DIFFERENTIAL SUBJECT MARKING
Studies in Natural Language and Linguistic Theory VOLUME 72
Managing Editors Marcel den Dikken, City University of New York Liliane Haegeman, University of Lille Joan Maling, Brandeis University Editorial Board Guglielmo Cinque, University of Venice Carol Georgopoulos, University of Utah Jane Grimshaw, Rutgers University Michael Kenstowicz, Massachusetts Institute of Technology Hilda Koopman, University of California, Los Angeles Howard Lasnik, University of Maryland Alec Marantz, Massachusetts Institute of Technology John J. McCarthy, University of Massachusetts, Amherst Ian Roberts, University of Cambridge
The titles published in this series are listed at the end of this volume.
DIFFERENTIAL SUBJECT MARKING Edited by
HELEN DE HOOP Radboud University, Nijmegen, The Netherlands
and
PETER DE SWART Radboud University, Nijmegen, The Netherlands
A C.I.P. Catalogue record for this book is available from the Library of Congress.
ISBN 978-1-4020-6498-2 (HB) ISBN 978-1-4020-6497-5 (e-book)
Published by Springer, P.O. Box 17, 3300 AA Dordrecht, The Netherlands.
www.springer.com
Printed on acid-free paper
All Rights Reserved © 2008 Springer No part of this work may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, photocopying, microfilming, recording or otherwise, without written permission from the Publisher, with the exception of any material supplied specifically for the purpose of being entered and executed on a computer system, for exclusive use by the purchaser of the work.
TABLE OF CONTENTS Preface Contributors List of Abbreviations 1
2
3
vii ix xi
Cross-linguistic Variation in Differential Subject Marking Helen de Hoop, Peter de Swart
1
Differential Subject Marking at Argument Structure, Syntax, and PF Ellen Woolford
17
Quantitative Variation in Korean Case Ellipsis: Implications for Case Theory Hanjung Lee
41
4
Ergative Case-marking in Hindi Helen de Hoop, Bhuvana Narasimhan
63
5
DOM and Two Types of DSM in Turkish Jaklin Kornfilt
79
6
Differential Subject Marking in Polish: The Case of Genitive vs. Nominative Subjects in “X was not at Y”-constructions Joanna Błaszczak
7
Differential Argument Marking in Two-term Case Systems and its Implications for the General Theory of Case Marking Peter M. Arkadiev
8
Non-canonical Agent Marking in Agul Dmitry Ganenkov, Timur Maisak, Solmaz Merdanova
9
From Topic to Subject Marking: Implications for a Typology of Subject Marking Yukiko Morimoto
v
113
151
173
199
vi
10
TABLE OF CONTENTS
Grammaticalization and Strategies in Resolving Subject Marking Paradoxes: The Case of Tsimshianic Jason Brown and Tyler Peterson
11
Different Subjects, Different Marking Mark Donohue
12
Differential Marking of Intransitive Subjects in Kambera (Austronesian) Marian Klamer
223
247
281
PREFACE This volume has its genesis in the PIONIER Workshop “Differential Subject Marking”, held at the Radboud University Nijmegen in July 2004, and organised by Tessel van Balen and us. We would like to thank all participants of the workshop for making it a success, and the contributors to this volume for their enthusiastic cooperation. Thanks as well to the editors of the series for giving us the opportunity to make a book on this fascinating theme, and to the reviewers of the individual chapters as well as the book manuscript for many useful comments. We gratefully acknowledge the Netherlands Organisation for Scientific research (NWO) for financial support of our research, grants 220-70-003 (PIONIER project “Case crosslinguistically”) and 360-70-220 (“Animacy”). Special thanks go to Luuc van der Horst for his indispensable help in preparing the final manuscript for publication. Helen de Hoop and Peter de Swart Radboud University Nijmegen
vii
CONTRIBUTORS Peter M. Arkadiev Department of Typology and Comparative Linguistics Russian Academy of Sciences, Moscow
[email protected] Joanna Błaszczak Linguistics Department University of Potsdam, Potsdam
[email protected] Jason Brown Department of Linguistics University of British Columbia, Vancouver
[email protected] Mark Donohue School of Languages, Cultures and Linguistic Monash University, Victoria
[email protected] Dmitry Ganenkov Department of Caucasian Languages Institute of Linguistics, Moscow
[email protected] Helen de Hoop Department of Linguistics Radboud University, Nijmegen
[email protected] Marian Klamer Department of Languages and Cultures of SE Asia and Oceania Leiden University, Leiden
[email protected] Jaklin Kornfilt Department of Languages, Literatures, and Linguistics Syracuse University, Syracuse
[email protected] Hanjung Lee Department of English Language and Literature Sungkyunkwan University, Seoul
[email protected] ix
x
LIST OF CONTRIBUTORS
Timur Maisak Institute of Linguistics Russian Academy of Sciences, Moscow Solmaz Merdanova Institute of Linguistics Russian Academy of Sciences, Moscow Yukiko Morimoto Zentrum für Allgemeine Sprachwissenschaft, Berlin
[email protected] Bhuvana Narasimhan Max-Planck-Institute for Psycholinguistics, Nijmegen Tyler Peterson Department of Linguistics University of British Columbia, Vancouver Peter de Swart Department of Linguistics Radboud University, Nijmegen Ellen Woolford Department of Linguistics University of Massachusetts, Amherst
[email protected]
LIST OF ABBREVIATIONS 1 2 3 A ABIL ABL ABS ACC AF AGR AgrP ALL ANTIP APPL ART ART.PERS ASC ASP AUX AV CAUS CNC CN.DET CNJ CNT COL COM.APPL COMIT COMP COND CONV COP CORE CTR DAT DEI DEM DIR
first person second person third person agent abilitative ablative absolutive accusative anti-focus agreement agreement phrase allative antipassive applicative article personal article associative aspect auxiliary actor voice causative common noun connective common noun determiner conjunction contiguous collective comitative applicative comitative complementizer conditional converb copula core case marker of control sentence dative deictic element (space/time) demonstrative direct
DIST DOM DSM DUR DV EMPH ENCL ERG EXCL EXP F FN FP FUT FUTFN FV GEN GENER HABIT I IMP INCEPT INCL INDIC IND INF INS INTENS INTR INTRADIR IPFV ITER KP LK LOC LOC.APPL LV xi
distal differential object marking differential subject marking durative dative voice emphasis marker enclitic ergative exclamation experiencer feminine factive nominalization sentence-final particle future future factive nominalization final vowel genitive generic present habitual irrealis imperative inceptive inclusive indicative individuated infinitive instrumental intensifier intransitive intradirective imperfective iterative aspect marker case phrase linker locative locative applicative locative voice
xii M ModP MOD MSD N NEG NOM NP nP NFN NML NON-IND NPST NumP OBJ OM OBL P PA PART PASS PFPART PFV PL PNC
LIST OF ABBREVIATIONS masculine modifier phrase mood marker deverbal noun (masdar) neuter negation nominative noun phrase little N-phrase non-factive nominalization nominalizer non-individuated non-past number phrase object object marker oblique patient paucal participle nominalizing marker passive perfective participle perfective plural proper noun connective
PN.DET POSS PP PREP PREV PROG PRS PST PVB Q R RED REF REFL REL RES RESPART S SG SI SIM SM SUBJ TOP TEMP TR vP
proper noun determiner possessive P-prefix preposition previous mention progressive present past preverb question marker realis reduplication referential reflexive relativization participle resultative resultative participle intransitive subject singular S-infix simultaneous subject marker subject topic marker temporal transitive little V-phrase
HELEN DE HOOP AND PETER DE SWART
CHAPTER 1 Cross-linguistic variation in differential subject marking
1. INTRODUCTION The phenomenon known as Differential Subject Marking (DSM) may take many forms. First of all, languages differ in which conditions govern DSM. Some languages differentiate their subjects on the basis of the form, such as being a pronoun or not, others on the basis of semantic features such as being a real agent (volitional, in control) or not, and still others distinguish their subjects on the basis of clausal features such as tense/aspect/mood or the main/dependent clause distinction. Secondly, DSM comes in different formal guises: case marking, agreement, inverse systems, voice alternations. Although relatively much is known about the cross-linguistic variation we find in the marking of subjects (see Dixon 1994; Aikhenvald, Dixon, and Onishi 2001 for an overview), relatively little attempt has been made to formalize the facts (see Aissen 1999). The present volume is an attempt to unify formal approaches to language with the typological enterprise. In the spirit of this project both specific case studies of DSM and theoretical approaches to the data are presented. In this introductory chapter, we will demonstrate how the respective chapters fit this general enterprise. 2. DIFFERENTIAL CASE MARKING AND PROMINENCE DISTINCTIONS There are many languages that show differential object marking (DOM) or differential subject marking (DSM) or both. These alternations are often, yet not always, related to a difference in prominence of the arguments. One highly influential OT approach that is developed to deal with argument prominence and differential case marking is Aissen’s harmonic alignment account of DSM and DOM (1999, 2003). The functional motivation behind this type of differential case marking would be to avoid ambiguity as to what is the subject and what is the object in a transitive clause (cf. de Swart 2005). Aissen’s approach accounts for crosslinguistic patterns in which case is assigned to subjects which are low in prominence (and thus less typical subjects). However, a problem for Aissen’s analysis of differential case marking, as pointed out by Woolford (2001) and de Swart (2003), is that it predicts DSM to mirror differential object marking in the sense that case is assigned to those subjects which are low in prominence (and thus less typical 1 H. de Hoop and P. de Swart (eds.), Differential Subject Marking, 1–16. © 2008 Springer.
2
HELEN DE HOOP AND PETER DE SWART
subjects) rather than to those which are high-prominent. In fact, however, DSM can also occur in subjects of intransitive verbs, and certain constraints show up crosslinguistically that seem to require case-marking on typical (e.g., highly agentive) subjects rather than on the atypical ones (cf. de Hoop and Malchukov, 2007). Comrie (1989) claims that the most natural type of transitive construction is one where the transitive subject (A) is high in animacy and definiteness, and the transitive object (P) is lower in animacy and definiteness. However, consider the following example from West Greenlandic, discussed in Næss (2006) (Fortescue 1984, glosses from Cooreman 1994): (1) inuit tuqup-pai people.ABS kill-TR.INDIC.3SG.ERG.3PL ‘He killed the people.’ (2) (inun-nik) tuqut-si-vuq people-INS kill-ANTIP-INTR.INDIC.3SG.ABS ‘He killed people.’ Sentences (1) and (2) show a difference in the definiteness or specificity of the object. In (2) the object is lower in definiteness than in (1). Hence, from Comrie’s point of view, (2) should be a more ‘natural’ transitive construction than (1). But from a morpho-syntactic point of view, only (1) is a transitive sentence. Sentence (2) is called an antipassive construction, and from a morpho-syntactic point of view antipassives, like passives, are intransitive rather than transitive. Generally, in antipassive constructions, the P argument is not realized as a ‘core’ object argument (in (2) it receives instrumental case), it can be omitted, and the verb is morphologically ‘detransitivized’ (this is also witnessed in (2)). Obviously, DOM also occurs without explicit detransitivizing morphology on the verb, but then it is usually not analysed as involving an antipassive construction (cf. Kittilä 2002). Naess (2006) wonders how one can maintain that a typical object in a transitive clause (i.e., its meaning) is low-prominent in definiteness and/or animacy, when at the same time this type of object is often not a grammatical object from a morphosyntactic perspective point of view (i.e., its form), as in antipassives like in (2), but also in object incorporation constructions. It is well-known that in transitive clauses subjects are in general more prominent than objects in terms of animacy and definiteness, but when a subject becomes ‘much more prominent’ than the object, an antipassive construction might be used (cf. Sells 2001). Legendre et al. (1993) provide an Optimality Theoretic (OT) analysis of case marking alternations that are triggered by prominence distinctions in the input argument structure. In their view, prominence distinctions trigger voice alternations and these voice alternations go hand in hand with case marking alternations. They claim that the input of a transitive sentence is an argument structure with two highprominent arguments, a high-prominent subject and a high-prominent object. Thus, unlike Comrie (1989) and Aissen (1999, 2003), Legendre et al. ignore the general pattern that subjects outrank objects in prominence (animacy, definiteness). They represent a transitive input as AP (where the capital letters stand for high-prominent arguments in the input). Accordingly, they argue that a passive sentence is triggered
CROSS-LINGUISTIC VARIATION IN DIFFERENTIAL SUBJECT MARKING
3
by the input aP (that is, with a low-prominent A-argument), while an antipassive sentence is triggered by the input Ap (that is, with a low-prominent object). Thus, in their account the intransitive antipassive form in sentence (2) is the result of optimization of an input Ap, while the transitive sentence (1) would be the optimal expression of the input AP. Hence, if we represent Comrie’s (1989) ‘natural transitive’ as the input Ap, this should always give rise to an antipassive construction in the approach by Legendre et al. (1993). But obviously, although this might give the correct result for the West Greenlandic sentences in (1) and (2), not all inputs where the object is low-prominent (e.g., indefinite and/or inanimate) are optimally realized as antipassive constructions. Indeed, Comrie is right in that in most canonical transitive sentences the subject outranks the object in prominence. Besides, as pointed out by de Hoop (1999) and Sells (2001), determining the input ‘prominence’ of the subject and the object is in fact not an easy matter. Also, Sells (2001) points out that the overt expression of prominence distinctions is not only a matter of case and voice marking, but can also involve word order and intonation. This brings us to an important generalization, one that is not accounted for by Legendre et al. (1993), which is that passives are found mostly in nominativeaccusative languages, while antipassives are found mostly in ergative languages. Hence, while the input AP leads to a transitive output in both nominative-accusative and ergative languages, the input Ap leads to an antipassive construction in ergative languages, and the input aP leads to a passive construction in nominative-accusative languages. But then we may wonder what a prominence distinction in the object leads to in nominative-accusative languages, and what a prominence distinction in the subject leads to in ergative languages. Malchukov (2006) observes that a prominence distinction in the subject may lead to DSM in ergative languages, while a prominence distinction in the object can trigger DOM in nominative-accusative languages. Malchukov claims that it is not a coincidence that both antipassive constructions and DSM are found mostly in ergative languages, while passive formation and DOM are found mostly in nominative-accusative languages. He provides an OT analysis of these co-occurrences, based on the interaction of two basic constraints, a faithfulness constraint that requires the marking of a prominence distinction in the input, and an economy constraint PAIP that penalizes marking of the ‘unmarked’ argument (the ‘unmarked’ argument refers to the nominative argument in nominative-accusative languages and the absolutive argument in ergative languages). Hence, consider the following table (cf. Malchukov 2006): Table 1. The relation between voice and case alternations (Malchukov 2006)
Input A/a prominence distinction P/p prominence distinction
Output in a nominativeaccusative language Active/passive voice alternation Differential object marking
Output in an ergativeabsolutive language Differential subject marking Active/antipassive voice alternation
4
HELEN DE HOOP AND PETER DE SWART
From this table we see that DSM occurs as the result of a prominence distinction in the subject. In ergative languages, the transitive object is the unmarked argument, but the transitive subject can bear ergative case. A subject case alternation is therefore easy to realize. Case morphology can simply be omitted (dropped) in certain circumstances. Hence, we account for languages where a prominence distinction in the subject corresponds to a case alternation (DSM). In nominativeaccusative languages, however, the subject is in the unmarked case. Hence, if a language wants to express a prominence distinction in the subject by a case alternation, this would involve ‘adding’ case morphology, thus violating Malchukov’s economy principle PAIP (Malchukov 2006). However, nominativeaccusative languages can apply the alternative strategy of a voice alternation when they want to express a prominence distinction in the subject. Of course, a voice alternation is not without ‘costs’ either, it will violate another economy constraint. But at least, PAIP is satisfied, and the prominence decrease of the agent is expressed in the output (satisfaction of the faithfulness constraint). The same story holds for prominence distinctions in the object, but now the other way around. In a nominative-accusative language, the object is case-marked, so a prominence distinction can easily be expressed by omitting (dropping) the case morphology from the object (DOM). In an ergative language, however, expressing a prominence distinction through a case alternation would involve ‘adding’ case morphology, since the object of a transitive clause is in the unmarked (absolutive) case. The use of an antipassive construction is an alternative strategy to express a prominence distinction, while satisfying the economy constraint PAIP. To sum up the above discussion, prominence distinctions in the input argument structure may trigger voice or case alternations in the output. Following Malchukov’s (2006) OT analysis, we account for the fact that DSM is the result of a prominence distinction in the subject, and that it is mostly found in ergative languages where a case-marker for subjects is already available. Thus, in ergative languages the subject is generally case marked, which makes it less costly to allow for a case alternation in the subject position, since the case-marking can simply be dropped. Apart from ergative case-marking in ergative languages, we find subject case alternations in certain clausal positions that involve genitive case-marking. Here we would also predict that DSM is an economical way to mark a prominence distinction in the subject, since case morphology can be dropped. Obviously, this is a very general picture. We refer the reader to the remainder of this book for more fine-grained data and analyses of DSM cross-linguistically. In our view, the facts and analyses presented in this volume clearly show that DSM is the result of an interplay of markedness and faithfulness constraints. Above, we argued, following Malchukov (2006), that faithfulness requires prominence distinctions in the input to be expressed in the output, and that dependent on the type of language and markedness considerations, this may result in a case alternation (such as DSM in ergative languages) or a voice alternation (such as passive formation in nominative-accusative languages). In an OT framework, we can account for the fact that DSM and antipassive formation are mostly found in ergative languages, while DOM and passive formation are mostly found in accusative languages (Malchukov 2006). As pointed out by de Hoop and Narasimhan (this
CROSS-LINGUISTIC VARIATION IN DIFFERENTIAL SUBJECT MARKING
5
volume), two functions of case-marking can be distinguished that play a role in natural language. One function of case is to express (‘identify’) certain semantic or pragmatic information, such as agentivity or volitionality. This function of case naturally leads to the case-marking of high-prominent subjects in DSM contexts. However, the ‘identifying’ function of case marking is not necessarily linked to the identification of thematic roles, as there is no one-to-one correspondence between prominence and agentivity, nor between prominence and subjecthood. This is also shown in the studies of Morimoto (this volume) and Donohue (this volume) on instantiations of topic-marking that are independent of thematic roles or grammatical functions. Note by the way that the studies of Morimoto and Donohue deal with agreement rather than case-marking and the functions of case-marking need not be the same as those of agreement, or head-marking. In fact, case-marking and agreement can even be in conflict (Brown and Peterson, this volume). The other function of case is to mark the difference between grammatical functions such as the subject and the object. This ensures case-marking of subjects that are similar to objects in certain respects, and thus leads to case-marking the lowprominent, i.e., inanimate and/or indefinite subjects. An example of this function of case-marking is attested in colloquial Korean (Lee, this volume). Asymmetries between different case-systems can be derived from the different functions of casemarking. For Turkish, Kornfilt (this volume) argues that in nominalized clauses genitive case-marking shows up on high-prominent (specific) subjects, but syntactic constraints may overrule this general pattern. Also, for Polish, it is argued that DSM only occurs in a highly restricted context (Błaszczak, this volume). Thus, we see that DSM is a cross-modular phenomenon, that is not triggered or constrained by semantic or pragmatic features in the input alone. Rather, it is the optimal outcome of a conflict between certain constraints, which can be syntactic, semantic, pragmatic, morphological or phonological in nature (Woolford, this volume). When the case system in a language is relatively rich, we may expect it to be easier to satisfy faithfulness constraints that require prominence distinctions in the input to be expressed in the morpho-syntactic output. However, as shown by Arkadiev (this volume), faithfulness can even be ranked high in languages that have poor case systems. When a case system is relatively rich, however, we can witness even more fine-grained distinctions to become expressed in the output. For example, in Agul a locative case is used to mark an intransitive subject that is potentially agentive but at the same time less agentive than would usually be the case, a rather subtle meaning difference (Ganenkov et al., this volume). Similarly, in Kambera, when the subject of an active intransitive verb is not as much in control as might be expected, accusative case marks the subject (Klamer, this volume). In the next section, we will give an overview of the chapters to follow.
6
HELEN DE HOOP AND PETER DE SWART 3. VARIATION IN DSM
The main question addressed by Ellen Woolford in her chapter on Differential Subject Marking at Argument Structure, Syntax, and PF is whether all DSM effects that involve case marking have the same cause. Woolford’s answer to this question is negative and she argues that because there are many different types of DSM effects with diverse causes, they cannot have a unified theoretical account. Woolford distinguishes four types of DSM effects: 1. DSM effects depending on the lexical selection properties of the particular verbs or the argument structure positions of the arguments; 2. DSM effects triggered by different syntactic contexts such as transitivity alternations; 3. DSM effects that are the result of phonological constraints on the morphological realization of certain case features; 4. DSM effects that involve person/animacy features. For this latter type of DSM effects, Woolford provides an alternative to the analysis in Aissen (1999, 2003), by relating these effects to the general markedness of ergative case. Aissen (1999, 2003) captures the idea of markedness in the interaction between on the one hand iconicity constraints such as ‘Avoid unmarked animate objects’, or to put it differently, ‘Case-mark animate objects’, and on the other hand an economy constraint that penalizes morphological case marking. In a canonical transitive construction, the object is lower than the subject in animacy/definiteness, and thus when the object is animate/definite it is marked (for an object) which means it should be case-marked, while a subject which is inanimate/indefinite is marked (for a subject) which means it should be case-marked as well. Woolford, on the other hand, predicts that the cases at the marked end of the Case Markedness hierarchy are more likely to be morphologically realized. Thus, ergative case is the most likely to be morphologically realized, and nominative the least, in the following hierarchy: ergative > dative > accusative > nominative. Woolford uses independently motivated constraints against morphological realization of ergative case in combination with marked features and feature combinations, such as first and second person pronouns (*ERGATIVE/1ST,2ND, or ‘Avoid ergative case on first and second person pronouns’) or animates (*ERGATIVE/ANIMATE, or ‘Avoid ergative case on animate noun phrases’). If one of these constraints is ranked above the opposing faithfulness constraint that requires morphological expression of ergative case of abstract case features inherited from syntax, ergative case will be blocked for the feature combination referred to in the constraint. For example, when this is the case for the constraint *ERGATIVE/1ST,2ND, a pattern is predicted with ergative case on third person pronouns only, as in Dyirbal. Woolford’s OT analysis is a simpler way to account for these DSM effects than Aissen’s (1999, 2003) model, as it does not require constraint conjunction, and interestingly, it does not require reference to the grammatical functions of subject and object at all. Instead, the effects are produced in terms of case markedness. Woolford’s constraints predict the same type of DSM effects that are predicted in Aissen’s (1999, 2003) model, namely DSM effects that are the mirror image of DOM effects, thus where the low-prominent subjects (in terms of animacy, definiteness, or person) are case-marked and not the high-prominent ones. However,
CROSS-LINGUISTIC VARIATION IN DIFFERENTIAL SUBJECT MARKING
7
Woolford notes that the mirror image pattern between DSM and DOM effects not always holds. Woolford hesitates to augment her OT approach in order to allow it to generate systems with exactly the opposite sort of DSM effects than those expected under her and Aissen’s model. She notes that “before we add this power, we need to know more about the extent of cross-linguistic variation in such DSM effects”. We believe this is exactly what the present volume offers. The chapter by Hanjung Lee on Quantitative Variation in Korean Case Ellipsis: Implications for Case Theory provides us with clear evidence for Aissen’s model of DSM. In Korean, low-prominent subjects (inanimates) get case-marked, while the high-prominent (human/animate) ones can remain caseless (case ellipsis). Case ellipsis in languages such as Japanese and Korean is a multi-factor phenomenon that is affected by a variety of pragmatic, semantic, syntactic, and phonological factors. Lee investigates a relatively unexplored aspect of case ellipsis, namely the parallel between cross-linguistic variation in differential case marking on the basis of animacy and definiteness, and the stylistic variation found within a single language, namely Korean. Person, animacy, and definiteness have categorical effects on case marking systems in various languages, but in colloquial Korean frequency evidence suggests that those hierarchies trigger case ellipsis within the language. The rate of case ellipsis for third person subjects is significantly lower than the rate for first and second person subjects. Human and animate subjects exhibit a higher rate of case ellipsis than inanimate subjects, while the results for direct objects are the converse: human and animate objects are overtly marked by the accusative case marker more often than inanimate objects. Similarly, definiteness interacts with the choice of case-marked or unmarked forms of subjects and direct objects in colloquial Korean. The case ellipsis rate for subject pronouns and proper names is significantly higher than the rate for other subjects, while the case ellipsis rate is significantly lower for object pronouns and proper names than for other object noun phrases. Lee claims that the same marked forms of subjects and objects that are avoided categorically in DSM and DOM languages, that is casemarked high-prominent subjects and case-marked low-prominent objects, also tend to be less frequent in Korean. To sum up, cases are most frequently omitted from objects low in animacy and from subjects high in animacy, and the same holds with respect to person and definiteness features. Thus, Lee’s findings in Korean clearly support the mirror image analysis between DSM and DOM effects as proposed by Aissen (1999, 2003). However, in the two subsequent chapters, on DSM effects in Hindi and Turkish respectively, the mirror image is challenged. Helen de Hoop and Bhuvana Narasimhan in their chapter on Ergative Casemarking in Hindi argue that ergative subjects in Hindi cannot be characterized in terms of low-prominence (atypical subjects), but rather they can be viewed as highprominent arguments (typical subjects). According to de Hoop and Narasimhan, Hindi thus provides clear evidence against the fact that differential case-marking on subjects is always motivated by the need to disambiguate subjects from objects. Rather, ergative case-marking in Hindi seems to mark the high-prominent subjects, hence coding prototypical subject properties. They conclude that apart from a distinguishing function (case-marking to distinguish subjects from objects), case-
8
HELEN DE HOOP AND PETER DE SWART
marking can also have the function of marking high-prominent or typical subjects or objects. In Hindi, subjects of highly transitive perfective predicates are marked by ergative case. In addition, volitional subjects of a small class of intransitive verbs bear ergative case. De Hoop and Narasimhan argue that although the latter subjects are not inherently high-prominent in Hindi, they get ergative case through the mechanism of bidirectional optimization. Jaklin Kornfilt discusses DOM and Two Types of DSM in Turkish. In Turkish, two factors influence DSM: the degree of specificity or referentiality of the subject (in our terms, high-prominence) and clausal type (verbal versus nominal). Kornfilt shows that there are two instances of neutralization of DSM. Firstly, while nonspecific subjects normally lack overt (genitive) case-marking in nominalized clauses, this can be overruled by certain morpho-syntactic constraints, such that nonspecific subjects get overtly case-marked (despite their non-specificity). Secondly, in certain instances, when the genitive case is expected on a specific subject, it does not show up because the syntax of the clause does not license the genitive case. Thus, Kornfilt shows that Turkish has DSM neutralization in both directions: non-specific (low-prominent) subjects which ought to lack overt case-marking can get it when it is required by the syntax, and specific (high-prominent) subjects which ought to bear overt case-marking do not bear it when it is blocked by the syntax. She notes that the first type of neutralization is also found in the domain of DOM in Turkish, but the second type is restricted to DSM. Kornfilt furthermore demonstrates that the neutralization of DOM as well as DSM markers with respect to the semantic feature of specificity takes place due to similar formal requirements. She therefore concludes that not only with respect to the expression of the semantic feature per se are the DOM and DSM phenomena similar (since both high-prominent objects and high-prominent subjects receive overt case-marking), but also the neutralization of DOM and DSM proceeds similarly. At least for Turkish, case-marking is in principle used to mark high-prominent arguments (both subjects and objects), but can be overruled in both directions by syntactic requirements. Kornfilt thus strongly rejects Aissen’s (1999, 2003) mirror-image approach to DOM and DSM. To sum up, both de Hoop and Narasimhan (this volume) and Kornfilt (this volume) show that at least not all DSM effects can be attributed to the distinguishing function of case, which would rather predict less typical, hence low-prominent subjects to receive case-marking, to distinguish them from objects which are generally less prominent than subjects. Instead, both Hindi and Turkish DSM effects can be more adequately accounted for under the view that high-prominent subjects receive overt case-marking, i.e., ergative case in Hindi, and genitive case in Turkish. Similar to the previous chapter, Joanna Błaszczak examines the alternation between genitive and nominative subjects in her chapter on Differential Subject Marking in Polish: The Case of Genitive vs. Nominative Subjects in “X was not at Y”-constructions. The language under examination is Polish. In Polish, the subject of a negated (existential-)locative sentence bears genitive or nominative case depending on the aspectual properties of the verb BE. Thus, the aspectual marking of the verb triggers a subject case alternation, but this is restricted to a specific (negated, existential) context, as in affirmative contexts the subject is marked for nominative case independent of the aspectual marking. Błaszczak shows that the
CROSS-LINGUISTIC VARIATION IN DIFFERENTIAL SUBJECT MARKING
9
DSM effect in this type of intransitive construction is in fact due to a prominence distinction in the argument input, related to the unergative or unaccusative status of the verb. She argues that the unergative and unaccusative frames depend on two factors, namely the aspectual properties of the predicate and the agentivity of the intransitive subject. For the verb być ‘to be’ to count as unaccusative and thus to have a genitive marked subject, it must be perfective and its argument must be nonagentive (hence, a low-prominent subject). The question is where this type of genitive case comes from. Błaszczak shows that the genitive case-marking on the subject of a negated existential-locative clause seems a cross between the usual genitive case-marked objects under negation in Polish, and partitive case triggered by perfective aspect. Given that the existential-locative verb has two arguments, a locative argument generated in the subject position, and a nominal argument generated in the object position, the configuration thus created is similar to a canonical transitive one. Since the genitive of negation in Polish applies to the direct object of negated transitive verbs, it does not come as a surprise that it applies to the ‘subject’ (or rather, object) of the negated existential-locative verb as well. In negated locative sentences with nominative subjects, the subjects are usually human, and they act as the ‘controller’ of the event. Hence, they are more agentive and sit in the canonical subject position of an unergative verb. Therefore, they are not affected by the genitive of negation. Roughly, until this point in the volume we have seen instances of DSM involving overt case-marking of the high-prominent subjects (as in Hindi and Turkish), as well as instances of DSM involving overt case-marking of the lowprominent subjects (as in Korean and Polish). This already suggests that at least with respect to some DSM effects the motivating factor of case-marking cannot be to distinguish the subject from the object, as that would predict only low-prominent (object-like) subjects to be case-marked in a DSM context. In the following chapter this commonly assumed distinguishing function of case-marking is explicitly rejected. Peter M. Arkadiev, in the chapter Differential Argument Marking in Two-term Case Systems and its Implications for the General Theory of Case Marking, claims that although the distinguishing function can still be one of the constraints that regulate case-marking, it is certainly not the primary function of case-marking. He shows that at least in several languages, the function of marking specific semantic or pragmatic information is more important. Arkadiev investigates case-marking in Vafsi, an Iranian language with only two cases: unmarked (direct) and marked (oblique) case. This language exhibits DSM as a tense/aspect split and DOM on the basis of animacy/definiteness of the object. The alleged need to morphosyntactically discriminate between the subject and the object seems to be irrelevant here. Animate and/or definite objects are case-marked and transitive subjects of perfective clauses are case-marked. But because there are only two cases available, we get sentences where both the subject and the object bear (the same) case as well as sentences where both the subject and the object lack case. In fact, exactly the same semantic features trigger case-marking in Hindi/Urdu, but here the subject and the object bear different types of cases, and therefore are more easily distinguished. However, as Arkadiev argues, this difference between Vafsi and Hindi/Urdu is only
10
HELEN DE HOOP AND PETER DE SWART
a consequence of the more or less arbitrary number of core case-markers in the respective languages, but not the result of a different function of case-marking. Arkadiev also discusses Uto-Aztecan languages that have a two-term case system. While the dominant pattern of case-marking is accusative, in subordinate clauses the marking of the embedded subjects may switch to oblique case. In the case alternations discussed, the case marking patterns are the result of two interacting factors: 1) subordinate clauses are headed by verbal nouns triggering possessor case-marking on their subjects, and 2) in these languages there are only two cases on nouns, and therefore nominal possessors (subjects of subclauses) and transitive objects get similar case-marking. But such a situation does not arise in languages where differential case marking coincides with a rich case system. Arkadiev emphasizes that the case inventory of a particular language is the result of various and often conflicting tendencies, many of which are better looked upon as historical ‘accidents’. The data from the two-term case systems support such a view. A rather rich case system is described by Dmitry Ganenkov, Timur Maisak and Solmaz Merdanova in their chapter on Non-canonical Agent Marking in Agul. They present data from Agul, a language from the Lezgic branch of the East Caucasian (Nakh-Daghestanian) family. Agul is an ergative language where the marking of the arguments is clearly semantically motivated, and there are no voice alternations such as passive or antipassive formation. The subject of a transitive verb is marked by ergative case, while the subject of an intransitive verb and the object of a transitive verb are in the absolutive form. There are, however, some minor clause types in Agul where other means of marking subjects and objects are used. Ganenkov et al. present a number of uses of the adelative and adessive cases, which are both originally locative, but have non-locative uses in the constructions described. What all these constructions appear to have in common is that the locative case is used on an agent which lacks certain properties of a typical agent. Thus, we can say that in Agul, low-prominent agents in certain constructions are marked by a locative case. This is not related to the subcategorization frame of a given verb. Rather, the adelative in Agul covers a coherent semantic domain. The participants marked by this case are all agentive (canonically marked by the ergative), but they are not fully agentive. Thus, the adelative (and adessive) case typically marks less typical agents in Agul. For example, in the case of a causative construction, there is a case alternation on the causee. Transitive verbs prefer the adessive/adelative marking on the causee. This may have a syntactic explanation: if the causee of a transitive verb would keep its original ergative case-marking, we would get two noun phrases marked by ergative case, which may be dispreferred. Intransitives prefer the (original) absolutive case-marking on the causee. But when the intransitive verb is clearly agentive (such as to run away), the causee can be marked by the adessive/adelative case as well. The use of the adessive/adelative is not allowed for transitive and intransitive verbs denoting uncontrolled situations, especially when the causee is not human. In those cases the causee must be in the absolutive case. Thus, the use of the adessive/adelative on a causee indicates that we deal with an inherently agentive participant, but at the same time it points at an agentivity decrease.
CROSS-LINGUISTIC VARIATION IN DIFFERENTIAL SUBJECT MARKING
11
To sum up, Ganenkov et al. argue that in Agul the two locative cases (in particular the adelative) can be viewed as a general means to express low agentivity of an agentive participant, one which is just ‘not agentive enough’ to count as completely agentive. We can conclude that this type of DSM is triggered by a prominence distinction in the input, and that here the subject which is agentive but low-prominent is marked by a special, oblique, case. Hence, we seem to deal with a conflict between marking a specific grammatical role (the agent) and marking its low-prominence (decrease in agentivity) and this conflict is resolved by the use of oblique case-marking. The last four chapters of the book deal with head marking (agreement) rather than case-marking. However, we will see that similar considerations play a part in these instantiations of DSM as in the previous chapters. In addition, two of the chapters deal with diachronic changes in subject marking. The chapter by Yukiko Morimoto, entitled From Topic to Subject Marking: Implications for a Typology of Subject Marking, explicitly deals with the conflict between marking the grammatical role and marking a prominence distinction. Morimoto argues that in a subset of Bantu languages there is agreement with the topic of the sentence, rather than with the subject. The topic is the argument which is most ‘salient’ or most prominent in the discourse. Topic agreement is a marked option (as opposed to subject agreement) in that there is no one-to-one correspondence between the agreement morphology and grammatical function – it licenses either the subject or the object. The observed variation in agreement properties across the synchronic Bantu grammars suggests a historical path from topic-based to subject-based agreement. This change can be motivated by the marked status of topic agreement which triggered reanalysis of the preverbal topic as the subject and concomitantly topic agreement as subject agreement. The crucial evidence for topic agreement in Bantu languages comes from a construction known as subject-object reversal. This subject-object reversal is restricted along the dimension of the animacy hierarchy. Predicates where the subject is inanimate and the object is human do not allow subject-object reversal. Also, when the arguments are of equal animacy and there is potential ambiguity in interpretation, the subjectobject reversal reading is not available. But when only one argument can be interpreted as the subject (as in the knife cut the bread), the reversal reading is allowed again. Otherwise, the subject-object reversal is only permitted when the subject outranks the object in animacy. This can probably be related to a general requirement of distinguishability. In these languages which have topic agreement rather than subject agreement, and where we can have preverbal object topics, we still need to be able to distinguish the subject from the object. This is achieved by the animacy restriction on the subject-object reversal. Morimoto suggests that the animacy restriction on subject-object reversal represents a stage in the gradual loss of this construction, that goes hand in hand with the shift from topic to subject agreement. Another result of a diachronic shift is discussed by Jason Brown and Tyler Peterson in their chapter on Grammaticalization and Strategies in Resolving Subject Marking Paradoxes: The Case of Tsimshianic where they present two case studies of ergative/nominative paradoxes. Both the Tsimshian and Salish languages display
12
HELEN DE HOOP AND PETER DE SWART
a (split) ergative pattern in verbal agreement, where the transitive subject is marked distinguishing it from the object and the intransitive subject. The paradox arises in the Salish languages when ergative agreement simultaneously occurs with nominative agreement, and in the Tsimshian languages when ergative agreement cooccurs with a transitive subject that is marked with nominative case. Brown and Peterson hypothesize that the introduction of a nominative/ergative paradox will often be accompanied by a new development of DSM. These phenomena represent a diachronic snapshot of how languages relieve the nominative/ergative paradox through different strategies in DSM. Ergative systems of cross-referencing agreement are considered typologically unusual. Also, while there are several examples of languages that combine an ergative-absolutive pattern for case-marking with a nominative-accusative pattern for agreement (thus, the subject of a transitive verb is in the ergative case, but the verb agrees with the argument in the nominative case), the inverse is rare or even non-existent. However, some of the Tsimshian languages represent this gap, as in subjunctive clauses nominative case is accompanied by ergative agreement. The resolution to this paradox is the neutralization of ergative agreement (replaced by nominative agreement) in subjunctive clauses. An important side-effect of this analysis is the readjustment of the ergative split in Tsimshian. Where the previous split was on the basis of clause type, ergative agreement is now only found in first and second person subjunctive clauses, while third person is nominative in both clause types. The two case studies in this chapter approach the issue of grammaticalization and DSM. They demonstrate that the reorganization of case and agreement morphology into paradoxes will be accompanied by a new paradigm of DSM. In Different Subjects, Different Marking by Mark Donohue it is argued that in Tukang Besi subject marking on the verb follows different paradigms. Tukang Besi is an Austronesian language of Southeast Sulawesi in central Indonesia. The subject of a transitive verb may be marked with different agreement markers, independently of their referential properties and without any relation to the semantics of the predicates. The determining factor is verbal diathesis. The verbal diathesis involves the alternation of the grammatical functions subject and object, hence like a voice alternation, but without demotion of the subject or the object. Case-marking is more consistent than agreement marking, with the nominative case marking the subjects in all main clauses and the genitive marking the subjects in all subordinate clauses. In some transitive sentences the verb shows canonical prefixal agreement for the Aargument, but there is an enclitic that shows agreement with the P. Nominative case also appears with P, while the non-nominative core case appears with P. Word order is freeer in this clause type than in the clause type without P agreement. Donohue argues on the basis of several subjecthood tests that the argument bearing nominative case is always the grammatical subject of the clause. Therefore, the Pargument can be the subject of a transitive sentence. In Tukang Besi the subject is determined pragmatically. The language has a grammatical subject that is clearly related to the pragmatic notion of topic, which is more or less true in most Austronesian languages. This means that roles such as agent or patient cannot be automatically assigned to a particular grammatical
CROSS-LINGUISTIC VARIATION IN DIFFERENTIAL SUBJECT MARKING
13
function (subject or object) just on the basis of their relative position in the argument structure. In a transitive sentence the agent is not necessarily the subject. Hence, more overt morphological marking is required to identify the argument structure in Tukang Besi. It is found that various factors play a part in determining the morphological marking of subjects, as well as the diathesis system of the language. The fact that the determination of the subject in a transitive clause depends on pronominal features and their marking, makes the animacy hierarchy relevant as well. There are transitive sentences with agreement for the P and others lacking this agreement, but in case of a first or second person pronoun, this will always be the subject, and if there are two first or second person pronouns, the subject will always be the P. Donohue claims that Tukang Besi does not have a constraint like Aissen’s *OBJ/AGT (‘Avoid agent objects’), but alternatively we can of course say that this constraint is low-ranked in Tukang Besi. In Tukang Besi we get a DSM system in which a given pronominal agreement paradigm, the S,A prefixes, retain constant reference to the argument they mark, while indicating different grammatical functions (subject or object) depending on the voice construction, direct or inverse, in which they occur. Marian Klamer discusses Differential Marking of Intransitive Subjects in Kambera (Austronesian), one of the indigenous languages spoken in the eastern region of the island of Sumba in Eastern Indonesia. This chapter presents no less than five different ways in which the subject of an intransitive clause in Kambera may be cross-referenced on the verb by pronominal clitics; there is no case-marking on noun phrases in Kambera. The grammatical functions assumed for Kambera are subject (comprising subjects of transitive and intransitive clauses) and object. Objects are only cross-referenced if they are definite. Definites are cross-referenced on the predicate and optionally doubled, indefinite objects are not cross-referenced, and are either left implicit (to be inferred from the context) or expressed as indefinite (bare) noun phrases. The subject of intransitive predicates can be marked in five different ways: 1) nominative; 2) genitive; 3) genitive plus dative; 4) nominative plus accusative; 5) accusative. These different markings are determined by a variety of factors, including the discourse function of a clause, its aspectual properties, and the degree of control the subject has over the event. Yet, none of these subject marking morphemes are selected on the basis of lexical information (selectional criteria) of the verb alone. Both transitive and intransitive subjects are canonically nominative. Nominal clauses have a genitive subject. These clauses have the external syntax of noun phrases, although their internal structure is clausal. Their discourse function is to present the background information to the main narrative line. The particular combination of a genitive and dative clitic expresses continuative aspect. Using a nominative and an accusative pronominal clitic marks epistemic modality: it functions to express the speaker’s belief or expectation, or an obligation. The last type of marking, called accusative marking, marks the subject as the object. Klamer notes that she has found some traces of an earlier ergative alignment system, and that the use of the accusative to mark the subject may still reflect a mixed stage of ergative versus nominative alignment. The subject of a nonverbal intransitive predicate is always accusative. But the accusative marks the subject of an intransitive in a number of verbal contexts too: in imperatives, with
14
HELEN DE HOOP AND PETER DE SWART
generic or impersonal referents, and with stative verbs modified for degree. In all these cases, the subject is not the controller of the situation, but rather a participant that is part of or undergoes it. In addition, the accusative is an option for marking intransitive subjects when they are less agentive than one would normally expect. Regarding the lexical representation of intransitive verbs, Kambera does not present evidence for a formal distinction between classes of intransitive verbs such as unaccusative versus unergative ones. Every intransitive verb may occur with all five different ways of marking, including the accusative one. Also, the role of the subject does not link to the morphological marking directly. Nominative is the default marking, but it can mark agents (as in to run), themes (as in be small) or patients (as in to fall). When the expected control (agentivity) of the subject of an active verb is absent, the accusative is chosen to mark the subject. The choice for the other subject morphemes depends on other sources of information, such as modality, aspect, and discourse. 4. CONCLUSION Aissen’s (1999, 2003) influential OT model of differential case-marking straightforwardly accounts for cross-linguistic patterns in which case is assigned to subjects which are low in prominence (and thus less typical subjects). A problem for this analysis, however, is that it predicts DSM to mirror DOM in the sense that case is assigned to those subjects which are low in prominence (and thus less typical subjects) rather than to those which are high-prominent. In fact, however, DSM effects have different causes (Woolford, this volume). The prediction that only lowprominent subjects are case-marked in DSM is borne out in some languages, such as for example Korean, where low-prominent (inanimate and indefinite) subjects cannot drop their case marking in colloquial speech, while high-prominent subjects can (Lee, this volume), and Polish, where low-prominent (less agentive and objectlike) intransitive subjects are marked with genitive case (Błaszczak, this volume). However, the prediction is falsified in other languages, in which the subjects high in prominence are marked rather than the low-prominent ones. Examples are Hindi where ergative case is used for subjects of perfective highly transitive predicates (de Hoop and Narasimhan, this volume), and Turkish where only specific subjects of nominal clauses bear genitive case (Kornfilt, this volume). Languages differ in the type of marking systems they have at their disposal. In certain languages with a poor case inventory (two cases only), such as Vafsi and Uto-Aztecan languages, case-marking certain semantic or syntactic features of subjects and objects may result in a violation of the constraint that requires subjects and objects of transitive clauses to be marked differently. Hence, we get sentences where both the subject and the object bear (the same) case as well as sentences where both the subject and the object lack case (Arkadiev, this volume). But such a situation does not as easily arise in languages where differential case marking coincides with a rich case system. In Agul, the subject which is agentive but lowprominent is not marked by either ergative or absolutive case, which are the canonical cases for subjects, but by a special, locative case. Hence, a conflict
CROSS-LINGUISTIC VARIATION IN DIFFERENTIAL SUBJECT MARKING
15
between marking a specific grammatical role (the agent) and marking its lowprominence (decrease in agentivity) is resolved by the use of a third type of subject case-marking (Ganenkov et al., this volume). The conflict between marking the grammatical role and marking a prominence distinction results in a shift from topic agreement (prominence marking) to subject agreement (marking of grammatical function) in Bantu languages (Morimoto, this volume). Another result of a diachronic shift is found in the Salish and Tsimshian languages. Here the reorganization of case and agreement morphology is accompanied by a new paradigm of DSM (Brown and Peterson, this volume). In Tukang Besi the agent in a transitive sentence is not necessarily the subject. The subject is pragmatically determined and depends on the verbal diathesis (Donohue, this volume). In Kambera all intransitive predicates, irrespective of lexical considerations, can mark their subjects in five different ways (Klamer, this volume). In our view, this latter type of five-way differential case-marking on intransitive subjects provides the clearest piece of evidence against the suggestion that DSM is always driven by the necessity to morphologically distinguish a transitive subject from a transitive object, firstly because distinguishability is not an issue in intransitive constructions, and secondly because distinguishability alone cannot account for the choice among different overt case-markers. Of course, the constraint of distinguishability (i.e., ‘Avoid ambiguity as to what is the subject and what is the object’) can still trigger DSM in certain contexts, but it definitely differs in strength across languages, and it has to interact with faithfulness constraints (of the type ‘Mark prominence distinctions’) as well as economy constraints (such as, ‘Avoid morphological marking’). We believe that the research reported in this study and the numerous illuminating discussions of DSM in the following chapters provide an important step forwards in our understanding of the complex phenomenon of DSM. As such, we hope that the present volume may function as a source of inspiration for future investigations of DSM. REFERENCES Aikhenvald, A.Y., R.M.W. Dixon and M. Onishi (eds.) (2001). Non-canonical Marking of Subjects and Objects. Amsterdam/Philadelphia: John Benjamins. Aissen, J. (1999). Markedness and subject choice in optimality theory. Natural Language and Linguistic Theory 17, 673-711. Aissen, J. (2003). Differential object marking: iconicity vs. economy. Natural Language and Linguistic Theory 21, 435-483. Comrie, B. (1989). Language Universals and Linguistic Typology. Chicago: University of Chicago Press. Cooreman, A. (1994). A functional typology of antipassives. Voice: Form and Function. Ed. by B. Fox and P.J. Hopper. Amsterdam/Philadelphia: John Benjamins, 49-88. Dixon, R.M.W. (1994). Ergativity. Cambridge: Cambridge University Press. Fortescue, M., 1984. West Greenlandic. London: Croom Helm. de Hoop, H. (1999). Optimal case assignment. Linguistics in the Netherlands 1999. Ed. by R. van Bezooijen and R. Kager. Amsterdam/Philadelphia: John Benjamins, 97-109. de Hoop, H. and A.L. Malchukov (2007). On fluid differential case marking: a bidirectional OT approach. Lingua 117, 1636-1656. Kittilä, S. (2002). Transitivity: Towards a Comprehensive Typology. Turku/Åbo, Åbo Akademis Trykeri.
16
HELEN DE HOOP AND PETER DE SWART
Legendre, G., W. Raymond, and P. Smolensky (1993). An Optimality-Theoretic typology of case and grammatical voice systems. Proceedings of the 19th Meeting of the Berkeley Linguistics Society. Berkeley. Malchukov, A.L. (2006). Transitivity parameters and transitivity alternations: constraining co-variation. Case, Valency and Transitivity: a Cross-linguistic Perspective. Ed. by L. Kulikov, A. L. Malchukov and P. de Swart. Amsterdam: John Benjamins, 330-357. Næss, Å. (2006). Case semantics and the agent-patient opposition. Case, Valency and Transitivity: a Cross-linguistic Perspective. Ed. by L. Kulikov, A. L. Malchukov and P. de Swart. Amsterdam: John Benjamins, 309-327. Sells, P. (2001). Form and function in the typology of grammatical voice systems. Optimality-theoretic Syntax. Ed. by G. Legendre, J. Grimshaw, and S. Vikner. Cambridge (MA): MIT Press, 355-391. de Swart, P. (2003). The Case Mirror. MA-thesis, Radboud University Nijmegen. de Swart, P. (2005). Noun phrase resolution: the correlation between case and ambiguity. Competition and Variation in Natural Languages: the Case for Case. Ed. by M. Amberber and H. de Hoop. Oxford: Elsevier, 205-222. Woolford, E. (2001). Case patterns. Optimality-theoretic Syntax. Ed. by G. Legendre, S. Vikner, and J. Grimshaw. Cambridge (MA): MIT Press, 509-543.
ELLEN WOOLFORD
CHAPTER 2 Differential Subject Marking at Argument Structure, Syntax, and PF
In a broad sense, a language may be said to have Differential Subject Marking (DSM) if some subjects have a different Case, agree differently, or occur in a different position than others. In a narrower sense, such differences are thought of as DSM effects only if they depend on the features of the subject in some way, and in the narrowest sense, DSM effects refer to situations in which only subjects with features toward the lower end of the Person/Animacy Hierarchy (third person, inanimate) are morphologically marked. In this last sense, DSM effects conform to the model in Silverstein (1976) where DSM effects are taken to be the mirror image of DOM (Differential Object Marking) effects where morphological marking of objects is associated with features at the higher end of the Person/Animacy Hierarchy. This paper will focus on DSM effects involving Case. The primary question to be addressed is this: Is there a special type of grammatical rule or principle whose exclusive role is to produce DSM (and DOM) effects, or do DSM effects follow from independently motivated principles of Case Theory? The position to be argued for here is that the DSM effects involving Case have diverse causes; what they have in common is only a descriptive unity, in that they all involve alternations in the Case of subjects. To the extent that can be determined at this point, it appears that DSM effects involving Case follow from independently motivated principles of grammar, and that there is no special type of grammatical rules or principle devoted exclusively to producing DSM effects. Nevertheless, there are still some kinds of DSM effects that we cannot yet account for with existing principles of Case Theory. The primary goal of this paper is to show that distinct DSM effects are associated with each of three different grammatical levels: argument structure (or the vP phase), syntax, and PF (morphological spell-out). Section 1 focuses on DSM effects caused by the use of lexical or inherent Case on some subjects. Such nonstructural Case licensing is associated with θ-marking, which takes place at the level where arguments are mapped to internal and external argument positions and θmarked accordingly. This level was D-structure in earlier versions of the theory (Chomsky 1986), but is now identified as argument structure or the vP phase in the Minimalist Program of Chomsky (2000). Dative subjects are licensed at this level and it is also at this level that ergative Case licensing takes place, under the view that ergative is an inherent Case licensed to external arguments (Woolford 1997, 2006 and the references cited therein). We will see that cross-linguistic differences in the exact range of thematic roles that get mapped to each argument structure position
17 H. de Hoop and P. de Swart (eds.), Differential Subject Marking, 17–40. © 2008 Springer.
18
ELLEN WOOLFORD
produce concomitant differences in exactly which subjects get ergative, dative, or nominative Case. DSM effects involving factors associated with external arguments such as volitionality (e.g. de Hoop and Narasimhan, this volume) should ultimately follow from the principles that determine which arguments get mapped to the external argument position in each language. In section 2, we turn to DSM effects in syntax. Kornfilt (this volume) discusses an interesting type of DSM effect that occurs in nominalized clauses in Turkish. This DSM effect actually parallels DOM effects (rather than behaving as the mirror image of DOM effects as predicted by the Silverstein model). We will see that a very similar DSM effect occurs in Hindi (Bhatt 2006). We then turn to a rather different syntactic DSM effect involving a Case change of the subject from nominative to ergative that occurs only when the object moves out of the VP. This occurs in Inuit/Inuktitut (Bittner 1994) and in Nez Perce (Rude 1982, 1986, 1988; Woolford 1997, to appear), and we will see a third example here from Niuean (Massam 2000). Section 3 describes DSM effects at PF, the level at which abstract Case features (and other features) are spelled out morphologically. It appears that most of the DSM effects that conform to the Silverstein mirror-image pattern are PF effects, involving incomplete spell out of Case features in the presence of marked features such as first and second person. We will see that there are parallel DOM effects where Case is marked only in third person. This sets the stage for section 4, which focuses on the narrowest definition of DSM effects, those which conform to the mirror-image relationship with DOM effects observed by Silverstein (1976). Here I propose a simpler alternative to the formal analysis in Aissen (1999, 2003). I will show that these effects follow from independently motivated PF constraints that regulate the spell-out of combinations of marked features. The morphological spell-out of ergative Case is blocked in combination with marked person features (first or second) in languages such as Dyirbal, a classic markedness effect. The same family of PF constraints accounts for the absence of morphological Case differentiation in first and second person object clitics in Romance languages (Grimshaw 2001). 1. DSM EFFECTS AT ARGUMENT STRUCTURE One source of DSM effects is the use of dative or ergative Case on some subjects and nominative on others. This occurs in languages such as Hindi: (1) a. Siitaa-ko larke pasand the be(PST.M.PL) Sita-DAT boys(NOM) like ‘Sita likes the boys.’ (Mahajan 1991 (7)) b. Siitaa-ne laRkii-ko dekhaa Sita(FEM)-ERG girl-DAT see (PERF.3SG.M) ‘Sita saw the girl.’ (Mahajan 1990:87) c. Siitaa kelaa khaatii thii Sita(NOM.FEM) banana(M) eat(IMP.F) be(PST.F) ‘Sita (habitually) ate bananas.’ (Mahajan 1990:72)
DSM AT ARGUMENT STRUCTURE, SYNTAX, AND PF
19
We see the same sort of DSM pattern in Basque: (2) a. Ni-ri zure oinetako-a-k-Ø gustatzen zaizkit AUX I-DAT your shoes-DET-NOM like ‘I like your shoes.’ (Austin and Lopez 1995:12) b. Gizona-k kurritu du AUX man-ERG ran ‘The man ran.’ (Levin 1989:57) c. Miren-ek atea-Ø ireki du Miren-ERG door-NOM open AUX ‘Miren opened the door.’ (Levin 1989 (20)) d. Atea- Ø ireki da door-NOM open AUX ‘The door opened.’ (Levin 1989 (21)) In section 1.1, we will see how these inherent Cases, dative and ergative, are related to argument structure (vP structure). In section 1.2, we will see that there are differences in the exact range of subjects that is marked with inherent dative and ergative Case cross-linguistically. 1.1 Argument structure and non-structural case The ergative Case, as well as the more predictable instances of the dative Case (on goals and experiencers) are appropriately referred to as inherent Cases because they are inherently associated with a particular θ-position.1 Ergative Case is licensed to the external argument position (where agents occur, along with some range of other thematic roles, depending on the language, as we will see in section 1.2). The external argument is θ-marked by a little/light v head in a vP projection above the VP proper (Hung 1988; Kratzer 1996; Chomsky 1995).
1
See Woolford (1997, 2001, and 2006) for arguments that ergative is an inherent Case, as opposed to a structural Case like nominative or accusative. This work includes references to other literature in support of this view. Unlike structural Cases, the ergative is θ-related, it is preserved under NP Movement to the external subject position, passive and raised subjects never become ergative, and like the dative, ergative may be transparent for the licensing of nominative objects. There is a large body of research indicating that many occurrences of the dative Case are fairly predictable (e.g. Zaenen, Maling, and Thráinsson 1985, Svenonius 2002, Jónsson 2003, and additional work cited in Woolford 2006).
ELLEN WOOLFORD
20 (3)
vP external argument
v
VP V
internal argument
This little v can also license ergative Case to the external argument, following Massam (2002) and Legate (2006) (who implement and update the view from Woolford (1997) that ergative Case is the inherent Case associated with agents). Dative Case marks many goals and experiencers (with cross-linguistic differences to be discussed in section 1.3). There are many parallels in the behavior of the more regular uses of the dative Case and the ergative Case (Woolford 2006); those parallels are consistent with the view of McGinnis (1996, 1998, 2001), developing work by Marantz (1989), that direct argument goals/experiencers are licensed in the specifier position of another little v head, located just above the VP proper, and that this little v may also license dative Case to that argument position (see Woolford 2006 for additional evidence and references in support of this view). I label these two little v heads as vA and vG in the tree below in (4): (4)
vP external argument
vA
vP
exp/goal
vG
VP V
internal argument
We see this inherent dative in true ditransitive constructions, as in the following example from Icelandic:
DSM AT ARGUMENT STRUCTURE, SYNTAX, AND PF
21
(5) Þeir gáfu konunginum ambáttina [Icelandic] they-NOM gave king-the-DAT slave-girl-the-ACC ‘They gave the king the slave-girl.’ (Maling 2002 (44a)) Not all datives are predictable, however. There is another type of non-structural Case that is appropriately called lexical Case because it is lexically selected by particular verbs, prepositions, or adjectives. Icelandic is well-known for having a range of verbs that select different lexical Cases for their subjects, including genitive, dative, and lexical accusative: (6) a. Jóns nýtur ekki lengur John-GEN enjoys not longer ‘John is no longer available.’ b. Bátnum hvolfdi boat-DAT capsized ‘The boat capsized.’ c. Bátinn rak á land boat-ACC drifted to shore ‘The boat drifted to the shore.’
við at (Jónsson 2003 (1c)) (Levin and Simpson 1981 (1b)) (Jónsson 2003 (66a))
Lexical Case is, of course, not restricted to subjects; some Icelandic verbs select a lexical Case for an object: (7) Börnin söknuðu foreldranna parents-GEN children-NOM missed ‘The children missed their parents.’
(Jónsson 1996:105)
Icelandic allows an inherent dative and a lexical dative to co-occur in the same clause, as in the following ditransitive example where the theme takes a lexically selected dative: (8) Ég skilaði henni peningunum I returned her-DAT the money-DAT (Zaenen, Maling, and Thráinsson 1985 (42a)) The alternative transitive version of this sentence, shown below, provides additional evidence that the inherent dative is not directly associated with the goal thematic role, but rather with the vP structure position which is restricted to direct argument goals; in this example, the goal is realized as an indirect argument inside a PP, and it does not take the dative Case: (9) Ég skilaði peningunum til hennar I returned the money(DAT) to her(GEN) (Zaenen, Maling, and Thráinsson 1985 (43a))
22
ELLEN WOOLFORD
When there are only two arguments present in a clause, it is much more difficult to tell whether the dative on an object is inherent or lexical, because the object could be located in either the specifier of the lower little v or inside the VP proper.2 However, when the same verb is observed to take a dative object across many languages, as the verb ‘help’ does, that is one indication that the dative is an inherent dative. (10) Fjölskyldan hjálpaði þér family-NOM helped you-DAT ‘The family helped you.’
(Jónsson 1996:105)
Returning now to subjects, let us turn to the discussion of cross-linguistic differences in DSM patterns that can be traced to differences in the exact range of thematic roles that is mapped to each of the three kinds of direct argument positions in a particular language. 1.2 Cross-linguistic differences in argument structure Ergative Case marks only external arguments, but languages are known to differ as to the exact range of thematic roles that is mapped to the external argument position. This always includes agents, but languages differ as to whether inanimate instruments can be mapped to the external argument position. This is possible in English and in Basque, (11) Giltzak atea ireki zuen key-ERG door(NOM) opened aux ‘The key opened the door.’
(Uriagereka ms (30b))
but not in Hindi (de Hoop and Narasimhan, this volume) nor Japanese (Watai 1996),
2
It is not always obvious from the meaning of a verb whether it will take a dative object. Maling (2002) gives examples of pairs of Icelandic verbs that appear to have the same meaning, and yet only one member of the pair takes a dative object. A reviewer refers us to the work of Svenonius (2002) who proposes that the event structure determines whether a verb will take a dative or an accusative object in Icelandic: “Accusative case in Icelandic is available when the two subevents introduced in a transitive verb phrase are temporally identified with each other, and dative case is available when the two parts are distinct.” (Svenonius 2002, 197). For those unpredictable dative objects that must involve lexical selection by individual verbs, Svenonius proposes that we replace the idea that verbs can directly lexically select the Case of their object, with the idea that verbs lexically select something about their event structure that, in turn, determines that the object will get dative Case.
DSM AT ARGUMENT STRUCTURE, SYNTAX, AND PF (12) *Kagi-ga doa-o aketa key-NOM door-ACC opened ‘The key opened the door.’
23
(Watai 1996:39)
Factors such as volitionality and animacy are clearly important for determining which arguments will be categorized as external arguments (cf. de Hoop and Narasimhan, this volume), but other factors are relevant as well. 1.2.1 Frighten class verbs Jónsson (2003) argues that the well-known generalization that agents never take non-structural Case in Icelandic is shown to hold not only of true agents, but also of “subjects that could be described as agent-like.” Jónsson observes that in psych-verb constructions of the frighten class, as in the examples below, (which initially appear to have a theme subject and an experiencer object), the subject never takes idiosyncratic lexical Case like other themes can. (13) a. Listaverkið hneykslaði marga many-ACC the artwork-NOM shocked ‘The work of art shocked many (people).’ b. Tónlistin róar sjúklingana the music-NOM calms the patients-ACC ‘The music calms the patients.’ (Jónsson 2003 (14b,c)) Jónsson suggests that the reason that the subjects of frighten class verbs share this property with agent subjects is that the subjects of frighten class verbs are agents. However, he notes (personal communication) that such verbs do not passivize, in contrast to true agent subject verbs in Icelandic. Baker (1997:78) suggests that the subject in frighten class constructions, while not actually an agent in a literal sense, is a sort of agent in a fuzzy, prototype sense. If we modernize these analyses, these researchers are arguing that the subject of such frighten class verbs is mapped to the external argument position, even though it is not an agent. Treating the subjects of frighten class verbs as themes, rather than as agents (as Grimshaw 1990 argues is correct), but nevertheless mapping them to the external argument position (because they are causers in some sense, as has often been noted) resolves a longstanding problem of why a theme argument occurs in a structurally higher position than an experiencer argument in these constructions. This fact is puzzling if we assume that word order is determined by individual thematic roles, ordered by the thematic hierarchy. The puzzle disappears if it is the argument structure positions that determine the base order of arguments, but that a particular thematic role such as a theme can be mapped to the external argument position under certain conditions in certain languages. This idea that the theme subject of frighten class experiencer verb constructions is mapped to the external argument position in Icelandic (and other languages) is supported by the fact that some ergative languages such as Basque mark the subjects of frighten class verbs with ergative Case, treating them like external arguments.
24
ELLEN WOOLFORD (14) Mikelek ni haserretu nau Michael-ERG I-NOM angry-PERF aux ‘Michael angered me.’ (Laka, personal communication)
However, not all languages map such causal themes to their external argument position; other ergative languages such as Marathi and Assamese do not. As a result, these languages have no frighten class verbs and instead use alternative constructions to express the same meaning. Instead of a verb like ‘frighten/scare’, Assamese uses a more complex construction that literally means something like ‘feed fear to’ or ‘cause to eat fear’. There is an ergative subject, but it is an argument of the verb ‘feed/cause to eat’.3 (15) a. Gan-tu-e xap-tu-k bhoi khuale song-class-ERG snake-class-DAT fear fed/ate ‘The song frightened the snake.’ b. Xap-e tak bhoi khuai snake-ERG him-DAT fear feed/eat ‘Snakes scare him.’ In place of English verbs such as ‘anger’ or ‘calm’, Assamese uses a light verb construction with the added verb ‘make/do’ which takes an agent subject: (16) a. Gan-tu-e xap-tu-k khɔŋal korile song-class-ERG snake-class-DAT anger made/did ‘The song angered the snake.’ b. Boroxun-e Ram-ok xantɔ korile rain-ERG Ram-DAT calm made/did ‘The rain calmed Ram.’ It thus appears that Assamese has stricter requirements for what can be an external argument than English, Icelandic, or Basque do; Assamese will not allow causal themes to be mapped to the external argument position. 1.2.2 Experiencers as external arguments Jónsson (2003) finds that about half of the Icelandic verbs with experiencer subjects take the dative Case, while the other half are treated as external arguments, which take only nominative Case in Icelandic. He explores several possible ideas for how to distinguish these two classes of verbs, and one of his ideas is that experiencer
3
These conclusions about Marathi and Assamese are based on fieldwork that I conducted with native speakers of these languages at the University of Massachusetts. I would like to thank Vishal Kashyap for all of the Assamese examples in this paper. Note that vowel length is not marked in these examples.
DSM AT ARGUMENT STRUCTURE, SYNTAX, AND PF
25
verbs that represent stronger emotions are more likely to be treated like external arguments, such as the verb ‘despise’.4 (17) Sveinn fyrirlítur svona Sveinn-NOM despises such ‘Sveinn despises such men.’
menn men-ACC (Jónsson 2003 (25a))
There are cross-linguistic differences, however, in what percentage of experiencer subjects are treated as external arguments. In contrast to Icelandic, I found that Assamese treats nearly all experiencer subjects as external arguments, marking them with ergative Case. Even the verb ‘like’, which takes a dative subject in Icelandic and Hindi, takes an ergative subject in Assamese:5 (18) Ram-e bhat Ram-ERG rice-NOM ‘Ram likes rice.’
bhalpai likes
In fact, I found only one verb that marks its experiencer subject with the dative in Assamese: (19) Ram-ok bhat Ram-DAT rice-NOM ‘Ram wants rice.’
lage wants
A full account of DSM effects involving inherent Case marking at argument structure (vP structure) awaits a fuller understanding of how and why different languages map different arguments to each of the three argument structure positions. 2. DSM EFFECTS IN SYNTAX There are well-known situations in which the Case of the subject is determined by its syntactic environment. For example, the subject of an embedded clause in English alternates between nominative or accusative Case, depending on whether the clause is tensed, or tenseless and in an ECM context: (20) a. He expects that she will win. b. He expects her to win. This subject Case alternation fits only the broadest definition of a DSM effect given in the introduction to this paper. In this section, we will focus on some less well-known DSM effects that fit the narrower definitions of DSM effects. The first 4 5
See also Maling (2002) and Svenonius (2002) for discussion of semantic factors that distinguish arguments that take the dative from those that do not. Assamese, like Hindi, marks certain objects with (what looks like) the dative Case. This is a DOM effect. Pronouns and proper names are always so marked when they occur as objects in Assamese.
26
ELLEN WOOLFORD
involves specificity marking of subjects, paralleling the well-known specificity marking of objects in DOM effects. The second involves an alternation between nominative and ergative subjects driven by object shift. 2.1 Specificity marking of subjects Kornfilt (this volume) discusses an interesting DSM effect that parallels specificity marking of objects. In Turkish nominalized clauses, there are minimal pairs such as the following where the subject is marked with genitive Case if it is specific, while non-specific subjects get no morphologically overt Case. (21) a. [köy-ü bir haydut-un bas-tığ-ın]-ı duy-du-m village-ACC a robber-GEN raid-FN-3SG-ACC hear-PST-1SG ‘I heard that a (certain) robber raided the village.’ (specific for all speakers) b. [köy-ü haydut bas-tığ-ın]-ı duy-du-m village-ACC robber raid-FN-3SG-ACC hear-PST-1SG ‘I heard that robbers raided the village.’ (non-specific, generic reading as the only reading) (Kornfilt (this volume) (3a,b)) As Kornfilt points out, this DSM effect is interesting because it parallels the DOM effect in Turkish, rather than forming a mirror-image pattern as one would expect under the Silverstein (1976) model. The fact that specificity marking can apply to subjects as well as to objects illustrates the general point also made in Woolford (1995) that whatever the process is that produces DOM effects, it does not exclusively target objects.6 A similar DSM effect occurs in nominalized clauses in Hindi. The data below appear in Bhatt (2006), credited to Peter Hook (personal communication). Here the specific subject is marked with the genitive -kaa, while the non-specific subject takes a Case which is not marked in overt morphology. (22) a. [yahã: paisõ-kaa mil-naa] mushkil hai here money-GEN find-INF difficult be.PRS ‘Finding the money here is difficult.’ (specific interpretation only) b. [yahã: paise mil-naa] mushkil hai here money find-INF difficult be.PRS ‘Finding/getting money here is difficult.’ (non-specific interpretation available) 6
As to the question of why specificity marking targets objects so much more often than subjects, another general point that Kornfilt makes is relevant: independent factors can effectively ‘neutralize’ a potential DSM effect by preventing the alternation that would make it possible, or at least visible. If DOM effects require the alternation of position inside and outside the VP, as argued in Woolford (1995), building on the work of Diesing (1992), then we will not see similar DSM effects in clauses unless subjects are also free to occur inside and outside the VP in the same language, and if that change in position is associated with a Case change. In Turkish, one often sees non-specific subjects adjacent to the verb, and one might argue that these remain in the VP, but we do not see a DSM effect involving Case because the subject gets nominative Case regardless of its position.
DSM AT ARGUMENT STRUCTURE, SYNTAX, AND PF
27
Bhatt argues that the genitive marked subject has moved to a higher position, paralleling the derivation of specificity marked objects in Hindi. These Turkish and Hindi data fit the second most narrow definition of a DSM effect, because they make reference to the features of the subject. They do not fit the most narrow definition because the features associated with the marked subject are high in the hierarchy, instead of low as the Silverstein model predicts. 2.2 DSM effects related to object shift Let us now turn to a DSM effect that is related to object shift. In many languages, objects with features high on the person/animacy hierarchy (e.g. specific, human) move out of the verb phrase (and may also get a marked Case, producing a DOM effect). In some such languages, there is a corresponding change in the subject Case from nominative to ergative. We will see examples of this kind of DSM effect in Inuit/Inuktitut and in Niuean. 2.2.1 Object shift and ergative subjects in Inuit/Inuktitut In Inuit/Inuktitut, the subject gets ergative Case only when the object moves out of the VP. The following pair of examples from Inuktitut [Qairnirmiut dialect] are taken from Johns (2001).7 (23) a. arnaq anguti-mik taku-juq woman-NOM man-INS.ACC see-INTR.3SG ‘The woman sees the/a man.’ b. arna-up angut taku-jaa woman-ERG man-NOM see-TR.3SG/3SG ‘The woman sees the man.’ (Johns 2001) As in Turkish and other languages discussed in Diesing (1992), Bittner (1994) argues that non-specific objects remain inside the VP, as in the (a) version, while specific objects move out of the VP, as in the (b) version. (24) a. [IP subject b. [IP subject object
[VP object V ]] [VP V ]]
The associated DSM effect, a change from the nominative subject in (a) to an ergative subject in (b), is unusual because it is a DSM effect that depends, not on the features of the subject itself, but rather on the features of the object.
7
In this example, I have altered the gloss of absolutive Case to nominative, in conformity with Bittner (1994). As for the object Case, Johns (2001) maintains that the Case that is traditionally labeled instrumental is really accusative in at least some dialects.
28
ELLEN WOOLFORD
This kind of DSM effect is often considered to belong to the larger domain of transitivity effects, and many would assume that the reason for the above subject Case change is that the (a) version must be intransitive. However, as Hopper and Thompson (1980) point out, two clauses can both be transitive (under the classic definition of having two arguments), and yet one may be more transitive than the other along a different dimension of transitivity, such as one involving the features of the object. That is, a clause with a specific object counts as more transitive than a clause with a non-specific object in their terms. In Niuean (Massam 2000), to which we now turn, this type of DSM effect involves an alternation between two types of ‘transitive’ clauses, one with object shift and one without object shift. 2.2.2 Niuean Object Shift and Ergative Case In Niuean, there is a word order difference in clauses depending on the features of the object, and Massam (2000) argues convincingly that this is a result of object shift. A complication is that both word orders involve VP fronting, so that both word orders are verb initial. When the object is non-specific and thus remains inside the VP, it fronts along with the rest of the VP, producing a VOS word order: V object ] [ subject inu kofe a Sione PST drink coffee NOM Sione ‘Sione drank coffee.’
]]
(25) [VP (26) Ne
[Niuean VOS order]
In contrast, when the object is specific and must shift out of the VP, that object is left behind when the VP fronts, producing a VSO word order: V] [ subject object ]] inu e Sione e kofe [Niuean VSO order] PST drink ERG Sione NOM coffee ‘Sione drank the coffee.’
(27) [VP (28) Ne
There is an associated change in the Case of the subject from nominative to ergative. (The two Case morphemes look alike in the above example because Niuean marks proper names differently than common nouns.) (29) Niuean Case Morphology pronouns/proper names common nouns
Ergative e he
NOM/ACC
a e
Massam (2000) argues that the VOS version of the clause with a nominative subject does not involve incorporation (which would make it an intransitive) as has sometimes been assumed. She shows that the VOS order is possible not only with a
DSM AT ARGUMENT STRUCTURE, SYNTAX, AND PF
29
single non-specific noun, as in the example above, but also when the object is complex, as in the example below where the object is a coordinate structure. (30) Ne
kai [sipi mo e ika mitaki ] a Sione eat chip COMIT ACC fish good NOM Sione ‘Sione ate good fish and chips.’ PST
It is not necessary to claim that the VOS clause is intransitive in order to explain why this subject Case change from nominative to ergative occurs. This Case change associated with object shift is actually predicted by independently motivated Case locality principles developed in Woolford (2003, to appear) to explain superficially unrelated data in other languages. In Woolford (to appear), I argue that the same DSM effect in Nez Perce is due to a locality restriction that comes into effect when two Cases occur in close proximity (that is, both inside or both outside the VP). The intuitive generalization relevant for Niuean is that some languages bar two structural Cases together in the same local area (e.g. phase). We see this effect in languages that prohibit two accusatives inside the VP. I claim that languages such as Inuit, and Niuean bar two structural Cases on arguments that both occur outside the VP. (31) *[DP-nominative DP-accusative [VP]] To avoid this configuration of two structural Cases in such close structural proximity, these languages change the Case pattern so that one of the Cases is not structural. In Inuit and Niuean, this is done by changing the Case of the subject to ergative. Ergative is always potentially available from v, but these languages use ergative only as a ‘last resort’ when they need it to avoid a double structural Case configuration outside the VP. When the object remains inside the VP, the two Cases do not occupy the same local structural domain, so they can both have structural Case and the ergative is not needed. For the details of this approach with respect to Nez Perce, using Optimality Theory to formally capture this notion of ‘last resort’, see Woolford (to appear). To conclude this section, we have now seen examples of two very interesting kinds of DSM effects in syntax. What we have not seen is a syntactic alternation that meets the narrowest definition of a DSM effect, one in which subjects with features low on the Person/Animacy Hierarchy get a morphologically overt Case, while subjects with high features do not. It appears that most DSM effects of this narrowest sort that fit the Silverstein (1976) model are PF effects, to which we now turn. 3. DSM (AND DOM) EFFECTS AT PF PF (or morphological spell-out) is the level where decisions are made concerning the morphological realization of (abstract) features from syntax. These decisions often involve markedness. In fact, the very terms marked and unmarked stem from one
30
ELLEN WOOLFORD
kind of PF effect.8 Plural, for example, is called marked with respect to singular because cross-linguistically, plural is more likely to be morphologically marked than singular. We see this pattern in English: (32)
cat cat+s
singular plural
Similarly, first and second person are more marked than third person and more likely to be realized morphologically. We see this kind of PF effect on agreement, as in Lakhota where first and second person agreement is expressed with overt morphemes, but third singular agreement is not. (33) a. Ø-Ø-kté 3SG-3SG-kill ‘He killed it/him.’ b. ma-yá-kte 1SG-2SG-kill ‘You killed me.’
(Van Valin 1985 (3))
However, not all PF effects favor the spell-out of marked features. The morphological spell-out of marked features may be blocked when they occur in combination with other marked features. We see this effect in English, where gender is not morphologically marked on pronouns when it occurs in combination with marked person features (first and second); but gender is spelled out morphologically in combination with the least marked person, third: (34) first person second person third person
I you he, she, it
no gender marked no gender marked masculine, feminine, neuter marked
In this section, we will look at DSM (and DOM) effects on Case of both types, favoring and blocking the spell-out of marked features. Section 3.1 shows examples of the first type, where only the more marked Cases are spelled out morphologically. Section 3.2 shows examples of the second type, where Case is not spelled out morphologically in combination with marked person/number features. In both types of situations, PF inherits an abstract Case feature from syntax (if one was present in syntax) and these PF effects involve decisions as to whether or not to spell out that abstract Case feature with an overt Case morpheme.9 These PF decisions concerning Case spell-out cannot affect agreement patterns, which depend on abstract Case features in syntax.
8 9
See Battistella (1996) for a nice summary of the history of markedness. I take the restrictive view that abstract and morphological Case are not independent, in the sense that an argument cannot have one abstract Case and a different morphological Case. Morphological Case is just a descriptive term for overt Case morphemes that spell out abstract Case features.
DSM AT ARGUMENT STRUCTURE, SYNTAX, AND PF
31
3.1 Marking the marked cases In examples of the first type of PF effect discussed above, Cases at the more marked end of the Case Markedness Hierarchy are more likely to be morphologically marked. (35) Case Markedness Hierarchy Ergative > Dative > Accusative > Nominative It is common for nominative to be zero marked cross-linguistically, whereas the ergative and dative are usually spelled out morphologically. We see this pattern in Hindi:10 (36) Hindi Case Morphology (Mahajan 1990) ergative -ne dative -ko nominative (zero) However, some languages mark all Cases with overt morphological Case, as in Japanese. (37) Japanese Case Morphology dative -ni accusative -o nominative -ga 3.2 Avoiding marked combinations involving case In examples of the second type of PF effect discussed above, the more marked Cases are not spelled out morphologically when they occur in combination with marked features. We see this second type of PF effect in Marathi, where ergative Case is not morphologically overt when it occurs in combination with marked person features. In Marathi, abstract ergative Case is not overtly realized in combination with first or second person, (38) Mi-Ø acɘvlɘ I(ERG) handwashed(N.3SG) ‘I washed my hands.’
(Comrie 1984 (16))
but ergative Case is morphologically realized in the third person (Comrie 1984).
10
I omit the accusative from this list because there is controversy over which arguments are accusative in Hindi.
32
ELLEN WOOLFORD (39) Ram-ne acɘvlɘ Ram-ERG handwashed(N.3SG) ‘Ram washed his hands.’
(Comrie 1984 (15))
As Comrie points out, we can distinguish instances of zero Case morphology that are ergative from those which are nominative in Marathi by their effect on agreement: ergative subjects do not agree in Marathi, whereas nominative subjects do.11 In the two examples above, the neuter 3sg agreement is default agreement; it does not match in features with either the first person ergative subject in (38), nor with the masculine subject in (39). Similarly, the first person ergative subject in (40) below does not control the agreement, and instead, the agreement is controlled by the plural nominative object. (40) Mi kame keli I(ERG) jobs(NOM) did(N.3PL) ‘I did the jobs.’
(Comrie 1984 (12))
What matters for agreement is abstract Case, regardless of whether that Case is morphologically realized. When the abstract Case of a first person subject is nominative, it looks morphologically identical to a first person ergative subject, but it controls the agreement as any nominative does in Hindi, as we see below in example (41):12 (41) Mi khup goSTi bollo I(NOM) many things(ACC)said(M.1SG) ‘I said many things.’ (Comrie 1984 (14)) The agreement pattern is determined by the abstract Case pattern in syntax, and this is unchanged by whether or not that abstract Case feature is spelled out at PF. Kabardian shows us that this kind of PF effect is not limited to the ergative Case. In Kabardian, we see the same spell-out pattern for ergative, dative, and even nominative Case (Kabardian does not use the accusative Case). The pattern is that third person pronouns are morphologically marked for all Cases, even nominative; in contrast, first and second person pronouns are not morphologically marked for any Case, ergative, dative, or nominative (Colarusso 1992:65-66):
11
Deo and Sharma (2006) confirm this pattern in Marathi, that nominative subjects control agreement but ergative subjects do not, even those first and second person ergative subjects which are not morphologically marked with ergative Case. However, they report that in one dialect of Marathi, Gowari, described in Volume VII of the Linguistic Survey of India (Grierson 1905), the morphologically unmarked first and second person forms really are nominative in abstract Case and they do trigger agreement. 12 The verb in (41) is one that never takes an ergative subject in Marathi even in the perfective aspect.
DSM AT ARGUMENT STRUCTURE, SYNTAX, AND PF
33
(42) Kabardian Pronouns (Colarusso 1992) 1sg 2sg 3sg
Ergative sa wa a-bə
Dative sa wa a-bə
Nominative sa wa a-r
Kabardian also shows what Marathi shows, but in a different way: we cannot simply assume that any argument without overt morphological Case is necessarily nominative. The Romance languages show us that this PF effect in which abstract Case is not morphologically spelled out in first or second person is not limited to subjects. We see this same PF effect on object clitics in Italian, French, and Spanish. These object clitics are morphologically distinguished for Case only in the third person, but not in first or second person (Grimshaw 2001). (43) The third-person Romance clitics (from Grimshaw 2001) him/it (acc) to him/it (DAT) her/it (ACC) to her/it (DAT)
Italian lo gli la le
French le lui la lui
Spanish lo le la le
(44) The Romance clitics in first and second person (from Grimshaw 2001) me (ACC) to me (DAT) you (ACC) to you (DAT)
Italian mi mi ti ti
French me me te te
Spanish me me te te
Since Case is realized on these object clitics only when they have features at the low end of the person hierarchy, this is an example of a DOM effect that goes against the predictions of the Silverstein model, to which we now turn. 4. THE MIRROR-IMAGE MODEL OF DSM AND DOM EFFECTS No discussion of DSM effects would be complete without consideration of the mirror-image model of the relationship between DSM and DOM effects proposed by Silverstein (1976). Under this model, a transitive subject is more likely to be marked with ergative Case if its features are at the unmarked end of the Person/Animacy Hierarchy (e.g. third person), whereas exactly the reverse pattern holds for objects. That is, objects are more likely to be morphologically marked with Case when they have features at the marked end of the hierarchy (e.g. first and second person). Silverstein proposes a functional motivation for this pattern: that subjects and objects need to be marked with morphological Case to flag them when they have
34
ELLEN WOOLFORD
a typical features, to avoid the danger that a subject will be misidentified as an object or vice versa. The features that are said to be typical for subjects and objects under this model are at opposite ends of the Person/Animacy Hierarchy, with first and second person more typical for subjects, and third person more typical for objects. “It is natural for third persons to function as patient (O) and for first and second person to function as agent (A), but not vice-versa. The marked Cases, ergative and accusative, formally express the violations of these principles.” (Silverstein 1976:152)
The basic model is that, for each language, there is a point in the Person/Animacy Hierarchy that divides marked from unmarked transitive subjects, and a (usually different) point in the hierarchy that divides marked from unmarked objects. I will mark these points with S and O, to indicate how these parameters might be set for a particular language: (45) Person/Animacy Hierarchy 1PL > 1SG > 2PL > 2SG > 3HUM.PL > 3HUM.SG > 3ANIM.PL > 3ANIM.SG > S O 3INAN.PL > 3INAN.SG The idea is that transitive subjects with features to the right of the subject point in the hierarchy will be flagged with an ergative case morpheme to indicate that they have features that are atypical for subjects. The placement of the S marker in (45) indicates that all third person subjects will be marked ergative, but no first or second person subjects will be. Objects with features to the left of the object point in the hierarchy will be marked with accusative to flag them as atypical objects. The placement of the O marker in (45) indicates that all human objects will get an overt Case morpheme. Silverstein (1976) points out numerous exceptions to this basic model, which we will discuss below. But first, let us turn to a well-known formalization of the basic model. 4.1 Aissen’s OT formalization of the mirror-image model Aissen (1999, 2003) formalizes the basic mirror-image model using Optimality Theory (Prince and Smolensky 1993). Aissen uses harmonic alignment to express the ideal features of subjects and objects proposed in Silverstein (1976). With the help of constraint conjunction, she then constructs a set of constraints to directly express Silverstein’s idea that subjects with atypical features should have overt Case morphology. For example, the constraint that produces the pattern where only third person subjects are marked ergative is a conjunction of a constraint against third person subjects and a constraint against lacking morphological Case: (46) *SUBJECT/THIRD & *ZERO CASE This constraint penalizes a third person subject that lacks overt Case morphology. To capture different subject marking patterns cross-linguistically, Aissen varies the ranking of such constraints above and below an opposing constraint that penalizes having overt morphological Case: *STRUCTURE(CASE).
DSM AT ARGUMENT STRUCTURE, SYNTAX, AND PF
35
There are several problems with this model as noted by Aissen. Some are related to the use of constraint conjunction, which is a controversial device within Optimality Theory. One problem is that the constraints that are conjoined do not necessarily have independent motivation as constraints on their own. For example, *SUBJECT/THIRD predicts that some languages should prohibit all third person subjects. A second problem is that the model predicts languages with exactly the opposite mirror-image pattern; this follows assuming that one can conjoin *SUBJECT/THIRD with *STRUCTURE(CASE). However, Aissen is correct that Optimality Theory is required to capture the kind of cross-linguistic variation that Silverstein (1976) describes. 4.2 A simpler OT approach In this section, I would like to show that the DSM effects that Silverstein (1976) describes can be produced in a different, simpler way, still using Optimality Theory, but without constraint conjunction. Independently motivated PF constraints of the type we saw in section 3.2 above will do the job. These contextually restricted constraints block the morphological realization of ergative Case when it occurs in combination with marked features. For example, we can formalize the PF constraint against spelling out ergative Case in combination with first and second person (which are often referred to as a group as ‘local’ persons) as follows:13 (47) *ERGATIVE/1ST,2ND
(or more simply, *ERGATIVE/LOCAL)
In this OT approach, the opposing constraint is the PF constraint that requires faithfulness to the Case features inherited from syntax; such faithfulness requires morphological spell-out at PF: (48) FAITH(ERGATIVE)
(or more accurately, MAX(ERGATIVE))
The OT equivalent of marking a point in the Person/Animacy Hierarchy to get different language patterns is to rank the appropriate markedness constraint(s) above this faithfulness constraint. Let’s look at this idea in more detail. 4.3 PF Constraints and DSM Effects To illustrate this model, let us formulate a range of relevant PF constraints that block the spell-out of ergative Case in the presence of various marked features and feature combinations. Let us take the approach of basing these constraints on separate standard universal fixed hierarchies: (49) Person Hierarchy: (50) Number Hierarchy: 13
first > second > third plural > singular
The constraint *ERGATIVE without this contextual restriction also occurs (Woolford 2001).
36
ELLEN WOOLFORD (51) Animacy Hierarchy:
human > animate > inanimate
Using these hierarchies, we can produce a set of contextually restricted markedness constraints that prohibit the spell-out of ergative Case in combination with features at the marked end of one or more of the other hierarchies: (52) PF Constraints14 *ERGATIVE/1ST *ERGATIVE/1ST, 2ND (or *ERGATIVE/LOCAL) *ERGATIVE/1STPL *ERGATIVE/1ST, 2NDPL (or *ERGATIVE/LOCAL.PL) *ERGATIVE/HUMAN *ERGATIVE/ANIMATE If any one of these contextually restricted markedness constraints is ranked above the opposing faithfulness constraint, MAX (ERGATIVE), ergative spell-out will be blocked for the feature combination referred to in the constraint. This allows us to generate a range of language patterns discussed in Silverstein (1976) which conform to the predictions of the mirror-image model: (53) Rankings and Predictions Constraint *ERGATIVE/1ST *ERGATIVE/LOCAL *ERGATIVE/1STPL *ERGATIVE/LOCAL.PL *ERGATIVE/HUMAN *ERGATIVE/ANIMATE
Pattern Predicted erg on 2nd and 3rd erg on 3rd erg on 1sg, 2, 3 erg on 1sg, 2sg, 3 erg on non-human erg on inanimate
Language in Silverstein (1976) Dyirbal Bandjalang Dhirari
If more than one of these constraints is ranked above MAX(ERGATIVE), the result will usually not change, because of the more general constraints often subsume the effect of the less general constraints. Using such contextually restricted markedness constraints at PF is a simpler way to get these DSM effects within Optimality Theory than in the model in Aissen (1999, 2003), as it does not require constraint conjunction. Moreover, it requires no reference to grammatical relations such as subject. As we saw above in section 3.2, this kind of PF effect is not limited to subjects or to ergative Case, but may occur with other Cases and with objects, as we saw above in Kabardian and Romance. To the extent that we see these PF effects more often with ergative Case than with other Cases, it is because ergative is at the marked end of the Case Markedness Hierarchy. We need no special constraints for such DSM effects. 14
Formulating the constraints in this way, anchored at the marked edge of the hierarchy and referring to a range of features starting at that edge (rather than having a constraint *ergative/2nd) eliminates the need to require a fixed hierarchy for these constraints, following De Lacy (2002).
DSM AT ARGUMENT STRUCTURE, SYNTAX, AND PF
37
4.4 Exceptions to the mirror-image model for DSM effects Silverstein notes that the mirror-image pattern does not always hold: “there are numerous ‘holes’ in the pattern, and these mean we have the opportunity for further constraint of the system.” (Silverstein 1976:125)15 A particularly spectacular hole in the pattern can be seen in Aranda (Silverstein 1976:127). At the left end of the hierarchy, the first person singular stands alone in being marked ergative, while at the right end of the hierarchy, inanimate subjects are also marked ergative. Nothing in between is overtly marked ergative. (54) Person/Animacy Hierarchy (marked for Aranda subjects) 1PL > 1SG > 2PL > 2SG > 3HUM.PL > 3HUM.SG > 3ANIM.PL > 3ANIM.SG > 3INAN erg Ø Ø Ø Ø Ø Ø erg Ø This discontinuous pattern is extremely problematic for the assumption that the features associated with ergative marking should make reference to a continuous span of the hierarchy. Silverstein suggests a possible two-step solution to this problem. First, he breaks the hierarchy into two separate hierarchies, one for first and second person pronouns and another for all third person elements. The hierarchy for third person elements is then relatively straightforward, with animates zero marked, and inanimates marked ergative. For the first and second pronouns, Silverstein suggests that the hierarchy be adjusted by moving the first person elements from the top of the hierarchy to the bottom, below second person: (55) Person/Animacy Hierarchy (for local pronouns in Aranda) 2pl > 2sg > 1pl > 1sg (Silverstein1976)) Ø Ø Ø erg However, Silverstein does not maintain that this is the definitive solution to Aranda, but only says that his schematization “at least provides a basis for seeking further information on this” (Silverstein 1976:127). The proper formal treatment of this extremely discontinuous pattern in Aranda depends on the other kinds of discontinuous patterns that occur cross-linguistically. Do these patterns always involve irregular marking for 1st singular (the speaker)? Or can any spot on the hierarchy be exceptional in this way? If the speaker (1st singular) is special, then such exceptions could involve politeness conventions in which the speaker treats himself as a lowly inanimate, consistent with Silverstein’s proposed solution. However, if such exceptions can involve all sorts of points on the hierarchy, then it suggests that contextually restricted faithfulness constraints also exist. If so, there could be an interleaving of contextually restricted faithfulness and markedness constraints at PF that alternatively favor and block the spell-out of ergative Case in combination with different features. The following constraints in the following ranking would produce the Aranda pattern: 15
A reviewer refers us to a recent paper by Filimonova (2005) which points out a range of exceptions to the strict predictions of the hierarchy.
38
ELLEN WOOLFORD (56) *ERGATIVE/1PL >> MAX ERGATIVE/1ST >> *ERGATIVE/ANIMATE >> MAX ERGATIVE
These constraints, ranked in this order, would (in effect) turn ergative spell-out on and off as one travels down the Person/Animacy hierarchy. This would be a formal way within Optimality Theory to produce the effect of adding more subject points to Silverstein’s model; that is, instead of one point dividing the hierarchy into a region where ergative is off versus on, we would now have several points where ergative realization is alternatively turned off and on: (57) Person/Animacy Hierarchy (marked for constraints applying in Aranda) 1PL > 1SG > 2PL > 2SG > 3HUM.PL > 3HUM.SG > 3ANIM.PL > 3ANIM.SG > 3INAN Ø erg Ø Ø Ø Ø Ø Ø erg ↑ ↑ ↑ ↑ However, adding a series of contextually restricted faithfulness constraints would allow this OT model (like that of Aissen 1999, 2003) to generate systems with exactly the opposite sort of pattern of DSM effects than those expected under Silverstein’s model. So before we add the power of contextually restricted faithfulness constraints at PF, we need to know more about the nature of the crosslinguistic variation in such DSM effects at PF. 5. CONCLUSION DSM (Differential Subject Marking) effects involving Case do not constitute a unified phenomenon. They come in diverse types, requiring different kinds of theoretical accounts. In this paper, we have seen DSM effects at three levels of grammar: changes involving lexical/inherent Case associated with argument structure (vP structure), Case changes in subjects depending on syntax, and restrictions on the morphological spell-out of abstract Case at PF. These effects appear to follow from independently motivated principles and constraints that hold at these grammatical levels, and they are not restricted to subjects. We have seen no evidence for special principles or constraints designed to specifically target subjects and produce DSM effects. Of special interest in this domain is the question of the extent to which DSM and DOM effects conform to the mirror-image pattern observed by Silverstein (1976), and how to account for this pattern to the extent that it does hold. Silverstein’s model, as formalized in Aissen (1999, 2003), takes a unified approach to these DSM and DOM effects, with opposing constraints that specifically target subjects or objects, and penalize those with atypical features if they lack morphological Case. In this paper, we have seen evidence that the DSM effects that conform to the mirrorimage model appear to be PF effects (involving only the spell-out of Case), in contrast to the conforming DOM effects which appear to be syntactic (involving object shift). We have seen that these DSM effects at PF follow from independently
DSM AT ARGUMENT STRUCTURE, SYNTAX, AND PF
39
motivated constraints on the spell-out of combinations of marked features, so that no special principles or constraints are needed. Although formulated within Optimality Theory, as in Aissen’s approach, the account proposed here is simpler that Aissen’s in that it does not require constraint conjunction, nor any reference to grammatical relations such as subject. The exceptional patterns discussed by Silverstein (1976) have yet to be fully explored and accounted for, but they are relevant to the question within Optimality Theory of whether the grammar has both contextually restricted faithfulness constraints and contextually restricted markedness constraints. REFERENCES Aissen, J. (1999). Markedness and subject choice in Optimality Theory. Natural Language and Linguistic Theory 17, 673-711. Aissen, J. (2003). Differential object marking: Iconicity vs. economy. Natural Language and Linguistic Theory 21, 435-483. Austin, J. and L. Lopez (1995). Nominative, absolutive and dative languages. NELS 25, 1-15. Baker, M. (1997). Thematic roles and syntactic structure. Elements of Grammar. Ed. by L. Haegeman. Dordrecht: Kluwer Academic Publishers, 73-137. Battistella, E.L. (1996). The Logic of Markedness. New York: Oxford University Press. Bhatt, R. (2006). Variable case marking, argument structure, and interpretation. Paper presented at the University of Connecticut. Bittner, M. (1994). Case, Scope, and Binding. Dordrecht: Kluwer. Chomsky, N. (1986). Knowledge of Language. New York: Praeger. Chomsky, N. (1995). The Minimalist Program. Cambridge (MA): MIT Press. Chomsky, N. (2000). Minimalist inquiries: the framework. Step By Step. Ed. by R. Martin, D. Michaels and J. Uriagereka. Cambridge, (MA): MIT Press, 89-155. Colarusso, J. (1992). A Grammar of the Kabardian Language. Calgary, University of Calgary Press. Comrie, B. (1984). Reflections on verb agreement in Hindi and related languages. Linguistics 22, 857864. De Lacy, P. (2002). The Formal Expression of Markedness. Doctoral dissertation, University of Massachusetts, Amherst. Diesing, M. (1992). Indefinites. Cambridge (MA): MIT Press. Deo, A. and D. Sharma (2006). Typological variation in the ergative morphology of Indo-Aryan languages. Linguistic Typology 10, 369-418. Filimonova, E. (2005). The noun phrase hierarchy and relational marking: Problems and counterevidence. Linguistic Typology 9, 22-113. Grierson, G.A. (1905). Linguistic Survey of India VII: Indo-Aryan Family, Southern Group (Marathi), vol. 7. Delhi: Motilal Banarsidass. Grimshaw, J. (1990). Argument Structure. Cambridge (MA): MIT Press. Grimshaw, J. (2001). Optimal clitic positions and the lexicon in Romance clitic systems. Optimality Theoretic Syntax. Ed. by G. Legendre, J. Grimshaw and S. Vikner. Cambridge (MA): MIT Press, 205-240. de Hoop, H. and B. Narasimhan (this volume). Ergative case-marking in Hindi. Hopper, P.J. and S.A. Thompson (1980). Transitivity in grammar and discourse. Language 56, 251-299. Hung, H. (1988). Derived Verbs and Nominals in Malagasy. MS, McGill University, Montreal. Johns, A. (2001). An inclination towards accusative. Linguistica Atlantica 23, 127-144. Jónsson, J.G. (1996). Clausal Architecture and Case in Icelandic. Doctoral dissertation, University of Massachusetts, Amherst. Jónsson, J.G. (2003). Not so quirky: On subject Case in Icelandic. New Perspectives on Case Theory. Ed. by E. Brandner and H. Zinsmeister. Stanford (CA): CSLI Publications, 127-163. Kornfilt, J. (this volume). DOM and two types of DSM in Turkish. Kratzer, A. (1996). Severing the external argument from its verb. Phrase Structure and the Lexicon. Ed. by J. Rooryck and L. Zaring. Dordrecht: Kluwer Academic Publishers, 109-137.
40
ELLEN WOOLFORD
Legate, J. (2006). Split absolutive. Ergativity: Emerging Issues. Ed. by A. Johns, D. Massam and J. Ndayirague. Dordrecht: Springer, 143-171. Levin, B. (1989). The Basque verbal inventory and configurationality. Configurationality. Ed. by L. Marácz and P. Muysken. Dordrecht: Foris, 39-62. Levin, L. and J. Simpson (1981). Quirky Case and lexical representations of Icelandic verbs. Chicago Linguistics Society 17, 185-196. Mahajan, A. (1990). The A/A-Bar Distinction and Movement Theory. Doctoral dissertation, MIT, Cambridge. Mahajan, A. (1991). Clitic doubling, object agreement and specificity, NELS 21, 263-277. Maling, J. (2002). Verbs with dative objects in Icelandic. Íslenskt mál 24: 31-105. Marantz, A. (1989). Relations and Configurations in Georgian. MS, University of North Carolina. Massam, D. (2000), VSO and VOS: Aspects of Niuean word order. The Syntax of Verb Initial Languages. Ed. by A. Carnie and E. Guilfoyle. Oxford: Oxford University Press. Massam, D. (2002). Fully internal Cases: Surface ergativity can be profound. AFLA 8. MIT Working Papers in Linguistics 44, 185-196. McGinnis, M. (1996). Projection and position. Proceedings of ConSole IV. Ed. by J. Costa, R. Goedemans and R. van der Vijver. Leiden: HIL, 203-220. McGinnis, M. (1998). Case and locality in L-Syntax: Evidence from Georgian. MITWPL 32, 139-158. McGinnis, M. (2001). Semantic and morphological restrictions in experiencer predicates. Proceedings of the 2000 CLA Annual Conference. Ed. by J.T. Jensen and G. van Herk. Cahiers Linguistiques d’Ottawa, Department of Linguistics, University of Ottawa, 245-256. Prince, A. and P. Smolensky. (1993). Optimality: Constraint Interaction in Generative Grammar. RuCCS Technical Report #2, Rutgers University Center For Cognitive Science. Rude, N. (1982). Promotion and topicality of Nez Perce objects. Berkeley Linguistics Society 8, 463-483. Rude, N. (1986). Topicality, transitivity, and the direct object in Nez Perce. IJAL 52, 124-153. Rude, N. (1988). Ergative, passive, and antipassive in Nez Perce. Passive and Voice. Ed. by M. Shibatani. Amsterdam: John Benjamins, 547-560. Silverstein, M. 1976). Hierarchy of features and ergativity. Grammatical Categories in Australian Languages. Ed. by R. M. W. Dixon. New Jersey: Humanities Press, 112-171. Svenonius, P. (2002). Icelandic case and the structure of events. Journal of Comparative Germanic Linguistics 5, 197-225. Uriagereka, J. Government Restrictions and Basque Movement. MS., University of Maryland. Van Valin, R.D. (1985). Case marking and the structure of the Lakhota clause. Grammar Inside and Outside the Clause: Some Approaches to Theory from the Field. Ed. by J. Nichols and A.C. Woodbury. Cambridge: Cambridge University Press, 363-413. Watai, F. (1996). Two subject positions and a functional category predicate. MA Thesis, University of Calgary, Calgary. Woolford, E. (1995). Object agreement in Palauan: Specificity, humanness, economy and optimality. Papers in Optimality Theory. Ed. by J. Beckman, L. Walsh Dickey & S. Urbanczyk. University of Massachusetts Occasional Papers in Linguistics 18. Amherst (MA): GLSA, 655-700. Woolford, E. (1997). Four-way Case systems: Ergative, nominative, objective, and accusative. Natural Language and Linguistic Theory 15, 181-227. Woolford, E. (2001). Case patterns. Optimality Theoretic Syntax. Ed. by G. Legendre, J. Grimshaw and S. Vikner. Cambridge, (MA): MIT Press, 509-543. Woolford, E. (2003). Nominative objects and Case locality. Formal Approaches to Slavic Linguistics 11. Ed. by W. Browne et al. Ann Arbor: Michigan Slavic Publications, 539-568. Woolford, E. (2006). Lexical Case, inherent Case, and argument structure. Linguistic Inquiry 37, 111130. Woolford, E. (to appear). Case locality: Pure domains and object shift. Lingua. Zaenen, A., J. Maling and H. Thráinsson. (1985). Case and grammatical functions: The Icelandic passive. Natural Language and Linguistic Theory 3, 441-483.
HANJUNG LEE
CHAPTER 3 Quantitative Variation in Korean Case Ellipsis: Implications for Case Theory1
It is well-known that hierarchies of person, animacy and definiteness have effects on case marking systems in various languages, where certain classes of subjects and objects are marked, but not others. This paper presents evidence of frequency effects of those hierarchies on case ellipsis in Korean. The two major aims of this paper are the following. First of all, the significance of variable case ellipsis patterns of Korean, as found in the CallFriend Korean corpus (LDC 1996), will be demonstrated when looked at from a functional-typological perspective: variation in case marking between style levels within a single language reflects variation across languages. In a second step, the findings from a comparative study of Korean and other languages are integrated into a coherent theoretical framework – Stochastic Optimality Theory (OT) (Boersma 1998; Boersma and Hayes 2001). It is shown that quantitative patterning found in Korean case ellipsis can be analyzed within the stochastic OT framework in a way analogous to an account of categorical differential case marking effects proposed by Aissen (2003). In this analysis, categorical differential case marking found in various languages is viewed as conventionalization of the same universal pragmatic tendency to mark disharmonic elements, which is also present in the variable case-marking systems of languages like Japanese and Korean. 1. INTRODUCTION In the last two decades, the field of generative linguistics has seen a great deal of research on cross-linguistic variation. Yet there has been a near absence of detailed research in this tradition on within-language variation, such as variation between dialects, sociolects or style levels (‘registers’) of the same language. With a few notable exceptions (e.g., Kroch 1978; Selkirk 1972), work on language-internal variation has received far more attention in dialectology, sociolinguistics and 1
Versions of this paper were presented at the 38th Meeting of the Chicago Linguistic Society, University of Chicago (April 2002), the Gradedness Conference, University of Potsdam (October 2002), and at the PIONIER Workshop on Differential Subject Marking, University of Nijmegen (July 2004). I am grateful for critical feedback to the audiences at all occasions of presentation, two anonymous reviewers, and to Helen de Hoop and Peter de Swart, who gave feedback and advice all along the way. Any errors of fact or interpretation are due to my own shortcomings.
41 H. de Hoop and P. de Swart (eds.), Differential Subject Marking, 41–61. © 2008 Springer.
42
HANJUNG LEE
historical linguistics than in (non-historical) generative linguistics. The present study explores parallels between cross-linguistic variation and stylistic variation2 within a single language, through a comparative study of stylistic variation in case marking in Korean and cross-linguistic patterns in differential case marking. The two major aims of this paper are the following. First of all, the significance of variable case ellipsis patterns of Korean, as found in the CallFriend Korean corpus (LDC 1996), will be demonstrated when looked at from a functionaltypological perspective: variation in case marking between style levels within a single language reflects variation across languages. In a second step, the findings from a comparative study of Korean and other languages are integrated into a coherent theoretical framework – Stochastic Optimality Theory (Boersma 1998; Boersma and Hayes 2001). Finally, implications that the results of the present study have for theories of linguistic variation and case will be discussed. 2. ELLIPSIS OF CASE MARKERS IN JAPANESE AND KOREAN The case study of stylistic variation taken in this work is ellipsis of case markers in Korean. An example of case ellipsis (henceforth CE) is given in (1): (1) Onul-un swul masi-myen Today-TOP alcoholic drink-ACC drink-COND ‘(I) shouldn’t drink today.’
an
toy
NEG AUX
In (1), the object swul ‘alcoholic drink’ appears without the following accusative case marker –ul, which would normally indicate the object of the verb. In colloquial Korean, this kind of ellipsis is extremely frequent. The same phenomenon has been observed in Japanese and studied extensively in the literature (e.g., Hinds 1983, Tsutsui 1984, Masunaga 1987, 1988, Matsuda 1996, Yatabe 1999, Minashima 2001, Fry 2001, D. Lee 2002, K. Lee 2002, among others). Previous studies on the phenomenon of CE in Korean and Japanese have identified a number of linguistic and non-linguistic factors that affect CE.3 They include (i) stylistic and pragmatic factors, (ii) discourse/semantic factors and information status and (iii) other factors such as the grammatical construction type (question, idiom), length of the NP (in syllables), utterance length (in words), and proximity of the NP to the predicate. As the brief discussion of these factors will show, CE is a multi-factor phenomenon. As such, any account of CE that relies only on a single categorical principle or constraint is doomed to inadequacy. Rather, a 2
3
We use the term ‘stylistic’ variation to refer to linguistic variation conditioned by the degree of the formality of communicative settings, modes and genres. Such style factors are distinct from status factors such as age, sex, socioeconomic class and sociolinguistic integration. As Preston (1991) points out, style factors are volatile and read from the immediate environment, whereas status factors are (semi-)permanent or long-term variables which are pre-loaded on linguistic items. It should be noted that not all cases that were counted as CE in previous studies will be counted as instances of CE in this work. Our particular operationalization of cases of CE will be discussed in section 2.1.2 in detail.
QUANTITATIVE VARIATION IN KOREAN CASE ELLIPSIS
43
comprehensive account would have to take into account the interplay among different factors, without privileging one or another factor. Although very interesting, such a broad investigation far exceeds the scope of the present study. Our own purpose is to investigate relatively unexplored aspects of CE, i.e., its parallels to cross-linguistic patterns in differential case marking, and to develop a formal account of this that sheds new light on the issue of variation. Stylistic and pragmatic factors Several studies have shown that rates of CE vary with the formality of the extra-linguistic context and the familiarity among interlocutors. Ko (2000), in her study of ellipsis of the accusative case marker –(l)ul in Korean, examined rates of –(l)ul ellipsis in three different speech styles of discourse – telephone conversations between friends, newsgroup narratives, and TV news broadcast scripts – which vary in terms of the degree of situational and topic formality. She found that there is a close correlation between CE and the style of the discourse: the more formal the discourse is, the lower the rate of use of CE. Lee and Thompson (1985) propose that –(l)ul ellipsis in Korean is subject to what they call the principle of ‘sharedness between communicators.’ According to them, the greater the amount of ‘sharedness’, i.e., shared experience, shared context and shared knowledge and cultural background, the less necessary it is to specify grammatical roles played by arguments. In their analysis, the conditions that allow the ellipsis of -(l)ul can be summarized as follows: (2) Contexts for the ellipsis of -(l)ul (Lee and Thompson 1985) a. Compounds (an incorporation of a direct object to a light verb ha‘do’) (…) ilekhey nul sayngkak ha-ku iss-ta-ku such always thought do-LK exist ‘I always think so.’ b. Pragmatic particles: -man ‘only’, -nun ‘topic’, -to ‘also’, etc. pap-un calchac-a mekna eat meal-TOP well find-LK ‘whether (somebody) is eating well’ c. Grammatically and lexically determined bare nominals ku ke kaps com pat-a tal-la-y that thing price a little receive give ‘please get the price for me.’ C. Lee (1995) proposes a similar analysis of CE in Korean. He contends that an NP form with the accusative case marker -(l)ul signals that the event involved is rather unexpected and attention is paid to the NP. On the other hand, deletion of the case marker occurs when the event involving the NP referent is rather expected. Semantic relationship between an NP argument and the predicate D. Lee (2002) observes that the kind of semantic role that a nominal argument bears affects the
44
HANJUNG LEE
naturalness of CE in Japanese. While subjects and direct objects of transitive verbs may often appear without case markers, NPs which are semantically more restricted normally do not appear without being case-marked. Other grammatical factors It is often claimed in the literature in Japanese that CE correlates significantly with a set of other factors. These factors include the grammatical construction type (question, idiom), length of the NP (in syllables), utterance length (in words), and proximity of the NP to the predicate. See Fry (2001) for a detailed discussion. Discourse /semantic factors and information status It is often claimed that CE in Japanese and Korean is constrained by discourse and semantic factors such as focus and exclusive/exhaustive interpretations. The notion of focus, although not always very well defined, has been claimed to be one of the strongest factors affecting CE. This claim has been formulated in various ways by different researchers: My assumption is that whenever the pertinent NP is “deemphasized” or “defocused”, the case marker can be deleted (Masunaga 1988:147). The ellipsis of the case particles (CP) of an NP-CP is unnatural if the NP-CP conveys the idea of exclusivity (Tsutsui 1984 (cited from Yatabe 1999)). The nominative case particle ga in Japanese cannot be dropped when the expression it marks is focused, i.e., when the expression it marks is interpreted as contrasting with some other object(s) of the same type (Yatabe 1999:79). I suggest that –lul is a focus marker in the sense of the alternative semantics of Rooth (1985) which is elaborated in Valludí and Vilkuna (1998): a set of alternatives for the ‘focused’ constituent is generated as an additional denotation (Ko 2000).
Ko (2000) contends that –lul functions as an operator just like other delimiters such as -man ‘only’ and –to ‘also’ in Korean. The following examples from Ko (2000) illustrate this function of –lul. In (3a), os ‘clothes’ is a ‘shared knowledge’ between the two speakers which does not generate a membership set. In (3b), however, os-ul identifies an element from a ‘wh-set’, analogous to the set of alternatives.
QUANTITATIVE VARIATION IN KOREAN CASE ELLIPSIS
45
(3) a. A: Nana-hanthey os ponay-ess-ni? ‘Did you send the clothes to Nana?’ B: I’m sorry. I forgot to do it. b. A: What’s in the packet? B: Nana-ka os-ul ponay-ess-e. ’Nana sent clothes.’ Other studies in the generative tradition which analyze Korean case markers such as –i/ka and –(l)ul as a morphological realization of underlying informational features include Kim (1990), Choi (1995) and Chae (1999), among others. These studies would predict that CE becomes unacceptable or unnatural when the NP marked by –i/ka and -(l)ul is focused.4 Note that this style of analysis takes the position that ellipsis of the nominative and accusative case markers is essentially the same phenomenon, motivated by the same factor of the de-emphasis of presupposed or unimportant materials. However, as we will show in section 3, CE for subjects and objects does not always pattern in the same way. Surprisingly, as we will show in section 3, CE for subjects and objects shows the opposite pattern with respect to animacy, definiteness and person. On the basis of this finding, we will argue that the notion of focus is not sufficient to account for CE and that the mechanism of interacting markedness hierarchies is independently required. More specifically, the relevant patterns of variation in CE reflect certain universal tendencies across human languages: CE in Korean is subject to the same grammatical constraints such as animacy, definiteness and person that characterize cross-linguistic variation in differential case marking. While the effects of those constraints are far from categorical in Korean, their patterns are robust. Yet this is not clearly predicted and cannot be handled by a system limited by ‘hard’ constraints. 3. CE IN COLLOQUIAL KOREAN SPEECH Hierarchies of person, animacy and definiteness have categorical effects on case marking systems in various languages, where certain classes of subjects and objects are marked, but not others. In this section we present evidence of frequency effects of those hierarchies on case ellipsis in Korean.
4
For direct empirical evidence that CE is constrained by focus, Fry (2001) examined the effect of prosodic focus on CE in the CallHome Japanese (CHJ) corpus, annotated on acoustic as well as syntactic/semantic dimensions. Somewhat surprisingly, Fry (2001) reports that prosodic focus does not have a significant effect on rates of CE. Arguments that are prosodically focused, as measured by sentence-maximum peak F0 values, do not appear with case markers significantly more often than other arguments in Fry’s data. It is not clear whether this disagreement as to the focus effect on CE is due to the apparent difference in the definitions of focus adopted in different studies. It remains to be determined experimentally whether the focus effect on CE is best understood in semantic or prosodic terms.
46
HANJUNG LEE
3.1 Methodology 3.1.1 Data source The data set for the present study is the CallFriend Korean (CFK) corpus of telephone speech collected by the Linguistic Data Consortium (LDC). The 1996 release used in this study consists of 60 unscripted telephone conversations spoken by 200 native speakers of Korean, lasting between 5 and 30 minutes. The CFK conversations are an ideal data source for studying CE because of the natural and elliptical style of spontaneous speech. Casual conversations between close friends or family members like the CFK conversations are usually about interlocutors themselves or entities and events that are familiar to both speakers. The result is a large number of human references and the very frequent occurrence of CE. This makes the CFK data particularly suitable for studying the effects of referential properties of NPs, such as person and animacy, on CE. 3.1.2 Coding Of 60 telephone conversations, 35 conversations were transcribed by two native speakers of Korean. Each transcript was then scanned for all instances of overt subject and object NPs regardless of whether or not they occurred with a case marker. For each token, factors that have been claimed in the literature to affect case marking patterns cross-linguistically were coded (person, animacy and definiteness), in addition to the grammatical function of the NP (subject and object).5 These factors are listed in (4). The reliability of coding indicated by the kappa statistic (Carletta 1996) greater than 0.80 was achieved for all these dimensions. (4) a. b. c. d.
Grammatical function: subject or object Person: first, second or third Animacy: human, animate or inanimate Definiteness: personal pronoun, proper name or other6
The main methodological issue was how to operationalize cases of CE: in other words, what should be counted as CE and what should not. In the present study, we 5
In the present work we examine CE only in subjects and direct objects of transitive verbs which are usually marked with nominative and accusative case markers respectively. These are the two grammatical functions which exhibit the highest rates of CE in the CFK corpus (28.1% and 46.8%, respectively). More oblique functions (e.g., indirect objects of ditransitive verbs, experiencers, goals, and sources) and adjuncts (e.g., expressions of time) were not considered because there are too few cases of CE in our annotations to allow us to make statistically meaningful generalizations about the condition under which CE is most likely to occur. 6 In contrast to English, Korean does not possess articles which grammatically encode the (in)definiteness of nouns. Due to the difficulty of automatically determining the definiteness and specificity of nouns, we did not attempt to differentiate the various kinds of noun phrases which do not belong to the two most definite NP categories, i.e., personal pronouns and proper nouns.
QUANTITATIVE VARIATION IN KOREAN CASE ELLIPSIS
47
define instances of CE in subjects and objects as those cases where subject and object nouns are not overtly marked by case markers. We will not, however, count all cases in which case markers are missing in subject and object NPs as instances of CE. Cases in which subject and object nouns are not marked by case markers but by focus and discourse markers like –(n)un (topic), -to ‘also’, and –lato ‘even’ were not counted as instances of CE. The justification for this was that in these cases nominative and accusative markers cannot appear after the nominal head because of a purely formal requirement coming from the morphology: the focus and discourse markers and the case markers (nominative, accusative and genitive) occupy the same morphological slot in the nominal structure and hence cannot co-occur (Yang 1972; Cho and Sells 1995). This is a morphological process that is distinct from CE, which is conditioned by the degree of formality of the extra-linguistic context. In addition, the light verb construction of the form ‘noun(-ACC) + ha’ (e.g., kongpwu(-lul) ha ‘do study’) was excluded from our annotations. The use of the accusative case marker after the complement noun of the light verb ha is considered optional in all genres and styles of Korean speech and writing. For this reason, the light verb construction was not considered to avoid unnecessary complications. In this way, a total of 1956 overtly expressed subject and object NPs was collected. The breakdown of NPs by the presence and absence of following case markers is given in Table 1. CE indicates the ellipsis (or absence) of case markers (nominative and accusative) following nouns; N-CASE and N-~CASE refer to nouns that are marked by case markers and other particles that are not case markers (e.g., focus and discourse markers) respectively. Table 1. frequency of tokens by the presence and absence of case markers
CE N-case N-~case Total
Subject 355 (28.1%) 489 (38.9%) 416 (33.0%) 1260 (100%)
Direct object 326 (46.8%) 175 (25.2%) 195 (28.0%) 696 (100.0%)
Perhaps the most notable feature of Table 1 is the high rate of CE in direct objects (46.8%), compared to that in subjects (28.1%).7 Given that proximity of the NP to the verb is a significant factor that induces CE in Japanese (Fry 2001), it is tempting to think that direct objects appear without case markers more often than subjects, since they are closer to the verb in the unmarked order in Korean (SOV). I will not attempt to verify this speculation in the present study.
7
The same strong tendency to omit the accusative case marker has been observed in Japanese (Matsuda 1996; Fry 2001; Minashima 2001).
48
HANJUNG LEE
3.2 Results of the statistical analysis Now let us see whether person, animacy and definiteness correlate with CE in subjects and direct objects in the CFK data. 3.2.1 Person effects Table 2 shows the effect of person on CE. The leftmost column gives the three categories of person. We estimate the number of times each person was expressed in non-case marked and case-marked forms. These numbers are given with the rates of CE and case marking. Table 2. Interaction of case marking and person in CFK
Person 1 2 3
Subject CE 47 (52.2%) 51 (46.8%) 257 (39.8%)
Direct object N-NOM 43 (47.8%) 58 (53.2%) 388 (60.2%)
Subject CE 11 (55.0%) 16 (53.3%) 299 (66.3%)
Direct object N-ACC 9 (45.0%) 14 (46.7%) 152 (33.7%)
Examples of CE for local person arguments and third person arguments are given in (5) and (6): (5) CE for third person subject (a) and object (b) kipwukum manhi nayss-e a. sangsa ipen-ey boss this time contribution a lot paid ‘The boss paid a lot of contributions this time.’ b. kensel hoysa-nun namca senho hay prefer construction company-TOP men ‘Construction companies prefer men.’ (6) CE for local person subject (a) and object (b) a. choykun-ey wuli younghwa-lul nemwu cacwu pwa often watch nowadays we movie-ACC too ‘Nowadays we watch too many movies.’ choday hays-e b. Youngmi-ka wuli moim-ey meeting-LOC invitation did Youngmi-NOM us ‘Youngmi invited us to the meeting.’ Although the percentage of first and second person pronouns in spoken conversational Korean is very small,8 the person/subject marking effects are significant. In particular, the rate of CE for third person subjects is significantly lower than the rate for local person (first and second person) subjects (χ2 = 37.55, p < 0.05). In the case of direct objects, the CE rate for third person objects is higher 8
This is due to the strong tendency to omit first and second person references in Korean.
QUANTITATIVE VARIATION IN KOREAN CASE ELLIPSIS
49
than the rate for local person objects, although this difference is not significant (χ2 = 2.98). 3.2.2 Animacy effects Examples of CE for animate arguments and inanimate arguments are given in (7) and (8): (7) CE for animate subject (a) and object (b) a. ipen hanki-nun sensayng-nim kwamok-ul ses kaluchi-si-e course-ACC three teach this semester-TOP teacher ‘The teacher teaches three courses this semester.’ b. Wuri hoysa-to naynyun cikwon camchwuk hay our company-also next year employee reduction do ‘Our company too will reduce the number of employees next year.’ (8) CE for inanimate subject (a) and object (b) a. say cadongcha kilum-ul manhi sopi hay consume new car gas-ACC a lot ‘The new car consumes a lot of gas.’ b. kun kenmwul-i cenlyek tel sse-ya hay less consume must big building-NOM energy ‘Big buildings should consume less energy.’ As Table 3 shows, human and (non-human) animate subjects exhibited the higher rate of CE than inanimate subjects. Once again, the results for direct objects are the converse: human and animate objects are overtly marked by the accusative case marker more often than inanimate objects. For both argument roles, the interaction of case marking and animacy was significant in the CFK data. In particular, the rate of CE for human and animate subjects is significantly higher than the rate for inanimate subjects (χ2 = 26.72, p < 0.05). Conversely, in the case of direct objects, the CE rate for human and animate NPs is significantly lower than the rate for inanimate NPs (χ2 = 25.14, p < 0.05). However, there was no statistically significant difference in the behavior of human arguments and non-human animate arguments for both argument roles. In other words, the data suggest that the distance between elements in the animacy hierarchy (Human > Animate > Inanimate) is not equal; the rate of CE for human NPs is far closer to that of non-human animate NPs than that of inanimate NPs.
50
HANJUNG LEE Table 3. Interaction of case marking and animacy in CFK
Animacy Human Animate Inanimate
CE 221 (52.7%) 68 (36.0%) 66 (28.0%)
Subject N-NOM 224 (47.3%) 126 (64.0%) 170 (72.0%)
Direct object CE N-ACC 88 (54.7%) 73 (45.3%) 35 (50.7%) 34 (49.3%) 203 (74.9%) 68 (25.1%)
These results are consistent with the findings of Fry’s (2001) study of CE in Japanese. Fry found that subject marking and object marking in Japanese exhibit a reversed pattern with respect to animacy, although in his data the animacy effect is significant in subjects but not in objects. 2.2.3 Definiteness effects Examples of CE for strongly definite arguments (pronouns and names) and low definite arguments are given in (9) and (10): (9) CE for strongly definite subject (a) and object (b) a. Ce hywuga-lul daum tal-lo milwu-ess-e-yo I vacation-ACC next month-LOC postponed ‘I (humble) postponed my vacation to next month.’ b. Wuyenhi hakkyo-eyse Hyunswu mannass-e accidentally school-LOC Hyunsoo met ‘(I) came across Hyunsoo at school.’ (10) CE for low definite subject (a) and object (b) a. etten haksayng sihum-ul an machiko nays-e-yo some student exam-ACC not complete submitted ‘Some student submitted the exam without completing it.’ b. gongchay-eyse yeca wuday hapnita women give priority Hiring-LOC ‘We give priority to women in hiring.’ We found that definiteness interacts with the choice of case-marked and unmarked forms of subjects and direct objects in the CFK data. As Table 4 shows, the CE rate for strongly definite subject NPs (pronouns and names) is significantly higher than the rate for other low definite NPs (χ2 = 6.96, p < 0.05). Turning to direct objects, we find stronger results in the opposite direction. In particular, the ECM rate for strongly definite object NPs is significantly lower that the rate for other object NPs (χ2 = 52.6, p < 0.05).
QUANTITATIVE VARIATION IN KOREAN CASE ELLIPSIS
51
Table 4. Interaction of case marking and definiteness in CFK
Definiteness Pronoun Name Other
CE 129 (47.6%) 36 (46.8%) 190 (38.3%)
Subject N-NOM 142 (52.4%) 41 (53.2%) 306 (61.7%)
Direct object CE N-ACC 24 (32.4%) 50 (67.6%) 15 (42.9%) 20 (57.1%) 287 (73.2%) 105 (26.8%)
In sum, the relative frequency of the choice of unmarked forms over casemarked forms in the CFK data increases with subjects high in person, animacy and definiteness9 and objects low in those dimensions, and decreases with low prominence subjects and high prominence objects. These results strongly suggest that the tendency to case-mark non-canonical argument types is not specific to split ergative/accusative languages and that it is also present in the grammars of pure nominative-accusative languages like Korean. In other words, it can be said that in terms of case marking patterns (rather than case feature systems), nominativeaccusative languages like Korean are much more similar to split ergative/accusative languages than generally recognized. 4. AN OPTIMALITY THEORETIC ANALYSIS OF CE This section presents a unified formal analysis of Korean case ellipsis and categorical differential case marking phenomena using Optimality Theory (Prince and Smolensky 1993). We will show that quantitative variation found in the CFK data can be analyzed within the framework of Stochastic Optimality Theory (Boersma and Hayes 2001) in a way analogous to an analysis of differential case marking phenomena that has been proposed by Aissen (2003). 4.1 Categorical effects of prominence hierarchies on case marking Person, animacy and definiteness are important properties in the systems of voice, direction and case marking of a number of languages. In particular, that they have effects on case marking has been demonstrated by work on split case marking phenomena in various languages, where certain classes of subjects and objects are marked, but not others. The standard example of a subject marking split is from Dyirbal, in which first and second person are overtly case-marked when they function as objects, but are not overtly marked as subjects (Dixon 1972, 1979; Silverstein 1976). Conversely, third persons are overtly case-marked when they function as subjects (of transitive clauses), but are not overtly case-marked as objects. Aissen (2003), following Bossong (1985), refers to splits in subject and 9
We can determine the relative weights of each of these three factors by using logistic regression (An excellent explanation of logistic regression is given in Paolillo (2002)). However, I will not discuss a logistic regression analysis because the primary issue here is to investigate factors that correlate significantly with case ellipsis and not to determine the relative strength of their influence.
52
HANJUNG LEE
object marking as differential subject marking (DSM) and differential object marking (DOM) respectively. An intuition which recurs in the literature on DOM and DSM is that they arise from the need to maximally differentiate subject and object. In other words, as emphasized by the functional approach of Silverstein (1976), subjects prototypically have features that are high on the hierarchies of person, animacy and definiteness, while objects prototypically have low features. When objects have high features, they need to be marked as atypical, because they are the ones most likely to be confused with subjects if left unmarked. Conversely, subjects which have low features are the ones most in need of being distinguished from objects because they are the ones most likely to be confused with objects if left unmarked. Dyirbal and other Australian languages exhibit both DSM and DOM. It is also common for languages with overt case marking to exhibit either DSM or DOM, but not both. The former possibility is realized in languages like Fore, where neither personal pronouns nor names may be marked in subject function, but inanimates must be (Scott 1978; Donohue 1999). Case marking systems which exhibit DOM only are found in a wider range of languages. In DOM languages, the split is very often triggered along the lines of animacy and definiteness. These dimensions are often given on referentiality scales, where the cut-off point is taken to vary from language to language: (11) a. Animacy hierarchy: Human > Animate > Inanimate b. Definiteness hierarchy: Personal pronoun > Proper noun > Definite NP > Indefinite specific NP > Non-specific NP (Aissen 2003:437; cf. Silverstein 1976) The general pattern, observed in more than 300 DOM languages, can be summarized as the following implicational universal: if a given object O can be case-marked in language A, then objects which are more prominent than O on one or both of the hierarchies in (18) can also be marked in A. Working within the framework of Optimality Theory (OT) (Prince and Smolensky 1993), Aissen (2003) formalizes this generalization using markedness constraints resulting from the operation of hierarchy alignment10 in conjunction with 10
Harmonic alignment is an operation which combines two scales to form a universally impermutable constraint hierarchy. It was introduced originally by Prince and Smolensky (1993) to express the relation between syllable position and sonority (the more prominent position (the nucleus) attracts segments which are more sonorous, while less prominent positions (the margins) attract less sonorous segments) and later applied to morphosyntax in Aissen’s influential work (Aissen 1999, 2003). The formal definition of harmonic alignment is given below (Prince and Smolensky 1993: 136): Harmonic Alignment: Suppose a binary dimension D1 with a scale X > Y on its element {X, Y}, and another dimension D2 with a scale a > b > … > z on its elements. The harmonic alignment of D1 and D2 is the pair of Harmonic scales: H X : X/a > X/b > … > X/z H Y : Y/z > … > Y/b > Y/a
QUANTITATIVE VARIATION IN KOREAN CASE ELLIPSIS
53
iconicity and economy constraints pertaining to case. The constraint hierarchies which form the basis for her analysis of animacy-sensitive DOM are given in (12): (12) a. *Subj/Inan(imate) >> *Subj/Anim(ate) >> *Subj/Hum(an) b. *Obj/Hum(an) >> *Obj/Anim(ate) >> *Obj/Inan(imate) The high-ranking constraints in (12a,b) penalize inanimate subjects and human objects respectively. Crucially, constraint hierarchies derived by hierarchy alignment, including (12a,b), are universal subhierarchies, which are hypothesized to be present in every grammar in the same relative order and must be interleaved with other constraints. In DSM/DOM languages, marked configurations like those penalized by the high-ranking constraints in (12a,b) are marked or flagged as atypical, but not excluded entirely. Aissen (2003) implements this flagging by the formal operation of constraint conjunction. Conjunction of the iconicity constraint *ZEROC (“Avoid null expression of case feature”) with the subhierarchies in (12) results in two new subhierarchies of iconicity constraints, which have the effect of favoring case marking: (13) Iconicity constraints: a. *Subj/Inan & *Zeroc >> *Subj/Anim & *Zeroc >> *Subj/Hum & *Zeroc b. *Obj/Hum & *Zeroc >> *Obj/Anim & *Zeroc >> *Obj/Inan & *Zeroc The subhierarchy in (13a) penalizes null expression of an inanimate referring subject over null expression of an animate or human referring subject. The subhierarchy in (13b) penalizes null expression of a human referring object over null expression of an animate or inanimate referring object. These hierarchies logically entail implicational universals. That is, they predict that if a human referring subject is case-marked, inanimate or non-human subjects must also be case-marked; and if an inanimate referring object is case-marked, so must be human or non-human animate objects case-marked. The effects of the subhierarchy in (13a) can be clearly seen in a DSM system like that of Fore, where only inanimate subjects must be casemarked, while case marking of animate subjects is (sometimes) possible but not obligatory and human subjects are rarely marked (Scott 1978; Donohue 1999). A wide typological range of object marking splits is also surveyed in Aissen (2003). Her survey includes DOM systems which are predicted by interactions of iconicity constraints including those in (13a,b) and the economy constraint in (14). See Aissen (2003) for a detailed discussion. (14) Economy constraint: *STRUCT(URE)C (“Avoid case structure”)
The constraint alignment is the pair of constraint hierarchies: C X : *X/z >> … >> *X/b >> *X/a C Y : *Y/a >> *Y/b >> … >> *Y/z
54
HANJUNG LEE
In sum, fundamental patterns and limits of variation found in DSM and DOM systems across languages can be accounted for in terms of interactions between two kinds of markedness constraints in OT that are in conflict: (i) the iconicity constraints which can be motivated in terms of maximization of contrasts for ease of comprehension from the point of view of the hearer, and (ii) the economy constraint, which embodies minimization of effort from the point of view of the speaker. 4.2 A Unified OT Account of CE and categorical differential case marking The results of our explorations in the CFK corpus suggest that the same marked forms of subjects and objects that are avoided categorically in DSM/DOM languages, i.e., case-marked forms of high-prominence subjects and unmarked forms of high-prominence objects, tend to be less frequent in Korean than less marked forms that are cross-linguistically favored. This parallel between categorical properties of grammar and usage preferences has already been recognized by Givón (1979) and stated clearly by Bresnan et al. (2001) as follows: “The same categorical phenomena which are attributed to hard grammatical constraints in some languages continue to show up as statistical preferences in other languages”. Bresnan et al. (2001) argues that Stochastic Optimality Theory (StOT) (Boersma 1998; Boersma and Hayes 2001) offers an elegant way to understand this phenomenon, since it allows universal grammatical constraints to emerge as either hard or soft in particular languages. StOT, pioneered by Boersma (1998), is one of a family of probabilistic OT frameworks that can provide a unified account of categorical and variable data (see Boersma and Hayes (2001) for reviews). It differs from standard OT in that it presupposes a continuous scale of constraint rankings rather than a discrete ordinal scale. That is, constraints are not simply ordered, but they have a value on the continuous scale of real numbers. Hence, constraints differ not only in dominance but in distance. Also, in StOT evaluation is stochastic. At each evaluation the value of each constraint is perturbed by temporarily adding to its ranking value random noise drawn from a normal distribution. The value permanently associated with a constraint is called a ranking value, while a constraint’s value in any given evaluation is called the selection point. Boersma and Hayes (2001) exemplify StOT with various empirical test cases of phonological variation, and recently it has been increasingly applied to the study of syntactic variation as well (see Bresnan and Deo 2001, Bresnan et al. 2001, Bresnan and Nikitina 2003, Clark 2004, Dingare 2001, Koontz-Garboden 2001, among others). Below, I will illustrate how the StOT model can be used to describe the patterns and limits of variation in case marking through the analysis of CE in Korean and DOM in Sinhalese (Gair 1970). The StOT analysis throughout incorporates the constraints of Aissen (2003). For reasons of space limitation, I will focus only on the animacy effect on case marking in Korean and Sinhalese. Recall from section 2.2 that animacy-conditioned DOM is a statistical tendency in colloquial Korean: the relative frequency of the choice of unmarked forms over case-marked forms in the CFK data increases with objects that are low in
QUANTITATIVE VARIATION IN KOREAN CASE ELLIPSIS
55
prominence in the animacy hierarchy (see Table 3). In contrast, in Sinhalese, animate referring objects are optionally case-marked, while inanimate objects may not be case-marked (Gair 1970). The comparison of these two languages is very useful because it shows that a stochastic OT grammar based on the constraints laid out above can account for CE and categorical DOM in a unified way. We used the Praat system (Boersma and Weenink 2000), which includes an implementation of the Gradual Learning Algorithm (GLA; Boersma 1998; Boersma and Hayes 2001), to determine the constraint ranking values and output distribution of grammars. The GLA is able to approximate the frequency distributions present in the learning data. For Korean, the GLA was confronted with a frequency distribution shown in table 5. Since there is no evidence from the CFK data that there are significant differences between human and non-human animates (see the discussion in section 3.2), they were collapsed into a single category ‘Hum-Anim’ for simulation purposes. Correspondingly, the two high-ranking iconicity constraints in the subhierarchy in (13b), i.e., *OBJ/HUM & *ZEROC and *OBJ/ANIM & *ZEROC, were collapsed into a single constraint, *OBJ/HUM-ANIM & *ZEROC. Table 5. Simulated Korean input/output distribution
Input Human-Animate Inanimate
% OBJ 53.5 74.9
% OBJ-CASE 46.5 25.1
Starting from an initial grammar in which all constraints had the same ranking values (=100.00), the GLA assigned the ranking values in table 6.11 Observe that the ranking value of the economy constraint *STRUCTC is close enough to the ranking values of the two iconicity constraints (within the near-categorical distance of 10; see fn. 8) to predict variable outputs for both inputs. Table 6. Ranking values for constraints (Korean)
*STRUCTC *OBJ/HUM-ANIM & *ZEROC *OBJ/INAN & *ZEROC
100.782 100.430 98.782
The output distributions of the learned grammar closely matched the frequency distributions in the CFK data shown in table 7:
11
The standard deviation of ranking variation here is fixed at 2.0. When the ranking distance between two crucial constraints is 10 units (= 5 standard deviations), the outputs are near-categorical; when the distance is 18 units (= 9 standard deviations), reversal of constraint dominance reduces to 1 in 10 billion (less than once in a lifetime) (Boersma and Hayes 2001:50).
56
HANJUNG LEE Table 7. Output distributions of grammar (Korean)
% OBJ 54.99 74.10
Input Human-Animate Inanimate
% OBJ-CASE 45.01 24.30
For Sinhalese, the GLA was presented with the frequency distribution given in table 8. (This is not based on real data.) We assume 0%/100% frequency for humans and inanimates and 50%/50% frequency for non-human animates. That is, in the learning data 100% of the outputs for the [human] input were case-marked; 100% of the outputs for the [inanimate] input were non-case marked; and 50% of the outputs for the [animate] input were case-marked (the other 50% were not). Table 8. Simulated Sinhalese input/output distribution
% OBJ 0 50 100
Input Human Animate Inanimate
% OBJ-CASE 100 50 0
The GLA assigned the ranking values in table 9. Observe that the iconicity constraint *OBJ/HUM & *ZEROC, penalizing the non-case marked form of a humanreferring object, is more than 10 units above the economy constraint *STRUCTC. This ranking yields the obligatory case marking of a human-referring object. In contrast, the iconicity constraint for a non-human animate object is ranked very close to *STRUCTC. The close ranking distance between these two constraints yields the alternative outputs (unmarked and case-marked forms of an animate object) through frequent constraint dominance reversal under stochastic evaluation. Sinhalese also differs crucially from the partial Korean grammar in table 6 in that the *STRUCTC is more than 10 units above *OBJ/INAN & *ZEROC . This ranking yields the (near-) categorical unmarked output for the [inanimate] input, capturing the categorical influence of animacy on object marking. Table 9. Ranking values for constraints (Sinhalese)
*OBJ/HUM & *ZEROC *OBJ/ANIM & *ZEROC *STRUCTC *OBJ/INAN & *ZEROC
114.059 99.481 99.354 87.107
Once again, the output distributions of the learned grammar approximate the frequency distributions present in the learning data:
QUANTITATIVE VARIATION IN KOREAN CASE ELLIPSIS
57
Table 10. Output distributions of grammar (Sinhalese)
Input Human Animate Inanimate
% OBJ 0.0 48.6 100.0
% OBJ-CASE 100.0 51.4 0.0
In sum, OT grammar with stochastic evaluation can provide a unifying account for variable case marking in Korean and cross-linguistic categorical DSM/DOM phenomena in terms of the same system of iconicity and economy constraints that are typologically motivated. This exemplifies the overarching point that variation within grammars can reflect variation between grammars (Bach 1974; Kortmann 1999; Bresnan and Deo 2001). 5. DISCUSSION: STYLISTIC VARIATION IN OT This section discusses implications that the results of the present study have for accounts of stylistic variation and theories of case. A fundamental difference between CE in Korean and DCM in other languages is that the former is stylistically conditioned. StOT seems quite promising for modelling this difference. In particular, in StOT the rankings of groups of grammatical constraints may covary with the speech style. This covariation can be formally modelled by the following equation (Boersma and Hayes 2001:83): (15) selection Pointi = ranking Valuei + styleSensitivityi · Style + noise Here Style is a continuous variable ranging from 0 (in maximally casual speech) to 1 (in maximally formal speech). styleSensitivity is a constraint-specific value: the ranking of constraints with positive values is boosted in formal speech, while the ranking of constraints with negative values is boosted in casual speech; constraints with zero values of styleSensitivity are stylistically neutral. For instance, in the case of Korean, the relative frequencies of case-marked and unmarked object forms in different speech styles can be modelled by the movement in strength of the iconicity constraints and the economy constraint pertaining to case along the ranking scale. That is, in formal speech, the iconicity constraints, favoring case marking, have positive values for styleSensitivity and their rankings are boosted, so case-marked forms would prevail over unmarked ones. In maximally formal style, the evaluation constraint ranking for Korean would look like the regular ranking in languages like Dhalandji, where all objects are case-marked (Austin 1981). In casual speech, on the other hand, the ranking of the economy constraint, penalizing complexity of case structure, is boosted, so that it is ranked quite close to the iconicity constraints (see table 6). As a result, variation between style levels within a single language samples the typological space of possible grammars. The StOT model is in many ways a more plausible model for stylistic variation than other models prevalent in pre-OT generative grammar and standard OT.
58
HANJUNG LEE
First, in StOT smooth changes in the relative frequencies of variants (e.g., casemarked and unmarked NP forms) in different styles can be captured by small incremental shifts in the distance between (groups of) constraints. Thus, under StOT, we can model style shifting and the process of historical change (Clark 2004) in a unified way by the movement in strength of constraints along the continuous ranking scale. Secondly, StOT offers a restrictive and explanatory model of stylistic variation. Many types of odd styles (those that categorically violate universal subhierarchies of markedness constraints) are not predicted and learnable under the theory. Any models where markedness constraints are not part of the grammar would fail to make this general prediction. Finally, the stochastic grammar is more economical than that of standard ordinal OT (e.g., van Oostendorp 1997; Legendre 1998). For example, in the OT approach of van Oostendorp (1997), style levels (or ‘registers’) all constitute subtly different grammars of their own. Although he passes over the quantitative aspect, his approach is illustrative of a grammar competition model prevalent in generative grammar (Kroch 2000). In StOT, stylistic variation arises in a single grammar without having to posit competing grammars and radical rerankings of constraints to account for style shifting. In our work in progress, we compared the frequency pattern in CFK with that in newsgroup narratives available from the internet server Chollian (http://www.chol.com). Newsgroup narratives offer a nice midway point between casual and formal style. As Table 5 shows, argument phrases tend to get casemarked more frequently in newsgroup narratives, but the reversal of relative frequency of CE in subjects and objects continues: cases are most frequently ellipsed on objects low in animacy and on subjects high in animacy. Table 11. Interaction of CE and animacy in newsgroup narratives
Animacy Human Animate Inanimate
CE 95 (33.0%) 28 (31.0%) 39 (18.0%)
Subject N-NOM 194 (67.0%) 61 (69.0%) 178 (82.0%)
Direct object CE N-ACC 20 (27.6%) 51 (72.4%) 17 (29.1%) 42 (70.9%) 97 (47.2%) 109 (52.8%)
These results confirm that language-internal probabilistic variation in case marking is constrained by the same grammatical factors that govern categorical differential case marking across languages. It remains to show whether the different stylistic data could indeed be predicted using actual style sensitivity values. 6. CONCLUSION This paper has presented evidence of frequency effects of hierarchies of person, animacy and definiteness on case ellipsis in Korean. I have shown that quantitative patterning found in Korean case ellipsis can be analyzed within the stochastic OT
QUANTITATIVE VARIATION IN KOREAN CASE ELLIPSIS
59
framework in a way analogous to an account of the categorical differential case marking phenomena proposed by Aissen (2003). In this analysis, categorical split case marking found in various languages is viewed as conventionalization of the same universal pragmatic tendency to mark disharmonic elements (e.g., highprominence objects and low-prominence subjects), which is also present in the variable case-marking systems of languages like Japanese and Korean.12 During the last three decades, the field of linguistics has shown a steady evolution towards fragmentation, resulting in a number of competing theories of grammar and variation from different disciplines, with language-internal variation becoming the object of sociolinguistics and historical linguistics. The present study of variable case marking patterns in Korean shows that we cannot fully understand even the patterns and limits of variation found in a single phenomenon without joint research efforts by different disciplines of linguistics. It is to be hoped that a broadening of the perspective in recent generative linguistics and pragmatics and a much improved data situation will allow us to make substantial advances in synthesizing the results of these distinct disciplines and integrating them into a larger theoretical frame. 7. REFERENCES Aissen, J. (1999). Markedness and subject choice in Optimality Theory. Natural Language and Linguistic Theory 17, 673-711. Aissen, J. (2003). Differential object marking: iconicity vs. economy. Natural Language and Linguistic Theory 21, 435-483. Austin, P. (1981). Case marking in southern Pilbara languages. Australian Journal of Linguistics 1, 211226. Bach, E. (1974). Syntactic Theory. New York: Holt, Rinehart and Winston. Boersma, P. (1998). Functional Phonology. The Hague: Holland Academic Graphics. Boersma, P. and B. Hayes (2001). Empirical tests of the gradual learning algorithm. Linguistic Inquiry 32, 45-86. Boersma, P. and D. Weenink (2000). Praat computer program. Online, Institute of Phonetic Sciences, University of Amsterdam: http://www.fon.hum.uva.nl/praat. Bossong, G. (1985). Differentielle Objektmarkierung in der Neuiranischen Sprachen. Tübingen: Gunter Narr Verlag. Bresnan, J. and A. Deo (2001). Grammatical constraints on variation: ‘be’. The Survey of English Dialects and (Stochastic) Optimality Theory. Ms., Stanford University. Bresnan, J., S. Dingare and C. Manning (2001). Soft constraints mirror hard constraints: voice and person in English and Lummi. Proceedings of the LFG 01 Conference. Ed. by M. Butt and T. Holloway King. Online, CSLI Publications: http://www-csli.stanford.edu/publications. Bresnan, J. and T. Nikitina (2003). On the Gradience of the Dative Alternation. Ms., Stanford University. Carletta, J. (1996). Assessing agreement on classification tasks: the kappa statistic. Computational Linguistics 22, 249-245. Chae, H. (1999). Adverbial nominals and the -ul/-lul marker. Harvard Studies in Korean Linguistics 8, 223-234. Cho, Y.Y. and P. Sells (1995). A lexicalist account of inflectional suffixes in Korean. Journal of East Asian Linguistics 4, 119-174.
12
This process of conventionalization of pragmatic principles can be naturally modeled in stochastic OT by the movement in strength of violable constraints along the continuous ranking scale, as shown by Bresnan et al. (2001).
60
HANJUNG LEE
Choi, H. (1995). Topic and focus in Korean: The information partition by phrase structure and morphology. Japanese/Korean Linguistics 6, 545-561. Clark, B.(2004). A Stochastic Optimality Theory Approach to Syntactic Change. Doctoral dissertation, Stanford University. Dingare, S. (2001). The Effect of Feature Hierarchies on Frequencies of Passivization in English. MA thesis, Stanford University. Dixon, R.M.W. (1972). The Dyirbal Language of Northern Queensland. Cambridge: Cambridge University Press. Dixon, R.M.W. (1979). Ergativity. Language 55, 59-138. Donohue, C. (1999). Optimizing Fore Case and Word Order. Ms., Stanford University. Fry, J. (2001). Ellipsis and ‘wa’-Marking in Japanese Conversation. Doctoral dissertation, Stanford University. Gair, J. W. (1970). Colloquial Sinhalese Clause Structures. The Hague: Mouton. Givón, T. (1979). On Understanding Grammar. New York: Academic Press. Hinds, J. (1983). Topic continuity in Japanese. Topic Continuity in Discourse. Ed. by T. Givón. Amsterdam: John Benjamins, 43-93. Kim, K. (1990). Where do the contrastive and focus readings come from? Japanese/Korean Linguistics 1, 395-412. Ko, E. (2000). A discourse analysis of the realization of objects in Korean. Japanese/Korean Linguistics 9, 195-208. Koontz-Garboden, A. (2001). A stochastic OT approach to word order variation in Korlai Portuguese. CLS 37, 347-361. Kortmann, B. (1999). Typology and dialectology. Proceedings of the 16th International Congress of Linguists, Paris 1997. CD-ROM. Amsterdam: Elsevier Science. Kroch, A. (1978). Toward a theory of social dialect variation. Language in Society 13, 145-204. Kroch, A. (2000). Syntactic change. The Handbook of Contemporary Syntactic Theory. Ed. by M. Baltin and C. Collins. Amsterdam: John Benjamins, 699-729. LDC. (1996). CallFriend Korean corpus. Linguistic Data Consortium, University of Pennsylvania. http://www.ldc.upenn.edu. Lee, C. (1995). Definite/specific and case marking in Korean. Theoretical Issues in Korean Linguistics. Ed. by Y. Kim-Renaud. Stanford: CSLI Publications, 325-341. Lee, D. (2002). The function of the zero particle with special reference to spoken Japanese. Journal of Pragmatics 34, 645-682. Lee, H.S. and S.A. Thompson (1985). A discourse account of the Korean accusative marker. Studies in Language 13, 105-128. Lee, K. (2002). Nominative case marker deletion in spoken Japanese: an analysis from the perspective of information structure. Journal of Pragmatics 34, 683-709. Legendre, G. (1998). Why French stylistic inversion is optimal. Ms., Johns Hopkins University. Masunaga, K. (1987). Non-thematic Positions and Discourse Anaphora. Doctoral dissertation, Harvard University. Masunaga, K. (1988). Case deletion and discourse context. Papers from International Workshop on Japanese Syntax. Ed. by W. Poser. Stanford: CSLI Publications, 145-156. Matsuda, K. (1996). Variable Zero-Marking of (o) in Tokyo Japanese. Doctoral dissertation, University of Pennsylvania. Minashima, H. (2001). On the deletion of accusative case markers in Japanese. Studia Linguistica 55, 175-190. van Oostendorp, M. (1997). Style levels in conflict resolution. Variation, Change and Phonological Theory. Ed. by F. Hinskens, R. van Hout and W.L. Wetzels. Amsterdam/Philadelphia: John Benjamins, 35-68. Paolillo, J.C. (2002). Analyzing Linguistic Variation: Statistical Models and Methods. Stanford: CSLI Publications. Preston, D.R. (1991). Sorting out the variables in sociolinguistic theory. American Speech 66, 33-56. Prince, A. and P. Smolensky (1993). Optimality Theory: Constraint Interaction in Generative Grammar. RuCCS Technical Report #2. Rutgers University Center for Cognitive Science. Rooth, M. (1985). Association With Focus. Doctoral dissertation, University of Massachusetts, Amherst. Scott, G. (1978). The Fore Language of Papua New Guinea. Canberra: School of Pacific Studies. Selkirk, E. (1972). The Phrase Phonology of English and French. Doctoral dissertation, MIT.
QUANTITATIVE VARIATION IN KOREAN CASE ELLIPSIS
61
Silverstein, M. (1976). Hierarchy of features and ergativity. Grammatical Categories in Australian Languages. Ed. by R.M.W. Dixon. Canberra: Australian Institute of Aboriginal Studies, 112-171. Tsutsui, M. (1984). Particle Ellipsis in Japanese. Doctoral dissertation, University of Illinois at UrbanaChampaign. Valludí, E. and M. Vilkuna (1998). On rheme and kontrast. The Limits of Syntax (Syntax and Semantics 29). Ed. by P. Culicover and L. McNally. San Diego: Academic Press, 79-108. Yang, I. (1972). Korean Syntax: Case Marking, Delimiters, Complementation and Relativization. Seoul: Paek Hap Sa. Yatabe, S. (1999). Particle ellipsis and focus projection in Japanese. Language, Information, Text, vol. 6. Department of Language and Information Sciences, University of Tokyo, 79-104.
HELEN DE HOOP AND BHUVANA NARASIMHAN
CHAPTER 4 Ergative Case-marking in Hindi
1. INTRODUCTION1 Sometimes, not all subjects or all objects of transitive clauses are case-marked, but only those that are ‘less prototypical’ subjects or objects (Aissen 1999, 2003). In other words, only those objects that have certain characteristics of typical subjects (such as being animate or specific) are case-marked, or only those subjects that have certain characteristics of typical objects (such as being a noun phrase rather than a pronoun or being a third person pronoun rather than a first or second one). Overtly marking a less typical subject with ‘subjective case-marking’ or a less typical object with ‘objective case-marking’ helps to tag the grammatical function of one argument with respect to the other. Citing the work in Silverstein (1976) and Comrie (1989), Aissen argues that in languages in which differential case-marking on objects occurs, it occurs whenever the object is ‘higher in prominence’ (higher than is usually the case for an object): ‘The higher in prominence a direct object, the more likely it is to be overtly case-marked’ since ‘it is those direct objects which are most in need of being distinguished from subjects that get overtly case-marked’ (Aissen 2003:3). This idea nicely accounts for a situation where accusative case is assigned to objects that are animate and specific, because these are high in prominence and thus less typical objects, hence confusable with subjects (Lee, this volume). Similarly, we can explain a situation where ergative case is assigned to subjects that are full noun phrases rather than pronouns, because they are low in prominence and thus less typical subjects, hence confusable with objects. This pattern of casemarking the less typical object and the less typical subject is found in Dyirbal (cf. Dixon 1979). In this language, subjects are case-marked when they are third person pronouns, proper names or common nouns, whereas objects are case-marked when they are first or second person pronouns. Clearly, Dyirbal supports Aissen’s (1999, 2003) idea that typical subjects and objects may often remain without case-marking while atypical subjects and objects are case-marked. 1
We are grateful to Miriam Butt and to the audience of the PIONIER workshop on Differential Subject Marking (Nijmegen, July 2004) for their comments. Two anonymous reviewers made some very helpful suggestions on an earlier version. Helen de Hoop furthermore acknowledges the Netherlands Organisation of Scientific Research (NWO) for financial support, grants 220-70-003 (PIONIERproject ‘Case cross-linguistically) DN 30-609 (bilateral cooperation programme NWO-DFG ‘Incremental interpretation of case and prominence’), and 360-70-220 (NWO-project ‘Animacy’).
63 H. de Hoop and P. de Swart (eds.), Differential Subject Marking, 63–78. © 2008 Springer.
64
HELEN DE HOOP AND BHUVANA NARASIMHAN
Differential object marking can be accounted for in this way in Hindi too (cf. Aissen 2003, de Swart 2003, de Hoop and Narasimhan 2005). Indeed, those objects that are most subject-like in the sense that they are animate and/or specific receive overt accusative ko marking.2 Differential object marking, between ko and nullmarking, shows up with a variety of transitive and ditransitive predicates. Strikingly, these are the same predicates that allow case alternations on the subject as well, between ergative ne and null-marking (cf. Davison 1999) (examples 1 to 4). Ergative case-marking marks the subjects of perfective highly transitive predicates, such as phaaD| ‘tear’ and toD| ‘break’, as illustrated in (1) and (3): (1) raam=ne kapD|e=ko / kapD|aa-∅ Raam=ERG cloth=ACC / cloth-NOM ‘Raam tore (a/the) cloth.’ (2) raam-∅ kapD|e=ko / kapD|aa-∅ Raam-NOM cloth=ACC / cloth-NOM ‘Raam tears (a/the) cloth.’ (3) raam=ne patthar=ko / patthar-∅ Raam=ERG stone=ACC / stone-NOM ‘Raam broke a/the stone.’ (4) raam-∅ patthar=ko / patthar-∅ Raam-NOM stone=ACC / stone-NOM ‘Raam breaks a/the stone.’
phaaD|-aa tear-PFV.SG.M phaaD|-taa tear-IPFV.SG.M
hae be.PRS.3SG
toD|-aa break-PFV.SG.M toD|-taa hae break-IPFV.SG.M. be.PRS.3SG
Ergative case-marking in Hindi is also found on the subjects of a range of perfective predicates that are lexically ‘less transitive’, such as as dekh ‘see’ and sUUgh ‘smell’ (cf. Malchukov, 2005 for a recent account of different classes of lexical verbs that subcategorize for different case marking patterns in relation to their degree of transitivity). This is illustrated in (5). (5) laDkii=ne ek kuttaa-∅ / kutte=ko girl=ERG one dog-NOM / dog=ACC ‘The girl saw a/the dog.’
dekh-aa see-PFV.SG.M
The predicates in (1)-(5) do not only share the property of ergative subject marking in the perfective, they also share the possibility of differential object marking (as is also illustrated in (1)-(5)), and they are similar in that they allow their objects to passivize. In de Hoop and Narasimhan (2005), we use the cover term ‘highly transitive’ for this lexical class of predicates. Subjects of low transitivity predicates, on the other hand, are not assigned ergative case, independent of whether they are in the perfective or not. Thus, the subject of a perfective transitive predicate such as mil ‘receive’ is in the dative and not in the ergative (6), the subject of an inability passive construction gets 2
The pattern is slightly more complex, since a nonhuman animate object can occur without –ko if it is non-specific, as example (5) shows (cf. Mohanan, 1994).
ERGATIVE CASE-MARKING IN HINDI
65
instrumental case (7), and the subject of constructions encoding metaphorical location bears locative case (8). (6) raam=ko ek kitaab-∅ mil-ii Raam=DAT one book-NOM receive-PFV.SG.F ‘Raam received a book.’ (7) raam=se Dibbe-∅ nahII khol-e ga-ye Raam=INS boxes-NOM not open-PFV.PL.M go-PFV.PL.M ‘Raam was not able to open the boxes.’ (8) ravii=par apne parivaar=kii saarii jimmedaarii-∅ hae responsibility-NOM be.PRS.3SG ravii=LOC self family=GEN all ‘Ravi is solely responsible (for the care of) self’s family.’ (Mohanan, 1994:176) To sum up, predicates at the high end of the transitivity continuum assign ergative case to their subjects when they are in the perfective, while transitive predicates at the lower end of continuum block ergative subjects. The question is whether differential subject marking in Hindi mirrors differential object marking. That is, can we characterize the ergative subjects in Hindi as subjects low in prominence or are they in any way atypical subjects, so that they need to be casemarked for the purpose of disambiguating them from objects? In the remainder of this chapter, we will argue that this is not the case (see also de Hoop and Narasimhan 2005, Kornfilt, this volume, Woolford, this volume, Arkadiev, this volume). We will show that in fact ergative case-marking in Hindi is associated with ‘strong’, hence typical subjects rather than with ‘weak’ or object-like or atypical subjects. 2. STRONG SUBJECTS IN HINDI In de Hoop and Narasimhan (2005) we have argued in favour of a model of casemarking augmented with the notion of argument strength (cf. also de Hoop 1996, 1999). The determination of argument strength is meant to distinguish between strong and weak arguments of a certain type. We suggest that in languages with differential case-marking, subjects and objects that are ‘strong’ are likely to be overtly marked as such. The question is how we measure the ‘strength’ of arguments, since languages seem to vary in this respect. We propose that the strength of arguments varies in accordance with the degree of transitivity (following Hopper and Thompson 1980). According to Hopper and Thompson, features that contribute to a high degree of transitivity include the number of participants (≥ 2), features of the subject, namely ‘volitionality’ and ‘high in potency’, features of the verb phrase, such as ‘telicity’, ‘action’, and ‘realis’, and features of the object, including ‘animacy’, ‘referentiality’ and ‘definiteness’. Hopper and Thompson argue that lowering any of these features may result in a decrease in formal transitivity. In the spirit of Hopper and Thompson, we may therefore define the strongest subject as the subject of an active, telic transitive or ditransitive clause that is volitional, high
66
HELEN DE HOOP AND BHUVANA NARASIMHAN
in potency, and that cooccurs with an animate and definite object. Note, by the way, that not only strong objects but also strong subjects can be thus characterized in terms of animacy, since only animate subjects can fulfil the feature of ‘volitionality’. The definitions of strong subjects and objects that are thus derived are also in accordance with proto-role properties (Dowty 1991). That is, prototypical subjects/agents are assumed to be volitional (in control of the action), whereas prototypical objects/patients are assumed to be affected by the action. Since animate objects can be argued to be more ‘affected’ than inanimate objects, it follows that animate objects are more patient-like (hence, more prototypical patients) than inanimate objects. In the remainder of this chapter, we will represent strong subjects and objects with capital letters A and P respectively, and weak subjects and objects with a and p, following Legendre et al. (1993) in this respect. When we consider differential subject marking, Hindi presents a serious problem for an analysis of differential subject marking in terms of distinguishability. First of all, whenever the subject and the object in a transitive clause are in need of being distinguished, then in principle, the case marking of the animate or specific object should already be sufficient to do the job. In that sense, there would be no need to additionally mark the subject of a transitive clause in the perfective with ergative case. Moreover, the ergative case-marking of the subject of a perfective transitive clause in Hindi is clearly not triggered by a decrease in strength of that subject. That is, subjects of perfective clauses are not less prominent than subjects of imperfective clauses, nor in any other way more similar to objects. Following Aissen’s account, we would expect non-canonical subjects (i.e. those subjects which are atypical in the sense that they are lower in prominence than usual, e.g. non-volitional, inanimate, non-specific) to receive ergative case-marking. But inanimate subjects hardly ever receive ergative case-marking, they are rarely even allowed to be subjects of transitives at all:3 (9) ??patthar=ne khiD|kii=ko toD| di-yaa window=ACC break give-PFV.SG.M stone=ERG ‘The stone broke the window.’ (10) khiD|kii-∅ patthar=se TuuT ga-yii stone= INS break go-PFV.SG.F Window-NOM ‘The window broke by (virtue of) the stone.’ Secondly, the subject of a perfective clause is usually conceived of as more of a typical agent than that of an imperfective clause, because the action was successfully completed (Dowty 1991). Besides, perfectivity is one of the factors leading to a high degree of transitivity in the approach of Hopper and Thompson (1980). To put it in our terminology, perfectivity can be argued to strengthen the subject but it certainly does not weaken it (cf. DeLancey 1981). Further evidence that the ergative case-marker ne marks properties associated with strong subjects rather than weak ones can be seen in constructions where ne is 3
Thanks to an anonymous reviewer for drawing our attention to this.
ERGATIVE CASE-MARKING IN HINDI
67
used optionally. Even though there is no second argument in intransitive predicates, case-marking is used to resolve a certain ambiguity allowed by the predicate semantics. In Hindi, the single argument of a small set of primarily ‘bodily emission’ verbs is optionally ergative. In such cases, ergative case-marking is argued to associate with an interpretation of the subject as being volitional, i.e., in control of the event the predicate refers to and as such more like a prototypical subject (Butt and King 2003; Mohanan 1994). (11) raam=ne chiikh-aa Raam=ERG scream-PFV.SG.M ‘Raam screamed (purposefully).’ (12) raam-∅ chiikh-aa raam-NOM scream-PFV.SG.M ‘Raam screamed.’ It appears that both subjects of perfective transitive sentences, as well as volitional subjects of some intransitive verbs, such as scream, are strong rather than weak. These subjects definitely do not need the ergative case marking to avoid ambiguity with respect to the object. To sum up, Hindi provides clear evidence against the fact that differential casemarking on subjects is always motivated by the need to disambiguate the subjects from the objects. It seems that we cannot account for differential subject marking in Hindi by appealing to the distinguishing role of case. Rather, ergative case-marking appears to mark strong subjects in Hindi, thus coding prototypical subject properties such as volitionality (see also Asudeh 2001 and Lee 2001 who incorporate Dowty’s proto-roles into a bidirectional OT analysis for Marathi and Hindi). But as we will argue below, we cannot account for differential subject marking in Hindi by (directly) relating ergative case to volitionality. Rather, as we will argue below, ergative case is reserved for strong subjects in Hindi, but volitionality is not among the factors that make a subject strong in Hindi. 3. ERGATIVE CASE IN HINDI Let us begin our examination of the ergative case with the assumption that ergative case is assigned to strong subjects in Hindi. In transitive predicates, we indeed find that a decrease in strength in one way or another may result in the loss of ergative case marking. For instance, in the ‘inability passive’ constructions (Mohanan 1994) where the subject is unable to accomplish the action encoded by the verb, it is marked with the instrumental case se. (13) Raam=se kaam-∅ nahII ki-yaa ga-yaa Raam=INS work-NOM not do-PFV.SG.M go-PFV.SG.M ‘Raam couldn’t do the work.’
68
HELEN DE HOOP AND BHUVANA NARASIMHAN
Obviously, despite the perfectivity of the predicate, the ‘action’ has not been completed in (13). In the framework of Hopper and Thompson, although ‘perfectivity’ is a factor of high-transitivity, other factors, such as ‘non-action’, ‘negation’ and ‘subject low in potency’ are recognized as factors of low-transitivity. Apparently, they win the competition here, and the subject comes out as weak rather than strong. 4 The ergative case-marking on the subject is also blocked with transitive predicates (e.g. khaa ‘eat’, pii ‘drink’) when combined with the intransitive light verb jaa ‘go’. This light verb is usually used with intransitive main verbs with a theme argument that undergoes change of location or state, as illustrated below: (14) siitaa-∅ kamre=se nikal ga-yii Siitaa-NOM room=ABL emerge go-PFV.SG.F ‘Siitaa left the room.’ (15) siitaa-∅ so ga-yii Siitaa-NOM sleep go-PFV.SG.F ‘Siitaa went to sleep.’ In (16) below, we may say that the cumulative effect of the highly transitive verb in combination with the intransitive light verb is a decrease in strength of the subject, again blocking the use of ergative case in favour of null-marking the subject, even in perfective contexts. (16) siitaa-∅ saaraa khaanaa-∅ Siitaa-NOM all food-NOM ‘Siitaa ate up all the food.’
khaa ga-yii eat go-PFV.SG.F
Note that the object in (16) can still alternate between null and ko marking of the object, reinforcing our point that it is the strength of the subject that influences its case here, not the case assigning properties of the main verb. (17) siitaa-∅ us baD|e seb=ko Siitaa-NOM that big apple=ACC ‘Siitaa ate up the whole big apple.’
puuraa khaa ga-yii all eat go-PFV.SG.F
To sum up the discussion so far, the examples suggest that a reduced strength of the subject blocks ergative case-marking in Hindi, independent of the case-marking of the object, thus in accordance with our initial assumption that ergative case4
It is important to note that the inability passive construction is a grammaticized means of indicating weak subjects of transitive clauses. It is not the case that any clause encoding non-completion of an event is accompanied by substitution of the ergative case-marking with instrumental case-marking (e.g. raam=ne / *raam=se Dibbaa-∅ nahII khol-aa ‘raam=ERG / *raam=INS box-NOM not openPFV.SG.M.’ ‘Raam did not open the box’). Also, perfective transitive verbs with modal auxiliaries occur with null case-marking on the subject whether or not they semantically imply successful achievement of a result (raam-∅ / *raam=ne Dibbaa (nahII) khol paa-yaa ‘raam-NOM / *raam=ERG box not open can-PFV.SG.M.’ ‘Raam could (not) open the box’.
ERGATIVE CASE-MARKING IN HINDI
69
marking in Hindi cannot be accounted for in terms of the distinguishing function of case. However, if we wish to argue that ergative case in Hindi is indeed used to mark (strong) subjects, then we should be able to characterize the exact property or set of properties that is associated with ergative case. In the examples (1)-(4) above we saw that it was the perfectivity of a highly transitive predicate that triggered ergative case marking on the subject. Examples (13) and (16) show, however, that a perfective highly transitive predicate can be overruled by other features, such as ‘inability’ or the combination with a certain type of light verb, and in those cases ergative case-marking gets blocked. As we pointed out above, in Hindi, the single argument of a small set of primarily ‘bodily emission’ verbs (e.g., khAAs ‘cough’, bhAUk ‘bark’, chiikh ‘scream’, muut ‘urinate’, thuuk ‘spit’, muskuraa ‘smile’) is optionally ergative (see examples (11)-(12) above). The case alternation in (11)-(12) seems to be clearly related to the strength of the subject, in this case volitionality. In some dialects of Hindi, the subject argument of ‘desire’ constructions can optionally receive ergative case-marking as well to imply volitionality (Butt and King 2003): (18) raam=ne ghar jaa-naa hae Raam=ERG home go-INF be.PRS.3SG ‘Raam wants to go home.’ By contrast, when the subject is marked with ko (dative case), the sentence is ambiguous between a volition and obligation reading: (19) raam=ko ghar jaa-naa hae Raam=DAT home go-INF be.PRS.3SG ‘Raam wants to/has to go home.’ The obligation reading suggests that the subject is non-volitional, or less in control, i.e., less of a typical subject. This might be argued to make the ergative case marking less adequate, and that is why this reading is only available with the dative case marker. Note, however, that the unmarked form (dative case) can be associated with both readings. Ergative case thus expresses the volitionality of the agent argument in the intransitive examples above, but volitionality is not necessarily realized as ergative case. This is also obvious when we consider other types of intransitive verbs, which do not optionally allow for ergative case. In fact, most intransitive verbs in Hindi only allow for an unmarked (nominative) subject, irrespective of whether the subject is volitional or not. The subject of bhaag ‘run’ is usually volitional, while the subject of gir ‘fall’ is usually not, yet both get a nominative subject. (20) mohan-∅ ghar bhaag-aa Mohan-NOM home run-PFV.SG.M ‘Mohan ran home.’
70
HELEN DE HOOP AND BHUVANA NARASIMHAN (21) aruNaa-∅ zamiin=par gir-ii AruNaa-NOM ground=LOC fall-PFV.SG.F ‘Aruna fell on the ground.’
The same holds for intranisitive verbs with volitional subjects such as tair ‘swim’, naac ‘dance’, and ucchhal ‘jump’. These verbs do not assign ergative case to their subjects. An example of a language where the type of predicate plays hardly any role in the assignment of case seems to be Kambera (Austronesian) as described by Klamer (this volume). Klamer discusses the linking of intransitive arguments in Kambera showing that there are five ways of case-marking the intransitive subject in Kambera. In general, the choice of one of these different ways of encoding the intransitive subject is not determined by the lexical properties of the verb nor by the syntactic structure, but rather depends on the context in which the verb is used. For example, while the intransitive subject in (22) is canonically marked by a nominative da-, which does not force a specific interpretation of the clause, the addition of an accusative marker -ha in (23) expresses obligation. (22) Da-laku pa-rama haromu 3NOM-go CTR-work tomorrow ‘They will go to work tomorrow.’ (23) Da-laku-ha pa-rama haromu 3NOM-go-3ACC CTR-work tomorrow ‘They must/have to go to work tomorrow.’ Note that the accusative marker on the intransitive subject does not replace the nominative one, but rather is added to it. The cumulative effect in (23) is a subject that is less volitional than the nominatively marked subject in (22). So, compared to Kambera, in Hindi the role of the verb class in determining the case-marking is much stronger. As we argued above, differential subject marking in transitive clauses is restricted by the lexical verb class and perfectivity, in the sense that subjects of perfective highly transitive predicates are marked with ergative case unless the ergative case-marking is blocked by overruling features such as ‘inability’ or the case-assigning properties of a light verb. Furthermore, in a small class of intransitive predicates, ergative case is used to express the volitionality of the subject, but in general, volitionality of the subject is not realized as ergative case in Hindi. Most volitional subjects of intransitive clauses are not assigned ergative case at all, and a volitional subject of an imperfective transitive clause does not bear ergative case either. Moreover, subjects of perfective predicates which are lexically characterized as highly transitive receive ergative case-marking, even if the subject is clearly nonvolitional, in fact a patient rather than an agent. This is shown in example (24), where the use of a highly transitive predicate, khaa ‘eat’, results in ergative case on the subject even when the subject in this idiomatic expression based on the verb khaa ‘eat’ has in fact all the properties of a patient and none of an agent. That is, the verb khaa can occur as the light verb in a compound verb such as maar khaa ‘suffer
ERGATIVE CASE-MARKING IN HINDI
71
beatings’ (lit. ‘eat beating’) or dhokhaa khaa ‘suffer deception’ (see Davison 1999). But even in those cases, the patient subject receives ergative case: (24) raam=ne apne shaitaanii=ke kaaraN bahut maar khaa-yaa Raam=ERG self bad.behaviour=GEN reason much beating eat-PFV.SG.M ‘Raam suffered much beating for his bad behaviour.’ Examples such as (24) as well as the example in (16) above might suggest that ergative case-marking on the subject is merely determined by the properties of the (light) verb and that the strength of the subject is irrelevant. However, the lexical and semantic contributions of the light verb are not easily teased apart. For instance, the compound verb construction in (24) differs from other constructions with patient-like transitive subjects ((28)-(30)) in being able to occur in the progressive, compatibility with an adverb of duration, and (marginally) passivization ((25)-(27)): (25) raam-∅ bahut maar khaa rahaa hae Raam-NOM much beating eat PROG be.PRS.3SG ‘Raam is getting a lot of beating.’ (26) raam=ne ek ghaNTe=ke liye bahut maar khaa-yaa for much beating eat-PFV.SG.M Raam=ERG one hour=GEN ‘Raam got a lot of beating for an hour.’ (27) ?raam=se aaj bahut maar khaa-yaa ga-yaa Raam=INS today much beating eat-PFV.SG.M go-PFV.SG.M ‘Raam was beaten a lot today.’ (28) *raam=ko seb-∅ pasand ho rahaa hae Raam=DAT apple-NOM liking be PROG be.PRS.3SG *‘Raam is liking the apple.’ (29) *raam=ko ek ghaNTe=ke liye maamlaa-∅ pataa hu-aa Raam=DAT one hour=GEN for matter-NOM know become-PFV.SG.M *‘Raam got to know the matter for an hour.’ (30) *raam=se seb-∅ pasand hu-aa ga-yaa Raam=INS apple-NOM liking become-PFV.SG.M. go-PFV.SG.M ‘The apple was liked by Raam.’ Since not all psych-predicate constructions necessarily pattern in a similar way with respect to these tests, the evidence is not conclusive that the possibility of ergative case-marking on the A argument of maar khaa can be linked to its higher ‘strength’ vis-à-vis the A argument of psych-predicate constructions. Nevertheless, the data suggest that the light verb may not be mechanically assigning case to the subject owing to arbitrarily specified lexical information, but it is indeed the interaction of (grammaticalized) predicate semantics and other semantic and aspectual features which eventually determines the strength (and thereby the casemarking) of the subject.
72
HELEN DE HOOP AND BHUVANA NARASIMHAN 4. ARGUMENT STRENGTH BETWEEN MEANING AND FORM
In this section we put forward the view that the mapping from semantic input to case morphology involves the determination of argument strength (cf. de Hoop 1999). Argument strength is useful as an abstract notion that is neither part of the meaning nor of the form. Languages may vary as to how important various scales are for case marking purposes, but also as to where the cut-off point is. The constraints on argument strength that we will use in order to account for ergative case-marking in Hindi are derived from what Hopper & Thompson (1980) characterize as a high degree of transitivity, as argued in section 1 above. Argument strength is not necessarily associated with case marking. When different languages are considered, it can be observed that some features trigger certain case alternations in one language or one grammatical construction, while similar features trigger agreement or preposition insertion or word order variation in another language or other configuration (cf. Silverstein 1976, Hopper and Thompson 1980). Palauan, as discussed by Woolford (1995), provides a nice example of such a pattern. As Woolford observes, in the context of a perfective predicate, there is object agreement when the object is either human or specific and singular. This is illustrated in (31)-(32): (31) Te-’illebed-ii a bilis a rengalek 3PL-PFV-hit-3SG dog children ‘The kids hit the dog.’ (32) Te-’illebed a bilis a rengalek children 3PL-PFV-hit dog ‘The kids hit a dog/ the/some dogs.’ Exactly the same combination of features (i.e., human or specific and singular) triggers the insertion of a preposition in the context of an imperfective predicate, as illustrated below: (33) Ng-milengelebed er a bilis PREP dog 3SG-IMP-hit ‘S/he hit the dog.’ (34) Ng-milengelebed a bilis dog 3SG-IMP-hit ‘S/he hit a dog/ the/some dogs.’ Thus, the crucial nominal features that distinguish different types of objects in Palauan, involve humanness, specificity and number. But the syntactic reflections differ as a function of aspect: for perfective verbs the nominal feature combination correlates with object agreement, while for imperfective verbs it correlates with preposition insertion. Linguistic phenomena that involve a ranking of argument properties and the morpho-syntactic consequences of this ranking have been attested in many
ERGATIVE CASE-MARKING IN HINDI
73
languages. The different distinctions cross-linguistically have been described in terms of a cut-off point in a certain type of hierarchy (referential hierarchy, animacy hierarchy, person hierarcy, topic hierarchy, to mention a few). In our view it may be helpful to use an abstract label to capture elements to the left and those to the right side of the ‘cut’. Therefore we have chosen the terms strong and weak to refer to certain properties of arguments. In Hindi, we might say that not all subjects will get the label strong, but only the subjects of perfective highly transitive predicates. Our approach differs from Wunderlich (1997), Wunderlich and Lakämper (2001), Kiparsky (2001), Stiebels (2000), as well as from Bittner and Hale (1996), who all postulate a kind of abstract argument structure in the input, in order to account for the differences in morphological case outputs. A general problem for this type of approach is that the input-output representations are very closely related, which means that the argumentation for having a certain argument structure in the input is partly driven by the data that it should explain. Hence, this type of analysis incorrectly predicts identical relations between argument properties and case assignment in languages with similar case systems. Furthermore, this type of analysis cannot straightforwardly deal with partly independent case alternations for subjects and objects (i.e., independent of the case realized on the other argument), such as we observe in Hindi. By contrast, in our approach, the strength of arguments is not really present in the input nor in the output. It is ‘hidden’ and the system through the optimization process must infer it. In the next section we will present our bidirectional OT analysis of differential subject marking in Hindi, making use of argument strength only in the process of evaluating form-meaning pairs. 5. OPTIMAL CASE-MARKING OF SUBJECTS IN HINDI In this section we present our bidirectional Optimality Theoretic analysis of ergative subject marking in Hindi. We assume that in Hindi a perfective highly transitive predicate makes a subject strong. In addition to factors such as perfectivity and verb semantics, the strength of the subject can be determined by additional sources, including, as discussed above, reduced transitivity (as in the inability passives) and the strict dominance of the light verb in compound verb constructions. We claim that volitionality itself is too weak to make a subject strong in Hindi (contra e.g., Lee 2003). The relevant constraints for our analysis are listed below: (35) IDENTIFY-A: [+ perfect] & [+ highly transitive] ↔ ERG (36) PAIP: ‘Primary Argument Immunity Principle’: don’t case-mark the unmarked – nominative – argument. (37) VOLITIONALITY : [+ volitional] ↔ ERG Our constraint IDENTIFY-A is used to case-mark and identify a ‘strong’ subject. That is, it is a truly bidirectional constraint, that states that all and only strong subjects should receive ergative case. Subjects of perfective highly transitive verbs
74
HELEN DE HOOP AND BHUVANA NARASIMHAN
are considered to be strong in Hindi, while volitional subjects are not (necessarily) strong in Hindi. PAIP is borrowed from Malchukov (2006). It is a constraint that penalizes morphological case-marking of an argument that could in principle remain unmarked. In Hindi, PAIP is satisfied when an object or subject is in the nominative (unmarked) case and it is violated when an object or subject bears an overt (ergative, accusative/dative, etc.) case-marker. We prefer to use PAIP rather than Aissen’s (1999, 2003) more general constraint *StrucC, since the latter assumes a special notion of ‘structural case’ which our account lacks. PAIP penalizes morphological case-marking on the argument that could (should) remain unmarked (the nominative argument), i.e. the obligatory case bearing argument in terms of Bobaljik (1993). With the help of two of the constraints formulated above, we can account for the nominative-ergative case alternation we find in the subject of transitive clauses in Hindi. The analysis is based on the assumption that in Hindi, as we pointed out above, in the context of a highly transitive predicate, perfectivity determines the strength of the subject. In Hindi, obviously, IDENTIFY-A is an influential constraint which is ranked higher than PAIP. Tableau 1. Form and meaning of the subject of a transitive clause
Subject of a highly IDENTIFY-A transitive predicate [∅, a] [ne, a] * * [∅, A] [ne, A]
PAIP
* *
The tableau reads as follows. Because the pair [∅, a] does not violate any constraints, it is super-optimal (cf. Blutner 2000). The second and third candidate form-meaning pairs are blocked by the first one. The fourth candidate violates P AIP in order to satisfy IDENTIFY-A. Because both the form and the meaning of the fourth pair differ from the form and meaning of the first super-optimal pair, this pair is super-optimal as well, i.e., a winner (cf. Blutner 2000). Thus, our bidirectional Optimality Theoretic analysis is able to explain the basic pattern of differential subject marking in transitive clauses in Hindi. The next step is to offer an account of the peculiar optional case-marking that we find with the restricted set of intransitive verbs which allow ne marking on the intransitive subject in perfective contexts when ‘volitionality’ can be attributed to the subject. An example of this alternation is repeated below: (38) raam=ne chiikh-aa Raam=ERG scream-PFV.SG.M ‘Raam screamed (purposefully)’ (39) raam-∅ chiikh-aa Raam-NOM scream-PFV.SG.M ‘Raam screamed’
ERGATIVE CASE-MARKING IN HINDI
75
At first sight, the pattern in (38)-(39) suggests that the constraint VOLITIONALITY plays an important role here, but as we have seen, this constraint must be ranked fairly low in Hindi. Hence, we do not expect it to be satisfied at all in either transitive or intransitive sentences. Reconsider the following examples that illustrate the fact that in most cases, a volitional intransitive subject is unmarked (40) just like a non-volitional one (41). (40) mohan-∅ ghar bhaag-aa Mohan-NOM home run-PFV.SG.M ‘Mohan ran home’ (41) aruNaa-∅ zamiin=par gir-ii AruNaa-NOM ground=LOC fall-PFV.SG.F ‘Aruna fell on the ground’ According to the constraint ranking that we proposed above to account for the transitive clauses, we do not expect the subject to be in the ergative, irrespective of its volitionality. This is illustrated in the following two tableaux, where VOLITIONALITY is ranked lowest. Tableau 2. Form and meaning of the subject of an intransitive clause
Subject of a verb like IDENTIFY-A run in the perfective [∅, volitional a] [ne, volitional a] *
PAIP
VOLITIONALITY *
*
Tableau 3. Form and meaning of the subject of an intransitive clause
Subject of a verb like fall IDENTIFY-A in the perfective [∅, non-volitional a] [ne, non-volitional a] *
PAIP
VOLITIONALITY
*
However, whenever we encounter a pattern in language where in the same context two forms are available as well as two meanings, we may find a one-to-one mapping of forms and meanings that is in accordance with a principle of markedness that states that the (more) marked form is used to express the (more) marked meaning (Horn 1984). This principle of markedness can be derived in bidirectional Optimality Theory (Blutner 2000). Clearly, intransitive predicates such as scream in Hindi do allow for two different forms and two different meanings of their subjects. In the following tableau, we evaluate the form-meaning pairs of subjects in the contexts of predicates of the scream-type against the three constraints introduced above. We know already that VOLITIONALITY is easily violated in Hindi. Hence, the subject of an intransitive verb comes out as weak, even when it is volitional. In the
76
HELEN DE HOOP AND BHUVANA NARASIMHAN
latter case, IDENTIFY-A is violated, but VOLITIONALITY is satisfied, when ergative case is assigned to such a subject. Tableau 4. Form and meaning of the subject of an intransitive clause
Subject of a verb like IDENTIFY-A scream in the perfective [∅, non-volitional a] [ne, volitional a] * [∅, volitional a] [ne, non-volitional a] *
PAIP
VOLITIONALITY
* *
* *
The picture above shows the derivation of two super-optimal pairs of form and meaning: the nominative form goes with the non-volitional weak subject, whereas the ergative form goes with the volitional weak subject. The third and fourth formmeaning pairs are both blocked by the first candidate, which does not violate any of the constraints. Crucially, the correspondence between a volitional meaning and ergative case does not mean that all of a sudden the constraint VOLITIONALITY is ranked higher. It still does not make the subject strong in Hindi and it is still lowest ranked. Although this might seem counterintuitive at first sight, it is exactly in accordance with our earlier findings, on the basis of which we in fact argued that the constraint VOLITIONALITY must be ranked very low in Hindi. 6. CONCLUSION While differential object marking in Hindi can be analysed in terms of the distinguishing role of case (following Aissen 2003, de Swart 2003, de Hoop and Narasimhan 2005), this approach cannot satisfactorily deal with differential subject marking in Hindi. In fact, Hindi provides clear evidence against the idea that differential case-marking on subjects is always motivated by the need to disambiguate the subjects from the objects. Rather, ergative case-marking seems to mark strong subjects, hence coding prototypical subject properties. We posit that an additional motivation for overt case-marking is to identify strong arguments (in this case subjects). We claim that a bi-directional Optimality Theoretic approach cannot only account for differential subject marking in transitive clauses in Hindi, but also for the fact that, quite unexpectedly, volitional subjects of a small class of intransitive verbs bear ergative case. REFERENCES Aissen, J. (1999). Markedness and subject choice in Optimality Theory. Natural Language and Linguistic Theory 17, 673-711. Aissen, J. (2003). Differential object marking: Iconicity vs. economy. Natural Language and Linguistic Theory 21, 435-483.
ERGATIVE CASE-MARKING IN HINDI
77
Arkadiev, P.M. (this volume). Differential Argument Marking in Two-term Case Systems and its Implications for the General Theory of Case Marking. Asudeh, A. (2001). Linking, optionality, and ambiguity in Marathi. Formal and Empirical Issues in Optimality-Theoretic Syntax. Ed. by P. Sells. Stanford: CSLI Publications. Bittner, M. and K. Hale (1996). Ergativity: Toward a theory of a heterogeneous class. Linguistic Inquiry 27, 1-68. Blutner, R. (2000). Some aspects of optimality in natural language interpretation. Journal of Semantics 17, 189-216. Bobaljik, J.D. (1993). Nominally absolutive is not absolutely nominative. Proceedings of the Eleventh West Coast Conference on Formal Linguistics. Ed. by J. Mead. Stanford: CSLI Publications, 44-60. Butt, M. and T.H. King (2003). Case systems: Beyond structural distinctions. New Perspectives on Case Theory. Ed. by E. Brandner and H. Zinsmeister. Stanford: CSLI Publications. Comrie, B. (1989). Language Universals and Linguistic Typology. Chicago: University of Chicago Press. Davison, A. (1999a). Ergativity. Functional and formal issues. Functionalism and Formalism in Linguistics. Ed. by M. Darnell, E. Moravcsik, F. Newymeyer, M. Noonan and K. Wheatley. Amsterdam: John Benjamins, 177-207. Davison, A. (1999b). 'Dependent structural case' as a consequence of VP structure. Paper presented at the Perspectives on argument structure: 1999 Texas Linguistics Society Conference. DeLancey, S. (1981). An interpretation of split ergativity and related phenomena. Language 57, 627-657. Dixon, R.M.W. (1979). Ergativity. Language 55, 59-138. Dowty, D. (1991). Thematic proto-roles and argument selection. Language 67, 3, 547-619. de Hoop, H. (1996). Case Configuration and Noun Phrase Interpretation. New York: Garland. de Hoop, H. (1999). Optimal Case Assignment. Linguistics in the Netherlands 1999. Ed. by R. van Bezooijen and R. Kager. Amsterdam: John Benjamins, 97-109. de Hoop, H. and B. Narasimhan (2005). Differential case-marking in Hindi. Competition and Variation in Natural Languages: The Case for Case. Ed. by M. Amberber and H. de Hoop. Oxford: Elsevier. Hopper, P.J. and S.A. Thompson (1980). Transitivity in grammar and discourse. Language 56, 2, 251299. Horn, L. (1984). Towards a new taxonomy of pragmatic inference: Q-based and R-based implicature. Meaning, Form, and Use in Context: Linguistic Applications. Ed. by D. Schiffrin. Washington: Georgetown University Press, 11-42. Kiparsky, P. (2001) Structural case in Finnish. Lingua 111, 315-376. Klamer, M. (this volume) Differential subject marking in Kambera (Austronesian). Kornfilt, J. (this volume). DOM and two types of DSM in Turkish. Lee, H. (2001). Markedness and word order freezing. Formal and Empirical Issues in OptimalityTheoretic Syntax. Ed. by P. Sells. Stanford: CSLI Publications. Lee, H. (2003). Parallel optimization in case systems. Nominals: Inside and Out. Ed. by M. Butt and T.H. King. Stanford, CSLI Publications. Lee, H. (this volume) Quantitative variation in Korean case ellipsis: implications for case theory. Legendre, G., W. Raymond, and P. Smolensky (1993). An Optimality-Theoretic typology of case and grammatical voice systems. Proceedings of the 19th Meeting of the Berkeley Linguistics Society. Berkeley, Berkeley Linguistics Society. Malchukov, A.L. (2005). Case pattern splits, verb types, and construction competititon. Competition and Variation in Natural Languages: The Case for Case. Ed. by M. Amberber and H. de Hoop. Oxford: Elsevier. Malchukov, A.L. (2006). Transitivity parameters and transitivity alternations: Constraining co-variation. Case, Valency, and Transitivity. Ed. by L.I. Kulikov, A.L. Malchukov and P. de Swart. Amsterdam: John Benjamins, 329-357. Mohanan, T. (1994). Argument Structure in Hindi. Stanford: Stanford University. Silverstein, M. (1976). Hierarchy of Features and Ergativity. Grammatical Categories in Australian Languages. Ed. by R.M.W. Dixon. Canberra: Australian Institute of Aboriginal Studies, 112-171. Stiebels, B. (2000) Linker inventories, linking splits and lexical economy. Lexicon in Focus. Ed. by B. Stiebels and D. Wunderlich. Berlin: Akademie Verlag, 211-245. de Swart, P. (2003). The Case Mirror. MA-Thesis, University of Nijmegen. Woolford, E. (1995). Object agreement in Palauan: Specificity, humanness, economy and optimality. Papers in Optimality Theory. University of Massachusetts Ocasional Papers 18. Ed. by J.N. Beckman, L.W. Dickey, and S. Urbanczyk. GLSA, Amherst.
78
HELEN DE HOOP AND BHUVANA NARASIMHAN
Woolford, E. (this volume) Differential subject marking at argument structure, syntax, and PF. Wunderlich, D. (1997) Case and the structure of verbs. Linguistic Inquiry 28, 27-68. Wunderlich, D. and R. Lakämper (2001) On the interaction of structural and semantic case. Lingua 111, 277-418.
JAKLIN KORNFILT
CHAPTER 5 DOM and Two Types of DSM in Turkish1
1. INTRODUCTION Turkish has a number of interacting systems of Differential (or Differentiated) Subject Marking (DSM). This paper attempts a characterization of the main types, while also looking at Differential Object Marking (DOM). I use the terms DSM and DOM in a descriptive way here, to refer to the fact that some subjects and objects are morphologically marked for Genitive and Accusative, respectively, while some other subjects and objects are not marked in this way. However, this is not how these terms have come to be used in most of the literature that appeals to these terms. As we shall see in section 2 of the text, there is a good deal of work (some of which represented in this volume), where the term ‘differential’ has a related, but distinct use within a particular functionalistic approach, based on views of archetypical subjects versus archetypical objects, viewing particular tokens of subjects and objects as ‘differentially’ marked with respect to such archetypes. I argue in this paper that Turkish facts of overt (structural) case marking, taken in some of the relevant literature as illustrative of this relativistic, functional view, actually should not be characterized in this way, but rather in an ‘absolute’ fashion, with formal conditions accounting for the presence versus absence of overt structural case markers. I limit myself to DSM as overt, i.e. morphological, Case marking on the subject, and to DOM as its counterpart on the object. Two main types of factors mentioned in the introduction to this volume — factors that can result in DSM — are both found in Turkish: differentiation of subjects depending on the subject’s own semantic features (here, specificity/referentiality — this differentiation being one that is also found on objects with respect to DOM), and differentiation on the basis of clausal features (here, 1. differentiation between the subjects of argument and 1
I would like to thank Helen de Hoop for inviting me to the workshop on DSM, held in Nijmegen in July 2004, that led to this volume, and to the writing of this paper; I am further indebted to her for her help with finding some bibliographic references as well as their sources. I am indebted to her, to Ellen Woolford, and to other members of the audience at that workshop, for useful comments. I am also grateful to Peter de Swart, for his close reading of a previous draft of this paper and his many insightful comments, as well as to two anonymous internal reviewers and to two anonymous external reviewers, for their comments. Any mistakes, misconceptions and awkward analyses are my responsibility entirely. Tessel van Balen deserves deep appreciation for her unique organizing skills. I am grateful to the various travel funding resources in Syracuse University that made my participation at the DSM workshop possible.
79 H. de Hoop and P. de Swart (eds.), Differential Subject Marking, 79–111. © 2008 Springer.
80
JAKLIN KORNFILT
adjunct clauses of a particular nominalization type and 2. differentiation between the Nominative subjects of fully verbal clauses and the Genitive subjects of fully nominal clauses).2 I then turn to instances of neutralization of DSM. First, I show that non-specific subjects must be overtly marked with the appropriate morphological case marker (despite their semantic feature which normally precludes such case marking), when the presence of the marker needs to be overt due to certain morpho-syntactic constraints; I also show that the same is true for non-specific direct objects, as well. In other words, in these instances, the morphological expression of structural Case, i.e. of the subject as well as the object Case, shows up where it is not expected. I then discuss another kind of neutralization which is in a sense the mirror-image of the first: I show that in certain instances, the second type of DSM (i.e. DSM which depends on clausal features) leads to the neutralization of the first type (i.e. of DSM which depends on semantic features): The nominal subject case, i.e. the Genitive, is expected on a specific subject, yet it doesn’t show up on such a subject, when the syntax of the clause doesn’t license the Genitive. (This is the case, as we shall see, with a certain type of gerundive adjunct clause). Thus, we have a neutralization of DSM (i.e. of the overt realization of specificity via case marking) in both logically possible directions: Non-specific subjects which ought to reject overt case marking but do exhibit such marking, and specific subjects which ought to bear overt case marking but don’t. Note that the first type of neutralization is also found with respect to DOM, while the second type is found only with respect to DSM. Presumably, this is because clausal features are realized, in Turkish, at levels higher than VP. The following generalization emerges: formal (clausal as well as phrasal) reasons for differential subject (and object) marking override factors for DSM (and DOM) based on semantic features, at least in Turkish. Whether this generalization is valid cross-linguistically I leave for future research. Finally, I argue in this paper that DSM (and thus also DOM) can be analyzed as absolute phenomena for Turkish, i.e. without comparing archetypical subjects and objects. If a relative view of DSM and DOM is viable for Turkish at all, then it would be a view that for DSM compares specific (and thus potentially topicalizable) subjects (‘strong’ subjects, in the terminology of de Hoop and Narasimhan 2005 and this volume) with other subjects. The same view, with respect to DOM, would again compare specific (and thus also potentially topicalizable and ‘strong’) objects with other objects. Again, in de Hoop and Narasimhan’s terminology, this view would embody the identificational function of differential case marking, and crucially not the distinguishing function. This, then, could be formalized as an absolute view which is motivated by information-structural considerations — very plausible for a language such as Turkish whose free word order is likewise driven by such considerations, and is equally formalizable — in this instance, via phrase structural
2
Here, I mention the difference in 2. only in passing. I concentrate on the interactions between the two types of subjects mentioned in 1. with respect to clausal features, and on the DSM with respect to marking of semantic features.
DOM AND TWO TYPES OF DSM IN TURKISH
81
architecture that includes, for example, a Topic Phrase among its highest functional projections. The paper is structured as follows: Differential case marking is introduced in section 2; first DOM, and then DSM, both as based on semantic features, are illustrated. Section 3 discusses formal, morpho-syntactic conditions on both DOM and DSM, and illustrates how these conditions neutralize both types of differential marking based on semantic features. In this section, we also see that partitives can be non-specific, against the claim discussed at length in Enç (1991), and adopted very widely since then, that partitives must be specific (and that specific DP are, in a sense, partitive). Section 4 discusses the second type of DSM, namely the type based on clausal properties, and illustrates how such properties override the first type of DSM, namely the type based on semantic features. In this same section, a proposal is sketched that provides an account for the directions in which these neutralizations proceed. A subsection (4.5.) discusses a related approach to DSM and DOM, and argues that this account, which is otherwise developed within a similar spirit as the account here, cannot be applied to the Turkish differential marking — unless that account is modified. Section 5 offers concluding remarks. 2. DIFFERENTIATION BASED ON SEMANTIC FEATURES 2.1 DOM based on semantic features I start with DOM, as it is the phenomenon that is more widely known for Turkish, and will then turn to DSM: (1) a. (ben) kitab-ı oku-du-m I book-ACC read-PST-1SG ‘I read the book.’ b. (ben) bir kitap oku-du-m I a book read-PST-1SG ‘I read a book.’ c. (ben) bir kitab-ı oku-du-m I a book-ACC read-PST-1SG ‘I read a certain book.’
(definite) (indef. non-spec.) (indef. spec.)
This triplet shows that the relevant notion behind the overt Accusative marker is not definiteness, but specificity: A specific direct object bears Accusative marking, even if it is indefinite, as in (1c).3 Similar facts concerning correlations between semantic features and overt case marking are found in a variety of languages (cf. Comrie 1975). The semantic features in question may be different from language to language; thus, other than specificity, one can find animacy or definiteness as the determining factors behind object case marking: whenever the relevant feature (or features) in a particular 3
Similar observations have been made in Enç (1991), Erguvanlı-Taylan (1984), Kornfilt (1984).
82
JAKLIN KORNFILT
language have positive value, the object case marking shows up; whenever the relevant feature(s) has/have negative value, the object case marking is missing. (For example, Lee, in this volume, addresses the issue of how animacy and definiteness are factors in Korean Case Ellipsis.) In some typological literature, such facts have been described as showing that the markers in question are not object Case markers, but rather markers for the semantic features in question (e.g. for Turkish, this would mean that -(y)I would be a specificity marker rather than an Accusative marker — or, perhaps, it would be a combined specificity-Accusative marker; Lewis (1967) — for whom the relevant notion is definiteness, rather than specificity — goes in this direction, when he refers to the Accusative in Turkish as a ‘defined accusative’, which he contrasts with the ‘indefinite accusative’). Some scholars (most notably, Comrie 1975 as an early and insightful instance, and Silverstein 1976, in a much-quoted paper) have analyzed facts of this type as an expression of a system of overt markers, whereby the existence or absence of the object marker in question can be understood only in relation to the nature of the subject, as a means to differentiate between 1. non-archetypical objects and archetypical subjects, and 2. between non-archetypical objects and archetypical ones. The term ‘Differentiated Object Marking’ (DOM) is generally used to characterize this approach, which is based on a relative view of object marking. The proposal in such work is that in languages with DOM-effects, only a direct object which is ‘too similar’ to the subject with respect to one or more relevant semantic features will bear a special ‘object case’ marker. An archetypical subject is assumed to have positive values for as many features such as definiteness, specificity, animacy as possible. In a Nominative/Accusative language, an ‘archetypical’ direct object (i.e. a direct object which has negative values for the same features for which an archetypical subject has positive values) will not have any special case marking and will appear as though it were Nominative, because it will be inherently different from a typical subject (and thus won’t need any marking to differentiate it from such a subject). If in such a language the Nominative case marker is a null morpheme (as it is in Turkish), an archetypical direct object won’t carry overt case marking. In contrast, an ‘untypical’ (i.e. definite and/or specific and/or animate) direct object in such a language would receive a special Accusative marker to set it apart from a typical subject. This is indeed what we appear to find in Turkish, as illustrated in (1): the relevant semantic feature is specificity. The subject in these examples is indeed specific, and thus ‘archetypical’, and it is in the Nominative. The direct object is specific in both (1a) and (1c), and thus is a non-typical direct object which therefore, under the approach just mentioned, needs to be differentiated both from an archetypical subject and from an archetypical object. Therefore, it bears a special case marker: -(y)I. In contrast, the direct object in (1b) is non-specific and is thus a typical object. It therefore needn’t be differentiated from the subject, and thus it shows up in the same shape with respect to case marking as the subject, namely without any case marking (and thus, at least apparently, in the Nominative). I will show in the course of this paper that appearances notwithstanding, this functional approach to DOM, based on notions of archetypical subjects and objects, has to be either limited much more strictly or else has to be abandoned — at least for
DOM AND TWO TYPES OF DSM IN TURKISH
83
languages such as Turkish, once DSM (in the descriptive sense) is considered, as well. 2.2 DSM based on semantic features Some languages have Differential Subject Marking (DSM), as well; in other words, not all subjects are marked in the same way. In languages that have both DOM and DSM, approaches that are based on a functionalistic view that bases morphological marking on differentiating objects and subjects from their respective archetypes and from each other would predict that DSM should be the mirror image of DOM: An archetypical subject should have positive values for relevant features such as animacy, specificity etc. Thus, a typical subject should not bear any special, ‘differential’ marking; however, an untypical subject that has negative values for the relevant features should bear such differential marking.4 Interestingly, Turkish allows for subject marking in clearly defined contexts. However, specific subjects are morphologically marked, while non-specific ones are not. While the subject in a matrix sentence cannot be case marked (which is another way of stating that in Turkish, the Nominative case is morphologically null), the subject of an embedded clause can be case marked by the Genitive suffix. The subject must bear overt Genitive marking in nominalized indicative (argument) clauses (which are quite similar to English gerundive clauses when those are argument clauses), as well as in nominalized subjunctive clauses, which have sometimes also been referred to as inflected infinitival clauses.5 Johanson (1977:1196) and Kornfilt (1984, 1997:215) notice that if the subject is not in a position directly before the verb, it must get case, as in (2). Here the contrast between specific and non-specific is neutralized for many speakers.6 If the subject takes the position directly preceding the verb, Genitive case marking indicates specificity for most speakers, as illustrated by (3). 4
As a matter of fact, Aissen (2003:473) discusses DSM under this view, i.e. with the explicit expectation that DSM should enforce morphological marking of indefinite, non-specific, inanimate subjects. 5 As we shall see in some detail in section 4, the two types of nominalized embedded clause mentioned here differ with respect to the case marking on their subjects. While the nominalized subjunctives, which I later claim are homogeneously nominal in their functional architecture, always have Genitive subjects (provided that those subjects are specific), the nominalized indicatives (which I claim are categorially hybrid in the sense of having mixed features in their functional projections, with verbal functional projections dominated by nominal functional projections; cf. Kornfilt 2003 and 2006) have Genitive subjects only when these clauses are themselves arguments of the matrix predicate; when these hybrid clauses are adjuncts, their subjects must be bare, i.e. cannot be Genitive, even if they are specific. An account of this alternation which is independent from semantic features will be attempted in section 4. 6 This appears to be the case for all of Johanson’s informants. While some of my own informants appear to have similar intuitions, things are somewhat more complex for the majority of my informants, and for myself. We share the intuition that direct objects that have no Accusative marking cannot appear separated from the verb with the exception of some ill-understood circumstances when bare nonspecific objects can be separated from the verb; however, for us, direct objects that do bear Accusative morphology are interpreted as specific, even they are separated from the verb; in other words, the neutralization between specific and non-specific readings doesn’t take place.
84
JAKLIN KORNFILT (2) [bir haydut-un köy-ü bas-tığ-ın]-ı duy-du-m a robber-GEN village-ACC raid-FN-3SG-ACC hear-PST-1SG ‘I heard that a robber (specific for all speakers, and additional non-specific reading for some speakers) raided the village.’ (3) a. [köy-ü bir haydut-un bas-tığ-ın]-ı duy-du-m robber-GEN raid-FN-3SG-ACC hear-PST-1SG village-ACC a ‘I heard that a (certain) robber raided the village.’ (specific for all speakers) b. [köy-ü haydut bas-tığ-ın]-ı duy-du-m village-ACC robber raid-FN-3SG-ACC hear-PST-1SG ‘I heard that robbers raided the village.’ (non-specific, generic reading as the only reading)
The following examples behave similarly; we see here that a non-specific subject which is positioned immediately before the verb does not bear overt Genitive (4a). If it does exhibit Genitive case marking in this position as in (4b), it gets interpreted as specific by most speakers, as the primary reading. Further, we see that such a subject, when it is separated from the verb, can be interpreted as non-specific by some speakers even when it bears Genitive marking as in (4c) — however, as explained in footnote 6, for most speakers, Genitive subjects are specific, irrespective of their position. Finally, such a subject cannot (usually) show up without the Genitive marking in a position removed from the verb as in (4d): (4) a. [yol-dan bir araba geç-tiğ-in]-i gör-dü-m car pass-FN-3SG-ACC see-PST-1SG road-ABL a ‘I saw that a car (non-specific, non-referential) went by on the road.’ (The subject may be focused, but it does not have to be.) b. [yol-dan bir araba-nın geç-tiğ-in]-i gör-dü-m car-GEN pass-FN-3SG-ACC see-PST-1SG road-ABL a ‘I saw that a car (indefinite, but specific) went by on the road.’ c. [bir araba-nın yol-dan geç-tiğ-in]-i gör-dü-m road-ABL pass-FN-3SG-ACC see-PST-1SG a car-GEN ‘I saw that a car (indefinite, and specific or non-specific) went by on the road.’ d. *[ bir araba yol-dan geç-tiğ-in]-i gör-dü-m a car road-ABL pass-FN-3SG-ACC see-PST-1SG Intended reading: ‘I saw that a car (indefinite and non-specific, nonreferential) went by on the road.’ Note the contrast with the grammatical (a). Similar facts hold in existentials — this is expected, as the ‘semantic’ subjects are non-specific:7
7
One of the reviewers suggests that the alternation between Genitive marking and zero marking can be reduced to the alternation between overt and zero Accusative marking, based on an analysis in
DOM AND TWO TYPES OF DSM IN TURKISH
85
(5) a. [Garaj-da beş araba ol-duğ-un]-u bil-iyor-um garage-LOC five car be-FN-3SG-ACC know-PROG-1SG ‘I know that there are five cars in the garage.’ b. *[ Beş araba garaj-da ol-duğ-un]-u bil-iyor-um five car garage-LOC be-FN-3SG-ACC know-PROG-1SG Intended reading: ‘I know that there are five cars in the garage.’ The restrictions for case marking are very similar for the Genitive case suffix in embedded sentences and for the Accusative case suffix. The case suffix is obligatory in general if the NP/DP is not to the immediate left of the verb, while in a position left-adjacent to the verb the case suffix signals (for most speakers, and as a primary reading) specificity, and its absence non-specificity. This observation contradicts the prediction of a functional, archetype-based approach to DOM/DSM. In that theory the respective markers of structural (i.e. non-lexical) case indicate two types of difference: 1. a difference between the morphologically marked direct object or subject and their respective kinds, i.e. the classes of archetypical direct objects and subjects, respectively; 2. a difference between the morphologically marked object and the subject, or between the morphologically (un)marked subject and the direct object. Comrie (1975) and later work that was conducted in a similar spirit argue that morphological structural case markers essentially indicate that the object or subject on which they appear is different from what is expected. In other words, the appearance of these markers is determined by relativistic considerations based on archetypes, under such an approach. In contrast with the predictions made by this relativistic approach to DOM and DSM, the observations from Turkish strongly suggest that the structural case markers (i.e. Accusative and Genitive) indicate specificity (unless this is overridden by formal considerations, as we shall see later). These markers do not reflect a semantic difference from the expected form, but rather express overtly the semantic feature of specificity in a similar way for both objects and subjects (when they are not neutralized). This doesn’t mean that no functional explanation is available at all — just that a relativistic, archetype-based explanation is not available. For example, one could propose that topics are usually definite, specific, animate etc., and that
Kennelly (1997), where it is claimed that such subjects are always objects. Indeed, the examples in (4) and (5) appear with unaccusative predicates, and their subjects can thus be analyzed as complements rather than specifiers. However, the examples in (3), where the verb is transitive, are of course problematic for such an analysis, and they are not exceptional. Other such examples with transitive predicates can be found, where the subject displays a similar alternation between exhibiting Genitive morphology or not, with the concomitant interpretations of specificity versus lack of specificity. The generalization is that non-specific and morphologically bare subjects of such nominalized clauses are (generally) confined to pre-verbal position. While this makes them similar to non-specific objects, this doesn’t mean that they are objects. Instead, these facts suggest a Case-based analysis such as the one in Kornfilt (1984). Transposed into more recent approaches, this would mean that bare direct objects get their structural Case feature checked by v, while bare subjects get their structural case feature checked by INFL (=Tense and/or Agreement as a bundle of phi-features) in situ, i.e. in their basegenerated VP-internal position, under long-distance Agree.
86
JAKLIN KORNFILT
therefore both subjects and objects that have topic-oriented features will have overt structural Case markers that mark them as potential topics, enabling them to move out of vP/VP. This proposal fits the facts of Turkish much better than any archetypebased account. I shall not formalize this alternative here, referring the reader to Jelinek and Carnie’s (2003) discussion, which is developed in a similar spirit as the account I sketch here.8 Here, my goal is only to present the Turkish facts and show that they argue against a relativistic, archetype-based account of those facts. 3. MORPHO-SYNTACTIC CONDITIONS ON DOM AND DSM AND THEIR NEUTRALIZATION I propose the following morpho-syntactic conditions on DP/NPs in Turkish: 1. Turkish DP/NPs must have an overt nominal head; 2. When that head is realized as overt Agreement in apparently headless DP/NPs, it necessitates overt structural Case (probably due to the pronominal nature of Agreement, which therefore is likely to include default, formal specific features, independently from semantic specificity — see also footnote 13; note also that Morimoto, this volume, argues that agreement markers in Bantu correlate with topicality, which might well be a related phenomenon). Since such instances of structural Case are required by this formal condition, the expression of specificity is neutralized, and consequently a nonspecific expression can be marked with overt structural Case. I shall illustrate this phenomenon via partitive constructions in Turkish. Note that, as a side issue, this means that partitive expressions can be non-specific, contra Enç (1991); cf. Kornfilt (2001) and van Heusinger & . Kornfilt (2005). 3.1 Non-specific partitives with and without structural Case markers It can be shown that subset expressions of partitive constructions can be non-specific (and lack Accusative marking), even with specific supersets (and that non-specific expressions can be partitive, i.e. can function as subset expressions in partitives): (6) Ali kadın -lar-dan iki kişi tanı-yor-du Ali woman-PL-ABL two individual know-PROG-PST ‘Ali knew two (unspecified) individuals of the women.’ çocuk-lar-dan iki kız al-acak (7) Ali büro-ya Ali office -DAT child-PL-ABL two girl take-FUT ‘Ali will hire (any) two girls of the (group of) children.’ Note that despite the semantic as well as syntactic partitivity of the italicised expressions, the constituents do not bear overt Accusative marking, and they are, 8
De Hoop and Narasimhan (this volume) and (2005) offer similar arguments based on Hindi, where DOM effects (as shown by alternations in Accusative marking) and DSM effects (as illustrated by alternations in Ergative marking) are based on similar conditions (referred to as argument strength by the authors), and do not exhibit mirror-image effects.
DOM AND TWO TYPES OF DSM IN TURKISH
87
indeed, non-specific. In other words, although the ablative superset part of the partitive is specific, the subset expression (which is the syntactic head of the partitive) is non-specific; therefore, just as expected, the partitive expression — a direct object in these examples — lacks Accusative (i.e. object) morphology. A similar point can be made by illustrating ‘bare’ ablative partitives: these are ablative partitives with phonologically unrealized heads: (8) Ali şarap-tan iç-ti Ali wine-ABL drink-PST ‘Ali drank (an unspecified amount) of the wine.’ (9) Ali balık-tan ye-di Ali fish-ABL eat-PST ‘Ali ate (an unspecified amount) of the fish.’ (10) Biz-de bu kitap-tan var /yok /kal-ma-dı we-LOC this book-ABL exist /NEG+exist /remain-NEG-PST ‘We have/don’t have any (copies) of this book; we don’t have any (copies) of this book left.’ Note that the subset of this expression is interpreted as being non-specific, despite the obvious partitivity of the expression.9 I now turn to examples used by Enç (1991) to establish her claim that specificity is (or rather, in many instances, can be) based on partitivity. Enç argues that partitive expressions must be specific, and that therefore in Turkish, partitives which are direct objects must bear overt Accusative (which she takes as an unequivocal diagnostic for specificity); where such marking is missing, the result is ill-formed. This is illustrated in the following examples, taken from Enç (1991): (11) a. Ali kadın-lar-ın iki-sin-i (=E.’s 28) Ali woman-PL-GEN two-AGR(3)-ACC ‘Ali knew two of the women.’ b. *Ali kadın-lar-ın iki-si tanıyordu (12) a. Ali kadın-lar-dan iki-sin-i (=E.’s 29) Ali woman-PL-ABL two-AGR(3)-ACC ‘Ali knew two of the women.’ b. *Ali kadın-lar-dan iki-si tanıyordu
tanı-yor-du know-PROG-PST tanı-yor-du know-PROG-PST
Note that the a.-examples (where overt Accusative marking shows up on the partitive direct object) are fine, while the b.-examples lacking such marking are illformed. Enç interprets these contrasts as evidence for her claim that partitive expressions must be specific: since in Turkish specific direct objects must bear Accusative (as she claims — and, in general, this is indeed so, as we have seen earlier), lack of
9
For discussions of partitives in Turkish, and, more specifically, of Turkish bare partitives, the reader is referred to Kornfilt (1984) and (1996a).
88
JAKLIN KORNFILT
Accusative, according to her, leads to ungrammaticality, assuming that these partitives must be specific. However, we just saw earlier in this section that there can be partitives which are non-specific and which can lack overt Accusative. Therefore, Enç’s explanation for the ungrammaticality of (11b) and (12b) can’t be right; the ill-formedness of these examples must be due to a different reason. I now turn to a discussion of what such a different reason, or reasons, might be. Accusative marking, while in general a reliable expression of specificity, may be misleading: it can conceal a non-specific expression. This is so when the appearance of the Accusative marking is due to formal reasons, e.g. to morphological requirements: the nominal agreement marking (in those constructions where it does show up, such as in certain partitive constructions that are direct objects) has to be followed by the Accusative in a transitive context. The agreement marking itself is an expression of specificity in general, as has been observed in the literature previously; the reason for this might be the pronominal features of the agreement morphology. However, this morphology can also conceal non-specificity when it appears in head-noun positions of nominal phrases — this in constructions when a nominal head is required but no lexical (nominal) head is available: (13) Kitap-lar-ın iki-sin-i al, book-PL-GEN two-AGR(3)-ACC buy geri-sin-i kutu-da bırak box-LOC leave remainder-AGR(3)-ACC ‘Take (any) two of the books and leave the remainder [of the books] in the box.’ (14) Kitap-lar-dan iki-sin-i al, book-PL-ABL two-AGR(3)-ACC buy geri-sin-i kutu-da bırak box-LOC leave remainder-AGR(3)-ACC Same readings as in (13) In these examples, the Accusative subset may be specific or, crucially, nonspecific for both the Genitive partitive and the Ablative partitive construction (as shown by the English translations). The Accusative is obligatory in both constructions (for the non-specific reading as well as the specific reading), as shown in (15) and (16): (15) *Kitap-lar-ın iki -si al, geri-sin-i kutu-da bırak book -PL-GEN two-AGR(3) buy remainder-AGR(3)-ACC box-LOC leave (16) *Kitap-lar-dan iki-si al, geri-sin-i kutu-da bırak two-AGR(3) buy remainder-AGR(3)-ACC box-LOC leave book-PL-ABL Intended readings: Same as the non-specific readings in (13) and (14), for both examples.
DOM AND TWO TYPES OF DSM IN TURKISH
89
Note that the Accusative marker is missing. Note also that these examples are illformed even under an intended non-specific reading of the partitive subset; thus, the missing Accusative leads to ill-formedness independently from specificity. Similar facts also obtain for partitives that are subjects in (relevant) nominalized embedded sentences and which therefore should be marked with the Genitive: (17) [Kitap-lar-ın iki-sin-in kaybol-duğ-un]-u bil-iyor-um book-PL-GEN two-AGR(3)-GEN disappear-FN-3SG-ACC know -PROG-1SG (ama hangileri olduğunu bilmiyorum) ‘I know that (any) two of the books got lost (but I don’t know which ones).’ (18) [Kitap-lar-dan iki-sin-in kaybol-duğ-un]-u bil-iyor-um two-AGR(3)-GEN disappear-FN-3SG-ACC know-PROG-1SG book-PL-ABL (ama hangileri olduğunu bilmiyorum) Same readings as in (17) Just like the Accusative for direct objects, the Genitive is obligatory here for subjects, even for non-specific readings: kaybol-duğ-un]-u bil-iyor-um (19) *[Kitap-lar-ın iki-si10 book-PL-GEN two-AGR(3) disappear-FN-3SG-ACC know-PROG-1SG (ama hangileri olduğunu bilmiyorum) (20) *[Kitap-lar-dan iki-si kaybol-duğ-un]-u bil-iyor-um two-AGR(3) disappear-FN-3SG-ACC know-PROG-1SG book-PL-ABL (ama hangileri olduğunu bilmiyorum) Intended readings: Same as the non-specific readings in (17) and (18), for both examples. I propose that the reason for the ill-formedness of (15) and (16), as well as of (19) and (20) has nothing to do with the semantic feature of specificity per se. Rather, this ill-formedness is linked to the fact that in all of these examples, the third person singular agreement morpheme is required, due to the following constraint (which holds of Turkish, and probably a good number of other languages): THE OVERT NOMINAL HEAD CONSTRAINT (ONHC) NP/DPs must have an overt head, occupied by nominal features. We thus have in these examples an agreement element which takes the place of a missing lexical N in head position; in other words, when an NP/DP lacks a lexical nominal head, an Agr morpheme shows up instead, motivated by the ONHC11 and,
10 11
The morpheme-final n of the third person singular agreement morpheme is deleted in word-final position. When the ‘superset’ expression of the partitive construction is not a third person plural expression, as in these examples, but rather is first or second person plural, the required agreement marker bears the phi-features of the superset. In such constructions, the superset is usually pro.
90
JAKLIN KORNFILT
in turn, this morpheme requires an overt Case marker (i.e. Accusative or Genitive, depending on the object or subject status of the entire DP/NP). To see this clearly, compare the Ablative partitive constructions, which do not exhibit an agreement morpheme when they have a lexical nominal head (such as in (6) and (7)) with (12a), (14), and (18), where the partitive phrase has no lexical head. The agreement morpheme shows up in the latter group only, and it is obligatory, as illustrated by (22b) and (23), where there is neither a lexical noun nor the expletive agreement morpheme to head the partitive phrase; note that even the presence of the appropriate Case marker cannot save these examples from being ill-formed.12 The requirement of the overt agreement marker is similar to English, when a nominal NP/DP head like one is necessary: (21) I compared various models and bought the new *(one).13 The agreement element in (13) and (14) as well as in (17) and (18) functions as the morpheme one in English, i.e. the overt nominal head of the nominal phrase, 12
The ONHC holds of NP/DPs in general, not just of partitive phrases; e.g.: (i) a. Üç kitap al-dı-m b. Üç tane al-dı-m c. *Üç al-dı-m three book buy-PST-1SG three ‘unit’ buy-PST-1SG three buy-PST-1SG ‘I bought three books.’ ‘I bought three ‘entities.’ Intended reading: ‘I bought three.’ Note that in (ic), the Turkish equivalent to the grammatical English translation is ill-formed. The closest one can get to the English translation of (ic) is (ib), with a morpheme matching the numeral in the feature ‘countable’. It is also possible to have, in the context of these examples, the numeral with a third person singular agreement morpheme: (i) d. Üç-ün-ü al-dı-m three-3SG-ACC buy-PST-1SG ‘I bought three of them.’ In other words, while the unmarked third person singular (nominal) agreement marker can satisfy the ONHC, as we have seen in the text, it presupposes a partitive superset. These facts, among others, had originally motivated me to propose (cf. Kornfilt 1984) a functional projection of AgrP for nominal phrases such as partitives and possessives, possibly translatable into DPs under more recent approaches. For our purposes, then, we could say that when a DP/NP (in Turkish) has no lexical head, an Agr morpheme is one way of satisfying the constraint; it is inserted into the N-position and raises to D. Some informants have reported to me that they accept violations of the ONHC with adjectival modifiers lacking an overt nominal head: (ii) Beyaz-ı yıka-dı-m, ama mavi-yi yıka-ma-dı -m white-ACC wash-PST-1SG but blue-ACC wash-NEG-PST-1SG ‘I washed the white (one) but didn’t wash the blue (one).’ For me, as well as for most of my informants, utterances such as (ii) are ill-formed. The question is, then, why some speakers do accept such examples (however only with adjectival modifiers, not with numeric or otherwise quantificational ones). I suspect that this is because in Turkish, adjectives are in general usable as nouns; e.g. büyük ‘big, old’, but also ‘elder’; genç ‘young’, but also ‘the young’. Thus, for speakers who accept utterances such as (ii), the adjective is itself a noun and occupies the head N-position of the NP/DP. 13 Some English speakers have reported to me that they find this example (in the version without ‘one’) barely acceptable in certain limited discourse contexts. All speakers I have consulted agree that such examples are problematic, while the corresponding utterances with the pronoun one are completely well-formed.
DOM AND TWO TYPES OF DSM IN TURKISH
91
and, as we just saw, itself requires the overt structural case, whether Accusative or Genitive.14 Such examples are, as just mentioned, ill-formed without the agreement morpheme, even when overt Case marking is present as in (22b): (22) a. Çeşitli model-ler-i karşılaştır-dı-m ve yeni-sin-i various model-PL-ACC compare-PST-1SG and new-AGR(3)-ACC al-dı-m buy-PST-1SG ‘I compared various models and bought the new (one).’ b. *Çeşitli model-ler-i karşılaştır-dı-m ve yeni-yi al-dı-m various model-PL-ACC compare-PST-1SG and new-ACC buy-PST-1SG ‘I compared various models and bought the new (one).’ This example becomes ungrammatical, when the agreement morpheme is omitted (even if the Case marker is present) — entirely comparable to the effect that omission of one has in the English example in (21). The same is true for examples with the Genitive instead of Accusative. Similar facts also obtain in partitive constructions with ‘bare’ quantifier heads, where the Case marker is not sufficient to save the construction whose illformedness is due to lack of a nominal head: (23) a. *Kitap-lar-dan iki/iki-yi al-dı-m two/two-ACC buy-PST-1SG book-PL-ABL Intended reading: ‘I bought two of the books.’ The way to ‘save’ such an example is to attach agreement morphology to the quantifier (again, the agreement morphology requires the appropriate structural Case marker): (23) b. Kitap-lar-dan iki-sin-i al-dı-m book-PL-ABL two-AGR(3)-ACC buy-PST-1SG ‘I bought two of the books.’ (specific or non-specific) As the well-formed examples show, the structural Case marker (i.e. Accusative or Genitive) may conceal a non-specific reading. While these markers correlate with specificity in general, they are unreliable in this capacity when they are needed for other reasons — as here, where the markers are due to the presence of agreement morphology, itself needed to provide a nominal head to the partitive phrase. In other words, when agreement morphology is used to meet the requirement of DP/NPs to have a nominal head, i.e. when this morphology is needed for formal reasons and
14
The reason for this requirement might well be that agreement has pronominal features, and pronouns are, in general, specific. If this explanation is on the right track, then the Accusative would be necessitated by formal specificity features rather than semantic specificity, because, as we saw in the text, all of the well-formed examples in this section have a non-specific interpretation.
92
JAKLIN KORNFILT
thus itself requires Case morphology, the structural Case markers are neutralized in their function as specificity markers. As a peripheral issue for our purposes here, we also have an answer to the question of why Enç’s examples are ill-formed: (11b) and (12b) are ungrammatical, but not because the partitive expressions are specific and thus need the Accusative, but rather because the partitive expressions in those examples don’t have a lexical nominal head. Instead, they have agreement as a provider of a nominal head, and it is because of the presence of agreement in this capacity that the structural Case marker is needed here, rather than as a marker for specificity. Indeed, in Enç’s wellformed examples, just as in other similar examples, there is a second reading under which the subset expression — and thus the entire partitive phrase — is non-specific (despite the presence of Accusative). Furthermore, the partitive expressions in (6) and (7) are well-formed despite lack of overt structural Case; this clearly shows that partitive expressions, just like other DP/NPs, may be non-specific, and when they are, they reject structural Case — unless, as we saw above, they need such Case for formal reasons; when they do, they can still maintain a non-specific reading, at least as one of the possible interpretations. Another instance of the formal usage of agreement morphology requiring the existence of Case morphology has to do with Genitive partitive constructions that do have lexical heads. This discussion will require a brief detour into some of the types of partitive constituents in Turkish. 3.2 Types of Turkish partitive constructions at the intersection of Agreement and Case As we have seen in the course of the discussion above, Turkish has two main types of partitive constructions: one type where the superset expression bears the Ablative, and one where the superset expression shows up in the Genitive. (A third kind which I called ‘bare partitive’ and which was also illustrated above is a subtype of the Ablative partitive.) When it has a lexical nominal head, the Ablative partitive has no agreement morpheme on its head (as illustrated in (6) and (7) as well as in (24) below); however, the Genitive partitive with a lexical nominal head requires overt agreement (as do constituents with Genitive specifiers in general): 15 (24) Kitap-lar-dan iki tane al-dı-m book-PL-ABL two ‘item’ buy-PST-1SG ‘I bought two of the books.’ (lit. ‘I bought two (entities) of the books’) The Ablative partitive does not need (and thus does not exhibit) an agreement morpheme here (in contrast to the earlier examples such as (23b), where agreement morphology is needed in the absence of a lexical nominal head).
15
For a detailed discussion of the correlation between Genitive specifiers and overt agreement, the reader is referred to Kornfilt (2003).
DOM AND TWO TYPES OF DSM IN TURKISH
93
The lexically headed Ablative partitive contrasts with its Genitive counterpart which does require overt agreement: (25) a. *Kitap-lar-ın iki tane(-yi) al-dı-m book-PL-GEN two ‘item’(-ACC) buy-PST-1SG Intended reading: ‘I bought two of the books.’ Even the overt Accusative cannot save this example, as we see; overt agreement is necessary (due to the presence of the Genitive) — and once again, once we have overt agreement, we also need overt structural Case, which is the Accusative in this instance: (25) b. Kitap-lar-ın iki tane-sin-i al-dı-m book-PL-GEN two ‘item’-3SG-ACC buy-PST-1SG ‘I bought (any) two of the books.’ As expected by what has been said so far about structural Case markers being neutralized with respect to acting as expressions of specificity when their presence is due to syntactic and/or morphological requirements, the overt Accusative marker in the Genitive partitive construction does not act as a straightforward specificity marker: the partitive here is ambiguous between a specific and non-specific reading. (While the specific reading is stronger, obviously reinforced by the specificity function of the Accusative elsewhere, the non-specific reading can be reinforced by enriching the context; e.g. the example can be continued with something like the following: ‘...but because I lost them immediately, I don’t know which ones they were.’) The unreliability of overt structural Case markers as specificity (i.e. semanticsrelated differential) markers is also shown by the pair (26) and (27) for the Accusative, and by the pair (28) and (29) for the Genitive: (26) Bu vasıf-lar-a sahip ol-an bir insan/ person/ this property-PL-DAT owner be-REL a bir kütüphaneci arı-yor-um a librarian seek-PROG-1SG ‘I am looking for a person/a librarian (non-specific) who has these properties.’ Clearly, the italicised expression is non-specific; the speaker is looking for any person or any librarian who fulfills certain properties; it is obvious that the speaker does not have any particular person or librarian in mind; such an expression could also be used as a job description. Not surprisingly, the non-specific direct object bears no overt Accusative marker. The Case marking properties of the following example, very similar to (26) in its meaning, are surprising:
94
JAKLIN KORNFILT (27) Bu vasıf-lar-a sahip ol-an bir-in-i arı-yor-um this property-PL-DAT owner be-REL one-AGR(3)-ACC seek-PROG-1SG ‘I am looking for someone (specific or non-specific) who has these properties (i.e. someone with these properties).’
This example means essentially the same as the previous example, with the difference that it is ambiguous between a specific and a non-specific reading for the direct object. Yet, it exhibits the Accusative marker — even under its non-specific reading. This is possible, because the presence of the Accusative marker is due to the presence of the agreement marker, whose presence, in turn, is due to the absence of a lexical head of the NP/DP. Similar pairs can be constructed easily for the Genitive instead of the Accusative: (28) [Bu vasıf-lar-a sahip ol-an bir insan/ person/ this property-PL-DAT owner be-REL a bir kütüphaneci ara-n-dığ-ın]-ı duy-du-m a librarian seek-PASS-FN-3SG-ACC hear-PST-1SG ‘I heard that a person/a librarian (non-specific) who has these properties was being looked for.’ (29) [Bu vasıf-lar-a sahip ol-an bir-in-in this property-PL-DAT owner be-REL one-AGR(3)-GEN ara-n-dığ-ın]-ı duy-du-m seek-PASS-FN-3SG-ACC hear-PST-1SG ‘I heard that someone (specific or non-specific) who has these properties was being looked for.’ In this pair, too, the DP/NP (in these instances, a subject) with a lexical nominal head doesn’t exhibit a structural Case marker (here, a Genitive), when it is nonspecific (cf. (28)). Therefore, for such subjects, the Genitive is a semantically based DSM marker (again, for my purposes, the term ‘DSM’ is used here descriptively, to refer to semantically-based alternations between the presence versus absence of the overt Genitive morpheme); here, then, the Genitive is indeed correlated with specificity. However, where such a lexical nominal head is missing, the agreement marker acts as such a head, and because the agreement requires overt Case, the Genitive marker’s presence is formally enforced (cf. (29)). In this latter instance, the Genitive does not block a non-specific reading. We have seen in this section that neutralization of DOM as well as DSM markers (in the same descriptive sense just mentioned) with respect to the semantic feature of specificity takes place due to similar reasons: the formal requirement for the presence of overt agreement (whether due to the need for a nominal head in an NP/DP phrase, or whether due to the presence of a Genitive specifier within a phrase) in turn necessitates an overt structural Case marker. This formal requirement leads to neutralization of such a marker with respect to expression of the relevant semantic feature. We thus see that not only with respect to the expression of the semantic feature per se are the DOM and DSM phenomena similar (rather than mirror-image of each other), but the neutralization of this semantic feature in the
DOM AND TWO TYPES OF DSM IN TURKISH
95
presence of DOM and DSM markers proceeds similarly, as well (rather than in a mirror-image fashion). I now turn to a different type of DSM, namely a type where the alternation between presence versus absence of overt Genitive marking expresses clausal properties (rather than the semantic features of NPs). 4. SECOND TYPE OF DSM: BASED ON PROPERTIES OF DIFFERENT TYPES OF CLAUSES Embedded clauses in Turkish typically are not tensed, and are traditionally said to be nominalized in varying degrees. For our purposes, the most interesting property of such clauses is the fact that the subject is marked with the Genitive, in contrast to the Nominative subjects of fully verbal clauses. We have seen both types of examples earlier in this paper, and we shall see more of them later on. Two points need to be made here: 1. The contrast is not one between root and embedded clauses per se; instead, the contrast is between nominal versus verbal features of the functional projections of these clauses; 2. The nominalized clauses are not lexically derived nominals, but rather regular clauses with full lexical as well as functional architecture.16 In this paper, it is essentially the first point which will be of interest to us. 4.1 Nominalized gerundive clauses as arguments, and their subjects Due to space constraints, I shall limit myself to a very brief description of the main facts, referring the reader to Kornfilt (2003) and (2006) for detailed discussions of the relevant facts and for an account. Turkish has two main types of nominalized clauses; those are mainly found as arguments, but they can function as adjuncts, as well. These two types have been labeled as factive or indicative for one type, and non-factive or subjunctive for the other. They are illustrated below: Factive (indicative) nominalized clause: (30) a. [Sen-in dün sabah ev-de yemek pişir-diğ-in]-i cook-FN-2SG-ACC you-GEN yesterday morning home-LOC food duy-du-m /san-dı-m hear-PST-1SG /believe-PST-1SG ‘I heard/believed that you had been/were cooking/cooked/had cooked food at home yesterday morning.’
16
One of the clearest indicators of this claim is the fact that when a transitive verb is the predicate of these clausal nominalizations, the (specific) direct object is in the Accusative. In other words, the verb retains its property as a Case-checker (presumably, in conjunction with little v). This contrasts with lexically derived verbal nominals, where the complement of the verb gets externalized and thus shows up in the Genitive; cf. Kornfilt and Greenberg (2000).
96
JAKLIN KORNFILT Non-factive (subjunctive) nominalized clause: (30) b. [Sen-in yarın ev-de yemek pişir-me-n]-i cook-NFN-2SG-ACC you-GEN tomorrow home-LOC food isti-yor-um want-PROG-1SG ‘I want for you to cook food at home tomorrow.’
It is the first, factive or indicative type of nominalized clause which is of interest to us, for reasons that will become clear presently. Note, for the time being, that this type can also be marked for future tense and can thus be independent from the matrix clause with respect to its tense: (30) c. [Sen-in ev-de yemek pişir-eceğ-in]-i cook-FUTFN-2SG-ACC you-GEN home-LOC food duy-du-m /san-dı-m hear-PST-1SG /believe-PST-1SG ‘I heard/believed that you will cook food at home.’ No such tense distinction can be made on the non-factive or subjunctive type; that type has no means to distinguish tense and thus takes on the tense of the matrix clause: (30) d. [Sen-in ev-de yemek pişir-me-n]-i cook-NFN-2SG-ACC you-GEN home-LOC food isti-yor-um /iste-di-m /isti-yeceğ-im want-PRPROG-1SG /want-PST-1SG /want-FUT-1SG ‘I want/wanted/will want for you to cook food at home.’ These distinctions, as well as some additional contrasts17 discussed in Kornfilt (2003) and (2006) lead me to propose that while nominalized subjunctive clauses are fully nominal, nominalized indicative clauses have the vestiges of verbal features and thus of verbal functional projections. Since the agreement markers on the predicates are the same on both the subjunctive and the indicative nominalizations, and come from a fully nominal agreement paradigm (in the sense that these same nominal agreement markers are also found on the head of clearly nominal possessive phrases), I proposed that the nominalized indicative clauses are categorially hybrid, exhibiting verbal functional projections in the lower architecture of the clause, while exhibiting nominal agreement in the higher architecture. I further proposed that the Genitive subject which is found in both types of nominalized clauses is licensed by the nominal Agr marker (which expresses the 17
Some of the relevant contrasts are as follows: Indicative nominalized clauses can host WH-operators (both in embedded WH-questions and in relative clauses), while subjunctive nominalized clauses cannot; indicative nominalized clauses cannot pluralize, nor can they admit a determiner, while subjunctive nominalized clauses admit both for many speakers, but in varying degrees of wellformedness. Thus, indicative nominalized clauses have some properties of TP/CPs, while subjunctive nominalized clauses do not, instead exhibiting properties of DP/NPs.
DOM AND TWO TYPES OF DSM IN TURKISH
97
subject’s phi-features as well as the clause’s categorial features) on the predicate. We also saw earlier in the present paper that this Genitive marking is sensitive to the semantic feature of specificity, thus acting as a semantically-based DSM marker with respect to subjects. When we turn to similar gerundive clauses that are adjuncts, however, the facts are different. 4.2 Nominalized gerundive clauses as adjunct clauses, and their subjects The gerundive, i.e. nominalized indicative clauses can show up as adjuncts which are usually accompanied by postpositions; however, these clauses can also occur without a postposition: (31) [[ Sen yemek pişir-diğ-in] için] ben konser-e cook-FN-2SG because I concert-DAT you[SG] (NOM) food gid-ebil-di-m go-ABIL-PST-1SG ‘Because you cooked, I was able to go to the concert.’ (32) [[ Sen yemek pişir-diğ-in]-e göre] hepiniz cook-FN-2SG-DAT according to all+you you[SG] (NOM) food ev-de kal-acak-sınız home-LOC stay-FUT-2PL ‘Given that you cooked, all of you will stay at home.’ (33) [[Ben yemek pişir-diğ-im]-den dolayı] konser-e cook-FN-2SG-ABL because concert-DAT I (NOM) food gid-e-me-di-m go-NegABIL-NEG-PST-1SG ‘Because I cooked, I was unable to go to the concert.’ Note that in these factive nominalizations, the subject is ‘bare’, i.e. in the default (or, morphologically, Nominative) Case rather than in the Genitive, although a nominal Agr element is present. Similar facts hold in the absence of postpositions, too: (34) [[Ben (*-im) yemek pişir-diğ-im]-den konser-e cook-FN-1SG-ABL concert-DAT I(NOM) (-GEN) food gid-e-me-di-m go-ABIL-NEG-PST-1SG ‘Because I cooked, I was unable to go to the concert.’ (35) [Sen (*-in) konser-e git-tiğ-in]-de ben ev-e home-DAT you[SG](NOM) (-GEN) concert-DAT go-FN-2SG-LOC I dön-üyor-du-m return-PROG-PST-1SG ‘When you were going to the concert (lit. at your going to the concert), I returned home.’
98
JAKLIN KORNFILT
Note that in this set of examples, the subjects are first or second person pronouns. They are thus definite as well as specific; yet, they are not, and can’t be, marked with the Genitive. The same is true for third person specific subjects in such clauses: (36) [Oya (*-nın) konser-e git-tiğ-i(n)]-de ben ev-e home-DAT Oya(nom) (-GEN) concert-DAT go-FN-3SG-LOC I dön-üyor-du-m return-PROG-PST-1SG ‘When Oya was going to the concert (lit. at Oya’s going to the concert), I returned home.’ In contrast, nominalized subjunctive clauses do not exhibit this argument/adjunct asymmetry; in other words, the subjects of such clauses are in the Genitive, irrespective of whether the clause is an argument or an adjunct; the following representative example should suffice to illustrate this point: (37) [[Sen-in yemek pişir-me-n] için] ben ev-de kal-dı-m cook-NFN-2SG for I house-LOC go-PST-1SG you-GEN food ‘I stayed at home so that you should cook (lit. for you to cook).’ I will now open a big parenthesis and offer a brief account of subject Case, sketching a proposal to explain these contrasts. After this sketch, I shall return to our main concern, namely the different ways in which DSM and DOM can be neutralized, and the ways in which these neutralizations interact with each other. 4.3 A brief account of subject Case in Turkish The account in a capsule: A. Agr, in order to license subject Case, must itself be licensed. B. Where subject Case is not licensed, default (=Nominative in Turkish and in many other, but not all, languages) Case applies as a last resort operation. (See Kornfilt 2003 and Schütze 2001 for discussion of default Case.) In Kornfilt (2003) and (2006), it is proposed that such licensing of Agr can happen in a number of ways: 1. The Agr element has the same categorial features as the functional heads it dominates and of the head which it (co-)occupies. This is so for verbal Agr (which is only found in fully verbal clauses and which licenses Nominative subject Case) and for nominal Agr (which licenses Genitive subject Case) when it is found in fully nominal clauses (such as the nominalized subjunctive clauses mentioned briefly in the previous section and discussed in detail in Kornfilt 2003; see also footnote 17) as well as in simple nominal DPs, i.e. in possessive phrases:
DOM AND TWO TYPES OF DSM IN TURKISH
99
(38) Hasan [Ali-nin kitab-ın]-ı oku-du Hasan Ali-GEN book-3SG-ACC read-PST ‘Hasan read Ali’s book.’ (39) Hasan kitab-ı [[ Ali-nin kız-ı] için] al-dı Hasan book-ACC Ali-GEN daughter-3SG for buy-PST ‘Hasan bought the book for Ali’s daughter.’ Note that in these clearly nominal domains, the specifier (i.e. possessor) of the possessive phrase bears the Genitive, irrespective of the argument (cf. (38)) or adjunct (cf. (39)) status of the entire phrase. From this point of view, these fully nominal domains are similar to nominalized subjunctive clauses, whose relevant properties in this respect we saw in the previous section (cf. example (37)). These new facts offer additional support for analyzing nominalized subjunctive clauses differently from nominalized indicative clauses, and for viewing them similar to possessive phrases, i.e. for analyzing them as consistently nominal. In such fully and homogeneously nominal domains, the (nominal) Agr is domain-internally licensed with respect to its category, and consequently licenses the nominal (=Genitive) subject/specifier Case; it does not have to be licensed domain-externally, and therefore the status of the domain as an argument or adjunct is irrelevant to the subject Case’s licensing. (Similarly, a verbal Agr would be licensed domaininternally in a fully verbal domain, thus licensing verbal subject Case (=Nominative).) 2. In contrast with such fully and homogeneously nominal domains, the gerundive (nominalized) indicative clauses are categorially hybrid, as mentioned earlier; their Agr is nominal, but the ‘nominalized’ Tense/Aspect marker is verbal, as already sketched. Further, following Kornfilt (2006), I propose that such gerundives are dominated by an nP-shell (somewhat in parallel to the vP-shell), with the nominal Agr raising into the n-head (similarly to the partitives we saw earlier in the paper, where Agr occupies the position of a missing lexical N-head). Finally, I propose that the n-head achieves a category shift, turning the categorially hybrid gerundive clause into a fully nominal one. The nominal Agr is now licensed categorially (after raising into the n-head) and, in turn, licenses the nominal subject Case, i.e. the Genitive. (I assume that the subject DP raises into the Spec position of the nP-shell in order to have its Genitive Case checked by the nominal Agr element now occupying the head-n position.) How does this explain the sensitivity of such gerundives to being an argument versus an adjunct domain? I claim that nP cannot dominate (categorially hybrid) adjunct clauses, while it does dominate similarly hybrid argument clauses. In this, I follow a central part of Rubin’s (2002) and (2003) proposal that adjunct clauses are ModPs, and that ModPs cannot be CPs or AgrPs (i.e. ModPs on the one hand and CP/AgrPs on the other are in complementary distribution). If we assume, as I have proposed here, that nPs are
100
JAKLIN KORNFILT
the nominal counterpart of CP/AgrPs, then it follows that ModPs and nPs are likewise in complementary distribution. It follows from this that in a categorially hybrid ModP, the nominal Agr has no n-head to raise to. It is thus not licensed as a subject Case licenser. Nominal subject Case (=Genitive) cannot show up — default Case applies instead. My claim that nP cannot act as a shell for categorially hybrid adjunct clauses, while it does act as a shell to corresponding argument clauses might appear to be adhoc. It will be difficult to disperse this suspicion without going into a large amount of detail. While being unable to do so, due to space considerations, I will nonetheless try to sketch the motivation for my proposal. This sketch will include some additional detail of Rubin’s proposals just mentioned above. 4.4 Motivation for ModP versus nP I follow Rubin (2002) and (2003) in assuming that modifier phrases and modifier clauses are surrounded by a ModP shell. Rubin also claims, as just mentioned, that the functional projections ModP and CP/AgrP are in complementary distribution. Rubin has two main types of reasons for advancing this new functional projection of ModP, and for his claim about its complementary distribution with CP/AgrP, i.e. the functional projections surrounding argument clauses: one is theory-internal, and the other is empirical. The theory-internal type of motivation has to do with Chomsky’s approach to Merge within the Minimalist Program; e.g. in Chomsky (2001), it is proposed that complements are merged under ‘regular’ set-Merge, while adjuncts are merged under ‘pair-Merge’. How will we recognize, asks Rubin, which portions are to be merged in what fashion, without running into problems with strict cyclicity? Before the actual Merge, we don’t know yet what ‘bit’ of structure will end up as an adjunct, and which bits as an argument. It is only after Merge that we can recognize such relationships. The proposed new projection, and its complementary distribution with CP/AgrP resolves this problem: ModPs are merged under pair-Merge, and CP/AgrPs under set-Merge. Either type of functional projection is immediately recognizable as the type of constituent which has to undergo the appropriate type of Merge. Turning to empirically-based motivations, one type is based on the observation that many languages have dedicated morphemes marking modifiers — depending on whether these are phrasal or clausal. Turkish, for example, has markers that can be analyzed in this fashion (although Turkish is not one of the languages surveyed by Rubin in this context); e.g. the suffix -kI18 on locative and temporal modifiers of nominal heads, as well as a variety of nominalizing suffixes (e.g. –(y)IncA, – (y)ArAk) found on the predicates of clausal adjuncts; the latter are suffixes, most of 18
Following general Turkological practice, I use capital letters to represent underspecified bundles of phonological features. In the case of vowels, Vowel Harmony determines the ultimate values of the segment; in the case of consonants, local rules such as voicing assimilation fill in the missing values. Segments in parentheses are deleted in certain phonological environments. -kI harmonizes only with a preceding ü in adverbial expressions.
DOM AND TWO TYPES OF DSM IN TURKISH
101
which (but not all) are found only on adjunct (rather than argument) clauses (and are often referred to in traditional literature as suffixes deriving converbs): (40) a. raf-ta-ki kitap shelf-LOC-KI book ‘the book on the shelf’ b. dün-kü haber yesterday-KI news ‘the news of yesterday.’ (41) a. [Ali ev-den çık-ınca] Oya konser-e git -ti Ali house -ABL exit-’when’ Oya concert-DAT go-PST ‘When Ali left the house,19 Oya went to the concert.’ b. [Ali ev-den çık-arken] Oya konser-e gid-iyor-du Ali house -ABL exit-’while’ Oya concert-DAT go-PROG-PST ‘While Ali left/was leaving the house, Oya was going to the concert.’ (42) [[Ali ev-den çık-tık-tan] sonra] Oya konser-e git -ti Oya concert-DAT go-PST Ali house-ABL exit-FN-LOC after ‘After Ali left the house, Oya went to the concert.’ The suffix -kI, as well as -(y)IncA, -(y)ArAk and some others are found attached to modifier domains only. Hence, it would make sense, as Rubin does for a variety of corresponding markers in other languages, to analyze them as occupying the head Mod-position of a functional category ModP. The suffix -DIK is found elsewhere, as well, as we have seen: it is the indicative nominalizer. However, its properties here are different; similarly to -(y)IncA and its kin, it is not followed by any Agr morphology in (42). I propose to analyze it as a Mod-morpheme in such instances, in parallel to my analysis of the morphemes in (41). Note also that we do not find here the strict correlation, otherwise widely attested in Turkish, between overt subjects and Agr (cf. Kornfilt 1984, 2003) — a correlation which I have interpreted to mean that Agr is responsible for subject Case, thus licensing overt subjects. Examples (41a) and (41b), as well as (42) are problematic for this generalization and its interpretation. I therefore have proposed the assignment of default Case to the subject in such instances, as an operation of last resort (cf. Kornfilt 2003 and 2006, as well as Schütze 2001). An additional piece of evidence that this is the right direction to go is that the otherwise complementary distribution between overt subjects and PRO breaks down in these adjunct clauses:
19
In this English translation and the other translation in this set of examples, we find tense expressions that are not present in the Turkish examples. In English, translations using the corresponding adjunct gerundives would have been closer to the Turkish examples, but they were not used here, because they would not allow for overt subjects — and we are centrally interested in those here.
102
JAKLIN KORNFILT
çık-ınca] Oya ayağ-ın-ı burk-tu (43) a. [PRO ev-den house -ABL exit-’when’ Oya foot-3SG-ACC sprain-PST ‘When she left the house [when leaving the house], Oya sprained her ankle.’ çık-arken] Oya ayağ-ın-ı burk-tu b. [PRO ev-den house -ABL exit-’while’ Oya foot-3SG.ACC sprain-PST ‘While she left/was leaving the house [while leaving the house], Oya sprained her ankle.’ çık-tık-tan] sonra] Oya konser-e git -ti (44) [[PRO ev-den Oya concert-DAT go-PST house-ABL exit-FN-LOC after ‘After she left the house [after leaving the house], Oya went to the concert.’ Since PRO rejects Case20 (or else, according to Chomsky and Lasnik 1991, has a special Case licensed in certain contexts such as infinitives, while in the examples in (41) and (42), the Case of the overt subject in the embedded clause appears to be Nominative), I conclude that the subject position in these adjunct clauses has no licensed Case.21 The overt NP/DP subjects that can show up in the same position as PRO therefore have no licensed Case, either; since they do need Case, this is provided by default Case, as a last resort. Note that the syntactic properties of adjunct clauses that consist of our ‘regular’ nominalized indicative clauses with Agr are similar to these of adjunct clauses without Agr and with a dedicated Mod affix, because their subject shows up in the default Nominative Case (rather than in the licensed Genitive Case), as we have seen. The only difference, other than the particular subject Case, is in fact the Agr. Therefore, I propose that Agr is not a Case licenser here, because it is in a different syntactic position than its counterpart in argument clauses. In argument clauses, nominal Agr is situated, as I claimed earlier, in the small n-head of the nP-shell. In adjunct clauses, the nominal Agr does not have its own phrase-structural position; rather, it 20
This tenet of the original version of the Government and Binding model has been challenged in recent literature; however, at least descriptively, this tenet does seem to hold for many languages, and certainly in Turkish (cf. Kornfilt 2003); I disregard languages such as Icelandic where PRO does systematically occupy positions with Case. 21 We might ask here whether the empty category in subject position is indeed PRO in these examples. The only other logical possibility would be pro — a not unlikely possibility, given that the empty subjects in these adjunct clauses are not in complementary distribution with overt subjects, in contrast with PRO subjects in infinitival argument clauses, where such complementary distribution is obligatory. At first glance, therefore, analyzing the empty subject in Turkish adjunct clauses as pro might seem to be plausible; furthermore, pro is not in complementary distribution with overt subjects in a Null Subject language; Turkish is indeed an NSL. On the other hand, there are serious arguments against such a pro-based analysis: There is an extensive body of evidence, showing that not only must pro be licensed by Agr, its phi-features must be identified by full Agr locally; even impoverished Agr is insufficient as an identifier and licenser (cf. Kornfilt 1984, 1996 b.). The complete absence of Agr would make pro in these adjunct clauses impossible. An additional fact that argues against pro, and in favor of PRO, is that pro, as a genuine pronoun, can also refer to a DP in the discourse, outside of the sentence. This is not possible in these examples; here, Oya is the only possible referent, as is expected if the subject of the adjunct clause is PRO, controlled by Oya. (Some speakers report that such discourse-antecedents might be possible under certain scenarios, but even then, this would be a very weak reading. I would interpret such intuitions as showing that for such speakers, the PRO is pragmatically controlled by a discourse antecedent.)
DOM AND TWO TYPES OF DSM IN TURKISH
103
is, together with -DIK (which, as we saw in this subsection, can also be a ‘converb’, i.e. a Mod morpheme) located in the head-Mod position. Therefore, nominal Agr is not licensed to act as a (nominal) subject Case licenser (given that it is not in nposition): in these adjunct clauses, the nominal Agr is placed, as it were, in an adverbial position (by virtue of being in Mod) — and adverbs don’t license Case. We therefore need to take recourse to default Case — this being realized as Nominative in Turkish (as it is in many other languages); we saw this in Agr-less adjunct clauses, as well.22 These remarks conclude the sketch of subject Case, and the alternations between Genitive and Nominative Case in embedded nominalized clauses in Turkish. Before returning to the issue of neutralization of subject Case alternations, I will briefly discuss an alternative account of the ‘clausal DSM’ facts. 4.5 An alternative account of the Turkish subject Case alternations in nominalized indicative argument versus adjunct clauses My treatment of Turkish nominalized clauses and of subject Case, as well as of the correlations between subject Case and Agr, originated in Kornfilt (1984), where I proposed the functional projection of AgrP for clauses and for certain phrases with overt Agr; the syntactic position AGR could host nominal versus verbal features and could license corresponding nominal or verbal subject Case. Abney (1987), when developing his proposal for DPs, translated my nominal AgrP into DPs when discussing Turkish facts (cf. his discussion of Turkish AgrPs in Abney 1987:50-53, 196). Kennelly (1990) adopted Abney’s approach for Turkish. She assumes in her paper that indicative nominalized clauses are DPs, not CPs. She further assumes that subject Case is licensed not by Agr, but by INFL — which would be realized as the Tense/Aspect marker in verbal clauses. In nominalized indicative clauses, she assumes that it is the nominalizer -DIK that also licenses subject Case — in fact the 22
Overt subjects can alternate with empty subjects in these adjunct clauses, too. However, the empty category in these clauses that do exhibit overt Agr is pro, not PRO; we can tell this by the referential properties of the empty subject: (i) Oya [[Pro ev-e erken dön-düğ-ü] için ] dinlen-ebil-di rest-ABIL-PST Oya house-DAT early return-FN-3SG because ‘Because s/he returned home early, Oya was able to rest While under the likeliest reading, pro does refer to Oya, it can also easily refer to a different entity mentioned in the discourse; thus, for example, if her husband had been mentioned in the previous discourse, (i) might well be interpreted as ‘because he, i.e. her husband, returned home early, Oya was able to rest’. This additional reading is possible, because pro is a pronoun rather than a controlled (or even arbitrary) PRO. (Thus, the referential properties of this example with overt agreement is different from the type of example discussed in the previous footnote, where such discourse-based reference is impossible even with ingenious scenarios, or at best very unlikely.) Furthermore, and importantly, the referent of the subject in examples such as (i) would change by changing local Agr markers; thus, the referent can be a first or second person singular or plural subject, depending on the shape of the local Agr morpheme. This is an interesting situation, because if my analysis is correct, this would provide a rare instance where Agr licenses and identifies pro, but does not license subject Case. PRO, then, can’t be the subject of such clauses, because the empty category must be pro — following Jaeggli (1984), according to whom an empty category must be pro, if it can be pro (cf. also the application of Jaeggli’s principle to empty categories in Turkish RCs in Kornfilt 2000).
104
JAKLIN KORNFILT
Nominative. However, when a clause with -DIK is an argument clause, Kennelly assumes that the Case-assigning ability of -DIK is blocked (I shall not go into the mechanism of this blockage). Therefore, motivated by the Case Filter, the subject raises from Spec, IP into Spec, DP. The D, due to its nominal nature, licenses the Genitive. In adjunct clauses, the blockage does not take place, hence -DIK can license Nominative subject Case. However, a number of problems arise. The nature of the blockage is keyed to Theta-government (following Williams 1994, or rather a pre-publication version of that work). Kennelly proposes that a ‘verby’ element (in her terminology) that can license Nominative is blocked from performing this function when it is thetagoverned by another ‘verby’ element. However, in Turkish, there are constructions with fully verbal argument clauses: (45) Ben [Ali kitab-ı oku-du] I Ali book-ACC read-PST ‘I believe Ali read the book.’
san-ıyor-um believe-PRPROG-1SG
The ‘verby’ PAST tense inflection obviously still licenses Nominative subject Case, even though it would be theta-governed by the equally ‘verby’ root predicate, under Kennelly’s analysis. It might be claimed here that the ‘verby’ embedded inflection is in fact blocked from assigning Nominative (due to as yet undetermined reasons), and that the subject raises into the specifier position of a higher functional projection, say of a verbal AgrP, receiving Nominative there, due to the verbal nature of the functional head. But then, a similar blockage should take place, yet it doesn’t. Note that if some ad-hoc way should be devised so as to lift the blockage from the theta-governed verbal AGR, the account would become indistinguishable from mine, where in examples such as (45), the subject of the embedded clause would have its Nominative Case licensed by the verbal features of AGR. Yet another problem with Kennelly’s account concerns nominalized indicative adjunct clauses — and this problem was touched upon in the previous subsection. If it is the -DIK inflection which licenses Nominative subject Case, and if it can exercise this ability because it is not theta-governed by virtue of being contained in an adjunct clause, then we should expect PRO not to be able to act as the subject of such clauses (if those clauses have no Agr — cf. the earlier discussion about prolicensing). However, we saw that adjunct clauses with -DIK, as well as Agr-less ‘converb’ clauses with a variety of predicate inflections, allow both Nominative overt subjects and PRO. 4.6 Back to neutralization of differential Case marking Note now that my account as developed so far does not make any reference to the semantic feature of specificity. The clausal syntax either dictates or prohibits licensed subject Case. Where subject Case is dictated (because it is licensed by the clausal syntax, as just discussed), it can act as a (semantic) differential subject
DOM AND TWO TYPES OF DSM IN TURKISH
105
marker (again, in the merely descriptive sense I have been using this term throughout this paper) as we saw earlier, i.e. the (syntactically) licensed Genitive shows up on a specific subject DP, and it does not show up on a non-specific subject DP — but this is so only in categorially hybrid argument clauses, or both in argument and adjunct domains, if they are categorially homogeneous as well as nominal, such as on the subjects of nominalized subjunctive clauses, or on the specifiers in possessive phrases. At the same time, we can say that the Genitive acts as a clausal differential subject marker in these hybrid argument clauses (with the proviso that the subject DP is specific), signaling the argument versus adjunct status of the entire clause. (In nominalized subjunctive clauses, which are homogeneously nominal, the Genitive does not have this ability). As a consequence, the (by now) familiar function of the Genitive as an expression for the semantic feature of specificity is neutralized (in hybrid adjunct clauses), when the syntax does not allow for the Genitive to show up at all. In these clauses, clearly specific, definite subjects cannot bear the Genitive. To put it differently, the Genitive as a DSM marker for the specificity feature comes to the fore only when the syntax allows for this; here, this would be for all the types of nominalized subjunctive clauses, as well as for argumental nominalized indicative clauses. Thus, we see that at least for Turkish DSM, syntactic DSM has primacy over semantic DSM. This is similar to what we saw previously for both DSM and DOM, but in the converse direction: there, even non-specific subjects and objects were marked with Genitive and Accusative, respectively, when morpho-syntactic reasons necessitated the presence of these structural Case markers, thus neutralizing the effects of both (semantically based) DSM and DOM. I now propose the following (sketchy) account for all the facts we have seen in this paper: The PF module represents relevant semantic features by acting as an interpretive component of narrow syntax. Specificity features in Turkish (and perhaps, more generally, any semantic features relevant for differential Case marking in other languages) are encoded in the D-head of a DP phrase and are read off the K-head of a KP phrase by the PF-component. That K-head takes a DP (or NumP) as its complement, which, in turn, takes an NP as its complement:
106
JAKLIN KORNFILT
(46)
KP K’ DP
K D’
NP
D N’ N
When the NP/DP is specific, the N-head (and, in the absence of a lexical noun in this position, the Agr of an AgrP in the specifier position of the NP, as in some of our partitive examples) raises to D, and then to K. If the entire KP is licensed for structural (i.e. subject or object) Case syntactically, the contents of K will be ‘read’ by the PF component as in need of overt Case; the syntactically licensed Case will be morpho-phonologically expressed. When the NP is non-specific, there are two options: 1. there is no DP-layer, in which instance there is also no KP, and thus the N wouldn’t have any functional heads to raise to — PF has no K to interpret; no structural Case is realized; this would be the situation for generic NPs without any determiner; 2. there is a determiner (which, in Turkish, would be in the specifier, rather than head, position of the DP), and the D hosts non-specific semantic features in its Dhead; the N raises into D, is marked non-specific (by the features residing there), and therefore cannot raise to K. Since the K remains empty (or, in an alternative analysis, is not even realized when the D bears a non-specific feature), PF cannot read off a syntactically licensed structural Case, either. This would be the situation of nonspecific NP/DPs with a determiner.23 Incidentally, for NP/DPs with oblique/lexical Case, I assume that the corresponding morphological case markers are lexically realized on the N, and are 23
These two instances of non-specific NP/DPs lacking structural Case need to be distinguished from each other, because in the first instance, no pronoun can follow in the sentence or discourse that can refer to the bare NP, while in the second instance, a co-referential pronoun can follow (cf. ErguvanlıTaylan 1984). This observation is easy to capture under the proposed analysis, if pronouns can refer to DPs, but not to NPs.
DOM AND TWO TYPES OF DSM IN TURKISH
107
not a representation due to PF-interpretation of the syntactic structure. Thus, they can appear on an N which heads a specific or a non-specific NP; in the first instance, there will be a D and a K for the N + lexical case marker combination to raise to; in the second instance, there either would be no D to raise to, or a non-specific D to raise to. In either instance, the lexical case marker is attached to the N. This analysis accounts easily for the fact that lexical Case is not sensitive to the specificity of the NP/DP that it appears on. Thus, what we have said about differential usage of Case marking applies to structural Case only (i.e. to Genitive, Accusative, and, abstractly, to the Nominative). This would mean that bare Case, i.e. for the NP/DP being bare of morphological (as distinct from phonological) Case, is something different from Nominative Case, which is either a licensed Case or the default Case (even though as a licensed or default Case, its morphological expression is null, i.e. it is a null morpheme, as opposed to being non-existent, as it is in instances of bare Case), and it is the syntactic, ‘verbal’ equivalent of the Genitive Case, which is ‘nominal’; both Nominative and Genitive are licensed subject Cases, and, as just mentioned again, Nominative can be a default Case, as well.24 This is consistent with the general approach taken here, namely that of saying that both DOM and DSM are identificational in Turkish, and not distinguishing25 — the latter between objects and subjects.26 More generally, this approach enables us to treat the primacy of formal (syntactic) factors determining differential Case markers over semantic ones, at the same time providing support for PF as a separate, interpretive component of grammar. PF interprets certain heads in certain positions, along with their syntactically licensed (structural) Cases, and provides the appropriate marking. Where the Case is not licensed (as, e.g., with subjects of categorially hybrid adjunct clauses), or where the head is not in an appropriate (e.g. K-head) position, the PF will not provide the appropriate marking. Because the relevant semantic features are represented in distinct syntactic positions, and because structural Case is licensed syntactically, we have PF interpreting syntactic structures and syntactic features (i.e. structural Case), and not semantic features (directly). Thus, where structural Case is 24
One of the referees asks whether making a distinction between bare Case and the Nominative as a default Case is not problematic. It is not, because default Case applies to full KPs, which are in a syntactic environment where no licensed Case can be checked. In contrast, bare Case is found with smaller nominal phrases, as explained in the text: NPs, and/or DPs whose D has non-specific features, and which either lack a KP projection or else whose N-head cannot raise as high as KP. Typically, these NPs are in a syntactic environment where syntactic Case is checked potentially, but where, due to their non-specific features, PF cannot comply with its interpretive job. 25 I am using identificational and distinguishing in the sense used by de Hoop and Narasimhan (2005), where identificational Case marking appears on ‘strong’ subjects and objects alike, with the relevant features having positive values for the ‘strong’ arguments. The distinguishing function of Case marking is in the service of distinguishing objects from subjects and vice-versa, i.e. the function that I have been arguing against, at least for Turkish and languages similar to it in this respect. 26 When both object and subject are non-specific in a nominalized clause, either the subject or the object must bear overt Case, i.e. Genitive and Accusative, respectively. (More often, it will be the Genitive on the subject which will be realized as the only overt Case in such instances.) This can be seen as the result of the distinguishing function of either DOM or DSM — and if this is correct, it would be the only domain in Turkish Case marking where this function plays a role.
108
JAKLIN KORNFILT
not licensed syntactically, PF has nothing to interpret, even if the relevant semantic features are present, given that PF has no direct access to the semantic component. Likewise, where structural Case is licensed syntactically, and where a relevant head position such as K is filled (by N or Agr), PF will provide overt Case marking, even where the relevant semantic feature is missing. Thus, we explain the different types of neutralizations of ‘differential’ Case markings we have seen, while also providing support for PF as a component that interprets syntax and has no direct access to semantics.27, 28 4.7 A related approach to DSM (and DOM) A related, while distinct, way of looking at DSM is discussed by de Hoop and Narasimhan (2005). The authors look at instances of case alternations in Hindi, where facts similar to Turkish obtain — similar at least in that subject and object case alternations appear to be driven by similar, rather than mirror-image, factors. Thus, de Hoop and Narasimhan propose a criterion of argument strength that applies to subjects and objects in similar ways, albeit with different motivations. ‘Strong’ objects are agent-like and undergo case-marking to distinguish them from subjects. ‘Strong’ subjects, i.e. agents, are case-marked under the identifying function of case, i.e. identifying them as agents, rather than under the distinguishing function of Case, since under its distinguishing function, Case would be expected to show up on ‘weak’, patient-like subjects, to distinguish them from objects, thus yielding a mirror-image effect to DOM. The identifying function is ranked higher in de Hoop and Narasimhan’s OT approach than another constraint which requires that syntactically ‘unmarked’ arguments, i.e. Nominatives, remain morphologically unmarked. These two constraints on subject Case determine the sensitivity of subject Case to different kinds of predicates (e.g. different lexical verbs, distinct aspects). Can this approach work for Turkish?
27
For details of a related account, see Woolford (this volume). However, our views on the nature of PFbased subject (and object) Case marking differ, inasmuch as for Woolford, PF-based DSM (and presumably DOM) has no syntactic consequences. For me, subject and object Case marking and their alternations can have such consequences, as PF is ‘fed’ by syntax and interprets the syntactic structures that give rise to the observable alternations. If these syntactic structures have distinct syntactic behavior, then we would have a situation where distinct PF-representations also have distinct syntactic properties. Indeed, in Turkish, we have seen that ‘bare’ DPs, i.e. DPs which are not overtly Case-marked, do not scramble as easily as DPs that do bear (structural) Case marking. While the approach proposed here does not mean that, strictly speaking, PF has syntactic consequences (as it does not feed narrow syntax), it does mean that distinctions at PF can have syntactic correlations. 28 One of the reviewers suggests that one might treat all structural Case alternations in Turkish as part of syntax — not only those we saw in section 4, under ‘clausal DSM’, but also those in section 2. The alternations apparently due to semantic factors such as specificity would be, under this view, epiphenomenal, arising from a combination of various factors and conditions. At this point, such a direction is too vague for me to pursue. I would hesitate to take the facts of NP/DPs with non-lexical heads as basic, and those with lexical heads as epiphenomenal. However, future research might convince me otherwise, and I thank this referee for suggesting this as a possible alternative.
DOM AND TWO TYPES OF DSM IN TURKISH
109
The constraint (‘Primary Argument Immunity Principle [PAIP]’) that the ‘unmarked’ primary argument should not be overtly marked (thus yielding the result that Nominative is null) would indeed work for root clauses and embedded verbal clauses. It would also work for adjunct nominalized indicative (i.e. categorially hybrid) clauses. (We might also say that this works for generic subjects of homogenious nominalized clauses and for subjects of argumental hybrid clauses, if we are willing to analyze lack of Genitive as Nominative, rather than as lack of subject Case.) However, this approach, even enriched with the notion of argument strength, would not work for the entire array of Turkish subjects, given the productivity of nominalizations with their Genitive (rather than Nominative) subjects, and given that Genitive subjects are not more agentive than their Nominative counterparts in verbal sentences (or in adjunct, categorially hybrid clauses), as illustrated by the pair below. We have to once again go back to one of the proposals made earlier, i.e. that DSM expressing clausal features (in other words, in this instance, Genitive subjects expressing the nominalization of the clause) overrides other considerations: (41) a. Ali kitab-ı oku-du Ali[-NOM] book-ACC read-PST ‘Ali read the book.’ b. [Ali-nin kitab-ı oku-duğ-un]-u duy-du-m Ali-GEN book-ACC read-FN-3SG-ACC hear-PST-1SG ‘I heard that Ali read the book.’ Here, the Genitive subject of the embedded sentence in (41b) is not more prominent in any way than the Nominative subject of the root clause in (41a). The PAIP is thus overridden, not by argument strength (and thus by the identifying function of case marking), but by the syntactic requirement of marking the subject of a nominal clause (with a Case-licensing nominal agreement). That marking will have to be that of the nominal subject Case, i.e. the Genitive. These examples and their proper analysis also have repercussions for DOM. Note that both in (41a) and (41b), the object is Accusative. In other words, even when the subject is (morphologically) marked, the object must be marked, as well (as long as it bears the semantic feature relevant for the particular language — here, specificity, in whatever way defined), although it doesn’t have to be distinguished from the subject morphologically. In other words, if such an object is not marked with the Accusative and is therefore bare, it can still be distinguished from the subject, due to the subject’s Genitive marking. Of course, it is clear why the object must be marked with the Accusative: the object must be distinguished from other objects which are non-specific. Thus, it appears that at least in Turkish, DOM effects are, at least primarily, identificational (just as semantically based DSM), rather than distinguishing. This conclusion, too, is one we had arrived at previously, on independent grounds.
110
JAKLIN KORNFILT 5. CONCLUSIONS
None of these considerations seriously challenge the proposals mentioned here, made for Hindi and other languages. However, the discussion in the last section suggests that DSM and DOM (in the descriptive sense used in this paper) can have different functions and motivations in different languages, along (very few) parameters.29 At least for Turkish, then, both DSM and DOM are identificational with respect to semantic features (of argument prominence or potential topicality), and both can be overridden by syntactic requirements. As we saw, this can result in Case marking despite lack of the relevant semantic feature. We also saw that, with respect to DSM, syntactic (clausal) feature requirements can also lead to the opposite situation, i.e. one where subject Case is not marked, even though the subject does bear the relevant semantic feature. A more general issue that arises is one about the status of functionally based analyses. The proposal about the distinguishing function of morphological Case marking is obviously driven directly by functionalistic considerations. But what about Case-marking with identificational function? How different is this from a formal approach to morphological case marking which would claim that, in certain languages, morphological Case (possibly limited to structural Case, as it appears to be in Turkish) is licensed when not only syntactic Case is licensed, but also when the licensee bears certain semantic features (e.g. [animate], [specific] etc.)? At least in Turkish, such an approach appears to be sufficient, as I hope to have shown here. The typological challenge, then, would be to characterize languages whose differential case markings are more ‘functionalistic’ than their counterparts in other languages — a worthwhile enterprise. REFERENCES Aissen, J. (2003). Differential object marking: iconicity vs. economy. Natural Language and Linguistic Theory 21, 435-483. Chomsky, N. (2001). Derivation by phase. Ken Hale: A Life in Language. Ed. by M. Kenstowicz. Cambridge (MA): MIT Press, 1-52. Chomsky, N. and H. Lasnik (1991). Principles and parameters theory. Syntax: an International Handbook of Contemporary Research. Ed. by J. Jacobs, A. von Stechow, W. Sternefeld and T. Vennemann. Berlin: de Gruyter, 506-569. Comrie, B. (1975). Definite and animate direct objects: a natural class. Linguistica Silesiona 3, Katowice, 13-21. Enç, M. (1991). The semantics of specificity. Linguistic Inquiry 22, 1-25. Erguvanlı-Taylan, E. (1984). The Function of Word Order in Turkish Grammar. [University of California publications in Linguistics, vol. 106.] Berkeley and Los Angeles: University of California Press. van Heusinger, K. and J. Kornfilt (2005). The case of the direct object in Turkish: Semantics, syntax and morphology. Turkish Languages 9, 3-44. de Hoop, H. and B. Narasimhan (this volume). Ergative case-marking in Hindi. de Hoop, H. and B. Narasimhan. (2005). Differential case-marking in Hindi. Competition and Variation in Natural Languages: The Case for Case. Ed. by M. Amberber and H. de Hoop. Oxford: Elsevier. 29
Woolford (this volume) also proposes that DSM is not one single, unified phenomenon, but rather consists of a class of phenomena. She arrives at this proposal in the course of detailed cross-linguistic observations.
DOM AND TWO TYPES OF DSM IN TURKISH
111
Jaeggli, O. (1984). Subject extraction and the null subject parameter. NELS 14, 132-153. Jelinek, E. and A. Carnie (2003). Argument hierarchies and the mapping principle. Formal Approaches to Function in Grammar. Ed. by A. Carnie, H. Harley, M. A. Willie. Amsterdam/Philadelphia: John Benjamins, 265-296. Johanson, L. (1977). Bestimmtheit und Mitteilungsperspektive im türkischen Satz. ZDMG (= Zeitschrift der Morgenländischen Gesellschaft) Suppl. 3/2, 1186-1203. Kennelly, S. (1990). Theta government in Turkish. Paper presented at the GLOW workshop, SOAS, London. Kennelly, S. (1997). Non-specific external arguments in Turkish. Dilbilim Araştırmaları 7, Istanbul, 58-75. Kornfilt, J. (1984). Case Marking, Agreement, and Empty Categories in Turkish. Doctoral dissertation, Harvard University. Kornfilt, J. (1996a). Naked partitive phrases in Turkish. Partitives. Ed. by J. Hoeksema. Berlin: Mouton de Gruyter, 111-125. Kornfilt, J. (1996b). Turkish and configurationality. Current Issues in Turkish Linguistics 1. Ed. by B. Rona. Ankara: Hitit, 111-125. (Copyright London: SOAS, 1996; related conference held in 1990.) Kornfilt, J. (1997). Turkish. London: Routledge. Kornfilt, J. (2000). Some syntactic and morphological properties of relative clauses in Turkish. The Syntax of Relative Clauses. Ed. by A. Alexiadou, P. Law, A. Meinunger and C. Wilder. Amsterdam/Philadelphia: John Benjamins, 121-159. Kornfilt, J. (2001). Non-Specific Partitives and the Unreliability of Specificity Markings. LSA annual meeting. Washington (DC) and ms., Syracuse University. Kornfilt, J. (2003). Subject case in Turkish nominalized clauses. Syntactic Structures and Morphological Information. Ed. by U. Junghanns and L. Szucsich. Berlin: Mouton de Gruyter, 129-215. Kornfilt, J. (2006). Agreement: The (unique and local) syntactic and morphological licenser of subject case. Studies on Agreement. Ed. by J. Costa and M.C. Figueiredo Silva. Amsterdam/Philadelphia: John Benjamins, 141-171. Kornfilt, J. and G. Greenberg (2000). Changing argument structure without voice morphology: a concrete view. Studies on Turkish and Turkic Languages: Proceedings of the Ninth International Conference on Turkish Linguistics, 1998. Ed. by A. Göksel and C. Kerslake. Wiesbaden: Harrassowitz. Lee, H. (this volume). Quantitative variation in Korean case ellipsis: implications for case theory. Lewis, G. L. (1967). Turkish Grammar. Oxford: Oxford University Press. Morimoto, Y. (this volume). From topic to subject marking: implications for a typology of subject marking. Rubin, E. (2002) The Structure of Modifiers. University of Utah manuscript, to be published by MIT Press, in the Linguistic Inquiry Monograph series. Rubin, E. (2003). Determining Pair-Merge. Linguistic Inquiry 34, 4, 660-668. Schütze, C.T. (2001) On the Nature of Default Case. Syntax 4, 3, 205-238. Silverstein, M. (1976). Hierarchy of features and ergativity. Grammatical Categories in Australian Languages. Ed. by R.M.W. Dixon. Canberra: Australian Institute of Aboriginal Studies, 112-171. Williams, E. (1994). Thematic Structure in Syntax. Cambridge (MA): MIT Press. Woolford, E. (this volume). Differential subject marking at argument structure, syntax, and PF.
JOANNA BŁASZCZAK
CHAPTER 6 Differential Subject Marking in Polish: The Case of Genitive vs. Nominative Subjects in “X was not at Y”-constructions
1. THE ISSUE It is by now a well-known fact that the interpretation of a nominal phrase (henceforth NP1) in a sentence might be determined or influenced by various factors. Often a specific interpretation of a given NP correlates with a specific case marking. For example, object NPs which are interpreted as specific might be marked in a special way, cf., e.g., the -(y)i- or ko-marking for specific objects in Turkish or Hindi, respectively (Enç 1991, Kornfilt, this volume, de Hoop and Narasimhan, this volume). In a similar vein, animate (or definite) objects might require another case marking than inanimate (or indefinite) ones, e.g., the a-marking for animate objects in Spanish (cf. von Heusinger and Kaiser 2003). From a cross-linguistic point of view, it is often the case that a particular case marking not only tells us how a given NP is going to be interpreted, but also might bear on the aspectual interpretation of a verbal phrase (VP); in other words, a special case marking might have both NP- and VP-related functions. A case in point is the Partitive case in Finnish (cf. Kiparsky 1998; see also Vainikka and Maling 1996). An intrinsically telic (bounded) verb such as saan ‘get’ takes a Partitive marked object when the object is quantitatively indeterminate (an indefinite bare plural or mass noun), otherwise the object is marked for ACC(usative). In other words, the Partitive tells us something about the specific interpretation of the NP in such cases. In contrast, with an aspectually unmarked verb such as ampua ‘shoot’ (i.e., a verb that is neither intrinsically telic/bounded nor intrinsically atelic/unbounded), the alternation between the ACC and Partitive marking of the object gives rise to an aspectual distinction (an accomplishment reading ‘to shoot dead’ with an ACC object and an activity reading ‘to shoot at’ with a Partitive object) (cf. Kiparsky 1998:267). This is the VP-related (aspectual) function of the Partitive. In this paper I will discuss another case of dependency between the aspectual properties of the predicate and the case marking of its nominal argument. The case to be discussed below is interesting and rather unusual since, unlike the cases of Partitive marking mentioned above, which are not restricted to just one particular 1
I will use throughout the chapter the general term ‘nominal phrase’ (NP) to refer to any nominal phrase to avoid the discussion about the NP- vs. DP-distinction, which is especially problematic in languages such as Polish which do not have the category article.
113 H. de Hoop and P. de Swart (eds.), Differential Subject Marking, 113–149. © 2008 Springer.
114
JOANNA BŁASZCZAK
verb, but, on the contrary, show quite a broad (but by no means unrestricted) distribution, the Polish case concerns exclusively negated (existential-)locative constructions. Depending on the aspectual properties of the verb być ‘to be’, the ‘subject-NP’2 is marked either for NOM(inative) or GEN(itive). Whereas the nominal argument of a negated (existential-)locative sentence is normally marked for GEN in Polish, on the habitual (or iterative) reading it is marked for NOM, as the contrast between (1a) and (1b) shows.3 In the affirmative variants the nominal is marked for NOM in both cases; cf. (2a) and (2b), respectively (see section 4.1 for more details). (1) a. Jana nie było na przyjęciu John-GEN NEG BE-3SG.N.PST at party ‘John was not at the party.’ (Lit.: ‘There was no John at the party.’) b. Jan nie bywał na przyjęciach John-NOM NEG BE-3SG.M.PST.HABIT at parties ‘John didn’t use to come to parties.’ (Lit.: ‘John was not (from time to time) at parties.’) (2) a. Jan był na przyjęciu John-NOM BE-3SG.M.PST at party ‘John was at the party.’ b. Jan bywał na przyjęciach John-NOM BE-3SG.M.PST.HABIT at parties ‘John used to come to parties.’ (Lit.: ‘John was from time to time at parties.’) That the aspectual marking of the predicate in (1) seems to matter for the case marking of its nominal argument is somewhat surprising given the Polish facts. Even though – as we will see below in section 1 – Polish has both GEN marked ‘subjects’ and the phenomenon called Genitive of Negation, which is a GEN marking of the internal argument triggered by negation, neither of them seem to be affected by the aspectual properties of the predicate. On the other hand, Polish also has Partitive objects which in fact show some affinity for perfective aspect. However, as we will see below, such Partitive NPs have a special interpretation which seems to be lacking in the case of the GEN marked NP in (1a).
2
The term “subject” is used here in a purely descriptive, pre-theoretical sense. It will be made more precise in the course of the argumentation; see especially section 4.2 and section 6. 3 It should be noted that whereas in cases like (1a) (i.e., with definite, animate subjects) a NOM marking is also possible (resulting in a different interpretation), the GEN marking is totally excluded in (1b); cf. (i-a) and (i-b), respectively. I will return to these issues in more detail in section 4.1.3. (i) a. Jan nie był na przyjęciu John-NOM NEG BE-3SG.M.PST at party ‘John was not at the party.’ b. *Jana nie bywało na przyjęciach John-GEN NEG BE-3SG.N.PST.HABIT at parties
DIFFERENTIAL SUBJECT MARKING IN POLISH
115
While examples like (1) above might be somewhat unexpected from a strictly Polish point of view, there is in fact nothing strange about them once a broader crosslinguistic perspective is taken into account. That the subject might be differently marked depending on the aspectual properties of the predicate is found in so-called split-ergative languages like, e.g., Hindi. In this language the ‘special’ case marking (here: the ERG(ative)) is triggered by perfective aspect as opposed to the unmarked (NOM/ABS(olutive)) marking in imperfective aspect; cf. (3) (Mahajan 1994:318, 323). (3) a. raam vah kitaabē parhtaa thaa Ram-NOM.M those books-PL.F read-IPFV.SG.M be-PST.SG.M ‘Ram used to read those books.’ b. raam-ne vah kitaabē parĩĩ thĩĩ Ram-ERG.M those books-PL.F read-PFV.PL.F be-PST.PL.F ‘Ram had read those books.’ The main goal of the chapter is to investigate in more detail the conditions responsible for the GEN vs. NOM split observed in (1). The main questions to be answered are the following: Q1: What factors determine the case marking of the nominal arguments in (1)? Q2: Why does the nominal argument in (1a) occur in GEN and why is such a marking not possible in cases like (1b)? Q3: Where does the GEN marking come from? Q4: Do the sentences in (1a) and (1b) have the same or rather different syntactic structures? Q5: Is there any deeper connection between the Polish facts in (1) and the facts found in split-ergative languages (as in (3))? The chapter is organized as follows. I start in section 2 with the question of the origin and nature of the GEN marking in (1a). The result of this investigation is that the GEN marking in (1a) must be regarded as resulting from the specific structural properties of the construction in question and cannot be reduced to some general mechanism of GEN or Partitive marking of nominal arguments in Polish. Since, as already pointed out above, the GEN/NOM split in (1) at first sight seems to be triggered by different aspectual properties of the respective predicates, in section 3, I will look in more detail at the aspectual differences between być in (1a) and bywać in (1b). Then, in section 4 I will investigate the question of whether there are some other factors responsible for the GEN/NOM split in (1) apart from aspect. It will be shown that the GEN vs. NOM marking of the nominal argument in negated (existential-)locative clauses as in (1) in fact correlates with further differences concerning the interpretation and structural properties of the nominal argument. In section 5, I will try to answer the question of why there should be a correlation between aspect and syntactic structure. It will be shown that there is in fact a deeper connection between the split-ergative structures like the one in (3) and the Polish sentences in (1). Based on these insights, I will propose in section 6 what
116
JOANNA BŁASZCZAK
the underlying syntactic structures for (1a) and (1b) should be like. Finally section 7 will conclude the chapter. The main result of the chapter is that in Polish negated (existential-)locative constructions, bywać ‘to beHABIT’ with its NOM NP is an unergative verb, whereas być ‘to be’ can be both an unergative (when it has an agentive NOM NP) and an unaccusative (when it has a nonagentive GEN NP). Thus, unergative and unaccusative frames depend on two crucial factors equally (i.e., both of these factors must be present): (i) the aspectual properties of the predicate, and (ii) the agentive interpretation of the nominal argument. Crucially, for być ‘to be’ to be an unaccusative, it must be a perfective and its NP must have a nonagentive interpretation. It should be stressed that the claim about the link between the perfective aspect and unaccusativity is made only for BE in Polish. No claim is made in this chapter about the unaccusative/unergative status of (im)perfective intransitive verbs in general. This is a matter for future research. Given the observation that być ‘to be’ can be an unaccusative or unergative verb, depending on the interpretation of the NP, this essentially amounts to the claim that Polish has at least two verbs for BE (three if one includes bywać ‘to beHABIT’).4 In this sense the present chapter can also be seen as a contribution to the discussion of a rather controversial issue, namely the question of how many different BEs should be distinguished in general and in Slavic languages in particular.5 2. ON THE ORIGIN AND NATURE OF THE GEN MARKING IN NEGATED (EXISTENTIAL-)LOCATIVE CLAUSES Given that subjects are normally marked for NOM in Polish, the question that we need to answer first is where the GEN marking for the nominal argument in negated (existential-)locative sentences, as in (1a), comes from. Potentially, the GEN in such cases could be (i) an instance of Genitive of Negation, (ii) an instance of a GEN marked subject or (iii) an instance of an optional Genitive Partitive in Polish. I will discuss these alternatives in turn below. As far as option (i) is concerned, this seems in fact the most plausible one. Why? The nominal argument is marked for GEN in negated, but not in affirmative (existential-)locative sentences (recall (2a)), hence the GEN marking obviously has something to do with the presence of negation. Given this observation, it would be plausible to assume that the GEN marking in (1a) is an instance of the case alternation triggered by negation, the well-known ‘Genitive of Negation’ (GoN). But 4
The present chapter does not discuss the canonical copular constructions with BE and an AP or NP predicate (as, e.g., ‘John is nice’ or ‘John is a student’) in Polish. This is also a question for future research whether the BE in such constructions is a different BE from the one used in the (existential-) locative constructions discussed in this chapter. But see Błaszczak (2004) and Błaszczak (2005) for some suggestions. 5 The opinions vary here from the view that there is no lexical verb BE (BE is rather just the spell-out of various functional heads in the syntax) (cf. Freeze 1992, Kayne 1993, den Dikken 1995, Kondrashova 1996) to the opinion that there are actually two or more different verbs BE (cf., e.g., Chvany 1975, Franks 1995, Geist 1999, Pereltsvaig 2001); see Harves (2002:169ff.) for a detailed discussion.
DIFFERENTIAL SUBJECT MARKING IN POLISH
117
note that in making such an assumption we would immediately encounter a problem. GoN in Polish is restricted to just one configuration, namely to the direct object position of transitive verbs; cf. (4). More importantly, GoN does not apply in Polish to internal arguments of unaccusative verbs; cf. (5). Note that even the default, nonagreeing form of the verb does not improve the acceptability of the GEN marked NP; cf. (5b). (4) a. Ewa czyta gazety /*gazet Ewa reads newspapers-ACC /*newspapers-GEN ‘Ewa reads/is reading newspapers.’ b. Ewa nie czyta gazet /*gazety Ewa NEG reads newspapers-GEN /*newspapers-ACC ‘Ewa does not read/ is not reading newspapers.’ (5) a. (W tym szpitalu) [Ŝaden pacjent] /* [Ŝadnego pacjenta] (in this hospital) [no patient]-NOM.M.SG /* [no patient]-GEN.M.SG nie umarł NEG die-3SG.M.PST ‘(in this hospital) no patient died.’ b. *(W tym szpitalu) [Ŝadnego pacjenta] nie umarło (in this hospital) [no patient]-GEN.M.SG NEG die-3SG.N.PST Now, given that the GEN marked NP in (1a) is most likely the internal argument of the unaccusative verb BE, and, more importantly, given the fact that internal arguments of unaccusative verbs (‘unaccusative subjects’) cannot be marked for GEN under negation in Polish, this would mean that the GEN-marking in (1a) cannot be taken to be an instance of the regular GoN in Polish (but see the discussion in section 6).6 This leads us directly to option (ii). In Polish there are verbs (verbs of addition, loss, lack, sufficiency) that mark their internal arguments with GEN. Such a GEN NP is traditionally called ‘logical subject’ (Polish: podmiot logiczny), as opposed to grammatical subject (see Przepiórkowski 1999:129-30 and the references cited there; cf. also Witkoś 1998:252f.). However, in such cases – unlike in the case of (existential-)locative być – the GEN marking is not affected by negation; compare (6a) to (6b). (6) a. Ubyło wody w rzece decrease-3SG.N.PST water-GEN in river ‘There was less water in the river.’ b. Nie ubyło wody w rzece NEG decrease-3SG.N.PST water-GEN in river ‘There was not less water in the river.’ In addition, unlike what we observe in cases like (1) above, the GEN marking of the nominal argument in (6) does not seem to be affected by the aspectual properties 6
The GEN marking in (1a) also differs from the GoN of the Russian sort; see Błaszczak (2004) for discussion.
118
JOANNA BŁASZCZAK
of the predicate; cf. (7). Note that the same is true for GoN (cf. (8)): independently of the aspectual properties of the predicate, and, more importantly, even of the habitual reading in (8c), the direct object of a negated transitive verb is marked for GEN in Polish.7 (7)
(8)
Wody ubyło / ubywało w rzece water-GEN decrease-3SG.N.PST.PFV / decrease3SG.N.PST.IPFV in river ‘There was less water in the river.’ / ‘There was less and less water in the river.’ a. Nie czytałam tej gazety NEG read-1SG.F.PST.IPFV [this newspaper]-GEN ‘I didn’t read this newspaper.’ b. Nie przeczytałam tej gazety NEG read-1SG.F.PST.PFV [this newspaper]-GEN ‘I didn’t read (completely) this newspaper.’ c. W młodości nie czytywałam gazet NEG read-1SG.F.PST.HABIT newspapers-GEN in youth ‘In my youth I didn’t use to read newspapers.’
Thus, given the independence of the GEN marking in cases like (6) or (8) both from negation and aspect, also option (ii) as an explanation of the origin of the GEN in (1a) does not seem very convincing. What about the last option? Unlike the cases of obligatory GEN marking, which are independent of the aspectual properties of the predicate, as in (7) and (8), there are cases in Polish where the internal argument of a verb can optionally be marked for Partitive (GEN) instead of ACC. In such cases, unlike in the case of (1), the Partitive marking in fact seems to be affected by the aspectual properties of the predicate.8 In other words, the perfective aspect seems to facilitate the Partitive (GEN) marking; cf. (9) (Rozwadowska and Willim 2004:133; see also Brown and Franks 1995:250). However, unlike the GEN marked NP in (3a), the Partitive argument in cases like (9a) might not be accepted by everyone and for those who accept it, it has a special interpretation (i.e., the Partitive argument is interpreted quantitatively in the sense of ‘part of’ or ‘a little’, cf., e.g., Saloni and Świdziński 1985:145-6, Swan 2002:333, or is ‘quantitatively indeterminate’, cf. Rozwadowska and Willim 2004:141). Such a partitive/quantificational interpretation seems to be absent in the case of (1a).
7
Aspect in Polish (like in other Slavic languages) is a morphological category of the verb, i.e., most verbs are morphologically marked either for being perfective or imperfective. It should be noted, however, that the aspectual marking in Polish (Slavic) is a very complex issue. As Franks (1995:282, fn. 16) points out, “there is no unique marker of perfective or imperfective, nor are the various morphological indicators of Russian [Polish] aspect uniquely prefixal or suffixal, and some verbs are biaspectual, while others may lack either one or the other aspectual form.” See Borik (2002) and Młynarczyk (2004) for a more recent discussion of these issues in Russian and Polish, respectively. 8 It should be pointed out that Polish, unlike, e.g., Finnish, does not have distinct Partitive morphology. In other words, what some scholars refer to as a Partitive (GEN) in Polish is morphologically nondistinct from the “standard” GEN case in Polish.
DIFFERENTIAL SUBJECT MARKING IN POLISH
119
(9) a. On kupił chleb / chleba. he buy-3SG.M.PST.PFV bread-ACC / bread-GEN ‘He (has) bought a/the loaf of bread/the bread. / ‘He (has) bought (some) bread.’ b. On kupował chleb /* chleba. he buy-3SG.M.PST.IPFV bread-ACC /* bread-GEN ‘He was buying bread/a/the loaf of bread/the bread.’ To sum up the discussion so far, the GEN in negated (existential-)locative być sentences seems to have what might be called ‘a hermaphroditic nature’. In other words, the GEN in such sentences appears to be a cross between the regular GoN in Polish and the optional Partitive marking of the internal argument. 9 Like the former it is triggered by negation (i.e., it is restricted to negated sentences) and like the latter it is affected by the aspectual properties of the predicate. However, unlike the regular GoN in Polish, the GEN marking in negated (existential-)locative być sentences does not seem to obey the configurational condition on GoN assignment in Polish. Recall that the rule of GoN in Polish only applies to internal direct arguments of transitive verbs. Thus, it seems that for an internal argument to be GEN marked under negation in Polish, there must be another (external) argument present in the structure. The similarity between the optional Partitive marking, as in (9), and GEN marking in negated (existential-)locative być sentences, as in (1a), is also far from being perfect. It has been pointed out that the Partitive marked arguments have a special quantificational interpretation which seems to be missing in the case of the GEN marked argument in negated (existential-)locative być sentences. What is, however, common to both cases is the fact that aspect seems to matter for the question of the case marking of the nominal argument. Given that the optional Partitive marking, as in (9), seems to be dependent on the perfective aspect, the question arises as to whether the GEN marking in negated (existential-)locative być sentences could also be viewed as dependent on some perfective properties of the predicate. In order to answer this question, let us look more closely at what aspectual value the verb BE has in Polish (Slavic). 3. THE GEN VS. NOM SPLIT AND ASPECTUAL DIFFERENCES BETWEEN BYĆ AND BYWAĆ Usually, native speakers of a Slavic language do not have problems with establishing the aspectual value of a given verbal form. Given this, it is very surprising that determining the aspectual value of the verb BE in Slavic turns out to be nontrivial. In fact, there is no agreement in the literature on this point: while some authors, as, e.g., Schoorlemmer (1995), take byt’ ‘to be’ in Russian to be clearly 9
See also Witkoś (1998:274-5, fn. 38) for some discussion on the nature of GEN in negated (existential-)locative sentences in Polish. Unlike in the analysis proposed in this chapter, Witkoś does not make a connection between the GEN marking of the “subject” NP and the aspectual properties of the predicate in the cases under discussion.
120
JOANNA BŁASZCZAK
imperfective, other researchers, as, e.g., Franks (1995), Junghanns (1997) or Eriksen (2000), take just the opposite view and claim byt’ to be perfective. There is also a middle position, which takes byt’ to be either aspectually unspecified or compatible with both the perfective and imperfective viewpoints; cf., among others, Matushansky (2001) and Borik (2002); see also Chvany (1975). Why is it so difficult to establish the aspectual value of BE in Russian or Polish? The reason for this could be the fact that this verb seems to have both imperfective and perfective properties. On the one hand, like imperfective verbs, it is compatible with durative adverbials like for an hour (cf. (10)), and can appear in the complement position of phase verbs like begin, cease or continue that otherwise only precede imperfective infinitives; cf. (11) (see Borik 2002:153 for Russian data). Moreover, BE patterns with imperfective verbs in yet another respect, namely as far as the participle formation is concerned (see Borik 2002:39ff., Piñón 1994:349ff. for detailed discussion). Polish has two types of adverbial participles: the ‘present participle’ (imiesłów przysłówkowy współczesny) and the ‘perfect participle’ (imiesłów przysłówkowy uprzedni). Of these two participles, imperfective verbs have only present participles (-ąc forms) (cf. (12a)), while perfective verbs have only perfect participles (-(w/ł)szy forms) (cf. (12b)) (Piñón 1994:350f.). Example (13) shows that only a present participle (cf. (13a)), but not a perfect participle (cf. (13b)) can be formed from być, hence być behaves like an imperfective verb (see Borik 2002:153 for the corresponding Russian data). (10) Jan był w domu przez dwie godziny John was-3SG.M at home for two hours ‘John was at home for two hours.’ (11) a. Jan przestał być nauczycielem John-NOM stopped BE-INF teacher-INS ‘John stopped being a teacher.’10 /*przeczytać ksiąŜki b. Jan przestał OKczytać John-NOM stopped OKread-INF.IPFV /*read-INF.PFV book-ACC.PL ‘John stopped reading books.’ *przeczytając-PFV ‘reading’ (12) a. OKczytając-IPFV ‘having read’ b. OKprzeczytawszy-PFV *czytawszy-IPFV (13) a. OKbędąc ‘being’ b. *bywszy (intended: ‘having been’) However, there is also evidence pointing to exactly the opposite conclusion: BE in Polish and Russian is perfective. 10
One should be careful with the application and interpretation of such tests. Notice that, unlike in the case of the predicative być, as in (11a) above, this test does not work with (existential-)locative być; cf. (i). See Błaszczak (2004) and Błaszczak (2005) for more discussion on this point. (i) #Jan przestał być w domu. John stopped BE-INF at home (intended: ‘John stopped being at home.’)
121
DIFFERENTIAL SUBJECT MARKING IN POLISH
Notice first of all that just as there are contexts which are compatible only with imperfective verbs (like, e.g., ‘phase’ verbs (recall (11b)) or the future auxiliary, which is discussed below; cf. (17)), there are environments which are compatible only with perfective verbs, e.g., the complex conjunction zanim nie ‘lit. before not’ (or poka ne in Russian; cf. Eriksen 2000:29, referring to Ferrell 1953); cf. (14a). As example (14b) shows, być is compatible with this context, thus it seems to pattern with perfective verbs in this respect. (14) a. Nigdzie nie wyjdę, zanim nie nowhere NEG go-1SG.PRS.PFV before NEG OK napiszę /*piszę [tego listu] OK write-1SG.PRS.PFV /*write1SG.PRS.IPFV [this letter]-GEN ‘I won’t go out before I have written this letter.’ b. Nie zadzwonię do nikogo zanim nie będę NEG phone-1SG.PRS.PFV to noone before NEG be-1SG.PRS w domu at home ‘I won’t call anyone before I get home.’ There is also other (in fact more solid) evidence that BE (in Polish and Russian) is perfective. It has been noted in the literature (see, e.g., Eriksen 2000:3) that the agreement suffixes employed by what are traditionally called future tense forms of byt’ (‘to be’) in Russian (i.e., the budet-forms) are actually ordinary present tense suffixes of Russian verbs. The same holds for the Polish ‘future forms’ of być, the będzie-forms; cf. (15). (15)
a. Russian
b. Polish
1.SG
bud-u
piš-u (write-PRS)
będ-ę
pisz-ę (write-PRS)
2.SG
bud-eš’
piš-eš’
będzi-esz
pisz-esz
3.SG
bud-et
piš-et
będzi-e
pisz-e
1.PL
bud-em
piš -em
będzi-emy
pisz-emy
2.PL
bud-ete
piš-e te
będzi-ecie
pisz-ecie
3.PL
bud-ut
piš-ut
będ-ą
pisz-ą
Nevertheless, the budet-forms, despite being, morphologically speaking, present tense forms, have exclusively future time reference; cf. (16a). This is characteristic of perfective verbs in Slavic; cf. (16b).
JOANNA BŁASZCZAK
122
(16) a. Ona budet vračom (Russian) she-NOM be-3SG.PRS doctor-INS ‘She will be a doctor.’ b. Ona pišet /napišet pis’mo she-NOM write-3SG.PRS.IPFV /write-3SG.PRS.PFV letter-ACC ‘She is writing/will write a letter.’ In fact, there is diachronic evidence that the budet-forms used to be perfective. In Old Church Slavonic the verb byti used to have two separate present-tense paradigms: an imperfective paradigm (→ est’-forms) and a perfective one (→ budetforms) (see Junghanns 1997, van Schooneveld 1951; see also Borik 2002:151f.); cf. Table 1. Table 1. Present-tense paradigms of Old Church Slavonic byti
imperfective
perfective
singular
plural
dual
singular
plural
dual
1
st
jesmь
jesmъ
jesvě
bǫdǫ
bǫdemъ
bǫdevě
2
nd
jesi
jeste
jesta
bǫdeši
bǫdete
bǫdeta
3
rd
jestъ
sǫtъ
jeste
bǫdetъ
bǫdǫtъ
bǫdete
Moreover, as pointed out by Franks (1995:233), this view of BE as formally perfective also explains the structure of the periphrastic future in Russian (Polish); cf. (17): ‘The future meaning is derived from the perfective character of the auxiliary, while the verb itself remains imperfective in aspect’ (ibid.). (17) a. Ona budet pisat’/*napisat’ pis’mo she-NOM be-3SG.PRS write-3SG.PRS.IPFV/*PFV letter-ACC ‘She will write a letter.’ *budet + infinitive-PFV b. OKbudet + infinitive-IPFV
(Russian)
Another piece of evidence for the perfective status of BE, mentioned by Franks (1995:233), is the impossibility of the periphrastic future form of BE; cf. (18a). This is due to the fact that byt’ is a perfective infinitive (cf. (18b)) and – as already mentioned above – in the periphrastic future forms the auxiliary is followed exclusively by imperfective verbs. (18) a. *Ona budet byt’ vračom she-NOM be-3SG.PRS BE-INF doctor-INS b. *budet + byt’-PFV ( cf. (17b) *budet + infinitive-PFV)
DIFFERENTIAL SUBJECT MARKING IN POLISH
123
BE is unusual in yet another respect: unlike most other verbs, BE in Polish (also in Russian) has a separate iterative/habitual paradigm; see Table 2.11, 12 Table 2. Aspectual forms of BE ASPECT ITERATIVE
bywać 13
IMPERF?/PERF?
być
będzie
FUTURE
będzie bywać
PRESENT
bywa
jest
PAST
bywał/-a/-o
był/-a/-o
Judging from the diagnostics mentioned above, the iterative bywać is clearly imperfective. Like imperfective verbs, (i) it can appear in the complement position of a phase verb like begin or cease (recall (11)) (cf. (19a)), (ii) it allows for a present participle (the -ąc form) but disallows a perfect participle (the -wszy form) (recall (12)) (cf. (19b)), (iii) the present tense form of bywać, unlike perfective verbs, does not have future time interpretation (recall (16)) (cf. (19c)), and (iv) it can follow the auxiliary in the periphrastic future forms (cf. (19d)) (recall (17)).14 (19) a. Jan przestał bywać na przyjęciach BE-INF.HABIT at parties John-NOM stopped ‘John stopped coming to parties.’ ‘being from time to time’ b. OKbywając *bywawszy (intended: ‘having been from time to time’) c. Jan bywa na przyjęciach John-NOM BE-3SG.M.PRS.HABIT at parties ‘John goes (from time to time) to parties.’ (*‘John will go (from time to time) to parties.’) 11
Normally, it is just an imperfective form that is used to express an iterative meaning of a given verb; cf. (i). (i) Jan często chodził na przyjęcia John often go-3SG.M.PST.IPFV to parties ‘John often went to parties.’ 12 In more general studies of verbal aspect (e.g., Comrie 1976), “the habitual” is analyzed as a subcategory of imperfective aspect. See, however, Brinton (1987) for the suggestion that for English at least, habitual aspect seems more aligned with perfective aspect than with imperfective aspect. See also Smith (1997:33-35) for a general discussion. 13 będzie bywać is a periphrastic future form; see (19d). 14 As far as the first diagnostic is concerned, there seems to be some imcompatibility between the inherently durative semantics of the adverb and the inherent iterativity/habituality of the predicate; cf. (i). (i) ?Jan bywał na przyjęciach przez całe dnie. John-NOM be-3SG.M.PST.HABIT at parties for whole days Lit: ‘John used to be at parties for days.’
124
JOANNA BŁASZCZAK d. Jan będzie bywać na przyjęciach John-NOM be-3SG.PRS BE-INF.HABIT at parties ‘John will go (from time to time) to parties.’
To sum up the discussion in this section, while bywać is clearly an imperfective verb, być seems to have both imperfective and perfective features and therefore it is difficult to establish its aspectual value. This conflict could be resolved, as suggested by Franks (1995:283, fn. 24), “with a proper understanding of the relationship between grammatical aspect and extensional aspect. Roughly speaking, the imperfective extensional properties of byt’ [‘to be’] derive from its intensional stativity, despite the fact that it is grammatically perfective.” Unfortunately, Franks is not very clear on this point, but what this is intended to mean is (most probably) the following: byt’ in Russian (and for that matter also być in Polish) has the semantics of a stative verb. That is, from the point of view of its semantic/lexical aspect (i.e., ‘situational aspect’ or ‘situation type’), it refers to situations which hold for a moment or an interval, i.e., which have the temporal features [Static] and [Durative] (Smith 1997:32). States are types of situations which do not evolve over time (states simply are). Unlike events, which are dynamic situations requiring a constant input of energy to maintain them (since they occur in successive stages which are located at different moments), “states consist of an undifferentiated period, and continue unless something happens to change them” (Smith 1997:36). Or to phrase it in Comrie’s words: dynamic situations but not states have to continually be “subject to a new input of energy” to maintain them (Comrie 1976:49). Moreover, unlike events, which are discrete, bounded entities, states are cumulative and unbounded: “They have a uniform part structure, as shown by the fact that the sub-interval property holds of them,” which means that when a state holds for an interval it holds for every sub-interval of that interval (Smith 1997:36, 32). This stative semantics is naturally compatible with the imperfective viewpoint since the imperfective viewpoint presents part of a situation, with no information about its endpoint.15 In other words, when an imperfective viewpoint is taken, the situation (or eventuality, to use Bach’s 1981 terminology) is understood to hold throughout and potentially to extend beyond the Topic Time (i.e., a time for which the speaker wishes to make an assertion), that is, the situation is presented as unbounded. Now, given that byt’ in Russian and być in Polish are stative verbs and given that the stative semantics is naturally compatible with the imperfective viewpoint, it follows that byt’ and być can naturally occur in ‘imperfective environments’, which gives rise to the impression that they are imperfective verbs. This impression notwithstanding, they are – as far as their grammatical/morphological aspect is concerned – perfective verbs. It is this ‘perfective status’ that determines their morphosyntactic behavior (recall the discussion above about the structure of the periphrastic future in Russian/Polish). Later, in section 6, it will be argued that it is precisely this ‘grammatical perfective
15
Smith (1997:36) states that “the global class of stative sentences includes all sentences with the imperfective viewpoint.”
DIFFERENTIAL SUBJECT MARKING IN POLISH
125
aspect’ which is relevant to the GEN-marking in negated (existential-)locative sentences. Assuming that the above observations are correct, the aspectual difference between być and bywać can be described as follows: bywać has properties of an imperfective verb as far as both its lexical semantic properties (situation type aspect16) and grammatical behavior (grammatical/morphological aspect) are concerned. In contrast, być has semantic properties characteristic of an imperfective verb, but from the grammatical/morphological point of view it behaves as a perfective verb.17 4. FURTHER FACTORS RESPONSIBLE FOR THE GEN/NOM SPLIT In the previous section, the first factor responsible for the GEN/NOM split in (1) has been identified, namely, the aspectual properties of the predicate: the GEN marking of the ‘subject’ appears to be dependent on the perfective aspect of the predicate (in the strictly grammatical/morphological sense), while the NOM marking of the subject correlates with the clearly imperfective aspect (in terms of habituality/iterativity; cf. also footnote 12). The question of why there should be a correlation between the case marking of the nominal argument and the aspectual properties of the predicate is the subject of the discussion in section 5. Before we go to this issue, let us first further examine the properties of the constructions with bywać (1b) and those with być (1a) in order to identify further factors (in addition to the already identified aspectual differences) responsible for the GEN/NOM split. 4.1 The interpretation of the ‘subject’ 4.1.1 Unmarked word orders One obvious difference between the construction with bywać (1b) and that with być (1a) is the fact that their nominal arguments are differently case marked. As will be shown below, this difference in the case marking is by no means just a purely formal distinction but has clear interpretive effects.
16
Recall that bywać is habitual (or iterative) (Comrie 1976:27, fn. 1 remarks that “in Slavonic linguistics, habitual forms are often referred to by the term ‘iterative’ (Russian mnogokratnyj), e.g. Russian pivat’, znavat’, the habitual counterparts of pit’ ‘drink’, znat’ ‘know’”). Habitual sentences are semantically stative. To see this, one should consider their truth conditions. As Smith (1997:34) points out, when “investigating the truth of a habitual, one asks whether there is a pattern which holds over an interval, not whether a particular situation occurred. The temporal schema holds for the interval, as is typical of statives.” 17 If this conclusion is correct, the situation in Polish described above would, in fact, in some sense resemble the situation found with stative verbs in Finnish. Kiparsky (1998:283) points out that verbs like omistaa ‘to own’ in Finnish behave – as far as their morphosyntactic properties (especially their case syntax) are concerned – as telic verbs, even though semantically they remain atelic verbs.
126
JOANNA BŁASZCZAK
The subject of bywać is always marked for NOM regardless of the presence of negation in the clause; cf. (20a/b) vs. (20c). Moreover, the subject tends to be preverbal in both affirmative and negative variants; cf. (20a) and (20b). (20) a. Jan bywał na przyjęciach John-NOM BE-3SG.M.PST.HABIT at parties ‘John used to come to parties.’ (Lit.: ‘John was from time to time at parties.’) NP-NOM V PP b. Jan nie bywał na przyjęciach John-NOM NEG BE-3SG.M.PST.HABIT at parties ‘John didn’t use to come to parties.’ (Lit.: ‘John was not at parties.’) NP-NOM NEG V PP c. *Jana nie bywało na przyjęciach John-GEN NEG be-3SG.N.PST.HABIT at parties Unlike the subject of bywać, the ‘subject’ in (existential-)locative clauses with być has a different case marking depending on the presence of negation. In an affirmative clause it is marked for NOM, and in the negated variant thereof it is marked for GEN; cf. (21a) versus (21b). Moreover, the subject in an affirmative (existential-)locative clause tends to occupy a postverbal position (cf. (21a)), whereas the subject in a negated (existential-)locative clause usually shows more freedom of word order, i.e., it does not necessarily need to occupy a postverbal position (cf. (21b) vs. (21c)); see Borschev and Partee (2001), Harves (2002), among others, for a similar observation regarding the Russian facts. (21) a. Na stole była ksiąŜka on table BE-3SG.F.PST book-NOM.SG.F ‘There was a book on the table.’ PP V NP-NOM b. Na stole nie było ksiąŜki on table NEG BE-3SG.N.PST book-GEN.SG.F ‘There was no book on the table.’ PP NEG V NP-GEN c. KsiąŜki nie było na stole book-GEN.SG.F NEG BE-3SG.N.PST on table Lit. ‘There was not the/a book on the table.’ (‘The book was not on the table.’) NP-GEN NEG V PP 4.1.2 (Non)agentivity Not only does the position of the nominal argument differ but there is also a clear difference in the interpretation of the NOM vs. the GEN NP. The first thing to
DIFFERENTIAL SUBJECT MARKING IN POLISH
127
observe is that the subject of bywać has to be [+animate] (preferably [+human]), both in the affirmative and negative variants; cf. the well-formed cases in (20) versus the unacceptable examples in (22).18 (22) a. *Na stole (nie) bywa banan. on table (NEG) BE-3SG.PRS.HABIT banana-NOM.SG b. *Banan (nie) bywa na stole. banana-NOM.SG (NEG) BE-3SG.PRS.HABIT on table For być sentences the situation is more complicated. In an affirmative clause both [+animate] and [-animate] ‘subjects’ are possible; cf. (23). The same holds of the negated variants with a GEN ‘subject’; cf. (24). (23) a. W ogrodzie jest dziecko BE-3SG.PRS child-NOM.SG in garden Lit.: ‘In the garden there is a child.’(‘There is a child in the garden.’) banan b. Na stole jest on table BE-3SG.PRS banana-NOM.SG Lit.: ‘On the table there is a banana.’ (‘There is a banana on the table.’) (24) a. W ogrodzie nie ma [Ŝadnego dziecka]19 NEG have-3SG.PRS [no child]-GEN.SG in garden Lit.: ‘In the garden there is no child.’ (‘There is no child in the garden.’) b. Na stole nie ma (Ŝadnego) banana on table NEG have-3SG.PRS [(no) banana]-GEN.SG Lit.:‘On the table there is no banana.’(‘There is no banana on the table.’) Apart from the clear preference for animate (or [+human]) subjects in the case of bywać as opposed to the absence of similar preferences in the case of (existential-) locative być, there is a clear difference in the interpretation of NOM vs. GEN nominal arguments. Whereas a NOM subject can be understood as an agent, having control over a given situation, a GEN ‘subject’ can be characterized by the absence of agentivity or volition/controllability (see Borschev and Partee 2001 and Harves 18
19
To express the intended meaning another construction has to be used, as shown in (i). bywać is used here in the meaning of ‘to happen, to occur’ and is subcategorized for a sentential (situational) complement in the form of a Ŝe (that)-clause. Note that the acceptability of examples with bywać with inanimate subjects seems to improve as long as such subjects can be interpreted “situationally”; e.g., in (ii) kłótnie ‘quarrels’ can be used to refer to ‘quarrelling’ situations, i.e., situations in which the neighbors were having quarrels. This issue has to await further research. (i) Bywa, Ŝe na stole jest banan. happens that on table BE-3SG.PRS banana-NOM.SG ‘It happens that there is a banana on the table.’ (ii) ?U sąsiadów bywają kłótnie. at neighbors BE-3PL.PRS.HABIT quarrel-NOM.PL Lit.: ‘There are from time to time quarrels at neighbors’ place.’ (‘It happens that the neighbors have quarrels.’) Note that the negated form of the existential-locative być in the present tense is actually nie ma, lit.: ‘not has’. The importance of this fact will become clear in the discussion in section 6.
128
JOANNA BŁASZCZAK
2002, referring to Padučeva 1992, for a suggestion along similar lines). Evidence for this comes from the observation that only a NOM subject is compatible with agentoriented intentional adverbs (cf. Grimshaw 1990:51, who uses a similar test to detect the presence of a ‘true agent subject’ in event nominals). Compare examples (25a/a’) and (26a) versus (25b). An agent-oriented intentional adverb like chętnie ‘willingly’ is perfectly fine with the NOM marked subject in bywać sentences (both in the affirmative and negated variants); cf. (25a/a’). The same holds for the NOM subject in być sentences (again in both the affirmative and negated variants; see section 4.1.3 for more discussion on this point); cf. (26a/a’). In contrast, the use of an agentoriented intentional adverb with a GEN ‘subject’ leads to ungrammaticality; cf. (26b).20 It is interesting to note that the interpretation of a NOM marked subject in an affirmative (existential-)locative sentence depends on its position in the clause: in postverbal position, the subject is interpreted (under normal circumstances) ‘nonagentively’; in preverbal position, on the other hand, as already pointed out above, the subject might acquire properties of a prototypical subject (agentivity, volition, controllability). Compare the contrast between (26a) and (26b).21 (25) subject(agent)-oriented intentional adverbs a. Jan chętnie bywał na przyjęciach John-NOM willingly was-HABIT at parties Lit.: ‘John was willingly at parties.’ (‘John willingly went from time to time to parties.’) a.’ Jan nie bywał chętnie na przyjęciach John-NOM NEG was-HABIT willingly at parties Lit.: ‘John was not willingly at parties.’ (‘John did not go willingly to parties.’) b. *Jana nie było chętnie w pracy John-GEN NEG was willingly at work (26) a. Jan chętnie był w domu John-NOM willingly was at home ‘John was willingly at home.’ a.’ Jan nie był chętnie w domu John-NOM NEG was willingly at home ‘John wasn’t willingly at home.’ 20
It should be stressed that in negated (existential-)locative sentences, the interpretation of the GEN “subject” does not depend on its position in the clause. In other words, both in preverbal and postverbal position, the interpretation of the subject is the same (in terms of nonagentivity, nonvolitionality, noncontrollability); what, however, presumably changes (in relation to word order) is the interpretation of the subject in terms of definiteness or discourse (information structural) related properties; see Borschev and Partee (2001) for a recent discussion of these issues. 21 In order to make example (26b) grammatical, a contrastive context has to be used; cf. (i). (i) W domu chętnie był Jan, a nie Piotr. at home willingly was John-NOM and not Peter-NOM Lit.: ‘At home willingly John was and not Peter.’ (‘It was John who was willingly at home and not Peter.’)
DIFFERENTIAL SUBJECT MARKING IN POLISH b. ??W domu at home
chętnie willingly
129
był Jan was John-NOM
The contrast detected in this section in the interpretation between GEN and NOM ‘subjects’ in terms of agentivity is further supported by facts discussed below. 4.1.3 The choice between a NOM vs. GEN ‘subject’ in negated locative być clauses In the beginning of the chapter (recall footnote 3), while discussing the puzzling facts about the Polish examples in (1), I pointed out that while the subject of bywać cannot be marked for GEN, the subject of być may sometimes (but not always, see below) appear in NOM; cf. (27) vs. (28): (27) a. Jan nie bywał na przyjęciach John-NOM NEG BE-3SG.M.PST.HABIT at parties ‘John didn’t use to come to parties.’ (Lit.: ‘John was not from time to time at parties.’) b. *Jana nie bywało na przyjęciach John-GEN NEG BE-3SG.N.PST.HABIT at parties (28) a. Jan nie był na przyjęciu John-NOM NEG BE-3SG.M.PST at party ‘John was not at the party.’ b. Jana nie było na przyjęciu John-GEN NEG BE-3SG.N.PST at party ‘John was not at the party.’ (Lit.: ‘There was no John at the party.’) Not every subject may appear in NOM in negated locative być-clauses. On the contrary, the NOM-marked subjects in negated locative być-clauses are subject to the following restriction: they must obligatorily be definite, preferably [+human]; in the case of indefinite, [-human] ‘subjects’ only GEN is possible, as the following contrast in (29) vs. (30) shows (Dziwirek 1994:152; see also Klebanowska 1974). (29) a. Jana nie było w sklepie (Dziwirek 1994:152) John-GEN NEG BE-3SG.N.PST in store ‘John wasn’t in the store.’ (Lit.: ‘There was not John in the store.’) b. Jan nie był w sklepie John-NOM NEG BE-3SG.M.PST in store ‘John wasn’t in the store.’ (30) a. Sera nie było w sklepie (Dziwirek 1994:152) cheese-GEN NEG BE-3SG.N.PST in store ‘There was no cheese in the store.’ b. *Ser nie był w sklepie cheese-NOM NEG BE-3SG.M.PST in store
130
JOANNA BŁASZCZAK
Given that [+definite]/[+human] NPs might in fact appear in GEN or in NOM in negated (existential-)locative być-clauses, the question arises as to what determines the use of the GEN or the NOM case for such nominal arguments. According to Dziwirek (1994:173-4), the choice between nominative and genitive […] is related to volitionality, as shown in [(31a)] and [(31b)]. In [(31b)] the genitive nominal is incompatible with the adverb celowo ‘on purpose’. The nominative implies willfull absence (hence is impossible with non-humans), while genitive does not. (31) a. Celowo nie byłem na przyjęciu u Ewy at Eve on purpose NEG BE-1SG.M.PST at party ‘I wasn’t at Eve’s party on purpose.’ b. *Celowo nie było mnie na przyjęciu u Ewy at Eve on purpose NEG BE-3SG.N.PST me-GEN at party Thus, the NOM subject is quasi-agentive, having control over the situation; the GEN ‘subject’ is nonagentive, implying the absence of control. This is further confirmed by the ‘polite’ (GEN) vs. ‘less polite’ (NOM) use of the ‘subject’ (see Borschev and Partee 2001:37, referring to Chvany 1975:158); cf. (32). Example (32a) has the interpretation that the reason for the speaker’s absence at the concert is beyond his or her control (some external force is responsible for it); this somehow implies that if it was not for this ‘external force’, the speaker might perfectly well come to the concert. (32b), in contrast, implies that the speaker’s absence at the concert is somehow intended by him or her; in other words, he or she could in principle come to the concert, but has decided not to. (32) a. Nie będzie mnie na twoim NEG BE-3SG.PRS me-GEN at your ‘I will not be at your concert.’ b. (Ja) nie będę na twoim I-NOM NEG BE-1SG.PRS at your ‘I will not be at your concert.’
koncercie (nonagentive) concert koncercie (agentive) concert
4.1.4 Further evidence for (non)agentivity The discussion in the previous sections has shown that while GEN ‘subjects’ cannot be interpreted as agents, NOM subject allows for such an interpretation. This claim is additionally supported by two other tests, which are discussed below, namely (i) occurrence as imperatives and (ii) occurrence as ‘pofectives’. The first test is illustrated in (33) below.22 It shows that it is possible, though not very common, to use bywać and być in imperative forms as long as the subject is understood as an agent, i.e., as someone who is able to perform what is required of him or her. Thus, for instance, one cannot use a sentence like Bądź na stole ‘Be on 22
Actually, this test has been cited by Lakoff (1966) as one of the tests for isolating stative verbs. Levin and Rappaport Hovav (1995:170f.) point out, however, that the tests cited by Lakoff, rather than isolating stative verbs, “turn out to isolate agentive from nonagentive verbs.”
DIFFERENTIAL SUBJECT MARKING IN POLISH
131
the table’, if the addressee of the imperative is an inanimate object, e.g., a newspaper (unless, of course, the newspaper is anthropomorphized). Particularly interesting are the examples (33d) and (33e). The regular imperative form in (33d), in which the unexpressed subject would correspond to the NOM form (cf. TyNOM nie bądź ‘lit.: You2.SG.NOM not be’ (‘Don’t be’)) has an agentive interpretation: the addressee has control over his being or not being at home, or, to put it differently, he or she can volitionally cause the required state (of being or not being at home). In contrast, (33e), where the hortatory particle niech is used (cf. Swan 2002:242), cannot be understood agentively. Rather, (33e) is understood as a (somewhat funny) wish that it might so happen (‘let it be so’) that the addressee will not be at home. Notice that if the ‘addressee’ is a third person, as in (33f), this is a perfectly licit sentence. (33) a. Bywaj częściej u nas! BE-2SG.IMP.HABIT more often at us Lit.: ‘Be more often at our house!’ (≈ ‘Come to visit us more often!’) b. Nie bywaj tak często w nocnych lokalach! NEG BE-2SG.IMP.HABIT as often in nightspots Lit.: ‘Don’t be so often at nightspots!’ (≈ Don’t spend so much time at nightspots!’) c. Bądź jutro koniecznie w domu BE-2SG.IMP tomorrow necessarily at home (mamy waŜnych gości)! have-1PL.PRS [important guests]-ACC ‘Whatever happens tomorrow be at home (we are having important guests)!’ d. Proszę cię, wieczorem nie bądź w domu, beg-1SG.PRS you-ACC evening-INS NEG BE-2SG.IMP at home bo oczekuję waŜnego gościa! because expect-1SG.PRS [important guest]-GEN ‘I beg you, don’t be at home tomorrow in the evening because I am expecting an important guest!’ e. ?Niech nie będzie cię jutro w domu! NEG BE-3SG.PRS you-GEN tomorrow at home let ‘Let it be so/let is so happen that you won’t be at home tomorrow!’ f. Niech Jana nie będzie jutro w domu! let John-GEN NEG BE-3SG.PRS tomorrow at home ‘Let it be so that John won’t be at home tomorrow!’ The second test, i.e., the occurrence in ‘pofective’ forms, is based on the analysis proposed by Piñón (1994).23 According to Piñón (1994), pofective po- attaches to imperfective verbs denoting processes and entailing an agent participant. However, 23
The term ‘pofective’ refers to (perfective) forms derived from corresponding imperfective predicates with the prefix po-, which is a marker of temporal delimination (the intuition being that pofective verbs describe situations that last only a relatively short time (Piñón 1994:341f.; see also Schoorlemmer 1995).
132
JOANNA BŁASZCZAK
there are cases in which pofective forms are derived from imperfective verbs denoting only states, as, e.g., siedzieć ‘sit’ → posiedzieć ‘sit for a while’; leŜeć ‘lie’ → poleŜeć ‘lie for a while’; być ‘be’ → pobyć ‘be for a while’. To account for this fact, Piñón (1994:347) assumes that some stative verbs like siedzieć, leŜeć or być are thus ambiguous between process and stative readings, depending on whether the subject argument is animate or not; cf. (34) (cf. also See also Levin and Rappaport Hovav 1995:126ff.). On their process reading such verbs entail an agent participant that controls the initiation and termination of a process denoted by the given verb, thus complying with the condition on pofective formation in (35) (Piñón 1994:348). (34) a. #Moja ksiąŜka pobyła wczoraj u Irenki yesterday at Irenka my book-NOM was ‘My book was at Irenka’s place for a while yesterday.’ b. BoŜenka pobyła wczoraj u Irenki yesterday at Irenka BoŜenka-NOM was ‘BoŜenka was at Irenka’s place for a while yesterday. (35) Pofective po- applies to any imperfective verb that a. denotes a property of processes b. entails an agent participant Now, given that it is possible to form a pofective form from być, we have evidence that być can in fact have an agentively interpreted nominal argument. Importantly, though, such an agentive interpretation is totally incompatible with a GEN marked nominal argument (cf. (36)), which shows once again that the GEN is in a sense a marker of nonagentivity (see the discussion above). (36) a. BoŜenki nie było wczoraj u Irenki BoŜenka-GEN NEG was-3.N yesterday at Irenka ‘BoŜenka was not at Irenka’s place yesterday.’ b. *BoŜenki nie pobyło wczoraj u Irenki BoŜenka-GEN NEG was-3.N yesterday at Irenka (intended: ‘BoŜenka was not at Irenka’s place for a while yesterday.’) 4.2 Subject properties The conclusion we arrived at in the last section was that NOM and GEN ‘subjects’ differ in terms of agentivity: while the GEN ‘subjects’ in negated (existential)locative być sentences never show agentive properties, the NOM subjects in (negated and affirmative) bywać and być sentences may very well have such properties. We also noticed that the postverbal NOM subject NP in affirmative (existential-)locative być sentences patterns with the GEN subjects in the respect that it also seems to be devoid of an agentive interpretation (recall example (26b)). In this section I investigate the question of whether the presence or absence of agentive properties correlates with the presence or absence of other typical subject properties. In other words, the question is whether there is a correlation between (non)agentivity
DIFFERENTIAL SUBJECT MARKING IN POLISH
133
and (non)subjecthood of a given nominal argument in the examples under discussion. Dziwirek (1994) argues that the grammar of Polish must recognize (three) different concepts of subject, which can be tested by using different diagnostics.24 The broadest notion of subject is that of a nominal heading a 1-arc. This construct does not make reference to a particular stratum and includes all nominals which head a 1-arc (i.e., bear the grammatical relation of subject in some stratum). Nominals heading a 1-arc can, among other things (see Dziwirek 1994:31ff for details), (i) antecede reflexives and (ii) control the fixed expression po pijanemu ‘while drunk’. The application of these tests to the nominal arguments of być and bywać is illustrated in (37) and (38) below. bywał u swoichi / *jegoi rodziców (37) a. Jani / *his parents John-NOM was-3SG.M.HABIT at REFL ‘John was from time to time at his parents home.’ był w swoimi / *jegoi pokoju b. Jani / *his room John-NOM was-3SG.M in REFL ‘John was in his room.’ Jani c. ?W swoimi pokoju był REFL room was-3SG.M John-NOM in Lit.: ‘In his room (there) was John.’ Jani but: c.’ *W jegoi pokoju był in his room was-3SG.M John-NOM nie było w jegoi / *swoimi pokoju25 d. Janai room John-GEN NEG was-3SG.N in his / *REFL ‘John wasn’t in his room.’ (38) a. Jan (nie) bywał w domu [po pijanemu] was-3SG.M.HABIT at home [while drunk] John-NOM ‘Johni used (didn’t use) to be at home while hei was drunk.’ b. Jan (nie) był w domu [po pijanemu] was-3SG.M at home [while drunk] John-NOM ‘Johni was (was not) at home while hei was drunk.’ c. ?W domu był Jan [po pijanemu] at home was-3SG.M John-NOM [while drunk] d. *Jana nie było w domu [po pijanemu] John-GEN NEG was-3SG.N at home [while drunk] ‘Johni wasn’t at home while hei was drunk.’
24
25
Dziwirek works in the framework of Relational Grammar, which treats grammatical relations as the primitives of lexical theory. A clause is represented as a relational network composed of arcs headed by elements which bear grammatical relations to the clause in a sequence of one or more strata. These judgments are quoted from Dziwirek (1994:154). Some researches, like, e.g., Przepiórkowski (1999:106) and Witkoś (1998:260), do not judge the variants with a reflexive as being completely ungrammatical.
134
JOANNA BŁASZCZAK
The two tests show that GEN NPs do not bear the grammatical relation of subject (cf. (37d) and (38d)). Rather, given their case marking and ‘thematic properties’ (lack of agentivity, see the discussion above; a ‘theme-like’ interpretation), these NPs seem to behave as (direct) objects (see Dziwirek 1994:154 for a suggestion along these lines). Thus, just like (direct) objects, they cannot serve as the antecedent of a reflexive pronoun (cf. (39a) and (37d)), nor can they control the fixed expression po pijanemu ‘while drunk’ (cf. (39b) and (38d)). o jeji / *swojąi mamę (39) a. Zapytałam Ewęi mother asked-1SG.F Eve-ACC about her / *REFL ‘I asked Eve about her mother.’ (Dziwirek 1994:33) b. Jan pobił Ewę [po pijanemu] John-NOM beat-up-3SG.M Eve-ACC [while drunk] OK ‘John beat up Eve while he was drunk.’ *‘John beat up Eve while she was drunk.’ (Dziwirek 1994:36) In contrast, the preverbal NOM NPs clearly have subject properties; cf. (37a/b) and (38a/b). As far as the postverbal NOM NP is concerned, the tested sentences in (37c) and (38c) are – in comparison with the perfectly well-formed examples with the preverbal NP subject in (37a/b) and (38a/b) – rather marked. This latter fact could be taken to mean that the postverbal NOM nominals in the cases at hand – despite their NOM marking and agreement with the finite verb, which are classified by Dziwirek as tests for final 1-hood26 – are too low in the surface structure to serve as antecedent or controller for certain elements. In other words, in order to establish the right binding or control configuration some additional ‘repair’ strategy is required, hence the markedness of (37c) and (38c). In sum, there seems in fact to be a correlation between agentivity and subjecthood. Those nominals that allow for an agentive interpretation (i.e., first of all, nominal arguments of bywać, and also, in certain cases – restricted to [+human] referents and preverbal position – nominal arguments of być) clearly show subject properties. In contrast, those nominal arguments that do not allow for an agentive interpretation either have no subject properties at all (as in the case of GEN marked NPs in negated (existential-)locative być clauses) or have a marked subject status (as in the case of postverbal NOM NPs in affirmative (existential-)locative być clauses). Given these correlations, we would expect (existential-)locative constructions to display an ergative/unaccusative syntax, and constructions with bywać (and also with locative być with a preverbal NOM subject, for that matter) to have an unergative syntax (cf. also Pereltsvaig 2001, Harves 2002; see also Borschev and Partee 2001 for discussion). In the next section I will provide some evidence showing that the above expectation might in fact be correct.
26
This concept includes only those nominals that bear the grammatical relation of subject in a final stratum.
DIFFERENTIAL SUBJECT MARKING IN POLISH
135
4.3 (Un)ergativity: Some diagnostics Usually BE is taken to be an unaccusative verb par excellence (cf. Babyonyshev 1996, Brown 1996; see also Moro 1997). Cetnarowska (2000) argues that there is a convenient deep unaccusativity diagnostic in Polish, namely the existence of resultative adjectives terminating in -ły. She notices furthermore that there is a correlation between the occurrence of resultative adjectives terminating in -ły and the nonoccurrence of related verbs in the impersonal -no/-to constructions. These latter constructions are taken to be a diagnostic for unergativity. Resultative adjectives can be derived from telic verbs only; some examples are given in (40). The only exceptions, which are related to stative verbs, are to a large extent lexicalized; cf. (41) (Cetnarowska 2000:87). (40) resultative adjectives a. przybyły ‘arrived’ b. upadły ‘fallen’ c. zmarły ‘dead’ (41) a. rosły ‘tall’ (cf. rosnąć ‘to grow’) b. były ‘former’ (from być ‘to exist’) c. bywały (w świecie) ‘experienced, knowledgeable’(cf. bywać ‘to frequent’) Since the forms in (41) have idiomatic meanings, this test cannot be used to decide on the difference between być and bywać. The second test, i.e., the -no/-to test, offers a better result. As already pointed out, -no/-to constructions are taken to be a diagnostic for unergativity. They can be built from transitive verbs (42a) and from unergative verbs (42b), but not from unaccusative verbs (42c).27 Applying this test to być and bywać, we observe that it works in the case of bywać (cf. (43b)), but not in the case of być (cf. (43a)). This would mean that bywać indeed has properties of an unergative verb.28
27
Rozwadowska (1992) independently argues that the -no/-to construction selects predicates with sentient and volitional subjects, i.e., with prototypical agentive arguments. This restriction is in fact similar to that imposed on the impersonal passives in Dutch/German; see Wurmbrandt (2004:994, fn. 2) for an interesting discussion. 28 Note, however, that there is something special about the iterative/habitual interpretation; cf. (i) vs. (42c). Example (i) actually shows that if an unaccusative verb has an iterative or a habitual reading, it is in fact possible to build a -no/-to form from it. This seems to indicate that on the habitual/iterative interpretation, an unaccusative verb no longer behaves as “unaccusative”. Why this should be the case is not clear to me at this point. More research is needed here to offer some insightful explanation. But see Aljović (2000) for the observation that in Serbo-Croatian “imperfectivized unaccusatives do not behave any longer as typical unaccusative verbs” (p. 14). (i) Podczas wojny umierano z głodu. (Cetnarowska 2000:90) during war no-died-IPFV from hunger ‘People would die from hunger during the war.’
136
JOANNA BŁASZCZAK
(42) impersonal -no/-to constructions (Cetnarowska 2000:83, 90) a. transitive Zbudowano szpital (*przez Ŝołnierzy) no-built hospital-ACC (*by soldiers) ‘They built a hospital.’ b. unergative Zatańczono (*przez Jana) no-danced (*by John) ‘They danced.’ c. unaccusative *Umarto z głodu to-died-PFV from hunger ‘They died of hunger.’ (43) a. *Byto na przyjęciu to-was at party b. Bywano na przyjęciach no-was-HABIT at parties ‘They were at parties./They used to be at parties.’ Before concluding this section, let me briefly mention another potential test for unaccusativity which is based on Grimshaw’s (1990) theory of nominals. According to Grimshaw (1990), nominalization – just like passivization – is an operation that affects an external argument (but see Rozwadowska 1997 for criticism). More specifically, the external argument of the base verb is suppressed in both passivization and nominalization. From this it follows that verbs with no external argument will neither passivize nor nominalize (into complex event nominals). Thus unaccusative verbs are expected not to be able to nominalize into complex event nominals (cf. Grimshaw 1990:112). Having said this, let us observe that one can perfectly well form a complex event nominal from bywać (cf. (44a)), which could be taken as evidence that this verb cannot be unaccusative, i.e., it must be unergative. In contrast, one cannot form a complex event nominal from (existential-)locative być when it refers to an inanimate subject; cf. (44b). The example is completely ungrammatical. Interestingly, the acceptability of such examples increases (though they remain quite bad29) once an animate (human) subject is used; cf. (44c). Notice that in cases like the one illustrated in (45) with a pronominal possessive subject the acceptability improves even more. If these observations are correct, we would have evidence that (existential-)locative być when constructed with inanimate subjects is clearly unaccusative, though it is probably not unaccusative (unergative?) when constructed with an animate [human] agentive subject; see the discussion below in section 6. (44) a. Bywanie Jana w nocnych klubach (zaskoczyło being-HABIT John-GEN in night clubs (surprised ‘John’s frequent going to nightclubs surprised all.’ 29
wszystkich) all)
The grammaticality judgements of the examples in (44) are due to BoŜena Rozwadowska (p.c.).
DIFFERENTIAL SUBJECT MARKING IN POLISH
137
b. **Bycie gazety na stole (zaskoczyło mnie) me-GEN) being newspaper-GEN on table (surprised intended: ‘newspaper’s being on the table surprised me.’ c. *Bycie Jana w domu (zaskoczyło mnie) me-GEN) being John-GEN at home (surprised intended: ‘John’s being at home surprised me.’ (45) (?) Twoje ciągłe bycie w domu denerwuje mnie your constant being at home irritates me-ACC ‘Your constant being at home irritates me/makes me nervous.’ 4.4 Partial conclusions We can now answer the first of the five questions posed in the beginning of the chapter, namely: Q1: What factors determine the case marking of the nominal arguments in (1)? The discussion so far has shown that the difference in the case marking of the nominal argument correlates with further important differences between (existential-)locative być sentences and bywać sentences. The first difference concerns the aspectual properties of the involved predicate. While bywać is clearly an imperfective verb, być has been argued to be ‘grammatically perfective’ (recall the discussion in section 3). The second difference concerns the subject properties of the nominal argument: Whereas the NOM marked nominal argument in bywać sentences passes all subjecthood diagnostics, the GEN marked nominal argument in (existential-)locative być sentences shows no subject properties. The GEN marks – in some sense – the absence of the ‘syntactic’ subject (or looking from another perspective: the presence of a dummy subject in the construction at hand; see section 6 for more details); the whole construction has, then, an ‘impersonal’ flavor. The difference in the subjecthood properties is related to the third difference between the constructions under discussion, namely the difference in the interpretation of the nominal argument. While the NOM nominal in bywać sentences is normally [+human], [+sentient], [+volitional], thus having typical properties of an agent argument, the GEN nominal in (existential-)locative być sentences does not need to be [+human]; on the contrary, even [+human] GEN-marked nominals lack agentive properties such as control and volitionality, thus showing typical properties of an internal (theme) argument. The agentive vs. nonagentive interpretation of the nominal argument seems to correlate with the fourth and last difference between (existential-)locative być and bywać, namely (un)ergativity. More precisely, on the basis of, among others, the deep unaccusativity diagnostic, the -no/-to test, bywać was classified as an unergative predicate and być as an unaccusative predicate. The factors determining the GEN versus the NOM marking of the nominal argument in (1a) and (1b), respectively, are summarized in Table 3 below.
138
JOANNA BŁASZCZAK Table 3. Factors determining the case marking of the nominal arguments in (1)
Aspectual
Subjecthood
Interpretation
Underlying
properties of
properties of
of the nominal
syntactic
the predicate
the nominal
argument
structure
lack of
unaccusative
argument GEN marked
‘grammatically no subject
nominal in (1a) perfective’
properties
agentive properties
NOM marked nominal in (1b)
imperfective
clear subject
presence of
properties
agentive
unergative
properties Now that we know what factors determine the NOM/GEN split in (1), we can start answering the remaining four questions. I will begin with the last question. Q5: Is there any deeper correlation between the Polish facts in (1) and the facts found in split-ergative languages (as in (3))? 5. THE CORRELATION BETWEEN ASPECT AND SYNTACTIC STRUCTURE: SOME OBSERVATIONS In the beginning of the chapter I pointed out that the GEN/NOM split in the Polish examples in (1) resembles in some sense the ERG/ABS split found in split-ergative languages in that both kinds of split seem to be conditioned by the aspectual properties of the predicate. As for split-ergative languages, it has been observed that whenever a language shows a split conditioned by aspect, it is always the perfective aspect which shows ERG(ative)/ABS(olutive) marking, whereas the imperfective aspect goes together with NOM/ACC marking (see, e.g., Trask 1979, Dixon 1994). Thus, there appears to be a close connection between perfectivity and ergativity. Why should this be the case? In order to understand this, let us assume, following Grimshaw (1990), that the argument structure “is a structured representation which represents prominence relations among arguments. The prominence relations are jointly determined by the thematic properties of the predicate (via the thematic hierarchy) and by the aspectual properties of the predicate” (ibid., p. 4). As for the latter, let us assume, again following Grimshaw, that the aspectual dimension “is a projection of an abstract event structure (e), which always includes two subparts, an activity (act) and a state or change of state (s/cos)” (ibid., p. 40); cf. (46).30 30
Grimshaw assumes that there is an event-structure template “which is fixed for all predicates rather than being projected from the lexical semantic representation of the individual predicate” (p. 40).
DIFFERENTIAL SUBJECT MARKING IN POLISH
139
(46)
e act
s/cos
Note that the template in (46) represents the structure of a complex situation type like accomplishments, which consist of activity and resulting state. In contrast, simple situation types correspond to one of the subparts: activities always fit the first slot in the template in (46), while states (or changes of state) always fit the second slot. Simplifying the picture a little bit, let us further assume that the unergative and the unaccusative predicates belong to particular aspectual classes: the former express activities while the latter express states or changes of state; cf. (47) (cf. Grimshaw 1990:26ff., 39f. and the references cited there). (47)
e accomplishments act
s/cos
activities
states/changes of state
unergatives
unaccusatives
external argument AGENT
internal argument THEME
Now, given Grimshaw’s assumption that an external argument is an argument that is maximally prominent in both (thematic and aspectual) dimensions (p. 33), in conjunction with the generalization that “an argument which participates in the first sub-event in an event structure is more prominent than an argument which participates in the second sub-event” (p. 26), it follows that the single argument of
140
JOANNA BŁASZCZAK
an unaccusative will never count as maximally prominent and will never qualify as external, while the single argument of an unergative will always count as external. With this background let us go back to the question of why there should be a connection between perfectivity and ergativity. In somewhat simplistic terms, it seems that both perfectivity and ergativity focus most naturally on what was identified in (46) above as the second subpart of the event-structure template: state/change of state.31 Given the scheme in (47), if the focus lies on state/change of state, this would mean that it is also the internal argument/THEME that is focused (or somehow gets into ‘the foreground’), while the external argument/AGENT somehow gets into ‘the background’. And finally, to bring the abovementioned split ergative languages into this picture, observe that what we have just said about the connection between perfectivity and ergativity is closely mirrored by what Trask (1979) writes about the origin of the Type B of ergative languages (i.e., those with tense/aspect split): Now the key word is stative. In a stative formation there is no action and hence no agent.32 The stative of an intransitive verb (e.g. gone, fallen) naturally refers to the NP which would be the subject of the corresponding active, while that of a transitive verb (e.g. broken, seen) equally naturally refers to the NP which would be the patient of the corresponding active transitive. The agent of the verb in this case occupies a peripheral position, being only secondarily connected with the state as the performer of the action resulting in the state. Thus in the stative aspect (in contrast with the situation in all finite forms of the verb referring to activity) ergativity is unmarked. When the stative form is used to make a predication, the NP most closely connected with the verbal adjective will be in the unmarked case; if and when an agent phrase comes to be attached to the stative of a transitive verb, it will be in some oblique case and carry an overt marking. I propose that this is the origin of Type B ergativity: the incorporation into the 31
Note that change of state would probably have to be understood to comprise both final state (outcome) resulting from some preceding activity and beginning of a state. In other words, the perfective aspect either focuses on the final boundary, i.e., the fact that the change was (or will be) attained, as, e.g., in the case of telic predicates entailing an inherent endpoint (an inherent change) that necessarily terminates the denoted eventuality (cf. (i)), or on the initial boundary (the inchoative phase), as, e.g., in the case of stative verbs (cf. (ii)). See Richardson (2003:24), referring to Filip (1994), and Rozwadowska (2003:865ff.) for more discussion on this point. (i) Jan napisał list. ‘final boundary’ John-NOM write-3SG.M.PST.PFV letter-ACC ‘John wrote a/the letter.’ (ii) Jan zrozumiał zadanie. ‘initial boundary’ John-NOM understand-3SG.M.PST.PFV problem-ACC ‘John started/got to understand the problem.’ 32 To understand what is meant by this, it is necessary to read the following passage about the origin of the Indo-European perfect: “Now the state of affairs attested by the historical languages is clearly this: the perfect and the (medio)passive are genetically related. This is borne out by the similarity of endings, and by the old intransitive value of the perfect and its strong association with the present mediopassive […] These formal and semantic facts authorize us to consider the Indo-European perfect as a verbal form denoting a state (resulting from a preceding action) and intimately related to the mediopassive. The semantic feature common to these two categories was the intransitive value, the differentiating contrast being state (perfect) versus action (in the mediopassive) […]” Trask (1979:396-7, quoting Kuryłowicz 1964:72).
DIFFERENTIAL SUBJECT MARKING IN POLISH
141
inflectional paradigm of a nominalized deverbal form with stative force. […] Since the verbal adjective is a nominalized form, it quite often happens that the agent phrase (or what is later interpreted as an agent phrase) is attached by means of a possessive construction, as seems to have been the case in Old Persian and Old Armenian […]; a consequence of this is that the ergative case is often identical to the genitive, dative or locative, the cases most often used to express possession. (Trask 1979:397f.).
6. ANALYSIS Returning now to question Q5 above, I think that there is indeed a deeper connection between the GEN/NOM split in Polish and the ERG/ABS split of the Hindi kind. The key to understanding both kinds of splits is the idea mentioned above that ergative structures are ‘stative’ and that the agent is often understood here as a ‘possessor’ of a state. Let us make this idea more precise. The situation (eventuality) denoted by the predicate in the Polish example (1a) is clearly stative: the predicate describes the existence of an entity at a particular location. The nominal argument is generated as the internal argument of the verb, i.e., it occupies the underlying direct object position. This is in accordance with Levin and Rappaport Hovav’s “Existence Linking Rule,” cited in (48) below: (48) EXISTENCE LINKING RULE (Levin and Rappaport Hovav 1995:153) The argument of a verb whose existence is asserted or denied is its direct internal argument. However, unlike Levin and Rappaport Hovav (1995), who assume that verbs of existence or verbs of spatial configuration in their simple position meaning have two internal arguments, a nominal argument and a locative argument (p. 82) (or unlike Hoekstra and Mulder 1990, who propose a small clause analysis: the NP and PP form a small clause, which is itself an internal argument of the verb), I would like to propose that the locative phrase is generated in some sense ‘externally’ to the verb. In this sense it is similar to the Ergative marked agent phrase in perfective sentences in split-ergative languages. In the latter case, as has been pointed out above, the Ergative marked agent phrase might be understood as a possessor of the state described by the participial clause (or location at which this state occurs). In the Polish example the locative phrase might also be understood as a possessor. In other words, the situation that some entity exists/does not exist at some location can be understood in such a way that the location contains/does not contain some entity (cf. Zamparelli 1995 for a similar analysis with respect to existential sentences in English and Italian).33 We can thus assume the following underlying structure of (existential-)locative być sentences in Polish:
33
Notice that in many languages of the world the verb ‘to have’ is actually used in existential constructions; cf., e.g., the French il y a constructions. See Freeze (1992) for some general observations concerning the relation between location, existence and possession.
142
JOANNA BŁASZCZAK
(49)
vP
PP-LOC
v’
VP
v
V
NP
The nominal argument is generated VP-internally in the direct object position, and the locative argument (understood as a possessor; see above) VP-externally in the specifier position of some light verb. Importantly, this light verb does not have any causative/agentive semantics (unlike the small v of canonical transitive verbs), i.e., the predicate is basically unaccusative. In a sense, the small v in (49) above resembles the stative small v assumed by Bennis (2004:86), following Kratzer (1994), for cases like John knows the audience, where the external argument generated in [Spec, v] is interpreted as Possessor, due to the fact that v is stative: “The thematic role Possessor indicates that the argument possesses the state denoted by the VP” (Bennis 2004:86, fn. 1); cf. (50). The presence of v in such cases thus follows from the principle/implication in (51) (Bennis 2004:88). (50) If v is stative: [Spec, v] is Possessor (John knows the audience) (51) If an external argument is present, v has to be generated. In affirmative sentences the locative phrase moves to Spec,TP to satisfy the Extended Projection Principle (EPP).34 The nominal argument remains in situ and undergoes long distance agreement with Tense. This explains the unmarked word order of such sentences (cf. section 4.1.1) and the ‘quasi’-subject behavior of the NP in such cases (understood in terms of agreement and NOM case marking; recall Dziwirek’s tests for final 1-hood) (cf. section 4.2).35 34 35
In this sense the structure is similar to that of locative inversion in English, for which it is also usually assumed that the locative phrase moves to Spec,TP to satisfy the EPP. The question that still needs to be clarified is what prevents the small “stative” v in (49) from having an ACC case feature. Note that the stative v obviously has this feature in cases like (50) above. One reason could be the actual “make-up” of the Possessor: a nominal Possessor correlates with an Accusative object while a locative Possessor does not. One could also think of a solution along the lines formulated by Freeze (1992) and the subsequent literature that a possessive verb HAVE (assigning the ACC case to its object) arises as the result of a syntactic incorporation operation of an abstract locative Preposition into BE. Thus, in cases like (49) one would have to assume that such an
DIFFERENTIAL SUBJECT MARKING IN POLISH
143
In negated (existential-)locative sentences, a NegP is inserted above vP and the verb moves to Neg°, which is due to the clitic nature of the negative marker in Polish (see Błaszczak 2001 for a discussion of the structure of negated clauses in Polish); cf. (52). The nominal argument which is marked for GEN in this case (see below) cannot undergo a long distance agreement with Tense. To implement this one could assume that only NPs with unchecked (or unvalued) case features can undergo an Agree relation (see Chomsky 1998, 1999). Given that the case feature of the NP has already been checked (or valued for GEN), the NP is no longer able to undergo a checking relation with Tense. As a consequence, the unchecked (unvalued) phi-features of Tense can only assume a default agreement (3.Sg.Neuter). In other words, the default agreement is a form of nonagreement; see Harves (2002) for an assumption along similar lines.36 As far as the Spec,TP position is concerned, it can be assumed that – just as in the previous case – the locative phrase can move there to satisfy the EPP. In cases in which it is the GEN NP that appears in preverbal position (recall (21c)) it could be assumed that this movement is driven by the relevant discourse properties of the GEN NP. But more research is certainly required on this point. (52)
NegP
Neg
vP
PP-LOC
v’
VP
v
V
36
NP
incorporation operation cannot take place (see, among others, Mahajan 1994 for a related discussion). I leave this question for further research. Alternatively, it could be assumed that the surface subject position, the Spec,IP (or Spec,TP), is filled by a dummy (cf. Dziwirek 1994), or in Witkoś’ (2000) terms, by an expletive pro of the it-type that is equipped with the categorial [+D] feature, 3rd person singular neuter agreement feature and the [+NOM] case feature, checking thus the relevant features of T and yielding the default agreement.
144
JOANNA BŁASZCZAK
At this point we can answer the question Q3, namely the question of where the GEN marking of the nominal argument comes from. It was pointed out in section 2 that the GEN marking in negated (existential-)locative clauses is in some sense a cross between the usual GoN in Polish (since it is triggered by negation) and the partitive (since it seems to be triggered by perfective aspect). The proposed analysis provides an answer to the ‘hermaphroditic’ nature of the GEN in negated (existential-)locative clauses. Given that the (existential-)locative być has two arguments: a locative argument generated as an ‘external’ argument (see the discussion above) and a nominal argument generated in the direct object position, this creates a configuration characteristic of canonical transitive verbs. Given this configuration, the rule of GoN can apply in this case. Recall that GoN applies only to the direct object of negated transitive verbs; in our case to the internal nominal argument of the ‘quasi-transitive’ negated (existential-)locative być. The importance of (grammatical) perfectivity comes into play in the present tense. Recall from section 4.1.2 (ex. (23) and (24)) that in the present tense instead of the regular negated form of the present tense form of być, the negated form of the verb mieć ‘to have’ is used (i.e., nie ma ‘not has’ and not the expected nie jest ‘not is’ form). I would like to suggest that this is so because the present form jest is not perfective in any obvious sense. Recall from section 3 that the actual present ‘jest’-forms derive from the Old Church Slavonic imperfective subparadigm of the present tense paradigm of the verb byti (see Table 1). I would like to suggest that this is precisely why the defective być paradigm is supplemented by the verb (nie) ma, which due to its (inherent) transitive nature – just like any other negated transitive verb in Polish (cf. section 2) – has the property of assigning the GEN case under negation.37 Note that due to its (inherent) transitive nature the GEN-assigning capacity of (nie) ma does not depend on its aspectual properties – just as is the case for any other negated transitive verb in Polish (recall the discussion in section 2). In other words, unlike in the case of negated (existential-)locative sentences with być (see below), there is no requirement for nie ma to be perfective in order to assign the GEN case to its internal argument. In sum, for a nominal argument in (existential-)locative być sentences to be marked for GEN two requirements have to be fulfilled: (i) the right configuration: the nominal argument has to be generated as a direct internal argument of the negated ‘quasi-transitive’ verb (this corresponds to the GoN rule in Polish), and (ii) the right aspectual properties of the predicate: the verb has to be grammatically perfective (this corresponds to the condition for the assignment of the optional Partitive case in Polish; cf. section 2). Given these two conditions responsible for the GEN marking in negated (existential-)locative sentences, the ‘hermaphroditic’ nature of the GEN in such cases receives a natural explanation. 37
This conclusion corresponds in some respect to the conclusion reached by Witkoś (2000) in his analysis of the negative locative copula nie ma. He analyzes nie ma in negated existential-locative clauses as “a defective verb, which can be regarded as an equivalent of the transitive verb mieć, ‘have’, but taking a locative/prepositional argument and devoid of the external argument,” and which is composed of several features, among others, the case feature [+Objective] (pp. 301-2). Notice that, unlike the analysis proposed in this chapter, the locative phrase is analyzed by Witkoś as an internal argument of the negated copula.
DIFFERENTIAL SUBJECT MARKING IN POLISH
145
What about negated locative bywać and być sentences with NOM subjects? Unlike the negated (existential-)locative sentences with GEN NPs, negated locative bywać and być sentences do not describe just a stative situation. The subject, which is usually a human entity, is in fact a controller or ‘internal causer’ (cf. Levin and Rappaport Hovav 1995) of the described eventuality. Piñón (1994) actually refers to such a reading as a process reading (recall section 4.1.4). Given this, the nominal argument will not be generated as an internal argument, but rather as an external argument in accordance with Levin and Rappaport Hovav’s ‘Immediate Cause Linking Rule’, cited in (53) below: (53) IMMEDIATE CAUSE LINKING RULE (Levin and Rappaport Hovav 1995:135) The argument of the verb that denotes the immediate cause of the eventuality described by that verb is its external argument. The structures of affirmative and negated locative bywać and być sentences with NOM subjects are given in (54) and (55), respectively. In both (54) and (55) it is the nominal argument that moves to Spec, TP to satisfy the EPP and at the same time undergoes an Agree relation with Tense (hence the NOM marking). This also explains the unmarked NP-V-PP word order of such sentences (see section 4.1.1). (54)
vP
NP
v’
VP
v
V
PP-LOC
146
JOANNA BŁASZCZAK
(55)
NegP
Neg
vP
NP
v’
VP
v
V
PP-LOC
Given the structure in (55) and given what we said above about the conditions of GEN assignment in negated (existential-)locative clauses, it becomes clear why the nominal argument in cases like (55) above cannot be marked for GEN (question Q2). The configuration required for GEN assignment is not satisfied in such cases since the nominal argument is generated as an external argument. In addition, in the case of negated bywać the predicate does not have the required aspectual properties. It should be noticed that the aspectual properties alone are a necessary but not sufficient condition for the GEN marking. In section 3, the verb być has been classified as grammatically perfective. However, despite this aspectual property, on the agentive reading the nominal argument of być – in accordance with the Immediate Cause Linking Rule in (53) – is generated externally, which precludes its GEN marking. In other words, unlike the (existential-)locative być, which is an unaccusative verb, bywać and być on their agentive reading are unergative verbs. Hence, the examples (1a) and (1b) have different syntactic structures (schematized in (52) and (55), respectively). This answers the last of the five questions posed in the beginning of the chapter, namely question Q4. 7. CONCLUDING REMARKS In the previous section we arrived at the conclusion that the question of whether an intransitive verb has an unergative or an unaccusative frame depends on both the aspectual properties of the predicate and the agentive interpretation of its nominal argument. This conclusion corresponds to the one reached by Arad (1998). In her paper she proposes a principle called ‘Structure-Interpretation Correspondence’ (SIC), which states that argument positions have fixed (semantic/aspectual)
DIFFERENTIAL SUBJECT MARKING IN POLISH
147
interpretations which are available only to arguments in that position (p. 17). For example, the aspectual interpretation associated with the subject position is that of originator of an event (causer, agent, etc.), and the aspectual interpretation associated with the object position is that of measurer of an event (affected argument, patient). In other words, SIC says that there are certain meanings which are only available to arguments if they occupy a certain position. “One of the consequences of [SIC] is that a certain position may not be associated with its particular interpretation. For example, there can be an object which is not a measurer. However, a given position can never be associated with another interpretation, which is available in (or associated with) another position. For example, you cannot have an agentive interpretation assigned at the object position.” (ibid. 17f.).
Applying this principle to the Polish examples, this would mean that in negated bywać and być on its agentive reading the nominal argument has to be generated in the subject position since otherwise it could not be interpreted as an agent. In contrast, the GEN nominal argument in negated (existential-)locative być sentences (and for that matter also the postverbal NOM nominal argument in affirmative (existential-)locative być sentences), which never have an agentive interpretation, cannot be generated in the subject position, but instead must be generated in the object position. As a consequence, (negated) bywać and być on its agentive reading have a syntactic structure typical of unergative verbs, while negated (existential)locative być sentences (and for that matter also affirmative (existential-)locative być sentences with the postverbal NOM nominal argument) have a structure typical of unaccusative verbs. REFERENCES Aljović, N. (2000). Unaccusativity and aspect in SerBoCroatian. Proceedings of CONSOLE 8. Ed. by C. Czinglar, K. Köhler, E. Thrift, E.J. van der Torre, and M. Zimmermann. Leiden: SOLE, 1-15. Arad, M. (1998). Are unaccusatives aspectually characterized? MIT Working Papers in Linguistics 32, 121. Babyonyshev, M. (1996). Structural Connections in Syntax and Processing: Studies in Russian and Japanese. Ph.D. Dissertation, MIT. Bach, E. (1981). On time, tense and aspect: An essay in English metaphysics. Radical Pragmatics. Ed. by P. Cole. New York: Academic Press, 63-81. Bennis, H. (2004). Unergative adjectives and psych verbs. The Unaccusativity Puzzle: Explorations of the Syntax-Lexicon Interface. Ed. by A. Alexiadou, E. Anagnostopoulou and M. Everaert. Oxford/New York: Oxford University Press, 84-113. Błaszczak, J. (2001). Investigation into the Interaction between the Indefinites and Negation. Berlin: Akademie Verlag [Studia grammatica 51]. Błaszczak, J. (2004). Some notes on aspect, (un)ergativity, and “X was not at Y”-constructions in Polish. Formal Approaches to Slavic Linguistics: The Ottawa Meeting 2003. Ed. by O. Arnaudova, W. Browne, M.L. Rivero and D. Stojanovic. Ann Arbor (MI): Michigan Slavic Publications, 37-57. Błaszczak, J. (2005). Two BEs or not two BEs? Beiträge der Europäischen Slavistischen Linguistik (POLYSLAV) 8. Ed. by M. Bayer, M. Betsch and R. Zimny. München: Sagner, 25-34. Borik, O. (2002). Aspect and Reference Time. Ph.D. Dissertation, Utrecht University. Borschev, V. and B. Partee (2001). The Russian genitive of negation in existential sentences: the role of theme-rheme structures reconsidered. Ms. Travaux de Cercle Linguistique de Prague (novelle série) [Working Papers of the Prague Linguistic Circle (new series)], v. 4. Ed. by E. Hajičova and P. Sgall. Amsterdam: John Benjamins.
148
JOANNA BŁASZCZAK
Brinton, L.J. (1987). The aspectual nature of states and habits. Folia Linguistica (Acta Societatis Linguisticae Europaeae) 21, 195-214. Brown, S. (1996). The Syntax of Negation in Russian. Ph.D. Dissertation, Indiana University. Brown, S. and S. Franks (1995). Asymmetries in the scope of Russian negation. Journal of Slavic Linguistics 3, 2, 239-287. Cetnarowska, B. (2000). The unergative/unaccusative split and the derivation of resultative adjectives in Polish. Formal Approaches to Slavic Linguistics: The Philadelphia Meeting 1999. Ed. by T.H. King and I. Sekerina. Ann Arbor (MI): Michigan Slavic Publications, 78-96. Chomsky, N. (1998). Minimalist inquiries: the framework. MIT Occasional Papers in Linguistics 15. [Published 2000 in Step by Step. Ed. by R. Martin, D. Michaels and J. Uriagereka.] Cambridge (MA): MIT Press, 89-155. Chomsky, N. (1999). Derivation by phase. MIT Occasional Papers in Linguistics 18. [Published 2001 in Ken Hale: A Life in Language. Ed. by M. Kenstowicz] Cambridge (MA): MIT Press, 1-52. Chvany, C. (1975). On the Syntax of BE-Sentences in Russian. Cambridge: Slavica. Comrie, B. (1976). Aspect: An Introduction to the Study of Verbal Aspect and Related Problems. Cambridge: Cambridge University Press. [Reprinted 1995.] den Dikken, M. (1995). Copulas. Manuscript, Vrije Universiteit Amsterdam/HIL. Dixon, R.M.W. (1994). Ergativity. Cambridge: Cambridge University Press. Dziwirek, K. (1994). Polish Subjects. New York/London: Garland Publishing. Enç, M. (1991). The Semantics of Specificity. Linguistic Inquiry 22, 1-25. Eriksen, P.K. (2000). On the semantics of the Russian copular verb byt’. Meddelelser Nr. 84, SlaviskBaltisk Avdeling, Universitetet i Oslo. Ferrell, J. (1953). On the aspect of byt’ and on the position of the periphrastic imperfective future in contemporary literary Russian. Word 9, 4, [Slavic Word 2], 362-376. Filip, H. (1994). Aspect and the semantics of noun phrases. Tense and Aspect in Discourse. Ed. by C. Vet and C. Vetters. Berlin/New York: Mouton de Gruyter, 227-256. Franks, S. (1995). Parameters of Slavic Morphosyntax. New York/Oxford: Oxford University Press. Freeze, R. (1992). Existentials and other locatives. Language 68, 553-595. Geist, L. (1999). Kopula byt’ (sein) eine funktionale und/oder eine lexikalische Kategorie? [Copula byt’ (BE) as a functional and/or lexical category?] ZAS Papers in Linguistics 14, 1-39. Grimshaw, J. (1990). Argument Structure. Cambridge (MA): MIT Press. Harves, S. (2002). Unaccusative Syntax in Russian. Ph.D. Dissertation, Princeton University. Heusinger, K. von and G. Kaiser (2003). The interaction of animacy, definiteness, and specificity in Spanish. Proceedings of the Workshop “Semantic and Syntactic Aspects of Specificity in Romance Languages”. Ed. by K. von Heusinger and G. Kaiser. Konstanz: Fachbereich Sprachwissenschaft der Universität Konstanz, 41-65. (Arbeitspapier Nr. 113.) Hoekstra, T. and R. Mulder (1990). Unergatives as copular verbs: locational and existential predication. The Linguistic Review 7, 1-79. de Hoop, H. and B. Narasimhan (this volume). Ergative case-marking in Hindi. Junghanns, U. (1997). On byt’ (and byti). Formale Slavistik. Ed. by U. Junghanns and G. Zybatow. Frankfurt am Mnz.:Vervuert Verlag, 251-265. Kayne, R. (1993). Towards a modular theory of auxiliary selection. Studia Linguistica 47, 3-31. Kiparsky, P. (1998). Partitive case and aspect. The Projection of Arguments. Lexical and Compositional Factors. Ed. by M. Butt and W. Geuder. Stanford: CSLI Publications, 265-308. Klebanowska, B. (1974). ‘Nie ma’, ‘nie było’, ‘nie będzie’. [‘There is not’, ‘There was not’, ‘There will not be’]. Prace Filologiczne [Philological Papers] XXV, 155-160. Kondrashova, N. (1996). The Syntax of Existential Quantification. Ph.D. Dissertation, University of Wisconsin at Madison. Kornfilt, J. (this volume). DOM and two types of DSM in Turkish. Kratzer, A. (1994). The Event Argument and the Semantics of Voice. Manuscript, University of Massachusetts. Kuryłowicz, J. (1964). The Inflectional Categories of Indo-European. Heidelberg: Winter. Lakoff, G. (1966). Stative adjectives and verbs in English. Mathematical Linguistics and Automatic Translation, I-1–I-16. Report NSF-17. Ed. by A.G. Oettinger. Cambridge (MA): The Computational Laboratory, Harvard University. Levin, B. and M. Rappaport Hovav (1995). Unaccusativity: At the Syntax-Lexical Semantics Interface. Cambridge (MA): MIT Press.
DIFFERENTIAL SUBJECT MARKING IN POLISH
149
Mahajan, A. (1994). The ergativity parameter: Have-Be alternation, word order and split ergativity. Proceedings of NELS 24, 317-331. Matushansky, O. (2001). The instrument of inversion: instrumental case in the Russian copula. Proceedings of WCCFL 19, 288-301. Młynarczyk, A. (2004). Aspectual Pairing in Polish. Ph.D. Dissertation, University of Utrecht. Moro, A. (1997). The Raising of Predicates. Cambridge: Cambridge University Press. Padučeva, E.V. (1992). O semantičeskom podxode k sintaksisu i genitivnom sub``ekte glagola BYT’. [On the semantic approach to syntax and the genitive subject of the verb BYT’ ‘BE’]. Russian Linguistics 16, 53-63. Pereltsvaig, A. (2001). On the Nature of Intra-Clausal Relations: A Study of Copular Sentences in Russian and Italian. Ph.D. Dissertation, McGill University. Piñón, C. (1994). Aspectual composition and the ‘pofective’ in Polish. Formal Approaches to Slavic Linguistics: The MIT Meeting 1993. Ed. by S. Avrutin, S. Franks and L. Progovac. Ann Arbor: Michigan Slavic Publications, 341-373. Przepiórkowski, A. (1999). Case Assignment and the Complement-Adjunct Dichotomy: A ConstraintBased Approach. Ph.D. Dissertation, University of Tübingen. Richardson, K. (2003). The Case for Meaningful Case: The Interaction of Case, Aspect, and Case in Russian. Ph.D. Dissertation, Harvard University. Rozwadowska, B. (1992). Thematic Constraints on Selected Constructions in English and Polish. Wrocław: Wydawnictwo Uniwersytetu Wrocławskiego. Rozwadowska, B. (1997). Towards a Unified Theory of Nominalizations. External and Internal Eventualities. Wrocław: Wydawnictwo Uniwersytetu Wrocławskiego. Rozwadowska, B. (2003). Initial Boundary and Telicity in the Semantics of Perfectivity. Investigations into Formal Slavic Linguistics. (Contributions of the Fourth European Conference on Formal Description of Slavic Languages FDSL IV, held at Potsdam University, November 28-30, 2001). (Linguistik International 10). Ed. by P. Kosta, J. Błaszczak, J. Frasek, L. Geist and M. śygis. Frankfurt am Mnz.: Peter Lang, 859-872. Rozwadowska, B. and E. Willim (2004). The role of the accusative/partitive alternation in aspectual composition in Polish. Poznań Studies in Contemporary Linguistics (PSiCL) 39, 125-142. [Papers from the Syntax Session at the 34th Poznań Linguistic Meeting, ed. by P. Tajsner and J. Witkoś.] Saloni, Z. and M. Świdziński (1985). Składnia współczesnego języka polskiego. [The Syntax of Contemporary Polish]. Warszawa: Państwowe Wydawnictwo Naukowe. van Schooneveld, C.H. (1951). The aspect system of the Old Church Slavonic and Old Russian verbum finitum byti. Word 7, 2, 96-103. Schoorlemmer, M. (1995). Participial Passive and Aspect in Russian. Ph.D. Dissertation, Utrecht University. Smith, C.S. (1997). The Parameter of Aspect. Dordrecht: Kluwer Academic Publishers. Swan, O.E. (2002). A Grammar of Contemporary Polish. Bloomington: Slavica. Trask, R.L. (1979). On the Origins of Ergativity. Ergativity: Towards a Theory of Grammatical Relations. Ed. by F. Plank. London: Academic Press, 385-404. Vainikka, A. and J. Maling (1996). Is partitive case inherent or structural? Partitives: Studies on the Syntax and Semantics of Partitive and Related Constructions. Ed. by J. Hoeksema. Berlin/New York: Mouton de Gruyter, 179-208. Witkoś, J. (1998). The Syntax of Clitics: Steps towards a Minimalist Account. Poznań: Motivex. Witkoś, J. (2000). Nominative-to-genitive shift and the negative copula nie ma: implications for checking theory. Journal of Slavic Linguistics 8, 1-2, 295-327. Wurmbrandt, S. (2004). Two types of restructuring – lexical vs. functional. Lingua 114, 991-1014. Zamparelli, R. (1995). Layers in the Determiner Phrase. Ph.D. Dissertation, University of Rochester.
PETER M. ARKADIEV
CHAPTER 7 Differential Argument Marking in Two-term Case Systems and its Implications for the General Theory of Case Marking1
1. INTRODUCTION In this paper I present a view of case marking that explicitly rejects a commonly assumed position that its primary function is to merely distinguish arguments from one another (cf. Comrie 1978, 1989; Dixon 1979, 1994), while marking them according to their specific semantic or pragmatic functions is a secondary phenomenon. In order to show that such a view (which has already been challenged by many linguists, see section 2) is untenable, I will investigate data from argumentencoding variations in languages which possess only two cases, and will compare them with similar phenomena from languages with richer case systems. As it will be seen, ‘nondiscriminative’ coding strategies found in two-term case systems, though typologically unusual, can be easily accounted for under the assumption that case marking of a particular argument is subject to ‘local’ ‘indexing’ rules and constraints dealing rather with this particular argument, than with the overall ‘global’ relational structure of the clause. The ‘discriminatory’ function, though retaining its importance, is, in this view, no more than just one of the constraints relevant for argument marking, whose ranking with regards to other such constraints is not always and not necessarily high. Also, I am going to argue that, contrary to some recent Optimality-theoretic proposals (see e.g., Woolford 2001), the case inventory found in a particular language cannot be always derived from a universal set of constraints (see Wunderlich and Lakämper 2001 for a similar proposal). As it will be shown, in order to account for case marking patterns in the languages with two-term case systems it is inevitable to regard the case inventory as a part of the input, and not as a feature of the candidates. In section 2 I will briefly outline the ‘discriminatory’ theory of case marking and summarize some arguments against it which have already been discussed in the literature. In section 3 I will discuss data from argument encoding variations (in particular, the interrelationships between tense/aspect conditioned ‘split ergativity’ 1
I am grateful to Geoffrey Haig, Helen de Hoop, Alexander E. Kibrik, Tracy Holloway King, Leonid Kulikov, Alexander Letuchij, Vladimir Plungian, Andrew Spencer, Peter de Swart, Donald Stilo, Jakov Testelec, and two anonymous reviewers for various help and useful comments and suggestions I received from them while preparing this paper. All faults and shortcomings are mine.
151 H. de Hoop and P. de Swart (eds.), Differential Subject Marking, 151–171. © 2008 Springer.
152
PETER M. ARKADIEV
and ‘differential object marking’) in the Indo-Iranian languages, and will show that their evidence is crucial with respect to my argument. In section 4 I will discuss subject marking in main vs. subordinate clauses in some Uto-Aztecan languages, which is also relevant for this topic. In section 5 I will present an Optimalitytheoretic conception of case marking which, I believe, is able to account for these facts. 2. THE ‘DISCRIMINATORY’ THEORY OF CASE MARKING The conception of case marking most commonly assumed among typologists is based on the assumption, clearly formulated by Comrie (1978:379-380; 1989:124127), and pursued to different extents also, among others, by Dixon (1979, 1994), Givón (1984:184), Legendre et al. (1993), Woolford (2001), Aissen (2003), that one and perhaps the most prominent function of case marking is to distinguish between those arguments which may be confused when simultaneously present in the clause (I will later call it the Syntagmatic Discrimination Principle, SDP). Thus, in transitive sentences, when both the A(gent) and the P(atient) arguments are present, it is necessary to mark either of them so that one might not confuse them. Consider the formulation of the D ISTINGUISHABILITY constraint by de Hoop and Narasimhan (2005), de Hoop and Lamers (2006): (1) DISTINGUISHABILITY The two arguments of a transitive relation should be distinguishable. On the contrary, it is not necessary to mark the S(ingle) argument of the intransitive predicate in any special way, because it cannot be confused with anything in its clause, and its marking would be redundant and uneconomical. So, we can state the following two constraints on the possible argument marking strategies: (i) SDP, which precludes A from being encoded identically to P, and (ii) *I(ndependent)-S, an economy-driven constraint, which prohibits special marking of the ‘intransitive subject’. Recasting these well-known insights in an Optimalitytheoretic fashion, we get the following predictions concerning the possible types of case marking strategies, see Table 1, where the input is a pair of clauses (a canonical transitive sentence, e.g., ‘The man killed the bear’, and an intransitive sentence, e.g., ‘The man is walking’), and all possible alignment types serve as candidates. Tableau 1. An OT-like account of case marking strategies
Vtr: A,P; Vitr: S a. accusative: S + A vs. P b. ergative: S + P vs. A c. neutral: S + A + P d. tripartite: S vs. A vs. P e. ‘quasi-neutral’: S vs. A + P
*I-S
SDP *
* *
*
DIFFERENTIAL ARGUMENT MARKING IN TWO-TERM CASE SYSTEMS
153
Both functionally motivated constraints predict that the accusative and the ergative systems of case-marking, neither of which violates any of the constraints in question, will be the most wide-spread case-marking patterns attested in the languages of the world, which is actually the case2 (see Nichols 1992:90 for the relevant statistics). A considerable difference in the frequency of occurrence between the neutral coding strategy (5 %), which marks all arguments similarly (or, to be more precise, usually does not mark them at all), and the tripartite coding strategy (< 1 %), which marks all three basic clausal arguments differently, suggests the following tentative ranking of our constraints: *I-S >> SDP. The quasi-neutral, as I call it (or ‘double-oblique’, see Payne 1979, 1980, 1989), coding strategy, which does not distinguish between A and P, but marks them separately from S and thus violates both constraints, is predicted to be the rarest one if not non-existent at all (cf. e.g., Kibrik 1979:63-66). Under the ‘discriminatory’ theory of case marking it is also predicted that A and P must be marked differently only when the roles of the nominals are not inferable from their lexical semantics, otherwise they may be left undistinguished. Thus, in a transitive clause with an animate and an inanimate participant the former will be the A and the latter the P with much greater probability than vice versa. The important generalization by Silverstein (1976; see also Moravcsik 1978a) concerning the person/animacy split of case marking is explained in this vein by Comrie (1978) and Dixon (1979), who explicitly argue that since it is ‘natural’ for As to be animate and for Ps to be inanimate, the ‘unnatural’ animate Ps and inanimate As get marked, while animate As and inanimate Ps remain unmarked. Also, in some languages, e.g., Hua (Papuan), a special marker is used only when there is a possibility of confusion between A and P, so it is claimed that in this language SDP is the only factor determining case marking (see Comrie 1978:384-385). The same holds, as it were, for the widely attested ‘differential object marking’ (Bossong 1985, Aissen 2003), when definite and/or animate Ps are encoded differently from indefinite and/or inanimate, which usually pattern with unmarked As (cf. Comrie 1979). A formulation of SDP pertaining to case marking only may be found in (2) (cf. Wunderlich and Lakämper’s (2001) UNIQUENESS and de Swart’s (2003) MINIMAL SEMANTIC DISTINCTNESS constraints): (2) SYNTAGMATIC DISCRIMINATION PRINCIPLE (for case marking) The two arguments of a transitive relation which are not distinguishable by their semantic/referential properties must be distinguished by case marking. It should be noted that (2) is less general than (1), in that it pertains only to morphological distinguishability (i. e., by means of case marking) of semantically and pragmatically similar clausal participants (see de Hoop and Lamers 2006 for a 2
Here I disregard the fact that, according to Nichols, the most frequent pattern of nominal case-marking is actually the ‘nondiscriminative’ neutral one. This fact is not surprising, however, since most languages in her sample simply lack any case marking on nouns. Neutral is the dominant ‘global’ alignment type only in a small, but noticeable (5 %, according to Nichols) number of languages with morphological case marking. Also, there must be independent reasons for ergative alignment being, contrary to the prediction of Table 1, almost two times less frequent than accusative.
154
PETER M. ARKADIEV
more comprehensible account of factors pertaining to distinguishability of As and Ps, such as agreement, word order etc.). The conception of case marking outlined above, as is well known, correctly predicts and attempts to coherently explain some important cross-linguistic generalizations concerning the types and distribution of argument-marking strategies. However, this theory is subject to objections of both a theoretical and an empirical nature. First of all, it has been often shown that cases usually have more or less coherent and more or less abstract semantic functions, which dictate the patterns of their usage (see, inter alia, Wierzbicka 1980, 1981, 1983, 1988:Ch. 8; Mallinson, Blake 1981:Ch. 2; Du Bois 1987; Primus 1999; Wunderlich and Lakämper 2001, Song 2001:156-167; Testelec 2003:64-68) regardless of the apparent need to distinguish between arguments. This is in accordance with the grammaticalization theory which claims and convincingly shows that ‘grammatical’ cases arise diachronically from morphemes with more ‘concrete’ functions (see Lehmann 1995/1982:66-107); it is hardly conceivable how one could place the discriminatory function on any of the well documented grammaticalization paths of case morphemes. The discriminatory function may thus be a mere by-product of the process of grammaticalization, but not its primary driving force. Second, there are large pieces of evidence that languages perfectly tolerate almost any degree of Agent-Patient ambiguity (see, e.g., Moravcsik 1978b, Plank 1980; cf. also de Hoop and Lamers 2006) and that quite a number of languages (5 % in the Nichols’ (1992) sample) lack any kind of grammaticalized discriminatory devices (Riau Indonesian, see Gil 1994, 1999, is a well-known example). Third, this theory incorrectly predicts that both ‘optimal’ case marking strategies, viz. accusative and ergative, must show comparable frequency of occurrence. Why such a prediction fails is not obvious; for a tentative explanation see Primus (1999). Therefore, one has to admit, I believe, that however appealing the ‘discriminatory’ theory may seem, it cannot be adopted without serious amendments. Some steps towards a more balanced theory of case-marking have been made in current literature (e.g., Kibrik 1997), especially in the framework of Optimality Theory, see the already mentioned contributions by Aissen (1999, 2003), Primus (1999, 2003), Wunderlich and Lakämper (2001), de Hoop and Narasimhan (2005), Malchukov (2005, 2006), and some other (e.g., a non-OT-based paper Butt and King 2002a). In their approaches, different though they are, it is assumed that there are two main functions of case marking (see the discussion in de Hoop and Narasimhan 2005): the already discussed ‘discriminatory’ function, and the so called ‘indexing’ function, which pertains to the encoding of particular semantic features of arguments. These functions quite often compete with one another and their tension is resolved by the relative ranking of the relevant constraints. OT approaches to case also assume (tacitly or explicitly) that not only case marking patterns in individual languages, but also language-particular case inventories may be predicted by universal constraints and their different rankings. Such an approach is advocated, e.g., by Woolford (2001), see also Aissen (1999:685-686). However, I believe that such a view is basically incorrect, and I will present some data which unequivocally contradict it.
DIFFERENTIAL ARGUMENT MARKING IN TWO-TERM CASE SYSTEMS
155
In the following sections I attempt to show that the data usually disregarded by typologists may present important empirical evidence against the primary role of SDP in argument encoding. Conversely, I outline a conception of case marking which, in my view, is free of the disadvantages of the ‘discriminatory’ theory, though it incorporates its basic insights. 3. DIFFERENTIAL ARGUMENT MARKING IN INDO-IRANIAN The modern Indo-Iranian languages possess a variety of argument marking patterns, which, however, have not been subject of detailed investigation of those linguists who advocate the ‘discriminatory’ theory of case marking. A closer look at those facts reveals that case marking in these languages is hard to explain in terms of SDP. In this section I present the relevant facts, paying special attention to the data from the languages with two-term case systems. Let us look at the patterns of argument marking in Vafsi, an Iranian language with two cases: unmarked Dir(ect) and marked Obl(ique). Like quite a number of its neighbors it exhibits two types of differential argument marking: a tense/aspect split, which affects the marking of A, and a animacy/definiteness split, which is relevant for the marking of P. In the non-perfect tenses, S and A are encoded by Dir, which has zero exponence (Stilo 2004:231, 243): (3) tæ æten bæ- ssæ in kelj- i palu you(DIR) now PFV- go this girl- OBL to ‘Now you go to this girl.’ (4) tæ in xæri næ- ruš- i? you(DIR) this donkey- OBL NEG- sell- 2SG ‘Won’t you sell this donkey?’ P must be marked by Obl if it is both animate and specific (individuated, in Lazard’s (1984, 1994) sense), as in (4), or left unmarked otherwise (Stilo 2004:243): (5) bæ- ss- e yey xær ha- gir- e PFV- went- 3SG one donkey(DIR) PVB- take- 3SG ‘He went to buy a donkey.’ In the perfect tense, however, A is invariably encoded by Obl, and S by Dir (Stilo 2004:244, 226): (6) in lutian yey xær= esan this wise.guy- OBL.PL one donkey(DIR)= 3PL ‘These wise guys were selling a donkey.’ (7) qondaq bidara næ - væ ? swaddled(DIR) wake.up NEG- become ‘Didn’t the infant wake up?’
æ- ruttæ DUR-sell.PST
156
PETER M. ARKADIEV
However, the differentiation of animate vs. inanimate Ps remains intact in the perfect tense, resulting in the following ‘nondistinctive’ quasi-neutral structure (Stilo 2004:244): (8) luas- i kærge= s bæ- værdæ fox- OBL chicken- OBL= 3SG PFV- take.PST ‘The fox took the chicken.’ Similar patterns can be found in other languages of this region, cf. the following examples from the Southern Tati dialect Chali (Yar-Shater 1969:75-76): (9) bar beškias door(DIR) broke ‘The door broke.’ ešō bāšind (10) varziar- ō barrōn3= peasant- OBL.PL spade- OBL.PL= 3PL threw.away ‘The peasants threw away their spades.’ and from Roshani, a Pamir language, where only demonstrative pronouns decline for case (Payne 1980:155): (11) dāδ xawrič- ēn= an tar Xaraγ sat PL= 3PL to Xorog go.PST these.DIR boy‘These boys went to Xorog.’ (12) duf xawrič- ēn um kitōb xēyt PL this.OBL book read.PST these.OBL boy‘These boys (have) read this book.’ In spite of the rarity (quasi-neutral patterns are, to my knowledge, attested only in the Iranian and perhaps also Dardic languages, (see Skalmowski 1974, Payne 1979, 1980, 1989, Bossong 1985) and apparent ‘disfunctionality’ of these structures, it is evident that they are clearly motivated (see, e.g., DeLancey 1981, Tsunoda 1981, Lazard 1994 for important insights into the nature of functional motivation of various types of ‘split case marking’), notably not by the SDP. The alleged need to morpho-syntactically discriminate between the syntagmatically co-occurring arguments seems to be altogether irrelevant here. What counts is, on the contrary, the semantic/pragmatic properties of the arguments themselves: individuated (animate and/or definite) Ps are marked, while their less individuated counterparts are left unmarked. Similarly, those As which coincide with the ‘aspectual point of view’ (in DeLancey’s 1981 sense), i.e. As of imperfective/non-past clauses, are unmarked, whereas those which do not, viz. As of clauses with perfective aspect or past tense, bear overt case markers. Let us compare Vafsi data with that of Hindi/Urdu, which uses postposition-like case markers for the encoding of core grammatical functions (see Mohanan 1994, 3
The final -n of the Oblique Plural ending appears before a vowel.
DIFFERENTIAL ARGUMENT MARKING IN TWO-TERM CASE SYSTEMS
157
Butt and King 2002a,b, Lee 2003, and de Hoop and Narasimhan 2005, this volume, for detailed accounts), its case system thus radically differing from two-term case systems common in the Iranian languages. Hindi is similar to Vafsi, however, in one crucial respect: it possesses the same two types of argument encoding alternations, viz. a tense/aspect ergativity split and differential object marking. The difference lies only in how these patterns are realized by the surface morphology. As in Vafsi, in Hindi/Urdu both S and A are unmarked in the non-perfective tenses (Mohanan 1994:72, 59): (13) Rām gā rahā hai sing DUR COP.PRS.3SG Ram(NOM) ‘Ram is singing.’ (14) Ravī kelā khā rahā thā Ravi(NOM) banana(NOM) eat DUR COP.PST ‘Ravi was eating a banana.’ If P is individuated, it is marked by the postposition =ko, which is also used to mark the Recipient/Addressee in ditransitive constructions; I will gloss it Acc(usative) (Mohanan 1994:59): (15) Nīnā bacce= ko uthāyegī Nina(NOM) child.OBL4= ACC pick.up.FUT ‘Nina will pick the child up.’ In the past tenses the A argument is marked by the postposition =ne glossed Erg(ative) (Mohanan 1994:59): (16) bacce= ne kītāb padhī child.OBL= ERG book read.PFV ‘The child read a book.’ It is necessary to mention here that not only transitive Agents are marked by =ne in Perfective tenses, but some intransitive Agents as well; here the Ergative encoding is clearly semantically motivated: the presence of =ne implies volitionality and control on the part of the subject, cf. the following pair (Mohanan 1994:72; see also de Hoop and Narasimhan, this volume): (17) a. rām= ne cillāyā. b. rām cillāyā Ram(NOM) scream.PFV Ram= ERG scream.PFV ‘Ram deliberately screamed.’ ‘Ram screamed (e.g., because of fright)’
4
Just as many Iranian languages, Hindi/Urdu has retained an older distinction between a Direct (unmarked) and an Oblique morphological case; however, it has developed a new ‘layer’ of grammaticalized case markers from former postpositions; see Masica (1991:230-248) for a comprehensive survey.
158
PETER M. ARKADIEV
Like in Vafsi, ‘differential object marking’ in Hindi/Urdu is present in all tenses, regardless of whether there is any need to distinguish P from A; cf. the following example, where both Ergative and Accusative postpositions are present (Mohanan 1994:80): (18) Īlā= ne bacce= ko uthāyā Ila= ERG child.OBL= ACC lift.PFV ‘Ila lifted the child.’ It is clearly seen from the examples above, that the ‘global’ strategies of argument marking which exist in Vafsi and other Iranian languages and Hindi/Urdu are no more than the automatic consequence of (i) functionally motivated argumentmarking rules, assigning special case markers to As in the past/perfective tenses and to individuated Ps, and most importantly (ii) the actual number of core case markers in these languages. In Vafsi there is only one non-zero case marker, which is thus used as the output of both rules; in Hindi/Urdu there are many different postpositions, thus no quasi-neutral pattern arises. The patterns of argument marking in these languages are summarized in Tables 2 and 3. Table 2. Patterns of argument marking in Vafsi
S Dir Dir Dir Dir
A Dir Dir Obl Obl
P Dir Obl Dir Obl
strategy neutral accusative ergative quasi-neutral
conditioning factor non-past; non-individuated P non-past; individuated P past; non-individuated P past; individuated P
Table 3. Patterns of argument marking in Hindi/Urdu
S Nom Nom Nom ~ Erg Nom ~ Erg
A Nom Nom Erg Erg
P Nom Acc Nom Acc
strategy neutral accusative ergative tripartite
conditioning factor imperfective; non-individuated P imperfective; individuated P perfective; non-individuated P perfective; non-individuated P
It is thus evident that both the ‘nondistinctive’ quasi-neutral pattern of Vafsi, Chali and Roshani and the ‘over-distinctive’ tripartite pattern of Hindi/Urdu are conditioned by the very same functional-semantic factors and differ only in their surface realizations, which is merely a consequence of a more or less arbitrary factor such as the number of core case markers in a particular language.
DIFFERENTIAL ARGUMENT MARKING IN TWO-TERM CASE SYSTEMS
159
4. MAIN VS. SUBORDINATE CLAUSE SPLIT IN THE UTO-AZTECAN Some Uto-Aztecan languages have two-term case systems for nouns. The dominant pattern of case marking is accusative: S/A is encoded by Dir, and P by Obl. A typical example is Chemehuevi (Press 1979:73, 78): (19) maŋ nakwi- j PRS he(DIR) run‘He is running.’ (20) maŋ puŋkuc- i kiɁi- vi he(DIR) dogOBL bit- PST ‘He bit the dog.’ However, in many types of subordinate clauses (relative clauses, complement clauses, and adjunct clauses) the marking of the S/A argument switches to Obl (Press 1979:108, 11, 115): (21) [puŋkuc- i havitu- g] aipac ay tika- vi OBL singSIM boy(DIR) that eat- PST dog‘While the dog sang, the boy ate.’ aipac- i kwipa- vi (22) waampakwic [nini paka- mpa- n] OBL stingPST scorpion(DIR) I.OBL kill- FUT- NML boy‘The scorpion I am going to kill stung the boy.’ kiaw taya- kai- n] (23) John [Ann- i karitia- j OBL chair- OBL yesterday kick- PFV- NML John Annputucuga- j PRS know‘John knows that Ann kicked the chair yesterday.’ Example (23) is of particular interest here. Both arguments of the subordinate clause are marked by the same Obl case, thus resulting in a ‘non-distinctive’ neutral structure. The same holds for another language of this family, Yaqui, which, too, has only two fully grammaticalized cases on nouns (personal and demonstrative pronouns have, apart from that, also a separate possessive form). In main clauses, argumentencoding follows the accusative pattern: S/A is unmarked, and P gets Oblique case (Lindenfeld 1973:11, 54): (24) itom čuuɁu bem kari= po we.POSS dog(DIR) they.POSS house= in ‘Our dog is playing in their house.’ (25) inepo em misi- ta biča- k I(DIR) you.POSS cat- OBL see- PFV ‘I saw your cat.’
yeewe play
160
PETER M. ARKADIEV
The Obl case also fulfills the function of the genitive, marking possessors (Lindenfeld 1973:56): (26) itom pareta kari si weela we.POSS priest- OBL house(DIR) very old ‘Our priest’s house is very old.’ Like in Chemehuevi, Yaqui encodes its subjects with Obl in subordinate clauses, cf. the following examples from Lindenfeld (1979:65, 81, 103): (27) hu kari [in ačaita hinuk- aɁu] weče-k NML fall- PFV this house .DIR I.POSS father- OBL buy‘The house my father bought fell down.’ (28) [hu- ka oɁoo- ta yepsa- k-o] itepo saha- k this- OBL man- OBL arrive- R-TEMP we.DIR go- PFV ‘When this man arrived we left.’ (29) na= a biča ke [hu- ka usi- ta čuɁu- ta kipwe- Ɂu] I.DIR= it see that this- OBL child- OBL dog- OBL have- NML ‘I see that this child has a dog.’ Similarly to Chemehuevi, the case marking of P is not affected by the case alternation on the subject, and thus a ‘non-discriminative’ pattern, like in (29), arises. The explanation for this pattern of case marking in subordinate clauses in both languages is relatively straightforward: as is well attested cross-linguistically (see, e.g., Keenan 1985:160-161, Lehmann 1988:195-200), verbs in subordinate clauses are often nominalized, and the marking of their subjects patterns with that of possessor NPs. Consider similar structures in English: (30) John visited Bill. [John’s visiting Bill] was a disaster. That this is the case in the Uto-Aztecan languages as well is proven by evidence from Yaqui, where the pronominal subject of the embedded clause is encoded similarly to the pronominal possessor, and not to the pronominal direct object, cf. the following example (Lindenfeld 1973:72): (31) ini- ka bači- ta [em hinu- k- aɁu] nee this- OBL corn- OBL you.POSS buy- R- NML I.OBL ‘Give me the corn that you bought.’
maka give
Here only the possessive form of the pronoun (em ‘your’, as in (25)) is possible, and not the form used for direct/indirect objects (enči ‘you.OBL’).
DIFFERENTIAL ARGUMENT MARKING IN TWO-TERM CASE SYSTEMS
161
From the evidence presented in this section it is possible to arrive at a conclusion similar to that of the previous one where I dealt with case-marking alternations in the Indo-Iranian languages: neutral argument marking pattern in embedded clauses in Yaqui and Chemehuevi is no more than an epiphenomenon of the interaction of two factors: (i) the typologically natural situation when subordinate clauses are headed by verbal nouns triggering possessor inflection on their subjects, and (ii) the fact that in these languages there are only two cases on nouns, and therefore nominal possessors and nominal Ps get similar marking. Such a situation does not arise in languages where a matrix/embedded split coincides with a richer case system which has different cases for Ps and adnominal dependents. 5. TOWARDS A UNIFIED APPROACH TO CASE MARKING In order to account for the facts presented in the previous two sections, I would like to outline a conception of case marking which, on the one hand, would capture the major insights of the ‘discriminatory’ view, and, on the other hand, would be devoid of its disadvantages. The main idea behind my proposal is that the Syntagmatic Discrimination Principle (SDP) is only one of the factors which may determine the actual patterns of argument encoding in a particular language and across languages, and that it may be (and often is) overridden by other principles. The most important rival of SDP that operates in all the languages I have surveyed is best regarded as a family of rules constraining the marking of particular arguments in particular contexts, which are itself determined by universal functional tendencies (cf. Lazard 1994 and Kibrik 1997 for comprehensive surveys of such motivations). For instance, in the Indo-Iranian languages there is a rule PERFA which requires the A argument to be marked differently according to the tense/aspect of the verb, and a rule INDIVP which is responsible for the dissimilar encoding of indefinite/inanimate (less individuated) vs. definite/animate (more individuated) Ps. A corresponding Uto-Aztecan rule is EMBEDSB, which assigns non-nominative case to the subjects of embedded clauses. It is important that the surface realization of these rules, viz. the actual case marking device which is used to mark the argument they apply to, is determined on the language-particular level and crucially depends on the inventory of formal means a language possesses. What all these rules (which, of course, operate in a large number of different languages, not just in those surveyed here) have in common is the fact that their application creates context-sensitive alternations in the encoding of a single argument (A, P, S/A, etc.) regardless of the properties of other NPs present in the clause. I believe them to be possible instantiations of a more general principle, which I will call the Paradigmatic Discrimination Principle (PDP). The rigorous formulation of PDP is not so easy to arrive at, so I will only attempt to give an informal characterization:5 5
The distinction between SDP and PDP is akin to the difference between the Syntagmatic and Paradigmatic Argument Selection Principles of Ackerman and Moore (2001); it is probable that similar semantic factors operate in both domains.
162
PETER M. ARKADIEV
PARADIGMATIC DISCRIMINATION PRINCIPLE (PDP) The argument with a given semantic role may be encoded differently depending on contextual and/or inherent factors, such as its position on empathy and referentiality hierarchies, viewpoint distinctions (≈ tense/aspect of the clause), or the independent/embedded status of the clause (see DeLancey 1981, Tsunoda 1981). What is most important about PDP, is that, unlike SDP, it is not a rule straightforwardly applying to any possible clause. As a general principle, it can be regarded only as a typological tendency; which PDP-driven rule or rules are operating in a particular language is a matter of its history,6 just as the presence or absence of nominal case marking. Each instantiation of PDP may be seen as a constraint contributing to the determination of the actual inventory and distribution of argument encoding types in a given language. SDP also plays an important role here, but the relative ranking of these constraints varies from language to language, and it is not always the case that SDP is higher in rank than any of the PDP-constraints. Let us see how the data surveyed above may be accounted for in these terms, by casting these constraints in the functionally-based Optimality-theoretic framework (see Aissen 1999, 2003, Aissen and Bresnan 2002 and others). First let us formulate the relevant PDP-based constraints. In order to do this it is necessary to bear in mind not only that there are certain (gradual or binary) distinctions relevant for differential argument marking, such as semantic role, grammatical function, animacy, person, definiteness, volitionality, perfective/imperfective, independent/embedded, etc., but also the fact that there are more ‘prototypical’ or less ‘marked’ configurations of these properties and less ‘prototypical’ and correspondingly more ‘marked’ constellations of them (see especially DeLancey (1981) for an initial proposal and Aissen (1999, 2003) for an enlightening OT account). For instance, it is known that it is less typical for Ps to be animate and/or definite (‘individuated’) than inanimate and/or indefinite (‘non-individuated’). Also, following DeLancey (1981), I assume that it is unmarked for As to be ‘viewpoint foci’, that is to occur in Imperfective clauses where the activity of the A is foregrounded, rather than in Perfective clauses which underscore the change of state undergone by the P. Observations of a similar kind follow also for subject of syntactically embedded (and thus pragmatically backgrounded) and independent (pragmatically foregrounded) clauses. These generalizations may be stated as in the following harmonic scales and corresponding constraint hierarchies (where ‘>’ should be read as ‘more harmonic than’), see Table 4.
6
This certainly does not contradict the view that the functional constraints motivating these rules are universal. Cross-linguistically valid functional constraints have language-particular instantiations and language-particular rankings, just as it does not follow from the universality of grammaticalization paths that all languages must grammaticalize all possible categories. I thank the editors for pointing out to me that it is necessary to clarify this point.
DIFFERENTIAL ARGUMENT MARKING IN TWO-TERM CASE SYSTEMS
163
Table 4. PDP-based harmonic scales and constraint hierarchies
Harmonic scales NonIndiv/P > Indiv/P Imperf/A > Perf/A Indep/Sb > Embed/Sb
Constraint hierarchies *INDIV/P >> *NONINDIV/P *PERF/A >> *IMPERF/A *EMBED/SB >> *INDEP/SB
In the languages surveyed both marked and unmarked constellations of argument properties are permitted and differ only in their relative formal markedness. Therefore it is necessary to supplement the constraint hierarchies in Table 4 with additional constraints which would have as a result that individuated Ps are casemarked while non-individuated ones are not, and similarly with As in imperfective and perfective clauses. Here, I follow proposals by Aissen (1999, 2003) and Primus (2003) postulating the following markedness constraints, see table 5. *X/∅ should be understood as ‘the feature constellation X should not receive default (Nominative/Direct) case marking’.7 Table 5. Constraint hierarchies conjoined with *∅
*INDIVP/∅>> *NONINDIVP/∅ *PERFA/∅ >> *IMPERFA/∅ *EMBEDSB/∅ >> *INDEPSB/∅ In order to account for those patterns of case marking where one or several arguments get unmarked (Nominative/Direct) case I assume, again following Aissen, the constraint *STRUC which penalizes assignment of marked cases; this constraint may be interpolated at various points of the relevant hierarchies thus predicting a sufficient range of cross-linguistic variation in differential argument marking. At last, SDP as formulated in (2) is violated by all transitive structures which simultaneously satisfy the following two conditions: (i) both arguments are prominent on animacy/definiteness scale, e. g. both are animate, or one of them is animate, and another definite; (ii) these arguments bear the same case marking. Thus, SDP is violated even if there are other grammatical clues, such as verb agreement of word order, which help to distinguish the two arguments. Another very important and already mentioned aspect of the account I propose here is the treatment of the language-particular case inventories. Contrary to current OT practice which attempts at accounting for the number and character of cases in individual languages on the basis of universal constraints and their languageparticular rankings, I propose to treat case inventory of each language as a part of 7
Aissen (op. cit.) and de Hoop & Narasimhan (2005) assume that ‘default’ marking is identical to ‘no case marking at all’, which view I am somewhat reluctant to adhere to (e.g. because there are languages, such as Japanese or Aleut, in which Nominative case is not formally unmarked).
164
PETER M. ARKADIEV
the input,8 and not just as a feature specification of the candidates, generated by GEN. I believe that this approach is more consistent with the actually attested data, which proves to be more ‘messy’ than one could conclude without taking into account the whole range of attested possibilities. The case inventory of a particular language is the result of various and often conflicting tendencies, many of which are better looked upon as historical ‘accidents’ rather than instantiations of universal ‘laws’. The data from two-term case systems, in my opinion, support such a view. Let us now turn to how the actual case marking patterns found in the languages surveyed in sections 2 and 3 can be accounted for under these assumptions. Let us begin with the Uto-Aztecan patterns, where only one PDP-driven rule is operating, namely EMBEDSB. Let the input be an embedded clause with both Agent and Patient; there are only two cases in these languages (Dir and Obl). The set of candidates I consider here consists of (i) the actual one, where both A and P are marked by Obl, and where SDP is violated (but only if both A and P are animate/definite); (ii) the one where SDP is satisfied by leaving A unmarked, thus violating *EMBEDSB/∅; and (iii) where both *EMBEDSB/∅ and SDP are satisfied by suspending case marking of the P, which violates either *INDIVP/∅ or *NONINDIVP/∅, which in these languages are both ranked over *STRUC thus securing that all Patients are marked,9 see constraint ranking in (34) and Tableau 6.10 (34) *EMBEDSB/∅, *INDIVP/∅ >> *NONINDIVP/∅ >> *SDP >> *STRUC Tableau 6. Case-marking of embedded clause subjects in Yaqui and Chemehuevi
8
As the editors point out, it is actually not correct to regard the case inventory as a part of the input per se; rather, it is a more general language-particular specification (most probably pertaining to the lexicon, since it is, in my view, possible to treat language-particular case grammemes on a par with ordinary lexical items) constraining the possible candidates; i. e., for a language with a two-term case system, like Vafsi, GEN simply does not generate candidates marked with, say, Genitive or Accusative. The rigorous formulation of this point is a topic for further research. Also, I do not take into account candidates where, e. g. A gets Accusative case or P gets Ergative; those I consider to be excluded by higher-ranked faithfulness constraints checking whether the semantic role of the argument is compatible with the specifications of the case grammeme, see e. g. Wunderlich and Lakämper (2001). 9 It is not important which of the PDP-driven constraints, viz. subject-oriented *EmbedSb/∅ or objectoriented *IndivP/∅ is ranked higher; here it is possible to assume that they are not ranked with respect to each other. However, as the discussion of the Indo-Iranian data will show, sometimes the relative ranking of different PDP-driven constraints is crucial. 10 I do not omit SDP from my tableaux even though it does not really play a role in the evaluation of the candidates, because what I want to show explicitly is precisely that in Yaqui, Vafsi and Hindi/Urdu it is irrelevant whether SDP is satisfied or violated.
DIFFERENTIAL ARGUMENT MARKING IN TWO-TERM CASE SYSTEMS
165
Let us now turn to the more complicated Indo-Iranian case. In Vafsi and other Iranian languages, as has already been shown, the ‘need’ to differentially mark As according to aspect and Ps according to their degree of individuation is stronger than the ‘need’ to discriminate between As and Ps; thus we may state that higher-ranked PDP-driven constraints (*PERFA/∅ and *INDIVP/∅) are again ranked above the SDP, but not with respect to each other; since only individuated Ps and As in the perfective are marked in these languages, *STRUC is ranked above the lower-ranked *IMPERFA/∅ and *NONINDIVP/∅ see (35): (35) *PERFA/∅, *INDIVP/∅ >> *STRUC >> SDP >> *IMPERFA/∅, *NONINDIVP/∅ Let us first see how these constraints account for the most ‘unmarked’ transitive clauses, where the aspect is imperfective and the Patientive argument is nonindividuated; since there are only two cases in these languages, the possible set of candidates is straightforwardly derivable. See Tableau 7. Tableau 7. ‘Unmarked’ transitive clause in Vafsi
Note that *PERFA/∅ and *INDIVP/∅ are of no relevance here, and neither is SDP, which is satisfied due to the non-individuated character of the P, which is thus distinguishable from the A on semantic/pragmatic grounds (let us assume for the time being that A in the inputs is invariably high in animacy). The candidates with one or both arguments marked are ruled out by *STRUC; the resulting pattern may be regarded as an instance of the ‘emergence of the unmarked’ structure (McCarthy and Prince 1994). Let us now look at the most complex case, where both PDP-driven rules operate, viz. clauses with both perfective aspect and individuated Patient, see Tableau 8. Tableau 8. Perfective transitive clause with individuated Ps in Vafsi
166
PETER M. ARKADIEV
From Tableau 8 it is clearly seen that the optimal candidate, viz. the ‘double oblique’ pattern of case marking is the sole candidate which does not violate any of the highest-ranked PDP-driven constraints; other candidates, most importantly those which satisfy SDP, are ruled out either by *PERFA/∅ or by *INDIVP/∅. If we now turn to Hindi/Urdu, we find that the constraint ranking in (35) applies to this language as well; the only difference between Hindi and the Iranian languages lies in the realm of the input, where the case inventory is registered. See tableaux 9 and 10. Tableau 9. ‘Unmarked’ transitive clause in Hindi/Urdu
Tableau 10. Perfective transitive clause with individuated Ps in Hindi/Urdu
Optimal output candidates for the most marked input in Vafsi and Hindi differ only with respect to SDP: double-oblique pattern in Vafsi violates the low-ranked SDP, while tripartite pattern in Hindi satisfies it (which is marked by in the tableau). Nevertheless, both these case marking patterns, although superficially different, are clearly motivated by identical functional constraints. Evidently, PDP-driven constraints are not invariably ranked higher than SDP. First of all, instances when these principles may come into conflict are not so widespread, and are probably limited to languages with relatively poor case systems, such as modern Iranian. Second, while in Vafsi, Southern Tati and Roshani two PDP-driven rules straightforwardly apply in all suitable contexts, it is also possible for one of such rules to be suspended in favor of the SDP; that is, in many languages PDP-driven rules operate only in those cases where this does not lead to violations of SDP. Let us look at some examples from another Iranian language with a two-term case system, viz. Zaza. Here also both the tense/aspect split ergativity and animacydriven differential object marking are present, but the latter applies only to non-past tenses, thus not creating the notorious quasi-neutral patterns. Consider the following examples from Selcan (1998:277-279):
DIFFERENTIAL ARGUMENT MARKING IN TWO-TERM CASE SYSTEMS
167
(36) televe kitav cên- o student(DIR) book(DIR) take- PRS.3SG ‘The student is taking the book.’ (37) televe malım- i vinen- o PRS.3SG student(DIR) teacher- OBL see‘The student sees the teacher.’ (38) televe- y kitav di student-OBL book(DIR) saw ‘The student saw the book.’ (39) televe- y malım (*-i) di student- OBL teacher (*OBL) saw ‘The student saw the teacher.’ As can be seen from the examples, the INDIVP rule does not operate in the past tenses in Zaza, resulting in the ‘pure ergative’ construction. This pattern may be accounted for by reranking SDP higher than *INDIVP/∅, see the ranking in (40):11 (40) *PERFA/∅ >> SDP >> *INDIVP/∅ >> *STRUC >> *IMPERFA/∅, *NONINDIVP/∅ Let us see how this ranking predicts the mirror-like case-marking patterns of (37) and (39), where the Patient is animate (that is, individuated), and only tense switches from Present to Past. The relevant evaluations are shown in Tableaux 11 and 12. Tableau 11. Present tense and individuated Patient in Zaza
Tableau 12. Past tense and individuated Patient in Zaza
11
Another possible ranking, viz. *INDIVP/∅ >> SDP >> *PERFA/∅, is, to my knowledge, not attested, at least all putative examples from the Indo-Iranian languages are not very reliable; as far as I can judge, nothing should preclude languages with such ranking.
168
PETER M. ARKADIEV
As is seen from Tableaux 11 and 12, in the present tense, when A is left unmarked, nothing prohibits the individuated P to get Obl case, thus satisfying both PDP-driven *IndivP/∅ and SDP. However, in the Past tense, when A is marked by Obl (otherwise it would have fatally violated the highest-ranked *PerfA/∅, as do candidates (a) and (b)), P cannot get Obl case due to the fatal violation of SDP. The Zaza example constitutes an important argument for my proposal that case inventories of individual languages are to be regarded as features of the input rather then as epiphenomena of the constraints and their ranking. Indeed, the independently motivated constraint ranking in (40) is perfectly compatible with a Hindi-like system with separate Accusative and Ergative. See Tableau 13 for an evaluation of ‘PseudoZaza’ past tense clauses with an individuated P, where instead of Direct and Oblique the case system includes at least three cases: Nominative, Accusative, and Ergative. Tableau 13. ‘Pseudo-Zaza’ with a three-case system
Tableau 13 clearly shows that since SDP is now irrelevant, it is the tripartite structure rather than the ergative structure that wins. Thus, if we were to derive the Zaza two-term case system merely from a constraint ranking like (40), in order to account for the fact that it is actually candidate (c), and not (d), that wins, we would need to stipulate some other economy-based constraint, which would be ranked low in Hindi and Vafsi, and high in Zaza. Such a solution is certainly always possible, but I am reluctant to overload the theory outlined here with further probably ad hoc constraints. From the preceding discussion it is clear that SDP and PDP-driven rules may be ranked differently in the grammars of particular languages, resulting in various patterns of case marking. It is a matter of empirical investigation whether the universally preferred ranking is SDP >> PDP or PDP >> SDP. If the former is the case, than the ‘discriminatory’ theory of case marking is proven to be basically true. But this would not, I believe, undermine the claim that the ‘discriminatory’ function of case marking, though it plays its role, is not the sole and not always the principal determinant of existing argument encoding structures. In any case, the universal ranking of the two basic functions of case marking is an empirical, and not an a priori, issue. 6. CONCLUSION In this article I have presented data from argument encoding variations in the languages with two-term case systems, which constitute strong evidence against the
DIFFERENTIAL ARGUMENT MARKING IN TWO-TERM CASE SYSTEMS
169
widely assumed ‘discriminatory’ conception of case marking, which states that the main factor determining the ‘global’ strategies of argument marking in the languages of the world is the Syntagmatic Discrimination Principle (SDP): A and P in a transitive clause must be somehow distinguished from one another. As the data from Indo-Iranian and Uto-Aztecan languages show, however, this simplistic view is empirically disconfirmed, since these languages tolerate severe violations of the SDP. In order to account for these facts I proposed that another factor is at work here, namely the Paradigmatic Discrimination Principle (PDP): arguments with similar semantic or grammatical functions may be encoded differently if there are strong contextual factors favoring their differentiation. Since different such factors may be grammaticalized in different languages, there are various instantiations of PDP. Also, the comparison of languages with different case systems shows that language-particular surface realization of various PDP-driven rules depends on the inventory of formal devices the language possesses. SDP and PDP may have different ranking, which results in variable distribution of particular argumentencoding strategies. It is not a priori obvious which of the two principles, both of which seem to be functionally motivated, cross-linguistically tend to be ranked higher. An Optimality Theory based framework, which incorporates both kinds of factors as universal constraints, and regards language-particular case inventories as relevant characteristics of the input, seems to be useful for the uniform description of these facts. REFERENCES Ackerman, F. and J. Moore (2001). Proto-Properties and Grammatical Encoding. A Correspondence Theory of Argument Selection. Stanford: CSLI Publications. Aissen, J. (1999). Markedness and subject choice in optimality theory. Natural Language and Linguistic Theory 17, 673-711. Aissen, J. (2003). Differential object marking: Iconicity vs. economy. Natural Language and Linguistic Theory 21, 435-483. Aissen, J. and J. Bresnan (2002). Optimality and functionality: Objections and refutations. Natural language and linguistic theory 20, 81-95. Bossong, G. (1985) Empirische Universalienforschung: Differentielle Objektmarkierung in den neuiranischen Sprachen. Tübingen: Narr. Butt, M. and T.H. King (2002a). Case systems: Beyond structural distinctions. New Perspectives on Case Theory. Ed. by E. Brandner, H. Zinmeister. Stanford (CA): CSLI Publications, 53-87. Butt, M. and T.H. King (2002b). The status of case. Unpublished manuscript, University of Konstanz. Comrie, B. (1978). Ergativity. Syntactic Typology. Studies in the Phenomenology of Language. Ed. by W.P. Lehmann. Austin: The University of Texas Press, 329-394. Comrie, B. (1979). Definite and animate direct objects: A natural class. Linguistica Silesiana, 3, 13-21. Comrie, B. (1989). Language Universals and Linguistic Typology. Oxford: Blackwell. DeLancey, S. (1981). An interpretation of split ergativity and related patterns. Language 57, 626-567. Dixon, R.M.W. (1979). Ergativity. Language 55, 59-138. Dixon, R.M.W. (1994). Ergativity. Cambridge: Cambridge University Press. Du Bois, J.W. (1987). The discourse basis of ergativity. Language 63, 805-855. Gil, D. (1994). The structure of Riau Indonesian. Nordic Journal of Linguistics 17, 179-200. Gil, D. (1999). Riau Indonesian as a pivotless language. Tipologija i teorija jazyka: Ot opisanija k objasneniju [Typology and Linguistic Theory. From Description to Explanation]. Festschrift for Alexander E. Kibrik. Ed. by E.V. Rakhilina, and J.G. Testelec. Moscow: Jazyki Russkoj Kul’tury, 187-211.
170
PETER M. ARKADIEV
Givón, T. (1984). Syntax. A Functional-Typological Introduction. Vol. I. Amsterdam/Philadelphia: John Benjamins. de Hoop, H. and B. Narasimhan (2005), Differential case-marking in Hindi. Competition and Variation in Natural Languages: The Case for Case. Ed. by M. Amberber and H. de Hoop. Amsterdam: Elsevier Publishers, 321-345. de Hoop, H and B. Narasimhan (this volume). Ergative case-marking in Hindi. de Hoop, H. and M. Lamers (2006). Incremental distinguishability of subject and object. Case, Valency, and Transitivity. Ed. by L.I. Kulikov, A.L. Malchukov and P. de Swart. Amsterdam/Philadelphia: John Benjamins, 269-287. Keenan, E.L. (1985). Relative clauses. Language Typology and Syntactic Description. Vol. II. Complex Constructions. Ed. by T. Shopen. Cambridge: Cambridge University Press, 141-170. Kibrik, A.E. (1979). Canonical ergativity and Daghestan languages. Ergativity: Towards a Theory of Grammatical Relations. Ed. by F. Plank. London: Academic Press, 61-78. Kibrik, A.E. (1997). Beyond subject and object: Toward a comprehensive relational typology. Linguistic Typology 1, 279-346. Lazard, G. (1984), Actance variations and categories of the object. Objects: Towards a Theory of Grammatical Relations. Ed. by F. Plank. London: Academic Press, 269-292. Lazard, G. (1994). L’actance. Paris: Presses Universitaires de France. Lee, H. (2003). Parallel optimization in case systems. Nominals: Inside and Out. Ed. by M. Butt and T.H. King. Stanford: CSLI Publications, 15-58. Legendre, G., W. Raynold and P. Smolensky (1993). An Optimality-Theoretic typology of case and grammatical voice systems. Proceedings of the 19th Annual Meeting of the Berkeley Linguistics Society. 464-478. Lehmann, C. (1988). Towards a typology of clause linkage. Clause Combining in Grammar and Discourse. Ed. by J. Haiman and S.A. Thompson, Amsterdam/Philadelphia: John Benjamins, 181226. Lehmann, C. (1995). Thoughts on Grammaticalization. LINCOM Europa, München and Newcastle. (Appeared as ms. in 1982) Lindenfeld, J. (1973). Yaqui Syntax. Berkeley, University of California Press. Malchukov, A.L. (2005), Case pattern splits, verb types and construction competition. Competition and Variation in Natural Languages: The Case for Case. Ed. by M. Amberber and H. de Hoop. Amsterdam: Elsevier Publishers, 73-118. Malchukov, A.L. (2006), Transitivity parameters and transitivity alternations: Considering co-variation. Case, Valency and Transitivity. Ed. by L.I. Kulikov, A.L. Malchukov, P. de Swart. Amsterdam/Philadelphia: John Benjamins, 329-357. Mallinson, G. and B.J. Blake (1981). Language Typology. Cross-linguistic Studies in Syntax. Amsterdam, North Holland. Masica, C. (1991). The Indo-Aryan Languages. Cambridge: Cambridge University Press. McCarthy, J. and A. Prince (1994). The emergence of the unmarked. Optimality in Prosodic Morphology. Proceedings of NELS-24. 333-379. Mohanan, T. (1994). Argument Structure in Hindi. Stanford: CSLI Publications. Moravcsik, E.A. (1978a). On the distribution of ergative and accusative patterns. Lingua 45, 233-279. Moravcsik, E.A. (1978b). On the limits of subject-object ambiguity tolerance. Papers in Linguistics 11, 255-259. Nichols, J. (1992). Linguistic Diversity in Space and Time. Chicago/London: The University of Chicago Press. Payne, J.R. (1979). Transitivity and intransitivity in the Iranian languages of the U.S.S.R. The Elements: A Parasession on Linguistic Units and Levels, Including Papers from the Conference on Non-Slavic Languages of the USSR (The 15th Annual Meeting of the Chicago Linguistic Society). Ed. by P.R. Clyne, W.F. Hanks and C.L. Hofbauer. Chicago, 436-447. Payne, J.R. (1980). The decay of ergativity in Pamir languages. Lingua 51, 147-186. Payne, J.R. (1989). Pāmir languages. Compendium Linguarum Iranicarum. Ed. by R. Schmitt. Wiesbaden: Reichert, 417-444. Plank, F. (1980). Encoding grammatical relations: Acceptable and unacceptable non-distinctness. Historical morphology. Ed. by J. Fisiak. The Hague: Mouton, 289-325. Press, M.L. (1979). Chemehuevi: A Grammar and Lexicon. Berkeley: University of California Press. Primus, B. (1999). Cases and Thematic Roles. Ergative, Accusative and Active. Tübingen: Niemeyer.
DIFFERENTIAL ARGUMENT MARKING IN TWO-TERM CASE SYSTEMS
171
Primus, B. (2003). Proto-roles and case selection in Optimality Theory. Arbeiten des SFB 282 “Theorie des Lexikons” 122. Selcan, Z. (1998). Grammatik der Zaza-Sprache. Nord-Dialekt (Dersim-Dialekt). Berlin: Wissenschaft und Technik. Silverstein, M. (1976). Hierarchy of features and ergativity. Grammatical Categories in Australian Languages. Ed. by R.M.W. Dixon. Canberra: Australian Institute for Aboriginal Studies, 112-171. Skalmowski, W. (1974). Transitive verb constructions in the Pamir and Dardic languages. Studia Indoeuropejskie, Polska Akademia Nauk - Oddział w Krakowie. Prace Komisji Językoznawstwa 37, 205-212. Song, J. (2001). Linguistic Typology. Morphology and Syntax. London: Longman. Stilo, D.L. (2004). Vafsi Folk Tales. Wiesbaden: Reichert. de Swart, P. (2003). The Case Mirror. MA Thesis, Radboud University, Nijmegen. Testelec, J.G. (2003). Grammaticheskie ierarxii i tipologija predlozhenija. [Grammatical Hierarchies and the Typology of the Clause] Doctoral dissertation. Russian State University of Humanities, Moscow. Tsunoda, T. (1981). Split case-marking patterns in verb-types and tense/aspect/mood. Linguistics 19, 389438. Wierzbicka, A. 1980). The Case for Surface Case. Ann Arbor: Karoma. Wierzbicka, A. (1981). Case marking and human nature. Australian Journal of Linguistics 1, 43-80. Wierzbicka, A. (1983). The semantics of case marking. Studies in Language 7, 247-275. Wierzbicka, A. (1988). The Semantics of Grammar. Amsterdam/Philadelphia, John Benjamins. Woolford, E. (2001). Case patterns. Optimality-theoretic Syntax. Ed. by G. Legendre, J. Grimshaw, S. Vikner. Cambridge (MA): MIT Press, 509-543. Wunderlich, D. and R. Lakämper (2001). On the interaction of structural and semantic case. Lingua 111, Special Issue on the Effects of Morphological Case. Ed. by Helen de Hoop et al., 377-418. Yar-Shater, E. (1969). A Grammar of Southern Tati Dialects. The Hague/Paris, Mouton.
DMITRY GANENKOV, TIMUR MAISAK AND SOLMAZ MERDANOVA
CHAPTER 8 Non-canonical Agent Marking in Agul1
1. INTRODUCTION2 This paper investigates the means of Agent encoding in Agul, a language from the Lezgic branch of the East Caucasian (Nakh-Daghestanian) family. Agul is spoken by approximately 30 thousand speakers in the Agul and Kurah districts of Southern Daghestan, Russia. Our study is based on the data from the ʜup̅uq’ dialect (spoken by some 600 speakers). Agul is an ergative language with predominantly agglutinative morphology and SOV basic word order. It is a clear example of a ‘role-dominated’ language in terms of Foley & Van Valin (1984): the marking of arguments is semantically motivated, and there are no syntactic mechanisms of changing this marking by means of any voice-like operations like the passive or the antipassive. The Agent of a transitive verb (A) is marked by the Ergative case, the core argument of an intransitive verb (S) and the Patient of a transitive verb (O) are marked by the Absolutive: (1) ze dad mask̅aw.di-as χab aldarku-naa Moscow-IN.ELAT back return-RES my father.ABS ‘My father has come back from Moscow.’ (2) dad.a guni ʕut’u-ne father.ERG bread.ABS eat-PST ‘Father ate bread.’ The Ergative and the Absolutive can be regarded as ‘canonical’ means of encoding core arguments in the two major clause types. There are, however, some 1
We thank Michael Daniel, Alice C. Harris, Martin Haspelmath, Seppo Kittilä, Yury Lander, Wolfgang Schulze, and two anonymous reviewers for their valuable comments on earlier versions of the paper, as well as for correcting our English. We are also grateful to all participants of the PIONIER workshop ‘Differential subject marking’ (July 2004, Nijmegen) for fruitful discussion. 2 Abbreviations: AD = location near a landmark, ELAT = motion from a landmark, IN = location inside a landmark, LAT = motion towards a landmark, POST = location behind a landmark, SUB/CONT = location under a landmark / in tight contact with a landmark, SUPER = location on a landmark. Singular and essive forms of nouns are not indicated (being formally unmarked).
173 H. de Hoop and P. de Swart (eds.), Differential Subject Marking, 173–198. © 2008 Springer.
174
DMITRY GANENKOV, TIMUR MAISAK, SOLMAZ MERDANOVA
minor clause types in Agul where other means of marking core arguments are used. As the Absolutive argument is normally present in such clauses, they would be probably treated as ‘extended intransitive’ in R.M.W. Dixon’s approach (see e.g. Dixon and Aikhenvald 2000:3). All other case forms that encode core arguments can be conventionally labelled ‘non-canonical’ means of argument marking. First of all, it is the Dative that marks Experiencers with a number of sentience verbs like ag ̥as ‘see’, unxas ‘hear’, ʜaa ‘know’, k̅andea ‘love, want, need’ and others, for example: (3) za-s we ruš agu-ne I-DAT your daughter.ABS see-PST ‘I saw your daughter.’ The Possessor is expressed by one of the two locative cases3 – the Ad-essive (originally referring to location near a landmark) and the Post-essive (referring to location behind a landmark), which distinguish between actual and permanent possession respectively: (4) za-w nis=na guni fa-a I-AD cheese.ABS=and bread.ABS AD.be-PRS ‘I have cheese and bread with me.’ (So, we can take a snack now) (5) za-q ʡu ruš=na sa gada qa-a I-POST two daughter.ABS=and one son.ABS POST.be-PRS ‘I have two daughters and one son.’ In fact, many other locative cases can be used as non-canonical means of core argument marking as well: in (6), for example, the Super-essive (originally referring to location on the upper surface of a landmark) is used to encode the participant of the situation ‘fall asleep’, expressed by the lexicalized compound aχun alč aʕ˳as: (6) šünük̅.i-l aχun alčaj-ne child-SUPER fall asleep-PST ‘The child fell asleep.’ Note that in many cases the use of a locative form is correlated with the presence of a certain prefix on the verb (verbal prefixes in Lezgic languages usually have the same origin as locative case markers): cf. above za-w f-aa ‘I have (with me)’, za-q qaa ‘I have (permanently)’, šünük̅i-l al-čajne ‘child fell asleep’. It means that the use of locative cases for encoding of core arguments stems from the conceptual 3
Like the majority of the Lezgic languages, Agul possesses a rich system of locative cases. Locative case markers consist of two parts, the first one specifies the localization of a trajector with respect to a landmark (‘inside’, ‘above’, ‘below’, ‘near’, ‘behind’, etc.), the second one points at the direction of movement (‘to’ vs. ‘from’) or absence of movement (‘at rest’, zero marked). In total, there are 25 case forms in Agul (Ergative, Absolutive, Dative, Comitative and Genitive, and 20 locative forms).
NON-CANONICAL AGENT MARKING IN AGUL
175
reinterpretation of the original locative construction (thus, in example (4) temporary possession is represented as location near the Possessor as the landmark).4 However, there are locative cases which have undergone a considerable extension of meaning and do not seem to show clear traces of original locative semantics. One can say that these cases are on their way from purely locative to ‘grammatical’, or ‘syntactic’ (like the Ergative or the Dative), using the terminology common to the East Caucasian studies. Among them, it is probably the Ad-elative that shows the widest distribution in the domain of non-locative arguments marking (its locative meaning being ‘motion from location near a landmark’). Thus, in (7) below the Ad-elative NP encodes the participant who is involved in the situation unintentionally, in (8) it marks the participant who takes part in the action deliberately, but mistakenly, in (9) it introduces the participant who is capable of doing something, and in (10) it expresses the Causee (in the latter example, the Adessive case is also possible):5 (7) ruš.a-f-as berʜem kura-se girl-AD-ELAT dress.ABS get dirty-FUT ‘The girl will unwittingly soil the dress.’ (the little girl was told that she has to be careful, but she cannot remember about that all the time, and she will most probably soil the new dress while playing) (8) za-f-as gi-s unaq’u-b xu-ne I-AD-ELAT that-DAT call-MSD become-PST ‘It so happened that I had to invite him.’ (I did not plan to do this, but when I was inviting other people, he was near, and it would have been impolite not to invite him) (9) za-f-as k’eǯ lik’a-s xu-ne I-AD-ELAT letter.ABS write-INF become-PST ‘I managed to write a letter.’ (10) baw.a ruš.a-f-as || ruš.a-w xed χa-s q’u-ne girl-AD water.ABS bring-INF do-PST mother.ERG girl-AD-ELAT ‘Mother made the girl bring water.’ Note that it is not the case that the Ad-elative encodes a particular participant according to the subcategorization frame of a given verb. Rather, the Ad-elative argument appears in particular constructions, having their own semantics and their own lexical restrictions. Each of the four constructions illustrated in (7)-(10) will be discussed below in more detail: the Involuntary Agent Construction in section 2, the Undesirable Action Construction in section 3, the Possibilitive Construction in section 4, and the Causative Construction in section 5.
4
See also Schulze (to appear) for the discussion of the metaphorical potential of locative forms in Lezgic languages (especially Agul and Udi). 5 Locative markers -w (in the Ad-essive) and -f (in the Ad-elative and in verbal prefix) are in complementary distribution and go back to one and the same historical source.
176
DMITRY GANENKOV, TIMUR MAISAK, SOLMAZ MERDANOVA
The fact that one and the same formal means, viz. the Ad-elative case, is employed in the constructions listed above does not seem to us accidental. We argue that the Ad-elative in Agul covers a coherent semantic domain, as the participants marked by this case are all agentive in nature. At the same time they are not real (‘canonical’) Agents, which are encoded by the Ergative. In contexts like those presented in (7)-(10) the participants in question lack some of the characteristic agentive features, or Proto-Agent entailments, using the terminology of Dowty (1991). Thus, in a sense, the Ad-Elative should be treated as a canonical means of ‘non-canonical Agent’ marking in Agul. It can be readily observed that in all examples given above (apart from (10)) the non-canonically marked NP corresponds to grammatical subject in the English translation (the same would be true if the Russian translations were given). However, in the overview given above we prefer to speak about (canonical or noncanonical) ‘core arguments marking’, and not about ‘subject marking’. The reason for this is that the notion of grammatical subject as such (or pivot, in Dixon’s terms) is problematic in Agul. Being a typical role-dominated Daghestanian language, Agul tends to encode clause arguments consistently with their semantic roles, and it is possible that the very notion of syntactic relations is not applicable to languages of this type (as argued, e.g., in Kibrik 1997, among other works). Nevertheless it seems important to understand whether it is possible to treat the Ad-elative argument as a grammatical subject on the basis of its syntactic behaviour (the alternative may be that it is a sort of ‘external’ clause argument, like the agentive adjunct in passives). In section 6 below we compare syntactic properties of the Ad-elative argument with that of ‘canonical’ Ergative and Absolutive NPs (in A, S and O functions). The conclusion that can be drawn from the comparison is that the Ad-elative argument (especially expressing the Involuntary Agent) does not show different behaviour from the most ‘subject-like’ NPs in Agul. Finally, the general discussion in section 7 concludes the paper. 2. THE INVOLUNTARY AGENT CONSTRUCTION The Involuntary Agent Construction has been previously described for the Lezgian language by Haspelmath (1993:291-293) as the construction “in which the Agent is in the Adelative case and the additional meaning is ‘involuntarily, unwittingly, or in a very indirect manner.’” Such a construction is attested in virtually all East Caucasian languages, including Lezgic, Avar-Andic, Tsezic and Lak. The only work dealing with the Involuntary Agent Construction in a typological perspective is Kittilä (2005). Semantically, the Involuntary Agent is an Agent-like participant, but in contrast to the Agent in standard transitive clauses it exhibits a very low degree of control and volitionality. Morphologically, the Involuntary Agent is usually expressed in East Caucasian languages by means of a locative case, which denotes motion from a
NON-CANONICAL AGENT MARKING IN AGUL
177
landmark.6 The same situation is found in Agul, where the actor of the Involuntary Agent Construction is encoded by the Ad-elative case, which in its spatial use describes motion from location near a landmark, cf. (11): (11) cil.i-f-as hat-̅ u čuwal! wall-AD-ELAT take away-IMP sack.ABS ‘Take away the sack from the wall!’ The following examples illustrate a canonical transitive clause in Agul (12) and the Involuntary Agent Construction with the same verb (13): (12) baw.a nek̅ at̅uzu-ne mother.ERG milk.ABS pour out-PST ‘Mother poured out the milk.’ (13) baw.a-f-as nek̅ at̅uzu-ne mother-AD-ELAT milk.ABS pour out-PST ‘Mother accidentally spilled the milk.’ The general interpretation of the Involuntary Agent Construction ‘to do something accidentally, unintentionally’ can come out in three different variations: (i) The participant affects the Patient accidentally, without noticing what s/he is doing, (ii) The participant involuntarily lets something happen by overlooking and not making enough efforts to prevent the situation, (iii) The participant finally (as a result of efforts) succeeds in doing something, although it is not quite expected. These three readings of the Involuntary Agent Construction are illustrated in the following example: (14) ruš.a-f-as rak̅ daqu-ne girl-AD-ELAT door.ABS open-PST a. The girl accidentally opened the door (because she pushed it with her elbow while playing with her toys on the floor). b. (Father told the girl to hold the door so that the wind could not open it, but her efforts were not enough) The girl accidentally opened the door // let the door open. c. (All the children tried but no one could open the tightly closed door, however it so happened that) The girl managed to open the door. What the three interpretations have in common is that in all of them the actor is not in complete control of the situation, but the lack of control affects different parts of events: in case of the first two readings it is obviously the initiation of an event, 6
Only Lak and Bagwalal possess distinct case markers whose main function is to mark the Involuntary Agent.
178
DMITRY GANENKOV, TIMUR MAISAK, SOLMAZ MERDANOVA
which is out of control, while in the third reading it is the completion of an event. The difference between the first two readings concerns a degree to which an instigator’s attention is focused on the event instigated. According to the first reading, the Involuntary Agent is a participant who acts fully unintentionally (s/he even may be not aware of the situation taking place), while in the second reading the focus is on the fact that the participant tries to prevent the initiation of an event, but fails to do this. Note that the last reading stands slightly apart from the first and the second, since in this case the action is not involuntary in the strict sense; on the contrary, the participant’s will and efforts are aimed at achieving the result. However, the polysemy found in Agul seems to be quite widespread cross-linguistically. Kittilä (2005) reports that it is also characteristic of similar constructions in a number of languages including, for example, Bagwalal, Finnish and Thompson River Salish. There are quite strong restrictions on the lexical verb and on the Adelative argument in the Involuntary Agent Construction. First of all, not every verb can appear in this construction. The set of verbs that allow the Involuntary Agent includes: (i) intransitive verbs denoting a change of state, cf. Table 1 and example (15), and (ii) labile verbs (that is, S=O ambitransitives), cf. Table 2 and examples (13)(14) above. (15) a. kitab gulu-ne book.ABS get lost-PST ‘The book got lost.’ b. gada.ji-f-as kitab gulu-ne boy-AD-ELAT book.ABS get lost-PST ‘The boy lost the book.’ c. *gada.ji kitab gulu-ne boy.ERG book.ABS get lost-PST ‘The boy lost the book.’ Table 1. Examples of intransitive change-of-state verbs in Agul
ac’as alaʕ ḁ s gulas ket’as kuras q’ešas ruq̅as ruʁas rutas t’aʜas
‘fill (intr)’ ‘boil over (about milk)’ ‘get lost’ ‘awake’ ‘get dirty’ ‘get wet’ ‘get dry’ ‘get cold’ ‘curdle (about milk)’ ‘swell’
NON-CANONICAL AGENT MARKING IN AGUL
179
Table 2. Examples of labile verbs in Agul
arʕas
‘break (tr, intr)’
atu̅ zas at’usas daqas čurqas
‘spill (tr, intr)’ ‘stop burning (tr, intr)’ ‘open (tr, intr)’ ‘explode (tr, intr)’
č’urχas
‘tear apart (tr, intr)’
č’ut’as k’es rüxes uc’as ugas
‘bend (tr, intr)’ ‘kill/die’ ‘boil, cook (tr, intr)’ ‘melt (tr, intr)’ ‘burn (tr, intr)’
On the contrary, ordinary transitive and non-change-of-state intransitive verbs cannot appear in the Involuntary Agent Construction, cf. (16) and (17): Transitive verb (16) a. ruš.a k’eǯ lik’i-ne girl.ERG letter.ABS write-PST ‘The girl wrote a letter.’ b. *ruš.a-f-as k’eǯ lik’i-ne girl-AD-ELAT letter.ABS write-PST ‘The girl accidentally wrote a letter.’ Intransitive non-change-of-state verb (17) a. ruš χul.a-s ušu-ne girl home-DAT go-PST ‘The girl went home.’ b. *ruš.a-f-as χul.a-s ušu-ne girl-AD-ELAT home-DAT go-PST ‘The girl accidentally went home.’ These examples show that the alternation between the Ergative and the Adelative marking is possible only for labile verbs, cf. at̅uzas ‘pour out’ with a transitive valence pattern in (12) and an intransitive pattern with the Involuntary Agent in (13). In fact, this also shows that the Involuntary Agent does not just ‘replace’ the canonically marked Agent, but instead is introduced in an event as an ‘external’ instigator of the whole event, described by the verb. However, even in this case the Involuntary Agent cannot appear in clauses with transitive and non-changeof-state intransitive verbs in any of the three readings, cf. examples (16c) and (17c):
180
DMITRY GANENKOV, TIMUR MAISAK, SOLMAZ MERDANOVA
(16) c. *za-f-as ruš.a k’eǯ lik’i-ne I-AD-ELAT girl.ERG letter.ABS write-PST ‘I accidentally made the girl write a letter.’ (17) c. *za-f-as ruš χul.a-s ušu-ne I-AD-ELAT girl.ABS home-DAT go-PST ‘I accidentally made the girl go home.’ At the same time, the Involuntary Agent Construction was judged possible by native speakers at least with one transitive verb – ʕut’as ‘eat’. In this case the construction is possible only with the third reading (‘manage to do’), cf. example (18): (18) za-f-as ruš.a guni ʕut’u-ne I-AD-ELAT girl.ERG bread.ABS eat-PST ‘I managed to feed the girl with bread.’ One can suppose that the acceptability of (18), unlike examples like (16c) and (17c), is due to the fact that the Involuntary Agent with the verb ʕut’as ‘eat’ is not just an instigator of the event, but is involved in the caused event as well. The Involuntary Agent is a participant that affects the Patient accidentally and unintentionally. Therefore, it is no surprise that it can be expressed only by NPs denoting human beings. Nouns denoting animals – like cats or dogs – are in principle possible, but less felicitous, as it is hard to say whether an animal does something intentionally or not. Cf. the following example: (19) ??kitan.i-f-as / *ǯaq˳ʼ.ala-f-as dak’ar arʕu-ne bird-AD-ELAT window.ABS break-PST cat-AD-ELAT ‘The cat / bird accidentally broke the window.’ In contrast to the standard transitive pattern, the Involuntary Agent Construction does not allow to add adjuncts presupposing a controlling and deliberately acting Agent, such as Instruments or Benefactives (cf. (20)) and adverbs of manner like qülias ‘for spite; to spite smb’, ʡašq̅unaldi ‘willingly, readily’, teʕdidi ‘quickly’ (cf. (21)). za-s arʕu-ne (20) gada.ji (*gada.ji-f-as) χew ʁ ̥an.di-l-di7 / boy.ERG boy-AD-ELAT nut.ABS stone-SUPER-LAT I-DAT break-PST ‘The boy broke the nut with a rock / for me.’ (21) ruš.a (*ruš.a-f-as) teʕdidi / qülias / ʡašq̅unaldi berʜem č’urχu-ne girl-ERG girl-AD-ELAT quickly / for spite / readily dress.ABS tear-PST ‘The girl quickly / for spite / readily tore the dress in pieces.’ 7
The Super-lative case is a regular means of the Instrument encoding in Agul.
NON-CANONICAL AGENT MARKING IN AGUL
181
On the other hand, an NP expressing Force (22) or external circumstances (23) can appear in the construction. Usually, in this case the construction gets the second reading – ‘the participant lets something happen by overlooking and not doing what should be done’ (cf. (22)). (22) gada.ji-f-as kurušk̅a kulak.i-q-as arʕu-ne boy-AD-ELAT mug.ABS wind-POST-ELAT break-PST ‘The boy unwittingly broke the cup because of the wind.’ (e.g. a strong wind blew suddenly, and the boy lost hold of the cup) (23) gada.ji-f-as baw gašila k’i-ne son-AD-ELAT mother.ABS of starvation kill/die-PST ‘Mother died of starvation because of her son.’ (e.g., he didn’t looked after her) All these restrictions seem to be clear evidence of low agentivity of a participant marked by the Ad-elative and prove that the behavior of the Involuntary Agent differs to a considerable extent from the behavior of the prototypical Agent. One can conclude from these differences that the Involuntary Agent Construction is a means of decreasing the degree of agentivity. In other words, it is a means of presenting the situation as occurring with the participation of an Agent, but an Agent with attenuated agentivity. Note that this construction is relevant only for prototypical Agents, which are human beings. So, it can be used in case when it is indicated explicitly that a participant can be considered as a true Agent according to his/her inherent parameters, but at the same time in this particular situation s/he lacks one of the most important features of Agents, that is the total control of the situation. Thus it turns out that the ‘Involuntary Agent’ is not, strictly speaking, an Agent as such (cf. (13) and interpretations i-ii), and it is not necessarily unintentional (cf. interpretation iii above). It is rather a particular participant with a low degree of control which can be introduced in an intransitive event. 3. THE UNDESIRABLE ACTION CONSTRUCTION Another construction which is in some respect similar to the Involuntary Agent Construction is formed with the deverbal noun (Masdar) of a verb and the auxiliary xas ‘become; happen’.8 Example (24b) illustrates this construction, which in contrast 8
The verb xas ‘become; happen’ is used as an inchoative copula, which is illustrated in (i)-(ii): (i) ze gada müʕelim xa-se my son.ABS teacher.ABS become-FUT ‘My son will become a teacher.’ (ii) če ʕümür četin xu-ne our.INCL life.ABS hard become-PST ‘Our life became hard.’ As one of the main auxiliary verbs (along with the stative copulas e ‘be’ and a ‘be located’), xas also forms a number of periphrastic tenses, e.g. the Future Imperfective and the Future Perfective – cf.
182
DMITRY GANENKOV, TIMUR MAISAK, SOLMAZ MERDANOVA
to its basic counterpart (24a) has a connotation of an undesirable event, that should have been prevented. The structure of (24b) can be translated literally as ‘the girl’s seeing of a gift happened’. (24) a. ruš.a-s p̅adark̅a agu-ne see-PST girl-DAT gift.ABS ‘The girl saw a gift.’ b. ruš.a-s p̅adark̅a agu-b xu-ne see-MSD] become-PST [girl-DAT gift.ABS ‘It so happened that the girl saw a gift.’ (although she was not supposed to do so) There are two variants of the Undesirable Action Construction, with difference in argument marking. In the first variant, the main participant is encoded by the case that is required by the lexical verb: cf. the Absolutive with the intransitive verb uʁas ‘to fall down (of rain or snow)’ in (25), the Ergative with the transitive verb ʕut’as ‘to eat’ in (26), or the Dative with the sentience verb ag˳as ‘to see’ in (24b) above. In this case the construction has an impersonal reading ‘(independenly of anyone’s will) it so happened that’, and is usually used to explain why something that had been previously planned didn’t take place: (25) Context: – Have you managed to build your new house this year? – No, we didn’t manage to do this: ixp̅-ar uʁu-b xu-ne [snow-PL.ABS fall-MSD] become-PST ‘It so happened that it snowed (so we had to stop building).’ {= lit. Falling of snow happened.} (26) Context: – Father was keeping the fast, but one day he forgot this, and dad.a guni ʕut’u-b xu-ne [father.ERG bread.ABS eat-MSD]become-PST ‘It so happened that he ate bread.’ {= lit. Father’s eating bread happened.} From the syntactic point of view, this variant of the construction can be probably represented as a one-place predicate xas ‘become; happen’ with a sentential complement (see the bracketing in examples above). There seem to be no restrictions on the semantic or syntactic class of the lexical verb and the semantic role of its main argument with this reading. The only restriction concerns the temporal reference: as the construction in question describes situations which run against the speaker’s expectations, it is most appropriate in the past tense. Present or future time reference is doubtful for the Undesirable Action Construction: ruχaj xase ‘will be reading’ (with the Imperfective converb of the main verb) and ruχuna xase ‘will have read’ (with the Perfective converb of the main verb).
NON-CANONICAL AGENT MARKING IN AGUL
183
(27) ??ruš.a-s p̅adark̅a agu-b xa-se see-MSD] become-FUT [girl-DAT gift.ABS ‘It will happen so that the girl will see a gift.’ In the second variant of the Undesirable Action Construction the core argument is marked by the Ad-elative. In this case the construction describes an action in which the participant takes part deliberately, but mistakenly. This is possible in two cases: (i) the participant does not know that s/he should not do something, cf. (28): (28) dad.a-f-as kurušk̅a arʕu-b xu-ne father-AD-ELAT mug.ABS break-MSD become-PST ‘Father (deliberately) broke the cup.’ (it turned out later that another cup should have been broken, but father did not know that) {= lit. From father breaking the cup happened.} (ii) the participant does know that s/he should not do something, but the circumstances are such that s/he has to do this, cf. (29): (29) za-f-as ušu-b xu-ne ge-wur.i-n χul.a-s I-AD-ELAT go-MSD become-PST that-PL-GEN house-DAT ‘I went to their place.’ (I knew that I should not visit them, but it so happened that I had to do this) {= From me going to their house happened.} In contrast to the first type of encoding, here the construction becomes more ‘personal’, in the sense that there is a particular participant responsible for the situation. While in the first case (cf. exx. (24)-(26)) the situation as a whole is in focus, construction with the Ad-elative case focuses on the participant who is ‘guilty’ of the situation and regrets it. The second strategy of argument marking imposes considerable restrictions both on the lexical verb and on the participant’s semantic role. First of all, only agentive verbs can appear in this variant of the construction (which is in clear contrast to the Involuntary Agent Construction, as shown above). Cf. example (30), which is ungrammatical, since the verb alurq’as ‘fall down, tumble down’ does not allow a controlled interpretation: (30) *gada.ji-f-as alurq’u-b xu-ne boy-AD-ELAT fall down-MSD become-PST ‘It so happened that the boy fell down.’ (although he was not supposed to do so) As for restrictions on the participant, only NPs denoting human beings can appear in the Undesirable Action Construction; animals are highly doubtful in this context (cf. (31)), and inanimate objects are ruled out altogether:
184
DMITRY GANENKOV, TIMUR MAISAK, SOLMAZ MERDANOVA
(31) ?? kitan.i-f-as nek̅ uχu-b xu-ne cat-AD-ELAT milk.ABS drink-MSD become-PST ‘It so happened that the cat drank the milk.’ The Ad-elative marking is possible neither for the Experiencer of sentience verbs (cf. (32)), nor for other types of non-canonically marked arguments: (32) *ruš.a-f-as p̅adark̅a agu-b xu-ne see-MSD become-PST girl-AD-ELAT gift.ABS ‘It so happened that the girl saw a gift.’ (although she was not supposed to do so) In sum, the Ad-elative argument in the Undesirable Action Construction is used to express a participant who controls the situation and can be regarded as a typical Agent according to his/her inherent properties. At the same time, its evaluation as a true Agent is attenuated as the participant in question has to act not following his/her own will, but being forced by the circumstances, or s/he does not realize the consequences of his/her action. As the examples given above show, the semantics of the Undesirable Action Construction is rather similar (albeit not identical) to that of the Involuntary Agent Construction. However, lexical restrictions on the verb in the Involuntary Agent Construction and the Undesirable Action Construction seem to be in complementary distribution: in the former case only non-agentive verbs can appear in the construction, in the latter case agentive ones. This may point at the possibility of interpreting both constructions as variants of one and the same basic construction. 9 It can be, for example, that the difference in interpretation stems from the difference in the meaning of the verb: while non-agentive change of state verbs favour the involuntary reading (like in ‘accidentally pour out the milk’), it is not available for agentive verbs with which the action is rather presented as intentional but inappropriate. Note the same effect with the use of adverbials like the English accidentally or the Russian slučajno ‘accidentally, by accident’: with verbs denoting non-controlled situations they emphasize the involuntary involvement, and the preferred reading with agentive verbs is ‘do something on purpose but mistakenly’. It is also possible for an agentive verb to have both readings, cf. the following examples from Russian: (33) a. Ja slučajno razbil etu čašku ‘I have accidentally broken this cup.’ {I did not notice it standing there.}
9
Martin Haspelmath seems to be close to this decision in his description of Lezgian data. Pointing that the Lezgian Adelative appears in the Involuntary Agent Construction, he continues with the following observation: “When the involuntary action is expressed by a transitive verb, a periphrastic construction with xun ‘become, happen’ and the Masdar has to be used” (Haspelmath 1993: 91). Unfortunately, apart from three examples no discussion is provided that could clarify this point.
NON-CANONICAL AGENT MARKING IN AGUL
185
b. Ja slučajno pročital eto pis’mo ‘I have accidentally read this letter.’ {I thought it was addressed to me.} c. Ja slučajno otkryl etu dver’ ‘I have accidentally opened this door.’ {a. I touched it with my elbow while passing by; b. I deliberately opened the door, but it turned out later that it should have been kept closed.} Still, it does not seem appropriate to treat the Involuntary Agent Construction and the Undesirable Action Construction in Agul as variants because of two reasons. Firstly, verb classes that are allowed by the two constructions are not complementary in the strict sense. It is possible to use some verbs both in the Involuntary Agent Construction and in the Undesirable Action Construction with slight difference in meaning: (34) a. za-f-as berʜem č’urχu-ne I-AD-ELAT shirt.ABS tear-PST ‘I accidentally tore the shirt.’ b. za-f-as berʜem č’urχu-b xu-ne I-AD-ELAT shirt.ABS tear-MSD become-PST ‘It so happened that I tore the shirt.’ (e.g., I needed to dress a wound and used my shirt for this) In the second place, even if the distribution of verbs in both constructions was strictly complementary, the formal relation between the two constructions is idiomatic. In particular, it is not quite clear why it is the combination of the Masdar with the verb xas ‘become; happen’ that is used for the Undesirable Action Construction (and why it is not the case, for example, that the Ad-elative argument of the Involuntary Agent Construction is extended to all types of verbs instead). Thus, is seems reasonable to keep the two constructions apart, although they both certainly belong to a wider family of ‘Agent-attenuating’ means in Agul. 4. THE POSSIBILITIVE CONSTRUCTION In a number of Lezgic languages (e.g. Lezgian, Agul, Tabassaran, or Udi) there are no modal verbs corresponding in meaning to can or may in English. Possibility is usually described by means of a special construction with the verb ‘become; happen’, in which the sentential complement is headed by the Infinitive and the main participant is encoded by one of the locative cases from the ‘elative’ series – that is, ‘he could do it’ is expressed literally as ‘it became/happened from him to do it’. In Agul it is the Ad-elative case that marks the main participant of the Possibilitive Construction. Examples (35) and (36) show that this construction can be used to express both participant-internal (‘can, be able’) and participant-external
186
DMITRY GANENKOV, TIMUR MAISAK, SOLMAZ MERDANOVA
possibility (‘may, be allowed’), using the terms introduced in Auwera & Plungian 1998.10 (35) ze gada.ji-f-as wa-s kümek aq’a-s xa-se my son-AD-ELAT [you.SG-DAT help.ABS do-INF] become-FUT ‘My son will be able to help you.’ (36) ilsan.di-f-as ̅ dua-s xa-fe-wa? allah.ti-qaj person-AD-ELAT [Allah-COMIT compete-INF] become-GENER-Q ‘Is it possible (permitted) for a human being to compete with Allah?’ The Ad-elative argument of the Possibilitive Construction is not obligatory: when it is absent the sentence is normally interpreted as agentless (‘it is possible that’): (37) a. ruš.a-f-as kitab ruχa-s xu-ne girl-AD-ELAT [book.ABS read-INF] become-PST ‘The girl managed to read the book.’ b. kitab ruχa-s xu-ne [book.ABS read-INF] become-PST ‘It became possible to read the book.’ (e.g. written in an unknown language) The main participant of the Possibilitive Construction does not have to be human (cf. acceptable sentences like ‘the cow managed to jump over a fence’, etc.); however, inanimate participants are definitely ruled out. Thus, there is no alternation of argument marking in the Possibilitive Construction: if present, the main participant is invariably marked by the Ad-elative. There is a formal similarity between the Possibilitive Construction and the Involuntary Agent Construction in that the low agentive animate participant can be omitted from the sentence. However, in this case the situation is not presented as happening autonomously (without any Agent), but is interpreted impersonally – ‘it is possible for someone/anyone to fulfill an action’. 5. THE CAUSATIVE CONSTRUCTION We now pass to another use of the Ad-elative (and the Ad-essive)11 which is a bit different from those described so far. At the same time, the semantic contrast which 10
Epistemic possibility (‘the proposition is judged by the speaker to be probable’) cannot be expressed by means of this construction in Agul; in this case the Future tense of the verb must be used. 11 The Ad-essive differs from the Ad-elative in that the former describes the location near a landmark (sometimes the motion toward the goal), not the motion from it. An example of the Ad-essive marking of core non-locative argument was given in (4) above, where it encodes the temporary Possessor. Apart from the temporary Possessor and the Causee, there seem to be no other non-locative uses of this case.
NON-CANONICAL AGENT MARKING IN AGUL
187
is associated with the Ad-elative/Ad-essive in the Causative Construction also deals with the degree of agentivity and provides further evidence for treating the Adelative as a specialized ‘Agent-attenuating’ device. The only productive causativization pattern in Agul is a construction with the auxiliary verb (a)q’as ‘do, make’ and the Infinitive of the main verb. Semantically, ‘do’-causatives normally express indirect, or ‘distant’ causation (a more detailed description is provided in Daniel et al. 2004). The Causer is invariably marked by the Ergative; as for the Causee, there are two strategies for its encoding. According to the first one, the actor keeps the case marking assigned by the main predicate; this is illustrated in exx. (38) through (40), where sentences in (b) are periphrastic causatives of sentences in (a): Intransitive verb (38) a. kitan hiši-ne cat.ABS run away-PST ‘The cat ran away.’ b. baw.a kitan hiša-s q’u-ne mother.ERG [cat.ABS ran away-INF] do-PST ‘Mother made the cat run away.’ (e.g. she made a noise and the cat woke up) Sentience verb (39) a. dad.a-s wuri unxu-ne father-DAT everything.ABS hear-PST ‘Father heard everything.’ b. baw.a dad.a-s wuri unx.a-s q’u-ne mother.ERG [father-DAT everything.ABS hear-INF] do-PST ‘Mother made father hear everything.’ Transitive verb (40) a. gada.ji šurpa ʕut’u-ne boy.ERG soup.ABS eat-PST ‘The boy ate the soup.’ b. baw.a gada.ji šurpa ʕut’a-s q’u-ne mother.ERG [boy.ERG soup.ABS eat-INF] do-PST ‘Mother made the boy eat the soup.’ (e.g. she threatened not to let him go for a walk) There is, however, a possibility of encoding the Causee by means of the Adessive and the Ad-elative cases. Both of them can mark the Causee when an agentive intransitive or a transitive verb appear in the Causative Construction (there seem to be no evident semantic difference between the two cases here): Intransitive verb (41) baw.a gada.ji-w // gada.ji-f-as hiša-s q’u-ne boy-AD-ELAT run away-INF do-PST mother.ERG boy-AD ‘Mother made the boy run away.’
188
DMITRY GANENKOV, TIMUR MAISAK, SOLMAZ MERDANOVA
Transitive verb (42) baw.a gada.ji-w // gada.ji-f-as šurpa boy-AD-ELAT soup.ABS mother.ERG boy-AD ‘Mother made the boy eat the soup.’
ʕut’a-s q’u-ne eat-INF do-PST
The second strategy of the Causee marking is not possible for verbs with noncanonical marking of core arguments – e.g. for sentience verbs with the Experiencer in the Dative: (43) baw.a *dad.a-w // *dad.a-f-as wuri mother.ERG father-AD father-AD-ELAT everything.ABS ‘Mother made father hear everything.’
unx.a-s q’u-ne hear-INF do-PSG
It is also not allowed for verbs denoting uncontrolled situations, especially when the Causee is not human. While the first, ‘neutral’ strategy is perfectly available here, the Ad-essive and the Ad-elative are impossible or at best questionable: (44) dad.a ruš // ??ruš.a-w // *ruš.a-f-as it̅arxa-s daughter-AD daughter-AD-ELAT fall ill-INF father.ERG daughter.ABS q’u-ne do-PST ‘Father made the daughter fall ill.’ (45) gada.ji dar // *dar.ala-w // *dar.ala-f-as adarxa-s q’u-ne tree-AD-ELAT fall-INF do-PST boy.ERG tree.ABS tree-AD ‘The boy made the tree fall down.’ The use of the Ad-elative is even more restricted than that of the Ad-essive, as only the latter is allowed when the Causee is an animal (cf. (41) above): (46) baw.a kitan.i-w // *kitan.i-f-as hiša-s q’u-ne cat-AD-ELAT run away-INF do-PST mother.ERG cat-AD ‘Mother made the cat run away.’ With verbs that allow both types of the Causee encoding, the following semantic contrast can be observed: while the original Ergative or Absolutive marking leaves the opportunity for the autonomous acting of the Causee (making even the permissive reading possible), the choice of the Ad-essive/Ad-elative clearly reduces his/her independence and underlines the dominant role of the Causer. Cf. the following examples, where in sentences under (b) the causee displays less control of the situation than those under (a) (for short, only the Ad-essive variants are given):
NON-CANONICAL AGENT MARKING IN AGUL
189
Transitive verb (47) gi ze šünük̅-ar.i wak̅.a-n jak̅ ʕut’a-s q’u-ne that.ERG my children-PL.ERG pig-GEN meat.ABS eat-INF do-PST a. ‘S/he let my children eat pork.’ (e.g. s/he forgot that they were not supposed to do that and cooked pork for them) b. ‘S/he permitted my children to eat pork.’ (48) gi ze šünük̅-ar.i-w wak̅.a-n jak̅ ʕut’a-s q’u-ne that.ERG my children-PL-AD pig-GEN meat.ABS eat-INF do-PST ‘S/he made my children eat pork’ (although they did not want to) Intransitive verb (49) dad.a ruš raʁ.una-k ʁut’a-s q’.u-ne father.ERG daughter.ABS sun-SUB/CONT stand-INF do-PST a. ‘Father made daughter stand under the sun.’ (e.g. he had forgot to leave her the key and she could not enter the house) b. ‘Father permitted daughter to stand under the sun.’ (50) dad.a ruš.a-w raʁ.una-k ʁut’a-s q’.u-ne father.ERG daughter-AD sun-SUB/CONT stand-INF do-PST ‘Father put daughter under the sun.’ (e.g. as a punishment) The two strategies of the Causee marking (one keeping original encoding of the argument, the other changing its case to the Ad-essive/Ad-elative) differ with respect to the preferences of their distribution among types of verbs. Transitive verbs prefer the Ad-essive/Ad-elative marking, exemplified in (42) above,12 while intransitives prefer original encoding, as shown in (38b). This may have a syntactic explanation: the occurrence of two Ergative NPs in sentences like (40b) clearly shows that we deal with two transitive clauses here, and the Ad-essive/Ad-elative marking of the Causee reflects a natural change of the Causative Construction from biclausal to monoclausal structure (like in (42)). However, a semantic explanation seems to us more illuminating: as exx. (44)(46) show, the Ad-essive/Ad-elative marking is dispreferred when the Causee shows low agentivity. The Ergative arguments of transitive verbs in Agul are always (real) Agents, so it is quite appropriate to mark them by the Ad-essive/Ad-elative in the Causative Construction. On the contrary, only those intransitives allow this strategy, whose participants are highly agentive (which is not the case in (44) and (45)). The 12
The Ad-essive marking is possible even for the inanimate Causee. Transitive verbs do not normally allow inanimate Ergative NPs in Agul. However, we are aware of one example of such kind: cf. (i), where the same metaphorical conceptualization of ‘catching fire’ is used, as in English: (i) a. k’ur-ar.i c’aj facu-ne firewood-PL.ERG fire.ABS catch-PST ‘The firewood caught fire.’ b. baw.a lat’u-na, naft̅ mother.ERG kerosene.ABS pour-CONV k’ur-ar.i // k’ur-ar.i-w c’aj faca-s q’u-ne [firewood-PL.ERG firewood-PL-AD fire.ABS catch-INF] do-PST ‘Having poured the kerosene, mother made the firewood catch fire.’
190
DMITRY GANENKOV, TIMUR MAISAK, SOLMAZ MERDANOVA
impossibility of the Ad-essive/Ad-elative encoding with sentience verbs also favors this explanation. Summarizing, the Ad-essive and the Ad-elative cases can both mark the Causee in the Causative Construction; however, this is possible only in cases when the Causee is highly agentive. The Ad-elative seems to be associated with even higher degree of agentivity, as hinted by native speaker judgments in examples like (44) and (46). Thus, in the case of the Causative Construction the use of the Ad-essive/Adelative shows that we deal with an inherently agentive participant. At the same time, the choice of this type of marking (and not the original Ergative or Absolutive encoding) points at the agentivity decrease, as the role of the Causee in the situation becomes more subordinate. 6. SYNTACTIC STATUS OF AD-ELATIVE NPS MARKING AGENTS As a typical role-dominated language, Agul tends to encode clause arguments consistently with their semantic roles. So, by and large the marking of core arguments is clearly motivated by their semantics. It is nevertheless interesting to find out whether the Involuntary Agent and other Ad-elative NPs are true syntactic subjects (like, e.g., A marked by the Ergative in transitive clauses), or not. However, many of the traditional syntactic tests for subjecthood (for a list of the tests see, e.g., Onishi 2001) do not apply in Agul. For example, the criterion of antecedent control over reflexive pronouns and relativization do not show even the difference between A/S and O in Agul, and thus cannot shed light on the status of the Ad-elative argument. Some other criteria are applicable only to the verb xas ‘become; happen’, whose core argument is invariably marked by Ad-elative in the Possibilitive Construction (see section 4 above). Following Onishi 2001, we divide subject properties into coding properties and behavioral properties. The former include case marking, constituent word order of core arguments and verbal agreement, while the latter include various syntactic phenomena, such as the ability to control reflexivization, deletion in second conjuncts and controlled infinitives, etc. Now let us look at these tests in more detail. 6.1 Coding properties 6.1.1 Case marking The Ad-elative is a morphologically peripheral case, so it is not expected to code the grammatical subject. But it is not obvious whether either of the two core cases that canonically mark S, A and O (i.e. the Absolutive and the Ergative) always identify the grammatical subject. It is noteworthy that the Absolutive argument is almost always present in canonical clauses (both transitive and intransitive). This might suggest that the Absolutive is a case of grammatical subject. However, even this criterion fails, since
NON-CANONICAL AGENT MARKING IN AGUL
191
there are a number of verbs that do not select for arguments in the Absolutive at all. These non-canonical valence patterns are mainly exhibited by sentience predicates, cf. the following examples: (51) za-s mik’eldi-a I-DAT be cold-PRS ‘I am cold.’ ̅ u-s ̅ (52) ze čuč bengiš-ar.i-q-as
guč’a-adawa my brother-DAT bear-PL-POST-ELAT be afraid-PRS.NEG ‘My brother is not afraid of bears.’
Another piece of evidence comes from the valence pattern of the verb jar as. This verb has an Absolutive argument which denotes the Instrument, as example (53) demonstrates: (53) dad.a degi-s ʁ ̥an jarʜu-ne father.ERG donkey-DAT stone.ABS hit-PST ‘Father hit the donkey with a rock.’ However, this Absolutive argument can be freely omitted (just as the Instrumental NPs in English or Russian) and the following pattern seems to be statistically predominant: (54) dad.a degi-s father.ERG donkey-DAT ‘Father hit the donkey.’
jarʜu-ne hit-PST
All these examples show that the Absolutive cannot be considered as the case of the grammatical subject nor even as an obligatory argument. Thus, in general, we can say that there is no such thing as ‘the case of the grammatical subject’ in Agul, so this criterion does not help us. 6.1.2 Constituent word order of core NPs The Ad-elative NP expressing the Involuntary Agent in most neutral contexts precedes other NPs and the verb. The same is true about other arguments corresponding to subjects in English, such as the Ergative expressing A, the Absolutive expressing S, the Dative expressing the Experiencer, the Post-essive expressing the Possessor. But in general, word order in Agul is flexible, and the first position of all these NPs is rather due to the topicality of corresponding semantic roles. As Haspelmath (1993:301) suggests discussing the same situation in Lezgian, word order is determined by information structure rather than by grammatical relations. All variants of basic word order in Agul are in principle possible, so this
192
DMITRY GANENKOV, TIMUR MAISAK, SOLMAZ MERDANOVA
criterion does not determine the grammatical subject. Cf. examples (55a)–(55f) illustrating all variants of word order in a transitive clause: (55) a. timur.a guni ʕut’a-a SOV Timur.ERG bread.ABS eat-PRS b. timura ʕut’aa guni SVO c. guni
timura
ʕut’aa
OSV
d. guni
ʕut’aa
timura OVS
e. ʕut’aa
timura
guni
VSO
f. ʕut’aa guni timura VOS ‘Timur is eating bread.’ Thus with respect to word order the Involuntary Agent is close to A/S (being most frequent in sentence-initial position), but the criterion itself is weak due to the discourse-pragmatic nature of word order rules. 6.1.3 Verbal agreement Unlike the vast majority of the East Caucasian languages, Agul as well as Lezgian has no agreement on the verb at all (neither noun class agreement, like in most languages of the family, nor person and number agreement like in Tabassaran, Dargi, Lak, Udi, and Akhvakh). 6.2 Behavioral properties 6.2.1 Antecedent control over reflexive pronouns This parameter is often used to determine subjecthood. However, in Agul the criterion cannot be used for identifying subjects, as it is not necessarily the subject or even a core argument that controls the reflexive pronoun. In general, it seems that the ability of an argument to control reflexive pronouns in Agul is not syntactically determined, but rather depends on pragmatics (particularly the topicality of the corresponding NP). (56) a. gada.jii uči boy.ERG REFL.ABS
k’i-ne die/kill-PST
b. gadai uč.ii k’i-ne boy.ABS REFL.ERG die/kill-PST ‘A boy has killed himself.’ (57) a. gada.ji-f-asi uči boy-AD-ELAT REFL.ABS
k’i-ne die/kill-PST
NON-CANONICAL AGENT MARKING IN AGUL
193
k’i-ne b. gadai uč.i-f-asi boy.ABS REFL-AD-ELAT die/kill-PST ‘A boy has accidentally killed himself.’ uč.i-q-t̅ii (58) a. dad.ai father.ERG REFL -POST-LAT
unaq’u-ne gada.ji-s call-PST boy-DAT
b. dad.a-q-t̅ii uč.ii unaq’u-ne gada.ji-s father-POST-LAT REFL-ERG call-PST boy-DAT ‘Fatheri called the boy (to himselfi).’ As examples (56)–(58) show, any argument including the Ad-elative can control a reflexive pronoun in its own clause. The only important constraint is that the reflexive pronoun must follow the full NP. 6.2.2 Relativization In Agul, participles are used to form relative clauses. The criterion cannot be used for identifying subjecthood, as every core and virtually every non-core argument (incl. locative adverbials) can be relativized. (59) A (Ergative) a. [šünük̅.i-s q̅enfet i ] xir child-DAT candy give.PFPART woman ‘a woman who gave a candy to a child’ O (Absolutive) b. [xir.a šünük̅.i-s i ] q̅enfet woman.ERG child-DAT give.PFPART candy ‘candy which a woman gave to a child’ (60) Experiencer (Dative) a. [šahar agu ] šünük̅ city.ABS see.PFPART child ‘a child who saw a city’ Stimulus (Absolutive) b. [šünü k̅.i-s agu ] šahar child-DAT see.PFPART city.ABS ‘a city which was seen by a child’ (61) S (Absolutive) a. [dar.a-l uq’unaje ] ǯaq˳ʼ tree-SUPER sit.RESPART bird.ABS ‘a bird which sits on a tree’
194
DMITRY GANENKOV, TIMUR MAISAK, SOLMAZ MERDANOVA Locative (Super-essive) b. [ ǯaq˳ʼ uq’unaje ] dar bird.ABS sit.RESPART tree.ABS ‘a tree on which a bird sits’
6.2.3 Controller of coreferential omission This test is applicable only to the verb xas ‘become’ in the Possibilitive Construction, and in this case the control is not only admissible but obligatory (cf. examples (35)–(37) above). Other verbs with the Ad-elative argument do not take complement clauses. That is, the criterion points at a subject-like behaviour of the Ad-elative NP, but it is somewhat weak, being applicable only to one verb governing this case. 6.2.4 Target of coreferential omission The test is again applicable only to the verb xas ‘become’ in the Possibilitive Construction (62), since omission of subject NP leads to the default agentive reading with labile verbs (63) and to the default agentless reading with other verbs (64). It is impossible to get an involuntary reading in examples (63) and (64):13 (62) gada.ji-si son-DAT
agu-ne Øi dar.ala-l alʁuč’a-s xu-f see-PST [AD-ELAT tree-SUPER climb up-INF become-PART]
‘The boyi saw that (hei) could climb up a tree.’ dak’ar arʕu-f (63) gada.ji-si agu-ne Øi son-DAT see-PST [ERG/*AD-ELAT window.ABS break-PART] ‘The boyi saw that (hei) broke the window.’ rutu-f (64) a. gada.ji-si agu-ne nek̅ son-DAT see-PST [milk.ABS curdle-PART] ‘The boyi saw that the milk curdled.’ nek̅ rutu-f b. *gada.ji-si agu-ne Øi son-DAT see-PST [AD-ELAT milk.ABS curdle-PART] ‘The boyi saw that the milk curdled (on himi).’ 6.2.5 Controller of ‘same subject’ constraints As true coordinated clauses are almost absent in Agul, we shall look at the syntactic behaviour of arguments in a construction with the perfective converb -na expressing a sequence of actions carried out by the same participant. Whereas high priority 13
The same is true for another potential subject property, viz. the possibility of being used as the addressee of the imperative.
NON-CANONICAL AGENT MARKING IN AGUL
195
arguments – the Absolutive NP of an intransitive clause (65a), the Ergative NP of a transitive clause (65b), and the Experiencer marked by the Dative (65c) – can all control omission of a coreferential argument in a linked clause, O cannot do so (65e). The Involuntary Agent in this respect is an argument with high priority (65d). ʁarxu-ne (65) a. gadai Øi adi-na boy.ABS [ABS come-CONV] fall asleep-PST ‘The boy came and fell asleep.’ ruš.a-s jarʜu-ne b. gada.jii Øi adi-na boy.ERG [ABS come-CONV] girl-DAT hit-PST ‘The boy came and hit the girl.’ ruš agu-ne c. gada.ji-si Øi adi-na boy-DAT [ABS come-CONV] girl.ABS see-PST ‘The boy came and saw the girl.’ šüše arʕu-ne d. gada.ji-f-asi Øi adi-na boy-AD-ELAT [ABS come-CONV] bottle.ABS break-PST ‘The boy came and accidentally broke the bottle.’ uzu-ne e. *gada.ji ʜünii Øi qaj-na boy.ERG cow.ABS [ABS come back-CONV] milk-PST ‘The cow came back and the boy milked it.’ 6.2.6 Antecedent control over demonstrative pronouns This test gives a sort of ‘inverse’ result: only O can be referred to anaphorically by a demonstrative pronoun, whereas high priority arguments manifest stronger restrictions: in a subordinate clause, they must be anaphorically referred to by a reflexive pronoun.14 The Involuntary Agent as well as another non-canonical subject marked by the Dative, were judged as hardly acceptable controllers, although not impossible altogether. uč.i-si / *gi-si fat’im.i kümek q’u-b (66) a. ʡajšati adi-ne Aishat come-PST REFL-DAT that-DAT Fatima.ERG help do-MSD badala in order to ‘Aishat came so that Fatima would help her.’ i-ne dad.a-l-di kitab ti b. ʡajšat.ai Aishat.ERG give-PST father-SUPER-LAT book.ABS this.ERG
uč.i-si / REFL-DAT
*gi-si ge ruχu-b badala that-DAT that read-MSD in order to ‘Aishat gave a book to father so that he would read it to her.’ 14
This property of demonstrative pronouns is discussed in more detail in Cardinaletti & Starke (1999). We are grateful to Yakov Testelec for bringing this criterion to our attention.
196
DMITRY GANENKOV, TIMUR MAISAK, SOLMAZ MERDANOVA kitab c. ʡajšat.a-si agu-ne uč.i-si / ??gi-si fajnaje Aishat-DAT see-PST REFL-DAT that-DAT bring.RESPART book.ABS ‘Aishat saw the book which was brought to her.’ gulu-ne uč.i-si / ??gi-si fajnaje kitab d. ʡajšat.a-f-asi Aishat-AD-ELAT lose-PST REFL-DAT that-DAT bring.RESPART book.ABS ‘Aishat lost the book which was brought to her.’ fat’im.i uči.-si / gi-si e. ʡajšat.a hatu-ne ruši Aishat.ERG send-PST daughter.ABS Fatima.ERG REFL-DAT that-DAT berʜem ag ̥arq’u-b badala dress.ABS show-MSD in order to ‘Aishat sent her daughter so that Fatima would show to her the dress.’
Thus, according to two behavioral criteria, the Involuntary Agent is a high priority argument like A and S (and not like O in transitive clauses). The second (patientive) argument in the Involuntary Agent Construction according to these criteria seems to pattern syntactically as O rather than as A/S: in contrast to A/S, it is very doubtful as a controller of coreferential omission (cf. (65f)), but it can be an antecendent of a demonstrative pronoun in dependent clause (cf. (66f)): Øi adi-na k’i-ne (65) f. ??gada.ji-f-as kitani χula-s boy-AD-ELAT cat.ABS [house-DAT ABS come-CONV] kill/die-PST ‘The cat came into the house and the boy accidentally killed it.’ gulu-ne uči / gei ʁušu-nde gada.ji-f-as (66) f. ʜünii cow.ABS get lost-PST [REFL / that take-RESPART] boy-AD-ELAT ‘The cow was lost by the boy who bought it.’ The results of the syntactic tests for S, A, and O, as well as for experiential Datives and Involuntary Agents are summarized in Table 3.
NON-CANONICAL AGENT MARKING IN AGUL
197
Table 3. Syntactic properties of core arguments and the Involuntary Agent15
S (Abs) +
A (Erg) +
Exp (Dat) +
IA (Ad-El) +
O (Abs) +
Antecedent control over reflexive nouns Antecedent control over demonstrative pronouns Relativization
+
+
+
+
+
-
-
±
±
+
+
+
+
=A
+
Controller of coreferential omission Target of coreferential omission Controller of ‘same subject’ constraints
+
n/a
+
+ (PC)
n/a
+
+
+
=A
-
+
+
+
+
-
Word order (first position)
7. CONCLUSION We have discussed a number of uses of the Ad-elative and, to a lesser degree, Adessive cases in Agul. Both of these cases are originally locative, however in the constructions described above they have non-locative uses, viz. they encode one of the main (and predominantly human) participants of the situation. What all these constructions have in common is that one of the ‘Ad’ cases denotes a participant which is inherently agentive, but which in a given situation lacks some important Agent properties. The Involuntary Agent takes part in the situation accidentally and unintentionally, that is s/he does not control it (totally or in part). The Causee in the Causative Construction is also agentive, however s/he acts not by his/her own free will, but being forced but the Causer (the choice of one of the strategies of the Causee encoding reflects the degree of agentivity). The Agent of the Undesirable Action Construction is a bit different as s/he acts deliberately; however, his/her acting is regarded as inappropriate and the whole situation is presented as being forced by ‘the circumstances’. And the participant of the Possibilitive Construction is close to this broad ‘low agentivity’ zone in that it is presented as only potentially capable of doing something (or as one that is allowed to do it). In sum, all these participants seem to have some problems with ‘volitional involvement in the event or state’, which is the main Proto-Agent entailment in Dowty’s (1991) influential approach. Thus the two ‘Ad’ cases in Agul (in particular the Ad-elative) can be considered as a general means of expressing low agentivity of a participant, a means of encoding a participant who is ‘not agentive enough’ to be regarded as a real Agent. 15
Abbreviations: ‘n/a’: not applicable; ‘=A’: is identical to A; ‘PC’: only relevant for the Possibilitive Construction.
198
DMITRY GANENKOV, TIMUR MAISAK, SOLMAZ MERDANOVA
The strategy of expressing semantic contrasts in ‘degree of Agentivity’ by means of “an alternation between canonically-marked subjects and other more oblique encodings” (Ackerman & Moore 2001:141) is well represented in the languages of the world, cf. the data discussed by Ackerman & Moore in chapter 6 (‘Subject alternations’) of their book, among other works. N ote also that the arguments marked by one of the ‘Ad’ cases have different status in different constructions. In the case of the Causative Construction and the Undesirable Action Construction we deal with a true ‘paradigmatic alternation’ in Ackerman & Moore’s terms (as there is a contrast between two types of marking). However, the Involuntary Agent is a sort of ‘extra’ participant of the situation that can occur spontaneously without it and includes at least one S-like participant. As to the Possibilitive Construction, the Ad-elative case is the only means to encode its main participant. REFERENCES Ackerman, F. and J. Moore (2001). Proto-Properties and Grammatical Encoding: A Correspondence Theory of Argument Selection. Stanford: CSLI Publications. Auwera, J. van der and V. Plungian (1998). Modality’s semantic map. Linguistic Typology 2.1, 79-124. Cardinaletti, A. and M. Starke (1999). Responses and demonstratives. Clitics in the Languages of Europe. Empirical Approaches to Language Typology. Ed. by H. van Riemsdijk. Berlin: Mouton de Gruyter. Daniel, M., T. Maisak and S. Merdanova (2004). Causatives, decausatives and lability in Agul. International Symposium on the Typology of Argument Structure and Grammatical Relations in Languages Spoken in Europe and North and Central Asia (LENCA-2). Kazan State University, Tatarstan, Russia, May 11-14. (http://www.ling.helsinki.fi/uhlcs/LENCA/LENCA-2/information/ abstracts.html) Dixon, R.M.W. and A.Y. Aikhenvald (2000). Introduction. Changing valency: Case studies in transitivity. Ed. by Dixon, R.M.W. and A.Y. Aikhenvald. Cambridge: Cambridge University Press. Dowty, D. (1991). Thematic proto-roles and argument selection. Language 67, 547-619. Foley, W. and R. van Valin, Jr. (1984). Functional Syntax and Universal Grammar. Cambridge: Cambridge University Press. Haspelmath, M. (1993). A grammar of Lezgian. Berlin: Mouton de Gruyter. Kibrik, A.E. (1997). Beyond subject and object: Towards a comprehensive relational typology. Linguistic Typology 1.3, 279-346. Kittilä, S. (2005). A typology of involuntary agent constructions. Word 56.3. Onishi, M. (2001). Introduction: Non-canonically marked subjects and objects. Parameters and Properties. Non-canonical Marking of Subjects and Objects. Ed. by A.Y. Aikhenvald, R.M.W. Dixon and M. Onishi. Amsterdam/Philadelphia: John Benjamins. Schulze, W. (to appear). A new model of metaphorization: Case semantics in East Caucasian. Metonymy and Metaphor in Grammar. Ed. by Barcelona, A., K. Panther, G. Radden and L.L. Thorburg. Amsterdam/Philadelphia: John Benjamins.
YUKIKO MORIMOTO
CHAPTER 9 From Topic to Subject Marking: Implications for a Typology of Subject Marking1
1. INTRODUCTION According to typologists, agreement systems evolve from a topic construction, in which a full (morphologically unreduced) pronoun is used to refer to the topic NP anaphorically. The anaphoric pronoun is then reduced to a clitic-like element, while still retaining the pronominal content. It is then further reduced to a morphologically dependent affix, with the subsequent loss of the pronominal content to a mere agreement marker. The evolution of subject agreement can thus be characterized schematically as in (1) (adapted from Givón 1979). (1) Topic pronoun the man, he came TOP
PRO
Clitic pronoun the man he-came
Subject agreement the man he-came
TOP
SUBJ
PRO
AGR
Givón (1979) argues that “grammatical agreement is fundamentally a topic related phenomenon, arising from anaphoric pronominalization in topical discourse contexts.” (p.185). The discourse basis for the grammaticalization of subject marking is quite transparent in the present-day Bantu agreement systems. Yet many of the synchronic analyses of Bantu agreement fail to recognize the role of topicality. In this chapter I examine the variation in the patterns of argument linking across Bantu languages, and argue that once we recognize the crucial role topicality plays in the grammars of these languages, we can provide a principled account of the variation observed in subject agreement and a set correlating facts, which would otherwise be treated independently of the system of agreement. This chapter relates to the central theme of this volume in the following way: the Bantu data on topic vs. subject marking and a possible diachronic path from topic to
1
Part of the present work was presented at the PIONIER workshop on Differential Subject Marking in Nijmegen, July 2004. I am grateful to Helen de Hoop and the members of the PIONIER research group for inviting me to the workshop, and to the audience of the workshop for useful questions and comments. For valuable discussions and suggestions on earlier versions of this work, I also thank Peter Sells, Dieter Wunderlich, and Laura Downing. Part of this work is also published in Morimoto (2006). Finally, I benefitted greatly from comments and suggestions from the editors and reviewers. I am solely responsible for any errors or misrepresentation.
199 H. de Hoop and P. de Swart (eds.), Differential Subject Marking, 199–221. © 2008 Springer.
200
YUKIKO MORIMOTO
subject marking presented here seem to suggest that there may also be some discourse basis for (the development of) differential subject (or object) marking. 1.1 Agreement in Bantu Bantu languages display rich verbal agreement morphology comprising of 16-18 noun classes that cross-reference the verb’s core arguments (subject and primary object). One of the most detailed studies of the agreement morphology within generative syntax is offered by Bresnan and Mchombo (1987), who focus on the properties of subject and object marking in the Bantu language Chicheŵa. They show that while the object marker (OM) in Chicheŵa functions only as a pronominal argument and not as agreement, the subject marker (SM) functions both as grammatical agreement and as a topic-anaphoric pronoun. A closer look at subject marking in other Bantu languages reveals that facts in those languages diverge from the findings reported by Bresnan and Mchombo on Chicheŵa, and they also make several observations. They note, for example (p. 778), that in Dzamba, although all arguments inside VP can be questioned in situ, as in (2b), the subject cannot be questioned in its initial position, as shown by the ungrammaticality of (2c). In order to question the subject, a relative clause must be used, as shown in (2d).2 (2) a. ó-Nebo a-imol-aki ó-Biko e-kondo loo mé. ‘Nebo told Biko a story/tale today.’ b. ó-Nebo a-imol-aki nzányí e-kondo loo mé? (lit.) ‘Nebo told who a story today?’ c. *nzányí a-imol-aki ó-Biko e-kondo loo mé? ‘Who told Biko a story today?’ d. ó-Moto ó-wimol-aki ó-Biko e-kondo loo mé nzányí? (lit.) ‘The person who told Biko a story today is who?’
Dzamba
Based on such data, Bresnan and Mchombo speculate that unlike Chicheŵa, in which the subject can be questioned in its initial position, in Dzamba, all subject NPs must necessarily be grammaticized topics. They further note, citing Bokamba (1981), that Dzamba has ‘nominal pre-prefixes’ which are obligatory on definite NPs. For example, in (2), we have the pre-prefix ó- on the proper nouns (ó-Nebo, óBiko). Such pre-prefixes are obligatory on subjects. This supports their hypothesis that all subjects are obligatorily grammaticized topics. Taking this initial observation as a starting point, in this chapter I explore the idea that what has uniformly been taken to be the subject marker in Bantu languages in fact represents two types of agreement systems: subject versus topic agreement.
2
In the examples taken from published sources, I have kept the original use of ones and glosses, which, in some cases, results in inconsistency throughout this chapter. For example, a final vowel (FV) in Chicheŵa (e.g. Bresnan and Mchombo 1987) is glossed as ASP(ECT) in Kinyarwanda (e.g. Kimenyi 1980); the morpheme ra is glossed as anti-focus (AF) in Kirundi (Njayiragije 1999) but present tense (PRS) in Kinyarwanda (e.g. Kimenyi 1980).
FROM TOPIC TO SUBJECT MARKING
201
The crucial evidence for topic agreement in Bantu languages comes from a construction known as subject-object (S-O) reversal in a subset of Bantu languages, exemplified in (3) from Kinyarwanda (Kimenyi 1980).3 In the canonical SVO sentence in (3a), the relevant agreement marker a (boldfaced) agrees in class 1 of the logical subject Umuhu ‘boy’. In the reversal sentence in (3b), the agreement marker ki agrees in class 7 of the lower argument igitabo ‘book’. As shown in the translation, the fronted patient (‘book’) is a topic, and the post-verbal logical subject receives a contrastive focus interpretation.4 (3) a. Umuhuûngu a-ra-som-a igitabo 1boy 1-PRS-read-ASP 7book ‘The boy is reading the book.’ b. Igitabo ki-som-a umuhuûngu 7book 7-read-ASP 1boy ‘The boy (FOC) is reading the book (TOP)’
SVO OVS Kinyarwanda
The agreement pattern we see in (3b) is the unique property of this construction that raises the question as to the grammatical relation of the fronted patient to the verb. Unlike the more familiar left-dislocation, the object marker agreeing with the fronted patient is never present in S-O reversal, as shown in (4). (4) Igitabo ki-som-a umuhuûngu 7book 7- (*OM)-read-ASP 1boy
There are two possible solutions for the agreement pattern. The predominant solution has been to maintain the standard assumption about agreement that it licenses core argument functions such as subjects and (primary) objects. Under this assumption, the agreement marker in question must be a subject marker; S-O reversal such as (3b) is then a grammatical relation changing operation whereby the patient gets linked to the grammatical subject, and the logical subject to the grammatical object. I refer to this solution as the subject analysis. An alternative solution, referred to as the topic analysis, assumes that no grammatical relation change takes place in S-O reversal. Rather, the agreement marker (ki in (3b)) is analyzed as a topic marker licensing the topical object. These two alternative solutions naturally lead to diverging predictions. I show evidence to support the topic analysis and argue that the topic function is indeed among the functions that can trigger agreement in so-called topic-prominent languages. 3 4
The data without citation have been generously provided by Alexandre Kimenyi for Kinyarwanda and Juvénal Ndayiragije for Kirundi. Tones are unfortunately left out in the elicited examples. Abbreviations used in glosses: Number N = noun class. Languages: Dz=Dzamba, Krw=Kinyarwanda, Kdi=Kirundi
202
YUKIKO MORIMOTO
The discussion in the rest of the paper proceeds as follows. In section 2, I review some arguments for the previously proposed subject analysis, and point out why the analysis, while being able to maintain the traditionally accepted view of agreement, is nonetheless untenable for empirical reasons. In section 3 I provide the topic analysis within the constraint-based framework of Lexical-Functional Grammar (LFG; Bresnan 1982, 2001, Dalrymple et al. 1995). In section 4 I try to relate the data on S-O reversal and other correlating facts to the historical path these languages apparently have undergone with respect to their agreement system. The final section concludes the discussion with some remarks on the implications for a typology of subject marking. 2. AGAINST THE SUBJECT ANALYSIS OF REVERSAL Earlier work on S-O reversal (e.g. Bokamba 1976, 1979, 1985, Kimenyi 1980, 1988, Morimoto 1999) has generally assumed that the construction involves grammatical relation change, mainly due to the fact that the preverbal patient in reversal apparently triggers subject agreement: the patient (object) is fronted to the subject position and assumes the subject function, and the agent (logical subject) is postposed to the VP-internal object position. The grammatical function (GF) reversal analysis is somewhat analogous to the analysis of inversion in other languages: English, Chicheŵa, and Chishona, for example, exhibit inversion of the theme subject and locative complement (e.g. Out of the bushes appeared a giant bear; cf. Bresnan and Kanerva 1989, Harford 1990, Bresnan 1994); Sesotho and Setswana additionally allow inversion that involves (intransitive) unergative predicates (e.g. equivalent of In the field are grazing the cattle; see Machobane 1995, Demuth and Mmusi 1997, Lødrup 1999). In these constructions, the preverbal locative is analyzed as the grammatical subject and the logical subject in postverbal position as object, inducing a grammatical relation change; see Morimoto (1999) for an earlier OT account of all three language types including S-O (agent-patient) reversal in Kirundi and Kinyarwanda. The post-verbal agent in S-O reversal has also been analyzed not as an object but as a demoted argument (e.g. chômeur in Kimenyi’s (1980) analysis in Relational Grammar). Studies within the Principles and Parameters approach have analyzed S-O reversal as involving raising of the object to some functional projection usually reserved for the subject (e.g. SpecIP in Kinyalolo (1991), SpecTP in Ura (1996), Ndayiragije (1999)) and verb raising (to I or T). The logical subject either remains in its VP-internal position (cf. Kinyalolo,Ura) or raises to a specifier of FocusP above VP (cf. Ndayiragije). The latter (FocusP) alternative is intended to capture the focalization effect of this construction. Apart from agreement, however, little evidence has been provided for the grammatical function status of the reversal arguments. Kimenyi (1980), for example, proposes that S-O reversal is a type of ‘subjectivization process’ akin to passive: the non-agent argument canonically realized as a non-subject is realized as subject, and the logical subject is demoted. The demotion analysis is supported by the observations that the postverbal agent cannot be relativized, passivized, clefted or
203
FROM TOPIC TO SUBJECT MARKING
pronominalized. These are taken to be standard diagnostics for termhood within Relational Grammar adopted in Kimenyi’s work. Kimenyi, however, admits that the preverbal NP lacks the properties of the usual subjects apart from agreement: “NPs advanced to subject by the [S-O] reversal rule do not acquire the properties of basic subjects, such as raising, deletion under identity, and ha-insertion; the only subject property they acquire is verb agreement.” (Kimenyi 1980:145).
Despite the apparent formal similarities between locative inversion and S-O reversal and attempts at a unified analysis of these phenomena, I show that there are several advantages for not adopting the subject analysis, and that the alternative analysis pursued in the present work which crucially relies on the grammaticized notion of TOPIC enables us to explain a broader range of related facts. The correct analysis of S-O reversal is critical to the understanding of the agreement systems in Bantu. Although in my analysis I assume that there is no actual GF ‘reversal’, I will continue to refer to this construction as S-O reversal for descriptive purposes. 2.1 Lack of Evidence for Grammatical Relation Change The argument that S-O reversal cannot involve GF alternation comes from a series of syntactic tests for subjecthood. A number of syntactic tests used by Bresnan and Kanerva (1989) and Bresnan (1994) to argue that locative inversion in Chicheŵa involves GF reversal all fail to apply to S-O reversal. Here I consider VP attribution, gapping in coordinate structure, and subject-to-subject raising. 2.1.1 VP attribution Bresnan (1994) shows that Chicheŵa has a non-finite verb that can be used to modify NPs, as in (5) (Bresnan, p.93 ex (58)). pa m-pando ] (5) a. m-sodzi [VPw-ó-ík-á nsómbá 1-fisherman 1-ASC.INF-put-FV 10.fish 16 3-chair ‘a fisherman putting fish on the chair’ pá m-pando] b. nsómbá [VPz-ó-ík-í-ídw-á 10.fish 10-ASC.INF-put-PASS-FV 16 3-chair ‘fish being put on a chair’
Chicheŵa
In both active and passive examples in (5), the subject is attributed by the modifying VP. In Chicheŵa locative inversion, Bresnan (p.94, ex (59)) shows that the inverted locative subject can also be attributed, providing evidence for the subject status of the inverted locative: (6) m-nkhalangó m-ó-khaál-á mi-kângo 18-9.forest 18-ASC.INF-live-FV 4-lion ‘in the forest where there lives lions’
Chicheŵa Bresnan (1994:93.94)
204
YUKIKO MORIMOTO
In Kirundi, there is no construction analogous to VP attribution like one found in Chicheŵa. The closest construction is one in which a verbal complex is used to modify NPs, as in (7). ku-gur-a ico gitabo], a-a-ciye (7) Uyo muhungu, [VP á-maze he-PST-cut that boy 3SG-finish INF-buy-ASP that book a-ja muhira ku-gi-soma he-go home INF-it-read ‘The boy, having bought the book, went home to read it.’ In this construction, S-O reversal is also acceptable: ], ki-a-ciye ki-soma Yohani (8) Ico gitabo, [VP kí-maze ku-gur-u-a that book it-finish INF-buy-PASS-ASP it-PST-cut it-read John ‘The book, having been purchased, was read by John.’ The Kirundi sentences in (7)-(8), however, are not quite the same as the Chicheŵa examples in (5)-(6): according to Ndayiragije (p.c. May 2000), the Kirundi counterparts are traditionally referred to as a ‘conjunctive clause’. a-maze ku-gur-a in (7) translates to ‘once he/she has finished buying’ or ‘after having bought’.5 Thus, these conjunctive clauses are better analyzed as clausal adjuncts. Unlike the Chicheŵa attributive VP, the Kirundi counterpart does not contain an associative morpheme which derives the gerundive form of the verb. This also suggests that these conjunctive clauses are not modifying N, as in the attributive VP in Chicheŵa. This construction, then, does not show that there is subject control. 2.1.2 Extraction out of coordinate constituents Extraction out of coordinate constituents is used to distinguish among grammatical functions. The generalization that holds for English is that “subject gaps at the top level of one coordinate constituent cannot occur with any other kind of gap the other coordinate constituent”, as succinctly stated in Bresnan 1994:98 (cf. Williams 1977, Gazdar 1981, Falk 1983, Woolford 1987). This is illustrated by the examples in (9) (Bresnan 1994:98, exs (71)-(72)). (9) a. b. (10) a. b.
5
She’s someone that loves cooking and hates jogging. (S-S) She’s someone that cooking amuses and jogging bores. (O-O) *She’s someone that cooking amuses and hates jogging. (O-S) She’s someone that cooking amuses and I expect will hate jogging. (O-embedded S)
Such conjunctive clauses are clearly distinguished from relative clauses by the position of the grammatical high tone: the former is indicated by a high-tone on the first syllable of the verbal complex (in the brackets), while the latter is identified by a high tone on either the final or the penultimate syllable of the finite verb, depending on the length of the word. Thanks to Juvénal Ndayiragije for information and discussion on this point.
FROM TOPIC TO SUBJECT MARKING
205
Bresnan (1994) shows how the inverted locative in locative inversion in English (e.g. Out of the bushes appeared a monster) observes the same constraint with respect to extraction from coordinate constituents, thereby providing evidence that the inverted locative is the grammatical subject (Bresnan 1994:98, exs (73)-(74)): (11) a. That’s the old graveyard, in which ___ is buried a pirate and ___ is likely to be buried a treasure. (subj-subj) b. That’s the old graveyard, in which workers are digging ___ and a treasure is likely to be buried ___. (nonsubj-nonsubj) (12) a. ??That’s the old graveyard, in which workers are digging ___ and they say ___ is likely to be buried a treasure. (nonsubj-subj) b. That’s the old graveyard, in which workers are digging ___ and they say ___ is buried a treasure. (obl-embedded subj) This test has been used to test for subjecthood in other languages as well (cf. Joshi 1993 for Marathi, Kroeger 1993 for Tagalog), as this allows us to distinguish between subjects and non-subjects. In Bantu languages, however, we observe no such asymmetry: one conjunct with a subject gap can be coordinated with another with a nonsubject gap, as shown in (13). As we see in the translation, the English counterpart is ungrammatical in (13b). (13) a. Yohani ni we ___ a-kunda ku-iruka ___ a-ka-anka 1SM-ka-hate/refuse John is him S 1SM-PRS.like INF-run S ku-soma INF-read ‘John is the one who likes running while/but hates reading.’ b. Yohani ni we umwarimu a-kunda ___ ___ a-ka-anka John is him teacher 1SM-PRS.like O S 1SM-ka-hate ku-soma INF-read *’John is the one that the teacher likes but hates reading.’ Kdi This test fails to show anything about grammatical functions in Bantu languages for obvious reason: whenever there is a gap (null elements), the subject/object agreement morphology on the verb functions as topic-anaphoric agreement. Hence there can never be a real gap in the coordinate structure. 2.1.3 Subject-to-subject raising Kirundi and Kinyarwanda have no raising predicates such as seem, expect, and likely. The construction in (14) from Kinyarwanda (Kimenyi, p.c, June 2000) which involves a modal verb is apparently taken to be putative raising (Kimenyi 1980; also p.c, June 2000).
206
YUKIKO MORIMOTO
(14) Abaana ba-shobor-a ku-nywa children 2SM-can-IPFV to-drink ‘The children can drink milk.’
amata milk
Putative raising Krw
This construction allows negation on each verb, indicating that it involves biclausal structure. Example (15a) is raising with negation on the modal and (15b) has negation on the lower verb. (15) a. Abaana nti-ba-shobor-a ku-nywa amata milk children NEG-2SM-can-IPFV to-drink ‘Children cannot drink milk.’ b. Abaana ba-shobor-a ku-ta-nyw-a amata children 2SM-can-IPFV to-NEG-drink milk ‘Children can [not drink] milk.’
Raising
Krw
Kimenyi (1980) shows that in this construction, S-O reversal is possible. In Kimenyi’s (1980) Relational Grammar analysis, this is taken to be evidence for the subject status of the preverbal reversal NP. (16) Amata a-shobor-a ku-nywa abaana children milk 6-drink-IPFV to-drink ‘The milk can be drunk by children.’
Reversal Krw
It can be shown, however, that there is an analysis of these putative raising data which is perfectly consistent with the analysis that involves no GF reversal (see Morimoto 2000, chapter 3 for such an analysis). In a parallel structure approach like LFG, grammatical functions are not defined configurationally, but are assigned through lexical mapping of predicate-argument structure to grammatical functions. Thus, we can identify the subject in the domain of ku-nywa ‘drink’ with the subject of the higher verbal domain by structure-sharing, and functionally associate the object with a TOPIC function at the level of functional structure, regardless of the structure position of these NPs. In short, the standard tests for subjecthood illustrated above all fail to show that the preverbal reversal NP is the grammatical subject. 2.2 Unexplained facts The subject analysis put forth by various researchers also leaves some facts unexplained. Here I focus on the properties of the post-verbal agent in S-O reversal and variation across Bantu languages in the properties of agreement.
FROM TOPIC TO SUBJECT MARKING
207
2.2.1 The post-verbal agent has no object properties One of the puzzling facts that remain unexplained under the subject analysis is that the post-verbal agent of a reversal sentence does not display usual object properties. For example, it can neither be relativized (17) nor pronominalized (18). (17) *N-kund-a umuhuûngu igitabo ki-som-a 7book 7-read-ASP I-like-ASP 1boy ‘I like the boy (foc) who is reading the book (top).’ Kinyarwanda (18) a. Ba-ra-gi-som-a 2-AF-7-read-ASP ‘They are reading it.’ b. Cyi-(ra)-ba-som-a 7-(AF)-2-read-ASP ‘#It is reading them.’ (non-reversal reading) (reversal reading) ‘*They (FOC) are reading it (TOP).’ The post-verbal agent is also not in the canonical, immediately post-verbal, object position. In (19a), the post-verbal agent is positioned after the adverbial phrase buhurobuhuro ‘slowly’, presumably adjoined to VP. Placing it in the canonical object position inside VP results in ungrammaticality as illustrated in (19b). This postverbal agent is therefore neither positionally nor morphologically licensed as the grammatical object. (19) a. Ibitabo bi-a-som-ye buhurobuhuro Yohani John book 7-PST-read-PFV slowly ‘John (FOC) read the book (TOP) slowly.’ b. *Ibitabo bi-a-som-ye Yohani buhurobuhuro The post-verbal logical subject also could not be a demoted oblique as analyzed by Kimenyi (1980): in the languages considered here (e.g. Kirundi, Kinyarwanda, Dzamba), oblique arguments are generally marked by prepositions such as n’ in (20) from Kinyarwanda. If the post-verbal agent were a demoted oblique, this would be an odd exception, for which no explanation has been provided. (20) a. Úmwáalímu a-ra-andik-a n’ ííkárámu teacher 1-PRS-write-ASP with pen ‘The teacher is writing with a pen.’ b. Úmwáana y-a-rir-aga n’ âgahiinda keênshi much child 1-PST-cry-ASP with sorrow ‘The child was crying with much sorrow.’
Kinyarwanda
Kimenyi (1980)
If the post-verbal agent is not an object or oblique, then we are left with the assumption that it is the grammatical subject.
208
YUKIKO MORIMOTO
2.2.2 Reversal and non-reversal languages show varying properties of agreement A closer look at the properties of the agreement morphology in question across the Bantu family reveals systematic variation that correlates with the presence/absence of S-O reversal. The relevant data include the patterns of wh-questions and idiomatic expressions involving the verb and object. These data both point towards the different properties of the agreement morphology across Bantu. Wh-elements: Wh-phrases are inherently focused, given that they are new information. If the reversal languages employ topic rather than subject agreement and requires the agreeing NP to be always topic, then these languages would not allow a wh-element to agree. We saw this earlier in (2c) from Dzamba, repeated here in (21). The same holds for other reversal languages like Kinyarwanda as shown in (22). Chicheŵa on the other hand does not permit S-O reversal, and it allows wh-subjects in situ, as shown in (23)).6 (21) *Nzányí ó-wimol-aki ó-Biko e-kondo l mé? Dz. wh-subject in situ ‘Who told Biko a story/tale today?’ (Bokamba 1981) (22) *Nde y-a-som-ye igitabo? who SM-PST-read-ASP book ‘Who read the book?’ Krw wh-subject in situ (23) (Kodí) chíyâni chi-ná-ónek-a? Q what(7) SM(7)-PST-happen-INDIC ‘What happened?’ Chicheŵa wh-subject in situ If we uniformly assume subject agreement across the whole family, then these facts would be left unexplained. If we assume, instead, that reversal languages display topic agreement, the ungrammaticality of (21)-(22) would be an obvious consequence of the system. Verb-object idioms: In idiomatic expressions consisting of a verb and its object such as (24) in German, the object is non-referential, and as such they can never be topic. (24) a. das Kind mit dem Bade ausschütten ‘to throw the baby with the bath water’ b. Viele Köche verderben den Brei ‘(Too) many cooks spoil the meal.’ If these idiomatic objects can be passive subjects and still maintain the idiomatic interpretation, then it would show that the language allows non-topical, nonreferential subjects. Now, German has a non-referential expletive subject es, as in es 6
Sam Mchombo (p.c, January 2003) points out that the verb in (23) has high tones as indicated, which signal extraction of the wh-element. In other words, this question still appears to retain the tonal property of the cleft construction (which is built on relativization in Bantu, according to Bresnan and Mchombo 1987) that is typically used to form a subject wh-question in Bantu. It could be that the tonal distinction on the verb remains merely as a historical source of what used to be the cleft required for a subject wh-question.
FROM TOPIC TO SUBJECT MARKING
209
regnet ‘it is raining’. And we see in (25) that the objects in these verb-object idioms can be passivized subjects without losing the idiomatic interpretation. (25) a. Das Kind wurde mit dem Bade ausgeschüttet ‘The baby got thrown with the bath water.’ b. Der Brei wird von vielen Köchen verderben ‘The meal gets spoiled by (too) many cooks.’ In Chicheŵa, we also find verb-object idioms, as exemplified in (26a). The grammaticality of (26b) shows that Chicheŵa allows the objects of verb-object idioms to be passivized (Bresnan and Mchombo 1987:763). (26) a. Chifukwá chá mwáno wake Mavútó tsópáno because of rudeness his Mavuto now a-ku-nóng’ónez-a bôndo SM-PRS-whisper.to-INDIC knee ‘Because of his rudeness, Mavuto is now whispering to his knee (= feeling remorse).’ b. Bôndo li-ná-nóng’onez-ĕdw-a knee(5) SM(5)-PST-whisper.to.PASS-INDIC ‘The knee was whispered to (= the remorse was felt).’ Chicheŵa None of the Bantu languages under discussion has idiomatic subjects analogous to those found in English, such as The cat has got his tongue (= ‘he can’t speak.’). Chicheŵa, however, has a few proverbs that are used as fixed expressions, exemplified in (27) from Bresnan and Mchombo (1987). (27) a. Kalulu a-na-mu-omba maondo SM-PST-OM-knock knees Hare ‘The hare knocked him in the knees.’ (= He has stiff knees which cannot bend.) b. Njovu i-na-ponda-po Elephant 9SM-PST-step-LOC.ENCL ‘The elephant stepped on it (hence rendered it dysfunctional).’ Chicheŵa Again in such expressions, the subject is non-referential; hence it cannot be topical. These data therefore show that subjects in Chicheŵa need not be topical or even referential. This is not true of reversal languages, as illustrated in (28) from Kirundi. The data in (28) are consistent with those of another reversal language, Dzamba, which requires the agreeing preverbal NP to be topic as we have already seen in (21). (28) a. Yohani a-a-ra-hend-ye umunwa John 1SM-PST-AF-cheat/mislead-PFV mouth (lit.) ‘John cheated/misled the mouth.’ (= ‘John ate almost nothing.’)
210
YUKIKO MORIMOTO b. *Umunwa u-a-ra-hend-u-ye (na Yohani) SM-PST-AF-cheat/mislead-PASS-PFV by John mouth (lit.) ‘The mouth was cheated/misled (by John).’ (= ‘Almost nothing was eaten (by John).’)
Kdi
In short, these data on idiomatic expressions also reveal different agreement properties that are unexplained by the uniform subject analysis. 2.2.3 Agentive object without verbal morphology Apart from the above facts that are unaccounted for in the subject analysis, it is also worth pointing out that it is typologically quite rare for an agent to be linked to object without any morphology accompanying such unusual linking (also argued by Lødrup 1999). And this would be precisely the claim made by the subject analysis. Now in ergative languages, the more patient-like argument (O) is linked to the most unmarked absolutive case, along with the sole argument of the intransitive predicate (S). The agent of the transitive (A) is linked to the marked ergative case. The linking patterns of A, S, and O are thus said to be the inverse of those in nominative-accusative languages (cf. Manning 1996). Similarly in Austronesian languages, the unmarked linking pattern is one in which the patient-like argument is linked to nominative (e.g. ang-marking in Tagalog; cf. Kroeger 1993). These inverse linking types, however, display rich case/verbal morphology. Bantu languages share no typological properties with these language types. It is therefore not reasonable to assume that ergative/Austronesian type linking is shared by (a subset of) Bantu languages without any morphology. To summarize, in this section I hope to have shown that the predominantly held view of S-O reversal represented by the subject analysis is untenable, firstly because there is no reliable evidence for the assumed grammatical relation change, and secondly because it fails to provide a coherent explanation of the set of covariation observed in the reversal and non-reversal languages. What we see instead, as I explicate further below, is an example of prominencemarking, such that the preverbal topic shows agreement with the verb independent of its grammatical role (see also Donohue, this volume). This clearly corresponds to identifying function of case (or here, rather agreement), as pointed out in de Hoop and Narasimhan (this volume): topic marking is an instance of identifying the topic of the sentence, even when the topic is the grammatical object. 3. SALIENCE-BASED AGREEMENT VS. ROLE-BASED AGREEMENT In this section I provide a sketch of the alternative topic analysis of S-O reversal and agreement patterns across Bantu more generally. The key idea is that Bantu languages exhibit two types of agreement: salience-based and role-based. Furthermore, the observed variation in agreement properties across the synchronic
FROM TOPIC TO SUBJECT MARKING
211
Bantu grammars suggests a historical path from salience-based to role-based agreement. The idea explored in the topic analysis is that just as arguments are ranked according to the argument hierarchy, they are ranked according to the topicality hierarchy. This hierarchy is determined by relative ‘salience’ of the arguments, and is distinguished from discourse prominence. The notions of ‘topic salience’ and ‘discourse prominence’ are defined in (29). (29) a. TOPIC SALIENCE Arguments are ranked according to topicality. An argument that is most prominent in discourse is also the most salient in terms of topicality. In the absence of such an argument, by default the nominative argument will be most salient (cf. Givón 1976, Keenan 1976, Li and Thompson 1976). b. DISCOURSE PROMINENCE Any element that can bear stress can be discourse-prominent. It may be topical or focal. There can be at most one argument that is most salient in terms of topicality, while there can be multiple constituents that are discourse-prominent. The claim in (29a) that the nominative argument will be the most topically salient by default is supported by the cross-linguistic characterization of subjects and the historical source of subjects’ topic prominence. Givón (1976), for example, notes that subjects in many languages (e.g. Malagasy and Kinyarwanda) must be either definite or generic (see also Keenan 1976, Li and Thompson 1976). This property is reminiscent of (dislocated) topics, which, according to Givón, are universally restricted to be definite or generic (also see Alsagoff 1992:192, Lambrecht 1994, Birner and Ward 1998). Malagasy is an VOS language, and the nominative NP is placed sentence finally. However, like most verb-initial languages, the position for discourse-prominent constituents is clause-initial. Even in so-called subject-prominent languages such as English, which tolerate indefinite, non-referential subjects, subjects are nonetheless overwhelmingly definite and referential (Givón pp. 154-155). In English, a continuing topic is generally expressed as a subject, but a new topic (discourse-prominent) is expressed in a leftdislocated position. For example in an answer to the question in A below, B would be more natural than B’, as shown in (30), where they (family) is established information: (30) A: B: B’:
Where does your family live? They live in N.Y. #My family, they live in N.Y.
On the other hand, as a continuation of the above exchange, if B wanted to mention where his/her cousin lives as opposed to his family, then an utterance like that of B’ in (30) becomes most felicitous. In this context, the cousin in (31) is introduced as a new topic contrasting with the old topic family.
212
YUKIKO MORIMOTO
(31) B: (Now) my cousin, he lives in Chicago. B’: #(Now) he lives in Chicago, my cousin. B’’: (Now) my COUSIN lives in Chicago. (OK with appropriate intonation) Thus, a dislocated topic is used as a new/contrastive topic, while an old topic tends to be expressed as a topic of topicalization or the subject in a non-dislocated subject construction. Birner and Ward (1998) make a similar pragmatic difference between English dislocation and topicalization based on a large number of corpus data. I assume therefore that topic salience is universally part of the argument linking domain. Following the system of abstract case features familiar in Lexical Decomposition Grammar (e.g. Wunderlich 2000), I represent topic salience by a binary feature [-ht], as shown in (32a,b). The feature [±ht] states that ‘there is no argument that is higher in topic salience’, and specifies the element in highest topic salience. Conversely, [+ht] states that ‘there is an argument that is higher in topic salience.’ These features form a universal scale as shown in (32c), which I refer to as the TOPICALITY HIERARCHY. (32) Encoding topic salience (a la Wunderlich 2000) a. [-ht]: ‘There is no other argument that is higher in topicality.’ b. [+ht]: ‘There is another argument that is higher in topicality.’ c. Topicality hierarchy: [-ht] > [+ht] In terms of these feature specifications, topic-prominent languages can be said to link [-ht] (topically most salient) to subject; subject-prominent languages, on the other hand, link [-hr] (highest argument) to subject. Within this system, we can say that the subject is cross-linguistically the default topic because [-ht] is assigned to the [-hr] role by default when context identifies none of the arguments as topically salient (or non-salient). We can now characterize the Bantu agreement systems using the notion of topic prominence defined above. S-O reversal languages employ salience-based topic marking in the sense that the agreement prefix encodes the most topical, but not necessarily the highest, argument. We can capture this property of the topic agreement by assuming a lexical entry like that in (33). These functional schemata, familiar from LFG, specify that there is some feature structure (f 1) that contains a grammaticized discourse topic ‘TOP’, whose value is another feature structure (f 2). This inner feature structure contains the information specified by the topic marker (TOP), gender and number. The optional semantic content SEM = ‘pro’ abbreviates two feature structures, one with ‘pro’, which represents a TOP functioning as a topic pronoun, and the other without ‘pro’, representing a TOP as topic agreement.
FROM TOPIC TO SUBJECT MARKING
213
(33) Lexical entry of the topic marker (TOP) TOP-: Vinfl (f1 TOP) = f2 (f2 GEND) = α (f2 NUM) = β = ‘pro’) ((f2 SEM) The instantiations of these functional schemata result in the feature structure shown in (34). The curved line from f 2, the value of TOP, to GF in (34) indicates that the TOPIC is associated with one of the argument selected by the predicate (realized as a core GF). This is ensured by the extended coherence condition of LFG (Bresnan 2001, chapter 4), which requires that grammaticized discourse functions TOPIC and FOCUS be associated with one of the argument functions selected by the predicate. (34)
f1
TOP
f2
(SEM
‘proi’)
GEND
α
NUM
β
GF
V
TOP
-verb
Reversal lgs. Non-reversal languages like Chicheŵa employ role-based subject marking: the subject marker (SM) encodes the highest argument in the argument hierarchy. Thus, the lexical entry may contain the information shown in (35). Here the SM specifies that there is some feature structure (f1) that contains a SUBJ, whose value is identified with another feature structure (f2). The information included in the inner fstructure is identical to that of the TOP in (33). The f-structure in (36) shows the instantiations of these functional schemata. (35) Lexical entry of the subject marker (SM) SM-: Vinfl (f1 SUBJ) = f2 (f2 GEND) = α (f2 NUM) = β = ‘pro’) ((f2 SEM)
214
YUKIKO MORIMOTO
(36)
V
(PRED ‘proi’) f1
SUBJ f2
GEND
α
NUM
β
SM
-verb
Chicheŵa Now there is a third type, namely, languages such as Sesotho and Setswana that do not allow S-O reversal like the salience-based agreement languages, but also do not allow non-topical subjects to agree like the role-based agreement languages (cf. Demuth and Mmusi 1997). In fact, I believe the majority of Bantu languages are described in the literature as belonging to this type. Within the proposed systems of agreement, we can characterize this third type as having a basically role-based agreement system. The only restriction is that the agreeing subject must be topically salient ([-ht]). The f-structure in (37) shows that the TOPIC function is always associated with SUBJ. (37)
f1
TOP
SUBJ
f2
(SEM
‘proi’)
GEND
α
NUM
β
V
SM -verb
Sesotho/Setswana Now, one might expect that in a language where we find both subject and topic marking, as I claim for the Bantu data in this chapter, the constraint of DISTINGUISHABILITY must be low ranked, since DISTINGUISHABILITY requires the subject and the object of a transitive sentence to be distinguishable. If the agreement marker can be used to mark the topic rather than the subject, we might predict ambiguity (hence, a violation of DISTINGUISHABILITY) as to what is the subject and what is the object in a given transitive sentence. However, as I will show below, DISTINGUISHABILITY between the subject and object is satisfied after all, even if not through agreement but through the referential properties of the arguments. Looking at the above typology from the diachronic perspective, we might interpret the Sesotho/Setswana type as an intermediate stage in the shift from topic to subject agreement. One piece of supporting data for the supposed historical path that gave rise to the two agreement systems in the present day Bantu languages comes from Meeussen’s (1967) reconstruction of Proto-Bantu. Meeussen (1967:120) reconstructs a syntactic pattern in Proto-Bantu that is analogous to S-O reversal (and locative inversion), as illustrated in (38) and (39). These data lead us to infer that the system of topic agreement already existed in Proto-Bantu.
FROM TOPIC TO SUBJECT MARKING
215
(38) a. nkíma jí-í¸ji buénge búá-mití S (exp) SM-V O (theme) b. buénge búá-mití bú-í¸ji nkíma SM-V S (exp) O (theme) ‘The monkey knows the (cleverness of) trees.’ (39) a. njogu¸ jí-ákú¸i¸de mu-duí¸ LOC S (exp) SM-V ‘The elephant died in the river.’ b. mu-duí¸ mu-ákú¸i¸de njogu¸ LOC SM-V S (exp) ‘In the river (there) died an elephant.’ Topic agreement is a marked option (as opposed to subject agreement) in that there is no one-to-one correspondence between the agreement morphology and grammatical function - it licenses either the subject or object. It seems plausible that the change from salience-based to role-based agreement was motivated by such markedness of topic agreement, in which the preverbal topic got reanalyzed as the subject and concomitantly topic agreement as subject agreement. We can then also expect a further historical path where all Bantu languages eventually come to employ the role-based agreement system. The expected concomitant change is, of course, the loss of S-O reversal. Additional data on S-O reversal suggest that this may already be happening, as discussed below. 4. GRAMMATICAL CHANGE ALONG REFERENTIAL HIERARCHIES When a new system replaces an old, it typically does so gradually in a predictable fashion along referential hierarchies. A case in point is split ergativity. Split ergativity is viewed as a transitional synchronic stage in a change from the ergative/absolutive system to nominative/accusative system (cf. Dixon 1994). In ergative languages, it is claimed that the higher the NP on the nominal hierarchy in (40), adopted by Dixon (1994), the more likely it is to be in A rather than O function (e.g. Du Bois 1987, Dixon 1994). (40) Nominal Hierarchy 1st > 2nd > 3rd > Name > Human > Animate > Inanimate more likely to be in A than in O function The first person pronoun is therefore the most prototypical or unmarked A, while the inanimate NP is the most unmarked O. Assuming the general markedness principle that marked forms are also morphosyntactically marked, what we expect in a split ergative language along this nominal hierarchy is presence of morphological marking of an NP from the lowest element of the hierarchy when it is in A function, but from the highest element in O function. According to Dixon (1994:85), a number of languages have split case-marking systems exactly on this principle: a
216
YUKIKO MORIMOTO
morphologically overt (marked) ergative case is used with NPs from the lowest end, up to some point in the middle of the hierarchy, and an accusative case from that point on, overt to the highest end of the hierarchy. Dixon (p.87, Table 4.3) provides an example of this type of split from the ergative language of Australia Yidiny, shown in table 1. Table 1. Split ergative pattern in Yidiny
A S O
Ø Ø
ERG
ERG
ERG
ERG
Ø
ACC
ACC
1&2 pronouns
human deictics, interrog.
Ø (ACC) inanimate deictics, names, kin terms
Ø Ø inanimate intterrog.
Ø Ø common Ns and As
Table 1 shows that in Yidiny, first and second person pronouns have a nominative-accusative (-Ø vs. -ny ~ -:ny) paradigm, while at the other end of the scale, common nouns show absolutive-ergative marking (-Ø vs. -ngu ~ -du etc). In short, split ergativity illustrates how a shift from one system to another takes place systematically and predictably along a referential hierarchy. Returning to the Bantu languages with the salience-based agreement system, we see that S-O reversal is restricted along the dimension of the animacy hierarchy (= a simpler version of the nominal hierarchy in (40)) given in (41). (41) Animacy hierarchy: Human > Animate > Inanimate As in the nominal hierarchy in (40), given two arguments, subject and object, the highest element in the hierarchy (human) is the most prototypical, or unmarked, subject, while the lowest element (inanimate) is the most unmarked object. The most unmarked transitive sentence, then, would be one such as the boy is reading a book where the subject is human, and the object, inanimate. On the other hand, the most marked transitive sentence would be one like loud noise disturbed the baby, where the subject is inanimate, and the object, human. As reported by Morimoto (2001), it is these predicates with marked animacy relations that block S-O reversal. This is exemplified in (42) and (43).7 Example (42) represents the most marked sentence type, where the subject is inanimate and the object is human. As shown in the translation for (42), the reversal reading is blocked. The b sentence is grammatical only in the non-reversal, non-sensical reading. In (43), the subject is also lower in animacy, and again, the reversal reading is blocked in the b sentence.
7
The examples in (42)-(45) are taken from Morimoto (2001).
217
FROM TOPIC TO SUBJECT MARKING (42) a. Urushiinge ru-ra-joomb-a umwaana needle it-AF-pierce-ASP child ‘The needle will pierce the child.’ b. Umwaana a-joomb-a urushiinge child he-pierce-ASP needle *‘The needle(FOC) will pierce the child(TOP).’ (43) a. Akayabu ka-a-ra-fyese umuhungu cat it-PST-AF-lick:PFV boy ‘The cat licked the boy.’ b. Umuhungu a-a-fyese akayabu boy he-PST-lick:PFV cat *’The cat -FOC licked the boy-TOP.’
S=inanimate O=human *reversal Kinyarwanda S=animate O=human *reversal Kirundi
When the arguments are of equal animacy and there is potential ambiguity in interpretation, as in (44), the reversal interpretation is unavailable just from the string (in a null context); the only possible meaning in (44b) is one in which umukoôwa ‘girl’ is the subject/agent and umuhuûngu ‘boy’ is the object/patient. (44) a. Umuhuûngu y-a-som-ye umukoôwa 1boy 1-PST-kiss-ASP 1girl ‘The boy kissed the girl.’ b. Umukoôwa y-a-som-ye umuhuûngu 1girl 1-PST-kiss-ASP 1boy ‘The girl kissed the boy.’ *‘The boy(FOC) kissed the girl(TOP).’
S=human O=human
(no reversal) (reversal) Kinyarwanda
On the other hand reversal is allowed in an example like (45) where only one of them can be the likely subject/agent due to the core meaning of the verb. (45) a. Icyuma cy-a-kas-e umugaati knife it-PST-cut-ASP bread ‘The knife cut the bread.’ b. Umugati w-a-kas-e icyuma bread it-PST-cut-ASP knife ‘The knife(FOC) cut the bread(TOP).’
S=inanimate O=inanimate √ reversal Kinyarwanda
In short, S-O reversal is permitted when the subject outranks the object in animacy, but not when the subject is lower in animacy than the object. When there is equal animacy and ambiguity results, the reversal interpretation becomes unavailable. These facts are summarized in (46).
218
YUKIKO MORIMOTO
(46) Grammatical Function-Animacy Association in S-O Reversal
HUM ANIM INAN
HUM
ANIM
INAN
yes* no no
yes yes* no
yes yes yes*
yes*: only if there is no ambiguity
The rationale for the facts summarized in (46) seems straightforward: on the one hand, the configuration in which object outranks subject in animacy is nonprototypical, or more marked, than the one in which subject outranks object. On the other hand, S-O reversal sentences are more marked, or non-canonical, both syntactically (non-canonical OVS order) and pragmatically (patient is topical and agent is focal), than the non-reversal counterparts. The descriptive generalization, then, is that the marked animacy relations cannot be expressed in the marked syntactic construction (reversal).8 This kind of markedness is susceptible to change/loss over time (cf. Croft 1990). Animacy also figures in the domain of differential object marking (DOM; cf. Bossong 1985, Aissen 2003), in which objects that are high in animacy are casemarked, while those low in animacy are unmarked. Like split ergativity, DOM is also said to be a stage in the diachronic process whereby overt case-marking is being lost from the low end of the referential hierarchy. In short, the animacy restriction on S-O reversal suggests this construction is being lost from the most marked end of the association between the referential and argument hierarchy (e.g. inanimate-subject, human-object). And this change must be concomitant to the change in the properties of the agreement morphology. Properly recognizing the two systems in the Bantu family, the salience-based and role-based, then enables us to relate the relevant synchronic data like the animacy restriction on S-O reversal, inventory of non-topical subjects to the issue of agreement and to a diachrony of the agreement systems. From the perspective of the DISTINGUISHABILITY constraint, S-O reversal is only possible in those cases where DISTINGUISHABILITY between the subject and the object is already guaranteed by their relative ranking on the animacy scale. That is, only when the object is lower than the subject in animacy, and therefore clearly recognizable as the grammatical object, topic marking is allowed. In all other cases, the preverbal noun phrase that agrees with the verb is necessarily interpreted as the subject. Therefore, even though argument linking in Bantu languages suggests that it is more important to identify the prominent (topical) argument than to distinguish between the subject and the object, this is only apparently so, since a preverbal object can only be marked as the topic if it is clear from its relative animacy that it is indeed the grammatical object. So, the constraint on DISTINGUISHABILITY between the subject and the object does not have to be satisfied via agreement. When
8
See Morimoto (2003) for an OT analysis of these facts.
FROM TOPIC TO SUBJECT MARKING
219
agreement does not mark the subject in order to distinguish between the two arguments, animacy will. 5. CONCLUDING REMARKS In this chapter, I have argued for the two-fold agreement systems in Bantu: saliencebased agreement and role-based agreement. The crucial evidence for salience-based agreement comes from the existence of S-O reversal, where the preverbal patient triggers topic, rather than subject, agreement. The subject analysis of this construction that assumes grammatical relation change was rejected on both empirical and theoretical basis. I have shown that the proposed view of agreement better explains the set of co-variation related to the presence/absence of S-O reversal. Although the proposal about topic marking might seem like a radical departure from the standard assumption (especially in generative syntax) that agreement crossreferences core arguments (subject and object), from the typological perspective, topic marking is perhaps not as peculiar as it first appears: salience-based topic marking can be viewed as being analogous to the prominence-based (pragmatically conditioned) agreement marking in Tukang Besi (and, perhaps to a lesser extent, in other Austronesian languages) discussed by Donohue (this volume). The notion of DISTINGUISHABILITY discussed by de Hoop and Narasimhan (this volume) was also brought to bear on the data on the animacy restriction of S-O reversal. The topic marking on objects, an identifying function of agreement, is available only when there is no ambiguity and the grammatical relation is obvious when the subject outranks the object in animacy, and hence DISTINGUISHABILITY is satisfied. Thus, the violation of the DISTINGUISHABILITY constraint (in order to prioritize the identifying function) is only apparent. I have also tried to relate the synchronic data to the historical path from topic to subject agreement, which I suggest is a particular realization of the more general functional shift from topic to subject. Referential properties like animacy seem to figure in diachronic change, as attested in split ergativity and differential object marking. The animacy restriction on S-O reversal thus is likely to represent a stage in the gradual loss of this construction, concomitant with the shift from topic to subject agreement. In conclusion, in order to provide a coherent explanation of the set of facts related to S-O reversal and to relate those facts to a broader set of agreement phenomena (split ergativity, DOM), it is crucial that we recognize the role of topicality in the core grammar, as manifested in topic agreement. REFERENCES Aissen, J. (2003). Differential case marking: Iconicity vs. economy. Natural Language and Linguistic Theory 21, 3, 435-483. Alsagoff, L. (1992). Topic in Malay: The Other Subject. Doctoral dissertation, Stanford University, Stanford (CA).
220
YUKIKO MORIMOTO
Birner, B.J. and G. Ward (1998). Information Status and Non-canonical Word Order in English. Amsterdam/Philadelphia: John Benjamins. Bokamba, E.G. (1976). Question Formation in Some Bantu Languages. Doctoral dissertation, Indiana University, Indiana. Bokamba, E.G. (1979). Inversions as grammatical relation changing rules in Bantu. Studies in the Linguistic Sciences 9, 2, 1-24. Bokamba, E.G. (1981). Aspects of Bantu Syntax. Preliminary edition. Department of Linguistics, University of Illinois, Urbana-Champaign, Illinois. Bokamba, E.G. (1985). Verbal agreement as a non-cyclic rule in Bantu. African Linguistics: Essays in memory of M.W.K. Semikenke. Ed. by D.L. Goyvaerts. Amsterdam/Philadelphia: John Benjamins, 954. Bossong, G. (1985). Differentielle Objektmarkierung in den neuiranischen Sprachen. Tübingen: Gunter Narr Verlag. Bresnan, J. (1982). The Mental Representation of Grammatical Relations. Cambridge (MA): MIT Press. Bresnan, J. (1994). Locative inversion and universal grammar. Language 70, 1, 72-131. Bresnan, J. (2001). Lexical-Functional Syntax. Oxford: Blackwell Publishers. Bresnan, J. and J.M. Kanerva (1989). Locative inversion in Chicheŵa: A case study of factorization of grammar. Linguistic Inquiry 20, 1, 1-50. Bresnan, J. and S.A. Mchombo (1987). Topic, pronoun, and agreement in Chicheŵa. Language 63, 4, 741-782. Croft, W. (1990). Typology and Universals. Cambridge: Cambridge University Press. Dalrymple, M., R.M. Kaplan, J.T. Maxwell and A. Zaenen (eds.) (1995). Formal Issues in LexicalFunctional Grammar. Stanford (CA): CSLI Publications. Demuth, K. and S. Mmusi (1997). Presentational focus and thematic structure in comparative Bantu. Journal of African Languages and Linguistics 18, 1, 1-19. Dixon, R.M.W. 1994. Ergativity. Cambridge (MA): Cambridge University Press. Donohue, M. (this volume). Different subjects, different marking. Du Bois, J.W. (1987). The discourse basis of ergativity. Language 63, 4, 805-855. Falk, Y.N. (1983). Subjects and long-distance dependencies. Linguistic analysis 12, 245-270. Gazdar, G. (1981). Unbounded dependencies and coordinate structure. Linguistic Inquiry 12, 2, 155-184. Givón, T. (1976). Topic, pronoun, and grammatical agreement. Subject and Topic. Ed. by C. Li. New York: Academic Press, 149-188. Givón, T. (1979). On Understanding Grammar. New York: Academic Press. Harford, C. (1990). Locative inversion in Chishona. Current Approaches to African Linguistics. Ed. by J.P. Hutchison and V. Manfredi. Vol. 11. Dordrecht: Foris Publications, 137-144. de Hoop, H. and B. Narasimhan (this volume). Ergative case-marking in Hindi. Joshi, S. (1993). Selection of Subjects and Objects in Marathi. Doctoral dissertation, Stanford University, Stanford, CA. Keenan, E. (1976). Towards a universal definition of ‘subject’. Subject and Topic. Ed. by C. Li. New York: Academic Press, 303-333. Kimenyi, A. (1980). A Relational Grammar of Kinyarwanda. Berkeley (CA): University of California Press. Kimenyi, A. (1988). Passives in Kinyarwanda. Passive and Voice. ed. by M. Shibatani. Amsterdam/Philadelphia: John Benjamins, 355-386. Kinyalolo, K.K.W. (1991). The SPEC-Head Agreement Hypothesis in KiLega. Doctoral dissertation, University of California Los Angeles, Los Angeles, CA. Kroeger, P. (1993). Phrase Structure and Grammatical Relations in Tagalog. Stanford (CA): CSLI Publications. Lambrecht, K. (1994). Information Structure and Sentence Form. Cambridge: Cambridge University Press. Li, C. and S.A. Thompson (1976). Subject and topic: A new typology of language. Subject and Topic. Ed. by C. Li. New York: Academic Press, 457-489. Lødrup, H. (1999). Linking and Optimality in the Norwegian presentational focus construction. Nordic Journal of Linguistics 22, 205-230. Machobane, ‘M.‘M. (1995). The Sesotho locative constructions. Journal of African Languages and Linguistics 16, 2, 115-136.
FROM TOPIC TO SUBJECT MARKING
221
Manning, C.D. (1996). Ergativity: Argument Structure and Grammatical Relations. Stanford (CA): CSLI Publications. Meeussen, A.E. (1967). Bantu grammatical reconstructions. Africana Linguistica III Tervuren Belgique Royal Museee de L’afrique: Sciences Humaines Annales, 61, 79-121. Morimoto, Y. (2000). Discourse Configurationality in Bantu Morphosyntax. Doctoral dissertation, Stanford University, Stanford (CA). Morimoto, Y. (1999). An Optimality account of argument reversal. Proceedings of the LFG '99 Conference. Ed. by M. Butt and T.H. King. Stanford (CA): CSLI Publications. Morimoto, Y. (2003). Markedness hierarchies and optimality in Bantu. Perspectives in Linguistics: Papers in Honor of P.J Mistry. Ed. by R. Laury, G. McMenamin, S. Okamoto, V. Samiian and K.V. Subbarao. New Delhi: Indian Institute of Language Studies, 261-280. Morimoto, Y. (2006). Agreement properties and word order in comparative Bantu. ZAS Working Papers in Linguistics, Vol. 43, Berlin: ZAS,161-188. Ndayiragije, J. (1999). Checking economy. Linguistic Inquiry 30, 3, 399-444. Ura, H. (1996). Multiple Feature Checking. Doctoral dissertation, MIT, Cambridge, MA. Williams, E.S. (1977). Across-the-board application of rules. Linguistic Inquiry 8, 419-423. Woolford, E. (1987). An ECP account of constraints on across-the-board extraction. Linguistic Inquiry 18, 1, 166-171. Wunderlich, D. (2000). Predicate composition and argument extension as general options. A study in the interface of semantic and conceptual structure. Lexicon in Focus. Ed. by B. Stiebels and D. Wunderlich. Berlin: Akademie Verlag, 249-272.
JASON BROWN AND TYLER PETERSON
CHAPTER 10 Grammaticalization and Strategies in Resolving Subject Marking Paradoxes: The Case of Tsimshianic
1. INTRODUCTION This paper presents a case study of what we will call ergative/nominative paradoxes, which we claim are found in the Tsimshianic language family, but which are possibly found in other languages as well. Such paradoxes are said to arise when both nominative and ergative morphology is simultaneously indexed or related to the same subject. Although these languages manifest this subject marking paradox in different ways, we conjecture that the paradox itself is the result of an ergative system decaying into an accusative one, and that one strategy for languages to relieve the pressure of such a paradox is to develop new paradigms of differential subject marking. Morphological ergativity is typically manifested through the morphological case marking of noun phrases, reflecting their semantic role and grammatical relationship to the head in a clause. This can pattern independently of any of the various ergative/nominative-type operations in the syntax. Blake (2001:13) also describes a head-marking relationship where there can be a pronominal representation of core grammatical relations that are independent from representations transmitted through noun phrases. These representations are in the form of grammatical agreement on the verb predicate, and are a prevalent areal feature of the Northwest Coast language area where the Tsimshianic language family is situated.1 Tsimshianic displays a (split) ergative orientation in verbal agreement, where there is an inflectional paradigm that marks a transitive subject distinguishing it from an intransitive subject or object. In Tsimshianic, a subject marking paradox manifests itself when ergative agreement represents the same transitive subject that is morphologically marked with nominative case. Another predictable subject marking paradox, where ergative 1
The Tsimshianic languages are spoken on the northwest coast of Canada almost entirely in the province of British Columbia and in adjacent areas of the interior and Alaska. Examples not cited are from original fieldwork. Thanks go to Henry Davis, Martina Wiltschko, our language consultants, Mrs. Doreen Jensen, Mrs. Barbara Sennott, and the audiences at California State University, Fresno and at the Workshop on Differential Subject Marking (Nijmegen, 2004). Funding for this research was provided by grants from The Phillips Fund for Native American Research of the American Philosophical Society, awarded to each of the authors. The authors’ names appear in alphabetical order. All errors are our own.
223 H. de Hoop and P. de Swart (eds.), Differential Subject Marking, 223–245. © 2008 Springer.
224
JASON BROWN AND TYLER PETERSON
agreement simultaneously occurs with nominative agreement, arises in the Salish languages. These situations can be schematized as follows: (1) Ergative agreement → Ergative agreement →
Subject Subject
← ←
Nominative case Nominative agreement
We propose a preliminary hypothesis regarding this type of paradox: The Paradox Hypothesis, which states that the introduction of a nominative/ergative paradox will often be accompanied by a new development of differential subject marking. By comparing the ergative properties of the Tsimshianic languages, we can see that the ergative/nominative paradoxes arise from the unique trajectories involved in the decay of ergativity. This type of comparative analysis also illustrates the point that languages faced with an ergative/nominative paradox have certain strategies available for resolving the paradox: in the Tsimshianic languages a completely new paradigm emerges in a syncategorematic class of morphemes known as the connectives (Boas 1911; Dunn 1979a/b; Tarpent 1987; Mulder 1994; Peterson 2003, 2004). This paper is organized as follows: §2 presents a detailed comparative analysis of the agreement and connective systems in the Tsimshianic languages, where the nominative/ergative paradox in case and agreement is morphologically transparent and isolable in one half of the family, but relieved through a process of grammaticalization in the other half of the family, resulting in the emergence of a new morphological paradigm: the connectives (see also Peterson 2003, 2004). §3 then turns to a brief examination of other potential cases of differential subject marking by means of marking paradoxes or potential paradoxes. These cases include Halkomelem Salish and the Iranian languages, where agreement systems are potentially in conflict in marking the same subject. The last section, §4, will offer some speculations into the Paradox Hypothesis we propose, specifically, that these phenomena in fact represent a diachronic snapshot of how languages relieve the nominative/ergative paradox through different strategies in differential subject marking. 1.1 Agreement and multiple cases It is a fairly common property of languages to display case concord, or agreement. For instance, case marking can appear on nouns, and also on dependents to those nouns. Take for example Ancient Greek, which allows case marking on adjectives and determiners which are nominal dependents (Blake 2001):
GRAMMATICALIZATION AND STRATEGIES IN RESOLVING SUBJECT MARKING
225
(2) Ancient Greek case concord (from Plato, cited in Blake 2001:7) Ho aneksetastos bios ou biōtos the.NOM.SG unexamined.NOM.SG life.NOM.SG not livable.NOM.SG anthrōpō man.DAT.SG ‘The unexamined life is not liveable for man.’ Here, the case marked noun is bios, and the predicative adjective biōtos is marked for case in agreement with the noun. Dependents of nouns can also be marked for case independently of case agreement when there occur situations where dependents are marked for case, and there is also case agreement marked on the very same dependents. This is termed double case marking, and it is the outcome of two basic strategies that languages employ in marking case: the marking of dependents and the use of concord or agreement. Double marking is banned in Indo-European languages (Blake 2001:103-4), but Georgian provides a clear example: (3) Georgian (Mel’cuk 1986:69, cited in Blake 2001:102) sarel-ita man-isa-jta name-INS father-GEN-INS ‘with father’s name’ Finally, languages can make use of Multiple Case. As Blake (2001:107) notes, multiple case marking is not very common. Blake states that ‘Most but not all instances involve an inner layer of adnominal case plus an outer layer of adverbial case’ and that ‘in Warlpiri, for instance, a locally marked adjunct may take ergative case marking in a transitive clause (Hale 1982:266)’ (Blake 2001:107): (4) Ngarrka-ngku ka maliki ngurra-kurra yilya-mi PRS dog.NOM camp-ALL send-NPST man-ERG ‘The man is sending the dog to the camp.’ Ngarrka-ngku ka kuyu ka-nyi ngurra-kurra (-rlu) PRS meat.NOM carry-NPST camp-ALL(-ERG) man-ERG In (4), ‘the noun phrase in the role of destination is marked for allative case as one would expect. However, if a verb for carry is substituted for a verb meaning send, then it is possible to further mark the phrase for ergative’ (Blake 2001:107). While these languages have not followed similar trajectories in their case histories as the languages that will be described next, they set the stage for what is possible when it comes to marking case in more than one way in a single language and the interaction that case marking can have with case agreement. In the following section we will demonstrate what happens when items are paradoxically assigned case by two separate sources.
226
JASON BROWN AND TYLER PETERSON 2. SUBJECT MARKING AND GRAMMATICALIZATION IN THE TSIMSHIANIC LANGUAGES: THE ‘CONNECTIVES’
There are two subgroups within the Tsimshianic family, each containing two languages: the Interior Tsimshianic (henceforth IT) subgroup is made up of Gitksan and Nisga’a, which are similar enough to be considered to be dialects of the same language. The Coast Tsimshia (henceforth CT) subgroup is divided into Sm’algyax and Sgüüxs, which are also very similar to each other grammatically. Ergativity in the Tsimshianic languages is a prevalent feature, as they have been classified as not only strongly morphologically ergative but also syntactically ergative (Rigsby 1986, 1989; Tarpent 1982, 1987 for IT; and Dunn 1979 and Mulder 1994 for CT). The Tsimshianic languages have highly ergative properties of pronoun and agreement distribution. This arises when a transitive subject has a distinctive agreement/pronoun paradigm, which sets it apart from an intransitive subject or the object of a transitive, which in turn share the same agreement/pronoun paradigms. The pronoun/agreement paradigms are given in Table 1:2 Table 1. IT Agreement/Pronouns (Rigsby 1986; Tarpent 1987)
Ergative
Nominative
SG
PL
SG
PL
ni mi t
tip (mi)…sim t
-y’ -n -t
-m’ -sim’ -tiit
Consider the following examples with an intransitive verb in 6 and 7, and a transitive verb in 8 and 9: (6) yukw=hl limx-n PROG=CNC sing-2SG.NOM ‘You’re singing.’ (7) yukw=hl limx-y’ PROG=CNC sing-1SG.NOM ‘I’m singing.’ (8) yukw ni hlimoo-n PROG 1SG.ERG help-2SG.NOM ‘I’m helping you.’ 2
It is proposed that this arrangement defines an ergative/nominative split as opposed to an ergative/absolutive one (Franchetto 1990; Murasugi 1992; Bittner and Hale 1996; Ura 2000): suffixal agreement on the verb can represent both an intransitive and transitive subject (with an unrepresented object). When ergative morphology is triggered (i.e. ergative agreement in the transitive subjunctive – see below for details), the suffixal nominative agreement switches to representing the object in the presence of ergative agreement – there is no distinct agreement paradigm representing a transitive object (i.e. accusative or absolutive). Independent Pronouns are actually morphologically complex containing the nominative person markers plus a pronoun determiner.
GRAMMATICALIZATION AND STRATEGIES IN RESOLVING SUBJECT MARKING
227
(9) yukw mi hlimoo-y’ PROG 2SG.ERG help-1SG.NOM ‘You’re helping me.’ There are two canonical clause types, roughly defined by the mood and tense/aspect of the clause (and additionally by animacy in CT), which is also the locus of an ergative split in pronominal/agreement marking. The examples in (6) – (9) illustrate so-called Subjunctive clauses, where the predicate is preceded by the progressive aspect and the agreement morphology is fully ergative. In so-called Indicative clauses in (10) and (11), the predicate is the first element in the sentence and agreement is nominative: (10) hlimoo-i-y’ n’ii-n help-TR-1SG.NOM DET-2SG.NOM ‘I help/helped you.’ (11) limx n’ii-y’ sing DET-1SG.NOM ‘I sing/sang.’ Agreement and pronouns in CT pattern nearly identically to their counterparts in IT: the ergative series in CT have similar shape, and pattern correspondingly to those in IT as preverbal (often phonologically independent) ergative clitics. These are given in Table 2: Table 2. CT Agreement/Pronouns (Dunn 1979a)
Ergative
Nominative
SG
PL
SG
PL
n m t
tip m-…-sm t
-u -n -t
-m -sm -t
The nominative affixes also parallel closely the nominative series in IT, having a similar shape and distribution. However, CT also diverges from IT in several significant ways. Beginning with subjunctive clauses, as in IT, a pronominal subject in a transitive clause is marked with the preverbal ergative clitic, and a pronominal object is marked with the nominative suffix: (12) ada wil m way-u and then 2SG.ERG find-1SG.NOM ‘and then you found me.’
(Boas 1911:384)
228
JASON BROWN AND TYLER PETERSON
Like subjunctive clauses in IT, ergative agreement occurs with a transitive subject, as in (12). However, unlike IT, there is no object agreeing suffix unless it represents a pronoun (cf. (12)): (13) yágwa-ti naxnúu=da yúuti=a PROG-3.ERG listen to=CNC man=CNC ‘The man was listening to the woman.’ (14) hla n wil niidz=as Meli INCEPT 1SG.ERG COMP see=PNC Mary ‘I’ve just now seen Mary.’
hanák woman (Dunn 1979b:133) (Dunn 1979a:65)
As in IT subjunctive clauses, (15) shows that a nominative suffix functions as a pronoun, but unlike IT, (16) and (17) show that it cannot function as nominative agreement with an intransitive subject: (15) hla
dm
ba-n run-2SG.NOM ‘You’re about to run.’ (16) yagwa ba=a wan PROG run=CNC deer ‘The deer is running.’ (17) yagwa yawxg=as Ami PROG eat=PNC Amy ‘Amy is eating.’ INCEPT FUT
(Dunn 1979a:62) (Dunn 1979a:60) (Mulder 1994:99)
Also as in IT, in Independent transitive and intransitive clauses there is no nominative agreement on the predicate and a pronominal subject is a bound pronoun: (18) sa-yüü=s Marjorie hlioon CAUS-hide=PNC Marjorie fried bread ‘Marjorie hid the fried bread.’ (Stebbins 2003:97) (19) hla xúupl manxyáa=s üünal TEMP dark walk.up=PNC Arnold ‘Arnold walked up in the dark.’ (Dunn 1979b:134) There are further animacy and discourse conditions resulting in a rather complex distribution of pronouns and agreement in CT clauses, however the basic generalization holds: the distribution of pronouns in CT is ergative in subjunctive clauses, but nominative in indicative clauses. 2.1 Tsimshianic connectives A fundamental feature of all the Tsimshianic languages is the form, function and distribution of a class of morphemes that have been conventionally labelled as the
GRAMMATICALIZATION AND STRATEGIES IN RESOLVING SUBJECT MARKING
229
‘connectives’ (Boas 1911; Dunn 1979a,b; Rigsby 1986; Tarpent 1987; Mulder 1994). Connectives appear to mark nouns for their referential properties, number and definiteness, and have thus led to various proposals that treat them as determiners or argument markers (Beck 2002:49). However, in addition to encoding these semantic properties, connectives can be conditioned by factors including the grammatical or semantic role of the noun it precedes, as well as clause type. Peterson (2003, 2004) analyzes the Tsimshianic connective system through reconstructing them as the remnants of a morphological case system, as well as also uncovering the previously assumed lack of nominative agreement in CT, both of which are present in IT (cf. (15)-(19)). This was achieved by demonstrating that CT connectives are actually morphologically complex, the segmentation of which reveals an isomorphic relationship between CT and IT agreement/pronoun patterns, and determiner shape and distribution. It is claimed here that this is also the locus of a subject marking paradox in the Tsimshianic languages, where ergative agreement co-occurs with nominative case. This analysis is reviewed and placed in the context of the present discussion, thus offering the opportunity to observe the emergence of the nominative/ergative subject paradox in IT case/agreement marking, and its resolution in the CT connective system. 2.1.1 Interior Tsimshianic connectives Both Nisga’a and Gitksan connectives distinguish two noun classes that correspond to the grammatical distinction between proper and common nouns. Proper nouns are additionally marked for number: Table 3. Interior Tsimshian Connectives (Tarpent 1987)
Proper Noun Connective (PNC)
Common Noun Connective (CNC)
SG
PL
SG/PL
(=)t(=)
=tip
=hl
IT connectives are not sensitive to the grammatical or semantic function of the noun they mark, and this might be what led Tarpent (1988:2) and Beck (2002:51) to treat all connectives in Nisga’a as determiners in function, including the morpheme =s. Hunt (1993) attempts to reduce the apparent redundancy of doubly marking proper nouns by claiming that =s is in fact a type of case marking (as it is glossed in this paper), a claim motivated by the fact that, unlike the other connectives, =s is indeed sensitive to transitivity, clause type and the semantic role of the proper noun it precedes. For example, in indicative sentences, the distribution of this case marker appears to be ergative, in that it precedes A but not S or O. Consider the Gitksan examples:
230
JASON BROWN AND TYLER PETERSON
(20) gub-ə-(t)=[s (t)= John] =hl smax eat-TR-3=CASE PNC= John =CNC meat ‘John ate the meat.’ (21) *w’itx=[s t=John ] (cf. w’itx t=John) PNC-John come ‘John came.’ (22) hləmoo-yə-(t)=s (t)= Tom=[*s t=John ] PNC=John help-TR-3=CASE PNC= Tom ‘Tom helped John’ However, this pattern holds only in indicative clauses. In subjunctive clauses, =s precedes any proper noun argument that immediately follows the verb. This also extends to objects: when there is a pronominal subject such as in (25), =s will precede the object. Thus it can potentially appear before A, S or O. (23) yukw=hl litsxxw-(t)[=s (t)=John]SUBJECT PROG=CON read-3=CASE PNC=John ‘John is reading.’ (24) needii-t hlimoo-t[=s (t)=John]AGENT t=Peter NEG-3 help-3-CASE PNC=John PNC=Peter ‘John didn’t help Peter.’ (25) needii=təp gya’-(t)=[s (t)=John]OBJECT NEG-1PL see-3=CASE PNC=John ‘We didn’t see John.’ Hunt goes on to analyze =s ‘CASE’ as a type of generic case that is only assigned to nominals that are adjacent to a lexical case assigning head, but does not elaborate on its actual status in the grammar. Table 4 outlines the distribution of the proper noun =s CASE-marker in IT based on the examples (20)-(25): Table 4. Proper noun=s case-marking of semantic roles in IT
Indicative Subjunctive
Intransitive S
Transitive A A, O
One key generalization can be extracted from the IT examples above: in all cases =s ‘case’ accompanies post-verbal nominative agreement. This is also what sets up the subject marking paradox under examination: the distribution of =s ‘case’ appears to be ergative in indicative clauses but it occurs with nominative agreement, which cross-references the transitive subject in indicative clauses, and is absent in intransitive clauses because of the lack of nominative agreement with intransitive subjects. However, a neutral orientation of =s emerges in subjunctive clauses: it can mark any semantic role, regardless of the grammatical relation of the argument nominative agreement represents.
GRAMMATICALIZATION AND STRATEGIES IN RESOLVING SUBJECT MARKING
231
Ergative agreement is designated as such because in subjunctive clauses it distinguishes transitive subjects from intransitive subjects and objects, which are represented by nominative agreement. Given the link between nominative agreement and =s ‘case’ , can this generalization be extended to designate the function of =s as a case marker in IT? In other words, can =s be described as essentially nominative case because of its link to nominative agreement? There are two general issues with this possibility: first, unlike nominative agreement, =s does not stand in opposition to any other type of morphological case marking in IT. Secondly, how can we account for the neutralized distribution of =s in subjunctive clauses? The ability of =s ‘case’ to mark objects in a subjunctive environment would also challenge the typological generalization that there are no reported languages that have ergative agreement and also morphologically mark accusative case (Woolford 2000). This is exactly what would surface in (25) if we were to interpret =s as accusative case. Comrie (1978:340) notes that there are several languages that have an ergative/absolutive system for nominal case-marking while having a nominative/accusative system of verbal agreement, such as Warlpiri. However, the inverse is rare or non-existent. If =s case can indeed be analyzed as nominative case, the IT languages would represent this gap, where in subjunctive clauses, nominative case is accompanied by a system of ergative/nominative agreement. In addition, CT subjunctive clauses seem to challenge the typological generalization that a language will not have ergative agreement unless it also has nominative agreement (Woolford 2001:8). It will be demonstrated that CT does indeed have nominative object agreement, but that it has undergone fusion with the other post-verbal elements such as the determiner system and historical case. In order to explore this hypothesis, we will now turn to the Tsimshianic system of connectives, and an examination and analysis of which will later be juxtaposed with the agreement system just presented. What will be pertinent to the following discussion is the agreement alternations associated with the indicative and subjunctive clause types, specifically third person subject and object nominative agreement – why IT has it and CT lacks it – and how they interact with the connective system and the yet to be determined =s ‘case’ marker in IT. Before discussing these issues further, let us turn to connectives \ in CT. 2.1.2 Coast Tsimshian connectives CT connectives are phonologically similar to those in IT in a number of respects: they also encliticize to the preceding word and distinguish between proper and common nouns (but not for number). However, unlike their IT counterparts, CT connectives are sensitive to the semantic role of the nominal they mark and clause type. This sets up a considerably larger paradigm of connectives, displayed in Table 5:
JASON BROWN AND TYLER PETERSON
232
Table 5. CT plain connective systems (Dunn 1979b; Mulder 1994)
A S O
Sm’algyax Proper Noun (PNC) =as/=dit =as =at
Common Noun (CNC) =a/=da =a
Sgüüxs Proper Noun (PNC) =as/=dit =as =it
Common Noun (CNC) =i/=di =i
The marking of an intransitive or transitive subject or object for both proper or common nouns is naturally subject to the transitivity of the clause. With proper nouns, a nominative orientation in connective marking is outlined in the indicative clauses in (24) and (25): Both the transitive and intransitive subject are marked with =as, while the transitive object is marked with =at. (26) CT: Sm’algyax niic=(a)s Nadine=(a)t Isabelle see=PNC Nadine=PNC Isabelle ‘Nadine saw Isabelle.’ (Stebbins 2001:19) (27) nah ts“lm-”wiihawtg=as Madzi da na-waab-u PST into.from-cry=PNC Margie PREP POSS-house-1SG ‘Margie came to my house crying.’ (Mulder 1994:57) In both common and proper noun agent-marking connectives, there are two forms available, the choice of which depends on the clause type. The aspect marker yágwa (PROG) in (28) and (29) triggers subjunctive clause ergative agreement with the transitive subject. The selection of the proper noun connective =dit marking ǘünal ‘Arnold’ and common noun =di marking ól ‘bear’ is dependent on this agreement relation: (28) CT: Sgüüxs yágwa-t bí’i’ ghl=dit ǘünal sá’awansk PROG-3 tear=PNC Arnold paper ‘Arnold is tearing the paper.’ (Dunn 1979b:133) óli-i hoon (29) yágwa-ti níis=di PROG-3 glare at=CNC bear=CNC fish ‘The bear glared at the fish.’ (Dunn 1979b:133) Using common nouns as an example, the transitive subject is marked distinctly from the intransitive subject and transitive object, yielding an ergative orientation in connective marking:
GRAMMATICALIZATION AND STRATEGIES IN RESOLVING SUBJECT MARKING (30) dzagha dá’uhl=i y’axw t’ei across across go=CNC man ‘A man went across the inlet.’
233
lu ‘asdi nak on both sides (Dunn 1990:2)
On the other hand, the tense/aspect complex nah (PST) hla (PFV) in (31) heads an indicative clause, hence no ergative agreement with the transitive subject and the PNC =as and CNC =a in (32) surface. (31) CT: Sm’algyax nah hla dzab=as Norman na-homework PST PFV do=PNC Norman POSS-homework ‘Norman has just finished his homework.’ (Mulder 1994:115) (32) nah t’uus=a y’uut=a hanak” PST push=CNC man=CNC woman ‘The man pushed the woman.’ (Dunn 1979a:63) (32) and (33) demonstrate the neutral distribution of CNC =a marking A, S or O: (33) yagwa hadiks=a sts’ool da ts’m t’aaks PRS swim=CNC beaver PREP in pond ‘A beaver is swimming in the pond.’ (Dunn 1979a:63) What emerges from these examples is four unique groupings, given in Table 6, of the semantic roles from the juxtaposition of the proper/common noun distinction and clause type: Table 6. Distribution of Semantic Roles in CT Connectives
Indicative Subjunctive
PNC
CNC
A/S:O (nominative) /=as/A/S /=at/O A:S:O (contrastive) /=dit/A /=as/S /=at/O
A/S/O (neutral) =/a/A/S/O A:S/O (ergative) /=da/A /=a/S/O
What is notable in CT is that certain interactions of clause type and transitivity yield typologically unusual morphological groupings or pivots of the semantic roles that cannot be adequately characterized as either ergative or nominative: CNCs pattern ergatively in subjunctive clauses, while making no distinction in indicative clauses. The distribution of PNCs somewhat inversely ‘mirrors’ the distribution of the CNCs: indicative clause connectives pattern nominatively while subjunctive clauses display a tripartite distinction, marking each semantic role with a different connective. However unusual, a system such as this may indeed be captured by the generalization that in indicative clauses the proper noun nominative and common noun neutral groupings simply morphologically distinguish the Agent in subjunctive
234
JASON BROWN AND TYLER PETERSON
clauses, while leaving the Subject-Object relations intact: A/[S:O] → A:[S:O] and A/[S/O] → A:[S/O]. These alternations would be supported by the general ergative split between indicative and subjunctive clauses in both CT and IT. While this functional explanation is tenable, it will be demonstrated in the next section that the source of this complexity – and ultimately the emergence of the connective system itself in CT – is the result of CT trying to relieve the ergative/nominative paradox that is still isolable in IT. In order to show this, we first review an analysis of Peterson (2003, 2004) in reconstructing the patterns of agreement and connectives in both CT and IT. 2.2 The decomposition of CT connectives: =s ‘CASE’ and subject marking Dunn (1979b), Stebbins (2001) and Peterson (2003, 2004), following observations made by Boas (1911:354-59), suggested that CT connectives may be morphologically complex, analyzing them as a series of elements containing one to three positions and listing their (semantic) function in any given clause: s d (32) = V t (where V = {i, a}) ∅ ∅ Beginning with the most complex connective in CT, the proper noun connective =dit marks A in a transitive clause: rules of stop voicing in Tsimshianic render the din the decomposed form d-V-t as voiceless underlyingly, and only voiced on the surface: /t-V-t/. From this, it is possible to relate the connective-initial ‘t-’ to the pattern of 3rd person object agreement (marked on the predicate by the nominative suffix), and connective-final ‘-t’ to the proper noun determiner (now re-glossed ‘PN.DET’) as marking the transitive subject, the same of which is found in a corresponding subjunctive IT clause: (34) CT: /dit/ → /t-i-t/ yagwa=ti t’uus=dj-i [-t PROG=3 push=3-? -PN.DET ‘John is pushing Mary.’ (35) IT: (Gitksan) yukw=ti hlimoom-(tj)=s=[(t) PROG-3 help-3=CASE=PN.DET ‘John is helping Mary.’
Dzoni]=at Melij John=PN.DET Mary (Dunn 1979b:67) Johni]=t Maryj John=PN.DET Mary
Comparing (34) with (35) illustrates how in IT the post-verbal affix/clitic complex made up of suffixal nominative agreement -t, the enclitic =s CASE, and enclitic determiner =t marking the proper noun transitive subject might have undergone fusion in CT, the result of which is the post-verbal connective. This reanalysis, schematized in (36), also implies the isomorphic relation of the CT connective-internal /-V-/ segment, /-i-/ (glossed as ‘?’ for the moment), to the IT =s
GRAMMATICALIZATION AND STRATEGIES IN RESOLVING SUBJECT MARKING
235
CASE marker. This process of fusion, and the implications of relating /-i-/ to =s will be discussed below.
(36) IT: [VERB]–[AGR]=[CASE]=[DET] CT: [VERB]=[AGR.CASE.DET] In transitive indicative clauses the proper noun connective =(a)s marking a transitive subject, and the connective =(a)t marking a transitive object can be decomposed into their component segments: /(a)-s/ and /(a)-t/ respectively. Parallels in position are found with the IT =s CASE marker and the proper noun determiner =t, which can be observed marking the proper noun object: (37) CT (Sm’algyax): /(a)s/ → /s/; /(a)t/ → /t/ niic=(a)s Nadine=(a)t Isabelle see=CASE Nadine=PN.DET Isabelle ‘Nadine saw Isabelle.’ (Stebbins 2001:19) (38) IT (Nisga’a) hlimoo-yə-(t)=s (t) Tom=t Mary help-TR-3=CASE PN.DET Tom=PN.DET Mary ‘Tom helps/helped Mary.’ (Tarpent 1987) Stebbins (2001) originally transcribed (37) as niic=s Nadine=t Isabelle, and in fact, transcriptions of CT clauses often omit the pre-consonantal ‘a’ in the connectives ([(a)-C]), but never a post-consonantal (or word-final) ‘a’ ([C-a]). This may be adduced as preliminary evidence that the post-consonantal ‘a’ in all of the CT connectives may be analyzed as a common noun determiner (now re-glossed ‘CN.DET’), based on the isomorphy with the CN.DET ‘hl’ in IT, and the fact that it is phonetically stable. The function of the pre-consonantal ‘a’ , on the other hand, may be epenthetic in order to break up consonant clusters, a common phonological process in the Tsimshianic languages. Under this reanalysis, both CT and IT mark the transitive subject in indicative clauses with =s. However, CT appears to lack the transitive subject (nominative) agreement found in IT transitive indicative clauses, as well as the proper noun determiner. Both Rigsby’s transcriptions of Gitksan and Tarpent’s of Nisga’a frequently omit both the subject nominative agreement and the proper noun determiner (denoted with bracketing), thus exactly paralleling the same construction in CT. Compare (39) with (40) and (41): (39) CT: niic=s Nadine=t Isabelle (40) IT (Gitksan) hlimoo-yə-(t)=s (t)=Tom t=Mary help-TR-3=CASE PN.DET=Tom PN.DET=Mary ‘Tom helped Mary.’
236
JASON BROWN AND TYLER PETERSON
(41) IT (Nisga’a) hlimoom-ə-(t)=s (t)=Tom t=Mary help-TR-3=CASE PN.DET=Tom PN.DET=Mary ‘ Tom helped Mary.’ (Tarpent 1988:108) So far, we’ve been able to plausibly reconstruct and track the distribution of =s and the proper noun determiner in a transitive Tsimshianic clause: =s always marks a proper noun transitive subject, while the =(a)t marking an object in CT is a cognate of the proper noun determiner in IT =t, as well as the connective-final [t] (/tV-t/) reconstructed form the subject connective =dit. Because =t marks both the subject and object, we can forward the generalization that there is no distinct morphological marking of an object in a Tsimshianic clause (ruling it out as a potential accusative/absolutive case marker): it is simply marked with a determiner, the distribution of which is not expected to be sensitive to grammatical or semantic roles. What will become crucial in the analysis to follow is the observation that 3rd person nominative agreement and =s CASE appear to be in complementary distribution in the examples above. Based on this, a preliminary claim can be made that either agreement or morphological case can mark a transitive subject, but not both. By treating =s as the morphological marking of case, we should be able to see a pivot emerge in comparing the marking of (proper noun) transitive subjects with intransitive subjects. In intransitive subjunctive clauses, =s marks the proper noun subject in both CT and IT, again in complementary distribution with third person nominative agreement (and the optional proper noun determiner in IT): CASE
(42) CT (Sgüüxs) hla xúupl manxyáa=s üünal TEMP dark walk up=CASE Arnold ‘Arnold walked up in the dark.’ (Dunn 1979b:134) (43) IT (Nisga’a) Mary yùkw=hl yuuxkw-(t)=s (t) PROG=CN.DET eat-3=CASE PN.DET Mary ‘Mary is eating.’ (Tarpent 1988:105) The reanalyzed CT connective =(a)s marking the agent in a transitive clause also marks the subject of an intransitive clause, yielding a nominative alignment. However, this pattern does not extend to IT indicative intransitive clauses, where there is no =s CASE marking the subject: (44) CT (Sm’algyax) nah ts’ lm-w’iihawtg=(a)s Madzi da na-waab-u PST into.from.side-cry=CASE Margie PREP POSS-house-1SG ‘Margie came to my house crying.’ (Mulder 1994:57)
GRAMMATICALIZATION AND STRATEGIES IN RESOLVING SUBJECT MARKING (45) IT (Nisga’a) Mary yúuxkw t PN.DET Mary eat ‘Mary ate.’
237
(Tarpent 1988:107)
This is where IT and CT diverge: recall that in IT transitive indicative clauses the distribution of the =s CASE marker appears to be ergative, in that it precedes A but not S or O (cf. Table 4). In CT transitive indicative clauses, a nominative S/A grouping emerges as both an intransitive and transitive subject are marked with =s (as reconstructed in comparing (42) with (44)). What is of interest is the distribution of =s in a subjunctive clause: IT marks both an intransitive and transitive subject with =s, and while CT also marks an intransitive with =s, it was unclear whether =s marks a transitive subject. The source of this ambiguity is the proper noun connective marking a transitive subject in a subjunctive clause in CT, =dit (VERB-/ti-t/), where the connective-internal /-V-/ segment, /-i-/, would have to be reconstructed as =s in order to be isomorphic with IT =s (or VERB-t=s=t). If it is not reconstructed as =s, then the =s CASE marking of a transitive subject is lacking from CT transitive subjunctive clauses. The difficulties in such a phonological reconstruction are obvious. It was claimed above that the apparent ergative distribution of =s CASE in IT is superficial, as =s CASE can be observed as accompanying nominative agreement. The lack of nominative agreement in indicative intransitives in IT was due to independent reasons in the syntax. This can be characterized as ‘agreement-linked’ case. Because nominative agreement in IT shifts between the transitive subject (in indicatives), the intransitive subject (in subjunctives), and the object (in subjunctives with a pronominal subject), =s CASE may be designated through this link as being nominative. This is adequate in characterizing all cases in IT except for transitive subjunctive clauses with a lexical subject, where ergative agreement and =s CASE both mark the subject. In these examples a nominative-ergative ‘paradox’ would arise in case and agreement: the lexical subject would be marked with both nominative-linked =s CASE and ergative agreement: (46) IT: (Gitksan) hlimoom-(tj )=s=(t) Lucyi=t Maryj yukw=ti PROG-3.ERG help-3.NOM=CASE=PN.DET L.=PN.DET M. ‘Lucy is helping Mary.’ There are two potential solutions to resolving this paradox: the simplest involves reinterpreting =s CASE as ergative. This is immediately problematic because =s can mark an intransitive subject in subjunctive clauses. The other solution is reexamining the agreement paradigms in both IT and CT. Note that the third person agreement is the same in both the ergative and nominative paradigms: /-t/ (with the exception of the plural in IT, /-tiit/). Considering that this subject marking paradox only arises in ergative-nominative agreement with a lexical subject, it is proposed
238
JASON BROWN AND TYLER PETERSON
that ergative agreement is actually neutralized in this environment, and what in fact surfaces is doubled (post- and pre-verbal) nominative agreement: hlimoom-(ti )=s=(t) Lucyi=t Maryj (47) yukw=ti PROG-3.NOM help-3.NOM=CASE=PN.DET L.=PN.DET M. ‘Lucy is helping Mary.’ Evidence for doubled nominative agreement can be adduced from what were previously classified as exceptional cases of agreement, involving subjunctive transitive clauses with a third person plural pronominal subject: stil-tiiti (48) needii-ti NEG-3.NOM accompany-3PL.NOM ‘They didn’t accompany John.’
t=Johnj DET=John
Equipped with this analysis, we can now re-examine the CT transitive subjunctive connective =dit (/t-i-t/), as this is the parallel locus of the nominativeergative paradox in CT. Notice how the question of reconstructing the connectiveinternal /-V-/ segment, /-i-/, as =s in IT is now somewhat moot. The connectiveinitial dit (/t-i-t/) reconstructed as nominative agreement is now co-referent with the subject and not the object, and the preverbal agreement, formerly ergative, is now also nominative. This revises our analysis of (34) and (35) above. Compare (47) with (49): (49) CT (Sm’algyax): /dit/ → /t-i-t/ t’uus=di-i-t Dzoni=(a)t Melij yagwa=ti PROG=3.NOM push=3.NOM-CASE-PN.DET J.=PN.DET M. ‘John is pushing Mary.’ (Dunn 1979b:67) Under this analysis, CT has employed a morphological strategy of deleting the =s CASE marker in order to not only ease the nominative-ergative paradox, but also to conform to the strong tendency for languages not to doubly mark arguments with both case and agreement (Woolford 2000). Doubled nominative agreement still surfaces as in CT, but /-i-/ can be analyzed as simply epenthetic in order to break up the agreement and proper noun determiner /-t/ obstruent cluster, thus saving the necessity of developing a language-internal phonological reconstruction of /-s-/ > /-i-/. 3. OTHER CASES OF SUBJECT MARKING PARADOXES With the case and agreement patterns of Tsimshianic outlined, we will now turn to speculate about other case/agreement patterns found in languages. A case similar to the Tsimshian patterns analyzed above of can be found in the Tupí-Guarani language family. As in the Tsimshianic languages, historical fluctuations in case and agreement morphology can be observed in constructions involving the proto
GRAMMATICALIZATION AND STRATEGIES IN RESOLVING SUBJECT MARKING
239
morpheme *–a, which has ‘diffused’ from what was probably a case marker to a variety of functions across the Tupí-Guarani family, such as marking the head of a propositional phrase, possession in genitive constructions, tense, adjectival and equative constructions (see Cabral 2001 for details; Rodrigues 2001). Relevance to the present discussion can be found in examining the synchronic behaviour of –a in the Tupí-Guarani language Jo’é. Cabral and Rodrigues designate –a as an ‘argument marker’ (and gloss it as such), however, one that is sensitive to the grammatical role of the argument it marks. In (50a) –a marks a transitive object, and an intransitive subject in (50b), thus yielding an absolutive orientation: (50) Jo’é a. tapiɁír-a a-jukjé kwahé tapir -ARG 1-kill yesterday ‘I killed a tapir yesterday.’ b. pitãg-a kã o-bohẽb little.child-ARG COL 3-arrive ‘The little children arrived.’
(Cabral 2001:151) (Cabral 2001:151)
These cases represent the inverse of the Tsimshianic patterns: while the TupíGuarani languages have a split ergative agreement system, these examples are of the nominative variety, and thus absolutive arguments in Jo’é are marked –a while being simultaneously represented with nominative agreement (in the intransitive case in 50b.) on the predicate. This sets the stage for the kind of functional instability observed with the =s case marker in Tsimshianic, which lost its ability to distinguish grammatical roles in subjunctive clauses (cf. 23 - 25). This functional instability is manifested by the generalized function of –a across the Tupí-Guarani family, and can also be observed in the following Jo’é examples in (51) and (52), where –a is freely optional: (51) a-esák=potát de ∅-bebỳt(-a) CNT-(mother’s)son/daughter(-ARG) 1-see=want 2.POSS ‘I want to see your mother’s son/daughter.’ (Cabral 2001:152) (52) ji e ∅-pajuát(-a) o-jukjé 1 1 CNT-husband(-ARG) 3-kill ‘My husband killed.’ (Cabral 2001:152) Whereas the subject marking paradox finds resolution in the creation of a new paradigm in the Tsimshianic languages, we suggest that the Tupí-Guarani case marker –a chooses an alternative strategy in diffusion: the specific functional properties of –a as a case marker have diffused from a case marker into what can be described as a generic relational marker across the Tupí-Guarani family. The schema in (1) highlights the potential for subject marking paradoxes to arise independently of the case marking patterns observed above in the Tsimshianic and Tupí-Guaraní families, manifesting itself instead in the agreement system of a
240
JASON BROWN AND TYLER PETERSON
language. One such potential case can be found in Upriver Halkomelem (Coast Salish). Halkomelem exhibits a split-ergative system of morphologically marking agreement (Gerdts 1988, Wiltschko 2000, 2006) conditioned by person and clause type. 1st and 2nd person subjects have a nominative orientation in subordinate and matrix clauses. 3rd person is also nominative in subordinate clauses, but there is distinct morphological agreement that marks 3rd person transitive subjects in matrix clauses, yielding an ergative orientation (Gerdts 1988). This system is illustrated in Table 7: Table 7. Split Ergativity in Halkomelem (adapted from Wiltschko 2000:251)
A 1/2 P 3P
Subordinate Matrix Subordinate Matrix
S
O
NOM ERG
ACC ABS
The matrix intransitive and transitive examples of 53a.-c. show that 3rd person S and O agreement is morphologically unmarked, while the transitive subject in example 53c. is marked with –es. This outlines an ergative/absolutive pivot.: (53) Halkomelem: 3rd person agreement (Wiltschko 2000; Galloway 1980) a. yó:ys-tsel b. yó:ys-∅ work-3.S work-1SG.S ‘I work.’ ‘S/he works.’ c. máy-t-∅-tsel d. may-th-óx-es help-TRANS-1SG.O-3.S.ERG help-TR-3.O-1SG.S ‘I help him.’ ‘S/he helps me.’ There are two ‘layers’ of agreement in Halkomelem: one can be characterized as predicative agreement which is suffixed to the verb, and another called subjunctive agreement which doubles the subject and is suffixed to the preverbal auxiliary. (54) Halkomelem (Galloway 1993:186 Wiltschko 2000) a. éwe tsel lí-l tl’íls-th-òmè NEG 1SG.S AUX-1SG.SS.NOM want-TR-2SG.O ‘I don’t like you.’ b. éwe (∅) lí-s í:mex NEG AUX-3.SS.NOM walking ‘He is not walking.’ c. éwe (∅) lí-s tl’ils-th-óx-es NEG AUX-3.SS.NOM want-TR-1SG.O-3.S.ERG ‘He doesn’t like me.’
GRAMMATICALIZATION AND STRATEGIES IN RESOLVING SUBJECT MARKING
241
The example in (54c) illustrates the situation where both ergative predicative agreement and nominative subjunctive (auxiliary) agreement both mark the transitive subject: He doesn’t like me. This presents a system of ergative predicative agreement, which is prevalent throughout the family, faced with the competing marking of the nominative subjunctive agreement. The result is the tension between these competing systems being absorbed by a third system of subject marking, the subject clitic series. This provides a type of evidence for the Paradox Hypothesis stated above. The larger idea is that ergativity is the result of subject suffixes being replaced by subject clitics in a different position (Davis 1999, Kroeber 1999, Koch 2005). As Koch (2005) has stated, ‘When we travel through the Salish language family, from Nhlkepmxcin (Thompson River Salish) in the Northern Interior of the Salishan area, to Lushootseed in the Central (Coast) Salish area, subject marking in transitive clauses shifts from entirely suffixal to entirely composed of clitics (Davis 1999, Kroeber 1999).’ Thus the breakdown of ergativity in Central Coast Salish can be attributed to this replacement of suffixes to clitics. The interesting thing about the situation in Halkomelem is the added pressure that what Wiltschko (2006) terms the ‘negative-ergative paradox’ places on the system, and which we speculate is an additional diagnostic of ergative decay. Other relevant examples include the development of the present-day Iranian languages from Old Iranian, discussed at length by Payne (1979). Payne has shown how the participial construction in Old Iranian has evolved into an ergative-based system in some of the daughter languages. In some daughter languages, the shifting of subject marking has resulted in a typologically unstable tripartite case system, which has then collapsed back into the original accusative system of Old Iranian due to the levelling of subject marking. These types of pressures on differential subject marking allow us to draw parallels with the situation in Tsimshianic, where we are left to speculate about the stability of such systems. Finally, there is the interesting case of historical change surrounding the presentday passive in Polynesian. As explained by Jacobs (1976:118), ‘Students of the Polynesian sub-branch of Austronesian have argued that Proto-Polynesian was an accusative language whose passives have decayed into ergative constructions in some now-ergative daughters. Other Polynesianists take the opposite tack, contending that the modern English-like passives in some daughters are innovations, and that Proto-Polynesian, perhaps even Proto-Oceanic, had a Philippine-like focus system.’ While our knowledge of the situation in Polynesian is limited, it is certainly worth exploring the possibility that the decay from one system into another in Polynesian had something to do with differential subject marking and paradoxes similar to those we have seen already. 4. DISCUSSION AND SUMMARY Little is known about how languages acquire or lose their syntactic pivots marking either morphological or syntactic ergativity (Trask 1979:205). These typically
242
JASON BROWN AND TYLER PETERSON
include some shift in the realization of passives, nominalization, discourse factors or even phonological considerations. This study presents an innovation in how languages that display degrees of morphological ergativity obscure their nominative vs. ergative alignment in trying to cope with some middle stage in their development, where both ergative and nominative forms converge on the same subject. The extreme result of this was the four-way split in the distribution of CT connectives (cf. Table 6). Comrie (1978:379) notes in his examination of the diachronic function of ergative case-marking that ‘there are indeed possibilities for a natural development from nominative/accusative to ergative/absolutive, but equally … a natural development in the opposite direction’ . The dynamics of morphosyntactic change in these languages can now be placed in the much broader context of grammaticalization. Take for instance the development of genitive marking on participial subjects in Finnish. As Timberlake (1977) has pointed out, the phonological neutralization of case morphemes in Finnish led to a surface ambiguity between nominative and genitive subjects. In the older Finnish, participial subjects were assigned ‘object’ case: partitive, accusative, or nominative. The participial subject was reanalyzed as genitive, and then the reanalysis spread. The result of this was the loss of nominal-type inflection, and what Hopper and Traugott (2000:104-105) term ‘decategorialization’. What we can see in this case is the development of a decategorialized morpheme through the reanalysis of a paradigm of subject marking. Although it does not exactly parallel the case studies presented above, it does illustrate the point that the reorganization of case systems often produces changes in differential subject marking. Returning to the Tsimshianic languages, despite the implicit nominative designation of =s CASE, through its agreement-linked relationship with nominative agreement, the ultimate goal is not necessarily to define the ultimate function of =s CASE as a morphological case marker in the sense of being ‘ergative’ , ‘nominative’ or ‘accusative’ etc. What is of ultimate interest is to track the merging of the remnants of a case system with other local morphemes such as agreement and determiners in terms of comparing two genetically related languages, and what drives this grammaticalization process. For example, in IT all of these morphemes affix/encliticize to the predicate, which can be considered as a staging ground for the grammaticalization that appears to have already taken place in the verb-encliticizing CT connectives. However, once decomposition and reanalysis is applied to the Tsimshianic connective systems, a problem is presented in accounting for the typological gap that originally motivated the generalization that ergative agreement is parasitic on ergative case: there are no reported instances of ergative agreement in a nominative/accusative case system (Anderson 1977, Dixon 1994). This was represented in the ergative/nominative subject marking paradox in (44), where a subject is represented by ergative agreement but also marked by nominative agreement-linked =s CASE. The resolution to this paradox is grammaticalization in CT, and the neutralization of ergative agreement (replaced by nominative agreement) in subjunctive clauses in both CT and IT. An important side-effect of this analysis is the readjustment to the ergative split in Tsimshianic. Where the previous split was aligned to clause type, ergative agreement is now only found in 1st
GRAMMATICALIZATION AND STRATEGIES IN RESOLVING SUBJECT MARKING
243
and 2nd person subjunctive clauses, while 3rd person is nominative in both clause types (where it is accompanied by nominative-linked =s CASE). The Tsimshianic languages also confirm the generalization that if a language has agreement at all, it will have agreement with nominatives (Woolford 2000), as this is borne out in Tsimshianic transitive indicative clauses and subjunctive clauses with a lexical subject. The examples in (42) and (43) have ergative subject pronouns, therefore, object agreement is expected and does indeed occur. Woolford (2000:4) notes the generalization that ergative agreement has not been observed occurring in languages with morphologically marked accusative case. Rather, ergative agreement can occur in languages with a nominative/accusative case system, as long as the accusative is not morphologically marked. The Tsimshianic languages now conform to both of these generalizations under this analysis: there is no distinct morphological accusative case. Objects are morphologically =s CASE-marked, but only as linked to nominative agreement (cf. (49)-(51)). Jo’é presents the inverse of this: absolutive case occurs in a nominative agreement system, and thus the paradox manifests itself in intransitives which agree with the verb but are marked with what would historically be absolutive case. Affixes typically undergo phonological processes of deletion, reduction and fusion with other neighbouring affixes. One of the effects of this is the reduction in the differentiation of morphological case distinctions (Blake 2001:169). Languages can make use of different ‘repair’ strategies to compensate for phonological changes and erosion, such as word order or adpositions to mark syntactic relations. Tsimshianic offers another option, represented by a kind grammaticalization process through fusion and neutralization – the result of which is the connective system. This paper approaches the issue of grammaticalization and differential subject marking. In particular, special attention has been given to what results from the reanalysis of case and agreement phenomena. It was conjectured that the reorganization of case and agreement morphology into paradoxical arrangements, what we have termed ‘ergative/nominative’ paradoxes, will be accompanied by the emergence of a new paradigm of differential subject marking. The Tsimshianic data outlined above illustrates this: when ergative/nominative paradoxes arise, some strategy must be employed to relieve the pressure of doubly-marking arguments. The strategy seen here is to take the functional load off of a single paradigm of subject marking by creating another. It should also be noted that not all cases of multiple case/agreement differential subject marking give rise to subject marking paradoxes. For example, Donohue (this volume) offers a plausible structural explanation for potential mismatches in case and agreement paradigms in Tukang Besi. While there have been many works dedicated to the evolution of ergativity (cf. Anderson 1977, Dixon 1994, Dryer 1986, Trask 1979), there has been little discussion of the competing trajectories of nominative vs. ergative systems. This paper seeks to fill this gap by offering a preliminary research agenda.
244
JASON BROWN AND TYLER PETERSON REFERENCES
Anderson, S.R. (1977). On mechanisms by which languages become ergative. Mechanisms of syntactic change. Ed. by C.N. Li. Austin: University of Texas Press, 317-363. Beck, D. (2002). Tsimshianic from a Central Northwest areal perspective: I*. ICSNL XXXVII: The ThirtySeventh International Conference on Salish and Neighbouring Languages. Ed. by C. Gillon, N. Sawai and R. Wojdak. Vancouver: University of British Columbia Working Papers in Linguistics (UBCWPL), 35-60. Bittner, M. and K. Hale (1996). Ergativity: toward a theory of heterogenous class. Linguistic Inquiry 27, 531-604. Blake, B.J. (2001). Case (2nd ed.). Cambridge: Cambridge University Press. Boas, F. (1911). Tsimshian. Handbook of American Indian languages, Bulletin No. 40, Pt. I, 287-422. Cabral, A.S. (2001). Observações sobre a história do morfema –a de família Tupí-Guaraní.’ Des noms et des verbes en tupi-guarani: état de la question. LINCOM Studies in Native American Linguistics 37. Ed. by F. Queixalós. LINCOM EUROPA, 135-162. Comrie, B. (1978). Ergativity. Syntactic Typology. Ed. by W. Lehmann. Austin: University of Texas Press, 329-94. Davis, H. (1999). Subject inflection in Salish. UBCWPL volume 1, current research on language and linguistics. Ed. by M.Caldecott, S. Gessner and E. Kim, 181-240. Dixon, R.M.W. (1994). Ergativity. Cambridge: Cambridge University Press. Dryer, M. (1986). Primary objects, secondary objects, and antidative. Language 62, 808-845. Dunn, J.A. (1979a). A Reference Grammar for the Coast Tsimshian Language. Canadian Ethnology Service, Mercury Series Paper No. 55. Ottawa: National Museum of Man. Dunn, J.A. (1979b). Tsimshian Connectives. International Journal of American Linguistics 45, 131-140. Dunn, J.A. (1990). The government of oblique/local grammatical relations in Lower Tsimshian. Handout from a paper presented at Conference on American Indigenous Languages, New Orleans, 1990. Franchetto, B. (1990). Ergativity and nominativity in Kuikúro and other Carib languages. Amazonian linguistics: Studies in lowland South American languages. Ed. by D.L. Payne. Austin: University of Texas Press, 407-427. Galloway, B. (1980). The Structure of Upriver Halkomelem, A Grammatical Sketch and Classified Word List for Upriver Halkomelem. Sardis (B.C.): Coqualeetza Education Training Center. Galloway, B. (1993). A grammar of Upriver Halkomelem. Berkeley: University of California Press. Gerdts, D.B. (1988). Object and absolutive in Halkomelem Salish. New York: Garland. Hale, K.L. (1982). Some essential features of Warlpiri verbal clauses. Working Papers of the Summer Institute of Linguistics – Australian Aborigines Branch. A6, 217-315. Hopper, P. and E. Traugott (2000). Grammaticalization (2nd ed.). Cambridge: Cambridge University Press. Hunt, K. (1993). Clause Structure, Agreement, and Case in Gitksan. Unpublished Ph.D. dissertation, University of British Columbia. Jacobs, R.A. (1976). A passive continuum in Austronesian. Papers from the parasession on diachronic syntax. Ed. by S.B. Steever, C.A. Walker and S.S. Mufwene. Chicago Linguistic Society, 118-125. Koch, K.A. (2005). Double subject marking in NÅeʔkepmxcin. Papers for the fortieth International Conference on Salish and Neighbouring Languages. (UBCWPL) vol. 16, 125-139. Kroeber, P. (1999). The Salish language family: reconstructing syntax. Lincoln: University of Nebraska Press. Mel’cuk, I.A. (1986). Toward a definition of case. Case in Slavic. Ed. by R.D. Brecht and J.S. Levine. Columbus: Slavica Publishers, 35-85. Mulder, J. (1994). Ergativity in Coast Tsimshian Sm’algyax. University of California Publications in Linguistics 124. Berkeley: University of California Press. Murasugi, K. (1992). Crossing and nested paths: NP movement in accusative and ergative languages. Ph.D. dissertation, MIT. Payne, J.R. (1979). Transitivity and intransitivity in the Iranian languages of the U.S.S.R. The Elements: A Parasession on Linguistic Units and Levels. Ed. by P.R. Clyne, W.F. Hanks and C.L. Hofbauer Chicago: Chicago Linguistic Society, 436-447.
GRAMMATICALIZATION AND STRATEGIES IN RESOLVING SUBJECT MARKING
245
Peterson, T. (2004). The (Re)organization of Semantic Roles in Tsimshian Connectives’ ICSNL XXXIX: Proceedings of the 39th International Conference on Salish and Neighbouring Languages. Ed. by J.C. Brown and T. Peterson. Vancouver: UBCWPL, 323-340. Peterson, T. (2003). Issues of Morphological Ergativity in the Tsimshian Languages: Determiners, Agreement and the Reconstruction of Case. Case, Valency and Transitivity in Studies in Language Companion Series 77. Ed. by L. Kulikov, A. Malchukov and P. de Swart. Amsterdam: John Benjamins, 65-90. Rigsby, B. (1986). Gitksan Grammar. Ms, University of Queensland. Rigsby, B. (1989). A Later View of Gitksan Syntax. General and Amerindian Ethnolinguistics. Ed. by M. R. Key and H. Hoenigswald. Berlin: Mouton de Gruyter, 245-260. Rodrigues, A. (2001). Sobre a Natureza do Caso Argumentativo. Des noms et des verbes en tupi-guarani: état de la question. LINCOM Studies in Native American Linguistics 37. Ed. by F. Queixalós. LINCOM EUROPA, 135-162. Stebbins, T. (2001). Sm’algyax. Grammatical sketch prepared for the Research Center for Linguistic Typology, La Trobe University. Stebbins, T. (2003). Fighting language endangerment: community directed research on Sm'algyax (Coast Tsimshian). Osaka: The Endangered Languages of the Pacific Rim Project. Tarpent, M.-L. (1982). Ergative and accusative: A single representation of grammatical relations with evidence from Nishga. Working Papers of the Linguistic Circle of Victoria 2/1, 50-106. Tarpent, M.-L. (1987). A Grammar of the Nisgha Language. Unpublished Ph.D. dissertation, University of Victoria. Tarpent. M.-L. (1988). Below the surface of Nisgha syntax: arguments and adjuncts. Paper presented at the International Conference on Salish and Neighbouring Languages. Timberlake, A. (1977). Reanalysis and actualization in syntactic change. Mechanisms of syntactic change. Ed. by C.N. Li, 141-177. Trask, R.L. (1979). On the origins of Ergativity. Ergativity: Towards a theory of grammatical relations. Ed. by F. Plank, 385-404. Ura, H. (2000). Checking theory and grammatical functions in generative grammar. Oxford: Oxford University Press. Wiltschko, M. (2000). Is Halkomelem (split) ergative? Proceedings of the 35th International Conference on Salish and Neighbouring Languages. Mount Currie, BC, 249-268. Wiltschko, M. (2006). On ergativity in Halkomelem (and how to split and derive it). A. Johns, D. Massam and J. Ndayiragije. Ergativity: Emerging Issues. Dordrecht: Springer, 197-228. Woolford, E. (2000). Ergative agreement systems. University of Maryland Working Papers in Linguistics 10, 157-191. Woolford, E. (2001). Case patterns. Optimality-Theoretic Syntax. Ed. by G. Legendre et al. Cambridge (MA): MIT Press.
MARK DONOHUE
CHAPTER 11 Different Subjects, Different Marking
In Tukang Besi the selection of subject is marked on the verb with the use of different agreement markers, while case marking is constant, thus leading to a system in which, depending on the diathesis employed, subject marking on the verb follows completely different paradigms. Some minor sentence types, and differently-marked subordinate sentence types, are also examined, and I explore the implications for the separation of argument structure and grammatical functional structure. 1. HOW SUBJECTS CAN BE DIFFERENT Tukang Besi, an Austronesian language of Southeast Sulawesi in central Indonesia, is a language in which the grammatical subject of a bivalent clause may be marked in different ways independently of their referential properties and without any relation to the semantics of predicates. As with other more northern Austronesian languages, such as Tagalog, the determining factor is verbal diathesis. Case marking is (usually) an explicit indicator of subject status, while verbal agreement is not always indicative of the functional status of the argument it references. Before attempting to show the ways in which subjects are uniformly or nonuniformly indicated in the morphology and syntax, I must clearly define what I mean by the term ‘subject’. The notion of ‘subject’ has been used in several different ways by linguists. Without reviewing the extensive literature from a variety of different theoretical persuasions, we need to review the distinctions that were popularised by Schachter in two seminal papers (Schachter 1976, 1977) which detailed the difference between what can in modern terminology be described as argument structure positions and pragmatically/syntactically salient arguments. In argument structure terms we can identify three ‘positions’ for monovalent and bivalent predicates, arranged in terms of the ordering of the different arguments of that predicate:1 1
The arguments are ordered in terms of the thematic hierarchy; I assume the version of the hierarchy described in Bresnan and Kanerva (1989): agent > beneficiary > goal/experiencer > instrument > theme/patient > locative The details and labels of the hierarchy are not important for the purposes of this exposition, simply the relative position in the ordered hierarchy. Alternative ways to rank arguments without reference to
247 H. de Hoop and P. de Swart (eds.), Differential Subject Marking, 247–279. © 2008 Springer.
248
MARK D ONOHUE
(1) the highest role (2) a position that is both the highest and the lowest ( / second) role; that is, it is the sole argument in the subcategorisation frame. (3) the lowest role ( / the second highest role)2 Following (approximately) Comrie (1978) and Andrews (1985:68) I shall use the abbreviations A, S and P to refer to the positions (loosely) described in (1), (2) and (3), respectively. An S is the single argument of a monovalent verb; it is the highest role in its subcategorisation frame, and can simultaneously be described as the lowest role in the frame. An A is the most agent-like argument of a polyvalent verb, the highest role, which is not simultaneously the lowest role. The label P refers to the non-A argument in a prototypical bivalent verb, and to the argument in a trivalent (or quadrivalent) predicate which shows the same morphosyntactic behaviour. It is the lowest ( / second highest; see footnote 2) role in its subcategorisation frame, but it is not simultaneously the highest. These are syntactic roles in the sense of relationships existing at argument structure that can frequently be shown to affect morphosyntactic categories in languages, and they are not descriptors of grammatical status, though the identity of grammatical functions such as subject and object may be defined, after examining the properties of appropriate constructions in the language, in terms of the appropriate groupings of these roles. Using this terminology we can describe the syntactically privileged argument in most English syntactic constructions as being the S or A (as appropriate for the verb) of the clause; this is the category that Philippinists such as Schachter call ‘actor’, and which in case terms is the nominative grouping (S and A, as opposed to P, the accusative).3 On the other hand the syntactically most privileged argument in many Mayan languages (such as Mam; see England 1983) or Inuit (see the summary in Manning 1996) is the S of a monovalent clause or the P of a bivalent clause; authors such as Gerdts (1988) describe this syntactic grouping as absolutive, the same label used to describe case or agreement referring only to this grouping of arguments, and in contrast to ergative, which refers to an A. Schachter shows that while some syntactic constructions are restricted to the S or A in the clause, other constructions refer to an argument which in a polyvalent clause might be the A, but might also be
2
3
semantic roles have been presented by Kiparsky (1997) and Wunderlich (1997). While these models do not need to refer to particular semantic roles and their ordering, they require a stipulatively ordered set of semantic explications to achieve a similar model. The difference between specifying ‘lowest role’ and ‘second role’ emerges only when we examine three-place verbs. In some languages, including Tukang Besi, it is the second argument of the three core arguments, the recipient in an
frame, that is treated in the same way as the second and lowest role in the subcategorisation frame of a two-place verb (<__ , __>). In some other languages (very few in number) it is the theme in such a subcategorisation frame that behaves as the second and lowest argument in a bivalent frame. For these languages we need to recognise a third ‘position’, one that is neither highest nor lowest in the subcategorisation frame. Falk (2000a, 2000b) uses the label ‘G^F’ to refer to this (apparently universally relevant, though not universally privileged) grouping, to avoid implications of morphological marking that might be present in the label ‘nominative’. ‘G^F’ should be read as ‘grammatical function that is defined by being the highest role in argument structure.’
DIFFERENT SUBJECTS, DIFFERENT MARKING
249
a different argument, such as the P. These latter properties are, following Kroeger (1993), taken to be tests for ‘subject’. The constructions that can be described in terms of argument structure configurations alone, and which do not vary with the overtly marked verbal diathesis, are not taken to be evidence for a separate module of grammatical functions, and so cannot be interpreted as tests for subject status. Only those constructions which are sensitive to the verbal diathesis, and which are not easily explicated in terms of argument structure positions, are assumed to be subject tests. I shall take ‘object’ to refer to any direct argument of the verb that is not the subject; as a less-salient participant in the discourse, grammatical tests for object status are less common than those that test subject status. This asymmetry has been noted at least since Plank (1984). Variation in subject marking is normally described as being found in terms of the semantic content of the predicate, the status of the predicate as a main or subordinate element of the sentence it is in, and the tense-aspect-mood of the clause. In Tukang Besi we find that all of these factors play a part in determining the morphological marking of subjects, as well as the diathesis system of the language, which is marked in a way that leads to a typologically unusual morphosyntactic system. 2. THE INFLECTIONAL MORPHOSYNTAX OF TUKANG BESI Tukang Besi is spoken natively on the Tukang Besi archipelago of Southeast Sulawesi, in central Indonesia, and in various trading communities between Singapore and New Guinea (Donohue 1999, 2004). It is a verb-initial, subject-final language. The syntax shows many similarities with the Philippine languages, but also a significant number of discrepancies. The verbs show obligatory agreement for the S or A by verbal prefixes, and optional agreement for the P by enclitic.4 Case marking choices are limited to the ‘nominative’ na, the genitive nu, the oblique i / di (the former used in future/irrealis contexts, the latter and underspecified form used in other cases), and the underspecified (core) case te.5, 6 The agreement and case marking systems interact as shown in the following examples. While there is only one marking choice for monovalent clauses, described in 2.1, bivalent clauses have two variants, which are discussed in 2.2 and 2.3, as well as more detail in section 3.
4 5
6
These pieces of morphology cannot be simply considered instances of ‘incorporated pronouns’; for arguments against this stance, see Donohue (1999:123-129). Tukang Besi examples are transcribed in the phonemic transcription used in Donohue (1999). The conventions are broadly equivalent to IPA standards, with the following exceptions: b represents a voiced imploded bilabial stop, d represents a voided imploded dental stop, ' represents a glottal stop, ng represents a velar nasal, and u represents a high back unrounded vowel. The gloss ‘nominative’ is used here in the tradition found in, amongst others, Bell (1976, 1983) and Kroeger (1993), and specifically does not refer to a grouping of S and A. Rather, it refers to a case that marks the grammatical subject of the clause. The use of this gloss pre-empts the analysis to be presented in section 3, but is maintained here for the sake of consistency.
MARK D ONOHUE
250 2.1 Marking Ss
In (4) the subject of the monovalent clause is marked by the agreement prefix ku- on the verb and by the use of the nominative case marker na on the NP (with local pronominal arguments the use of free NPs is at best optional, and is strongly dispreferred).7 (4) No-waliako=mo (na ia) i kampo 3R.S/A-return=PFV NOM 3SG OBL village ‘S/he returned to the village.’ The same agreement and case marking patterns can be used with all monovalent verbs, regardless of their semantic content/event structure. Evidence of this can be seen in the following examples.8 agent S (5) No-rau na ana 3R.S/A-yell NOM child ‘The child is yelling.’ experiencer S (6) No-malino na ana 3R.S/A-lonely NOM child ‘The child is lonely.’ patient S (7) No-tunu na kau 3R.S/A-burn NOM wood ‘The wood is burning.’ The prefixal agreement paradigm varies in realis versus irrealis contexts, for all but first person singular Ss. The differences between the two prefix sets can be seen in the following paradigm for mo’aro ‘be hungry’. (8) 1SG 2SG 3SG 1PA 1PL 2PL 3PL
7 8
Realis ku-mo'aro 'u-mo'aro / nu-mo'aro no-mo'aro / o-mo'aro ko-mo'aro to-mo'aro ki-mo'aro no-mo'aro / o-mo'aro
Irrealis ‘…was / is hungry.’ ku-mo'aro ko-mo'aro na-mo'aro / a-mo'aro ka-mo'aro ta-mo'aro i-mo'aro na-mo'aro / a-mo'aro
The free DP na ia is grammatical in this sentence, but is unlikely to appear without a strongly contrastive pragmatic context. See 2.4. For monovalent verbs with experiencer subjects there is an alternative agreement option using the genitive enclitics that will be described in 2.4. This will be discussed in relation to table 5.
DIFFERENT SUBJECTS, DIFFERENT MARKING
251
As can be seen there is no distinction between singular and plural for third person, though there are two variants for this person, with and without the n. The second person has two variants in the realis, ‘u- and nu-; the second person also distinguishes two numbers, singular and nonsingular (= plural), while first person distinguishes a singular, paucal and plural category. There are different prefixes for realis and irrealis for all but the 1SG S, and the relationship between the realis and irrealis forms is not consistent; in some cases there is a vowel alternation, while in others the consonant of the prefix varies. This is clearly a difference in subject marking based on the mood of the clause. Nonetheless, all Ss show the same behaviour, with the same number of distinctions maintained in both realis and irrealis. The fact that all monovalent verbs show the same coding options means that we are not dealing with a ‘split-S’ system, in which the single arguments of some monovalent predicates are marked in one way (the same as the A of a bivalent verb), and some are marked differently (the same as the P of a bivalent verb). In fact, as we shall see, the prefixal agreement coding strategy seen with Ss is the same as the agreement found with As. 2.2 Marking As In (9) a bivalent clause with similar verbal prefix and nominative case for subject is found. (10) shows that the order of the two core arguments is important in the bivalent clause. V P A (9) No-'ita te ana (na ia) 3R.S/A-see CORE child NOM 3SG ‘S/he saw a child.’ V A P (10) * No'ita na ia te ana The same prefixes that are used to mark Ss are used with As. As with Ss, different prefixes are used for realis and irrealis clauses, as shown in (11) with the verb manga ‘eat’. The paradigm is identical to that used to mark agreement with Ss. (11) 1SG 2SG 3SG 1PA 1PL 2PL 3PL
Realis ku-manga 'u-manga / nu-manga no-manga / o-manga ko-manga to-manga ki-manga no-manga / o-manga
Irrealis ‘…ate / eats.’ ku-manga ko-manga na-manga / a-manga ka-manga ta-manga i-manga na-manga / a-manga
252
MARK D ONOHUE
The same prefixal coding is found regardless of the semantic role of the A, as shown in the following different predicates: agent A (12) No-manga te bae na ana 3R.S/A-eat CORE rice NOM child ‘The child ate rice.’ recipient A (13) No-'awa te bae na ana 3R.S/A-receive CORE rice NOM child ‘The child received the rice.’ experiencer A (14) No-'ita te bae na ana 3R.S/A-see CORE rice NOM child ‘The child saw the rice.’ instrument (effector) A (15) No-motiti='e na 'oloo kau te 3R.S/A-dry=3P NOM wood CORE sun ‘The sun dried the wood.’ The enclitic =‘e in the last example will be explained in the following section. 2.3 Marking Ps In addition to the bivalent clauses seen in 2.2 an alternative bivalent clause coding option exists. In this clause type the verb shows the same prefixal agreement for the A that was seen in 2.1 and 2.2, but also has an enclitic that shows agreement with the P. The case marking is also changed, the nominative na that was found with subject Ss or As now appearing with the P, and the (non-nominative) core case te that appeared with Ps in 2.2 is associated with the A in the double-agreement clause type. Word order is freer in this clause type that in the clause type without P agreement seen in section 2.2.9 V A P (16) No-'ita='e (te ia) na ana 3R.S/A-see=3P CORE 3SG NOM child ‘S/he saw the child.’ V P A (17) ~ No'ita'e na ana te ia
9
Note the use of a definite article with ‘child’ in the English translation of (16), as opposed to the indefinite in (9). There is a consistent difference between clauses with, and without, P enclitics in that those with P enclitics show greater definiteness, specificity, telicity, affectedness, or a combination of these features.
DIFFERENT SUBJECTS, DIFFERENT MARKING
253
Unlike the prefixes used to mark agreement for an S or A, there is no variation in the form of the P enclitics for realis or irrealis mood, as can be seen in (18) in which the A is kept constant as a third person, marked with a third person prefix no- (o- is also possible as a free variant in all cases, but has not been shown here for reasons of space) on the verb moniasi ‘miss, feel homesick for’, and the person of the P is changed paradigmatically. Again there are different numbers of number distinctions in the different persons, following the same patterns seen with the prefixes. (18) 1SG 2SG 3SG 1PA 1PL 2PL 3PL
Realis no-moniasi=aku no-moniasi=ko no-moniasi='e no-moniasi=kami no-moniasi=kita no-moniasi=komiu no-moniasi='e
Irrealis ‘They missed / miss …’ na-moniasi=aku na-moniasi=ko na-moniasi='e na-moniasi=kami na-moniasi=kita na-moniasi=komiu na-moniasi='e
In addition to these regular P enclitics, which may be used to mark the P of any bivalent verb, an additional set of enclitics are optionally found with Ps that are recipients or beneficiaries. (19) shows that the verb hoti ‘donate (food or clothing) to someone charitably’ can take either the P enclitics described above, or the ‘dative’ enclitics. Most other verbs, such as siasia ‘beat’ cannot appear with the dative enclitics, shown in (20). (19) 1SG 2SG 3SG 1PA 1PL 2PL 3PL (20) 1SG 2SG 3SG 1PA 1PL 2PL 3PL
Regular P enclitics no-hoti=aku no-hoti=ko no-hoti='e no-hoti=kami no-hoti=kita no-hoti=komiu no-hoti='e Regular P enclitics no-siasia=aku no-siasia=ko no-siasia='e no-siasia=kami no-siasia=kita no-siasia=komiu no-siasia='e
‘dative’ P enclitics ‘They donated to …’ no-hoti=naku no-hoti=nso no-hoti=ne no-hoti=nsami no-hoti=nggita no-hoti=ngkomiu no-hoti=ne ‘dative’ P enclitics ‘They beat …’ *no-siasia=naku *no-siasia=nso *no-siasia=ne *no-siasia=nsami *no-siasia=nggita *no-siasia=ngkomiu *no-siasia=ne
This dative enclitic option is considered by most speakers as archaic, but is used by many speakers productively even with verbs that only have a beneficiary P by virtue of applicative derivation, as seen in (21). Here the same verb siasia is derived with a benefactive applicative =ako, and the beneficiary P is marked by the dative enclitics. It is also possible for the regular P enclitics to be used, as in (22).
254
MARK D ONOHUE
(21) No-siasia=ako=ne te ikami 3R.S/A-beat=APPL=3DAT CORE 1PA ‘They beat us for her/him.’ te ikami (22) No-siasia=ako='e 3R.S/A-beat=APPL=3P CORE 1PA Examples of Ps of various semantic roles with enclitic agreement, and the conditions on their use, are shown in (23) - (36). The dative enclitics may only be used with Ps that bear beneficiary or recipient roles, as mentioned above. agent P (derived only) (23) No-wila-ngkene=aku te ana 3R.S/A-go-COMIT.APPL=1SG.P CORE child ‘The child went with me.’ (24) *No-wila-ngkene=naku te ana 3R.S/A-go-COMIT.APPL=1SG.DAT CORE child beneficiary P (25) No-hoti=aku te ana 3R.S/A-give.charitably=1SG.P CORE child ‘The child gave (food or clothing) to me.’ (26) No-hoti=naku te ana 3R.S/A-give.charitably=1SG.DAT CORE child Beneficiaries are also found in derived verbs derived by applicatives, such as was seen in (21). Recipient Ps are only known to occur in two verbs, hu’u ‘give’ and kahu ‘send’. recipient P (27) No-hu'u=aku te boku te ana 3R.S/A-give=1SG.P CORE book CORE child ‘The child gave me a book.’ (28) No-hu'u=naku te boku te ana 3R.S/A-give=1SG.DAT CORE book CORE child While instruments are rare as the lexical P of a verb they are found in a small number of verbs for which they are inherent, and are productively found when an instrumental applicative (identical in morphological form to the benefactive applicative) is used with a verb (as seen in (31) and (32)). (intermediary) instrument P (29) No-heka-batu='e na po'o te ana 3R.S/A-INTENS-stone=3P NOM mango CORE child ‘The child threw a mango (as if it were a stone).’ (30) *No-heka-batu=ne na po'o te ana 3R.S/A-INTENS-stone=3DAT NOM mango CORE child
DIFFERENT SUBJECTS, DIFFERENT MARKING
255
na kau te ana (31) No-siasia=ako='e 3R.S/A-beat=APPL=3P NOM wood CORE child ‘They beat the child with a stick.’ (32) *No-siasia=ako=ne na kau te ana 3R.S/A-beat=APPL=3DAT NOM wood CORE child theme P te ana (33) No-'ita=aku 3R.S/A-see=1SG.P CORE child ‘The child saw me.’ te ana (34) *No-'ita=naku 3R.S/A-see=1SG.DAT CORE child patient P (35) No-pa-mate=aku te ana 3R.S/A-CAUS-dead=1SG.P CORE child ‘The child killed me (temporarily).’10 (36) *No-pa-mate=naku te ana 3R.S/A-CAUS-dead=1SG.DAT CORE child While there are no verbs which lexically have a locative P, there is a variety of related locative applicative suffixes which appear with both monovalent and bivalent verbs. locative P (37) No-kede-mi=aku te ana 3R.S/A-sit-LOC.APPL=1SG.P CORE child ‘The child sat on me.’ (38) *No-kede-mi=naku te ana 3R.S/A-sit-LOC.APPL=1SG.DAT CORE child All polyvalent verbs require the use of prefixes showing agreement with their A. As seen in 2.2, agreement with P is not obligatory for most verbs. Some exceptions exist; in (15) the verb motiti ‘dry’, when used bivalently, must appear with P agreement, and this is true of verbs with instrumental As generally, as well as a small number of other verbs. The following section summarises the information on case, constituency and agreement patterns of Tukang Besi clauses.
10
The morphologically complex verb pa-mate ‘kill’ (literally, ‘cause to die’, with the generic causative prefix pa-) is used in cases where the person has been revived, medically or magically following death. A permanent or irreversible death will be marked with the verb hoko-mate ‘kill, murder’, showing the same root mate ‘die, be dead’ prefixed with hoko-, the factitive prefix. See Donohue (1999:205-211).
MARK D ONOHUE
256
2.4 The inflectional patterns of Tukang Besi The agreement, case-marking and word orders seen in 2.1-2.3 are summarised in the schemas shown in table 1. Table 1. Basic agreement, word order and case marking in Tukang Besi
Clause type
Variants
Agreement
Word order
Bivalent
P indexed on V P not indexed on V
A-V=P A-V S-V
V A P (~ VPA) VPA VS
Monovalent
Case marking A S P te na na te na
The full set of pronominal affixes and free forms is given in table 2. The agreement prefixes that index an S or A have variant forms depending on the mood, realis or irrealis, of the clause, while the other major verbal paradigm, marking a P, are invariant (and much closer phonologically to the independent pronouns, which may appear in any syntactic role). The dative P enclitics are clearly related to the generic P enclitics with the addition of a nasal component; in some cases the k of the general P enclitics is cognate with an s, showing the same sound change that is also found in the 1SG.GEN form =su, a sound change that has only marginally affected Tukang Besi in these morphemes, but which is widespread in other languages of western Buton. The genitive enclitics in the second last column have not been introduced yet, but will become relevant in table 5, where they appear in a variety of minor clause types to indicate arguments of the verb, and in an optional role presented later in this section. While there are three different sets of pronominal enclitics, a single word may only host one enclitic.11 Table 2. Pronominal forms in Tukang Besi
Position: Role: (Mood): 1SG 2SG 3SG 1PA 1PL 2PL 3PL
Pre-root S,A realis irrealis kuku'u-/nukono-/ona-/akokatotaikino-/ona-/a-
Post-root Post-root P Dative P
Post-root nominal POSS’R subordinate A, S, P =naku =su =nso ='u =ne =no =nsami =nto =nggita =mami =ngkomiu =miu =ne =no
Independent (any)
=aku =ko ='e =kami =kita =komiu ='e
iaku iko'o ia ikami ikita ikomiu amai
The only TAM-conditioned split involves the use of different sets of prefixes, as shown above. All Ss are marked on the verb with the same prefixes that are used to 11
This means that polyvalent verbs will show agreement for at most two arguments, the A and the first P; Tukang Besi is a primary object/secondary object language.
DIFFERENT SUBJECTS, DIFFERENT MARKING
257
mark As, and none of them use the P enclitics. Compare (7), repeated below, with the ungrammatical clauses in (39) - (41), which show the same predicate with P, dative P, and genitive agreement markers. (7) No-tunu na kau 3R.S/A-burnNOM wood ‘The wood is burning.’ (39) *Tunu='e na kau burn=3P NOM wood (40) *Tunu=ne na kau burn=3DAT NOM wood (41) *Tunu=no na kau burn=3GEN NOM wood While the most P-like Ss, those with theme or patient roles, show no difference in marking compared to the most A-like Ss, we do find some alternative coding with experiencer Ss. In addition to the normal prefixal agreement option, shown in (6) repeated below, an experiencer S may also be marked on the verb by means of genitive enclitics, as in (42).12 (6) No-malino na ana 3R.S/A-lonely NOM child ‘The child is lonely.’ (42) Malino=no na ana lonely=3GEN NOM child The use of the genitive agreement option on verbs is only found with Ss, not experiencer As. It is worth noting, however, that a number of bivalent psych predicates, such as gau ‘desire’ and hada ‘want’, are more frequently used in nominal clauses than in verbal clauses, in which case the indication of the experiencer is found by means of genitive clitics. In (43) and (44) we can see that a regular verbal clause is not grammatical with genitive agreement marking on the verb. Verbal clause with prefixes po'o (43) Ku-hada te 1SG.S/A-want CORE mango ‘I want a mango.’ Verbal clause with genitive enclitics po'o (44) *Hada=su te want=1SG.GEN CORE mango ‘I want a mango.’
12
P-enclitic or dative enclitic marking on the verb remains ungrammatical: *malino=‘e, *malino=ne. This genitive predication is only found with experiential verbs, not with any other kinds of predicates.
258
MARK D ONOHUE
In (45) the nominal clause, of the form [ te NP te NP ], takes genitive marking on hada to indicate the experiencer of ‘want’. It is not grammatical to use verbal prefixes in a nominal clause, as seen in (46). Nominal clause with genitive enclitics (45) Te hada=su te po'o CORE want=1SG.GEN CORE mango ‘I want a mango.’ (Literally, ‘My want(ing) is a mango.’) Nominal clause with prefixes po'o (46) *Te ku-hada te CORE 1SG.S/A-want CORE mango ‘I want a mango.’ While this is not the same use of genitive enclitics to mark an argument of a verbal clause as was the experiencer S option, it indicates a preference, or at least consistent tendency, to avoid the prefixal forms with experiencers. Note that, even allowing for the genitive coding option available with experiencer Ss, we would not want to classify the Tukang Besi agreement system as showing a ‘split-S’ or ‘splitintransitive’ pattern, since genitive marking is not a feature of the agreement system for Ps. It is true that both the P agreement paradigm and the genitive paradigm are enclitic, but they are formally distinct sets of morphemes, and the use of P enclitics to index an S is ungrammatical, as shown in (39) and (40). An agreement subset more similar to a split-S system is found when motion verbs are serialised with numeral verbs. Either of these verb types on their own can only show agreement by means of the prefixal agreement markers. The grammaticality of prefixal agreement with a motion verb is shown in (47) - (50). Motion verb with pronominal prefixes (47) Ko-mai=mo 1PA.S/A-come=PFV ‘We have come.’ Motion verb with pronominal enclitics (48) *Mai=kami come=1PA.P (49) *Mai=nsami come=1PA.DAT (50) *Mai=mami come=1PA.GEN Analogously to the motion verb, the prefixal agreement pattern of a numeral verb are shown in (51)-(54). Motion verb with pronominal prefixes (51) Ko-gana=mo 1PA.S/A-be.four= PFV ‘We are now four.’ / ‘There are now four of us.’
DIFFERENT SUBJECTS, DIFFERENT MARKING
259
Motion verb with pronominal enclitics (52) *Gana=kami be.four=1PA.P (53) *Gana=nsami be.four=1PA.DAT (54) *Gana=mami be.four=1PA.GEN When these verbs are serialised together the only grammatical agreement pattern involves the use of both the prefixes and P enclitics; the enclitics may be taken from the set of plain P enclitics shown in (55), or from an additional set of enclitics that are found only in these constructions, illustrated in (56) and shown in full in (57). The verb + numeral verb serialisation is found only with nonsingular numbers, and so there are no special enclitic forms for 1SG, 2SG or 3SG.13 The dative or genitive enclitics may not be used in this construction. Other instances of verb serialisation do not require (or allow) coreferent prefixes and enclitics. Motion verb and numeral verb serialised (55) Ko-mai-gana=kami 1PA.S/A-come-be.four=1PA.P ‘The four of us came.’ (56) Ko-mai-gana=ngkami 1PA.S/A-come-be.four=1PA.INTRADIR.P ‘The four of us came.’ (57) 1PA 1PL 2PL 3PL
Regular P enclitics ko-mai-gana=kami to-mai-gana=kita i-mai-gana=komiu no-mai-gana='e
Intradirective‘P’ enclitics ‘Four of…came’ ko-mai-gana=ngkami to-mai-gana=ngkita i-mai-gana=ngkomiu no-mai-gana='e
A possible explanation for the use of both prefixal and enclitic agreement markers for these predicates can be found in the nature of motion verbs: motion verbs, and some posture-assuming verbs, are predicates in which the one participant is simultaneously the instigator of the event, the agent, and the participant that undergoes a change of location, a theme. It is thus not completely unexpected that the marking for such a predicate should involve both an element that is typical of agentive agreement, the prefix, and an element that is associated with a P, a syntactic role that is most typically associated with a theme or a patient. The unexplained aspect is why this double-marking pattern should emerge only when the 13
In order to emphasise that only a single participant was involved in the action the adverb ala’a ‘only’ or pe’esa ‘self’ +genitive clitics may be used, as in (i) and (ii). ala'a. pe’esa='u (i) ‘U-mai (ii) ‘U-mai 2SG.R.S/A-come only 2SG.R.S/A-come self=2SG.GEN ‘Just you came.’ or ‘You just came.’ ‘You came on your own.’
MARK D ONOHUE
260
motion verb is serialised with a numeral verb. For this conundrum I have no explanation, except to hypothesise that an earlier general double-marking pattern with intradirectives has been preserved only with the serial verb construction, while the simple motion verb construction has normalised to the general nominativeaccusative pattern. From the data in sections 2.1-2.4 we summarise the different agreement paradigms in Tukang Besi and their functions: (58)
prefixal sets P enclitic dative enclitic genitive enclitic prefix + (intradirective) enclitic
S,A agreement prefixes, obligatory on verbs P agreement clitics, optional on most bivalent verbs optionally used in place of the accusative agreement markers, only when marking a P that is beneficiary or recipient with main clause verbs only used to show agreement with an intransitive experiencer used with motion verbs that are serialised with numeral verbs
The case marking system for core arguments is quite simple, as shown in table 1: an S is marked with na, while in a polyvalent clause na can mark either the A, if the verb shows agreement only for the A and not for the P, or the P, if there is a pronominal enclitic on the verb. Structurally we can determine that an S or an A are always external to the VP, while a P is external to the VP only if the verb is marked with P agreement enclitics. A locative adjunct can only follow a VP, and not intrude in it. With a monovalent predicate, in which the S is external to the VP, this means that a location can intrude between an NP subject and the verb. In addition to (4), repeated below, we can also find the grammatical (59). V S Location (4) Ku-waliako=mo (na iaku) i kampo 1SG.S/A-return= PFV NOM 1SG OBL village ‘I returned to the village.’ V Location S (59) Ku-waliako=mo i kampo (na iaku) 1SG.S/A-return= PFV OBL village NOM 1SG ‘I returned to the village.’ With a bivalent verb with single agreement for A, but not P, we find that a locative adjunct can appear in the following positions, based on (9). The adjunct may appear following or preceding the A, but may not intrude between the P and the verb.
DIFFERENT SUBJECTS, DIFFERENT MARKING
261
V P A Location (60) Ku-'ita te ana (na iaku) di kampo 1SG.S/A-see CORE child NOM 1SG OBL village ‘I saw a child in the village.’ V P Location A (61) Ku'ita te ana di kampo (na iaku) V Location P A (62) *Ku'ita di kampo te ana (na iaku) When a bivalent verb is marked for agreement with both the A and the P a different set of possibilities is found. The adjunct may appear anywhere following the verb, not respecting the close constituency between the verb and the P that was evident in (60)-(62). V A P Location (63) Ku-'ita='e (te iaku) na ana di kampo 1SG.S/A-see=3P CORE 1SG NOM child OBL village ‘I saw the child in the village.’ V A Location P (64) Ku'ita'e (te iaku) di kampo na ana V Location A P (65) Ku'ita'e di kampo (te iaku) na ana V P A Location (66) Ku'ita'e na ana (te iaku) di kampo V P Location A (67) Ku'ita'e na ana di kampo (te iaku) V Location P A (68) Ku'ita'e di kampo na ana (te iaku) A further difference between Ps in the two different bivalent clause types can be seen in the grammatical positions available for time expressions. A time expression must always follow an S, or a na-marked A, but may precede or follow a na-marked P. This results in the following grammaticality judgements for clauses in which the verb lacks agreement for P. V P A Time (69) Ku-'ita te ana (na iaku) dinggawi 1SG.S/A-see CORE child NOM 1SG yesterday ‘I saw a child yesterday.’ V P Time A (70) *Ku'ita te ana dinggawi (na iaku) V Time P A (71) *Ku'ita dinggawi te ana (na iaku)
MARK D ONOHUE
262
When there is agreement for the P on the verb, the following judgements are found. Once again there is complete grammaticality for the time adverbial in any position following the verb. V A P Time (72) Ku-'ita='e (te iaku) na ana dinggawi 1SG.S/A-see=3P CORE 1SG NOM child yesterday ‘I saw the child yesterday.’ V A Time P (73) Ku'ita'e (te iaku) dinggawi na ana V Time A P (74) Ku'ita'e dinggawi (te iaku) na ana V P A Time (75) Ku'ita'e na ana (te iaku) dinggawi V P Time A (76) Ku'ita'e na ana dinggawi (te iaku) V Time P A (77) Ku'ita'e dinggawi na ana (te iaku) These data concerning locative and temporal adjuncts, taken together, imply roughly the following structural positions.14 (78)
?P IP VP V
te DPS,A, na DPP na DPS,A
te DPP
Further tests for the constituency of the P and the non-agreeing verb can be found in Donohue (1999, 2004).15 Note that an alternative morphological realisation of the nominative argument is available. It may appear preverbally, yet not in a topic position external to the clause. In this case it is marked with te, the case marker that otherwise marks non-nominative core arguments. This split in marking emphasises 14
The identify of the ‘?P’ is not certain. Topicalisation is found to the left of the clause, and there are other reasons to feel that this is not a CP in the sense of a CP being an extra-clausal projection. Nonetheless, there are certain similarities between a CP and the ?P constituent shown here. 15 Topicalisation also affects the case marking strategy, in that the nominative case may only appear postverbally. A nominative argument in a preverbal position must be marked with the underspecified case te.
DIFFERENT SUBJECTS, DIFFERENT MARKING
263
the more general nature of te, and the more restricted function of na in the linker inventory of Tukang Besi. We have now thoroughly examined the morphological and structural properties of core arguments in verbal clauses in Tukang Besi, finding considerable differences between marking strategies, revealing a nominative-accusative pattern in verbal agreement and a complex set of patterns in case marking and phrase structure. In the next section I shall show how these patterns are related to the question of subjecthood. 3. DETERMINING THE IDENTITY OF THE SUBJECT The problems inherent in identifying a subject have been recognised by many authors (for instance, Keenan 1976, Schachter 1976, 1977, Foley and Van Valin 1977, 1984, Andrews 1985, Van Valin and La Polla 1997, and many others). While languages such as English and many western European languages display a subject that can be identified on the basis of argument structure positions (‘select the argument that ranks highest in the thematic hierarchy’ is a good rule-of-thumb way of identifying the argument that counts as subject in English), (western) Austronesian languages show a syntax that is more directly sensitive to pragmatic factors, and a split in ‘subject properties’ between the highest-ranked argument and the one that is morphologically marked as unique. Having examined the morphosyntax used to mark the core arguments, subjects and objects, in Tukang Besi we must now turn to the question of identifying the subject. It seems unproblematic that the sole argument of a (typical) monovalent verb is the subject of that predicate, and I shall take some of the syntactic properties of this type of argument as a guide in evaluating the grammatical status of the arguments of bivalent verbs. The constructions that shall be examined here include two that are useful in identifying the subject, and one, restrictions on heading external relative clauses, that is dependent on the position of the argument in a thematic hierarchy, and so has no bearing on the identification of the subject but which is methodologically relevant to illustrate the different marking afforded to subjects of subordinate clauses. When we examine the syntactic properties of the single subcategorised argument of a monovalent predicate, we find a number of tests that we can apply to determine the status of the arguments of polyvalent predicates. The constructions listed here are only a subset of those that are discussed in Donohue (1999), but serve to illustrate the argumentation.16 1. Control floating quantifiers Only the S in a monovalent predicate can control a floating quantifier, as seen in the single interpretation of saba’ane in (79). 16
From here on I shall omit the gloss ‘S/A’ with the verbal prefixes, letting their unique position differentiate them from the three enclitic pronominal paradigms, which shall continue to be labelled as ‘P’, ‘DAT’ and ‘GEN’.
MARK D ONOHUE
264
kua kampo (79) [QUANT Saba’ane] no-wila na mia all 3R-go NOM person ALL village ‘All of the people went to the village(s).’ *‘The person/people went to all of the villages.’ 2. Control conjunction reduction Conjunction reduction preferentially applies between core arguments in adjacent clauses. A textual example is presented later in this section, but the following sentences illustrate the principle. (80) No-wila=mo na mia kene no-helawe 'obu, maka 3R-go= PFV NOM person COMIT dog and.then 3R-rest ‘The peoplei went with the dog(s)j, and then (they)i rested.’ *‘The peoplei went with the dog(s)j, and then (they)j rested.’ 3. External relative clauses An external relative clause is marked by the infix in the verb, which shows no agreement. Only the S of the clause may appear as the head of the relative clause. [RC wila kua kampo] go.SI ALL village ‘the person/people who went to the village(s)’ (82) *te kampo wila CORE person go.SI ‘the village(s) where (the people) went’17 (81) te
mia
CORE person
When we apply these tests to the two types of polyvalent clauses, we find the following results for subjecthood: 1. Controlling floating quantifiers Only the A in a polyvalent predicate without P agreement, and only the P in a polyvalent predicate with P agreement, can control a floating quantifier, as seen in the unambiguous interpretation of saba’ane in (83) and (84). Floated quantifier referring to an A kaluku na amai (83) [QUANT Saba'ane] no-lemba te all 3R-carry CORE coconut NOM 3PL ‘All of them carried coconuts.’ *‘They carried all of the coconuts.’ 17
To be expressed grammatically the verb must be derived with a locative applicative, and then the Phead relative clause prefixes used, as in (i). (i) te kampo i-wil(a)-isi=no CORE village PP-go-LOC.APPL=3GEN ‘the village that they went to’
DIFFERENT SUBJECTS, DIFFERENT MARKING
265
Floated quantifier referring to a P amai (84) [QUANT Saba'ane] no-lemba='e na kaluku te all 3R-carry=3P NOM coconut CORE 3PL ‘They carried all of the coconuts.’ *‘All of them carried coconuts.’ Only the argument that is eligible to be marked with nominative case, na, may control floating quantifiers. 2. Controlling conjunction reduction The A in a polyvalent predicate without P agreement, and the P in a polyvalent predicate with P agreement, is the preferred controller and target of conjunction reduction in adjacent clauses. Coreference between S and A, both nominative (85) No-waliako=mo di kampo, maka no-'ita te ana 3R-return= PFV OBL village and.then 3R-see CORE child ‘Øi returned home, and then Øi saw a childj.’ Coreference between S and P, both nominative (86) No-waliako=mo di kampo, maka ana) no-'ita='e (te 3R-return= PFV OBL village and.then 3R-see=3P CORE child ‘Øi returned home, and then (a child/Ø)j saw Øi.’ Only the argument that is eligible to be marked with nominative case, na, may control or be the target of zero anaphora between clauses. 3. Heading external relative clauses The relative clause marked with the infix may only be used with an A head, not a P. As with an S head, the verb is not marked with nominative agreement prefixes, but may optionally use P enclitics. kau ] [RC koho te chop.SI CORE wood ‘the person/people who chopped the wood’ (88) te mia [RC koho='e na kau ] CORE person chop.SI=3P NOM wood ‘the person/people who chopped the wood’ (89) *te kau koho CORE wood chop.SI ‘the wood that (the person) chopped’ (90) te kau [RC i-koho ] CORE wood PP-chop ‘the wood that was chopped’ (87) te
mia
CORE person
MARK D ONOHUE
266
Only the A may head a relative clause marked with the infix; a P may only head a relative clause marked by prefix, shown here in (90) as i-. The P-headed relative clauses are discussed in more detail in table 5. The data on floating quantifiers, conjunction reduction and external relative clauses may be summarised as follows: (91)
external relative clause floating quantifiers
conjunction reduction
shows the same nominative-accusative distinctions as are made with the verbal agreement sets, and does not show a correlation with case marking follows the distinctions in case marking in that the argument marked with nominative case na is the argument that can be the restriction of a floating quantifiers follows the distinctions in case marking in that the argument marked with nominative case na is the argument that can control or be the target of conjunction reduction
Other tests will also single out the na-marked argument as being syntactically privileged (Donohue 1999, 2004). Most importantly, while constructions such as external relative clauses can be accounted for by appealing to argument-structure positions, the fact that the selection of the na-marked argument is divorced from considerations of argument structure hierarchies, either of the arguments of a bivalent predicate being eligible for this marking and this set of syntactic properties, means that we can only account for its syntactic status by assuming a privileged position in an ordered grammatical functions hierarchy. The conclusion, therefore, is that the na-marked argument is the subject of the clause it is in. Note that the notion of subject in Tukang Besi is not one that can be defined in terms of the labels A, S, and P. It is true that an S will be the subject of its clause, as the sole core argument in the subcategorisation frame, but in a bivalent clause we cannot uniquely identify subject with either A or P: rather than being syntactically predetermined, it is determined by pragmatic considerations, and monitored morphologically on both the verb and the choice of case marker. Interestingly, but not importantly for the purposes of this paper, we should note that the diathesis found in bivalent clauses does not involve demotion: both the single-agreeing verbs described in 2.2 and the doubly-agreeing verbs detailed in 2.3 have two core arguments, a subject and an object. The syntactic situation is identical to that of the better-described Philippine-type languages such as Tagalog, a point to which I return in section 5.18 18
Evidence for the argument status of both A and P in both the clauses described in 2.2 and those seen in 2.3 can be found through an examination of reflexive constructions: (i) Ku-’ita te karama=su na (ii) Ku-’ita='e karama=su 1SG-see CORE self=1SG.GEN 1SG-see=3P NOM self=1SG.GEN ‘I saw myself.’ ‘I saw myself.’
DIFFERENT SUBJECTS, DIFFERENT MARKING
267
We can therefore draw the following correlations between agreement, case and structural position, on the one hand, and grammatical function and syntactic role on the other. Table 3. Coding properties of SUBJECTs in different syntactic roles
Syntactic role A S P
Case marking na na na
Verbal agreement prefix prefix enclitic
Word order / structural position VP-external VP-external VP-external
Table 4. Coding properties of OBJECTs in different syntactic roles19
Syntactic role A S P
Case marking te na te
Verbal agreement prefix (prefix) -
Word order / structural position VP-external VP-external VP-internal
Objects Ss are only found in passives, as discussed in Donohue (2005b). The implications for these facts, syntactic and morphological, on a discussion of differential subject marking will be presented in the next section. 4. IMPLICATIONS We have seen that the agreement morphology in Tukang Besi is essentially nominative-accusative, grouping S and A together as opposed to P. At the same time the case marking system tracks the grammatical subject, monitored by the amount of agreement marking on the verb, and correlating roughly with the structural position of the argument. We have, therefore, different marking for subjects in terms of the different inflectional components, agreement and case marking, not aligning together, and in terms of the paradigm used to mark the subject. In a bivalent verb with agreement marked for both A and P the agreement paradigm that marks the subject is the set of P enclitics, and the P nominal is marked with the na case. On the other hand a bivalent verb with agreement marked for the A alone will show agreement with subject by means of the prefixes; the case marking na on the A nominal is consistent in marking subject. While the agreement markers are not consistent in their marking of subject or object, they are consistent in marking argument structure positions, marking an S or A in the case of the prefixes and a P in the case of the enclitics. The discrepancy arises in the assignment of grammatical functions to argument structure positions,
19
(Some speakers, mainly younger ones, prefer orungu ‘body’ as a reflexive nominal.) See Arka and Manning (1998), or Donohue (1999), for discussion of the significance of these data. For ‘S as object’, see the description of passives in Donohue (2005b). Non-passive Ss do not show any evidence of being objects
MARK D ONOHUE
268
the principles for which are shown in (92) and which correlate with the appearance of enclitics on the bivalent verb (following suggestions in Manning 1996). This makes clear the difference between argument structure positions and grammatical functions: while there is a relationship, generally that shown in (92a), in all languages, it is possible for the inverse mapping to be found, as in (92b). In this case the P (theme, lowest role) is the subject. This is found locally in Tukang Besi, but is apparently the default mapping found in languages such as Mam (England 1983) and certain Eskimo languages (see Manning 1996 for detailed argumentation of this point). (92) A-structure / GF-mapping for the pronominal voice alteration a. No enclitics
〈 __A , __P 〉 SUBJ
b. P enclitics 〈 __A , __ P 〉
OBJ
SUBJ
OBJ
An approach such as that advocated in Sells (2001), in which the P is raised in a clause with P clitics on the verb, would more formally capture the generalisations illustrated here, and would agree with the evidence for constituency shown in 2.4. Note that, in Tukang Besi, rather than being covert the pronominal incorporation on the verb posited by Sells is overt, in the form of the P agreement clitics. The fact that the determination of subject in a bivalent clause depends on pronominal features and their marking means that the complications associated with pronominal arguments in terms of their position on the animacy hierarchy come into play. While we have seen that there are bivalent clause types with agreement for the P and others lacking this agreement, when we deal with local persons it is almost unheard of to not show agreement on the verb. This means that, in a clause with local persons, a local person will always be subject, and if there are two local persons the subject will always be the P. It is not ungrammatical for an pronominal argument that is indexed on the verb to also be represented by a free pronoun in the same clause; this is extremely rare, but is attested in texts. Such constructions are usually associated with pragmatically marked information, and so are judged as being more natural if the case-marked pronoun appears preverbally, as in (94) and the textual (96) (see the discussion following (78) for the licensing of case in preverbal positions). Nonetheless, examples such as (93) and (95), with postverbal na-marked subjects, are not judged to be ungrammatical, though they are unusual, and contain a sense of contrastive focus. (93) No-'ita=aku te ana 3R-see=1SG.P CORE child ‘The child saw me.’ (94) Te iaku no-'ita=aku CORE 1SG 3R-see=1SG.P ‘The child saw me.’
( #!na NOM
te
iaku) 1SG ana
CORE child
DIFFERENT SUBJECTS, DIFFERENT MARKING
269
(95) #To-waliako=mo i kampo na ikita 1PL.R-return= PFV OBL village NOM 1PL (96) … te iaku habuntu (')u-hu'u=aku te kuli=no … CORE 1SG in.fact 2SG.R-give=1SG.P CORE skin=3GEN ‘…in fact you’ve just given me the peel (of the bananas) …’ Sentences such as (97), in which a local person is not indexed on the verb, are often judged to be grammatical, but are at best marginally felicitous, and are not part of any corpus of naturally-occurring speech. This implies a constraint requiring local person Ps to be the subject of their clause; morphologically, local persons should be marked on the verb. This does not apply to all local persons, regardless of syntactic role; (98) shows an instance of a non-nominative local person A; the choice to code loka ‘bananas’ as nominative forces the object coding of the A. The fact that te iko’o ‘you’ is overtly coded as a DP in the clause is mildly surprising, and probably reflects contrastive focus. (97) */# No-'ita te iaku na ana 3R-see CORE 1SG NOM child ‘The child saw me.’ (98) “Oho, 'u-pidi='e te iko'o na loka saba'ane='e.” yes 2SG.R-rubbish=3P CORE 2SG NOM banana all=3P ‘Yes, you’ve rubbished all of the bananas.’ Other evidence for the origins of the grammatical subject of Tukang Besi lying in pragmatic conditions can be found by examining sentences with question words. We find that there is a condition on the case marking (and thus clause type) that may be associated with question words, that can be expressed as in (99). Constraints on question word case marking (99) A question word cannot be cast syntactically in a way that would lead to it being assigned nominative case (na) in a verbal clause; that is, a question word may not be the grammatical subject. The implications of this are easily illustrated. Question words appear in situ, as in the questioned non-nominative P in (100). (100)
No-sampi te paira na amai? CORE what NOM 3PL 3R-pick ‘What did they pick?’
Questioning the A in this clause is only possible if the verb shows P enclitics, that is, casts the A in the non-nominative case, te. It is not possible to question an A that bears the nominative case.20 20
That this is a syntactic, and not simply morphological, restriction can be shown by examining the argument in a preverbal position, commonly used with focussed information. Here we find that the
270 (101) (102)
MARK D ONOHUE emai na No-sampi='e te 3R-pick=3P CORE who NOM ‘Who picked the bananas?’ *No-sampi te loka na CORE banana NOM 3R-pick ‘Who picked the bananas?’
loka? banana emai? who
Similarly, the P of a clause with P enclitics may not be questioned with a content word: (103)
amai na paira? *No-sampi='e te 3R-pick=3P CORE 3PL NOM what ‘What did they pick?’
When it comes to a monovalent clause we cannot question in situ, since there is no non-nominative coding option for the S. The only way to question such an argument is with a cleft construction, a possibility that is available (and frequently used) for questioned As and Ps as well, shown in (107) and (108).21 Questioning Ss (104) No-waliako na ana NOM child 3R-return ‘The child returned.’ (105) *No-waliako na emai NOM who 3R-return (106) Te emai na waliako CORE who NOM return.SI ‘Who returned?’ (= ± ‘Who is it that returned?’)
21
participant that would have been eligible for nominative marking may not be questioned, while a participant that would have appeared in non-nominative te case may be questioned. This is shown in (i), parallel to (101), and (ii), parallel to (102). (i) No-sampi=‘e na loka te emai? 3R-pick=3P NOM banana CORE who ‘Who picked the bananas?’ (ii) *No-sampi te loka na emai? 3R-pick CORE banana NOM who ‘Who picked the bananas?’ Less frequently a non-subcategorised for participant may be questioned in a cleft, following applicativisation, but more commonly adjuncts are questioned in situ. Questions with in situ question words are shown in (i) - (iii), while grammatical and ungrammatical clefts are shown in (iv) and (v). (i) No-kede di ‘umpa? (ii) No-kede di paira? 3R-sit OBL where 3R-sit OBL what ‘Where did they sit?’ ‘What did they on?’ (iii) No-kede-mi te paira? (iv) *Te paira na di-kede=no? CORE what NOM PP-sit=3GEN 3R-sit-LOC.APPL CORE what ‘What did they on?’ ‘What did they on?’ (v) Te paira na di-kede-mi=no? CORE what NOM PP-sit-LOC.APPL=3GEN ‘What did they on?’
DIFFERENT SUBJECTS, DIFFERENT MARKING
271
Cleft questions for As and Ps (107) Te emai na sampi te loka CORE who NOM pick.SI CORE banana ‘Who picked the bananas?’ (= ± ‘Who is it that picked the bananas?’) (108) Te paira na di-sampi? CORE who NOM PP-pick ‘What was picked?’ (= ± ‘What is it that was picked?’) The simplest explanation that can account for these restrictions is one that assumes that the pragmatic function FOCUS, with which question words are inherently associated, is incompatible with the specification of the nominative case/ subject, which implies that some element of the pragmatic function TOPIC adheres to the grammatical function SUBJECT.22 Arguments against treating the na-marked element as simply being a topic can be found in Donohue (1999, 2004), and are similarly described for Tagalog in Kroeger (1993). In short, we have a diathesis that involves the alternation of the grammatical functions subject and object, without demotion. This does not affect the relative positions of the arguments in argument structure, but the pragmatic and syntactic consequences are reflected in the morphosyntax of the language.23 Table 5 shows the different agreement strategies used to mark subjects in different constructions. A dash – indicates that the syntactic role in question is not eligible to appear in that construction. Table 5. Differential subject agreement marking
default main clause experiencer motion+numeral SVC existential passive nominalisation adverbial temporal P-relative clause
22
S prefix prefix or genitive prefix + P enclitic (prefix)/(P enclitic) (prefix) genitive genitive -
A prefix prefix genitive genitive
P P enclitic P enclitic Ø24 genitive genitive genitive25
This is similar to the analysis presented of Chichewa agreement in Bresnan and Mchombo (1987). There are cases in which these principles are violated. We have already seen some cases of atypical verbal marking in the optional genitive marking found with experiencer subjects, seen in 2.4, and with the double marking patterns found with motion+numeral serial verb constructions, also seen in 2.4. Additionally, there is no clearly discernible ‘subject’ in existential, passive and meteorological clauses, and there is no distinction between different arguments in nominalised clauses and clauses that use nominal morphology, such as P-relative clauses and adverbial temporal clauses. See Donohue (1999:75-78, 274-281, 379-385, 411-413, 472-473, 478-480; 2005b) for further details. 24 A secondary P may appear as a core argument in a passive of a trivalent verb (as in ‘they were given the books’), but cannot be marked on the verb with P-enclitic agreement. This contrasts with the closely related Muna (van den Berg 1989), which allows two suffixal agreement positions. 23
272
MARK D ONOHUE
In addition to the major split in subject marking found in Tukang Besi, that between marking a subject that is an S or A, by prefix, or a subject that is P, by enclitic, we have also examined three minor sentence types, those involving experiencer Ss, motion verbs serialised with numeral verbs, and existential clauses, and seen that there are additional patterns of marking. Experiencer verbs may index an S by means of genitive clitics, a marking strategy that dominates the subordinate clause marking. Motion verb + numeral verb serialisations show a double marking pattern in which both S-agreement and P-(like) agreement marking is found. The existential construction is the least fixed, with agreement only optional, and potentially marked by either the prefixes or the enclitics, depending on pragmatic conditions, or by the multifunctional ke(ne) ‘comitative (/ instrumental / ‘and’). The variability of marking options in the existential construction correlates with the fact that there is no grammatical subject in this construction, and this has been shown in table 5 by the bracketing about the data. The verb in a passive construction, which also lacks a subject, only optionally shows agreement for the S. Table 6 shows the case marking strategies available to the different kinds of subjects (and Ss). Table 6. Differential subject case marking
default main clause experiencer motion+numeral SVC existential passive nominalisation adverbial temporal P-relative clause
S na na na (na or ke ‘comit’) na genitive genitive -
A na genitive genitive
P na te ‘core’ genitive genitive genitive
Case marking is more consistent than agreement marking, with the ‘nominative’ na appearing to mark subjects in all main clause types, and the genitive nu in all subordinate clause types. This is true regardless of the marking strategy employed, again the existential showing variation, correlating with the lack of subject in this construction. The passive, by contrast, shows the case marking patterns expected of an S subject, even though the S argument in this construction lacks subject privileges. The final section explores the issues raised in section 4 from a more crosslinguistic perspective.
25
A secondary P may appear in a P-relative clause, not heading the relative clause but as a genitive complement of it. An example of this can be found in (144).
DIFFERENT SUBJECTS, DIFFERENT MARKING
273
5. MODULES OF GRAMMAR Tukang Besi is the south-easternmost language in the Western Malayo-Polynesian branch of Austronesian, and is thus the last language to be (relatively) closely genetically related to the Philippine-type languages, famous for their unusual and controversial subject properties. It is also one of the westernmost languages in Indonesia to show pronominal head-marking, a feature that is widespread in eastern Indonesia, but absent in the west (and in the Philippines). It is worth comparing the morphosyntax of the Philippine-type languages, and of a more familiar language such as English. In Tagalog a clause with a predicate that selects for two arguments, such as kita ‘see’, allows for either of those arguments to be coded as subject, indicated by the use of nominative forms, in these examples the determiner ang, and the choice of prefix on the verb. See Kroeger (1993) for explicit arguments about the status of these arguments as subjects, building on discussion in Schachter and many others.26 (109) (110)
Naka-kita ng aso ang bata AV-see GEN dog NOM child ‘The child saw a dog.’ Na-kita ng bata ang aso PV-see GEN child NOM dog ‘The child saw the dog.’
While the alternation in (109) and (110) could be (and has been) analysed as an instance of a ‘familiar’ voice alternation such as that seen between an active and a passive clause, or an active and an antipassive, there are substantial differences, which may be summarised as follows: • in the Philippine case seen here in Tagalog the amount of morphological marking is equal for both voice choices; no voice selection is morphologically unmarked with respect to the other. • in the Philippine case the voice alternation does not involve ‘demotion’ of the Agent when the Patient is coded as subject. In both voice selections we have two core arguments, one of which is subject. • in the Philippine-type voice systems there are voice choices for more arguments than are subcategorised for by the predicate. This can be seen in (109) and (110), which show that a monovalent predicate like lunsad ‘alight’ may select a participant that is not subcategorised for by the verb, if the verb is marked with the appropriate morphology, in this case -an ‘dative voice’.
26
I am simplifying (though not misrepresenting) the facts of Tagalog morphology somewhat dramatically for the purposes of exemplification here, since the argumentation is secondary to the main discussion of this chapter. In particular the degree of lexical specification found in affixal possibilities is not mentioned at all in this chapter; McFarland (1976) and De Guzman (1978) provide discussion on this topic.
MARK D ONOHUE
274 (112) (113)
Lunsad=siya mula sa kotse alight=3SG.NOM from DAT car ‘He alighted from the car.’ Lunsar-an=mo ang kotse alight-DV=2SG.GEN NOM car ‘Alight from the car.’
Sells (2000) argues for a raising interpretation for the nominative arguments, in which the nominative argument, siya in (112) and ang kotse in (113), is in a structurally higher position than the rest of the arguments, and so is more visible for a range of grammatical processes. Similarly, ang bata in (109), and also ang aso in (110), appear in a position different to that of the rest of the clause. This is shown in outline in (114), which models (110). Structural model of (110) (114)
S’ S
Nakita
ang aso ng bata
As with Tukang Besi, the case marking option (‘nominative’) is the same for all subjects, but the choice of morphology used on the verb to index that argument varies depending on the semantic and syntactic role of that argument. While the Tagalog voice morphemes are reflected in Tukang Besi, they are not found in main clause uses, being relegated to subordinate functions. In their place, the pronominal agreement system has been coopted, showing an (eastern) Indonesian tendency towards head marking replacing the older Austronesian system.27 This is not such a radical departure from the earlier; recall that Sells (2001) analyses the voice system in Tagalog as involving (abstract) pronominal elements.28 In Tukang Besi we have simply seen the replacement of these abstract pronominal elements with overt ones in main clauses. In doing so we have ended up with a system in which the paradigm used for marking subject on the verbs varies with the voice of the clause. The syntactic role, S and A on the one hand or P on the other, is monitored consistently by the agreement marking, while the grammatical function is indicated consistently
27
28
See Donohue (2002) for discussion. Of the morphemes seen here, Tagalog ‘actor voice’ is cognate with Tukang Besi ‘S-(and A) infix (in subordinate clauses)’, and Tagalog -an ‘dative voice’ is cognate with Tukang Besi -‘a ‘nominaliser’. A development of this analysis, which treats the ‘voice morphology’ on the verb as pronominal, is currently being pursued by the present author, in collaboration with Alastair Butler.
DIFFERENT SUBJECTS, DIFFERENT MARKING
275
by the case marking, thus showing that the functions of the two subject-marking systems are quite distinct. The fact that there is no dedicated voice-monitoring morphology might come as a suprise to linguists acquainted with the voice systems of western Eurasia. There are, however, analogous voice systems found elsewhere, showing a similar lack of dedicated morphology. Lango (Noonan and Bavin Woock 1978; Foley and Van Valin 1984; Noonan 1992) has a voice system without any dedicated voice morphology, as shown in (115) and (116). Here there is consistent agreement on the verb for both arguments, but positional variation correlates with subject status. Active subject object (115) Dákó ò-jwát-ò lócà woman 3SG-hit-3SG man ‘The woman hit the woman.’ Passive subject object (116) Lócà dákó ò-jwát-ò man woman 3SG-hit-3SG ‘The man was hit by the woman.’ In Palu’e and Manggarai, both languages from Flores, in southern Indonesia (Donohue 2005a; Arka and Kosmas 2005), a voice alternation operates with no verbal morphology. In Manggarai there is a change in the VP-final agreement clitic that marks subject, and prepositional marking of the Agent, in the passive construction, shown in (118). The active construction in (117) shows a VP that agrees with the Agent as subject. Active subject object (117) Aku [VP cero latung ]=k 1SG fry corn=1SG ‘I fry/am frying corn.’ Passive subject oblique (118) Latung hitu [VP cero l=aku ]=i corn that fry by=1SG=3SG ‘The corn is (being) fried by me.’ In Palu’e, as in Lango, word order alone codes the difference in grammatical functions. There is no change in the marking (or absence of marking) on the verb, and no change in the use of bare NPs. Active subject object (119) Ata wai lie ata laki person woman see person man ‘The woman saw the woman.’
276
MARK D ONOHUE
Passive subject oblique (120) Ata laki ata wai lie person man person woman see ‘The man was seen by the woman.’ The Manggarai case, in which there is at least marking for non-core status of the Agent in the passive clause, has a direct parallel in the non-literary passive constructions of many Chinese languages. The Hokkien examples in (121) and (122) illustrate the active-passive alternation, showing the use of the multifunctional morpheme ho to mark the passive. (123) shows that the agent in (122) is an oblique, not an adjunct, since it is not optional in the passive construction. Active subject object (121) I phah liau hit-e hit PERF that-MOD 3SG ‘He hit that person.’ Passive subject oblique (122) Hit-e lang ho i that-MOD person HO 3SG ‘He hit that person.’ (123) *Hit-e lang ho phah liau.
lang person phah liau hit PERF
Comparing with English, or other languages of the western European model, we are struck by the relative autonomy of the argument-structure determined syntactic roles and the case marking, and this is due to the fact that the voice system of Tukang Besi, and other Western Malayo-Polynesian languages, is non-demotional, as described in section 4. This means that a bivalent clause remains bivalent, regardless of which voice, A-as-subject or P-as-subject (which could just as well be labelled ‘direct’ and ‘inverse’), is selected. This has the consequence of bivalent clauses with subject Ps having a second core argument, the A, which must be thought of as the object of the clause. Since the verbal agreement for this argument is with the same set of prefixes that are used for agreement with Ss and As in all clause types, we can see that, essentially, the same agreement prefixes can be used to mark either a subject or an object, depending on the clause type. By contrast the P enclitics are only used to mark a subject on the verb, since they are not present in a clause that does not have a subject P. In terms of case marking, na is essentially used to mark subjects, and te for objects (with the existential and passive constructions as exceptional cases). In contradistinction to the materials presented in Aissen (1999), we see that Tukang Besi (as well as many other western Austronesian languages) does not have a high-ranked constraint of the form *OBJ/AGT, and that this non-subject ‘Agt’ (= A) is case-marked identically to an object ‘Pat’ (= P) (taking ‘object’ to be roughly equivalent to ‘non-subject term’). I suggest that the extreme regularity of subject marking in languages such as English, languages which have played the major part in informing modern syntactic theory, is to a large part an artefact of the absence of object As. In a language with a demoting voice system, in which the ‘Agt’ is coded
DIFFERENT SUBJECTS, DIFFERENT MARKING
277
as oblique and the ‘Pat’ is the single core argument of the predicate, there is no mechanism for agreement with the ‘Agt’. Since the ‘highest role’ feature, the feature that defines S and A as a group, is relevant to the description of all languages in at least some constructions it follows that the Agt might retain marking as a (core) category regardless of the voice choice, if the voice is non-demotional.29 This is especially so if the language has a grammatical subject that is clearly related to the pragmatic notion of topic, a characterisation that applies to Tukang Besi, as seen in section 4 (and is true, to a greater or lesser extent, in most Austronesian languages, certainly in those west of New Guinea). Unlike a more syntactically-centred grammatical function, one that can be defined in terms of the syntactic roles A, S and P, the Tukang Besi subject is determined pragmatically. This means that syntax does not play a significant role in allowing, through the constructions that are restricted on the basis of grammatical functions, determination of a relationship between the grammatical function subject and the syntactic roles A and P in a bivalent clause, and thus semantic notions such as actor / proto-agent / ‘Agt’, and undergoer / proto-patient / ‘Pat’ cannot be automatically assigned to a particular grammatical function just on the basis of their relative position in argument structure. In a language such as English knowledge of the identity of the subject will automatically give information about the identity of the most agentive core argument in the predicate In Tukang Besi this ‘agent’identifying (or at least ‘agent’-confirming) side-effect of the well-construed subject argument is absent, and so some ‘freeing up’ of the degree to which subject marking on the verb is performed by a unique paradigmatic set can be expected, since we require more overt morphological means to identify the syntactic grouping that cannot, in this language, be easily identified through grammatical function identity. In short, the unusual grammatical function selection system of Tukang Besi, with its lack of demotion in the pronominal voice system, sets up the conditions necessary for the differential subject marking system in which a given pronominal agreement paradigm, the S,A prefixes, retain constant reference to the argument they mark, while indicating different grammatical functions (subject or object) depending on the voice construction, direct or inverse, in which they occur. REFERENCES Aissen, J. (1999). Markedness and subject choice in Optimality Theory. Natural Language and Linguistic Theory 17, 4, 673-711. Andrews, A. (1985). The major functions of the noun phrase. Language Typology and syntactic description: Volume I, clause structure. Ed. by T. Shopen. Cambridge: Cambridge University Press, 62-154. Arka, I W. and J. Kosmas (2005). Passive without passive morphology? Evidence from Manggarai. Voice in Western Austronesian languages. Ed. by I W. Arka and M. Ross. Canberra: Pacific Linguistics.
29
Constructions such as control structures with verbs like ‘want’, and (to a lesser extent) textual cohesion in narratives, show S,A pivots in (nearly) all languages, and so cannot be taken as evidence for the grammatical function ‘subject’. For further discussion, see (amongst others) Manning (1996) and Falk (2000a).
278
MARK D ONOHUE
Arka, I W. and C.D. Manning (1998). Voice and Grammatical Relations in Indonesian: A New Perspective. Ed. by M. Butt and T.H. King. On-line LFG proceedings. Available at http://wwwcsli.stanford.edu/publications/ LFG3/lfg98-toc.html. Bell, S.J. (1976). Cebuano Subjects in Two Frameworks. PhD thesis, Massachusetts Institute of Technology. Bell, S.J. (1983). Advancements and Ascensions in Cebuano. Studies in relational grammar 1. Ed. by D.M. Perlmutter. Chicago: The University of Chicago Press, 143-218. Berg, R. van den, (1989). A Grammar of the Muna Language. KITLV, Verhandelingen 139. Dordrecht/Providence: Foris. Bresnan, J. and J. Kanerva (1989). Locative inversion in Chichewa: a case study of factorisation in grammar. Linguistic Inquiry 21: 1-50. Bresnan, J. and S.A. Mchombo (1987). Topic, pronoun, and agreement in Chichewa. Language 63, 4, 741-782. Comrie, B. (1978). Ergativity. Syntactic Typology: Studies in the Phenomenology of Language. Ed. by W.P. Lehmann. Sussex: The Harvester Press, 329-394. De Guzman, V.P. (1978). Syntactic Derivation of Tagalog Verbs. Oceanic linguistics special publications No. 16. Honolulu: The University Press of Hawaii. Donohue, M. (1999). A Grammar of Tukang Besi. Berlin: Mouton de Gruyter. Donohue, M. (2002). Voice in Tukang Besi and the Austronesian voice system. The History and Typology of Western Austronesian Voice Systems. Ed. by F. Wouk and M. Ross. Canberra: Pacific Linguistics 518, 81-99. Donohue, M. (2004). Voice oppositions without voice morphology. Proceedings of AFLA 11, ZAS, Berlin 2004. Ed. by P. Law. University of Newcastle. ZAS Papers in Linguistics Nr. 34 – October 2004: 7388. Berlin: Zentrum für Allgemeine Sprachwissenschaft, Typologie und Universalienforschung, Berlin. Donohue, M. (2005a). The Palu’e passive: from pragmatic construction to grammatical device. The Many Faces of Austronesian Voice Systems: Some New Empirical Studies. Ed. by I W. Arka and M. Ross. Canberra: Pacific Linguistics, 59-85. Donohue, M. (2005b). Structure is not syntax: passive functions in Tukang Besi. Proceedings of the 12th Annual Conference of the Austronesian Formal Linguistics Association. Ed. by J. Heinz and D. Ntelitheos. University of California at Los Angeles Working Papers in Linguistics 12, 73-89. Available online at http://www.linguistics.ucla.edu/faciliti/wpl/issues/wpl12/wpl12.htm England, N. (1983). A Grammar of Mam, a Mayan Language. Austin: University of Texas Press. Falk, Y. (2000a). Pivots and the theory of grammatical functions. Proceedings of the LFG00 conference, University of California at Berkeley. Ed. by M. Butt and T.H. King. Stanford: CSLI Publications, 123-139. Falk, Y. (2000b). Philippine subjects in a monostratal framework. The Proceedings of the Austronesian Formal Linguistics Association VI. Ed. by C. Smallwood and C. Kitto. Toronto Working Papers in Linguistics, 133-136. A more complete version of this (sadly truncated) paper is available at http://pluto.mscc.huji.ac.il/~msyfalk/PhilippineSubjects.pdf. Foley, W.A. and R.D. Van Valin (1977). On the viability of the notion of subject in universal grammar. Proceedings of the Berkeley Linguistic Society 3. Ed. by K. Whistler et al., 293-320. Foley, W.A. and R.D. Van Valin (1984). Functional Syntax and Universal Grammar. Cambridge: Cambridge University Press. Gerdts, D.B. (1988). Object and Absolutive in Halkomelem Salish. New York: Garland Publishing. Keenan, E.L. (1976). Towards a universal definition of Subject. Subject and Topic. Ed. by C.N. Li. New York: Academic Press, 303-333. Kiparsky, P. (1997). The rise of positional licensing. Parameters of Morphosyntactic Change. Ed. by A. van Kemenade and N. Vincent. Cambridge: Cambridge University Press. Kroeger, P. (1993). Phrase Structure and Grammatical Relations in Tagalog. Stanford: CSLI publications. Manning, C. (1996). Ergativity: Argument Structure and Grammatical Relations. Stanford: CSLI Publications. McFarland, C.D. (1976). A Provisional Classification of Tagalog Verbs. Study of languages and cultures of Asia and Africa monograph series No.8. Tokyo: Toyo Shuppan and Tokyo Gaikokugo Daigaku: Institute for the study of languages and cultures of Asia and Africa. Noonan, M. (1992). A Grammar of Lango. Berlin: Mouton De Gruyter.
DIFFERENT SUBJECTS, DIFFERENT MARKING
279
Noonan, M. and E. Bavin-Woock. (1978). The passive analog in Lango. Proceedings of the Fourth Annual Meeting of the Berkeley Linguistics Society. Berkeley: Berkeley Linguistics Society, 128-139. Plank, F. (1984). Objects: Towards a Theory of Grammatical Relations. New York: Academic Press. Schachter, P. (1976). The Subject in Philippine languages: Topic, Actor, Actor-Topic or None of the above. Subject and Topic. Ed. by C.N. Li. New York: Academic Press, 491-518. Schachter, P. (1977). Reference-related and role-related properties of subjects. Syntax and Semantics 8: Grammatical Relations. Ed. by P. Cole and J.M. Sadock. New York: Academic Press, 279-305. Sells, P. (2000). Raising and the Order of Clausal Constituents in the Philippine Languages. Formal Issues in Austronesian Linguistics. Ed. by I. Paul, V. Phillips and L. Travis. Dordrecht: Kluwer Academic Publishers, 117-143. Sells, P. (2001). Form and function in the typology of grammatical voice systems. Optimality-Theoretic Syntax. Ed. by G. Legendre, J. Grimshaw and S. Vikner. Cambridge: MIT Press, 355-391. Van Valin, R.D. and R.J. LaPolla (1997). Syntax: Structure, Meaning and Function. Cambridge: Cambridge University Press. Wunderlich, D. (1997). Cause and the structure of verbs. Linguistic Inquiry 28: 27-68.
MARIAN KLAMER
CHAPTER 12 Differential marking of intransitive subjects in Kambera (Austronesian)1
1. INTRODUCTION Kambera is one of the five or six indigenous languages spoken in the eastern region of the island of Sumba in Eastern Indonesia. It has approximately 150,000 speakers, and it is classified as belonging to the Central Malayo-Polynesian (CMP) subgroup of Austronesian languages.2 Native speakers refer to the language as hilu Humba, the ‘Sumba language’ (in contrast to hilu Jawa ‘Indonesian’). In the literature it has been referred to as ‘Sumbaneesch’ (Wielenga 1909), ‘Sumba(a)sch’ (Onvlee 1925), ‘Kamberaas’ (Onvlee 1984), and ‘Bahasa Sumba/Kambera’ (Kapita 1982). Klamer (1998a) is a recent grammar of the language.3 The information presented in this chapter is based on a corpus of 12-hours of spontaneous speech, plus additional elicitation, collected in Sumba during 12 months of fieldwork between 1991 and 1994 in one village, Katàka. This paper presents five different ways in which the single arguments of intransitive clauses in Kambera may be cross-referenced on the verb by pronominal clitics. The term ‘subject’ of the title of this paper refers to these clitics; there is no case marking on NPs in Kambera.4 The paper is structured as follows. Section 2 presents a brief grammatical overview of the language, focussing on the argument marking function of pronominal clitics. Section 3 describes the five markings of S 1
Parts of this paper appeared as Klamer (1998b), and I would like to thank Helen de Hoop for suggesting that the data discussed in that paper might be a useful contribution to the 2004 Nijmegen workshop and the present volume. Two anonymous referees gave insightful comments and suggestions; their input is acknowledged with thanks. I also wish to thank Andrej Malchukov and Peter de Swart for their helpful comments. 2 Blust (1993) is the most recent proposal regarding the constituency of the CMP subgroup, and contains references to earlier work on the subgrouping of languages of Eastern Indonesia. Note that unlike Tukang Besi, which belongs to the West Malayo Polynesian branch (Donohue, this volume), Kambera is only remotely related to well-known Phillipine-type languages like Tagalog, and does not share any of the unusual subject properties of it. 3 Readers interested to follow up on the Kambera data may also want to consult publications on Kambera that appeared subsequent to the 1998 grammar, including Klamer (2000, 2001, 2002a, 2002b, 2004, 2005, 2006). 4 This is one of the more significant morpho-syntactic contrasts between Kambera and Tukang Besi (Donohue, this volume) and other Phillipine-type languages such as Tagalog.
281 H. de Hoop and P. de Swart (eds.), Differential Subject Marking, 281–299. © 2008 Springer.
282
MARIAN KLAMER
arguments in Kambera. Section 4 presents a summary and discusses some general implications of the data. 2. GRAMMATICAL OVERVIEW Kambera is a head-marking language. A Kambera sentence is built on the basis of a ‘nuclear’ (or ‘minimal’) clause, which consists of a predicate phrase (PredP) (a verbal or nominal phrase that functions as the predicate of the clause) as well as a clitic cluster attached to that PredP.5 In the diagram in (1), the nuclear clause is dominated by the lowest S node. The diagram shows that this S, as well as the one above it, is a non-configurational structure, while the structure higher up the tree is clearly more hierarchical. (1)
S’ Topic
S’ Conj
S Neg
NP
S NP
One proclitic
S
PredP
NP
NP
PP
Several enclitics
(Modifier) Head (Modifier) At the top of the diagram, we find a position for a topicalised, left-dislocated constituent; followed by positions for a conjunction and a negation. Adjoined to the
5
The PredP is a phrase since, apart from its head, it may contain up to two modifiers. In verbal predicates, modifiers (adverbs) are separate words occurring directly adjacent to the head (verb), while the clitic cluster attaches to the outer edge of the phrase that comprises head and modifier(s). An example of a complex PredP is the first clause of (17), which contains a head verb (ita ‘see’) and two modifiers (lalu ‘too’, dí ‘Emphasis’).
DIFFERENTIAL MARKING OF INTRANSITIVE SUBJECTS IN KAMBERA
283
nuclear clause S, there are positions for optional NPs: maximally two precede S, maximally two follow it. Postpredicate NPs are followed by PP adjuncts. These NPs and PPs are within the scope of the negation and conjunction; they can only occur outside this domain when they are topicalised. (For motivation of (1), cf. Klamer 1998a:77-89). The grammatical relations assumed for Kambera are subject (comprising S/A) and object (O).6 Kambera argument alignment uses a paradigm of free pronouns, four simple pronominal clitic sets, and one complex one which expresses S in continuative aspect. The paradigms are given in (2): (2) Kambera pronouns and pronominal clitic paradigms GEN ACC DAT pronoun NOM nyungga ku-nggu -ka -ngga 1SG nyumu (m)u- -mu -kau -nggau 2SG nyuna na-na -ya -nya 3SG nyuta ta-nda -ta -nda 1P.I nyuma ma-ma -kama -nggama 1P.E nyimi (m)i-mi -ka(m)i -ngga(m)i 2PL nyuda da-da -ha -nja 3PL
Cont.aspect 7 -nggunya -munya -nanya -ndanya -manya -minya -danya
Synchronically, these are distinct paradigms, but observe the formal relations that pertain between the paradigms: the genitive enclitics are prenasalised forms of the nominative proclitics (except for 3pl.NOM da– and 3pl.GEN –da), and the dative clitics are prenasalised forms of the accusative ones (though 3pl.ACC –ha and 3pl.DAT –nja involves more than just simple prenasalisation).8 The clitic cluster may contain up to nine clitics, and apart from the pronominal clitics, it contains modal and aspectual clitics in various shapes and combinations, for example marking emphasis, and aspect (ka ‘perfective’, pa ‘imperfective’, i ‘iterative’). 9 Kambera does not mark tense grammatically. Kambera has two types of O: direct O (Patients, Themes), and indirect O (Recipients, Benefactives, Goals, Locations), and both may be marked (also simultaneously) on the PredP. In a transitive declarative clause, A can be nominative or genitive, and a direct O can be accusative or dative:10 6
A = most agent-like argument in a transitive clause, O = most patient-like argument in a transitive clause, S = single argument of an intransitive clause (cf. Dixon 1979, 1994, Andrews 1985). Structural motivation to view Kambera S/A as ‘subject’ and O as ‘object’ is presented in Klamer (1998a: 72-77). 7 This paradigm diachronically derives from a combination of the Genitive paradigm plus a 3sg Dative clitic; see the discussion in section 3.3. below. 8 For more discussion of the paradigm forms, see Klamer (1998a: 62). 9 It is beyond the scope of this paper to present a formal account of the very complex patterns of clitic placement in a Kambera clause. In Klamer (1997) the placement of the Kambera clitics is analysed as the result of a morphological spell-out of morphosyntactic feature bundles at the interface between syntax and prosody: the postlexical level. 10 Notational conventions: In the notation of the examples a clitic is separated from its (syntactic) host by a dash [-] and an affix is separated from its base by a dot [.] when this is relevant for the discussion.
284
MARIAN KLAMER
(3) a. Ku-tàru-ya 1SG.NOM-watch-3SG.ACC ‘I watch him.’ b. Tàru-nggu-nya watch-1SG.GEN-3SG.DAT ‘I am watching him.’ The canonical marking of direct objects is accusative, as in (3a), while the canonical marking of the indirect object is dative, as in (4a). In case of a ditransitive verb, as in (4), the indirect O is always cross-referenced if it is definite, (4a).11 In addition, the direct O may also be cross-referenced if it is definite. In such cases, it follows the indirect O marking clitic, as in (4b). In this position, it must be dative because of clitic cluster restrictions. (4) a. I
Ama na-kei-nja rí father 3SG.NOM-buy.for-3PL.DAT vegetable ‘Father buys them vegetables.’ (indefinite Patient) b. I Ama na-kei-ngga-nya ART father 3SG.NOM-buy.for-1SG.DAT-3SG.DAT ‘Father buys it for me.’ (definite Patient) ART
Direct and indirect O share the property of only being cross-referenced if they are definite. The grammatical definiteness of an NP is marked by the presence of an article (na for singulars, da for plurals, i for humans). The absence of the article renders an NP indefinite. Definite NPs are cross-referenced on the predicate and optionally doubled, indefinite object NPs are not cross-referenced, and are either left implicit (to be inferred from the context), or expressed as indefinite (‘bare’) NPs. Syntactically, there are two major clause types: clauses with a verbal predicate (of which we saw some examples above), and clauses with a non-verbal (nominal, numeral, or locational) predicate. Kambera has no copular verb; S is simply attached to the nominal (or numeral, or locational) PredP. The S of non-verbal predicates is always marked as O, with either an accusative, as in (5) and (6), or with a dative, (7).
Accents on vowels mark contrastive vowel length. Note on translations: 3sg pronominals in Kambera are neutral with respect to gender but are translated as male, unless the context demands otherwise. Verbs are not marked for tense; the tense used in the translations was determined by the original context of the utterances. 11 Though clauses with indefinite indirect objects are rare in Kambera, they do exist. In such clauses, the verb has an applicative suffix, and the direct object is also indefinite or implicit (cf. Klamer 1998: 198, 203). For example: (i) Jàka ngga-nggamu bia, nda na-wua.ng-a If RED-who just NEG 3SG.NOM-give.APP-MOD ‘He doesn’t give it to just anyone.’ (ngga-nggamu lit. ‘whoever’)
DIFFERENTIAL MARKING OF INTRANSITIVE SUBJECTS IN KAMBERA
285
(5) [Tau hàmu ]NP-ya person be.good-3SG. ACC ‘He’s a good person.’ (6) [Lai nú]PP-ya LOC DIST-3SG.ACC ‘He/she/it is there.’ (7) [Mbapa-nggu]-nya husband-1SG.GEN-3SG.DAT ‘He is my husband.’ The dative marking of S in (7) is another instance of the idiosyncratic constraint on pronominal clitic clusters, according to which the second pronominal enclitic must always be dative. In the following sections, I discuss various additional ways in which the S, the argument of intransitive predicates, can be marked. I show that the differential subject marking in Kambera consists of five distinct markings of S: nominative (3.1), genitive (3.2), genitive plus dative (3.3), nominative plus accusative (3.4), and accusative (3.5). We will see that the variable markings depend on information from various subcomponents of the grammar, so that a formal account cannot be uniform for all markings. Classic ways to explain why S’s are marked like O’s such as the account of Burzio (1981), assume that the lexical entry of intransitive verbs contains information about the internal/external status of their single argument, as well as its semantic role. According to these properties, intransitive verbs are assigned to distinct lexical classes (e.g. ‘unergative’ verbs such as run versus ‘unaccusative’ verbs such as fall). It will be shown that the Kambera data cannot be accounted for along these lines. 3. THE MARKING OF S IN KAMBERA In Kambera, there are five different ways to mark S with pronominal clitics, and these markings are determined by a variety of factors, including the discourse function of a clause, its aspectual properties, and the amount of ‘control’ S has over the action or event. We will see that none of the S-marking morphemes are selected on the basis of information encoded in the lexical entry of the verb alone. 3.1 Nominative Subjects, both transitive (A) and intransitive (S), are canonically nominative. S is nominative with both active and non-active verbs: (8) Da-tama la kurung 3PL.NOM-enter LOC room ‘They enter the room.’
286
MARIAN KLAMER
(9) Na
ài na-tambuta wood 3SG.NOM-drop.out ‘That tree is uprooted.’ ART.SG
dàngu with
amung root
3.2 Genitive In Kambera, many clauses have a S/A that is marked with a genitive enclitic. I refer to these as nominal clauses, and simple examples are the first clauses in (10) and (11): (10) [Bidi njoru-na na ài]CLAUSE [ba talànga just.now fall-3SG.GEN ART.SG wood CNJ while nàhu-ngga] CLAUSE12 move.away-1SG.DAT ‘The tree fell when I walked by.’ na bi (11) [Ka tama-du-na] CLAUSE [hi na-wanga-ya CNJ enter-EMP-3SG.GEN CNJ 3SG.NOM-open-3SG.ACC ART.SG real ngara-na] CLAUSE mouth-3SG.GEN ‘So he goes inside then he opens its beak...’ In clausal sequences such as these, the second clause expresses the events that constitute the main narrative line, while the first, nominal clause presents the background information. Although nominal clauses are dependent in discourse, syntactically they are not: they may be juxtaposed or coordinated to another clause, or govern a controlled clause. Kambera nominal clauses have the external syntax of possessed NPs. They can be clefted, or occur in comparisons: (12) Hama pingu-mi dàngu ama-mi father-2PL.GEN be.same know-2PL.GEN and ‘You (pl) and your fathers are equally bright.’ (lit. (The) same (is) your knowing and your father.’13 Nominal clauses may be marked for definiteness with an article (sg. na, pl. da), as illustrated in (12). If a nominal clause is definite, it can be crossreferenced as an argument of the main verb, as in (13), (16) and (17). 12
13
Incidentally, this sentence illustrates an additional marking of S which only applies to the S of four directional motion verbs. These verbs are derived from deictic elements by the addition of a suffix .ng (ni.ng(u) ‘be (at speaker)’, na.ng(u) ‘come (towards addressee)’, nàmu.ng ‘move towards speaker’ and nàhu.ng ‘move (past/away from speaker)). The S of these verbs is obligatorily marked with a dative clitic, and cannot be marked otherwise, cf. Klamer (1998a:148-151). The literal translation of this sentence cannot be ‘the knowing of [you and your father] is the same’, since pingu-mi ‘your knowing’ is an indefinite nominal clause. What is juxtaposed here is the proposition ‘your knowing’ and ‘your father’ (rather than ‘you’ and ‘your father’).
DIFFERENTIAL MARKING OF INTRANSITIVE SUBJECTS IN KAMBERA
287
(13) Na-muda-a nuna, jàka jia [na pala-ndaS] EXIST ART.SG cross-3PL.GEN 3SG.NOM easy-just DIST.3SG if ‘That’s easy for us to cross.’ (lit.: ‘It is easy that one, if (it's) our crossing’) Internally, nominal clauses are indeed clauses: they may contain mood and aspect clitics, as in (14a), as well as negations, as in (15). Such grammatical elements cannot occur inside possessed NPs, as illustrated in (14b). Nominal clauses may also contain full NPs; for example, (15) contains the NP da ana-na ‘his children’. (14) a. Hili mandai-ma-naS-i... again be.long-EMPH-3sg.GEN-ITER ‘It (was) some time later...’ b. *Uma-ma-na-i house-EMP-3sGEN-ITER (15) Panau-nya nyuna ka14 àmbu palu-na-nja-i CNJ NEG.I hit-3SG.GEN-3PL.DAT-ITER tell-3SG.DAT he ana-na child-3SG.GEN ‘Tell him that he shouldn't hit his children (anymore).’
da ART.PL
Nominal clauses may function as syntactic complements when they are crossreferenced as the S or O of a main verb. Such ‘complement’ nominal clauses are a tiny minority in my database – normally nominal clauses occur as independent main clauses. Two examples of ‘complement’ nominal clauses are (16) and (17). In (16) the nominal clause is a definite NP that follows the main verb and is marked as the S of that verb (hàmu ‘be good’). In (17) we find a nominal clause in O function. ndoku (16) Nda naS-hàmu NEG 3SG.NOM-be.good NEG.EMPH [na ludu-na na tau la rudung] ART.SG sing-3sg.GEN ART.SG person LOC night ‘It's not nice at all that people sing at night.’ [na karuhi-na banda] (17) Nda ku-pí-nyaO NEG 1SG.NOM-know-3SG.DAT ART.SG demand-3SG.GEN cattle ‘I do not know about his demanding cattle.’ Example (18) contains two nominal clauses, one being the main clause, the other functioning as the O of the verb ita ‘see’, being crossreferenced with –nya:
14
Despite the fact that it translates as ‘that’ in English, ka is glossed as a conjunction (like ba, hi, jàka, etc.). All of these are coordinating conjunctions, Kambera has no subordinating conjunctions or complementisers. See Klamer (1998a:143, and section 8.2).
288
MARIAN KLAMER
(18) ...ba lalu ita dí-na-nyaO-i-ka nú [na lalu CNJ too see EMPH-3SG.GEN-3SG.DAT-ITER-PFV DIST ART.SG too mbuha-na-nya na ana njara parai-na nyuna like-3SG.GEN-3SG.DAT ART.SG child horse work-3SG.GEN he yena i Umbu Mada]O this.one ART.PERS Sir Mada ‘...because only too well did he see... the big liking of the foal by Sir Mada.’ (i.e. that Sir Mada liked the foal very much) This example is another illustration that a nominal clause is more than a nominalised verb: the entity crossreferenced on the main verb here is the constituent [na lalu....... i Umbu Mada]. This is a nominal constituent which contains an article, an adverb, a verb, two pronominal clitics, a Patient NP (na ana njara) and an NP that contains the Agent i Umbu Mada. (For more discussion, see Klamer 2006.) In sum, many Kambera clauses (whether or not syntactically independent) have a genitive S/A. These nominal clauses have the external syntax of NPs, but their internal structure is clausal. Despite their usual syntactic independency, the discourse status of nominal clauses is dependent – their core function is to present the background information in the discourse, instead of the expressing the main narrative line. Typically (though not exclusively) nominal clauses express irrealis mood, and are non-agent oriented. In other words, the genitive marking of S or A relates to various subcomponents of the grammar: not only the discourse function of a clause, but also its modal (irrealis) properties, and whether the clause is agent-oriented or not (see Klamer 1998a, section 4.2.1, and 5.3-5.5). 3.3 Genitive and dative The third way to mark S is by using a combination of a genitive and a dative enclitic. The genitive marks person and number of S, while the dative is always the same 3sg form -nya. Both active and stative verbs can take such a complex S marking: (19) Laku-nggu-nya go-1SG.GEN-3SG.DAT ‘I’m going’ (20) Poki-na-nya? be.blind-3SG.GEN-3SG.DAT ‘Is he blind?’
Mm, poki-na-nya yes be.blind-3SG.GEN-3SG.DAT ‘Yes, he’s blind.’
This particular clitic combination expresses ‘continuative’ aspect, that is, it marks the event specifically as continuous, as in (21a), also in combination with any of the three aspect enclitics, as in (21b-d). In contrast, marking the S with nominative (22a-c) or genitive (23) allows for various interpretations, including past, present, future, completed, and uncompleted, depending on the grammatical context.
DIFFERENTIAL MARKING OF INTRANSITIVE SUBJECTS IN KAMBERA
289
(21) a. Mutung-na-nya na uma burn-3SG.GEN-3SG.DAT ART.SG house ‘The house is burning/aflame.’ b. Mutung-na-nya-ka na uma burn-3SG.GEN-3SG.DAT-PFV ART.SG house ‘The house has been burning/aflame.’ c. Mutung-na-nya-pa na uma burn-3SG.GEN-3SG.DAT-IPFV ART.SG house ‘The house is still burning/aflame.’ d. Mutung-na-nya-i na uma burn-3SG.GEN-3SG.DAT-ITER ART.SG house ‘The house is burning/aflame again.’ (22) a. Na-mutung na uma 3SG.NOM-burn ART.SG house ‘The house burns/is burned/is burning/will burn.’ etc. (depending on context) b. Na-mutung na uma jàka u-pajulu wàngu epi fire 3SG.NOM-burn ART.SG house if/when 2SG.NOM-play use ‘The house will burn down if/when you play with fire.’ c. Na-mutung-ka na uma 3SG.NOM-burn-PFV ART.SG house ‘The house is burned (down).’ (23) Muda’a mutung-na na uma ba u-pajulu wàngu epi fire easy burn-3SG.GEN ART.SG house CNJ 2SG.NOM-play use ‘When you play with fire the house may burn down easily.’ (Lit. ‘It is easy for the house to burn down when you play with fire.’) 15 The continuative aspect construction is formally related to the nominal clause. The GEN-DAT marked forms are diachronically derived from nominal predicates, where the head is a nominal clause rather than a noun. For example, Mbapa-nggunya ‘husband-1sg.GEN-3sg.DAT’ in (7) is a clause with a predicate that consists of the possessed NP mbapa-nggu ‘my husband’. This nominal predicate occurs in the equative nominal construction ‘He (is) my husband’. The occurrence of the dative clitic rather than the accusative, which is normally the S marker on nominal predicates (see (5) and (6)), arises from a linear restriction on clitic co-occurrence, which states that second pronominal enclitics can only be dative. This implies that a genitive S-marking enclitic cannot be linearly followed by an accusative, but rather must be followed by a dative, even though the clitics belong to different syntactic constituents (i.e., NP versus clause). Possessed nominal predicates may have a verbal head as well. In such cases, the nominal predicate is in fact a nominal clause (as discussed in section 3.2). For
15
In (23) the initial clause (‘the house may burn down easily’) is a nominal clause, in (22b) it is not. In (23), ‘playing with fire’ is thus presented as the main event in discourse, the nominal clause being its possible result, while in (22b) both clauses have the same discourse status.
290
MARIAN KLAMER
example, in (24), the head of the nominal predicate is the nominal clause ‘the burning of the house’, and the S is accusative -ya: (24) [Mutung-na na uma]-ma-ya, hi na-meti tau.... burn-3SG.GEN ART.SG house-EMPH-3SG.ACC CNJ 3SG.NOM-die person ‘It (is) [because of] the house burning down that people died...’ Because of clitic cluster restrictions, the accusative -ya is replaced by dative -nya when it is linearly adjacent to the genitive: (25) [Mutung-na]-ya Burn-3SG.GEN -3SG.ACC ‘?’
clitic cluster restrictions >
Mutung-na-nya burn-3SG.GEN-3SG.DAT lit. ‘It (is) its burning.’
Clauses with a GEN-DAT clitic sequence have been grammatically reanalyzed as constructions with a particular aspectual function, the continuative aspect (cf. the paradigm in (2) above). The reanalysis involved a development where equative nominal constructions such as ‘it (is) [its burning]’ were reinterpreted as constructions marking continuative aspect ‘it (is) burning’.16 3.4 Nominative and accusative The fourth way to mark S is by using a nominative and an accusative pronominal clitic simultaneously, as in: la pinu tana Miri Yehu]j naj-mài-yaj Lord Jesus 3SG.NOM-come-3SG.ACC LOC top earth ‘The Lord Jesus did come down to earth.’
(26) [I
ART.PERS
Both clitics have the same referent, namely the S of the clause, I Miri Yehu: when the S is pluralised, both clitics become plural.17 This double S marking marks epistemic modality: it functions to express the speaker’s (certain) belief, (26), the speaker’s expectation, (27a) (compare 27b)), or an obligation, (28a), (compare (28b)).18
16
Nominal constructions that developed into constructions with particular aspectual functions have been attested cross-linguistically. In Dutch, for instance, the progressive aspect is expressed by a construction expressing equation between a subject and a PP containing an NP: Hij is [aan [het rennen]NP]PP ‘He is [to [the run]]’ > ‘He is running’. For more discussion on this pattern of grammaticalisation in Kambera, see Klamer (2000: 60 ff.). 17 Since only core arguments are cross-referenced on the Kambera verb, and locations are expressed as adjuncts, the clitics do not refer to PPs like la pinu tana ‘on earth’, nor to the N within PPs. 18 A sense of obligation is also expressed by non-canonical case marking in Urdu (Butt and King 1991), when A is marked dative rather than ergative.
DIFFERENTIAL MARKING OF INTRANSITIVE SUBJECTS IN KAMBERA (27) a. E!
Na-mbata-ya-ka
291
nú!
EXCL 3SG.NOM-be.broken-3SG.ACC-PFV DIST
‘Hey! It is almost breaking/will surely break.’ b. E! Na-mbata-ka nú! EXCL 3SG.NOM-be.broken-PFV DIST ‘Hey! It’s broken!’ (28) a. Da-laku-ha pa-rama haromu 3PL.NOM-go-3PL.ACC CTR-work tomorrow ‘They must/have to go to work tomorrow.’ b. Da-laku pa-rama haromu 3PL.NOM-go CTR-work tomorrow ‘They go/will go to work tomorrow.’ Native speakers observed that the construction does not always express such epistemic moods; in some contexts it may be used as an alternative with no special semantics. For example, (29a,b) are alternative constructions with no semantic difference. (29c) shows that in case one of the two clitics is omitted, it must be the accusative, not the nominative. (29) a. Da-tama-ha la kurung ba ku-yaulu-ha 3PL.NOM-enter-3PL.ACC LOC room CNJ 1SG.NOM-chase-3PL.ACC ‘They entered the room when I chased them.’ b. Da-tama la kurung ba ku-yaulu-ha 3PL.NOM-enter LOC room CNJ 1SG.NOM-chase-3PL.ACC ‘They entered the room when I chased them.’ c. *Tama-ha la kurung ba ku-yaulu-ha enter-3PL.ACC LOC room CNJ 1SG.NOM-chase-3PL.ACC Intended reading: ‘They entered the room when I chased them.’ The question may be asked if this construction can be analyzed as a type of ‘reflexive’ S argument. I do not have good arguments for such an analysis. Kambera (transitive and intransitive) reflexives use a construction with the possessed nominal wiki ‘self/own’, as illustrated in (30), (31) and (32a). The NP with wiki is the O of a transitive construction, and as such may be indefinite or definite. If it is indefinite, it is not cross-referenced on the verb, as in (30), if it is definite, it is cross-referenced, as in (31).19 wiki-nggu (30) Ku-pa.ita.ng20 1SG.NOM-CAUS.see.APPL self-1SG.GEN ‘I showed/revealed myself (to someone).’ 19
The wiki NP cannot be omitted, lest the sentence looses its reflexive reading, as in: Ka ta-kinju-ha nyuta ha’atu-ha’atu CNJ 1PL.NOM-examine-3PL.ACC we RED-each.one ‘Let each one of us examine them.’ 20 Pa.ita.ng ‘CAUS.see.APPL’ > ‘show’ is a causative and applicative derivation based on the root ita ‘see’. For details on the derivation of causatives and applicatives, see Klamer (1998a: 177-190, 197-213).
292
MARIAN KLAMER
(31) Ka
ta-kinju-ha da wiki-nda 1PL.NOM-examine-3PL.ACC ART.PL self-1PL.GEN nyuta ha’atu-ha’atu we RED-each.one ‘Let’s examine ourselves, each one of us.’
CNJ
When the verb is intransitive, the NP containing wiki cannot be definite, and neither can it be cross-referenced as the O of the verb, compare (32a,b): (32) a. Imbu ndingir wiki-mu seek stand.up self-2SG.GEN ‘Try to be independent.’ (Lit. ‘Try yourself (to) stand up.’) b. *Imbu ndingir-ya na wiki-mu seek stand.up-3SG.ACC ART.SG self-2SG.GEN In other words, double marking of S does not occur in sentences with a reflexive reading. Neither can we take a sentence with a double S marking and make it (explicitly) reflexive by adding a wiki-NP, as shown in (33). (33) *Na-mài-ya na wiki-na la pinu tana 3SG.NOM-come-3SG.ACC ART.SG self-3SG.GEN LOC top earth In other words, the accusative enclitic is not used to refer reflexively to S, and it cannot be used to crossreference the canonical reflexive wiki-NP in intransitive clauses. The double-S construction has a restricted use, as it is mainly used in specific registers, poetic and/or religious texts and is considered archaic. It was used more widely at the beginning of this century; Wielenga (1909:47, 51-53) gives several examples that were considered grammatical at the time, but are judged as ungrammatical by present-day speakers; e.g.: (34) *Na-manandang-ya na uma-nggu 3SG.NOM-be.beautiful-3SG.ACC ART.SG house-1SG.GEN ‘My house is beautiful.’ Other examples of Wielenga are still considered grammatical today, though the use of only the nominative clitic is preferred, e.g.: (35) Hi
da-beli-ha la uma 3PL.NOM-return-3PL.ACC LOC house ‘And they went back home.’
CNJ
In the following section (3.5) I will discuss a number of grammatical contexts where S is obligatorily marked like O, with an accusative, and that in addition, Kambera allows for optional accusative S’s. This pattern may be considered as a kind of absolutive-ergative alignment that applies to certain circumscribed domains
DIFFERENTIAL MARKING OF INTRANSITIVE SUBJECTS IN KAMBERA
293
of the grammar. In addition, in Kambera morphology we find some traces of an earlier absolutive-ergative alignment system (cf. Klamer 1998a:76, 262-270). This suggests that there may have been a stage in the language’s development where it mixed an absolutive-ergative alignment system with a nominative-accusative one. The use of the nominative-accusative to mark S may reflect such a ‘mixed’ stage. Note that the construcion is only used marginally and has mostly archaic connotations. The nominative marking of S has now clearly become the canonical one; i.e. the mixed alignment system of which Wielenga documented some examples has largely been replaced by a nominative-accusative pattern. 21 However, there are still a number of grammatical domains where the absolutive-ergative pattern prevails, as will be explained in the next section. 3.5 Accusative In section 2 we saw that the S of non-verbal predicates is always accusative. Such predicates are inherently states rather than events. In the present section we will see that accusative marks S in a number of other contexts too: in imperatives, with ‘foregrounded’ predicates, with generic or impersonal referents, and with stative verbs modified for degree. In these contexts, it is obligatory to mark S accusative (3.5.1). In these cases, S is typically a non-volitional participant that is not in control of the situation. In addition, the accusative is an option for all intransitive verbs to express an S that is less in control than it would canonically be expected (3.5.2).22 3.5.1 Contexts with obligatory accusative In transitive imperatives, the accusative marks O, while the A (addressee) is left unexpressed: (36) Kinju-ha! examine-3PL.ACC ‘Examine them!’ S’s in imperative clauses are always marked as O, with an accusative, as in (37); the addressee of intransitive imperatives cannot be expressed otherwise: (37) Katuda-kau nàhu! sleep-2SG.ACC now ‘Go to sleep now!’ 21
22
One reviewer asks whether there are typical verbs that use this mixed pattern. Unfortunately, I do not know; all I can say is that my database contains less than a handful of spontaneous occurrences of double S marking, which suggests that it is a very marginal structure, and that they typically occur in religious or formulaic expressions. Kambera is one of the nine (Austronesian and Papuan) languages of eastern Indonesia analysed in Klamer (in press) to have a ‘semantic’ alignment system. In these languages, the marking of S in verbal clauses is primarily determined by the semantic characteristics of S.
294
MARIAN KLAMER
This pattern can be explained by the fact that an imperative subject is treated as someone who is not fully in control of the activity: the addressee of an imperative is told by someone else to do something. Secondly, we find accusative S’s with predicates that are ‘fore-grounded’ by e.g. repetition and/or left-dislocation of the verb, as in: (38) Tembang, nda tembang-a-ya-pa i be.stupid NEG be.stupid-MOD-3SG.ACC-IPFV ART.PERS ‘(As for) being stupid, Windi is no longer stupid.’
Windi Windi
Thirdly, when stative verbs are modified for (excessive) degree, as in (39), their S is also accusative, since in these contexts too the emphasis is on the state expressed by the verb, while the argument is portrayed as an entity that is part of it: (39) Dira mayila ailulu-kama extremely be.needy very-1PL.ACC ‘We are so very, very poor.’ And finally, an S with a generic or impersonal referent, expressed with a 3sg enclitic, as in (40), is also marked accusatively. In such contexts, too, the emphasis is on the verb, while the impersonal or generic S is a referentially underspecified participant: (40) Jàka
nda nyumu, meti-ya-ka làti NEG you die-3SG.ACC-PFV in.fact ‘Without you, one/we would die/have died.’ CNJ
In sum, the common denominator of all the grammatical contexts where S is obligatorily marked with an accusative, is that they all emphasise the situation of which S is a part. S is cast as an entity that is part of that situation, rather than an actively controlling or volitionally involved participant. 3.5.2 Contexts with optional accusative In addition to contexts where S is obligatorily accusative, there are also contexts where the choice for an accusative S is optional, and semantically determined. Consider the following sentences which only contrast on the marking of S: (41) a. Hí-ma-a-ya-ka i Umbu Mada una EMPH.3SG cry-EMPH-MOD-3SG.ACC-PFV ART.PERS Sir Mada ‘Sir Mada just cried and cried.’ (i.e. could do nothing else) b. Hí-ma-a-na-nya-ka i Umbu Mada cry-EMPH-MOD-3SG.GEN-3SG.DAT-PVF ART.PERS Sir Mada una EMPH.3SG ‘Sir Mada was crying.’ (but could have chosen not to)
DIFFERENTIAL MARKING OF INTRANSITIVE SUBJECTS IN KAMBERA
295
As indicated by the translations, the contrast between -ya and -nanya in these two sentences is that the accusative S is less in control than the S marked with a genitive and dative enclitic. In other words, S can (optionally) be presented as a less controlling participant in the state of affairs by marking it accusative. Though I have not tested this for all intransitives, the productivity of the pattern for those that I did test suggests that all Kambera intransitive verbs would allow for an optionally accusative S, and all of these accusative S’s are interpreted as ‘less controlling’ than they canonically are expected to be. Verbs attested with accusative S include activity verbs (pabànjar ‘chat’), directional verbs (mài ‘come (towards speaker)’), as well as verbs denoting events (meti ‘die’, hí ‘cry’), processes (kalit ‘to grow dark’), or states (hàmu ‘be good’, hangunja ‘sit idly, sit doing nothing’). There is, however, one morphological class of intransitive verbs that systematically does not take accusative S’s: the verbs derived with the prefix ta., illustrated in (42). Ta. is a productive prefix that derives intransitive verbs from transitive and intransitive bases. The derived forms express uncontrolled, unintentional, involuntary or unexpected achievements: (42) bunggah lunggur lukur nggàjir mbutuh
‘X open Y’ ‘X scrape Y’ ‘to be huddled’ ‘to shake’ ‘to slip off’
ta.bunggah ta.lunggur ta.lukur ta.nggàjir ta.mbutuh
‘Y is open (accidentally, etc.)’ ‘Y is sore (accidentally, etc.)’ ‘Y is huddled (involuntarily)’ ‘Y shakes (involuntarily)’ ‘Y slips off (accidentally, unexpectedly)’
The S of ta-verbs cannot be accusative: (43) *Na ài nuna tambuta-ya-ka dàngu ART.SG tree that.one drop out-3SG.A-PFV with
amung root
Because the ta.verbs are uncontrolled, unintentional, involuntary achievements, their S is a non-controlling participant by default, and since the nominative is used as the default case - accusative only being used to specifically indicate that the expected control of S is not present - , the S of ta. verbs is marked as nominative.23 This will be further discussed in section 4. In sum, in this section, we have seen that S is obligatory accusative in those a syntactic contexts where the emphasis is on the entire state of affairs, with S being cast as an entity that is part of it. With all intransitive verbs, S can optionally be
23
The argument marking of these verbs is different from underived intransitives in other respects too. For instance, S cannot be marked with a genitive, or with the genitive-dative combination (the continuative aspect construction). Often the S is not cross-referenced on the verb at all, as in (i): (i) Na ài nuna ta.mbuta-ka dàngu amung ART.SG tree that.one drop.out -PFV with root ‘That tree is uprooted.’
296
MARIAN KLAMER
marked accusative to present it as less actively controlling. The class of ta. verbs are an exception: their non-controlling S can never be accusative, only nominative. 4. SUMMARY AND DISCUSSION The data discussed above lead to a number of general conclusions on: (i) the variation in the morphological marking of S (ii) how the case marking of S relates to its default/unmarked semantic role (iii) the lexical representation of intransitive verbs In section 4.1 I summarise point (i), in section 4.2 I discuss point (ii) and (iii). In section 4.3, I explain how the pattern of S marking for underived intransitives is applied to two classes of derived intransitive verbs. 4.1 Variation in the morphological marking of S S can be marked in five different ways, each with its own function, briefly recapitulated here for the verb meti ‘to die, to be dead’: (44) Nominative: default unmarked expression of S Jàka nda nyumu, da-meti-ka làti CNJ NEG you 3SG.NOM-die-PFV in.fact ‘Without you, they would die/they would have died.’ (45) Genitive: Irrealis mood, non-agent orientation, dependent discourse function. Mbàda meti-na-ka? already die-3SG.GEN-PFV ‘Is he dead already/has he died already?’ (46) Genitive plus 3sg dative: continuative aspect. Ba na-habola tuna-ka nú, CNJ 3SG.NOM-give.birth thus-PVF DEI meti-ma-a-na-nya nyuna yena this.one die-EMPH-MOD-3SG.GEN-3SG.DAT she ‘While/when she thus gave birth, she died/was dying.’ (47) Nominative and accusative: epistemic modality; often special register/archaic. Jàka nda nyumu, da-meti-ha-ka làti NEG you 3PL.NOM-die-3PL.ACC-PFV in.fact if ‘Without you, they would die/have died for sure.’ (48) Obligatory accusative S: non-verbal predicates, imperatives, generic/impersonal referents, with stative verbs modified for degree, with ‘foregrounded’ predicates. Optional accusative S: less controlling. Jàka nda nyumu, meti-ya-ka làti CNJ NEG you die-3SG.ACC-PFV in.fact ‘Without you, we would die/have died.’ (lit. ..one would have died)
DIFFERENTIAL MARKING OF INTRANSITIVE SUBJECTS IN KAMBERA
297
We have seen that the choice for one of the various morphemes for S depends on information from various subcomponents of the grammar, including those representing notions of modality and aspect, as well as discourse. 4.2 The lexical representation of intransitive verbs Regarding the lexical representation of intransitive verbs, Kambera does not present evidence for a formal distinction between classes of intransitive verbs (e.g., ‘unaccusative’ versus ‘unergative’ verbs). Every intransitive verbs may in principle occur in all of the five configurations, including the one with an accusative S, so none of the constructions is connected to a particular class of verbs. Furthermore, in embedded syntactic structures (e.g. control, relativization) as well as in morphological derivations (not discussed here, but see Klamer 1998a,b) all intransitive verbs (both active and stative ones) behave alike. That is, there is no structural evidence to assume a particular class of verbs whose S patterns like O (‘unaccusatives’) and another class whose S patterns like A (‘unergatives’). Thus the lexical argument structure of Kambera intransitives does not distinguish between internal and external arguments, and neither does the semantic/thematic content of the single argument of intransitives (as e.g. PATIENT, THEME or AGENT) link directly to the morphological case marking of S.24 In other words, nominative and accusative, as well genitive and the continuative genitive+dative, may alternate with most intransitive verbs. Nominative being the unmarked case, it is used as the default marking of S, whatever its semantic role, including e.g. AGENT in active controlling intransitives such as run, dance, scream; THEME in statives such as be small, be red, or PATIENT in non-agentive events such as die, fall. When one wants to specifically indicate that the expected control of the S of active verbs is absent, the accusative is chosen to mark S instead. In addition, we saw in 4.1 that the choice for one of the other S morphemes depends on information from subcomponents of the grammar that represent notions of modality and aspect, as well as discourse. 4.3 The S of morphologically derived verbs The perspective that the default case marker for S is nominative also accounts for the morphologically derived intransitive verbs. The first group are the verbs derived with ta. that express involuntary, incidental, or accidental events. By default, the S of these verbs is a non-controlling, non-volitional entity, and the default marking of it is nominative. In fact, nominative is the only marking that is allowed: intransitives derived with ta. do not allow S to be accusative, because the variable interpretation of S as a less controlling, less volitional entity is not available.
24
These data are problematic for most analytical approaches that assume a close relation between abstract argument structure and morphological case, such as e.g. Bittner and Hale (1996). See de Hoop and Narasimhan (this volume, section 4), for similar observations.
298
MARIAN KLAMER
The second group of derived intransitives are the anticausative verbs. The anticausative prefix is a nasal that modifies the initial stop consonant of a transitive base verb, resulting in a derived verb with a prenasalized initial stop:25 (49) kodang ‘X move Y’ buta ‘X pluck/weed Y’ pàda ‘X extinguish Y’
nggodang ‘Y is loose/moving’ (e.g. tooth) mbuta ‘Y is plucked/weeded’ mbàda ‘Y is gone out/X is extinguished’
Like the ta. derivations, the anticausative derives non-active, non-controlled intransitive verbs from transitive base forms. Unlike the ta. derivations, however, the morphological marking of the S of anticausatives shows the same variation as the S of underived verbs. How can we explain this? One relevant factor is the productivity of the derivational process involved. Unlike the ta.derivation, the anticausative is no longer a productive morphological process – though there are many semantically transparent pairs of transitive ~ anticausative verbs. Related to their unproductivity is the fact that anticausatives are semantically less regular than the ta. verbs. For example, anticausatives may, or may not, imply agents: in mbuta ‘to be plucked/weeded’ an actor is implied because weeding can only be done by an actively involved participant. However, the verb mbàda simply indicates the achievement that a fire is no longer burning – this may be the result of having gone out ‘by itself’ or by an agent extinguishing it. Anticausatives thus have a more variable, irregular, interpretation than ta.verbs. In addition, while speakers always consider ta. verbs as morphologically complex, they analyse anticausatives as morphologically simple forms. Since they are analysed as underived intransitives, anticausative verbs mark their S following the same patterns that are allowed for underived intransitives, using the nominative as the default case, the accusative for a less controlled S, and the other markings in their their respective contexts. REFERENCES Andrews, A. (1985). The major functions of the Noun Phrase. Language Typology and Syntactic Description: Clause Structure. Ed. by T. Shopen. Cambridge: Cambridge University Press, 62-154. Bittner, M. and K. Hale (1996). The structural determination of case and agreement. Linguistic Inquiry 27, 1, 1-68. Blust, R. (1993). Central and Central-Eastern Malayo-Polynesian. Oceanic Linguistics 32, 2, 241-293. Burzio, L. (1981). Intransitive Verbs and Italian Auxiliaries. Cambridge (MA): MIT dissertation. Butt, M. and T.H. King (1991). Semantic case in Urdu. Papers from the 27th Regional Meeting of the Chicago Linguistic Society. Chicago: CLS, 31-45. Dixon, R.M.W. (1979). Ergativity. Language 55, 59-138. Dixon, R.M.W. (1994). Ergativity. Cambridge: Cambridge University Press. Donohue, M. (this volume). Different subjects, different marking. de Hoop, H. and B. Narasimhan (this volume). Ergative case marking in Hindi. Kapita, O.H. (1982). Kamus Sumba/Kambera-Indonesia, Waingapu: Gereja Kristen Sumba. Klamer, M. (1997). Spelling out clitics in Kambera. Linguistics 35, 895-927. Klamer, M. (1998a). A grammar of Kambera. Berlin/New York: Mouton de Gruyter. 25
For more information on this process, as well as example sentences, cf. Klamer (1998a: 262-265).
DIFFERENTIAL MARKING OF INTRANSITIVE SUBJECTS IN KAMBERA
299
Klamer, M. (1998b). Kambera intransitive argument linking. Studia Linguistica 52, 2, 77-111. Klamer, M. (2000). Continuative aspect and the dative clitic in Kambera. Formal issues in Austronesian Linguistics. Ed. by I. Paul, V. Phillips and L. Travis. Dordrecht: Kluwer, 49-63. Klamer, M. (2001). Phrasal emotion predicates in three languages of Eastern Indonesia. Yearbook of Morphology 2000. Ed. by G. Booij and J. van Marle. Amsterdam: Kluwer Academic Publishers, 97-121. Klamer, M. (2002a). Semantically motivated lexical patterns: a study of Dutch and Kambera expressives. Language 78, 2, 258-286. Klamer, M. (2002b). Report constructions in Kambera (Austronesian). Reported discourse. Ed. by T. Güldemann and M. von Rucador. Amsterdam: Benjamins, 323-340. Klamer, M. (2004). Multi-categorical items as underspecified lexical entries: The case of Kambera wàngu. Up and Down the Cline: The Nature of Grammaticalization. Ed. by O. Fisher, M. Norde and H. Perridon. Amsterdam: Benjamins, 299-323. Klamer, M. (2005). Kambera. The Austronesian Languages of Asia and Madagascar. Ed. by A. Adelaar and N.P. Himmelmann. London: Routledge, 709-734. Klamer, M. (2006). Kambera complement clauses and complementation strategies. Complementation - A Cross-Linguistic Typology [Explorations in Linguistic Typology 3]. Ed. by R.M.W. Dixon and A. Aikhenvald. Oxford: Oxford University Press. Klamer, M. (in press). The semantics of Semantic Alignment in eastern Indonesia. Ed. by Mark Donohue and Søren Wichmann. Typology of Active-Stative Languages. Oxford: Oxford University Press, 221-251. Onvlee, L. (1925). Eenige Sumbasche Vertellingen. Leiden: Brill. Onvlee, L. (1984). Kamberaas (Oost-Soembaas) - Nederlands Woordenboek. Dordrecht: Foris. Wielenga, D.K. (1909). Schets van een Soembaneesche spraakkunst (naar 't dialect van Kambera). Batavia: Landsdrukkerij.
INDEX iterative, 114, 123, 125, 135, 283 process, 131, 132, 145 state, 68, 73, 124, 131, 132, 138– 142, 197, 294 telic, 65, 125, 135, 140 Assamese, 24, 25 Avar-Andic, 176
Adposition postposition, 97, 156–158 preposition, 21, 72, 142, 207, 275 Agentivity, 2, 5, 9–11, 14, 15, 109, 116, 126–132, 134–138, 142, 146, 147, 176, 181, 183, 184, 186, 187, 189, 190, 194, 197, 198, 259, 277, 297 Agreement, 5, 11–13, 15, 30, 32, 72, 85, 86, 88–94, 96, 103, 109, 119, 121, 134, 142, 143, 154, 163, 190, 192, 199–279 object, 72, 205, 231, 234, 243 role-based, 210, 211, 214, 215, 219 salience-based, 210, 214, 216, 219 subject, 11, 15, 202, 208, 214, 215, 219, 271 topic, 11, 15, 201, 208, 212, 214, 215, 219 Agul, 5, 10, 11, 14, 173–198 Akhvakh, 192 Aleut, 163 Anaphor, 195, 205, 265 Animacy, 2, 6, 7, 9, 11, 13, 23, 27, 29, 33, 35–38, 45, 46, 48–56, 58, 65, 66, 73, 81–83, 153, 155, 162, 163, 165, 216–219, 227, 228, 268 Antecedent, 102, 134, 190, 192, 195, 197 Aranda, 37, 38 Argument external, 18–20, 22–25, 119, 136, 139, 140, 142, 145, 146, 297 internal, 20, 114, 117–119, 139– 141, 145 Aspect activity, 138–140, 162 atelic, 125 durative, 120, 123, 124 habitual, 18, 114, 118, 123, 125, 135
Bantu, 11, 15, 86, 199–221 Basque, 19, 22–24 Blocking, 30, 68 Case ellipsis, 7, 41–61, 82 Case (marking) ablative, 87, 88, 90, 92, 93 absolutive, 3, 4, 10, 12, 14, 27, 174, 176, 182, 188, 191, 193, 195, 210, 215, 216, 226, 231, 236, 239, 240, 242, 243, 248, 290, 293 applicative, 253–255, 264, 284, 291 comitative, 174, 270 dative, 6, 13, 18–25, 31–33, 64, 69, 74, 141, 174, 175, 182, 188, 191–196, 253–260, 273, 274, 283–290, 296, 297 genitive, 4, 5, 8, 9, 12–14, 21, 26, 27, 47, 79–150, 160, 164, 179, 239, 242, 249, 250, 256, 257, 260, 271, 272, 283–290, 296, 297 inherent, 19, 20, 25, 38 instrumental, 2, 27, 65, 67, 68, 191, 254, 255, 272 locative, 5, 8–11, 14, 65, 100, 113–149, 174–176, 185, 186, 193, 194, 197, 202–205, 214, 255, 260, 262, 264, 278 oblique, 9–11, 146, 150, 153, 156–159, 168, 198, 207, 249, 275–277 structural, 19, 23, 29, 74, 80, 85, 86, 91–94, 105–108, 110 301
302 tripartite system, 152, 153, 158, 168, 233 Causative, 10, 142, 175, 186–190, 197, 198, 255, 291, 298 Chali, 156, 158 Chemehuevi, 159–161, 164 Chiche a, 200, 202–204, 208, 209, 213, 214 Chishona, 202 Clitic, 12, 13, 18, 33, 143, 227, 231, 234, 242, 247–279, 282–292, 294, 298 Connective, 223–245 Converb, 101, 103, 104, 182, 194 Dargi, 192 Definiteness, 2, 6, 7, 9, 41–61, 65, 81, 82, 128, 155, 162, 163, 229, 252, 284, 286 Demonstrative, 156, 159, 195–198 Determiner, 96, 106, 224, 226, 229, 231, 234–238, 242, 273 Detransitivized, 2 Diachronic, 11, 12, 15, 122, 154, 214, 218, 219, 224, 242, 283, 289 Dialect, 27, 32, 69, 156, 226 Diathesis, 12, 13, 15, 249, 266, 271 Discourse, 11, 13, 14, 42–44, 47, 90, 102, 103, 106, 128, 143, 192, 200, 211–213, 228, 242, 249, 285–289, 296, 297 Distinguishability, 11, 15, 66, 153, 154, 214, 218, 219 Dyirbal, 6, 18, 36, 51, 52 Dzamba, 201, 207–209 Economy (constraint), 3, 4, 6, 53–57, 152, 168 Ellipsis, 7, 41–61, 82 English, 22, 24, 25, 30, 46, 83, 90, 91, 123, 141, 142, 160, 184, 185, 189, 191, 199–221, 248, 263, 273, 276, 277
INDEX
Ergativity, 135, 138, 140, 157, 215– 219, 224, 226, 240, 242, 243 split ergative, 51, 140, 215–219, 239, 240 Expletive, 90, 143, 208 Factive, 95–97 Faithfulness, 3–6, 15, 35–39, 164 Finite, 134, 204 Finnish, 118, 125, 178, 242 Focus, 44–47, 140, 201, 213 French, 33, 141 Frequency, 7, 41–61, 153, 154 Gender, 30, 212, 284 feminine, 30 masculine, 30, 32 neuter, 30, 32, 143 Georgian, 225 German, 135, 208 Grammatical function, 5, 6, 11–13, 46, 156, 162, 169, 202, 205, 206, 215, 218, 248, 249, 266, 267, 271, 274, 277 Grammaticalization, 12, 154, 162, 223–245 Greek, 225 Halkomelem, 224, 240 Harmonic alignment, 52 Hierarchy animacy, 11, 13, 27, 29, 35–38, 49, 52, 55, 216, 268 argument, 211, 213, 218 nominal, 215, 216 number, 35 person, 33, 35 thematic, 23, 263 topicality, 212 Hindi, 7–9, 14, 18, 22, 25–27, 31, 32, 63–78, 86, 108, 110, 115, 141, 156–158, 164, 168 Hokkien, 276
INDEX
Icelandic, 20–25, 102 Iconicity (constraint), 6, 53–57 Idiom, 42, 44, 70, 135, 185, 208–210 Incorporation, 2, 28, 43, 140, 142, 143, 268 Individuation, 165 Infinitive, 122, 185, 187 Inflection, 104, 141, 161, 242, 256, 267 Intonation, 3, 212 Inuit, 18, 27, 29, 248 Inuktitut, 18, 27 Inversion, 142, 202, 203, 205, 214 Italian, 33, 141 Japanese, 7, 22, 31, 42, 44, 47, 50, 59, 163 Jo’é, 239, 243 Kabardian, 32, 33 Kambera, 5, 13–15, 70, 281–299 Kinyarwanda, 201–211, 217 Kirundi, 201–209, 217 Korean, 5, 7, 9, 14, 41–61, 82 Labile, 178, 179, 194 Lak, 177, 192 Lango, 275 Lezgian, 176, 184, 185, 191, 192 Mam, 248, 268 Manggarai, 275–277 Marathi, 24, 31–33, 39, 67, 205 Markedness, 4, 6, 15, 18, 29–31, 35– 39, 45, 52, 54, 58, 75, 134, 163, 215, 218 Modality/mood, 13, 14, 198, 227, 249, 251, 253, 256, 287, 288, 290, 291, 296, 297 epistemic, 13, 186, 286, 290, 291, imperative, 13, 130, 131, 194, 293, 294, 296 indicative, 83, 95–97, 99, 101– 105, 109, 227–237, 243
303 irrealis, 249–251, 253, 256, 288, 296 modal verb, 185, 205 realis, 250, 251, 253, 262 subjunctive, 12, 83, 95–99, 105, 226–243 Negation, 9, 68, 114, 116–119, 126, 206, 282, 283, 287 Nez Perce, 18, 29 Nisga’a, 226, 229, 235–237 Niuean, 18, 27–29 Nominalization, 80, 95–97, 109, 136 Number, 30, 35, 72, 192, 212, 229, 231, 251, 253, 259, 288 plural, 30, 32, 35, 89, 96, 103, 122, 156, 237, 238, 251, 284, 290 singular, 30, 35, 37, 72, 89, 90, 103, 122, 143, 250, 251, 259, 284 Object shift, 26–29, 38 Obligation, 13, 69, 70, 290 Optimality Theory, 1–78, 151–171 bi-directional OT, 8, 67, 73–76 Stochastic OT, 42, 51, 54–58 Optimization, 3, 8, 73 Palu’e, 275 Participle, 120, 123, 193 Partitive, 9, 81, 86–93, 99, 106, 114– 119, 242 Person first, 7, 18, 30–37, 48, 51, 215, 216, 250, 251 second, 6, 7, 12, 13, 18, 30–37, 48, 51, 89, 98, 103, 216, 251 third, 6, 7, 12, 18, 30–37, 48, 51, 89, 90, 98, 131, 231, 236–238, 253 PF (phonological form), 6, 17–40, 105–108 Polish, 5, 8, 9, 14, 113–149
304 Possessive, 90, 96, 98, 99, 105, 136, 141, 142, 159, 160 Prominence, 1–16, 51, 54, 55, 59, 65, 66, 110, 138, 210–212, 219 Proto-role, (see thematic role) Prototypicality, 7, 52, 66, 67, 76, 128, 135, 162, 181, 215, 216, 218, 248 Psych-verb, 23 Quantifier, 91, 263–266 Question, 42, 44, 96, 208, 269–271 Raising, 99, 202–206, 274 Referentiality, 8, 52, 65, 162 non-referential, 208, 211 referential, 12, 103, 211, 214–219, 229 Reflexive, 133, 134, 190, 266, 291 Relativization, 190, 208 Reversal (S-O), 11, 201, 210 Roshani, 156, 158 Russian, 117, 118, 119, 184 Salish, 11, 178, 224, 240 Sesotho, 202, 214 Setswana, 202, 214 Sgüüxs, 232, 236 Sinhalese, 54 Sm’algyax, 226, 232, 233, 235, 236, 238 Spanish, 33 Specificity, 2, 8, 26, 72, 80, 83, 85, 86, 92, 104 non-specific, 8, 26, 28, 52, 66, 80, 82, 85, 86, 91, 105 specific, 7, 26, 52, 64, 72, 80, 86, 92, 98, 105, 155 Style (register), 42, 46, 57 Synchronic, 11, 211, 215, 218, 239 Tabassaran, 185, 189 Tagalog, 205, 210, 266, 271, 273, 274 Tati, 156
INDEX
Telicity (see aspect) Tense, 9, 85, 96, 103, 121, 140, 142, 155, 162, 167, 182, 227, 233, 239, 249, 283 future, 96, 121, 182, 249, 288 imperfective, 66, 70, 115, 120, 131, 156, 162 past, 104, 123, 156, 167, 182, 288 perfective, 8, 14, 64, 72, 73, 114, 120, 125, 137, 146, 156, 162, 194 present, 121, 167, 182, 288 Thematic roles, 5, 19, 22 agent, 4, 10, 15, 23, 66, 108, 127, 139, 147, 154, 164, 173, 176, 184, 196, 202, 206, 210, 217, 230, 233, 236, 250, 254, 273, 276, 297, benefactive, 180, 253, 283 experiencer, 19, 23, 25, 174, 184, 188, 191, 193, 195, 250, 257, 272 goal, 21 instrument, 191, 252, 254 location, 68, 141, 260, 261 patient, 12, 70, 147, 154, 164, 177, 180, 201, 217, 250, 255, 257, 259, 273, 284, 288, 297 proto-role, 66 recipient, 157, 252, 254, 260 theme, 21, 23, 68, 139, 202, 215, 255, 257, 259, 268, 297 Topic (see also agreement), 5, 11, 15, 43, 81, 124, 201, 203, 206, 208, 211, 219, 262, 271, 277 Topicality, 86, 110, 191, 211, 219 Topic salience, 211, 212 Transitivity, 6, 28, 64–68 72, 229, 232 ditransitive, 20, 21, 64, 157, 284 intransitive, 2–5, 8–10, 14, 28, 67, 75, 116, 140, 146, 152, 157, 174, 178, 187, 190, 195, 210, 226, 230, 236, 260, 285, 291, 296
305
INDEX
transitive, 2, 9, 12, 21, 28, 33, 44, 51, 64–76, 88, 117, 135, 140, 152, 157, 163, 169, 176, 179, 190, 192, 195, 210, 214, 225, 228, 230–240, 243, 283, 285, 291, 298 Tsezic, 176 Tsimshianic, 11, 224, 226, 228, 231, 234, 236, 238, 242, Tukang Besi, 12, 15, 219, 243, 249, 255, 258, 260, 263, 267, 272, 276 Turkish, 5, 7, 8, 14, 18, 26, 80 Udi, 175, 185, 192 Unaccusative, 9, 14, 116, 134, 136, 139, 142, 146, 285, 297 Unergative, 9, 14, 116, 134, 136, 139, 146, 202, 285, 297 Urdu , 9, 156, 158 Vafsi, 9, 14, 155–158, 165 Voice, 2, 10, 12, 51, 268, 273
active, 2, 5, 14, 65, 140, 203, 273, 275, 285, 288, 297 antipassive, 2, 10, 273 passive, 2, 10, 64, 67, 203, 208, 271, 275 Volitionality, 5, 18, 23, 65, 69, 73, 130, 137, 157, 162, 176 Warlpiri, 225, 231 Word order, 3, 12, 23, 28, 72, 80, 126, 142, 145, 154, 163, 190, 243, 267, 275 canonical, 3, 6, 9, 10, 14, 142, 152, 176, 190, 201, 207, 227 Yaqui, 159, 164 Yidiny, 216 Zaza, 167
Studies in Natural Language and Linguistic Theory Managing Editors Liliane Haegeman, University of Geneva Joan Maling, Brandeis University James McCloskey, University of California, Santa Cruz
Publications 1. L. Burzio: Italian Syntax. A Government-binding Approach. 1986 ISBN Hb 90-277-2014-2; Pb 90-277-2015-0 2. W. D. Davies: Choctaw Verb Agreement and Universal Grammar. 1986 ISBN Hb 90-277-2065-7; Pb 90-277-2142-4 3. K. É. Kiss: Configurationality in Hungarian. 1987 ISBN Hb 90-277-1907-1; Pb 90-277-2456-3 4. D. Pulleyblank: Tone in Lexical Phonology. 1986 ISBN Hb 90-277-2123-8; Pb 90-277-2124-6 5. L. Hellan and K. K. Christensen: Topics in Scandinavian Syntax. 1986 ISBN Hb 90-277-2166-1; Pb 90-277-2167-X 6. K. P. Mohanan: The Theory of Lexical Phonology. 1986 ISBN Hb 90-277-2226-9; Pb 90-277-2227-7 7. J. L. Aissen: Tzotzil Clause Structure. 1987 ISBN Hb 90-277-2365-6; Pb 90-277-2441-5 8. T. Gunji: Japanese Phrase Structure Grammar. A Unification-based Approach. 1987 ISBN 1-55608-020-4 9. W. U. Wurzel: Inflectional Morphology and Naturalness. 1989 ISBN Hb 1-55608-025-5; Pb 1-55608-026-3 10. C. Neidle: The Role of Case in Russian Syntax. 1988 ISBN 1-55608-042-5 11. C. Lefebvre and P. Muysken: Mixed Categories. Nominalizations in Quechua. 1988 ISBN Hb 1-55608-050-6; Pb 1-55608-051-4 12. K. Michelson: A Comparative Study of Lake-Iroquoian Accent. 1988 ISBN 1-55608-054-9 13. K. Zagona: Verb Phrase Syntax. A Parametric Study of English and Spanish. 1988 ISBN Hb 1-55608-064-6; Pb 1-55608-065-4 14. R. Hendrick: Anaphora in Celtic and Universal Grammar. 1988 ISBN 1-55608-066-2 15. O. Jaeggli and K. J. Safir (eds.): The Null Subject Parameter. 1989 ISBN Hb 1-55608-086-7; Pb 1-55608-087-5 16. H. Lasnik: Essays on Anaphora. 1989 ISBN Hb 1-55608-090-5; Pb 1-55608-091-3 17. S. Steele: Agreement and Anti-Agreement. A Syntax of Luiseño. 1990 ISBN 0-7923-0260-5
Studies in Natural Language and Linguistic Theory 18. E. Pearce: Parameters in Old French Syntax. Infinitival Complements. 1990 ISBN Hb 0-7923-0432-2; Pb 0-7923-0433-0 19. Y. A. Li: Order and Constituency in Mandarin Chinese. 1990 ISBN 0-7923-0500-0 20. H. Lasnik: Essays on Restrictiveness and Learnability. 1990 ISBN 0-7923-0628-7; Pb 0-7923-0629-5 21. M. J. Speas: Phrase Structure in Natural Language. 1990 ISBN 0-7923-0755-0; Pb 0-7923-0866-2 22. H. Haider and K. Netter (eds.): Representation and Derivation in the Theory of Grammar. 1991 ISBN 0-7923-1150-7 23. J. Simpson: Warlpiri Morpho-Syntax. A Lexicalist Approach. 1991 ISBN 0-7923-1292-9 24. C. Georgopoulos: Syntactic Variables. Resumptive Pronouns and A’ Binding in Palauan. 1991 ISBN 0-7923-1293-7 25. K. Leffel and D. Bouchard (eds.): Views on Phrase Structure. 1991 ISBN 0-7923-1295-3 26. C. Tellier: Licensing Theory and French Parasitic Gaps. 1991 ISBN 0-7923-1311-9; Pb 0-7923-1323-2 27. S.-Y. Kuroda: Japanese Syntax and Semantics. Collected Papers. 1992 ISBN 0-7923-1390-9; Pb 0-7923-1391-7 28. I. Roberts: Verbs and Diachronic Syntax. A Comparative History of English and French. 1992 ISBN 0-7923-1705-X 29. A. Fassi Fehri: Issues in the Structure of Arabic Clauses and Words. 1993 ISBN 0-7923-2082-4 30. M. Bittner: Case, Scope, and Binding. 1994 ISBN 0-7923-2649-0 31. H. Haider, S. Olsen and S. Vikner (eds.): Studies in Comparative Germanic Syntax. 1995 ISBN 0-7923-3280-6 32. N. Duffield: Particles and Projections in Irish Syntax. 1995 ISBN 0-7923-3550-3; Pb 0-7923-3674-7 33. J. Rooryck and L. Zaring (eds.): Phrase Structure and the Lexicon. 1996 ISBN 0-7923-3745-X 34. J. Bayer: Directionality and Logical Form. On the Scope of Focusing Particles and Wh-in-situ. 1996 ISBN 0-7923-3752-2 35. R. Freidin (ed.): Current Issues in Comparative Grammar. 1996 ISBN 0-7923-3778-6; Pb 0-7923-3779-4 36. C.-T. J. Huang and Y.-H. A. Li (eds.): New Horizons in Chinese Linguistics. 1996 ISBN 0-7923-3867-7; Pb 0-7923-3868-5 37. A. Watanabe: Case Absorption and WH-Agreement. 1996 ISBN 0-7923-4203-8 38. H. Thráinsson, S. D. Epstein and S. Peter (eds.): Studies in Comparative Germanic Syntax. Volume II. 1996 ISBN 0-7923-4215-1
Studies in Natural Language and Linguistic Theory 39. 40. 41. 42. 43. 44. 45. 46. 47. 48. 49. 50. 51. 52.
53. 54.
55. 56. 57.
58.
C. J. W. Zwart: Morphosyntax of Verb Movement. A Minimalist Approach to the Syntax of Dutch. 1997 ISBN 0-7923-4263-1; Pb 0-7923-4264-X T. Siloni: Noun Phrases and Nominalizations. The Syntax of DPs. 1997 ISBN 0-7923-4608-4 B. S. Vance: Syntactic Change in Medieval French. 1997 ISBN 0-7923-4669-6 G. Müller: Incomplete Category Fronting. A Derivational Approach to ISBN 0-7923-4837-0 Remnant Movement in German. 1998 A. Alexiadou, G. Horrocks and M. Stavrou (eds.): Studies in Greek Syntax. 1998 ISBN 0-7923-5290-4 R. Sybesma: The Mandarin VP. 1999 ISBN 0-7923-5462-1 K. Johnson and I. Roberts (eds.): Beyond Principles and Parameters. Essays in Memory of Osvaldo Jaeggli. 1999 ISBN 0-7923-5501-6 R. M. Bhatt: Verb Movement and the Syntax of Kashmiri. 1999 ISBN 0-7923-6033-8 A. Neeleman and F. Weerman: Flexible Syntax. A Theory of Case and Arguments. 1999 ISBN 0-7923-6058-3 C. Gerfen: Phonology and Phonetics in Coatzospan Mixtec. 1999 ISBN 0-7923-6034-6 I. Paul, V. Phillips and L. Travis (eds.): Formal Issues in Austronesian Linguistics. 2000 ISBN 0-7923-6068-0 M. Frascarelli: The Syntax-Phonology Interface in Focus and Topic Constructions in Italian. 2000 ISBN 0-7923-6240-3 I. Landau: Elements of Control. Structure and Meaning in Infinitival Constructions. 2000 ISBN 0-7923-6620-4 W. D. Davies and S. Dubinsky (eds.): Objects and other Subjects. Grammatical Functions, Functional Categories and Configurationality. 2001 ISBN 1-4020-0064-2; Pb 1-4020-0065-0 J. Ouhalla and U. Shlonsky (eds.): Themes in Arabic and Hebrew Syntax. 2002 ISBN 1-4020-0536-9; Pb 1-4020-0537-7 E. Haeberli: Features, Categories and the Syntax of A-Positions. Cross-Linguistic Variation in the Germanic Languages. 2002 ISBN 1-4020-0854-6; Pb 1-4020-0855-4 J. McDonough: The Navajo Sound System. 2003 ISBN 1-4020-1351-5; Pb 1-4020-1352-3 D. E. Holt (ed.): Optimality Theory and Language Change. 2003 ISBN 1-4020-1469-4; Pb 1-4020-1470-8 J. Camacho: The Structure of Coordination. Conjunction and Agreement Phenomena in Spanish and Other Languages. 2003 ISBN 1-4020-1510-0; Pb 1-4020-1511-9 I. Toivonen: Non-Projecting Words. A Case Study of Swedish Particles. 2003 ISBN 1-4020-1531-3; Pb 1-4020-1532-1
Studies in Natural Language and Linguistic Theory 59. 60.
61. 62. 63. 64. 65. 66. 67. 68. 69. 70. 71. 72.
D. Adger, C. de Cat and G. Tsoulas (eds). Peripheries. Syntactic Edges and their Effects. 2004 ISBN 1-4020-1908-4; Pb 1-4020-1909-2 C. Goria: Subject Clitics in the Northern Italian Dialects. A Comparative Study Based on the Minimalist Program and Optimality Theory. 2004 ISBN 1-4020-2736-2; Pb 1-4020-2737-0 V. Dayal and A. Mahajan (eds.): Clause Structure in South Asian Languages. 2004 ISBN 1-4020-2717-6; Pb 1-4020-2718-4 P. Kempchinsky and R. Slabakova (eds.): Aspectual Inquiries. 2005 ISBN 1-4020-3032-0; Pb 1-4020-3035-5 M. Arad: Roots and Patterns. Hebrew Morpho-syntax. 2005 ISBN 1-4020-3243-9; Pb 1-4020-3245-5 P. Štekauer and R. Lieber (eds.): Handbook of Word-Formation. 2005 ISBN 1-4020-3595-0; Pb 1-4020-3597-7 A. Johns, D. Massam and J. Ndayiragije (eds.): Ergativity. Emerging Issues. 2006 ISBN Hb 1-4020-4186-1; Pb 1-4020-4187-X F. Lee: Remnant Raising and VSO Clausal Architecture. A Case Study from San Lucas Quiavini Zapotec. 2006 ISBN 1-4020-4300-7 Olga M: Balkam Sprachbund Morpho-Syntactic Features. 2006 ISBN 978-1-4020-4487-8 Kiss, Katalin É. (eds.): Event Structure and the left Periphery. 2006 ISBN 978-1-4020-4753-4 Harbour, Daniel: Morphosemantic Number, from Kiowa Noun Classes to ISBN 978-1-4020-5037-4 UG Number Features. 2007 Pereltsvaig, Asya: Copular Sentences in Russian A Theory of Intra-Clausal Relations. 2007 ISBN-978-1-4020-5792-2 William D. Davies & Stanley Dubinsly (eds.): New Horizons in the AnalyISBN-978-1-4020-6175-2 sis of Control and Raising. 2007 Helen de Hoop and Peter de Swart (eds.): Differential Subject Marking. 2008 ISBN-978-1-4020-6498-2
springer.com