Agreement from a Diachronic Perspective 9783110399967, 9783110373349

Citation preview

Jürg Fleischer, Elisabeth Rieken, Paul Widmer (Eds.) Agreement from a Diachronic Perspective

Trends in Linguistics Studies and Monographs

Editor Volker Gast Editorial Board Walter Bisang Jan Terje Faarlund Hans Henrich Hock Natalia Levshina Heiko Narrog Matthias Schlesewsky Amir Zeldes Niina Ning Zhang Editor responsible for this volume Volker Gast

Volume 287

Agreement from a Diachronic Perspective

edited by Jürg Fleischer Elisabeth Rieken Paul Widmer

ISBN 978-3-11-037334-9 e-ISBN (PDF) 978-3-11-039996-7 e-ISBN (EPUB) 978-3-11-040009-0 ISSN 1861-4302 Library of Congress Cataloging-in-Publication Data A CIP catalog record for this book has been applied for at the Library of Congress. Bibliographic information published by the Deutsche Nationalbibliothek The Deutsche Nationalbibliothek lists this publication in the Deutsche Nationalbibliografie; detailed bibliographic data are available on the Internet at http://dnb.dnb.de. © 2015 Walter de Gruyter GmbH, Berlin/Boston Typesetting: PTP-Berlin, Protago-TEX-Production GmbH, Berlin Printing and binding: CPI books GmbH, Leck ♾ Printed on acid-free paper Printed in Germany www.degruyter.com

Editors’ preface The present volume originates from an international workshop on “Agreement from a diachronic perspective” co-organized by the editors at Marburg University, October 4–5, 2012. It took place in the context of our joint research project “Diachrone Entwicklung von Kongruenzsystemen in vier flektierenden indogermanischen Sprachen”, funded by the Deutsche Forschungsgemeinschaft (RI 1730/4, April 2011–March 2014). In the project, together with our colleague Erich Poppe, we conducted diachronic corpus studies in well-attested Indo-European languages (namely Hittite, Ancient Greek, German, and Welsh), and interpreted the findings from a typological perspective in order to establish generalizations of change in agreement systems. The workshop, which was organized as part of the project, proved to be extremely stimulating with its lively discussions between linguists of various different fields of expertise. The present volume collects a selection of thoroughly reworked papers presented at that occasion, and it is our hope that the contents of the present volume will enhance our understanding of the diachrony of agreement systems and provide a useful point of reference for future studies on this both fascinating and intricate field of research. The papers reflect a broad range of research specialties, and they demonstrate the innovative impact of studies in diachronic typology. We would like to take this opportunity to acknowledge our debt of gratitude to a number of people and institutions: first, we would like to thank the Deutsche Forschungsgemeinschaft for funding our research project, which included generous financial support for the workshop. Marburg University provided the institutional frame both for our research project and the workshop. Magnus Breder Birkenes and Florian Sommer did a great job in helping us organize the workshop and commenting upon papers. Thanks to their valuable editorial assistance, a speedy publication of the present volume was possible. Our sincere thanks are given to the participants of the workshop, listeners and speakers, and once again, we would like to express our heartfelt gratitude to the contributors who provided us not only with their important contributions but also submitted to the tight schedule of publication. Finally, we wish to thank the managing editor of the Trends in Linguistics. Studies and Monographs series Volker Gast for accepting the volume for publication, as well as the editorial staff at De Gruyter Mouton, Birgit Sievert in particular, for their professionalism and support. Marburg and Zurich, October 2014 Jürg Fleischer, Elisabeth Rieken, Paul Widmer

Contents Editors’ preface | v Jürg Fleischer, Elisabeth Rieken and Paul Widmer Introduction: the diachrony of agreement | 1

Part 1: Verbal and adpositional agreement Balthasar Bickel, Alena Witzlack-Makarevich, Taras Zakharko and Giorgio Iemmolo Exploring diachronic universals of agreement: alignment patterns and zero marking across person categories | 29 Manuel Widmer The transformation of verb agreement into epistemic marking: evidence from Tibeto-Burman | 53 Lameen Souag How to make a comitative preposition agree it-with its external argument: Songhay and the typology of conjunction and agreement | 75

Part 2: (Pro-)nominal agreement Michele Loporcaro The impact of morphology on change in agreement systems | 103 Lien De Vos Pronominal gender agreement: a salience-based competition | 127 Kyle Jerro and Stephen Wechsler Person-marked quantifiers in Kinyarwanda | 147 Aaron Griffith Degrees of agreement in Old Irish | 165

viii   

   Contents

Part 3: Mismatch constellations and resolution contexts Greville G. Corbett Hybrid nouns and their complexity | 191 Hans-Olav Enger When friends and teachers become hybrids (even more than they were) | 215 Damaris Nübling Between feminine and neuter, between semantic and pragmatic gender: hybrid names in German dialects and in Luxembourgish | 235 Guðrún Þórhallsdóttir (University of Iceland) Gender agreement in 19th- and 20th-century Icelandic | 267 Antje Dammel One plus one make(s) – what? | 287 Cyril Brosch Agreement patterns of coordinations in Hittite | 327 Index of languages | 353 Index of subjects | 354

Jürg Fleischer, Elisabeth Rieken and Paul Widmer

Introduction: the diachrony of agreement 1 Notions of agreement Synchronic research has led to many interesting generalizations concerning the architecture of agreement systems. Arguably, the most important contribution has been made by typologically oriented studies (e.g. Corbett 1979, 1991, 2003, 2012; Aikhenvald 2003), which enabled researchers to formulate empirically well-founded generalizations such as the Agreement Hierarchy (Corbett 1979, 1991: 225–241, 2006: 203–237, 2012: 93–99). But agreement phenomena have recently been discovered as interesting research topics within more formal frameworks as well. For instance, valuable insights have been provided by representatives of Lexical Functional Grammar (e.g. Bresnan and Mchombo 1987), Head Driven Phrase Structure Grammar (e.g. Wechsler and Zlatić 2000, 2003), and the generative paradigm (e.g. Baker 2008). For the sake of clarity, a terminological definition of the term agreement and its use in this introductory chapter is in order. In what follows we basically adopt the concepts and the terminology of Corbett (2006:  4–8). Thus, the very basic notion of agreement is co-variance (as per Steele 1978: 610). The terms for the elements that are involved in an agreement relation are well established: controller refers to the element that determines the values of a co-varying element, the element that co-varies is called the target, and the syntactic environment in which agreement is triggered is called the domain. Feature refers to the category showing co-variation (e.g. gender, number, or person), and the value of a feature is a particular member of that category (e.g. feminine, singular, etc.). Elements that have an impact on agreement, but are not directly involved in co-variation, are called conditions. There is no unanimity in the literature as to the categories that should or can be included in the set of agreement features. Here again, we follow the approach of Corbett (2006) and will deal with “canonical” agreement features only, viz., person, gender, and number; agreement in case is thus not dealt with.

2 Agreement and diachrony Although significant progress in research on agreement systems has been made from a synchronic point of view, a diachronic perspective is still missing in most accounts of agreement (cf. the brief remarks in Corbett 2006: 264–274). This is

2   

   Jürg Fleischer, Elisabeth Rieken and Paul Widmer

regrettable since the interest in language change in general has increased considerably in the last years. One reason for this diachronic renaissance is that language change can provide interesting insights when it comes to assessing the explanatory power of synchronically motivated theoretical concepts (cf. e.g. Bickel et al., this volume). The lack of informed statements on the diachrony of agreement is due to the scarcity of empirical studies in this field. This volume is intended to begin filling this gap by presenting empirically well-founded research dedicated to the understanding of the diachronic development of agreement systems. In the remainder of this introductory chapter, we will briefly discuss the methodological problems specific to research on the history of agreement systems (section 3). We will then illustrate some changes observed in the diachrony of agreement systems, drawing on our own research and that reported on in the present volume, and thus provide a phenomenology of agreement change (section 4). Finally, the chapter concludes with a short section on the scope of this book (section 5).

3 Methodological issues Diachronic linguistics requires somewhat different methods than those used in purely synchronic approaches. This applies not only to the problem of sampling data for empirical studies, but also to the reliability of the data. Basically, there are two ways of obtaining diachronic data. For some of the world’s languages, older stages are attested in the form of a textual record that can be analyzed in the form of corpus studies (especially prominent in this respect are of course the Indo-European and Semitic language families). However, this is not the case for the majority of the world’s languages. There diachronic data can be obtained via comparison of closely related living linguistic systems (“dialects” or closely related languages), as has already been recommended by Harris and Campbell (1995: 12). But also for languages that do have older textual records it makes sense to look at “living” data, since older texts pose certain problems (see below). Thus, beside corpus analyses “comparative” or “variationist” studies can add valuable evidence to diachronic research (they are, in fact, inherently diachronic). In general, scholars use both approaches to obtain diachronically relevant data (corpus-based and comparative), depending on the availability of written records and the varieties attested. As concerns historical records, they notoriously contain “bad data” in the sense of Labov (1994: 11). In the absence of native speakers, they cannot provide negative evidence, and the textual transmission often leads to a blurred picture

Introduction: the diachrony of agreement   

   3

because it is often not clear whether a certain form is original or introduced by a scribe copying or/and adapting the text. In addition, the corpus is all too often quite limited and contains gaps. As Labov (1994:  10) puts it: “Historical documents survive by chance, not by design.” In the remainder of this section, we will focus on two serious problems scholars often have to deal with when analyzing older written sources. The following examples illustrate how cautious we must be when interpreting agreement data from historical records.

3.1 Translation syntax It is common knowledge that the so-called committee nouns (see Corbett 2006:  211), that is, collective nouns, can trigger plural agreement in many languages (cf. the well-known British English example the committee decide). As discussed by Birkenes and Sommer (2015) in more detail, committee nouns triggering plural agreement are also attested in both the older stages of German and in Ancient Greek. Sometimes we find coordinated structures of a singular and a plural verb as committee nouns’ agreement targets. In these cases, the committee noun subject triggers singular agreement on the verb of the same clause, but plural agreement in the following coordinated clause. This is illustrated by example (1) from Luther’s Early New High German Bible translation: (1)

Early New High German (Luther’s Bible translation 1545; Ex 32,6) Darnach satzt sich das Volck zu essen Thereafter sat.sg refl art.sg people.sg to eat vnd zu trincken / vnd stunden auff spielen and to drink and stood.pl up play ‘and the people sat down to eat and to drink, and rose up to play’ (King James Version)

As a matter of fact, we find exactly the same type in Ancient Greek, as the following parallel example from the Septuagint shows: (2)

Ancient Greek (Septuagint; Ex 32,6) kaì ekáthisen ho laòs phageȋn kaì and sat_down.sg art.sg people.sg eat and pieȋn kaì anéstēsan paízein drink and stood_up.pl play

4   

   Jürg Fleischer, Elisabeth Rieken and Paul Widmer

Indeed, the same pattern is also found in the Vulgate and in the Hebrew original text (see Birkenes and Sommer 2015: 214–215): (3)

Latin (Vulgate; Ex 32,6) et sedit populus comedere ac bibere and sat.sg people.sg eat and drink et surrexerunt ludere and rose.pl play

(4)

Biblical Hebrew (Ex 32,6) wa-y-yēšɛḇ hå̄ -ʿå̄m lɛ-ʾɛ̆ḵōl and-sat.sg the-people.sg to-eat wa-y-yå̄qumūw lə-ṣaḥēq and-rose.pl to-play

wə-šå̄ ṯōw and-drink

When we examine German and Ancient Greek original texts, this construction hardly appears at all. As argued by Birkenes and Sommer (2015:  216), we thus conclude that this special agreement pattern was simply transmitted with the translation from Hebrew into Greek, from Greek into Latin, and, depending on the text underlying Luther’s translation, from Hebrew, Greek, and/or Latin into German. Since this pattern only appears in translated texts, one should speak of translated, not borrowed syntax. Its productivity status within the respective grammatical systems seems doubtful at best.

3.2 Prescriptivism It is well known that (Standard) Icelandic takes neuter plural forms in gender resolution contexts. For instance, couples and other mixed-gender groups are normally referred to by neuter plural forms. However, in generic contexts and when reference is made to lexical hybrids, prescriptive grammars have demanded use of masculine forms since the 19th century. Þórhallsdóttir (this volume) analyzed 1640 private letters written in the 19th century by writers with little formal education. In opposition to the standard language, neuter forms prevail in these contexts, as illustrated by the following example:

Introduction: the diachrony of agreement   

(5)

   5

Icelandic (19th century private letter; cf. Þórhallsdóttir, this volume: 279) Krakk-ar=nir mín-ir er-u kid-(m).nom.pl=def poss-nom.pl.m be[prs]-3pl efnileg og heldur lagleg promising[nom.pl.n] and rather pretty[nom.pl.n] ‘My kids are promising and quite pretty’

The masculine plural subject krakkar ‘kids’ triggers neuter agreement in the predicative adjectives efnileg(-Ø) and lagleg(-Ø). If the 19th and 20th century standard had been followed, the masculine plural forms efnileg-ir and lagleg-ir would have been used. Although the neuter plural agreement was eliminated from the standard by prescriptive activities in the 19th century, it was retained to some extent in spoken non-standard varieties. Scholars dealing with the history of agreement systems, and the history of language in general, need to be aware of prescriptive tendencies and of the selective perception of elites producing the prescriptive norms in question.

4 A phenomenology of change in agreement In this section, we provide a brief phenomenology of change in agreement systems by using mostly (though not exclusively) historical data collected to a large extent in the context of the Marburg research project and the contributions of the present volume. We will do so by taking into account the concepts of Corbett’s (2006) agreement model outlined above and look at the different elements of agreement with respect to linguistic change.

4.1 Controller Since agreement is defined as co-variance between a controller and its target(s), it is almost trivial to remark that changes in the controller might lead to changes in the agreement system as a whole. For instance, if a noun (being a potential controller) changes its gender, targets will change their agreement values accordingly. Such a change of noun class (which may or may not be overtly marked on the noun itself, depending on the morphological marking of nominal classes on the nouns themselves in general) means a change in the lexical entry of the noun. A particularly interesting case in point is the diachronic development of lexical hybrids, i.e. nouns whose formal features may conflict with their seman-

6   

   Jürg Fleischer, Elisabeth Rieken and Paul Widmer

tics, e.g., between (grammatical) gender and (natural) sex of their referents. In what follows we will illustrate that such lexical hybrids are subject to “dehybridization”, that is, they can become “regular” nouns, and that the opposite process of “hybridization”, that is, of “regular” nouns becoming lexical hybrids, is attested as well. In several old West Germanic languages (and many modern ones), the word for ‘woman; wife’ (Old High German wīb, Old Saxon, Old Frisian and Old English wīf) is neuter in terms of grammatical gender. Throughout the history of (written) German, attributes to this noun tend to take neuter forms in agreement relations, whereas anaphoric pronouns mostly, but not exclusively show feminine agreement; in the relative pronoun either gender is attested (cf. Fleischer 2012). This change is thus structured by the Agreement Hierarchy. Figure 1 illustrates the frequency of grammatical and semantic agreement for five different target types. The twelve corpora used for this investigation cover the time span from the early 9th to the 20th century.

Figure 1: Semantic and grammatical agreement to OHG wīb/NHG Weib ‘woman; wife’ with five different target types (adapted from: Fleischer 2012: 189)

As argued by Fleischer (2012), from a synchronic point of view, the fact that the individual lines do not cross one another means that the Agreement Hierarchy holds for each single corpus, albeit with different figures. This can be regarded as additional synchronic empirical evidence for the validity of the Agreement Hierar-

Introduction: the diachrony of agreement   

   7

chy. In a diachronic perspective, it can easily be seen that, for the time span covered here, there is no single direction of change: while semantic agreement has a broad peak between the corpora of Kaiserchronik (late 12th century) and Prosalancelot I (13th/15th century), with 100 % of possessive, personal, and relative pronouns, it becomes less frequent during the following period, but by no means does it diminish fully. On the other hand, semantic agreement is somewhat less frequent at the beginning of the attested history of German. Thus, we seem to be dealing with a “there and back” change (Corbett 2006: 273) rather than with a change pointing in one direction. However, it is not chaotic, but structured according to the different positions of the Agreement Hierarchy. Crucially, although substantial changes can be observed, the hybrid agreement behavior of Old High German wīb is retained continuously, even in its modern reflex, New High German Weib. However, this is not the case in different, closely related, linguistic systems. In some Yiddish varieties as well as in many North Frisian dialects, the noun in question behaves like an ordinary feminine noun. Given the absence of unambiguous formal marking of noun classes on the noun itself in these languages, this behavior can be deduced only from the fact that all agreement targets only display feminine gender forms. For instance, in the following Yiddish example (taken from the story In a fargrebter shtot ‘In a boorish town’, by the Soviet-Yiddish writer Dovid Bergelson [1884–1952], born in Oxrimovo, present-day Ukraine), in addition to the personal pronouns, even the article and the postpositioned possessive pronoun display feminine gender; thus, there is no longer any clue to the former hybridity of this noun, which, as a consequence, has arguably been “de-hybridized”: (6)

Yiddish (Dovid Bergelson, In a fargrebter shtot) frier shmeykhlt zi azoy tsu mir, earlier smiles she so to me di vayb zayn-e, un zogt: def:nom.sg.f wife his-nom.sg.f and says “vart, zogt zi, ikh vel forn” wait says she I will drive ‘earlier she smiled to me, his wife, and says: “Wait, she says, I will drive.”’

Note that in this Yiddish variety the neuter still exists:¹ not only do we find in the same texts instances of neuter articles with non-hybrid neuter nouns such as dos bukh ‘the:nom.sg.n book’, dos shtetl ‘the:nom.sg.n shtetl’ or dos holts ‘the:nom. 1 Thus, this variety is different from North Eastern Yiddish, where the entire neuter gender has disappeared (see e.g. Herzog 1965: 102). For North Eastern Yiddish, it is of course to be expected

8   

   Jürg Fleischer, Elisabeth Rieken and Paul Widmer

sg.n wood’, but the hybrid agreement behavior of meydl ‘girl’ (cf. the discussion of cognate German das Mädchen ‘the:nom.sg.n girl’ in Corbett 1991: 227–228) has also been retained. This is attested by the following example, taken from the same text, where we find the neuter article dos, but the feminine anaphoric personal pronoun zi, both targets of meydl: (7)

Yiddish (Dovid Bergelson, In a fargrebter shtot) dos meydl iz mit im nit def:nom.sg.n girl is with him not geven bakant, zi hot nor gevolt zen … been known she has only wanted see ‘the girl was not known to him, she only wanted to see …’

In some North Frisian dialects the same phenomenon can be observed. In Old Frisian wīf is attested with the neuter form of the definite article (but see below). In the modern North Frisian Mooring dialect, the reflex of Old Frisian wīf is listed as a “regular” feminine used with the feminine article jü (instead of neuter dåt);² compare the following Old Frisian and Mooring examples: (8)

Old Frisian (Skeltana Riucht XXXVI) ief thet wif kweth if def:nom.sg.n woman says ‘if the woman says’

(9)

Mooring (North Frisian dialect; Jörgensen 1978: 12) jü wüf def:f woman ‘the woman’

Interestingly, we find feminine attributive agreement forms already in Old Frisian, as already mentioned by Jacob Grimm ([1837] 1898, 4:  317) and Richt-

that earlier neuter lexical hybrids denoting female persons have become feminines (not masculines, the other remaining gender), as indicated by Herzog (1965: 106). 2 Note that in Mooring, different from some insular North Frisian dialects such as Fering, the neuter is retained as a grammatical category; interestingly, in addition to many neuters such as dåt ääse ‘food, eating’ there is a neuter lexeme that seems to be related to wüf, having a pejorative meaning, namely, dåt wüset (Jörgensen 1978: 13). If this lexeme, as one would expect, can co-occur with feminine anaphoric pronouns, we are dealing with a retained lexical hybrid, in this case.

Introduction: the diachrony of agreement   

   9

hofen (1840: 1147–1148; cf. also Rauch 2007, although she does not quote these early observations). This is illustrated by the following example: (10)

Old Frisian (Londriucht VI) bi there wiue at def:dat.sg.f woman:dat.sg ‘at the woman’s place’

Judging from the examples provided by Richthofen (1840:  1147–1148), it seems that feminine agreement is frequent in the oblique cases, particularly in the dative. Since strong feminine and neuter nouns (including wīf) both have a dative singular ending -e (cf. Boutkan 2001: 622), this syncretism could be the starting point for the spread of feminine attributive agreement forms. The Yiddish and North Frisian examples discussed above illustrate that previously hybrid nouns can become “de-hybridized”. The opposite development, “hybridization” of a noun, is attested as well: as argued by Enger (this volume), in present-day Norwegian there seems to be a tendency for lexemes such as lærer ‘teacher’, whose grammatical gender is (or used to be?) masculine, to be used now with feminine agreement targets. This is a new development, earlier there was a now obsolete feminine derivation lærerinne for denoting a female teacher. If this tendency, as of yet quite recent, solidifies, an originally masculine noun will become a lexical hybrid such as the standard textbook example of the Russian masculine noun vrač ‘(medical) doctor’, which can be used to denote female doctors as well (and, accordingly, triggers feminine agreement in some instances; see e.g. Corbett 1983: 30–39, 1991: 231–232, 2006: 210–211). The Norwegian case provides us with the opportunity to observe the process of change in detail. The Norwegian examples also testify to the development of new hybrids motivated by their lexical semantics starting from the pronominal end of the Agreement Hierarchy. Corbett (this volume) now presents examples where it is the mismatch of form that generates split hybrids. These split hybrids have (like other hybrids) morphological characteristics such as endings that would normally indicate a certain grammatical gender which is in conflict both with their lexical semantics and their assigned grammatical gender, but this conflict is avoided in part of their paradigm by the assignment of morphologically justified agreement. Corbett argues that in split hybrids there can be no question of matches or mismatches prompted by the lexeme’s lexical semantics, which is the same throughout the paradigm, leaving the split unaccounted for. We are led to conclude, then, that the split started from the attributive end of the Agreement Hierarchy by gender assignment according to the inflectional type of the noun in categories where the morphological mismatch was most obvious in order to

10   

   Jürg Fleischer, Elisabeth Rieken and Paul Widmer

obtain alliterative agreement, as seen in Corbett’s example from the Nordreisa dialect of North Norwegian: (11)

Norwegian (cf. Corbett, this volume: 208) mama-n diː-n mum-def.sg your-m ‘your mum’

Another structure in which the controller often plays an important role in prompting a change of agreement rules is agreement with conjoined noun phrases. For instance, number resolution triggering plural agreement in the targets is frequent, but the singular also occurs (see Corbett 2006:  168–170). The starting point for change is again the choice of a given language between competing sets of rules, in this case between formal partial agreement (agreement with the nearest conjunct), semantic resolution (agreement with the addition of the conjuncts), and morpho-syntactic resolution (agreement with the combined conjuncts viewed as a single entity, as defined by Dammel, this volume). Each one is illustrated by the following Hittite examples taken from Brosch (this volume): (12)

Hittite (KUB 56.1+ i 11; cf. Brosch, this volume: 336) 2 LÚ 1 MUNUS akkanz 2 man 1 woman die:ptcp.nom.sg.c ‘Two men (and) one woman are (“is”) dead.’

(13)

Hittite (IBoT 1.36 i 63; cf. Brosch, this volume: 337) BELUTIM=ya=kan UGULA LIMTI=ya GAL-yaz lords=and=lp commander 1000=and big:abl katta paiskanta down go:ipfv:prs.3pl.mp ‘Both the lords and the commander of thousand (troops) go down through the main (gate).’

(14)

Hittite (KBo 17.105+ iii 7–8; cf. Brosch, this volume: 342) ⌈ an⌉da=ma=kan āssuwa mīyawa hat⌈ta⌉nta in(to)=cnj=lp good:n/a.pl.n soft:n/a.pl.n wise:n/a.pl.n wid[du] come:imp.3sg.act ‘Good, pleasant, and wise things shall come in.’

Introduction: the diachrony of agreement   

   11

Hittite, as analyzed by Brosch (this volume), exhibits a complex system for the application of formal agreement and various resolution rules, such as dependency on gender, number, animacy, and lexical semantics. The stability of this system during all of the attested history of Hittite (350 years) is, however, remarkable. In contrast, German displays a change of system twice within 700 years. In her detailed investigation, Dammel (this volume) shows that the changes originate in variation depending on the conditions of animacy and word order and are caused by the external development from an oral to a written standard.

4.2 Target Agreement requires morphology. Hence, the development or disappearance of agreement depends on the development or disappearance of agreement morphology. For instance, it is a well-established fact that pronominal elements provide the most important source for the emergence of verbal agreement morphology. Pronouns may proceed along the established paths of grammaticalization from full pronouns, to clitics, to inflectional morphemes (Givón 1976). The results of this can be observed in both the creation of agreement systems and in the renewal of already existing agreement systems. As Corbett (2006:  266) notes, it is difficult to provide data illustrating the actual birth of an agreement system where none existed before. He provides the following example from Palu’e, an Austronesian language without any kind of agreement in its more recent history: (15)

Palu’e (Austronesian; Donohue in Corbett 2006: 266) a. ak=pana 1sg=go ‘I went’ b. aku pana 1sg go ‘I went’

The bound clitic ak= in example (15a) is obviously a reduced form of the personal pronoun aku in example (15b). It might develop into an agreement marker (more precisely: a verbal prefix). Currently, however, this is not yet the case: the bound clitic ak= cannot combine with the full pronoun aku, which means that it still retains its referentiality; cf. example (15c). Therefore, for the time being, it cannot be regarded as an agreement morpheme.

12   

(15)

   Jürg Fleischer, Elisabeth Rieken and Paul Widmer

Palu’e (Austronesian; Donohue in Corbett 2006: 266) c. *aku ak=pana 1sg 1sg=go ‘I went’

There are other cases in which an earlier clitic pronoun has proceeded further on its way to becoming an agreement marker. In the following example from a Highest Alemannic dialect spoken in Northern Italy, which has displayed verbal agreement throughout all of its known history, the full second person plural pronoun ir can combine with its originally enclitic form =er (the same phenomenon can also be observed with other persons in this dialect). In this case we are dealing with the (cyclical?) renewal of an earlier agreement system: (16)

Highest Alemannic dialect of Gressoney (Vallée d’Aoste, Italy; Zürrer 1999: 320) ir tie-d=er 2pl do-2pl=2pl ‘You do.’

As Zürrer (1999:  316–385) discusses in detail, the enclitic subject pronouns of these Alemannic dialects are on the verge of becoming verbal agreement markers (the same phenomenon is also documented for other German varieties, see Zürrer 1999: 365–366). Here, different from Palu’e, the original clitic, having become a verbal suffix, can combine with the full pronoun. Another case in point is the free, enclitic, and affixal pronominal marking of Old Irish. In this language, phonological and morphological reconstruction has enabled us to track down the relevant categories into pre-history. In his contribution, Griffith (this volume) investigates the correlation of the phonological and morphological clines with the whole syntactic cline for the shift of pronominal arguments to non-arguments (= agreement markers) and offers a diachronic interpretation of the synchronic snapshot of the Old Irish system. The rise of agreement markers from independent pronouns may end up in entirely unexpected targets, as illustrated by Souag (this volume), presenting previously unreported data from Songhay languages in West Africa. Pronouns that had formerly been part of an afterthought construction like X (subject NP) …, Prox and Y were reanalyzed as external agreement markers of a comitative: X … Agrx-with Y. The bipartite comitative is cross-linguistically highly unusual and its development seems to have been induced by contact with Berber, which exhibits the same rare construction.

Introduction: the diachrony of agreement   

   13

More than once, scholars have claimed that the structure of a paradigm is affected by universal principles such as Watkins’ Law, according to which personal agreement marking develops in such way that third person markers are more likely to exhibit zero endings than first and second person markers (Watkins 1962). Bickel et al. (this volume) also demonstrate the cross-linguistic validity of a tendency towards this distribution of zero markers and propose a plausible path of development. They assume that the distribution of agreement markers is not based on ‘natural’ or universal principles underlying the structure of paradigms, but rather on the recurring grammaticalization process by which individual agreement markers arise from pronominal controllers and the higher frequency of zero third person pronouns. Just as cliticization and grammaticalization of originally free to bound morphemes can correlate with agreement morphology, so may simplification and the reduction of morphology lead to the partial or total loss of agreement. For instance, in most Germanic varieties the verb regularly agrees with the subject in terms of person and number. As is well known, there are different types of syncretism in Germanic verbal morphology, in some cases even total. The gray shading in Table 2 indicates syncretisms in five old and modern Germanic languages. If a syncretism is total, agreement disappears altogether from the domain. Table 2: Verbal syncretisms in Germanic languages: present indicative of the verb ‘to lead, to guide’

pres

OHG

OE

ModE

Norwegian (Bokmål)

Afrikaans

1.sg

leitu

lǣde

lead

leder

lei

2.sg

leitis

lǣdest

lead

leder

lei

3.sg

leitit

lǣdeð

leads

leder

lei

1.pl

leitemēs

lǣdað

lead

leder

lei

2.pl

leitet

lǣdað

lead

leder

lei

3.pl

leitent

lǣdað

lead

leder

lei

As can be seen, there is no subject verb-agreement in Afrikaans or in Norwegian. This is the case for most Mainland Scandinavian varieties and holds true not only for the verb illustrated here, but for the entire system. (In Standard Bokmål not even the highly frequent auxiliaries have preserved varying inflected forms for person and number; in the dialects, however, some number distinctions can still be found.) Interestingly, the development of these syncretisms in Afrikaans and Norwegian is the convergent result of different paths of development.

14   

   Jürg Fleischer, Elisabeth Rieken and Paul Widmer

In the case of Afrikaans, there is no overt morphology (thus, one might indicate the absence of morphology by writing lei-Ø). In fact, no verbal morphological distinctions remain at all, with perhaps a single exception. Both the infinitive and the imperative also have the form lei, but the past participle is gelei. However, this is arguably not a proper verbal form. So we are dealing here with the total, or almost total, destruction of verbal morphology, which obviously has consequences for agreement. In the case of Norwegian, as indicated by Table 2, we find the overt suffix -er in all present tense personal forms, but we do not find this suffix in the infinitive led-e, in the imperative led(-Ø), in the preterite led-te, or in the past participle led-t. Thus, in Norwegian there is still some morphology left, but interestingly (and crucially for agreement) not in the domains of the person and number distinctions. The Norwegian suffix -er was originally a second person singular verbal ending that expanded into all present indicative forms, but not beyond. As a consequence, in this case a former verbal personal ending has acquired a new function, namely the indication of present indicative tense. In this case, original agreement morphology is preserved, but acquires new functions outside the domain of agreement, which is a path of development quite different from the one seen in Afrikaans (for a similar example from Old Irish cf. Schumacher 2004: 61–62, with references) As argued by M. Widmer (this volume) former verbal agreement morphology can also develop into epistemic marking, viz. “assertor’s involvement marking”. Evidence for this comes from two cognate Tibeto-Burman languages, Dolakha Newar and Bunan. In conclusion of this section we can state that, apart from disappearing completely, agreement morphology can also acquire various new functions outside of agreement proper.

4.3 Domains Change of agreement behavior is also observed with respect to the agreement domain. For instance, in early Old High German texts, the adjective agrees in both the attributive and predicative position. This is illustrated by examples (17a) and (17b), in which the same strong masculine plural form is used: (17)

Old High German (Tatian 244,17; Tatian 224,6) a. blint-e leitida blind-nom.pl.m leaders(m) ‘blind leaders’

Introduction: the diachrony of agreement   

   15

b. daz sie sin blint-e that pers.3pl.m be:3pl.pres.subj blind-m.pl ‘that they (m) be blind’ The loss of agreement in the predicative adjective began quite early in the history of German. Even in the earliest Old High German texts not all predicative adjectives are inflected, and in Late Old High German it is only a small minority (cf. Fleischer 2007). The result of this development can be seen in New High German. Here, although adjectives are still morphologically sensitive to case, number, and gender (gender distinctions were lost in the plural though, but remain intact in the singular; cf. section 4.4.1), the inflected forms only appear in the attributive domain, while there is no longer any overt agreement morphology in the predicative adjective. The following New High German examples, imitating their Old High German equivalents, illustrate this. (18)

New High German a. blind-e Führer blind-nom.pl leaders (m) ‘blind leaders’ b. New High German dass sie blind-Ø seien that pers.3pl blind be:3pl.pres.subj ‘that they be blind’

In summary, agreement of adjectival targets was lost in the predicative domain, but not in the attributive domain. Note that this change cannot have happened as a consequence of morphological syncretism, since the endings themselves have not been lost. A different effect concerning the domains can be seen in Neo-Hittite. In this language, canonical agreement operates throughout the domain of the clause (with the famous exception of the skhêma attikón, see section 4.5 below). Both controllers and targets are marked overtly for number. Only the class of neuter nouns in -r- and -l- makes no distinction between the nominative/accusative singular and plural, neither of which displays an overt morpheme. However, there is an incipient change to be observed in which they acquire a new overt ending -i in the plural (for an analysis and an interpretation of the data see Rieken 2012[2014]). The interesting fact about the use of these endings is that they are used in contexts where there is already another overt marking of the plural on at least one target within the same clause. In these cases the lack of overt marking

16   

   Jürg Fleischer, Elisabeth Rieken and Paul Widmer

and the formal asymmetry of the system were particularly obvious and were remedied within the domain of the clause by the use of the new ending. In the following example the controller arkuwar with zero-ending has been replaced with the new form arkuwarr-i, which has -i as an overt plural marker: (19)

Hittite (KUB 6.45 I 26–27, 13th century B.C.) nu=mu kē arkuwarr-iḪI.A cnj=pers.1sg dem.nom/acc.pl.n prayer(n)-nom/acc.pl ištamašten listen:2pl.imp ‘And listen to these prayers of mine!’

On the other hand, in an older stage of the same language, agreement across the boundaries of the clause domain is reduced by the elimination of formal number agreement between the quantifier ḫumant- ‘all, every’ and its plural controller in a preceding clause. Instead, the singular is used as a default (see Rieken and Widmer 2014). Both of these changes in the Hittite agreement system, which was otherwise remarkably stable, make the clause domain more important for agreement.

4.4 Values and features: increase and decrease, addition and deletion Change in agreement systems can be due to the increase and decrease of values in an established feature system and to the addition or deletion of certain features altogether. We will first discuss cases of value reduction and feature deletion and then proceed to the opposite processes.

4.4.1 Value reduction and feature deletion Agreement systems may change because the values of the features participating in an agreement relation change. An interesting example is provided by gender distinctions in the plural and by gender resolution in Germanic. Most of the older Germanic languages displayed a three-way gender distinction in the pronominal and adjectival inflections, not only in the singular, but also in the plural (this is still the case in modern Icelandic and Faroese). Compare the relevant forms of early 9th century Old High German in Table 3:

Introduction: the diachrony of agreement   

   17

Table 3: Plural gender distinctions in Old High German (Braune and Reiffenstein 2004: 243, 247, 220) masc.

neut.

fem.

personal pronoun

sie

siu

sio

demonstrative/relative pronoun

die

diu

dio

strong adjective

blinte

blint(i)u

blinto

The plural gender distinctions of this system were gradually given up during the history of German (see Fleischer 2007:  291–293). First, masculine and feminine forms merged (more precisely, the masculine form spread to feminine contexts; examples for the personal pronoun can be found as early as in the second half of the 9th century, the demonstrative pronoun followed somewhat later), yielding a two-way distinction of masculine/feminine vs. neuter. Eventually, this distinction was given up as well, leading to a system without gender distinction in the plural, see Table 4 for the personal pronoun: Table 4: Development of gender distinctions in the German plural pronominal system (3rd plural personal pronoun)

stage I: three-way distinction stage II: two-way distinction stage III: no distinction

masc.

fem.

neuter

sie

sio

siu

sie

siu sie

The three morphological systems illustrated above provide interesting constructions in gender resolution contexts. In stage I (the oldest stage of Germanic), the co-ordination of a masculine and a feminine noun is usually resolved by the neuter plural (Behaghel 1928: 39). In stage II, after the merger of the masculine and feminine plural, there are examples where the masculine/feminine form appears in what used to be a gender resolution context; see the following example from late Old High German (10th/11th century), in which a personal pronoun (and, agreeing with it, a predicative adjective) appears in the masculine/feminine form:

18   

(20)

   Jürg Fleischer, Elisabeth Rieken and Paul Widmer

Late Old High German (Notker, Psalter 77, 13–14; Fleischer 2007: 298) prénne mine glúste unde mîne gedáncha burn my lusts(f) and my thoughts(m) daz sîe únreht-e ne-sîn that they(m/f) unright-m/f neg-be:3pl:subj ‘burn my lusts and my thoughts that they be not unrightful’

With the value reduction in the plural gender system, the gender resolution context no longer requires a special gender resolution form because the masculine/feminine form can be used. As a consequence, the problem for the earlier morphological system posed by the gender resolution context has disappeared.³ In stage III, finally, with the total disappearance of the gender distinction in the plural, the gender resolution contexts disappear completely.

4.4.2 Value addition, feature increase Opposite developments such as the addition of values of already existing features or the introduction of new features are also attested. One example of value 3 Interestingly though, there can still be found neuter plural gender resolution forms referring to a masculine+feminine conjunct in stage II, the morphological merger of masculine and feminine plural notwithstanding. In the following example, the nouns spîse ‘food’ (feminine) and slâf ‘sleep’ (masculine) are referred to by the neuter plural form of the demonstrative pronoun. (i) Middle High German (Iwein 4818–4819; quoted from Paul 2007: 384) guot spîse und dar nâch sanfer slâf good food(f) and thereafter gentle sleep(m) diu wâren im bereit dem:nom.pl.n were him ready ‘good food and gentle sleep thereafter, they were ready for him’ This is an interesting example to illustrate that gender resolution forms can persist even if there is no formal need for a special resolution form (it would be worth discussing whether we are dealing with “junk” in the sense of Lass 1990 here). Actually, there are even examples in which the neuter forms are used in cases of agreement with the co-ordination of the nouns of the same (non-neuter) gender, as can be seen in the following Old High German example where two masculine plural nouns control a neuter plural form in the predicative adjective: (ii) Old High German (Tatian 205, 17; Fleischer 2007: 298) mîne ferri Inti paston sint arslagan-u my oxen(m) and fatlings(m) are slain-nom.pl.n ‘my oxen and my fatlings are killed’ (Matthew 22,4) As these intriguing examples show, further research is needed here.

Introduction: the diachrony of agreement   

   19

addition is found in adjectival agreement in Ancient Greek. Here, many adjectives display a three-way morphological gender contrast (e.g. masculine ákr-os, feminine ákr-a, neuter ákr-on ‘highest, topmost’). However, a subset of (mostly compound) adjectives lacks a formal contrast for masculine and feminine (e.g. masculine/feminine bárbar-os, neuter bárbar-on ‘strange, foreign’). In Ancient Greek, the number of distinct gender values in this target class was already sporadically increased by introducing a morphologically distinct form -a/-ē for the feminine gender by analogy with the common pattern. Over time, the two-way gender distinction of Ancient Greek in this particular subclass of adjectives was replaced with a three-way contrast by means of a systematic addition of a distinct expression for the feminine gender (Horrocks 2010: 289–290). Thus, a distinction between masculine and feminine was newly introduced, providing an example of the addition of a new distinct value to an already existing feature. Furthermore, there are examples of the introduction of new features into particular target classes where they did not exist before. In Indo-European languages the canonical feature “gender” is typically realized in nominal targets such as pronominals and adjectives, but may also spread to other target types. One such development in recent times is reported for dialectal Italian by Loporcaro (this volume): here, gender agreement has spread to the verb. This is illustrated by the following example, in which gender agreement is not only found on the participle, but also on the auxiliary: (21)

Italian, dialect of Ripatransone, province of Ascoli Piceno, Marche (Loporcaro, this volume: 106) noja s-em-i dat-i 1pl be.prs-1pl-m.pl give:ptp-m.pl ‘we(m) have given’

In example (21) the finite auxiliary verb s-em-i agrees with the subject not only in number and person (-em- first person plural), but also in gender (-i masculine plural). This kind of gender agreement is confined to periphrastic perfective constructions, where the past participle (dat-i) could function as a starting point for the development. In this construction, then, gender agreement has spread to verbal targets, which are not specified for this feature elsewhere. Another example of the introduction of a new feature onto a particular target is discussed by Jerro and Wechsler (this volume): in Kinyarwanda, a Bantu language, the feature ‘person’ was introduced into quantifiers, where it had not existed before. The following example illustrates the person agreement for the quantifier ese ‘all’ with the second person plural subject pronoun mwe.

20   

(22)

   Jürg Fleischer, Elisabeth Rieken and Paul Widmer

Kinyarwanda (Bantu; Jerro and Wechsler, this volume: 148) (Mwe) mw-ese mw-agi-ye ku i-duka 2pl 2pl-all 2pl-pst.go-perf to cl5-store ‘All of you went to the store.’

4.5 Conditions Finally, the conditions of an agreement relation may change. For instance, in Ancient Greek, a language known for its highly canonical agreement system, neuter plural subjects normally fail to trigger plural agreement on the verb, as illustrated by example (23). This construction is known as skhêma attikón in traditional grammar (see, for example, Cooper 1998:  1015–1016). But it is not difficult to find clauses where, contrary to the skhêma attikón, the verb agrees with its neuter plural subject, as shown by example (24) taken from the same author: (23)

Attic Greek (Aristotle Anal. post. 98b 36) tà déndra phyllorreî def:nom.pl.n tree-nom.pl.n shed_leaves:3sg ‘the trees shed their leaves’

(24)

Attic Greek (Aristotle Rh. 1408b 25) prolambánousi tà paidía catch_up:3pl def:nom.pl.n child-nom.pl.n ‘the children catch up’

The variation observed here between the plural and singular agreement of a verb with its neuter plural subject can, as a matter of fact, be shown to be influenced by the degree of animacy of the subject referent: the higher its degree of animacy, the more likely the verbal target will have a plural form. This can be seen in Table 5, which illustrates the frequency of singular and plural agreement in three semantic groups with differing degrees of animacy. Plural agreement is shown to be most frequent for the two nouns highest on the animacy scale, i.e. tékna/paidía ‘children’. The opposite is the case for the inanimates, i.e. phýlla ‘leaves’ and déndra ‘trees’. Therefore, the Animacy Hierarchy (see, e.g., Silverstein 1976, Bossong 1998) can be considered a crucial factor for the frequency of plural agreement:

Introduction: the diachrony of agreement   

   21

Table 5: Agreement of verbs with selected neuter plural subjects in Attic and Koiné Greek (500 BCE–300 CE) tékna ‘children’ paidía ‘children’

zō̃ia ‘living beings’ thēría ‘animals’

phýlla ‘leaves’ déndra ‘trees’

Sg. agreement

204 (77.9 %)

435 (86.5 %)

417 (97.4 %)

Pl. agreement

58 (22.1 %)

68 (13.5 %)

11 (2.6 %)

The way in which a system changes may be closely related to the conditions under which variation arose in the first place. As has been discussed above, in Attic and Koiné Greek the verbal agreement of neuter plural subjects partly depends on referential properties of the controller: neuter subjects with human referents are much more prone to trigger plural agreement on the verb than subjects with nonhuman referents. In Table 6, the frequency of plural agreement is indicated for three different periods. Table 6: Plural agreement of verbs with selected neuter plural subjects in Attic and Koiné Greek in three periods tékna ‘children’ paidía ‘children’

zō̃ia ‘liv. beings’ thēría ‘animals’

phýlla ‘leaves’ déndra ‘trees’

500BCE–250BCE

30.8 %

17.7 %

2.2 %

250BCE–50CE

29.6 %

11.6 %

0.4 %

50CE–300CE

13.3 %

 9.8 %

6.4 %

Plural agreement triggered by animate nouns was most frequent in the earliest period, but the impact of animacy decreased considerably over time, so that the difference between the three types of controllers nearly levels out in the third period. More generally speaking, the trend toward singular agreement of the skhêma attikón-type can be interpreted as a loss of the impact of the controller’s semantic properties on agreement behavior, while the formal rules of the grammaticalized skhêma attikón win out. The role of the conditions is thus reduced.⁴ Conversely, when formal agreement loses its dominant role, conditions seem to gain more importance for the realization of agreement rules. De Vos (this 4 Note that in Modern Greek plural verb agreement for neuter plural nouns is the only option (Holton, Mackridge and Philippaki-Warburton 2012:  608–609), while the variation caused by differences in animacy has been given up entirely.

22   

   Jürg Fleischer, Elisabeth Rieken and Paul Widmer

volume) reports that formal agreement of gender-marked personal pronouns is being lost rapidly in Southern Dutch dialects with neuter pronouns developing as the default. However, during this on-going process, the discourse salience of the controller exerts influence on the use of formal agreement on the pronominal target. Two factors contributing to discourse salience have been shown to be relevant, viz. syntactic prominence in terms of subjecthood and semantic prominence in terms of verbal actionality. Thus, a controller in subject position triggers formal agreement on a target pronoun by a higher percentage than a controller in object position. Similarly, for controllers being subject of clauses with action verbs the likelihood of formal agreement on target pronouns is significantly higher than, for instance, for controllers in subject position of clauses with copular verbs. As a tentative conclusion we can state that the examples discussed here indicate that there is a correlation between the dynamics of the change of a given system and the importance of agreement conditions.

5 The scope of this book The present volume contains three thematically defined parts – verbal and adpositional agreement, (pro-)nominal agreement, and mismatch constellations and resolution contexts. In the section on verbal and adpositional agreement, subject-verb agreement figures prominently, but a different, typologically rare, type of agreement is also taken into account (see the contribution by Souag). In the contributions dealing with (pro-)nominal agreement, different nominal targets such as adjectives and pronouns are analyzed. The interaction with morphological changes motivated outside the architecture of the agreement system figures prominently here. The parts on verbal and (pro-)nominal agreement cover what might be labeled the core areas of agreement. In the third part papers treating mismatch constellations and resolution contexts are united; it is an especially interesting question whether such contexts, intrinsically unstable as they are, can be made responsible for the initial stages of change in agreement systems as a whole. The contributors to this volume explore a wide range of research topics. They focus on problems of methodology (Bickel et al.), questions of register and prescriptive activities (Þórhallsdóttir), the functional shift of agreement markers (M. Widmer, Nübling), the spread of agreement morphology to new targets (Loporcaro, Jerro and Wechsler, Souag), the disappearance of agreement in sub-systems (De Vos), the observation of an incipient change (Enger), the grammaticalization cline (Griffith), the agreement hierarchy (Brosch), the distinction of and switch

Introduction: the diachrony of agreement   

   23

between various types of agreement such as formal/morpho-syntactic, semantic, referential, and pragmatic agreement (Dammel, Nübling), and the emergence of split agreement (Corbett). The various problems are tackled either from the controller point of view (resolution contexts: Dammel, Brosch, Þórhallsdóttir; lexical hybrids: Corbett, Enger; person names: Nübling; pronouns: Griffith) or from the target side (verbal agreement: Bickel, M. Widmer, Loporcaro; pronominal agreement: De Vos; quantifier agreement: Jerro and Wechsler; comitative agreement: Souag). The kind of data sampled for the papers of the present volume varies: historical data collected by analysis of textual records serves as the empirical basis in the papers by Brosch, Dammel, Griffith, and Þórhallsdóttir. Data drawn from field work and/or questionnaire surveys is analyzed in the papers of Enger, Nübling, Souag and M. Widmer. Reference grammars, text collections, and/or samples are used by Corbett, Bickel et al., De Vos, Loporcaro, Souag and Jerro and Wechsler. The languages treated range from Germanic (Early New High German and New High German: Dammel; High German dialects: Nübling; Dutch dialects: De Vos; Norwegian [present-day spoken Bokmål]: Enger; 19th century and present-day Icelandic: Þórhallsdóttir), Romance (Italian dialects: Loporcaro), and Celtic (Old Irish: Griffith), to less well researched language families such as Berber (Souag), Bantu (Jerro and Wechsler), and Tibeto-Burman (M. Widmer), while other papers derive their generalizations from large, areally and/or genetically balanced samples (Bickel et al., Corbett).

Acknowledgements We are grateful to Magnus Breder Birkenes and Florian Sommer for valuable comments on a previous version of this introduction. We also would like to thank Sara K. Hayden for smoothing out our English. Needless to state, all remaining errors are ours.

References Aikhenvald, Alexandra. 2003. Classifiers. A typology of noun categorization devices. Oxford: Oxford University Press. Baker, Mark C. 2008. The syntax of agreement and concord. Cambridge: Cambridge University Press.

24   

   Jürg Fleischer, Elisabeth Rieken and Paul Widmer

Behaghel, Otto. 1928. Deutsche Syntax: eine geschichtliche Darstellung. Band III: die Satzgebilde. Heidelberg: Winter. Birkenes, Magnus Breder & Florian Sommer. 2015. The agreement of collective nouns in the history of Ancient Greek and German. In: Chiara Gianollo, Agnes Jäger & Doris Penka (eds.), Language change at the syntax-semantics interface, 183–221. Berlin & Boston: De Gruyter Mouton. Bossong, Georg. 1998. Le marquage différentiel de l’objet dans les langues de l’Europe. In: Jack Feuillet (ed.), Actance et valence dans les langues de l’Europe, 193–258. Berlin & New York: Mouton de Gruyter. Boutkan, Dirk F. H. 2001. Morphology of Old Frisian. In: Horst Haider Munske (ed.), Handbuch des Friesischen/Handbook of Frisian Studies, 620–626. Tübingen: Niemeyer. Braune, Wilhelm & Ingo Reiffenstein. 2004. Althochdeutsche Grammatik I: Laut- und Formenlehre. 15. Auflage bearbeitet von Ingo Reiffenstein. Tübingen: Niemeyer. Bresnan, Joan & Sam Mchombo. 1987. Topic, pronoun, and agreement in Chichewa. Language 79. 741–782. Cooper, Guy L. 1998. Attic Greek prose syntax. Volume 2. Ann Arbor: University of Michigan Press. Corbett, Greville G. 1979. The Agreement Hierarchy. Journal of Linguistics 15. 203–224. Corbett, Greville G. 1983. Hierarchies, targets and controllers: agreement patterns in Slavic. London: Croom Helm. Corbett, Greville G. 1991. Gender. Cambridge: Cambridge University Press. Corbett, Greville G. 2006. Agreement. Cambridge: Cambridge University Press. Corbett, Greville G. 2012. Features. Cambridge: Cambridge University Press. Fleischer, Jürg. 2007. Das prädikative Adjektiv und Partizip im Althochdeutschen und Altniederdeutschen. Sprachwissenschaft 32. 279–348. Fleischer, Jürg. 2012. Grammatische und semantische Kongruenz in der Geschichte des Deutschen: eine diachrone Studie zu den Kongruenzformen von ahd. wīb, nhd. Weib. Beiträge zur Geschichte der deutschen Sprache und Literatur 134. 163–203. Givón, Talmy. 1976. Topic, pronoun and grammatical agreement. In: Charles N. Li (ed.), Subject and topic, 149–188. Amsterdam: Benjamins. Grimm, Jacob. [1837] 1898: Deutsche Grammatik 4. Herausgegeben von Gustav Röthe. Gütersloh [Reprint 1967 Hildesheim: Olms.] Harris, Alice C. & Lyle Campbell. 1995. Historical syntax in cross-linguistic perspective. Cambridge: Cambridge University Press. Herzog, Marvin I. 1965. The Yiddish language in Northern Poland. Bloomington: Indiana University & The Hague: Mouton. Holton, David, Peter Mackridge & Irene Philippaki-Warburton. 2012. Greek: a comprehensive grammar of the Modern Language. 2nd ed. London: Routledge. Horrocks, Geoffey. 2010. Greek: a history of the language and its speakers. Chichester: WileyBlackwell. Jörgensen, V. Tams. 1978. Kort spräkeliir foon dåt mooringer frasch [Short grammar of Mooring Frisian]. 4. aplååge. Bräist: Noordfriisk Instituut. Labov, William. 1994. Principles of linguistic change. Volume 1: internal Factors. Oxford: Blackwell. Lass, Roger. 1990. How to do things with junk: exaptation in language evolution. Journal of Linguistics 26. 79–102.

Introduction: the diachrony of agreement   

   25

Paul, Hermann. 2007. Mittelhochdeutsche Grammatik. 25. Auflage, neu bearbeitet von Thomas Klein, Hans-Joachim Solms und Klaus-Peter Wegera, mit einer Syntax von Ingeborg Schröbler, neubearbeitet und erweitert von Heinz-Peter Prell. Tübingen: Niemeyer. Rauch, Irmengard. 2007. Gender semiotics, Anglo-Frisian wīf, and Old Frisian noun gender. In: Rolf H. Bremmer Jr., Stephen Laker & Oebele Vries (eds.), Advances in Old Frisian philology, 357–366. Amsterdam: Rodopi. Richthofen, Karl. 1840. Altfriesisches Wörterbuch. Göttingen: Diederichs. Rieken, Elisabeth. 2012[2014]. Der Nom.-Akk. Pl. n. auf -i im Hethitischen: zu den Mechanismen der Ausbreitung von Kongruenzsystemen. Historische Sprachforschung 125, 285–294. Rieken, Elisabeth & Paul Widmer. 2014. Kongruiert alles? – Zu den Kongruenzmustern des Pronominaladjektivs in der Bedeutung ‘all, jeder, ganz’ im Griechischen und Hethitischen. In: Norbert Oettinger & Thomas Steer (eds.), Das Nomen im Indogermanischen. Morphologie, Substantiv vs. Adjektiv, Kollektivum, 325–359. Wiesbaden: Reichert. Silverstein, Michael. 1976. Hierarchy of features and ergativity. In: R. M. W. Dixon (ed.), Grammatical categories in Australian languages, 112–171. Canberra: Australian Institute of Aboriginal Studies. Steele, Susan. 1978. Word order variation: a typological study. In: Joseph H. Greenberg, Charles A. Ferguson & Edith Moravcsik (eds.), Universals of human language IV: syntax, 585–623. Stanford: Stanford University Press. Schumacher, Stefan. 2004. Die keltischen Primärverben. Innsbruck: Institut für Sprachen u. Literaturen. Watkins, Calvert. 1962. Indo-European origins of the Celtic verb. Volume I: the sigmatic aorist. Dublin: Institute for Advanced Studies. Wechsler, Stephen & Larisa Zlatić. 2000. A theory of agreement and its application to SerboCroatian. Language 76: 799–832. Wechsler, Stephen & Larisa Zlatić. 2003. The many faces of agreement. Stanford, Calif.: CSLI Publications. Zürrer, Peter. 1999. Sprachinseldialekte. Aarau: Sauerländer.

Part 1: Verbal and adpositional agreement

Balthasar Bickel, Alena Witzlack-Makarevich, Taras Zakharko, and Giorgio Iemmolo

Exploring diachronic universals of agreement: alignment patterns and zero marking across person categories Abstract: Two principles shaping agreement paradigms have been implicitly assumed to constitute diachronic universals: (i) ergativity is assumed to be more likely to develop or be maintained in third than in non-third person; (ii) zeros are assumed to develop and be preserved more commonly in third than in non-third person. Estimating probabilities of diachronic change in a worldwide database and controling for areal diffusion effects, we find no evidence for (i). Principle (ii) receives no support either when examining how paradigms develop as systems, but we observe a weak cross-paradigm effect which is likely to be caused by frequency patterns during grammaticalization.

1 Introduction As has been repeatedly noted in the typological literature, most – perhaps even all – statistical universals are not really synchronic in nature, but are rather the result of underlying diachronic mechanisms that cause languages to change in preferred or ‘natural’ ways (e.g. Greenberg 1978, Bybee 1988, Hall 1988, Croft 2003, among others). Diachronic universals are not only of interest to typology and inquiries into the cognitive factors that shape human language, but they are also essential for historical linguistics and the reconstruction of the prehistory of individual languages or language families: when choosing between possible reconstructions, there is good reason to prefer ‘natural’ over ‘unnatural’ sound laws (Blevins 2004). A good illustration of this principle comes from work on Proto-Polynesian (Donohue & Oppenheimer 2012). The cognate word for ‘nose’ in Central Pacific languages shows an intervocalic /h/ in the majority of languages, as in Tongan ihu, while only one language of the group, namely Emae, shows intervocalic /s/, as in isu. If reconstruction were to follow a linguistically uninformed ‘majority-wins’ or maximum parsimony guideline, one would be tempted to reconstruct */ihu/. However, as noted by Donohue & Oppenheimer (2012:542), given what we know about the typology of sound changes, one would ceteris paribus reconstruct */isu/ and hypothesize a natural rule of lenition rather

30   

   Balthasar Bickel, Alena Witzlack-Makarevich, Taras Zakharko and Giorgio Iemmolo

than the other way around (also cf. Campbell 1999:116). ‘Natural’ here means in line with theoretically-motivated and universal expectations about diachrony. Establishing diachronic universals is no trivial enterprise. Using only reconstruction is not a viable option, since reconstruction often relies on assumptions about what is natural and universal in the first place. Using historically attested changes gets us further, but quickly runs into a severe sampling problem because there are only very few families in the world of which we know the history with any confidence. One way out that is currently being explored (e.g. Maslova 2000, Dunn et al. 2011, Bickel 2015) is based on statistical explorations of synchronic data with the goal of estimating the most likely diachronic trends and biases in the histories behind these data. In this paper, we will use one such method in order to evaluate the validity of two widely-known principles that have been claimed to affect the structure of agreement paradigms. These are principles that govern the distribution of patterns of role alignment (accusative, ergative etc.) and the distribution of zero marking in paradigms. In each case, we interpret these principles as candidates for diachronic universals. The first principle is what we call “Silverstein’s Law”. It predicts that paradigms naturally and universally develop in such a way that ergativity is more likely in the third than in the first and second person. The second principle we test is “Watkins’ Law”, which predicts that naturally and universally paradigms develop in such a way as to have zero forms in the third rather than in the first and second person. Before proceeding, however, an immediate proviso is in order: although the original formulations of these principles might not have been assumed to have true universal validity (as in the case of Watkins’ Law), or true diachronic validity (as in the case of Silverstein’s Law) by their proponents, several scholars have taken their status as diachronic universals for granted (see for example Koch 1995, Siewierska 2009 on the distribution of zeros, or Kiparsky 2008 on the distribution of ergative alignment). Independently of these interpretations in the literature, we do find it interesting to assess the validity of such principles as diachronic universals: if they turn out to be valid, they could have far-reaching consequences for our understanding of how languages change over time. The paper is organized as follows. Section 2 introduces the general method used to test the validity of the two principles and Section 3 describes our data source. Section 4 deals with the first case study, namely Silverstein’s Law, while Section 5 discusses Watkins’ Law. Section 6 summarizes the study and draws general conclusions.

Alignment and zero-marking across persons   

   31

2 Methods Universal trends in diachrony can be estimated on the basis of synchronic distributions. One method for such estimates is the Family Bias Method (Bickel 2015). The basic idea is the following: each family is evaluated as to whether its daughter languages are biased towards a certain structure (such as paradigms with ergative alignment in the third rather than the first or second person), as revealed by a statistical test. If there is such a bias, this means that daughter languages have preferentially innovated in the same direction, or they kept what was already there in the proto-language. Either way, a bias suggests that – for whatever reason – there was a systematic preference in the development of the given family. The absence of a bias suggests random fluctuation in development. The presence or absence of biases can be determined straightforwardly in families with enough representatives, using for example a binomial (or multinomial) test. But what about smaller families, or families with just one member, i.e. isolates? These constitute critical data because about half of the world’s language families are small, mostly in fact having one member. A solution comes from extrapolation algorithms: for this, we can use the information on biases in large families in order to estimate the biases that are likely to have been behind small families as well: if, say, 60 % of large families are biased towards some specific structure (e.g., biased towards Silverstein-style alignment patterns in paradigms) rather than balanced between structures (i.e. with about as many daughters with such patterns as daughters that contradict the pattern), we estimate a rough .6 probability that the known members of small families come from larger unknown families with a bias as well, in whatever direction, as opposed to families without any bias. In some small families, the known or only members will be representative of the bias in the unknown larger family from which they ultimately derive, and so we can take their type to reflect the bias. In other cases, the known or only members will happen to be deviates. The probability of being representative can be estimated from the strength of the bias in attested large families: for example, if among biased large families, biases tend to be very strong (e.g. on average covering over 90 % of members), we can estimate a high probability that the known members of small biased families are representative of the larger unknown family from which they derive, and only a small proportion is expected to be non-representative. Using the probabilities of bias and of representativeness based on large families, we can estimate the proportion of small families that come from larger biased as opposed to unbiased families, and if they are estimated to come from biased families, we can estimate whether the known members represent indeed

32   

   Balthasar Bickel, Alena Witzlack-Makarevich, Taras Zakharko and Giorgio Iemmolo

the biases of their families or deviate from them. These extrapolation estimates introduce error, but through multiple extrapolation we can arrive at a mean value that is fairly reliable. The method as summarized here is described and justified in detail in Bickel (2013), and it is implemented in an R (R Development Core Team 2012) package (Zakharko & Bickel 2011ff.). Principles are universal only if they hold independently of where families are located on the globe. This is particularly important with regard to abstract typological features, such as the distribution of zeros or of ergativity in agreement paradigms, because abstract patterns like these are known to have spread in large areas, often in the course of thousands – perhaps even tens of thousands – of years (Dryer 1989, Nichols 1992). Large areas constitute a confounding factor that needs to be controlled for. We will do this here by estimating family biases separately within large areas and conclude universality only if family biases show a significant trend that holds world-wide and that is statistically independent of areas.

3 Data The data for the two case studies come from a survey of individual agreement paradigms in 314 languages. By ‘agreement’, we understand here only grammatical agreement in the sense of Bickel & Nichols (2007), i.e. verbal markers of argument properties that can in principle co-occur with a coreferential noun phrase in the same clause (regardless of whether this co-occurrence is frequent or rare in discourse). In contrast to such markers, cliticized or incorporated pronouns that cannot co-occur with co-referential noun phrases in the same clause were not analyzed as instances of agreement. By ‘paradigm’, we understand here a set of agreement forms that shares a unique value in non-agreement categories, such as polarity, tense, aspect, direct vs. inverse, main vs. dependent clause status etc. When agreement patterns are conditioned by tense, aspect or polarity, a language is represented by two or more paradigms. For example, in Chortí (ISO639.3:caa; Mayan; Quizar 1994) completive and incompletive aspects have non-identical distribution of zeros and alignment patterns. For practical reasons, however, we limited our sample to the paradigms found in the main clause. Thus, for instance, in the case of Algonquian languages we considered only the independent order paradigms and excluded the conjunct order paradigms, etc. As a result of these considerations, our sample comprises more paradigms than languages, in total 352 paradigms. When paradigms are split into several conjugation classes, we generalize over them as long as they show the same distribution of alignments and zero mark-

Alignment and zero-marking across persons   

   33

ers. When the distribution is not uniform, we choose the pattern that is most frequent, either because it is found in most conjugation classes or because the classes are the largest in the lexicon (depending on what information is available in the source). For instance, Latvian has three conjugation classes with several subclasses. Class II (also referred to as long) and the overwhelming majority of subclasses in class I (also called short) do not mark the second person singular present overtly, whereas the verbs of the class III (mixed) use the suffix -i in this context. In addition, the most productive and numerous classes are I and II. Thus, the exemplar paradigm selected for Latvian has no overt marker in the second person singular present (Holst 2001, Mathiassen 1997, Nau 1998).¹ We limit our attention to agreement among lexical predicates that qualify as open, default classes of their language and exclude agreement paradigms of predicates that are deficient or deponent or that show special alignment patterns (such as experiencers coded like objects), any other special behavior or lexical constraints of any kind. There is only one language in our database, where no default intransitive predicate class can be established for agreement. This is Choctaw (ISO639.3:cho; Muskogean) and in order to keep the coding consistent, we excluded this language from further analysis. A further restriction is that we only look at person distinctions, and even more simply at only a binary distinction between first and second vs. third person. When there are subdistinctions within person, such as gender or honorificity, we coded the category that is said to be unmarked in the source (usually masculine, non-honorific). We did this in order to factor out confounding effects from these categories.²

4 Silverstein’s Law 4.1 Background and hypothesis The alignment of generalized roles (S, A, P etc.) is well-known to be split sometimes within languages, so that some parts align ergatively, some accusatively. If

1 We did not assess the relative frequency of paradigms in terms of discourse frequency. This would be a very expensive project, and we doubt that results would have a major impact on our general findings since the number of relevant splits is too rare to begin with. 2 The dataset used in this study is available for download at http://dx.doi.org/10.1515/ 9783110399967.suppl, once coded for alignments and once coded for zeros, following the coding principles explained below.

34   

   Balthasar Bickel, Alena Witzlack-Makarevich, Taras Zakharko and Giorgio Iemmolo

such a split is governed by a difference in person, it is commonly expected that we find accusative alignment in the first and second and ergative alignment in the third person. This idea was developed in the 1970s (e.g. , Silverstein 1976) and was originally formulated in terms of a synchronic generalization.³ Later on, it has also been rephrased in terms of an absolute, non-violable property of Universal Grammar (Aissen 1999) that constrains how languages can or cannot change (Kiparsky 2008). Here we test this prediction as a statistical diachronic universal, calling it ‘Silverstein’s Law’. We expect that, if paradigms have a person-conditioned split in alignment, they are more likely to develop in such a way that ‘S=A’ alignment (i.e. accusative or neutral) is found in the first and second person and ‘S=A’ ̸ alignment (i.e. ergative) in the third person, rather than the other way round. If a paradigm shows the preferred pattern, it is expected to be preserved as such over time; if a paradigm violates the pattern, it is expected to develop the preferred pattern. This is visualized in Figure 1, where paradigms are represented by boxes and where the thickness of arrows depicts the expected probability of diachronic developments between paradigms. 1/2: S=A ̸

1/2: S=A

3:

3:

S=A

S=A ̸

Fig. 1: Hypothesis of Silverstein’s Law as a diachronic universal affecting agreement paradigms

Apart from person, other features – especially animacy and number – are expected to affect the distribution of ergativity as well. However, we limit our attention in the following to person only because person is by far the most widespread agreement category in the world. If Silverstein’s Law leaves a detectable statistical signal anywhere, one would expect this to be first and foremost in person agreement.

3 Note that not all claims about the distribution of alignment types over person focus on or even include agreement systems. Comrie (1981) for example limits the claim to case marking and does not consider agreemeent systems.

Alignment and zero-marking across persons   

   35

4.2 Alignment coding For determining alignment types, we focus on the coding of S, A, and P argument roles and exclude arguments of three-argument verbs from our present purview. S, A, and P are defined by numerical valency and semantic entailment properties of lexical predicates, following earlier proposals of ours (e.g., Bickel & Nichols 2009, Bickel 2011, Witzlack-Makarevich 2011). We understand alignment type as a particular grouping of S, A, and P argument roles in a specific morphosyntactic operation, here agreement marking: if an agreement marker, or a set of agreement markers, treats S and A alike, as it does for example in English or Latin, S aligns with A, while P is kept apart insofar as it does not trigger agreement marking. In English, the third person triggers s-agreement with both S arguments (he work-s) and A arguments (he see-s me), but not with P arguments (*I see-s him). Such a pattern (i.e. S=A=P) ̸ is traditionally known as accusative alignment. Other possibilities are S=P=A ̸ (ergative alignment), S=A=P (neutral alignment), S=A ̸ =P ̸ (tripartite alignment), and, finally, S=A=P ̸ (horizontal alignment). Determining alignment in agreement paradigms can be non-trivial, mostly because individual agreement markers can pattern in a way that differs from the basic question of which arguments ever trigger agreement (Siewierska 2003, Bickel et al. 2013). For example, in an agreement paradigm like that of Swahili (ISO639.3:swh; Benue-Congo) individual agreement markers pattern accusatively: the S and A arguments of the third person singular in noun class I require the agreement prefix a- ‘3s.I.S/A’, whereas the P argument requires the prefix mw‘3s.I.P’. On the other hand, when one asks which arguments trigger agreement, the answer would be S, A, and P alike (i.e. neutral) because they all trigger agreement of some kind: (1)

a. a-li-mw-ona mbuzi. 3sI.S/A-PST-3sI.P-see goat ‘S/he (A) saw the goat (P).’

b. a-li-kimbia. 3sI.S/A-PST-run ‘S/he (S) ran.’

Whenever there is such a discrepancy, we determine alignment on the basis of individual agreement markers as they appear or fail to appear in specific morphological positions/slots, following the procedure described in detail in Bickel et al. (2013). Thus, the data in (1) are coded as follows: in the first prefix position, we get accusative alignment for the third person since S and A are marked by a- while P is not marked. In the third position, i.e. after the tense marker, the third person is again accusatively aligned but for a different reason: here it is based on the fact that P is marked by mw- while S and A are treated alike by not being marked. The

36   

   Balthasar Bickel, Alena Witzlack-Makarevich, Taras Zakharko and Giorgio Iemmolo

pattern in (1) is not the only one in Swahili; under certain semantic and pragmatic conditions, there is no overt P agreement: (2)

a-li-ona mbuzi. 3sI.S/A-PST-see goat ‘S/he (A) saw a goat (P).’

In this case, Swahili has neutral alignment in the third position since now all arguments are alike in failing to show any agreement exponence in that slot. We determined alignment separately in each condition that a language may put on its agreement paradigms. Computing alignments in this way has the advantage that the specific morphological structure of paradigms is respected. The disadvantage is that in languages like Swahili one looses the generalization that only S and A are marked in the first and only P argument in the third slot, so that one could derive the accusative alignment in the third slot from the accusative alignment in the first slot. However, the situation in Swahili is very special, and generalizations of this kind are far from being universal. There are many languages where the distribution of S, A and P markers is not tied to specific positions. In Pipil (ISO639.3:ppl; UtoAztecan; Campbell 1985) or Puma (ISO639.3:pum; Kiranti branch of Sino-Tibetan; Bickel et al. 2007), for example, the same positions cover S, A and P markers alike. Consider first person singular agreement in the Puma past affirmative paradigm: a. cind-u-ŋ teach[-PST]-3sP-1sA ‘I (A) taught him/her (P)’

c. p -cind-oŋ 3sA-teach-1sS/P.PST ‘s/he (A) taught me (P)’

b. t -cind-oŋ 2-teach-1sS/P.PST ‘you (A) taught me (P)’

d. phind-oŋ jump-1sS/P.PST ‘I (S) jumped’

v

(3)

v

First person alignment is neutral in the prefix position as all roles are alike in having zero realization, except for the second person, which is realized as t - (3b), and the third person which is realized as p - (3c). This differs from the final suffix position, where the first person shows ergative alignment: it is registered as -oŋ in S and P function (cf. 3b and 3c for P, and 3d for S) but as -ŋ in A function (3a). Unlike in Swahili, the alignment in the prefix position cannot be predicted from the alignment in the suffix position, or vice versa. Thus, in order to be able to capture alignment in all agreement paradigms, we coded alignment separately for each person in each morphological position. In order to achieve consistency, we do this v

v

Alignment and zero-marking across persons   

   37

in all languages, including languages like Swahili, where one could in principle also characterize the alignment patterns in terms of generalized positions. The Puma data point to yet another problem: the alignment pattern in a specific person category can depend on the presence of other person markers in the same form. For the second person, alignment in Puma is neutral in the prefix position since all roles show t - here. Compare the following forms (and also 3b, which is parallel to 4a, with t - marking A): v v

a. t -cind-i 2-teach-3sP.PST ‘you (A) taught him/her (P)’

c. t -phind-a 2-jump-PST ‘you (S) jumped’ v

v

(4)

b. t -cind-a 2-teach-PST ‘s/he (A) taught you (P)’ v

These forms show that t - occurs in all three roles. However, when the second person combines with a first person, the pattern is different, as there is no prefix in this case: v

(5)

cin-na-a teach-1>2-PST ‘I (A) taught you (P)’

Here, second person P is realized by means of the suffix -na ‘first person acting on second person’. This form contrasts with t - which in the same first person context covers the A role (3b) and which also covers the S role (4c), thus constituting accusative alignment. This kind of split, which is conditioned by coarguments, is very common in Puma and similar languages (Bickel et al. 2013). For example, third person in Puma shows ergative alignment in the prefix position when co-occurring with a first person: there is a marker p - in the form for ‘s/he taught me’ (3c) while the third person lacks overt marking in the prefix position when it is in the P function (as in the ‘I taught him/her’ form in 3a) or in the S function (phind-a ‘jump-PST’, i.e. ‘s/he jumped’). However, when the third person co-occurs with a third person, it also lacks overt marking in the prefix position (cind-i ‘teach-3sP.PST’, i.e. ‘s/he taught him/her’), and so, alignment is neutral here. (The distribution is again different in the suffix position, where third person aligns accusatively). In response to these issues, we coded alignment not only per person in each slot but also relative to the co-arguments that a form might have. This increases v

v

38   

   Balthasar Bickel, Alena Witzlack-Makarevich, Taras Zakharko and Giorgio Iemmolo

the number of alignment statements considerably, as combinations multiply. For example, third person S can be compared to third person A when co-occurring with first person P and to third person P when co-occurring with first person A; or to third person A when co-occurring with second person P and to third person P when co-occurring with first person A, and so on. Summary characterizations of paradigms then need to be statistical, e.g., via the empirical distribution of alignment types, or the proportion of accusative vs. ergative alignments, etc. In the following we focus on the distribution of S=A(=P) vs. S=A ̸ alignments, in line with the hypothesis put forward above.

4.3 Results

6 4 0

2

Frequency

8

10

Of the 352 paradigms, only 52 show an S=A vs. S=A ̸ split, distributed over 42 languages. Almost all cases come from Algonquian (10 languages) and the Kiranti subgroup of Sino-Tibetan (19). The rest is found in Mayan (7 languages), Southern Nilotic (Teso, ISO639.3:teo), Pano-Tacanan (Reyesano, ISO639.3:rey), MacroGe (Bororo, ISO639.3:bor), Sepik (Namambu, ISO639.3:mle) and in the isolates Ainu (ISO639.3:ain) and Zuni (ISO639.3:zun). In order to assess the distributions of the alignments, we computed for each paradigm the proportion of S=A vs. S=A ̸ alignments within all first and second persons and, separately, within all third person forms (i.e. across all number and other distinctions and across all co-arguments that they combine with). The results of this are reported in Table 1 and summarized in Figure 2. The last column in the table reports whether S=A alignments are more common in the first and second person than in the third. According to Silverstein’s

-0.5

-0.4

-0.3

-0.2

-0.1

0.0

0.1

0.2

0.3

0.4

0.5

Fig. 2: Histogram of differences between the proportion of S=A alignments in the first and second vs. in the third person in Table 1. Negative values contradict the predictions of Silverstein’s Law and are printed in grey.

Branch

Algonquian Algonquian Algonquian Algonquian Algonquian Algonquian Algonquian Algonquian Algonquian Algonquian Bororo Gr. Kanjobalan Gr. Tzeltalan Gr. Tzeltalan Gr. Tzeltalan Quichean-Mamean Quichean-Mamean Yucatecan Yucatecan Southern Nilotic Tacan Middle Sepik (Ndu) Kiranti Kiranti Kiranti

Stock

Ainu Algic Algic Algic Algic Algic Algic Algic Algic Algic Algic Macro-Ge Mayan Mayan Mayan Mayan Mayan Mayan Mayan Mayan Nilotic Pano-Tacan Sepik Sino-Tibetan Sino-Tibetan Sino-Tibetan

Ainu Arapaho Atikamekw Blackfoot Cheyenne Cree (Plains) Menomini Micmac Munsee Ojibwa (Eastern) Passamaquoddy Bororo Jacaltec Chontal Maya Chortí Chortí Quiche Tzutujil Itzaj Yucatec Teso Reyesano Mambu Athpare Bahing Bahing

Language

IND PST.IND.AFF NPST.IND.AFF

COMPL COMPL

PFV NCOMPL COMPL

INDEP INDEP.NPST INDEP.AFF INDEP.PRS.IND INDEP INDEP INDEP.IND INDEP.NPST INDEP.PRS INDEP.PRS

Paradigm

Table 1: Proportion of S=A (as opposed to S=A) ̸ alignments in paradigms with splits

0.811 0.762 0.879 0.814 0.801 0.887 0.845 0.868 0.806 0.832 0.839 0.000 0.000 0.556 0.833 0.333 0.000 0.000 0.200 0.200 1.000 0.706 0.000 0.878 0.909 0.914

Prop(S=A) in 1/2 1.000 0.719 0.789 0.923 0.690 0.691 0.846 0.639 0.676 0.799 0.788 0.500 0.500 0.667 0.667 0.333 0.250 0.250 0.167 0.167 0.671 0.986 0.500 0.946 0.889 0.894

Prop(S=A) in 3

−0.189 0.043 0.091 −0.109 0.111 0.196 −0.001 0.229 0.130 0.033 0.051 −0.500 −0.500 −0.111 0.167 0.000 −0.250 −0.250 0.033 0.033 0.329 −0.281 −0.500 −0.069 0.021 0.020

Diff.

Alignment and zero-marking across persons       39

Branch

Kiranti Kiranti Kiranti Kiranti Kiranti Kiranti Kiranti Kiranti Kiranti Kiranti Kiranti Kiranti Kiranti Kiranti Kiranti Kiranti Kiranti Kiranti Kiranti Kiranti Kiranti Kiranti Kiranti Kiranti Kiranti

Stock

Sino-Tibetan Sino-Tibetan Sino-Tibetan Sino-Tibetan Sino-Tibetan Sino-Tibetan Sino-Tibetan Sino-Tibetan Sino-Tibetan Sino-Tibetan Sino-Tibetan Sino-Tibetan Sino-Tibetan Sino-Tibetan Sino-Tibetan Sino-Tibetan Sino-Tibetan Sino-Tibetan Sino-Tibetan Sino-Tibetan Sino-Tibetan Sino-Tibetan Sino-Tibetan Sino-Tibetan Sino-Tibetan Zuni

Table 1: (continued)

Bantawa Belhare Camling Chintang Dumi Hayu Hayu Jero Kõic Kõic Koyi Koyi Kulung Kulung Limbu Limbu Lohorung Lohorung Puma Puma Thulung Thulung Wambule Yakkha Yamphu Zuni

Language

IND.AFF

PST.IND.AFF NPST.IND.AFF NPST.IND.AFF PST.IND.AFF NPST.IND.AFF PST.IND.AFF PST.IND.AFF NPST.IND.AFF NPST.IND.AFF PST.IND.AFF NPST.IND.AFF PST.IND.AFF NPST.IND.AFF PST.IND.AFF

IND.AFF NPST.IND.AFF NPST.IND.AFF PST.IND.AFF NPST.IND.AFF

Paradigm 0.839 0.870 0.808 0.847 0.837 0.937 0.898 0.756 0.954 0.972 0.835 0.864 0.854 0.844 0.855 0.849 0.822 0.822 0.877 0.877 0.911 0.917 0.756 0.881 0.836 0.500

Prop(S=A) in 1/2 0.853 0.963 0.838 0.905 0.606 0.915 0.879 0.768 0.852 1.000 0.859 0.894 0.964 0.964 0.946 0.946 0.904 0.904 0.870 0.870 0.949 0.949 0.657 0.953 0.961 0.500

Prop(S=A) in 3

−0.013 −0.093 −0.030 −0.059 0.231 0.021 0.019 −0.011 0.102 −0.028 −0.023 −0.030 −0.110 −0.120 −0.091 −0.097 −0.082 −0.082 0.007 0.007 −0.038 −0.033 0.100 −0.071 −0.125 0.000

Diff.

40       Balthasar Bickel, Alena Witzlack-Makarevich, Taras Zakharko and Giorgio Iemmolo

Alignment and zero-marking across persons   

   41

Law, this should tend to be the case. However, as the histogram of differences in Figure 2 shows, there are far more paradigms that contradict this, resulting in negative numbers, viz. 60 % paradigms (31 out 52). Only 40 % of the paradigms follow Silverstein’s Law. The histogram also suggests that most differences cluster around 0, i.e. most paradigms simply show only a marginal difference between the proportions of S=A alignments across persons. Aggregating the proportions per family allows us to estimate diachronic biases along the lines described in the methods section, although given the small number of relevant families, we do not extrapolate to small families here. The results show that Algonquian is the only large family with a bias towards obeying Silverstein’s Law, with 8 out 10 paradigms following it. But even in Algonquian, the trend is only borderline significant (one-sided binomial test, p = .055), and we are therefore reluctant to infer a diachronic bias in the family based on this. We conclude that there is no support for Silverstein’s Law in agreement paradigms. The other families with splits do not show the slightest bias in line with the hypothesis from Figure 1. Mayan and Kiranti even have slight biases against the hypothesis. In Mayan, 5 out 8 paradigms contradict the hypothesis; in Kiranti, 19 out of 28 paradigms contradict it (which is in fact even statistically significant at a 5 % rejection level). Among the families for which we have only one representative each, Ainu, Bororo, Reyesano, and Manambu also contradict the hypothesis (Table 1); only the Nilotic language Teso supports it, while Zuni is neutral in this regard (having S=A ̸ in the plural in all persons and neutral alignment in the singular).

4.4 A non-paradigmatic interpretation of Silverstein’s Law? While there is no support for Silverstein’s Law as explicated in Figure 1, the question arises whether there is support for it under a different interpretation. So far we have looked at paradigm structures, counting different alignment types within each paradigm separately (Table 1) and assessing family biases on the paradigms in a second step only. Under an alternative, ‘non-paradigmatic’ interpretation, Silverstein’s Law does not operate on paradigm structures but predicts a higher probability for S=A alignment in first and second than in third persons more generally and cross-cutting paradigms. To test whether Silverstein’s Law has more empirical support under such an interpretation, we performed a family bias analysis without respecting paradigms. We first split the data into data on alignment in the first and second person and data on alignment in the third person. Using the method explained in Section 2, we then estimated the extent to which there are family biases towards S=A align-

42   

   Balthasar Bickel, Alena Witzlack-Makarevich, Taras Zakharko and Giorgio Iemmolo

bias towards S=A

1/2 3 Africa

1/2 3 Eurasia

1/2

3 Pacific

1/2 3 Americas

bias towards S A

Fig. 3: Family bias estimates ignoring paradigm structures (based on all 314 languages)

ment in each dataset, controlling for large-scale macro-areas.⁴ The results of these estimates (including now extrapolations to small families and isolates) is shown in Figure 3.⁵ The figure leaves out families that are diverse because the absence of any bias gives no evidence of the direction of diachronic developments (Bickel 2013). The figure shows that, in each area, there are about as many families with biases in each direction in the first or second person as families with biases in each direction in the third person. This suggests that Silverstein’s Law has no empirical support under a non-paradigmatic interpretation either. The findings confirm earlier conclusions by Bickel (2008), which were based on a different and smaller dataset (that of Bakker & Siewierksa 2006), adopted a different method for coding alignments (focusing on overall types of splits, rather than on alignment proportions), and relied on a more qualitative than quantitative approach.

5 Watkins’ Law 5.1 Background In its original formulation, Watkins’s Law concerns the analogical reanalysis and subsequent reorganization of inflectional paradigms on the basis of the third person singular. The law was proposed by Watkins (1962) for a number of ancient

4 We did this by running the familybias function (Zakharko & Bickel 2011ff) directly on a binary re-code of the alignment statement, differentiating only between S=A vs. S=A. ̸ We used the default setting of the function. 5 Here and in all other figures below, we use the ‘mosaic’ visualization technique of Meyer et al. (2006), so that the sizes of tiles in the graph are proportional to the number of families in each condition.

Alignment and zero-marking across persons   

   43

Indo-European languages as well as in the reconstruction of Proto-Indo-European (see Watkins 1969). Watkins (1962:90–96) proposes a scenario according to which third person markers are reanalyzed as part of the verbal stem, thus giving rise to zero marking in the third person. An illustration of this process comes from the reanalysis of the third person singular ending -t as part of the verbal stem from Proto-Iranian to Persian (cf. Watkins 1962:94). The process is represented in Table 2. Table 2: Singular agreement forms of ‘to be’ from Proto-Iranian to Persian Proto-Iranian 1sg 2sg 3sg

*as-mi *as-i *as-ti

≻ *as-m ≻ *as-i ≻ *as-t ≻ *ast-0

Persian ≻ *ast-m ≻ *ast-i ≻ *ast-0

≻ hast-am ≻ hast-i ≻ hast-0

The reanalysis comprises several aspects. First, -t is reanalyzed as part of the verbal stem ast. The new stem then becomes the basis for the rest of the paradigm. This in turn leaves the third person form ast identical with the stem, i.e. without a proper ending. The paradigmatic opposition to the first and second person finally establishes a zero exponent in this form, specialized for third person agreement. The same analysis applies, mutatis mutandis, to other languages across the IndoEuropean family (see for example Bybee & Brewer 1980 on Provençal, Haiman 1977 on Vallader Rumantsch, among others). As stated by Watkins (1962:178), the reanalysis of third person as having zero exponence constitutes a “general linguistic phenomenon”. Koch (1995:64) has indeed proposed that such processes should be seen as driven by a general diachronic principle, according to which a form which expresses via non-zero marking a feature that is typologically expected to be zero marked (such as a third person) is likely to be reanalyzed as having zero exponence. The validity of this process as a general diachronic principle, however, has been questioned by, e.g., Hock (1986:218–222), who considers the evidence for the reanalysis of third person as “meager”. Reanalysis is not the only possible process through which zero exponence arises. Another possibility is that a language never developed overt marking for the third person. In some language families the evidence in fact points towards a “non-development” scenario, that is a scenario where overt marking for third person never arose. Such a scenario is exemplified by Uralic languages. In ProtoUralic, only first and second person suffixes are reconstructable for S/A (Collinder 1965:58, Körtvély 2005). Third person is very often unmarked in modern Uralic

44   

   Balthasar Bickel, Alena Witzlack-Makarevich, Taras Zakharko and Giorgio Iemmolo

languages: when overtly marked, there is considerable heterogeneity in the innovated markers (see also Comrie 1981). The classical explanation for the lack of overt marking in the third person goes back to Jakobson’s (1932) and Benveniste’s (1946:227–232) characterization of the third person as a “non-person”, negatively defined. Zeros are more likely to be found in the third person as opposed to first and second because of its “impersonal, non-referring nature”, as Benveniste has it. The restructuring of paradigms based on Watkins’ Law is explained, e.g., by Koch (1995), in terms of iconicity: this is due to the tendency for paradigms to encode via zero the “non-person” as opposed to first and second. Under either the reanalysis or the non-development scenarios outlined above, Watkins’ Law predicts that languages should preferentially develop and maintain paradigms with zero-marked third person, as for example in the Persian Aorist or in the Puma Intransitive Past, summarized in Table 3 (cf. Section 4.2 for details on Puma). Watkins’ Law should disfavor paradigms such as the one found in the English Present tense, where only the third person is overtly marked, as well as paradigms such as the one in the Persian present tense, where all persons have overt exponence (cf. the rightmost columns in Table 3). Note that the reason for being disfavored are different in the English and the Persian Present tense paradigms: the English paradigm is disfavored because it has zeros ‘in the wrong place’; the Persian paradigm is disfavored because it would be expected to undergo reanalysis along the lines sketched in Table 2 – not categorically, but as a worldwide trend. Table 3: Favored and disfavored paradigms according to Watkins’ Law Favored Persian Aorist (‘go’)

Puma Intransitive Past (‘jump’)

English Present

Persian Present (‘go’)

raft-am raft-i raft-0

phind-oŋ t -phinda 0-phinda-0

like-0 like-0 like-s

mi-rav-am mi-rav-i mi-rav-ad

v

1sg 2sg 3sg

Disfavored

Following the scheme used to visualize Silverstein’s Law in Section 4, Figure 4 visualizes Watkins’ Law as a diachronic universal operating on paradigms (where again paradigms are represented by boxes and probabilities of diachronic development by arrow thickness; ‘x’ stands here for any overt marker).

Alignment and zero-marking across persons   

1/2: x/0

1/2: x

3:

3:

x

   45

0

Fig. 4: Hypothesis of Watkins’ Law as a diachronic universal affecting agreement paradigms

5.2 Coding For this study, we specified for each person feature whether or not it has any overt marking in its agreement morphology. If paradigms also register number, we limit our attention to the way person is marked in the singular number. Non-singular number strongly tends to induce overt marking because it is a structurally marked category (Greenberg 1966b). Any such effect would cancel out possible signals from Watkins’ Law. When agreement is sensitive to co-arguments in the way explained in Section 4.2, we coded all relevant singular forms, for example third person singular, when occurring as S, when occurring as A acting upon first person singular, dual, etc. P, or on second person singular, dual etc., when occurring as P being acted open by a first person singular, dual, etc. A argument, and so on. Like in the previous study, this has the consequence that, for some paradigms, there is a substantial number of entries in the database.

5.3 Results For each agreement paradigm, we computed the proportion of overt markers in first and second person as opposed to third person and then determined for each paradigm whether it obeys Watkins’ Law, i.e. whether it has more zero markers in the third than in the other persons. Only 35 % of the paradigms in our dataset follow the law. All others have what is expected to be a disfavored structure in Table 3 and what appears on the left side in Figure 4. This is in line with what Cysouw (2003:53) observes in a survey of paradigms involving syncretism (like English, where first and second person syncretize): there is no evidence for zeros to be more common among third than among non-third persons. Estimates of family biases on whether paradigms obey vs. violate Watkins’ Law are given in Figure 5. The results show that there is no support for Watkins’ Law as a diachronic universal: families are not more likely to be biased towards developing and maintaining paradigms in line with the law than they are likely to be biased in the opposite direction. In terms of the visualization in Figure 4, this means that all arrows should have about the same thickness, or that indeed

46   

   Balthasar Bickel, Alena Witzlack-Makarevich, Taras Zakharko and Giorgio Iemmolo

the arrows from the favored (on the right-hand side) to the disfavored (on the left-hand side) paradigm should be thicker than those from the disfavored to the favored paradigm type (except perhaps in the Americas where this trend seems to be weaker).

bias against fitting Watkins' Law bias towards fitting Watkins' Law Africa

Eurasia

Pacific

Americas

Fig. 5: Family bias estimates on paradigm structures

The only large families (taken here to be families with at least five members) in our sample that abide by Watkins’ Law are Mayan and Kiranti. For Mayan, the bias is complete: all Mayan paradigms in our dataset comply with Watkins’ Law. For Kiranti, the bias is incomplete, but strong and statistically significant (90 % fit, N = 30, p < .001 under a one-sided binomial test). In addition, there is a borderline trend in Algonquian (75 % fit, N = 12, p = .073). Beyond these three cases, all other large families in our dataset show a bias against Watkins’ Law (e.g. Oceanic, Berber, Chimbu-Wahgi, Dravidian, Nakh-Daghestanian, Omotic, Torricelli, Tucánoan, Uralic, West Papuan). This also includes Indo-European, the family for which it was originally formulated. Of all the 28 Indo-European languages in our dataset (sampled from all branches except Tocharian), only 5 languages (Latvian, Slovene, Provençal, Nepali, and of course Persian) have at least one paradigm that complies with Watkins’ Law. All others are of the type that would be disfavored by the law, as illustrated above in Table 3, most of them with overt agreement markers in all persons (like in the Persian Present tense paradigm in the table). We conclude that Watkins’ Law receives no support when tested against a large cross-linguistic sample: there is no cross-linguistic tendency for languages to develop paradigms with zero-marked third person and overt marking for first and second either via reanalysis or by non-development. Note that this does not invalidate Watkins’ Law as a description of what happened in individual paradigms such as the Persian Aorist; but it means that such a process is not motivated by a universal principle – that is, it cannot be said to be a “natural” development.

Alignment and zero-marking across persons   

   47

5.4 A non-paradigmatic interpretation of Watkins’ Law Like in the study of Silverstein’s Law, the question arises whether Watkins’ Law has better support if it is interpreted differently. In its original formulation and in the way we approached it so far (cf. Figure 4), the law is interpreted as a force driving the structuring of paradigms. However, it is also possible to interpret the law as affecting the distribution of zero marking independently of paradigm structures. The hypothesis then is this: zero markers are more likely to be found in the third than in the first or second person, across all paradigms in a family. The motivation for such a hypothesis derives from grammaticalization theory. It has often been proposed that agreement markers develop from independent pronouns (see, e.g., Givón 1976, Ariel 2000). The distribution of third person pronouns leads one to expect that they end up as zeros when they grammaticalize into agreement morphology. Bybee (1985) suggests that this higher likelihood for zero agreement markers in third person is due to the lower frequency of overt third person pronouns in discourse. An alternative account, making opposite assumptions on frequency, is given by Greenberg (1966a:65–69). According to this account, third person forms are usually realized as zeros because of their higher frequency in discourse, given that third person forms accompany lexical NPs while first and second do not. Then, higher frequency leads to phonetic reduction and loss or reanalysis of overt marking. In a third account, elaborated by Ariel (2000), third person forms are more frequently zero not because of their relative frequency, but rather because of their lower accessibility in discourse compared to first and second person. Whatever explanation has ultimately the best empirical and theoretical support, a grammaticalization-based theory is plausible and worth testing. To test the hypothesis, we proceeded in the same way as with the nonparadigmatic version of Silverstein’s Law: we split the data on individual agreement makers into a set for first and second person and a set for third person. Within these, we estimated family biases towards the development or persistence of zero marking. Figure 6 shows the results. Like before, we leave out families that are diverse, i.e. families that lack a bias in any direction, because they give no support for or against the hypothesis.⁶ For families with a bias, the hypothesis predicts that there should be significantly more families biased towards zero-marking among third person forms than

6 Note that the number of diverse families is different when looking at first and second as opposed to third person forms. As a result, the total counts of biased families differs in some areas. This is reflected by unequal tiles in the plot.

48   

   Balthasar Bickel, Alena Witzlack-Makarevich, Taras Zakharko and Giorgio Iemmolo

bias towards overt marking

bias towards -marking 1/2 3 Africa

1/2 3 Eurasia

1/2

3 Pacific

1/2 3 Americas

Fig. 6: Family bias estimates on individual agreement forms

among first or second person forms, independently of areas. This is statistically supported, although the effects are relatively week (as the figure shows). A likelihood ratio test on loglinear models shows that the interaction between person × bias direction is significant (χ 2 = 5.26, p = .022) and at the same independent of a further three-way interaction with area (χ 2 = .79, p = .85).

6 Conclusions In this paper we assessed the validity of two hypotheses on possible principles that shape agreement morphology over time. Our data lend no support for what we call Silverstein’s Law in agreement morphology: the distribution of S=A ̸ and S=A alignment in the structure of agreement morphology does not appear to be subject to general principles but seems to be the result of individual processes. Watkins’ Law, by contrast, shows more empirical success but only if the law is interpreted not as a factor that operates on paradigms but as a factor in the grammaticalization of individual agreement markers, independently of and cutting across paradigms: across the board, third person markers are more likely to develop and maintain zero exponence than first and second person markers. This finding suggests that paradigm structures are perhaps generally less open to universal factors than individual markers (or constructions). A possible explanation for this could come from the observation that paradigm structures tend to be simultaneously affected by a vast range of independent developments, from phonological change to analogical leveling. As a result of this, general principles cannot easily exert systematic effects that leave clear and detectable signals in daughter languages. The grammaticalization of individual markers, by contrast, can proceed in a relatively independent manner, so that general principles can channel these developments along universal pathways. It is an

Alignment and zero-marking across persons   

   49

open issue for future research to determine whether this difference is of relevance beyond the phenomena we looked at in the present study. Our finding also invites further research on the theoretical background of why the grammaticalization of third persons tends to end up in zeros more often than the grammaticalization of first and second person. As the discussion in Section 5.4 suggests, progress in this crucially depends on a better understanding of frequency distributions in discourse and their effects on language change. This ultimately requires large-scale corpus research across a larger variety of languages and genres than have traditionally been looked at so far.

Acknowledgement This research was supported by Grant Nrs. BI 799/3-1/2 and BI 799/5-1 from the Deutsche Forschungsgemeinschaft. A previous version was presented at the workshop on ‘Agreement from a diachronic perspective’, Marburg, October 3–5, 2012 and at the 21st International Conference on Historical Linguistics, Oslo, August 5–9, 2013. We thank these audiences as well as two reviewers, the editors, and Michael Cysouw for insightful comments and helpful questions. All remaining shortcomings are our own.

Bibliography Aissen, Judith. 1999. Markedness and subject choice in Optimality Theory. Natural Language and Linguistic Theory 17. 673–711. Ariel, Mira. 2000. The development of person agreement markers: from pronouns to higher accessibility markers. In Michael Barlow & Suzanne Kemmer (eds.), Usage-based models of language, 197–260. Stanford: CSLI. Bakker, Dik & Anna Siewierksa. 2006. The agreement database. http://www.lotschool.nl/ Research/ltrc/agreement.htm; accessed August 7, 2006. Benveniste, Émile. 1946. Structure des relations de personne dans le verbe. In Problèmes de linguistique générale, 1, 225–236. Paris: Gallimard 1966. Bickel, Balthasar. 2008. On the scope of the referential hierarchy in the typology of grammatical relations. In Greville G. Corbett & Michael Noonan (eds.), Case and grammatical relations: papers in honor of Bernard Comrie, 191–210. Amsterdam: Benjamins. Bickel, Balthasar. 2011. Grammatical relations typology. In Jae Jung Song (ed.), The Oxford handbook of language typology, 399–444. Oxford: Oxford University Press. Bickel, Balthasar. 2013. Distributional biases in language families. In Balthasar Bickel, Lenore A. Grenoble, David A. Peterson & Alan Timberlake (eds.), Language typology and historical contingency, 415–444. Amsterdam: Benjamins.

50   

   Balthasar Bickel, Alena Witzlack-Makarevich, Taras Zakharko and Giorgio Iemmolo

Bickel, Balthasar. 2015. Distributional typology: statistical inquiries into the dynamics of linguistic diversity. In Bernd Heine & Heiko Narrog (eds.), The Oxford handbook of linguistic analysis, 2nd edition, 901–923. Oxford: Oxford University Press. Bickel, Balthasar, Martin Gaenszle, Arjun Rai, Prem Dhoj Rai, Shree Kumar Rai, Vishnu S. Rai & Narayan P. Sharma (Gautam). 2007. Two ways of suspending object agreement in Puma: between incorporation, antipassivization, and optional agreement. Himalayan Linguistics 7. 1–18. Bickel, Balthasar, Giorgio Iemmolo, Taras Zakharko & Alena Witzlack-Makarevich. 2013. Patterns of alignment in verb agreement. In Dik Bakker & Martin Haspelmath (eds.), Languages across boundaries: studies in memory of Anna Siewierska, 15–36. Berlin: De Gruyter Mouton. Bickel, Balthasar & Johanna Nichols. 2007. Inflectional morphology. In Timothy Shopen (ed.), Language typology and syntactic description, 169–240. Cambridge: Cambridge University Press (Revised second edition). Bickel, Balthasar & Johanna Nichols. 2009. Case marking and alignment. In Andrej Malchukov & Andrew Spencer (eds.), The Oxford handbook of case, 304–321. Oxford: Oxford University Press. Blevins, Juliette. 2004. Evolutionary phonology: the emergence of sound patterns. Cambridge: Cambridge University Press. Bybee, Joan. 1988. The diachronic dimension in explanation. In John A. Hawkins (ed.), Explaining language universals, 350–379. Oxford: Blackwell. Bybee, Joan L. 1985. Morphology. Amsterdam: Benjamins. Bybee, Joan L. & Mary Alexandra Brewer. 1980. Explanation in morphophonemics: changes in Provençal and Spanish preterite forms. Lingua 52. 201–242. Campbell, Lyle. 1985. The Pipil language of El Salvador. Berlin: Mouton de Gruyter. Campbell, Lyle. 1999. Historical linguistics: an introduction. Cambridge, Mass.: MIT Press. Collinder, Bjorn. 1965. An Introduction to the Uralic languages. Berkeley: University of California Press. Comrie, Bernard. 1981. Language universals and linguistic typology. Chicago: University of Chicago Press. Croft, William. 2003. Typology and universals. Cambridge: Cambridge University Press [2nd edition]. Cysouw, Michael. 2003. The paradigmatic structure of person marking. Oxford: Oxford University Press. Donohue, Mark, Tim Denham & Stephen Oppenheimer. 2012. Consensus and the lexicon in historical linguistics. Rejoinder to “Basic vocabulary and Bayesian phylolinguistics”. Diachronica 29. 538–546. Dryer, Matthew S. 1989. Large linguistic areas and language sampling. Studies in Language 13. 257–292. Dunn, Michael J., Simon J. Greenhill, Stephen C. Levinson & Russell D. Gray. 2011. Evolved structure of language shows lineage-specific trends in word-order universals. Nature 473. 79–82. Givón, T. 1976. Topic, pronoun, and grammatical agreement. In Charles N. Li (ed.), Subject and topic, 149–188. New York: Academic Press. Greenberg, Joseph H. 1966a. Language universals, with special reference to feature hierarchies. The Hague: Mouton.

Alignment and zero-marking across persons   

   51

Greenberg, Joseph H. 1966b. Synchronic and diachronic universals in phonology. Language 42. 508–517. Greenberg, Joseph H. 1978. Diachrony, synchrony and language universals. In Joseph H. Greenberg, Charles A. Ferguson & Edith A. Moravcsik (eds.), Universals of human language I: method and theory, 61–92. Stanford: Stanford University Press. Haiman, John. 1977. Reinterpretation. Language 53. 312–328. Hall, Christopher J. 1988. Integrating diachronic and processing principles in explaining the suffixing preference. In John A. Hawkins (ed.), Explaining language universals, 321–349. Oxford: Blackwell. Hock, Hans-Heinrich. Principles of historical linguistics. Berlin: de Gruyter. Holst, Jan Henrik. 2001. Lettische Grammatik. Hamburg: Helmut Buske Verlag. Jakobson, Roman. 1932. Zur Struktur des russischen Verbums. In Selected writings, vol. 2, 3–15. The Hague: Mouton 1971. Kiparsky, Paul. 2008. Universals constrain change; change results in typological generalizations. In Jeff Good (ed.), Linguistic universals and language change, 23–53. Oxford: Oxford University Press. Koch, Harold. 1995. The creation of morphological zeros. In Geert Booij (ed.), Yearbook of morphology 1994, 31–71. Dordrecht: Kluwer. Körtvély, Erika. 2005. Verb conjugation in Tundra Nenets. Szeged: SzTE Finnugor Tanszék. Maslova, Elena. 2000. A dynamic approach to the verification of distributional universals. Linguistic Typology 4. 307–333. Mathiassen, Terje. 1997. A short grammar of Latvian. Columbus, Ohio: Slavica Publishers. Meyer, David, Achim Zeileis & Kurt Hornik. 2006. The strucplot framework: visualizing multiway contingency tables with vcd. Journal of Statistical Software 17. 1–48. Nau, Nicole. 1998. Latvian. Munich: Lincom Europa. Nichols, Johanna. 1992. Linguistic diversity in space and time. Chicago: The University of Chicago Press. Quizar, Robin. 1994. Motion verbs in Ch’orti’. Función 15–16. 211–229. R Development Core Team. 2012. R: a language and environment for statistical computing. Vienna: R Foundation for Statistical Computing, http://www.r-project.org. Siewierska, Anna. 2003. Person agreement and the determination of alignment. Transactions of the Philological Society 101. 339–370. Siewierska, Anna. 2009. Person asymmetries in zero expression and grammatical functions. In Franc Floricic (ed.), Essais de linguistique générale et de typologie linguistique offerts au professeur Denis Creissels à l’occasion de ses 65 ans, 425–438. Paris: Presses de L’Ecole Normale Supérieure. Silverstein, Michael. 1976. Hierarchy of features and ergativity. In R. M. W. Dixon (ed.), Grammatical categories in Australian languages, 112–171. New Jersey: Humanities Press. Watkins, Calvert. 1962. Indo-European origins of the Celtic verb. Vol. I. The sigmatic aorist. Dublin: Institute for Advanced Studies. Watkins, Calvert. 1969. Geschichte der indogermanischen Verbalflexion. Heidelberg: Carl Winter. Witzlack-Makarevich, Alena. 2011. Typological variation in grammatical relations. Leipzig: University of Leipzig dissertation. Zakharko, Taras & Balthasar Bickel. 2011ff. familybias: Family bias estimation. R package, https://github.com/IVS-UZH.

Manuel Widmer

The transformation of verb agreement into epistemic marking: evidence from Tibeto-Burman Abstract: Tibeto-Burman languages of the Himalayas are renowned for their complex epistemic verbal systems. In the course of the past decades, a wealth of descriptive studies has considerably enhanced our synchronic understanding of such systems. However, the diachronic processes that give rise to epistemic verbal categories are still poorly understood. The article addresses this research gap and provides evidence for a diachronic link between the grammatical domains of person indexation and conjunct-disjunct marking. Evidence for this diachronic link will be drawn from the Tibeto-Burman languages Dolakha Newar (East Nepal) and Bunan (North India), both of which possess “hybrid verb systems” that display features of person agreement systems and conjunct-disjunct systems. Based on a comparison of Dolakha Newar and Bunan, it will be argued that the two languages bear witness to the functional reanalysis of person indexation as conjunct-disjunct marking. In the course of this process, a ternary person distinction (“first person” vs. “second person” vs. “third person”) is reduced to a binary epistemic distinction (“conjunct” vs. “disjunct”).

1 Introduction¹ Since Hale’s (1980) pioneering description of the verbal system of Kathmandu Newar, it has become evident that so-called “conjunct-disjunct systems” (i.e. verbal systems that encode whether or not the speaker possesses privileged access to the knowledge contained in a proposition; see below for a more detailed definition) do not only occur in the Greater Himalayan region (Tibeto-Burman, Mongolic), but also in South America (Barbacoan), Papua New Guinea (Trans-New Guinean), and the Caucasus (Nakh-Dagestanian) (cf. Creissels 2008: 3). Moreover, the publication of various descriptive studies has considerably enhanced

1 I would like to thank the editors, Scott DeLancey, Marius Zemp, Mark Post, Marc Matter, and two anonymous reviewers for helpful comments on earlier drafts of this article. Any remaining errors are my own responsibility. I would also like to acknowledge the Swiss National Science Foundation for its support of the research project “A Descriptive Grammar of Bunan” (grant P1BEP1_148871).

54   

   Manuel Widmer

our understanding of the grammatical category over the past few decades. Little research, however, has been dedicated to diachronic processes that give rise to such epistemic systems. This is regrettable, as insight into these mechanisms would undoubtedly improve our synchronic understanding of conjunct-disjunct marking. In the small number of articles published on the topic so far, two possible origins of conjunct-disjunct systems have been discussed: the development from copulas (DeLancey 1992: 59–60, 2011: 3–6) and from formerly infinite participial forms (Creissels 2008: 14–18; Genetti 1994: 130–136). In this article, I provide evidence for a third diachronic source that has not been described in the literature so far: the development of conjunct-disjunct systems from verb agreement systems through the reanalysis of the verbal category of person. Evidence for this process will be drawn from two Tibeto-Burman languages, Dolakha Newar and Bunan. The present article is structured in the following way: In section 2, I will provide a definition of conjunct-disjunct marking based on Hargreaves’ (1991, 2005) concepts of “privileged access” and “epistemic source”. In section 3, I will discuss the verb systems of Dolakha Newar and Bunan and show that the two languages possess “hybrid verb systems” that display both syntactic agreement and epistemic marking.² In section 4, the verbal systems of the two languages will be related to each other. It will be demonstrated that both languages bear witness to a similar functional relation between syntactic and epistemic markers. On this basis, it will be argued that these correspondences are the result of the same reanalysis in the course of which person agreement develops into conjunctdisjunct marking. Section 5 summarizes the findings of this study.

2 Conjunct-disjunct marking The term “conjunct-disjunct” was introduced by Hale (1980) to describe a previously unknown verbal category in Kathmandu Newar.³ Hargreaves (2003: 376) characterizes the conjunct-disjunct system of Kathmandu Newar with the following words:

2 The term “epistemic marking” was apparently first used by Bickel (2008). I use the term to refer to any verbal category that serves the primary function of indicating the source of knowledge or qualifying the access to knowledge. 3 Kathmandu Newar is a Tibeto-Burman language spoken in Central Nepal by approximately 690,000 speakers (Hargreaves 2003: 371).

The transformation of verb agreement into epistemic marking   

   55

A verb will have a conjunct form whenever: 1 The verb is finite, and 2 the event is construed as involving an intentional action by the actor, and 3 the speech act is: a declarative with a first person subject, or b interrogative with a second person subject, or c reported speech where the main clause subject and reported speech complement clause subject are coreferential. Disjunct suffixes occur in all other finite environments except those outlined above.

Hale (1980) originally interpreted the conjunct-disjunct system of Kathmandu Newar as a peculiar type of person agreement system that is sensitive to certain pragmatic factors, i.e. speech act deixis and intentional action. However, DeLancey (1986, 1990) made clear that conjunct-disjunct systems do not encode syntactic agreement but epistemic distinctions.⁴ Subsequent to DeLancey’s studies, Hargreaves (1991, 2005) developed a descriptive account of the verbal category that is based on the notions of “privileged access” and “epistemic source”. These two concepts will be briefly discussed in the following, as I consider them fundamental to the understanding of the functional motivation of conjunct-disjunct marking. According to Hargreaves (2005: 31), the notion “privileged access” refers to the fact that one’s own mental state is only directly accessible to oneself. Other persons are not able to access one’s mind in a similarly direct manner but can merely draw inferences about one’s internal state. This ontological restriction represents the functional foundation of conjunct-disjunct marking, as Hargreaves argues. In other words, conjunct-disjunct systems qualify the speaker’s access to the information contained in a proposition. Conjunct marking indicates that the relevant information is based on personal and internal knowledge that is only accessible to the speaker, while disjunct marking indicates that the information is based on common knowledge that is accessible to any other discourse participant as well. The conjunct-disjunct system of Kathmandu Newar operates on a rather narrow definition of privileged access, as conjunct marking is only licensed if an event is portrayed as willfully instigated (see Hargreaves’ quote above).⁵

4 Note that DeLancey (1986) had not yet adopted Hale’s term “conjunct-disjunct”, but instead referred to the category as “volitionality”. 5 It is important to note that such a narrow definition of privileged access is by no means universal. The considerable cross-linguistic diversity of conjunct-disjunct systems seems to be a consequence of the fact that languages operate on different definitions of privileged access (cf. Bickel 2008; Widmer forthcoming).

56   

   Manuel Widmer

Hargreaves (2005: 31) additionally introduces the notion “epistemic source” to capture the fact that conjunct marking can only be triggered by a number of specific speech act participants, i.e. the speaker in declarative contexts, the addressee in interrogative contexts, and the reported speaker in reported declarative contexts.⁶ The fact that these discourse participants are subsumed under one category is a direct consequence of the principle of privileged access (see above). The aforementioned speech act participants assume a privileged epistemic position in the corresponding pragmatic contexts, as they possess privileged access to the knowledge on which the respective speech acts are based. Based on Hargreaves’ model, conjunct-disjunct can thus be defined as a binary verbal category that indicates whether or not the epistemic source possesses privileged access to the knowledge contained in a proposition.⁷ With regard to the languages described in the present study, we may more specifically define privileged access as “direct access to the intentional instigation of an event”. Accordingly, conjunct markers indicate that the epistemic source is the wilfull instigator of an event, whereas disjunct markers indicate that this is not the case. Admittedly, this definition represents a slight simplification in the case of Bunan (cf. Widmer forthcoming). However, as this article is a diachronic rather than a descriptive study, the peculiarities of the Bunan conjunct-disjunct system will not be addressed in full detail.

3 Evidence for hybrid verb systems 3.1 Dolakha Newar 3.1.1 The verbal system of Dolakha Newar Dolakha Newar is a Tibeto-Burman language spoken by about 5,000 people in the Janakpur Zone of Northeastern Nepal. The most important contact language is Nepali, which serves as a national lingua franca and is spoken as a second language by the vast majority of Dolakha speakers. However, it is likely that Tibetan and Tibeto-Burman languages of Eastern Nepal also played an important role as

6 As a matter of fact, one should also add the “reported addressee in reported speech contexts” as a fourth subcategory (cf. Widmer forthcoming). 7 Note that this definition entails that conjunct-disjunct marking is distinct from the verbal category of evidentiality, which specifies the source rather than the quality of one’s knowledge (cf. Aikhenvald 2004).

The transformation of verb agreement into epistemic marking   

   57

regional languages of communication in historical times (Genetti 2007: 17–22). Dolakha Newar is closely related to the Newar dialects spoken in the Kathmandu valley. The wider genetic affiliation of the Newaric subgroup is unclear (Genetti 2007: 3–5). According to Genetti (2007: 308), Dolakha Newar exhibits a full-fledged verb agreement system with verbs being inflected for person, honorific status, and number. A complete verb paradigm is given in the table below: Table 1: Present tense paradigm of hat- ‘to say’ (Genetti 2007: 161) Singular

Plural

1

hat-a-gi

hat-a-gu

2

hat-a-n

hat-a-min

2HON

hat-a-gu

hat-a-gu

3

hat-a-i

hat-a-hin

As can be seen, the present-tense stem hat-a- (consisting of the root hat- and the present tense marker -a-) takes different agreement morphemes depending on person and number of the subject. Apart from a syncretism between first person plural and second person honorific singular / plural, the individual forms are distinct. Dolakha Newar thus seems to exhibit a verb agreement system that looks trivial from a typological point of view. However, Genetti has described functional properties that clearly set the Dolakha system apart from agreement systems of other languages. A Dolakha verb allows for “disagreement in person” (Genetti 2007: 172–174), that is to say, first and second person subjects are able to trigger third person morphology on a verb. In such cases, the person distinction is no longer used to encode syntactic agreement between a verb and its subject. Rather, agreement morphology is exploited to express an epistemic distinction. This is illustrated by the following example: (1)

ji=ŋ sir-eu. ji chana nāpa tuŋ 1SG=EXT die-3SG.FUT 1SG 2SG.GEN together FOC ‘I will also die. I will die with you.’ (Genetti 2007: 172)

sir-i. die-1SG.FUT

The example above is taken from a traditional Dolakha story. In the relevant passage, the female protagonist is grieving about the imminent death of her friend, a magical goat. As the girl does not want to live on without her companion, she desperately utters the two sentences, in which the verb sir- ‘to die’ is

58   

   Manuel Widmer

attested twice together with the subject pronoun ji ‘1SG’. Interestingly, the verb sir- occurs in two different forms, taking the third person future suffix -eu in the first sentence, but the first person future suffix -i in the second. According to Genetti (1994: 107, 2007: 173), these two verb forms express a difference in terms of volitionality. The verb form sir-eu (showing syntactic disagreement) portrays the girl’s death as an inevitable consequence of the death of her friend. The verb form sir-i (showing syntactic agreement), on the other hand, describes the girl’s death as an intentional act and thus emphasizes her strong wish to die together with her companion. The use of third person agreement markers has also been documented in combination with second person subjects. In such cases, second person agreement markers express intentional actions, while third person agreement markers refer to non-intentional actions, as illustrated by the following example sentences: (2)

chi tul-eu. 2SG fall-3SG.FUT ‘You will fall.’ (Genetti 2007: 174)

(3)

chi tul-ina. 2SG fall-2SG.FUT ‘You will fall intentionally (e.g. as we have planned).’ (Genetti 2007: 174)

The verbal category of person thus seems to have acquired two distinct grammatical functions in Dolakha Newar. The category can either be used to encode syntactic agreement between a subject and a verb, or it can be used to distinguish between intentional and non-intentional actions with first and second person subjects. Accordingly, the verb agreement system of Dolakha Newar exhibits functional traits that are reminiscent of conjunct-disjunct marking. These similarities are not confined to the distinction of intentional and non-intentional actions, but also extend to reported speech constructions. Here, syntactic mismatches can be exploited to express coreference between the subject of the matrix clause and the subject of the complement clause, a phenomenon that has been referred to as “logophoric agreement” by DeLancey (1992: 58). (4)

rekā=n jin rājā=ta nāplat-ki Reka=ERG 1SG.ERG king=DAT meet-1SG.PST ‘Rekai said “Ii met the king”.’ (Genetti 1994: 109)

haŋ-an hat-cu. say-PART say-3SG.PST

The transformation of verb agreement into epistemic marking   

(5)

   59

rekā=n jin rājā=ta nāplat-cu haŋ-an hat-cu. Reka=ERG 1SG.ERG king=DAT meet-3SG.PST say-PART say-3SG.PST ‘Rekai said that Ij met the king.’ (Genetti 1994: 109)

The use of first and third person agreement markers in the examples above parallels the use of verbal suffixes in languages that encode a conjunct-disjunct opposition. It also becomes apparent that there is a clear pattern of correspondence between the two different functions of Dolakha agreement morphology: First / second person suffixes are associated with conjunct-like function, whereas third person suffixes are associated with disjunct-like function. Accordingly, it seems that Dolakha Newar exhibits a hybrid verb system that comprises the two functional domains of syntactic agreement and epistemic marking. However, Genetti (2007: 174) states that agreement morphology is only rarely used to express epistemic distinctions: While it is possible to manipulate the agreement system in this way, it is not at all common, and it has certainly not grammaticalized in the sense of becoming a regular or required feature of the grammar of the language. One can certainly use first-person morphology with non-control verbs without any added implication of heightened volition […]. It is the use of the third-person morphology with first-person subjects which is marked, and which emphasizes the lack of volition.

In summary, the Dolakha system is essentially a verb agreement system that exhibits marginal traits of a conjunct-disjunct system. However, the question arises as to which system is more conservative, especially in the light of the fact that the Newar dialects of the Kathmandu valley exhibit conjunct-disjunct systems without any traces of verb agreement (Hale 1980). This issue will be addressed in the following section.

3.1.2 The historical status of verb agreement in Dolakha Newar The sharp contrast between the verb agreement system of Dolakha Newar and the epistemic system of the Newar varieties spoken in the Kathmandu valley was already noticed by Genetti (1988: 173–174). Since then, several scholars (van Driem 1992: 29–33; DeLancey 1992: 53, 56–57; Genetti 1994: 130–136) have discussed the question of which of the two systems is older, and all of them have argued in favor of the reconstruction of a verb agreement system for Proto-Newar. First, there are structural arguments for this assumption. As Genetti (1994: 131) points out, the mere complexity of the Dolakha agreement system – verbs

60   

   Manuel Widmer

are inflected for person and number throughout all tenses and moods – implies that it is an archaism and not an innovation. If the agreement system had only developed recently, this would presuppose that the verbal system had undergone fundamental changes affecting every single aspect of it and leaving only minor traces of conjunct-disjunct marking. Second, there is philological evidence for this claim. Jørgensen (1941: 60) describes an interesting pattern of variation in the usage of past tense endings in early Classical Newar.⁸ According to him, early Classical Newar distinguishes between a past tense suffix -aṃ, which is mostly found in combination with the third person, and a past tense suffix -o, which occurs less frequently and is mostly attested with the first and second person. According to Genetti (1994: 132), this opposition, which is not attested in any modern Newar variety, might represent a reflex of a former person distinction that disappeared during the Classical Newar period. Third, there are comparative reasons for the reconstruction of a verb agreement system for Proto-Newar. Genetti (1994: 132–34) notes that there is no obvious source from which the Dolakha verbal endings could have been grammaticalized. In particular, there is no etymological connection to the personal pronouns, which are considered to be the primary basis for the development of agreement morphology (cf. Givón 1976). Moreover, several Dolakha agreement markers seem to be cognate with verbal endings found in other Tibeto-Burman languages, which suggests that at least some of the verb agreement morphology in Dolakha Newar may be of considerable age. In this context, it is also interesting to note that some of the finite verbal endings in Kathmandu Newar have etymological cognates in Dolakha Newar. However, the corresponding Dolakha forms are not finite verbal endings but nominalizing suffixes, which is why Genetti (1994: 136) infers that at least some of the Kathmandu Newar finite verbal morphology may have developed from infinite forms. In sum, there is compelling evidence that conjunct-disjunct marking in Kathmandu Newar is an innovation and that the verb agreement system of Dolakha Newar is conservative. This, however, suggests that the epistemic use of person suffixes in Dolakha Newar is a recent development and represents an early stage of a functional shift towards epistemic marking. Further evidence for this assumption will be provided in the following section, where I will discuss another instance of a hybrid verb system.

8 Classical Newar was the literary language of the Newar people. The language is attested in sources written between the 14th and 20th century and represents an old stage of the contemporary Newari varieties of the Kathmandu Valley (van Driem 2001: 745–746).

The transformation of verb agreement into epistemic marking   

   61

3.2 Bunan⁹ 3.2.1 The verbal system of Bunan Bunan is spoken by approximately 3,500 speakers in the North Indian state of Himachal Pradesh. The national language Hindi is spoken as a second language throughout the Bunan community, but western Tibetan varieties were the most important contact languages until the second half of the 20th century, which is documented by a large number of Tibetan loanwords in the Bunan lexicon. Bunan is commonly assigned to the subgroup of West Himalayish languages (van Driem 2001: 938; see Widmer forthcoming for a detailed discussion of the genetic affiliation). One hundred years ago, Francke published a first brief account of the Bunan verbal system, which he described with the following words: The three languages of Lahoul [i.e. Bunan, Manchad, and Tinan] have very full systems of conjugation, with terminations for the different persons, singular and plural, whilst the Tibetan verb hardly ever distinguishes persons. Thus ‘I made, thou madest,’ etc., is conjugated in Bunan and Tibetan in the following way: Bunan ligiza ligzana ligza liitsa ligtsani ligtsa

Tibetan chospin chospin chospin chospin chospin chospin

(Francke [1907] 1998: 135–136; italics original)

Apparently, Francke described a verb agreement system based on the categories of person and number for Bunan. This assumption is supported by his detailed morphological analysis of the Bunan verbal system, which was published two years later (Francke 1909). The data presented in this article consists of various tense paradigms all of which are fully inflected for person and number. Accordingly, Francke’s description strongly suggests that Bunan exhibited a verb agreement system at the beginning of the 20th century. When I first conducted fieldwork on the language in 2010, I found that the verbal system of modern Bunan is rather different from the system described 9 The Bunan data presented in this article were gathered by the author during four field trips from 2010 to 2013. I would like to thank the family of Sonam Tashi, Rinchen Zangpo, and Norbu Ram for their hospitality as well as my consultants Tshering Dorje, Tashi Morup, and Sonam Angrup for sharing their knowledge about their mother tongue with me.

62   

   Manuel Widmer

by Francke. To be sure, the language exhibits certain features of a verb agreement system. The dominant verbal categories are, however, clearly epistemic in nature. The present tense subsystem of contemporary Bunan closely resembles the conjunct-disjunct system of Kathmandu Newar (see Hale 1980; Hargreaves 2005), while the past tense subsystem is reminiscent of the complex epistemic systems that have been described for several Tibetan varieties (see among others, DeLancey 1990; Haller 2000; Huber 2005). As a detailed description of the entire verbal system of Bunan lies beyond the scope of this article, I confine myself to the discussion of the present tense system in the following. In the present tense system, conjunct endings are used with the first person in declarative statements and the second person in questions if the verb denotes an intentional action: (6)

gi=niː iŋgi=ki jato epo tsuk-s-ɕ-i 1SG=TOP myself=GEN friend good bring.up-DETR-MID-ACT.PART kjaŋka dʑot-k-ek. always sit-INTR-PRS.CJ.SG ‘As for me, I always stay with my well-educated friends.’ (The Prodigal Son 54)

(7)

han guj dʑot-k-ek. 2[SG] where sit-INTR-PRS.CJ.SG ‘Where are you staying?’ (TD 322.2 [elicited])

Disjunct endings are used in declarative statements with the second and third person. They are also used with the first person if the event / state denoted by the verb is not controllable for the speaker: (8)

han ŋaro tat-ka […] dʑot-k-are. 2[SG] noise prepare-PROG.PART […] sit-INTR-PRS.DJ.SG ‘Making noise […], you sit here.’ (The Lama and the Owl 95)

(9)

tʰe nira dzokna dʑot-k-are. this daytime motionless sit-INTR-PRS.DJ.SG ‘This one sits without moving all day long.’ (The Lama and the Owl 11)

The transformation of verb agreement into epistemic marking   

(10)

   63

gi dzuk=tok=tɕi gjal-k-are. 1SG pain=DAT=ABL recover-INTR-PRS.DJ.SG ‘I am recovering from the sickness.’ (TD 322.4 [elicited])

As might be expected, the conjunct-disjunct opposition can also be exploited to express coreference of “subjects” in contexts of reported speech: (11)

gi ra-k-ek riŋ-k-are. 1SG come-INTR-PRS.CJ.SG tell-INTR-PRS.DJ.SG ‘Hei says “Ii am coming”.’ (TL 1.14 [elicited])

(12)

gi ra-k-are riŋ-k-are. 1SG come-INTR-PRS.DJ.SG tell-INTR-PRS.DJ.SG ‘Hei says that Ij am coming.’ (TL 1.6 [elicited])

The examples considered above imply that Bunan marks a straightforward conjunct-disjunct opposition in the present tense. However, the system is in fact more complex than the data discussed so far suggest, as there is a third present tense ending -ana, which has not been taken into consideration so far. This ending exhibits an extremely limited distribution. In the present tense, it only occurs in interrogative speech acts in combination with second person singular subjects. (13)

han guj dʑot-k-ana. 2[SG] where sit-INTR-PRS.2SG ‘Where are you staying?’ (TD 216.4 [elicited])

In declarative statements that contain a second person singular subject and have present tense reference, the ending -ana cannot be used. In such contexts, only the present disjunct ending -are is allowed to occur. However, the use of -ana is grammatical in declarative statements with future tense reference.¹⁰ This is demonstrated by the following examples.

10 Note that present tense suffixes may have future tense reference in the intransitive and middle conjugation.

64   

   Manuel Widmer

(14)

han dzaŋdzaŋ lik-tɕ-are / *lik-tɕ-ana! 2[SG] insincere.refusal make-TR-PRS.DJ.SG / *make-TR-PRS.2SG ‘You are insincerely refusing (the food that I am offering you)ǃ’ (TD 325.7 [elicited])

(15)

kjumar lis-dʑi ni! han bret-k-ana! roof freeze-CVB.SG EX.NON1SG 2[SG] slip-INTR-PRS.2SG ‘The roof is icyǃ You are going to slipǃ’ (TD 329.1 [elicited])

These strong restrictions on the use of the ending -ana seem rather peculiar. However, the morpheme is remarkable in yet another respect, as it is only used by speakers belonging to the oldest generation of the Bunan community. In my corpus of natural language data, it is only attested in the speech of my oldest consultant (*1936). Most of my young and middle-aged consultants are familiar with the form, but do not actively use it themselves. However, I have also come across young speakers that do not know the ending -ana and, accordingly, consider the sentence given in (13) to be ungrammatical. My young and middle-aged consultants who were familiar with the suffix told me that the sentences given in (7) and (13) essentially have the same meaning, but that (13) sounded old-fashioned and obsolete to them. All of this suggests that the suffix -ana is an archaic second person ending that has only survived in the generatiolect of old speakers. The fact that some young speakers are no longer familiar with the form further indicates that the morpheme will disappear from the spoken language in the near future. The Bunan data considered so far makes clear that there is a considerable difference between Francke’s account of the Bunan verbal system and my own analysis. Francke analyzed Bunan as a language with a verb agreement system. My own data, however, suggests that Bunan possesses a full-fledged conjunct-disjunct system, although there are peculiar second person agreement forms that might be remnants of a former agreement system. This gives rise to the question as to whether it is possible to interrelate Francke’s material with my own data. A comparison of the two data sets indeed reveals an interesting pattern: Francke’s first and third person endings correspond to conjunct and disjunct endings in contemporary Bunan, while Francke’s second person endings correspond to the obsolete second person endings characteristic of the generatiolect of old speakers. Remarkably, the pattern of correspondence emerging from the table above bears striking similarities to the phenomenon of “disagreement in person” described for Dolakha Newar. Recall that in Dolakha Newar first person suffixes can assume a conjunct-like function, whereas third person suffixes can serve a disjunct-like function. The only difference is that in Dolakha Newar second

The transformation of verb agreement into epistemic marking   

   65

Table 2: A comparison of Francke’s (1909: 74) present tense endings with the author’s data Francke’s data*

Personal data

-kyeg

‘-PRS.1SG’

-k-ek

‘-INTR-PRS.CJ.SG’

-kyana

‘-PRS.2SG’

-k-ana

‘-INTR-PRS.2SG’

-kyare

‘-PRS.3SG’

-k-are

‘-INTR-PRS.DJ.SG’

-khye

‘-PRS.1PL’

-k-ʰek

‘-INTR-PRS.CJ.PL’

-khyagni

‘-PRS.2PL’

-k-ʰakni

‘-INTR-PRS.2PL’

-khyag

‘-PRS.3PL’

-k-ʰak

‘-INTR-PRS.DJ.PL’

g

* Francke (1909) used superscript g to transcribe unreleased velar stops, which have allophonic status in Bunan and only occur in syllable-final position. The glide y that occurs in Francke’s endings is the result of a morphophonological process and is thus not represented in my phonological transcription.

person suffixes can also serve a conjunct-like function in combination with second person subjects. This possibility does not exist in Bunan, where second person suffixes are restricted to a small number of specific pragmatic contexts. The obvious parallels between Bunan and Dolakha Newar suggest that the two languages show the effect of the same diachronic process that transforms person agreement into conjunct-disjunct marking. This hypothesis is, however, only plausible if it can be convincingly demonstrated that Francke’s account of the Bunan verbal system was correct and that the conjunct-disjunct system of contemporary Bunan indeed developed from a former person agreement system. This issue will be addressed in the following section.

3.2.2 The historical status of verb agreement in Bunan There is ample evidence that Bunan once exhibited a full-fledged verb agreement system. First of all, this claim is corroborated by comparative research. Several West Himalayish languages were described in the Linguistic Survey of India (Grierson 1903–1928, 3: 427–567), where they were referred to as the “Western subgroup” of the so-called “Complex Pronominalized Languages”. The attribute “pronominalized” reveals that the presence of verb agreement systems was considered to be a characteristic trait of these languages. In an introductory note, Konow explicitly stated that “[i]n all of them we find the same tendency to distinguish the person of the subject by means of a pronominal suffix added to the verb.” Of course, we have to take into account the possibility that Konow’s state-

66   

   Manuel Widmer

ment may have been based on erroneous analyses of the verbal systems of these languages. However, recent research suggests that verb agreement is indeed a common feature within West Himalayish. Verb agreement systems have been described for the West Himalayish languages Rongpo (Zoller 1983: 66–71), Kinnauri (Takahashi 2001: 109–112), Byangsi (Sharma 2007: 55–63), Darma (Willis 2007: 346–359), and Shumcho (Huber 2013: 228‒239), and I was able to confirm the presence of similar systems in Manchad (Francke 1909: 78–86) and Tinan (Francke 1909: 78–97) by means of personal fieldwork. Furthermore, comparative studies by Saxena (1997) and Takahashi (2009) suggest that some of the agreement markers found in West Himalayish languages are cognate. The Bunan conjunct ending -ek, for example, is clearly related to the first person singular suffixes found in other West Himalayish languages. This is demonstrated by the following table. Table 3: First person endings in selected West Himalayish languages Language

First person singular suffix

Source

Manchad

-g / -ga

(Sharma 1996: 86‒87)

Tinan

-g

(Sharma 1996: 90‒91)

Kinnauri

-k

(Takahashi 2001: 109)

Shumcho

-kʰ

(Huber 2013: 229)

Rongpo

-ki

(Zoller 1983: 68)

The Bunan second person ending -ana likewise possesses obvious cognates in other West Himalayish languages, as the table below illustrates. Table 4: Second person endings in selected West Himalayish languages Language

Second person singular suffix

Source

Manchad

-n / -na

(Sharma 1996: 86‒87)

Tinan

-n

(Sharma 1996: 90‒91)

Kinnauri

-n

(Takahashi 2001: 109) *

Shumcho

-na

(Huber 2013: 229)

Rongpo

-n

(Zoller 1983: 68)

* Note that the Shumcho second person singular ending -na is honorific, the non-honorific ending being -n.

The transformation of verb agreement into epistemic marking   

   67

Based on morphological correspondences such as the ones given above, Saxena (1997: 88) reconstructs a first person singular suffix *-k / *-ga as well as a second person singular suffix *-na for Proto-West Himalayish. The latter of the two morphemes has clear cognates in other branches of Tibeto-Burman and can be traced back to a Proto-Tibeto-Burman second person singular suffix *-na (cf. van Driem 1993: 321). The first person singular suffix *-k is more difficult to relate to agreement morphology in other Tibeto-Burman subgroups. Most probably, this morpheme represents a West Himalayish innovation, although it might be connected to the Proto-Tibeto-Burman first person plural suffix *-k (cf. van Driem 1993: 325). In any case, this brief comparative survey demonstrates that the epistemic morphology of Bunan is cognate with agreement suffixes in closely related West Himalayish languages and more distantly related Tibeto-Burman languages. This strongly suggests that the epistemic system of contemporary Bunan developed from a former verb agreement system. In addition, language-internal evidence supports that assumption, as instances of syntactic agreement can still be found in contemporary Bunan. The case of person agreement has already been illustrated above based on the present tense ending -ana, which expresses syntactic agreement with a second person singular subject in questions and in declarative statements with future tense reference. However, the present / future tense is not the only domain of the Bunan verbal system where vestiges of person agreement have been retained. Obsolete first and second person agreement morphemes can be found in other parts of the verbal system. Moreover, person agreement has also been retained in certain copulas. However, a detailed discussion of these person agreement phenomena lies beyond the scope of this article (see Widmer forthcoming for a detailed description). Another obvious instance of syntactic agreement that is still robustly attested in contemporary Bunan is number agreement between a verb and its subject. This is illustrated by the following sentences. (16)

otɕi gi kelaŋ=maŋ el-k-ek. tomorrow 1SG Keylong=ALL go-INTR-PRS.CJ.SG ‘Tomorrow I will go to Keylong.’ (TD 173.1 [elicited])

(17)

eraŋ nuŋ el-k-ʰek. 1PL.INCL there go-INTR-PRS.CJ.PL ‘We will go there.’ (Tulshug Lingpa 117)

68   

   Manuel Widmer

Number agreement is well-established in the Bunan verbal system. The majority of inflectional endings distinguish between a singular and a plural form. However, a glance at Francke’s (1909) data reveals that number distinctions in inflectional endings were still more pervasive at the beginning of the 20th century, which in turn implies that some distinctions have disappeared in the course of the past hundred years. The process is still ongoing in contemporary Bunan, with young speakers generally making fewer number distinctions than old speakers. This gradual loss of number agreement may be one of the diachronic pathways that lead from a verbal system based on syntactic agreement to a system based on epistemic marking. The evidence considered in this section thus strongly suggests that Francke’s (1909) account of the Bunan verbal system was correct and that the language still exhibited a full-fledged agreement system one hundred years ago. In the course of the 20th century, the person distinction was then regularly transformed into a conjunct-disjunct opposition. This functional shift will be discussed in the following section.

4 From syntactic agreement to epistemic marking In the preceding section, the verbal systems of Dolakha Newar and Bunan were briefly described. It was demonstrated that Dolakha Newar exhibits a full-fledged verb agreement system based on the categories of person and number, while Bunan possesses an epistemic system that primarily encodes a straightforward conjunct-disjunct opposition, although there are remnants of number agreement and, to a lesser extent, person agreement. At first sight, the two verbal systems seem to be based on rather different functional principles. However, on closer examination it becomes clear that there may be more similarities between the two systems than initially thought, as both of them provide evidence for a functional transition between the grammatical categories of person and conjunct-disjunct. In Dolakha Newar, person agreement morphology may either be used to express syntactic agreement or epistemic distinctions in terms of volitionality. Accordingly, the language provides us with a synchronic link between the two grammatical categories. In Bunan, there is no such synchronic multifunctionality, but a comparison between Francke’s (1909) material and my own data reveals that the conjunct-disjunct system of contemporary Bunan must have evolved from a former person agreement system. When comparing the functional relations between person agreement and epistemic marking in Dolakha Newar and Bunan, it becomes clear that both lan-

The transformation of verb agreement into epistemic marking   

   69

guages bear witness to the same functional transition: First person indexing is associated with conjunct marking, while third person indexing is associated with disjunct marking. These functional correspondences are depicted in the form of a diachronic process in the table below. Table 5: The diachronic relation between person agreement and conjunct-disjunct marking Person agreement

Conjunct-disjunct marking

first person

→

conjunct

second person

→

∅

third person

→

disjunct

It seems justified to assume that these striking parallels are due to the same diachronic process that transforms the category of person into the category of conjunct-disjunct according to the pattern outlined above. The two languages would then provide evidence for different stages of the process. In Dolakha Newar, the category of person is still clearly prevalent and only rarely exploited to express epistemic differences. This is additionally indicated by the fact that second person agreement morphemes are still fully functional within the agreement system. Hence, the language seems to stand at the very beginning of this diachronic process. In Bunan, on the other hand, the grammatical category of conjunct-disjunct is pervasive, whereas the grammatical category of person has almost vanished. There are still vestiges of old second person endings, but these agreement morphemes do no longer fulfill an essential function in the epistemic dichotomy of the now fully established conjunct-disjunct system and are likely to become entirely obsolete in the near future. This suggests that Bunan has almost completed the functional transition from syntactic agreement to epistemic marking. Of course, these considerations give rise to the question as to what triggers the functional reanalysis of person agreement morphemes as epistemic markers. DeLancey (1992: 56–57) has pointed out that language contact with Tibetan speaking communities is likely to play a crucial role in the development of epistemic verbal systems. Such contact is attested for both Dolakha Newar (Genetti 1994: 128) and Bunan (Jäschke 1865: 94). Accordingly, it is highly probable that the functional transition was at least partially motivated by Tibetan influence. In addition, DeLancey (1992: 58) has noticed a functional parallelism between logophoric agreement and conjunct-disjunct marking. There is indeed evidence that such constructions represent a starting point for the functional reanalysis

70   

   Manuel Widmer

of person agreement as conjunct-disjunct marking (cf. Widmer 2013). However, discussing this aspect would go beyond the scope of this article.

5 Conclusion In this article, I have discussed instances of hybrid verb systems that display both syntactic agreement and epistemic marking. First, I analyzed data from Dolakha Newar, a language that possesses a full-fledged agreement system based on the categories of person and number, but also exploits the category of person to express differences in terms of volitionality. Second, I discussed the case of Bunan, a language with an epistemic verbal system that, however, seems to have developed from a former verb agreement system. A comparison of the two systems revealed that there are clear parallels between Dolakha Newar and Bunan: In both languages, there is evidence for a transformation of an opposition between first person and third person into a conjunct-disjunct system. This transformation is documented as synchronic variation in Dolakha Newar and can be established diachronically in Bunan by comparing old and new grammatical descriptions. In view of this remarkable parallelism, I have argued that the hybrid systems of the two languages are the result of a process of language change that transforms the category of person into the category of conjunct-disjunct. Several additional questions linked to the transformational process have not been addressed in this article due to the lack of space. Instead, I have confined myself to describing the fundamental basics of the process based on the evidence that is currently available. Future research may provide more descriptions of languages that show similar diachronic developments, which will hopefully lead to a more fine-grained picture of the process.

The transformation of verb agreement into epistemic marking   

   71

6 Abbreviations 1

first person

FOC

focus

2

second person

FUT

future tense

3

third person

GEN

genitive

ABL

ablative

INCL

inclusive

ACT.PART

active participle

INTR

intransitive conjugation

ALL

allative

MID

middle conjugation

CJ

conjunct

PART

participle

CVB

converb

PL

plural

DAT

dative

PROG.PART

progressive participle

DETR

detransitivized

PRS

present tense

DJ

disjunct

PST

past tense

ERG

ergative

SG

singular

EX

existential copula

TOP

topic

EXT

extension particle

TR

transitive conjugation

References Aikhenvald, Alexandra. 2004. Evidentiality. Oxford: Oxford University Press. Bickel, Balthasar. 2008. Verb agreement and epistemic marking: A typological journey from the Himalayas to the Caucasus. In Brigitte Huber, Marianne Volkart & Paul Widmer (eds.), Chomolongma, Demawend und Kasbek: Festschrift für Roland Bielmeier zu seinem 65. Geburtstag, 1–14. Halle: International Institute for Tibetan and Buddhist Studies. Creissels, Denis. 2008. Remarks on so-called “conjunct/disjunct” systems. Paper presented at the 3rd Syntax of the World’s Languages conference, Free University of Berlin, 25–28 September. www.deniscreissels.fr/public/Creissels-conj.disj.pdf (accessed 22 July 2014). DeLancey, Scott. 1986. Evidentiality and volitionality in Tibetan. In Wallace Chafe & Johanna Nichols (eds.), Evidentiality: The linguistic coding of epistemology (Advances in Discourse Processes 20), 203–213. Norwood: Ablex Publications. DeLancey, Scott. 1990. Ergativity and the cognitive model of event structure in Lhasa Tibetan. Cognitive Linguistics 1(3). 289–321. DeLancey, Scott. 1992. The historical status of the conjunct/disjunct pattern in Tibeto-Burman. Acta Linguistica Hafniensia 26. 39–62. DeLancey, Scott. 2011. Finite structures from clausal nominalization in Tibeto-Burman. In Foong Ha Yap, Karen Grunow-Hårsta & Janick Wrona (eds.), Nominalization in Asian languages:

72   

   Manuel Widmer

Diachronic and typological perspectives (Typological Studies in Language 96), 343–360. Amsterdam: John Benjamins. Driem, George van. 1992. The Newar verb in Tibeto-Burman perspective. Acta Linguistica Hafniensia 26. 23–43. Driem, George van. 1993. The Proto-Tibeto-Burman verbal agreement system. Bulletin of the School of Oriental and African Studies 56(2). 292–334. Driem, George van. 2001. Languages of the Himalayas: An ethnolinguistic handbook of the Greater Himalayan region, 2 vols. (Handbuch der Orientalistik, Abt. 2, Indien 10). Leiden: Brill. Francke, August H. 1909. Tabellen der Pronomina und Verba in den drei Sprachen Lahoul’s: Bunan, Manchad und Tinan. Zeitschrift der Deutschen Morgenländischen Gesellschaft 63. 65–97. Francke, August H. 1998 [1907]. A history of Western Tibet: One of the unknown empires, rev. edn. Delhi: Motilal Banarsidass. Genetti, Carol. 1988. A contrastive study of the Dolakhali and Kathmandu Newari dialects. Cahiers de Linguistique Asie Orientale 17(2). 161–191. Genetti, Carol. 1994. A descriptive and historical account of the Dolakha Newari dialect (Monumenta Serindica 24). Tokyo: Institute for the Study of Languages and Cultures of Asia and Africa. Genetti, Carol. 2007. A grammar of Dolakha Newar (Mouton Grammar Library 40). Berlin: Mouton de Gruyter. Givón, Talmy. 1976. Topic, pronoun and grammatical agreement. In Charles N. Li (ed.), Subject and topic, 149–188. New York, San Francisco & London: Academic Press. Grierson, George A. (ed.). 1903–1928. Linguistic Survey of India, 11 vols. Calcutta: Superintendent of Government Printing. Hale, Austin. 1980. Person markers: Finite conjunct and disjunct verb forms in Newari. In Stephen A. Wurm (ed.), Papers in South East Asian Linguistics 7 (Pacific Linguistics A 53), 95–106. Canberra: Australian National University. Haller, Felix. 2000. Dialekt und Erzählungen von Shigatse (Beiträge zur tibetischen Erzählforschung 13). Bonn: VGH Wissenschaftsverlag. Hargreaves, David. 1991. The concept of intentional action in the grammar of Kathmandu Newari. Eugene: University of Oregon dissertation. Hargreaves, David. 2003. Kathmandu Newar (Nepāl Bhāśā). In Graham Thurgood & Randy J. LaPolla (eds.), The Sino-Tibetan languages (Routledge Language Family Series). London & New York: Routledge. Hargreaves, David. 2005. Agency and intentional action in Kathmandu Newar. Himalayan Linguistics 5. 1–48. Huber, Brigitte. 2005. The Tibetan dialect of Lende (Kyirong): A grammatical description with historical annotations (Beiträge zur tibetischen Erzählforschung 15). Bonn: VGH Wissenschaftsverlag. Huber, Christian. 2013. Subject and object agreement in Shumcho. In Thomas Owen-Smith & Nathan W. Hill (eds.), Trans-Himalayan linguistics: Historical and descriptive linguistics of the Himalayan area (Trends in Linguistics. Studies and Monographs 266), 221–274. Berlin: Mouton de Gruyter. Jäschke, Heinrich A. 1865. Note on the pronunciation of the Tibetan language. Journal of the Asiatic Society of Bengal 34(1). 91–100.

The transformation of verb agreement into epistemic marking   

   73

Jørgensen, Hans. 1941. A grammar of the classical Newari (Historisk-filologiske Meddelelser 27(3)). Copenhagen: Munksgaard. Saxena, Anju. 1997. Towards a reconstruction of the Proto West Himalayish agreement system. In David Bradley (ed.), Papers in South East Asian linguistics 14 (Pacific Linguistics A 86), 73–94. Canberra: Australian National University. Sharma, Suhnu Ram. 1996. Pronouns and agreement in West Himalayan Tibeto-Burman languages. Indian Linguistics 57. 81–103. Sharma, Suhnu Ram. 2007. Byangsi grammar and vocabulary. Pune: Deccan College Post Graduate and Research Institute. Takahashi, Yoshiharu. 2001. A descriptive study of Kinnauri (Pangi dialect): A preliminary report. In Yasuhiko, Nagano & Randy J. LaPolla (eds.), New research on Zhangzhung and related Himalayan languages (Bon Studies 3 = Senri Ethnological Reports 19), 97–119. Osaka: National Museum of Ethnology. Takahashi, Yoshiharu. 2009. On the verbal affixes in West Himalayan. In Yasuhiko, Nagano (ed.), Issues in Tibeto-Burman historical linguistics (Senri Ethnological Reports 75), 21–49. Osaka: National Museum of Ethnology. Widmer, Manuel. 2013. Semi-direct speech as a source of assertor’s involvement marking. Paper presented at the 19th Himalayan Languages Symposium, Australian National University, September 6–8. Widmer, Manuel. forthcoming. A grammar of Bunan. Bern: University of Bern dissertation. Willis, Christina M. 2007. A descriptive grammar of Darma: An endangered Tibeto-Burman language. Austin: University of Texas dissertation. Zoller, Claus P. 1983. Die Sprache der Rang Pas von Garhwal (Raṅ Pɔ Bhāsa): Grammatik, Texte, Wörterbuch (Neuindische Studien 8). Wiesbaden: Harrassowitz.

Lameen Souag

How to make a comitative preposition agree it-with its external argument: Songhay and the typology of conjunction and agreement Abstract: This article describes two hitherto unreported comitative strategies exemplified in Songhay languages of West Africa – external agreement, and bipartite – and demonstrates their wider applicability. The former strategy provides the first clear-cut example of agreement for a previously unattested target-controller pair. Based on comparative evidence, this article proposes a scenario for how these could have developed from the typologically unremarkable comitative and coordinative strategies reconstructible for protoSonghay, in a process facilitated by contact with Berber. The grammaticalisation chain required to explain this has the unexpected effect of reversing a much better-known one previously claimed to be unidirectional, the development COMITATIVE > NP-AND.

1 Introduction In recent years, the domain of conjunction has received increasing attention from typologists, as illustrated by such overviews as Stassen (2003), Lehmann and Shin (2005), Stolz et al. (2006), Haspelmath (2007), Arkhipov (2009), and Palancar (2012). This work provides a fairly comprehensive typology of the domain. Stassen (2003) points out two principal strategies for the encoding of NP conjunction, which he treats as extreme positions on a continuum: coordinative, in which the two NPs involved receive equal structural rank and are both assigned the same theta-role, with extraction possibilities limited by the Coordinate Structure Constraint, and prototypically forcing non-singular agreement; and comitative, in which the two NPs involved receive unequal structural rank and one is treated as oblique, resulting in a structure exempt from the Coordinate Structure Constraint and prototypically allowing singular agreement. The typology of comitative strategies itself has been further elaborated by Lehmann and Shin (2005:43–54): concomitant predication (with a converb/coverb ‘be with’); adpositional marking (‘with’); case marking (‘-with’); verb derivation (‘with-V’, a type of applicative); and incorporation. The commonest of these, according to Stassen, is adpositional or case marking.

76   

   Lameen Souag

Stassen’s typology, and in particular the distinction he draws between comitatives and coordinatives, implicitly presupposes that, in structures satisfying the prototypical conditions for the comitative strategy, each noun phrase is expressed only once within the conjunction, allowing unambiguous comparison of their relative ranks. The alternative would involve doubling of at least one noun phrase; to the author’s knowledge no example of this phenomenon has been reported in existing surveys of agreement, e.g. Corbett (2006), or of adpositions, e.g. Hagège (2010). However, this article will demonstrate that this alternative is in fact attested: at least three Songhay languages of North and West Africa systematically use comitative strategies – confirmed as such by being exempt from the Coordinate Structure Constraint and taking agreement with only one noun phrase – in which the higher-ranked noun phrase is doubled by a pronominal copy (free or bound agreement) forming a formally coordinative phrase with the lower-ranked noun phrase. A priori, while neither Stassen’s nor Lehman and Shin’s typologies include this possibility, they provide a plausible means for such a development to occur: by a comitative-expressing converb / serial verb retaining agreement. Surprisingly, however, this path turns out not to be a plausible source for the examples of this strategy documented here. Rather, comparative Songhay evidence indicates that these have developed – partly under Berber influence – from an earlier situation, still attested in closely related languages, in which the comitative/coordinative particle was a preposition rather than a verb. This development reverses a better-known grammaticalisation path, COMITATIVE > NP-AND, and contradicts Haspelmath’s (2007:29) claim that “Theoretically, one could imagine the reverse diachronic process, from coordinator to comitative, also giving rise to the same synchronic polysemy, but this never happens.” After briefly setting the family-internal context, this article will open with a discussion of the most extreme case of a doubling strategy – a comitative preposition agreeing with its external argument, as found in Kwarandzyey (Songhay, Algeria)¹. On common assumptions about the emergence of agreement (e.g. van Gelderen 2011), such a phenomenon presupposes a prior stage in which a 1 All Kwarandzyey data derives from the author’s fieldwork, with grateful thanks to his consultants – in particular Smail and Madani Yahiaoui and Mohamed Ayachi – and to the Arts and Humanities Research Council. Thanks are also due to the British Academy and the Centre National de la Recherche Scientifique for funding post-doctoral research, to Catherine Taine-Cheikh for Zenaga examples, and to anonymous reviewers for comments. On the grammar of Kwarandzyey, see Souag (2010a) and the references given there, notably Cancel (1908); for sound shifts, see Souag (2010b). The transcription used here is based on the former; however, the second element of affricates is transcribed superscript, and elided vowels are written with an overstrike line (e.g. a) rather than being omitted from the transcription. Abbreviations in glosses added by the au-

Comitative preposition agreement   

   77

pronoun doubling the higher rank noun phrase was used, which I term the bipartite comitative; re-analysis of existing data for other Songhay languages demonstrates that this stage is indeed attested there, though only in the more northerly varieties. This makes it possible to propose a tentative outline of the grammaticalisation process involved. The languages in question are independently known to have undergone Berber influence, and comparison to Berber then helps elucidate an otherwise improbable early step in this development. Finally, a few less well documented examples of the same two strategies in languages outside the region are discussed.

2 The split reflexes of Songhay nda ‘with/and’ in Kwarandzyey In almost all members of the closely knit Songhay family of West Africa (mainly Mali and Niger), a word nda or da has the following polysemy: 1. if (complementiser with a phrasal complement); 2. and (conjunction linking noun phrases); 3. with (instrumental preposition); 4. with (comitative preposition). Such polysemy is cross-linguistically frequent. Identity of instrumentals and comitatives is a relatively widespread pattern, displayed by about a quarter of all languages (Stolz, Stroh & Urdze 2011), and identity of comitatives and noun phrase conjunctions is found in nearly half of all languages (Stassen 2011). The connection of conditional complementisers to “and”, while less well attested, is confirmed by cases like middle English and (early modern an) ‘if’ < ‘and’ (Kurath 2001). It thus appears that this polysemy reflects a common etymon, and not coincidental convergence from multiple sources. Kwarandzyey, spoken at the oasis of Tabelbala in Algeria a thousand kilometres from the nearest Songhay-speaking town, has undergone intense Berber and Arabic influence and prolonged isolation from other Songhay varieties, resulting in numerous divergences from mainstream Songhay through both innovation and retention. In Kwarandzyey, three of the functions of pan-Songhay nda are reflected by ndza (the shift of d > dz is regular before non-emphatic vowels):

thor follow the Leipzig Glossing Rules, with the addition of Emph ‘emphatic (pronoun)’, ABS. GEN ‘absolute genitive’, INC ‘inceptive’, AntiAgr ‘anti-agreement’.

78   

   Lameen Souag

(1)

‘if’: ndza lħšiš bạ a-ka nə-m-dzuɣ-a if plant exist 3Sg-LOC 2S-IRR-uproot-3Sg ‘If there are weeds in it, you uproot them.’

(2)

‘NP-and’: yu=yu ndza fə̣ṛka=yu camel=Pl and donkey=Pl ‘camels and donkeys’

(3)

‘with (instrumental)’: nə-m-yạḍạ a-ka ndza 2Sg-IRR-step 3Sg-LOC INS ‘you step on it with your foot’

nə-n tsi 2Sg-GEN foot

In certain contexts, ndza has also acquired various spatial usages (‘from’, ‘(passing) by’), not relevant here. However, in contrast to other Songhay languages, ‘with (comitative)’ is expressed not by ndza, but by the preposition indza preceded by an agreement marker: (4)

ʕa-m-ka ʕa-m-gwạ ʕa-indz-ana 1Sg-IRR-come 1Sg-IRR-sit 1Sg-COM-3SgEmph ‘I will come and sit with him.’

In order to understand this unexpected development, it is necessary first to examine the synchronic facts more closely.

3 The morphology and syntax of indza 3.1 The controller of the agreement prefix As the previous example illustrates, the agreement marker prefix of indza does not agree with the object of the preposition; rather, it expresses the other participant in the comitative relation. A comitative relation involves two parties: the accompanied, and the accompanier. In “I went with him”, “I” is the accompanied party, and “him” the accompanier; in “I left you with them”, “you” is the accompanied party, and “them” the accompanier. In English, the accompanied is left unstated

Comitative preposition agreement   

   79

within the prepositional phrase, and is deduced from the rest of the sentence. But in Kwarandzyey, the agreement prefix on indza agrees in number and person with the accompanier. Typically, this coincides with the subject of the clause, e.g.: (5)

xʷəd gga nə-b-yəxdəm n-indza nə-n bạ=yu… when PST 2Sg-IPFV-work 2Sg-COM 2S-GEN friend=Pl ‘When you (sg.) were working with your friends…’

(6)

ana=a ggʷərgʷəy Ø-indza ɣuna 3SgEmph=FOC fight 3-COM that ‘It’s him that fought with that person.’

(7)

iškədda=ɣu, ks y-aʕam-dzyəy y-indz-a child=this, let 1Pl-FUT-talk 1Pl-COM-3SG ‘This little kid, let’s talk with him.’

(8)

gga tsuɣ ndz-a-b-dza ndz-indz-a? PST what 2Pl-PROG-IPFV-do 2Pl-COM-3Sg ‘What did you (pl.) used to do with² them [=locusts, a collective singular in Kwarandzyey]?’

(9)

i-m-ka Ø-indz-a 3Pl-IRR-come 3-COM-3Sg ‘They’ll come with him.’

However, examples can be found where it coincides with the morphological object rather than the subject: (10)

ʕa-kkəs-ni n-indz-a 1Sg-leave-2Sg 2Sg-COM-3Sg ‘I left you with him.’

or even with the indirect object:

2 Despite the ambiguity of the English translation, this is comitative, not instrumental; it was answered with “When locusts came, we would make smoke against them”, i.e. what they would do when locusts were present with them, not what they would do using locusts.

80   

(11)

   Lameen Souag

tsuɣu=a yəṣṛa ni-ši n-indz-ana what=FOC happen you-DAT 2Sg-COM-3SgEmph ‘What happened to you with him/it?’

3.2 The paradigm of indza The full paradigm of indza, illustrated in the examples above, is given in Table 1: Table 1: Paradigm of indza. ʕ(a)-indza “1Sg-with”

y(a)-indza “1Pl-with”

n-indza “2Sg-with”

ndz-indza “2Pl-with”

Ø-indza “3Sg/Pl-with”

These prefixes are quite distinct from the free pronouns listed in Table 2 (although their common origin is still evident): Table 2: Kwarandzyey free pronouns. aɣəy “1Sg”

yayu “1Pl”

ni “2Sg”

ndzyu “2Pl”

ana “3Sg”

ini “3Pl”

They bear a far closer similarity to the subject agreement prefixes used with verbs (Table 3), differing from the latter only in not distinguishing 3rd singular from 3rd plural: Table 3: Kwarandzyey subject agreement. ʕ(a)- “1Sg”

ya- “1Pl”

nə- “2Sg”

ndz- “2Pl”

Ø-/a- “3Sg”

i- “3Pl”

Comitative preposition agreement   

   81

3.3 Establishing indza’s word class Both the use of subject prefixes and their form suggest that indza might be a verb. However, other tests rule this out. In Kwarandzyey, any verb can act as a predicate without any copula or existential verb, and can take mood and negation markers. indza cannot stand alone as a predicate; to form a predicate expressing a comitative relation, bạ “exist” must be used: (12)

*(n-bạ) n-indza-ɣəy *(2Sg-exist) 2Sg-COM-1Sg ‘You *(are)³ with me.’

Nor can indza take mood/aspect/negation markers: (13)

*nə-s-indza-ɣəy *2Sg-NEG-COM-1Sg (ungrammatical for any reading)

(14)

*nə-mm-indza-ɣəy *2Sg-IRR-COM-1Sg (ungrammatical for any reading)

Instead, comitative relations are negated as follows: (15)

nə-s-bạ n-indza-ɣəy 2Sg-NEG-exist 2Sg-COM-1Sg ‘You are not with me.’

(16)

bla-ɣəy without-1Sg ‘without me’ (bla is an Arabic loan)

and irrealis mood is expressed on the Arabic loan verb ikun ‘exist, always be’ (bạ ‘exist’ is defective, taking negation markers but not mood/aspect ones):

3 Throughout this paper, in accordance with common practice, the notation *(X) means that the omission of X results in ungrammaticality, whereas (*X) would mean that the insertion of X results in ungrammaticality.

82   

(17)

   Lameen Souag

nə-mm-ikun n-indza-ɣəy 2Sg-IRR-exist 2Sg-COM-1Sg ‘You will be with me.’

Indza cannot be classed as a noun either. Nominal complements always take a genitive case marking postposition n and precede the noun. Nominal adjuncts of nouns follow the noun, but take a postposition wani/wini. Thus: (18)

dwa gung=wani medicine stomach=ABS.GEN ‘stomach medicine’

(19)

tsạṛfəs *(n) dzəɣʷ=yu truffle GEN uproot=Pl ‘truffle-digging’

(20)

xaləd *(n) yimma Khaled GEN mother ‘Khaled’s mother’

The most striking confirmation of its non-verbal, non-nominal status is afforded by relativisation and WH-question formation. In relativisation and WH-questions, verbs unsurprisingly remain in situ when their objects are extracted – in other words, verbs are never pied-piped – as illustrated by the following examples (in which the expected position for the relevant argument in a declarative main clause is marked with t): (21)

ạṛ=dzi ʕa-ggwa t binuw man=REL 1Sg-see t yesterday ‘the man I saw yesterday’

(22)

tsiruw=dz=i ə-ggʷa-b-sku-ndza t bird=REL=Pl 3Sg-INC-IPFV-be caught-CAUS t ‘the birds he kept catching’

(23)

tsuɣu nə-nɣa t? what? 2Sg-eat t ‘What did you eat?’

Nor, for that matter, are genitive nouns pied-piped in relativisation – they remain in situ, using either a resumptive pronoun or a dative strategy:

Comitative preposition agreement   

(24)

   83

nə-bbəddəl an kə̣kkạbu ləqfər=dzi=si lock=REL=DAT 2Sg-change 3Sg.GEN key ‘the lock whose key you changed’ (lit. ‘the lock to whom you changed its key’)

Adpositions, by contrast, are regularly pied-piped. Pied-piping is obligatory for the case-marking postpositions si and ka, and optional for the instrumental preposition ndza: (25)

yə-ggənga ljaməʕ=[dzi=ka] mosque=[REL=Loc] 1Pl-pray ‘the mosque in which we prayed’

(26)

[tsuɣu=si]=a nə-ddziw-a? [whom?=Dat]=Foc 2Sg-send-3Sg? ‘To whom did you send it?’

(27)

stilu [ndza uɣudzi] əgga ʕa-b-iktəb pen [INS REL] PST 1Sg-IPFV-write ‘the pen [with which] I was writing’

(28)

[ndza tsuɣu] n-bạb-kạ? [INS what?] 2Sg-PROG-hit? ‘With what (=which hand) do you normally hit?’

although ndza is also allowed to appear in situ with a resumptive pronoun: (29)

ndz-a littsin tsəffạ=dzi əgga a-b-qətt knife=REL PST 3Sg-IPFV-cut with-3Sg oranges ‘the knife that you were cutting oranges with’

Just like ndza, indza may be pied-piped or left in situ with a resumptive pronoun, and shows normal agreement marking in either case: Pied-piped: (30)

ʕan bạ=yu [ʕ-indza uɣudzi] əgga ʕa-b-yəxdəm my friend=Pl [1Sg-COM Rel] PST 1Sg-IPFV-work ‘my friends with whom I was working’

84   

   Lameen Souag

(31)

zutsi ɣəy.si [n-indza uɣu] nə-ddər likul binuw bring 1Sg.Dat [2Sg-COM Rel] 2Sg-go school yesterday ‘bring me the one you went to school with yesterday’

(32)

[n-indza tsuɣu]=a nə-dri? [2Sg-COM what?]=Foc 2Sg-go? ‘Who did you go with?’

In situ: (33)

ʕa-nn-a ạṛ=dzi=si əgga ʕ-bạ ʕ-indz-a binuw 1Sg-give-3Sg man=REL=DAT PST 1Sg-exist 1Sg-with-3Sg yesterday ‘I gave it to the man I was with yesterday.’

The former possibility also rules out an analysis of indza as a conjunction under most assumptions, since extraction of the right conjunct would violate the Coordinate Structure Constraint (Ross 1986). Therefore, indza is best regarded as a preposition.

3.4 AGR+indza’s other function: in conjunction The conjunction ndza ‘and’ is distinct from indza not just functionally but also phonetically (by the lack of initial i-), morphologically (by the lack of an obligatory agreement prefix), and syntactically (not being used to mark adjuncts of the verb phrase). Examples of its usage include: (34)

[ayyub ndza lmahdi] i-ggʷərgʷəy ya-si. [Ayoub and Mahdi] 3Pl-fight 1Pl-Dat ‘[Ayoub and Mahdi] fought on us (i.e. fought, inconveniencing us.)’

(35)

ʕa-ggạ [ʕan əbba ndza ʕa-yəmma] 1Sg-find [my father and 1Sg-mother]. ‘I found [my father and my mother].’

However, when (and only when) the first conjunct happens to be a pronoun, the expected form ndza gets replaced by AGR+indza. Contrast the following cases, elicited in pairs to maximise parallelism:

Comitative preposition agreement   

(36)

atsəy smaʕil ndza ħəṃṃad i-ba-nnən Smail and Hammad 3P-PRF-drink tea ‘Smail and Hammad have drunk tea.’

(37)

aɣəy ʕ-indza ħəṃṃad y-a-nnən atsəy 1Sg 1Sg-COM Hammad 1Pl-PRF-drink tea ‘I and Hammad have drunk tea.’

(38)

a-kʷbʷəy smaʕil ndza ħəṃṃad 3Sg-meet Smail and Hammad ‘He met Smail and Hammad.’

(39)

a-kʷbʷəy n-indza ħəṃṃad 3Sg-meet 2Sg-COM Hammad ‘He met you and Hammad.’

   85

In such instances, the subject agreement marker matches in number and person the combination of the two conjuncts, e.g. “I and Hammad” (=we), rather than matching either of the two conjuncts, confirming that we are dealing with conjunction and not with a comitative adjunct. Thus: (40)

ʕabdəlqadər ndza xaləd i-kku Abdelkader and Khaled 3Pl-tall ‘Abdelkader and Khaled are tall.’

(41)

n-indz-a ndzə-kku 2Sg-COM-3Sg 2Pl-tall ‘You and he are tall.’

4 Comparative background 4.1 Family-internal reconstruction Songhay is a fairly close-knit family with no proven relatives, much less any relatives close enough to be relevant to reconstruction. It has two principal branches: Eastern Songhay (with relatively few shared innovations) to which most Songhay speakers and languages belong, and Northwestern Songhay (with many clear shared innovations), itself sharply divided into two distinct subgroups, Northern

86   

   Lameen Souag

Songhay (to which Kwarandzyey belongs) and Western Songhay (Souag 2012). While much descriptive work remains to be done on Songhay, there are already at least three Songhay languages for which sufficiently detailed grammars now exist to allow comparison, and more limited data is available for a wider range. From South to North, these are Tondi Songway Kiini (TSK) around Kikara in southern Mali (Heath 2005), Koyraboro Senni (KS) around Gao in eastern Mali (Heath 1999a) – both Eastern – and Koyra Chiini (KC) around Timbuktu in northeastern Mali (Heath 1999b), in the Western branch. The dialect of Timbuktu may be expected to be particularly relevant to comparison with Kwarandzyey, as the oasis of Tabelbala was historically a stop on the trade route between southern Morocco and Timbuktu. In the following examples, note that serial verbs in all non-Northern varieties are formed with reflexes of the preverbal non-finite particle *ká (in TSK also dí), unattested in combination with the comitative; this rules out a serial verb interpretation of the comitative anywhere in Songhay. In Kikara, the comitative is a plain preposition showing no agreement marking, homophonous with “and” and with the instrumental; Heath states that “The postverbal sequence dá X… can be instrumental… or comitative…” (2005:139), e.g.: (42)

ày góy dá nî: 1Sg work with 2Sg ‘I worked with you.’ (Heath 2005:139)

In relativisation, it is treated like a postposition, e.g. : (43)

hɔ̀rɔ́ á ăy kèré ká dá áy ' már ká kóy mótì Dem it-is 1Sg-Poss friend Rel with 1Sg Impf⁴ join Infin go Mopti ‘This is my friend, with whom I will go to Mopti.’ (Heath 2005:198)

This plain preposition strategy is well attested in other, less fully described Songhay varieties, from both principal Songhay subgroups. It is found in Zarma (Eastern Songhay, Niger):

4  abstract > (nominalizations) reduction: SV > VS

loss: SV/VS

Thus, a recent increase in formal agreement is fed by and gradually replaces an earlier gain in semantic agreement. This means that Behaghel’s hypothesis of an increase in formal agreement at the expense of semantic agreement towards

One plus one make(s) – what?   

   319

New High German can be maintained in part for the structural agreement conflict under consideration. Figure 8 summarizes the major developments discussed in this article in a different way. The relative strength of conceptual agreement (on the left) versus formal agreement (on the right) is indicated by partitioning the rectangle and through the depth of the dark shading. The time axis is represented by the vertical crack. Middle High German and Early New High German start out with a formally based agreement system with partial agreement and singular verbs especially in VS order. This preference for singular verbs has been explained in Section 4.2 by processing restrictions and tendencies of discourse grammar. It can be overridden by highly individuated referents in SV and even VS order. For these, semantic resolution yields plural verb agreement. The successive generalisation of plural verb agreement down the hierarchy of individuation leads to the reanalysis and successive emancipation of a purely morpho-syntactic additive rule of resolution (sg + sg → pl), which becomes increasingly insensitive to syntactic and semantic restrictions. However, in contrast to the old partial agreement rule, this new formal rule is in harmony with semantic agreement for a material part of the nominal lexicon (animate and concrete nouns).

CONCEPTUAL AGREEMENT

FORMAL AGREEMENT MH

PL

with conceptually unified/unbalanced referents

PL

with preceding, individuated referents

EN

SG

HG

hum a

Partial agreement Local: with nearest NP

G/

SG/ Semantic(-pragmatic) resolution

n

CON co n

cre t

SV o

TEX T

e

- G EN

ER A

LISA TIO

rde r rde r

(ab s

NH

G

VS o

N

t ra c

t) n o uns

Global: PL Morphosyntactic resolution

PL (SG)

Figure 8: Development of verb agreement with conjoined noun phrases in German in a nutshell

320   

   Antje Dammel

5 Outlook In this last Section, I take up some general questions discussed with respect to agreement change and relate them to the case study at hand. On the one hand, the change of verb agreement with conjoined noun phrases corroborates determining factors known from synchronic and typological research. These are above all the hierarchy of animacy/individuation and word order asymmetries, but also the notion of thetical sentences and the concept of natural coordination can be supported by cross-linguistic data. The tendency to singular verb agreement in VS order has been linked to parallel restrictions in the incremental processing of New High German percipients. On the other hand, the case study questions the assumption that number resolution must be purely semantic. In the diachrony of German, resolution became less and less sensitive to semantic restrictions, which suggests a grammaticalization path from semantic to formal, morpho-syntactic resolution. If one accepts this idea and the evidence provided (Sections 4.3, 4.4), Behaghel’s hypothesis of increasing formal agreement holds partly true for verb agreement with conjoined noun phrases. In quest of cross-linguistically valid factors determining the outcome of the competition between formal and conceptual agreement choices, the morphological richness of a language has been brought forth as one of the suspects (e.g., Berg 1998; Acuña-Fariña 2012): The more explicit morphological marking of grammatical categories is preserved, the stronger should be the tendency towards formal agreement; if formal marking is heavily reduced, a language should tend more towards semantic agreement. The diachrony of German agreement with conjoined noun phrases qualifies this hypothesis by revealing that explicit formal marking cannot be the only decisive factor. In the case at hand, a major change in verb agreement happens with constantly non-marked singular noun phrases. Of course, one can object here that reasoning based on formal marking should take relational information into account and that morphological number marking of plurals in German indeed has been reinforced since Early New High German; but then we have to assume a more elaborate hypothesis than the one mentioned above, one that includes paradigmatic information.²¹ Another point discussed as a determinant of semantic versus formal agreement is the direction of encoding (Acuña-Fariña 2012): Due to top-down process-

21 Beyond coordination, the hybrid noun OHG wīb / ENHG Weib might be a counterexample to the assumption. Historical varieties of German, albeit having richer morphology than New High German, show more semantic gender agreement in relative and personal pronouns than they do in the modern language (cf. Fleischer 2012).

One plus one make(s) – what?   

   321

ing, the production perspective should tend more to semantic agreement, while – due to bottom-up processing – the perception perspective should rely more on formal agreement. If this were true for the present case, the generalisation of plural agreement should have started out as an optimisation from the production perspective, while we saw in 4.2 that speakers of New High German still have difficulties in perception with the “new” resolution rule in VS order. However, the distinction might be not a dichotomy after all, as both perspectives can combine bottom-up and top-down strategies. I think another distinction could be more interesting with respect to the agreement change under consideration, namely online (incremental) versus offline (memory based) processing. The change could be interpreted as a development from a rule avoiding a local agreement mismatch and being thus apt for local, incremental processing (or: speaking and listening), towards a resolution rule that avoids a global mismatch and helps to recognize constituency, which would be advantageous for global offline processing (or: reading and writing). In the face of these ideas, it would be necessary to analyse spoken corpora of contemporary German in order to find out whether spoken varieties tend more to the Early New High German pattern than written German. Given that the new resolution rule is “unnatural” for non-individualized nouns and for incremental processing in VS order, it would also be a good idea to look at the works of historical German grammarians in order to find out about a possible influence of normative discourse.²² The development of resolution, especially the emergent morpho-syntactic resolution, blends well with other phenomena of syntactic change observable since Early New High German. Relevant phenomena are, e.g., the development of clear-cut subordination marking by generalisation of OV and the diversification of subjunctions, and the general increase in framing constructions, which have become a characteristic typological trait of German morpho-syntax (Ronneberger Sibold, e.g., 1991, 2010). The common denominator with the generalisation of plural verbs with conjoined noun phrases is the tendency to mark the boundaries of syntactic constituents as reliably as possible.

22 The same holds for the other usual suspect, translational influence from Latin: although the latter could be excluded for the historical agreement pattern, it should also be tested for the spread of the plural resolution rule by comparing translational and original Early New High German texts.

322   

   Antje Dammel

References Primary Texts and Software BibleWorks 7. Software for biblical exegesis & research. Galmy = Georg Wickram, Sämtliche Werke, vol. I: Ritter Galmy, ed. by Hans-Gerd Rohloff (Ausgaben deutscher Literatur des 15. bis 18. Jahrhunderts 10). Berlin: De Gruyter 1967 http://www.hs-augsburg.de/~harsch/germanica/Chronologie/16Jh/Wickram/wic_ga00. html (accessed 31 May 2012). Faustbuch = Historia vnd Geschicht Doctor Johannis Faustj. Herzog-August-Bibliothek Wolfenbüttel, Undatierte Handschrift des 16. Jahrhunderts. Digitale Ausgabe Ulrich Harsch 1999 auf der Grundlage der digitalen Ausgabe der Akademie der Wissenschaften in Göttingen. http://www.hs-augsburg.de/~harsch/germanica/Chronologie/16Jh/Faustus/ fau_intr.html (accessed 31 May 2012). Fehd = Götz von Berlichingen. Mein Fehd und Handlungen, ed. by Helgard Ulmschneider, (Forschungen aus Württembergisch-Franken 17), Sigmaringen: J. Thorbecke Verlag 1981 [Text according to the Rossacher manuscript (before 1567); searched with TITUS, accessed 14 April 2013] Luther's Bible 1545 = Martin Luther. Biblia: das ist: die gantze heilige Schrifft: Deudsch. Wittenberg 1545. Electronic version: Die Luther-Bibel: Originalfassung 1545 und revidierte Fassung 1912 (Digitale Bibliothek 29), Berlin: Directmedia Publishing, 2000. Moscovia = Sigismund von Herberstein: Moscovia der Hauptstat in Reissen […] Wien 1557. Electronic version transcribed by Frank Kämpfer: http://www.hs-augsburg.de/~harsch/ germanica/Chronologie/16Jh/Sigismund/sig_m000.html (accessed 31 May 2012). Nachbarn = Georg Wickram, Sämtliche Werke, vol. IV: Von guten und bösen Nachbarn, ed. by Hans-Gerd Rohloff, Berlin: De Gruyter 1967 (Ausgaben deutscher Literatur des 15. bis 18. Jahrhunderts). Electronic version: http://www.hs-augsburg.de/~harsch/germanica/ Chronologie/16Jh/Wickram/wic_na00.html (accessed 31 May 2012). Prosalancelot I/ Lancelot und Ginover I, according to the Heidelberger Handschrift Cod. Pal. Germ. 147 ed. by Reinhold Kluge, supplemented by Ms. allem. 8017-8020, Bibliothèque de l’Arsenal Paris, ed. by Hans-Hugo Steinhoff, Frankfurt/Main: Deutscher Klassiker Verlag, 1995, Stuttgart: Reclam 1993, pp. 1–385. [searched using TITUS based on the Heidelberger Codex, March 2013] Tatian. Lateinisch und altdeutsch mit ausführlichem Glossar, ed. by Eduard Sievers. 2nd revised ed. 1892. Reprint Paderborn: Schöningh 1966 (Bibliothek der ältesten deutschen LiteraturDenkmäler, V), S. 3-292. [searched using TITUS, April 2013] TITUS = Thesaurus indogermanischer Text- und Sprachmaterialien. http://titus.uni-frankfurt. de/indexd.htm (accessed March/April 2013).

One plus one make(s) – what?   

   323

Other References Acuña-Fariña, Juan Carlos. 2012. Agreement, attraction and architectural opportunism. Journal of Linguistics 48. 257–295. Aoun, Joseph, Elabbas Benmamoun & Dominique Sportiche. 1994. Agreement, word order, and conjunction in some varieties of Arabic. Linguistic Inquiry 25. 195–220. Behaghel, Otto. 1928. Deutsche Syntax. Eine geschichtliche Darstellung, vol. III. Heidelberg: Winter. Berg, Thomas. 1998. The resolution of number conflicts in English and German agreement patterns. Linguistics 36. 41–70. Birkenes, Magnus Breder & Florian Sommer. In press. The agreement of collective nouns in the history of Classical Greek and German. In Chiara Gianollo, Agnes Jäger & Doris Penka (eds.), Language Change at the Syntax-Semantics Interface. Berlin/New York: De Gruyter Mouton (Trends in Linguistics. Studies and Monographs 278). Branigan, Holly P., Martin J. Pickering & Mikihiro Tanaka. 2008. Contributions of animacy to grammatical function assignment and word order during production. Lingua 118. 172–189. Burkhardt, Petra, Gisbert Fanselow & Matthias Schlesewsky. 2007. Effects of (in)transitivity on structure building and agreement. Brain Research 1163. 100–110. Corbett, Greville. 1983. Hierarchies, targets, controllers. Agreement patterns in Slavic. London: Croom Helm. Corbett, Greville G. 2000. Number. Cambridge: Cambridge University Press. Corbett, Greville. 2006. Agreement. Cambridge: Cambridge University Press. Dal, Ingrid. 1966. Kurze deutsche Syntax. Auf historischer Grundlage. 3. edn. Tübingen: Niemeyer. Dalrymple, Mary & Ronald M. Kaplan. 2000. Feature indeterminacy and feature resolution. Language 76. 759–798. Demske, Ulrike. 2001. Merkmale und Relationen. Berlin & New York: De Gruyter. Dittmer, Arne & Ernst Dittmer. 1998. Studien zur Wortstellung – Satzgliedstellung in der ahd. Tatianübersetzung. Göttingen: Vandenhoeck & Ruprecht. Donhauser, Karin & Svetlana Petrova. 2009. Die Rolle des Adverbs tho bei der Generalisierung von Verbzweit im Deutschen. In Monika Dannerer, Peter Mauser, Hannes Scheutz & Andreas E. Weiss (eds.), Gesprochen – geschrieben – gedichtet. Variation und Transformation von Sprache (Philologische Studien und Quellen 218), 11–24. Berlin: Erich Schmidt. Doron, Edit. 2005. VSO and left-conjunct agreement: Biblical Hebrew vs. Modern Hebrew. In Kiss, Katalyn È. (ed.), Universal Grammar in the Reconstruction of Ancient Languages, 239–264. Berlin: Mouton de Gruyter. http://pluto.huji.ac.il/~edit/edit/doronleft-conjunct-agreement.pdf (accessed 9 July 2012). Duden-Grammatik. 2009 = Duden. Die Grammatik (Duden 4), 8. revised edn. Mannheim: Dudenverlag. Duden. Richtiges und gutes Deutsch. 2011. Das Wörterbuch der sprachlichen Zweifelsfälle Duden 9), 7. edn. Mannheim: Dudenverlag. Ebert, Robert Peter. 1978. Historische Syntax des Deutschen. Stuttgart: Metzler. Enger, Hans-Olav & Tore Nesset. 2011. Constraints on diachronic development: the animacy hierarchy and the relevance constraint. Sprachtypologie und Universalienforschung (STUF) 3.193–212.

324   

   Antje Dammel

Findreng, Ådne. 1976. Zur Kongruenz in Person und Numerus zwischen Subjekt und finitem Verb im modernen Deutsch. Oslo: Universitetsforlaget. Fleischer, Jürg. 2011. Poetische Texte (ch. 3), Entwicklungen im Bereich Kongruenz (ch. 7). In Jürg Fleischer & Oliver Schallert, Historische Syntax des Deutschen. Eine Einführung, 49–58, 103–120. Tübingen: Narr. Fleischer, Jürg. 2012. Grammatische und semantische Kongruenz in der Geschichte des Deutschen: eine diachrone Studie zu den Kongruenzformen von ahd. wīb, nhd. Weib. Beiträge zur Geschichte der deutschen Sprache und Literatur 134.163–203. Hinterhölzl, Roland, Svetlana Petrova, & Michael Solf 2005. Diskurspragmatische Faktoren für Topikalität und Verbstellung in der ahd. Tatianübersetzung (9. Jh.). Interdisciplinary Studies on Information Structure 03.143–182. Jaeger, Christoph. 1992. Probleme der syntaktischen Kongruenz. Theorie- und Normvergleich im Deutschen. Tübingen: Niemeyer. Johannessen, Janne Bondi. 1997. Coordination. New York, Oxford: Oxford University Press. Kathol, Andreas. 1999. Agreement and the syntax-morphology interface in HPSG. In Robert D. Levine & Georgia M. Green (eds.), Studies in contemporary phrase structure grammar, 223–274. Cambridge: Cambridge University Press. Kazana, Despina. 2011. Agreement in Modern Greek coordinate noun phrases. University of Essex dissertation. http://courses.essex.ac.uk/lg/lg617/theses/Despina_Kazana_Thesis. pdf (accessed 10 September 2012). King, Tracy Holloway & Mary Dalrymple. 2004. Determiner agreement and noun conjunction. Journal of Linguistics 40(1).69–104. Klein, Wolf Peter. 2004. Koordination als Komplikation. Über eine strukturelle Ursache für die Entstehung syntaktischer Zweifelsfälle. Deutsche Sprache 32.357–375. Levi, Jaakov. 1987. Die Inkongruenz im biblischen Hebräisch. Wiesbaden: Harrassowitz. Marten, Lutz. 2005. The dynamics of agreement and conjunction. Lingua 115.527–547. Önnerfors, Olaf. 1997. Verb-erst-Deklarativsätze. Grammatik und Pragmatik. Stockholm: Almqvist & Wiksell. Paul, Hermann. [1919] 1968. Deutsche Grammatik, vol. III, part IV: Syntax (Erste Hälfte). Tübingen: Niemeyer. Plank, Frans. 1991. Ellipsis and inflection of determiners and modifiers in coordinate NPs. EUROTYP Working Papers. Technical Reports 11. http://ling.uni-konstanz.de/pages/home/ plank/for_download/publications/78_Plank_1991.pdf (accessed 08 April 2013). Prell, Heinz-Peter. 2000. Die Stellung des attributiven Genitivs im Mittelhochdeutschen. Zur Notwendigkeit einer Syntax mittelhochdeutscher Prosa. Beiträge zur Geschichte der deutschen Sprache und Literatur 122.23–39. Reiten, Håvard. 1964. Über die Kongruenz im Numerus im Mittelhochdeutschen. Oslo. Ronneberger-Sibold, Elke. 1991. Funktionale Betrachtungen zu Diskontinuität und Klammerbildung im Deutschen. In Norbert Boretzky, Werner Enninger, Benedikt Jeßing & Thomas Stolz (eds.), Sprachwandel und seine Prinzipien. Beiträge zum 8. Bochum-Essener Kolloquium über Sprachwandel und seine Prinzipien, 206–236. Bochum: Brockmeyer. Ronneberger-Sibold, Elke. 2010. Die deutsche Nominalklammer. Geschichte, Funktion, typologische Bewertung. In Ziegler, Arne (ed.), Historische Textgrammatik und historische Syntax des Deutschen. Traditionen, Innovationen, Perspektiven, vol. 1: Diachronie, Althochdeutsch, Mittelhochdeutsch, 85–120. Berlin & New York: De Gruyter. Sadler, Louisa. 2003. Coordination and asymmetric agreement in Welsh. In Miriam Butt & Tracy Holloway King (eds.), Nominals: inside and out, 85–118. Stanford: CSLI Publications.

One plus one make(s) – what?   

   325

85–118. http://privatewww.essex.ac.uk/~louisa/newpapers/nomvolcoord.pdf (accessed 07 September 2012) Sasse, Hans-Jürgen. 1995. Theticity and VS order. A case study. Sprachtypologie und Universalienforschung 48(1–2).3–31. Schrodt, Richard. 2005. Kongruenzprobleme im Numerus bei Subjekt und Prädikat: Die Termqualität geht vor. In Ludwig M. Eichinger & Werner Kallmeyer (eds.), Standardvariation. Wie viel Variation verträgt die deutsche Sprache? 231–246. Berlin & New York: De Gruyter. Wälchli, Bernhard. 2007. Co-compounds and natural coordination. Oxford: Oxford University Press. Wechsler, Stephen & Larisa Zlatić. 2003. The many faces of agreement. Stanford: CSLI Publications. Wegera, Klaus-Peter. 2000. Grundlagenprobleme einer mittelhochdeutschen Grammatik. In Besch, Werner, Anne Betten, Oskar Reichmann & Stefan Sonderegger (eds.), Sprachgeschichte. Ein Handbuch zur Geschichte der deutschen Sprache und ihrer Erforschung. 2nd ed., 1304–1320. Berlin & New York: De Gruyter. Wehr, Barbara. 1984. Diskurs-Strategien im Romanischen. Ein Beitrag zur romanischen Syntax. Tübingen: Narr.

Cyril Brosch

Agreement patterns of coordinations in Hittite Abstract: Based on new in-depth philological studies on Hittite, an ancient IndoEuropean language, the paper gives an overview over the number, gender and person features shown by targets having conjoined noun and pronoun phrases as controllers. Moreover, some special cases and linguistic pecularities, which otherwise would blurr the clear rule-based opposition of two agreement strategies, are discussed. It turns out that the most common pattern for the domain of the clause and the text is agreement with the nearest controller, with only animate controllers triggering resolution. Except for a few special cases, one seemingly always finds plural resolution in the domain of the noun phrase, however, which could mean that Hittite violates Corbett’s agreement hierarchy, showing semantic agreement at its left end the while syntactic in the other parts.

1 Introduction Until recently, for Hittite there has been no widely accessible research on agreement in general or on coordinations specifically.¹ There were some remarks in grammars (see Hoffner & Melchert 2008: 235‒239) or other types of publications (see Patri 2007, esp. 54‒55), but new systematic philological studies based on our advanced understanding of and larger access to the Hittite texts were not conducted until 2012/13 – independently, but by and large with similar results – by H. Craig Melchert (see Melchert forthcoming, cf. also Rizza 2011 & 2012) and the author of the present paper (see Brosch 2014a and 2014b). This article aims to present these new findings from a typological point of view. As it focuses on the agreement patterns, the discussion of problematic cases, exact numbers of attestations, and philological details are as much as possible confined to the mentioned 2014 papers.

1 There is an excellent pioneer PhD thesis covering all the topics of agreement in Hittite (Drohla 1934/1949), which was however never published and got known to me only after finishing my own research. Remarkably Drohla, basing his work on a much smaller corpus and lacking the nowadays available philological resources in many important points obtained results according with mine and Melchert’s ones. Divergences between Drohla and the findings presented in this paper are presented in detail in Brosch (2014a and 2014b).

328   

   Cyril Brosch

Some additional information on the Hittite language and culture will be given in section 1.1. For the same purpose all examples are glossed.² Section 1.2 contains a short overview of the theoretical background and the most common terms of agreement. Section 2 gives an overview of the corpus the research is based on, with some important information on the pecularities of a Cuneiformwritten language. Section 3 presents a collection of examples for coordinations in Hittite, separated into common (S. 3.1) and special cases (S. 3.2). Section 4 gives an account of the agreement patterns, both of the common and of the special cases found in section 3, according to the three main agreement features number, person, and gender. A summary of the results and questions open to further research are given in the concluding section 5.

1.1 The Hittite language Hittite is the most widely attested language of the Anatolian branch of the IndoEuropean language family and was spoken in central Anatolia in the second millenium B.C. In the ruins of the Hittite capital, Hattusa (200 km east of modern Ankara), and of some other Bronze age cities over 30,000 clay tablet fragments inscribed with a variant of the Mesopotamian Cuneiform Script (cf. S.  2) have been excavated since 1906. They convey in majority religious texts (rituals, myths, omina), instructions, political treaties, and annals and were written down between ca. 1530 and 1180 B.C. The extant corpus both in paleographic and in linguistic details shows a clear distinction between four diachronic states, labeled Old, Middle, New and Late New Hittite.³ After the end of the Hittite Empire, which also marks the end of the Cuneiform tradition in Anatolia, there are no further traces of the Hittite language, while the Hittite culture was continued to a certain degree by population groups speaking Luwian, a closely related language. Hittite⁴ is a fusional accusative language with dominant SOV word order. Hittite nouns can inflect in up to seven cases in singular and plural. Different from most other Indo-European languages, which display a three-gender system, the Anatolian languages had not developed a feminine gender, distinguishing 2 Additional abbreviations: c =common gender, cnj = conjunction, d/l = dative-locative, dir = directive case, lp = local particle, mp = mediopassive, n/a = nominative-accusative, spec = specification. 3 All attestations in this paper are dated, indicating the linguistic stage with OH, MH, NH, and LNH, and the script paleography with OS, MS etc., as far as they don‘t coincide. 4 For a modern synchronic description s. Hoffner & Melchert (2008); the two dictionary projects (Hethitisches Wörterbuch, Munich, and see Chicago Hittite Dictionary, Chicago) have not yet fully covered all of the Hittite lexicon.

Agreement patterns of coordinations in Hittite   

   329

only a common and a neutral gender (distinct only in the nominative and accusative case); natural sex is expressed lexically (cf. Hoffner & Melchert 2008: 64‒66). The common gender nouns can be animate and inanimate, the latter being more flexible in the assignment of case roles than the former. The Hittite verb distinguishes three persons, two number categories, two moods, two diatheses and two tenses in two conjugations. A distinctive property of Hittite is the particle chain, which introduces virtually every sentence and can contain clitics marking citations, pronominal subjects or objects, the reflexive, and local configurations. Another pecularity, which is of importance for the research on agreement, is the – in comparison with e.g. Greek or English – unusual structure of relative clauses (see Hoffner & Melchert 2008: 423‒426): In Hittite the relative clause in most cases precedes the main clause and is syntactically not embedded into the latter. It contains the head noun, which is later resumed in the main clause by a pronoun, and the relative adjective kui- agreeing in number and gender with its head noun (and having the same case assignment). So a structure like “[main clause head noun:c[ase]n[umber]g[ender]] [relative:ng relative clause] [main clause cd.]” in e.g. Latin looks like “[relative clause relative:cng head noun:cng] [anaphoric pronoun:ng main clause]” in Hittite (see exx. (2) and (31) below).

1.2 Agreement, especially in coordinations The present paper follows the theoretical framework of agreement outlined in Corbett (2006), according to which agreement is the systematic covariance of two forms, with the formal properties of the so called target (marked bold type in the examples) triggered by properties of the so called controller (underlined). The covariance of the targets can depend on the specific agreement features they can render (e.g. gender, number, and person⁵), and on their syntactical relation to the controller. The four possible canonical positions –appearing in the domains of the noun phrase, the clause, or the text– make up the agreement hierarchy (see Corbett 2006: 206‒237, esp. 207):

5 These three features are the most common and the only ones of importance for Hittite (or other Indo-European languages). NB that case in this framework is not an agreement feature (see Corbett 2006: 133‒135 for a justification) and is therefore not explored here.

330   

   Cyril Brosch

attribute > predicate > relative pronoun⁶ > personal pronoun [i.e. phoric elements] (> predicative) This hierarchy becomes important with regard to the two basic kinds of agreement: In syntactic (also called formal) agreement the form of the target depends on formal/lexical properties of the controller, while in semantic agreement the targets show feature values which are triggered by meaning components of the controller, e.g. inherent plurality of collectives (as in the famous example the committee have decided).⁷ Of course the features triggering syntactic agreement need not be marked overtly on the controller, e.g. gender as a lexical property of nouns often has no morphemic realisation. It is claimed that the more to the right a target appears on the agreement hierarchy, the greater is the probability that the expected syntactic agreement is overridden by semantic agreement (leading to a so called mismatch), and if a language at one position in the hierarchy allows a switch from syntactic to semantic agreement, all positions to the right of it can show semantic agreement, too. Besides the opposition of syntactic vs. semantic agreement, mismatches can appear in coordinations of two or more controllers, most commonly conjoined noun phrases (see Corbett 2006: 168‒170). In cases where the multiple controllers show different agreement feature values (e.g. one with masculine and one with feminine gender) the target usually must either confine itself to agree with only one of the controllers (either the first or the nearest, other positions are not attested cross-linguistically) or it must show resolution (see Corbett 2006: 238‒263), i.e., a form that does not necessarily match either of the controllers and is used in these cases (e.g. neuter target with mixed-gender controllers).

6 NB that in Hittite this position does not exist as a discrete category, as the relative pronoun is syntactically an attributive adjective of the head noun and not separated by a clause boundary, see S. 1.1. The predicate position has a subhierarchy consisting of verb > participle > adjective > noun, see Corbett (2006: 233). 7 As in English, also in Hittite a singular collective noun can trigger plural agreement, cf. KUB 22.70 rev. 51‒52 (NH): UN.MEŠ-tar=pat=kan kuit saknuwantes anda salikisker people:n/a.sg(n)=spec=lp because dirty:n/a.pl.c in(to) approach:prt.3pl.act ‘Because impure people have touched (them), …’

Agreement patterns of coordinations in Hittite   

   331

2 The corpus For the Marburg Agreement Project (cf. the introductory footnote) more than 8100 cases of agreement from all periods and genres of Hittite literacy (cf. fn. 3 and S. 1.1) have been collected, i.e. 8100 different targets, not necessarily 8100 controllers. In this data base one can find 477 valid attestations of conjoined controllers, 271 of which consist of singulars only, 96 of plurals only, 85 of number-mixed coordinations with singular in first and plural in last position, and 25 with plural first and singular last. The majority of the examples come from the New Hittite period. There are also a few dozen attestations which are superficially reminiscent of coordinations, but which actually represent different phenomena (see Brosch 2013a: 37–38). In many cases these are disjunctive coordinations containing an OR-expression, which means that in fact there is only one controller, and accordingly the targets do not show any kind of resolution but behave as if they agreed with a single controller (see ex. (1), the Cuneiform transcription conventions will be clarified on page 332). In a few other cases controllers can occur in two subsequent, parallel sentences and govern a common target (see ex. (2)). As it had not been clear a priori whether or not this kind of construction differs from coordinations, the attestations were analyzed in Brosch (2013a: 34). However, it turned out that they –other than real coordinations– always trigger plural resolution in the target. Neither type of pseudo-coordination will be discussed further in this paper (but cf. the conclusion).⁸ (1)

KBo 6.2 i 54’ (OH) [tak]ku ARAD-as nasma [GÉME]-as if male.slave:nom.sg.c or female.slave:nom.sg.c huwāi run:prs.3sg.act ‘If a male or female slave runs away, …’

8 The lack of a distinction between real and pseudo-coordinations is the essential problem in Drohla’s (1934/1949) work. Whithout it most divergences between his interpretations and the conclusions presented here would disappear.

332   

(2)

   Cyril Brosch

KBo 17.105+ iii 17‒19 (MH) KUŠ ANA dLAMMA kursas=si=ssan kuis dat Tutelary.Deity hunting.bag:gen(c)=3sg.d/l=lp rel:nom.sg.c SÍG BABBAR anda hmankanz wool white in(to) bind:ptcp.nom.sg.c ⌈par⌉huenas=(m)a=kan kuis DINGIR-ni anda (kind of grain):nom.sg.c=cnj=lp rel:nom.sg.c deity:d/l.sg(c) in(to) MUNUS ishiyanz n(u)=as=kan ŠU.GI DINGIRLIM-ni tie:ptcp.nom.sg.c cnj=3pl.acc.c=lp old.woman deity:d/l.sg(c) arha lāizzi away detach:prs.3sg.act ‘The Old Woman detaches the white wool, which is bound onto the Tutelary Deity of the hunting bag, and the p.-grain, which is tied onto the deity, from the deity.’

But there are also other pecularities diminishing the explanatory value of the number of 477 attestations, impressive at first glance. Besides 68 examples containing morphologically underspecified targets,⁹ the two major problems are the Cuneiform Script and the foreign origin of more than a third of the corpus of Hittite texts. The Cuneiform writing system was invented for Sumerian and extended for Akkadian, making its way from an ideographic script to a mixed logographic-syllabographic system. It was adopted without any greater changes for Hittite (see Hoffner & Melchert 2008: 9‒24), so besides syllabograms (e.g. a, tu, e/it, šar) the Hittite writers made use of Sumerian (written upper case, e.g. É ‘house’, MEŠ for plurality) and Akkadian (upper case italics , e.g. ANA ‘to’, -ŠU ‘his’) logograms (with and without Hittite phonetic complements, e.g. ANA LUGAL, LUGAL-ui = hāssui ‘to the king’), which can serve as signs for single words or written-only classifiers to indicate certain classes of nouns (superscript upper case, e.g. GIŠ ‘wood’ in front of any kind of instruments, d = Sumerian DINGIR in front of gods) – and sometimes it is not clear whether we are dealing with a logogram or a classifier. As it is often unclear which particular Hittite form or even lexeme is hidden behind a logogram, also for common nouns like MUNUS ‘woman’, DUMU ‘son’, or ANŠE.KUR.RA ‘horse’, it is in many cases impossible to determine the

9 In most cases it is the number-indifferent neutral anaphoric pronoun =at ‘it, they’ (MH+; OH had a plural form =e, which was lost in MH).

Agreement patterns of coordinations in Hittite   

   333

properties of a controller. This of course severely hampers any analysis, and so do numerous lacunae in the clay tablets, indicated by squared brackets.¹⁰ In the Middle and New Hittite time, there is a large-scaled cultural, especially religious influence form the southern lands of Kizzuwatna (the later Cilicia) and Syria, which were inhabited by a population speaking Luwian and/or Hurrian. As the majority of the Hittite texts are rituals and religious myths, one finds many of such documents written in Hittite, but possibly composed by people with another native language. This can be seen especially in the syntactic domain, where one finds luwianisms, so that for the purpose of the research presented here it was necessary to separate the 177 cases concerned (37% of the overall attestations!), because they could contain non-authentic patterns. And in fact some phenomena are confined to this group of documents.¹¹ The remaining attestations are still sufficient for a reliable analysis of the Hittite agreement patterns, as they contain combinations of all major semantic classes of nouns (abstract, inanimate, animate, collective), word classes (except for the rare interrogative pronouns) and genders. While there are examples for semantically mixed controllers, most controllers appearing together belong to the same class, and two thirds of the attestations consist of nouns (or pronouns) denoting people, collectives of people, or animals, while there are relatively few examples of inanimate or abstract nouns.

3 The attestations Most attestations, presented in this section, show a behavior governed by a few basic rules, which will be discussed in S. 4. There are, however, a few types of known pecularities, which will be treated separately in S.  3.2. As there are not many attestations for gender conflicts and even less for person conflicts, the focus will necessarily be on agreement resp. mismatches in number.

10 Half brackets indicate damaged signs, round brackets within squared brackets indicate text portions preserved in duplicate texts. 11 Examples from foreign texts are henceforth marked by “transl[ation]”, although according to the lack of written Luwian or Hurrian originals they may be original compositions by Luwian and/or Hurrian native speakers. They will only rarely be cited in the following section, and exclusively for illustrative purposes, never as evidence.

334   

   Cyril Brosch

3.1 Common cases The common cases will be presented according to the three domains of agreement. These are not attested with equal frequency, with the attributes making up less than ⅛ of the examples, while verbs (and predicatives) and phoric elements (in most cases anaphoric clitics) are nearly equally distributed. For this reason not all combinations of controllers are attested with all domains (for each of which one example is given in the following sections), there are also very few examples for pronominal controllers. In the domain of the noun phrase, with any kind of coordinated controllers one seemingly always finds plural targets (see ex. (3)‒(5) with common animate, common inanimate, and neutral/inanimate controllers, cf. also (28)), the singular is found only with a specific class of nouns discussed below in S. 3.2.2. (3)

KUB 21.38 obv. 47 (LNH) MUNUS.LUGAL-as=⌈z⌉ kuiēs DUMU.MUNUS KUR queen:nom.sg.c=refl rel:n/a.pl.c daughter land URU d ⌈ ⌉ URU GARA DUNIYA Š [DUMU.MUNUS KUR] AMURRI=ya Babylon daughter land Amurru=and dahhun … take:prt.1sg.act ‘The daughter of the land Babylon and the daughter of the land Amurru, whom I, the queen, had taken to me, …’

(4)

KUB 15.11+ iii 16‒17 (NH) nu 1 hu!ta!nnin KÙ.BABBAR 1 hutannin conn 1 (a vessel):acc.sg.c silver 1 (a vessel):acc.sg.c K[Ù.SI22?] IŠTU Ì.DÙG.GA sūwantes […] gold abl/instr fine oil full:nom/acc.pl.c ‘One h.-vessel made of silver (and) one h.-vessel made of gold(?), filled with fine oil […]’

(5)

KBo 39.8 iv 15‒16 (transl./MS) tuwarnattaru=war=at hūmanda break:imp.3sg.mp=quot=3sg/pl.n/a.n all:n/a.pl.n hurtāus=sa curse:n/a.pl.c=and ‘They shall be broken, all words and curses!’

ud[d]ār word:n/a.pl(n)

Agreement patterns of coordinations in Hittite   

   335

As predicates and predicatives are better attested, for nearly every semantic and formal combination of controllers an example can be found. Independently of the semantic class of the controllers, every kind of coordination with a nearest member in the singular triggers a singular target, cf. ex. (6)‒(10) with abstract, inanimate, mixed and animate controllers: (6)

VBoT 24 iii 43‒45 (transl./NS) KUŠ tug=(m)a=kan ANA dLAMMA kursas you[d/l.sg]=cnj=lp dat Tutelary.Deity hunting.bag:gen.c karpis kartimmiaz sāuwar rancor:nom.sg.c wrath:nom.sg.c anger[nom.sg](n) arha QATAMMA mertu away likewise disappear:imp.3sg.act ‘And from you, Tutelary Deity of the hunting bag, may disappear rancor, wrath, (and) anger in the same way.’

(7)

KBo 17.1+ iii 5 (OH) LUGAL-s=an king:nom.sg.c.=3.sg.acc.c ‘King and queen let it go.’

MUNUS.LUGAL=sa tarnas queen=and let:pst.3sg.act

(8)

KBo 17.1+ iii 1‒2 (OH) d [mā]hhanda dUTU-us IŠKUR-as nēpis as sun:nom.sg.c Stormgod:nom.sg.c heaven[n/a.sg](n) tē[(kan=na) ] uktūri earth[n/a.sg](n)=and eternal[n/a.sg.n]¹² ‘As Sungodess, Stormgod, heaven, and earth are eternal [ ], …’

(9)

KBo 17.3+ iv 23‒25 (OH) 3 NINDAharsaes ispantuzzi=ya 3 thick.bread:nom.pl.c libation[n/a.sg](n)=and marnuan kitta barley.beer?:n/a.sg(n) lie:prs.3sg.mp ‘Three loafs of thick bread and a libation of barley beer? lie (there).’

12 In a few cases the morphologically endingless form can be used as a plural, too. There are, however, no other cases of resolution of the number with concurrent retention of the neuter gender with animates, so that it seems to be safe to assume a singular here.

336   

(10)

   Cyril Brosch

KUB 56.1+ i 11 (NH) 2 LÚ 1 MUNUS akkanz 2 man 1 woman die:ptcp.nom.sg.c ‘Two men (and) one woman are (“is”) dead.’¹³

If, however, the nearest controller is a plural (common gender only, see 3.2.1), the target usually takes the plural.¹⁴ (11)

KUB 57.79 iv 33 (transl./MS) [nu=tt]a MU.KAM.ḪI.A ITU.KAM.ḪI.A cnj=2sg.d/l years months talukaēs asand[u] long:nom.pl.c cop:imp.3pl.act ‘Let your years, months, (and) days be long.’

U4.KAM.ḪI.A days

The plural can also appear if all members of the coordination are animate or collective nouns, independently of the number of the nearest controller (ex. (12)‒(16) with different combinations of singular, plural, animate and collective noun and with personal pronouns). In fact, with this kind of nouns, the plural is much more often (ca. 9 : 1) attested than the singular. (12)

KBo 4.4 ii 7 (NH) [mNÍG.BA-d10-a]s=ma KUR URUKinza=ya ammedaz EGIR-pa Niqmadu:nom.sg.c=cnj land Qadeš=and I:abl back wahnuer turn:pst.3pl.act ‘But Niqmadu and the land of Qadeš sided with me again.’

13 Parallel formulations in the same text show the plural form akkantes, see Otten/Souček (1965: 30, with fn. 16). 14 There are hardly any examples of plural in last position with inanimate controllers, since these are generally rather rare and even less often attested in the role of the subject; see, however, also ex. (18) (with anaphoric pronoun), which shows that in the non-special cases they trigger plural agreement.

Agreement patterns of coordinations in Hittite   

   337

(13)

IBoT 1.36 i 63 (MH) BELUTIM=ya=kan UGULA LIMTI=ya GAL-yaz katta lords=and=lp commander 1000=and big:abl down paiskanta go:ipfv:prs.3pl.mp ‘Both the lords and the commander of thousand (troops) go down through the main (gate).’

(14)

HKM 75 15‒17 (MH) m m m nu zik Kāssus Zilapiya Huillis cnj you[nom.sg] Kāssu:nom.sg.c Zilapiya:nom.sg.c Hoilli:nom.sg.c nuntarnutten hurry:imp.2pl.act ‘You – Kāssu, Zilapiya, (and) Hoilli –, hurry up!’

(15)

KUB 14.3 iv 7‒9 (NH) LUGAL KUR Hatti=wa(r)=nnas=kan uk (…) ku⌈rur⌉ king land Hatti=quot=1pl.obl=lp I[nom.sg] enmity[n/a.sg](n) esūen cop:pst.1pl.act ‘The king of the land Hatti (and) I were hostile against each other (…) .«’

(16)

KBo 17.1+ iii 7 (OH) LUGAL-us MUNUS.LUGAL-as=sa QATAMMA uktūries king:nom.sg.c queen:nom.sg.c=and likewise eternal:nom.pl.c asantu cop:imp.3pl.act ‘Let king and queen likewise be eternal.’

In the domain of phoric reference targets in most cases behave similarly to the predicate. With all kinds of coordinations, phoric pronouns – usually anaphoric pronouns after a clause boundary (but not always, cf. ex. (19))– agree with the nearest controller (ex. (17)‒(19) with different orders of singular and plural). However, again in ca. 90% of the cases animate or collective nouns trigger plural agreement (ex. (20)).

338   

   Cyril Brosch

(17)

KBo 17.1+ iii 10‒13 (OH) kās[(a LU)]GAL-i MUNUS.LUGAL-ri DUMU.MEŠ-mas=sa dem.1sg king:d/l.sg(c) queen:d/l.sg(c) children:d/l.pl(c)=and URU Hattusi ērma(n)=smet Hattusa:d/l.sg(c) illness:[n/a.sg](n)=poss.3pl.n/a.n ēsh[(ar=s)]met idālu=smet blood:n/a.sg(n)=poss.3pl.n/a.n evil:[n/a.sg](n)=poss.3pl.n/a.n hatuka(n)=smet hari[(enu)]n ta=at terrible:n/a.sg.n=poss.3pl.n/a.n bury: prs.1sg.act cnj=3sg.n/a.n ⌈ āppa srā lē wēzzi⌉ back up proh come:prs.3sg.act ‘For king, queen, and their children I’ve buried their illness, blood guilt, evil (and) horror here in Hattusa, and they shall not come up again.’

(18)

KBo 17.1+ iii 8‒9 (OH) wīlnas ÉRIN.MEŠ-an¹⁵ tessummius=sa taknā clay:gen.sg(c) troop(s):acc.sg(c) cup:acc.pl.c=and earth:dir.sg(n) hariemi t(a)=us tarmaemi bury:prs.1sg.act cnj=3pl.acc.c peg:prs.1sg.act ‘I bury the clay “troop” and the cup into the earth and fix them with a peg.’

(19)

KBo 15.10+ ii 34‒35 (MH) d d nu=ssi UTU-us IŠKUR-as=sa ANA BE[L]I cnj=3sg.d/l sun:nom.sg.c Stormgod:nom.sg.c=and dat my.lord ANA DAM DUMU.MEŠ-ŠU āssu TI-tar dat his.wife his.sons good[n/a.sg.n] life[n/a.sg](n) GIŠ mayan⌈datar⌉ TUKUL prā niyantan [na]m[m]a thriving[n/a.sg](n) weapon forth turn:ptcp.acc.sg.c further piskatten give:ipfv.imp.2pl.act ‘Furthermore, Sungodess and Stormgod, give good life, thriving, and pulled out weapon to them (lit. “him”) –my lord, wife, (and) his sons!’

15 I.e. a clay figurine of a soldier. The plural determinative MEŠ is part of the logogram ÉRIN.MEŠ and does not determine the number of the underlying noun in this case.

Agreement patterns of coordinations in Hittite   

(20)

   339

KUB 14.15+ iv 35 (NH) URU URU namma URUArsanin Sārauwan Impan=na further Arsani:acc.sg.c Sārawa:acc.sg(c) Impa:acc.sg(c)=and wetenun n(u)=as BÀD-esnanun build:prt.1sg.act cnj=3pl.acc.c fortify:prt.1sg.act ‘Then I built the cities Arsani, Sārawa, and Impa, and I fortified them.’

There is, however, a rare phenomenon, found mostly in texts from the Old Hittite period. After coordinations with multiple, mixed-gender controllers and at least two clause boundaries between controller and target the targets show common gender and plural, even when syntactical agreement should trigger (or even does trigger, if the target immediately after the first clause boundary, see. ex. (21)) a neuter and/or singular pronoun. (21)

KBo 17.1+ iv 21‒22 (OH) (offerings, 1 additional sentence) n(u)=e LUGAL-as cnj=3pl.n/a.n king:gen.sg(c) MUNUS.LUGAL-as=sa [(ke)]tkar=smet tēhhi queen:gen.sg(c)=and at.the.head=poss.3pl.n/a.sg.n put:prs.1sg.act sēr=(m)a=s⌈san⌉ GADA-an pessiemi at.the.top=cnj=lp blanket:acc.sg(c) throw: prs.1sg.act su=us [(LÚ-as n)]atta auszi cnj=3pl.acc.c man:nom.sg.c neg see:prs.3sg.act ‘(I fix a thread to every branch. A small [ ] (and) one clay figurine with mud-plaster and with [ ], araummi. Emmer ears are bound. All of these I put into a basket.) I put them (n.) at the head¹⁶ of king and queen. And on top I throw a blanket upon it/them, so that noone would see them (c.).’

16 The exact local configuration is not clear. The basket with offerings may be fixed somewhere in the area where the heads of king and queen are, or it may be even literally put on their heads. It also can not be excluded that the blanket does not cover the basket, but the royal couple, although this seems rather unlikely.

340   

   Cyril Brosch

(22)

KBo 17.3+ iv 23‒25 (OH; cf. ex. (9)) 3 NINDAharsaes ispantuzzi=ya 3 thick.bread:nom.pl.c libation[n/a.sg](n)=and marnuan kitta barley.beer?:n/a.sg(n) lie:prs.3sg.mp (2 sentences) t(a)=u(s)=sta srā tummēni cnj=3pl.acc.c=lp up take:prs.1pl.act ‘Three thick breads and a libation of barley beer? lie (there). (…) We pick them up from there.’

(23)

KBo 11.10 iii 27‒30 (transl./NS) ḪUL-un=na EME-an ḪUL-un MUŠEN-in evil:acc.sg.c=and tongue:acc.sg(c) evil:acc.sg.c bird:acc.sg.c QATAMMA warnuwandu n(u)=as QATAMMA likewise burn:imp.3pl.act cnj=3sg.nom.c likewise hastāi kisaru nu=us kissan prā bone[n/a.pl](n) become:imp.3sg.mp cnj=3pl.acc.c so forth ÍD-as pēdāu nu=us river:nom.sg.c bring.thither:imp.3sg.act cnj=3pl.acc.c harniktu destroy:imp.3sg.act ‘And let them likewise burn the evil tongue (and) the evil bird. Let them (lit. “it”) likewise become bones, and so let the river take them forth and destroy them.’

Another pecularity is found from Middle Hittite times on and concerns long lists, which are referred to by summarizing expressions like hūmant- ‘every, all’, often in combination with a demonstrative (see Rieken & Widmer forthcoming). While in Old Hittite these show regular syntactic agreement (cf. ex. (24)), in later times the anaphoric reference appears only in the singular, with common gender for animate controllers and neuter for inanimates (see exx. (25) and (26)). (24)

KBo 17.3+ iv 16‒17 (OH) (list of offerings) [ZÍZ.Ḫ]I.A=sa harsār ishiyanda emmer=and head[n/a.pl](n) tie:ptcp.n/a.pl.n kē=ssan hūmanda [pa]ttanī tēhhe dem.1ps.n/a.pl.n=lp all:n./a.pl.n basket:d/l.sg(n) put:prs.1sg.act ‘(I fix a thread to every branch. A small [ ] (and) one clay figurine with mud-plaster and with [ ], araummi). Emmer ears are bound. All of these I put into a basket.’

Agreement patterns of coordinations in Hittite   

   341

(25)

KUB 24.4+KUB 30.12 obv. 16‒17 (MH?) nu kurewanas KUR.KURTIM k[u]e arahzand[a] […] ? conn independent:gen countries rel:n/a.pl.n around ⌈ ⌉ nu h ūmanz sull[(ēt)] conn all:nom.sg.c become arrogant:pst.3sg.act ‘The independent countries around […], everyone started to defy.’

(26)

KUB 9.25+KUB 27.67 ii 9‒14 (transl./NS) (list of offerings) peran=ma=ssan KUŠsālas hminkanz in.front=cnj=lp thong:nom.sg.c bind:ptcp.nom.sg.c GI nu=ssan kī hūman paddanī cnj=lp dem.1ps.n/a.sg.n all[n./a.sg.n] basket:d/l.sg(n) dāi put:prs.3sg.act ‘(Balls and mixtures of dough. A bow made of reed?, (its) bowstring? is drawn; three arrows made of reed?; in front of them (are) balls of dough, a door-bolt? (and) a latch?). He/She puts this all into a basket.’

3.2 Special cases The patterns outlined in S. 3.1 hold true for the majority of the cases throughout the language history. As was said above, cases with apparent aberrant agreement are confined to a few types, which, however, are prominent because of their high token frequencies. They are diachronically stable, too, with the Schema Attikon (see the following section) being probably inherited from Proto-Indo-European.

3.2.1 Schema Attikon In Hittite, Ancient Greek, Old Avestan, and Vedic Sanskrit a neuter plural controller need or even must not trigger plural, but triggers singular agreement with the predicate. Since this behavior is anomalous regarding the agreement patterns of the relevant languages and mostly vanishing during their attested history, it is often seen as an archaism, going back to a prehistoric stage, when the element *-h2, which was to become the neuter plural ending, was some kind of a com-

342   

   Cyril Brosch

prehensive number.¹⁷ Unlike the other ancient Indo-European languages, which keep the phenomenon only to a certain and ever diminishing extent, Hittite very consequently applies this rule, even extending it from Old Hittite times on to the predicative noun (see van den Hout 2001: 172‒180 and ex. (37)). The predicate of a neuter coordination appears in the singular virtually without exceptions, too (cf. ex. (27)), while attributes to it as a rule show the canonical plural resolution (ex. (28)). Being fully grammaticalized, the Schema Attikon overrides all other agreement rules. (27) KBo 17.105+ iii 7‒8 (transl./MS) ⌈ an⌉da=ma=kan āssuwa mīyawa hat⌈ta⌉nta in(to)=cnj=lp good:n/a.pl.n soft:n/a.pl.n wise:n/a.pl.n wid[du] come:imp.3sg.act ‘Good, pleasant, and wise things shall come in.’ (28)

HKM 54 12‒14 (MH) ŠEAM Ù ZÍZ=ya [k]ue anniyan barley and emmer=and rel:n/a.pl.n sow:ptcp[n/a.sg.n] esta¹⁸ man ŪL apēz datta cop:pst.3sg.act irr neg dem.2ps.abl take:pst.2sg.act ‘If you hadn’t taken from the barley and emmer, which were sown, …’

3.2.2 Military terms Within the context of martial actions, like campaigns and plunderings, the nouns ÉRIN.MEŠ ‘troop(s), army’, ANŠE.KUR.RA.MEŠ/ḪI.A ‘horse troops’ (lit. “horses”), NAM.RA.MEŠ/ḪI.A ‘deportees’, GU4(.ḪI.A) ‘cattle’, and UDU(.ḪI.A) ‘sheep’ (i.e. big and small domestic animals), most of which do not have an assured Hittite reading, always trigger singular agreement,¹⁹ not only with the predicate and 17 See van den Hout (2001: 167‒171) with further references. The evidence outside Greek and Hittite, however, is very scarce and there are no compelling arguments to discard the alternative view of a secondary development. 18 The predicative and the predicate are non-probative for the Schema Attikon here, as the last controller seems to be a neuter noun, hence the coordination would trigger neuter and singular agreement according to the common rules (see S. 3.1), too. 19 The behavior of these words as single controllers is somewhat different, as they do trigger plural agreement when marked as plurals (i.e. ÉRIN.MEŠ.ḪI.A, KARAŠ.ḪI.A), see Naglik (2012: 14‒20, 36‒39). Hence it seems that coordinations like ÉRIN.MEŠ ANŠE.KUR.RA.MEŠ are highly lexicalized, following neither the rules of coordinations nor of their single components.

Agreement patterns of coordinations in Hittite   

   343

phoric expressions (exx. (29)‒(31)a), but even in the domain of the noun phrase (see ex. (31)b), where otherwise plural resolution is the rule (see. S. 3.1). (29)

KBo 3.4+ iv 35‒36 (NH) nu=mu namma ÉRIN.MEŠ ANŠE.KUR.RA.MEŠ ⌈ŠA⌉ KUR cnj=1sg.d/l further troop(s) horses gen land Azzi zahhiya ŪL tiyat Azzi fight:d/l.sg(c) neg step:pst.3sg.act ‘The foot soldiers (and) chariotry of the land Azzi did not encounter me again for a fight.’²⁰

(30)

KBo 5.2 ii 12‒13 (NH) nu=ssi=kan NAM.RA.ḪI.A GU4 UDU arha cnj=3sg.d/l=lp deported.population cow sheep away dahhun n(u)=an tuzziyas=mis take:pst.1sg.act cnj=3sg.acc.c army:nom.sg.c=poss.1sg.nom.sg.c saruwāet loot:pst.3sg.act ‘I took prisoners, cattle (and) sheep away from him, and my army looted them (lit. “it”).’

(31)

KBo 3.4+ iii 53‒54 (NH) KÙ.BABBAR-as=ma=z EN.MEŠa ÉRIN.MEŠa Hattusa:gen(c)=cnj=refl lords troop(s) ANŠE.KUR.RA.MEŠa=ya kuinb horses=and rel:acc.sg.c NAM.RA.MEŠb GU4b UDUb uwateta deported.population cow sheep lead.hither:pst.3sg.act ‘But the prisoners, cattle (and) sheep, which the lords, foot soldiers, and chariotry of Hatti had led here for themselves, …’ URU

Outside the agreement patterns, there is no additional evidence explaining this behavior, but one fact seems to suggest that this kind of controller is seen as one unity, independent of the actual forms: The concerning expressions do show canonical agreement according to the rules given in S. 3.1, if the controllers are not a unitary force, but e.g. different armies (cf. ex. (32)) or if they are seen outside the military context (cf. ex. (33) with a single controller).

20 Cf. the same singular verb in ex. (31).

344   

   Cyril Brosch

(32)

KBo 16.47 obv. 5’‒6’ (MH) nu=mu ÉRIN.MEŠ URUUrā ÉRIN.MEŠ URUMutamutasi [kattan] cnj=1sg.d/l troop(s) Urā troop(s) Mudamudasa with lahha iyantari campaign go:prs.3pl.act ‘The troops of Urā (and) the troops of Mudamudasa are going to campaign together with me.’

(33)

KBo 6.2 iv 12 (OH) takku GU4.ḪI.A A.ŠÀ-ni pānzi if cows field:d/l.sg.n? go:prs.3pl.act ‘If cattle enters a (foreign) field, …’

3.2.3 Numerals? There are no examples of numerals referring to all members of a coordinations, all numerals attested with coordinations refer to a single member (like five oxen and ten sheep). Nevertheless it cannot be excluded that a numeral may be able to influence the form of its controllers and so indirectly of the targets for the fact that, for reasons understood only now,²¹ after numerals higher than ‘1’ the controller may appear in the singular or in the plural, depending on animacy, gender, and contextual classification. There are, however, no probative examples where the usual agreement patterns would not apply because of a numeral (cf. exx. (9) and (10) with regular agreement), so this remains only a theoretical possibility.

4 Agreement patterns: number, gender, person In order to understand the pecularities of coordinations, one has to take into account the agreement patterns of single controllers first. They are straightforward and comparable to the ones found in other ancient Indo-European languages. Attributive nominals agree in gender and number with their controllers (see exx. (34), (37)), so do predicative ones (ex. (35)), while the verbal predicate agrees in number and person (exx. (36)‒(37)). The only regular exception is the

21 See Rizza (2011 and 2012) for preliminary results for Old Hittite and Rieken (2013b) for a thorough study with the basic rules and the (mostly motivated) exceptions. Cf. also Hoffner & Melchert (2008: 158‒163).

Agreement patterns of coordinations in Hittite   

   345

Schema Attikon, mentioned above (S. 3.2.1), according to which the predicat(iv)e of a neuter plural controller appears in the singular (ex. (37)). Phoric elements agree in gender, number, and, if applicable, in person with their antecedents (exx. (34)‒(37)). Controllers devoid of any feature values (like clauses) always trigger the neuter, seemingly as the less marked gender (ex. (38)). (34)

KBo 17.1+ ii 19‒20 (OH) mān MUŠENhāranan hus[(uwandan appanzi)] when eagle:acc.sg(c) living:acc.sg.c catch:prs.3pl.act n(u)=an udanzi cnj=3sg.acc.c bring.hither:psr.3pl.act ‘When they catch a living eagle, they bring it here.’

(35)

HKM 25 15‒17 (MH) nu=ssan mān halkiēs arantes cnj-lp when grain:nom.pl.c stand:ptcp.nom.pl.c n(u)=as=kan arha warsten cnj=3pl.acc.c=lp away reap:impl.2pl.act ‘When the grains are ripe, harvest them completely.’

(36)

KUB 8.81+ ii 11’‒12’ (MH) URU URU mān=asta LÚpatteanza Hattusaz K[izzuwa]t[ni] if=lp flee:ptcp.nom.sg.c Hattusa:abl(c) Kizzuwatna:d/l.sg(c) m paizzia Sunassurasb=ana ēptub go:prs.3sg.act Sunassura:nom.sg.c=3sg.acc.c seize:imp.3sg.act ‘If a fugitive comes over from Hattusa to Kizzuwatna, let Sunassura seize him.’

(37)

KUB 23.72+ rev. 2‒3 (MH) [nu] kuea uddāra tiy[an]a cnj rel:n/a.pl.n word[n/a.pl](n) set:ptcp[n/a.pl.n] ēstaa apāsb=ata=kan cop:prt.3sg.act dem.2ps.nom.sg.c=3sg/pl.n/a.n=lp hūmantaa sarrasb all:n/a.pl.n transgress:prt.3sg.act ‘He transgressed all the words, which were set (as treaty).’

346   

(38)

   Cyril Brosch

KUB 1.1+ i 73‒74 (NH) kuitman=ma=z DUMU-as esun nu=z while=cnj=refl boy:nom.sg.c cop:prt.1sg.act cnj=refl KUR.KUR LÚKÚR kue taruhhiskenun lands enemy rel:n/a.pl.n defeat:ipfv.prt.1sg.act n(u)=at ṬUPPU hantī DÚ-mi cnj=3sg/pl.n/a.n tablet extra make:prs.1sg.act ‘The countries of the enemy I defeated while a was young – I will make an extra tablet on it.’

Against this background, it becomes clear that coordinations in Hittite are a discrete phenomenon, since they not only do not work like simple additions of controllers (“singular + singular = plural”), but show intricate patterns depending on their semantic class (e.g. animate vs. inanimate), though apparantly not depending on lexical features like gender. First, a real coordination is a combination of two or more syntactically equal controllers within the same clause, which corresponds to the same amount of extralinguistic entities –in that way differing from pseudo-coordinations, which comprise disjunctive coordinations and parallel controllers (see S. 2, exx. (1) & (2)). Second, the targets of coordinations show two different agreement patterns outside the domain of the noun phrase: They can agree with the nearest controller (exx. (6)‒(11)),²² but they can also show number resolution resulting in plural (exx. (12)‒(16)). The latter case is only possible –and the usual one–, if all members of the coordinations are animate or at least collective nouns denoting animates. The examples of animates without plural resolution are predominantly confined to natural pairs like “king and queen” (see ex. (7)). Attributes always seem to take the plural (exx. (3), (5), and (28)), an exception being a closed class of military terms probably making up a kind of semantic unity (see S. 3.2.2).²³ There are, however, only two probative attestations (see exx. (4) and (28)), i.e. two inanimate singular controllers with a plural attribute.²⁴ If 22 As most targets are following their controllers this is usually the last member of the coordination, but cf. ex. (19). There are no convincing examples showing agreement with the farer first member of a coordination. As was said above in S. 3.2.1, the grammaticalized Schema Attikon overrides all other agreement rules. 23 For the lack of phonetic spelling it is unclear whether this phenomenon has a formal foundation, too. Some kind of (copulative) compound may be possible, however very unlikely. 24 As animate nouns usually trigger plural resolution in other domains, their attestations can’t be taken into account here. In the reliable texts there are no clear examples for the behavior of attributes to mixed-number coordinations.

Agreement patterns of coordinations in Hittite   

   347

they were discarded, the remaining examples (military terms only) would suggest agreement with the next controller as the only agreement strategy with attributes. But then, as there are no good reasons to do so and as the military terms, other than common collective or animate nouns never showing plural resolution, do seem to follow different rules, it still seems most reasonable to me to see obligatory plural resolution as the standard agreement strategy within the domain of the noun phrase. This finding has weighty consequences, as such a pattern goes against Corbett’s agreement hierarchy (see S. 1.2), claimed to be universal: The number resolution found in the domain of the attribute is semantically better motivated than the syntactic agreement found within the domains of the clause and text, the plain opposite of the typological prediction. Such clear counter-evidence to the agreement hierarchy has not yet been noted in Corbett’s data and should be taken seriously. Although Hittite has the grammatical category of gender without any indications of a loss of the distinction between common and neuter gender, lexical gender does not seem to play any role in the assignment of the number of the target outside the Schema Attikon (see S. 3.2.1). The few cases of gender conflict largely corroborate the patterns just explained: The target takes the gender of the nearest controller (also in the domain of the noun phrase, so there is no kind of resolution; cf. exx. (5), (8), and (11)). With animate controllers triggering number resolution it takes the common gender (ex. (16)).²⁵ There are, however, two additional phenomena concerning the domain of anaphoric reference: First, as an instance of diachronic change, from Middle Hittite times on long lists cease to be summarized according to the general rules, but a new animacy-based distinction between hūmanz (sg. c.) ‘everyone’ and hūman (sg. n.) ‘all’ is introduced (see Rieken & Widmer forthcoming and exx. (24)‒(26)). Second, mixed-gender or even pure neuter coordinations are referred to by common-gender pronouns, if separated by several clause boundaries (see exx. (21)‒(23)). The reason for this is not clear yet, but as in all examples the controllers are concrete nouns, it is possible that the common gender to the Hittites seemed more appropriate to express this semantic class of nouns than the neuter, which comprises more abstract and much less concrete nouns than the common

25 Animate nouns in Hittite overwhelmingly have common gender, while collective nouns can be of neuter gender (as utnē n. ‘land’). There is no attestation of a neuter plural predicative agreeing with a mixed-gender coordination, but this remains an argumentum e silentio, as long as there are no certain examples for a gender conflict in such coordinations, so that the eventual outcome of expressions like “The king of X (c.) and the land of X (n.) are Y-ic” remains unclear.

348   

   Cyril Brosch

gender.²⁶ Unfortunately, there are hardly any examples of single neuter controllers triggering common-gender agreement, so this remains a hypothesis.²⁷ While proximity is a deciding factor for the agreement features number and especially gender, for the category of person it does not seem to be relevant. There are very few attestations for conflicts in person (see. exx. (14) & (15)), but in all of them the first or second person enforce their respective agreement forms together with the number resolution expected with animate controllers, independently of the word order. In order to establish a person hierarchy examples of conflicts between first and second person are needed, but examples are lacking so far.

5 Conclusion Summing up, it has been shown that coordinations in Hittite are a complex phenomenon. Their attestations at first glance seemed not to have any rationale for their agreement values. However, a careful examination, separating real from pseudo-coordinations and analyzing the former according to semantic categories, revealed a clear set of rules based on animacy and word order. By marking the switch between two agreement strategies, i.e. agreement with the nearest controller (as formal agreement labeled ‘F’ in the following table) resp. resolution (labeled ‘S’ as an instance of semantic agreement), the category of animacy, known e.g. from case assignment or the use of hūmant- ‘all’ (see above), has again shown its importance in Hittite. There are open questions regarding the agreement of mixed-gender coordinations and conflicts in person, the solution of which depends on new data. The agreement patterns of the different types of controllers  – for the sake of clarity and comparison also of those not or not amply discussed in this paper – are collected in the following table, ordered horizontally according to the agreement hierarchy, and vertically according to the 26 A quick examination of about 1340 examples for these two types of nouns shows the following numbers: common gender: abstract 391 (53%), concrete 352 (47%), neuter: abstract 368 (61%), concrete 231 (39%). 27 There is, however, a phenomenon found in the so called KIN-oracles (see Beal 2002, Haas 2008), a group of texts describing the events of a not yet understood oracular technique (possibly involving small animals or lots). The events are usually formulated like “X (name of a person/ god) rises, takes Y1‒Yn (one or more abstract conceptions like “mildness”, “long years”, “defeat”) and gives it/them to Z (a person or location)” or “Y is taken and given to Z”. In about 40% of the attestations the pronouns or predicative participles referring to Y take the common instead of the expected neuter gender. As Y in fact are not real abstract nouns but objects representing these concepts, the mismatches may be caused by this discrepancy between signifier and the denotation. The syntactical circumstances are not identical, however.

Agreement patterns of coordinations in Hittite   

   349

complexity of the controllers (a‒b single, c‒d pseudo-coordinations, e‒h real coordinations):²⁸ Table 1: Formal (F) and semantic (S) agreement in number with different kinds of controllers in Hittite type of controller

attribute predicative predicate verb phoric expression

a single (pl.): common gender

F=S

F=S

F=S

F=S

F=S

F=S > F

F

F=S

F=S

F=S

F=S

(F=S)

(F=S)

S

1 S

S

S

S

2 ?

F

F

F

b single (pl.): neuter gender *

c disjunctive “coordinations” (S. 2) F=S d split controllers (S. 2)

**

e coordination: animate (incl. collective nouns), common gender (S. 3.1)

(F=S)

f coordination: inanimate/mixed, common gender (S. 3.1)

S?

F

F

F

g coordination: usually inanimate, neuter gender (S. 3.2.1)

S?

S>F

F

F

F

F

F

h coordination: military terminology F (animate, both genders) (S. 3.2.2)

* It is a theoretical problem whether one should count the nouns coordinated by OR-expressions as one or more controllers. As they never show any kind of resolution, they seem to be treated as one controller also syntactically. It is not fully certain what happens in cases where formal features of the expressions concerned collide. There is only one clear such attestation, showing agreement with the nearest of the two possible controllers: mān=ma ḪUR.SAG=ma kuiski nasma sinapsi suppa AŠRU kuitki ḪUL-ahhan “But if some mountain or some sinapsisanctuary – a pure place – is offended, …” (KBo 11.1 obv. 32; NH), where ḪUR.SAG is common gender, and sinapsi and AŠRU (= Hitt. pēda- n.) are neuter, with which the neuter participle ḪUL-ahhan agrees syntactically. ** Since the controllers appear in different clauses, only in the domain of the text there can be pecularities. However, it is of course expected that pronouns referring to multiple antecedents show up in the plural, otherwise the reference would be unclear.

28 For the sake of simplicity in this already complex table, plural single controllers are compared to singular conjoined controllers (and pseudo-coordinations) in order to show the differences. Of course other types of coordinations, e.g. pl. + pl., show forms where formal and semantic agreement coincide.

350   

   Cyril Brosch

With single plural controllers formal (F) and semantic (S) agreement should coincide (plural targets, hence F=S). This holds true for common gender (a), while with neuters (b) the grammatical rule called schema Attikon (see S. 3.2.1) demands a singular predicate, which is labeled ‘F’ here as it is not motivated semantically. The rule is extended even to the predicative in post-OH times (OH F=S changes to MH F, labeled here as S=F >F). The very same forms of the targets (i.e. according to the order in table 1 plural – plural > singular – singular – plural/number-indifferent) show up with conjoined neuter singular + singular controllers (g). With coordinations the plural of the targets, however, is to be seen as purely semantic agreement (as the controllers are not plurals), and hence labeled ‘S’. The table also shows some other facts: First, pseudo-coordinations behave like single controllers (c is like a, and d is simply two times a in consecutive sentences) and are not comparable to real coordinations, although they look similar at first glance. Second, there is a clear split of linguistic hierarchies between the cases a‒d and e‒h, respectively: While the agreement of the former is predictable by formal and/or lexical features of the controller (gender, number), for the latter one has to take into account basically semantic and pragmatic features (animacy, specialized lexicon), with a distinction between common and neuter gender only in the second place. Oddly enough, although the selection of the two possible patterns of number agreement has a semantical base, in the majority of the types the chosen pattern is the one of formal, not semantic agreement.²⁹ It has already been discussed in S. 4 that these agreement patterns surfacing in the context of coordinations (e2‒h, except for the major part of animates (e1) taking always resolution), i.e. probably semantically based plural resolution within the noun phrase vs. syntactically based agreement with the nearest member elsewhere, would contradict the allegedly universal Agreement Hierarchy, if one cannot explain away the two clear examples (4) and (28).³⁰ On the

29 Having not singled out the pseudo-coordinations, Drohla (1934/1949: 10‒11, 72‒75, 85‒87) comes to somewhat slightly diverging results. First and foremost for him no difference of behavior is visible between the domains of agreement (as there seem to be examples of lacking plural resolution with attributes), and there seemingly are cases of agreement with a farther first controller. However, none of his examples withstands scrutiny, as is shown in Brosch (2014a and 2014b). 30 I have no neat explanation for this behavior, but as Corbett (2006: 239) in fact has stated, resolution “produces examples which are non-canonical”, so it would be intriguing to think of the possibility that the agreement hierarchy were not cogent with coordinations. There could be a correlation between the distance of controller and target and the (lack of) resolution, too: While only the nearest of two farther controllers controls the agreement features of the target, an attribute is too near to both of them not to show plural resolution. But without additional evidence from more languages this remains speculative.

Agreement patterns of coordinations in Hittite   

   351

other hand, regarding gender there is no kind of resolution even for attributes, but the usual agreement with the nearest controller. Another intriguing fact is the general stability of the Hittite agreement system, which applies not only to coordinations, but, as the project has shown (with further publications to appear), to the overall language. In over 300 years of attestation there have been no significant changes in the agreement patterns, both with regard to syntactic and semantic agreement. Known changes like spread of the Schema Attikon (see above) or spontaneous recharacterization of neuter plurals (see Rieken forthcoming) are of minor scope (and independent of coordinations).³¹ Exhibiting non-trivial agreement patterns, the Hittite data for coordinations are valuable from a typological point of view, having been ignored until recently even by the Hittitologists themselves.

Acknowledgments The research this paper is based on was conducted between 2011 and 2013 within the University of Marburg project „Diachrone Entwicklung von Kongruenzsystemen in vier flektierenden indogermanischen Sprachen“, led by Jürg Fleischer, Elisabeth Rieken, and Paul Widmer (s. http://www.uni-marburg.de/fb10/iksl/ sprachwissenschaft/forschung/projekte/kongruenz [2013‒04‒15]) and funded by the Deutsche Forschungsgemeinschaft. I’d like to express my sincere thanks to H. Craig Melchert for sending me his unpublished paper and to Elisabeth Rieken, Florian Sommer, and an anonymous reviewer for much valuable advice concerning all parts of the paper. All remaining errors or inaccuracies are mine, of course.

References Beal Richard H. 2002. Hittite Oracles. In Leda Ciraolo & Jonathan Seidel (eds.), Magic and Divination in the Ancient World. Leiden et al.: Brill, 57–81. Brosch, Cyril. 2013a. Koordinierte singularische Substantive im Hethitischen und ihr Kongruenzverhalten. Altorientalische Forschungen 40/1, 20–41.

31 The odd common-gender reference to mixed-gender or even purely neuter controllers discussed in 4 may also amount to a diachronic issue, but there are hardly any post-OH examples. This lack may either be chance or a conscious avoidance of the concerning construction.

352   

   Cyril Brosch

Brosch, Cyril. 2013b. Gemischte und pluralische Koordinationen im Hethitischen und ihr Kongruenzverhalten. Altorientalische Forschungen 40/2, 314–336. Corbett, Greville G. 2006. Agreement. Cambridge: Cambridge University Press. Drohla, Wolfgang. 1934/1949. Die Kongruenz zwischen Nomen und Attribut sowie zwischen Subjekt und Prädikat im Hethitischen. Unpublished PhD thesis Marburg. Haas, Volkert. 2008. Hethitische Orakel, Vorzeichen und Abwehrstrategien. Berlin & New York: de Gruyter. HKM = Alp, Sedat. 1991. Hethitische Keilschrifttafeln aus Maşat. Ankara: Türk Tarih Kurumu Basımevi. Hoffner, Harry A. & H. Craig Melchert. 2008. A Grammar of the Hittite Language. Part 1: Reference Grammar. Winona Lake: Eisenbrauns. IBoT = İstanbul Arkeoloji Müzelerinde bulunan Boğazköy Tabletleri(nden Secme Metinler). Istanbul 1944‒1954, Ankara 1988. van den Hout, Theo. 2001. Neuter Plural Subjects and Nominal Predicates in Anatolian. In Onofrio Carruba & Wolfgang Meid (eds.). Anatolisch und Indogermanisch. Anatolico e Indoeuropeo. Akten des Kolloquiums der Indogermanischen Gesellschaft. Pavia, 22.‒25. September 1998. Innsbruck: Institut für Sprachen und Literaturen der Universität Innsbruck, 167‒192. KBo = Keilschriftttexte aus Boghazköy. Leipzig 1916‒1923, Berlin 1954‒. KUB = Keilschrifturkunden aus Boghazköy. Berlin 1921‒1990. Melchert, H. Craig. forthcoming. Agreement Patterns in Old and Middle Hittite. To appear in a forthcoming festschrift. Naglik, Benedikt. 2012. Das Kongruenz-Verhalten von Committee-Nouns im Hethitischen: Eine exemplarische Untersuchung. Master thesis, Marburg. Otten, Heinrich & Vladimir Souček. 1965. Das Gelübde der Puduḫepa and die Göttin Lelwani. Wiesbaden: Harrassowitz. Patri, Sylvain. 2007. L’alignement syntaxique dans les langues Indo-Européennes d’Anatolie. Wiesbaden: Harrassowitz. Rieken, Elisabeth. 2013. Die Numeruskongruenz nach Kardinalzahlen im Hethitischen. In: Indogermanische Forschungen 118, 321–338. Rieken, Elisabeth. forthcoming. Der Nom.-Akk. Pl. n. auf -i im Hethitischen: Zu den Mechanismen der Ausbreitung von Kongruenzsystemen. Historische Sprachforschung. Rizza, Alfredo. 2011. The ritual for the Royal Couple CTH 416. Syntax of non verbal predicates and numerals. In M. Barbera et al. (ed.), Anatolistica, indoeuropeistica e oltre nelle memorie dei seminarî offerti da Onofrio Carruba (anni 1997‒2002) al Medesimo presentate, Milano: Qu.a.s.a.r. s.r.l. 13‒37. Rizza, Alfredo. 2012. On the syntax of numerals in Hittite and in the ANE linguistic area. Atti del Sodalizio Glottologico Milanese N.S. 6, 235‒261. VBoT = Goetze, Albrecht (ed.). 1930. Verstreute Boghazköi-Texte. Marburg. Widmer, Paul & Elisabeth Rieken. forthcoming. Kongruiert alles? Zu den Kongruenzmustern des Pronominaladjektivs der Bedeutung ‘all, jeder, ganz’ im Griechischen und Hethitischen. In Norbert Oettinger (ed.). Das Nomen im Indogermanischen. Morphologie, Substantiv versus Adjektiv, Kollektivum. Akten der Arbeitstagung der Indogermanischen Gesellschaft, Erlangen, 14. bis 16. September 2011. Wiesbaden: Reichert.

Index of languages Acooli 95–96 Arabic, Modern Standard Standard 117 Bantu 19–20, 23, 105, 108, 148, 150–151, 153, 156, 158, 160–161, 163, 208–209 Berber 91–95 Bunan 61–68

Kikara 86 Kinyarwanda 19–20, 148–154, 156–163 Kwarandzyey 77–97 Lak (Daghestanian) 209 Latin 107 Lubukusu 160 Luxembourgish 237, 243, 251–255, 259–262

Chaoui 94 Dolakha Newar 65–60 Dutch 128–140, 196–198, 219, 228

Norwegian, Bookmål 215–233 – Nordreisa 207–208 Polish, Southern Polish 209–210

Englisch 195–196, 200, 228 – Old English 6 French 219 Frisian, North Frisian 7–9 – Old Frisian 6, 8–9 German 194, 200 – Early New High German 288–289 – Middle High German 288 – Modern German 287–325 – Old High German 6–7, 14–18, 117, 200–203, 208, 210, 241, 274, 292–293, 296, 309 Greek, Ancient Greek 20–21 Godié 118 Hombori 87 Icelandic 4–5, 16, 23, 203–204, 210, 267–281, 283–284 Irish, Old Irish 165–187 Italian 106–124, 219

Saxon, Old Saxon 6 Serbo-Croat 199, 203, 205–207 Shona 153 Slavonic, Old Church Slavonic 205–206 Songhay 77, 85–95 Spanish 105 Swahili 105 Swiss German 238, 243–247, 250 Tadaksahak 87, 90 Tamazight 93 Tashelhiyt 92 Turkish 118–119, 150, 161–162 Yiddish 7–9 Zarma 86 Zenaga 93–94 Zuaran 94

Index of subjects agreement conflict 288–292 agreement hierarchy 1, 6–7, 9, 22, 130, 191–195, 197, 199, 203, 205–209, 211, 217, 221, 227–229, 240–241, 268, 271–272, 274, 329–330, 347–348, 350 agreement, canonical 171–172 – external 77–97 – gender 19, 104, 109–112, 114, 118, 122–123, 127–128, 130, 149, 203, 239–240, 247–248, 253, 268–269, 274, 276, 279–280, 284, 303, 320 – person 19, 34, 43, 54–55, 58–59, 64–65, 67–70, 108, 149–151, 154, 157–158, 162–163 – prefix 78 – prepositional object 174–175, 182–183 – pronominal 127–140 – semantic 193, 215–233 – Structural Condition (SCOPA) 150–151 – subject-verb 173–174, 182–183 – verbal 287–325 – verbal object 175–179, 182–183 alignment 30–38, 41–42, 48, 111 animacy 11, 20–21, 34, 127, 129, 132–133, 135–136, 140–141, 156, 172, 175, 208–209, 215–216, 220, 223, 226–229, 237, 245, 258–259, 272, 288, 292, 294, 296–299, 301–304, 307, 311–313, 319–320, 329, 333–337, 340, 344, 346–350 areas 32, 42, 47–48, 95, 121, 208, 235, 239, 247 argument 165–187 asyndetic conjunction 290 clitic 166 cliticization 13, 149, 154, 158–159, 166–167 comitative 77–97, 290 committee nouns 195–197 conjunct-disjunct 54–56 controller 78 Coordinate Structure Constraint 84 coordination 287–325 count/mass 133

derogative neuter 256 disjunction 290 double marking 168, 179 emigrant letters 277, 284 Ergativity 30, 32, 34 feminist language reform 268, 276, 285 gender 108 gender assignment 9, 118, 194, 210, 236–239, 243, 245–247, 250–251, 255, 259–262 gender, common 270, 328–329, 336, 339–340, 347–351 – erosion 129–133 – referential 235–236, 239, 262 – pragmatic 238, 245, 255, 258, 261–263 generic masculine 270, 276, 281, 284 generic neuter singular 281 genitival marking 180–183 grammaticalisation 11, 13, 22, 47–49, 76, 161, 165, 182–183, 217–218, 227–229, 320 head- and dependent marking 167–168 hybrid names 237, 241–242, 256, 260–261 hybrid nouns 9, 191, 191–214, 200, 237, 239, 256, 260, 274 hybrids 216–217 – diachrony of 208–210 – extreme 216–232 – full 200–204 – lexical 4–6, 8, 23, 192, 269 – split 205–208 individuation 128, 288, 292, 294–304, 307, 311, 320 inflectional class 108–122 isolates 31, 38, 42 language change 2, 49, 70, 103, 267, 277 mismatch, meaning-form 196

Index of subjects   

   355

normative grammar 290, 321 notae augentes (Old Irish) 169–170 noun classes 7, 118, 148, 333, 347 number agreement 16, 67, 69, 114, 120, 123, 153, 182, 211, 268, 279, 281, 283, 290, 302–303, 314, 316, 318, 350

relative clause 88, 175, 273, 329, 330 relative pronoun 6–7, 17, 130, 193, 197, 201, 203, 206, 239–240, 242, 272–273, 330 resolution 4, 10–11, 16–18, 22–23, 134, 287–288, 291–292, 302–307, 311–314, 318–321

Otfried von Weißenburg 200–203

salience, discourse-related 137–141 schema attikon 15, 20, 341–342, 345–347, 350–351 specifier-head 149, 154, 158 standardisation 267

paradigms 13, 30, 31–36, 38, 41–42, 44–48, 61, 168, 269 politeness 209 polysemy 77 prominence, pragmatic 135–137 – syntactic 134 – thematic 139–141 proper names 235–236, 244, 249, 257, 259 pseudo-coordination 331, 346, 348–350

universals, diachronic 29, 30 word order 290–292, 294–297, 301–304, 307, 310, 314, 318, 320 zeros 30, 32–33, 44–45, 47, 49

quantifiers 19, 148–151, 156, 161–163 reanalysis 42–44, 46–47, 54, 69, 91, 123, 179, 311, 319 reinforcement 225–229