disyllabic nouns, and allows him/her to convert disyllabic nouns to trisyllables by adding an initial syllable, first undifferentiated, which will a few months later end up as a proto-article and finally as a standard article. On the other hand, trisyllables are analyzed by the child in this way, too. Some of the trisyllables are reduced to disyllables (as in muñeca -> keka), because they are fitted into the disyllabic foot and then they follow the fate of disyllables, in that an extra syllable is adjoined to the left in the construction of the word. This extra syllable is segmentally reduced into a "simpler" syllable, often the default syllable, equivalent segmentally to the proto-article. When constructing Prosodie Words, the child does a scanning of the pattern from right to left, so that the last position to the left is just filled with a default syllable. As mentioned above, the prosodie model adopted here is similar to that of Demuth's (1996), with a stage of minimal words, represented by Binary Feet, which in Spanish correspond to two-syllable Feet. But our Spanish data do not confirm the sequences of stages proposed by Demuth (1996). After the Binary Feet stage, Demuth proposes two more stages in the following sequence: Stage III.

Stress-Feet a. One Stress-Foot per word b. Two Feet per word Stage IV. Phonological Words Extrametrical syllables permitted In our data, stages Illb and IV are reversed, extrametrical syllables being introduced before two feet per (prosodie) word are allowed. As we saw above, in the Spanish data two-feet structures within a Prosodie Word, i.e. either within a lexical word or a phrase, are produced 8

Hayes (1995:1090) partially rejects extrametricality plus stray adjunction in favor of degenerate feet. Nevertheless, the issue is not settled and I leave it open what the most optimal mechanism to deal with these unstressed syllables of Spanish will be.

194

Conxita Lleó

later than extrametrical syllables. Two-feet structures are produced at the two-word period, much later than the extrametrical syllables corresponding to the proto-article or to the first syllable of trisyllables. It is not clear at this point whether the reversal of stages found in the Spanish data has to be interpreted as a language-specific phenomenon, supported by the specific prosodie characteristics of Spanish. A plausible alternative to be investigated is whether the purported universal precedence of stage Illb might be an artefact due to data collection and data analyses, in the sense suggested in §4.1. A certain lack of interest on the oneword stage due to syntax-centered analyses, might have obscured the fact that some functional words are already present before stage Illb. Furthermore, the first two-word utterances are produced with almost no filler syllables or functional words. It might be easy to miss the chronological precedence of these proto-functors, when looking at the two-word period, once functional words have been integrated there.

5. Conclusion A fine-grained analysis of early noun production reveals the presence of proto-articles in Spanish at a very early age, much earlier than it has ever been reported for any language up to date. Proto-articles appear at the one-word stage, so that they are evidence for the universal availability of functional heads at the threshold of language acquisition. The development of proto-articles from undifferentiated default syllables into a masculine and a feminine category offers evidence for the availability of the category 'gender' at the earliest stages, and for an early differentiation of fillers into masculine and feminine articles, on the one side, and into definite and indefinite articles, on the other. Functional categories are thus available, but they are manifested in the linguistic productions of young children, only if the prosodie structure of the target language stimulates their appearance. The early Spanish lexicon of the children examined here has many trisyllables, which forces the child to enlarge the basic disyllabic feet when constructing prosodie words by one more syllable to the left. This makes the child sensitive to the presence of functional words earlier than in the acquisition of languages with different prosodie conditionings.

References Demuth, Katherine (1992) Accessing functional categories in Sesoto. In Jürgen Meisel (ed.): The Acquisition of Verb Placement. Functional Categories and V2 Phenomena in Language Acquisition. Dordrecht, Boston, London: Kluwer. - (1996) Stages in the development of prosodie words. In Eve Claik (ed.): Proceedings from the 27th Child Language Research Forum. Stanford University: CSLI. Demuth, Katherine & Jane E. Fee (1995) Minimal words in early phonological development. Ms. Fikkert, Paula (1994) On the Acquisition of Prosodie Structure. Holland Institute of Generative Linguistics.

Proto-articles in the Acquisition of Spanish

195

Gerken, LouAnn & Bonnie J. Mcintosh (1993) Interplay of function morphemes and prosody in early language. Developmental Psychology 29, 448-457. Golston, Chris (1995) Syntax outranks phonology: evidence from Ancient Greek. Phonology 12, 343-368. Harris, James W. (1991) The exponence of gender in Spanish. Linguistic Inquiry 22, 27-62. Hawayek, Antoinette (1995). Acquisition of functional categories and syntactic structure. Probus 7, 147-165. Hayes, Bruce (1995) Metrical Stress Theory: Principles and Case Studies. Chicago and London: University of Chicago Press. Hyams, Nina M. (1986) Language acquisition and the theory of parameters. Dordrecht: Reidel. Lieo, Conxita (1996) To spread or not to spread: different styles in the acquisition of Spanish phonology. In Barbara Bernhardt, John Gilbert & David Ingram (eds.): Proceedings of the UBC International Conference on Phonological Acquisition, 215-228. Somerville: Cascadilla Press. - (1997) At the interface of phonology and morphology: the emergence of the article in the early acquisition of Spanish and German. Paper presented at TROPICS, Berlin 26-29, 1996. López Ornat, Susana (1994) La adquisición de ¡a lengua española. Madrid: Siglo XXI. Peters, Ann M. & Lise Menn (1993) False starts and filler syllables: ways to learn grammatical morphemes. Language 69, 742-777. Peters, Ann M. & Sven Strömqvist (1996) The role of prosody in the acquisition of grammatical morphemes. In James L. Morgan & Katherine Demuth (eds.): Sygnal to Syntax: Bootstrapping from speech to grammar in early acquisition, 215-232. Hillsdale, NJ: Lawrence Erlbaum. Pizzuto, Elena & Maria Cristina Caselli (1992) The acquisition of Italian morphology: implications for models of language development. Journal of Child Language 19, 491-557.

Dunstan Brown (University of Surrey)

Stem Indexing and Morphonological Selection in the Russian Verb: a Network Morphology Account 1

1. Introduction According to AronofF (1994:25) there is a level of pure morphological functions, the morphomic level, which is itself neither morphosyntactic nor morphonological. These functions are selected to give the required forms for a given morphosyntactic array. An example of this is the perfect participle in English, used both for the passive and the perfect, and where the featural content that the two constructions have in common is difficult to characterise. The use of morphological functions allows for the fact that these structures are syntactically dissimilar and yet always make use of the same formal element (Aronoff 1994: 24). As far back as the work of Trubetzkoy (1931:163) morphonology is viewed as a link between phonology and morphology. Its regularities of form are not entirely reducible to rules of phonology. One area that was not addressed was whether it was possible to generalise over the type of (morphosyntactic) category for a given alternation, although Trubetzkoy (1931:163) notes the difference in types of alternation between the verbal and nominal system of Russian. Aronoffs work deals primarily with affixation (Latin) and Semitic intercalation (the binyanim). In this paper I propose to consider the Russian verbal system in the light of Aronoffs work, because it serves to clarify the distinction between stems of 'a constant sound form and those characterised by their place in the morphological paradigm' (Aronoff 1994:59). Sadler,

1

This paper is the result of ongoing work on the Russian verbal system. It is based on earlier research presented at the Linguistics Association of Great Britain Autumn Meeting, Middlesex University, 13-15 September 1994, and the 'Linguistics at the End of the Twentieth Century' Conference, Moscow State University, 1-4 February 1995. A general outline of this approach to verbal morphology has appeared in Russian in Brown (1995). The work presented here contains different assumptions about jot-insertion and prefixation from that publication. The author should be contacted for copies of the DATR fragment on which this paper is based. I am grateful to a number of people who have helped with comments and discussion during the period I have been working on Russian verbal morphology, among them Greville Corbett, Ray Fabri, Norman Fraser, Andrew Hippisley, Rosemary Leonard, Tore Nesset, Albert Ortmann, Louisa Sadler, Andrew Spencer, Greg Stump, Alan Timberlake, Tanja Volz, Dieter Wunderlich and Ilse Zimmermann. Further thanks should go to the participants at the ESRC funded 'Frontiers of Research in Morphology' seminar series who have also influenced my thinking on this matter. Any errors or mistakes are mine. The research presented here has been supported in part by the Economic and Social Research Council under grants R000233633 and R000236063 and by the British Council under the Academic Research Collaboration Programme, which has enabled co-operation with Dieter Wunderlich's research group in Düsseldorf. I am grateful to both funding bodies for their generous support.

Stem Indexing in the Russian Verb: a Network Morphology Account

197

Spencer and Zaretskaya (1996) have already presented an account of deverbal nominalization in Russian which makes use of the Aronovian concept of indexed stems. The purpose of this paper is to clarify how indexing and morphonological selection are related. Russian is interesting because the realization of stem types changes as we move down a hierarchy. Initially, the requisite stems are created by affixation, but in the more marked cases, lower in the hierarchy, consonantal alternations are to be found. However, whether affixed or mutated, these stems have the same role in inflection.

2. Russian Verbs The Russian verbal system is interesting, because the morphology may involve reference to segmental replacement. Furthermore, the rules which refer to these segmental alternations could be considered 'morphomic' in that they are functions which create stems, but are not themselves the same as the consonantal alternations. The consonantal alternations of the Russian verbal system are not purely phonological, as they must involve reference to morphosyntactic context. In fact, it appears that four things are required: to describe the morphonological alternations; to describe different stems which select the morphonological alternations; mappings between the morphonological alternations and the stems which select them; selection of the appropriate stem for the appropriate inflectional (morphosyntactic) context. Spencer (1988) deals in detail with morphonological selection in terms of morpholexical rules, and argues that morpholexical rules are required to deal with selection of the appropriate stem and ending for some Czech nouns. With Russian verbs we find that the appropriate form of the stem and ending depends on membership of conjugation. Furthermore, stems are indexed to occur in particular inflectional slots, but may differ in the form for a particular index. 2.1 The data Russian has two conjugations, which appear to differ only in terms of the 'thematic ligature' which occurs in the present tense. We take the term thematic ligature from Timberlake (1993:850). The thematic ligature of the first conjugation is loi when it occurs under stress and tends to be schwa when not under stress.2 The thematic ligature of the second conjugation is Ν

2

This is an oversimplification. For a detailed criticism of an approach which transcribes the graphemes and of the first conjugation as loi see Kempgen (1989: 146). His basic argument is that there would be an unstressed allophone [ι] for the phoneme loi which occurs nowhere else but in veibal endings. A similar point is made by Pirogova (1969: 26). The examples in question involve unstressed vowels in post-tonic position after /j/ or a palatalised consonant. It is not the purpose of this paper to deal with this question. However, the solution that would fit in with our current analysis is to assume that the evaluation which de-

Dunstan Brown

198

and always occurs after a consonant with palatalisation ('softening') or the reflexes of historical palatalisation, where there is a change in the primary place of articulation. Table 1 gives example verbs in the present and past tenses, together with the infinitive, from the first and second conjugation. 3 Conjugation

First Conjugation

Second Conjugation

Infinitive Present sgl" sg 2nd sg 3rd pi 1st pi 2nd pi 3rd

dela-t'

govor'i-t'

dela-j-u dela-j-o-s dela-j-o-t dela-j-o-m dela-j-o-te dela-j-ut

'to do'

'to speak'

govor'-u govor'-i-s govor'-i-t govor'-i-m govor'-i-te govor'-at

Past sg mase sg fem sg neuter

govor'i-1 dela-1 dela-l-a govor'i-l-a govor'i-l-o dela-l-o 1 govor'i-l'-i dela-l'-i P Table 1: First and Second Conjugation In table 1 it can be seen that the only difference between the first and second conjugation in terms of endings is in the thematic ligature and the third person plural. Another possible contrast is to be found in the infinitive, where for the first conjugation the vowel before the ending -t ' is not the same as the thematic ligature, whereas it appears that for the second conjugation verb govor it''to speak' it is. However, there are other second conjugation verbs for which this is not true, such as smotret' 'to look (at)', which has the present smotr'u Ί look', smotr'is 'you look', smotr'it 'he looks' etc. In fact, Timberlake (1993:850) refers to the formative before the infinitive ending as a classificatory suffix. The important point to note is that it is not the same as the thematic ligature, or theme vowel, which occurs as a conjugation marker in the present tense. We must therefore distinguish the theme vowel, which indicates conjugation in the present tense, from the classificatory suffix that occurs in the allostem used for the infinitive and past tense. This suffix is generally a vowel, but need not be.

3

termines whether to insert a Ijl on the basis of information about the final element of the stem would also determine whether the stem-final element were palatalised and insert /e/ rather than loi. Nesset (In draft) discusses the phonological issues in detail, and an example can be found in Avanesov (1952: 151-152) in the transcription of the PuSkin poem Zimnee Utro Ά winter's morning'. An acute ' indicates that the preceding consonant is 'soft'. Aspect and participle formation is not dealt with here.

Stem Indexing in the Russian Verb: a Network Morphology Account

199

It therefore follows that there are two main variants of the stem, or two allostems. These are the stem used in the infinitive and the past tense, and the stem used in the present tense. Consideration of the data in table 1 might lead us to say that the infinitive and past stem of the first conjugation example is not identical with the stem of the present tense, because the present tense is augmented by the palatal glide /j/. However, the glide can be accounted for by stating that it is inserted between two vowels which are not root-internal. Whenever a classificatory suffix which is a vowel occurs next to the thematic ligature, also a vowel, the glide /j/ will be inserted. In this respect we follow the approach of Garde (1972). This means that there is a generalisation that states that by default the infinitive and past stem is the same as the present stem. A further generalisation states that two vowels which are on the periphery or outside the stem will have the glide /j/ inserted between them. This differs from an earlier approach by Brown (1995), which assumed that the glide was an integral part of the present stem. Furthermore, it contrasts with the Jakobsonian (1948) approach which treats the verbal forms as being derivable from one stem, and for which the glide /j/ is treated as underlying and dropped before the consonantal desinences of the past and infinitive. For the second conjugation it appears to be the case that the present stem is govor'-, and that the stem for the infinitive and past is govor '/'-. In addition to the examples given in table 1 there are other verbs which show even more radical differences. The verb ugr'umet' 'to become morose' behaves in the same way as delat ' except that the vowel Id stands in for /a/. There are other verbs which are augmented in different ways, as in those of the trebovat' kind given in table 2 below, together with the classification of the verbal types of Zaliznjak (1977) and Isacenko (1960) for comparison. Table 2 illustrates an additional difference between the present and the past/infinitive where the root is augmented. Verbs like trebovat' and ugr'umet' are still similar to those of the general delat' type in that their infinitive stem ends in a vowel. Furthermore, verbs like ugr'umet' follow the default generalisation that the present stem is the same as the infinitive. The -ova portion of trebovat' consists of -ov and the default vowel for infinitive and past stems /a/. This is supported by the fact that other verbs of this type, such as zabastovat' 'to strike', have nominal formations like zabastov-ka 'a strike', where the vowel is no longer to be found, but the formative -ov still is.

Dunstan Brown

200

Base Type

BASE3 Stem (Zaliznjak's Class 2) (Isacenko's Class ΙΠ)

Infinitive

treb-ov-a-t'

Present sg 1st sg 2nd sg 3rd pi 1st pi 2nd pi 3rd

treb-u-j-u treb-u-j-o-s treb-u-j-o-t treb-u-j-o-m treb-u-j-o-te treb-u-j-ut

Past sg mase sg fem sg neuter Pi

treb-ov-a-1 treb-ov-a-l-a treb-ov-a-l-o treb-ov-a-l'I

'to demand'

Table 2: The verb trebovat' 'to demand' As the analysis we present is couched within a declarative framework, we do not recognise the need for truncation (cf. Aronoff (1976:40 and 88-101) and Darden (1988)) or deletion, which goes hand in hand with Jakobson's (1948) single stem approach, and that of Halle (1963). It will be seen from the approach we adopt to stem formation in the verb that indexed stems can be made available for different purposes. The difference between a declarative approach which uses indexes and the derivational one, which allows truncation, is that the derivational one assumes that there is asymmetricality in stem-formation processes which involve truncation. The index-based approach claims that the relation is symmetrical. In Network Morphology, referrals (Zwicky 1985) are used to capture asymmetrical relations between compatible feature structures. However, a strong claim is made that stem indexing is required where feature contents may be incompatible and that this incompatibility will be found where other analyses have chosen to make use of truncation. Past and infinitive stems cannot be treated as having compatible features, because 'past' is a feature of the category TENSE, whereas 'infinitive' is not. Andersen (1980:299) argues on the basis of analysis of data from Gvozdev's (1949) child language acquisition study and historical research that segment addition is preferred to segment deletion. In our analysis a truncated stem is an entity parallel with that of another stem (the one from which it would be derived by truncation in procedural frameworks), but which merely shares some, rather than all, of the material of that stem.

Stern Indexing in the Russian Verb: a Network Morphology Account

201

2.2 Consonantal Alternations As indicated, the most general type of verb belonging to the first conjugation can be characterised in a relatively straightforward manner. The stem is the same for both the infinitive and past on the one hand and the present tense on the other. Where the classificatory suffix is a vowel, /j/ is inserted between it and the theme vowel in the present tense. For the most general type of verb belonging to the second conjugation it would appear that the theme vowel and classificatory suffix /i/, as a high front vowel, brings about palatalisation of the preceding consonant. The problem with this assumption is that the sounds [i] and [i] are in complementary distribution after soft and hard consonants respectively. Furthermore, /i/ is backed after hard consonants up to the word domain (Hamilton 1980:71). Rather than the vowel /i/ affecting the preceding consonant, it is affected by it. Palatalised consonants have to be recognised as phonemes, because they occur word-finally and therefore cannot be accounted for in terms of palatalisation by a following vowel. We shall propose an analysis which treats the palatalisation of the consonants as a question of morphonological selection. In the present tense of the first conjugation some verbs have a consonant appearing before the thematic ligature. These consonants show varying degrees of 'palatalisation'. More examples are given in table 3, labelled according to the node names given them in our representation, together with the classification of the verbal types of Zaliznjak (1977) and Isacenko (1960) for comparison.

Base Stem Type

OBASE_5 (Zaliznjak's Class 6_) (Isacenko's Class IV)

OBASE_5a (Zaliznjak's Class 3) (Isacenko's Class VI)

Infinitive

sos-a-t' 'to suck'

tolk-n-u-t' 'to push'

Present sg l rt sg 2nd -)rd sg 3 pll" pi 2nd pi 3"1

sos-u sos'-o-s sos'-o-t sos'-o-m sos'-o-te sos-ut

tolk-n-u tolk-n'-o-s tolk-n'-o-t tolk-n'-o-m tolk-n'-o-te tolk-n-ut

Past sg mase sg fem sg neuter Pi

sos-a-1 sos-a-l-a sos-a-l-o sos-a-l'-i

tolk-n-u-1 tolk-n-u-l-a tolk-n-u-l-o tolk-n-u-l'i

Dunstan Brown

202 OBASE6 (Zaliznjak's Class 6) (Isacenko's Class VI)

Base Stem Type

Infinitive

plak-a-t' 'to cry'

pl'as-a-t'

Present sg 1st sg 2nd sg 3rd pi 1st pi 2nd pi 3rd

plac'-u plac'-o-s plac'-o-t plac'-o-m plac'-o-te plac'-ut

pl'as-u pl'as-o-s pl'as-o-t pl'as-o-m pl'as-o-te pl'as-ut

Past sg mase plak-a-1 sg fem plak-a-l-a sg neuter plak-a-l-o plak-a-1 i Pi Table 3: Base Stem types of First Conjugation Verbs

'to dance'

pl'as-a-1 pl'as-a-l-a pl'as-a-l-o pl'as-a-l'i

In addition to these alternations there are also a number of other verbs which show varying degrees of idiosyncrasy as regards such phenomena as vowel alternation (ablaut), including the so-called fleeting vowels. It is important for our purposes to note here that the verbs pl'asat' 'to dance' and sosat' 'to suck' both have Isl in the infinitive, yet they show different alternations in the present tense. Further, we see that the verbs tolknut' 'to push' and sosat' have the unpalatalised stem in the first person singular and third person plural of the present tense. However, the stem of the third person plural must be the same as the first person singular, if it differs from the rest of the present tense. This means that we only need accept an additional allostem for the present tense. Generally, of course, the stem is the same throughout the present. In our analysis this allostem is a subtype of the present allostem. Further to the alternations in the first conjugation we note additional ones in the second conjugation. Unlike the more general type of verb in the first conjugation, it is always the case for the second conjugation that the verbs in the present tense have a consonant other than jot before the thematic ligature. They have this in common with the less general stems of the first conjugation. In addition to the apparently straightforward example oígovor it', we also have the alternations in table 4, again given according to the node labels which appear in the hierarchy in figure 4 in section 4 below. The classifications of Zaliznjak (1977) and Isacenko (1960) are also included for comparison.

Stem Indexing in the Russian Verb: a Network Morphology Account

203

I_BASEi (Zaliznjak's Class 4) (Isacenko's Class V)

Base Stem Type

Infinitive

pros'-i-t'

Present sg 1st sg 2nd sg 3rd pi 1st pi 2nd pi 3rd

pros-u pros'-i-s pros'-i-t pros'-i-m pros'-i-te pros'-at

l'ubl'-u l'ub'-i-s l'ub'-i-t l'ub'-i-m l'ub'-i-te l'ub'-at

pros'-i-l pros'-i-l-a pros'-i-l-o pros'-i-l'-i I_BASEja (Zaliznjak's Class 5) (Isacenko's Class VII)

l'ub'-i-l l'ub'-i-l-a l'ub'-i-l-o l'ub'-i-l'-i I_BASEa (Zaliznjak's Class 5) (Isacenko's Class VII)

Infinitive

kr'ic'-a-t'

sp-a-t'

Present sg 1st sg 2nd sg 3rd pi 1st pi 2nd pi 3rd

kr'ic'-u kr'ic'-i-s kr'ic'-i-t kr'ic'-i-m kr'ic'-i-te kr'ic'-at

Past sg mase sg fern sg neuter Pi Base Stem Type

Past sg mase sg fern sg neuter

'to ask'

'to shout'

kr'ic'-a-l kr'ic'-a-l-a kr'ic'-a-l-o kr'ic'-a-l'-i pi Table 4: Base Stem types of Second Conjugation Verbs

l'ub'-i-t' 'to love'

'to sleep'

spl'-u sp'-i-s sp'-i-t sp'-i-m sp'-i-te sp'-at

sp-a-1 sp-a-l-a sp-a-l-o sp-a-l'-i

For the verb kr'ic'at' 'to shout' it appears that the root has Ici as its final consonant, but it is paired with the (semelfactive) perfective kr 'iknut' 'to shout' and is obviously related to the noun kr ik 'cry'.

Dunstan Brown

204

The apparently bewildering array of alternations in table 4 is actually a lot less problematic than it seems. In table 5 each final element of a stem can be seen as having a 'zero', 'soft' or 'jotated' grade. Zero Grade

Soft Grade /ρ'/ ña'l Imi /Γ/ Iv'l Iii Id'l Is'/ ΙζΊ lì Ί Ini ΙτΊ Ici /ζ/ Iii Table 5: The three grades of root/base final elements /p/ Ibi Imi Iti M Iti Idi Isl IzJ Ν /η/ Irl /k/ /g/ /χ/

Jotated Grade /p'/ + IVI Ib'l + iïl Imi + ΙίΊ /f / + /l'I IVI + /IV Ici Iii Isl ΙζΙ IVI Ini ΙτΊ Ici m Isl

Note that for the velars and IV, /n/ and Irl the soft grade is the same as the jotated grade. For the velars softening involves both a change in place and manner of articulation. For all the other consonants it can be seen that the soft grade is the zero grade palatalised.4 If we now look back over conjugation II, we see that the alternations are very easily characterisable. For all but the least general class IBASEa, no matter what the vowel (classificatory suffix) in the infinitive, the consonant before it will belong to the soft grade. The consonant before the ending of the first person singular belongs to the jotated grade, and the consonant before the theme vowel or the third person plural ending belongs to the soft grade. As the stem of the past tense behaves in the same way as the infinitive the consonant before the vowel (classificatory suffix) belongs to the soft grade, just as the infinitive. In the first conjugation, in those cases where the thematic ligature is preceded by a consonant, we have a limited number of possibilities in the present tense: (i) the consonant before all endings (including thematic ligature) belongs to the jotated grade; (ii) the consonant before the first person singular belongs to the zero grade, and the consonant before the thematic ligature belongs to the soft grade, with the consonant before the third person plural ending behaving the same as the first person singular.

4

In fact, the soft grade gains a secondary palatalised articulation and loses a velarized one, as hard consonants are either velar or velarized (Jakobson 1948).

Stem Indexing in the Russian Verb: a Network Morphology Account

205

With type (i) the vowel classificatory suffix remains in the infinitive, as with the more general class of verbs. For type (ii) the vowel may be found in the infinitive, as for sosa-t ' 'to suck' and kribiu-t' 'to shout', or the infinitive ending may be preceded by the consonant of the root, such as with nes-t 'i 'to carry' or pas-t ' 'to fall', where the difference in infinitive ending, ' or t'i, is related to stress. Less general subtypes of (ii) may have the vowel of the root before the infinitive ending, such as the verb pec ' 'to bake', where -c ' is the infinitive ending for verbs which have a velar elsewhere in the paradigm. Other less general classes are outlined in Timberlake (1993 :852-3). For the less general verbal types, therefore, the stem of the present has to be distinguished from the stem of the past and infinitive. In sum, for the most general class of verbs, the present stem is identical with the infinitive and past stem, meaning that there is only one stem. Next, the present stem may be different from the infinitive/past stem, giving two stems (as illustrated by the O BASE 6 type in table 3). Verb stems such as I BASEi, IBASEja and I BASEa (table 4) require three stems: one for the infinitive and past, one for the present, one for the first person singular of the present. Verbs of the O B A S E 5 and 0_BASE_5a types also require three stems: one for the infinitive and past, one for the present, one for the first person singular and third person plural. Smaller groups which we treat as subtypes of the O BASE 5 type may also distinguish past from infinitive. A maximum of four stems is required, as indicated by figure 1.

Present Stem



Infinitive/Past Stem Figure 1: The two major stems

(First Sg or Third PI)

(Past)

In figure 1 the two major stems are labelled 'stem 1' and 'stem 2' respectively. Further splitting these, we have treated the stem found in the first person singular and third person plural of the present tense as a sub-indexed variant (where 'a' is the sub-index). Although we allow for four stems, the analysis which we present shows that we do not always need to make explicit statements about each of them. In this paper we shall show that there is a default statement that says that the 'stem 2' is the same as 'stem 1'. Further, in the absence of any other information, the shape of 'stem 2 a' can be inferred from the shape of 'stem 2', and the shape of 'stem 1 b' can be inferred from the shape of'stem 1'. The choice of the subindexes 'a' and 'b' for variants of'stem 1' and 'stem 2' is arbitrary, as is the choice of numbers for the two main stem types. The next section briefly introduces the Network Morphology framework so that we can go on to show how an analysis using morphomic indexing and morphonological selection might be treated within that framework.

206

Dunstan Brown

3. Network Morphology Network Morphology is a theoretical framework that is being developed as a tool for formalising typological generalisations about the morphology of natural language. It consists of a number of general principles for representing linguistic knowledge, plus a number of more specific theoretical principles which are intended to limit the possible types of structures which can be represented formally, thereby making more specific claims about the nature of linguistic morphology. Foundational work in Network Morphology is found in Corbett and Fraser (1993), where the benefits of a default inheritance account of the category of animacy and morphosyntactic mismatches are detailed. Brown and Hippisley (1994) detail the relation between stress and inflection class in determining the realization of the genitive plural in Russian. Fraser and Corbett (1995) account for complex interdependencies of semantics and phonology, among other things, in the assignment of gender and inflectional class in Russian. The work of Brown, Corbett, Fraser, Hippisley and Timberlake (1996) shows how the complexities of Russian nominal stress can be accounted for by assuming a Network Morphology approach to morphology as a network of parallel hierarchies. As indicated in Brown et al (1996), Network Morphology treats language as a network of linguistic information. The basic building blocks of the framework are nodes and connections between nodes. Information is stored at the nodes, and the connectivity of the network determines which information is available at which node. This is not a new view of language. Flickinger (1987), for example, treats lexical knowledge in terms of inheritance hierarchies in which information may be passed down from one node to another. Such an inheritance-based approach is used in Word Grammar (Hudson 1990; Fraser and Hudson 1992), and in a number of recent natural language processing computer systems (summarised in Daelemans et al. 1992). Furthermore, this approach can also be compared with the semantic nets approach for representing real world knowledge, which goes back at least as far as Quillian (1968) and is discussed in detail in Fahlman (1979). 3.1 Network Morphology Entities In this section we outline the theoretical constructs made use of by Network Morphology to model morphology. Illustrations of these and principles governing the representation of morphological knowledge are given using the DATR formalism developed by Evans and Gazdar (Evans and Gazdar 1989a, 1989b, 1996; Keller 1995).5 The formalism was developed for lexi-

5

This is not the only work to represent an analysis of Russian verbs using DATR. Volz (1994) uses DATR to represent her classification of Russian verbs.

Stem Indexing in the Russian Verb: a Network Morphology Account

207

cal knowledge representation and is based on default inheritance. It makes use of ordered attribute structures to represent knowledge, and this makes it well suited for encoding Network Morphology theories. Network Morphology theories contain facts which, in keeping with its declarative orientation, are statements about the properties of a particular language. In Network Morphology facts are given as path: value pairs. (1) is a fact about Russian nouns. (1)

"" _i

The nominative plural consists of the plural stem and the ending -/'. This fact consists of a path to the left of the == operator and a complex value to the right of the == operator. This complex value consists of a path "" and the symbol combination _i. This fact is a parochial statement about the morphology of Russian nouns. If we apply it to a Russian noun, such as the lexeme ZAKON 'law', the fragment will predict that it has the plural nominative form zakon-i (given in a phonological transcription). The fact that we have used to illustrate is specific to a particular word class, namely nouns. However, facts differ in how general they are. Other facts may hold for nomináis (adjectives, nouns and pronouns) or words. In order to capture this we need to make additional statements about the class of words to which particular facts apply. In order to do this the concept of node is made use of. While facts are statements about types of words, nodes are locations in a hierarchy of information where facts are to be found. Nodes represent classes of items which can be considered to have a number of properties or facts in common. An indefinite number o f f a c t s may be found at a particular node. Extending our example further, we name the node at which facts about the morphology of Russian nouns is found MORNOUN. We can then modify the equation in (1) to (2). (2)

MOR_NOUN :

== ""

_i

Now it is clear that this is not a fact about just any Russian word, but about the class of nouns. In order to account for other classes of word we must add further nodes to our hierarchy. We add a node for nomináis (adjectives, nouns and pronouns) from which the node MOR NOUN can inherit facts which nouns, adjectives and pronouns have in common (for instance, the instrumental, dative and locative plural, as described in Corbett and Fraser (1993)). MOR NOUN is in a hierarchy relation with the node MORNOMINAL, which means that MOR NOUN will inherit all other facts not declared at MOR NOUN from MOR NOMINAL. We represent hierarchy relations with the equation = = NODE, where NODE is the name of the node which is the source of inheritance. The equations for nouns in (2) can now be added to so as to give (3).

Dunstan Brown

208 (3)

MOR_NOUN : == MOR_NOMINAL

== "" _i MOR NOMINAL could be understood as the mother of MOR NOUN. Additional nodes can be added which make more general statements about words, or which apply to sub-groupings within the class of nouns. These sub-groupings can be understood in terms of the traditional concept of inflection classes. It can also be seen that this method can be applied to any class of words. We may represent the hierarchy relations which we have given as the equation == node graphically as a straight line between a higher and a lower node, as indicated in figure 2.

MOR_WORD == MOR WORD MOR NOMINAL MOR_NOMINAL

MOR ADJ

MOR NOUN

Figure 2: Hierarchy relations Note that in figure 2 the graphical representation indicates what the hierarchy relations are between nodes. It does not tell us which facts are found at those nodes, although we know that facts found at a higher node will be inherited by a lower node unless overridden. Only hierarchy relations are rendered as lines in diagrams such as figure 2. Where a set of nodes is connected by hierarchy relations, such as figure 2, those nodes constitute a hierarchy. We stated earlier that facts consist of path: value pairs. A path is an attribute structure which may consist of zero or more attributes. In the formal notation that we are using here paths are enclosed in angle brackets. Attributes come in a number of different flavours, of which there are at least three types, hierarchy identifiers, morphosyntactic features and indexes. Hierarchy identifiers Network Morphology allows for different levels of linguistic structure, such as semantics, syntax, and most importantly for our purposes, morphology. These levels of linguistic structure are modelled by the hierarchies mentioned earlier. Hierarchy identifiers indicate the level of linguistic structure modelled by a particular hierarchy. Morphosyntactic Features These correspond to the 'properties' of Matthews (1972:162), such as 'accusative', which belong to particular 'categories', such as CASE.

Stem Indexing in the Russian Verb: a Network Morphology

Account

209

Indexes These are arbitrary symbols, or symbol combinations, used to address other nodes in the network or to label stems which occur in morphosyntactic contexts which do not share similar morphosyntactic features. They can be understood as indexes in the spirit of Aronoff(1994). The ordering of attributes is important. Loosely speaking, moving from left to right along a path involves moving from more general to more specific attributes. As an example of how to interpret a path we take the example of the nominative plural equation from earlier. (4)

== "" _i

The path starts with the hierarchy identifier mor which gives the most general information that we are dealing with the level of morphology. Next the morphosyntactic feature p i specifies further that we are dealing with the morphology of plurals. Finally, the morphosyntactic feature nom further specifies the nominative case in the plural. This is how we account for the sub-types of the verbal stems mentioned in the previous section. As is less specific than , the value paired with is inferred, if there is no explicit information about . The principle is the same as with (4), only here we are dealing with indexes, rather than morphosyntactic features. Values, as we have indicated, may be complex or simple. That is, they may consist of one element, or more than one element. They may also involve reference to a value via another path. In the example for the nominative plural of Russian nouns we see that the path "" is referred to. This means that the value for the path refers to the value for the path "". The quotes enclosing the path are indirection markers. They indicate an indirect reference to a value The value referred to by the path "" will depend on the lexical item in question. Nodes need not be connected by hierarchy relations. They may be connected by network relations where a node inherits a particular type of information from another node, but not all possible types of information. This involves either evaluation of information already available and determination of inheritance source at an interdependency node, or a non-empty path referring to a node. Note that any fact which contains a path referring to another node must be a network relation if the path contains one or more attributes. If the path contains no attributes, then it must be a hierarchy relation. The two instantiations of network relations are illustrated in (5) below. (5)

VERB: == WORD == PREFIX == BASE 1

Dunstan Brown

210

In the example above, the node VERB belongs to a hierarchy of lexemes and inherits information from the node WORD. Information about prefixes can be found at the node PREFIX. Any information about the morphological operations carried out on bases can be found at the node BASE l. These are both network relations, as they involve reference to other nodes for a specific type of information, rather than all possible information, as would be the case if they were connected by a hierarchy relation. As network relations involve reference to other nodes for a specific kind of information, there is no possibility of inheriting contradictory information. Network relations are therefore examples of what is called orthogonal multiple inheritance (Touretzky 1986:73). Note also that the nodes are referred to by paths which contain one attribute, in this case the hierarchy identifiers prefix and base. A diagrammatic representation of the formal rendering above would look like that in figure 3. WORD PREFIX

BASE 1

VERB

Figure 3: Network Relations The hierarchy relation is rendered as an uninterrupted line, whereas the network relation is rendered as a dashed line. Looking at the formal notation for the nodes VERB and WORD does not tell us whether PREFIX and BASE l are also in hierarchy relations with them and therefore part of the same hierarchy. We would need to see the formal notation for the nodes PREFIX and Β ASE l. This leads us on to the next important point about the general architecture of Network Morphology theories. That is, not all nodes are in hierarchy relations and therefore not all nodes are part of the same hierarchy. However, all nodes are ultimately connected. The sum total of the connections between these nodes constitutes one network. This network consists of distinct hierarchies whose nodes may be connected to nodes outside their hierarchy by network relations. So far, we have really only seen nodes from the morphological hierarchy and the lexemic hierarchy. Our Network Morphology analysis of Russian verbs and consonantal alternations treats both of these in terms of hierarchies. The hierarchy of root and base final consonants treats the possible final elements as nodes in the hierarchy, with paths containing attributes for the different grades of alternation. The hierarchy of bases forms different verbal bases, and therefore stems. These bases are values which are paired with the paths containing the hierarchy identifier base and an index. The other way for network relations to be instantiated is for information to be evaluated and the source of inheritance determined at an interdependency node. For instance, in current Network Morphology analyses of both Polish and Russian semantic information about nomináis is evaluated and determined at an interdependency node, as illustrated in (6).

Stem Indexing in the Russian Verb: a Network Morphology Account (6)

NOMINAL : ==

211

""

NOUN: == NOMINAL

== DECLENSION:


At the node NOMINAL in the lexemic hierarchy it is stated that the path < m o r > , and therefore also more specific instances of that path, is referred to the path " < d e c i e n s i o n a i _ c i a s s > " for its values. However, at the NOUN node which inherits from NOMINAL we see that the path < d e c i e n s i o n a i _ c i a s s > is referred to the node DECLENSION. The path paired with the node does not contain an attribute, nor is it the empty path. Rather, it contains another path " < s e m s e x > " which must be referred to for a value This value is used as an attribute in the path at the DECLENSION node. As the path " < s e m s e x > " is enclosed in indirection markers the value will depend on the individual lexical item in question. The DECLENSION node then declares the nodes from which a lexical item inheriting from NOUN should inherit facts about its morphology on the basis of the semantic information evaluated, as illustrated in (7). (7)

DECLENSION: == N_I: == N_II:.

A noun should inherit its morphology from the node N_I, if it denotes a male, and N_II if it denotes a female. The DECLENSION node is an interdependency node. In diagrammatic representations of network relations, interdependency nodes are often left out. Diagrams therefore do not always distinguish between the two different types of instantiation of network relations we have discussed here. They are both represented by dashed lines! Having given a general outline of the underpinnings of Network Morphology we now go on to deal with the data and show how morphonological selection and indexing of stems are independent of one another.

4. Network Morphology and the Russian Verb There are two questions related to morphonology and the Russian verbal system. First, how is it that consonantal alternations are independent of word class and used by different word classes in different ways? Second, how does verbal stem-formation make reference to particular grades of alternation? In this paper a distinction is made between roots, bases and stems. Following the definition of Matthews (1991:131) the term 'base' is used to account for the material upon which morphological operations are carried out. Consonantal alternations are represented as a hierarchy.

212

Dunstan Brown

Operations on bases create stems which are indexed according to whether they are stems for the infinitive and past or they are stems for the present. Prefixes are added separately and play no role in the consonantal alternations which occur as a result of operations on bases. At each node in the hierarchy of bases a particular set of operations is defined for a particular class of verbs. Fully inflected forms of verbs must consist of a stem, optional theme vowel and ending. As well as inheriting information about operations on bases, lexical entries also inherit information about stems. By default, a stem is no more than a base on which no prefixation occurs. This generalisation is represented indirectly in that the lexemic hierarchy states that a stem is a prefix and a base. The form of the prefix defined at a separate node is determined by an evaluation of the value paired with the path . As this path is undefined by default, there is no prefix. As an answer to the question of how verbal stem formation refers to particular grades of alternation there are three dimensions to take into account. First, lexical entries contain facts stating what their roots are and where to find information regarding the final element of their roots. Second, information regarding the final element of roots is declared at nodes in the hierarchy of consonantal alternations, which defines the shape of different grades of alternation for a given final element. Third, the hierarchy that defines operations on bases refers to the different grades of final elements in the formation of indexed stems. As prefixation is orthogonal to the other base operations it cannot have a direct influence on consonantal alternations. Equally, where facts in the hierarchy of consonantal alternations involve reference to other paths, these paths contain the hierarchy identifier base to indicate that the alternations occur in that domain. This is because Russian does not use consonantal replacement, as defined by Hoeksema and Janda (1988), to realise inflectional categories. It does use it to realise particular stem types, however. There are varying degrees of generality for particular alternations. Calculations based on figures from Ilola and Mustajoki (1989:140) indicate that verbal stems which belong to the same class as the verb delat ', used as an illustration for conjugation I verbs, account for 38 % of all verbs in Zaliznjak (1977). Zaliznjak's (1977:91) type 1 includes verbs which we treat as a subclass of the delat ' type, those such as ugr 'umet' which have a different vowel before the infinitive ending but which behave in the same way otherwise.6 According to Ilola and Mustajoki (1989:141) Zaliznjak's type 1 accounts for 44.8 % of all verbs in the dictionary and 78.2 %

6

Veibs such as ugr 'umet' 'to grow morose' are treated as a class which inherits from the higher class of delat ' type verbs. We may note here that our treatment fits neatly between the classification of Karcevski (1922; 1927), where verbs with the -e classificatoiy suffix are a separate class and, for example, that of Zaliznjak (1977), where they are the same class. All of the information that can be shared between the two classes, namely everything apart from the different classificatoiy suffix, is shared, while at the same time the difference is captured.

Stem Indexing in the Russian Verb: a Network Morphology Account

213

of all verbs of imperfective aspect in the dictionary. These figures give some support to our treatment of BASE l as a default. The differing degrees of generality in verbal stem formation is captured by the hierarchy of bases, which defines differing operations which either add material to roots or replace consonants. The more general types of base operations, which are also associated with particular conjugation classes, occur higher in the hierarchy of bases. So BASE_1 in figure 4 defines the different stems for verbs such as delat', the most numerous verbal class according to the figures from Ilola and Mustajoki (1989:141). It is as we move down the hierarchy to a point beyond the node labelled CONSONANTAL that consonantal replacement is defined as an operation on bases. First and foremost for the classes of verb stems below this one, consonantal replacement is found in stems for the present tense. In fact, all verb stems used for conjugation II have consonantal replacement in the present tense. The nodes which represent these base classes start with the letter I in their names.

BASE Ι

Ο BASE 5aO_BASE_5bO_BASE_5cO_BASE_5e 0_BASE_5f 0_BASE_6a

O BASE_7

O BASE 5d

I_BASEe

IBASEja

I BASEa

O BASE 8 O BASE 11

O BASE 9

OBASEJO

Figure 4: The hierarchy of verbal bases Note that figure 4 is a hierarchy of bases and orthogonal to the lexemic hierarchy which brings together all information about lexical items. The node labels correspond to the labels we have given in tables 1-4. According to our classification there are three major divisions in verbal stems which make use of consonantal replacement. The first group are those verbs which have the soft grade in the present tense, except for the first person singular and third person plural, which have the zero grade. This group is represented by the node O B A S E 5 and all nodes below it. The second group are those verbs which have jotation throughout the present tense.

Dunstan Brown

214

The node 0 BASE 6 represents this strand. Third are the verbs which have the soft grade in the present and jotation in the first person singular of the present. Verbs of this latter group all belong to conjugation II and are represented by the node I BASEi and the nodes below it. 7 By default, verbs inherit base operations from a node B A S E l . This fact is defined at the VERB node in the lexemic hierarchy. Some of the facts to be found at the VERB node are given in (8). (8)

VERB : = = WORD < s y n cat> = - ν < s t e m > = = "" < p r e f i x > = = PREFIX < b a s e > = = BASE_1

""

These facts state the syntactic category and that definitions of prefixes and base operations are found at the nodes PREFIX and BASE l respectively. The node PREFIX contains the possible verbal prefixes which realise semantic extensions of the hierarchy identifier prefix. Stems consist of a reference to prefix values and base values. As prefixes are undefined by default, the most general case is for the stem and the base to be identical. As we have mentioned, there are two main allostems of the verb. The one that is used in the infinitive and past, and the one that is used in the present. These are indexed 1 and 2 respectively. At the node B A S E l a number o f f a c t s define properties of bases. These facts, inherited by default, are given in (9).

7

This makes our classification differ somewhat from that of Zaliznjak (1977), as veibs such as sosal ' 'to suck' belong to 0_BASE_5, because they have the soft grade in the present and the zero grade in the first person singular and third person plural of the present. In Zaliznjak's (1977: 112) classification this type of verb is a subtype of group 6 (namely 6_), although it is a subtype in virtue of its consonantal alternations. Our classification follows Zaliznjak (1977) in that O BASE 6 verbs are essentially the same as his type 6, with the exception of the group mentioned and veibs of the osmejat' 'to ridicule' type, which we place in class one, claiming that the root of the verb ends in /j/. Other nodes and their labels in figure 4 do not correspond to those labels of Zaliznjak (1977). As we have stated, it can readily be seen that any node which inherits from 0_BASE_5 will have the soft grade in the second and third person singular and first and second person plural of the present tense. Any veib with jotation throughout the present tense will belong either to 0_BASE_6 or 0_BASE_6a. Verbs with jotation in the first person singular of the present tense and the soft grade elsewhere in the present will inherit from I_BASEi or a node which inherits from it. This should enable anyone seeking a classification for a veib to limit it to a particular area of the hierarchy. It should also be noted that this hierarchy of bases differs from that of Brown (1995) in that the nodes are differently labelled and the hierarchy is configured differently. As it is not our purpose here to go into the intricacies of Russian veibal classification, but rather to discuss the relation between stem indexing and morphonological selection, we shall not take this any further.

Stem Indexing in the Russian Verb: a Network Morphology (9)

Account

215

BASE_1 :

a. = = "" b. == first C. == "" d. == "" e. cbase final v> == "" f. == "" "" "" g. == cbase 1> h. == _a

(9a) states that the default base for an operation is the root. Bases will belong to the first conjugation according to (9b). The final element of a base will be the final vowel according to (9c). The final consonant of the base, which need not be the final element of the base, will be the final element of the root, as stated by (9d). The assumption underlying this is that the default shape of verbal roots is (CV)CVC. According to Öurganova (1973:41) CVC is the minimal structure for nomináis, but verbs may have CV. Given that verbs of the del-a-t' type are the default verbal class, as the figures from Hola and Mustajoki (1989) indicate, we see that (CV)CVC must also be the default root structure for verbs. It is also clear that if we take the base of the verb del-a-t', namely del-α- (i.e. without the infinitive ending), the final consonant is the final consonant of the root, as (9d) declares. According to (9e) the final vowel of the base is what has been called here the base vocal. This is the 'classificatory suffix' (Timberlake 1993:850) before the infinitive ending. It is given for this default class in (9h) as the value a. The fact (9f) states that bases with index l, used in the infinitive and past, involve the concatenation of a default base with a base final consonant and a base final element. As (9a) states that roots are the default bases, the root of the verb is the first element in (9f) for this class of verbs. Further, the base final consonant is the same as the final element of the root according to (9d). So the final element of the root is added to the root. The final element of the base is added to this combination. By (9c) referring to (9e), and (9e) to (9h) we find that the final element of the base is a. Taking our example again, the effect of (9f) on the lexeme DELAT' 'to do' is to require the concatenation of de- with I- and a-. The fact (9g) states that the stem indexed 2, used in the present, is the same as the stem indexed l, which is determined according to (9f), as already discussed. This captures the default identity of the stems, which we mentioned, in our discussion of the data. Both allostems of the lexeme DELAT' are therefore of the shape del-α-, other verbs of this class follow the same pattern. The lexical entry of DELAT' is given in (10).

Dunstan Brown

216 (10) Delat': == VERB == to do == de == L:.

The final fact about the final element of the root addresses a node L for further information. Our use of the attribute root in (10) and elsewhere is somewhat non-standard, as it really refers to the root minus its final element. In the next section we illustrate the hierarchy of which L is a member and show how it determines the different grades of consonant independent of word class. Note that so far in our illustration of the formation of the most general stem type we have not had recourse to talking about grades of alternation. 4.1 Network Morphology and Morphonological Selection Verbs which do not inherit the default generalisation about bases must stipulate the node in the hierarchy of bases from which they inherit. As tables 4 and 5 show, less general classes of base operation involve consonant replacement, rather than the addition of a classificatory suffix. In (10) we saw that the final element of the root declared in the lexical entry was addressed to a node in a hierarchy of consonantal alternations. This hierarchy, which is based on a suggestion made by Andrew Spencer (personal communication), is represented diagrammatically in figure 5.

CONS soft = zero + softening jotated = base final soft

L

Ν

R

Κ

O X

Figure 5: A hierarchy of consonantal alternations At the node CONS there are facts which state that the default is for a root to end in a zero grade, as in (IIa). In (1 lb) the soft grade takes the zero grade of the final consonant of a base and softens it. This is indicated by the overlap operator _ (following Gibbon 1991:5) and the symbol for softening ' . This means that there is accompanying palatalisation. By default, the jotated grade is the same as the soft grade (11c).

Stern Indexing in the Russian Verb: a Network Morphology

Account

217

( 1 1 ) CONS: a. = = "" = = c b. < s o f t > = = "cbase final c 0>" _ ' C. < j o t a t e d > = = "".

The node labelled l inherits directly from CONS. It is given in (12). The only fact given apart from the hierarchy relation with CONS is that the zero grade has the value _l. (12) L: == CONS = = _ 1 .

The node in (12) inherits from CONS the fact (11a) that its least specific variant is the zero grade. Recall that the last fact in (10) in the previous section referred to the node l for all instances of the final element of the root. The attribute o therefore extends the path (attribute structure) " b. cbase 2> == ""

The node CONSONANTAL defines the total base for bases which involve consonant replacement as a combination of the root and its final element (which will be a consonant). This is given in fact (14a). The use of the number index in the path "", referred to as part of the complex value for this fact, is there for verbs which may have a change in vowel quality within their root. It need not concern us here, but works in a similar way to the consonant replacement being discussed. Fact (14b) states that the stem indexed 2 will be the soft variant of the entire base. Verbs of the plakat' type, as table 3 shows, have the jotated grade throughout the present tense. They inherit from the node CONSONANTAL the fact (14a) which defines all of the base as consisting of the root and its final element. In contrast to (14b), however, nouns like plakat ' have jotation in their stem indexed 2, rather than the soft grade. For verb stems of the O B A S E 6 class fact (14b) is overridden. The node for O B A S E 6 is given in (15) below. (15)

0_BASE_6: == 0_BASE_5 cbase 2> == cbase all jotated>.

According to the last fact in (13) the final element of the root ofplakat' is defined at the node κ in the hierarchy of alternations. The attribute jotated in cbase all jotated> in (15)

Stern Indexing in the Russian Verb: a Network Morphology

Account

219

extends the path in (14a). As (14a) states that combines the root and its final element, referred to by , the extension of is also an extension of the path croot finai>. This means that the value for plakat', and other verbs like it, involves reference to the paths and at the lexical entry. The path croot 2 jotated> matches with the path in (6), giving pia-. The path matches with the path which refers to the node κ. As the attribute jotated extends the path it will also extend the path referred to at the node κ. This means that the value of the final element of the root used for cbase 2> is the one referred to by the path < j otated> at the node κ. Recall that the node κ given in (16) is a node in the diagrammatic representation in figure 5. (16)

K:

== CONS

== _k ==

.

Cursory examination of the node in (16) indicates that there is no path for the jotated grade. This is inherited by default from the node CONS, which we have already seen in (11). Fact (1 lc) says that the jotated grade is the same as whatever the soft grade of the base final consonant is. Readers with a good memory should remember that (9d) states that the base final consonant is the same as the final element of the root. This means that the attribute soft in the path "cbase final c soft>" referred to in (11c) is an extension of "croot finai>" referred to in (9d) and that the value of the jotated grade is the one referred to by the path csoft> at the node κ. This path occurs in the last fact defined at that node. The value of the last element of the stem in question is Ici. Combining the information tells us that the stem indexed 2 for the verb plakat ' is pia-c '-. This is the stem used in the present tense. In sum, it can be seen that the hierarchy of bases refers to the different grades of consonant, which are defined in a separate hierarchy. This approach can be contrasted with a morphemebased one such as that of Wunderlich (1995) which makes use of a floating segment [pal] and a function CPAL to derive alternations. The use of the latter requires stipulation by each item. Our approach states both the function and its context of application. The hierarchy of bases indicates varying degrees of generality. While one could argue that the smaller classes of bases with a few members could be treated as arbitrary facts, it is not desirable to have to state in each lexical entry for, say, the 8000 or so verbs (see Mustajoki and Ilola 1989:140) belonging to the second conjugation that there is a C PAL stem variant to account for the first person singular (see table 4). In our approach the lexical entry itself mediates between the base hierarchy and the hierarchy of consonantal grades. As we shall see, it is stated as a fact that particular base types go with either the first or second conjugation. Therefore, the choice of a particular stem paradigm determines the choice of conjugation.

Dunstan Brown

220 4.3 Inflection and stems

So far we have shown how indexed stems are created. Before we move on to show how the appropriate stem is selected for the appropriate morphosyntactic context, an additional piece of information is required regarding stems. The paths in the hierarchy we have shown begin with the identifier base. However, there is also a node PREFIX which defines prefixes for addition to bases. As mentioned at the beginning of this section, stems are by default the same as bases. Some example facts from the node PREFIX are given in (17). (17) a. b. C. d. e.

PREFIX: == == undefined == == v_ == vi_

Fact (17a) states that the choice of prefix is dependent on the semantics of the prefix, which will be specified at a lexical entry. (17b) states that the semantics of the prefix is, by default, undefined. (17c) states that if the semantics of the prefix is undefined, then there is no prefix. This accounts for the default generalisation that there is no prefixation. Facts (17d) and (17e) state values for prefixes with the semantics IN and OUT respectively. ι

For the discussion to follow it should be noted that any extension of the path will involve referral to the index extensions of the path , because of the fact == "" "". If any fact in the hierarchy of inflection makes reference to , for example, then this will also refer to any value paired with the path . As we have seen, these are defined differently for the verbs plakat' and delat'. Furthermore, prefixation requires evaluation of semantics specified in a lexical entry and has nothing to do with indexation. Consequently, verbal prefixation has no effect on consonantal alternations. Hence, according to this approach the prefixed and unprefixed pairs vibrai' 'to select' and brat' 'to take' should have the same consonantal alternations, and prefixation in Russian does not have any influence on morphonological selection or the indexing of stems. As shown in table 1, verbal inflection consists of two conjugation classes, which we shall label V I and V II respectively. If we separate up the endings from the thematic ligature, or theme vowel, of the conjugation class, we find that the major difference between the two classes is, in fact, the theme vowel and the ending of the third person plural. The third person plural is -ut for V I and -at for V II. Note that the stem-indexing approach does not require us to posit homophonous but different endings for the first person singular on the basis that the stem consonant may have jotation in V II, but need not for V I. Hence, the first person singular forms l'ubi'-u Ί love' and dela-j-u have the same ending and require a stem with the same

Stem Indexing in the Russian Verb: a Network Morphology

Account

221

index. The determination of that stem's index is where the difference lies. Equally, as Stump (1996) shows with data from Sanskrit to illustrate his Indexing Autonomy Hypothesis, stems may be formed in the same way but have different indexes. This is true of the default class, such as the stem del-α-, where the stems are the same, but the indexing is not. The various classes of stem defined by base operations are associated with one of the two conjugation classes. Information at the word class node for verbs is given in (18), with facts additional to those in (8). (18) VERB: == WORD == ν a. == "" b. == CONJ: == "" "" == "" == "cbase stress>" C. == "cbase conjugation;»" == PREFIX == BASE_1.

In (18a) it is stated that the 'syntactic aspect' is determined according to the 'semantic aspect' of the verb. Russian verbs can be either imperfective or perfective in aspect. Currently, we allow for verbal semantics which state that an event is either 'bounded' or 'undefined'. As the default value for any fact is 'undefined', we only need state if verbs denote bounded events, if they have perfective aspect. Otherwise, they are assigned imperfective aspect in virtue of the default. We assume that aspect is syntactic, because it affects the realization of the future tense. If the verb is imperfective, then it is realised by an auxiliary plus the infinitive. If it is perfective, then it is realised by a single form that would have a present tense reading, if it were imperfective. The fact (18b) states that the morphology of the verb depends on an evaluation of the conjugation class associated with the stem. This is in turn dependent on the particular base of the stem, as stated by (18c). Finally, in (19) we see the node for verbal morphology from which the classes V I and V II inherit. Note that the value for is not specified, as this is stipulated at V I and V II, as is the value for the third person plural, and these nodes are in a hierarchy relation with MORVERB. For our purposes we see that the indexed stems are referred to for each morphosyntactic combination. The node JOT, referred to as part of the complex value for each combination, defines whether or not a palatal glide may occur, depending on the final element of the stem indexed 2.

Dunstan Brown

222 (19)

MOR_VERB: == "" JOT: _u == "" JOT:: "" J = = "" JOT:: " " _t = = "" JOT:: " " m = = "" JOT:: "" te == "" _1' PRED: == "" _1 PRED:

The actual shape of the stem is therefore irrelevant for verbal inflection, with the possible exception that there is a need to separate a W combination with a palatal glide. Furthermore, the treatment of stem indexing as autonomous from morphonological selection means that there is no need to resort to phonological diacritics in order to account for different grades of palatalisation before the same ending. For instance, conjugation II verbs will always have the jotated grade before the first person singular ending, whereas conjugation I verbs may choose from a number of possibilities. Stem indexing also has linguistic reality connected with it in a way that phonological diacritics appear not to. Indexed stems can be seen to have a distribution within a paradigm, analogous to the way that cases or other generally recognised categories have a particular distribution in syntax. Our use of indexation demonstrates the relation between the conjugation I sos-u Ί suck' and the conjugation II lubl'-u Ί love', which have a different form in the first person singular from most of the present tense. The difference between indexation and rules of referral is that the latter capture asymmetrical relations between feature combinations. While there is good reason to assume asymmetrical referrals where a whole word form is concerned, there is little reason to do so for stems.

5. Conclusion This paper has attempted to answer the question of how what Aronoff has termed the 'morphomic level' and morphonological selection might be related. Stem indexing is motivated by the fact that particular stem forms may be required in particular slots in the paradigm. These slots need not, however, be characterised in terms of compatible morphosyntactic features. For instance, it is not clear that the default identity of the stem in the past and the infinitive can be accounted for by assuming that there is some feature that they have in common. Morphonological selection is determined by a hierarchy which generalises over final elements of the base. This is independent of the indexing of stems, thereby allowing for different realisations of identically indexed stems. The approach we have outlined here also demonstrates how préfixa-

Stern Indexing in the Russian Verb: a Network Morphology Account

223

tion can have no direct influence on indexing of stems. Consonantal alternations are inherited by default from the base.

References Andersen, Henning (1980) Russian Conjugation: Acquisition and Evolutive Change. In: E. Traugott, R. Labrum and S. Shepherd (eds.): Amsterdam Studies in the Theory and History of Linguistic Science IV. 285-301. Amsterdam: John Benjamins. (= Current Issues in Linguistic Theory 14) Aronoff, Mark (1976) Word Formation in Generative Grammar. Cambridge, Mass: M.I.T. Press. - (1994) Morphology by itself: Stems and Inflectional Classes. Cambridge, Mass: M.I.T. Press. Avanesov, R. I. (1952) Russkoe literaturnoe proiznoSenie. Moscow: Gosudarstvennoe ucebno-pedagogiceskoe izdatel'stvo ministerstva prosveäienija RSFSR. Brown, Dunstan (1995) Setevaja morfologija i russkaja glagol'naja sistema. Vestnik Moskovskogo Universiteta, ser. 9. Filologia no. 6. 91-108. Brown, Dunstan, Greville Corbett, Norman Fraser, Andrew Hippisley and Alan Timberlake (1996) Russian noun stress and network morphology. Linguistics 34, 53-107. Brown, Dunstan and Andrew Hippisley (1994) Conflict in Russian Genitive Plural Assignment: A Solution Represented in DATR. Journal of Slavic Linguistics 2, 48-76. Corbett, Greville G. and Fraser, Norman M. (1993) Network Morphology: A DATR account of Russian nominal inflection. Journal of Linguistics 29, 113-42. Öurganova, V. G. (1973) Oäerk russkoj morfonologii. Moscow: Nauka. Daelemans, Walter, Koenraad De Smedt and Gerald Gazdar (1992) Inheritance in Natural Processing. Computational Linguistics 18, 205-18. Darden, Bill J. (1988) Truncation and/or Transderivational Constraints in Russian Word-formation. In: Lynn Macleod, Gary Larson and Diane Brentari (eds.): Papers from the 24th Meeting of the Chicago Linguistic Society. 88-99. Evans, Roger and Gazdar, Gerald (1989a) Inference in DATR. Proceedings of the 4th Conference of the European Chapter of the Association for Computational Linguistics. Manchester, England. 66-71. - (1989b) The semantics of DATR. In: A. G. Cohn (ed.): Proceedings of the Seventh Conference of the Society for the Study ofArtificial Intelligence and Simulation of Behaviour, 79-87. London: Pitman. Evans, Roger and Gerald Gazdar (1996) DATR: A Language For Lexical Knowledge Representation. Computational Linguistics 22, 167-216. Fahlman, Scott E. (1979) Representing and using real-world knowledge. In Patrick H. Winston and R. H. Brown (eds): Artificial Intelligence: An MIT perspective, Volume 1, 451-70. Cambridge, Mass: M.I.T. Press. Flickinger, D. P. (1987) Lexical rules in the hierarchical lexicon. Unpublished doctoral dissertation, Stanford University. Fraser, Norman M. and Greville G. Corbett (1995) Gender, animacy and declensional class assignment: a unified account for Russian. In Geert Booij and Jaap van Marie (eds.): Yearbook of Morphology 1994, 123-50. Dordrecht: Kluwer. Fraser, Norman M. and Hudson, Richard A. (1992) Inheritance in Word Grammar. Computational Linguistics 18, 133-58. Garde, Paul (1972) La distribution du hiatus et le statut du phonème /j/ dans le mot russe. In: D. Worth (ed.): The Slavic Word. The Hague: Mouton. Gibbon, Dafydd (1991) Lexical Signs and Lexicon Structure: Phonology and Prosody in the ASL-Lexicon. ASL-MEMO-20-91/UBI, University of Bielefeld. Gvozdev, A. N. (1949) Formirovanie u rebenka grammatiieskogo stroja russkogo jazyka, /-//. Moscow: Izdatel'stvo Akademii pedagogiCeskix nauk RSFSR.

224

Dunstan Brown

Halle, Morris (1963) O pravilax russkogo spijaiemja. In: American Contributions to the Fifth International Congress of Slavists 1, September 1963, Sofia, 113-132. Mouton: The Hague. Hamilton, William S. (1980) Introduction to Russian Phonology and Word Structure. Columbus: Slavica. Hoeksema, Jack and Richard D. Janda (1988) Implications of process-morphology for categorial grammar. In: R. T. Oehrle, E. Bach and D. Wheeler (eds.): Categorial Grammars and Natural Language Structures. Studies in Linguistics and Philosophy 32, 199-249. Dordrecht: Reidel. Hudson, Richard (1990) English Word Grammar. Oxford: Blackwell. Hola, Eeva and Mustajoki, Arto (1989. Report on Russian Morphology as it appears in Zaliznyak's Grammatical Dictionary. Helsinki: Helsinki University Press. Isaëenko, A. V. (1960) Grammatiöeskij stroj russkogo jazyka ν sopostavlenii s slovackim. Bratislava: Izdatel'stvo slovackoj akademii nauk. Jakobson, Roman (1948) Russian Conjugation. Word IV, 155-167. Karcevski, Serge (1922) Études sur le système verbal du russe contemporain. Slavi a 1. - (1927) Système du verbe russe (essai de linguistique synchronique). Prague: Plamja. Keller, Bill (1995) DATR Theories and DATR Models. Proceedings of the Association for Computational Linguistics '95. Kempgen, Sebastian (1989) Grammatik der russischen Verben. Wiesbaden: Otto Harrasowitz. Matthews, P. H. (1972) Inflectional Morphology. Cambridge: Cambridge University Press. - (1991) Morphology. Cambridge: Cambridge University Press, [second edition] Nesset, Tore In draft. Russian Conjugation Revisited. Doctoral Thesis, University of Oslo. Pirogova, L. (1969) O klassifikacii russkix glagolov. Russkij jazykv nacional'noj Skole 4, 26-33. Quillian, M. Ross (1968) Semantic memory. In: Minsky, Marvin. Semantic Information Processing. Cambridge, Mass: M.I.T. Press. Sadler, Louisa, Andrew Spencer and Marina Zaretskaya (1996) The 'Third Stem1 in Russian. Yearbook of Morphology. To appear. Spencer, Andrew (1988) Arguments for morpholexical rales. Journal of Linguistics 24,1-29. Stump, Gregory T. (1996) Stem Formation, Stem Indexing, and Stem Choice. Ms, University of Kentucky. Timberlake, Alan (1993) Russian. In: Bernard Comrie and Greville G. Corbett (eds.): The Slavonic Languages, 827-886. London: Routledge. Touretzky, David S. (1986) The Mathematics of Inheritance Systems. London: Pitman. Trubetzkoy, Nikolai S. (1931) Gedanken über Morphonologie. Travaux du Cercle linguistique de Prague 4, 160-163. Volz, Tanja (1994) Zu einer Klassifizierung des russischen Verbmaterials anhand des Mechanismus der Default-Vererbung. Staatsexamenszulassungsaibeit, Universität Konstanz. Wunderlich, Dieter (1995) Verb inflection in Russian. Ms, Heinrich-Heine-Universität Düsseldorf. Zaliznjak, A. A. (1977) Grammati öeskij slovar' russkogo jazyka. Moscow: Russkij jazyk. Zwicky, Arnold M. (1985) How to describe inflection. BLS11, 372-86.

Wolfgang Ullrich Wurzel (Zentrum Allgemeine

Sprachwissenschaft,

Berlin)

Drei Ebenen der Struktur von Flexionsparadigmen

Einleitung Unter einem Flexionsparadigma versteht man bekanntlich die Gesamtheit der Flexionsformen eines Lexems. Es besteht aus einer lexikalischen Grundform und in der Regel mehreren abgeleiteten Formen.1 Dabei stellt sich natürlich die Frage, in welchem Verhältnis die einzelnen Flexionsformen eines Paradigmas zueinander stehen, auf welche Weise Flexionsparadigmen innerhalb des Flexionssystems organisiert sind. Das gilt speziell für Flexionssysteme mit mehreren konkurrierenden Paradigmen bzw. Flexionsklassen bei Wörtern der gleichen Wortart, wie wir sie beispielsweise im Deutschen, Isländischen oder Lateinischen, nicht aber im Türkischen finden. Aus einer Reihe von Gründen (die weiter unten deutlich werden) erweist sich gerade auch die deutsche Substantivflexion als ein fur die Fragestellung sehr gut geeigneter Faktenbereich.2 Wir wollen also im folgenden die Struktur von Flexionsparadigmen anhand der deutschen Substantivdeklination untersuchen, wobei wir an frühere Arbeiten zur gleichen Problematik anschließen können.3

1. Die Struktur der Einzelparadigmen In Grammatiken, auch wissenschaftlichen, treten uns die Flexionsparadigmen im allgemeinen als eine konventionalisierten Abfolge von Flexionsformen entgegen, in der deutschen Deklination beispielsweise mit der Abfolge der Kasuskategorien Nominativ - Genitiv - Dativ -Akkusativ und mit der Abfolge der Numeruskategorien Singular - Plural, so daß sich die bekannte traditionelle Reihenfolge der Formen vom N.SG. bis zum A.PL. ergibt; vgl. dazu z.B. das 'starke' Maskulinum Wolf. (1) N.SG. G. D. A.

2 3

{der) (des) (dem) (den)

Wolf PL. Wolfle)s Wolf Wolf

(die) (der) (den) (die)

Wölfe Wölfe Wölfen Wölfe

Die Minimalausstattung eines Flexionsparadigmas besteht aus einer Grundform und einer abgeleiteten Form, wie sie z.B. die Substantive in vielen Bantusprachen zeigen, vgl. Suaheli mtoto 'Kind' - Plural watoto, doch solche Paradigmen sind hinsichtlich ihrer Struktur natürlich recht uninteressant. Zu diesen Gründen vgl. speziell Wurzel (1994: 29f ). Vgl. Wurzel (1984: 116ff.), (1987), (1990) und (1994).

Wolfgang Ullrich Wurzel

226

Diese Abfolge der Formen ist insofern nicht völlig arbiträr, als daß der N.SG. die lexikalische Grundform des Wortes und damit die Ausgangsform für die Deklination darstellt, alles Übrige in der Reihenfolge ist einfach traditionell, nicht durch die Struktur der Sprache selbst begründet. Man vgl. dazu z.B. die im Isländischen übliche Reihenfolge der Kasus Nominativ - Akkusativ - Dativ - Genitiv, die etwa den in den Flexionsparadigmen vorkommenden Synkretismen (übrigens auch denen des Deutschen) besser Rechnung trägt als die für das Deutsche konventionalisierte. Die Reihenfolge der Flexionsformen, jedenfalls wenn sie die hierarchischen Strukturverhältnisse in den Flexionsparadigmen widerspiegeln soll, ist aber durchaus nicht arbiträr. Das betrifft zunächst natürlich den hervorgehobenen Status der Grundform im Paradigma, der allerdings auch nur selten in Frage gestellt wird (wobei - das soll hier nur erwähnt werden - die Ermittlung der Grundform nicht immer so trivial ist wie beim deutschen Substantiv). Doch auch die übrigen Formen eines Flexionsparadigmas sind normalerweise nicht einfach gleichwertig oder gleichgewichtig. Das zeigt bereits ein Blick in ein beliebiges Wörterbuch mit Flexionsangaben für Sprachen wie Deutsch, Lateinisch oder Russisch. So erscheinen beim deutschen Substantiv im allgemeinen neben dem N.SG. als Grundform zusätzlich als Kennformen der G.SG. und der N.PL. (Wolf - Wolf(e)s, Wölfe), beim lateinischen Verb neben dem Infinitiv noch die l.PS.PRÄS.IND.AKT., die l.PS.PERF.IND.AKT. und das PART.PERF. (facere 'machen' - faciô, fêci, factum) usw. Aus den angegebenen Flexionsformen kann der Benutzer dann alle übrigen Formen des Paradigmas implikativ ableiten. Untersucht man den Aufbau der Flexionsparadigmen in Sprachen mit Flexionsklassen, so zeigt sich deutlich, daß die Paradigmen, zumindest in sehr starkem Maße, implikativ strukturiert sind. Wenn innerhalb eines Paradigmas in einer bestimmten Kategorie eine bestimmte Form erscheint, dann impliziert das das Auftreten einer bestimmten anderen Form in einer anderen Kategorie bzw. in anderen Kategorien. So kann beim deutschen Substantiv Wolf zwar die Form des N.PL. nicht strikt aus der Grundform abgeleitet werden (wir kommen darauf zurück), doch auf der Grundlage des N.PL. lassen sich dann wie bei sämtlichen regulären Substantiven des Deutschen alle übrigen Flexionsformen des Paradigmas implikativ gewinnen. Für Wörter dieses Typs ergibt sich die folgende Struktur des Flexionsparadigmas:

(2)

N.SG. Wolf N.PL. Wölfe G.SG. Wolf(e)s A./G.PL. Wölfe 1 D./A.SG. Wolf

1 D.PL. Wölfen

Drei Ebenen der Struktur von Flexionsparadigmen

227

Aus der Form des N.PL., die mit dem Marker -e (und Umlaut) gebildet ist, ergibt sich, unter der Maßgabe, daß Wolf die im Lexikon spezifizierte syntaktische Eigenschaft hat ein Nichtfemininum zu sein, die Form des G.SG. Das funktioniert aber nicht in umgekehrter Folge, denn i-Genitive kommen ja nicht nur bei Substantiven mit e-Pluralen, sondern auch bei solchen mit 0-, er-, s- und «-Pluralen vor; vgl. (des) Ankers, Mann(e)s, Parks, Staat(e)s, aber (die) Anker, Männer, Parks, Staaten. Aus dem s-Genitiv läßt sich die markerlose Form des D.SG. und des A.SG. gewinnen (die D.SG.-Formen auf -e wie (dem) Wolfe sind heute veraltet). Da zwischen diesen beiden Formen keine eindeutige Ableitungsrichtung auszumachen ist, ordnen wir sie an der gleichen Position ins Paradigma ein. Desweiteren ergeben sich aus der Form des N.PL. auch die anderen Pluralformen des Paradigmas, sowohl die mit dem N.PL. formal übereinstimmenden Formen des G.PL. und A.PL., für die das gleiche gilt wie für D.SG. und A.SG., als auch die formal abweichende Form des D.PL. Wie leicht zu sehen unterscheidet sich die formale Struktur des Paradigmas beträchtlich von der traditionellen Darstellung der substantivischen Deklinationsparadigmen im Deutschen. Interessant ist dabei, daß abweichend vom Usus der Wörterbücher nur eine einzige zusätzliche Kennform notwendig ist, um zu wissen, wie das Wort und die anderen Wörter dieser Klasse flektieren; und diese Kennform ist die Form des N.PL., anhand derer auch das Flexionsverhalten im Singular erschließbar ist. Das zeigt übrigens, daß die deutsche Substantivflexion nicht (wie häufig behauptet) in voneinander völlig oder doch weitgehend unabhängige Singular- und Pluralparadigmen 'auseinandergefallen' ist, sondern weiterhin im Rahmen einheitlich funktionierender Paradigmen organisiert ist. Hier kommt zugleich ein wichtiger Aspekt der implikativen Strukturbildung von Flexionsparadigmen zum Ausdruck. Die formale Struktur der Paradigmen folgt nicht notwendigerweise semantischen Kriterien; die Form des N.PL, ist (zusammen mit bestimmten Eigenschaften der lexikalischen Grundform) auch die Basis fur die implikative Ableitung der Kasusformen des Singular. Eine solche nichtsemantische morphologische Strukturbildung kommt relativ häufig vor, übrigens auch in Derivationsparadigmen, was anhand des Verhältnisses von Ländernamen, Bezeichnungen für die Einwohner und den zugehörigen Adjektiven ini Deutschen herausgearbeitet wurde.4 Die Substantive des Typs Wolf gehören zu denen, die im Deutschen die komplexeste Paradigmenstruktur aufweisen. Das beruht zum einen darauf, daß ihre N.PL.-Form nicht aufgrund der Grundform strikt zu implizieren ist, und zum anderen darauf, daß im Paradigma immerhin vier formal distinkte Wortformen auftreten, nämlich hier Wolf Wolfle)s, Wölfe und Wölfen.

Vgl. dazu Fuhrhop (1997): So gibt es ein weitgehend implikatives Verhältnis zwischen dem Ländernamen und der Bezeichnung für den Einwohner sowie zwischen der Bezeichnung fur den Einwohner und dem entsprechende Adjektiv nach dem Muster Türkei o Türke türkisch, aber nicht umgekehrt. Von der Einwohnerbezeichnung Türke (auf -e) läßt sich dagegen nicht der Ländernamen ableiten, vgl. Franke Franken und Grieche - Griechenland, ebenso nicht vom Adjektiv auf -isch die Einwohneibezeichnung, vgl. belgisch - Belgier. Das gilt, obwohl sich das jeweilige Adjektiv natürlich semantisch auf den Ländernamen und nicht (ausschließlich) auf die Einwohneibezeichnung bezieht.

Wolfgang Ullrich Wurzel

228

Die deutschen Substantive unterscheiden sich in dieser Hinsicht beträchtlich, wie noch im einzelnen zu zeigen sein wird. Als Gegenstück zum diskutierten Fall sei noch ein Femininum mit «-Plural wie Katze angeführt, das die im Deutschen einfachste mögliche Paradigmenstruktur aufweist: (3)

N.SG. PL. G./D./A.SG.

Katze 1 Katzen 1 Katze

In diesem Paradigma treten nur zwei formal distinkte Wortformen in systematischer Verteilung auf Singular und Plural auf, Katze und Katzen. Desweiteren kann die Pluralform anders als im Fall Wolf strikt aus der Grundform impliziert werden: Ein Femininum, das in der Grundform auf -e endet, kann seine Pluralform nur auf -n bilden. Die Flexion solcher Wörter ist also voll prädiktabel. Hier gibt es im übrigen innerhalb des Paradigmas keine Evidenz, daß die obliquen Singularformen aus der Pluralform abgeleitet werden, doch da in den komplexeren Paradigmen (wie u.a. beim Typ Wolf) die Pluralform eindeutig als Ausgangsform für die obliquen Singularformen fungiert, nehmen wir auch hier diese implikative Abfolge an. In diesem Zusammenhang noch eine kurze Erläuterung zur nicht unwichtigen Unterscheidung zwischen Flexionsklassen und Teilflexionsklassen: Das Paradigma Wolf steht für die gesamte Flexionsklasse der starken Nichtfeminina mit Umlaut (e-PL. und s-G.SG). Dagegen steht das Paradigma Katze nur für eine Teilflexionsklasse der schwachen Feminina («-PL. und 0-G.SG.), nämlich die der auf -e endenden Wörter. Man vgl. dazu das Paradigma eines auf Konsonant endenden Femininums (solche auf Diphthong und Langvokal funktionieren ebenso) der gleichen Flexionsklasse: (4)

N.SG. PL. G./D./A.SG.

Bahn Bahnen

i

Bahn

Dieses Paradigma hat nicht die gleiche Struktur wie das Paradigma Katze. In beiden Paradigmen treten zwar die gleichen Kategorienmarker auf, doch das Paradigma Bahn unterscheidet sich vom Paradigma Katze dadurch, daß hier die Pluralform nicht strikt aus der Grundform impliziert werden kann. Feminina auf Konsonant, auf Diphthong oder Langvokal (die übrigens phonologisch gesehen die natürliche Klasse der auf einen verzweigenden Reim endenden Wörter bilden) können ja im Plural auch den Marker -e aufweisen und damit zur Klasse der 'star-

Drei Ebenen der Struktur von Flexionsparadigmen

229

ken' Feminina gehören wie z.B die Wörter Gans, Sau und Kuh. Noch in einer weiteren Hinsicht unterscheiden sich beide Flexionsparadigmen: Das Paradigma Bahn liefert (anders als Katze) wiederum Evidenz für die Ableitung der Singularformen aus dem N.PL. Wohl impliziert bei einem Femininum auf Konsonant ein «-PL. Einen obliquen 0-Singular, aber nicht umgekehrt, denn der 0-Singular kommt ja auch kombiniert mit dem e-Plural (und Umlaut) vor; vgl. (der) Gans - (die) Gänse.Die Paradigmen wie in (2), (3) und (4) aufgeführt spiegeln (ganz wie es die traditionellen Flexionsparadigmen sollen) die Paradigmenstruktur des jeweiligen einzelnen Wortes und damit indirekt aller Wörter der jeweiligen Flexionsklasse oder Teilflexionsklasse wider. Sie erfassen die implikativen Relationen zwischen den Flexionsformen des einzelnen Paradigmas. Auf diese Weise konstituieren sie die erste Ebene der Struktur von Flexionsparadigmen.

2. Die Struktur der Paradigmen im Rahmen des Flexionssystems Man kann nun sämtliche voneinander unterschiedenen Paradigmen eines Flexionssystems als Repräsentanten der vorhandenen Flexionsklassen und Teilflexionsklassen ermitteln. Damit hat man dann zwar die jeweilige Struktur aller im Flexionssystem vorkommenden unterschiedlichen Flexionsparadigmen mit ihren internen Strukturrelationen erfaßt, aber nicht die über die Paradigmengrenzen hinausreichenden Strukturzusammenhänge im Flexionssystem. Will man diese erfassen, sind entsprechende Generalisierungen notwendig. Dazu nur ein Beispiel: Es gibt drei verschiedene nichtfeminine Paradigmen mit dem Wortausgang -e in der Grundform, nämlich die Typen Bote, Auge und Funke. Sie zeigen unterschiedliche Marker im G.SG.: (des) Boten, Auges, Funkens. Doch sie haben gemeinsam den Marker -η im Plural. Man kann also entsprechend über die Grenzen der einzelnen Paradigmen hinaus generalisierend konstatieren, daß alle Nichtfeminina auf -e im Singular im Flexionssystem ihren Plural mit dem Kategorienmarker -n bilden. Solche Strukturzusammenhänge lassen sich in recht angemessener Weise als Bedingungen über den Aufbau der Flexionsparadigmen in einem Flexionssystem, kurz Paradigmenstrukturbedingungen (PSB), formulieren.5 Die Paradigmenstrukturbedingungen gehen aus von den für die Flexion relevanten unabhängig gegebenen Eigenschaften der Wörter und leiten aus diesen das Flexionsverhalten der Wörter ab. Solche Eigenschaften können sein:

Die Paradigmenstruktuibedingungen sind also Strukturregularitäten und als solche nicht mit Flexionsregeln zu verwechseln; für Details vgl. Wurzel (1990).

230

-

Wolfgang Ullrich Wurzel

phonologische Eigenschaften, z.B. 'Wortausgang auf -e', vgl. Bote, syntaktische Eigenschaften, z.B. 'Femininum', vgl. Gans, semantische Eigenschaften, z.B. 'Belebt', vgl. Mensch, morphologische Eigenschaften (Eigenschaften der Wortstruktur), z.B. 'Wortstruktur ge+ +e\ vgl. Gebirge.

Da die Paradigmenstrukturbedingungen die Strukturzusammenhänge über die Grenzen der Einzelparadigmen hinaus fixieren, formulieren sie nicht implikative Relationen zwischen den Flexionsformen, sondern solche zwischen den jeweiligen Kategorienmarkera. Auf diese Weise lassen sich dann auch Implikationen zwischen mehreren Markern ein und derselben Flexionsform erfassen. So impliziert im Deutschen bekanntlich der Pluralmarker -er das Auftreten von Umlaut im Plural (vorausgesetzt der Vokal ist umlautbar); vgl. Männer, Wir wollen weiterhin annehmen, daß die Paradigmenstrukturbedingungen grundsätzlich nach dem Defaultprinzip funktionieren. Eine Bedingung spezifiziert nur dann einen Marker, wenn für die entsprechende Kategorie (wie Plural) oder das entsprechende Kategorienbündel (wie G.SG.) noch kein anderer Marker spezifiziert ist; wenn bereits ein Marker für die Kategorie oder das Kategorienbündel spezifiziert und somit die Position des Markers besetzt ist, ist ihre Anwendung blockiert (genaueres dazu weiter unten). Daraus ergeben sich zwei wichtige Konsequenzen. Erstens lassen sich aufgrund des Defaultprinzips nicht nur strikte, sondern auch tendentielle Implikationen im Flexionssystem erfassen. Solche Zusammenhänge sind in den Flexionssystemen sehr häufig, sind für ihren Aufbau geradezu typisch. So haben beispielsweise Nichtfeminina auf -e (außer den Wörtern mit der Struktur ge+ +e) im allgemeinen einen η-Plural; vgl. Bote - Boten, Auge - Augen und Funke - Funken. Es existieren jedoch einige wenige solcher Wörter wie Piefke und Steppke, die abweichend davon i-Plurale aufweisen. Wenn die Paradigmenstrukturbedingungen nach dem Defaultprinzip funktionieren und die Wörter des Typs Piefke im Lexikon als ^-Plurale gekennzeichnet sind, kann dennoch eine entsprechende Bedingung formuliert werden, die Nichtfeminina auf -e den Pluralmarker -n zuweist. Damit haben wir zugleich auf recht plausible Weise die Unterscheidung von markierter und unmarkierter Flexionsklassenzugehörigkeit vorgenommen: Markierte, vom Sprecher intuitiv als weniger normal empfundene Flexionsklassenzugehörigkeit ist für die Wörter lexikalisch spezifiziert, unmarkierte, vom Sprecher als normal empfundene Flexionsklassenzugehörigkeit wird den Wörtern 'automatisch' durch die Paradigmenstrukturbedingungen zugewiesen. Zweitens sichert das Defaultprinzip eine disjunktive Ordnung der Paradigmenstrukturbedinungenen. Wenn also der Marker für eine bestimmte Kategorie oder ein bestimmtes Kategorienbündel durch eine vorangehende Bedingung spezifiziert ist, ist seine Position besetzt und kann dann nicht mehr durch eine folgende Bedingung spezifiziert werden, auch wenn ihre Ausgangsbedingungen erfüllt sind. Auf diese Weise läßt sich das System der Paradigmenstrukturbedingungen redundanzfrei und damit auf eine einfache Weise erfassen, was sicher auch der Strategie des Sprechers entspricht.

Drei Ebenen der Struktur von Flexionsparadigmen

231

Die Bedingungenen beziehen sich (wie gesagt) nicht auf Relationen zwischen den Flexionsformen, sondern auf relationen zwischen den Kategorienmarkern. Das bedeutet, daß wenn in einer Kategorie oder einem Kategorienbündel kein Marker auftritt, auch keine Bedingung angenommen werden muß. Das Nichtauftreten eines Markers braucht normalerweise nicht implikativ eingeführt zu werden. Ausgenommen davon sind lediglich Fälle, wo durch eine 0-Spezifizierung für eine Kategorie/ein Kategorienbündel die Anwendung folgender Paradigmenstrukturbedingungen für die gleiche Kategorie bzw. das gleiche Kategorienbündel verhindert wird. Beispielsweise brauchen durch die Voranstellung der Implikation Venn ein Wort ein Femininum ist, so hat es keinen Marker im G.SG.': (5) [Fem] 3 [ 0 / G . S G . ] in vier folgenden Bedingungenen die Feminina nicht von den jeweiligen Markerzuweisungen für den G.SG. explizit ausgenommen zu werden. Unter den diskutierten Voraussetzungen läßt sich das System der Paradigmenstrukturbedingungen der Substantivflexion des Deutschen einschließlich der entsprechenden lexikalischen Flexionsspezifikationen folgendermaßen formulieren (-V v # steht für Wortausgang auf 'Vollvokal', d.h. Vokal ungleich kurzes -e, phonetisch [È]; -eL# steht für Wortausgang auf kurzes -eplus Sonorkonsonant, d.h. Liquid bzw. Nasal, womit auch die Wörter auf -lein [laen] erfaßt sind): Paradigmenstrukturbedineungen der deutschen Substantivflexion I. Lexikonspezifikationen [n/PL]: [e/PL]: [0/PL]: [s/PL]: [er/PL]: [s/PL]: [n/PL]: p/PL]_ n/PL ns/G.SG 0/A.SG

Typ Villa/Drama, unmarkiert Typ Oma/Teddy/Kino - Omas/Teddys/Kinos Typ Gans, unmarkiert Typ Bahn - Bahnen Typ Mutter, unmarkiert Typ Schwester - Schwestern Typ Band/Single (beides feminin); unmarkiert Typ Bahn/Schwester - Bahnen/ Schwestern Typ Mann/Rind, unmarkiert Typ Wolf/Bein - Wölfe/Beine Typ Park, unmarkiert Typ Wolf Typ Raffke-, unmarkiert Typ Bote Typ Bär/Staat/Muskel, unmarkiert Typ Wolf Typ Untertan; unmarkiert Typ Herold QDoIctor) Einzelfall Käse, unmarkiert Typ Auge/Funke - Augen/Funke Einzelfall Herz, unmarkiert Typ Bein - Beine, schwächer markiert Ohr - Ohren

Wolfgang Ullrich Wurzel

232

[Uml/PL]: Einzelfälle Floß!Lager (neben, umlautlosem Plural), Kloster, unmarkiert Brot/Polster - Brote/Polster [0Uml/PL]: Typ Hund/Sommer, unmarkiert Wolf/Vater II. Pluralsuffixe (a) [-Vv#] => [s/PL]

Typ Oma/Teddy/Kino

(b) [Fem] 3 [n/PL]

Typ Katze/Bahn/Schwester

(c)fpNeut J #ge+

+e#

3 [0/PL]

[[V... eL#]

Typ Gebirge (vs. Geselle/Gemeinde-Gesellen/Gemeinden) Typ Vater/Wagen/Segel/Mädchen/Büchlein

(d) [-e#] 3 [n/PL]

Typ Bote/Auge/Funke

(e) [

Typ Wolf/Bein

] 3 [e/PL]

III. Pluralumlaut (f) [er/PL] 3 [Uml/PL]

Typ Mann/Lamm

(h)

Typ Brot/Polster

[Neut] 3 [0Uml/PL]

Typ Fahrer

[+er#]

Typ Wolf/Gans

(i) [Íe/PL] 3 [Uml/PL]

Typ Vater/Mutter

) [0/PL]

IV. Kasusmarker (j) [Fem] 3 [0/G.SG]

Typ Frau/Oma/Gans

(k) Mask -Bel -e#

[ns/G.SG]

Typ Funke

Drei Ebenen der Struktur von

(1)

n/PL Bei 'V...VKi#

(m) n/PL Bei (n) [

233

Flexionsparadigmen

3 [s/G.SG]

=> [n/G.SG]

] 3 [s/G.SG]

Typ Untertan(/Professor)

Typ Bote/Bär Typen Wolf/Vater/Mann/Park, Staat/Ohr/Auge Typ Bote/Bär

3 [n/D.A.SG] Typ Funke (P) >#]PL]

3 [n/D.PL] Typen

Wolf/Vater/Mann, Gans/Mutter (vs. Bär Bär-en, Bote - Bote-η, Wagen - Wagen; Park/Teddy/Oma - Parks/Teddys/Omas)

Das hier formulierte System von Paradigmenstrukturbedingungen erfaßt die gesamte Flexion der deutschen Substantive mit nativer Wortstruktur, ohne allerdings die neueren «-losen D /A.SG.-Formen bei Wörtern des Typs Bär (dem/den Bär) und die häufigen s-losen G.SGFormen bei Wörtern des Typs Amboß/Iltis (des Amboß/Iltis) zu berücksichtigen. Entsprechende Modifizierungen ließen sich leicht einarbeiten In den Lexikoneinträgen der Wörter sind diejenigen Marker spezifiziert, aufgrund deren sich ihr Flexionsparadigma insgesamt ergibt. Interessant ist in dieser Hinsicht der Einzelfall Käse, im Lexikon als [0/PL] spezifiziert; im übrigen funktioniert das Wort dann völlig regulär: Durch PSB (n) erhält es den Marker -s des G.SG. und durch PSB (p) den Marker -η des D.PL. Das Wort Herz (in seiner traditionellen Flexion die Herzen; des Herzens, dem Herzen, das Herz) hat im Lexikon neben der Spezifizierung [n/PL], wie sie z.B. auch bei Ohr auftritt die beiden zusätzlichen Angaben [ns/G.SG] und [0/A.SG]. Nur der Marker des D.SG. wird durch die PSB (o) spezifiziert. Die Paradigmenstrukturbedingungen für die Pluralsuffixe nehmen ausschließlich Bezug auf unabhängig gegebene phonologische, syntaktische, semantische und morphologische Eigenschaften der Wörter in ihrer Grundform. Die letzte davon benötigt aufgrund der Ordnung der Bedingungen keine Ausgangsspezifizierung mehr; sie gilt für alle verbleibenden Fälle. Für die Paradigmenstrukturbedingungen, die das Auftreten des Pluralumlauts regeln, sind die Pluralsuffixe entscheidend, doch auch hier spielen unabhängige syntaktische und morphologische Eigenschaften der Wörter eine Rolle. Im Zusammenhang mit dem Pluralumlaut wird deutlich, daß die Substantive des Deutschen zwei Positionen ('slots') fur Kategorienmarker haben, eine Suffixposition und eine Vokalwechselposition. Entsprechend gilt hier keine Di-

234

Wolfgang Ullrich Wurzel

stinktivität, die Marker der beiden vorkommenden Typen blockieren sich folglich nicht gegenseitig. Ein Wort wie Mutter, das im Lexikon durch [0/PL] gekennzeichnet ist, was bedeutet, daß es kein Pluralsuffix annimmt, kann sehr wohl Pluralumlaut aufweisen, vgl. Mütter, ein Wort wie Hund, das im Lexikon durch [0Uml/PL] gekennzeichnet ist, kann sehr wohl ein Pluralsuffix aufweisen, vgl. Hunde; und Suffix und Umlaut sind natürlich entsprechend kombinierbar, vgl. Männer und Gänse. Die beiden ersten Gruppen von Bedingungen leiten nicht die Marker fur den N.PL., sondern fur eine einheitliche Pluralform ab, die im N./G./A. und zum einen Teil auch im D.PL. so erhalten bleibt, und zum anderen Teil durch einen Marker des D.PL., genauer des Dativs im Plural., dann weiter modifiziert wird. Die Bedingungen für die Kasusmarker gehen entweder nur von unabhängigen Eigenschaften der Grundform aus (so PSB (j), (k)), nur von vorher abgeleiteten Markern (so PSB (o)) oder von beidem zugleich (so PSB (1), (m)). Die allgemeinste Paradigmenstrukturbedingung für den G.SG. kommt wiederum ganz ohne Ausgangsspezifizierung aus. Von speziellem Interesse aufgrund ihrer Ausgangsspezifizierung ist die PSB (p). Ihre Formulierung, die offenbar die einfachst mögliche ist, geht aus von phonologischen Eigenschaften, aber nicht der entsprechenden Grundform, sondern der Pluralform, aus der der D.PL. additiv abgeleitet wird. Nicht die Form des Pluralmarkers ist entscheidend für das Auftreten des D.PL.-Markers -«, sondern die Phonotaktik der Pluralform: (den) Wagen und (den) Gärten fünktionieren in gleicher Weise wie (den) Boten und (den) Bären. Wenn man die Ordnung der Paradigmenstrukturbedingungenen im System betrachtet, dann zeigt sich, daß sich diese nicht konsequent intrinsisch dem Kriteriums 'Spezielleres vor Allgemeinerem' folgt. Das resultiert daraus, daß die Bedingungen nicht aufgrund eines einheitlichen Korpus von Eigenschaften oder Merkmalen, etwa vergleichbar den phonologischen Merkmalen, operieren. Als 'Merkmale' fünktionieren (wie zu zeigen war) phonologische, syntaktische, semantische und morphologische Eigenschaften der Wörter sowie bereits spezifizierte Marker, so daß in vielen Fällen keine Inklusionsverhältnisse zwischen den Ausgangsspezifizierungen der Bedingungen existieren. Wo jedoch solche Verhältnisse auftreten, konstituieren sie intrinsische Ordnungen. Man vgl. die PSBen (1) und (m): Die Spezifikationen von (m) sind eine Untermenge von (1), deshalb geht PSB (1) 'automatisch' PSB (m) voraus. Das gleiche gilt natürlich für PSB (e) und alle anderen Bedingungen für Pluralsuffixe sowie für PSB (n) und alle anderen Bedingungen für Kasusmarker. Wo solche Relationen nicht gegeben sind, müssen extrinsische Ordnungen angenommen werden, sollen sich die Bedingungen nicht als unangemessen komplex darstellen. Die Paradigmenstrukturbedingungen sind also Generalisierungen über die Struktur der verschiedenen im Flexionssystem vorkommenden Paradigmen, die wiederum für Flexionsklassen bzw. Teilflexionsklassen stehen. Sie betreffen die implikativen Strukturzusammenhänge zwischen den konkreten Kategorienmarkern im Flexionssystem, d.h. sowohl innerhalb der Para-

Drei Ebenen der Struktur von Flexionsparadigmen

235

digmen als auch darüber hinaus. Die Paradigmenstrukturbedingungen erfassen damit die zweite Ebene der Struktur von Flexionsparadigmen.

3. Die kanonische Struktur der Flexionsparadigmen Wenn man die Paradigmenstrukturbedingungen der deutschen Substantivdeklination einschließlich der zugehörigen Lexikonspezifikationen im Zusammenhang betrachtet, so zeigt sich, daß hier durchaus weitere generelle Strukturzüge vorhanden sind. Erstens wollen wir festhalten, welcherart von Flexionsspezifizierungen in den Lexikonrepräsentationen der Wörter überhaupt vorkommen. Das ist leicht auszumachen: Abgesehen von einem Fall sind das bei den hinsichtlich ihrer Flexionsklassenzugehörigkeit markierten deutschen Substantiven auschließlich Pluralmarker. Man vgl. dazu z.B. die Typen Villa/Drama mit [η-PL], Mann/Rind mit [er/PL] und das Wort Käse mit [0/PL]. Auch letzteres paßt sich also, obwohl es ein Einzelfall ist, entsprechend in das Flexionssystem ein. Dagegen erscheint beim Wort Herz nicht nur der Pluralmarker, sondern es treten auch Kasusmarker im Lexikon auf, was ansonsten nie der Fall ist. Beide Einzelwörter unterscheiden sich darin, daß Käse wie Villa oder Mann ein regulär-markierter Fall ist, während Herz eine echte Ausnahme im Flexionssystem bildet. Zweitens wollen wir zusammenfassen, von welchen unabhängig gegebenen Eigenschaften die Paradigmenstrukturbedingungen im einzelnen Gebrauch machen, d.h. welche phonologischen, syntaktischen, semantischen und morphologischen Eigenschaften der Wörter deren Flexionsweise beeinflussen. Von den Bedingungen wird insgesamt auf fünf verschiedene phonologische Strukturen der Wörter Bezug genommen, auf vier der Grundform, nämlich -Vv#, -e#, V. ..eL# und -'V. . .VKi#, und eine der Pluralform, nämlich -n# bzw. -s#.6 Die Bedingungen betrachten also die phonologische Wortstruktur ausschließlich vom Wortausgang her. Ob ein Wort Anfangsbetonung hat oder nicht, spielt z.B. überhaupt keine Rolle. Das ist für Suffixsprachen wie das Deutsche sicher nicht ungewöhnlich, aber auch eben nicht trivial. Die deutsche Verbflexion nimmt bekanntlich Bezug sowohl auf die Anfangsbetonung (relevant für das Auftreten von ge- im PART.PERF.: 'übersetzen - übergesetzt vs. über'setzen - übersetzt) als auch auf auf den Stammvokal (relevant für die Spezifizierung der unterschiedlichen Ablautvokale). Was die syntaktischen Eigenschaften betrifft, so ist faktisch das gesamte System strukturiert durch die Distinktion von Femininum und Nichtfemininum. Hingegen spielt die Unterscheidung von Maskulinum und Neutrum nur (noch) eine untergeordnete Rolle. Sie ist relevant

6

An dieser Stelle sei wenigstens erwähnt, daß die Form des D.PL. im Deutschen anders als die übrigen Flexionsformen eine formale 'Weiterbildung' der Pluralform darstellt; vgl. Hund - Plural Hund-e - D.PL. Hund-e-n, Kind - Plural Kind-er - D.PL. Kind-er-n. Hier liegt damit faktisch eine agglutinierende Struktur vor, vgl. dazu türkisch türk - Plural tiirk-ler - D.PL. tiirk-ler-e.

236

Wolfgang Ullrich Wurzel

für die Zuweisung des Pluralumlauts bei den e-/0-Pluralen: unmarkierte Neutra zeigen keinen Umlaut, vgl. Brote, Polster, unmarkierte Maskulina zeigen Umlaut, vgl. Wölfe, Väter, und in zwei Fällen für die Suffixspezifizierung: nichtbelebte Maskulina auf -e haben einen G.SG. auf ns, vgl. (des) Funkens, entsprechende Neutra einen G.SG. auf -s, vgl. (des) Auges-, Neutra mit der Wortstruktur 'ge+_ +e' haben einen 0-Plural, vgl. (die) Gebirge, entsprechende Maskulina einen «-Plural, vgl. (die) Gefährten. In den Paradigmenstrukturbedingungen tritt eine einzige semantische Distinktion auf, die zwischen Belebt und Unbelebt. Sie ist relevant bei der Spezifizierung der Kasusmarker für Nichtfeminina mit «-Plural, vgl. (des) Funkens (nach PSB (k)), (des) Untertans (nach PSB (1)), (des) Boten, (des) Bären (nach PSB (m)) und (des) Staates, (des) Auges (nach PSB (n)). Auch zwei morphologische Wortstrukturen (als solche aufgrund der Relevanz der Morphemgrenzen von den phonologischen Wortstrukturen unterscheidbar) spielen für die Flexionszuweisung eine Rolle, eine die sich wiederum auf den Wortausgang bezieht: +er#, und eine, die sich auf die Gesamtstniktur des Wortes bezieht: #ge+ +e#. Darüber hinaus weisen im allgemeinen auch Wörter mit anderen Derivationsuffixen eine einheitliche Flexion auf (vgl. Altertum - Altertümer), worauf wir nicht näher eingehen wollen. Noch ein kurzer Blick auf die Kombinationen von unterschiedlichen Eigenschaften, die für die Auswahl der unterschiedlichen Marker eine Rolle spielen. Hier ist zumindest ein Fall bemerkenswert, nämlich die Kombination von Maskulinum, Wortausgang auf -e und Nichtbelebtheit (in PSB (k)), die wenige Wörter wie Funke, Name und Same betrifft. Sie ist für das Deutsche höchst untypisch, da Maskulina auf -e in der Überzahl der Wörter belebt sind; man vgl. Bote, Hase, Sachse, Preuße, aber auch viele nichtnative Wörter wie Chinese, Kanute, Chaote und Pantomime. Damit ist die Substantivdeklination des Deutschen bezüglich der genutzten unabhängigen Eigenschaften der Wörter ziemlich inkohärent; viele andere Flexionssysteme sind in dieser Hinsicht viel kohärenter. So basiert beispielsweise die russische Substantivdeklination nahezu ausschließlich auf dem phonologischen Wortausgang, nur je einmal spielen zusätzlich die Distinktionen Belebt vs. Unbelebt, vgl. vrac'Arzt' - A.SG. vraöa vs. lue 'Strahl' - A.SG. lue, und Maskulin vs. Feminin, vgl. dozd' 'Regen' - G.SG. dozd'a vs. tetrad' 'Heft' - G.SG. tetrodi, in den Paradigmenstrukturbedingungenen eine Rolle. Drittens wollen wir uns die Richtung der in den Paradigmenstrukturbedingungen fixierten Implikationen noch einmal etwas näher ansehen. Es ist leicht festzustellen, daß die des deutschen Substantivs eine einheitliche Ableitungsrichtung haben. Das ist keinesfalls trivial. Die Bedingungen führen, ganz unabhängig davon, welche konkrete phonologische Form die jeweils spezifizierten Kategorienmarker haben, von der Grundform, dem N.SG., zum Plural und von dort einerseits zum G.SG. und weiter zu den anderen Singularkasus und andrerseits zum D.PL. Die Relationen zwischen den Markern sind nicht in allen Fällen die gleichen: Die Implikationen vom N.SG. zum Plural haben meist Defaultcharakter, wenn es davon auch bestimmte Ausnah-

Drei Ebenen der Struktur von Flexionsparadigmen

237

men mit strikter Implikation wie den Typ Gebirge gibt.7 Die weiterführenden Implikationen zu den Kasusmarkern und zwischen diesen gelten dagegen jeweils strikt. Von diesen hat wiederum die Relation zwischen den Markern des A.SG. und des D.SG. einen besonderen Status. Wenn hier in einem Kasus ein Marker auftritt, dann tritt derselbe Marker auch im jeweils anderen auf, und wenn in einem Kasus kein Marker auftritt, dann auch im jeweils anderen; d.h. es herrscht Formengleichheit (diese Relation gilt natürlich erst seit dem Abbau des Markers -e im D.SG.; man beachte auch die erwähnte Ausnahme Herz - D.SG. (dem) Herzen - A.SG. (das) Herz). Insgesamt gesehen stellt sich die kanonische Paradigmenstruktur des deutschen Substantivs dann folgendermaßen dar ('M' steht für den jeweiligen Marker, der auch gleich 0 sein kann): (6)

N.SG.

Mi/PL.

Mj/G.SG.

Mm/D.PL.

I Mk/A.SG.

Μ,/D.SG. Der Einfachheit halber wird hier nicht zwischen Pluralsuffix und Pluralumlaut differenziert; 'Mj/PL' steht also für beides. Wie bekannt müssen nicht alle Positionen in dieser Struktur durch Marker besetzt sein. So erscheinen in verschiedenen Paradigmen keine Kasusmarker im D./A.SG. (dem/den Wolf), im ganzen SG. (der/der/die Gans), im D.PL. (den Boten), im D./A.SG. und im D.PL. (dem/den Garten, den Gärten) und in sämtlichen Kasusformen (der/der/die Katze, der/den/die Katzen). Dagegen ist es nicht möglich, daß im G.SG. kein Marker, aber im D./A.SG. ein Marker auftritt. Die Abfolge der Kasusmarker kann also nicht unterbrochen werden. Einen speziellen Status hat in dieser Hinsicht jedoch der Plural. Wir haben oben festgestellt, daß die Pluralform die Kennform der Substantivparadigmen ist. Das gilt auch, wenn diese ohne Marker gebildet ist: Ein 0-Plural impliziert einen G.SG. auf -s und ei-

7

Faktisch gilt eine solche strikte Implikation auch für die Feminina auf -e vom Typ Katze (vgl. Abschn. 1). Doch die Formulierung einer entsprechenden Paradigmenstniktuibedingung erübrigt sich, da fur alle Feminina außer denen auf 'Vollvokal' die n-Pluralbildung ohnehin den Defaultfall darstellt, was durch PSB (b) erfaßt wird.

Wolfgang Ullrich Wurzel

238

nen D.PL. auf -n, er bewirkt also keine Unterbrechung der implikativen Ableitbarkeit der Formen. Die implikativen Relationen zwischen den Kategorienmarkern ungeachtet ihrer konkreten phonologischen Form lassen sich wie folgt durch kanonische Paradigmenstrukturbedingungnen (KPSB) formulieren: Kanonische Paradigmenstrukturbedingungen der deutschen Substantivflexion (i) GF =)dft Mi/PL (ii) Mi/PL 3 Mj/G.SG (iii) Mj/G.SG 3 Mk/A.SG (iv) Mk/A.SG o

Mi/D.SG

(v) M¡/PL 3 Mm/D. PL Die kanonischen Paradigmenstnikturbedingungen fixieren die dritte Ebene der Struktur von Flexionsparadigmen. Es ist theoretisch und praktisch von Bedeutung, daß die (einfachen) Paradigmenstnikturbedingungen durchaus nicht in allen Flexionssystemen einer generellen implikativen Ableitungsrichtung folgen wie in der Substantivflexion des Deutschen. In solchen Fällen existiert dann keine dritte Ebene der Struktur der Flexionsparadigmen. Das gilt beispielsweise für die lateinische Substantivflexion, wo im Bereich der /-Deklination (Typ puppis 'Achterdeck'), der konsonantischen Deklination (Typ rêx 'König') und ihrer drei 'Mischformen' (Typen ignis 'Feuer', auris 'Ohr' und civis 'Bürger') die Ableitungsrichtung teilweise vom A.SG. über ABL.SG. und A.PL. zum G PL. und teilweise genau gegenläufig vom G PL. über A.PL. und ABL.SG. zum A.SG. verläuft. Doch auch in der deutschen Verbflexion lassen sich einerseits aus dem Auftreten der 'starken' Charakteristika des Vokalwechsels und der Suffixlosigkeit im IMP.SG. bzw. des Vokalwechsels in der 2./3.SG.PRÄS. IND. (vgl. helfen/saufen: hilf/sauf bzw. hilft/säuft) alle anderen 'starken' Kategorienmarker bis hin zu den Markern des PART.PERF. (Suffix -en und Ablaut, geholfen/gesoffen) implizieren, während andererseits aus dem 'schwachen' Marker des PART.PERF. (Suffix -t: gesagt) alle anderen Marker der 'schwachen' Verben bis hin zum IMP.SG. (Suffix -e: sage) implizierbar sind, aber eben jeweils nicht umgekehrt.8 Hier gibt es also strukturell-typologische Unterschiede im Aufbau von Flexionssystemen.

g

Vgl. dazu die ausführliche Darstellung dieser Problematik in Bittner (1996).

Drei Ebenen der Struktur von Flexionsparadigmen

239

4. Paradigmenstruktur und Wandel In den vorangehenden Abschnitten sind wir anhand des Beispiels der Substantivflexion des Deutschen der Frage nachgegangen, wie Flexionsparadigmen und darauf aufbauend Flexionssysteme strukturiert sind bzw. strukturiert sein können. Ausgehend von der Struktur der Einzelparadigmen haben wir zwei weitere Ebenen der Paradigmenstruktur ermittelt, die durch die Formulierung von entsprechenden Regularitäten, d.h. Paradigmenstrukturbedingungen und kanonischen Paradigmenstrukturbedingungen spezifiziert wurden. Wenn diese Regularitäten korrekt sind, sollten sie nicht nur die Paradigmen und das Flexionssystem insgesamt synchron strukturieren, sondern sollten zugleich erfassen, wie sich die Paradigmen im Rahmen des Systems diachron verändern. Im folgenden soll anhand einiger einschlägiger Beispiele gezeigt werden, auf welche Weise das geschieht. Betrachten wir zunächst Veränderungen in der Flexion von Wörtern, die die Ebene der Paradigmenstrukturbedingungen einschließlich der dazugehörigen lexikalischen Spezifikationen betreffen. Erstens. Wenn fur die Flexionsklassenzugehörigkeit unabhängige phonologische, syntaktische, semantische und morphologische Eigenschaften der Wörter in der angenommenen Weise verantwortlich sind, sollte eine Veränderung von relevanten Eigenschaften tendentiell auch zur Veränderung der Flexion fuhren. Solche Veränderungen sind in der Tat recht häufig zu beobachten. - Wenn Kollektiva des Typs Gebirge das auslautende -e der Grundform verlieren wie u.a. Gebüsch, Gestell und Genick, dann gilt für sie nicht mehr die PSB (c), und sie werden folgerichtig von PSB (e) erfaßt, aus (das) Gebüsche - (die) Gebüsche wird (das) Gebüsch - (die) Gebüsche. - m Deutschen kommen bestimmte feminine Wörter wie Kartoffel und Trüffel, die normalerweise ihre w-Pluralmarker durch PSB (b) zugewiesen bekommen, in neuerer Zeit auch als Maskulina vor (im Fall von Trüffel auch vom Duden als 'umgangssprachlich' akzeptiert). Sie bilden dann entsprechend PSB (c) die Pluralform suffixlos; vgl. (der) Trüffel - (die) Trüffel. - Das Wort Bär ist ein belebtes Nichtfemininum mit «-Plural und hat demzufolge auch im G.SG. den Marker -η: (des) Bären. Durch Komposition wurde aus dem Wort das nichtbelebte Substantiv Rammbär abgeleitet, das nach PSB (n) dann den G.SG. (des) Rammbärs zeigt. Gerade andersherum wird im moderneren Sprachgebrauch das Wort Typ, ursprünglich nur unbelebt und mit s-G.SG. nach PSB (n), häufig auf Personen angewandt und zeigt dann eine starke Tendenz zu den Formen des, dem, den Typen gemäß den PSBen (m) und (o). Zweitens. Wenn die lexikalische Spezifizierung der Flexionsklassenzugehörigkeit durch die Angabe von Pluralmarkern markiert und die NichtSpezifizierung unmarkiert ist, dann sollten solche lexikalischen Angaben zum Abbau tendieren und die Wörter in die für sie unmarkierte

240

Wolfgang Ullrich Wurzel

Klasse übertreten. In der Geschichte des Deutschen und auch in der Gegenwart gibt es viele solche Klassenübergänge. Dafür nur zwei Beispiele: - Maskulina wie Blitz, Hahn, Mond und Schwan, jünger auch Hagestolz und Pfau, haben ihre lexikalische Spezifizierung als «-Plural (vgl. noch Hahnenkampf, Mondenschein usw.) aufgegeben und entsprechend PSB (e) den für sie unmarkierten e-Plural angenommen; vgl. Blitze, Hähne usw. - Viele Feminina haben im Laufe der Sprachgeschichte ihre lexikalische Spezifizierung als ePlurale, aus der nach PSB (i) Umlaut folgt, abgebaut, allein in diesem Jahrhundert völlig bzw. weitgehend u.a. die Wörter Flucht, Gruft, Sucht und Zucht. Sie erhalten die für sie unmarkierte Flexion mit »-Plural durch PSB (b) zugewiesen; vgl. (die) Fluchten usw. Drittens. Wenn im Lexikon eine große Anzahl von Wörtern für ihre Flexionsklassenzugehörigkeit spezifiziert sind und diese haben zufallig eine gemeinsame unabhängig gegebene Eigenschaft, die für die Fixierung der Klassenzugehörigkeit bisher nicht genutzt wurde, dann ergibt sich die Tendenz zur Nutzung ebendieser Eigenschaft mit den entsprechenden Klassenübertritten. Auch dafür gibt es Beispiele beim deutschen Substantiv: - Im frühen Althochdeutschen wurden die auf -a endenden Feminina teilweise als ô -Stämme mit dem N.PL.-Marker -â wie èra - Ν.PL. ërâ und teilweise als «-Stämme mit dem N.PL Marker -ûn wie zunga - N.PL. zungün flektiert. Die beiden Klassen sind ungefähr gleichstark belegt. In einem solchen Fall macht es keinen Sinn, eine davon als unmarkiert und die andere als markiert zu betrachten. Das wird dadurch bestätigt, daß es zunächst auch keine ein heitliche Übergangsrichtung von einer in die andere Klasse gibt. Die Wörter der beiden Klassen sind im Lexikon als â- bzw. ««-Plurale spezifiziert, d.h. sie sind beide gleichermaßen markiert. Eine solche Konstellation ist, auch crosslinguistisch gesehen, ziemlich ungewöhnlich und belastet das Lexikon stark. Hier setzen entsprechende Veränderungen zum Abbau von Markiertheit ein. Der Ausgangspunkt dafür ist eine zufällige ungleichmäßige Verteilung von Konkreta und Abstrakta auf die beiden Flexionsklassen. In der «-Deklination überwiegen die Konkreta stark, u.a. gehören auch alle femininen Personenbezeichnungen auf -a wie diorna 'Jungfrau', quena 'Frau' und wituwa 'Witwe' in diese Klasse. Hingegen überwiegen in der ô-Klasse die Abstrakta. Unter diesen Voraussetzungen kann damit die Eigenschaft der Konkretheit/Abstraktheit für die Fixierung der Flexionsklassenzugehörigkeit genutzt werden, eine semantische Eigenschaft, die bisher im Flexionssystem überhaupt keine Rolle gespielt hat. Als normal, also unmarkiert, werden die konkreten Substantive des Typs zunga und die abstrakten Substantive des Typs èra gewertet, die damit jeweils keine Flexionsklassenspezifizierungen im Lexikon mehr benötigen. Das betrifft dann den Großteil aller Feminina auf -a. Die damit verglichen recht kleinen Teilklassen der Abstrakta der «-Deklination vom Typ fasta 'das Fasten' und der Konkreta der ôDeklination des Typs wamba 'Bauch' bilden die jeweils weniger normalen, d.h. markierten Teilklassen, deren Wörter im Lexikon entsprechende Flexionsangaben haben. Wie nicht anders zu erwarten hat diese Konstellation zahlreiche Klassenübertritte zur Folge. Im Mittelhochdeut-

Drei Ebenen der Struktur von Flexionsparadigmen

241

sehen (wo das auslautende -e zu -a geworden ist) sind dann vor allem viele ursprüngliche ôKonkreta wie brücke, erde und strâje in die «-Klasse übergetreten. Wie stark diese Tendenz zum Klassenwechsel ist, zeigen speziell als Personenbezeichnungen gebrauchte Abstrakta wie ère 'Ehrerbietung, Zierde' und minne '(freundliches) Gedenken, Liebe', die mittelhochdeutsch konsequent nach der «-Deklination flektiert werden. Sehen wir uns jetzt noch einige Veränderungen in der Flexion von Wörtern, die die Ebene der kanonischen Paradigmenstrukturbedingungen einschließlich der Typen von lexikalischen Spezifizierungen betreffen. Erstens. Wenn es im Flexionssystem Wörter gibt, die einer oder mehreren der kanonischen Paradigmenstrukturbedingungen nicht entsprechen und infolgedessen unsystematische, regulär nicht zugelassene Flexionsspezifizierungen im Lexikon aufweisen, so sollte die Flexion dieser Wörter gemäß diesen Bedingungen verändert werden. - Das ist der Fall beim bereits mehrfach erwähnten Einzelwort Herz mit der Flexion die Herzen, aber des Herzens, dem Herzen, das Herz. In diesem Paradigma ist zunächst der Marker ns des G.SG. nicht entsprechend der KPSB (ii) implikativ aus dem Pluralmarker -« zu gewinnen. Der G.SG.-Marker muß im Lexikon angegeben werden. Spezifiziert sein muß im Lexikon weiter der 0-Marker des A.SG., der nicht nach KPSB (iii) aus dem G.SG. abgeleitet werden kann. Damit widerspricht die Flexion des Wortes zugleich auch der KPSB (iv), die die Formengleichheit von A.SG. und D.SG. festlegt. Entsprechend gibt es eine starke Tendenz zur Anpassung der Flexion an die kanonischen Paradigmenstrukturbedingungen, d.h. an die Tilgung der irregulären, weil sonst nicht vorkommenden Spezifizierungen von Kasusmarkern im Lexikon. Gegenwärtig ist eine entsprechende Veränderung zu beobachten. Ihr Ergebnis ist, daß Herz heute schon überwiegend ganz wie Ohr flektiert, wohl weiterhin markiert fur den nPlural, aber nicht mehr im Widerspruch zu den kanonischen Paradigmenstrukturbedingungen; vgl. Herz, die Herzen, des Herzes, dem/das Herz. Zweitens. Wenn es im Flexionssystem eine Flexionsklasse gibt, deren PSBen von Eigenschaften bzw. von Kombinationen von Eigenschaften ausgehen, die der Systematik des Systems nicht entsprechen, d.h. in dieser Hinsicht markiert sind, so sollte diese Klasse abgebaut werden. Hier findet sich zumindest ein einschlägiges Beispiel: - Wie bereits festgestellt ist die Kombination der Eigenschaften Maskulin, Endung auf -e und Nichtbelebt beim deutschen Substantivs sehr untypisch, markiert; bei Maskulina signalisiert die Endung auf -e normalerweise Belebtheit. In der Sprachgeschichte sind schon früher Substantive mit dieser Kombination von Eigenschaften abgebaut worden, entweder durch Veränderung des Wortausgangs wie bei mittelhochdeutsch balke, garte, slite > neuhochdeutsch Balken, Garten, Schlitten bzw. durch Überfuhrung zum Femininum wie bei mittelhochdeutsch der bluome, brille, slange > neuhochdeutsch die Blume, Brille, Schlange.9 So tendieren auch die wenigen 9

Auf diese Weise ergeben sich auch die bekannten Minimal pairs des Typs Franken - Franke und Lappen Lappe.

242

Wolfgang Ullrich Wurzel

noch verbliebenen nichtbelebten Maskulina auf -e wie Funke, Name und Same zur Beseitigung der markierten Konstellation von Eigenschaften durch Übernahme der D./A.SG.-Form in den N.SG.; vgl. Funken, Namen und Samen. Damit fallen die PSB (k) und der zweite Teil der PSB (o) weg; das Flexionssystem wird einfacher und systematischer. Die betroffenen Wörter flektieren dann wie der Typ Wagen entsprechend den PSB (c) und (n): (der) Funken - (des) Funkens usw. Hier wird damit also eine der Systematik widersprechende Flexionsklasse abgebaut, obwohl ihre Flexion aufgrund unabhängiger Eigenschaften voll prädiktabel ist. Drittens. Wenn Substantive ihre Flexion verändern, dann sollte das entsprechend der von den kanonischen Paradigmenstrukturbedingungen festgelegten implikativen Hierarchie geschehen. Auch dafür ein Beispiel: - Es wurde bereits festgestellt, daß aufgrund von Markiertheitsabbau hinsichtlich der Flexionsklassenzugehörigkeit Maskulina wie Blitz, Mond und Pfau zur e-Pluralbildung übergetreten sind; vgl. Blitze, Monde, Pfaue. Da der Pluralmarker den G.SG.-Marker impliziert, nehmen solche Wörter auch den entsprechenden G.SG.- Marker -s an. (des) Blitzes, Mond(e)s, Pfaus. Es gibt desweiteren eine zweite Gruppe von Fällen wie Untertan, Vetter und Nachbar, die den G.Sg.-Marker -s angenommen hat, ohne ihren «-Plural aufzugeben (nach PSB (1)); vgl. (des) Untertans, Vetters, Nachbars, aber (die) Untertanen, Vettern, Nachbarn. Die dritte Möglichkeit des Übergangs wäre hier, daß Maskulina mit -n im Plural und im G.SG. einen e-Plural annehmen, aber den G.SG. auf -n beibehalten. Doch dieser theoretisch mögliche Wandel tritt, anders als die beiden anderen, nicht ein. Er kann aufgrund der durch die kanonischen Paradigmenstrukturbedingungen festgelegten implikativen Hierarchie nicht eintreten, da ein bestimmter Pluralmarker immer auch einen bestimmten G.SG.-Marker impliziert. So hat eben ein ePlural bei den Nichtfeminina 'automatisch' auch immer den Marker -s im G.SG. zur Folge. Damit ist die Herausbildung einer theoretisch mögliche Flexionsklasse im Übergangsfeld von den 'schwachen' zu den 'starken' Maskulina ausgeschlossen; vgl.: (7) e/PL. η/PL.

s/G.SG. Hund Mond [< -en, -en] Staat, Untertan [< -en, -eri\

η/G.SG. 0 Mensch

Soviel zur Struktur der Flexionsparadigmen im Rahmen des Flexionssystems und ihren Konsequenzen für den Sprachwandel. Es bleibt noch die zusammenfassende Feststellung, daß die Flexionsparadigmen, wie zu zeigen war, keine bloßen Konstrukte des Linguisten sind, sondern realiter existieren. Entsprechend sind auch die einzelnen Regularitäten der Paradigmenbildung, die Paradigmenstrukturbedingungen, real, sind Bestandteil des Lexikons der Sprechergramma-

Drei Ebenen der Struktur von Flexionsparadigmen

243

tik. Eine wissenschaftliche Grammatik, die die Grammatik der Sprecher in angemessener Weise rekonstruieren will, muß diese Regularitäten - in welcher Form auch immer - widerspiegeln.

Literatur Bittner, Andreas (1996) Starke 'schwache' Verben - schwache 'starke' Verben. Deutsche Veibflexion und Natürlichkeit. Tubingen: Stauffenburg (= Studien zur deutschen Grammatik 51). Duden (1991) Rechtschreibung der deutschen Sprache, 20. Auflage. Mannheim/Leipzig/ Wien/ Zürich. Fuhrhop, Nanna (1997) Von Stämmen und Wurzeln. Dissertation. Berlin. Wurzel, Wolfgang Ullrich (1984) Flexionsmorphologie und Natürlichkeit. Berlin: Akademie-Verlag. (= Studia grammatica XXI). - (1987) Paradigmenstrukturbedingungen: Aufbau und Veränderung von Flexionsparadigmen. In: A.G. Ramat et al. (eds.), Papers from the 7th Conference on Historical Linguistics, Pavia 1985, 629-644. Amsterdam: Benjamins. - (1990) The mechanism of inflection. - In: W.U. Dressier et al. (eds.), Contemporary Morphology, 203-216. Berlin/New York: Mouton de Gruyter. - (1994) Gibt es im Deutschen noch eine einheitliche Substantivflexion? Oder: Auf welche Weise ist die deutsche Substantivflexion möglichst angemessen zu erfassen? - In: K.M. Köpcke (Hg.), Funktionale Untersuchungen zur deutschen Nominal- und Verbalmorphologie, 29-44. Tübingen: Niemeyer (= Linguistische Arbeiten 319)

Martin Neef - Universität zu Köln

The Reduced Syllable Plural in German

Among the morphological phenomena of the German language, the formation of the nominal plural is one of the fiercly discussed areas which appears to be an insoluble miracle. This becomes especially evident in the attempt of learning grammars to teach German as a foreign language, as the following pessimistic assessment by Kufner (1962:56) indicates: „There is no way in which the plural formation of German nouns can be made easy. The best we can tell our students is that, after they have learned the commonest one thousand nouns or so, they will find that most other masculines and neuters add -e and most other feminines -(e)n. This is small comfort to a student who is just starting to learn the commonest one thousand."

Since then, scientific advance might have deepened the understanding of this matter, but even in present days, Bauer (1996:252) supposes that „the plural stem for most German nouns has to be lexically listed." The aim of this paper is to develop an analysis of the formal properties of the nominative plural in an a-morphous and declarative model of morphology called 'Word Design' (cf. Neef 1996a), which allows to take into account especially prosodie constellations that are mostly ignored in the literature.1 This will lead to a better understanding of the mental representation of the plural system. As a result, I will show that plural inflection is organized differently from what has been assumed traditionally. Finally, I will sketch a simple default-based system concerning the learnability of the nominative plural. The analysis will be based on the availability of gender information. This assumption, however, does not deny the existence of formal and semantic regularities that enable to predict gender, at least to a certain extent, as Zubin and Kopeke (1986) and Kopeke and Zubin (1996) have argued for.

1. Prosodie Properties of Plural Forms Grammars of German typically list five regular forms of plural formation, named after the five suffixes zero, schwa (or -e), -er, -(e)n, and -s. Zero, schwa and -er can be combined with umlaut.2 The brackets in -(e)n indicate that this suffix has two realizations. Usually, this difference between -en and -n is attributed to somewhat unspecified morphophonemic operations. Taking

2

An example for such an approach to the German plural that ignores prosodie information is the otherwise very instructive and detailed study by Köpcke (1993). See Mugdan (1977:108f.) for a survey of the number of supposed regular plural classes of linguistic analyses which ranges from 3 to 8 (ignoring the s-plural).

The Reduced Syllable Plural in German

245

a prosodie perspective, I will claim that the first four suffixes are markers of one general plural class and, moreover, that the zero plural and the schwa plural constitute one and the same pattern. The first point reflects the fact that for a long time only these plurals were regarded as elements of the regular German plural system (cf. Rettig 1972:99ff ), the latter point is widespread in theoretical analyses,3 but it is typically not reflected in grammars of German. In what follows I will therefore argue that there are three distinct types of plural forms in Modern Standard German (MSG): the regular type indicated by the four suffixes mentioned above, the j-plural type, and one heterogeneous class of irregular plural nouns. Regular German plurals must end in a specific syllable type, usually referred to as 'schwa syllables' (cf. e.g. Wiese 1996:61). Since many of these so-called schwa syllables contain no schwa (at least not on the surface in which I am interested primarily) as can be seen in Haketi 'hook' in the standard pronunciation [ha:.krj| (besides formal [ha:.kan]) or m Bruder 'brother' in the only possible pronunciation [bRu:.de], I prefer to call this syllable type 'reduced syllables' (following Vennemann 1991). Reduced syllables are defined by their syllable peak (i.e. the most sonorous element of a syllable): reduced syllables are centered around the vowel schwa or a (syllabic) sonorant, whereas their counterpart — which are called 'full syllables' — are centered around a full vowel. A property corresponding to this distinction concerns stress: whereas full syllables always bear stress, at least to a certain extent, reduced syllables are never stressed at all. Orthographically, reduced syllables are always represented by the letter (which, by the way, can also indicate a full vowel). Within the framework of Prosodie Morphology, Wiese (1996:106) criticizes the notion of 'pseudo-suffix' which is employed in work by Augst (1979), Kopeke (1988, 1993) and others to cover the similar behaviour of the stem endings -el, -en, and -er in words like Vogel 'bird', Wagen 'car', and Vater 'father' in that they (and almost only they) allow a zero plural. Wiese argues that the notion of pseudo-suffix can be derived from the independently motivated notion of prosodie foot and claims that German (regular) plurals must end in a bisyllabic foot with the second syllable being a schwa syllable. Wiese's analysis gives rise to at least two empirical problems: First, his notion of prosodie foot is much broader than the notion of pseudo-suffix since it also covers obstruent-final reduced syllables like in Kirmes 'kermis' as well as schwafinal reduced syllables like in Hase 'rabbit'. Remarkably, nouns with these endings typically do not allow a zero plural.4 Employing the notion of bisyllabic foot, the different behaviour of the pseudo-suffixes and the other reduced syllables has to be explained on independent grounds (Wiese incidentally ignores this point). I will come back to this issue in section 3.

3

4

See for example Curme (1922:70, 80), Steche (1927:129), Kufner (1962:55), Bech (1963:179), Werner (1969:97), Rettig (1972:25f.), Mugdan (1977:85), Carstairs (1987:238), Russ (1989:59), Wurzel (1994:34), Wegener (1995:18), and Golston and Wiese (1995:156). Cf. (8) below for some exceptions.

Martin Neef

246

Second, in Neef (1996a:62) I gave the following simple and complex nouns ending in a trisyllabic or quadrisyllable metrical foot in the plural as counterevidence to Wiese's bisyllabic foot hypothesis. The criticism is based on Wiese's definition of foot (1996:56) as a sequence of one or more syllables with the first one carrying stronger stress than the subsequent ones:

(1)

a.

b.

singular 'Arbeit 'Heimat 'Plural 'Predigt 'Abfahrt 'Ablösung 'Bahnhof 'Lehrling

plural 'Arbeiten 'Heimaten 'Plurale 'Predigten 'Abfahrten 'Ablösungen 'Bahnhöfe 'Lehrlinge

gloss 'work' 'home' 'plural' 'sermon' 'descent' 'detachment' 'railway station' 'apprentice'

In Neef (1996a:62f), I argued that the main factor combining the regular plural forms is not a final bisyllabic foot, but exactly one reduced syllable at the end of the word. Nevertheless, there exist some systematic counterexamples to this claim: (2)

a.

b. c. (3)

a. b.

(4)

(der) Abend (die) Gegend (die/das) Tausend (die) Tugend (die/das) Hundert (die) Kirmes

Abende Gegenden Tausende/n Tugenden Hunderte/n Kirmessen

evening 'region' 'thousand' 'virtue' 'hundred' 'kermis'

(der) Versicherer (der) Zauberer (der) Tausender

Versicherer Zauberer Tausender

insurer 'magician' (noun connected with the number thousand, like a bill or a mountain)

(die) Sieben

Siebenen (or. Sieben) '(the) seven'

Singular nouns ending in an obstruent-final reduced syllable like in (2) are extremely rare in MSG. Interestingly, the corresponding plural forms end without exception in two reduced syllables.5 The same is true for nouns already ending in two reduced syllables in the singular like the ones in (3). The items in (a) represent a productive pattern of agent nominalization to verbs ending orthographically in -em like versichern 'to insure'. The noun in (4) is the only case of a feminine noun ending in -en in the singular and, moreover, it is the only sonorant-final noun with one final reduced syllable in the singular and two in the plural (at least as one possibility; according to Duden (199621), both options listed in (4) are occurring). These data indicate that regular plurals may end in one or two reduced syllables.

Some studies falsely claim that plurals with two final reduced syllables are principally ungrammatical and non-occurring. See for example Mugdan (1977:86), Wegener (1995:19), and Golston and Wiese (1995:151).

The Reduced Syllable Plural in German

247

2. Word Design: A Declarative Model of Morphology The analysis I will present is based on the theory of Word Design (Neef 1996a) which is a declarative model as it takes language competence to be more adequately described by invoking constraints for surface forms than by rules transforming stipulated deep structures (or underlying forms) to perceptable surface structures. In its constraint-based character, Word Design is similar to other declarative approaches such as Optimality Theory (OT, Prince and Smolensky 1993) and Declarative Phonology (DP, Scobbie, Coleman and Bird 1996). In other respects, Word Design differs from these theories: First, following e.g. Becker (1990) and Anderson (1992), Word Design is 'a-morphous' in that it assumes a fundamental difference between stems on the one hand and (derivational and inflectional) affixes on the other. Whereas stems are considered to be the formal part of lexical entries, affixes are reconstructed as wellformedness conditions (i.e. constraints) in the morphological component. Second, in accordance with DP but contra OT, constraints in Word Design are considered unviolable and, consequently, unordered. Third, as opposed to DP and OT, certain constraints do not hold for the whole vocabulary of a given language but only for a morphologically defined subclass.6 These morphological constraints (termed 'design conditions') are at the heart of Word Design. A further central assumption in this approach is that morphological categories in the sense of Becker (1990:66) (i.e. specific derivational and inflectional patterns) are not exclusively characterized by certain phonological segments (usually conceived as morphemes) but also by other formal aspects which are mostly prosodie in nature. The prosodie constellations that are relevant for specific morphological categories are also modelled as design conditions. DP and OT, on the other hand, try to minimize the morphological (and presumably language-specific) component in favour of (preferably universal) phonological constraints. Design conditions are formulated relative to the Main Principle of Morphology (cf. Neef 1996a:285ff.) which is assumed to be universal — in contrast to design conditions which are language-specific and, moreover, specific for morphological categories: (5)

Main Principle of Morphology Grammatical words are ideally identical with the phonological realization of their base.

This principle states that a dependent grammatical word (which are surface forms following Aronoff 1994) must be as similar to its base as possible, This guarantees a pragmatically relevant connection between a grammatical word and a lexeme (as an abstract unit of form and meaning). Design conditions, on the other hand, state which formal conditions a grammatical word must meet to be able to function as an instance of a particular morphological category. In

6

Only a strong version of OT avoids morpheme-specific constraints. In more detailed studies, the need of language- and morpheme-specific constraints is discussed. Cf. Golston and Wiese (1995) who assume a constraint specific for plurals.

Martin Neef

248

many cases, design conditions lead to a formal difference between base and dependent grammatical word, thereby allowing to identify function via form which is also of pragmatical relevance. Obviously, design conditions typically lead to a violation of the Main Principle, which is not a well-formedness condition. Design conditions describe what knowledge language users must acquire to handle the formal properties of a morphological category in an adequate way on the assumption that every difference between base and instances of morphological categories must be learned in an explicit way. Since design conditions are unviolable (except by forms which are considered irregular), I assume that they are learnable. Conditions for well-formedness of surface forms are part of the linguistic knowledge of language users. Stems, which are abstractions from grammatical words, are also part of this linguistic knowledge. Stems, as the formal part of a lexeme, are strings of segments that comprise the information contained in all variants of that stem in the language. What is crucial with respect to well-formedness is that a stem must be identifiable in every grammatical word. Thus, the nominative plural can be well-formed only if it contains a noun stem.

3. The Pure Reduced Syllable Plural Within the Word Design approach, the observations concerning the prosodie structure of regular plural forms can be expressed through the following condition:7 (6)

Design Condition of Lower Limit for Reduced Syllables The grammatical word must end in at least one reduced syllable.

Design conditions of this kind are morphological in that they only hold for a morphologically defined subclass of the vocabulary. The condition in (6) holds for a subclass of nominal plurals which I will refer to as the 'reduced syllable plural type'8 (which is more or less identical to what so far has been referred to as the 'regular plural'). The Lower Limit for Reduced Syllables Condition is also valid for subjunctive forms (cf. Neef 1996a) and the nominalization pattern ending in -er like in Lehrer 'teacher'. In addition, reduced syllable conditions in slightly different formulations play a central role in different areas of German morphology such as the verbal infinitive (cf. Neef 1996a, 1997b) or the lexeme formation pattern Gehopse 'hopping' (cf. Neef 1996b). The Lower Limit for Reduced Syllables Condition is a kind of meta-condition for the reduced syllable plural type which is divided into three sub-groups that are subject to other 7

Golston and Wiese (1995:151) formulate an OT-constraint NON-FINALITY ('Inflected words do not end in a stressed syllable') which is also able to explain the occurrence of final schwas in the plural forms under consideration, but which obviously is not surface-true, not even for all German plurals. For descriptive reasons only I will occasionally retain the traditional notions of morpheme-based plural names.

The Reduced Syllable Plural in German

249

conditions as well. These conditions first of all concern the segmental endings which characterize these sub-groups. The first sub-group has no specific segmental ending. Therefore, I will call this plural class the 'pure reduced syllable plural', covering the traditional notions of 'zero plural' and 'schwa plural'. If the nominative singular ends in a full syllable, the corresponding plural form ends in schwa, thus resulting in the so-called 'schwa-plural'. The vowel schwa is the phonological default to form a reduced syllable if there is one needed for morphological reasons. In the following examples I ignore the occurrence of umlaut, but see section 5. (7)

a. b. c.

(der) Arm (der) Hut (das) Schaf (die) Kraft

Arme Hüte Schafe Kräfte

'arm' 'hat' 'sheep' 'strength'

Another instance of the pure reduced syllable plural is the so-called 'zero plural' which is found only in nouns whose singular forms already end in a reduced syllable.9 a.

b.

c.

(der) Vogel (der) Wagen (der) Bruder (der) Käse (das) Bündel (das) Laken (das) Lager (das) Gebirge (die) Sieben (die) Mutter (die) Tochter

Vögel Wagen Brüder Käse Bündel Laken Lager Gebirge Sieben Mütter Töchter

'bird' 'car' 'brother' 'cheese' 'bundle' 'sheet' 'camp' 'mountain' '(the) seven' 'mother' 'daughter'

These patterns are not equal in distribution: There are only three feminine nouns (8c) corresponding to this pattern; the regular plural form of feminine nouns ending in a sonorant-final reduced syllable is the «-plural.10 For neuter and masculine nouns of this pattern, the zero plural is the regular type. I only found nine nouns with a sonorant-final reduced syllable in the singular taking a different plural form, namely the «-plural (with the first one being the only example of a neuter noun): (9)

10

a.

(der/das) Friesel (der) Fussel (der) Muskel (der) Pantoffel (der) Stachel (der) Treidel (obs.)

Frieseln Fusseln Muskeln Pantoffeln Stacheln Treideln

'pimple' 'fluff 'muscle' 'slipper' 'thorn' 'tow rope'

Some irregular exceptions to this claim are listed in section 7 below. It may be for functional reasons that feminine nouns typically do not allow a zero plural: Kopeke (1993:151, following Werner 1969:98) argues that this restriction is due to the fact that the definite article is identical for feminine nouns in the nominative singular and plural. On this level, a plural noun must always differ from the singular form. This is what Bojunga (1890:8ff.) considers one of the central differences between Middle High German and New High German (cf. Law of Number Differentiation).

Martin Neef

250 b.

(der) Bauer (der) Gevatter (der) Vetter

Bauern Gevattern Vettern

'farmer' 'godfather' 'cousin'

The picture is different for singular nouns ending in schwa: Except for the masculine noun Käse^1 and some instances of the pattern Gehopse like Gebirge12 as mentioned in (8), all relevant nouns have an «-plural: (10)

(der) Bote (das) Auge (die) Straße

Boten Augen Straßen

'messenger' 'eye' 'street'

Äugst (1979:224) captures this pattern in an explicit rule stating that nouns ending in schwa systematically take the (e)n-plural. In my approach, this is a somewhat peculiar regularity: Why should a noun which is schwa-final in the singular have no zero plural, whereas a noun ending in a sonorant-final reduced syllable in the singular can and typically does take the zero plural? It is worth mentioning that schwa as the final element of the citation form of nouns, namely the nominative singular, is not part of the stem. This can be seen in derivations from the nouns under consideration like Botschaft 'message', Äuglein 'eye (diminutive)', and Sträßchen 'street (diminutive)'. Here, there is no schwa at the end of the string representing the nominal stem. Even with plural inflection, a final schwa in the nominative singular is not necessarily preserved in the plural. The plural of Auge is Augen, which is pronounced with schwa in explicit articulation only as [?ay.gan] but without schwa as [?au.grj in standard articulation. Therefore, in an approach that relates the well-formedness of the nominative plural to the stem, the form of the nominative singular can not be decisive. What may be important is the distribution of schwa-final nouns across genders. In addition to the instances of the pattern Gehopse, there are only four neuter nouns that end in schwa: (11)

Auge Ende Erbe Interesse

Augen Enden {no plural) Interessen

'eye' 'end' 'heritage' 'interest'

Except for Käse, all other schwa-final masculine nouns are weak masculines that mark all inflectional forms except the nominative singular by a nasal-final reduced syllable {(der) Bär 'bear' — Bären). The majority of schwa-final nouns is feminine (about 90% of non-derived nouns according to Augst 1979:225). Feminine nouns, regardless of their internal make-up, take the ^«-plural in about 95% of all cases (according to Duden 1995s:234). Schwa-final

11

12

For Käse, in fact, the plural is marginal in standard speech and rather expressed through Käsesorten 'types of cheese'. Whereas regular items of this pattern have no plural forms, irregular items show four different possibilities of the reduced syllable plural (as well as all three genders), cf. Neef (1996b). Only neuter nouns ending in schwa in the singular show a zero plural. There exist about 50 of these items with about 25 being part of the core vocabulary (cf. Neef 1998b).

The Reduced Syllable Plural in German

251

feminine nouns with a front vowel as in Wiese 'meadow' can not take the pure reduced syllable plural because otherwise the nominative plural would sound identical to the nominative singular (umlaut is not available for front vowels). Therefore, the relevant feminine nouns have to mark their plurals with -en (-er never appears with feminine nouns). Thus, it seems that the «-plural for schwa-final nouns has not to be stated as an explicit rule but follows from independent classifications of masculine and feminine nouns. The irregular «-plural for the few neuter nouns in (11) may be due to analogical transfer.13 However, there is one problem to be solved: if a plural of this class is well-formed if it ends in a reduced syllable, then why does the singular noun Abend not suffice as a plural? This problem concerns all the nouns in (2), repeated as (12):14 (12) a.

b. c.

(der) Abend (die) Gegend (das) Tausend (die) Tugend (das) Hundert (die) Kirmes

Abende Gegenden Tausende Tugenden Hunderte Kirmessen

'evening' 'region' 'thousand' 'virtue' 'hundred' 'kermis'

What makes the difference between nouns like (das) Segel 'sail' which allow a zero plural and those in (12) which do not? Obviously, all of the nouns in (12) end in an obstruent, whereas nouns allowing a zero plural end in a sonorant. This is stated by the following condition:13 (13) Design Condition of Ban of Final Obstruents The grammatical word must not end in an obstruent. As will become clear, this condition is relevant for the morphological class of nouns taking the reduced syllable plural. It is surface true for this class since all instances of the reduced syllable plural end in schwa or in a sonorant, and it motivates the final schwas for the nouns in (12). The condition reconstructs the above-mentioned notion of pseudo-suffix in some respect because it explains why nouns ending in an obstruent-final reduced syllable can not have a zero plural. In contrast to the notion of pseudo-suffix, (13) in principle allows vowel-final nouns to have a zero plural. Plurals ending in a full vowel, however, are independently excluded by the reduced syllable condition in (6), and the rareness of schwa-final nouns taking the zero plural has been discussed above.

13

A conceivable strategy for schwa-final nouns could be to mark the plural by an additional schwa. Two adjacent schwas, however, are impossible in German which is explainable in terms of a phonological constraint to be discussed in section 6 below.

14

13

There are only few nouns in German ending in -em like Atem "breath'. Despite the recent loanword Item 'item' (pi. Items), they have no corresponding plural forms. This condition is similar to the OT-constraint SON]pl stating that plurals end in a sonorant, as formulated by Golston and Wiese (1995:152). Whereas the OT-constraint is valid for all the words in the world (but is violable), my constraint only holds for German nouns taking the reduced syllable plural (in this class, it is unviolated).

252

Martin Neef

4. The Extended Reduced Syllable Plural The other two sub-groups of the reduced syllable plural which I will subsume under the name of the 'extended reduced syllable plural' are characterized by a particular segmental ending, namely Irl and INI respectively. As a segmental ending, Irl obligatorily appears in the syllable rhyme where it has to be pronounced vocalic (cf. Neef 1996a:74ff). The nasal,16 on the other hand, assimilates to a preceding consonant if the form is realized without schwa (which is the normal case in standard articulation). The following condition covers these observations which are usually stated in terms of morphemes: (14) Design Condition of Segmental Ending The grammatical word must have a specific segmental ending. Clause: The segmental ending must be [e] / [N], With these segmental endings, the final reduced syllable is obligatory in many cases, independent of the reduced syllable condition. If, for example, the noun Held 'hero' is to end in a nasal, this nasal must stand in a reduced syllable because it is more sonorous than the noun-final plosive and can thus not be integrated into a monosyllabic word form. On the other hand, there are some nouns like Rebell 'rebel' and Schi 'ski' with their respective plural forms Rebellen and Schier that could be realized without a final reduced syllable as far as sonority is concerned as "[Re.'beln]17 and *[si :b], '#' marks these forms as morphologically not-well-formed, i.e. they are no possible plural forms in MSG.

5. Umlaut-Peculiarities For the extended reduced syllable plural ending in [e], another condition is relevant. All nouns belonging to this plural pattern show umlaut. Usually, this observation is restricted to those vowels that potentially can be umlauted, namely back vowels. In a declarative approach in which only the surface form (or the question inasmuch a grammatical word meets a specific condition) is taken into consideration, umlauting can be approached in a more homogeneous way: the -er-plural demands a front vowel as the last full vowel. Thus, plural forms like Münder 'mouths' (sg. Mund) and Kinder 'children' (sg. Kind) equally obey this condition. The

16

The capital letter is meant to indicate that the nasal is underspecified with respect to the place of articulation.

17

This form actually occurs in colloquial speech as a plural form. Obviously, the design conditions for colloquial speech differ from those for standard speech. Strictly speaking, I assume that the design conditions relevant for a morphological category in colloquial speech are a proper sub-group of the design conditions relevant for the same morphological category in standard articulation.

253

The Reduced Syllable Plural in German

following condition reconstructs the morphological part of the notion "umlaut' in a declarative way: (15) Design Condition of a Front Vowel The grammatical word must have a front vowel as the last full vowel. The notion of full vowel is crucial in this formulation because umlaut is blind for both schwa and the vocalic Irl in that these vocalic elements neither are umlauted themselves nor prohibit umlauting of a preceding vowel as can be seen in a pair like Vater 'father' — Väter. In principle, the Front Vowel Condition could block the -er-plural for nouns with a back vowel. The phonological component of German, however, states that there is a strong affinity between vowels that differ only in the feature for backness. Thus, if a morphological category demands a front vowel in a particular position, such a front vowel is always evaluated as similar enough to the corresponding vowel of the base even if this base vowel is a back one. The Front Vowel Condition is part of the design conditions relevant for the -er-plural. It is clearly not part of the conditions for the fejrt-plural and the s-plural which never show umlaut.18 The situation is more complicated for the pure reduced syllable plural. Interestingly, with respect to the distribution of umlaut the zero plural and the schwa plural display homogeneous behaviour, indicating again that these plurals constitute one and the same class: a.

b.

c.

(der) Vogel (der) Hobel (der) Bart (der) Hund (das) Kloster (das) Lager (das) Floß (das) Brot (die) Mutter (die) Nacht

Vögel Hobel Bärte Hunde Klöster Lager Flöße Brote Mütter Nächte

'bird' 'plane' 'beard' 'dog' 'cloister' 'camp' 'raft' 'bread' 'mother' 'night'

Feminine nouns (cf. c.) obey the Front Vowel Condition in all cases except for a handful of words of the lexeme formation pattern Trübsal 'afflications'. This unproductive pattern, however, predominantly appears with neuter gender. In order to reveal a constant plural pattern, the feminine forms have to be irregular. Neuter nouns (cf. b.) almost never obey the Front Vowel Condition (with Kloster and Floß being the only exceptions in the core vocabulary). Masculine nouns (cf. a.), in contrast, show no clear tendencies.19 For them, I prefer to state umlauted forms as lexically marked exceptions, even though they are not rare. This assumption treats masculine and neuter nouns as one class and feminine nouns as another which corresponds to the basic opposition with respect to nominal inflection. Moreover, it accounts for the

18 19

The only exception is (die) Werkstatt 'garage' with its plural Werkstätten. Cf. Russ (1989) for a test concerning this subject which yielded no clear results. Köpcke (1994a) identifies some interesting tendencies that govern the distribution of umlaut in a subgroup of the relevant data.

Martin Neef

254

seemingly unproductiveness of umlauting in new plural forms which is verified by Kopeke (1988:325) who points out that umlaut is strongly undergeneralized in recent loanwords. This leads to the prediction that umlaut in masculine items of the pure reduced syllable plural should diminish. I must admit that I have no hints that this is really the case. Taking into account the notion of the pure reduced syllable plural, umlauting can be conceived as a homogeneous pattern with only one exception, namely masculine nouns taking the pure reduced syllable plural. This analysis obviously reduces the amount of irregularities to a limited class.

6. The Syllable Peak Adjacency Constraint As opposed to full syllables, reduced syllables are not free in their distribution. It has often been stated that nouns ending in a full vowel must take the s-plural (cf. van Megen 1992:129, Duden 19955:229). But there is a significant group of exceptions to this generalization: a. b. c.

singular (der) Floh (der) Schi (der) Zeh (das) Reh (die) Fee (die) Kuh

plural Flöhe Schier Zehen Rehe Feen Kühe

standard pronunciation ['fl0:.a] ['si:.e] E'tse:.n] ['Re:.a] [*fe:.n] ['ky:.s]

gloss 'flea' 'ski' 'toe' 'deer' 'fairy' 'cow'

Across genders, there are nouns — although ending in a full vowel — that take the reduced syllable plural. Adding the reduced syllable plural to a vowel-final noun results in a constellation with the peak of a reduced syllable being adjacent to a full vowel. The reduced syllable's peak can be the vowel schwa like in Flöhe, the vocalic /r/ like in Schier, or a syllabic nasal like in Zehen in the standard pronunciation given in (17). The central point is that all the singular nouns in (17) end in a vowel that bears main stress. Only nouns ending in an unstressed full vowel (i.e. a full vowel not bearing main stress) take the J-plural regularly. There is no single case attested where nouns of this formal class take the reduced syllable plural properly (but see (21) below): (18) a. b. c.

(der) 'Krimi (der) 'Uhu (das) 'Auto (das) 'Mobil[e] (die) 'Oma (die) 'Pizza

Krimis Uhus Autos Mobil[es] Omas Pizzas

*Krimi[a] *Uhu[e] *Auto[a] *Mobil[ea] *Oma[n] *Pizza[n]

'thriller' 'eagle-owl' 'car' 'mobile' 'grandma' 'pizza'

The examples in the third column are meant to represent the typical reduced syllable plural for each gender. All of these words are ill-formed. The phonological analysis of these data lies neither in the avoidance of a hiatus nor in the prohibition of reduced syllables with empty on-

The Reduced Syllable Plural in German

255

sets since the plurals in (17) render counterexamples for both these hypotheses. What is impossible is the constellation of a reduced syllable with an empty onset following immediately the peak of an unstressed (full or reduced) syllable. This is not only impossible for plural nouns as has been noticed by several researchers (cf. e.g. Werner (1969:96) and Kopeke (1993:133)), but it is a general condition of the phonology of German.20 Therefore, the generalization is stated by the following phonological constraint (which is — in contrast to design conditions — relevant for the whole vocabulary of German): ( 19) Syllable Peak Adjacency Constraint The syllable peak of a reduced syllable must not be right-adjacent to the syllable peak of an unstressed syllable. Domain: Phonological word. There are two kinds of exceptions to this phonological constraint. One exception is the formation of names of origin from cities like Londoner 'Londoner'. Here, words like Jenaer ['je:.na.e] or Osloer ['?ds.1o.b] exist. The other exception concerns words containing the combination of the letters (not representing a main stressed syllable) like Ferien 'holidays'. These letters suggest a trisyllabic pronunciation of the word under consideration (with initial primary stress). This pronunciation — contradicting the Syllable Peak Adjacency Constraint— in fact occurs, but rather in explicit (orthography-driven) than in standard articulation; on this stylistic level, the bisyllabic form ['fe:e.jan] is typical which accords with the constraint in (19) (cf. Neef 1998b for a detailed analysis of the Syllable Peak Adjacency Constraint). In consequence, for nouns ending in an unstressed full vowel the reduced syllable plural is not available. As a substitute, the s-plural stands in as the predictable form. This kind of plural usually occurs with marginal lexeme classes only as in (20) which are not part of the core lexicon, but of the peripheral lexicon. The following list is taken from Marcus et al. (1995:229):21 unassimilated borrowings onomatopoeic nouns names

Café Kuckuck Mann

product names

Kadett

nominalizedfunction words acronyms truncations

Wenn GmbH Sozi

Cafés 'café' Kuckucks 'cuckoo' Manns (but common noun: Männer 'men') Kadetts (but common noun: Kadetten 'cadets') Wenns '(the) if GmbHs 'corporation' 'socialist' Sozis

20

21

Wegener (1995:23) states that this constellation is 'unnatural' in German. Bornschein and Butt (1987:141) give a formulation of this 'phonotactic rale' which is closely related to mine. For a characterization of 'marked nouns' see also Bornschein and Butt (1987:140) and Wegener (1995:22ff.).

Martin Neef

256

On the basis of these data I claim that the s-plural is in principle favoured for nouns of this class to mark a difference between the peripheral and the core lexicon.22 Only nouns ending in an unstressed full vowel must take the s-plural independent of theif relative status in the lexicon. It should be noted that there is a (sub-)class of nouns ending in an unstressed full vowel that nonetheless takes the reduced syllable plural: (21) a. b. c.

(die)'Firma (die) 'Villa (das) 'Drama (das) 'Konto

Firmen Villen Dramen Konten

'company' 'villa' 'drama' 'account'

These nouns can not participate in the reduced syllable plural straightforwardly because of the Syllable Peak Adjacency Constraint. In order to avoid the ¿-plural, they substitute the final vowel with a reduced syllable ending in [N]. This is a somewhat regular pattern for loaned nouns ending in [a] (cf. Kopeke 1993:129f., Harnisch 1994:106). The nouns in (21) end in the typical German plural marker -(e)n, but they differ from regular patterns in that the stem is not completely preserved in the plural form. In this respect, these nouns violate the Main Principle of Morphology. Such a deletion, however, is not a productive pattern. Words that become part of the core lexicon do not prefer the reduced syllable plural over the j-plural if this would result in a violation of the Syllable Peak Adjacency Constraint. So, for example, a word like Auto 'car' (which is surely part of the core lexicon of MSG) takes the i-plural (Autos). Interestingly, the archaic form of this word Automobil 'automobile' takes the pure reduced syllable plural (Automobile). The s-plural is relatively free in its distribution. There is only one restriction concerning nouns ending in [s]: The plural of a Christian name like Jonas is Jonasse, indicating that the requirement that the final segment be [s] does not suffice. It seems that an s-plural form must not be identical to its base. This additional requirement is stated in (22), a condition that plays a role in other parts of German morphology as well (cf. Neef 1996a, b, 1997b).23 (22) Design Condition of Difference to the Base The grammatical word must have a different phonological structure than the phonological realization of the base. Due to the s-plural, the plural inflection of German nouns displays no source of blocking for conversion into nouns. Every phonologically well-formed string can function as a noun. Verbs, on the other hand, must possess a well-formed infinitive which is formally restricted since it 22

23

In addition, Wegener (1995:23) argues that the s-plural is favoured for marked nouns to preserve their sound structure since the reduced syllable plural typically changes the syllable structure of the singular form. Certain aspects of this plural type are still unsettled. Condition (22) in connection with the Design Condition of Segmental Ending with the clause [s] suggests that the plural form for Jonas should be *Jonasses. This form may be not occurring because obstruent-final reduced syllables are illicit in German plural nouns, and a schwa-final word has a perfect plural form.

The Reduced Syllable Plural in German

257

must end in exactly one reduced syllable (cf. Neef 1996a, 1997b). Were the reduced syllable plural the only way for nouns to form their plural, nouns ending in an unstressed full vowel would have no plural at their disposal. This, too, could block conversion into nouns if nouns were forced to have a plural form.

7. Irregular Plural Forms Analyzing a part of the German plural formation as regular comprises the stipulation of what is regarded as irregular. This section is devoted to the remaining irregular plurals. The Main Principle of Morphology demands that the base be preserved in a grammatical word. There is at least one regular morphological pattern in German allowing truncation explicitly, namely the Compi-type (hypocoristic form of Computer 'computer', cf. Neef 1996a:278ff). Unsystematic truncation can be observed in some irregular plural nouns like in (23), all of which are foreign words that are mainly of technical terminology. The nouns in question are borrowed from Latin, Greek, or Italian together with their specific plural form. Arguably, these lexemes are part of the learned vocabulary. Some of these nouns are also used with regular plural forms. (23)

Appendix Cello Examen Genus Kasus ['ka:.zUs] Lexikon

Appendizie]s Celli Examina Genera Kasus [ka.'zu:s] Lexika

also: Appendixe also: Cellos also: Examen also: Lexiken

'appendix' 'cello' 'exam' 'gender' 'case' 'lexicon'

For a noun like Cello, there is no reason to assume an abstract stem /tsel/ since in German there is no regular pattern that makes a noun end in [o] in the nominative singular.24 All that a language user can abstract from the surface form is the stem /tsslo/ which never leads to the corresponding plural form in a predictable way. The regular way to form a nominal plural is to take a complete stem and make it end in at least one reduced syllable. The examples in (23) deviate from this pattern in both respects. In German, there are also irregular nouns that deviate in only one point. There is a group of plural nouns that contain a complete nominal stem but do not end in a reduced syllable. What is characteristic is the occurrence of the plural marker η with the preceding vowel being a full vowel as in (24a, b). The same holds for the (native) diminutive pattern in (24c), but here the nasal is already present in the stem.

24

Hämisch (1994:106ff.) focusses on cases like these to motívate his assumption of general stem inflection in German nouns.

Martin Neef

258 (24) a. b. c.

Herr Konsul Nachbar Ungar Blümlein

Herrn25 Konsuln Nachbarn Ungarn Blümlein26

'master' 'consul' 'neighbour' 'Hungarian' 'flower (diminutive)'

Except for Herrn, all nouns in (24) end in a full syllable not bearing main stress. Thus, these (non-compounded) plural forms meet the condition of ending in an unstressed syllable, but this syllable happens to be a full syllable. Wegener (1995:19) assumes that nouns with a final unstressed full syllable taking the (e)n-plural never end in a reduced syllable if the stem final consonant is a liquid. According to Mugdan (1977:213) and Marcus et al. (1995:248), there exist seven nouns in German following this pattern (the ones in (24b) and four uncommon items such as Satyr 'satyr'). Possible counterexamples to this claim are the three nouns Lorbeer 'laurel1, Vorfahr 'ancestor', and Abfuhr 'rebuff all of which take the full -(e)w-plural with main stress on the initial syllable. Wegener concedes that the regularity she observed only holds for nouns with the last full vowel being a short one. Since there is no perceptible difference in vowel quantity or quality between the last vowels in Nachbar and Vorfahr, I take the regularity as an undecided matter. In spoken German, however, the nouns I consider to be counterexamples are often pronounced without a final reduced syllable in the plural, thus supporting Wegener's assumption. For the moment, I take the plurals in (24) as irregular ones belonging to the learned vocabulary. The second group of nouns deviating from the regular pattern is constituted by some plurals ending in a reduced syllable; the stems of these plurals differ from the stems of the singular forms with respect to vowel quality, stress pattern, additional or missing segmental material or a combination of these characteristics. Again, these irregular plural forms are supposed to be listed in their corresponding lexical entries. a. b. c. d. e.

25

26

Stadt [stat] Pro'fessor Material Bau Datum 'Atlas Cha'rakter 'Embiyo 'Klima

Städte [stas:.ta] Professoren Materialien Bauten Daten At'lanten Charak't[e:]re Embry'onen Kli'mate

also: Atlasse also: Embryos also: Klimas

'town' 'professor' 'material' 'building' 'date' 'atlas' 'character' 'embryo' 'climate'

The monosyllabic form Herrn is typical for colloquial speech. In standard articulation, the bisyllabic form Herren is the regular one. In colloquial speech this pattern is also used with an ¿-plural.

The Reduced Syllable Plural in German

259

8. Representational Characteristics of Plural Forms in a Word Design Grammar In what follows I will sketch the model of grammar in the Word Design approach and summarize the results of the analysis of the representational characteristics of the German plural. For the subject of plural inflection, two components are relevant, namely the lexicon and morphology. Lexemes or lexical entries are stored in the mental lexicon. They are associated with an abstract stem representation, grammatical features and semantic information. What is central with respect to plural formation is the grammatical information encoded in the lexical entries. Most nouns are unmarked in this regard, that is to say that they display a regular plural form in a predictable way. The final section of this paper is devoted to the question how regular nouns are related to their plural class since there exists more than one regular plural class. Many nouns are marked for a formally regular plural class they unpredictably belong to. Some nouns are marked as irregular in that the plural form is stored in the lexical entry. Finally, a remarkably number of nouns is marked for not taking a plural form at all. The second point of relevance is the morphological component of grammar. The preceding analysis has shown that there are four formally regular plural classes. Three of these classes can be subsumed under the heading of the reduced syllable plural, the fourth one is the s-plural. Word Design as a theory of mental representation of morphological knowledge states which design conditions are relevant for each of these plural classes. The plural system as resulting from the preceding analysis is sketched in the following table: (26) Plural system for German nouns Base: nominal stem REDUCED SYLLABLE PLURAL

Design Condition of Lower Limit for Reduced Syllables Design Condition of Ban of Final Obstruents EXTENDED REDUCED SYLLABLE PLURAL /N/-TYPE

Design Condition of Segmental Ending, Clause [N] EXTENDED REDUCED SYLLABLE PLURAL /Γ/-ΤΥΡΕ

Design Condition of Segmental Ending, Clause [e] Design Condition of a Front Vowel ¿-PLURAL

Design Condition of Segmental Ending, Clause [s] Design Condition of Difference to the Base

Martin Neef

260

Knowing the plural form of a noun of the extended reduced syllable plural of the /n/-type is equivalent to knowing the base form and the three relevant design conditions. Take, for example, the feminine noun Jagd 'hunt' which takes the (e)n-plural by default. Its stem is represented as /jagd/. The following table shows how the grammar marks the plural form Jagden as the only grammatical possibility (in standard articulation):27 (27)

plural formation for Jagd /jagd/

[jagd] [ja:kt]

LOWER LIMIT

BAN OF

SEGMENTAL

FOR REDUCED

FINAL

ENDING [N]

SYLLABLES

OBSTRUENTS

*

*

*

*

*

*

[ja:k.ds]

PHONOLOGY

* (Final Devoicing)

*

—>[ja:k.dn] This table resembles evaluation tables in OT, but is different in central respects. First, any star indicating a violation of a condition is fatal. Second, and following from that, the conditions are not ranked. Third, there are two blocks of conditions: the morphological ones relevant for the morphological category of the (e)n-plural, and the phonological ones. The latter only appear in abbreviated form as 'Phonology'. Were (27) to be taken as an exhaustive analysis, it should contain all the phonological constraints which are relevant for MSG. However, this would demand an elaborated declarative phonological theory of German and, furthermore, is not crucial to the analysis presented. The Main Principle of Morphology demands that a grammatical word be as similar to the base as possible. In this respect the design conditions are positive conditions that state explicitly in which aspects a difference to the base is allowed (without forcing a difference in every instance). It should be noted, again, that the Main Principle of Morphology is not a condition for well-formedeness, but a basic principle. Therefore, it does not appear in (27).

9. Defaults and Learning Rules The Word Design analysis shows what kind of knowledge has to be represented mentally for a speaker to be able to handle adequately the category of the nominative plural. The main result 27

Regularities concerning the alternation of syllabic sonorants with sequences containing schwa are ignored here. This alternation is attributed to a difference of the phonological component between standard and explicit articulation. The phonology of the latter stylistic level contains a further constraint that demands to avoid syllabic sonorants; cf. Neef 1997a.

The Reduced Syllable Plural in German

261

is that among the regular forms there are two principally different types of plurals in German, namely the reduced syllable plural and the Λ-plural. This distinction is reflected in the whole plural paradigm. Nouns taking the reduced syllable plural have identical forms in the nominative, the genitive and the accusative plural whereas the dative is always marked by an [N], This nasal can also be present in the nominative, thus giving rise to identical forms in the whole plural paradigm for nouns ending in a nasal-final reduced syllable in the nominative plural like Haken 'hooks' or Frauen 'women'. This result can not be achieved in models invoking morphemic segmentation. On the other hand, nouns taking the s-plural and irregular nouns that do not take the reduced syllable plural like Celli 'cellos' have identical forms in the whole plural paradigm. This makes the s-plural more similar to the irregular plurals than to the regular reduced syllable plural (cf. Neef 1998a). A problem closely connected to that of representation is that of learnability. In this area, several ways have been pursued: Some studies highlight the s-plural as the only regular one, contradicting the just given assessment. Following Wiese 1986 and Wunderlich 1986, Marcus et al. (1995:231) argue that this plural marker can be characterized as the 'default plural' that „bespeaks the operation of a rule referring to a symbol for an entire category that applies unless blocked by a competing stored form." Although of relatively low frequency — Marcus et al. (1995:229) estimate this frequency through a comparison of several investigations of less than 10% in type frequency and less than 5% in token frequency — they conclude that -s is the default because it combines with marginal lexeme classes that are arguably not part of the learned vocabulary. The aim of Marcus et al. (1995) is to prove that there are rules which are symbolconcatenating mental operations (besides stored items). The assumed mental rule for plural formation would read as 'add -s to a noun'. In case there is a stored lexical item with a different plural form, this rule is blocked. But as Kopeke (1988:325) points out, only about 50% of 182 recent (monosyllabic) German loanwords are formed with -s. For example, the English word sketch has been borrowed in the 20th century without the English plural suffix -es, the German plural form is Sketche. It remains unclear why the presumed rule does not apply to all, or at least most of all recent loanwords even though these new words were not stored in the beginning of their use. Furthermore, the assumptions made by Marcus et al. (1995) impose a great amount of learning onto speakers of German because nearly the complete core lexicon is said to be irregular and hence to be learned item by item.28 Marcus et al. (1995) are forced to posit exactly one default for all nouns. If'default' is understood as the regular or normal case like in Bittner (1994:75), Wurzel (1994:40f ), or Wegener (1995:24, 31) where different nominal subclasses are associated with different de-

28

For a criticism of this line of argumentation, cf. also Wegener (1995:11, 39). Interestingly, Marcus et al. (1995:238) subsequently weaken their claims in such a way as to allow novel nouns to form their plural by analogy to existing irregular nouns. My analysis tries to handle this putative analogy in strict terms.

262

Martin Neef

faults, the identification of defaults comes close to the goal of learning grammars which seek to minimize the set of plural rules in order to make the learning as easy as possible.29 In this spirit, I will sketch the defaults of plural assignment that arise from the preceding analysis. For nouns belonging to the peripheral lexicon (cf. (20)), the default is the i-plural. The picture is indistinct for peripheral nouns ending in [s]. For nouns belonging to the core lexicon, the reduced syllable plural is the default. Exceptions are core nouns ending in an unstressed full vowel that regularly take the s-plural due to the Syllable Peak Adjacency Constraint. The reduced syllable plural appears in different shapes. The classification of core nouns into different subgroups mainly depends on gender distinctions. The default for feminine nouns is the extended reduced syllable plural ending in [N], whereas for masculine and neuter nouns it is the pure reduced syllable plural. The sub-group of the weak masculines behaves differently in that it takes the extended reduced syllable plural ending in [N], The notion of a 'weak masculine' is crucial for the whole nominal inflection and, thus, the question of how masculines are categorized as weak vs. non-weak is an important one.30 There are exceptions to this default system. First, the notion 'peripheral lexicon' is not clear. Why does (der) Schal 'shawl' behave as an unassimilated borrowing in that it takes the i-plural and not the regular pure reduced syllable plural although it is frequent in MSG and phonologically assimilated? Second, the -er-plural is in no case the regular one although it has a type frequency of 2,3% (according to Duden 19955:233f). The problem is especially crucial for the 2.000 so-called core words31 which are morphologically simple and end in a full syllable in the singular. Of these core words, 21% of the neuters and 2% of the masculines unpredictably take the -er-plural. At any rate, there is only one recent loanword taking the -er-plural, namely Schi 'ski' (according to Kopeke 1988:325), indicating that nowadays -er is not a productive plural marker. Third, there exist some nouns that do not behave as predicted: 25% of the feminine core words take the e-plural (with an absolute type frequency of 1,3%), and 9% of the masculine and 4 % of the neuter core words take the (e)n-plural. Finally, there are some lexeme formation patterns that behave irregular. The feminine pattern ending in -nis like Kenntnis 'knowledge' and the partly feminine pattern ending in -sal like Trübsal 'abdications' take the e-plural (without umlaut) and not the (e)n-plural, the (mostly) neuter pattern ending in -tum like Herzogtum 'dukedom' takes the -er-plural and not the e-plural, and, as mentioned above, the neuter diminutive pattern ending in -lein takes the zero plural, thus even ignoring the reduced syllable condition typical for German plurals. 29

30

31

t

Such rale systems are presented in Augst (1979:224), van Megen (1992:125f.), Duden (19955:229), and Wegener (1995:24-31). Kopeke (1995) identifies several phonological, grammatical and semantic criteria to characterize the weak masculines along prototypical lines, but without being able to give clear-cut delimitations. According to Duden (1995s:233), the type frequency of the weak masculines is 3,6% and the token frequency is 1,6% of all nouns. Cf. Duden (1995s:227fif.) for this notion and for the following ratings.

The Reduced Syllable Plural in German

263

Finally, I want to discuss a problem that results from learning systems. The plural can only be regarded as an inferred word form if there is a base. Grammars and some linguistic studies (for example analyses carried out in Natural Morphology, cf. Wurzel 1994) take the nominative singular as the base. Generative studies, on the other hand, typically assume an abstract underlying form which may differ from the surface form of the singular. One reason for adapting the generative perspective has to do with final devoicing: From the nominative singular [?alp], two forms with an (e)n-plural can be inferred, namely [?al.pan] and [?al.bsn]. Both words exist in German with Alp meaning 'alpine pasture' and Alb meaning 'elf. The stem, however, can only be inferred from the plural forms. This indicates that the nominative singular is not in every case a perfect clue for plural formation. It might be said that orthography comes into play here, but plural inflection is typically acquired prior to the acquisition of orthography. Moreover, the nominative singular sometimes can not be inferred from the plural-based stem. This is crucial, for example, for nouns that predominantly appear in the plural. Duden (1995s:215f.) gives a list of nouns that only or almost only appear in the plural like Faxen '(the) fooling about' that is marked for sometimes being used in the singular. But how can a speaker decide whether the nominative singular is Fax or Faxe (and how can the gender be fixed)? Thus, it seems that neither gender nor the nominative singular nor the nominative plural alone are in all cases sufficient to know all forms of the inflectional paradigm.32

References Anderson, Stephen R. (1992): A-Morphous Morphology. Cambridge: Cambridge University Press (= Cambridge Studies in Linguistics 62). Aronoff, Mark (1994): Morphology by Itself. Stems and Inflectional Classes. Cambridge/ Mass., London: MIT Press (= Linguistic Inquiry Monograph 22). Augst, Gerhard (1979): Neuere Forschungen zur Substantivflexion. Zeitschrift ftlr germanistische Linguistik 7, 220-232. Bauer, Laurie (1996): Derivational paradigms. Yearbook of Morphology 1996, 243-256. Bech, Gunnar (1963): Zur Morphologie der deutschen Substantive. Lingua 12, 177-189. Becker, Thomas (1990): Analogie und morphologische Theorie. München: Fink. Bittner, Dagmar (1994): Die Bedeutung der Genusklassifikation für die Organisation der deutschen Substantivflexion. In: Köpcke (ed), 65-80. Bojunga, Klaudius (1890): Die Entwicklung der neuhochdeutschen Substantivflexion ihrem inneren Zusammenhange nach in Umrissen erfaßt. Leipzig: Hirschfeld. Bornschein, Matthias and Matthias Butt (1987): Zum Status des s-Plurals im gegenwärtigen Deutsch. In: Abraham, Werner and Ritva Arhammar (eds.): Linguistik in Deutschland. Akten des 21. Linguistischen Kolloquiums, 135-154. Tübingen: Niemeyer. Curme, George A. (1922): A Grammar of the German Language. New York/London: MacMillan.

32

Thanks to Dagmar Haumann, Robert Kemp, Albert Ortmann, Heinz Vater, Dieter Wunderlich, and Wolfgang Ullrich Wurzel for providing clarifying comments. I would like to thank the Sonderforschungsbereich 282 'Theorie des Lexikons' for financial support during my work on plural inflection.

264

Martin Neef

Duden (1995s): Grammatik der deutschen Gegenwartssprache. Mannheim/Leipzig/Wien/Zürich: Bibliographisches Institut (= Duden 4). - (199621): Rechtschreibung der deutschen Sprache. Mannheim/Leipzig/Zürich/Wien: Bibliographisches Institut (= Duden 1). Golston, Chris and Richard Wiese (1995): Zero morphology and constraint interaction: subtraction and epenthesis in German dialects. Yearbook of Morphology 1995,143-159. Harnisch, Rüdiger (1994): Stammerweiterung im Singular — Stammflexion im Plural. Zum Bautyp der deutschen Substantivdeklination. In: Köpcke (ed.), 97-114. Kopeke, Klaus-Michael (1988): Schemas in German Plural Formation. Lingua 74, 303-335. - (1993): Schemata bei der Pluralbildung im Deutschen. Versuch einer kognitiven Morphologie. Tübingen: Narr (= Studien zur deutschen Grammatik 47). - (1994a): Zur Rolle von Schemata bei der Pluralbildung monosyllabischer Maskulina. In: Köpcke (ed.), 81-96. - (ed.), (1994b): Funktionale Untersuchungen zur deutschen Nominal- und Verbalmorphologie. Tübingen: Niemeyer (= Linguistische Arbeiten 319). - (1995): Die Klassifikation der schwachen Maskulina in der deutschen Gegenwartssprache. Zeitschrift für Sprachwissenschaft 14, 159-180. Köpcke, Klaus-Michael and David Zubin (1996): Prinzipien für die Genuszuweisung im Deutschen. In: Lang, Ewald and Gisela Zifonun (eds.) Deutsch — typologisch, 473-491. Berlin, New York: de Gruyter (= Jahrbuch des Instituts ftir deutsche Sprache 1995). Kufiier, Herbert L. (1962): The grammatical structures of English and German. A contrastive sketch. Chicago, London: University of Chicago Press. Marcus, Gary F., Ursula Brinkmann, Harald Clahsen, Richard Wiese and Steven Pinker (1995): German Inflection: The Exception That Proves the Rule. Cognitive Psychology 29, 189-256. Megen, Jan van (1992): Zur Pluralbildung der Substantive. Zielsprache Deutsch 23, 120-131. Mugdan, Joachim (1977): Flexionsmorphologie und Psycholinguistik. Tübingen: Narr (= Tübinger Beiträge zur Linguistik 82). Neef, Martin (1996a): Wortdesign. Eine deklarative Analyse der deutschen Verbflexion. Tübingen: Stauffenburg (= Studien zur deutschen Grammatik 52). - (1996b): Wortdesign: Das Wortbildungsmuster Gehopse und die Kopflosigkeit von 'Ableitungen'. Zeitschrift ftir Sprachwissenschaft 15, 61-91. - (1997a): Die Alternationsbedingung: Eine deklarative Neubetrachtung. In: Dürscheid, Christa, Monika Schwarz & Karl Heinz Ramers (eds.): Sprache im Fokus. Festschrift für Heinz Vater, 17-31. Tübingen: Niemeyer. - (1997b): Conversion into verbs: A declarative analysis of the German infinitive. Arbeiten des Sonderforschungsbereichs 282 „Theorie des Lexikons" Nr. 95, Universität Düsseldorf. - (1998a): A declarative path through the jungle of German nominal inflection. Ms. University of Cologne. - (1998b): Elemente einer deklarativen Wortgrammatik. Ms. University of Cotogne. Prince, Alan and Paul Smolensky (1993): Optimality Theory. Constraint Interaction in Generative Grammar. Ms., Rutgers University (= Rutgers University Center for Cognitive Science Technical Report #2). Rettig, Wolfgang (1972): Sprachsystem und Sprachnorm in der deutschen Substantivflexion. Tübingen: Narr (= Tübinger Beiträge zur Linguistik 32). Russ, Charles V.J): (1989): Die Pluralbildung im Deutschen. Zeitschrift für Germanistische Linguistik 17, 5867. Scobbie, James M., John S. Coleman and Steven Bird (1996): Key Aspects of Declarative Phonology. In: Durand, Jacques and Bernard Laks (eds.) Current trends in phonology: models and methods. Volume II, 685-709. Salford: European Studies Research Institute (= Proceedings of the Royaumont meeting 1995). Steche, Theodor (1927): Die neuhochdeutsche Wortbiegung unter besoderer Berücksichtigung der Sprachentwicklung im 19. Jahrhundert. Erster Teil: Die Wortklassen — Die Beugung. Breslau.

The Reduced Syllable Plural in German

265

Vennemann, Theo (1991): Syllable structure and syllable cut prosodies in modern Standard German. In: Bertinetto, Pier Marco, Michael Kenstowicz and Michele Loporcaro (eds.): Certamen Phonologicum II. Papers from the 1990 Cortona Phonology Meeting, 211-243. Turin: Rosenberg & Sellier. Wegener, Heide (1995): Die Nominalflexion des Deutschen — verstanden als Lerngegenstand. Tübingen: Niemeyer (= Reihe Germanistische Linguistik 151). Werner, Otmar (1969): Das deutsche Pluralsystem. Strukturelle Diachronie. In: Moser, Hugo (ed.) Sprache — Gegenwart und Geschichte, 92-128. Düsseldorf: Schwann (= Sprache der Gegenwart 5). Wiese, Richard (1986): Schwa and the structure of words in German. Linguistics 24, 695-724. - (1996): The Phonology of German. Oxford: Clarendon (= The Phonology of the Languages of the World). Wunderlich, Dieter (1986): Zum Aufbau lexikalischer Repräsentationen. Ms. Universität Düsseldorf. Wurzel, Wolfgang Ullrich (1994): Gibt es im Deutschen noch eine einheitliche Substantivflexion? oder: Auf welche Weise ist die deutsche Substantivflexion möglichst angemessen zu erfassen? In Kopeke (ed.), 29-44. Zubin, David and Klaus-Michael Köpcke (1986): Gender and folk taxonomy: The indexical relation between grammatical and lexical categorization. In: Craig, Colette (ed.). Noun classification and categorization, 139-180. Philadelphia: Benjamins.

Damaris Nübling (Albert-Ludwigs-Universität

Freiburg)

Wann werden die deutschen Präpositionen flektieren? Grammatisierungswege zur Flexion

Einleitung In diesem Beitrag geht es darum, Flexion primär über die Abgrenzung zu ihren beiden morphologischen Nachbardomänen, die Derivation und die Klise, zu bestimmen. Aus diesen beiden morphologischen Typen entwickelt sich auch neue Flexion. Mit dem Vergleich von Flexion, Derivation und Klise und mit der Frage nach der Entstehung von Flexion sollen die Ziele und Prinzipien von Flexion sichtbar gemacht werden. Der zweite Schwerpunkt dieses Artikels besteht in einer detaillierten Analyse einer sich anbahnenden Flexivierung via Klitisierung im Deutschen: Mit den Präposition-Artikel-Verschmelzungen (im, ins, zur, aufm, in'ri) liegt ein Paradebeispiel derzeit beobachtbarer und sukzessive sich herausbildender Flexion vor. Diese Verbindungen sind zwar noch als Vorstufen der Flexion zu bewerten, doch läßt sich über die Untersuchung dieses komplexen Grammatisierungsprozesses diskutieren, was noch geschehen muß, damit im Deutschen Präpositionalflexion entsteht. Kapitel 1 befaßt sich kurz mit dem Begriff der Flexion, Kapitel 2 mit der Entstehung von Flexion aus Derivation und Klise. Kapitel 3 widmet sich dem Beispiel der deutschen Präposition-ArtikelVerschmelzungen.

1. Erscheinungsformen der Flexion Der Terminus Flexion wird in einem weiteren und in einem engeren Sinn verwendet. Im weiteren Sinn bezeichnet Flexion ein spezifisches Verfahren der morphologischen Kodierung grammatischer Kategorien. Traditionellerweise gliedert sich Flexion in die Konjugation, Deklination und Komparation. Als sog. Wortformbildung wird Flexion neben die Wortbildung gestellt, die sich aus Derivation und Komposition zusammensetzt und im allgemeinen Lexikoneinträge ergibt, d.h. die Lexik erweitert; Flexionsformen tun dies nicht. Flexion im engeren Sinn bezeichnet einen Subtyp der eben genannten Flexion und betriffi die konkrete Materialisierung der grammatischen Informationen. Hier wird sie der Agglutination gegenübergestellt: Während Agglutination eine 1:1Beziehung von Inhalt und Ausdruck (bzw. Morphem und Morph) und Transparenz, d.h. klare morphologische Segmentieibarkeit voraussetzt, zeichnet sich Flexion im engeren Sinn durch

Wann werden die deutschen Präpositionen flektieren?

267

Allomorphik und Informationsüberlagerungen (Portmanteaumorphe) aus, womit sich eine morphologische Segmentierung als schwierig bis unmöglich erweist. Ein prototypisches Beispiel fur Flexion liefert das Lateinische mit seinen umfangreichen Paradigmen. Formal lassen sich sechs Kasus, zwei Numeri und drei Genera - auf Portmanteaumorphe gebündelt - mit dem Substantivstamm verbinden; unterschiedliche Deklinationsklassen sorgen für einen hohen Grad an Allomorphik. Flexion im weiteren Sinn liegt insofern vor, als sich Kategorien(bündel) in systematischer Weise mit einem nominalen Lexem verbinden, wobei es sich in beiden Fällen um gebundene Morphe handelt: Im Fall von lat. filia 'Tochter1 können sowohl fili- als auch -a (Nom.Sg.Fem.) nicht selbständig stehen. Solche lateinischen Flexionsparadigmen kommen den einschlägigen Definitionen von Flexion sehr nahe. Fragen zur Flexion im engeren Sinn betreffen den Ausdruck der Kategorien am bzw. im Wort, also ob sie miteinander fusionieren (im lat. Beispiel stark ausgeprägt) oder sogar mit der (lexikalischen) Wurzel (hier nicht). Wurzelflexion ist im Deutschen stark ausgeprägt, wofür der Ablaut bei Veiben zum Ausdruck des Tempus und der Umlaut in allen drei Hauptwortarten genannt sei. Im Extremfall kann die Flexion jeglicher morphologischen Segmentierbarkeit entbehren. Die Einheiten des Paradigmas stehen in keinem Verhältnis gegenseitiger Ableitbarkeit, ihr Ausdruck erfolgt suppletiv. Dies gilt weitgehend für die deutschen Personalpronomina, die zwar systematisch die grammatischen Kategorien Person, Numerus und Genus (letzteres jedoch nur an die 3.Sg. gekoppelt) zum Ausdruck bringen, diese Kategorien jedoch kaum mit morphologischen Segmenten korrelieren lassen (im Nominativ: ich, du, sie/er/es, wir, ihr, sie). Segmentierbarkeit besteht allenfalls in bestimmten Paradigmenausschnitten wie bei ihm [i:m] - ihn [i:n] - ihr [i:e] (hierzu s. Howe 1996). In der Literatur wird die lateinische Flexion oft als prototypisches Beispiel angeführt. Grundsätzlich sind aber auch andere Konstellationen in den Blick zu nehmen, etwa daß auch die Wurzel grammatische Funktion hat und unter Umständen kürzer ist als ihr(e) Flexiv(e): So ist der definite Artikel im Nhd. in die Wurzel d- und die Flexive -er, -em, -as etc. zu segmentieren. Erst in Verbindung mit den vokalhaltigen Flexiven entsteht ein vollständiges Wort. Umgekehrt gibt es den Fall, daß die Wurzel auch ohne Flexiv(e) ein Wort bildet (sog. Grundformflexion): Tag-e, Auto-s. Manchmal entfällt auch die Flektierbarkeit der gesamten Wortart (engl, hard-er, aber more important), oder nur eine Paradigmenform weist Flexion auf (z.B. engl, -s in der 3 .Sg.Präs., sonst überall -0). Eine vollständige Typologie der Flexion unter der besonderen Fragestellung, was "ideale" bzw. "periphere" Flexionseigenschaften sind, steht noch aus. Im folgenden soll die Flexion zur Derivation und zur Klise hin abgegrenzt werden; dabei werden die konstitutiven Eigenschaften von Flexion ausführlich zur Sprache kommen.

268

Damaris Nilbling

2. Wege zur Flexion Bybee 1985 entwickelt eine Typologie der verschiedenen Ausdracksverfahren von Informationen. Die fünf wichtigsten Verfahren siedelt sie auf dem folgenden Kontinuum an (S. 12, dort ohne Fettdruck): lexical — derivational — inflectional — free grammatical — syntactic < greater degree of fusion Die Kategorie Kausativität kann beispielsweise lexikalisch (links auf der Skala) über ein ganz anderes Lexem realisiert werden, z.B. töten. Bei anderen Verben - und dies ist im Deutschen die Regel - wird dies syntaktisch gelöst (rechts auf der Skala): jmcbi zum Lachen bringen. Manche Sprachen halten für den Ausdruck von Kausativität einflexivischesAusdrucksverfahren bereit, etwa ein bestimmtes Affix. Reste eines solchen im Germanischen vorhandenen Verfahrensfindensich in den nhd. Formen tränken (zu trinken), führen (zu fahren), setzen (zu sitzen) etc. Die Kausativierung erfolgte im Germanischen bzw., da besser belegt, im Gotischen durch die Suffigierung von -j-(an) an die Wurzel: got. sat-j-an ~ nhd. setzen, got. *warmjan ~ nhd. wärmen (zu warm). Die heute noch im Deutschen konservierten Fälle lassen sich der Derivation zuordnen. Mit dem Verfahren "free grammatical" sind z.B. Periphrasen mit bestimmten Auxiliaren gemeint, etwa die deutsche Perfekt- oder Passivperiphrase. Entwickeln sich solche Funktionswörter zu gebundenen Morphen, entstehen Klitika bzw. klitische Verbindungen (hast du haste), auch solche Verschmelzungen werden noch unter "free grammatical" gefaßt, da ihre phonologische Bindung oft nicht obligatorisch ist und das Klitikon noch eine relativ freie, meist syntaktisch gesteuerte Distribution an den Tag legt (wenn du —> wermde). Im folgenden wollen wir die beiden Nachbarverfahren zur Flexion (in der Skalenmitte) beleuchten, die zu ihr in einem typologischen Ahnlichkeitsverhältnis stehen. Dabei wenden wir uns zuerst dem Spannungsfeld zwischen Derivation und Flexion zu (2.1), dann dem zwischen Klise und Flexion (2.2). 2.1 Derivation - » Flexion: Selten begangener Grammatisierungsweg Bybee äußert sich nicht explizit darüber, ob ihre Skala der fünf Ausdruckstypen auch eine diachrone Dimension beinhaltet, d.h. ob sich das eine Verfahren im Laufe der Zeit zu dem benachbarten Verfahren entwickeln kann. Ganz sicher gilt dies, wenn man die Skala von rechts nach links liest, womit dieser Sprachwandel unidirektional verläuft (wenn man von gewissen Ausnahmen absieht). Der Fusionsgrad der einzelnen Morpheme nimmt dabei sukzessive bis zum nicht mehr segmentierbaren lexikalischen Ausdruck zu. Wie dieser Abschnitt zeigen wird, kommt es kaum vor, daß derivationelle inflexivischeVerfahren übergehen. Zuvor sollten die wichtigsten Unterschiede zwischen Derivation und Flexion skizziert werden, wobei es sich als schwierig erweist, Kriterien zu

Wann werden die deutschen Präpositionen flektieren?

269

formulieren, die durchgehend auf die Unterscheidung Derivation/Flexion anwendbar sind; eher lassen sich Tendenzen finden bzw. Eigenschaften, die nur bei einem der beiden Verfahren auftreten können. Mit Bybees zentralen inhaltsseitig motivierten Begriffen der Bedeutsamkeit ("relevance") und Allgemeingültigkeit ("generality") gelangt sie zu einer zwar skalaren, doch auf alle morphologischen Ausdruckstypen anwendbaren Differenzierung. Das Bedeutsamkeitsprinzip definiert sie wie folgt: " A meaning element is relevant to another meaning element if the semantic content of the first directly affects or modifies the semantic content of the second." (13)

Hier handelt es sich also um semantische Beziehungen zwischen zwei Morphemen. Ein besonders hohes Maß an Bedeutsamkeit äußert sich in lexikalischem, ein etwas geringeres in derivationellem Ausdruck. Für die Basis weniger bzw. nicht bedeutsame Konzepte werden dagegen eher flexivisch bzw.

gar

syntaktisch

ausgedrückt.

Bevor

wir

zu

Beispielen

kommen,

sei

das

Allgemeingültigkeitsprinzip vorgestellt: "The second factor that needs to be taken into consideration in determining what can be an inflectional category is lexical generality. By definition, an inflectional category must be applicable to all stems of the appropriate semantic and syntactic category and must obligatorily occur in the appropriate syntactic context. If a semantic element has high content, i.e. is very specific, it simply will not be applicable to a larger number of stems." (16/17)

Die Verbindung beider Prinzipien bedeutet fur Flexive, daß sie einerseits gegenüber der Semantik der Wurzel so bedeutsam sein müssen, daß sich ihr morphologischer Ausdruck überhaupt lohnt, andererseits sollte ihre Bedeutung möglichst unspezifisch und allgemeingültig sein, damit alle Wurzeln mit dieser Kategorie kombiniert werden können. Diese inhaltlichen Prinzipien bewirken ganz charakteristische äußere Erscheinungsformen von Flexion: Ihre hohe Allgemeingültigkeit (bzw. ihre unspezifische Bedeutung) sorgt für ungehinderte Kombinierbarkeit mit der Basis, d.h. es lassen sich lange Paradigmen bilden, die sich meist sogar durch Produktivität auszeichnen; auch entlehnte Verben werden mit den Person/Numerus- und Tempusflexiven versehen (z.B. managen). Die beiden wichtigsten Parameter, Bedeutungskonstanz und Paradigmatizität, können bezüglich der Unterscheidung von Derivation und Flexion gemäß Figur 1 veranschaulicht werden. Typisch nominale Flexionskategorien sind Numerus, Kasus und Genus, die das durch das Substantiv benannte Objekt nicht in dessen Integrität berühren, im Gegensatz zur Diminution, die das bezeichnete Objekt in quantitativer Hinsicht modifiziert: Pflänzchen bezeichnet ein kleines Exemplar von Pflanze, berührt also die quantitative Seite des Objekts selbst, während der Plural Pflanzen nur das mehrfache Vorkommen der gleichbleibenden Einheit Planze bezeichnet. Desweiteren werden flektierte Wörter meist von der Syntax gefordert, d.h. es besteht bzgl. der Flexionsform keine Wahlfreiheit. So etwa paßt in den syntaktische Rahmen sie ... einen Baum nur ein Verb in der 3.Sg., wobei das Tempus in diesem Fall Präsens wie Präteritum sein kann: planzt(e), ebenso: dieser Baum ist (größer) als der andere. Die Flexionsformen sind also durch die syntaktische Umgebung festgelegt (Plank 1981) und unterliegen damit einem hohen Grad an Obligatorik. Dagegen sind Derivationsbildungen syntagmatisch autonom.

270

Damans Niìbling

Figur 1 : Bedeutungskonstanz und Paradigmatizität als zentrale Unterscheidungskriterien zwischen Derivation und Flexion [Bedeutungskonstanz]

Weitere prototypische, doch nicht obligatorische Eigenschaften sollen kurz genannt werden (s. hierzu Bloomfield 1933, Plank 1981, Wurzel 1984 und 1988, Dressler 1989, Lehmann 1989 und Booij in diesem Band). Flexionskategorien können im Satz mehrfach markiert werden (Kongruenz) (was fur die Derivation nicht gilt): die kleinen Hunde bellen. Durch die Derivation kann sich die Wortart verändern, durch die Flexion nicht. Derivative (im Sinn von Derivationsaffixen) sind oft noch von ihrer Basis abtrennbar (eß- und trinkbar, be- und entladen), Flexive nicht (*sie iß- und trinkt). Derivative stehen - wie erwähnt - näher bei der Basis als Flexive. Bei Sprachkontakt werden eher Derivations- als Flexionsaffixe entlehnt. Flexive sind prinzipiell wortartsensitiver als Derivative. Tendenziell sind Flexive eher unbetont und entsprechend mit Schwachdrackvokalen ausgestattet; ebenso sind die phonotaktischen Verknüpfungsregeln einfacher und das gesamte Flexiv kürzer. Dies alles gilt in geringerem Maße für Derivative. Portmanteaumorphefindensich eher in der Flexion als in der Derivation, d.h. der Fusionsgrad ist unter Flexiven höher. Auch werden Flexive - als Folge ihrer Allgemeingültigkeit, Obligatorik usw. - viel öfter verwendet als Derivative. Junkturen treten seltener in Flexions- als in Derivationsformen auf. Generell sind Flexive - und dafür sprechen die meisten genannten Eigenschaften - grammatisierter als Derivative. Zusammenfassend halten wir als die wichtigsten Eigenschaften von Flexion im Gegensatz zur Derivation die folgenden Punkte fest: Paradigmatizität (formale Parallelität): Im Idealfall sollte die gesamte Wortart nach einer bestimmten Kategorie flektieren; seltene Ausnahmen gibt es dabei immer auch in den klassischen Flexionsparadigmen (z.B. Singularia- bzw. Pluraliatantum bei Substantiven, keine flexivischen

Wann werden die deutschen Präpositionen flektieren?

271

Infinitive und Preterita bei engl. Modalverben). Keine Behandlung erfährt dabei die Frage, ob auch bei einem Kategorienbündel (z.B. Genus/Numerus/Kasus) sämtliche potentiellen Kombinationen flexivisch-affigierend realisiert werden müssen. Diese Frage wird ausfuhrlicher in Kap. 3 im Zusammenhang der zunehmenden Markierung nominaler Kategorien an der Präposition im Deutschen aufgegriffen. Im Idealfall zeichnen sich die Basis und die Flexive durch strikte Parallelität aus. Dies hat u.a. zur Folge, daß innerhalb der Flexion mehr Analogien wirken als innerhalb der Derivation. Produktivität: Auch bei neu entlehnten Basen sollte im Idealfall Flektierbarkeit bestehen, d.h. die Flexionsregeln sollten produktiv sein. Erstarrt solche Produktivität (wie im Fall der englischen Adjektivkomparation), ist der erste Schritt zur Deflexion vollzogen. Bedeutungskonstanz -> Vorhersagbarkeit (semantische Parallelität): Basis und Flexiv/e zeichnen sich jeweils durch ein Höchstmaß an gleichbleibender Bedeutung aus, d.h. es kommt nicht zu inhaltlichen Amalgamierungen, die zu einer u.U. ganz anderen, nicht erschließ- und vorhersagbaren Bedeutung fuhren. Mittlere Bedeutsamkeit bei hoher Allgemeingiiltigkeit: Semantisch sollten Flexive zur Basis in einem mittleren Relevanzverhältnis stehen (wie immer man dieses Maß bestimmen mag), so daß sich die obligatorische morphologische Bindung überhaupt lohnt. Das heißt, die Flexivbedeutung darf die Basisbedeutung nicht direkt affizieren. Prototypische Flexionskategorien sind daher von hohem Allgemeingültigkeitsgrad. Da sie immer mit der Basis kompatibel sind, lassen sich komplette Paradigmen bilden (Reihenbildung). Obligatorik: Flexionsformen werden meist durch die syntaktische Umgebung erfordert und unterliegen damit (entgegen den Derivationsformen) starker Obligatorik. Selektivität: Flexive richten sich in ihrer Distribution und Bindung an die Basis nach ganz bestimmten Wortarten (Derivative nicht unbedingt). Kongruenz: Flexive kongruieren oft untereinander. Portmanteau: Flexive fusionieren öfter untereinander (im Vergleich zu Derivativen). Entfernung zur Basis: Flexive schließen das Wort jeweils nach außen hin ab (etwaige Derivative kommen zwischen Basis und Flexiv/e zu stehen). Keine Movierbarkeit: Flexive verändern die Wortart nicht (Derivative oft).

272

Damaris

Nübling

Keine Koordinierbarkeit: Flexive sind nicht von ihrer Basis abtrennbar (Derivative öfter). Keine Paraphrasierbarkeit: Flexionsbildungen sind nicht umschreibbar (Derivationsbildungen viel eher). Materielle Eigenschaften: Flexive sind im Vergleich zu Derivativen kürzer, phonotaktisch einfacher strukturiert, öfter unbetont und unsilbisch. Hohe Tokenfrequenz: Flexive zeichnen sich durch extrem hohe Tokenfrequenzen aus (als Folge ihrer Allgemeingültigkeit, Obligatorik und Neigung zur Kongruenz). Hohe Grammatizität: Flexive sind prinzipiell extrem grammatisiert. Trotz der durchgängigen Skalarität und der vielen Affinitäten zwischen Derivation und Flexion gibt es erstaunlich wenige Beispiele fur den Übergang des einen in das andere Verfahren. Theoretisch wäre es durchaus denkbar, daß sich ein relativ allgemeingültiges und bedeutungskonstantes Derivativ aus seinen Restriktionen löst, an Frequenz gewinnt und produktiv wird. Dies ließe sich etwa für das deutsche Kollektivpräfix Ge- vorstellen, indem es sich zumindest auf die Konkreta ausweitet nach dem partiell reihenbildenden Muster Ast - Geäst, Berg - Gebirge, Busch - Gebüsch. Stattdessen sind auch hier Lexikalisierungen an der Tagesordnung (Wetter - Gewitter, Stute Gestüt, Flügel - Geflügel), ebenso Isolierungen (Gemüse - Mus?, Geländer - ?, Geweih - ?), ebenso Blockierungen (Baum - *Gebäum(e), Fels - *Gefelse) (s. hierzu Fleischer/Barz 1992). Dennoch gibt es hier zuweilen Ad-hoc-Bildungen wie ?Gefräß (zu Gesöff), die eine gewisse, wenn auch sehr schwache Produktivität und Bedeutungskonstanz erkennen lassen. Als ein Beispiel fur die Entwicklung Derivation —> Flexion wird gelegentlich die Entstehung der ahd. Pluralendungen am Substantiv aus einstigen stammbildenden Suffixen gesehen (Werner 1969, Lehmann 1989). Allerdings scheinen diese Suffixe zum Zeitpunkt ihrer "Flexivierung" keine festen Bedeutungen mehr gehabt zu haben, d.h. sie waren bereits desemantisiert und lieferten nur das Material für die späteren Pluralflexive. Damit findet keine eigentliche Umfünktionierung statt, d.h. kein eigentlicher Wandel von der Derivation zur Flexion. Als ein Beispiel fur die Entwicklung Flexion Derivation könnte die genannte deutsche Kausativbildung betrachtet werden, falls sie im Germanischen systematisch realisiert wurde (das Gotische liefert hierfür Anhaltspunkte). Als weiteres Beispiel kann die relativ systematische Adverbbildung im Ahd. mit -I ch gesehen werden, die sich heute zu einem fast unüberschaubaren Bedeutungsspektrum von derivationellem -lieh aufgesplittert hat. Plank 1981 führt das ursprünglich genitivische -s an, das zunehmend für die Bildung von Adverbien und Präpositionen (also zur Wortbildung) verwandt wird: nachts, namens, allerdings, falls (Adverbien); zwecks, jenseits, betreffs, seitens (Präpositionen; übrigens dann immer mit Genitivrektion). Doch auch der Weg Flexion —»

Wann werden die deutschen Präpositionen flektieren?

273

Derivation kann nicht als prototypisch angesehen werden. Vielmehr gilt: Flexive speisen sich aus der Syntax und Derivative aus der Wortbildung, genaugenommen aus Kompositionsgliedern. Diese prototypischen Pfade veranschaulicht Figur 2. Figur 2: Prototypische Übergänge zwischen unterschiedlichen Ausdrucksverfahren

^ Lexikon

Grammatik

Zu Figur 2 ist zu ergänzen, daß - wie die Pfeile andeuten - der Weg vom Lexem über die Allegroform und das Klitikon zum Flexiv unidirektional ist1. Ebenso ist auch der Weg vom Lexem über das Kompositionsglied zum Derivativ unidirektional; die in die entgegengesetzte Richtung weisenden Pfeile von der Komposition und Derivation zum Lexem sind so zu verstehen, daß die gesamte Verbindung (und nicht etwa ein einzelnes Kompositionsglied bzw. Derivativ) zu einem Simplex verschmelzen kann (so z.B. ahd. wer-alt 'Zeitalter' > nhd. Welt). Hier nicht vertieft werden können die direkten, abrupten Entwicklungen wie die vom Lexem zur Flexionsform (hier gestrichelter Pfeil, da relativ selten); hierunter fällt z.B. lexikalische Suppletion wie engl, go - went. Die im Titel dieses Beitrags genannte Schnittstelle zwischen Derivation und Flexion erweist sich also vornehmlich in typologischer Hinsicht als eine solche. Was die diachrone Dimension betrifft, so liegt hier eine nur schwer durchlässige Grenze vor, die in Figur 2 durch die unterbrochenen Pfeile

1

Zu der seltenen Ausnahme der Entwicklung eines Flexivs zurück zum Klitikon sei auf Nübling 1992:1 Π Ι 18 verwiesen.

274

Damaris Nübling

liegt hier eine nur schwer durchlässige Grenze vor, die in Figur 2 durch die unterbrochenen Pfeile symbolisiert wird. Vielmehr entwickelt sich Flexion prototypischerweise über die Morphologisierung ehemals syntaktischer Einheiten. Hiervon handelt der folgende Abschnitt. 2.2 Syntax —> Klise —> Flexion: Vielbegangener Grammatisierungsweg Mit der Entwicklungsrichtung Syntax —> Klise —> Flexion betreten wir nun einen prototypischen Grammatisierungspfad. Häufig auftretende Funktionswörter (meist Artikel, Pronomina, auch Präpositionen, Konjunktionen oder Hilfsverben) verschmelzen zuerst im Allegro-, später auch im Lentostil gesprochener Sprache mit benachbarten Wörtern. Dabei wird die Enklise, d.h. die Bindung an ein vorangehendes Wort, in den europäischen Sprachen eindeutig präferiert (zur Suffixpräferenz s. Wandruszka 1992). Je nachdem, wie fest eine solche Verschmelzung ist, ob sie obligatorisch erfolgt oder nicht, ob das Klitikon zu seiner (selbständigen) Vollform in einem synonymen oder in einem oppositiven Verhältnis steht, ob seine Distributionsregeln syntaktisch oder morphologisch gesteuert sind usw., unterscheidet man zwischen einfachen (weniger grammatisierten) und speziellen (stärker grammatisierten) Klitika. Erfolgt die klitische Bindung obligatorisch und erlischt jegliche Austauschbarkeit mit seiner (einstigen) Vollform, ist Flexion erreicht (s. Nübling 1992). Die in der Überschrift angedeutete Skala läßt sich durch die genannten Klisetypen und die Allegroverschmelzung präzisieren: Syntax —» Allegroverschmelzung —> einfache Klise —> spezielle Klise - » Flexion. Beispiele für solche vollzogenen Grammatisierungen (Flexivierungen) lassen sich reichlich und in unterschiedlichen Sprachen belegen. Einschlägige Fälle bestehen in der Herausbildung des synthetischen Futurs und Konditionals im Frühromanischen (vlat. cantare habeo > *ccmtar-abeo > cantar-aio > span, cantar-é, frz. (je) chanter-ai 'ich werde singen'), in der Entstehung des synthetischen Passivs in den skandinavischen Sprachen aus dem altnordischen Reflexivpronomen sik (sog. s-Passiv) und in der Suffigierung des definiten Artikels ebenfalls in den skandinavischen Sprachen. Schließlich sind auch die neuen, festen Personalendungen am Verb, die aus enklitischen Personalpronomina hervorgingen, zu nennen, wie sie für Walser Sprachinseln am Südrand der Schweiz bzw. am Nordrand Italiens gelten. Diese erstarrten Pronomina treten in Kongruenz zum Subjekt. Solche eindrucksvollen Pleonasmen sind bei Zürrer 1982 und 1997 belegt: ich tueni 'ich tue', du tuescht, e'r tueter, dschi tuetdsch, äs tuets, wir tieber, fr tieder, dschi tiendsch. Im Fall der 3 .Sg. wird sogar das Genus des Subjekts am Verb markiert (eer tue-t-er, dschi tue-t-dsch, äs tue-ts), d.h. mit dem morphologischen Aufbau ist auch ein kategorieller verbunden. Doch zeichnen sich bereits Tendenzen zum Abbau der Genuskongruenz zugunsten einer einheitlichen Neutrumform ab (eer/dschi/äs tuets) (zu Näherem s. Zürrer 1997, auch Dal Negro 1997). Offensichtlich ist die Bedeutsamkeit des Subjektgenus für die Verbalhandlung zu gering, als daß sich hierfür eine dauerhafte morphologische Verbindung lohnte.

Wann werden die deutschen Präpositionen

flektieren?

275

3. Die Präposition-Artikel-Enklise im gegenwärtigen Deutschen Einen Paradefall, an dem man die sukzessive Entstehung von Flexion studieren kann, liefert das gegenwärtige Deutsche einschließlich seiner Umgangssprachen und Dialekte. Da sich dieser grammatisch-morphologische Sprachwandel in voller Bewegung befindet, läßt er sich nur schwer greifen. Viele Deutsche - und dies ist deutliches Indiz für die Dynamik dieses Wandels - bewerten die Akzeptanz etwa der folgenden Verschmelzungsformen unterschiedlich: Sagt man untern Stuhl oder unter den Stuhl? Sie kommt mit dem, mit'm [mitm], mip'm [mipm] oder mim [mi'm]/[mim] Fahrradi Heißt es er arbeitet im oder in dem Garten seines Brüdenfi Kann man noch sagen sich bei dem Radfahren erkälten oder vielmehr beim Radfahren? Oft bestehen auch beide Möglichkeiten nebeneinander. Meist aber entzieht sich die Frage der obligatorischen, der potentiellen und der absolut unmöglichen Verschmelzung der bewußten Beobachtung durch die Sprecher. Die Frage, ob unsere Präpositionen schon flektieren, ist vorab mit einem klaren Nein zu beantworten, wenngleich in einigen Punkten bereits flexionsartige Verhältnisse erreicht sind. Ein zentrales Kriterium für Flexion ist z.B. ihre Paradigmatizität: Im Idealfall sollte die gesamte Wortart flektierbar sein. Von solchen Verhältnissen sind die Präpositionen mit ihren suffigierten Artikelresten noch deutlich entfernt. Vielmehr eröffiiet sich hier ein ganzes Spektrum an Möglichkeiten, das von unverschmelzbaren Syntagmen (in die/tinne Stadt) über einen großen Zwischenbereich mit beiden Optionen (vor dem/vorm Haus) bis hin zu obligatorischen, d.h. nicht mehr auflösbaren Verschmelzungen reicht (im/*in dem Schwarzwald). Diese lange Skala zwischen Verschmelzungsblockade einerseits und Verschmelzungsobligatorik andererseits wird durch ein ganzes Bündel von Parametern bestimmt, das die chaotisch erscheinenden Verhältnisse bewirkt und im folgenden entwirrt werden soll. Bevor wir durch die Anwendung der im vorangehenden Abschnitt gewonnenen Flexionseigenschaften auf die Verschmelzungsformen deren Flexivitätsgrad ermitteln (hierzu s. 3.2), beleuchten wir zuerst die Bedingungen, unter denen die Präposition mit dem Artikel verschmilzt. Schon die Existenz solcher Verschmelzungsbeschränkungen spricht gegen einen hohen Flexivitätsgrad der Verschmelzungsformen: Wenn nur ganz bestimmte Präpositionen mit ganz bestimmten Artikelformen verschmelzen und zusätzlich Faktoren frequenzieller, medialer und textueller Natur die Klitisierbarkeit beeinflussen, kann es sich nur um ein in seiner Entstehung begriffenes System handeln. Kein Flexionssystem steht in Abhängigkeit zu den folgenden Verschmelzungsparameter. 3.1 Determinanten bei der Verschmelzung von Präposition und Artikel Bei den folgenden Faktoren, die die Verschmelzbarkeit des definiteti, bedingt auch des indefiniten Artikels mit der Präposition steuern, sei auf Figur 3 hingewiesen.

276

Damaris Näbling

Figur 3 : Das Kontinuum der Präposition-Artikel-Verschmelzungen im Deutschen

tauschbarkeit

tauschWteit

• Medium: Prinzipiell finden sich in der gesprochenen Sprache weitaus mehr Verschmelzungsformen als in der geschriebenen Sprache. Wie es sich im einzelnen mit der sukzessiven Integration von Verschmelzungsformen in die Schreibung verhält, kann hier nicht ausfuhrlich thematisiert werden; die wichtigste Gruppe der orthographisch akzeptierten Verschmelzungsformen im, am, zur etc. befindet sich in der oberen Ellipsenhälfte in Figur 3, d.h. diese sind gleichzeitig extrem grammatisiert. Der indefinite Artikel wird in der Schriftsprache nicht mit der Präposition verschmolzen. • Allegro-ZLentostil : Innerhalb des Mediums der gesprochenen Sprache spielt außerdem das Sprechtempo eine Rolle: Viele (nicht alle) Verschmelzungen kommen nur bei Allegro vor - auch und vor allem solche des indefiniten Artikels: in'n, in(n)e, aufm, aufnem etc. • Nähe-/Distanzgrad des Textes: Wichtiger als die mediale ist die konzeptuelle Dimension eines Textes (zu dieser Unterscheidung s. Koch/Oesterreicher 1985 und 1994): Typisch (konzeptuell) nähesprachliche Textsorten wie vertraute Gespräche unter zwei Personen, Werbung, Sprechblasen in Comics etc. weisen eine viel höhere Rate und auch größere Bandbreite an Verschmelzungsformen auf als distanzsprachliche Texte (z.B. Fachtexte). Wie oft welche Präpositionen mit welchen

Wann -werden die deutschen Präpositionen flektieren?

277

Artikelformen verschmelzen, hängt also nicht nur vom Sprachsystem selbst ab, sondern auch von den äußeren Kommunikationsbedingungen. • Regionale Herkunft der Sprecher: Viele unterschiedliche Bewertungen bei der Akzeptanz von Verschmelzungsformen erklären sich aus der Herkunft, dem regio- bzw. dialektalen Hintergrund der betreffenden Person. Da die Präposition-Artikel-Enklise in Dialekten prinzipiell weiter gediehen ist als in der Standardsprache, ist bei Personen mit starkem dialektalem Hintergrund der Hang auch zur umgangssprachlichen Enklise größer. So erkennt man Ruhrdeutsch-Sprecher an der häufigen Verwendung von Formen wie inner 'in der* und irme 'in die'. • Definit-/Indefinitheit des Artikels: Der definite Artikel neigt viel stärker zu Verschmelzungen mit der Präposition, und zwar sowohl was die Häufigkeit betrifft - Formen wie im, vom, beim sind selbst in distanzsprachlichen, geschriebenen Texten überaus geläufig, meist sogar obligatorisch -, als auch was den phonologischen Verschmelzungsgrad betrifft: Die Enklise des definiten Artikels kann die formale Integrität der Präposition beeinträchtigen: zu [tsu:] + (de)m -» zum [tsum] (Vokalkürzung), i(n + de)m —> im, mi(t + de)m -> mim [mim] (Auslautreduktionen). Das heißt, die präpositionale Basis, an die sich das ArtikelsufEx heftet, bildet bereits eine gebundene Wurzel und kein selbständiges Wort mehr. In Verbindung mit dem indefiniten Artikel bleibt die Präposition dagegen immer unbeschadet, d.h. hier entsteht keine Wurzelallomorphik: zu [tsu:] + (ei)nem zu'nem [ltsu:nam], in + (ei)nem inÇn)em ['mam], mit + (ei)nem mit'nem ['mitiram]. In der Regel bleibt der indefinite Artikel silbisch. Dagegen hat der enklitische definite Artikel in vielen Fällen seinen Silbenstatus aufgegeben (zunem vs. zum; zuner vs. zur), und im Gegensatz zum indefiniten Artikel verschmelzen nicht alle seine grammatischen Formen gleichermaßen mit der Präposition: Verschmelzungen mit dem der- und vor allem mit dem iÄe-Artikel sind viel seltener. Damit kommen wir zu einem weiteren zentralen Parameter, der die Verschmelzbarkeit steuert, nämlich der • Flexionsform des definiten Artikels: Als potentielle Enklitika kommen nur die Dativ- und Akkusativformen des Artikels in Betracht; der Genitivartikel kann hier vernachlässigt werden. Die Objektformen des Artikels weisen große Unterschiede bezüglich ihrer Verschmelzungsfreudigkeit auf (s. die Zahlenangaben in Tab. 1): Am weitaus häufigsten und hier oft sogar obligatorisch verschmilzt (1) der dem-Artikel mit einer Präposition (can, hinterm, aufm), gefolgt vom (2) dasArtikel (ins, aufs). Seltener und nie obligatorisch verschmilzt der dew-Artikel (in'n, vorn, bei'ri). Dabei scheint der Mask.Akk. Sg.-Artikel (3) eher Verschmelzungen einzugehen als der homophone Dat.Pl.-Artikel (4): untern Tisch, aber übern (?) Wolken. Trotz der Vollformhomophonie verlaufen die Klitisierungen gestaffelt2. Noch seltener, d.h. stärker an die Faktoren Nähesprache, Mündlichkeit

2

Diese Beobachtung verdient eine systematische Untersuchung und Erklärung; offensichtlich scheinen diese beiden den-Artikel getrennt abgespeichert zu werden.

Damans Níibling

278

und Allegro gekoppelt ist die Enklise des (fer-Artikels (5) - abgesehen von der singulären, häufig vorkommenden und stark grammatisierten Form zur. Dagegen sind Verbindungen wie in'er weitaus weniger üblich, d.h. allenfalls an regionale Umgangssprachen gebunden (im Ruhrdeutschen etwa sind sie geläufig). Das Schlußlicht bildet der rf/e-Artikel sowohl im Singular (6) wie im Plural (7): * inné Stadt/*aufe Straße(n) (zumindest im Fem.Sg. sind auch solche Formen im Ruhrdeutschen geläufig). In allen Varietäten ist die Enklise des Akk.Pl.-Artikels am stärksten blockiert. Tabelle 1 : Objekt-Artikelformen und Angabe ihrer Enklisehäufigkeit

Sg.

Dativ

Akkusativ

Fem.

der 5

die 6

Mask.

dem 1

den 3 das 2

Neutr. PI.

den 4

die 7

Betrachtet man die numerierte Rangfolge in Tab. 1, läßt sich eine kategorienspezifische Steuerung der Verschmelzung nur ansatzweise erkennen. Bezüglich des Numerus läßt sich eine allgemeine Enklisepräferenz der Singular- gegenüber den Pluralformen erkennen (doch eher ?übem Wolken als *inne Stadt). Bezüglich des Kasus fuhrt zwar der Dat.Mask./ Neutr.-Artikel dem mit großem Abstand, doch folgt dann der Akk.Neutr.-Artikel das (weit vor dem Dat.Fem.-Artikel der). Und schließlich fuhrt auch die Frage nach dem Genus nur zu dem Befund, daß der Mask.- und Neutr. Artikel eher verschmelzen als der Fem.-Artikel (bis auf singuläres zur). Es ergibt sich eine gewisse Polarität, wenn man die extrem enklisefreundlichen Kategorien +mask./+neutr., +sg., +statisch (Dat.) den enklisefeindlichen Kategorien +fem., +pl., +direktional (Akk.) gegenüberstellt (wobei nur bei den wenigen Präpositionen mit Doppelrektion der Dat. mit +statisch und der Akk. mit +direktional korreliert; vgl. dagegen zu (+dir.) + Dat.). Doch herrschen zwischen diesen beiden Polen nur sehr bedingt skalare Verhältnisse, d.h. es läßt sich keine eindeutige kategorielle Determinierung der Verschmelzung beschreiben. Hier müssen weitere Parameter wirken. Ein solcher Parameter ist die • phonologische Struktur der Artikelform. Zwar beginnen alle definiten Artikelformen mit einem d-, doch ist der Auslaut jeweils unterschiedlich beschaffen. Von dem und das über den, der [dee] und die läßt sich eine gewisse Zunahme an Sonorität feststellen - mit Ausnahme von auslautendem s in das. Abgesehen von dieser Ausnahme könnte man die vorsichtige Regel aufstellen: Je sonorer der Artikelauslaut, desto geringer die Verschmelzbarkeit der Artikelform. Ein anderes denkbares Kriterium ist der Nexus, der entsteht, wenn man jeweils den Vokal tilgt: dem > *dm, das > *ds [ts], den > *dn. Während bei dem > *dm inhomorgane Konsonanten

Wann werden die deutschen Präpositionen flektieren?

279

aufeinandertreffen, besteht bei ts und dn Homorganität (Apikoalveolare). Möglicherweise spielt dieser phonotaktische Fakor eine Rolle. Schwerer zu artikulierende potentielle Phonotagmen (*dm) werden in der Enklise eher vereinfacht als leichter zu bildende ([d(9)u], [di(:)]). Insgesamt scheint jedoch die heutige phonologische Struktur der Artikelform keine dominante Rolle zu spielen, was damit zusammenhängt, daß der phonologische Verschmelzungsprozeß ins Mhd., teilweise sogar ins Ahd. zu datieren ist (s.u.) • Typ der Präposition: Es wäre viel zu kurz gegriffen, nur den Artikel für die Verschmelzbarkeit verantwortlich zu machen, denn die Artikelformen verschmelzen nicht mit allen Präpositionen gleichermaßen. So ist umgangssprachlich die Verschmelzung überm Haus ohne weiteres möglich, nicht jedoch mit der morphologisch komplexen Präposition gegenüber: */gegertüberm Haus. Als mit Abstand verschmelzungsfreudigste Gruppe erweisen sich die fünf Präpositionen in, an, von, zu und bei (s. Figur 3, oberste Ellipsenhälfte). Hierunter gibt es viele Verschmelzungen, die obligatorisch, also nicht mehr auflösbar sind (z.B. im/*in dem Schwarzwald). Alle diese Präpositionen sind alt, einsilbig, morphologisch eingliedrig strukturiert und extrem grammatisiert; oft fungieren sie auch als besonders desemantisierte Verbalpräpositionen (denken an, halten von, gehören zu). Gefolgt wird diese Gruppe von auf -r [B] auslautenden Präpositionen wie über, unter, vor, hinter, für. Weniger leicht verschmelzen jedoch ebenfalls morphologisch eingliedrige Präpositionen wie nach, aus oder mit. Dies fuhrt uns zum nächsten Kriterium, dem • Auslaut der Präposition: Gewisse Regularitäten ergeben sich, wenn man den Auslaut der Präposition betrachtet. Es hat den Anschein, daß hohe Sonorität erhöhte Verschmelzbarkeit bewirkt: zu und bei lauten vokalisch aus, die Gruppe auf -(e)r ebenfalls, wenn man die Vokalisierung von auslautendem -r berücksichtigt (vor [fem], über [y:bc]). Auch die auf Nasal auslautenden Präpositionen in, an, von verfugen über relativ sonore Auslaute - im Gegensatz etwa zu auf, aus, nach und mit, deren obstruente Auslaute die Verschmelzbarkeit eher behindern. Formen wie mit'm/mim, aufm, aus'm, nach'm sind in der Schreibung ausgeschlossen bzw. hochgradig markiert, im Gesprochenen jedoch durchaus geläufig3. Allerdings bleiben sie immer durch die entsprechenden Syntagmen austauschbar - im Gegensatz zu den viel stärker grammatisierten und auch orthographisch akzeptierten Verschmelzungsformen im, am und zum (im/*in dem Schwarzwald). Nun gibt es auch sehr sonor auslautende Präpositionen wie wegen, neben, ohne, die nicht in gleichem Maß zu Negern oder nebem verschmelzen wie von zu vom oder in zu im. Hier greift die Auslautregel offensichtlich nicht. Der entscheidende Faktor, der hier hinzutritt, ist der der • Tokenfrequenz der Präposition. Dem Frequenz-Wörterbuch von Ruoff21990 zufolge sind die zehn häufigsten Präpositionen die folgenden: in, mit, von, an, auf, zu, bei, nach, um, für. Dabei ist zu betonen, daß schon diese zehn Präpositionen fast 90% aller im laufenden Text vorkommenden Präpositionen stellen. Unter diesen Präpositionen befinden sich die sehr konsonantisch auslautenden und dennoch (im Gesprochenen) verschmelzenden Präpositionen auf nach und mit {aus folgt auf 3

Im standardsprachlichen (geschriebenen) Luxemburgischen gibt es - neben vielen anderen Verschmelzungsformen - die reguläre Enklise von no + dem —> nom 'nach dem1 und mat + dem zu mam 'mit dem', ohne daß regulärer f-Schwund bestünde.

Damaris Nilbling

280

Rang 11), d.h. Hochfrequenz fördert die Verschmelzung selbst dann, wenn der Auslaut wenig sonor und damit einer Verschmelzung eigentlich hinderlich ist. Umgekehrt verschmelzen wenig frequente und sehr sonor auslautende Präpositionen wie ohne oder neben kaum. Damit spielt die reine Gebrauchsfrequenz der Präposition eine entscheidende Rolle. Versucht man, auch die fünf Artikelformen dem, das, den, der und die nach ihren Gebrauchsfrequenzen anzuordnen, gelangt man jedoch kaum zu einem verschmelzungsfördernden Faktor: Meier 1968 ermittelt ausgerechnet die als die häufigste Artikelform, wobei hier auch die nominativischen Formen enthalten sind. Auf die folgen der, den, das und dem, d.h. die Gebrauchsfrequenz dieser Artikelformen steht ihrer Verschmelzungshäufigkeit diametral entgegen. Da hier jedoch, wie gesagt, auch die Nominativformen enthalten sind, ist es von zentraler Bedeutung, die sog. • Kookkurrenzfrequenz zwischen Präposition und Artikel zu betrachten, d.h. die Häufigkeit des gemeinsamen Auftretens von Präposition und Dativ- oder Akkusativartikel. Eigene Auszählungen hierzu (Nübling 1992:221/2) haben zu dem eindeutigen Befund geführt, daß der de/w-Artikel die mit großem Abstand häufigste Artikelform hinter einer Präposition ist. Dies beruht v.a. darauf, daß gerade die häufigsten Präpositionen fast immer den Dativ und deutlich seltener den Akkusativ regieren (s. Tab. 2). Die extrem häufige syntaktische Nachbarschaft zwischen Präposition und Dativartikel macht es also plausibel, weshalb gerade der ifem-Artikel besonders viele, tiefgreifende und oft obligatorische Verschmelzungsformen bildet. Tabelle 2: Kasusrektion der 10 häufigsten Präpositionen des Deutschen Frequenz

Präposition

Dativ

Akkusativ

0) (2) (3) (4) (5) (6) (v (8 (9) (10)

in mit von an auf zu bei nach um fir

+ + + + + + + +

+

—

+ +

—

— —

+ + — — —

Nun ist die Präposition-Artikel-Enklise nicht nur eine phonologisch und frequenziell determinierte Koartikulationserscheinung; vielmehr gibt es nicht wenige Fälle, in denen die Verschmelzungsform obligatorisch wird, d.h. sie tritt in Opposition zur unverschmolzenen Form: im vs. in dem Garten, zur vs. zu der Schule. • Funktion/Skopus des deflniten Artikels: Die folgenden semantisch-funktionalen Differenzierungen zwischen Verschmelzung und NichtVerschmelzung betreffen nur die Spitze des Eisbergs, d.h. hauptsächlich die folgenden acht extrem häufigen Verschmelzungsformen im, ins, am, ans,

Wann werden die deutschen Präpositionen flektieren?

281

vom, zum, zur und beim. Alle setzen sich aus denfrequentestenund gleichzeitig sonor auslautenden Präpositionen in, an, von, zu, bei und dem dem- und das-Artikel zusammen. Diese Verschmelzungsformen treten unter bestimmten Bedingungen obligatorisch auf, wobei sich nicht immer eindeutige Relationen zwischen der Funktion des Artikels und seiner Verschmelzbarkeit herstellen lassen. Dennoch besteht die deutliche Tendenz, daß der Artikel in der Regel vor Unika verschmilzt, d.h. vor Substantiven, die sich nur auf ein einziges Referenzobjekt beziehen: sie fliegt zum/*zu dem Mond. Diese Regel betrifft auch Eigennamen, sofern sie einen festen (stets definiten) Artikel bei sich haben: sie wohnt im/*in dem Schwarzwald. Ebenso finden sich in Phraseologismen, wo eine echte Referenz auf das Objekt entfällt, zahlreiche und ansonsten sogar eher unübliche Verschmelzungsformen: ums/*um das Leben kommen; jmdn hinters/*hinter das Licht führen; jmdn übers/*über das Ohr hauen. Auch substantivierte Infinitive erfordern immer den verschmolzenen Artikel: zum/*zu dem Arbeiten. Verschmelzung gilt auch bei generischem Gebrauch des Artikels: beim/*bei dem Hund handelt es sich um ein Säugetier. Es gibt also nicht wenige Konstellationen, in denen die Präposition-Artikel-Verbindungen fest eingerastet und damit nicht mehr durch die Vollform ersetzbar sind. Dagegen fordert die spezifische Verwendung des definiten Artikels meist seine Vollform, d.h. die Verschmelzung ist blockiert, und dies besonders dann, wenn das Referenzobjekt durch ein längeres Attribut oder gar durch einen Relativsatz determiniert wird: Hans arbeitet in dem/*im Garten, den er letztes Jahr seinem Bruder abgekauft hat4. Doch kann auch die Verschmelzung eines spezifischen Artikels stattfinden, wenn ihm keine oder nur eine sehr kurze Bestimmung folgt: Hans arbeitet im/?in dem Garten seines Bruders. Damit sind Verschmelzung und Nichtverschmelzung nicht nur referenzsemantisch gesteuert, sondern auch von der Länge der Bestimmung: Die semantische Reichweite, der Skopus des enklitischen Artikels, erfaßt keine längeren Attribute. Es sei noch einmal betont, daß diese funktionale Aufspaltung des Artikels nur einen kleinen Teil der formal möglichen Verschmelzungsformen betrifft. In Beispielen wie sie geht vors/vor das Haus besteht Austauschbarkeit des enklitischen durch den freien Artikel, nicht jedoch bei sie ist im ( in dem) Haus. Ein reines Variantenverhältnis betrifft alle Formen des indefiniten Artikels: Jede seiner Verschmelzungen kann ohne Bedeutungsverschiebung durch den vollen Artikel ersetzt werden (zu'nem = zu einem). Der definite Artikel ist damit deutlich grammatisierter als der indefinite, was auch mit seiner vielfrüherenEntstehung korreliert: Während sich der definite Artikel schon im Ahd. (aus dem Demonstrativum) entwickelt, beginnt die Herausbildung des indefiniten Artikels (aus dem Zahlwort 'eins') erst im Mhd. Ein genauerer Blick auf die Diachronie offenbart auch, daß schon im Ahd. Verschmelzungen zwischen Präposition und dem neuen definiten Artikel bestanden: zemo 'zum' < zi themo, zeru 'zur1 < zi theru, imo 'im' < in themo, auch zen 'zu den' < zi then etc. Im Mhd. mehren sich die Beispiele und

4

Vgl. auch die Duden-Grammatik s1995:§540.4: "Nach Präpositionen, die mit dem Artikel verschmolzen sind, kann bezeichnenderweise kein erläuternder Relativsatz folgen. Also nicht: Fritz ist jetzt im Haus, das er sich letztes Jahr gekauft hat."

282

Damans Niìbling

Kombinationsmöglichkeiten beträchtlich, wobei wir hier auch richtige Zwischenformen zu greifen bekommen wie anme/amme 'am', vonme/vomme Vom'. Auch heute noch rein umgangssprachliche Formen aufm, cms'm sind in mhd. Texten absolut geläufig (als ûfine/ûfem bzw. ûzme/ûzzeni). Hier also vollzieht sich der Verschmelzungsprozeß, nicht im heutigen Deutsch. Als der definite Artikel verschmolz, lautete er noch mit der Spirans th- [ö] an, und er war zweisilbig. Hier lohnt es sich, nach phonologjschen Verschmelzungsfaktoren zu suchen; diese Einsicht macht die vielen umständlich-komplizierten Versuche einer synchronen Ableitung von im < in dem obsolet. Bemerkenswert ist dabei, daß genau die "Spitze des Eisbergs", die besonders grammatisierten Verschmelzungsformen in Figur 3, auch die ältesten sind. Zur besseren Übersicht sei zusammenfassend das Kontinuum der Verschmelzungsformen in drei größere Etappen untergliedert (s. Figur 3): (a) Funktionalisierung von Enklise vs. Syntagma (Vollform): Dieser grammatisierteste Bereich der Verschmelzungsformen wird in Figur 3 durch die obere (fettgedruckte) Ellipsenhälfte symbolisiert. Hier besteht eine Opposition zwischen Enklitikon und Vollform: Der gebundene Artikel tendiert, vereinfacht gesagt, zur generischen und spezifischen Interpretation (sofern das Attribut nicht zu lang ist), während der selbständige Artikel ana- und kataphorische Bezüge zu leisten vermag: sie geht zur Schule vs. sie geht zu der Schule, die .... Am besten läßt sich der Unterschied über die unterschiedlichen Skopen greifen. In ohnehin spezifischen Kontexten, also vor Eigennamen, Unika, substantivierten Infinitiven und innerhalb von Phraseologismen (wo keine echte Referenz stattfindet) sind Verschmelzungen obigatorisch, d.h. die Verwendung der Vollform wäre ungrammatisch. Als besonders verschmelzungsfreudig erweisen sich dabei die fünf extrem frequenten und sonor auslautenden primären Präpositionen in, an, von zu, bei in Verbindung mit dem dem-, teilweise auch dem ¿¿is-Artikel Die Schreibung etc. respektiert und dokumentiert diese Obligatorik. (b) Fakultative Verschmelzung (s. die untere Ellipsenhälfte): Hier betreten wir den großen Bereich der funktional unerheblichen Austauschbarkeit des enklitischem mit dem freien Definitartikel (doch nicht immer umgekehrt): sie geht vors vor das Haus. Bei (ana-) phorischem Gebrauch verbietet sich die Enklise. Auf dieser Grammatisierungsetappe befinden sich deutlich mehr Präpositionen und Artikelformen - auch der indefinite Artikel. Verschriftet werden solche Verschmelzungsformen zunehmend in überwiegend nähesprachlichen Textsorten (Comics, Briefe, e-mail-Mitteilungen). Dabei zeugt häufige Apostrophsetzung (als Auslassungssignal) von dem mittleren Grammatisierungsgrad dieser Verbindungen und der damit zusammenhängenden Unsicherheit der Schreiber: neben etc. (c) Verschmelzungsblockade (Beginn der Grammatisierungsskala in Figur 3): In der überregionalen Umgangssprache weitgehend ausgeschlossen sind Verschmelzungen von

Wann werden die deutschen Präpositionen

flektieren?

283

selten(er)en Präpositionen und prinzipiell des der- und c/ze-Artikels (von singulärem zur abgesehen); dies gilt jedoch nicht fur den indefiniten Artikel. Doch gibt es durchaus Gebiete wie das Ruhrdeutsche, in denen auch solche Blockaden (von Präposition + enklitischem der- und ¿Äe-Artikel) nicht mehr bestehen (ruhrdt. umme Ecke 'um die Ecke', aufer/aufe Straße 'auf der/die Straße' etc.). Schriftsprachlich verbieten sich jedoch alle Verschmelzungen jenseits der beiden Ellipsenhälften in Figur 3. 3.2 Wann werden die deutschen Präpositionen flektieren? Um nun auf die eingangs gestellte Frage zurückzukommen, wann die deutschen Präpositionen flektieren werden, bietet es sich an, die in Abschnitt 2.1 aufgelisteten prototypischen Eigenschaften von Flexion anhand der Präposition-Artikel-Enklise zu überprüfen. Die nur aus dem Kontrast zur Derivation resultierenden Merkmale (z.B. fehlende Movierbarkeit der Flexive) können hier vernachlässigt werden. Paradigmatizität: Bei der Frage nach der Paradigmatizität stößt man auf das Problem, daß Präpositionen keine geschlossene Wortklasse bilden; gerade bei neuen Präpositionen ergeben sich Abgrenzungsprobleme zu den Adverbien, Verben, Substantiven etc. (zu dieser Problematik der verschiedenen Präpositionalitätsgrade s. Lindqvist 1994). Beschränkt man sich auf die sog. primären Präpositionen, die die älteste Schicht stellen, monomorphematisch strukturiert und kurz sind und nur den Dat. und/oder den Akk. regieren {in, bei, von auf...), so ist seitens der Präposition das Paradigmatizitätskriterium zumindest dann erfüllt, wenn die Verschmelzung mit wenigstens einer Artikelform ausreicht und wenn auchdie gesprochene Sprache berücksichtigt wird (z.B. fürs, ?für'n, aber *för'e). Faßt man unter Präpositionen auch die sekundären und tertiären (komplexen) Präpositionalien, nimmt die Erfüllung des Paradigmatizitätskriteriums sukzessive ab (z.B. entlang dem -> *entlang'm). Mindestens ebenso wichtig ist m.E. die Frage nach der Paradigmatizität der potentiellen Flexive: Wie gezeigt, legen die unterschiedlichen Artikelformen bzw. Kategorienbündel eine ganz unterschiedliche Verschmelzungsfreudigkeit an den Tag: Am häufigsten und leichtesten verschmilzt der úfe/w-Artikel, regelrecht blockiert ist dagegen der cöe-Artikel. Spricht dieses heterogene Verhalten schon gegen Flexion? Angenommen, der de/w-Artikel würde sich mit jeder Präposition gleichermaßen obligatorisch verbinden (wovon er realiter gar nicht so weit entfernt ist): Spräche man dann von einer Mask /Neutr.Dat.Sg.-Flexion der Präpositionen, so wie man von der Dat.Pl Flexion der Substantive (Kind-er-n) spricht? Oder sollte man die Präpositionalflexion erst dann als erreicht betrachten, wenn das gesamte Artikelparadigma an die Präposition gebunden werden kann? Die Antwort hängt vom zugrundegelegten FlexionsbegrifF ab, Strenggenommen sollte vollständige Flexion voraussetzen, daß sämtliche Artikelformen suffigieren, ebenso wie im Ahd. sämtliche Kasus flexivisch am Nomen realisiert wurden. Die heute verbleibende Dat.Pl.-"Flexion" ist davon nur ein

284

Damaris Nübling

Relikt, d.h. hier besteht genaugenommen Deflexion. Damit ist es verfehlt, die verbleibende Dat.PlFlexion - in Dialekten ohnehin längst abgebaut - als idealtypischen Fall und damit als Maßstab für Flexion zu bewerten. Vielmehr repräsentiert die nominale Dat.Pl.-"Flexion" die letzte Etappe einstiger Kasusflexion und entsprechend die präpositionale Dat.Sg.Mask / Neutr.-"Flexion" die erste Etappe zukünftiger Flexion. Weiter ist zu fragen, ob die Paradigmatizität des Flexivs auch den indefiniten Artikel einschließt; d.h. sollte das gesamte Artikelsystem oder nur das des definiten Artikelsflexivischrealisierbar sein? Mit Blick auf den am Substantiv suffigierten definiten Artikel in den skandinavischen Sprachen, der der Flexion zugerechnet wird, ist diese Frage zu verneinen. Hier wird nämlich der Indefinitartikel entweder - wie im Deutschen auch - dem Substantiv vorangestellt (z.B. schwed. ett hus 'ein Haus', aber huset 'das Haus'), oder er wird überhaupt nicht materialisiert (wie im Isländischen: hús 'ein Haus', aber húsió 'das Haus'). Die Frage nach der Paradigmatizität der Flexive/Klitika betrifft auch das Problem der sog. flektierenden Konjunktionen etwa im Fränkischen oder Bairischen: Gehören solche obligatorischen Erweiterungen wie bair. wermst willst 'wenn du willst' zur Flexion? Die Meinungen hierüber gehen weit auseinander, doch ist in Hinblick auf die Flexionskriterien folgendes zu bedenken zu geben: Erstens ist von der Personalsuffigierung nicht nur die Wortart der Konjunktionen betroffen, sondern jedes nebensatzeinleitende Element (also auch Relativpronomina, Fragepronomina bis hin zu Adverbien, Adjektiven und sogar Substantiven), d.h. die Distribution von -st ist nicht primär morphologisch, sondern syntaktisch gesteuert. Zweitens ist hier - zumindest mit dieser Obligatorik vor allem -st, also nur die 2.Sg., betroffen. Produktivität: Gerade jüngere und damit komplexere, weniger frequente Präpositionen sperren sich oft gegen eine Verschmelzung, selbst dann, wenn sie formal eine ansonsten verschmelzungsfreudige Präposition enthalten (s. das Beispiel über vs. gegenüber, überm Haus , doch 1gegenüberm Haus *). Auch regieren viele neue Präpositionen den Genitiv, und der Genitivartikel klitisiert nicht. Das Produktivitätskriterium ist damit als nur partiell erfüllt zu betrachten. Bedeutungskonstanz —> Vorhersagbarkeit: Dieses Kriterium ist teilweise erfüllt und hängt vom Grammmatisierungsgrad sowohl der Präposition als auch der Artikelform ab. Zumindest innerhalb der oben aufgeführten Grammatisierungsetappen (a) "Funktionalisierung" bis (c) "Verschmelzungsblockade" ist dieses Kriterium jeweils erfüllt. Semantisch-funktionales Verhältnis zwischen Präposition und Artikel: Bemerkenswert bei der Präposition-Artikel-Enklise ist die Tatsache, daß die Präpisition den Artikel formal an sich bindet, obwohl der Artikel als Teil der NP (innerhalb der PP) strukturell dem Nomen nähersteht. Daß der Artikel sich mitsamt seinem Bündel nominaler Kategorien mit der Präposition verbindet, geschieht wahrscheinlich weniger aus Gründen funktionaler Nähe. Vielmehr spielen hier typologische und

Wann werden die deutschen Präpositionen flektieren?

285

syntaktische Gesichtspunkte die entscheidende Rolle: Zum einen besteht im Deutschen eine ausgeprägte Präferenz zur Suffigierung; auch die Personalpronomina ziehen die Enklise am Verb der Proklise eindeutig vor. Des weiteren ist die Fügungsenge zwischen Präposition und Artikel ungleich stärker als die zwischen Artikel und Nomen (Attribution); d.h. durch die direkte und zudem häufige syntaktische Nachbarschaft von Präposition und Artikel bietet sich auch eine Morphologisierung eher an5. Schließlich fugt sich die "Flexion" der die PP eröffiienden Präposition gut in das im Deutschen stark ausgeprägte Klammerverfahren. Wie Ronneberger-Sibold 1993 und 1997 zeigt, sind hiervon nicht nur verbale, sondern auch nominale Konstruktionen betroffen. So läßt sich seit dem Mhd. die zunehmende Markierung des Initialelements von Nominalphrasen beobachten (sog. Monoflexion). In diesen typologischen Trend stellt sie auch die Präposition-ArtikelEnklise: "The same is trae [...] for the cliticization of the definite articles after prepositions: zum, zur instead of zu dem, zu der. Although the primary motivation is clearly phonological, the structural result is a first step towards inflected prepositions. The desirable side effect is that in the prepositional phrase, too, the strong adjective inflection marks the leftmost element of the whole construction." (Ronneberger-Sibold 1993:308)

Auch andere neuere Entwicklungen im Deutschen sprechen für den weiteren Ausbau des Klammerprinzips. Obligatorik: Wie bereits dargestellt, erfolgt ein bedeutender Teil der Präposition-Artikel-Verschmelzungen obligatorisch in dem Sinn, daß eine Auflösung in das entsprechende VollformSyntagma entweder einfach ungrammatisch wäre oder aber zu einer anderen Referenzialisierung/Interpretation führen würde. Doch ist durchgehende Obligatorik noch nicht vorhanden. Kongruenz: Der suffigierte Artikel kongruiert formal mit dem Substantiv und dem Adjektiv (das jedoch nicht zwingend auftreten muß): Bezüglich des Numerus kongruiert er mit beiden, bezüglich des Kasus mit dem Adjektiv und nur partiell mit dem Substantiv (etwa im Dat.Pl.), und bezüglich des Genus und der Definitheit mit dem Adjektiv. Diese prototypische (doch nicht obligatorische) Eigenschaft von Flexion wäre also erfüllt. Selektivität: Hier kann nur angedeutet werden, daß sich der Artikel auch mit anderen ihm vorangehenden Wortarten verbindet (er hat's Kind gesehen). Doch im Gegensatz zur Verbindung mit der Präposition bleibt dieser enklitische Artikel immer mit dem vollen (unbetonten) Artikel austauschbar, ohne daß sich die Bedeutung ändert. Das heißt, hier liegt nur das Stadium der Allegroverbindung, allenfalls der einfachen Klise vor. Obligatorisch und mit den genannten Funktionalisierungen, die keine freie Austauschbarkeit mit der Vollform mehr erlauben, verbindet 5

Für das verschmelzungsfördernde Kriterium der Fügungsenge spricht auch der suffigierte Definitartikel in den skandinavischen Sprachen (vgl. isl. hús-iá 'Haus-das'): Im Altnordischen traten Adjektive hinter den postponierten Artikel (bzw. das Demonstrativum): an. hits it gamia 'Haus das alte', womit Nomen und Determinant durch direkte syntaktische Adjazenz verbunden waren.

286

Damans Nübling

sich der Artikel nur mit der Präposition. Besonders die grammatisierten Artikelformen -m und -s zeigen deutliche Ansätze zu morphologischer Selektivität. Keine Koordinierbarkeit: Bei der Koordination zweier Präpositionen ist der volle Artikel auslaßbar, der enklitische jedoch nicht mehr: in und an dem See, aber: *in und am See, sondern: im und am See. In dieser Hinsicht verhält sich der Artikel wie ein Flexiv. Ebenso muß auch die Präposition wiederholt werden, wenn sie mit dem Artikel verschmolzen ist und ihr ein weiterer, nicht klitisierbarer Artikel folgt: *man sprach vom Erfolg [...] und den weiteren Plänen (sondern: von den) (s. die Duden-Grammatik s1995:§540.6). Entfernung zur Basis: Die Artikelsuffixe verbinden sich direkt mit der Basis. Bemerkenswert ist, daß sie sogar die Basis selbst phonologisch affizieren, indem diese in bestimmten Verbindungen Wurzelallomorphe entwickelt: zu dem [tsu:] —» zum [tsum]; in dem i-m, mit dem -> [mim]. Hier ist also bereits eine Art Wurzelflexion eingetreten im Gegensatz zur ansonsten bestehenden "Grundformflexion" (an-s, vor-m). In morphonologischer Hinsicht haben sich also schon bemerkenswerte Überlagerungen zwischen Präposition und Suffix herausgebildet, was normalerweise eine Eigenschaft "älterer" Morphologie ist. Interessant ist, daß diese Amalgamierungen nur in Verbindung mit der grammatisiertesten Artikelform -m eintreten. Portmanteau: Der Portmanteau-Charakter der Artikelklitika wurde bereits genannt. Während der volle Artikel immerhin noch in ein d- mit der Information '+ définit' und das Flexiv mit den drei Informationen Genus, Numerus und Kasus segmentierbar ist, rückt in der Verschmelzung auch die Kategorie der Definitheit auf den minimalen Artikelrest: in d-em. Die Portmanteaustruktur verstärkt sich also mit der Klitisierung. Auch die phonologischen Überlagerungen zwischen Präposition und Artikel steigern die Portmanteauhaftigkeit. Keine Paraphrasierbarkeit: Nur im Bereich der Allegroverschmelzung und einfachen Klise besteht freie Austauschbarkeit mit dem vollen Artikel. Für die grammatisierteren speziellen Klitika entfällt jegliche Paraphrasierbarkeit. Materielle Eigenschaften: Die suffigierten Artikelformen sind kürzer und auch phonotaktisch einfacher strukturiert als ihre entsprechenden Vollformen. Meist bestehen sie aus nurmehr einem Laut. Ebenso sind sie unbetont, auch unbetonbar und meist unsilbisch (im, zur, doch aufm). Dieses Kriterium ist also hochgradig erfüllt. Vertritt man einen strengen Flexionsbegriff, so werden die Präpositionen dannflektieren,wenn alle (auch neue) Präpositionen sich mit allen Artikelsuffixen verbinden, und dies obligatorisch, d.h. es darf keinefreieErsetzbarkeit des gebundenen mit demfreienArtikel bestehen. Damit orientiert sich

Wann werden die deutschen Präpositionen

flektieren?

287

der Flexionsbegriff am Prototyp der lateinischen Substantivflexion und nicht etwa an der rudimentären Kasusmarkierung am deutschen Substantiv. Wirft man einen Blick auf die deutschen Dialekte, so schlagen die Präpositionen tatsächlich den Weg zur Flexion ein: Im Alemannischen suffigieren mehr Artikelformen an mehr Präpositionen, und hiervon sind der definite wie auch der indefinite Artikel gleichermaßen betroffen. Kontrastiert man solche Verbindungen, gelangt man zu interessanten morphologischen Strukturen: alem. im 'im' vs. ime 'in einem'; üfem 'auf dem' vs. (feme 'auf einem'. Während die Kategorien Genus, Numerus und Kasus direkt auf die Präposition folgen (-(e)m-), wird Indefinitheit systematisch durch -e, Definitheit durch -0 realisiert, i-m-0 vs. i-m-e, üf-em-0 vs. Qf-em-e. Auch hat sich das Verhältnis zur Vollform grundlegend gewandelt: Enklitischem -em [am] entspricht keine ¿/-haltige Vollform mehr, sondern der allgemeine Dat.Sg.Mask./Neutr.-Artikel lautet in allen Positionen em [am]: em Lehrer [9m_ls:RB] 'dem Lehrer". Da im Alemannischen kein systematischer ¿-Schwund im Anlaut besteht, kann nur davon ausgegangen werden, das sich das präpositional gebundene und entsprechend reduzierte Enklitikon -em aus seiner Restriktion gelöst und in alle syntaktischen Positionen auf Kosten der ¿-Form ausgebreitet hat; so wurde eine Reduktionsform zum Ersatz für seine einstige Vollform. Auch der unbestimmte Artikel lautet durchgehend eme 'einem' (kontrahiert aus mhd. eineme), sei es in freier oder in enklitischer Position. Der Grammatisierungsprozeß ist im Alemannischen also deutlich weiter gediehen als in der Standardsprache; ebenso ist auch das Ruhrdeutsche der Präpositionalflexion sehr nahe (zum Alemannischen s. Nübling 1992 und 1993). Möglicherweise hemmt die Schriftlichkeit diesen Grammatisierungsprozeß. Daß die Fusion von Präposition und Artikel keine Angelegenheit exotischer Sprachen zu sein braucht, belegt Stolz 1990: So verfugen das Portugiesische und besonders das Italienische über besonders viele und grammatisierte Präposition-Artikel-Verschmelzungen. Regelrechte Präpositionalflexion weisen z.B. die modernen keltischen Sprachen auf, wo sich das Personalpronomen mit der Präposition verbindet (zu den Flexionsparadigmen s. Stolz 1990:341). Dabei kommt es auch zu unterschiedlichen Flexionsklassen, im Irischen gar zu Suppletivformen. Die keltischen Verbindungen (Präp. + Pers.Pron.) bilden jedoch vollständige Präpositionalphrasen, was für das Deutsche nicht gilt. Auch für das Ungarische beschreibt Stolz deklinierende Postpositionen. Im Unterschied zum Deutschen sind es zwar nicht einstige Artikelformen, die flexiviert wurden, doch ist das Gemeinsame, daß nominale Kategorien an der Präposition ausgedrückt werden. Diese Fakten sprechen gegen die so oft postulierte Indeklinabilität von Präpositionen (bzw. Adpositionen).

4. Schluß Trotz vieler typologischer Ähnlichkeiten zwischen Flexion und Derivation läuft der prototypische Entstehungsweg von Flexion nicht über die Derivation, sondern über die Morphologisierung syntaktischer Einheiten. Für solche vollzogenen Flexivierungen lassen sich viele Beispiele anführen. Aufschluß über die Einzelheiten dieses komplexen Prozesses liefert die derzeitige Präposition-

288

Damans Niibìing

Artikel-Verschmelzung im Deutschen, die als eine Flexivierung in statu nascendi zu begreifen ist: In vielen Punkten sind bereits flexivische Verhältnisse erreicht, in anderen jedoch noch nicht. Der Beitrag hat gezeigt, daß eine Vielzahl teilweise heterogener Parameter diese Grammatisierang steuert. Mit Blick auf die deutschen Dialekte läßt sich eine weitere Grammatisierung prognostizieren. Auch die Fakten anderer Sprachen weisen die Indeklinabilität von Präpositionen zurück.

Literatur Bloomfield, Leonard (1933): Language. Chicago: University of Chicago Press. Booij, Geert (1998): The Demarcation of Inflection. (In diesem Band) Bybee, Joan L. (1985): Morphology. A study of the relation between meaning and form. Amsterdam/Philadelphia: John Benjamins. Dal Negro, Silvia (1997): Morphosyntaktischer Sprachverfall bei den Waisern in Pomati. In: Ruoff, Arno & Peter Löffelad (Hgg.): Syntax und Stilistik der Alltagssprache. Beiträge der 12. Arbeitstagung zur alemannischen Dialektologie. Idiomatica Bd. 18,191-195. Tübingen: Niemeyer. Dressier, Wolfgang U. (1989): Prototypical Differences between Inflection and Derivation. ZPSK 42, 3-10. - (1996): Comments on Wurzel's "On similarities and differences between inflectional and derivational morphology". STUF 49,280-285. Duden-Grammatik s1995. Mannheim: Dudenverlag. Fleischer, Wolfgang & Baiz, Irmhild (1992): Wortbildung der deutschen Gegenwartssprache. Tübingen: Niemeyer. Howe, Stephen (1996): The Personal Pronouns in the Germanic Languages. A study of personal pronoun morphology and change in the Germanic languagesfrom the first records to the present day. Berlin/New York: de Gruyter. Koch, Peter & Wulf Oesterreicher (1985): Sprache der Nähe - Sprache der Distanz. Mündlichkeit und Schriftlichkeit im Spannungsfeld von Sprachtheorie und Sprachgeschichte. Romanistisches Jahrbuch 36,1543. - (1994): Schriftlichkeit und Sprache. In: Günther, Hartmut & Otto Ludwig (eds.): Schrifi und Schrifilichkeit. Ein interdisziplinäres Handbuch internationaler Forschung, Bd. 1, 587-604. Berlin/New York: de Gruyter. Lehmann, Christian (1989): Grammatikalisierung und Lexikalisierung. ZPSK 42,11-19. Lindqvist, Christer (1994): Zur Entstehung von Präpositionen im Deutschen und Schwedischen. LA 311. Tubingen: Niemeyer. Matthews, P. H. (1972): Inflectional Morphology. A theoretical study based on aspects of Latin verb conjugation. Cambridge: University Press. - (1974): Morphology. An introduction to the theory of word-structure. Cambridge: University Press. Meier, Helmut (1968): Deutsche Sprachstatistik. Hildesheim. Nübling, Damans (1992): Klitika im Deutschen. Schriftsprache, Umgangssprache, aleman-nische Dialekte. ScriptOralia 42. Tübingen: Narr. - (1993): Synthesetendenzen im Alemannischen: Die Klitisierung von Artikel und Personalpronomen. In: Schupp, Volker (ed.): Alemannisch in der Regio. Beiträge zur 10. Arbeitstagung alemannischer Dialektologen in Freiburg/Breisgau 1990, 97-112. Göppingen: Kümmerle Verlag. Plank, Frans (1981): Morphologische (Ir-) Regularitäten. Aspekte der Wortstrukturtheorie. Tübingen: Narr. Ronneberger-Sibold, Elke (1993): Typological conservatism' and framing constructions in German morphosyntax. In: Van Marie, Jaap (ed.): Historical linguistics 1991. Papers from the 10th International Conference on Historical Linguistics, 295-314. Amsterdam/Philadelphia: John Benjamins. - (1997): Typology and the diachronic evolution of German morphosyntax. In: Fisiak, Jacek (ed.): Linguistic Reconstruction and Typology. Trends in Linguistics 96,313-335. Berlin/New York: Mouton de Gruyter. Ruoff, Arno (21990): Frequenzwörterbuch gesprochener Sprache. Idiomatica Bd. 8. Tübingen: Niemeyer. Stolz, Thomas (1990): Flexion und Adpositionen, flektierte Adpositionen, adpositionelle Flexion. ZPSK 43, 3, 334354.

Wann werden die deutschen Präpositionen flektieren?

289

Wandruszka, Ulrich (1992): Zur Suffixpräferenz. Prolegomena zu einer Theorie der morphologischen Abgeschlossenheit. Papiere zur Linguistik 46,3-28. Werner, Otmar (1969): Das deutsche Plural system. Strukurelle Diachronie. In: Moser, Hugo (ed.): Sprache. Gegenwart und Geschichte. Jahrbuch des Instituts für deutsche Sprache 1968, Bd. V, 92-125. Düsseldorf: Schwann. - (1994): Auch Paradigmen entstehen und vergehen. In: Köpcke, Klaus-Michael (ed.): Funktionale Untersuchungen zur deutschen Verbal- und Nominalmorphologie, 5-28. Tübingen: Niemeyer. Wurzel, Wolfgang Ullrich (1984): Flexionsmorphologie und Natürlichkeit. Studia grammatica XXI. Berlin: Akademie-Verlag. - (1988): Derivation, Flexion und Blockierung. ZPSK 41,179-188. - (1995): Alterserscheinungen in der Morphologie. In: Boretzky, Norbert et al. (eds.): Natürlichkeit und Sprachwandel. Beiträge zum internationalen Symposium über "Natürlichkeitstheorie und Sprachwandel" an der Universität Maribor vom 13.5.-15.5.1993,67-91. Bochum: Universitätsverlag Brockmeyer. - (1996): On the similarities and differences between inflectional and derivational moiphology. STUF 49,267-279. Zürrer, Peter (1982): Wörterbuch der Mundart von Gressoney. Mit einer Einfiihrung in die Sprachsituation und einem grammatischen Abriß. Frauenfeld: Huber. - (1997): Systemveränderung in Südwalser Sprachinseldialekten. In: Ruoff, Amo & Peter LöfFelad (eds.): Syntax und Stilistik der Alltagssprache. Beiträge der 12. Arbeitstagung zur alemannischen Dialektologie. Idiomatica Bd. 18,155-170. Tübingen: Niemeyer.