The Yearbook of South Asian Languages and Linguistics: 2001 9783110245264

Citation preview

EDITORIAL BOARD CHIEF EDITOR: Rajendra Singh, Dept. de Linguistique, University de Montreal C.R 6128 Succ. Centre-ville, Montreal H3C 3J7, Canada. Voice: (514) 343-2113; Fax: (514) 343-2284; e-mail: [email protected] ASSOCIATE EDITOR: Probal Dasgupta, Centre for Applied Linguistics, University of Hyderabad, Hyderabad 500046, India. Voice: (91-40) 258-500; Fax: (91^0) 258-120; e-mail: pdgalts@ uohyd.ernet.in REGIONAL EDITORS: R.K. Agnihotri, University of Delhi, India; Tej K. Bhatia, Syracuse University, NY, USA; W.S. Karunatillake, University of Kelaniya, Sri Lanka; Baber SA. Khan, UAE Air Force Academy, Al-Ain, United Arab Emirates; Rajend Mesthrie, University of Cape Town, South Africa; France Mugler, University of the South Pacific, Fiji. Tsuyoshi Nara, Seisen University, Tokyo, Japan; John Peterson, Universität München, Germany; Tariq Rahman, National Institute of Pakistan Studies, Islamabad, Pakistan; Udaya Narayana Singh, University of Hyderabad, India; E. TifTou, Universite* de Montreal, Canada; Yogendra P. Yadav, Royal Nepal Academy, Kathmandu, Nepal. EDITORIAL ADVISORS: E. Annamalai, Mysore, India; Bernard Comrie, Max Planck Institute, Leipzig, Germany; Wolfgang U. Dressler, Universität Wien, Vienna, Austria; Aravind Joshi, University of Pennsylvania, Philadelphia, USA; Ashok R. Kelkar, Pune, India; Paul Kiparsky, Stanford University, USA; E.F. Konrad Koerner, University of Ottawa, Canada; Bh. Krishnamurti, Hyderabad, India; Jayant K. Lele, Queen's University, Kingston, Canada; Marvin Minsky, M.I.T., Cambridge, MA, USA; Pieter Muysken, Universiteit van Amsterdam, The Netherlands; N.S. Prabhu, Bangalore, India. EDITORIAL ASSISTANT: Alice Anugraham and Mohammad Enjiah, Universite de Montreal. South Asia is home to a large number of languages and dialects. While the number of linguists working on South Asia has grown considerably in the recent past, there is as yet no recognized international forum for the exchange of ideas among them. The Yearbook of South Asian Languages and Linguistics is designed to be just that forum. It will consolidate empirical and theoretical research and provide a testing ground for the articulation of new ideas and approaches grounded in a study of South Asian languages but which have universal applicability.

4·

The Yearbook of South Asian Languages and Linguistics

2001 Tokyo Symposium on South Asian Languages Contact, Convergence and Typology Guest Editors

PERI BHASKARARAO KARUMURI VENKATA SUBBARAO Chief Editor

RAJENDRA SINGH A*«·* R«n: SERIALS SUBSCRIPTION SERVICE (INDIA) SUBSCRIPTION AQBNT· BOOK SELLERS

PUBLISHERS · STOCKISTS OF BACK-VOLUME! 4, B.S.S. HAU, KHYBER PASS MARKET, CIVIL LINE* DELHI-110054 (INDIA), PH.: 3811659 FAX: 91-11-3812678. E-mail: ••flaltOtatyam.nft.ln BRANCH OFFICE : 4630/24, ANSARI ROAD, DARYA GAN NEW DELHI-110002 (INDIA). PH.: 3245225

Sage Publications New Delhi · Thousand Oaks · London

/ UNIVERSITATS ( BIBLIOTHEK ; \ LEIPZIG /

Papers from the International Symposium on South Asian Languages: Contact, Convergence and Typology, organized by the Institute for the Study of Languages and Culture of Asia and Africa, Tokyo University of Foreign Studies.

Copyright© Rajendra Singh, 2001 All rights reserved. No part of this book may be reproduced or utilized in any form or by any means, electronic or mechanical, including photocopying, recording or by any information storage or retrieval system without permission in writing from the publisher. First published in 2001 by Sage Publications India Pvt Ltd M-32 Market, Greater Kailash -I New Delhi 110048 Sage Publications Inc. 2455 Teller Road Thousand Oaks, California 91320

/^ rCj ^J

Sage Publications Ltd 6 Bonhill Street London EC2A 4PU

Published by Tejeshwar Singh for Sage Publications India Pvt Ltd, typeset by Line Arts, Pondicherry and printed at Chaman Enterprises, Delhi. ISSN: 0971-9539 ISBN: 0-7619-9560-9 (US-Hb) 81-7829-046^1 (India-Hb) Sage Production Team: Arpita Das, M.S.V. Namboodiri and Santosh Rawat

Contents Commentum Editoris Guest Editors9Preface

7 9

Two Grammars in One: Sentential Complements and Complementizers in Bengali and other South Asian Languages JOSEF BAYER

11

Language Mixing, Typology, and Second Language Acquisition TEJ Ê. ÂÇÁÔÐÁ and WILLIAM C. RITCHIE

37

Typology vs. Convergence: The Issue of Dravidian/Indo-Aryan Syntactic Similarities Revisited HANS HENRICH HOCK

63

Where do Compound Verbs Come From? (And Where are They Going?) PETER EDWIN HOOK

101

Typology of Verbal Categories of the Languages of Nepal TEJ R. KANSAKAR

131

Ergativity Attrition in the History of Western New Indo-Aryan Languages (Panjabi, Gujarati and Rajasthani) L.V KHOKHLOVA

159

Convergence Patterns in Tuluva: A New Scope for Comparative Studies NOZOMI KODAMA

185

The Definition and Significance of Linguistic Areas: Methods, Pitfalls, and Possibilities (with Special Reference to the Validity of South Asia as a Linguistic Area) COLIN P. MASICA

205

Personal Pronouns and Related Phenomena in the South Asian Linguistic Area: Convergent Features or Convergence-resisting Features TOSHIKI OSADA

269

The Role of Language of Religion in the Convergence of South Asian Languages RAJESHWARI V. PANDHARIPANDE

289

Contents Dative Subject Constructions in South Asian Languages MASAYOSHI SHIBATANI and PRASHANT PARDESHI

311

Morphological Diversity and Morphological Borrowing in South Asia RAJENDRA SINGH

349

Convergence and Resistance to Morphological Change in Agglutinative Languages of South and Central Asia ANDR£E E SJOBERG

369

Creolization and Convergence in Morphosyntax: Sri Lanka Portuguese and Sourashtra Nominal Marking Typology x IAN R. SMITH

391

Ergativity in Gujarati, Hindi, and Pashto: The Evidence from Causative Constructions STANLEY STAROSTA

411

Agreement in South Asian Languages and Minimalist Inquiries: The Framework KARUMURI VENKATA SUBBARAO

457

The Interplay of Typological and Genetic Factors in Determining the South Asian Language Area BORIS A. ZAKHARYIN

493

The Proto-Munda Verb System and some Connections with Mon-Khmer NORMAN H. ZIDE and GREGORY D.S. ANDERSON

517

About the Guest Editors and Chief Editor Notes on Contributors

541 543

Commentum Editoris Just when we were thinking that it would perhaps be a good idea to bring out occasional thematic issues of the Yearbook, Professor Peri Bhaskararao of the Tokyo University of Foreign Studies announced his important conference on contact and convergence in South Asia. He readily agreed with our suggestion that the proceedings of his conference should appear as a special issue of the Yearbook. He has co-edited this special issue with Professor Karumuri Venkata Subbarao of the University of Delhi. We are delighted to put that issue in your hands. It would not have seen the light of day, however, without the help of Arpita Das of Sage, whose professionalism and diplomatic skills made its timely production possible. We are grateful to her. We regret that Prof. K.P. Mohanan is leaving the Editorial Board. He had agreed to serve for 'a couple of years' and his current administrative responsibilities make it impossible for him to continue. We are grateful to the Shastri Indo-Canadian Institute, Calgary for the support it provided to our editorial assistants. The Editors

Guest Editors' Preface The international symposium on South Asian Languages: Contact, Convergence and Typology was organized by the Institute for the Study of Languages and Cultures of Asia and Africa, Tokyo University of Foreign Studies, Tokyo, Japan in December, 1999. It was sponsored by the Japanese Ministry of Education, Science, Sports, and Culture under its CoE program. This special issue of the Yearbook contains most of the papers presented at the symposium. These papers represent the current research in the areas of contact, convergence and typology in South Asian languages. The last two decades have witnessed a great interest in presenting evidence based on data from a wide variety of languages to substantiate specific theoretical issues. To build an explanatorily adequate theory, it is essential to have a sound data base. A fruitful combination of theory and data alone will yield the desired results. It has been argued elsewhere (Subbarao 1998; Subbarao & Saxena 1987) that there is a symbiotic relationship between linguistic theory and language typology which primarily emphasizes on choosing data from a wide variety of languages. While it is the linguistic theory that enables linguists to formulate a hypothesis in a given framework and helps them to look for the relevant data and provides them a direction in which to look, it is the data that not only substantiates the claims made in the theory but also provides a window to look for inadequacies, counter-examples to the claims made, thus helping the linguist revise the theoretical base, if necessary. An active interaction between data and theory is thus a desideratum. Any theory that is proposed without a sound data base and any theoretical claims made regarding the nature of human language without a strong theoretical base will not be explanatorily adequate. Lust et al. (2000:1) rightly put it: 'Without basic theory, the fundamental questions of the nature of human competence for language cannot be addressed. However, in the absence of real language data, proposed answers may not be relevant to the real questions.' The wide range and variety of topics covered at the symposium in the areas of typology and convergence are indicative of the concern of scholars to combine theory and empirical language data. These papers demonstrate the rich

10 · Guest Editors' Preface

contributions that South Asian languages can make to linguistic theory. Each individual author presenting her/his paper at the symposium was free to choose the framework she/he wished to present her/his findings in and there was no insistence on any specific theoretical framework for the papers to be presented. It is significant that in spite of choosing different theoretical frameworks, two papers in this volume (Shibatani & Pardeshi, and Subbarao) arrived at the same conclusion that the predicate in the dative/genitive subject construction should be intransitive and not transitive contrary to the existing view (Chomsky [forthcoming]; Ura 2000). The specific topics covered in the volume include papers in syntactic, morphological and phonological typology, issues related to typology and convergence, the origin, emergence or loss of a specific feature or construction in course of time, the notion and significance of 'linguistic area' and the interplay of typological and genetic factors in determining the South Asian language area, issues related to convergence and resistance to convergence, morphological diversity and borrowing in South Asia, creolization in morpho-syntax and code mixing in contact situations. We hope this volume fulfils the needs of scholars working in the areas of linguistic theory and typology. We are grateful to the Japanese Ministry of Education, Science, Sports, and Culture for sponsoring the symposium and to the Institute for the Study of Languages and Cultures of Asia and Africa for organizing it. • REFERENCES Chomsky, N. Forthcoming. Derivation by phases. Ken Hale: a life in language, ed. by M. Kenstowicz. Cambridge, MA: The MIT Press. Lust, Barbara C, Kashi Wali, James W. Gair and K.V. Subbarao. (eds.) 2000. Lexical anaphors and pronouns in selected South Asian languages: a principled typology. Berlin: Mouton de Gniyter. Subbarao, K.V. 1998. Linguistic theory and syntactic typology: a proposal for a symbiotic relationship. Vaagbhaarati: proceedings of the International Conference on South Asian Languages, ed. by L. Khokhlova and A. Sawant. Moscow: Moscow State University. Reprinted in Indian Linguistics 1999,60. 95-110. Subbarao, K.V. and A. Saxena. 1987. Language universals: inductive or deductive? Select papers from SALA 7, ed. by E. Bashir, M. Deshpande and P.E. Hook, 337-43. Indiana: Indiana University Linguistics Club, Indiana University. Ura, H. 2000. Checking theory and grammatical functions in universal grammar. Oxford: Oxford University Press.

Two Grammars in One: Sentential Complements and Complementizers in Bengali and other South Asian Languages ι JOSEF BAYER ι Indo-Aryan languages with Dravidian contact often show a dual system of sentential complementation with clause-initial complementizers for clauses in post-verbal position and clause-final complementizers (the so-called 'quotatives') for clauses mainly in pre-verbal position. The present article addresses parallels and differences between these two kinds of complements and makes suggestions as to an explanation of their heterogeneous distribution and properties with respect to mobility and operator scope. Of particular interest for current syntactic theories is the question how these complements are attached. It will be argued that CPs with initial complementizers are indirectly licensed via coindexing with a pronominal element that may also be nullified. CPs with final complementizers seem to be licensed directly. The article also contains data on negative sentences and negative polarity licensing which are of immediate relevance for theories of phrase structure which subscribe to the idea that linear order automatically mirrors hierarchical order such as c-command. The article closes with speculations about the status of languages which seem to follow conflicting word order parameters. Suggestions are made of how conflicts of this kind can be kept under control by the learner.

• 1. Introduction The bigger modern South Asian languages generally fall into the Indo-Aryan and the Dravidian group. The former show the Indo-European (IE) model of sentential complementation and relativization, i.e., complement or relative clauses have an articulated left periphery in which we see either a functional head such as a complementizer (C) or an operator like a relative pronoun or

12 · Josef Bayer

relative phrase. Sentential complements are uniformly positioned to the right of the heads that select them. The Dravidian model typically has clause-final affixal operators which bind variables to their left unselectively; the function of complementizers is performed by clause-final elements which are usually grammaticalized verbs of saying. In the unmarked case, sentential complements are positioned to the left of the heads that select them. In various languages on the Indian subcontinent, the two systems coexist in one and the same grammar. The languages in question, Marathi, Southern Hindi-Urdu (Dakkhini Hindi-Urdu), Oriya, Bengali, and Assamese, are geographically located in the South and in the East and North-East of India. It is to date highly controversial how the syntax of languages with mixed systems of complementation should be described. Various attempts have been made to reduce head-initial complements to head-final ones and vice versa. As can be shown, these attempts run into theory-internal conflicts of various sorts; in addition, they are incompatible with the development of IE. The complementizers of these languages come from radically different sources. In the IE type they are originally operators which seem to have been re-analyzed as heads like in various Western IE languages. They never appear in headfinal position. In fact, head-initial and head-final clausal complements are functionally at best overlapping but never identical. In addition, there is no tangible syntactic motivation for IPs to move either to the right or to the left of C. The syntax of operator scope also strongly indicates that the respective orders [C IP]] and [IP C] are basic orders that cannot be reduced to a single type. The mixed syntax of complementation in the languages at hand seems to be the result of intensive language contact. It cannot be an accident that the mixture is found where the Indo-Aryan model is weak and heavily influenced by the Dravidian model. Geographically, the relevant languages form a belt along the frontiers of Dravidian territory, with an extension into the NorthEastern regions. It will be argued that the syntax of these languages cannot meet with an explanatory account as long as the coexistence of the two typological models of complementation is not properly acknowledged. As the comparison with Germanic SOV shows, the IE model of complemetation can be integrated into the typological model of strict head-finality. The grammatical constraints that result from the import of head-initial structure into a head-final system are surprisingly stable, suggesting that mixed systems of the sort described nevertheless adhere to a single parametrical choice. This has important consequences for a theory of parameter setting. It will be argued that the child can deal with 'two grammars in one' as long as there is a containing grammar, and the contained subsystem is unequivocally signaled by a distinct functional vocabulary. Most of the material to be presented here comes from Bengali (or Bangla).1 As far as I know, however, the main generalizations seem to hold also for the other Indo-Aiyan languages with intensive Dravidian contact. This must

Two Grammars in One · 13

clearly be true for Oriya as one can conclude from Bal (1990) and for Marathi as one can conclude from Wali (1988) and Pandharipande (1997). The article is organized as follows: Section 2 will first introduce the relevant lexical elements, namely clause-final and clause-initial complementizers. We will then show that the two are in strictly complementary distribution, and that they differ in their respective feature structure as well as in their selection properties. Section 3 forms the core of the article. It will be shown there that the two complementizers give rise to very heterogeneous types of complements as signaled not only by their position but also by their mobility. In the second part of Section 3 the two types of complements will be investigated with respect to their interaction with the scope of wh-in-situ elements. Section 4 arrives at the conclusion that there may indeed be two grammars in one. The discussion will focus on the question how the learner of such a 'mixed' language can nevertheless set the relevant parameters.

• 2. Two Types of Complementizers Final complementizers (FC) differ from initial complementizers (1C) on all counts. In the South Asian languages, PCs are mainly degenerate verba dicendi\ they are traditionally called quotatives, because they seem to set the preceding discourse in quotes.2 ICs, on the other hand, are in IE very often degenerate operators. Consider Romance ehe, que, etc. Slavic shto, co, etc., Greek (o)tij all being lexically identical with the wh-pronoun for 'what'. The origin of ICs in the South Asian languages seems to be less clear, but it is suggestive that they too correspond lexically to operators. In the relevant languages these are ki ('what') and je (relativizer 'which').3 Table 1 gives an overview of the distribution of PCs and 1C in selected South Asian languages. Assuming that all the morphemes in question have head status in the modern languages, the Dravidian languages consistently show head-finality, Table 1: Complementizers in Selected South Asian Languages Language Telugu Tamil Kannada Malayalam Bengali Oriya Assamese Marathi Dakkhini H.-U.

Final complementizer ani (QUOT) enru (QUOT) anta (QUOT) ennu (QUOT) bole (QUOT) boli (QUOT) buli (QUOT) mhaNUn (QUOT), asa ('thus', QUOT), te (pronominal) bolke (QUOT), ki (OP)

QUOT is short for 'quotative', OP is short for Operator'.

Initial complementizer _ -

je (OP) je (OP)

je (OP) ki (?OP)

-

14 · Josef Bayer

whereas the Indo-Aryan contact languages show a mix of head-finality and head-initiality. Dakkhini Hindi-Urdu patterns with the Dravidian languages. According to Subbarao and Arora (1989), the grammar of Dakkhini HindiUrdu has undergone a process of 'extreme convergence* with its Dravidian contact language Telugu. Dakkhini does have the morpheme ki, but does not use it as an initial complementizer as it is found in standard Hindi and in various Northern Indo-Aryan languages. In Dakkhini, ki is rather used in a consistently parallel fashion with the suffixal disjunctive marker -oo of Telugu which appears in the marking of interrogative scope among various other functions. The quotative bolke is fully parallel with the Telugu FC ani. On the basis of the data that are provided in Subbarao and Arora (1989) one may well conclude that Dakkhini Hindi-Urdu is in fact a Dravidian language with an Indo-Aryan vocabulary.4 According to Hock (1991: 502ff), an equally strong case of convergence can be found in the Kupwar district between Maharashtra and Karnataka. In this region, the Indo-Aryan languages Urdu and Marathi have contact with the Dravidian languages, Kannada and Telugu at least for several hundred years. Although the speakers of these languages keep linguistic boundaries by sticking to the vocabulary of each language, the grammar of these languages has become 'virtually identical'. Such extreme convergence between Dravidian and Indo-Aryan is obviously not found in Bengali, Oriya, Assamese and Marathi. Nevertheless, these languages seem to employ a dual system in which heterogeneous chunks of grammar coexist. From the viewpoint of UG, this is not obviously an attractive perspective because the question arises how the child can set the parameter for his or her language. According to recent developments of syntactic theory, it may be the case that there are no word order parameters, and that the word order of languages can only vary as a result of movement which is conditioned by the satisfaction of different morphological requirements. But even then the question would arise how the child can find out which categories to move and which categories to leave in place. The old questions stay. We will now examine evidence which can plausibly serve as a lexical trigger to distinguish FC- from IC-clauses. • 2.1. Complementary Distribution Consider the following Bengali sentences which seem to be synonymous: (1) a. chele-Ta [or baba aS -be bole] Suneche boy-CF his father come-will QUOT heard 'The boy heard that his father will come.' b. chele-Ta Suneche [je or baba aS -be] boy -CF heard C his father come-will 'The boy heard that his father will come.'

Two Grammars in One · 15 There are no varieties of Bengali in which bole and je could be swapped. The following examples are ungrammatical in all known varieties of the language. (2) a. *chele-Ta boy -CF b. *chele-Ta boy -CF

Suneche [bole heard QUOT [or baba aS his father come

or baba aS -be] his father come -will -be je] Suneche -will C heard

The Bengali child has either the input [[... bole] V] or the input [V \je ...]] but never [V [bole ...]] or [[... je] V]. Thus, the lexical choice of the complementizer goes hand in hand with word order. This is an important clue toward establishing the different structures that are projected form the lexical items in question. According to Pandharipande (1997), the same is true in Marathi for the distribution of the 1C ki and the PCs mhaNUn ('having said') and asa ('thus'). • 2.2. Syntactic and Semantic Selection Although bole- and ye-clauses may occasionally be synonymous, they differ with respect to their property of being semantically selected. Broadly speaking, the 1C je seems to be more tolerant than the FC bole. For example, jeclauses can be selected by factive verbs, attitude verbs, verbs of perception (in their epistemic reading) as well as by nominals. ίofe-clauses, on the other hand, are restricted in various ways. Consider the contrast between (3) and (4). (3) [ram kolkata-y jacche bole] {janlaml bhablam/ Sunlam/ Ram Calcutta-LOC goes QUOT knew-I thought-I heard-I *dekhlaml *OSombhob} saw-I unlikely That Ram is going to Calcutta {I knew/I thought/I heard/*! saw/*is unlikely}.' (4) {janlaml bhablam I Sunlam/ dekhlaml OSombhob} [je knew-I thought-I heard-I saw-I unlikely C ram kolkata-y jacche] Ram Calcutta-LOC goes '{I knew/I thought/I heard/I saw/it is unlikely} that Ram is going to Calcutta.' We see that there is only an overlap in semantic selection. Predicates like idEkha and OSombhob are only acceptable withye but not with bole.5 That the (conflict with these predicates really stems from the presence of bole is made (dear by the fact that Bengali also allows complements without a comiplementizer (or perhaps with a 'zero complementizer'). In that case, the lexiical restriction disappears. Both ram kolkata-y jacche dekhlam and ram

16 · Josef Bayer

kolkata-y jacche OSombhob are acceptable. The reason for this seems to be that the latter versions can be understood with a dislocated clause, and with a silent pronoun in the matrix clause, whereas the versions with bole cannot. An overt pronoun is found in [ram kolkata-y jacche] ami ta dekhlam and in [ram kolkata-y jacche] ta OSombhob. Such a pronoun is impossible in *[[raw kolkata-y jacche] bole] ami ta dekhlam or in *[[ram kolkata-y jacche] bole] ta OSombhob. A natural explanation for this is that, due to its origin, bole has a verbal feature which is incompatible with the nominal feature inherent in ta. In those cases where no overt complementizer is found, or where the complementizer is underspecified, no feature conflict emerges. If one could show thatdEkha and OSombhob require a complement with the feature [+N], the deviant cases in (3) could be characterized as syntactic selection violations; it is however unclear why such a restriction should hold and how it could be derived from a semantically plausible basis. All I could show here is that the two types of complements are subject to overlapping but nevertheless different constraints on selection. • 2.3. Categorial Selection by C Bole in the literal sense of 'having said' is always possible with direct speech and non-propositional verbal utterances. In contrast to this Je cannot select anything but a finite clause. (5) a. colo colo bole Se taRataRi hoTte Suru korlo go go having-said (s)he quickly walk begin made 'Saying "let's go", (s)he quickly began to walk.' b. Se bollo: 'colo colo' (s)he said go go '(S)he said: "Go, go!"' c. *Se bollo je colo colo (s)he said C go go (5a) shows bole as a regular verb of speaking. Similar effects are achieved in (5b) by using bola as a main verb and adding the quote directly. As (5c) shows, direct speech is incompatible with the complementizer/e.6 While in (5a) bole can be seen as the participial form of the 'verbum dicendi' bola, such semantic transparency may not always be found. We have already seen clear signs of an expansion of its function according to which bole must have been subject to extensive grammaticalization. As Singh (1980) has pointed out, there are various uses of bole, e.g., as a postposition (amake didi bole Dakbe, (you) me eldersister BOLE call-will; 'You will call me didi9) or as a sentence final adverb with the meaning 'almost'. Given that in other South Asian languages with quotative FC-complements the quotative always requires a finite IP, it is perhaps justified to say that bole in its function as an FC is likewise restricted, and

Two Grammars in One · 17 that these other functions reflect different directions of grammaticalization. If true, we avoid the conclusion that bole is an FC which can select non-finite or even non-propositional phrases.7 As Singh also points out, bole-clauses can be used as reason adverbials.8 Fory'e-clauses this is never possible. Another important difference between bole and je is that bole must be underspecified for the feature [+wh], whereasy'e is always negatively specified as [-wh]. To see this, consider the data in (6): (6) a. ram ke aS -be (bole) jiggeS koreche Ram who come-will QUOT question made 'Ram asked who will come.' b. ram jiggeS koreche (*je) ke aS -be Ram question made C who come-will 'Ram asked who will come.' (6a) shows that the indirect question ke aSbe is compatible with bole, while it is incompatible with ye, as shown by (6b). (6b) is only grammatical without ye. In its negative specification for [wh] je differs from Hindi and Marathi Id. In Hindi we find interrogative as well as non-interrogative complements invariably with Id, as seen in Srivastav's (1991b) Hindi example tumjaante ho [ki [usne kyaa kiyaa]] ('You know what he did') or in Pandharipande's (1997) Marathi example mohan ne wicArle kl sudha kuThe gell {asa/mhaNUn} ('Mohan asked C where Sudha went QUOT') where we see in addition that interestingly the 1C and either one of the two PCs may occur simultaneously.9 The idea here is that the fa-clause will be assigned the wh-feature at LF by kyaa moving to scope position (which one may take to be the specifier of the fo'-clause). Under general considerations of economy it is expected that je and bole serve different functions, and that both of them retain certain distinct features which cannot be seen in languages in which C is uniformly either 1C or FC. • 2.4. Conclusion The considerations so far bring us to the conclusion that despite their functional overlap, bole and je are not only superficially distinct complementizers. They come from radically different genetic sources, and they show a number of lexical features, both as selectors and as selectees, which would act as a force against collapsing the two into one category with two accidentally different 'spell-outs'. According to Bal (1990) and Pandharipande (1997), Oriya and Marathi allow complements in which a finite complement clause is flanked by both an 1C and an FC simultaneously.10 Thus, for the child acquiring the language it seems to be unlikely that a parametrical conflict could arise. Being an FC, bole clearly falls into the standard typological set-up of a rigidly head-final language. However, the 1C je can easily be kept aside. It is

18 · Josef Bayer

not only phonologically a different lexeme, but is also different according to a host of lexical and syntactic properties, some of which may have been diachronically passed along as the coverage of the relative operator/e expanded. We will in the following section explore which kinds of deeper syntactic differences are induced by the two kinds of complementizers.

• 3. Two Types of Complements We have seen that PCs and ICs— here Bengali bole and je—give rise to rather heterogeneous constructions, and that this generalization holds beyond their lexical distinctness and also beyond their superficial appearance in different positions in the clause. We will now turn to some of their syntactic properties. In particular, we want to investigate the distribution of FC- and IC-clauses in complex sentences as well as their behavior with respect to the scope of wh-insitu element. • 3.1. The Position of FC- and IC-clauses In the languages of the world, FC-clauses appear almost invariably in strictly head-final languages, and they are regularly positioned in the space to the left of the matrix verb. The order is usually as in (7a) or in (7b), but (7c) is also possible, although it is often considered to be a marked option: (7) a. [...[CP....C*]V°] b. c. Singh (1980: 190) says that fcofe-clauses 'prefer to occur towards the western site (i.e., S-initially) of the whole sentence'. According to my own investigations, speakers of Bengali often judged sentences of type (7c) as deviant, unless bole can be interpreted as a reason marker, in which case the FC-clause is an adjunct and not an argument.11 Assume that FC-clauses in extraposed position are not ungrammatical, but just marked. The structure in (7a) is perfectly grammatical, but seems to be often avoided in practice; in all likelihood this is, however, due to the well-known performance problem with center embedding. Notice that (7c) is equally able to circumvent the performance problem. The fact that, according to Singh, the unmarked resolution (of a potential parsing problem) leads to (7b) rather than to (7c) suggests that bole· clauses arise to the left of V. In the languages of the world, IC-clauses regularly appear in head-initial languages, and they are regularly ordered in the right periphery of the verb of the matrix clause. In German, an attested, although not very strict, OVlanguage, finite CPs are banned from the position of the direct object. In those

Two Grammars in One · 19

iinstances in which such CPs are seen in the region to the left of the verb, either scrambling or topicalization is involved. Stowell (1981) has attributed this to a principle of Case Resistance: Phrases which cannot receive Case, are evacuated from positions in which Case is assigned. It does not matter in the present (context, to what extent Stowell's generalization should still be upheld, or how iit could be derived from more general principles. The important observation iis that Bengali ye-clauses can only appear to the right of V (or to the right of some other selecting head). They are invariably banned from the left side of tthe verb, be it an argument position, a scrambling position or a position targeted by a topicalization process. The distribution in (7) is reversed, the difference now being that any deviation from V-CP order is clearly ungrammatical, and not just marked.12 (8) a. [...V°[CPC°...]] b. * . . . C P C ° . . . 1 V ° t 1 c. The same ban against leftward movement can be found for Hindi ^/-clauses. To my knowledge such immobility of right-hand complements is not unique to South Asian ye-clauses and fa'-clauses. The same restriction can be found in Turkish fa'-clauses and in Persian ^-clauses.13 For a strictly head-final language this distribution is remarkable. Why should a complement obligatorily appear in a position which is normally not accessible (or at least not accessible for material which is not explicitly de-focused)? And why should such languages have IC-complements in the first place? There is a simple answer in recent syntactic theorizing which, however, does not strike me as particularly convincing, and which leads to a number of conflicts. This answer is that all languages are underlyingly head-initial, and that Bengali ;e-clauses simple show the order that UG provides as the Natural order'. Under this perspective, it is unexpected to observe thatye-clauses cannot even be topicalized, although Bengali can normally topicalize almost anything. According to this theory, feo/e-clauses must be shown to start out as head-initial CPs which are ordered to the right of the matrix verb. Two processes must apply to them: first the IP selected by the 1C bole must move to the left of bole, superficially turning bole into an FC. Then the entire CP has to undergo movement to the left of V This is the derivation that is forced by the assumptions of antisymmetric syntax with a universal SVO-base, as suggested by Kayne (1994) and subsequent work. The steps just described are s.hown in (9): (9) a. . . . V° [CP [C' [C bole] [IP ...]]] b. ... V° [CP [IP ...], [C [Cbole] t,]] c. [CP [IP . . .], [C' [C bole] ti]] 2 V° t2

basic order IP-raising CP-raising

20 · Josef Bay er

The problems I see with this account are at least the following: (a) It remains rather unclear what could drive leftward movement in these cases. As I have pointed out in Bayer (1999) it is extremely difficult to come up with arguments in favor of triggers for IP- and CP-movement. But such triggers are required, the theory being embedded in the larger theory of the Minimalist theory of language, in which economy principles would block movements that are not required by interface conditions, (b) Such a theory necessarily ascribes to a given language a single parameter. This follows trivially from the assumption that strictly speaking word order parameters do not exist. Variation has to be ruled exclusively by the strength of morphological features. These morphological features rest in functional heads. If they are strong, they attract a (near enough) phrase that is able to satisfy its need of being checked (and perhaps erased); if they are weak, the checking process is—depending on the actual version of the theory—either postponed past Spell-Out, i.e., until LF, or there is simply invisible feature movement. The theory in question would be supported by dialectal or cross-linguistic variation according to which in some dialect IP raises to SpecCP, while CP remains in post-verbal position, or IP fails to undergo movement to SpecCP while the entire CP is attracted by some higher functional head, or both IP and CP remain in their base position. The first derivation would yield sentences of type (7c), which my informants found either less acceptable or marked. The second derivation yields the structure [CP [C [C bole] [IP .. .]]]i V° t b the third one would yield V° [CP [C [C bole] [IP ...]]], both of which are ungrammatical; see our previous remarks in connection with (2a) and (2b). Without detailed information about Bengali, dialectal variation one cannot be absolutely sure, but nothing of this sort has been reported in the literature I had access to. In my view it is very unlikely that structures which are severely ill-formed in one variety should be grammatical in a closely related variety. In the case at hand the reason for such nonvariation seems to be that the distribution of je· andfco/e-clausesdoes not rest on parametric variation or any of its more recent implementations in terms of feature strength at all, and that the apparently confusing picture of the language stems from a mixture of two grammatical systems in which parts of a VO-model are incorporated but not integrated into a dominant OV-model. • 3.2. How is the IC-clause Attached? If we reject the idea that FC- and IC-clauses are uniformly merged with the matrix verb in the order head-complement, we have to specify an alternative. For FC-clauses I simply assume that they are merged as the complement of the verb, and that the basic order is CP-V. For IC-clauses, the picture looks a lot less obvious. That in South Asian languages IC-clauses are somehow more loosely attached has already been suggested in the literature. Dwivedi (1994) offers a theory of Hindi complementation which claims that fa'-clauses are completely disconnected from the main clause. The idea is that they are

Two Grammars in One · 21

licensed somewhat like coordinate constituents. According to Dwivedi, main «clause and subordinate clause are dominated not by the IP-node that dominates the main clause but by a node T (for 'text')—a concept borrowed from Heim (1982) which indicates that there is no syntactic relation at all between the main clause and the fa'-clause. One of Dwivedi's main arguments is that ki«clauses cannot be selected because their head ki remains invariant across all possible semantic clause types. In the typical case, she argues, selected CPs show a smaller or larger degree of variation in the form of the C-morpheme. It follows that /a-clauses, had they been selected, would not be properly 'typed' in the sense of Cheng (1991). Applied to Bengali ;e-clauses, this argument does not hold. As has been pointed out already Je is incompatible with embedded interrogatives. Dwivedi is right, however, in having realized that the ICclause is merged in a special way. The nature of the fa'-clauses in Hindi remains a controversial issue as long as we do not know more about their initial appearance in the language. The nature of the Bengali ;e-clause seems to be less mysterious, if we take it that the formal coincidence between the complementizerye and the relative pronoun ye is not accidental, the null theory being that they are the same. This was claimed to be the case for Oriya (see Bal 1990), a language which resembles Bengali in much syntactic detail. Bal suggested that only bole (or Oriya boli) is a complementizer, and that je is an operator, no matter where we see it. Although I believe that this radical view cannot be maintained, there is an important insight in Bal's theory, namely that they'e-clause has originally started out as a noun-modifying clause. To see this, one should realize that a very typical pattern forye-clauses (as well as for ^/-clauses) is such that there is a nominal correlate in the expected object position to the left of the verb. This correlate can be either a simplex pronoun like ßá or a demonstrative pronoun like eTa, or a dummy NP such as e kOtha ('this story'). An example is given in (10). (10) also indicates that once such a dummy NP is present, the presence of je is obligatory.14 (10) cheleTa c kOtha jane na *(je) baba aS -be boy -CL this story knows not C father come-will 'The boy does not know it that his father will come.' In (10) the;e-CP cannot be in an Α-position. The Α-position is occupied by the dummy NP e kOtha. Following Srivastav's (1991a, b) analysis of Hindi, we assume that Case and theta-role are assigned to this position. The;e-CP is an adjunct which is licensed by coindexation with e kOtha. In my view, this construction must play a key role in the analysis of sentential complementation in OV-languages. Assume that right-hand complements have started out as noun-modifying clauses in which the operator ye acts as a general relativizer comparable to English such that. Noun modification of this sort is not unique to Bengali. For instance, Matsumoto (1997: ch. 5) points out that this is a common pattern in Japanese.151 would look at constructions like that as reflecting

22 · Josef Bayer

a stage in the development of genuine forms of subordination: there is first ai dummy NP or a pronoun which completes the argument structure of the matrix verb, and a loosely attached clause which specifies the content of the dummy NP. At this stage we are between syntax and discourse since there is no argument clause yet, while on the other hand, the explicational clause is not simply added in the sense of an asyndetic text continuation; it is rather connected by the relativizing operator. Imagine now two processes that may have applied to the grammar: (a) The dummy NP is deleted.16 (b) The operatorye is re-analyzed as a complementizer. In this case we end up with a grammar that is suspiciously similar to a grammar in which a clausal object may be licensed just like an NP-object (if such is possible at all). The difference is now that after deletion of the dummy NP the adjoined CP would not be connected to the clause at all. On the other hand, deletion of the dummy NP would violate the Projection Principle which—informally speaking—requires that the valency of the verb cannot be altered in the syntactic derivation. The scenario allows a very plausible resolution of the conflict, though: the Projection Principle cam be satisfied, if the verb assigns its thematic role to the stranded CP itself rather than to the dummy NP. The transition from (lla) to (lib) shows which reanalysis must be assumed in order to resolve the conflict. (11) a. ... dummy NP! V° [CP OPi ...]i = delete NP!/re-analyze OP! -> b. ... 0 V°[CPC°...] Under the assumption that theta-marking is not dependent on the parametric choice of a language, but solely on the phrase structural configuration and the lexical property of the head involved, we predict that in (lib) the verb wiU theta-mark its right sister. We assume this to be the syntactic core of what Paul (1886) has referred to as the 'extension of the object relation to the clause'.17 • 3.3. An Explanation of the Immobility of the IC-clause Let us on this basis return to the question why FC-clauses are mobile while 1Cclauses are immobile to the extent that they are frozen in place. The facts have already been summarized in (8). In order to approach this question, I adopt a distinction that has been drawn in Chomsky (1995). Chomsky proposes two kinds of nullification, a weaker form which he calls deletion and a stronger form which he calls erasure. We are dealing with deletion, if the element in question has disappeared phonologically but is still present in the abstract syntactic representation. Erasure, on the other hand, amounts to the radical removal of the element in question from the representation, i.e., not only from the phonological but also from the abstract syntactic representation. Assume now that in the licensing of aye-clause the NP-correlate of the clause may have been deleted but not erased. If this is the case, the symbol 0 in (1 Ib) does not represent total emptiness but rather the absence of a phonological matrix.

Two Grammars in One · 23

En the abstract syntactic representation there would still be an empty NP of some sort. Within the context of the Minimalist theory of grammar this may be even necessary, if we assume that it is not they'e-clause itself but rather a nomimal element which is required to check the object feature of the transitive verb trhat is involved here. On the other hand, there is evidence that in (lib) theyeclause has indeed become the direct object of the verb. Therefore we do not want to say that in (1 Ib) 0 is simply an inaudible syntactic category that results from the deletion of its phonological matrix. Instead we assume that there is a direct selection relation between the categories V and CP, while the object feature of V is checked by a feature F that is associated with CP, say, by an iindex as usually. If this is the case, (lib) changes into (12) (12) ...Ρ,ν^ΟΡΟ 0 ...], Assume now that CP undergoes leftward movement, and that movement targets either the position where features of the object would be checked, or some higher position. The two possibilities are shown in (13a) and (13b) respectively.

(13) a. ...[CPC°...] 1 V°t 1 b. [cpC^./hF^t, ((13a) can be ruled out on the assumption that theye-CP itself does not have a feature that could check the verb's object feature. Feature checking in (13b) is as required, however. So why should theye-CP not be allowed to undergo leftward scrambing or topicalization? The answer is that its trace would fail to be Α-free. Notice that F checks a slot that is associated with the object- or transitivity-feature of the verb. In this sense, F shares an index with CP. But then the movement of CP to an Α-bar position across F amounts to a case of Crossover.18 To summarize so far, we have seen that FC-clauses differ from IC-clauses substantially, both according to their distribution in the clause and according to their mobility. FC-clauses appear more or less in situ as far as we can speak of the position to the immediate left of V as the 'in-situ' position in a headfinal language. From there they can undergo leftward movement, but less so rightward movement; rightward movement of FC-clauses seems to have t hroughout a very different function than any kind of leftward movement. ICclauses, on the other hand are frozen in the post-verbal position. In a strictly head-final language like Bengali or the other South Asian languages under discussion, these clauses are clearly the Odd-ones-out'. We have suggested that their properties derive from the following facts: (a) The complementizer j*e is diachronically a relativization operator. In the IE model, such operators undergo leftward movement to a designated operator position, (b) The origin

24 · Josef Bay er

of the ye-clause as a noun-modifying relative clause is supported by the fact thatye-clauses can be visibly linked to NP-correlate.19 (c) This correlate can be deleted, but it is in all likelihood not fully erased. If this is true, we find an explanation of why it is impossible for the/e-clause to move to the left. Its distribution is again similar to the distribution of a right-hand relative (*[Who\ hit mydog]\ I hate [theguy]\ ti). We will complete the syntactic discussion about 1C- and FC-clauses by taking a look at their wh-in-situ properties, in particular how far the scope of wh can be extended. • 3.4. FC- and IC-complements and Wh-scope It is widely known that wh-in-situ languages vary with respect to how far the scope of wh-elements can be extended. In Chinese, a wh-element in a complement clause may obtain scope over the matrix clause. It has been pointed out by Davison (1988), Srivastav (1991a, b) and Mahajan (1990a; 1994), although with different theoretical conclusions, that this is not possible in Hindi. A whelement in a Hindi fa'-clause is never able to obtain scope over the matrix: clause. As I have pointed out in Bayer (1990; 1996), the same is true for Bengali. I expect it to be true for comparable languages, i.e., head-final languages with right-hand IC-complements, although I have checked only a few of them.. As we have said earlier, Bengali differs from Hindi in disallowing the presence of aye-complementizer in a clause that should be interpreted as interrogative.. Therefore, no je will be involved in the following examples. (14) ora Suneche [ke aS -be] they heard who come-will i. igfoA (Canfield 1980) Nancy bich'i show anileeh Nancy to.3 show make.2 'You show it to Nancy.' (19) Warlpiri-Ercg/wA (Bavin and Shopen 1985) a. grow jam mi grow become NNPST To grow/ b. hold mani hold make To hold/ (20) Japanese-English (K. Nakamura, personal communication) kanai ga kimono o serekuto shimasu wife SBJCT kimono OBJCT select do 'My wife chooses a kimono.' In addition to these language pairs, the following are also reported in the literature as exhibiting this structure (matrix language first): Philippine Creole Spanish-English (Molony 1977), Philippine Creole Spanish-Tagalog (Molony 1977), Spanish-English (Pfaff 1976), Shona-English (Myers-Scotton 1993). For some of the language pairs referred to above, the Light Verb Construction appears to be less stable than for others. Because the data in Hindi are relatively clear-cut, we limit ourselves to this case, assuming that our discussion is extendable in some form to other cases while recognizing that languages may differ in the detailed treatment of this structure. • 2.3. 'Conjunct' Verbs and the Light Verb Construction Before proceeding to an analysis of the Light Verb Construction, we will consider a related construction in Hindi—the structure traditionally termed the 'Conjunct Verb Construction' as exemplified in (21)-(24) below with both monolingual Hindi and code-switched Hindi-English exemplars. (21) Monolingual Hindi merii patnii [NP saaRii kii pasand] kar -egii my wife saree of liking do -FUT.3.SG.FEM 'My wife will take a liking to a saree/ (22) Code-switched Hindi-English merii patnii [NP saaRii kii choice] kar -egii my wife saree of choice do -FUT.3.SG.FEM 'My wife will choose a saree/ (23) Monolingual Hindi ye davaaii [AP mujh ko accha] kar -egii this medicine me to better do -FUT.3.SG.FEM 'This medicine will make me better/

44 · Tej K. Bhatia and William C. Ritchie (24) Code-switched Hindi-English ye davii [APmujh ko better] kar -egii this medicine me to better do -FUT.3.SG.FEM

and Forthcoming). In Dravidian, by contrast, verb-initial structures such as s (1) earlier seem to be rare and lacking the discourse features of Sanskrit verb- initial structures. Moreover, as suggested by 1988 fieldwork, native speakers > of Modern Dravidian interpret structures of this sort (as in example [2]) as > resulting from multiple extraposition. (2) kotta avanu avanige eradu ettu give-sg.3 she he-sg.dat.m two beating-pl.n 'She gave him two beatings!' Perhaps significantly, Modern Indo-Aryan appears to have converged with Modern Dravidian by (a) tending toward relatively strict verb-finality in formal discourse and (b) employing verb-initial structures such as (3) only rarely, , without the discourse features of Sanskrit verb-initial structures, and with connotations very similar to those of their Dravidian counterparts (see the translation). (3) demge tumhe cämtä give-pl.lm you-sg.dat. slap-sg.m 'We = I will give you a slap!' It is, thus, only at the relatively modern stage that we find fairly convincing evidence for convergence. And while the loss of free verb-fronting in IndoAryan may be a case of convergence of Indo-Aryan toward the Dravidian pattern,4 the greater restriction of word order freedom is multiply ambiguous in interpretation. It could have been a Dravidian innovation which spread to Indo-Aryan—or vice versa; it could be a convergent development in both language families; or it could even result from parallel independent innovations At this point I am not aware of any evidence which might resolve this issue.

• 3. Subordination and Finiteness Since the days of Caldwell (1875 [1913]: 488), those advocating Dravidian sub version have tended to think of pre-contact Dravidian and Indo-Aryan/Indo European as fundamentally different in terms of the syntactic or morpho syntactic encoding of the pragmatic notion of subordination and in terms rf the distribution of finite verbs. Dravidian is claimed to express subordinatioi by non-finite means (relative participles, absolutives) and to have seven restrictions on verbal finiteness, while Indo-European is assumed to havi expressed subordination through dependent clauses with finite verbs, or δ least to have preferred such structures; see especially Kuiper 1967.5

Typology vs. Convergence m 69 Here, too, a careful re-examination of the evidence suggests a rather differt ent picture: Dravidian and Sanskrit/Indo-European were typologically simi1 lar, with both embedded non-finite subordination (relative participles and ; absolutives in Dravidian, participial structures and perhaps some kind of ;absolutives in Indo-European) and non-embedded finite subordination ((quasi-coordinate relative-correlative structures); see Hock 1988, 1989, and 1993a with references. On the Indo-Aryan/Indo-European side, the coexistence of both strategies ι is demonstrated by the examples in (4) for relative-correlatives, and (5) for ι non-finite constructions. Note especially the b.- and c.-examples, from early Indo-European languages outside India.6 A similar coexistence is found in ] Dravidian, as seen in the finite-verb relative-correlative example (6) beside \ the non-finite (7). (4) a. [tvam tarn ... b dhasva...]CC you-sg.voc. that-sg.acc.m bind-sg.2.impve.mid [yd no jigharnsati]RC who-rel.sg.nom.m. us-obl.clit. slay-desid.sg.3.indic 'You ... tie down that (evil-doer) who ... tries to slay us.' (RV 6.16.32) b. [nu ku[(UJ)]GAL-us te-ez-zi]RC now what-sg.acc.n king-sg.nom.m say-sg.3.indic. [nu a-pa-a-at i-ya~mi]CC now that-sg.acc.n do-sg.l.indic 'Now, what the king says, (now) that I will carry out.' (KBo 17.1-7: Vs.2.17-18; Hittite) c. [de bacanalibus quei foideratei \ ofbacchanal-pl.dat. who-nom.pl.m. associated-nom.pl.m. esent]RC [ita exdeicendum censuere]CC be-pl.3.subj. thus declare-gerund.acc decide-pl.3.perf. '(About those) who might be associated with the bacchanals, they decided to declare as follows.' (SCBacch.2-3; Old Latin) (5) a. punardaya brahmajayam krtvi devair give-back-abs. brahmin's-wife-sg.acc. do-abs. God-pl.instr.m nikilbisam \ ÜÞáçé prthivya bhaktvaya expiation-sg.acc. vigor-sg.acc. earth-sg.gen.f partake-abs. urug yam up sate wide-strider-acc.sg. pay-respect-pl.3.mid 'After they have returned the brahmin's wife, accomplished expiation through the Gods, and partaken of the earth's vigor, they pay their respects to the far-striding one.' Or: 'They return the brahmin's wife, accomplish expiation through the Gods, partake of the earth's vigor, and (then) pay their respects

70 · Hans Hennch Hock

Lit.: 'Having returned the brahmin's wife, having accomplished 1 expiation through the Gods, having partaken of the earth's vigor, , they pay their respects ....' (RV 10.109.7) b. lusai d'ouk edunanto parastadon loosen-inf. pcle.-not can-pl.3.impf. come-close-abs. After they came close, they could not loosen (him).' Or: 'They could not come close and loosen (him).' Lit.: 'They could not loosen (him), having come close.' (II. 15.22; Greek) ι (6) [e-vari nall-avar atavar]RC [α-ναπ nall-ai]CC which-place good-pl.3m men-pl.3m that-place good-sg.2 'You (are) good in a place where the men are good.' (Pura 183.3-4; Old I Tamil) (7) nan atai-c ceytu mutittu-p p nen (Tamil) I-sg.nom. that-sg.acc.n do-abs. finish-abs. go-sg.l.past 'After I had done that and finished I went away.' Or: Ί did that, finished, and (then) went away.' Lit.: 'Having done that, having finished, I went away.' • 3.1. Relative-correlative Structures7 Interestingly, while the non-finite absolutive of Sanskrit is commonly attributed to Dravidian subversion, the Dravidian relative-correlatives have been t considered to result from Indo-Aryan influence; see e.g., Burrow & Bhattacharya (1970); Nadkarni (1975); Sridhar (1981).8 Recent research in Dravidian historical/comparative linguistics suggests that relative-correlatives are inherited and not the result of Indo-Aryan subversion; see Lakshmi Bai (1985); Ramasamy (1981); and especially Steever (1988). As Downing 1978 notes, relative-correlatives are fairly common in SOV languages. And following ideas of Steever (1988), I have argued (Hock 1988,1989,1993a) that they are typologically compatible with the tendency of SOV languages to place restrictions on the occurrence of finite verbs in syntactically complex structures, if they are analyzed as syntactically coordinate, rather than subordinate. The generalizations to be assumed for languages of this type are given as 'Rule A (below). For Sanskrit, I provide ample evidence for this assumption in Hock (1989). (A more cautious alternative to Rule A would be to claim that relativecorrelative structures are compatible with SOV restrictions on finiteness, since their relative clauses are non-embedded, and finiteness restrictions only apply to embedded structures.) Rule A: One finite verb is permitted per sentence. However, in conjoined structures, each of the conjoined clauses is treated as a sentence in its own right and is therefore permitted to have its own finite verb. Relativecorrelatives syntactically are conjoined structures, permitting each of the two

Typology vs. Convergence · 71

sub-clauses to have its own finite verb. Elsewhere, i.e., in structures with syntactic embedding, only one finite verb is permitted per sentence, occurring in the 'matrix' clause; other verbs have to appear in non-finite shape (participles, absolutives, infinitives, or verbal nouns).9 Steever (1988) offers a very different account of Dravidian relativecorrelative structures, partly based on the evidence of modern southern Dravidian structures of the type (8), in which the relative clause is followed by a clitic marker, -e or -o. According to his Rule (1), predicates are permitted to be finite (i) in the 'root' (or matrix) clause, and (ii) in embedded structures ccommanded by elements licensing or 'shielding' finite verbs, such the clitic -o of relative clauses (see example [8]), while elsewhere predicates are non-finite (1988: 36).10 Moreover, according to Steever, this typology, including the 'clitic' relative clause type, is to be reconstructed for Proto-Dravidian. (8) SO[S1 [embe ni istam kinid]Sl = ban id?a]]SQl} where you-sg. desire do-sg.2.fut. + or there putimpve 'Put it wherever you want to.' (Konda) Based on Old Tamil evidence similar to that in (6) above,121 argued in Hock 1988 that the type with postposed clitic must be an innovation; and my claim is now supported for Tamil by Th. Lehmann (1998, the source for example [6]): 'In this correlative construction a finite clause is directly embedded under a head NP without any embedding device; later in Modern Tamil, a correlative clause must be embedded with a clitic such as = ö or = e.J Further support for considering the 'clitic construction' to be an innovation comes from Old Malayalam (9), from more northern, 'tribal' Dravidian languages (10)—(14),13,14 and interestingly also from Koraga (15), a language in southern Karnataka, which Bhat (1971) argues, belongs to the North Dravidian branch of the Dravidian language family. (9) a. \yatontu mahäräjaniyögam]RC [atu a-vannam]CC what.n-one maharaja-order that.n. that-manner 'What is the king's order, (let) that (be done) in that manner' (Old Malayalam; Pillai 1973: 165). b. [ävan adhika punyan]RC [avan = e who-nom.sg.m very righteous he-nom.sg.m. = emph. sevyari\CC venerable 'Who is very righteous, he indeed is venerable' (Old Kannada; Kittel 1903, § 267).15 (10) [ät cenmor ale]RC [an bele immodi where go.p42.polite pres address pcle I emph.pcle you-with

72 · Hans Henrich Hock

(11)

(12)

(13)

(14)

(15)

verran]CC go-sg 1 fut. 'Wherever you go, I will go with you' (Parji; Burrow & Bhattacharya 1953: 114, #77). [ed et kedavar dolnen]RC [ad at kedavar where where side fall-sg.3m.dur. there there side gO'lner]CC beat-pl.3.m.dur. On whichever side he fell, on that side they beat him' (Kolami; Emeneau 1961:168, #172). [ece kärdo puni änal]RC how much time.loc. full-moon become sg.S.nonmasc.fut. [äserj parbu tinaka\CC we.incl. festival cat-verbal noun At whatever time the moon is full, (at that time) we must eat the feast' (Pengo; Burrow & Bhattacharya 1970: Story 5, # 3). [embäsi e barhäyani payinesi]RC [evanaki who.sg.m this wild-pig.acc. kill-sg.3m.nonpast he.dat.sg.m. nä mäfiga blha kifi\CC I daughter marriage do-sg.l.nonpast shall give my daughter in marriage to whoever kills the wild pigs' (Kuvi; Israel 1979: 261). [endrannim neor]RC [ädin nimäge what-acc.you-pl.nom. ask-pl.2.m.fut. that-acc. you-pl.dat. ci'on]CC give-sg.l.fut 'Whatever you will ask, I will give to you' (Kurux; Hahn 1911: 111). [anna petige kala pöyerega eti jana raken]RC my box bring go-abs. how-many people flee-pl.3.past [ati janakana sipayli bara guddu so-much people.dat.(?) soldier come.abs. beating pedodu]CC put-sg.3.nonmasc.assertive 'Let as many soldiers as there are people who took away my box come and beat them' (Koraga; Bhat 1971: 61, line 1-2).

Old Malayalam and Old Kannada, thus, add to the evidence for clitic-less relative clauses in early southern Dravidian. The significance of the data from northern, 'tribal' languages for the same kind of relative clause structures lies in the fact that these are the languages which best preserve Steever's Serial Verb Construction' (see Steever 1988 and see also Section 4.2 below) and thus may be expected to also preserve other syntactic archaisms. In this regard, note further the evidence of 'quotatival', rather than fully quotative constructions in the same languages (Section 4 below). The fact that Koraga,

Typology vs. Convergence · 73

apparently a genetically North Dravidian language located in southern territory, likewise shows clitic-less relative clauses (as well as quotativals, see Section 4 below) is worthy of note and raises interesting questions about the earlier history of the language. The cumulative evidence of Old Tamil, Old Malayalam, Old Kannada, and the northern Dravidian languages (including Koraga) supports a reconstruction of clitic-less relative-correlative structures for Proto-Dravidian and thus suggests a closer typological similarity between early Dravidian and IndoAryan than commonly assumed. By contrast, the southern relative clause type with postposed clitics must be an innovation, a fact which confirms the suspicion raised in the introduction to this paper that the current linguistic situation in the southern Dravidian languages is not necessarily identical with the situation at the time of Rig-Vedic Indo-Aryan. This finding has consequences which extend beyond its implications for the question of prehistoric convergence or subversion. First, only the modern southern Dravidian languages adhere to Steever's perception of Dravidian as having a 'strict' finiteness typology. Earlier Dravidian, as well as modern northern Dravidian, have the less 'strict' OV typology captured by Rule A given earlier and therefore do not require a clitic to 'shield' the finite verb of the relative clause. The 'strict' OV typology of the modern southern Dravidian languages, thus, is the result of a typological shift. Second, as a consequence of this regionally restricted shift, the modern Dravidian languages exhibit a major typological division between the southern languages which adhere to the much 'stricter' OV typology of Steever's Rule (1) and the northern, 'tribal' languages which follow the 'looser' OV typology of Rule A above. This division between south and north is important because it is paralleled by a similar division as regards the marking of cited discourse by means of quotative vs. 'quotatival' constructions; see Section 4 below. • 3.2. Absolutives As examples such as (5) and (7) show, both Indo-Aryan/Indo-European and Dravidian also provide evidence for non-finite-verb strategies to encode pragmatic subordination and/or 'sequential and' constructions. As far as the use of participles is concerned, this claim is quite uncontroversial; and whatever controversies may exist about the use of infinitives in PIE, these are concerned with whether they were 'true infinitives' or whether they were verbal nouns (see e.g., Disterheft 1980; Gippert 1978). The case for Indo-European absolutives is less strong. And since Kuiper (1967), it has been customary among those advocating Dravidian subversion to claim that Proto-Indo-European had no absolutives and that the absolutives of Vedic Sanskrit are an innovation resulting from Dravidian subversion—even though Kuiper admits that the morphology of the Vedic absolutives is quite archaic and synchronically opaque.

74 · Hans Hennch Hock

As it turns out, the morphology of the Homeric Greek absolutive (in [5b] above) is likewise archaic and synchronically opaque, and as a category, the Old Greek absolutive is moribund. At least one form of this absolutive (in -on) can be compared to one form of the Vedic absolutive (in -am). Moreover, the suffix *-om of both of these absolutives has a morphological counterpart in the Oscan-Umbrian infinitive suffix -urn. Interestingly, the more common Vedic absolutive of the type -tv (ya), -tvi finds its closest morphological parallels in the Vedic verbal nouns in -ft/-, which become clear infinitives in the later language and whose PIE antecedents furnish the so-called supines of Old Church Slavic and Latin. They are also similar to gerundives in -/vo-, later -tavya- (note the quasi-absolutive use of the Old Latin supine in -tu); and the variant absolutive suffix -ya likewise appears morphologically related to a gerundive suffix -ya- That is, both the Homeric Greek and the Vedic absolutives exhibit morphological affinities with verbal noun/infinitive and gerundive structures and are thus not without Indo-European precedent. In this regard it is interesting to note that Old(er) Tamil shows a similar affinity between absolutive and infinitive, in that the infinitive in -a can also be employed as an absolutive. Moreover, like Vedic it exhibits a fair amount of morphological variation; see Beythan (1943: 104-6) and Natarajan (1977: 169-82); the latter cites more than 20 different formations of the Old Tamil absolutive. If the affinity of the Vedic absolutive suffixes to infinitive or verbal noun suffixes and their morphological variability are taken as evidence that these absolutives are of secondary origin, the same argument could be raised for Dravidian. An alternative that I find more attractive, given the strong evidence for the general typology of Rule A in both Proto-Indo-European and ProtoDravidian, is the assumption that languages of this typology exhibit a tendency to develop absolutives as a means of avoiding multiple finite verbs in the same (non-conjoined) clause, and that a common source for such absolutives (even if perhaps not the only one) lies in infinitives and verbal nouns. In fact, as noted in Hock (1996a), the Dravidian subversion hypothesis leaves an important aspect of the morphology of Indo-Aryan absolutives unexplained. There are two distinct and formally unrelated sets of variants for the (nonpresent) absolutive: one is the type -ßíá, -tv ya, -M\ the other is the type -ya. The exact historical reason for the variation is not clear, but it is associated with an eminently Indo-Aryan and general early Indo-European distinction, between simple and prefixed verbs (as in kr-tva 'having done/made' vs. sariiskr-tya 'having made pure'). To judge by the historical record, this distinction did not exist in early Dravidian and therefore cannot be attributed to Dravidian influence. Moreover, the very variation would be unexpected, if the formation of absolutives were due to Subversion: As is well known in historical linguistics, in contact-induced structural change, formal variation usually is abandoned; the introduction of variation in a contact-induced formation would be quite anomalous.

Typology vs. Convergence · 75

• 3.3. Conclusion on Subordination and Finiteness As in the case of major constituent order, it therefore appears as if, even before they came into contact, Dravidian and Indo-Aryan were very similar to each other in their general typology of subordination and finiteness16. And again, we can also note some differences. In Sanskrit, for instance, some genres (such as the epics and fables) have clear sub-genre preferences for relativecorrelatives vs. non-finite verbs, with non-finite verbs preferred in narrative passages, and relative-correlatives (as well as non-finite verbs) being more prominent in cited discourse; see Hock 1994a, b and forthcoming; TsiangStarcevic 1997. Moreover, in this case there are marked differences between early Indo-European languages. While Sanskrit, Greek, and Latin use both relative-correlatives (or their later embedded-relative replacements) and non-finite structures, Avestan and Old Persian, as well as early Germanic use non-finite structures more sparingly, and Hittite vastly prefers relativecorrelatives or simple asyndeton. What is shared by the early Indo-European languages, thus, is not necessarily the specific strategies for encoding subordination (which in fact differ from language to language), but the overall typology of Rule A above which precludes multiple finite verbs in nonconjoined structures. The specific strategies employed, then, can be considered to be different attempts at 'avoiding' the restrictions imposed by Rule A. A more detailed comparative study of Dravidian would be highly desirable in this regard, in order to get a clearer picture concerning the relative role of the two subordination strategies and any possible changes in their use. Discussion with Dravidianists suggests that the southern literary languages use relative-correlatives colloquially only where non-finite strategies would lead to excessive ambiguity, but that the structures are more common in scholarly discourse. By contrast, the northern Dravidian languages seem to employ relative-correlatives more freely; see the examples in (10)-(14), all of which come from folk narratives. Given that the southern languages have innovated by requiring clitic particles after relative clauses, it is tempting to attribute this difference in usage to southern innovation, too. To test this idea it would be helpful to examine in detail the use of relative-correlatives in the oldest (southern) Dravidian texts. While the overall typology of subordination and finiteness in early Dravidian and Indo-Aryan, thus, may have been more similar than commonly assumed, some differences in detail can be discerned even at this point. Thus, virtually all of the relative-correlatives of the northern Dravidian languages are of the generalizing ('whoever ...', 'whatever ...') type;17 their Indo-Aryan counterparts have no such restrictions. The fact that southern Dravidian languages can use relative-correlatives with non-generalizing function could possibly be attributed to Indo-Aryan influence. Another difference concerns the finiteness of the conditional/optative mode. The conditional mode of Dravidian is -morphologically non-finite,

76 · Hans Henrich Hock

while Sanskrit has the finite category of the optative (and of the conditional). In this regard it may be significant that modern Indo-Aryan languages employ a form of the clearly non-finite present participle in contrary-to-fact conditionals, such as Hindi turn jaldi ate to tumhe khänä miltä, '(if) you had come quickly, you would have gotten food'.18 This innovation may well be the result of Dravidian influence. As in the case of major constituent order, such matters of detail are more likely to furnish evidence for convergence or subversion than the more general patterns that are usually focused on.

• 4. Quotatives As noted in Hock (1996a), Of the syntactic features commonly attributed to Subversion, the [early Indo-Aryan] use of the quotative marker iti may be the best candidate.' For preliminary examples of Rig-Vedic and Dravidian quotatives, see (16) and (17). (16) [tvam stosäma...] \\ iti tvägne... you-sg.acc. praise-pl.l.aor.subj. Q you-clit. = Agni-sg.voc. fsayo 'vocan seer-pl.nom.m say-pl.3.aor. '"We will praise you ..." (thus) the seers said to you, Agni' (RV 10.115.8-9). (17) (Tamil) nan [ [avan inke vantän] enru] ninaikkiren I-sg.nom. he-sg.nom. here come-sg.3m.past Q think.sg.l.pres. think [(that) [he has come here] ].' • 4.1. Indo-Aryan and Indo-European Although in earlier publications (Hock 1982a, 1984) I have argued that there is ample evidence for parallel structures to (16) in other early Indo-European languages, on closer examination this claim turns out to be problematic. First, in early Vedic the order of the quotative marker iti with respect to the quote that it marks is not yet fixed; see (18). In this regard, early Vedic is similar to Avestan which has the markers uiti and iSa (< ) in remarkably similar constructions; see (19). The only difference is that in the Rig Veda the type (18b) occurs most frequently; in Avestan, the type (19a) is most common. Significantly, beyond their overall syntactic similarity, the markers employed Vedic iti and Avestan uiti/ also agree in their original meaning, which is 'thus'. (18) a.

///-initial: iti braviti vaktan raränah \ Q say-sg.3.pres. speaker-sg.m.nom. giving-sg.m.nom.

Typology vs. Convergence m 77

b.

c.

(19) a.

b.

c.

\yasoh vasutva käravah good-sg.n.gen. goodness-sg.n.instr. singer-pl.m.nom. anehah] guiltless-pl.m.nom. '(Thus) says the giving speaker "Through the goodness of the good, the singers are guiltless"/ (RV 10.61.12). iti-final: nakih vaktd [na dät] M no-one-sg.m.nom. say-sg.m.nom. neg give-sg.3.subj. Q 'No one is about to say "He shall not give"/ (RV 8.33.15). iti plus verb of speaking inserted: [idam udakam pibata] iti abravltana (\) this-sg.n.acc. water-sg.n.acc. drink-pl.2.impve Q say-pl.2.impf [idam vä ghä pibatä munjanejanam] this-sg.n.acc. or pcle. drink-pl.2.impve rinse-water-sg.n.acc. '"Drink this water," you said, "or drink this rinsewater"/ (RV 1.161.8). uai-initial: uityaojanä miOrai... [... ] Q-speak-pple.pl.m.nom. Mithra-sg.m.dat. 'Thus speaking (they cry to/address) Mithra ...: "..."/ (Yt. 10.1). -final: [dazda manarjhö] para im ida manarjhe dazda manaijho" for pron.sg.S.clit. Q mind-sg.n.dat. cinasti mean-sg.3.pres. '"Dazda manai]ho" (a quotation from 27 on which 19 is a commentary) teaches/means "for the thought/for thinking"/ (Y 19.13). uiti plus verb of speaking inserted: [usta ahmäi naire mainyai] \ hail that-sg.m.dat. man-sg.m.dat. authoritative-sg.m.dat. uiti mraot ahurö mazdä \ Q say-sg.3.impf. Ahura-Mazda-sg.m.nom. [äi... zara6ustra] O Zarathushtra.sg.voc.m "'Hail to the authoritative man," said Ahura Mazda, "O ... Zarathushtra"/ (Yt.10.137).

Note that Avestan is an Eastern Iranian language which seems to have participated to some (more limited) extent in the convergent phonological developments that introduced a contrast dental : alveolar : retroflex in Dravidian, much of early Indo-Aryan, and perhaps other languages as well (Hock 1996b). The development of quotative markers therefore may be claimed to have

78 · Hans Henrich Hock

been another development shared with early Indo-Aryan and attributable, directly or indirectly, to Dravidian contact. However, as I observed in earlier papers (Hock 1982a and 1996a), there is an important difference between early Rig-Vedic Sanskrit and Avestan on one hand and post-Rig-Vedic early Indo-Aryan on the other. While the placement in Rig-Vedic and Avestan was rather free (postposed, preposed, or inserted into cited discourse), the normal pattern of post-Rig-Vedic is the postposed one. And it is this postposed use which is comparable to the postposed Dravidian marker generally cited in this regard—an absolutive of a verb of speaking, such as Tarn. enru. Compare again the examples in (18) and (19). As I note in Hock (1996a), 'This post-Rig-Vedic change ... creates certain difficulties for the Subversionist explanation of the Rig-Vedic quotative: How can we attribute to Dravidian influence the early "free" use of to', if only the later postposed one mirrors that of the Dravidian quotative?' As I concluded in earlier publications, the best we can say at this point, then, is that quotative marking is a feature common to all of early eastern IndoIranian, and that it is only the (near-)obligatory postposition of to which is an Indo-Aryan innovation. This is a view which I would still maintain. However, my earlier arguments that other Indo-European languages, besides eastern Indo-Iranian, provide evidence that quotatives go back to Proto-Indo-European are, on reconsideration, much less well founded. True, Hittite does have a robust quotative marker wa(r) as in (20) (see most recently Fortson 1994), but its etymology (probably from wer- 'to speak') and its syntax (inserted as a second-position clitic into the direct discourse) differ from that of Indo-Iranian. As example (20) shows, the cited discourse is often preceded by an adverb meaning 'thus'; but this clearly is not the quotative marker. (20) nu man kissan kuiski memai now if thus someone-sg.m.nom. speak-sg.3.pres. [annisan]- war[-an LUGAL-izanni formerly Q he-sg.m.acc.clit. throne-loc.n.sg. kuwat tittanut] why place-sg.2past 'Now if someone speaks as follows, "Why did you formerly place him on the throne? ..."' (Apology of Hattusilis 3.73-77). Most important, the Homeric Greek evidence that I cited in earlier publications as evidence for the Proto-Indo-European origin of quotative marking is rather weak. True, the widely attested structure exemplified in (21) is similar to the eastern Indo-Iranian type (18)/(19) in using a particle meaning 'thus', namely hos. However, what seems to be at least equally characteristic of this structure is the occurrence of a finite verb of speaking preposed to the cited

Typology vs. Convergence m 79 discourse, and a postposed second verb of speaking which may be finite or non-finite. The Homeric type, thus, differs from the quotative structures in (18)-(20) by lacking a single, fixed quotative marker and, instead, exhibits a less fixed structure, with multiple verbs of speaking plus the particle hos 'thus'. It therefore cannot be considered a true quotative construction, but is at best "quotatival", and thus cannot be used to support the claim that Indo-Iranian quotatives are inherited from Proto-Indo-European. (21) Peleides d'... prosoipe... \ [ ]| Peleus-son-sg.m.nom. pcle say-sg.3.aor hos photo Peleides thus speak-sg.3.aor.mid Peleus-son-sg.m.nom.s 'But the son of Peleus spoke ... " " Thus spoke the son of Peleus (and ...)' (11.1.223-45). • 4.2. Dravidian (Quotatives, Quotativals, and Serial Verbs)19 Significantly, closer examination reveals that there are also problems with the common view that all of Dravidian has quotative structures marked by a postposed absolutive of the verb en- 'to say so', or at least did so originally. Thie, this is the normal pattern in the southern Dravidian languages. Moreover, the extent to which the absolutive of en- is a quotative marker is indicated not only by its apparent obligatoriness, but also by the fact that when the cited discourse is extraposed to the right, the quotative marker moves along with it; see the Tamil example (22). There appears to be only one exception to the use of the absolutive of en· as quotative marker, and that is the fact that conservative Modern Tamil does not use an absolutive of en- in the presence of a finite form of the same verb. (22) a. nän apporute connen-e [äpatt'äka I-sg.nom. then speak-sg.lpast + pcle dangerous mutiyum] enru end-fut. Q said it then: "It will end dangerously".' 4.2.1. Quotativals in Northern Dravidian In contrast to modern southern Dravidian, the northern languages have what may be called 'quotatival' constructions, which are remarkably similar in structure to those of Homeric Greek, employing a preposed finite verb of speaking (optionally preceded by a particle meaning 'thus'), and a second, postposed finite verb of speaking; see (22)-(25).20 Interestingly, here, too, the geographically southern but genetically North Dravidian Koraga exhibits the same pattern, as in (26). While the two verbs of what may be called the 'double finite' construction may be identical (23)/(24), they do not have to be (22), as

80 · Hans Henrich Hock

in Homeric Greek. Significantly, the double-finite verbs of speaking in (22)(26) are finite, in contrast to the non-finite absolutive of southern Dravidian languages, for which see examples (17) and (21) above. (To save space, cited discourse usually is not quoted in full, but its presence is indicated as [...].) (22) b. ...pokra ge [...] speak.sg.3nonmasc.past narr.pcle enda ge say-sg.3nonmasc.past narr.pcle '... she said "..." (said she)' (Parji, Burrow & Bhattacharya, story 1,10). (23) patlak eneq lakten[...] eneij laklen21 headman en.inf. say-sg.3m.past en.inf. say-sg.3m.past The headman said "..." (said he)' (Kolami; Emeneau 1961, story 1, 17-18). (24) kodiicat ke [...] icat cow say-sg.3nonmasc.past narr.pcle say-sg.3nonmasc.past 'The cow said "..." (it said)' (Pengo; Burrow & Bhattacharya 1970: story 10, 39-40). (25) ...raja ele icesi[...] icesi king thus say-sg.3m.past say-sg.3m.past 4 ... the king spoke thus "..." (he spoke)' (Kuvi; Israel 1979, story 1, #6). (26) pärne sollugi [...] nukid brahmin speak-sg.3m.past say-sg.3.m.past The Brahmin spoke "..." (he said)' (Koraga; Bhat 1971, p. 50, lines -6/5). In addition to structures with the double finite construction, the northern Dravidian languages, plus Koraga, offer a great variety of other structures— another sign of lack of fixity. These include structures with a preposed finite verb of speaking and a postposed non-finite one, structures with a finite verb of speaking preceding or following (but not both), structures with postposed non-finite SAY, and even structures without any verb of speaking at all. The full range of these additional constructions is illustrated in (27) (from Kuvi; Israel 1979). Note that Konda, which goes with the southern Dravidian languages in its relative-clause structures, lacks the type (23)-(26), but does have the entire range in (27). It thus forms a kind of bridge between the northern and southern Dravidian languages. (27) a.

SAY-fin. + CD + SAY-nonfm. + SAY-fin. ... burha onpitesi ke [...] injihl ke old man think-sg.3m.past narr.pcle en.abs. pcle. onpitesi ke think-sg.3m.past narr.pcle 'The old man thought"..." (saying he thought)' (Story 6, #5).

Typology vs. Convergence · 81

b.

SAY-fin. + CD + 0 ...ele ice [-.]0 thus say-sg.3.nonmasc.past '... thus she said "..."' (Story 1, #9; a rare type in most languages). c. ... CD + SAY-fin. ...burhi [...] ice old woman say-sg.3.nonmasc.past 'The old woman said "..."' (Story 1, #5). d. ... CD + SAY-nonfin. + SAY-fin. [...] injihi riskalesi en.abs. ask-sg.3m.past '"..." (saying) he asked' (Story 2, #17; a type which is found occasionally). e. No verb SAY either before or after CD 0 [...] 0 (That is, direct discourse without preceding or following verb; a rare type in Kuvi, but common in Parji and Koraga dialogues.) (Story 6, #54-55).

In spite of the great variation found in the northern languages and in spite of the different structural behavior of the northern and southern languages, one element that seems to unite (almost) all of them is the fact that forms reflecting Proto-Dravidian *en- are the unmarked verb of speaking. Thus, in the northern Dravidian languages they tend to be used as the second verb in double finite constructions and as the post-cited discourse absolutive in structures of the type [27a, d]; and an absolutive of the same root furnishes the quotative marker of southern Dravidian languages. In this regard, then, there is agreement between the two major geographical areas of Dravidian. One suspects therefore that *en- was the unmarked verb of speaking in ProtoDravidian and that, perhaps, it even tended toward quotatival usages. Beyond that, however, we see a major difference between the quotatival structures of the northern and the quotative structures of the southern Dravidian languages. This difference closely parallels the difference between clitic-less northern Dravidian relative clauses and the southern type with clause-final clitic, as well as the survival of Steever's Serial Verb construction in the north vs. its general loss in the south.22 More than that, the geographically southern, but genetically North Dravidian language Koraga agrees with the northern languages both in terms of quotatival (rather than quotative) structures and in terms of clitic-less relative clauses.23 Given that the northern Dravidian retentions of Serial Verb constructions and clitic-less relative clauses are archaisms, it seems a priori likely that the northern quotatival structures likewise are archaic, and that the southern quotative construction is an innovation. This is especially true for the rather idiosyncratic 'double finite' construction in (23)-(26) above. And it is this

82 · Hans Henrich Hock

construction which, I believe, provides the most promising starting point for a historical explanation of Dravidian quotatives and quotativals. The basis for this development may well have lain in something similar to the 'exuberance' of verbs of speaking in vernacular English expressions of the type So she said I'm not going home, she said .... Alternatively, the double finite construction may have arisen as a compromise between the types (27b) with preposed verb of speaking ('SAY') and [27c] with postposed SAY,24 or as an extension of "Linking" Constructions with finite verbs (see [32] below). What is significant, is that structures of this sort are open to reinterpretation as Serial Verbs in languages that permit such structures. To justify this claim and the conclusions to be drawn from it, it is useful to take a closer look at the range of structures that can be considered Serial Verb constructions (SVCs) in Dravidian. 4.2.2. Excursus on Dravidian Serial Verbs

Note that the range to be examined includes structures recognized as SVCs by Steever (all of [28] and [31 a, b]), but also similar structures not recognized by Steever. The latter are included because they share with Steever's SVCs the fact that they contain two (or more) finite verbs that agree with each other in the same syntactic domain.25 The range includes the following: Orthodox' SVCs, in which the two verbs are adjacent to each other [28]. Four different subtypes can be distinguished, of which the first two tend toward being grammaticalized (the 'Auxiliary' Type [28a] and the 'Iterative' Type [b]), the second set of two tends to be lexicalized (the 'Balance Verb' type [28c] and the fa'-echo Type [28d]), and only the third, 'Sequential' type [28e] is free from such tendencies toward specialization. Note that although it is possible to distinguish these five subtypes, example [29] suggests that they are freely combinable with each other. (See also [31a] below.) (28) Adjacent' Serial Verbs: a. Auxiliary'Type i. celvem allem go-pl.l.fut. neg.-pl.l 'We will not go' (Old Tamil; Steever 1988:42). ii. unded m&dad sit-sg.3m.past be-sg.3m.fut/pres 'He is sitting' (Parji; Burrow & Bhattacharya 1953: 58). b. "Iterative" Type: i. ar mandaycar mandaycar they plant-pl.3m.past plant-pl.3m.past '... (wherever) they kept planting ...' (Malto; Mahapatra 1979: 223, #16).

Typology vs. Convergence m 83 ii. ... embati?e burhi burha maceri maceri ... so wife husband stay-pl.3m.past stay-pl.3m.past '... so the wife and husband stayed on ...' (Kuvi; Israel 1979, Story 11, #218). c. "Balance Verb" i. are keyditj ÉáúÀÜßç and hand-sg.acc.nonmasc. foot-sg.acc.nonmasc. r ccal rdstat stroke-sg.3nonmac.past rub-stroke-sg.3nonmac.past 'And she stroked and rubbed (them) hand and foot' (Pengo; Burrow & Bhattacharya 1970, Story 1, #11). ii. ... ncttc vcddn dog run-sg.3nonmasc.pres. chase-sg.3nonmasc.pres. '... the dog chases after [the rabbit]' (Ollari; Steever 1988: 94, from Bhaskararao 1980, Story 1, #75). d. "/o-echo"Type avanukku-p pacUdratu kicUdratu he-sg.m.dat. hunger-sg.3n.pres. "fo'-nger"-sg.3n.pres 'He is always hungry, or some damn thing' (Tamil; Steever 1988: 49). e. "Sequential" Type i. iramcinem v rttinem worship-pl.l past live-pl.lpast 'We worshipped and lived.' = 'We lived worshipping' (Old Tamil; Natarajan 1977: 182, without translation). This type was known to the ancient Tamil grammarians who referred to the first verb of such structures as murreccam 'a non-finite verb (expressed as) a finite verb'; see Natarajan (1977: 182) and T. Lehmann (1998). ii.

jogyeh menjah awdah that mendicant hear-sg.3m.past say-sg.3m.past 'That mendicant heard and said.' = 'The mendicant said after hearing' (Malto; Mahapatra 1979: 187; similarly p. 224, #17). (29) Combination of [28a] (in absolutive form), [28c], and [28e] [cined erod] [[tindar undar] give-sg.3m.past be-abs. eat-pl.3m.fut/pres drink-pl.3m.fut/pres [abs. of Auxiliary Type] [[ Balance Verb ] madrar] sleep-pl.3m.fut/pres Sequential Type] 'He having given (boiled rice, gruel, and vegetables), they ate-anddrank and then slept' (Parji; Burrow & Bhattacharya 1953; Story 9, #189).

84 · Hans Henrich Hock Of the five subtypes in [28], only the lexicalized fa'-echo Type [28d] survives fully in the southern Dravidian languages; the Auxiliary Type tends toward univerbation (Steever 1988,1993); and the others tend to be replaced by structures with non-finite verb (generally absolutive) plus finite verb. This is in contrast to the oldest southern Dravidian for which a broad range of SVCs is attested (see [28a.i] and [28e.i]). The northern Dravidian languages tend to preserve examples of all subtypes, with some variation across languages in terms of which subtypes are found and with what frequency. (A determination of the full extent of variation is one of the desiderata of comparative Dravidian linguistics.) What is especially noteworthy is that according to Hahn (1911: 58), Kurux commonly uses SVCs (with or without intervening particles dam or kl) instead of the absolutive + finite verb constructions of other Dravidian languages. In other northern Dravidian languages, however, SVCs of the type [28], with the finite verbs adjacent to each other, are not very common, and absolutive + finite verb constructions seem to be preferred under adjacency. Steever also recognizes certain SVCs with non-adjacent verbs, such as [30a,b], presumably because [30a] parallels the fa'-echo Type [28d], and [30b] is similar to the Balance Verb type [28c]. On similar grounds, I believe, it is possible to add the examples in [30c,d] because of their parallelism with the Sequential Type [28e]. (30) Non-adjacent Serial Verbs: a. bomman embavana kane Bomma e/i-verb.noun see-neg.sg.l gi-mman embavana käne Gi-mma en-verb.noun see-neg.sg.l do not see = acknowledge anyone named Bomma, named Shmomma' (Medieval Kannada; Steever 1988:125, n. 3). b. päv patti cendar or patti cendar road take-abs. go-pl.3m.past path take-abs. go-pl.3m.past 'They went by the path' (Parji; Steever 1988: 88). c. tal cenda ge mother go-sg.3nonmasc.pres. narr. pcle. [...] enda ge say-sg.3nonmasc.pres. narr. pcle. 'The mother went and said "...".' (Parji; Burrow & Bhattacharya 1953, Story 1, #7). d. are vatat ke and go-sg.3nonmasc.past narr. pcle. samdur kulatiq astat ocean shore-loc. seize-sg.3nonmasc.past And she came (and) took hold of the ocean shore' (Pengo; Burrow & Bhattacharya 1970, Story 1, #10).

Typology vs. Convergence · 85

The parallelism between the Adjacent and Non-adjacent Types is further illustrated by the fact that the two types can be combined, as in [31]. (31) Combination of Adjacent and Non-adjacent Serial Verbs a. vizu ... [[ tinad unad] all eat-sg.3nonmasc.pres. drink-sg.3nonmasc.pres. [ Balance Verb ] vizu näkna poknad ] e all lick-reduced finite finish-sg.3nonmasc.pres. and Sequential Type] sonad pass-sg.3nonmasc.pres. '...everything it eats-drinks = consumes (and) everything it licks up completely' (Konda; Krishnamurti 1969, Story 7, # 15, with p. 304 on the 'reduced-finite' construction in näkna poknad e sonad; see also Steever 1988: 71-5). b. mide miyädara pöno onas manno then bathe-abs. go-sg.3habit. meal make-sg.3habit. banno come-sg.3habit. 'Then he would go to bathe, (he would) take his meal, and (he would) come (back)' (Koraga; Bhat 1971: 62). c. döriti himbonta pojitesi rope-sg.acc.nonmasc. cloth-sg.loc.nonmasc. cover-sg.3m.past tenga dekitesi vätesi injahä stick carry-sg.3m.past. go-sg.3m.past and burhini ele icesi [...] icesi wife-sg.dat.nonmasc. thus speak-sg.3m.past speak-sg.3m.past 'He covered the rope in a cloth, carried a stick, went (off), and then spoke thus to his wife "..." (he said)' (Kuvi; Israel 1979, Story 11, # 181-82). It is probably significant that the Non-adjacent Type, to my knowledge, has left no traces in the southern Dravidian languages. In the northern Dravidian languages, by contrast, it is, if anything, more 'vigorous' than the Adjacent Types. Thus, Bhat's Koraga texts have many examples of Non-adjacent SVCs, while the Adjacent Sequential type embedded in [31b] is quite rare. Before concluding this section it may be mentioned that yet another SVC type can be argued for in Dravidian, a type which shares with the one in [30] the fact that two (or more) verbs thaf agree with each other are found in the same syntactic domain. This is the 'Linking' Tfype in [32], in which a finite verb (± complements) repeats the final verb of the preceding sentence (again ± complements) and in so doing, 'links' or 'chains' the narration. This type is remarkable both because it may involve verbs of speaking (as in [32c]) just as

86 · Hans Hennch Hock freely as any other verb and because it goes against the common claim (going back to Bloch 1929) that only non-finite verbs are used for this purpose in Dravidian. While the latter claim does indeed seem to hold true for modern southern Dravidian, it evidently is not correct for the northern languages— although even here, non-finite constructions (usually with absolutives) are a common alternative. (32) 'Linking' Serial Verbs: a. bikse p dara bakki \ bakki tanno mala alms go-abs. return-sg.3m.past come-sg.3m.past own all biksedo lekka s ki alms-gen(?) account look-sg.3.m.past 'He went for begging and came back. He came and looked at the accounts of his begging' (Koraga; Bhat 1971: 48; a common construction). b. karde gajakan tan \ karde gajakan boy big-sg.m.nom. become-sg.3m.past boy big-sg.m.nom. Üßáç-e budiya icat become-sg.3m.past-clit. wife say-sg.3nonmasc.past '... the boy became big. When the boy had become big, the wife said ...' (Pengo; Burrow & Bhattacharya 1970, Story 2, # 1-2, with p. 60 on the fact that in this language -e is just a particle, not a linker). c. [...]| nukki | say-sg.3m.past a nukki kelesi bakki pera he say-sg.3m.past barber come-sg.3m.past back '"..." he said. He, the barber, said (and) came back' (Koraga; Bhat 1971:58; a common construction). d. [...] izi idtan \ say-abs. say-sg.3m.past idtanr e ... rostan say-sg.3m.past linker arrange-sg.3m.past '"..." he said. He said (and) arranged ...' (Konda; Krishnamurti 1969, Story 5, #17-18). The findings of this section can be summarized as follows. Older southern and modern northern Dravidian have a broad range of SVCs, with two or more finite verbs in the same syntactic domain and agreeing with each other. The range includes both Adjacent and Non-adjacent SVCs. Except for the faecho constructions (and a few other traces), the southern languages have replaced the SVCs by structures in which all verbs but the (canonically) final one are non-finite (typically, absolutives). In the northern languages, too, structures with absolutives are an alternative to SVCs. This is especially the case when the verbs are adjacent to each other.26

Typology vs. Convergence m 87 4.2.3. From Quotatival to Quotative Against this background, let me provide a speculative account of how the southern quotative can be derived from the quotatival double finite construction of northern Dravidian, under the assumption that the double finite construction, as in the Kuvi example (25), repeated here in fuller form as (33), was reinterpreted as an SVC. (33) ... raja ele icesi king thus say-sg.3m.past [mekn necu vesta?i] icesi pulling-out day show-sg.lnonpast say-sg.3m.past 'The king said thus, "I will tell you the pulling-out day" (he said).' Given this reinterpretation, an extension of the double finite construction to structures with embedded, rather than extraposed, cited discourse might be expected. And in fact, occasional instances of this type are met with, such as (34), a structure which Steever includes among his SVCs (1988: 78). (34) bugabuga icat umtat buga buga say-sg.3nonmasc.past fly-around-sg.3nonmasc.past 'Saying/making the sound "buga buga" it flew around' (Pengo; Burrow & Bhattacharya 1970, Story 11, # 66). However, as noted in the preceding section, adjacent SVCs have strong competition from structures with nonfinite verb + finite verb. Structures of the type (35), a fuller version of (27d) are therefore what one would expect. (35)

...CD + SAY-nonfin. + SAY-fin. [e heyä häritani hfyamu] that eggplant left-over-acc.nonmasc. give-pers.obj.impve. injihl riskatesi en.abs. ask-sg.3m.past '"Give me the extra eggplant" saying he asked' (Story 2, # 17; a type which is found occasionally).

While structures of this sort are by no means as frequent in northern Dravidian as the double finite type (33), they are precisely the kinds of structures that would be expected in the southern languages, with their much 'stricter' OV typology. First, by employing a non-finite verb of speaking, these structures are in accordance with the southern shift away from SVCs. Second, their embedded, rather than extraposed, cited discourse conforms to the strong tendency of (formal) modern southern Dravidian to avoid extraposition and thus to have the finite verb in canonical final position of the entire sentence. Third, the absolutive of the verb SAY in the resulting construction could then

88 · Hans Henrich Hock

be re-analyzed as a QUOTATI VE marker which, like the post-relative clause clitic, would 'shield' the embedded cited discourse from the consequences of the southern 'strict' OV/finiteness constraint; compare (36). (36) a. Inferrable pre-modern Tamil (Quotatival CD + SAY-nonfin. + SAY-fin.) nän [avan inke vantän] I-sg.nom. he-sg.nom. here come-sg.3m.past enru ninaikkiren en.abs. think.sg.lpres. '"He has come here.," (saying) I think.' b. Attested modern Tamil (Quotative CD + Q + SAY-fin.) nän [ [avan inke vantän] *nru\ I-sg.nom. he-sg.nom. here come-sg.3m.past Q ninaikkiren think.sg.lpres. think [(that) [he has come here] ].' As a consequence of this reinterpretation, finally, the quotative marker moves along with extraposed cited discourse in the modern southern languages, as in (37). (37) nän ninaikkiren [ [avan inke vantän] enru] I-sg.nom. think-sg.lpres. he-sg.nom. here come-sg.3m.past Q think [(that) [he has come here] ].' While the present hypothesis is admittedly speculative, it manages to relate the difference between the modern southern quotative and the northern quotatival to the overall typological difference between these two groups. Morever, it receives support from an apparently parallel development in what would at first appear to be an unlikely quarter—the Tibeto-Burman languages, a language family usually ignored in the debates over early South Asian subversion or convergence. 4.2.4. A Tibeto-Burman Parallel

A number of Tibeto-Burman languages have structures very similar to both the North Dravidian double finite quotatival construction and to the CD + SAY-nonfin. + SAY-fin. type in (35), see (38) and (39).27 Note especially (38b) beside (39) which shows variation between SVC and absolutive + finite verb in the same language. Some Tibeto-Burman languages have further developed genuine quotatives, employing an absolutive of a verb of speaking and exhibiting syntactic behavior entirely comparable to that of modern southern Dravidian; see for instance (40).

Typology vs. Convergence m 89 (38) a. Shi-chyam shi-phe tsun-wu-ai-ga [...] ngu-wu-ai (Kachin) he-to him again say-past + pcle say-past 'He said to him again "...", he said/ b. Tara äbä-se rhäng-lächäkara-tä päng-ji [...] päng-ji (Murmi) But father-erg.own servants-dat. say-past say-past 'But the father said to his own servants, "...", he said.' (39) Ngä re-chhi rhäng-lääbä chhyämni-si [...] I rise-abs. own father-dat. go-abs. pang-si päng-lä (Murmi) speak-abs. speak-fut. , rising, going to my own father "..." saying shall-say.' = shall rise, go to my father and say "...".' (40) chha-mha seoka-yä-ta sala-täo, [...] dha-ka ne-na (Newari) he servant-dat. call-abs. Q ask-past 'Calling a servant, he asked, "...".' The earliest attested language of this group, Classical Tibetan, shows even more complex constructions, with preposed SPEAK plus preposed di 'this' and postposed de 'that', elements such asskad(a) 'speech', pre- and postposed ces(a) 'thus', as well as pre- and postposed absolutival forms of SPEAK, such as (ba)sgoo 'saying' (Jδschke 1883: 38,84-5,10828). The lack of fixity of these structures and their morphosyntactic 'exuberance' suggest that here, too, we are dealing with quotativals, not with true quotatives. It appears, then, that the Tibeto-Burman languages have undergone changes very similar to those found in Dravidian, from quotativals to true quotatives. However, given the enormous geographic separation, the development of true quotatives in southern Dravidian and in Tibeto-Burman languages such as Newari clearly must have been independent developments. How, then, should we account for these changes? 4.2.5. A Typological Motivation for the Changes Although quotatives may also be found in SVO languages (see e.g., Lord 1976 on Kwa), they seem to be much more widespread in SOV languages (Hock 1984 with references; many additional examples in Haspelmath and Kφnig 1995: 40-1, 50,170-71, 400-1,434~35).29 This is a fact that requires explanation. Steever (1987,1988) and following him, Hock (1988), suggest that quotative marking is indirectly linked to the common tendency of SOV languages to place restrictions on finite verbs—it is a way of 'shielding' direct discourse from the finite-verb restrictions, thus permitting embedded cited discourse, the most unmarked form of discourse, to remain intact. See the Reformulated Rule A. (Earlier suggestions along the same lines, but more informal ones, are found in Hock 1984.)

90 · Hans Henrich Hock

Reformulated Rule A: One finite verb is permitted per sentence. However, in conjoined structures, each of the conjoined clauses is treated as a sentence in its own right and is therefore permitted to have its own finite verb. Relativecorrelatives syntactically are conjoined structures, permitting each of the two sub-clauses to have its own finite verb. In addition, cited discourse is treated as syntactically independent and thus entitled to its own finite verb(s); it may be marked by a quotative marker to signal that it is exempt from finite-verb restrictions. Elsewhere, i.e., in structures with syntactic embedding, only one finite verb is permitted per sentence, occurring in the 'matrix' clause; other verbs have to appear in non-finite shape (participles, absolutives, infinitives, or verbal nouns). If this reformulated hypothesis is correct, then the southern Dravidian and early Indo-Aryan quotatives may have developed independently from earlier 'quotatival' structures, as parallel attempts to 'shield' the finite verb(s) of embedded direct discourse, in response to the typological 'pressure' of Rule A which would otherwise require that embedded verbs be non-finite. As I note in Hock (1996a), 'The problem is that in spite of Steever's excellent work (especially 1988), the prehistory and early history of Dravidian syntax is still quite poorly understood. Here as elsewhere, it may be hazardous to rely on relatively recent, even modern, evidence in trying to establish prehistoric Subversion.' Significantly, however, a closer examination of the early Indo-European evidence, as well as the evidence of the northern, non-literary Dravidian languages and of Tibeto-Burman languages suggests that the usual comparison of the post-Rig-Vedic postposed /^-construction with the postposed type enru of the southern Dravidian languages may be inappropriate, since this similarity may be the result of secondary developments, motivated by typological tendencies. The evidence of quotatives, therefore, does not seem to be strong enough to resolve questions on the nature of early Dravidian/Indo-Aryan interaction in South Asia.

• 5. Conclusion As I hope to have shown in this paper, the syntactic (or morphosyntactic) evidence usually cited in favor of early Indo-Aiyan subversion by Dravidian turns out to be relatively weak, upon closer examination and re-examination of Vedic and early Indo-European and of the non-literary, northern Dravidian languages (as well as of the oldest southern Dravidian). The evidence, thus reexamined, suggests a remarkable degree of similarity in the overall syntactic typology of early Dravidian and Indo-Aryan/Indo-European: Both sets of languages agree on the well-known 'strict SOV restriction that sentences have only one finite verb; but they also agree on a Rule A that permits them to accommodate subordinate structures to this restriction, by resorting either to

Typology vs. Convergence m 91

non-finite subordination or to relative-correlative structures. For similar reasons, both sets of languages tend to develop quotative markers to set off cited discourse. The similarities in overall syntactic typology between early IndoAryan and Dravidian therefore are not likely to yield evidence for subversion, or for convergence within South Asia for that matter, but evidence for subversion or convergence must be sought in finer-grained aspects of syntactic structure.30 As I hope to demonstrate in this paper, however, without proper philological foundation, any hypotheses on the typological convergence of Sanskrit and Dravidian must be considered with a healthy dose of agnosticism. The findings of this paper, if on the right track, have even broader implications. If absolutives and quotatives should indeed have developed independently in Dravidian and Indo-Aryan because of the typological 'pressure'31 of Rule A, then it must be considered possible that elsewhere, too, linguistic developments may be shared by given languages, not just by common history (including contact) or chance, but also as a consequence of typological 'pressure'. For instance, quotative constructions are found not only in Dravidian and Indo-Aryan, and in other languages of Eurasia that might, with some imagination, be said to have been in contact with these languages (even if only indirectly). They are also found in some varieties of Quechua (Coombs et al. 1976:176-79; Stark et al. 1973:247) and in Luiseno (Davis 1973:257-58). For these languages, contact with Eurasian quotative languages can be safely ruled out; but these clearly are SOV languages that otherwise have a typology very similar to that captured by Rule A; for Quechua see e.g., CerronPalomino 1976:251-73 (for a non-quotative variety, Coombs et al. 1976:17172, Stark et al. 1973:24), and for Luiseno, Davis 1973:139,180, and elsewhere. The quotative structures found in these languages are certainly best explained as independent parallel responses to a typology with severe restrictions on finite verbs, similar to that of Rule A.

I Abbreviations 1 first person 2 second person 3 third person abs. absolutive ace. accusative aor. aorist CC correlative clause CD cited discourse clit. clitic dat. dative desid. desiderative

92 · Hans Henrich Hock

dur. enemph. erg. fin. fut. gen. habit. impf. impve. incl. indie. inf. instr. loc. m mid n narr. neg. nom. non-fin. nonmasc. obi. pcle. pers.obj. pi. pres. Q RC rel. SAY sg· subj. verb.noun voc.

durative reflex of PDrav. en- 'say (so)' emphatic ergative finite future genitive habitual imperfect imperative inclusive indicative infinitive instrumental locative masculine middle voice neuter narrative negative nominative non-finite nonmasculine oblique particle personal object conjugation plural present quotative relative clause relative verb of speaking singular subjunctive verbal noun vocative

NOTES Emeneau (1974) adds an argument for lexical-syntactic subversion of Sanskrit by Dravidian in terms of the use of Skt api to mirror that of Drav. -urn. This argument is questioned in Hock (1975) and Gil (1994) (who apparently had not read Hock 1975). Another feature, recently reintroduced into the literature by Abbi (1992), is lexical 'reduplication' or iteration (Skt amredita). She shows that this feature is interesting and relevant and suggests possible Munda

- Convergence m 93 origin; but as Hock (19935) shows, its historical antecedents and its history in the non-IndoAryan languages of South Asia are not clear. Non-verbal iteration is found in many early Indo-European languages, but the Sanskrit verbal variant utpluiya—utplutya ... 'continually jumping up' first appears in late Vedic. 2. Except for remnants such as the 'fa-echo construction'. 3. To distinguish the geographical from the genetic division of Dravidian, I use the terms 'northern' and 'southern' in reference to geography and terms such as 'North Dravidian' or 'South Dravidian' for genetic classification. 4. But note that languages such as English have lost free verb-fronting, too—without the help of Dravidian subversion. 5. Since Bloch (1929), a special use of absolutives as discourse linkers (of the type Skt... uväca ... (Uy) uktvä ... '... said ... (Thus) having spoken ...') has been commonly attributed to Dravidian subversion. Recent publications (Hock 1994a, b; Migron 1993) have independently demonstrated that non-finite verb forms as discourse linkers have parallels in other Indo-European languages and may therefore be inherited from Proto-Indo-European. See also Section 4.2.2. 6. On purely typological grounds, W. Lehmann (1974) denies the existence of relative clauses in Proto-Indo-European and, interestingly, on the basis of the same typological considerations, Friedrich (1975) takes the existence of relative clauses with relative pronouns in Proto-IndoEuropean as evidence that the language cannot have been OV. As had already been noted by Watkins (1976) and as supported by examples such as the ones in (4), early Indo-European did in fact have relative-correlatives; and as argued in Hock (1989,1993a), relative-correlatives are eminently compatible with OV syntax. W. Lehmann (1993: 89-90) now comes close to accepting this view. 7. In the following discussion I ignore Dravidian relative-correlatives with the verb of the relative clause in the potential mode, since the potential is morphologically non-finite and is thus compatible both with the older and northern type and the innovated southern type. In the modern southern languages, the non-finite verb of the relative clause is compatible with the greater restrictions on finiteness in these languages. In the northern languages and the older stages of southern Dravidian, it is acceptable since Rule A (as given later in Section 3.1) permits, but does not require, finite verbs in non-embedded clauses. For non-finite predicates in these languages, see example (6) above and (9) below, as well as the discussion in Steever (1988). 8. Ironically, Klaiman (1976) has attributed the Sanskrit relative-correlative structures to Dravidian subversion. 9. Steever (1993) objects to my account, faulting me for not having expert knowledge in Dravidian linguistics. But this is really beside the point; what matters is the linguistic evidence. (Steever's bibliographical reference suggests that he was only aware of my review in Language, not of my review article [Hock 1988] which provides extensive evidence from a large variety of Dravidian languages in support of my proposal.) 10. The preceding is a paraphrase of Steever's Rule (1). Note that his Rule also specifies verbs such as *en- 'say (so)' as elements 'shielding' cited direct discourse from the consequences of the finiteness restriction of part (i) of the Rule. The issue of cited discourse and quotatives is discussed in Section 4. 11. The bracketing in this example is that of Steever (1998). 12. The example that I cited in Hock (1988) (from Ramasamy 1981) is syntactically quite complex. Example (6) has the virtue of being considerably more transparent. 13. Malto may possibly have the same kind of construction, to judge by the following passage from Das (1973: 94): ikkeno dokni enki uglir eko bediQ what-loc live-sg.2f I-gen mind go-adv.pple want-sg.3nonmasc

94 · Hans Henrich Hock The most probable interpretation of the passage is 'Where you live, there my mind wants to go'; but Das's translation 'where do you live, my hearts [sic] wants to go there' causes difficulties. 14. Note however that Konda follows the 'southern' type, by having clitic -o after the relative clause; compare example (8) above, as well as Krishnamurti (1969: 323), the source of this example. The materials available to me do not make it possible to determine the exact dividing line between the 'northern' and 'southern' relative-clause types. 15. Kittel, however, analyzes these structures as juxtaposed questions and answers. The increasing evidence for relative-correlatives in early southern Dravidian makes such a forced interpretation unnecessary. 16. Note that the 'strict OV-language' Turkish, too, has both finite (relative-correlative) [a] and non-finite strategies [b] for subordination. Compare the following examples and the fuller discussion in Hock (1993a). (Note that the Turkish finite-verb relative-correlatives appear to be used only in generalizing function.) (a) kalkip gittik rise-abs, go-past.pl.2 'Having risen we went' = 'We rose and went.' (b) [ne kadar yukuridab inerse]RC [o kadar derine girer]CC what extent high-from descend-sg.3 that extent low-to enter-sg.3 'From whatever extent high it (the pile driver) descends, to that extent low it enters.' = 'The greater the height it falls from, the deeper it goes down.' 17. The generalizing function of these structures is, of course, eminently compatible with the fact that the pronouns used in the relative clause are not only relative and interrogative, but also function as generalizing, indefinite pronouns—another difference between Dravidian and Indo-Aryan, for the latter has a separate set of relative pronouns, distinct from the interrogatives, and interrogatives tend to function as indefinites only when combined with particles (as in ka$-cit 'whoever'). Interestingly, Dravidian clitics of the type -e and -o tend to reinforce the indefinite character of the interrogative/indefinite pronouns. Could it be that the southern relative clause type with post-clausal clitic -e or -o owes its existence to re-analysis of an earlier variant of the northern type in which the indefinite character of the pronoun was reinforced by a clause-final clitic? That is, could the southern typological shift be the consequence of a simple reinterpretation? 18. Nick Masica has pointed out that in Modern Hindi, the present participles of these structures can be considered finite. However the use of present participles as conditionals goes back to Middle Indo-Aryan, a period when present participles do not yet function as finite verbs (at this point, structures with present participles as apparent main verbs are no doubt to be interpreted as having an Omitted* verb 'to be'; the entire structure ordinarily has indicative, not modal value; and the tense value commonly is past). 19. In the following, the major focus for modern southern Dravidian will be on the quotative use of en- in Tamil. The alternative use of ä(k)-, the colloquial Tamil use of various verbs and nouns without quotative marker and its parallels in other Dravidian languages (Steever 1988), the quotative constructions in other southern Dravidian languages, and the relationship between the various marking strategies for cited discourse deserve a complete investigation in their own right. For the time being, Steever (1988) remains the most comprehensive discussion. 20. For Malto, the story material in Das (1973) and Mahapatra (1979) suggests a complementary distribution between awd- (< ??) preceding cited discourse and an- (< en- 'say (so)') or clitic -ay (< ??) following cited discourse. Given the very limited nature of the texts, it is not clear to me whether these are the only patterns that are found. If they are, then Malto would seem to be quite different from the non-North Dravidian northern languages. The texts in Hahn (1911) suggest a similar distribution for Kurux, but the evidence is quite limited.

Typology vs. Convergence m 95 21. Kolami greatly favors 'redundant' structures of this type, with a combination of non-finite verb plus finite verb. The non-finite verb commonly is derived from PDrav. *en- 'say (so)', the verb which furnishes the quotative marker of, say, Tamil; but it may also be from other verb roots, such as iddeij 'tell-inf.' (Story 1, # 130). 22. The area with surviving Serial Verbs, however, extends farther and also includes Konda, as well as relics in Toda and Kota. 23. For the question of Serial Verbs in Koraga, see Section 4.2.2 below. 24. If the evidence in Das (1973) and Mahapatra (1979) is reliable, this may be the more attractive interpretation, for as noted earlier, Malto seems to have only structures with SAY either preposed or postposed to cited discourse, but not both (ignoring structures with SAY as linker). 25. The term 'syntactic domain' is chosen to include Conjoined SVCs as in [31]. 26. Given the overall evidence for a (relatively) strict OV typology in Dravidian, the question arises of how to account for SVCs. Steever (1988) suggests that the agreement of the finite verbs in SVCs is a matter of morphology, not of syntax; and this appears to be the correct analysis for the modern southern languages. For the older and northern languages, my 1988 account in terms of syntactic coordination (permitted under Rule A) seems to be more appropriate. 27. Data from Grierson (1909). 28. I apologize for the perhaps unconventional transliterations of Jδschke's Tibetan-script examples. 29. As is well known, Turkish is one of the SOV languages that use the absolutive of a verb of speaking as quotative marker [a]. Interestingly, it also has the Dravidian type of Serial Verbs with two finite verbs showing person and number agreement [bj. Note that Serial Verbs also occur in early Indo-European, best preserved in Hittite [c]. (See Hock Forthcoming b for further details.) (a) [kirn bunu yapti] diye düpünüyordum who this do-past.sg.3 Q wonder-past.sg.l was wondering who had done this.' (b) Mustafa i§e yaramiyor degildi M. work-for fit-neg.sg.3 be-neg.past.sg.3 'Mustafa was not fit for work.' (c) [nu huma]n man apedani UD-ti adanna akuwafnna tarhteni now if that-day.loc.sg. eat-inf. drink-inf. can-pl.2 na-a]t ezzalin ekuttin (KUB XIII. 4-6,17-19:11. 1-2) now = it eat-pl.2 drink-pi.2 'Now if you can eat-(and-)drink (= consume) everything on that day, eat-(and-)drink it (= consume) it. 30. At the Symposium, Nick Masica expressed concerns that 'we still have to account for the fact that in Indo-European these [SOV] characteristics were lost elsewhere, and were preserved—and further developed—only in South Asia and its environs—presumably due to the typologically supportive areal environment (not necessarily just Dravidian, of course).' The fact that the typology is, in fact, preserved and even further developed in South Asia is certainly remarkable and raises interesting questions about the extent to which contact may be able to lead to common typological retentions. But note that the difference between the South Asian languages and most of the languages of Europe results at least equally from the fact that the latter languages underwent a common shift from SOV toward verb-second and/ or SVO typology (Hock 1982a), with languages such as Farsi and Armenian taking an intermediate position. 31. In this regard, note the fact that the SOV typology of Turkish is remarkably similar to those of Dravidian and Sanskrit, with both finite (relative-correlative) and non-finite subordination

96 · Hans Hennch Hock strategies (see footnote 16), as well as with quotatives (footnote 30). Even more specific, apparently typologically driven similarities can be observed within South Asia. In addition to the use of postposed forms of son- 'say, speak' (often the absolutive nus£(n)) to mark cited discourse and of absolutives for subordination, Burushaski also has relative-correlatives with finite verbs; see example [a]. What is most remarkable is that, like the modern southern Dravidian languages, Burushaski normally requires a generalizing particle (ke) to follow the relative clause (exceptions are rare and apparently result from recent Urdu influence). (See Berger 1998.) Since, as we have seen, the postposed -o of modern southern Dravidian is a relatively recent, regional innovation, limited to the extreme south, a subversion or convergence account for this similarity between these languages, spoken at extreme ends of South Asia, is highly unlikely. Rather, it seems that in both cases we are dealing with independent phenomena, implementing an OV typology with greater restrictions on the occurrence of finite verbs than the typology of Old and modern northern Dravidian, or that of early Indo-Aryan for that matter. (a) [amin dauckuin imanimi]RC ke [ine buaa dhap ... oo$coi]CC who hunter become-pret.sg.3m pcle. he cow-gen, meat neg-eat-pres.sg.3m 'Whoever has become a hunter does not eat beef/

• REFERENCES Abbi, Anvita. 1992. Reduplication in South Asian languages: an areal, typological, and historical study. New Delhi: Allied Publishers. Andersen, Paul Kent 1980. Typological and functional aspects of word order in ASoka. General Linguistics 20.1-22. . 1982-83. Amplified sentences in Asoka. Zeitschrift für vergleichende Sprachforschung 96.1729. Berger, Hermann. 1998. Die Burushaski-Sprache von Hunza und Nager. Teil I: grammatik, Teil II: texte mit Übersetzungen. Wiesbaden: Harrassowitz. Berneker, E. 1900. Die Wortfolge in den slavischen Sprachen. Berlin: Behr. Beythan, Hermann. 1943. Praktische Grammatik der Tamilsprache. Leipzig: Harrassowitz. Bhaskararao, Peri. 1980. Konekor Gadaba: a Dravidian language. Pune: Deccan College. Bhat, D.N. S[ankar]. 1971. TheKoraga language. Linguistic survey of India series, 7. Poona (Pune): Deccan College. Bloch, Jules. 1929. Some problems of Indo-Aryan philology. Bulletin of the School of Oriental and African Studies (London) 5. 719-56. Burrow, T[homas], and S. Bhattacharya. 1953. The Parji language: a Dravidian language ofBastar Herford: Austin & Sons. . 1970. The Pengo language: Grammar, texts, and vocabulary. Oxford: Clarendon Press. Caldwell, Robert 1875 (1913).X comparative grammar of the Dravidian or South-Indian family of languages. 2nd edition 1875.3rd edition 1913, slightly revised by J.L. Wyatt and T. Ramakrishna Pillai. Reprinted 1974, New Delhi: Oriental Books Reprint Corporation. Cerron-Palomino, Rodolfo. 1976. Gramatica quechua: Junin-Huanca. Lima: Ministerio de Education. Coombs, David et al. 1976. Gramatica quechua: San Martin. Lima: Ministerio de Education. Das, A. Sisir Kumar. 1973. Structure ofMalto. Annamalainagar: Annamalai University, Department of Linguistics. Davis, John Frederick. 1973. A partial grammar of simplex and complex sentences in Luiseno. Los Angeles: UCLA Ph.D. dissertation. Disterheft, Dorothy. 1980. The syntactic development of the infinitive in Indo-European. Columbus, OH: Slavica.

Typology vs. Convergence · 97 Downing, Bruce T. 1978. Some universals of relative clause structure. Universals of human language, ed. by J.H. Greenberg, Charles A. Ferguson and Edith A. Moravcsik, 4: 375—418. Stanford, CA: University Press. Emeneau, Murray B. 1954. Linguistic prehistory of India. Proceedings of the American Philosophical Society 98. 282-92. Repr. in Emeneau 1980. . 1956. India as a linguistic area. Language 32.3-16. Repr. in Emeneau 1980. . 1961. Kolami:A Dravidian language. Annamalainagar: Annamalai University Publications in Linguistics. . 1974. The Indian linguistic area revisited. International Journal of Dravidian Linguistics 3. 92-134. Repr. in Emeneau 1980. . 1980. Language and linguistic area: Essays selected by A.S. DU. Stanford, CA: University Press. Fortson, Ben. 1994. Prolegomena to a synchronic analysis of the Hittite quotative particle. Harvard Working Papers in Linguistics 4. 37-51. Friedrich, Paul. 1975. Proto-Indo-European syntax. Butte, Montana: Journal of Indo-European Studies. Gil, David. 1994. Conjunctive operators in South-Asian languages. SALA XV: Papers from the Fifteenth South Asian Language Analysis Roundtable Conference 1993, ed. by A. Davison and P.M. Smith. Iowa City: South Asian Studies Program, University of Iowa. Gippert, Jost. 1978. Zur Syntax der infinitivischen Bildungen in den indogermanischen Sprachen. Europδische Hochschulschriften, Reihe 21, Linguistik, 3. Frankfurt: Lang. Grierson, George. 1909. Linguistic survey of India, Vol. 3, Parts J and 2. Repr. 1967, Delhi: Motilal Banarsidass. Hahn, Ferd. 191 l.Kurukh grammar. Calcutta: Bengal Secretariat Press. (Repr. Delhi: Mittal [no date].) Haspelmath, Martin and Ekkehard König (eds.) 1995. Converts in cross-linguistic perspective: structure and meaning of adverbial verb forms—Adverbial Participles, Gerunds. Berlin: Mouton de Gruyter. Herring, Susan. Forthcoming. Poeticality and word order in Old Tamil. Textual parameters in older languages, ed. by S. Herring, P. van Reenen and L. Schoesler. Amsterdam: Benjamins. Heusler, Andreas. 1931. Altisländisches Elementarbuch. Heidelberg: Winter. Hock, Hans Henrich. 1975. Substratum influence on (Rig-Vedic) Sanskrit? Studies in the Linguistic Sciences 5. 2. 76-125. . 1982a. The Sanskrit quotative: a historical and comparative study. Studies in the Linguistic Sciences 12 (2). 39-85. . 1982b. AUX-cliticization as a motivation for word order change. SLS 12:1.91-101. . 1984. (Pre-)Rig-Vedic convergence of Indo-Aryan with Dravidian? Another look at the evidence. Studies in the Linguistic Sciences 14 (1). 89-107. . 1988. Review article. Finiteness in Dravidian. (Sanford B. Steever [1988]. The serial verb formation in the Dravidian languages.) Studies in the Linguistic Sciences 18 (2). 211-31. . 1989. Conjoined we stand: Theoretical implications of Sanskrit relative clauses. Studies in the Linguistic Sciences 19 (1). 93-126. . 1993a. Reconstruction and syntactic typology: a plea for a different approach. Explanation in historical linguistics, ed. by G.W. Davis and G.K. Iverson, 105-21. Amsterdam and Philadelphia: Benjamins. . 1993b. Review of Abbi 1992. Studies in the Linguistic Sciences 23(1). 169-92. . 1994a. Discourse linkage in Sanskrit narratives with special emphasis on the story of Nala. Papers from the Fifteenth South Asian Language Analysis Roundtable 1993, ed. by A. Davison and EM. Smith, 117-39. Iowa City, IA: South Asian Studies Program. . 1994b. Narrative linkage in the Mahδbhδrata. Modem evaluation of the Mahäbhärata: Prof. R.K. Sharma felicitation volume, ed. by S.P. Narang, 295-313. Delhi: Nag Publishers.

98 · Hans Henrich Hock Hock, Hans Henrich. 1996a. Pre-Rgvedic convergence between Indo-Aryan (Sanskrit) and Dravidian? A survey of the issues and controversies. Ideology and status of Sanskrit: contributions to the history of the Sanskrit language, ed. by J.E.M. Houben, 17-58. Leiden: Brill. . 1996b. Subversion or convergence? The issue of pre-Vedic retroflexion reconsidered. Studies in the Linguistic Sciences 23 (2). 73-115. . 1997. Chronology or genre? Problems in Vedic syntax. Inside the texts—beyond the texts: new approaches to the study of the Vedas, ed. by Michael Witzel, 103-26. Harvard Oriental Series, Opera Minora, 2. Cambridge, MA: Harvard University. . Forthcoming a. Genre, discourse, and syntax in Sanskrit. Textual parameters in older languages, ed. by S. Herring, P. van Reenen, and L. Schoesler. Amsterdam: Benjamins. . Forthcoming b. Vedic eta ... staväma—subordinate, coordinate, or what? Journal of IndoEuropean Studies. Israel, M. 1979. A grammar of the Kuvi language (with texts and vocabulary). Trivandrum (Thiruvananthapuram): Dravidian Linguistics Association. Jäschke, HA. 1883. Tibetan grammar. 2nd edn. London: Trόbner. Kittel, Ferdinand. 1903. A grammar of the Kannada language. Mangalore: Basel Mission and Tract Depository. Klaiman, M.H. 1976. Correlative clauses and IE syntactic reconstruction. Papers from the parasession on diachronic syntax, 159-68. Chicago: Linguistic Society. Krishnamurti, Bh. 1969. Konda or Kübi: a Dravidian language. Hyderabad: Tribal Cultural Research & Training Institute. Kuiper, F.BJ. 1967. The genesis of a linguistic area. Indo-Iranian Journal 10. 81-102. Repr. in IntemationalJoumalof Dravidian Linguistics 3. 135-53 (1974). Lakshmi Bai, B. 1985. Some notes on correlative constructions in Dravidian. For Gordon H. Fairbanks, ed. by V.Z. Acson and R.L. Leed, 181-90. Honolulu: University of Hawaii Press. Lehmann, Thomas. 1998. Old Tamil. Steever 1998: 75-99. Lehmann, Winfred P. 1974. Proto-Indo-European syntax. Austin: University of Texas Press. . 1993. Theoretical bases of Indo-European linguistics. London and New York: Routledge. Lord, Carol. 1976. Evidence for syntactic reanalysis: From verb to complementizer in Kwa. Papers from the parasession on diachronic syntax, 179-91. Chicago: Linguistic Society. Mahapatra, B.P. 1979. Malto—an ethnosemantic study. Mysore: Central Institute of Indian Languages. Marlow, Patrick Edward. 1997. Origin and development of the Indo-Aryan quotatives and complementizers: an arealapproach. Urbana: University of Illinois Ph.D. dissertation in linguistics. Migron, Saul. 1993. Catena and climax in Vedic prose. Die Sprache 35. 71-80. Miller, D. Gary. 1975. Indo-European: VSO, SOV, SVO, or all three? Lingua 37. 31-52. Nadkarni, M.V. 1975. Bilingualism and syntactic change in Konkani. Language 51. 672-83. Natarajan, T. 1977. The language of Sangam literature and Tolkappiyam. Madurai: Publishing House. Pillai, P.V. Velayudhan. \913.EarlyMalayalamprose: a study. Trivandrum (Thiruvananthapuram): University of Kerala. Ramasamy, K. 1981. Correlative relative clauses in Tamil. Dravidian syntax, ed. by S. Agesthialingom and N. Rajasekharan Nair, 363-80. Annamalainagar: Annamalai University Publications in Linguistics, 73. Sridhar, S.N. 1981. Linguistic convergence: Indo-Aryanization of Dravidian. Lingua 53.199-220. Starke, Louisa R. and Lawrence K. Carpenter (eds.) 1973. Elquichea de Imbabura: unagramatica pedagogica. Otovalo (Ecuador): Institute Interandino de Desarollo. Steever, Sanford B. 1987. Remarks on Dravidian complementation. Studies in the Linguistic Sciences 17 (1). 103-19. . 1988. The serial verb formation in the Dravidian languages. Delhi: Motilal Banarsidass. . 1993. Analysis to synthesis: the development of complex verb morphology in the Dravidian languages. New York: Oxford University Press.

Typology vs. Convergence · 99 Steever, Sanford B. 1998. (ed.). The Dravidian languages. London/New York: Routledge. Thomason, Sarah Grey and Terrence Kaufman. 1988. Language contact, creolization, and genetic linguistics. Berkeley and Los Angeles: University of California Press. Tikkanen, Bertil. 1987. The Sanskrit gerund: a synchronic, diachronic, and typological analysis. (Studio Orientalia 62) Helsinki: Finnish Oriental Society. Tsiang-Starcevic, Sarah Yvonne. 1997. The discourse functions of subordinate structures in Classical Sanskrit narrative texts. Urbana: University of Illinois Ph.D. dissertation in linguistics. Watkins, Calvert. 1976. Towards Proto-Indo-European syntax: Problems and pseudo-problems. Papers from the parasession on diachronic syntax, 305-26. Chicago: Linguistic Society.

Where do Compound Verbs Come From? (And Where are They Going?) ι PETER EDWIN HOOK ι One of the characteristic features of the Indo-Turanian linguistic area (see Masica, this volume) is the use of grammaticalized forms of verbs such as 'give', 'leave', 'go', 'throw', etc., as auxiliaries expressing deixis, aspect and speaker attitude. Not found in ΟΙΑ this feature must have arisen independently or been borrowed from one or another group of languages in the Indo-Turanian area. Early attestations of it both in Dravidian and in Altaic suggest that its emergence and/or diffusion lie beyond the horizon of recoverability. However, it is possible to identify and measure in early ΝΙΑ a dramatic inflation in its use and to correlate differences in its frequency from language to language with differences in its function.

• 1. Introduction One of the striking features of South and Central Asian languages and one that plays an important part in Masica's demonstration of the 'Indo-Turanian' or South-and-Central Asian linguistic area is the 'compound verb'. For the moment let us define a compound verb as a sequence of two verbs AB ('polar A plus 'vector B')1 that alternates with A (the 'polar') with little or no difference2 in meaning. In this pair of examples from Hindi-Urdu the compound verb a }ata fiae alternates with the non-compound ('simple') verb ata fiae: (1) a. k3bhi.ksbhi dopefier me bhi a fata fix Sometimes afternoon in also come GO is 'Sometimes he comes in the afternoon, too' (Premchand 1966: 86).

102 · Peter Edwin Hook (1) b. rat ko jzh roz pani pi-ne ata & night Dat here daily water drink-Inf come is 'Every night he comes here to drink water' (Premchand 1966: 86). Another pair, from the Central Dravidian language Gondi: (2) a. cucjur morri ... Izkk deif taksi fiot.tur young son ...distant country go WENT 'The younger son ... went to a faraway place' (Grierson 1919: 489). (2) b. ox tec&i spno: dadana muftis taLtur he rising his (own) father.Gen near went 'He got up and went to his father' (ibid.: 489-90). For Central Asia this pair of examples from Mongolian (Dagur dialect): (3) a. cua e tere gajire-de kure-ji ire-ji a:-bie: soldiers that country-Dat arriv-ing COM-ING be-Fut '...soldiers will be arriving in that place...' (Martin 1960: 98-9, sect. 125). (3) b. buni gere-t.e: kure-bie: tomorrow home-Dat arrive-Fut Ί will arrive home tomorrow...' (ibid.: 103, sect. 145). For Northeast Asia this pair of examples from Japanese: (4) a. sake-wa boku-ηα non.de fimaLta rice.wine-Top I-Nom drink PUTAWAY Ί drank up the sake' (adapted from Mizutani 1977:188). (4) b. sake-wa boku-ga non.da rice.wine-Top I-Nom drank Ί drank the sake.' A fifth pair, from the northeast Caucasian language Avar represents the extreme westward extension of the Central Asian lobe of Masica's Indo>Turanian linguistic area:

(5) a. o'a sun ana fire go.out WENT 'The fire went out' (Charachidze 1981:127). (5) b. έ'α suna fire went.out 'The fire went out' (ibid.: 127). The key phrase in the definition (and a troublesome one) is 'with little or mo difference in meaning'. It is necessary, however. Nearly every language in the

Where do Compound Verbs Come From? · 103 world employs sequences of two (or more) verbs to express modulations of single actions or events. Engl. get going, keep moving] Jp. hataraite iru 'is slaving'; Ru. bud'et zavtrakat* 'will breakfast'; Ch. kanjian 'see'; Ger. habe gesehen 'have seen'; Fr. // risque de pleuvoir 'It threatens to rain'; etc. If we allow all these sequences to qualify as instances of the 'compound verb', we will not be able to use the category to identify and distinguish a linguistic area. (See [in this volume] Masica's Pitfall # 23!). Many writers on the compound verb, even recent ones, neglect this point: Chelliah 1999; Katre 1976; Pokharel 1999; and Vale 1948. While a complete exposition of how the compound verb is to be defined is outside the scope of this paper,3 in almost every language in the Indo-Turanian linguistic area it is formed of the conjunctive participial form (aka 'convert)') of a main (or 'polar') verb accompanied by an auxiliary (or 'vector') verb which bears whatever desinence is required by the syntactic environment. Typical vectors include items homophonous with full lexical verbs meaning GO, COME, BECOME, GIVE, TAKE, THROW, PUT, PUT AWAY, INSERT, RISE, FALL, EMERGE, LEAVE, RELEASE, STRIKE and perhaps a dozen or two more. In a particular Central or South Asian language some of these items have also developed other specific grammatical uses that are simultaneously present, serving as passives, inceptives, modals, permissives, causatives, etc.4 Main-auxiliary sequences in these functions are not to be regarded as compound verbs. Related to the preceding is a problem that can best be understood by a specific illustration of it: Shina is one of the very few Indo-Aryan languages that does not have a compound verb of the kind that satisfies the condition that the auxiliary (or 'vector') must alternate with its absence. This is a fact of some significance, especially if one wants to plot the geographical distribution of the compound verb in South Asian languages. However, it happens that in the Shina of Gilgit the past tense forms of almost all polyvalent predicates derive (Was this under the influence of western Tibetan?) from a form of the conjunctive participle plus a (phonologically reduced) inflected form of the verb (6) go [.ε ga:s go[.er g-ee bodo kf*-ee -gas home went.lsg home go-CP much eat-CP -WENTlsg Ί went home. Going home I ate a lot.' If we accept sequences like kheegas 'I ate' as instances of the compound verb then we will have a peculiar situation in which almost all polyvalent verbs in Gilgiti Shina must be compound in the past tense but no verb may be compound in any other tense. With a condition on the definition of the compound verb requiring 'alternation with absence' we have a better chance to avoid the inclusion of formations like kheegas which resemble compound verbs in form but actually function as ordinary indicators of tense.

104 · Peter Edwin Hook

For some individual languages (Hindi-Urdu, for example) much effort has been put into developing a definition of 'compound verb'. For other languages, the category is still not very well defined. For the moment let us assume that the compound verb phenomenon (or at least available descriptions of it) in the various languages of South and Central Asia have enough in common to allow general agreement with Masica's inclusion of it as one of the features distinguishing South Asian and Central Asian languages (including Korean and Japanese) as a group from their neighbors to the west and east which lack it. (Less certain is its status in Chinese and Southeast Asia.5) Accepting the compound verb as a legitimate distinctive feature of the 'Indo-Turanian' Linguistic Area immediately raises questions of history: Where did it come from? Did it arise independently in several places? Or did it diffuse from one of the members of the Indo-Turanian linguistic area into the others? These are the questions I plan to address in this paper and, if not answer, at least formulate more precisely.

• 2. Origins • 2.1. Evidence from Frequency Counts Twenty-some years ago in an analysis of the present distribution of the compound verb in South Asia (Hook 1977). I attempted to approach these diachronic questions by synchronic means. On the assumption that those languages in which the compound verb phenomenon is most intense are likely to be those in which it has resided longest, I used the LSI as a data bank from which to extract information about compound verb frequencies in several hundred languages and dialects of the Indo-Aryan family. Numerical analysis of the LSI returns placed the center of intensity along the Gangetic plain from Bihar in the east to western Punjab, with peaks near Delhi and Rawalpindi. In view of remarks by S.K. Chatterjee (1926:1050) and others (Southworth 1974: 208, 312; Pofizka 1967-69: 65) who have suggested the possibility of a Dravidian origin for the Indo-Aryan compound verb this is a surprising result. The pole of its intensity lies far from the present-day Dravidian heartland. In fact the frequency of the Indo-Aryan compound verb is least in precisely those places that are either the ones most likely to represent Dravidian substrates or are the areas most likely to have been influenced by neighboring Dravidian languages: In Marwar, Gujarat, Maharashtra and Orissa the compound verb occurs no more than a third as often as it does in returns from the Gangetic Plain and Punjab. Because the territories included in the Madras Presidency were outside the scope of the Survey, I cannot provide parallel data from the Dravidian family as a whole. However, numerical analysis of the 50 or so LSI returns that are

Where do Compound Verbs Come From? · 105

AFGHANISTAN TIBET

'.2,4 ..2,4

INDIA

.4,7

.2,4

Map 1: The Distribution of Compound Verbs in Indo-Aryan Based on Data in the LSI.

106 · Peter Edwin Hook

available for northern and central Dravidian indicates a center of intensity in the Rajmahal hills of SE Bihar (see Map 2). Although parallel data are needed to complete the picture for southern and south-central Dravidian, from these limited data it seems that the range of frequency of the compound verb in the Dravidian languages spoken in the central and eastern parts of India mirrors the range of its frequency in the immediately surrounding Indo-Aryan languages. It is already clear from both frequency counts and the kinds of verbs that have been grammaticalized as compound verb auxiliaries or vectors that Munda is unlikely to be the source (See Hook 1991a). I am unable at present to evaluate Tibeto-Burman as a possible source except to note the sharp drop in frequency found again in those places that are either the ones most likely to represent Tibeto-Burman substrates or are the areas most likely to have been influenced by neighboring Tibeto-Burman languages: Assam, Bengal, Nepal, Kumaon, Garhwal, Himachal, Kashmir, Baltistan, Hunza, and Gilgit. Jn the absence of a corpus of transcribed, glossed texts from the area (to say nothing of a Linguistic Survey of Central Asia) it is very difficult to determine the frequency of the compound verb for a sufficiently large sample of Altaic and adjacent languages. Working with what is available I have found a percentage of 2.5 in Avar texts and a percentage of 9.5 in Mongolian texts (Dagur dialect). These figures for the Avar and the Mongolian compound verb are significantly lower than those for many Indo-Aryan languages such as HindiUrdu. See discussion below in Section 4 ('Relativization') of Stages. What can we conclude then from the present-day distribution of the compound verb in the languages of South Asia? Not much. The observed pattern of distribution may have resulted from an innovation in Indo-Aryan which later diffused into Dravidian, Mundan, and nearby Tibeto-Burman languages. Or the compound verb may have entered Indo-Aryan from ancient Dravidian, TibetoBurman, Mundan, or even other languages which were once spoken in the Indus Valley and upper Gangetic Plain but which have now disappeared. Or it may have entered South Asia from some Central Asian language and subsequently developed on its own, especially in Indo-Aryan. In fact, it is hard to imagine an origin scenario that is excluded by the LSI data shown in Maps 1 and 2! • 2.2. Evidence from Texts The earliest attested forms of the compound verb in the language families that belong to the Indo-Turanian Linguistic Area are indicated in the table below. Examples of these early attestations follow: Tfcble 1: Earliest Attestations of the Compound Verb in South and Central Asia Family

Language

Indo-Aryan Dravidian Altaic

Pali Tamil Orkhon Turkic

Text Jδtaka Tales Sangam Poetry Inscriptions

Date Early centuries CE? 0-300 CE? 750 CE

Location Sri Lanka Tamil Nadu Mongolia

Where do Compound Verbs Come From? · 107

AFGHANISTAN TIBET PAKISTAN

INDIA

Map 2: The Distribution of Compound Verbs in Dravidian Based on Data in the LSI.

108 · Peter Edwin Hook

(7) Pδli: dairuni afidrutva: dggim kstva: dssssti firewood bringing fire.Acc making GIVE.Fut3sg 'He'll bring wood and make a fire (for you)' (Trenckner 1879: 77). (8) Tamil: marantan-am turant-a kaz. forgot-lpl LEFT-adj seed 'The seed which we forgot...' (Narrinai 172.2 as quoted by Steever 1988: 466). (9) a. Orkhon: kyl agin qoji jil-qa jiti jegirmi-k& uc.di Kul Tigin Sheep Year-in seven twenty-in flew Trince Kul passed away on the 17th day of the Sheep Year' (TekinKTNE:!). (9) b. [qaya i\t jil onujic aj alti otuz-qa uc.a bar.di kagan Dog Year tenth mont six twenty-in fly WENT 4 [The Kagan] died the 26th day of the 10th month of the Dog Year' (TekinBKS:10). What can we conclude from these attestations? More than 1,500 years ago the compound verb was already widespread in South and Central Asia. This suggests either that we are dealing with independent innovations in two or more places in the Indo-Turanian linguistic area or that the compound verb is a feature whose diffusion from one part of the area to the other (and we cannot say which part was which!) predates the emergence of writing systems by many centuries. However, one significant piece of negative evidence has come down to us from ancient times: The compound verb does not occur in the Vedic corpus.7 If it existed in the spoken language in 1500 BCE it is difficult to imagine why it should have been excluded from the sacred texts. One must not draw the same conclusion from its absence in the later language. Unambiguous instances of the compound verb fail to occur anywhere in 'classical' Sanskrit, not in medieval, not in late medieval, not even in modern texts written by naive authors whose Sanskrit is otherwise thoroughly contaminated by later languages (viz., girvδiiapadamaίjari). Perhaps because of their exposure to texts from earlier periods, non-native (second-language) writers and speakers when writing or speaking in Sanskrit shun this particular feature of their mother tongues. This absence is paralleled by the absence of the compound verb in the forms of Indian English known to me. In short we cannot conclude anything about the existence or non-existence of the compound verb in the vernacular languages of any post-Vedic epoch by looking at Sanskrit (or for that matter Indian English) texts.

• 3. Flux Inflation I have borrowed the term 'inflation' from cosmology to denote a sudden increase in the text frequency of the compound verb. There is direct evidence

Where do Compound Verbs Come From? · 109

for inflation in Marwari (Hook 1993) and Marathi (Hook and Pardeshi, ms.) and indirect evidence for it in Bangla (Zbavitel 1970). In Marwari prose texts from about 1600 CE to the present, there is a 1200 percent increase. In Bangla texts we find a similar increase during roughly the same period of time, and in Marathi, an increase from 1300 to the present of again about 1200 percent (although the absolute percentages are smaller). 20 1816141210-

1300

1400

1500

1600

1700

1800

1900

2000

Figure 1: Line Graph Showing the Increase in Frequency of Use of the Marwari, Bangla, and Marathi Compound Verb from 1300 CE until the Present

With the possible exception of the North Dravidian language Kurukh there is no nearby non-Indo-Aryan language in which the compound verb is used as frequently as it is in modern Bangla and modern Marwari. With its small number of socially marginalized speakers, Kurukh seems an unlikely trigger of flux inflation. A further difficulty is that the compound verb in Kurukh is formally divergent from what we find in Indo-Aryan languages. In Kurukh the functional equivalent of the compound verb is actually a 'pair verb' (parnyjglagol) in which both elements (i.e., both the polar and the vector verb) are fully inflected for the tense and show agreement in person, gender, and number with the subject of the sentence. (Pair verbs—a sub-class of what Steever [1988] terms 'serial verb formations'—are also found in South Munda [Aze 1973; Zide and Anderson (this volume)] and Tibeto-Burman languages [Weidert and Subba 1985: 74ff] spoken in the same vicinity and, further to the west, in Brahui [Bray], Northern Tajik [Kerimova 1964], and, even further to the west, in Russian):

110 · Peter Edwin Hook Table 2: Pair Verbs (parnyje glagofy) in Non-Indie Languages Kurukh:

em kanum kanum ij?-kan I walking walking stop-MlsgPst GAVE-MlsgPst Ί stopped while walking' (Arun 1992: 105). Gorum (South Munda): mi ne~ga?~ru ne-la?-ru I Isg-eat-Pst Isg-STRIKE-Pst Ί ate vigorously.' (as cited by Zide & Anderson from Aze 1972). Limbu: pe%-e£&$ a-bir-et&e go-2du!mper.lsgBen Aug-GIVE-2duImper.lsgBen 'You (two) go for me!' (Weidert and Subba 1985: 316). Brahui: te:na: huUL· kask-as hina:s-as selfs horse died-Nsg WENT.Nsg '...my horse ... died' (Steever 1988: 106). North Tajik me.guzar-am me.rau-am pass.by-lsg GO-lsg Ί will pass by' (Kerimova 1964: 280). Russian vzfa'-la i sloma-U s'ib'e 'ruku TAKE.PRV-PstFsg and break.P-PstFsg self.Dat hand.Acc 'She went and broke her arm' (Zakharyin, personal communication).

To my knowledge no Indo-Aryan language has pair verbs. If my information is correct then it is even less likely that influence from Kurukh is behind the recent flux inflation in central Indo-Aryan compound verb systems. Moreover, there has been no invasion from outside India during the past four centuries which could explain the sudden expansion shown in Fig. L The conquest of northern India by the Mughals under Babar occurred before the 17th century began. Judging from Emperor Babar's journal the compound verb was infrequently used in his native Chagatay. Early Marwari annals show a higher CV frequency than the baabar-naamaa. While we may not ever be able to identify the original source of the compound verb in South and Central Asian languages, the source of the recent inflation in its use must be attributed to internal developments in Indo-Aryan.

• 4. Relativization When counts are made of compound verbs in connected texts one finds a striking range of variation in the ratio of compound verbs to total verb forms. Some variation may be due to differences in genre. For instance, the compound verb in belle-lettristic prose texts written in Hindi occurs roughly three to four times more frequently than it does in the Hindi translation of the Indian Constitution (Hook 1999). However, limiting ourselves to counts made of texts whose intention is primarily narrative (short stories, novels, folktales, biographical anecdotes and historical annals) we still find a 20-fold variation across languages in rates of CV occurrence:

Where do Compound Verbs Come From? · 111 Table 3: Variation in Ratios of Compound Verbs to Total Verbs in Connected Texts Family

Language

Text type

Text dates

Indo-Aryan

Hindi-Urdu Marwari Gujarati Bangla Gujarati Bhitrauti Marathi Middle Marwari Kashmiri Dagur dialect Chagatay Avar

stories stories stories prose stories folktale stories annals folktales narrative narrative folktale

since 1950 since 1950 since 1950 20th cent 20th cent 20th cent 20th cent 17th cent 19th-20th cent 20th cent 16th cent 20th cent

Mongolic Altaic NE Caucasian

Counts

CV(%)

7 2 2 4 3 1 2 2 3 1 1 1

15-20 13-18 14-16 10-13

8-9 8 3-4 1-3 0.5-1.5

8-9 1-2 2.5

It is not possible to ascribe the same repertoire of functions to a structure which varies from language to language by 20 times or more in its rate of occurrence. In recent work I have shown that the expression of perfective aspect, identified by a number of Indologists (Pof izka, Nespital, Hook) as one of the chief functions of the compound verb in Hindi-Urdu, is not one of its functions in Kashmiri (Hook and Koul 1992), in Marathi (Hook 1993), or in Middle Manvari (Hook 1991b). This functional variability brings us back to the problem of defining the compound verb in such a way that we do not include too much or too little territory in the relevant areal isogloss: a purely formal definition leads to overinclusion, while a functional one based on, say, the expression of perfective aspect results in a pattern of non-contiguous patches in South Asia and who knows what in Central Asia. In earlier work I have shown that this problem can be mitigated (if not solved) by relativizing the functionality of a given language's compound verb system to its frequency in connected texts. In order to function as an exponent of perfective aspect, the text frequency of the compound verb must exceed 10 percent. The 10 percent ratio of use marks a threshold in the evolution of a language's compound verb system from what is (primarily) a means of lexical enrichment to something that can be treated on a par with inflectional categories. Table 4 summarizes the differences between compound verbs of low text frequency (Stage I systems) and those of high text frequency (Stage III systems—I am reserving the term 'Stage ΙΓ for intermediate cases). Table 4: Correctable Characteristics of Compound Verb Systems Properties Text frequency Productivity

Stage I systems

< 10% Not every polar participates

Stage III systems

> 10% All polars participate Table 4 continued

112 · Peter Edwin Hook Table 4 continued Specialization Degree of bleaching Sensitive to context? (a) Negative contexts (b) Finiteness (c) Derivational suffixes Functions (a) Anterior clauses (b) Phasal complements (c) Fear-clauses

Wider variety of vectors Vectors must be factors Insensitive (a) CVs are freely negatable (b) Non-finite CVs freely occur (c) CVs may be internal CVs primarily extend lexicon (a) Do not favor CVs (b) CVs occur with phasals (c) Do not favor CVs

Narrower set of choices Vectors may be empty Highly sensitive (a) Negation disfavors CVs (b) Few non-finite CVs (c) CVs never internal aspectual; discoursal (a) All but require CVs (b) Phasals require SVs (c) Strongly favor CVs

In Table 5,1 apply these criteria for distinguishing Stage I and Stage III systems to Kashmiri and Hindi-Urdu, languages at two extremes in the presentday range of compound verb frequency in Indo-Aryan. Tfeble 5: Comparison of Compound Verb Systems in Kashmiri and Hindi-Urdu Properties Text frequency Productivity Specialization Degree of bleaching Sensitive to context? (a) Negative contexts (b) Finiteness (c) Derivational suffixes Functions (a) Anterior clauses (b) Phasal complements (c) Fear-clauses

Stage I system: Kashmiri

Stage III system: Hindi

In Kashmiri less than 2 percent Many polars don't participate Many vectors; none dominant

Varies: 15% to 20% All polars participate ^ de, le predominate de de, asja: are good Highly sensitive Negation disfavors CVs

relatively insensitive CVs are freely negated W9sjithpi3th is good tshun'ino.'LhS bi oiz fraVif* CVs primarily extend lexicon CVs express perfective mi bf5fi ml· m*o.7i Ramsaji. . . me lekm muf-sepefile mera:p9fOsL· before rose my neighbor... tietm IxSil· gatfinf is good ir as tsi ma: tsatakl· kul faίt- tu pec ka:t ns idlA3en>i he begins read.NonP read.Prfctv sentence 'He begins to read the sentence.'

114 · Peter Edwin Hook

(13) Hindi-Urdu: faq us-ko log-ne /(*logja:-ne) cold him-Dat feel-Inf (feel GO-Inf) began 'He began to feel the cold.' The ability of a language's compound verbs to appear as phasal complements is thus an indication that they do not have perfective aspectual value. (14) Kashmiri: p'W ffeVm tas tsamfi lasrJit* gatf-inf drops began him.Dat skin.Dat stick GO-Inf 'The (rain) drops began to stick to his skin' (Shauq 1996: 66). While the inability of a language's compound verbs to appear as phasal complements can be taken as satisfying a necessary condition for identifying perfective aspect; this is not a sufficient condition. That is, ability of a compound verb to occur as the complement of a phasal is direct evidence against its having perfective aspectual value; but inability to occur as a complement cannot in itself be taken as a reason for ascribing such value. Tear-clauses': In languages in which aspectual contrasts are expressed by the alternation of perfective and non-perfective forms, clauses which are complements of predicates expressing fear or anxiety strongly prefer perfective forms (Forsyth 1970: 258-61): (15) Russian: maV bAJaUsi kaLbi ji'jo sin n>i mother was.afraid lest her son NEG got.sick.Prfctv 'The mother was afraid her son might get sick' (Forsyth 1970:259). A strong preference for compound verbs to appear in clausal complements of predicates expressing fear or anxiety is positive evidence that they have perfective aspectual value as one of their functions. Using a database consisting of the text of Premchand's novel godam8 it is possible to establish that fear-clauses in Hindi-Urdu show just such a preference. In Premchand's novel relevant contexts occur 25 times. Out of these 25, 19 have the compound verb: (16) Hindi: msb i phjikjif Q:S tsi ma: tsai-akf1 kul us.Dat worry was you NEG cut-Fut2sg tree 'We were afraid you might cut down the tree' (O.N. Koul, p.c.). A function which appears to be presupposed by the presence of perfectivity is one I term 'relative completion' or 'anteriority'. If the compound verb in a language expresses perfective aspect then it obligatorily expresses anteriority of one action relative to a later one in structures in which event-times are explicitly contrasted. Such constructions include 'By the time that X occurred, Υ had (already) occurred' and 'X occurred before Υ occurred'. (21) a. Hindi: Anteriority: expressed in biclausal structures: J3b tdk turn jsfia pafioce tdb te/c msb ne use yjdt de when till you here arrived then till I Erg him letter give άήα t^a GIVEN was 'By the time you got here I had given him the letter.' (21) b. Hindi: Anteriority: expressed in a single complex clause: tumhare jdfia pdfiocne se pefile fii: mx ne use χοί your here arrive. Inf from before Emp I Erg him letter de dija ί*α give GIVEN was 'Before you got here I had (already) given him the letter.' If we reverse the temporal positions of the two events we find that preferences for compound versus non-compound verb forms are also reversed.

116 · Peter Edwin Hook

(22) a. Hindi: Anteriority: expressed in biclausal structures: JQb tek mab ne

use χζί

άηα tob tek turn jdfi

pafi c

when till I Erg him letter gave then till you here arrive goje fa ? GONE were 'By the time I gave him the letter had you gotten here?' (22) b. Hindi: Anteriority: expressed in a single complex clause: mere use γ$ί

dene

se

pefile

fii:

turn jofia pofioc

my him letter give.Inf from before Emp you here arrive gQJe fa ? GONE were 'Before I gave him the letter had you (already) gotten here?' It is a matter of empirical record that Stage HI systems also include anteriority or relative completion as one of their functions. The converse, however, does not hold. That is, there are languages in which the compound verb is used to express anteriority but in which compound verbs fail the tests for perfective aspect. Languages of this (intermediate) type are classified as 'Stage IF languages. Subordinate clauses introduced by conjunctions having the meaning 'until' usually9 present the compound verb in Stage HI systems. I regard a preference for compound verbs in this environment as strong if not decisive evidence for a system's being Stage III. Viewed cross-linguistically, compound verb systems show a hierarchy of functions: (23) Stage I ('lexical') > Stage II ('relative tense') > Stage HI ('perfective aspect') This hierarchy can be read either synchronically or diachronically. That is, Stage III systems include the relative tense function of Stage II and have evolved from earlier Stage II systems; while a Stage I system cannot enter Stage III without also developing the relative tense function characteristic of Stage II. Neither traditional grammarians nor most contemporary linguists are conscious of the importance of determining frequency of the use of each member of an alternating set of forms. But by developing a set of multiply correctable diagnostic properties, it becomes possible to make a global evaluation of a language's compound verb system even in the absence of some particular pieces of information. If our diagnostic tests are sound, we should be able to use them to discover sequentiality in the functional evolution of compound verb systems.

• 5. Indo-Aryan Let us begin by taking two more Indo-Aryan languages: Panjabi and Marathi. From the LSI data we already know that the compound verb is much more

Where do Compound Verbs Come From? · 117

frequent in Panjabi than it is in Marathi. Is that difference in flux consistent with the other differences discussed above? Table 6: Characteristics of the Compound Verb System in Panjabi Properties

Values in Panjabi (examples from Tej K. Bhatia, p.c.)

Text frequency Productivity Specialization Degree of bleaching

High (> 10%) Nearly every polar participates. Big 4? (GO, GIVE, TAKE, LEAVE?) High: Sequences of homophonous polar and vector exist: xot de ditto: sL· (letter give GAVE had)'... had given the letter.' Sequences of antonymous polar and vector also exist: o etfe a-gija-sL· (he here come WENT had) 'He had come here.' Sensitive to context? Yes. Most negative contexts favor non-compound verbs. (a) Negative contexts buatfa- fia-Si: mar gijo: versus koL· fio^L· nzL· msroi old elephant die WENT any elephant NEG died 'The old elephant died.' versus 'No elephant died.' (b) Non-finiteness disfavors CV: o m· nw xst de (*de) -ke vapas gijahe us to letter give (GIVE)-ing back went 'He gave us the letter and went back.' Functions Relative completion (a): yes. Perfective aspect (bed): yes. (a) Anterior clauses jod tok tusi etfe pujjoe mab ο nu: χζί de ditto sL· when till you here arrived LErg him Dat letter give GIVE had 'By the time you got here I had given him the letter.' (b) 'until'-clauses jsd tok ο tuanu χ& na de dave etfie miazar ksro when till he you letter NEG give GIVE here waiting do 'Wait here until he gives you the letter.' (c) Phasal complements: ssi' mifa-ija: tfa-ri (*?khale-ri) lagge we sweets eat Inf eat TAKE-Inf began 'We began to eat up the candy.' (d) Fear-clauses favor compound verbs: so: nw φί si: ke tussi ο nu: e xat de de-oge us to fear was that you him to this letter give GIVE-Fut2sgM 'We were afiaid thai you might give him the letter.'

Comparing Table 6 with Table 5, we can see that the properties of the Panjabi compound verb system are nearly identical to those of the Hindi-Urdu system. From Table 7 (below) we can see that the properties of the Marathi system contrast with the Panjabi system in a manner that is parallel to the contrast between Kashmiri and Hindi-Urdu.

• 6. Dravidian The four major Dravidian languages (Tamil, Malayalam, Kannada, and Telugu) happen not to be well represented in the LSI. Being literary languages

118 · Peter Edwin Hook Table 7: Characteristics of the Compound Verb System in Marathi Properties

Values in Marathi (examples from Prashant Pardeshi)

Text frequency Productivity Specialization Degree of bleaching

Sensitive to context? (a) Negative contexts (b) Non-finiteness does not disfavor CVs Functions: (a) Anterior clauses (b) 'until'-clauses (c) Phasal complements (d) Fear-clauses

Low (3 to 4%) Not every polar participates: je 'come', mil 'get'» etc· Less advanced than in Hindi-Urdu. (See Hopper 1991:27) Lower: Sequences of homophonous polar and vector absent: *deun dihfioiQ (give GIVEN had) '...had given....' Sequences of antonymous polar and vector are also absent jeun gela- (come WENT) *in sense of 'came' Much less than in Hindi-Urdu or Panjabi. accept CVs: dorva-dza u^o^un de-u nzje door open GIVE-Inf not.should '(You) shouldn't open the door.' to dsnxL'dza: u£&^un de.un ba er gelo: he door open GIVING outside went 'He opened the door and went out.' Relative completion (a): no. Perfective aspect (bed): no. tu (its) jei parjznt mi tja-lapair dih fiotQ you here come until I.Erg him.Dat letter given had 'By the time you got here I had given him the letter.' to pair (lifi-un) dei psrjont (ifa) famb he letter (write-and) give until (here) wait 'Wait here until he gives you the letter.' can be compound: ti $nne vacfun ffe-u lagli she food serve TAKE-Inf began (Chirmule 1967: 60) 'She began to dish out the food.' Do not favor compound verbs: (Prashant Pardeshi, p.c.) tu t'a-la pytr de-fil zfi canha:la b^iti υα·(9ί fioti you him letter give-Fut2sg such us.Dat fear feeling was 'We were afraid that you might give him the letter.'

with their own individual writing systems, they have escaped having representative texts published in Roman transcription with morph-by-morph glossing. This makes detailed study and evaluation of their compound verb systems difficult for the non-Dravidianist. For this reason I have had to rely on elicitation and questionnaires. Even with this methodological limitation it seems that the compound verb in at least one of them (Tamil) may be at Stage III. Table 8 lists the diagnostic characteristics of the compound verb in Tamil. Table 8: Characteristics of the Compound Verb System in Tamil Properties

Text frequency Productivity

Values in Tamil (examples are from Raj Raja, p.c., or from S.N. Vendadesan, p.c.)

High: CVs can be made from all polars. Table 8 continued

Where do Compound Verbs Come From? · 119 Table 8 continued Specialization Degree of bleaching Sensitive to context? (a) Negative contexts

Heavy preference for using υίφυ 'release' as vector. High: nan uandui υιφιηχη (I come RELEASE) Til come.' Yes. irdndui ui[[sdui => irokkzuillzi (Vendadesan, p.c.) die-RELEASED 'died' => died.NEG 'didn't die' ? (insufficient data) Relative completion (a): yes. Perfective aspect (bed): yes? ni kacfdam ko died.NEG 'didn't die'But see: maric'i-po:-kaii(f (forget-GO-NEG) 'Don't forget!' (p. 216).

(6) Finiteness

Compound verbs can be made from compound verbs. ap-es-ej (stop PUT PUT) 'Stop!' (K&G 1985: 216).

Functions

Relative completion (a): yes. Perfective aspect (bed): no.

(a) Anterior clauses

nuuvu uaccei-sanki ataniki uttaram icci-ve:s.7iu you come until him.Dat letter given PUT.IsgPsl 'By the time you got here I had given him the letter.' αίαφ uttaram icce:-uaraku ikkac^a κ/^φ he letter give until here be 'Wait here until he gives you the letter.'

(b) 'until'-clauses

Table 9 continued

120 · Peter Edwin Hook Table 9 continued (c) Phasal complements

(d) Fear-clauses do not show preference for compound verbs

may sometimes contain compound verbs: αίαφ tsampi uejzcjam modalu.pe[(e:(ju he kill PUT began 'He began boring me to death.' nuuuu ataniki Luttaram is-taw ani you him.Dat this letter give-Fut Quotative fear-suffered. Isg. Ί was afraid that you might give him the letter.'

In its repertory of vectors, in its ability to deploy them in non-finite contexts, in the possibility of using them in clausal complements of the phasal verb 'begin', and in the absence of a preference for them in the complements of verbs of fear, the system of compound verbs in Telugu appears not to have a Stage III system. Since it does show a preference for using them in the expression of relative completion, I will place it provisionally among the Stage II languages.

• 7. Central Asia The Altaic languages of Central Asia seem to have a less developed compound verb than the major Dravidian languages in South Asia. Let us look at Table 10 for Kyrgyz, the language in which the compound verb as a typological trait seems to be more intense than it is in other members of the Altaic familyHowever, Kyrgyz seems not to have a Stage III system. Table 10: Characteristics of the Compound Verb System in Kyrgyz Properties

Values in Kyrgyz (examples from Judaxin 1965)

Text frequency Productivity Specialization Degree of bleaching

No limit on polars that may participate in CVs 7 Fairly high: Antonymous sequences possible? kir.ip bar.di '(He) enter WENT (= entered quickly)'

Sensitive to context?

? (insufficient data)

Finiteness: CVs occur as conjunctive participles

ajdarbek ookonu oakir-t-ip al-ip ur-dur.gan Ajdarbek Chokon call-CAUS-CP TAKE-CP kill-CAUSed 'Ajdarbek had Chokon summoned and put to death.' Compound verbs can be made from compound verbs. kulap kele tafta.di (fall GO THREW) 'He almost fell.'

Functions

CVs seem not to express perfective aspect. ? (insufficient data)

Phasal complements Fear-clauses do not seem to favor CVs (ex. from Judaxin 1965: 869)

bir balaket kil-di-bi dep Mu-jm one mistake do-PST-? saying fear-lsg Ί am afraid he might misbehave/

Where do Compound Verbs Come From? · 121

Chagatay, the pre-modern literary register of what is now called Uzbek, was brought to South Asia by Babar and his armies in the 16th century. It seems to be a Stage I system: Table 11: Characteristics of the Compound Verb System in Chagatay Properties

Values in Chagatay (examples from Eckmann 1966)

Text frequency Productivity Specialization Degree of bleaching

Very low (< 1%) Only those polars that express movements participate. Many different factor verbs; very few vectors in use. Low: All CVs made from factors (except for her GIVE): jayi qaca berdi '(The) enemy flee GAVE (= fled quickly)' Sensitive to context? ? (insufficient data) Functions CVs primarily express direction of motion. (a) Phasal complements: ? (insufficient data) (b) Fear-clauses do not seem to favor CVs (ex. from Eckmann 1966: 210) qorqaram kim afkara bol-mayaj bu sirr.i pinfianim fear.lsg that revealed become-NEG this secret hidden. Isg Ί fear this hidden secret of mine might come to light...'.

The modern language (Uzbek) also seems to have a Stage I system (see Sjoberg 1963). This is puzzling in that the Tajiki dialects spoken in Uzbekistan have what appears to be a Stage HI system. It suggests that compound verb is a trait that may be borrowed from source Language A by borrowing Language B and then it may undergo a more rapid evolution in the borrowing language B than in the source A. Table 12: Characteristics of the Compound Verb System in Northern Tajik Properties

Values in N. Tajiki (examples from Kenmova 1964)

Text frequency Productivity Specialization Degree of bleaching

? Most polars participate? ? (Many different vectors in use.) Quite high: Meanings of polar and vector need not converge. xona ra* fan foda raft (room lit become WENT) '...room lit up.' Sequences of antonymous polar and vector exist: man dar bora-i masala-i ozodi-i zanon istzda meguzaram I on subjcct-of problem-of freedom-of woman stand PASS.Isg Ί will stop on the question of women's liberation.' Sensitive to context? ? (Seem to be not so sensitive to context as Hindi's CVs.) (b) Non-finiteness sur£ fuda rafta rujmol-ro girift SOfD farmida does not disfavor Soro embarrassed red become GOING handkerchief-Ace took CVs (Kerimova 'Embarrassed and turning red Soro took the handkerchief.' 1964: 236) Table 12 continued

122 · Peter Edwin Hook Table 12 continued Functions

Expression of perfective aspect among functions of CV.

(α) Until-clauses favor CVs (Kerimova 1964: 280)

to yud-aton jak omada na-raved in koc bud na-mefauad until self-your once come NEG-GO this job done NEG-be 'This job won't get done until you come once yourself.'

(b) Fear-clauses also kof.kL· α χοηα buromada na-meraft seem to favor lest out.of house exit NEG-GO '(Let us pray) that he doesn't go out of the house.' CVs (Kerimova 1964: 283)

• 8. Northeast Asia Diagnostic tests indicate that the Altaic10 languages of Northeast Asia have Stage II systems. The clearest evidence comes from Korean: Table 13: Characteristics of the Compound Verb System in Korean (Some examples as cited by E.K. Guseva 1961:83-90)

Properties Text frequency

?

Productivity

Most polars participate?

Specialization

Many vectors: pzri- THROW, Λ suffix, for example, can indicate past time reference or indecisive/indefinite modality in non-past time. The following examples from Chalise (pp. 45, 48) exemplify this usage: (11) a. aamaa-ce çáá-daa ken caa-puq~ji mother-ERG I-DAT rice eat-permit-PST 'Mother permitted me to eat rice', b. ijaa-laa aale dihm-ri mu-ji I-GEN brother house-LOC be-Indefinite 'My brother may or may not be at home'. In Tamang, the past tense expresses an event, and past Stative situations are normally expressed in terms of non-past tense. Examples: (12) a. the khren-baa-laa he hunger-NPT-PRF 'He has been hungry [stative]'.

Typology of Verbal Categories m 139 b. the khren-ji he hunger-PST 'He became hungry [eventive]'. Tamang verbs also encode purely modal suffixes like < -sye > and < nem > which, according to Chalise (p. 43) 'always indicate future potentiality and inference of past event respectively'. For example: (13) a. çáá-ni-ce cutaamthe-daasyet-sye I-PL(excl)-ERG this matter he-DAT tell-Potential 'We will tell him about this matter', b. aale-ceq ethurj-nem brother-ERG milk drink-Inference 'Brother has drunk the milk (There was milk in the glass but now it is empty)'. The prominence of mood in Tamang can also be seen in the pervasive uses of epistemic and deontic notions of judgment, evidentials, interrogative, imperative, request, etc., which are indicated by verbal suffixes as well as by case markers. Limbu, on the other hand, has a complex agreement system where verbs are marked for person and number, and the conjugations involve transitive, intransitive and reflexive verbs. The involvement of a specific person is also determined by co-occurring morphemes which may indicate the person as an agent, subject or patient. In such a verbal scenario, one would expect a predominant system of tense-aspect distinction in the language. This however is not borne out by facts when we consider that Limbu lacks the category of nonpast and a large number of verbs are not marked for tense. The following examples are from van Driem (1987: 78,82): (14) a. We (dual) will give it to him/her. a - b i - 0 - s - u - 0 - 0- 0 1 - give - NPT - dA - 3P - SP - i - PRF b. He will give it to her. /He gave it to her. 0 - 0 - pi.r - 0 - U - 0 - 0 3 - sAS - give - NPT/PT - 3P - sP - PF c. Where are you going? a.tto.ke - be.k - 0 - 0 - pa? where2 - go - NPT - sPS - IPRF In (14a) the first person morpheme < a- > co-occurs with the third person patient morpheme < -u > and this shows that the first person is the actant. In (14 b) the third person is marked and thus has an agentive role, but the nonpreterite/preterite (NPT/PT) distinction is not overtly marked on the verb.

140 · Tej R. Kansakar

The example (14 c) however is an intransitive conjugation which, according to van Driem (1987:80) 'inherently disallows more than one person marker. The second person involvement denoted by < ke- > in such a form is one of subjecthood'. The grammar of Limbu also incorporates a wide range of aspects and aspectivizers which gives one the impression that the language is aspectprominent. It may however be possible to analyze the aspectual differences in Limbu in conjunction with the modal distinctions such as optative, conditional, irrealis, interrogative, imperative, etc., that are available in the language. The following examples from van Driem (ibid.: 71) illustrate the alternation in the verb form kamma? 'to grow attached' when mood suffixes are attached to it. (15) a. ko.co.kam-å 'The dog has grown attached.' b. ko.co.kam-la?ba 'The dog will probably grow attached.' c. mirikkam-b! 'May he develop an attachment.' The paucity of data related to the modal category in T-B languages may pose problems to the typologist. One obvious problem is the emphasis given to tense-aspect descriptions and the minimal treatment of modal distinctions. The current practice of recognizing the markings on the finite and non-finite verbs as past/non-past or perfect/imperfect may actually reflect an irrealisrealis modal distinction in the T-B languages of Nepal. This bias in the presentation and analysis of data needs to be avoided if we are to obtain a better understanding of the nature and function of the modal category occurring in these languages. As noted earlier, Newar verbal affixes are used freely without reference to any definite tense, where the past conjunct suffix < -aa > is used for both perfective and proximate future: (16) a. ji-ú mhiga: chan-ta khan-aa I-ERG yesterday you-DAT see-PC Ί saw you yesterday', b. ji aa che-e won-e ten-aa I now house-LOC go-NPC begin-PC Ί am about to go home now'. The formal contrasts in finite past < -a > and non-past < -e > however are maintained in modal expressions. (17) a. wamhiga: hal-a ju-i he-ERG yesterday bring-PD be-NPD 'He probably brought it yesterday', b. wa kanhae won-e phu he tomorrow go-NPC may 'He may leave tomorrow'.

Typology of Verbal Categones m 141

Malla (1985:76) notes that 'as a verb-final language, Newar auxiliary verbs carry temporal and aspectual meanings. But tense and aspect are often combined in one and the same auxiliary. Tense and modality are also generally combined in one and the same auxiliary verb.' Some of the modal auxiliaries listed by him can be reclassified into their lexical and grammaticalized meanings.

(18)

Lexical meaning of modal auxiliaries

Modality

Grammaticalized meaning of modal

pha-ye ji-ye da-ye

'to be able' ability 'to be permitted' permission 'to be/become' possibility

won-e phu won-e jyu da-yaa con-a

kaa-ye

'to take'

possession

dhandaa kaa-ye

cho-ye 'to send' bi-ye won-e

'to send'

volitional 1

yaan-aa cho-ye

'to give' 'to go'

volitional 2 resultative 1

dha-yaa bi-ye thatheju: won-a

wo-ye caa-ye laa-yc

'to come' 'to feel' 'to snatch, to spread, to heal' 'to see' 'to need'

sensation 1 sensation 2 resultative 2

so-ye maal-e

lumanaa wol-a to caa-ye thatheyaa-ye laataeffort, attempt yaan-aa so-ye necessity jaebili maal-e

'may go' 'can go' (evid. It happens that) khan-e dal-a 'came into view' 'to worry, to be anxious'; thö: kaa-ye 'to be drunk' "to initiate action' 'to propose action' 'It turned out this way* 'came to remember' 'to feel anger' 'happened to do it like this' 'attempt to do' 'to be under pollution of childbirth; du: maal-e 'to be under pollution of a death in the family'

The prominence of mood in T-B is perhaps best exemplified in the greater number of modal distinctions as compared to tense and aspect distinctions. The two important concepts of evidentiality and volitionality are amply expressed in epistemic and deontic moods which are more clearly distinguished in T-B languages than in I-A languages where modal categories tend to merge together. In Nepali, for example, the modal auxiliaries sak-cha, par-cha and hol-aa express ambiguity in mood: (19) a. u bholi aaun-u sak-cha he tomorrow come-NPST able-am (i) Perhaps he will come tomorrow/(ii) He is permitted to come tomorrow.

142 · Tej R. Kansakar

b. u aaphis-ma hun-u par-cha he office-LOC be-NPST is-am (i) I am certain that he is in his office/(ii) He is obliged to be in his office. c. us-le lekh-e-ko hun-cha he-ERG write-PST be-am (i) He has it in writing (as you will see)/ (ii) What he has written is acceptable. We may contrast this situation with that of Newar in which the epistemic notions of possibility, probability, necessity, certainty, etc., and the deontic notions of permission, obligation, requirement, etc., are clearly distinguished. Epistemic notions: Modality (20) a. wa-he won-i: certainty with reduplication of the verb and emphatic marker go-EMPH go-NPD He will certainly go. b. bwä-laa bon-a certainty of an action with evidential particle, but read-EVIDread-PD implies doubtful result. He evidently read. c. yaa-ye jyu:-saa possibility with conditional marker do-NPCbe-CND If it can be done. d. won-i: laa-saa probability, but hopefully the action will not take place go-NPD occur-CND Probably he will go. e. cha won-e maa: necessity, compulsion you go-NPC must You must go. Deontic notions a. yaa-sääjyu permission with concessive marker do-CONS be (You) can do it (if you wish). b. cha-ä thalhe yaa-saa jyu suggestive, friendly advice with conditional marker you-ERG like this do-CND be (Will be good) if you can do it like this. c. wa-ä dhaa-sää restrictive with concessive marker he-ERG say-CONS Although he said.... d. dha-yaa bi: maal-i requirement to achieve a certain result tell-PC give need-NPD (You) need to tell (him/them)

Typology of Verbal Categories · 143

Additional mood markers include dhhathe kha: (definiteness = I confirm that...); khä chu dhaa-saa (factual = the fact of the matter is); fraru(rather = choice, preference); kha-tu(reservation = yes, but...). As pointed out earlier, the notions of evidentiality of an action/event or volitionality of the actant is far more important in T-B than the notion of transitivity which generally plays a central role in describing the formal grammatical relations of sentence structure. The relevance of grammatical relations as a descriptive device in all languages has been questioned by Bhat (1991), and it is quite possible that T-B languages require 'semantic relations' like the volitional actor or causer, animate patient (affected individual) and location denoted directly by case markers. These languages also require 'pragmatic relations' like topic and focus which are expressed either by word order or by certain case marking conventions. In T-B these two types of relations are kept distinct and the abstract entities like grammatical relations are redundant to relate semantic roles to their morphemic representations. Another interesting point that needs to be noted here is that factive complements can occur only with non-volitional (mostly epistemic) predicates expressed by verbs like loman-e 'to forget', wo-ye 'to come', kha-ye 'to be true', maal-e 'to be necessary', etc. The non-factive complements, on the other hand, can occur only with volitional (mostly deontic) predicates like dhaa-ye 'to say', nen-e 'to ask', khyaa-ye 'to frighten, to threaten', etc. Bhat (1998:28) notes that 'this distinction also cuts across the transitive-intransitive distinction', as can be seen in the following examples. (21) a. wa-ä dhaa-gu khä loman-a he-ERG say-NOM (factive) matter forget-PD (I) forgot what he said, b. wa wo-e-gu yaa: he come-NPC-NOM (factive) do He comes (habitually). (22) a. jiwa-yaala daa-e dhakaa khyaan-aa I-ERG he-DAT beat-NPC (now factive) COMP threaten-PC I threatened him that I would beat him. b. ji won-e-ta so-yaa I go-NPC-PURP (now factive) try-PC I tried to go. Bhat (1998:28) also raises the question of whether 'a split ergativity system' can be used as a label to describe ergative-accusative relations in T-B. He advocates the view that since the notion of volitionality is more crucial than transitivity, 'we may have to regard T-B languages as neither ergative nor accusative, but as active'.

144 · Tej R. Kansakar

m 3.2. The Position of Dhimal Our present knowledge of Dhimal, a T-B language spoken in the south and southeastern Nepal, indicates that it is quite similar to I-A languages with tense-aspect distinctions. The tense markers for finite and non-finite verbs can be seen in the following examples which also illustrate Dhimal as a pronominalized language. (23) a. kaa urn caang-kaa I rice eat-PRS.l.sg. I eat rice. b. naa urn caa-fii-naa you rice eat-PST-2.sg. You ate rice. c. kaa urn caa-khaa I rice eat-PRG.l.sg. I am eating rice. d. waa urn caa-gaai-waa/caa-doq-gaai he rice eat-PRG-3.sg He was eating rice. Khatiwada (1999) also includes other finite verbs like past habitual, hypothetical tense, and non-finite verbs with perfective participle marker < kaa >, conjunctive participle < -teq >, temporal participle < -laau > and iterative participle < -kaataaq >: (24) a. ghaa-ghaa nhaa-fii play-PHl.pl.-PST We used to play. b. ghaa-dhaa—ç—kaa play-HYPT-FUT-l.sg. (I) would have played. c. kaa urn caa-kaa fii-ghaa I rice eat-PERFbe-PST.l.sg. I have eaten rice. d. waa dhaawaa phe?-teij haate haane-fii he cloth wash-CP market go-PST. 3.sg. Having washed the clothes he went to the market. e. haate haane-laau kaa-waa sehe qdusu-ghaa market go-TP he-DAT visit-PST.l.sg. While going to the market I visited him. f. urn caa-kaataang caa-kaataarj balla waa ghaai-fii rice eat-ITR. eat-ITR hardly he satisfy-PST3.sg. Eating rice continuously he was hardly satisfied.

Typology of Verbal Categories · 145

On the basis of this data, we may safely conclude that Dhimal is different from most of the other T-B languages that are spoken in Nepal. As pointed out above, T-B languages in general are characterized by modal distinctions where verbs function pragmatically as active or inactive and do not follow any strict grammatical relations with the subject/agent or direct/indirect object. The position of Dhimal in this respect is interesting as it is not yet certain what aspects of its grammar are of genetic origin and what are the results of language contact influences from the Maithili-Bhojpuri-Avadhi group of I-A languages of Nepal Terai. • 3.3. Prominence of Aspect in Indo-Aryan Languages The prominence of aspect is derived from Indo-European, but this feature has been traditionally described as tense-based. The distinction in these languages is not a temporal one of past/non-past but an aspectual one of perfective/imperfective. The I-A tense and mood are expressed by adverbials and are not marked on the verbs. The official language of the nation Nepali represents typical aspectual contrasts in I-A: (25) Simple vs. progressive a. ma bhaat khaan chü I rice eat am I eat rice. b. ma bhaat khaan-dai chü I rice eat-PRG am I am eating rice. (26) Present vs. present perfect a. ma dekh chü I see am I see. b. mai-le us-laai dekh-e I-ERG he-DAT see-PRF I have seen him. (27) Perfect vs.imperfect a. us-le cithi lekh-yo he-ERG letter write-PRF He wrote a letter. b. u cithi likhi-raheko thi-yo he letter write-PRG be-PRF He was writing a letter. c. us-le cithi lekh-in cha he-ERG letter write-FUTam He will write a letter.

146 · Tej R. Kansakar

The grammatical category of 'perfect' has various kinds of ambiguous interpretations. It is defined both as a temporal or durative kind, typically applying to verb forms under the heading of tense and sometimes under aspect. For example, in Maithili the contrast between present perfect and plu-perfect or past perfect are expressed as follows: (28) a. hamaar paer-mE cot laag-alae-ch I:GEN leg-LOC hurt occur-PST3.PRS-Aux I have hurt my leg (It is still painful). b. hama paer-mE cot laag-al I leg-LOC hurt occur-PST I had hurt my leg (last week). c. raam likh-ne rah-at Ram write-PRF Aux-FUT.3.sg.nh. Ram will have written. In perfective aspect a contrast is typically seen as a whole, regardless of the time contrasts that may be a part of it. The imperfective or non-perfective aspect, on the other hand, typically draws attention to what Bhat refers to as 'the internal time-structuring of the situation or the incomplete or durative aspect of the situation'. The Maithili examples in (28) have two contrasting suffixes < -ae-ch > and < -al >, which denote present perfect and past perfect meanings respectively. Both (28 a,b) express past events, but the former represents an event of current relevance, while the latter has no relevance to the present. In (28 c) the future perfect is marked with the perfective suffix < ne > on the main verb and the future marker < -at > on the auxiliary to show the aspectual distinction. Maithili also has a series of imperfective aspectual distinctions to denote simple present, progressive/habitual past which show a prominence of tense-aspect link in the language. (29) a. raam likh-ait ch-al Ram write-IPRF Aux-PST.3.sg.nh Ram was writing/Ram used to write. b. raam likh-ait ai-ch Ram write-IPRF Aux-PRS.3.nh Ram writes. c. raam likh-ait rah-at Ram write-IPRF Aux-Fut.3.nh Ram will be writing. The perfective/imperfective aspect in Nepali can also be marked differently but apparently with the same meaning. The examples in (30) and (31) may thus have ambiguous implications:

Typology of Verbal Categories m 147

(30) a. us-le khaa-e-ko cha he-ERG eat-PRF am He has eaten. b. us-le khaa-i- sak-yo he-ERG eat-PRF-PST He has eaten. (31) a. u khaan-dai cha he eat-PRG is He is eating, b. u khaai-rahe cha he eat-PRG is He is eating. Examples (30 a, b) are both perfective regardless of possible time differences in the completion of the action. Examples (31 a, b), on the other hand, are both imperfective or durative aspect but the distinction here maybe more semantic/pragmatic in terms of a speech situation with direct participation of speaker and patient. While both (31 a, b) can be used in direct conversation, only (31 b) is acceptable in reported speech. If you were to say Ί saw him eating', the progressive marker < -dai > in (31 a) cannot be used. A grammatical description most often needs to be supplemented by contextual/pragmatic considerations of this kind to express its appropriate meaning. • 3.4. Prominence of Tense in Austro-Asia tic/Munda and Dravidian Languages The prominence of tense in A-A/M and D languages is fairly easy to demonstrate. In these languages, the tense markings occur in both finite and nonfinite verb forms. Among the A-A/M languages, Satar/Santhali was the only one recognized as spoken in Nepal. It now turns out that there are three other Munda languages: Kharia, Mundari and Koruwa spoken by minority groups of speakers settled in parts of the Terai region. In the recently identified Kharia language spoken by less than 100 families in Jhapa and Morang districts of southeastern Nepal, there is person agreement but no number and gender agreement, and all verbs inflect for tense. The language is pronominalized but has no ergative or accusative case. (32) a. irjpe nog-irj

Iriceeat-PRS.l.sg. I eat rice, b. ampe nog-em you rice eat-PRS.2.sg. You eat rice.

148 · Tej R. Kansakar

c. haa kaarpenog-e he rice eat-PRS.3.sg. He eats rice. (33) a. iqpenokh-oi

Iriceeat-PST.l.sg. I ate rice. b. ampenokh-op you rice eat-PST.2.sg. You ate rice. c. haarj kaarpenokkh-o he rice eat-PST.3.sg He ate rice. (34) a. iijnog-eq leat-FUT.l.sg. I will eat. b. aam ud-e you drink-FUT.2.sg. You will drink. c. amud-pe you drink-FUT.2.du. You (dual) will drink. d. hankar johaa-ye he watch-FUT.3.sg. He will watch. The non-finite verbs like participles and causatives also inflect for tense, as can be seen in the following examples: (35) a. han konsel kupru-te pe ob-nokh-o that woman man-DATrice CAUS-eat-PST.3.sg. That woman caused the man to eat rice. b. he konsel-ki kupru-te pe ob-nokh-o these woman-PLUman-DAT rice CAUS-eat-PST3.pl. These women fed rice to the man. c. mon kirog tola-te diyaar-kon kimir erj-ki one tiger village-LOC enter-PTP forest return-PST A tiger having entered the village, returned to the forest. A similar situation exists in another Munda language Santhali which has a series of tense-aspect and person markers. Ebert (1996: 7) points out that in Santhali 'tense markers, animate object infixes and the subject marker (suffixed to the word that precedes the verb) are retained'. She further notes that the subordinate clauses in Munda languages are 'minimally reduced' with sequential marker < -kaate > and the finite marker < -aa > on the final verb. The following example is from Ebert (1996: 8):

Typology of Verbal Categones · 149

(36) ädä abuk- bälä-kate ona gando-rä-y- ä durup-en-khan-da so wash-enter-SEQ that stool-LOC-(y)-he sit down-PT:itr-TEMP-TOP daka-utu-i agu-ad-e-a rice-curry-she bring-PT:tr-3.sg.O-FIN 'So when he had washed (his hands) and entered and sat down on the stool, she brought him rice and curry'. In the Dravidian language Kurux/Urδo, simple verbs and periphrastic verb sequences are fully inflected for tense, as can be seen in the following examples (Ebert 1996: 6,7): (37) Present Perfect a. masc. fern. 1 .sg esk-an bee 'e-d-an isk-in bee 'een have broken' 2.sg esk-ai bee'e-d-ai isk-ii bee'ii-dii 'You have broken' 3.sg esk-as bee'e-d-as isk-ii biiii 'He/she has broken' b. ii goll-as bar-c-as xane ema-ge amb-na ra 'a~c-a this landlord-m come-PST-m after we-DAT leave-VN be-PST-n After this landlord came here, we had to give up. These examples show that tense markers are obligatory in the verbs of Dravidian languages where both finite and non-finite verbs are marked for either deictic or non-deictic tense.

• 4. Correlated Grammaticalization Processes I have suggested, in the previous sections, a typological classification of the languages of Nepal on the basis of tense, aspect and mood-prominent language types. I have attempted to demonstrate how these distinctions are generally linked to particular language groups of various genetic origins such as the I-A, T-B, A-A/M, and D languages spoken in Nepal. I have also tried to provide principled arguments on how these characteristics and tendencies can give varying degrees of prominence to the verbal category, and it is useful to correlate these variables to the idealized language types established earlier as a typological norm. The present section will attempt to describe how prominence in a verbal category is subject to a greater degree of grammaticalization in that category. The process of grammaticalization in a way expands the scope of typology by relating synchronic language structure to language function and language change. It is thus worthwhile to investigate how a verbal category such as mood can become tense- and aspect-oriented, and how an aspect-prominent language can gradually recognize temporal distinctions. In these processes, fall lexical items become grammatical morphemes which may lead to changes in linguistic form and grammatical functions. Bhat (1999:102)

150 · Tej R. Kansakar

however contends that the prominent category is generally regarded as the most relevant category and other category distinctions as less relevant. The tendency of a tense-prominent language therefore will be to change perfect and progressive forms into past and present forms, and of aspect-prominent languages to change the same into perfective and imperfective aspect forms through grammaticalization. The languages of Nepal also display some of these tendencies as can be seen in the processes relating to the perfective aspect, the paths along which this is grammaticalized into temporal or aspectual category in different languages. There are also other areas of grammar such as nominalization and case-marking systems such as 'split ergativity' where tense, aspect and mood markers have been grammaticalized into inflectional (or derivational) auxiliaries or affixes. • 4.1. The Concept of Perfect The notion of perfect has been a controversial one in which various linguists have offered different interpretations. Although perfect is generally regarded as a part of the tense system (of past, present, future and plu-perfect), Comrie (1985: 78) takes the view that the difference between past and perfect is not one of tense. While both past and perfect refer to an event prior to the moment of speaking, the crucial difference is one of 'current relevance' of the perfect form. In the two sentences: lost my watch' and have lost my watch' the additional meaning of current relevance applies only to the second sentence and not to the first sentence with a simple past form. Other linguists like Salkie (1989) however argue that perfect is essentially tense-based and the notion of current relevance is merely one of implication. Another view is that of Givon (1984) and Brinton (1988) who assert that perfect is both temporal as well as aspectual. The Maithili examples (28 a, b) above clearly show aspectual distinctions, but in mood-prominent languages the distinction between past and perfect is not strictly maintained, and the aspectual categories of present perfect and past perfect are linked to the modal notions of relevance and irrelevance. Consider the following Newar examples: (38) a. wa-ä baakhä col-a he-ERG story write-PD He wrote/has written a story, b. wa~ä baakhä co-e dhun-ka-la he-ERG story write-NPC PRF-CAUS-PD He had written a story. In the following examples, the verbs in (39a, b) are both marked with the irrealis suffix < -i >, but one denotes the meaning of relevance (i.e., an event that would occur in the future), and the other denotes irrelevance as an event that would or could have taken place but did not.

Typology of Verbal Categories m 151 (39) a. wa fyaahää wo-i-gu kha: he back come-irrealis-NOM is (assertive) He would have come back, b. wa fyaahää wo-i tini he back come-irrealis will be He will be coming back. The use of an assertive auxiliary in (39a) and a dynamic or movement verb in (39b) provides the basis for the development of resultatives and completives which contrast with the realis notions of imperfective, progressive or the present. Bybee, Perkins and Pagliuca (1994) argue in this connection that past can be regarded as a more grammaticalized version of the perfective, and present, on the other hand, is considered by them to be a specialized imperfective, i.e., one which is restricted to the present moment. The generalization that one can derive from these various viewpoints is that the notion of perfect (or plu-perfect) is interpreted as a temporal category in tenseprominent languages, as an aspectual category in aspect-prominent languages, and as a modal category in mood-prominent languages. The connotations that can be given to this verbal category is summarized by Bhat (1998:19) as follows: 1. Temporal notion: past event with current (present) relevance. 2. Aspectual notion: completed (perfective) event with continuing (imperfective) relevance. 3. Modality notion: realis event with irrealis relevance (something that needs to be done). • 4.2. Ergativity split The T-B languages of Nepal and Tibet have often been cited as languages with 'split ergativity' patterns (DeLancey 1981; Givon 1984; Watters 1973, 1975, 1999). The split refers to the inconsistent uses of ergative in their case marking morphology in relation to verb agreement patterns. This means that case markings can be expressed as ergative in certain arguments and as nominative-accusative in other arguments. The split most commonly discussed is on 'aspectual split' where the tense-aspect in clauses manifest ergative type with perfective aspect or past tense, or nominative-accusative type case marking with imperfective aspect present or future tense. Sherpa and Newar of Nepal are cited as typical examples. Some other languages of the T-B family that have been described do have a similar pattern of aspectual split, but Watters (1999: 25) denies that 'an aspectual split is typologically characteristic of the language family'. In the following examples from Newar and Sherpa the three concepts symbolized as 'S' represent a single argument of intransitive sentences; the agent/controller of action and ' the patient represent the two

152 · Tej R. Kansakar

obligatory arguments of transitive sentences. The Sherpa examples in (41 a,b) below are from Sch ttelndreyer (1973: 76). (40) a. khicaa sit-a dog die-PD The dog died. b. manukh-a khicaa syaat-a man-ERG dog kill-PD The man killed the dog. (41) a. 'tangbo 'ngunlaaq yeTi-yi wot-u-yi-nok many years ago Yeti-one exist-Ims-cont-PD Many years ago there was a Yeti, b. tamaa yeTi 'ti-ki 'mi teriq set-up-i-nokq then Yeti that-Ag man all kill-Imp-cont-PD Then the Yeti killed all the men. DeLancey (1981, 1988) has argued for a different split-ergative typology that is relevant for the so-called 'pronominalizing' languages of T-B. In these languages the 1st and 2nd persons (referred to by DeLancey as speech-actparticipants or SAPs) take priority over 3rd person and express the 'natural viewpoint' of a speech act. In a verbal agreement system the agreement is with a SAP either as an agent or patient, and with patient if both agent and patient are SAPs. In these agreement patterns the ergative marking occurs where a transitive verb agreement is with a patient, and nominative type agreement elsewhere. Watters (1999:27) calls this kind of split a 'person-related' split-ergative as opposed to the familiar aspectual split. He regards this pattern in Kham as 'an outgrowth of an earlier direction marking system, not of tense-aspect or transitivity'. One part of the pattern can be seen in his Kham examples: (42) a. ijaa-0 laa: çáá-saih-ke I leopard Isg.-kill-ASP I killed a leopard. b. m:-0 laa: na-saih-ke you leopard 2sg.-kill-ASP You killed a leopard. c. no-e laa: saih-ke-o he-ERG leopard kill-ASP-3sg. He killed a leopard. DeLancey (1981:627) refers to this split ergative as the 'empathy hierarchy' pattern which is governed by notions like 'animacy, agentivity, topicality, salience or empathy'. DeLancey also describes a third type of split casemarking pattern, namely the 'active/stative split' where the case-marking varies according to the intentional or unintentional action expressed by an

Typology of Verbal Categones m 153 intransitive verb. In Tibetan, however, some verbs are predicated for volition (control) or involution (inadvertance) as can be seen in (43 a, b) for Tibetan and (44 a, b) for Newar. It may be noted that when the subject is non-first person, the auxiliary verb in Tibetan does not encode a contrast in volition (43 c). The Tibetan examples are from DeLancey (1987: 57): (43) a. rja-s dkaryol bcag-pa-yin I-ERG cup break-PRF/VOL I broke the cup (deliberately). b. qa-s dkaryol bcag-sorj I-ERG cup break-PRF/NVOL I broke the cup (inadvertently). c. kho-s deb der bzag-sorj he-ERG book there put-PRF He put the book there. (44) a. ji-ú kayo tachyaan-aa I-ERG cup break-PC/VOL b. ji-ú kayo tachyaat-a I-ERG cup break-PD/NVOL c. wasaphu: ana tal-a he-ERG book there put-PD The Tibetan and Newar examples in (43) and (44) demonstrate the importance of the notions of volitionality and control in the grammars of these two languages. In the Tibetan examples, in particular, the selection of the auxiliary is clearly sensitive to volition, not control. In Newar, the volition/non-volition distinction reflects a first person vs. non-first person marking on the verb. DeLancey refers to this as the 'empathy hierarchy' pattern where the first person outranks the second person, and the two together outranks the third person.

• 5. Conclusions and Implications for Further Research • 5.1 General Conclusions A general survey of representative data from the four families of languages spoken in Nepal provides a basis for a typological classification that identifies or differentiates these languages. This study is based on the hypothesis that the prominence in verbal categories of tense, aspect and mood can be typologically significant and the generalizations that we can arrive at may relate to languages belonging to a single family or spread across genetic boundaries. The processes of convergence and divergence also play key roles in language

154 · Tej R. Kansakar

change and the paths of grammaticalization may also bring about a shift in areal or typological focus. The Newar and Tamang languages, for example, can be traced to the proto-nucleus that is shared by the Gurung-Tamang group of western Himalayan Bodish section. This hypothesis also allows us to make a further claim that the morphosyntax of these languages from the earliest times had a simple verb agreement system based primarily on modal distinctions of experiencer-evidential or agent-patient grammatical relations which contrast most obviously with the complex inflectional morphology and syntax in the eastern Himalayan languages of the Rai-Kiranti group. This fact shows that clear typological distinctions can also exist within languages of the same genetic stock. As suggested earlier, these distinctions may be the result of different grammaticalization processes that have taken place in language groups. In aspect-prominent languages, the development of past/non-past or a three-way past-present-future distinction can usually be traced to earlier perfectiveimperfective aspectual categories. The mood-prominent languages, on the other hand, generally develop a future/non-future distinction from an earlier realis-irrealis modal categories where the past and the present represent the realis, and future the irrealis frame of reference. The mood-prominent languages also tend to develop quite an elaborate system of past and future tenses in terms of modal concepts like evidentiality, supposition, judgment, etc. In the Tamang example (lOb) 'Mother may have kept it' the suffix < laa > denotes the supposition as to what the agent may have done, while the suffix < -baa > indicates the speaker's perception of the event. Many other verbal affixes in Tamang like < -cil-ji > (see 11 a, b) can denote tense-aspect as well as modality distinctions, and hence tend to be grammatically ambiguous. The paths of grammaticalization, therefore, are quite crucial in understanding how aspect-prominent and mood-prominent languages incorporate features of a temporal system. While such changes are widely attested in the I-A and TB languages of Nepal, we should not lose sight of the genetically inherited features which often coexist with borrowed features in language systems. • 5.2. Implications for Further Research 5.2.2 Bias in Grammatical Descriptions

In this paper I have argued that prominence in the verbal categories of T-A-M can serve to establish typological distinctions among languages within or across family boundaries. The prominence given to a verbal category can usually be determined by 'the degree of grammaticalization, systematicity, obligatoriness and pervasiveness that is shown by that category in the languages concerned' (Bhat 1998:12). I have shown that in tense-prominent languages, such as D and A-A/M, tense markers occur pervasively in both finite and nonfinite verb forms. The T-B and I-A languages, on the other hand, pose

Typology of Verbal Categories · 155

problems in that most published grammars of these languages are generally biased towards tense. In other words, the aspectual or modal categories have often been disguised by extensive uses of tense-based terminology. What are described as past/non-past or past-present-future distinctions should have been expressed as perfective/imperfective, and the tense/aspect-based description of T-B languages may actually represent an irrealis-realis modal distinction that is more natural to languages of this family. For example, Tamang as a mood-prominent language can express several distinctions in evidentiality and judgement: -ro -ci/ji -taa-laa -baa -paal-baa-laa

reported event observed event unconfirmed event universal Stative, experiential

-paal-baa-laa -jiml-dm -mu-ji -laa-paa -gen

Stative, experiential inferential probability habitual immediacy

Newar also adds the suffix < -the > to denote doubt, and < -ha > to express a reported event or heresy, and < -ta > to show that the speaker presumes a proposition to be correct but is uncertain about it. These kinds of modal distinctions are usually described within the framework of tense-aspect categories. We need to avoid such tendencies so that individual grammars can reflect inherent correlations between syntax, semantics and pragmatics. 5.2.2. Areal Typology Each language family in Nepal generally consists of distinct groups of languages. For example, the group defined as the Himalayan languages of west Bodish includes Gurung, Tamang, Thakali, Magar, Ghale, Manangba and Kaike (commonly known as the Gurung group or the Tamang group). The Rai-Kiranti languages of western Nepal also form two distinct groups, while the I-A languages such as Maithili-Bhojpuri-Tharu-Avadhi are classified as the southern Terai group. These language groups offer immense scope for areal typological studies in which comparative morphological and syntactic data could establish a continuum of language contact and diversification. The Newar language, for example, has traditionally been classified as a central Himalayan language under the Tibetan group, but interpretations have differed whether the language is closer to the western or the eastern branch of Himalayan languages (cf. van Driem 1993; Kansakar 1998,1999). DeLancey (1992) has also linked Newar to the Tibetan group as a source of its conjunctdisjunct verb conjugation, but it is more likely that the conjunct-disjunct system is a reflex of a more comprehensive areal spread and not confined to a limited group of languages. Since speakers of many T-B languages under these groups have drifted away from their traditional homelands, it will be necessary to determine which languages share comparable features that are due either to common genetic origin or the result of language contact.

156 · Tej R. Kansakar 5.2.3. Diachronie Typology

The study of typology in relation to historical syntax is an extremely difficult area for which there is no reliable methodology. However it is possible to develop certain modest approaches that can explain syntactic changes in a typological perspective. For the languages of Nepal we could define areas such as word order or morpheme order typology, agentivity and ergativity, case-marking systems, verb conjugation, relativization and causativization strategies, etc., which languages share in varying degrees. These similarities may be due to the fact that two languages or a group of languages are genetically related or there have been direct or indirect borrowings from other languages in areal contact. Since the languages of Nepal in general have SOV structure, one would expect to find a close connection between morphological typology and word order typology. Greenberg (1966) shows that there is a close connection between the SOV word order and the existence of a case system. This case system can be seen very pervasively in the languages of Nepal with nominative-accusative or agent-patient distinction which correlates the S and O. The processes of grammaticalization and historical change, however, are most often explained in terms of data on verb morphology. DeLancey (1985: 379-80), for example, refers to deictic or motion verbs in Newar like wa- 'go', wal- 'come', bo- 'fly' which are first syntacticized in their use as specifiers of deictic orientation for other verbs, e.g., nyaase wz-walk-go = 'to walk away', jonaa wo/-carry-come = 'to bring'. Later, these syntacticized morphemes agglutinate into the verb complex where free morphemes are reduced to bound morphemes. The Newar verb w0/-was historically bimorphemic but has followed the cycles of syntacticization, morphologization and relexicalization. In other words, the loss of wa-/vvfl-/distinction as productive morphemes led to the second and third stages of change. In Kansakar (1994: 81-97) I had also discussed the grammaticalization processes in Classical and Modern Newar in terms of verb serialization, auxiliarization and morphologization for verbs of location/position (e.g., con- 'sit/stay', tad- 'put'), motion/direction (e.g., yen- 'take away', hal- 'bring'), and verbs of experience/perception (e.g., caal- 'feel', taal- 'hear'). Synchronically too, there are numerous variations in the syntactic devices employed by a group of languages of common genetic origin (like the TamangGurung-Manangba group) or which coexist in close geographical proximity like Dolakha-Pahari Newar and the Rai-Kiranti languages of eastern Nepal. Historically, however, we cannot make any judgments on the phonological, morphological or syntactic changes in these languages without in-depth diachronic studies, nor can we attempt a reconstruction of the syntax of the Proto-TGM or Proto-Newar. It would therefore be premature to conclude that the complex verb inflectional morphology in Dolakha Newar, for example, represents the Proto-Newar system without a fragment of historical/comparative data. The scope for future research in this direction is immense and highly relevant for several language sub-families in Nepal.

Typology of Verbal Categories m 157

Abbreviations 1 2 3 sg. du. pi. ACC AGT/ABL ASP CAUS COM COMP CONS CND CP DAT EMPH ERG EVID FEM FIN PUT GEN h. nh. HYP

first person second person third person singular dual plural accusative agentive/ablative aspect causative commitative complementizer concessive conditional conjunctive participle dative emphatic marker ergative evidential feminine marker finite verb future genitive honorific non-honorific hypothetical

Ims. INF IPRF LOG NPST NOM NPC NPD NPT NTG NVOL PC PD PH PST PRF PRG PRS PT PTP PURP Q SEQ SUP TOP VOL

Impersonal infinitive imperfective locative non-past nominalizer non-past conjunct non-past disjunct non-preterite neuter gender non-volitional past conjunct past disjunct past habitual past perfective progressive present preterite participial purpose question marker sequential superlative marker topic volitional

• REFERENCES Bhat, D.N.S. 1991. Grammatical relations: the evidence against their necessity and universality. London: Routledge. . 1998. A typology of Himalayan language. Keynote Address to the 4th Himalayan Languages Symposium, Punc, India. December 7-9. . 1999. The Prominence of tense, aspect and mood. Amsterdam: John Benjamins. Bilgiri, H.S. 1965. Kharia: phonology, grammar and vocabulary. Poona: Deccan College. Brinton, LJ. 1988. The development of English aspectual systems. Cambridge: Cambridge University Press. Bybee, Joan L., R. Perkins, and W. Pagliuca, 1994. The evolution of grammar: tense, aspect and modality in the languages of the world. Chicago: University of Chicago Press. Chalise, Krishna Prasad. 1999. Morpho-syntax of the Tamang verb. Kirtipur, Kathmandu: Central Department of Linguistics, Tribhuvan University MA dissertation. Comrie, Bernard. 1976. Aspect. Cambridge: Cambridge University Press. . 1985. Tense. Cambridge: Cambridge University Press.

158 · Tej R. Kansakar Comrie, Bernard. 1989. Language universal* and linguistic typology (second edition). Chicago University Press. Croft, William. 1990. Typology and universal*. Cambridge: Cambridge University Press. DeLancey, Scott 1981. An interpretation of split ergativity and related patterns. Language 57 (3). 626-57. . 1982. Aspect, transitivity and viewpoint. Tense and Aspect: Between Semantics and Pragmatics, ed. by Paul Hopper, 167-84. Amsterdam: John Benjamins. . 1985. The analysis-synthesis-lexis cycle in Tibeto-Burman: A case study in motivated change Iconocity in Syntax, ed. by J. Haiman, Amsterdam: John Benjamins. 367-89. . 1987. Transitivity in grammar and cognition' Coherence and Grounding in Discourse, ed. by Rόssel S. Tomlin, 53-68. Amsterdam: John Benjamins. . 1988. Transitivity and ergative case in Lhasa Tibetan. Berkeley Linguistic Society. 13-140. . 1992. The historical status of the conjunct/disjunct patterns in Tibeto-Burman. Acta Linguistica Hafniensia 25. 39-62. Driem, George van. 1987. A grammar ofLimbu. Berlin: Mouton de Gruyter. . 1993. 'The Newar verb in Tibeto-Burman perspective'.Xcto Linguistica Hafniensia 26.23-43. Ebert, Karen H. 19%. Finite forms in nonfmite function: the Tibeto Burman-Dravidian-Munda contact area, 14th European Conference on Modem South Asian Studies, Kobenhavn 21, 24 August 1996. Gfrvon, T. 1984. Syntax, Vol /. Amsterdam: John Benjamins. Greenberg, Joseph H, Charles A. Ferguson and Edith A. Moravcsik (eds.). 1966. Universals of human language (second edition). Cambridge, MA: MIT Press. Heine, Bernd, Ulrike Gaudi and Friederike Hunnemeyer. 1991. Grammaticalization: a conceptual framework. Chicago: University of Chicago. Kansakar, Tej R. 1994. Grammaticalization of verbs in classical and modern Newari. Linguistics of the Tibeto-Burman Area 17:1. 81-97. . 1998. The Syntactic typology of Newar and Tamang languages (Nepal). Paper presented to the Fourth Himalayan Languages Symposium, Pune, India. 7-9 December. . 1999. Verb agreement in classical and modern Newar dialects. Topics in Nepalese linguistics, ed. by Yogendra P. Yadava and Waresen W. Glover, 421-43. Kathmandu: Royal Nepal Academy. Khatiwada, Karunakhar. 1999. Verb morphology of Dhimal. Kirtipur, Kathmandu: Central Department of Linguistics, Tribhuvan University MA dissertation. Malla, Kamal R 1985. The Newari language: a working outline. Monumenta Serindica 14. Tokyo: Institute for the Languages and Cultures of Asia and Africa. Masica, Colin P. 1991. The Indo-Aryan language. Cambridge: University of Cambridge. Matisoff, James A. 1967. Verb concatenation in Lahu: the syntax and semantics of "simple" juxtaposition Acta Linguistica Hafniensia 12.69-120. Pokharel, Madhav P. 1998. Categorial splits in the use of < -le > in Nepali. Nepalese Linguistics 15.42-50. Salkie, R. 1989. Perfect and pluperfect. Journal of Linguistics 25.1-34. Schottelndreyer, Burkhard and Heiderose Schottelndreyer. 1973. Sherpa texts. Clause, sentence, and discourse patterns in selected languages of Nepal, ed. by Austin Hale and David Watters, 53176. Kathmandu: Summer Institute of Linguistics. Watters, David E. 1973. Cause patterns in Kham. Clause, sentence and discourse patterns in selected languages of Nepal·! ed. by Austin Hale. 39-202. Norman, Oklahoma: Summer Institute of Linguistics. . 1975. The evolution of a Tibeto-Burman pronominal verb morphology: a case study of Kham. LTBA2(l). 45-80. . 1999. The iconicity of direction marking in Kham. Notes on Tibeto-Burman-4,25-47. High Wycombe, Bucks, England: South Asia group.

Ergativity Attrition in the History of Western New Indo-Aryan Languages (Panjabi, Gujarati and Rajasthani)1 ι L.V. KHOKHLOVA ι The ergative pattern in Apabhramsha declension developed through two main case marking systems of ΝΙΑ: the accusative and the tripartite. The bound pronominal affixes in Western Panjabi followed general tendency towards the accusative pattern. The Rajasthani pronouns were behind nouns in the process of attrition of A/S opposition, and this contradicts the general tendencies existing in languages with split nominal systems. In late MIA and early ΝΙΑ the syntactic constraints on clause combination or on the coreferential deletion showed deviations from the accusative pattern, while Modern NLA follow the accusative standards. The Agent in Modern Western ΝΙΑ languages is no more freely omitted from any perfective transitive clause, as has been possible in the earlier stages: the marginal role of Agent typical for ergative languages is not retained in Modern ΝΙΑ.

• 1. The Scope and Aim of the Present Study There are many scholarly works devoted to the origin of the ergative construction in Indo-Aryan languages. Anderson (1977, 1988) puts these languages into the general framework of mechanisms by which languages of the world become ergative. Klaiman (1978,1987) analyzes the antecedents for the Modern Indo-Aryan ergative construction and ergativity hierarchy in modern ΝΙΑ languages. Hock (1986), when tracing back the ergative construction as part of the development of the whole syntactic system of the language, concentrates on Patient-oriented constructions. Andersen (1986) discusses the origin of the

160 · I.V. Khokhlova

ergative construction in MIA cast in the framework of Functional Grammar. Bubenik (1998) gives a detailed description of the morphosyntactic changes in the history of Indo-Aryan languages during their medieval period (6-12 c.). The specificities of the process of ergativity attrition in the course of development have not yet attracted the attention of many linguists. There exist good enough books, touching on a number of syntactic issues in Old NLA languages in a descriptive fashion (Tessitory 1914-16; Miltner 1964; Srivastav 1970) but there are no consistent studies of their typological development. Stump (1983) analyzed the conservative and eliminative tendencies in the history of Indo-Aryan ergative construction, showing the drift towards nominativity. His analysis is restricted to case marking and verbal agreement patterns. An attempt to describe the historical evolution of the reflexes of Sanskrit ta- participle construction as a part of the whole syntactic process has been made (Khokhlova 1992 and 1995). Here I shall try to give a more detailed description of the attrition of morphological ergativity in Western in connection with the general typological development of the described languages. The lower chronological point in this study will be the 15th century when Old Western Gujarati-Rajasthani started splitting into two languages: Gujarati and Marwari. The syntactic changes in will be analyzed while taking into account some syntactic properties of basically ergative languages serving as standard of ergativity (Kibrik 1992). The data analyzed here were obtained from prose texts writtem by Jain authors in Old Rajasthani and Old Gujarati. Preference has been given to prose texts as they are less liable to various distortions demanded by rhyme or metre. Besides that, poetic language has its own rules of word order and of ellipsis. Keeping in mind all these specific features of poetical texts, I have still used here as an additional source of data the famous Marwari ballad Dholä Märu räDühä written in the 15th century. The earliest prose texts in Panjabi (Janam Säkhl) belong to 17th-18th centuries. In order to investigate the morphological evolution of the language of 15th-16th centuries, I studied the Panjabi poetical texts included in the Adi Granth. Tradition says that this collection of Sikh scriptures was compiled by the fifth Sikh Guru Arjan Dev in 1604. It comprises the hymns of the first five Sikh gurus as well as the works of various religious poets (notably Kabir, Namdev, Ravidas and Sheikh Farid). Texts produced by these different authors had their own linguistic peculiarities, but the main trends in the system of noun declension and in verbal concord were the same. On account of grammatical similarities and common religious terminology, the language used by the religious poets of Northern India during the 15th and 16th centuries is usually defined as sädhukan bhäsä. Syntactic processes in modern Gujarati, Rajasthani and Panjabi were analyzed on the basis of prose texts and the data obtained from native speakers.2 The texts I have been able to work through and the number of native speakers I could consult provide the necessary data

Ergativity Attrition and Western New Indo-Aryan Languages m 161

for making some tentative hypotheses, though this material is incomplete for final conclusions.

• 2. The Morphological Evolution of Western Languages • 2.1. Bubenik (1998: 65-87, 142-43) has shown that an ideal ergative pattern in noun declension developed in Apabhramsha when Nominative vs. Accusative distinction had been destroyed by sound changes. Since that time the case marking as well as the verbal agreement in the constructions with the reflexes of the historical ta-participle were maximally ergative. This means that in ergative domain (in perfective tenses) the transitive object (O) was grouped together with the intransitive subject (S) against the transitive subject (A), and the unmarked transitive object (O) was controlling the verbal agreement. In the process of case merging, the Instrumental case ceased to exist in Rajasthani. This led to the attrition of A/S opposition. Early samples of Rajasthani prose demonstrate the first signs of ergative attrition in noun declension. A/S opposition is not always preserved in masculine stems ending in consonants (1), (2) and in all the feminine stems (3), (4). Ergative Agent may be marked by varying freely Nominative (1), (3) or Instrumental (2), (4) and this variation may take place within the same text: see the samples from 16thcentury Rajasthani: (1) kumär(A) tatkäl te phul prince + NOM/M/G immediately these flowers + NOM/M/PL suijghäryä

smell + CAUS + PtPpl + M/PL 'The prince immediately caused [her] to smell these flowers' (R.G. 39). (2) kumari (A) kahiu prince + INSTR/M/SG say + PtPpl + NEUT3 The prince said' (R.G. 37). (3) kuttinl (A) püchiu bawd + NOM/F/SG ask + PtPpl + NEUT The bawd asked' (R.G. 35). (4) kuffinii (A) vayarsena (O) äpnai bawd + INSTR/F/SG Vayarsena + NOM/M/SG self + LOG ghari änyau house + LOG bring 4- PtPpl + M/SG The bawd brought Vayarsena to her house' (R.G. 36).

162 · I.V. Khokhlova

Nominative case has also been used for marking S (5), (6) and (7), (8). The effect of this was that the singular forms of the masculine stems ending in consonants and the singular forms of the feminine stems have developed the neutral case marking (A = S = ). The problem of identifying the NPs of the sentence with A and thus could be solved only by taking into account the word order and the verbal agreement that remained consistently ergative (8): (5) kumär(S)... äwiu prince + NOM come + PtPpl 'The prince came.' (6) veSyä (S) äpnai nagari ävl courtesan self + LOG town + LOG come + PtPpl 'The courtesan came to her town' (R.G. 37). (7) ketlai käli vyäpär rahit dhan-vyay kartu so much time trade without wealth-consumption doing kumär (O) dekhi veSyäi" (A) püchiu" prince -I- NOM see + ABS courtesan + INSTR ask + PtPpl 'Having seen the prince spending money without trading, the courtesan... asked'(R.G. 36-37). (8) räjä (A) kuttinl (O) mükäwi King -I- NOM/M bawd + NOM/F absolve + PtPpl -I- F 'The king absolved the bawd' (R.F.39). The only singular noun paradigm that had retained A/S opposition was the masculine paradigm of nouns ending in = as in (9) and (10) below: (9) tin-rai pat edal dikro (S) baitho him-of throne Edal son + NOM/M/SG sit + PtPpl + M/SG 'In his throne sat his son Edal' (R.G. 42). (10) sürai (A) kahiu bottle bird + INSTR/M/SG (from sürä) say + PtPpl + NEUT 'The bottle bird said' (R.G. 33). In languages with mixed nominative-accusative and absolutive-ergative types of intraclausal marking ergativity is usually marked most strongly in the plural (Dixon 1994: 93). Old Rajasthani followed the general rule retaining A/S opposition in plural paradigms: "sons1 beta (NOM/M/PL) / bete (INSTR/M/ PL); 'thieves cor (NOM/M/SG = PL) / core (INSTR/M/PL); 'daughters' kanyä (NOM/F/SG = PL)/tarcyfle(INSTR/F/PL).4 Plural masculine nouns not ending in = followed the singular pattern with its tendency towards the attrition of A/S opposition, compare (11) with Agent in Instrumental (fifteenth century) and (12) with Nominative Agent (16th cent.): (11) loke püchiu" people + INSTR/PL PtPpl + NEUT 'People asked' (R.G. 16).

Ergativity Attrition and Western New Indo-Aryan Languages m 163

(12) bevai parjkhi(A)... ambä~nä both bird + NOM/M/SG = PL mango-of + M/PL phal (O) vayarsen-nai" fruiM- NOM/M/SG = PL Vayarsena-of lap + LOG utsarjgi müijkyä place -I- PtPpl + M/PL 'Both birds placed mango fruits into the lap of Vayarsena.' (R.G. 34). The same tendency towards marking both A and case may be observed in poetry (13):

by the same Nominative

(13) räjä prohit teriyau raja + NOM priest + NOM call + PtPpl 'Raja called the priest' (D.M.: 191). Pronouns in early Rajasthani prose texts retained ergative case marking, grouping and S against A: (14) hün(S)... te mäthä räjä-nai dekhärsu" I + NOM those heads king-to show V FUT/l/SG shall show those heads to the king' (R.G. 33). (15) sumitra nämi mantri tini (A) haum (O) tumh kanhai Sumitra by name minister he + INSTR I + NOM you to mokaliu send '+ PtPpl + M/SG 'There is a minister, Sumitra by name, by him I am sent to you/ ...he sent me to you' (R.G. 6). (16) mal (A) purä-purvi e yogl-nun ves (O) pahiriu" I + INSTR previously this yogin-of appearance put on + PtPpl nathl NEG have never before worn the attire of an ascetic' (R.G. 36). The fact that the pronouns, especially personal pronouns, were behind nouns in the process of attrition of A/S opposition contradicts the general tendencies existing in the languages with split nominal systems: wherever there exist contrasting systems for nouns and pronouns, personal pronouns inflect in an accusative paradigm. Even in 'classical' ergative languages like Dyirbal (Dixon 1994) personal pronouns follow accusative morphological pattern. This deviation from the general typological rule can be explained partly by the greater conservatism of the pronoun system and partly by the differences in morphological and syllabic structures of MIA and NLA nouns and pronouns: having no real restrictions on syllable structure, nouns can be subject to greater phonological loss in final syllables, compared to typically just disyllabic (or

164 · I.V. Khokhlova trisyllabic) pronouns. That may explain why nouns are more likely to undergo case syncretism. (H.H. Hock, p.c.) Having separated from Rajasthani, Gujarati has retained Instrumental case and as a result ergative case marking system: see below (17) from 16th-century Gujarati: (17) kidie cilatiputra nau deha ant + INSTR/F/PL Chilati's son of body + NOM/M/SG cäläni nl parii" lädhau sieve of manner + INST/F/SG do + PtPpl + M/SG 'The body of the son of Chilati was made like a sieve by the ants' (Dave 1935: 66). The morphological development of Old Panjabi shows many similarities with the evolution of case marking system in Rajasthani. If we do not take into account the deviations from the main trends that may be observed in verses belonging to various authors, the nominal system of Old Panjabi may be described as comprising three main case marking systems: (a) the ergative system: masculine nouns ending in = u like guru 'religious teacher' (I); (b) neutral case marking system: the rest of singular masculine nouns like pop* 'sinner' (III) and singular feminine nouns like sakhl 'friend (fern.)' (IV); 3) tripartite system: the singular masculine nouns ending in = ä like marjgta fakir' (II), and four plural paradigms (V), (VI), (VII), (VIII): (I) (H) (III) (IV)

S guru maijgtä päpi sakhi

A guri/guri maijgtai päp] sakhi

O guru maqge päpi sakhi

(V) (VI) (VII) (VIII)

S gur mangte päpi sakhr

A gurt1 maqgtl" päpn" sakhli»

O guran marjgti" päpiä" sakhia»

Similar to the situation in Rajasthani, the 'neutral' case marking in some masculine and feminine paradigms (III), (IV) of Old Panjabi characterized by high frequency resulted in identical morphological markings of A and of the sentence—compare (8) above and (18) below: (18) nandl (A) kantu (O) na young woman + F/SG husband + NOM/SG not rävio please + PtPpl + NOM/SG 'The young woman did not please her husband (i.e., did not play amorously with him)' (F. 93). Pronouns of Old Panjabi had the tripartite case marking system that was gradually changing into the accusative one. Within this general tendency there

Ergativity Attntion and Western New Indo-Aryan Languages · 165 were many deviations from the main trend. The pronouns varied not only in different texts belonging to different authors but even within one text belonging to the same author. For instance, in verses of Sheikh Farid, singular personal pronouns in some contexts form tripartite opposition where A*S*O (19-73): (19) hau (S) lori sahu äpnä tu (S) I + NOM need + PRES/I/SG lord self-poss you -I- NOM lorahl

(20)

(21) (22)

(23)

angiar

need + PRES/2/SG live coals search for/need my Lord, you search for/need live coals' (F. 75). mai (A) jäniä dukh mujh ku, I + OBL know + Ptppl + M/SG sufferingme to suffering dukhu sabäiai jagi whole + LOG world + LOG f l thought I am suffering [alone], [but there is] suffering in the whole world' (F. 97). tai (A) rabu visäriä you + OBL god + NOM/M/SG forget + PtPpl + M/SG 'God is forgotten by you/...you forgot the god' (F.I 14). bäriparäiai baisnä sal mujhai (O) door + LOG another's + LOG sit + INF Lord me + ACC na deh not give + IMPER 'Lord, do not let me sit at other's door' (F.74). sabhe vastü mithia" rab na all thing + NOM/F/PL sweet + F/PL god not pujani tudhu reach H- Pres.'+ 3/PL you + ACC All sweet things do not reach you (i.e., all sweet things of the world cannot be compared with reaching the God)' (F. 67).

There are many deviations from the above described tripartite system within the same collection ofshlok and shabd by Farid. The Instrumental form of first person sg. pronoun mai might be used not only in S function—instead of Nominative hau (24) (this was not a unique feature of Rajasthani only, as it was typical for the other languages also), but it could replace the Direct Object (25) and Dative subject (26) as well: (24) gunahi bhanä mai (S) phirä sin + INSTR filled I + OBL wander wander full of sins' (F. 86). (25) äpu savärahi mai milahi self purify + Pres/2/SG I ·*· OBL meet-4-Pres/2/SG '[If] you purify yourself, you [will] meet me' (F. 106).

166 · I.V. Khokhlova

(26) lokän äpo äpnl mai äpnl people -l· OBL self selve's/F/SG I + OBL selve's/F/SG pal fall + PtPpl/F/SG 4 People have their own [thoughts, interests, desires], I have my own (lit.'tome fell mine') (E106). The second person singular oblique pronoun tai might function not only as ergative subject (21), but also as Goal (27), (28). This fact shows that Old Panjabi retained the MIA model of grouping A and against S, as described by Bubenik (1998: 3), who characterizes such syntactic grouping as 'the main event in the development and restructuring of the pronominal systems of Ardha-Magadhi, Apabhramsha and early languages'. This is a typologically rare case, and it has also been attested in modern Pashto and in Kurdish (ibid.: 89). (27) jo tai märani mukmn who (rel.) you + OBL beat + Pres/3/PL fist + DIR = OBL/F/PL 'Who beat you with fists...' (F. 56). (28) dhan kük^di gor me" tai sah woman cry + PresPpl/F/SG grave in you -I- OBL Lord na miliäsu not meet + PtPpl + F/SG -I-1 'Woman cries in the grave: «I have not met you, Lord!» (F. 83). The oblique form of the second person sg. pronoun tai is used not only for expressing A and O in sentences with pronominal actants (see above), but also in constructions of comparison (taijevadu 'like you' [F.55]) and in those with postpositions (taipäsahu 'from you' [F. 92]). The wide functional repertoire of the oblique singular forms of the first and second pers. pronouns leads to grammatical homonymy. Together with lexical homonymy it creates a specific ambiguous style where an essential number of sentences or their parts may be understood in different ways. For instance, sah in (28) means both 'shahj and 'w//A', and tai may be used (and understood) in two ways: as the Direct object and as the oblique form preceding the postposition. Thus tai sah na miliäsu may be interpreted as did not meet you' (traditional interpretation of Sikh scholars) and as did not manage to unite with you'. The development of the 'neutral' case marking in some masculine and feminine nominal paradigms of Old Panjabi and of Old Rajasthani and—as a result—the identical morphological markings of A and O in sentences, the grammatical homonymy of personal pronouns—all these peculiarities have made the morphological case marking systems of the described period not distinctive enough. Due to this the postpositions had to appear as the means of identifying NPs with the syntactic roles typical for them.

Ergativity Attrition and Western New Indo-Aryan Languages · 167

• 2.2.

The tendency for special marking of those NPs that were higher in animacy and definiteness, cleared the way towards nominative morphology. The 'accusative' postposition appeared first in constructions with verbs in imperfective tenses5 and later penetrated also into the perfective domain.6 By approximately the 17th century, Rajasthani had developed two main case marking systems: the tripartite (A*S*O): animate masc. sg. nouns ending in = (I), all plural masculine nouns (II), (III), and the accusative (A = S*O): masculine singular nouns not ending in = (IV), singular (V) and plural (VI) feminine nouns. 1. tripartite(A*S*O): 2. accusative(A = S*O): (S) (A) (O) (A = S) (O) (I) räno ränai ränä + nai (IV) kumar kumär + nai (H) ränä ränä" ränä" + nai (V) vesyä vesyä + nai (III) kumär kumärä" kumärä" 4- nai (VI) vesya" vesya" + nai These case marking systems have survived till present times. Quite recently, the singular masculine (rδno-type) nouns and all the nominals of attributive classes formerly declined asrδ/10-type nouns, have started changing the tripartite case marking for the accusative one. Personal pronouns of Modern Rajasthani retain the tripartite system, while the other classes of pronouns (demonstrative, relative, interrogative) do demonstrate nowadays a clear tendency towards the accusative type of case marking. See the details and the examples in Khokhlova (1995). Unlike Rajasthani, Gujarati has preserved the Instrumental case. As a result of this, with the appearance of accusative postposition its ergative case marking system has developed into a consistent tripartite one for nominals of all classes (Cardona 1965). The earliest available Panjabi prose texts (Janam Sakhi) have been composed at the time of the system's turning towards nominative morphology. By that time the Instrumental case had stopped to exist, and there remained only two cases: the direct one and the oblique; due to this the gwrw-type nominals started being declined as those of -type (see the declensional paradigms of the Adi Granth above). The 'accusative' postposition that had appeared by that time, was still used irregularly even with pronouns. The constructions with ergative Patients expressed by pronouns in Nominative case were possible in Janam Sakhis (29), though in the later language they are not allowed: (29) hau (O) tere

päp

I + DIR your (second pers sg.poss.pron.) 4- M/PL sin + DIR/M/PL

gavävane

de väste paramesar (A)

do away with for

God 4- DIR = OBL/M/SG

168 · I.V. Khokhlova bhejiä hän send -h PtPpl H- M/SG be + Pres/1/SG have been sent by God to do away with your sins' (PPV 36). In this later language the accusative postposition has become obligatory with animate nominals. At the same time nominals in A function have acquired the ergative postposition ne. As a result of this process Panjabi has developed the tripartite case marking system for all the nominal classes excluding personal pronouns. These latter tend towards acquiring the accusative case marking system: A = S*O. Having differences in case marking, Rajasthani and Gujarati have developed similar verbal concord systems: in ergative domain the verb agrees with Patient, either marked or unmarked. In some Rajasthani dialects there exists a split in verbal concord control: the perfective participle agrees with O in gender and number, while the present auxiliary verb shows agreement in person with A (Magier 1983: 250). All the examples of this type I could find in the texts had the Agent expressed by 1st person sg. pronouns, occupying the lefthand end of the Agency hierarchy, see (30a). Contradictory examples with the 3rd person, aux. are also frequent in Marwari texts (30b): (30) a. mhai" saugan lai li I + ERG oath + DIR = OBL/F/SG take take + PtPpl + F hun aux -I- 1/SG have sworn an oath' (Mathur: 105). (30) b. rnhai" äp + rai kanvar n hityä I + ERG you -I- of son of murder + F/SG ki hai do + PtPpl + F aux + 3/SG = PI have committed the murder of your son' (Kaviyδ: 44). In Panjabi the verb agrees only with unmarked O. In case is marked by the accusative postposition, the verb is in its default form (of third pers. masc. sg.). Specially interesting is the system of bound pronominal affixes in western Panjabi. The preliminary results of the observations on usage of these affixes in the Adi Granth are given below. Further research is needed in analyzing these data. The first person singular affix = mhi/mi/mu cross-referenced: 1. S coded: bhavio = mi I was [F.63]. 2. A coded: ditho = mi have seen' [A.G. 966], diü = mu bhähi set fire' [A.G. 706].' 3. Passive subject coded: täriä = mu was put across (the ocean)' [A.G. 710]. 4. Oblique object: thlo = mi '(it) became/happened to me' [F. 63].

Ergativity Attrition and Western New Indo-Aryan Languages · 169

The corresponding pronoun itself might be omitted from the verbpredicate structure, (cf. [31] below) with the pronominal affix replacing S in bhaviomhi and Oblique object in thiomi: (31) farida ini Farid! these + INST bhaviomhi be + PtPpl + M/SG + I küjrä

niki jarjghiai thai dügar small feet + INSTR plain mountain aju faridai today Farid + DAT sai

kohän

earthen vessel + M/SG even mountains-[like] thiomi become + PtPpl + M/SG -I- to me , Farid: When I was a child I walked [with my] small feet across plains and mountains. Now [in old age] even an earthen vessel becomes [heavy] like mountains for me, Farid' [F. 63]. The second person affix hi/hu was used to cross-reference S(33) and O(32). Examples (32) and (33) show that the pronoun is not always omitted when cross-referenced in the verb. (32) farida kithai taide mäpiä, jinh'i Farid! where your parents who(reL) + INSTR tu(S) janio = hi you + NOM/M/SG bear + PtPpl + M/SG -I- you , Farid! Where are your parents by whom you were born/...who gave birth to you?' [F. 92]. (33) tu ajai na you + NOM/M/SG today + emph. not patmo = hi believe (v.i.) + PtPpl + M/SG + you Till nowadays you did not believe (in God)' [F. 92]. The third person affix = (a)nu or (a)ni cross-referenced A (34): (34) hukme dharti säj'ia = nu saci dharamsälä will + INSTR earth create + PtPpl -I- he real abode of Dharma 'By Will He[God] created the Earth, real abode of Dharma' [A.G. 787]. Affix = (alä)su or = (a/a)si was polysemantic. It cross-referenced first (35), and third (36) person S: (35) hau ävani jäni mui = asi I + F coming + LOG going + LOG die + PtPpl + F + I have died in coming and going' (Shackle 1983: 135).

170 · I.V. Khokhlova

(36) nandl kantu na rävio vadi th'i mm = äsu woman husband not pleased old was die + PtPpl + F/SG + she 'The woman did not please her husband, [when] she became old, she died.' The same affix could also mark inside the verb the third person pronoun functioning as Beneficiary (37) or Possessor (38): (37) sahi tikä dito = su Padishah + INSTR tika + NOM/M/SG give + PtPpl + M/SG + he jlvdai being alive 'Padishah [Guru Nanak] during his life time gave him [Guru Angad] a tika' [i.e., 'blessing']. (38) putr mit dhanu kichu na rahio = su son friend wealth something not stay + PtPpl + M/SG + his chodi gaiä sabh bhäi säk leave + ABS go + PtPpl all brother relative 'Son, friend, wealth—nothing belonging to him remained, all brothers and relatives left him' [A.G. 925]. The above given description shows that only the first person singular pronoun occupying the extreme left position at the Nominal Hierarchy scale (the hierarchy of NPs that are more likely to be in A than in function), had the same bound pronominal affix = mhi/mi/mu for both S and A. Other affixes were used for expressing one particular syntactic role each and marked either A alone (as = (a)nu/(a)ni)) or S alone (as = (alä)sul = (a/a)si), or grouped together S and (as = hi/hu). Deviations from this mostly ergative system are observed already in 'Janam Sakhis\ where affix = (alä)sul = (a/a)si is used not only with intransitives, but also with transitives (39) and not only in the past tense, but also in the present (40): (39) baithkar araj kiti = asu sit V ABS request make + PtPpl + F/SG + he 'He sat down and made a request' [PPV 38]. (40) nänak näu hai-su Nanak name be + Pres/3/SG + his 'His name is Nanak' [P.P.V. 38]. In Modern Western Panjabi the bound pronominal affixes can occur with a wide range of verb forms (Shackle 1976:152). For the most part they are not in complementary distribution with respect to the cross-referenced syntactic roles. The same third person singular affix may cross-reference S (41), A (42), Dative subject (43), Possessor as a pronominal modifier of S (44) and even Possessor as a modifier of (45):

Ergativity Attrition and Western New Indo-Aryan Languages m 171 (41) jananr jah te he = su woman like he be + Pres/3/SG + he 'He is like a woman' (Smirnov 1976: 99). (42) apJl dp kort tal kiti = (u)s appeal Chief Court to do + PtPpl + F/SG + he 'He appealed to the Chief Court' (ibid.: 120). (43) m te bhen y d gyoni = 5 mother and sister memory come-stem go + PtPpl/F/PL + to him 'He remembered [his] mother and sister' (Shackle 1976:152). (44) duhe« dhi n in both daughter + F/PL come + PtPpl + F/ + PL han = (i)s be + Pres/3/PL + his 'Both his daughters have come' (ibid.: 152). In case two pronouns are cross-referenced on the same verb, their syntactic roles are identified by the order of bound affixes—see (45), where both A and Ο are marked in the verb, (46) with the bound affixes for A and Addressee, and (47)—a rare case—where the bound affixes simultaneously mark A and the modifier of O: (45) ditho = mi = s see -I- PtPpl + M/SG + I -I- him Ί saw him' (Shackle 1976:152). (46) kit b dito = mi = s book give + PtPpl + M/SG + I -I- to him Ί gave him a book' (ibid.: 152). (47) khir pltai = mi = s milk drink + PtPpl + M/SG + be + Pres/3/SG + I + his Ί have drunk his milk' (ibid.: 152). The detailed comparison of the system of bound pronominal affixes in the Adi Granth, Janam Sakhis and in the dialects of Modern Western Panjabi not only goes beyond the scope of this paper, but seems in general hardly possible, as it is difficult to establish correspondences between the texts produced by the medieval authors with the concrete modern dialects, but the general trend is nevertheless quite evident: the ergative-type complementary distribution of pronominal affixes with respect to the cross-referenced syntactic roles is not retained in Modern Western Panjabi.

• 3. The Syntactic Evolution of Western ΝΙΑ • 3.1. The ergative construction in later MIA and in earlier ΝΙΑ suited perfectly the definition of the ergative system suggested by Comrie (1978: 29-95): the

172 · L.V. Khokhlova

'passive' syntax without the 'basic' active counterpart. This definition of ergativity was advocated by Hock (1986: 15-26) in his description of the main trends of syntactic development in later Sanskrit: the generalization of the 'passive pattern' (with its exclusive use of the Instrumental as Agent marker) and its spread onto the constructions with gerundive and with ία-participles; the ever increasing use of ta-participles and of gerundives as main verbs; absence of active-passive opposition in the perfective domain and its attrition in the imperfective domain with the development of A-demotional passives and of passives with modal readings; the development of a variant of the gerundivial construction implying Α-demotion even for intransitives. ' Many scholars describe the origin of ergativity in Indian languages through the reinterpretation of passive to-participle construction. (Anderson 1977; Bloch 1965; Masica 1991, etc.). Hook (1992:187) argues that in 'reanalysing/ restructuring passive as ergative' the late Sanskrit syntax has been unconsciously calking the ergatively patterned vernaculars. Klaiman (1978) gives strong arguments, supporting the claim that the ίαparticiple construction was ergative to begin with, and did not result from the 'reinterpretation of an earlier passive'. Simultaneously with the development of the 'passive' syntactic pattern the subject properties (topic position in a sentence; reflexivization and absolutivization control) shifted from the grammatical subject to the Agent (Hock 1986: 21-4; Hook 1992), etc. Taking into account the general tendency of the ergative languages to be more role-oriented than the accusative typology languages, this process of clustering of the majority of subject properties in Agent might be treated as one of the implications of the development towards ergativity. However, in 'classical' ergative languages subject properties are not connected with any particular syntactic role. The lack of direct connection between the ergative development and Subject-to Agent shift may explain the fact that the attrition of ergativity in ΝΙΑ has not changed the general role orientation towards Agent as the bearer of subject properties. Having originated in later Sanskrit, ergative development has reached its climax in later MIA. The detailed description of the syntactic history of MIA is given in Bubenik (1998). The diminishing of ergativity took place in the syntactic systems of the analyzed NLA languages during 17-20th centuries. Some important features of the later ΝΙΑ syntactic development are analyzed below. • 3.2.

The Agent of ergative construction in the Originally' ergative languages usually occupies a marginal position in the sentence and can easily be omitted. This phenomenon is observed in various ergative languages of the world. According to Dixon (1994:218), the majority of the ergative languages of Australia freely allow the A-NP to be omitted from any transitive clause. Kibrik

Ergativity Attrition and Western New Indo-Aryan Languages m 173

(1992) gives similar data from the Caucasian languages. In later MIA and in earlier the constructions with perfective participles of transitives freely allowed omission of the Agent—see below the examples (48) from Apabhramsha and (49) from 17th-century Panjabi: (48) carjgau jam sihoyaru dhariyau nice that Simhodara + NOM hold -I- PtPpl 'It is nice that Simhodara was apprehended' (Bubenik 1998:128). (49) tusän mal nün kithe bhejiä you -I- OBL I + OBL ACC where send + PtPpl + M/SG hai. mere kapre lähi tite be + Pres/3/SG my clothes tear off take + PtPpl -I- M/PL hain ate mere täF be + Pres/3/ = PL and I -I- POSS + OBL ACC bahäliä hai imprison + PtPpl -I- M/SG be -l- Pres/3/SG 'Where did you send me? My clothes are torn off and I am imprisoned' [PPV 34]. Compare also the examples (50)—with present—and (51)—with the omitted Agents—from 15th-century Rajasthani: (50) jini karani Sr'ipur-nagar-nayaki (A)... tiha" what (rel.) reason -I- INSTR Shripur city ruler + INSTR those corahan(O) mänvä nimittu amhe(o) thief -I- OBL/PL kill + INF + OBL for we mokaliyä chän send + PtPpl + M/PLbe + Pres -I- M/PL 'By what reason the ruler of the city of Shripur...sent us to kill those thieves'[R.G.: 11]. (51) ini kärani teh märäwiwä karani amhe (O) mokaliyä (by) this reason him for killing by reason we sent 'That is why we are sent to kill him' [R.G.: 9]. Objection may be raised that in Sanskrit and MIA any constituent recoverable from the text can be freely omitted, and Agent in the ta-participle construction has no privilege compared to any other NP. Wallace (1984: 167-87) examining the convergent syntactic behavior of gerundive, passive and taparticiple clauses in Classical Sanskrit text (Jambhaladatta's Vetalapancavim$ati) came to the conclusion that the /«-participle with its demoted agent matched more closely the pragmatic conventions of direct speech: agent-demotion signaled that the speaker wanted to show modesty or politeness. My point is that unlike O1A, where -deletion was elliptical, like the omission of the other constituents, and could be easily recovered from the text, -deletion in MIA

174 · I.V. Khokhlova and early ΝΙΑ may denote agentless events. In this case it is not elliptical and does not need to be recovered from the text. Compare examples (49) and (51) above where ία-participle constructions do not necessarily imply any Agent, with the elliptical constructions (52) and (53) below where the Recipient in (52) and the Goal in (53) are easily recoverable from the context: (52) gajjiu jam vasueu-kumarem dinnu boast + PtPpl when Vasudeva-prince + INSTR give + PtPPpl mah rahu sahwn jutt rem big-chariot with ' charioteer + INSTR 'When Vasudeva boasted [in this manner], [he] was given a big chariot with a charioteer'7 (Bubenik 1998:129). (53) galiya-pariyanu caiya-suhi-sayanu... Vinhussin! left-retinue abandoned-friend-relative Vi^nuSrl tuhum kahim gaiya caiu mamam? you ' where go + PtPpl abandon + ABS I + ACC If bhanantu diffau vinhussiri-juina QUOT speaking see + PtPpl V.-joined + INSTR nivaina kaha-vi bhamantu king + INSTR somehow roaming 'With the retinue gone, abandoned by the friends and the relatives, and exclaiming VisnuSrl, where did you go having abandoned me?' [he] was seen by the king, who was accompanied by VisnuSrl, roaming with toil' (Bubenik 1998:130). In the Modern Western ΝΙΑ the Agent NP is no more freely omitted from any perfective transitive clause, as was possible at the earlier stage. That means that Agent is no more playing a marginal role (a necessary feature of Originally* ergative languages). In the narrative style of Panjabi (as well as in Hindi) A-demotional perfective constructions are practically not used at all, in Gujarati they are rare, in Rajasthani their frequency is higher than in the neighbouring languages, though much lower than it used to be at the earlier stages.8 See below the example (54) of the perfective A-demotional construction from modern Rajasthani. Constructions of this type are used interchangeably with A-demotional passives (55) that have penetrated into the perfective domain: (54) r m l me" svar hai. pan j... innai log vyanjan r root in vowel is but today it + ACC people consonant W jy n bolai in v ste in n jag 'ri' 'ro* º ri as pronounce this for it of place 'ri' 'ro' emph. prayog kiyo hai ar innai chor diyo hai usage made is and it + ACC abandon give + PtPpl is

Ergativity Attrition and Western New Indo-Aryan Languages m 175

r' is originally a vowel, but today people pronounce it as consonant [plus vowel]eri',Vo'. That is why instead if it [r] ri\ ro' are used and it [r] is not used' [Bhandari: 15]. (55) f1 västai in sangraime" kei dhvani-cinh jornä this for this collection in several sound-symbol include pariyä hai, nai kei chor diyä gayä necessary was and several abandon give + PtPpl go + PtPpl hai be -f PRES 'That is why we had to include in this collection several sound symbols, while several[others] were abandoned' [Bhandari: 19]. • 3.3.

The examples given above show that while the A-demotional ta -participle construction has always had passive meaning (48), (49), (51), the constructions with the expressed Agent could be interpreted either actively or passively—see Rajasthani example (50) above and Apabhramsha (56) below: (56) hau jäyau janänie kaha I + NOM born mother + INSTR how 'How was I born of the mother?' could also mean 'How did mother give birth to me?'(Singh 1980: 42). Bubenik (1998:126-31) in his work dedicated to Apabhramsha syntax has suggested several criteria for passive interpretation of the constructions with perfective participles of transitives. Besides the A-demotional constructions like (48) that describe agentless events and naturally must have passive interpretation, four constructions with the overt Agent are also ascribed to 'passive': 1. Constructions with the argument low in animacy as an Agent (Instrument): (57) cauhim cay an turamgama ghäiya four -f INSTR four horses wound -f PtPpl 'The four horses were wounded with four [arrows]' (Bubenik 1998: 126). 2. Constructions with the Agent in marked postverbal position (58)9 (58) parirakkhiu gokula kanhem protect -f PtPpl gokula -f NOM KrsnaV INSTR 'The gokula was protected by Krsna' (Bubenik 1998: 129). 3. Constructions in which subject switching is avoided to maintain topic continuity: e.g., the ergative (i.e., active) interpretation of the main

176 · I.V. Khokhlova clause in (52) above implies some other agent, non-coreferential with the agent of the subordinate clause; as a result of this the topic continuity is disrupted: 'when Vasudeva (Top. 1) boasted [in this manner], [his father in law] (Top. 2) gave [him] a big chariot with a charioteer'; 4. Constructions with the goal as the vantage point for the narration: e.g., in (53) the vantage point for the narration is the omitted goal 'me', while the Agent NP 'by the king' is placed too far from the topic position to qualify for the ergative Agent. It seems that the very discussion concerning the choice between the active or passive interpretation of the newly formed ergative construction means the actual imposing of the syntactic models of accusative language on the language of different typology. The fact that passive transformation (changing case marking of Agent and Patient as well as word order) is used in accusative languages for 'topic-focus change' does not imply that the ergative construction where the Agent is placed into the postverbal (marked) position should have passive (opposed to active) interpretation. The type of discussion suggested by Bubenik may have sense only in connection with the English translation of a given Apabhramsha sentence. As regards the identical marking of Agent and Instrument (57), it is a typical feature of egrative syntax (Klimov 1983).

• 3.4. Examples 52-53 above deserve special attention, as they show that the rules of clausal combination and coreferential deletion in Apabhramsha were in some cases different from those in Modern . Dixon (1994:143-81) has divided languages typologically into three groups: syntactically accusative, in which the syntactic constraints on clause combination or on the omission of coreferential constituents in clause combination treat S and A in the same way (S/A pivot), while they treat differently; syntactically ergative, where the syntactic constraints treat S and O in the same way (S/O pivot) and A differently; pivotless languages, where coreferential deletion is controlled by the place of NP in the constituent order, as the examples (59) and (60) below (from Kannada) demonstrate: (59) kölu madakege tägi eradu tundu äyitu stick+ NOM pot' + DAT hit + PAST two piece become 'The stick hit the pot and (the stick) broke into two'. (60) madakege kötu tägi eradu tundu äyitu pot' + DAT stick + NOM hit + PAST two piece become 'The stick hit the pot and (the pot) broke into two' (Dixon 1994:144); (Bhat 1988:13,128; 1991:17).

Ergativity Attrition and Western New Indo-Aryan Languages m 177

When taking into account these syntactic constraints on clause combination or on the omission of coreferential constituents, the Indo-Aryan languages have always been treated as syntactically accusative (Dixon 1994: 175). The Apabhramsha examples (52-53) demonstrate deviations from the accusative pattern of clause combination: (52) shows the omission of the Beneficiary (Oi); (53), the omission of the Goal (Od), coreferential with the Agent (A). In (61) below we may see the omission of the Agent, coreferential with the Patient of the first clause:

(61) Ol = A2 mem... pesiya kimkara (Ol) pure king + INSTR send + PtPpl servant city + LOG ghare [A2 = Ol] avaloiu uvavane house + Loc search -I- PtPpl park + LOG 'The king sent the servants [and they] searched in the city in every house and in the park' (Puspadanta's //arivamSapurana 83. 6. 6-7, quoted by Bubenik [1998:158]). Early prose texts demonstrate various kinds of coreferential omissions. In the examples from the 15th-century Rajasthani given below the coreferential omissions follow the accusative (62), as well as the ergative (63), (64) patterns.

(62) (Al = S2) ti sarve(O) tini(Al) jitä, they all he + INSTR defeat + PtPpl + M/PL räjpuri äpni [ Al ]kari [S2 = Al ] baifhau Rajpuri self + POSS make + ABS sit'-I- PtPpl + M/SG 'They all are defeated by him, having appropriated Rajpuri, [he] sat [on the throne]' [R.G. 6]. (63) (Ol = S2) Srenik

sinhasani

baifhau

vidyä

parhai

Shrenik throne + LOK sit + PtPpl knowledge learn + pres ghan'i van -matarjgi" vidya(O) kahi, pun many time outcast + INSTR knowledge tell + PtPpl but [O2 = Ol]avai nah'f come -I- PRES + 3/SG not 'Raja Shrenik takes in knowledge (mantra) while sitting on the throne. The outcast has told mantra many times, but it(mantra) is not remembered'[R.G. 15]. (64) (Ol = O2) pachai mätaijgi(A) mäthä (Ol) tim kari later outcast + INSTR head + D1R/M/PL so make + ABS räjä nai dekhäryä jim rajai Raja ACC show V PtPpl + M/PL that Raja -I- INSTR

178 · I.V. Khokhlova [O2 =

l] j ni na saky know + ABS not can + PtPpl + M/PL 'Later chandala, having made the heads, had shown them to Raja, so that Raja could not recognise [them]' [R.G. 33]. Being neither consistently accusative nor consistently ergative (the latter typologically a very rare case), the rules of clausal combination and coreferential deletion in MIA and Early ΝΙΑ seem different also from the 'pivotless' syntax of modern Kannada, as the coreferential deletion seems not to be controlled by the place of NP in the constituent order. The rules of clause combination in MIA and early NLA are yet to be investigated. The only tentative hypothesis to be made here is that the intraclausal development that resulted in creating two syntactic pivots in ergative domain: A/S for controlling and S/O for coding syntactic properties might have influenced the inconsistencies in the kiterclausal processes. Later with the development of accusative marker and active/passive opposition, the rules of clausal combination and coreferential deletion became consistently accusative. Compare (61) with Modern Hindi (65) where omitted constituent may be coreferential only with A: (65) r j ne naukar " ko Sahar bhej raja + ERG servant + ACC city send 4- PtPpl every gharme" dekh aur udyanme" house + Loc see + PtPpl and park + Loc ^

^ < *e •^

*>^ ^

v

J2 g^

^>

ICO

§

^

S õ

_-

o 1 « -3 «

2 ^

s

«0

oT

JS

1

â

1

«J

1 cd cd cd PJ

i l i l i l l i l i I ll S" 2 s! 1

g

å

2

ico ù cj *co "ù

å 'S

I

^«j c

^ ^3

jo

å

co

N

-' ï

·—

å á. É ^Ï ºico .c 1—

cOicOJC

7> cd

CO

1Seo 's-

^

c cd

ICO

.co ,Ç ico «CO* »CO Ï

^

***·

åõcd

0

å

•S

gC c åo c åo c1o c åo c åo 3 Ë ^ 'Ë

'3

ó -c

'S

'2 c

j— o 'p e

"S ä vc. -^"

'^

"§

-S ·—

j= c cË c oËc o co J =ox : gc

åco

CO

ó

i Cd

«cd «co

"«Ú

1

å j= CO

^

Languages

I

._«ï

*5 ·§ ß

o J=

o

s«

3 0 C 0

"Ë c

ù

«õ

i

Scd .Si C

f 'S S S^ S" ."S

0

õ

Languages

á c

*0

cd

1

1 5»

Sanskrit

Ii

« -5

^

»cd

c^^rtw^S^.·^

3

118 i s

> g^ S ^ ' ^ •^ cd

^ o>

«i

rrj

2

j2 ÉË

,3

g

5

0

|s

^

S 8 S S 2, ù ù á ú ï ï

ce

^.^

cd

•s å 1 'c

o ^

•5 5

á

É É é É º À É É é* é Àé

t

iI

c*

j

^i

272 · ToshikiOsada

m 2.2. Dravidian I draw the following forms from Steever's (1998) generalization on personal pronouns in Dravidian. The personal pronouns of Proto-Dravidian are reconstructed in Table 1.7. (= the same as Osada cited later). Note that the oblique forms of the pronouns have short vowels while the corresponding nominative forms have long vowels. Note also that the first person plural distinguishes between an inclusive plural 'we and you' and an exclusive plural 'we but not you'. The exact shape of the reconstructed first person plural inclusive pronoun is uncertain; some scholars have proposed the reconstructed form as *nara. In the dissolution of Proto-Dravidian into the various daughter languages, the shape of several personal pronouns has naturally changed. In several South Dravidian and South-Central languages, for example, the first person singular and first person plural inclusive now begin with an initial /z, due to analogic restructuring with the first person plural inclusive and the second person pronouns so that all non-third person pronouns begin with n. This would have rendered the two first person plural pronouns homophonous; in fact the standard Modern Kannada pronoun nävu 'we' does not distinguish between inclusive and exclusive forms of the pronoun, even though the Havyaka dialect preserves reflexes of this distinction with exclusive yeqlu and inclusive nävu. In other languages, however, the first person inclusive plural was restructured in response to the potential loss of this distinction. In the development of Middle and Modern Tamil, a 'double plural' was formed by adding the plural marker-faz/ to the plural pronoun näm 'we' pronoun or, perhaps, the singular pronoun nän T, giving narjkal 'we, not you'. The formation of double plurals in the second person is well attested in the history of Tamil: the modern second person plural pronoun 'you' historically derives from nlr 'you' and plural marker-fa*/. In the northwestern dialects of Gondi, the inclusive plural, with forms such as äplo and äpan, has been borrowed from neighboring Indo-Aryan languages to maintain the inclusive-exclusive distinction. (Steever 1998: 21-2) The northwestern dialects of Gondi show us that the inclusive-exclusive distinction seems to be one of convergence-resisting features. I make a list of the Dravidian personal pronouns from Steever (1998) below. Title 2: Reconstructed Personal Pronouns of Proto-Dravidian (Steever 1998) Person

Singular Nominative

'

Plural

Oblique

First

*yan

*yan *(y)en

Second Third

*nin *tan

*nin *tan

Exclusive Inclusive

Nominative

Oblique

*yam *nam *nim *tam

yamnamnimtam-

Personal Pronouns and Related Phenomena · 273 Table 3: Personal Pronouns in Dravidian (Steever 1998)

1

Languages

1PL Inci

Old Tamil Tamil Kannada Tulu Telugu

yän, nän

Konda Gondi Kolami Gadaba Mallo Brahui

nän(u) nanna an an en

nän nanu yänui nenu

I

2PL

3PL

ExcL

näm näm nävu nama manam(u)

yarn nänkal nävu eqkulu mem(u)

ni ni ninu i niwu/nuwwu

nir,niyir nlgkal nivu nuwgulu mini

mät(u) mammat nend/nem am näm nan

mäp(u) mammat am am em nan

nin(u) nima nlv in nin ni

mir(u) nimat nir im nim num

tan tan tänu

tarn täqkal tämu

wädu/ atanu

wäUu/ wändju

tan

tarn

ö(d)

ofk

I illustrate only the nominative forms in each language. Further, most of the Dravidian languages use the demonstrative pronouns for third person pronouns. The demonstrative system in Dravidian is quite complicated. So I leave the column of third person in Tulu, Konda, Gondi, Kolami and Malto. • 2.3. Munda We have a 3 (1st, 2nd, 3rd) X 3 (SG, DU, PL) system in most of Munda languages. Sora, Gorum and Gutob lack a dual form. As for the inclusive and exclusive distinction, Juang, Sora, Gorum and Remo have no such a distinction. I have compiled the following table from several sources: Bodding (1929) for Santali, Osada (1992) for Mundari, Deeney (1975) for Ho, Drake (1903) for Korku, Biligiri (1965) for Kharia, Matson (1964) for Juang and Zide (1969) for South Munda (see Table 4). According to Norman Zide (comments on my paper), the canonical shape of Munda personal pronoun stems is: fl/nl

fNl

P 0

• 2.4. Convergent or Convergence-resisting? : Inclusive/Exclusive Distinction As is shown above, it seems that personal pronouns in Indian languages are convergence-resisting. Interestingly the northwestern dialects of Gondi still retain the distinction between inclusive and exclusive forms of the first person plural pronoun although the form is a borrowing from the neighboring Indp-

Ï Ï Ï 3 2 .× .* .* .× Ï

ü -S Å C

11

.S .E? .S* 'S 3 15rt 12 12 12 o A S Ë Ë Ë JS CO

I

é å å CO

I

'-s-s'sl ill

("

111 "Ë

á,å ï .Ëc^i ·- 'å "å á . ï . ï . ï . å"§22ùùù cO cO CO CO cO cO Ï .Ï C C u* (X Q»

á §

C/5

2

É

eo

Ë

e

.e

eo

Ë

.1

1

E E E E E E E g

I

1 si

1

é

ï ï õ «õ ù

iulf-ilf

•w

3

•C

åÏ å0

CO

c c c

M .-,

Ï

ù eo ç c c c

.__, ._ ï CO C

cd Ë C C

O CT 'C ^ ^

i-S-S-g-gS

"cO "eo "cO "co CO C

ß·1

:E'g-s:S:E'glt

'

ill

Personal Pronouns and Related Phenomena · 275 Aryan languages. As is cited above, it is very clear that this distinction occurs in Proto-Dravidian. Further, according to Pinnow (1965), this distinction is also found in Proto-Munda. As far as Indo-Aryan is concerned, this distinction is also found in some Indo-Aryan languages; e.g., Marathi, Gujarati, and some forms of Rajasthani (Marwari, Harauti).2 According to Masica (1991: 251) these languages 'conform to a Dravidian pattern in distinguishing first person exclusive vs. inclusive pronouns: Marathi ämhi/äpaN, Gujarati ame/apaN, Marwari mhe/äpä'. Have these languages borrowed this distinction from Dravidian? Masica does not mention explicitly the origin of this distinction in Marathi, Gujarati, Marwari and Harauti. Emeneau (1980) pointed out simply that 'Sindhi, like some other modern Indo-Aryan languages, has acquired it.' From where? He also didn't discuss it as far as I know. If this distinction is borrowed from Dravidian I wonder which language is the source. From a geographical point of view. Kannada would be the most likely candidate. However such a distinction is not found in Kannada, as I noted in Section 2.2. According to Gumperz and Wilson (1971), Kupwar Kannada has this distinction. But 'we do not know whether its presence in the Kupwar dialect is due to retention or to borrowing from Marathi' as Thomason and Kaufman (1988: 349) state. Furthermore, Southworth (1974: 209) gives the example of Marathi and Gujarati inclusive/ exclusive distinction of first person plural pronouns as a semantic convergence. He only mentions that 'this is a pan-Dravidian feature, but is not found in early IA'. He also never specifies the source of this distinction. Now I will suggest an alternative interpretation of this issue. This distinction occurs not due to contact-induced change but due to the socio-linguistic factors. We can find this distinction even in Japanese, which is said not to have such a distinction according to school textbooks of Japanese. We have a special exclusive form for first person plural in Japanese; e.g., temae-domo (watakushi-tachi or watashi-tachi is a general form). This is only used by salespersons. Of course we could consider this distinction as a contact-induced change, because the Ainu language has this distinction. If so, we should explain why this distinction is used only by salespersons. I guess that the acquisition of this distinction due to contact with the Ainu language is less likely than due to the sale strategy of business. It is easily understandable that business persons need the distinction of 'we including the customers' and 'we excluding the customer' at the sales scene. This socio-linguistic interpretation may be applied to the Indo-Aryan cases. As is well known, the Marwari- and Sindhi-speaking peoples are dominated by the business class. It is much suitable for my interpretation. The acquisition of this distinction due to sale strategy is probably not applicable to the cases in Marathi and Gujarati. Then we should take the inclusive form äpan in Marathi and Gujarati into consideration. This is explicitly derived from the reflexive pronoun as Southworth (1974: 210) has pointed out. Is it possible that the internal development from the reflexive pronoun to

276 · ToshikiOsada

the first person plural inclusive pronoun occurred in Marathi and/or Gujarati? We have an interesting example in the Osaka dialect of Japanese. The reflexive pronoun jibun 'self in Tokyo Japanese is used for the second person pronoun 'you' in Osaka Japanese. These are slightly different from the abovementioned case. But the first person plural inclusive pronoun means the inclusion of the second person in the deictic situation. This is a Japanese case. We, however, find the same cases in the South Asian context. For example äp 'you (polite)' in Hindi comes from atrnan 'self in Sanskrit. Besides the Hindi language, Turner (1966) illustrates the historical development of ätman 'self as follows: Nepali äphu 'self, respectful second sg. pronoun'; Assamese äpon 'respectful second sg. pronoun'; Bengali äpani, äpuni 'respectful second sg. pronoun'; Onyaäpan 'you (respectful)'; Old Marwari äpa 'self, you (honorific), we (inclusive)'; Gujarati äp 'self, you (respect)', äpan 'we (inclusive)'. It is very easy to propose the internal development in Indo-Aryan languages; i.e., reflexive pronoun > second singular honorific pronoun > first plural inclusive pronoun. In my conclusion the alternative interpretation is inferable. The distinction of inclusive/exclusive forms of first person plural in modern Indo-Aryan is considered not as a result of borrowing from Dravidian but as a result of internal development in the socio-linguistic factor; e.g., business settings. I have discussed the Indo-Aryan cases so far. However this distinction is widely spread. In the Austroasiatic languages the distinction between inclusive and exclusive forms of the pronoun is a common phenomenon. Pinnow (1965: 6) states the following: Exclusive and inclusive forms of the first pers. pi. (and dl.). This distinction occurs in MU (Munda), NIC (Nicobarese), PW (Palaung-Wa), MK (MonKhmer), and ML (Malacca languages = Aslian), but not in KHS (Khasi). In the case of NAH (Nahali) there are certain indications that the categories in question were formerly present. It must be pointed out, however, that most languages of the groups MK and ML possess no separate forms for incl. and excl. But the distinction exists in, e.g., Bahnar, which in this respect also proves close, because specially conservative, to MU and NIC. On Pinnow's proposal, Munda retained the archaic system of personal pronouns of the Austroasiatic languages. If so, you might ask me another question: is the Dravidian distinction of inclusive and exclusive derived from Munda? If the genetic relationship between Dravidian and Elamite, which lacks this distinction (Reiner 1969), can be taken to be proved (McAlpine 1981), we might suggest a possibility that Dravidian borrowed this distinction from Munda. But it is still a problem. Because we can also postulate that Elamite lost this distinction under the influence of neighboring languages. When we turn our attention to South Asia we find that the following languages have this distinction: Andamanese (Manoharan 1989), TibetoBurman languages: Ladakhi (Koshal 1979), Limbu (van Driem 1987), Dumi (van Driem 1993), Bahing (Michailovsky 1975), Khaling (Toba & Toba 1975),

Personal Pronouns and Related Phenomena · 277

Kinaur (Takahashi personal communication), Mikir (Grüßner 1978), etc. In this situation it is very difficult to specify the source of this distinction in the South Asian linguistic area.3 The contact-induced change from complex to simple is, I think, more natural. If the dominant language has a complex system, the governed languages can acquire a complex system by the force of the dominant language. The reversed case rarely occurs. As for the distinction of inclusive/exclusive forms of first person plural, the loss of this distinction is easily understandable. On the other hand, the acquisition of this distinction cannot be easily regarded as a result of contact-induced change. My conclusion is that this is probably due to the socio-linguistic factor as I illustrated in the Japanese case above. The history of the personal pronominal system is very complicated. The internal development may have easily occurred as Steever showed us in the history of Tamil. It seems to me that the personal pronominal system is a convergence-resisting feature. In this context I will discuss the subject/object marking system in Munda, which is unique in the South Asian linguistic area.4

• 3. Bound Subject/Object Pronominals in Munda We have discussed personal pronouns in Indian languages so far. In themselves the personal pronouns in Indian languages are not so different from a traditional taxonomic linguistic point of view. But subject/object marking systems in Munda are quite different from those in Indo-Aryan and Dravidian. Recently Zide (1997) and Anderson (1999b) have each written a paper on this issue. Zide focuses mostly on South Munda, while Anderson discusses the Proto-Munda subject/object marking system. My general view of subject/ object marking in Munda is in agreement with Zide (1997) and Anderson (1999b). • 3.1. A Description of Subject/Object Marking Systems in Munda First of all, I give the Mundari example of subject-object marking from my own data. The verbal morphology of Mundari is very complicated. The basic verbal structure in Mundari may be described as follows: (1) (NP = SUB) Verb base + (ASP marker) + (n) + a (= SUB) + (dl?) (+ OBJ) n: intransitive marker, d/?: transitive marker (morphological variants), a: indicative marker. A verbal base is formed by affixing a verbal stem. Verbal bases can be simple or complex; complex bases are formed by reduplication or serializing of the verbal stem. When the subject NP and object NP are animate, subject and object

278 · ToshikiOsada

agreement elements can be marked in the verbal morphology. The following personal pronominal suffixes are used for subject-object agreement: SG 1st (INC) (EXC)

2nd 3rd

-n -ml-me -el-il-e?l-i?

DU

PL

-lay -tig -ben -kir,

-bu -le -pe -ko

The same forms are used for subject and object, but occupy different slots. The subject agreement element is attached either to the end of the verb or as a clitic to the pre-verbal NP which may be the subject or a non-subject. The object agreement element occurs before the indicative marker -a, after the transitive marker -d, in unmarked sentences. Thus, ape

bir-re = n

lel-ke-d-pe-a

2PL forest-in = 1SG-.SUB see-COMPL-TR-2PL:OBJ-IND º saw you (pi.) in the forest' pulis-ko kumburu-kiq = ko sab-aka-d-kirj-a police-PL thief-DU = 3PL:SUB catch-CONT-TR-3DU:OBJ-IND The policemen have caught the two thieves.' In Mundari, animacy plays an important role for subject/object marking. I adopt here the terminology suggested by Dixon (1979,1994): S for intransitive subject, A for transitive subject, Ï for transitive object. Then I formulate the following rule for Mundari sentences: A NPs are animate nouns except for the object-experiencer construction. Thus the following sentence is not acceptable by this rule. *mandi hon-ko bisi-ja-d-ko-a food°° child-PL poison-INGR-TR-3PL:OBJ-IND The food has poisoned the children.' As the A NP mandi 'food' is inanimate this sentence is not acceptable. We use the instrumental postposition te normally, for instance, in a passive construction, mandi-te hon-ko = ko bisi-ja-n-a food-by child-PL = 3PL:SUB poison-INGR-INTR-IND The children have been poisoned by the food.' Because of the limited scope of this paper, we will not discuss the Mundari case in detail. This type of subject/object marking is dominant in Khenvarian languages.

Personal Pronouns and Related Phenomena · 279 Subject/object marking is common to all Munda languages although the subject marker occupies different slots; i.e., it is a prefix to the verb in South Munda, but a suffix or enclitic to the pre-verbal element in North Munda. I give a list of subject/object marking in Munda as cited from Anderson (1999b). Table 5: North Munda Referent Indexing Enclitics (Anderson 1999b: 22) Languages 1

Santali Mundari

1DL

-(i)fi

-Þ -iq -iq -in

Ho Bhumij Korku

1PL

Incl

Excl

Incl

-laq -laq -laq -laq -laq -lorn

-lin -liq -liq -liq -lin

-bo(n) -le

2

2DL 2PL

3

3DL

3PL

-e, -i -e/i(?)-å/-ö -?e -e'j

-kin -kiq -kiq -kin -kin

-ko ko -ko -ko -ku

2PL

3DL

3PL

-pe v i~ bo-

-kiyar -ki -kia -ki

Excl

-bu -bu -bu

-le -le -le -le

-bun

-ben -ben -m -ben -m -ben -mi(n) -pin

-m(e) -m(e)

-pe -pe -pe -pe -pe

Table 6: South Munda Subject Markers (Anderson 1999b: 12)

1DL

Languages 1

Kharia Juang Sora Gorum Gutob Remo Gta?

-n/q -V,-niq -(n)ii)

2DL

Incl. Excl.

Incl.

Excl.

-naq -jar

niq

-le

ba-

ba-

-ay ne-

1PL

nV(:,)- nVg-be a-.. -ay

-naq -naq

n-

ni-

-m

-bar mV(,) ha3-. .-å

le-

mo-

-nei -nay nae/çå-

-no na-

-nom

-pa pa-

-pe pe-

-ji -gi

-nen

-

-har-

Table 7: South Munda Object Markers (Anderson 1999b: 21) Languages 1

1DL Incl.

Kharia Juang Sora Gorum Gutob Remo Gta?

-(ni)n -in -iq

Excl.

-nba -ay

IPL Incl

3DL

2DL 2PL

3PL

Excl *

-nenin -Éåç/q -ileq

-(n)(o)m -pa -sm -om

-pe -ben -ibeq

-e

(-kia) (-ki) -ji (-gi)

By comparing these with the personal pronoun forms given in Section 2.3., we can see that the forms of North Munda referent indexing enclitics are clearly derived from the forms of personal pronouns. Subject/object markers of this kind are not found elsewhere in South Asia, as far as I know.5

280 · Toshiki Osada

• 3.2. Subject/Object Marking System as a Convergence-resisting Feature As we saw in our discussion of the person-marking systems in Munda above, complicated systems are found in all of Munda languages. Although the system in Munda is indeed divergent, we may be able to reconstruct this system for Proto-Munda along the lines Anderson has proposed. That is to say, the force of convergence towards Indo-Aryan and/or Dravidian on this matter seems to be weak. In this respect Steever (1986) has demonstrated that both subject marking and object marking occur in Dravidian languages in the Khondmal area; e.g., Kuvi, and simultaneously in the adjoining Munda language; e.g., Sora. In his conclusion Steever (ibid.: 284) pointed out that 'we have rejected the hypothesis of exercise of direct influence of SM( = South Munda) and SCDr(= South-Central Dravidian) on one another in the formation of these various exotic conjugations in the Khondmals.' Further Steever (1993: 4) has proposed that 'Dravidian morphology is fundamentally simple, and demonstrates that Contraction is the process responsible for the historical transformation of analytic forms into relatively more synthetic ones/ In his book Steever (ibid.: 220) has concluded that 'any theory of the historical development of Dravidian verb morphology, through contraction, compression, and reanalysis as mediated by markedness relations, must recognize that such a development is made possible by virtue of the extensive interaction of morphology and syntax in the grammar of these languages.' As far as the Munda influence on Dravidian languages is concerned, he has argued that A lack of adequate grammars for certain Dravidian and Munda languages has rendered many proposals of inter-influence speculative' (Steever 1993: 10). We might take the internal development into consideration first instead of contact influence. Therefore I would like to support strongly his view on the Munda influences. In modern Indo-Aryan, the existence of subject and object marking in Magahi has been described by Verma (1991). He has also demonstrated agreement with the Possessor NP as follows: i. ham okar intajär kailiai I he-3P (-Honor)-POSS waiting did-3P (-Honor) 'lit., I did his waiting', or waited for him.' ii. ham unkar intajär kailiain I he-3P (+Honor)-POSS waiting did-3P (+Honor) 'lit., I did his waiting' or waited for him.' This Magahi case is quite different from the Mundari case. As Verma (1991: 131) says: there is a big difference between Mundari and Magahi. Magahi has variant inflectional forms and not simply the actual pronominal forms occurring in

Personal Pronouns and Related Phenomena · 281 the verbal word. Furthermore, the Magahi situation looks much more elaborate than what the possible contacts could have engendered all by themselves. As has already been mentioned above, it seems to me that the subjectobject marking system in Munda is a convergence-resisting feature in the South Asian linguistic area. What is it convergence-resisting? I have a suggestion. According to Goodman (1985:131-32), 'Creoles may have a tendency to borrow free morphemes rather than bound, which would in turn have profound effects upon various aspects of their structure/1 understand well that the Indian situation is not a creole-genesis area. But I think that Goodman's suggestion is applicable to the Indian situation. I have demonstrated above that the subject/object agreement elements in Munda are bound morphemes which are mostly short forms of personal pronouns. Therefore the subjectobject marking system is convergence-resisting.6

• 4. Possessor Raising in Munda A striking feature of Munda is possessor raising. Santali examples are given first. In Santali the animate S, A, O are marked in the verbal morphology just like Mundari, as we have seen in Section 3.1. Besides these, a special set of affixes, the f- series, are used for marking the possessor of object or subject. For example, Subject/Object Marking nel-en-kan-a-e see-lSG:OBJ-PROG-IND-3SG:SUB 'He/she is seeing me.' Possessor of Object Marking in Transitive gai — ye akinrt-ked-e-t-ae-a cow = 3SG:SUB sell-COMPL-3SG:OBJ-POSS-3SG-IND 'He sold Aw cow.' Possessor of Subject Marking in Intransitive tengon~kan~t-in-a-e stand-PROG-POSS-lSG-IND-3SG:SUB 'My (animate) one is standing.' It seems that the possessor marking can occur only once. Thus, in the case of the following sentence; e.g., hopon-ingai = ye akuin-ked-e-t-ae-a 'my children sold his cows away', only the possessor of the object is marked. Further, I cite Sora and Gorum examples from Anderson (1999b) on this count.

282 · Toshiki Osada Sora

shave-head-PAST-2SG 'your head is shaven.' Gorum mir) pensil(-nom) ISO pencil(-2SG.POSS) º will see your pencil.'

ne-gi?~t-om lSG:SUB-see-PAST-2SG:OBJ

According to Anderson (1999b), similar constructions are found in numerous languages of the world although we cannot find anything similar in the South Asian linguistic area except in Burushaski. In his typological overviews, Anderson (ibid.: 28) suggested that many Possessor Raising constructions are found with verbs exhibiting incorporation of a nominal, frequently a body part. These kinds of incorporative complexes are motivated by the desire to maximize the highly salient, animate possessors in the formal referent-indexing machinery of the verb, which in head marking languages tends to be the central (and often the only obligatory) part of the clause.

• 5. Genitive Suffixes in Munda: The Effect of an Areal Feature; i.e., Word Order? As is seen in Section 4, a special set of affixes, the t- series, exist in Kherwarian which includes Santali, Mundari, Ho, Bhumij etc. This is used not only for marking the possessor of objects or subjects but it can also be suffixed to the noun to indicate the possessor; e.g., disum 'a country', diswn-tabu Our country' in Mundari. This form is found not in daily speech but in poetry. Besides this, we have another set of suffixes -a? as a genitive case marker; e.g., abu-a? disum 'our country'. We find this form in daily speech. Thus we have two forms for possessor pronouns in Mundari and other Kherwarian languages. Possessor raising is not productive in contemporary Mundari. The t- series is gradually being replaced by the a? suffixes in Kherwarian. According to Bhattacharya (1975), this suffix is found also in Kharia and Juang which belong to South Munda. But interestingly Kharia and Juang adjoin North Munda: i.e., Mundari, Ho and Bhumij. It seems to me that the a? suffix spread from Kherwarian instead of the t- series. In other words, the word order modified + modifier has been replaced by the reverse word order. According to Masica (1976: 25), Indian languages are left-branching—the modifier precedes the modified. The /- series is a counter example to this areal feature. From an Austroasiatic point of view Donegan (1993) suggests that the

Personal Pronouns and Related Phenomena · 283

word order in Mon-Khmer is Head-Dependent (SVO, NA, preposition) while the word order in Munda is Dependent-Head (SOV, AN, postposition). She uses the term Head and Dependent, which correspond to modified and modifier in Masica. Thus it is easily understandable that left-branching word order affected Munda languages in the South Asian linguistic area. This does not simply mean that the possessor pronouns change their place from postposition on the modified NP to preposition of the modified NP. The t- series possessor-pronouns are still found in Mundari poems but not marked in modern Mundari verbal morphology. We need further studies to look into the nature of convergence that has occurred in this respect. The left-branching constraint is clearly one of the factors in this change. But it seems to me that the convergence process is not straightforward and more detailed research is indeed required.

• 6. Conclusion In this paper, I have discussed personal pronouns in Indian languages and related phenomena; i.e., subject/object marking systems and possessor raising in Munda. These seem to be some of convergence-resisting features in the South Asian linguistic area. This contrasts with other convergent features which have been suggested by Masica (1976) and Emeneau (1980). For example, dative experiencer subjects in Indo-Aryan and Dravidian correspond to experiencer objects in Munda, especially in Mundari as was illustrated in Osada 1999. This exhibits the differences between Indo-Aryan and Dravidian, on the one hand, and Munda, on the other hand, although the semantic overlaps are copious among those languages. Furthermore, compound verbs, which have been proposed as a convergent feature by Masica, are also found in Munda. We, however, have subject/object marking in compound constructions in Munda while we don't have such marking in Indo-Aryan and Dravidian. This is very important. The subject/object marker occurs even in between VI and V2 in Gorum, which belongs to South Munda, according to Norman Zide (1997). Finally I summarize what I see as convergence-resisting features in the South Asian linguistic area from a Mundaist's point of view. • 6.1. Phonological Features The so-called checked (preglottalized) consonants: North Munda and South Munda (Bhattacharya 1965; Arlene Zide 1978; Norman Zide 1958) Low tones: Korku (Norman Zide 1966) Creaky vowels: Gorum (Arlene Zide and Norman Zide 1987)

284 · Toshiki Osada

• 6.2. Morphological Features Infixes -pV- (or -Vp-) (reciprocal) North Munda -b- (causative) South Munda -nK (or -Kn-) (nominalizing) North Munda Partial reduplication Cl VC2- > Cl VC1VC2 North Munda and South Munda Nominal combining forms: Gorum and Sora (Arlene Zide 1976), Gta? (Norman Zide and Mahapatra 1972) • 6.3. Morphosyntactic Features Subject/object marking: all Munda languages (Anderson 1999a, b) Possessor Raising: Kherwarian (except Mundari), South Munda (Anderson 1999b)

• NOTES 1. I would like to express my sincere thanks to Professor Peter Hook, Visiting Professor of ILCAA, Tokyo University of Foreign Studies, and Professor Nick Evans of Melbourne University, who read the earlier version of this paper. At the symposium, Professor Norman Zide of the University of Chicago (as the commentator) and other participants have given me insightful comments which are useful for revising my paper. Needless to say, I alone am responsible for any errors and inadequacies in this paper. 2. According to Peter Hook (personal communication), Hindi can also make the distinction in the possessive: hamärä Hindustan (inclusive) vs. apna Hindustan (exclusive) Our Hindustan*. I don't touch this case here. 3. See Masica's paper for the areal distribution in this volume. 4. Since Hodgson (1856) and Konow (1909) pointed out the similarity between Munda verbal morphology and Tibeto-Burman pronominalization, it is well known that there are subject/ object marking systems in the Tibeto-Burman languages. In this paper I, however, limit the scope among three language families; i.e., Indo-Aryan, Dravidian and Munda. See the following papers on this issue: Bauman (1974), Maspero (1948), Watters (1975). Further, according to Nichols and Peterson (19%: 336), 'personal pronouns with first person n and second person m have been claimed to be frequent in the native languages of the Americas, widespread there, and rare elsewhere*, but we found these forms in Munda languages. More cross-linguistic survey is required. 5. [Tibeto-Burman languages such as Hmar, Mizo, Aimol have subject/object pronominal clitics that occur with the verb—Editors}. 6. According to Nick Evans (personal communication), there are some cases of differences of bound pronominal structures in Australian languages. More cross-linguistic study on this issue is required for my conclusion.

• REFERENCES Anderson, Gregory D.S. 1999a. A new classification of the Munda language family: evidence from comparative verb morphology. Paper read at the 209th Meeting of the American Oriental Society.

Personal Pronouns and Related Phenomena · 285 Anderson, Gregory D.S. 1999b. Referent indexing in the Munda verb. Chicago: University of Chicago, ms. Bauman, Jim 1974. Pronominal verb morphology in Tlbeto-Burman. Linguistics of the TibetoBurmanArea 1 (1). 108-55. Bhatia, T.K. 1993. Punjabi: a cognitive-descriptive grammar. Routledge: London. Bhattacharya, Sudhibhushan. 1965. Glottal stop and checked consonants in Bonda. Indo~Iranian Journal 9. 69-77. . 1975. Studies in comparative Munda linguistics. Simla: Indian Institute of Advanced Study. Biligiri, H.S. 1965. Kharia: phonology, grammar and vocabulary. Poona: Deccan College. Bedding, P.O. 1929. Materials for a Santali grammar. Ehimka: The Santal Mission of the Northern Church. Bubenik, Vit. 1996. The structure and development of middle Indo-Aryan dialects. Delhi: Motilal Banarsidass. Campbell, George L. (ed.) 1995. Concise compendium of the world's languages. London and New York: Routledge. Deeney, John. 1975. Ho grammar and vocabulary. Chaibasa: Xavier Ho Publication. Dixon, R.M.W. 1979. Ergativity. Language 55.59-138. . 1994. Ergativity. Cambridge: Cambridge University Press. Donegan, Patricia Jane. 1993. Rhythm and vocalic drift in Munda and Mon-Khmer, Linguistics of the Tibeto-Burman Area 16(1). 1-43. Drake, John. 1903. A grammar of the Kurku language. Calcutta: Baptist Mission Press. Driem, George van. 1987. A grammar ofLimbu. Berlin: Mouton de Gruyter. . 1993. A grammar of Dumi. Berlin: Mouton de Gruyter. Emeneau, Murray B. 1956. India as a linguistic area. Language 32.3-16. . 1980. Language and linguistic area. Essays by Murray B. Emeneau. Selected and introduced by Anwar S. Dil. Stanford;. Stanford University Press. Goodman, Morris F. 1985. Review of Bickerton (1981) Roots of Language. International Journal of American Linguistics 51.109-37. Grußner, Karl-Heinz. 1918.ArlengAlam. Die Sprache der Mikir Grammatik und Texte. Wiesbaden: Franz Steiner. Gumperz, John and R. Wilson. 1971. Convergence and creolization: a case from the Indo-Aryan/ Dravidian border. Pidginization and creolization of languages, ed. by Dell Hymes, 151-68. Cambridge: Cambridge University Press. Hodgson, Brian H. 1856. Aborgines of the Niligiriswith remarks on their affinities. Journal of the Asiatic Society of Bengal 25. 498-522. Konow, Sten. 1909. Linguistic survey of India, Vol. 3. Calcutta: Superintendent of Government Printing. Koshal, Sanyukta. 1979. Ladakhi grammar. Delhi: Motilal Banarsidass. Mahapatra, Byay Prasad. 1976. Comparative notes on Juang and Kharia finite verbs. Austroasiatic Studies II. 801-14. Manoharan, S. 1989. A descriptive and comparative study ofAndamanese language. Calcutta: Anthropological Survey of India. Masica, Colin. 1976. Defining a linguistic area: South Asia. Chicago: Chicago University Press. . 1991. The Indo-Aryan languages. Cambridge: Cambridge University Press. Maspero, Henri. 1948. Notes sur la morphologic du tibeto-birman et du mounda. Bulletin de la Societe Linguistique de Paris 43. 155-85. Matson, D.M. 1964. A grammatical sketch of Juang, a Munda language. Madison: University of Wisconsin Ph.D. dissertation. McAlpine, Divide. \9Sl.Proto-Elamo-Dravidian. Philadelphia: America Philosophical Society. Michailovsky, Boyd. 1974. Hayu typology and verbal morphology. Linguistics of the TibetoBurmanArea 1 (1). 1-26. . 1975. Notes on the Kiranti Verb , Linguistics of the Tibeto-Burman Area 2 (2). 183.

286 · ToshikiOsada Nara Tsuyoshi. l919.Avahattha and comparative vocabulary of new Indo-Aryan languages. Tokyo: Institute for the Study of Languages and Cultures of Asia and Africa. Nichols, Johanna and David A. Peterson. 1996. The Amerind personal pronouns. Language 72 (2). 336-71. Osada Toshiki. 1992. A reference grammar of Mundari. Tokyo: ILCAA. . 1999. Experiential constructions in Mundari. Journal of the Linguistic Society of Japan 115. 51-76. Pandharipande, Rajeshwari V. 1997. Maralhi. London: Routledge. Pinnow, Heinz = Jürgen. 1965. Personal pronouns in the Austroasiatic languages: a historical study, Lingua 14.3-42. . 1966. A comparative study of the verb in the Munda languages, Studies in comparative Austroasiatic linguistics, ed. by Norman H. Zide, 96-193. Ray, Tapas S. 1999. Lexical anaphors and pronouns in Oriya. Lexical anaphora and pronouns in selected South Asian languages: a principled typology, ed. by B.C. Lust, K. Wali, J.W. Gair and K.V. Subbarao, 575-636. Berlin: Mouton de Gruyter. Reiner, Erica. 1969. The Elamite language. Handbuch der Orientaüstik, ed. by B. Spuler. Leiden: E.J. Brill. Southworth, Franklin C. 1974. Linguistic stratigraphy of North India. International Journal of Dravidian Linguistics 3 (1). 201-23. Steever, Sanford B. 1986. Morphological convergence in the Khondmals: (pro)-nominal incorporation. South Asian languages: structure, convergence, and diglossia, ed. by Bh. Krishnamurti, 270-S5. Delhi: Motilal Banarsidass. . 1993. Analysis to synthesis: the development of complex verb morphology in the Dravidian languages. New York: Oxford University Press. . (ed.) 1998. The Dravidian languages. London and New York: Routledge. Thomason, Sarah Grey and Terrence Kaufman. 1988. Language contact, Creolization, and genetic linguistics. Berkeley: University of California Press. Tbba Sueyoshi and Ingrid Toba. 1975. A Khaling-English dictionary. Kirtipur: Summer Institute of Linguistics. Turner, R.L. 1966. comparative dictionary of the Indo-Aryan languages. London: Oxford University Press. Ullrich-Baylis, Helen E. 1974. Morphological coexistence: a key to linguistic convergence. International Journal of Dravidian Linguistics 3 (2). 224-30. Verma, Manindra K. 1991. Exploring the parameters of agreement: the case of Magahi. Language Sciences 13 (2). 125-43. Wali, Kashl and Omkar N. Koul. 1997. Kashmiri—a cognitive-descriptive grammar. London: Routledge. Watters, David E. 1975. The evolution of a Tibeto-Burman pronominal verb morphology. Linguistics of the Tibeto-Burman Area 2 (I). 45-79. Zide, Arlene R.K. 1976. Nominal combing forms in Sora and Gorum. Austroasiatic Studies, ed. by Phillip N. Jenner, Lawrence C. Thompson and Stanley Starosta, 1251-94. Honolulu: Hawaii University Press. . 1978. A note on glottalization and release in Munda. Indian Linguistics 39. 70-75. Zide, Arlene R.K. and Norman H. Zide. 1987. A KM laryngeal as a conditional factor for s-loss in Sora-Juray-Gorum. Festshcriftfor Henry Hoenigswald: on the occasion of his seventieth birthday, ed. by George Cardona and Norman H. Zide, 417-31. Tübingen: Gunter Narr Verlag. Zide, Norman H. 1958. Final stops in Korku and Santali, Indian Linguistics 19: 44-8. . 1965. Gutob-Remo vocalism and glottalized vowels. Lingua 14.43-53. . 1966. Korku low tone and the Proto-Korku-Kherwarian vowel system. Studies in Comparative Austroasiatic linguistics, ed. by Norman H. Zide, 214-29. The Hague: Mouton. . 1968. Some comparative notes on Gata? pronouns. Pratidanam: Indian, Iranian and IndoEuropean studies, present to EBJ. Kuiper on his sixtieth birthday, ed. by J.C. Heesterman, 348-58. Mouton, The Hague.

Personal Pronouns and Related Phenomena · 287 Zide, Norman H. 1969. Munda and non-Munda Austroasiatic languages, Trends in Linguistics, ed. by Thomas Sebcok, Vol. 5, 411-30. The Hague: Mouton de Grufer. . 1997. Gutob pronominal clitics and related phenomena elsewhere in Gutob-Remo-Gta? Languages of tribal and indigenous peoples of India, ed. by Anvita Abbi, 307-34. Delhi: Motilal Banarsidass. Zide, Norman H. and Gregory D.S. Anderson. Forthcoming. Towards an analysis of the South Munda verbal system. Southeast Asian linguistics VJI, ed. by F.K. Lehman. Chicago: Southeast Asian Linguistic Society. Zide, Norman H. and Khageswar Mahapatra. 1972. Gta? Nominal combing forms. Indian Linguistics 33 (3). 179-202.

The Role of Language of Religion in the Convergence of South Asian Languages • RAJESHWARI V. PANDHARIPANDE · This paper discusses the convergence of Modern Indian languages in the context of the Sanskritized register which these languages commonly share and which has been developed as a result of the influence/power of Sanskrit in the domain of Hindu religion. It investigates the social motivations for this convergence and the impact of this convergence on the structure of converging languages. Additionally, the paper focuses on the following hitherto unanswered questions in this context: (a) When languages converge, why do some languages fail to participate in the process? (b) When languages develop a common register, for example, a Sanskritized register, do these Sanskritized registers across languages select the same features of Sanskrit? (c) Is convergence reversible? If it is, what are the factors responsible for it? The paper shows that in order to answer these questions, it is necessary to understand the notion 'language of power' and its influence on other languages.

• 1. Introduction Linguistic convergence has been widely studied from various perspectives which analyze the structure as well as the function of language. As a process, convergence has been investigated as language change and as a result, it has been examined as one of the factors which contribute toward creating linguistic areas and/or typology of languages. It is generally assumed that linguistic convergence occurs when genetically related or unrelated languages within a geographic area come to share linguistic features due to either or both of the following reasons: (a) mutual geographic contact for an extended period of time with or without bilingualism, and (b) influence of a language of power

290 · Rajeshwari V. Pandharipande

(e.g., English, Sanskrit, etc.) which motivates similar changes in the languages which may or may not be in direct geographic contact with one another. The major goals of the studies on convergence, typology of South Asian languages, and South Asia as a linguistic area (Bright 1968; D'Souza 1992; Emeneau 1956; Fergusson 1992; Hock 1986; Kachru 1983,1986; Krishnamurti and Mukherji 1984; Masica 1976,1992; Pandharipande 1986,1992,1998; Pandit 1972; Sridhar 1981; and Shapiro and Schiffman 1975, among others) have been to identify the features commonly shared by the languages, the origin and direction of their spread, social motivations and the significance of this spread for defining South Asia as a linguistic area. However, there are many questions related to the convergence of South Asian languages which need answers. Masica's (1992: 37) question, 'Do South Asian areal features find identifiable correlates in the South Asian society?' is relevant in the context of the convergence of South Asian languages as well. In other words, it is necessary to investigate whether it is possible to establish a causal connection between social conditions and the convergence of these languages. This paper discusses the convergence of Modern Indian languages in the context of the Sanskritized register which Modern languages commonly share and which has been developed as a result of the influence/power of Sanskrit in the domain of religion— Hinduism. The paper investigates the social motivations behind this convergence and the impact of this convergence on the structure of the converging languages. Additionally, the paper focuses on the following hitherto unanswered questions in this context: (a) when languages converge, why do some languages fail to participate in the process? (b) when languages develop a common register, for example, a Sanskritized register (as a result of the influence of Sanskrit in the context of Hinduism), do these Sanskritized registers across languages select the same features of Sanskrit?, (c) is convergence reversible? If it is, what are the factors responsible for it? The paper shows that in order to answer these questions, it is necessary to understand the notion "language of power" and its influence on other languages. It demonstrates that in a multilingual country such as India, the same language is not necessarily perceived as 'powerful' across time and space. Even within the same domain such as religion, Sanskrit is not the only language perceived as powerful across time as well as across different social groups in the speech community. It further shows that the attitudes of the speech community toward the 'power' of languages of religion registers have been primarily responsible for (a) the convergence and divergence of the religious registers of South Asian languages; (b) determining which languages would participate in the process of convergence; and (c) preventing some languages (e.g., English, and Arabic) from participating in this convergence in one geographic context (India) but allowing them to converge with Sanskrit in another (i.e., English-Sanskrit converges in the US). Finally, the paper points out that convergence creates variation and diglossia in the converging languages.

The Role of Language of Religion m 291

• 2. Sanskritization and de-Sanskritization of Modern Indian Languages It is a well-known fact that Hinduism does not prescribe one exclusive language for its expression. Sanskrit, as well as Modern Indian languages are viewed as legitimate languages of Hinduism. However, Sanskrit has always been considered the most sacred and prominent language of Hinduism to the extent that it has been called 'devavänV (divine language). Within the domain of Hinduism, Modern Indian languages converge as well as diverge. That is, they each have a Sanskritized register, which they commonly share and which shows a remarkable convergence at various levels, i.e., lexicon, metaphors, prosodic features, and religious/philosophical themes of Hinduism. In particular, the 'manipravala style' (literally, a mixture of gems and corals) in Tamil, Telugu, Kannada, and Malayalam specifically adopted by the Vaisnava school of Hinduism shows a mixture of Sanskrit and the Dravidian languages. Similarly, the Indo-Aryan languages each have a Sanskritized register. Thus, Indo-Aryan as well as Dravidian religious registers share a stock of Sanskrit lexicon, Hindu religious themes, metaphors, and prosodic features. However, Modern Indian languages have also developed corresponding native religious registers which typically lack linguistic features of Sanskrit and which significantly diverge from one another. In fact, it is observed that de-Sanskritization (which leads to regionalization, and, thereby, divergence of the languages) is as common among Modern Indian languages as their Sanskritization (which leads to their convergence). Although it may be claimed that the Sanskritized register is not limited to the domain of religion, but rather, it includes the non-religious domain as well, the religious registers differ from their non-religious counterparts in that they, unlike their non-religious counterparts, resist nativization of the Sanskrit material to a large extent and as a result, show a higher degree of convergence, i.e., the number of unassimilated Sanskrit lexical items is decidedly larger in the religious registers. Additionally, the religious registers share common themes, metaphors, and prosodic features which their non-religious counterparts do not. Moreover, the religious register is primarily used to express the Hindu identity and therefore, unlike its non-religious counterpart, it prohibits mixing of English and Arabic linguistic features since they mark Christianity and Islam respectively. The following discussion will examine the structure of the Sanskritized registers of the languages and the social motivations behind their emergence as well as the social motivations behind the emergence of the de-Sanskritized varieties.

• 3. Sanskritization of the Religious Register of Tamil: The Manipraväla Style The following discussion first describes the process of Sanskritization of Tamil and thereby the convergence of Sanskrit and Tamil in the religious register of

292 · Rajeshwari V. Pandharipande

Hinduism, and then proceeds to describe the process of de-Sanskritization of Tamil. The points of importance are: (a) the convergence is motivated primarily to assert the identity of the Tamil Hindu religiosity; (ft) the convergence is not homogeneous: there is variation within the structure of convergence; and (c) the convergence is not shared by all of the Hindu sects, rather, it is predominantly a feature of the Vaisnavite sect of Hinduism. Narayanan (1994) in her seminal work entitled, The vernacular Veda points out that the manipraväla style was a prominent feature of the Hindu Sriniväsa (vaisnava) literature. The beginning of the manipraväla style is traced to Tirukkurukai Pirän Fijian's commentary (llth century CE) on Tiruväymozi written in Tamil by Nammälvär. Narayanan (ibid.: 102) points out, Tiljän wrote his commentary in manipraväla, a new hybrid language of communication used in Srivaisnava circles. The Tamil of Nammälvär was translated and explained in a new "situational language." Manipraväla means "gems and corals" or "pearls and corals" and refers to a combination of Sanskrit and Tamil.' The reason for choosing to write the commentary in Tamil was to boldly establish the power of Tamil at par with Sanskrit for theological discourse. However, the mixing of Tamil with Sanskrit provided authenticity to the original work of Nammäjvär and expressed the identity of the Snniväsa Vaisnavite tradition. Narayanan (ibid.: 108) attributes the primary purpose of mixing Tamil with Sanskrit 'to show the equality of both Tamil and Sanskrit Vedas'. It should be noted here that the Tamil work Tiruväymozi is a treatise on Visisthädvaita philosophy of the Vaisnavites. According to Narayanan (1994), mixing Rämänuja's Sanskrit words from his original treatise on the philosophy of Visisthädvaita in his Tamil commentary is, 'an effort to show correspondence between the Tiruväymozi and Rämänuja's philosophy—and here the medium itself was the message: a harmonious combination of Tamil and Sanskrit'. In contrast to this, according to Thirugnansambandhan (1992) the motivation to mix Tamil with Sanskrit came from the desire to provide access to Sanskrit scholars (pandits) to the Tamil literature of the pre-Rämänuja period. According to Venkatacari (1978), the major impetus for the manipraväla style came from the 'consciousness of Tamil and Sanskrit as parallel religious languages in the writings of the Äjvärs. Although the ÄJvärs sing the praises of the Lord in Tamil, their mother tongue, they consider themselves part of the Vedic tradition.' According to this opinion, the mixture of Sanskrit and Tamil was intended to establish the dual (Aryan and Dravidian) heritage of the Tamilian Hindu religiosity1. Despite the difference in the opinions regarding the motivation behind the mixture of Sanskrit and Tamil, it is clear that the mixing placed Tamil at par with Sanskrit and established the religious identity of the Tamilian Hindus. What is interesting is that the structure of the mixture included enormous variation. For example, Pillan (the first Srmivasa to use this style) used Sanskrit words with Tamil endings (see Narayanan (1994)). The following example is from Pillän'sÄräyirapati (10:2:11) which is the first Vaisnava commentary in the manipraväla style.

The Role of Language of Religion m 293 (1) pnyatamarkalukku priyatamaikal bhogyamämäpöle tirunäteilullarkkum ittiruväymozi vallar bhogyamämävarenkirär. Tor the dear ones that which is desirable is enjoyable. In the same way, this composition of the composer of Tiruväymozi will be enjoyable for those who live in the divine place'. In the above example, the Sanskrit word priyatama which occurs twice in the first line is followed by the Tamil suffix ku 'for', and kal 'plural' respectively. Similarly, the other Sanskrit word, bhogya (which also occurs twice) is followed by the Tamil verbal suffix mä ('to be') -pole ('like'). In contrast to the above, a 15th-century manuscript located in the Adyar library which deals with the episode of killing of Kicaka in the Virätaparva of the Mahäbhärata, depicts a mixture of Sanskrit and Tamil where Sanskrit words notably lack Tamil suffixes and carry exclusively Sanskrit suffixes instead. The following example from Thirugnasambandhan (1992:119) illustrates this point. (2) muttukkalalum iva narpavalannaläum köttukkalanta tamill api samskrtena ettikkilum bhavatu häralateva baddhä puttikku matkrtir iyam sudrsam vibhäsä. 'May my composition where Tamil and Sanskrit are mixed and strung together like pearls and excellent corals mixed and strung together in a garland be an embellishment for the intellect everywhere.' In the above example, the Sanskrit and Tamil words exist side by side, each with their respective suffixes. For example, in the Sanskrit word bhavatu, the Sanskrit verb bhü 'happen' is followed by the Sanskrit optative, third person singular suffix -tu 'may happen'. Similarly, the past passive participle baddha (from the root bandh') 'having been strung together', is derived by adding the Sanskrit suffix -kta. In addition to these two extremes, there are various authors/poets who significantly differ from one another regarding the proportion of Sanskrit and Tamil vocabulary. Those differences further mark the sub-styles of the authors/poets of the manipraväla style. Some examples are given below2: the Sanskrit material in the following verses is underlined. Verse (3) is taken from Onpatinäyirappati (10:2:11) and Verse (4) is taken from Muppattäräyirappati (10:1:11). (3) mutivilläta pukalaiyutaiyanäyttiruvanantapurattile sannihitanäna jagatkäranabhütapurusanaisamrddhamäna tirunakariyaiyutaiya älvär arulicceyta äyiram liruväymoziyilum. The one who is responsible for creating the world, the one who is

294 · Rajeshwari V. Pandharipande

ever prospering, and who is endowed with eternal fame abides [literally, is placed] in Tiruvanantapura. This Älvär who belongs to Tirunagari, offers these thousand verses to him'. In the following example, there are only two Sanskrit words and the rest of the words are Tamil. (4) parama padattai vittu tiruvanantapurttile vantu ämtarkkäkak kanvalamtarulukaiyäle pukalakku orumutivillai yäyittu avarukku. 'He is of eternal fame who has given up the paramapada [Vaikuntha— the abode of gods] and has come to Tiruvanantapuram to protect and bless us who have taken shelter with him'. The manipraväla style shows variation in other Dravidian languages as well. For example, in the case of the mixture of Sanskrit with Malayalam, Sanskrit words are used with Sanskrit suffixes and Malayalam words with Malayalam suffixes.3 Moreover, the LÜätilakam, the grammar of the manipraväla, describes three types of styles: (a) uttama (the best): the one in which Malayalam words outnumber Sanskrit words, (b) madhyama (the medium): where Malayalam and Sanskrit words are equal in number, and (c) adhama (the lowest): where Sanskrit words outnumber Malayalam words. This categorization of the styles based on the proportion of the Sanskrit and Malayalam reflects the attitude of the people toward the converging languages. While the prominence of Malayalam is most desirable, the prominence of Sanskrit is considered most undesirable. This clearly shows that the major goal of the manipraväla was to establish Malayalam as the language of religion at par with Sanskrit. While the alignment with Sanskrit (in the mixture) gave legitimacy to the religiosity, the maintenance of the prominence of Malayalam established the regional identity of the Hindus in Kerala. The manipraväla style in Telugu and Kannada also shows variation in its structure. What is commonly shared is mixing with Sanskrit linguistic structure as well as the common Hindu religious themes from the Sanskrit religious literature such as the epics (theRämäyana and the Mahäbhärata), the Puränas, the Vedas, and the Upanisads. However, it is interesting to note that there seems to be a mixing of the indigenous (Dravidian) religious themes with the Indo-Aryan themes as well (for further discussion on the thematic convergence see Hart [1976], and Narayanan [1994], Pandharipande [forthcoming]). The manipraväla had a threefold impact on the Dravidian languages: (a) it created variation in the respective languages as well as within their respective religious registers; (b) it affected the structure of Tamil, i.e., it brought a new set of Indo-Aryan vocabulary (related to religious domain), phonology, syntax, and metaphors, thereby creating variation in the structure; (c) the mixed code functioned as an exponent of the mixed identity (Dravidian-Aryan) of

The Role of Language of Religion m 295 the speech community; (d) it further separated the elite from the non-elite: since Sanskrit was used exclusively among the elite Brahmin community, the Tamil-Sanskrit mixed code was also seen as a marker of the elite group as opposed to the common people within the Tamil religious community; (e) the manipraväla style is mostly restricted to the written religious texts and is not used as much in the spoken language (see Sridhar 1981; Thirugnanasambandhan 1992; and Zvelebil 1974); (/) 'Sanskritization' introduced new phonemes such as aspirated stops into the religious register of Dravidian languages (for further discussion, see Sridhar 1981: 206-210), and finally; (g) the Sanskritized variety of Tamil and other Dravidian languages converged with the Sanskritized varieties of Indo-Aryan languages. Thus, Sanskritization can be viewed as the Indo-Aryanization of Dravidian languages. In contrast to the manipraväla, which uses the Tamil-Sanskrit mixed code, there exists an indigenous Tamil religious register used in the Teväram (etymologically, private ritual worship') texts of Tamil Saiva religiosity expressed in Tamil (as opposed to any other language). This register shows a deliberate attempt to wipe out the influence of Sanskrit from the structure of Tamil. The following discussion focuses on the motivations behind the emergence of this purely Tamil register of Hinduism which can be viewed as the process of deSanskritization. The major motivations behind the emergence of the purely Tamil (as opposed to the Tamil-Sanskrit) register was (a) to consolidate and establish a regional as opposed to pan-Indian religious identity of the Tamil community, (b) to establish a Dravidian as opposed to Indo-Aryan religious identity, and (c) to establish the identity of the common people within the Tamil community. These assumptions are supported by Peterson's study (1989). According to Peterson (1989: 9), The Tamil movement is better characterized as a movement toward communal solidarity than an expression of social protest.' Peterson (1989: 5) further claims, As the first literary expressions of a popular regional religious culture in Hinduism, the Teväram hymns reflected many features of Tamil culture, and differed in many respects from the authoritative sacred texts of the Great tradition, composed in Sanskrit, the classical language of the Brahmanical civilization. The three major early poets of the Teväram texts are Appar, Campantar, and Cuntarai (6th-7th century CE). However, the canonization of the Teväram took place much later (14th century CE). In contrast to the, manipraväla, the Teväram prefers the use of Tamil although the context is primarily that of the traditional religious themes. For example, the Tamil word vinai for the Sanskrit word karma, vitu for moksa 'liberation', mal for Visnu (Peterson 1989: 83). What we see here is that two registers, one code-mixed and the other 'pure Tamil' (centamiz) coexist with mutually exclusive

296 · Rajeshwari V. Pandharipande

forms and functions. Consider the following example of Teväram taken from one of later (9th century) poet Manikkaväcakar (Tiruväcakam 2):4 (5) küvinapünkuyil; küvina köli; kurukukal iyampina, iyampina cankam; ovinatärakai, oli; oli utayat toruppatukinratu, vintppotunamakkut tevatarcenkalarrälinai kätläyl tirupperunturai uraicivaperumane\ yä varum arivariyäy! yemakkeliyäyl emperumän! palli eluntarulayel The tender koel's note is heard; the cocks have crowed; the little birds sing out; sound loud the tuneful shells; starlights have paled; day's lights upon the eastern hill are mustering. In favoring love show to us your two feet, anklet-decked, divinely bright; Lord Siva, in Perunturai's hallowed shrine who dwells! All find hard to know you; easy to us your own! Our mighty Lord! From off your couch in grace arise!' The Teväram represents a fairly common trend in Dravidian languages, i.e., there were several anti-elite movements to assert the local common people's religious identity and its expression in their own language in contrast to Sanskrit, which represented the elite, the Brahmins, and Indo-Aryan identity. Thus the Virasaiva (12th century) resistance to Sanskrit as an expression of anti-Brahmanism resulted in the composition of vacanas in Kannada which is relatively free from Sanskrit (for further discussion see Sridhar 1981). Similar developments are observed in Telugu, and Malayalam where the 'purist' movements succeeded in counteracting Sanskritization of the religious registers (for further discussion, see Zvelebil 1974). The above discussion shows that the same language is not perceived as powerful in all domains. While Sanskrit was perceived as powerful for expressing the elite Hindu identity, it was perceived as powerless for expressing the regional Tamil/Dravidian Hindu identity of the common people.

• 4. Sanskritization of the Religious Register of Marathi: The Panditikävya The following discussion focuses on the convergence of Marathi and Sanskrit in the context of the religious register of Hinduism. Early Marathi literature is traditionally divided into two major categories: Santakävya and Panditikävya (for further discussion, see Watwe 1964) based not so much on their content but on their linguistic form. While the former favors native Marathi linguistic

The Role of Language of Religion m 297 form (vocabulary, phonology, syntax, metaphors, and discourse strategies), the latter shows a marked predilection for the Sanskrit linguistic form. The term Santakävya primarily refers to the religious poetry (kävya) of the poetdevotees (Santa) of the Värkanpantha such as Jnänesvar, Nämdev, Eknäth, Tukäräm, and Rämdäs, as well as the women devotees such as Janäbäi, Muktäbäl, Bahinäbäl, and Venuäkkä. In contrast to this, the term Panditi kävya refers to the religious poetry (kävya) of the poets who claimed to be more Pandit (the learned ones who were well-versed in the Sanskrit language and the literary styles) than devotees. The goals of the two types of religious poetry significantly differed from each other. The Santakävya forcefully articulated that Marathi, similar to Sanskrit, was capable of expressing the religious experience of the devotees. Moreover, it argued for the need for bringing the Hindu religious philosophy and devotional Sanskrit literature to the common people (in their language, i.e., Marathi) who were deprived of the access to Sanskrit. Thus, in this poetry, there is a conscious effort to depart from the established Sanskrit linguistic paradigm/idiom of the expression of the religious thought. However, the intention of establishing a new indigenous religious sect was not necessarily shared by all the poet-devotees. Therefore, the Santakävya is viewed as an expression of Hindu religiosity in the native Marathi idiom. Most of the poetry written by the poet-devotees is either a commentary on the Sanskrit texts (e.g., Bhävärthadipikä—Jnänesvar's Marathi commentary on the Bhagavadgita) or description of their own religious experience. The major motivations for the use of Marathi as opposed to Sanskrit were varied: (a) to establish the power of the Marathi form at par with Sanskrit, (b) to express the religiosity of the common people in their own language, and (c) to bring the knowledge of the Hindu tradition described in the Sanskrit text to the common people in the Marathi speech community. The following examples clearly indicate the intent of the poets: Example (6) is taken from Jnänesvar's commentary (Bhävärthadipikä) on the Sanskrit text the Bhagavadgita:5 (6) mädzhä marhätäci bolu kautufce pari amrtätehl paidzä jinke aisi aksare rasike, melavlna. 'My speech (is) indeed Marathi but it will win the competition with ambrosia in admiration. I will present such speech to the connoisseurs (Jnanesvarl: 6:14). Example (7) is taken from Sridhara's Rämvijay: (7) abla na kale samskrta vani, dzaise ädätale nirmala päni pari dora päträvätsoni, asaktacanä kevi nighel 'There are people who are not capable of understanding Sanskrit [which is for them] like the pure water in the deep well. Without the rope and the pail, how can it [water] be brought to them?' (Rämvijay: 189)

298 · Rajeshwari V. Pandharipande

Example (8) is taken from Eknäth's Bhägvat: (8) samskrta väni deve kell, präkrta tan tsoräpäsum all 'Sankrit was created by the gods, and Prakrt came from the thieves?' (EknäthiBhägvat: l) Note that examples (6), (7), and (8) clearly articulate the motivations mentioned earlier behind the use of Marathi as opposed to Sanskrit. In contrast to the above, the pandit ka vl wanted to compose religious poetry for the vidvajjana 'the learned ones', vyutpanna 'the accomplished ones', and the rasajna 'who have inculcated the ability to appreciate the rasa (the sentiment) expressed in the poetry'.6 Therefore, the emphasis of the religious poetry of the pandit kavi such as Narendra (13th century CE), Bhäskarbhatta (fourteenth century CE), Vaman Pandit, Nagesh and Vitthal (16th century CE), and Moropanta (17th century CE) was on deliberate imitation of the Sanskrit structure at various linguistic levels such as vocabulary, phonology, syntax, metaphors, prosody, discourse strategies, etc. Thus, similar to the manipraväla style in the Dravidian languages, we find a marked mixture of Sanskrit and Marathi in the language of pandit kavL The following examples illustrate the Marathi-Sanskrit mixed structure. The poem is taken from Vaman Pandit's Nämasudhä:! (9) haricaranasaroji citsudhäsärapäne^ bhramara paramayogi gunjafi nämagäne, harati_ sakala dukkhä h[ sukhaci^ nidhäne vada vada vadajivhe! väsudeväbhidhäne_. At the lotus-feet of Hari, the saint-bees, having drunk the nectar of the divine consciousness, sing the songs of gods' names. These (names) which are the source of happiness, remove all pains. tongue! say, say, say, various names of Vasudeva'. haricaranasaroji (loc) 'at the lotus-feet of Hari'; citsudhäsärapäne (pl) 'because of drinking the nectar of 'cit (divine consciousness)'; paramayogi bhramara 'the saint-bees'; gunjati nämagäne 'sing the songs of (gods') names'; hi 'these'; sukhacl (poss) Of happiness'; nidhäne (pl) 'abodes'; harati (3. pl.) 'these remove'; sakala dukkhä 'all pains'; vada, vada, vada (voc) 'say, say, say'; jivhe , tongue!'; Vasudeväbhidhäne (pl) 'various names of Vasudeva' The above poem has all but the underlined material in Sanskrit. Marathi suffixes are attached to the Sanskrit nouns as well as verbs. Now note the following example where the style of compounding nouns in Sanskrit is observed in Marathi. The verse is taken from Moropanta's Kusalavopäkhyäna (3:11). (10) evam vadom bharate kele raraasi vandana thus say- part bharata did rama-dat salute

The Role of Language of Religion m 299

dzhälä asrujjalabhrastahrdayäliptacandana. became tears-water-wiped-chest-anointed-sandalwood 'Having said this, Bharata prostrated before Rama. The sandalwoodpaste on his chest was wiped out by tears/ The long compound asrujjalabhrastahrdayäliptacandana is a direct borrowing from Sanskrit which is quite unnatural in Marathi. The extreme case of this mixture is observed in (11): the Marathi verse has a complete sentence in Sanskrit (underlined). Consider the following example from äryäbhära: ädiparva: 31: 27. (11) aisemhanatä stricyä kanthl ghälunl mithl thus saywhen woman-poss neck-loc having placed embrace mhane ayite said o, you kathamevamuktam tadadayam kimanusthitam ksamo dayite. How- thus-said that-cruel q-carry out-part capable woman! 'when (she) said this, having embraced her he said, "how could you utter such cruel (speech)? How could (I) carry this out? (how will I be capable of carrying it out?).'" This Sanskritized register of Marathi not only created a contrast between religious and non-religious registers of Marathi, but it also marked a difference between Sanskritized and non-Sanskritized religious registers of Marathi. The non-Sanskritized register of Marathi can be seen in the following example from Janäbäi's poem from Nämdevämci gäthä: 948: (12) may a geli bäpa gelä, ätä sämbhäll viththalä mi mother died father died now take care Viththal-voc I tudze gä lekaru, nako malä avherü your indeed child, do not I-acc reject 'My mother and my father died; now O Viththalä! Take care of me. I am your child; do not reject me.' Note that there is not a single Sanskrit word in the above poem. However, there are other religious poems by many poets which fall between the two, that is, highly Sanskritized and non-Sanskritized registers. For example, Rämdäs' poem given below from Däsbodh (14:3:27) shows some Sanskrit vocabulary. However, the underlined Sanskrit vocabulary is very commonly used in nonreligious register and it is not recognized as non-native (Sanskrit in this context). (13) bhagavanü lägale mana, tethe näthave god-loc attached mind there neg-remember

300 · Rajeshwari V. Pandharipande

dehabhäna, samkä lajjä paloni dun thell, body-consciousness doubt shyness ran away remained to premaramge ramgalä he love-color-inst besmeared 4 When the mind got absorbed in the god, the body-consciousness was lost. Doubt and shyness also disappeared completely. He got immersed in the color of love [of god]'. Several points regarding the Sanskritized religious register need to be mentioned here: (a) The Sanskritized register was used by the * elite' poets who were well-versed in both Marathi and Sanskrit literary and religious traditions; (b) the audience was also expected to be the 'elite' (mostly Brahmins): in contrast to this, the non-Sanskritized register was used for the sake of the common people of all castes (including Brahmins) who did not understand Sanskrit; (c) in the Sanskritized register, the phonology of Sanskrit is maintained for the Sanskrit material: in particular, the word-final vowel is retained even in the absence of the consonant cluster preceding it, and the retroflex /r/, and Isl sounds are retained in all positions; (d) Marathi has maintained the contrast between Idzl and Ijl and between ltd and Id in the Marathi material in the mixed code; (e) two types of patterns are observed: Sanskrit nouns/verbs take Marathi suffixes, i.e., the Sanskrit vocabulary is nativized or the whole Sanskrit phrases (mostly compounds) or sentences are mixed with Marathi— In the case of the latter, rules of compounding in Sanskrit are applied; and, finally (/) Sanskritization created variation in Marathi both at the formal and the functional levels. Three major codes were used in the religious contexts as shown in (14). (14)

Religious register Sanskrit

Sanskritized Marathi

Non-Sanskritized Marathi

While Sanskrit was used exclusively in the context of Vedic rituals (marriage, funeral, naming ceremony, etc.), Sanskritized register was used for household religious rituals (daily prayers, religious festivals, etc.), as well as religious discourse and; non-Sanskritized register for local/regional rituals/ festivals of Maharashtra or regions within Maharashtra (Nagpur, Khandesh, Aurangabad, etc.). The use of English was completely prohibited in all religious contexts. It is very clear that each register was aligned with a particular kind of religious identity. Sanskrit with the elite, pan-Hindu identity, Sanskritized Marathi with the elite Maharashtrian identity, and non-Sanskritized Marathi with the more local religious identity. English was viewed as the language of the mlecchas 'barbarians' and it was considered to be "polluted" and therefore, its use was prohibited in the religious/sacred domain. It must be noted here that the registers were aligned with certain themes as well. While

The Role of Language of Religion m 301 the Sanskrit and Sanskritized registers showed considerable resemblance with the Vedic, Upanisadic, epic, and Puränic themes, the non-Sanskritized register showed a marked predilection for innovation and freedom regarding themes, local legends, worship of local deities. Also, regionalisms are observed at the formal levels of the language as well. The non-Sanskritized register shows relatively more dialectal features of phonology, morphology and syntax than its Sanskritized counterpart. The form of Marathi which is mixed with Sanskrit represents the contemporary form of the elite speech. The most fascinating feature of this functional distribution of the religious registers is the change in the attitudes of the people toward what is viewed as the elite canonical Hindu text. Jnanesvari, which was considered to be a text of lesser prestige (since it was written in Marathi as opposed to Sanskrit at the time of its inception), gained the position at par with the canonical Sanskrit (Hindu) texts such as the Bhagavadgiia over a period of time. Recitation of Jnänesvari, similar to the Bhagavadgita, is part of the rituals in the contemporary Maharashtra. This is not surprising in the context of Hinduism where a linguistic form can be authenticated as a legitimate vehicle of religious discourse if its user is accepted as a saint (who is believed to have had säksätkära 'direct experience of the divine'). Since Jnänesvar is viewed as one of the most revered saints of Maharashtra, it is not surprising that his language is viewed as sacred and his text a scripture. In addition to the above religious registers, Marathi has developed a popular register of bhajans (devotional songs), and klrtan which is a religious discourse in a mixture of prose and verse which illustrates religious themes in the context of the current social situation. The differentia of this register is that its language is the local dialect and the metaphors, and discourse strategies are based on the experiences of the people in the current social context. Linguistically, this is the least constrained religious register where various dialects and languages from the linguistic repertoire of the speech community are freely mixed.

• 5. Sanskritization, Convergence and Diglossia: The Case of Hindi In the following discussion, it will be pointed out that Hindi shares some features of religious registers of Marathi and Tamil, but it also significantly differs from them in that the linguistic repertoire of the respective speech communities varies across Hindi, Marathi, and Tamil speech communities and therefore, the nature of convergence and diglossia also varies across these languages. In my earlier work (Pandharipande 1992), it is pointed out that starting from the 9th century CE the use of Hindi as opposed to Sanskrit was inspired not so much as a revolt against Sanskrit, but rather, as a means to make the Hindu system of belief accessible to common people in their own linguistic medium, to defend or preserve Indian cultural or religious identity from the

302 · Rajeshwari V. Pandhanpande

power of Buddhism (9th to 10th century CE), Islam (1200-1700 CE), or Christianity (19th century CE), and to consolidate diverse sects of Hinduism under the umbrella of the same language (sampradäyaväd 'integrative approach'). Unlike Marathi and Tamil, religious literature in Hindi was complementary to the earlier Sanskrit literature. Vidyäpati (14th century CE) in his Kirtilatä clearly composed in Apabhramsa, makes this point when he says, 'Sanskrit appeals to the learned, but who does not grasp and relish natural speech? To everyone the speech of his region is sweet, and so one should speak in Avahattha'1 From the 9th century CE till today, Hindi religious literature is primarily based on the religious themes from the original Sanskrit scriptures such as the Vedas, Upanisads, Puranas and the epics. All through the sectarian (Saiva and Vaisnava) Hindu literature, Hindi maintains a Sanskritized style which involves vocabulary, syntax, and metaphors from Sanskrit. The following example is from Tulsidas' (15th century CE) Vinay Palrikä (pad: 43)8. (15) vedabodhita-karma-dharma-dharam-dhenu-vipra-sevaka· sadhumodakari Veda-instructed-action-duty-earth-cows-brähmins-servants-saintsjoy giver 'He makes the (whole world) earth, cows, brahmins, and servants happy by performing actions, [religious] duties which are instructed by the Vedas/ The following example illustrates modern religious prayer in Hindi: (16) jay jagadis harey svarni jay jagadis hare bhaktajanö ke samkatpal me dür kare. 'Victory to you Hari, who is the lord of the world. O Hari, you are the controller, the lord of the world. You instantaneously remove the troubles of your devotees.' In addition to the above, Hindi shows tremendous variation in the codes used in the religious domain. The term 'Hindi' as a language of Hinduism does not refer to a single code but rather, it refers to a group of languages which are used by various authors for their compositions. Sürdäs (1478 CE) chose Braj for his work Sürsägar devoted to Krsna; Miräbäi (1504 CE) used Braj mixed with Rajasthani for herpadas', Tulsldas (1554 CE) composed Rämcantmänas in Avadhl; and Vidyäpati (1350 CE) wrote his Padävall in Maithili. These languages share a number of common grammatical features. In the non-religious context, the speech communities view them as separate languages. However, in the context of religion, these are not perceived as separate languages. Thus, due to the shared religious beliefs, the formal divergence of these codes is overlooked and they seem to converge as varieties of the same code. It has

The Role of Language of Religion m 303 been argued before (Pandharipande 1992: 280) that shared religious beliefs should be included into the parameters for defining a speech community. Similar to the situation of Marathi and Tamil, the above codes (Sanskritized Hindi, Braj, Avadhl, Raäjasthänl, and Maithill are not used randomly, rather, their distribution is determined across religious themes and social/religious setting. In my earlier work (ibid.), I have illustrated the functional distribution of languages according to the theme and religious setting. This distribution is shown in (17) below (for further discussion see Pandharipande [1992: 282]). (17)

Theme

Setting

Language

1. Ram (Ultimate Reality with attributes)

(a) Vedic rituals or chants

Sanskrit

(b) Non-Vedic* rituals or chants

Sanskrit or (Sanskritized) Standard Hindi Avadhl or other dialects of Hindi Sanskrit

2. Krsna

3. Brahman

(c) Domestic* rituals or chants (a) Vedic rituals or chants (b) Non-Vedic* rituals or chants (c) Domestic* rituals or chants Vedic rituals and chants

Sanskrit and/or Sanskritized Standard Hindi Braj or other dialects of Hindi Sanskrit

* There is a great deal of overlap between these two settings. However, the important point is that the languages in question (Standard Hindi, Avadhi, Braj, etc.) are not freely interchangeable. Also, it should be noted that the themes in 1 and 2 represent the Ultimate Reality with attributes (saguna) while the theme in 3 represents the Ultimate Reality without attributes (nirguna).

• 6. English-Sanskrit Convergence: Hinduism in the US The above discussion shows that the convergence of Indian languages via their convergence with Sanskrit was primarily motivated by the attitudes of the speech communities of Modern Indian languages toward Sanskrit which has always been viewed as the most sacred language of Hinduism and thereby their Hindu identity. In contrast to this, English, Arabic, Urdu, and Persian

304 · Rajeshwari V. Pandharipande

were excluded from this convergence due to their affinity with other religions (i.e., Christianity and Islam). However, the English-Sanskrit convergence which has occurred in the US is making its inroads in India, though very slowly. At this point, this English-Sanskrit mixed code is freely used only at the Hare Krishna temples (which originated in the US), the Säl Bäbä centers as well as other Hindu temples in the US. In my recent work (Pandharipande 1998), I have shown that Hindu immigrants as well as the American-born (second/third generation Hindus and the native Americans) Hindus use English as their language of religion. While the major rituals at the Hindu temples or at homes are generally performed in Sanskrit by the Hindu priests, the devotees themselves do not hesitate to use English translations of the Hindu texts as scriptures and they freely read these translations as prayers or religious chanting. Similarly, in religious rituals during weddings, English is mixed with Sanskrit to make it easy for the rituals to be understood by the bride, groom, and other people who participate in the ritual. Similarly, in a religious congregation of Satya Sai Baba in the US, I have observed the devotees singing the prayers in English mixed with Sanskrit vocabulary, prosody in the tune of traditional Hindu prayers (bhajans). English is no more considered an 'impure language of themlecchas (barbarians)', rather, it is viewed as one of the legitimate languages of Hinduism. The convergence of English and Sanskrit in this case is most interesting. When the priest mixes English with Sanskrit, he imposes Sanskrit (or other Modern Indian language) phonology and intonation on the English material. In contrast to this, when the devotees mix Sanskrit with English, they impose English phonology and intonation on the Sanskrit material. The examples given below are from the text of the bhajans (religious prayersongs) devoted to a modern Indian Hindu saint of high repute, Satya Säi Bäbä.9 The bhajans include devotional songs in various languages of India including Sanskrit and English (in addition to other Modern Indian Languages). Those songs are sung alternatively in traditional bhajan rhymes and tunes. It is interesting that the English songs 'sound' like traditional Hindu bhajans in Sanskrit or other Indian languages. Another type of bhajans involves mixing Sanskrit with English and other languages. Example (18) is a prayer entirely in Sanskrit to the goddess Durgä. The prayer consists of various names which also serve as attributes of the goddess. (18) Sanskrit: Santa durge sänti svarüpim; syämalä komalä mana mohini; ädi anädi ananta svarüpim; moksa pradäyini näräyani peaceful Durgä! [you are the one] of the peaceful nature. You are the dark, delicate, and enchanting. You are the origin [of everything and yet] you are without the beginning and the end. O Näräyani! You are the granter of final liberation!'

The Role of Language of Religion m 305 The following example (19) is a prayer recited at the same congregation of Säi Bäbä. Note that the whole prayer is in English. (19) English We are calling Säi, Be with us [three times] In our hearts My Lord, you're the light [three times] lord you are the light You are Shining Lord, the Whole world through— Fill our Hearts with love, Expand the light [two times] Lord, Expand the light Example (20) is also a prayer which is included in the collection of prayers sung at the same congregation. Note that language clearly involves mixing of English with Sanskrit. The Sanskrit words/phrases are underlined and translated. The rest of the prayer is in English. (20) Code-mixing (English-Sanskrit) I am God, I am God, I am no Different from God am the infinite supreme, the One Reality am Satcidänanda Svarüpa (the one who is of eternal, conscious, and blissful nature) am Om tat Sat Om (that eternal existence Om) am love, I am Truth, I am Peace Eternally am Ever Pure Delight, I am Always Full and Free Fear or Grief Can Never Touch me I am Om tat Sat Om (that eternal existence Om)

• 7. Convergence and Divergence: Attitudes, and Impact The discussion in the preceding sections shows that the processes of Sanskritization (and thereby, convergence) and de-Sanskritization (and thereby, divergence) of South Asian languages are primarily motivated by the attitudes of the speech communities toward Sanskrit and the regional (Modern Indian) languages. When Sanskrit is perceived as 'powerful' in the religious context, Modern Indian languages are mixed with Sanskrit and the result is Sanskritized varieties of South Asian languages. The major reasons for the perception of Sanskrit as more powerful than the regional languages were (a) Sanskrit was the language of the ancient Hindu scriptures (the Vedas, Upanisads, epics, etc.); (b) it was the language of the elite/ priestly Brahmin class in the society, and; (c) Sanskrit indexed trans-regional or pan-Indian Hindu tradition. Thus, Sanskritization was perceived as the process of empowerment of regional languages, through their connection with the Great tradition beyond regional Hindu traditions. In Tamil, Marathi, and Hindi speech communities, Sanskritization also indexed legitimization of the use of regional languages (mixed with

306 · Rajeshwari V. Pandharipande

Sanskrit) in the context of religion at par with Sanskrit. Thus, it was not a surprise that English did not participate in the process of Sanskritization in India since English was perceived as totally powerless in the context of Hinduism. It was viewed as the language of the non-Hindus (Christians). However, in the US. where English is primarily the medium of expression of the second generation Hindus as well as the American followers of Hinduism (e.g., the Hare Krishna devotees), and where English serves as the major language for consolidating the linguistically diverse Hindu community, it is viewed as one of the languages used by the Hindus. In this case, Sanskritization of English is guided by two major factors: (a) empowering of English in the religious context, (ft) making the Hindu tradition of Sanskrit texts accessible to the Englishspeaking Hindu community in the US, and (c) maintaining authenticity of the Hindu tradition through'the use of original Sanskrit words, phrases and excerpts from the ancient scriptures. The following aspects of the impact of this convergence at the macro and micro levels need to be noted here: (a) The impact of Sanskritization is not necessarily the same across languages at the structural and functional levels. While 'Sanskritization' in Marathi and Hindi marks the retention of the wordfinal vowel in the absence of an immediately preceding consonant cluster (e.g., sara 'arrow', mastaka 'head', kam 'hand', etc.), and in Kashmiri, it refers to the introduction of voiced aspirates (bhaväni 'Siva's wife', mürdhä 'head', etc.). Also, it is important to note that the above process is not found in the speech of all speakers, rather, it is largely restricted to the elite class in the speech community, (b) This Sanskritization marks the register of Hinduism distinctly different from the Christian and Islamic registers of these languages (which are marked by Englishization and Arabicization respectively), and thereby, creating a diglossic situation in the converging languages, (c) In addition to the religious identity, this register further represents a caste (Brahmin, among Tamilians, Andhras, Kannadigas, and Malayalis, Maithils, Bengalis, etc.), or the 'elite' speech in general (as among the Maharashtrians, Gujaratis, etc.). (d) Since English, and Arabic have been perceived as languages of Christianity, and Islam respectively, they were not used for expressing Hinduism and consequently, the question of a Hindu religious register did not arise in India. However, the emergence of the Hindu register of English in the US is the result of Sanskritization of English in the context of Hinduism, (e) This process of convergence of the South Asian languages has created diglossia at various levels (caste: Brahmin vs. non-Brahmin, elite vs. non-elite, written vs. spoken, and most importantly, religious vs. non-religious registers). (/) additionally, Sanskritization does not perform the same social function across different speech communities. For example, while Sanskritization indexes IndoAryanization of Dravidian languages and Dravidian Hinduism (in addition to other functions mentioned above), it indicates the Hinduization of English in the US and Sanskritization of Indo-Aryan languages marks the effort to place regional languages at par with Sanskrit—the language of the elite.

The Role of Language of Religion m 307

The following discussion will focus on the process and impact of de-Sanskritization on the Modern Indian languages. As it is pointed out in earlier sections, most of the South Asian languages have a non-Sanskritized counterpart of the religious register which does not have any or only minimal features of Sanskrit. De-Sanskritization, similar to Sanskritization, was motivated by diverse socio-linguistic factors. As it is shown in the earlier discussion, the historical reasons for the development of non-Sanskritized registers vary across languages and their sociocultural contexts. Some of the major motivations can be summarized as follows: (a) to bring the religion to the masses (who did not have any access to Sanskrit) in their own language. For example, the emergence of the 'pure Tamil' variety of Teväram (as interpreted in Peterson 1989). Also, by using Marathi as a language of his commentary (bhävärthadipika) on the Bhagavadgitä, Jnänesvar brought the Bhagavadgita to the common people. (b) To provide legitimacy to the use of regional languages for the expression of religiosity of the common people. For example, the Varkari movement led by Jnänesvar (thirteenth century) boldly articulated the power of Marathi at par with Sanskrit, (c) to index the identity of the common people as opposed to the elite Brahmins. Thus, the Teväram literature in Tamil as well as Tukäram's and Rämdäs' religious compositions in Marathi are viewed as the linguistic as well as the social revolt against the Brahmanic olite classes to establish the power of the common people, (d) To express regional (Dravidian) as opposed to panIndian/Indo-Aryan Hindu tradition, (e) To express religious emotions/experience at a very personal/individual level which could only be done in one's own native/regional language (as opposed to Sanskrit). For example, as shown earlier, the devotional Hindi poetry in the 15th-16th centuries was composed in various dialects of Hindi as opposed to Sanskritized Hindi. The major impact of this de-Sanskritization was: (a) the languages developed religious registers which diverged from one another giving rise to diglossia within the domain of religion (the use of Sanskritized register in the Brahmin/elite classes within certain branches of Hinduism and the use of its nonSanskritized counterpart in the non-Brahmin/non-elite classes within certain other branches of Hinduism, (b) the process of de-Sanskritization involved various strategies such as using native/non-Sanskrit lexicon, nativization of the Sanskrit lexicon, native phonology, metaphors, prosody, and modification of the religious themes of the classical Sanskrit religious literature (e.g., the epics—Mahäbhärata, and the Rämäyana, Puränas, etc.), and (c) not all languages used all of these strategies; they used one or more of these, at times, mutually exclusive strategies.

• 7. Conclusion There are several implications of the discussion in the preceding sections: (a) The convergence of languages which results due to the individual, independent

308 · Rajeshwari V. Pandharipande

convergence of the each of the languages with a common language (Sanskrit in this case) differs from the convergence due to direct geographic proximity of the languages. In the former, there is not necessarily a mutual exchange of linguistic features, rather, what they commonly share are the features of the language (Sanskrit in this case) with which they independently converge. (b) It should be noted here that while Sanskritization involves convergence with Sanskrit of only the respective high varieties of the languages, deSanskritization is not restricted to the replacement of Sanskrit by the respective high varieties of the languages, rather, various varieties, dialects (with high or low prestige) are freely used in this process, (c) Although Modern Indian languages share a common Sanskritized register, they significantly vary in their degree and type of convergence with Sanskrit and thereby with one another. In other words, the discussion clearly shows that the Sanskritized registers of different South Asian languages do not necessarily share the same linguistic features of Sanskrit, i.e., adoption of Sanskrit vocabulary with its Sanskrit suffixes vs. adding native suffixes/prefixes to the Sanskrit vocabulary. Therefore, it is necessary to differentiate between the two types of convergence mentioned above, (d) Finally, the discussion in this paper points out that the attitudes of the people toward the relationship between the linguistic form and its function/meaning play a crucial role in determining the structure of convergence. In other words, the discussion shows that the 'power' of language is not a discrete category and therefore, the current research (Bourdieu 1991 and Fairclough 1989, 1998) on language of power needs to take into account the fact that the same language can be perceived as both powerful and powerless in different context of time and space. While Sanskrit is perceived as powerful in the context of pan-Indian Hinduism of the Great tradition, it is perceived as powerless to express the regional dimension of Hinduism of the common people. Similarly, over a period of time, the regional languages acquired the prestige of'sacred' languages (recall, the case of Avadhi (the language of Rämcantmänas, and the Marathi oiJnänesvari). Similarly, English which was perceived as powerless in the context of India, is perceived as a legitimate language of Hinduism in the US. It is also demonstrated that these perceptions of language of power influence language change, in particular, convergence in this context. For example, when it is believed that Sanskrit can be the only medium for expressing the religious meaning of Hinduism, the languages tend to develop Sanskritized registers, while the belief (expressed in the anti-Sanskrit movements) that any Indian language is capable of expressing religious meaning results in de-Sanskritization/divergence of religious registers. In the US, where the primary language of communication is English, the range of languages, which are perceived as legitimate for expressing the religious meaning, is further expanded and the convergence of English and Sanskrit becomes acceptable.

The Role of Language of Religion m 309 NOTES 1. It is necessary to note here that the manipraväla style was not restricted to the Hindu texts. The early Jain texts in Tamil used a mixture of Sanskrit and Prakrt (for further discussion see Venkatacari 1978). 2. These three examples are taken from Venkatacari (1978: 172-73). I am thankful to Ms Kalpagam Venkataraman for her help with the English translations of these Tamil verses. 3. In addition to manipraväla, there exists another style calledpattu in Malayalam where the Sanskrit words are used with Malayalam suffixes. This style is used in the lyrics with Malayalam rhymes. 4. This translation is taken from Manikkavacakar, Tiruväcakam or 'Sacred Utterances', by G.V Pope (1900:208-9). I am thankful to Professor Rajagopala Parthasarathy who has kindly provided the Tamil text as well as the translation of the text of this verse. 5. This verse is taken from Särtha Jnänesvari (6:14) Dandekar (1984: 138). 6. Examples 7, 8, 9, 10, 11, and 13 are taken from Watve (1964). Example 12 is taken from Bhingarkar (1989). 7. This quote is taken from McGregor (1984: 30). The original verse is as follows: sakkaya buhäna päü rasa komamma na pävai desila vayanajana tarn taisana jampaü avahatthä. 8. This verse is quoted in Omprakash (1971:126). 9. These three examples 16,17, and 18 are taken from the text (put together in 1977 in the mimeographed format) entitled, 'Sai Devotional Songs' which is distributed from the Säi Bäbä center around the world. This text is used at the regular meetings of the devotees.

• REFERENCES Bhingarkar, Damodar, B. 1989. Santa kavayiiriJanäbäi. Mumbai: Majestic Prakashan. Bourdieu, P. 1991. Language and symbolic power. Cambridge: Polity Press. Bright, William. 1968. Toward a cultural grammar. Indian Linguistics 29 (1). 20-29. Dandekar, Sakharam Waman. 1984. Särtha Jnänesvari. Pune: Svananda Prakashan. D'Souza, Jean. 1992. South Asia as a sociolinguistic area. Dimensions of sociolinguistics in South Asia: papers in memory of Gerald Kelly, ed. by E.C. Dimock, Jr, B.B. Kachru and Bh. Krishnamurti, 15-24. New Delhi: Oxford & IBH Publishing House. Emeneau, Murray B. 1956. India as a linguistic area. Language 32 (1). 3-16. Fairclough, Norman. 1989: Language and power. London: Longman. . 1998. Power and language. Concise encyclopedia of pragmatics, ed. by Jacob L. Mey. Amsterdam: Elsevier. Fergusson, Charles, A. 1992: South Asia as a Sociolinguistic area. Dimensions of sociolinguistics in South Asia: papers in memory of Gerald Kelly, ed. by E.C. Dimock, Jr, B.B. Kachru and Bh. Krishnamurti, 25-36. New Delhi: Oxford & IBH Publishing House. Hart, George Luzerne. 1976. A history of Indian literature: the relation between Tamil and classical Sanskrit literature. Gonda, Jan (ed.) Vol. 10, Fasc. 2. Wiesbaden: Otto Harrassowitz. Hock, Hans Henrich. 1986. Principles of historical linguistics. Berlin: Mouton de Gruyter. Kachm, Braj. B. 1983. The Indianization of English: the English language in India. Delhi: Oxford University Press. . 1986. The alchemy of English: the spread, functions and models of non-native Englishes. Oxford: Pergamon Press. Krishnamurti, Bhadriraju and Aditi Mukherji. 1984. Modernization of Indian languages in the news media. Hyderabad: Department of Linguistics, Osmania University.

310 · Rajeshwari V. Pandhanpande McGregor, Ronald Stuart. 1984. A history of Indian literature: Hindi literature from its beginnings to the nineteenth century. Gonda, Jan (ed.) Vol. 8, Wiesbaden: Otto Harrassowitz. Masica, Colin, P. 1976. Defininga linguistic area: South Asia. Chicago: University of Chicago Press. . 1992. South Asia as a Sociolinguistic Area? Dimensions of sociolinguistics in South Asia: papers in memory of Gerald Kelfy, ed. by E.C. Dimock, Jr, B.B. Kachru and Bh. Krishnamurti, 37-46. New Delhi: Oxford & IBH Publishing House. Narayanan, Vasudha. 1994. The vernacular veda. Columbia: The University of South Carolina Press. Omprakash. 1971. Präcin Hindi kävya. Delhi: Radha Krishna Prakashan. Pandhanpande, Rajeshwari. 1986. Language contact and language variation: Nagpuri Marathi. South Asian languages: structure, convergence, and diglossia, ed. by Bh. Krishnamurti, Colin Masica and A.K. Sinha, 219-31. Delhi: Motilal Banarsidass. . 1992. Language of religion in South Asia: the case of Hindi. Dimensions of sociolinguistics in South Asia: papers in memory of Gerald Kelfy, ed. by E.C. Dimock, Jr, B.B. Kachru and Bh. Krishnamurti, 271-84. New Delhi: Oxford & IBH Publishing House. . 1998. Constructing religious discourse in diaspora: American Hinduism. Paper presented at the Symposium on Language, Creativity and Identity in Diaspora. Center for Advanced Study. University of Illinois at Urbana-Champaign. 16 November. . Forthcoming. Language of religion in South Asia: theory and practice. Pandit, Prabodh B. 1972. India as a sociolinguistic area. Poona: University of Poona. Peterson, Indira Viswanathan. 1989. Poems to Siva. Princeton: Princeton University Press. Pope, G.V. 1900. Tiruvacakam or sacred utterances. Oxford: Clarendon Press. Sridhar, Shikaripur N. 1981. Linguistic convergence: Indo-Aryanization of Dravidian languages. Lingua 53. 199-220. Shapiro, Michael C. and Schiflman, Harold F. 1981. Language and society in South Asia. MLB D series in linguistics, Vol. 1, ed. by D. Jain. Columbia: South Asia Books. Thirugnansambandhan, P. 1992. Sanskrit Tamil contact. Thiruvananthapuram: International School of Dravidian Linguistics. Venkatacari, Koil Kandadai Appan. 1978. The Manipraväla literature of the Srivaisnava äcäryas. Madras: Premier Press. Watwe, Keshav, Narayan. 1964. Präcin Maräthi panditikävya. Pune: Joshi-Lokhande Prakashan. Zvelebil, Kami! V. 1974. Tamil literature. A history of Indian literature. Gonda, Jan (ed.) Vol. X, Fasc. 1. Wiesbaden: Otto Harrassowitz.

Dative Subject Constructions in South Asian Languages* • MASAYOSHI SHIBATANI and PRASHANT PARDESHI · Over the past twenty-five years, the so-called dative subject constructions have received focused attention from the specialists of South Asian languages, Japanese, Quechua, Icelandic, in which a similar construction exists. The past analyses under various theoretical persuasions generally agree that these constructions are transitive with the assumption that the two relevant noun phrases are (direct) arguments of the lexical predicates. In this paper, we attempt to show that these past analyses are mistaken and that these non-canonical constructions are basically intransitive. Specifically, we advance a hypothesis that they are to be analyzed as variants of doublesubject constructions, where only one noun phrase is a lexically selected argument, the other being sanctioned by a clausal predicate. Also explored are the semantico-pragmatic reasons for the elliptical nature of the relevant intransitive predications and the factors governing the distribution of subject properties over the 'large subject' and the 'small subject' of the double subject construction.

• 1. Introduction In addition to the possession of retroflex consonants and breathy voice, South Asian languages are well known for their non-canonically case-marked * The major portion of the research reported in this paper was carried out while the first named author was a Visiting Fellow at the Research Centre for Linguistic Typology in the Australian National University during the period of December 1998-March 1999. He is very grateful to Professors R.M.W. Dixon and Alexandra Aickenvald, Director and Associate Director fo the Research Centre, respectivly, for providing him with a marvelous opportunity for enjoying an ideal research environment as well as a unique combination of British, Australian and Russian hospitality. We also wish to thank Deepthi Kumara Henadeerage of the Linguistics Department at ANU for sharing his expertise on Sinhala and for providing us with many Sinhala examples useful to our discussion.

312 · Masayoshi Shibatani and Prashant Pardeshi

constructions. These constructions are popularly known as dative subject constructions for the reasons to be discussed presently. The following are typical sentences found in the area instantiating this construction type. (1) aaja ma-laaii jaaDo laag-yo. Today I-DAT cold feel-MASC 'I feel it cold today.' (Nepali; Clark 1963:17) (2) use gussaa aayaa. he.DAT anger came 'He became angry' (Hindi; adapted from Kachru 1990: 63). (3) mare jAvuu joiie. I-DAT go needed º want/need to go' (Gujarati; Lambert 1971). (4) rAm-lA jarman bhAshA bol-tA ye-t-e Ram-DAT German language.? speak-PTCPL come-PRES-F 'Ram can speak German' (Marathi). (5) maTA lamaya-wA penAwa I-DAT child-ACC see.PRES 'I see the child' (Sinhala; Kumara Henadeerage, p.c.). (6) avanige jvara bantu. he.DAT fever came 'He got a fever' (Kannada; Sridhar 1976:132). (7) avanukku muham malarndadu. he.DAT face bloom.PAST.it 'His face bloomed; he felt pleasure' (Tamil; Lindholm 1976:175). (8) niNaL-kkA pook-aam. you-DAT go-may 'You may go' (Malayalam; Jayaseelan 1990:279). (9) ji-ta dhebaa yawa maai. I-DAT money much need º need a lot of money' (Newari; Kazuyuki Kiryu, p.c.). (10) Amake Aiy ora hiju ketAna, you.DAT I house come have to 'You have to come to my house' (Mundari; Abbi 1990:259). We should note that in some languages—most prominently in Bengali—the genitive case is employed where other languages use the dative: (11) Bengali (Onishi, forthcoming) a. ama-r baba achen. 1SG-GEN father be/have.PRES.2/3HON º have a father'. b. ama-r jinis-Ta cai. 1SG-GEN thing-DEF need º need that thing.'

Dative Subject Constructions m 313 c. t -r

h si

pa-cch-e.

3HON.SG.GEN laugh get-IMPERF-PRES.3ORD 'S/he (HON) feels like laughing.' (12) Marathi a. mAza nehaml Doka dukh-t-a1 I-GEN.N always head.N ache-IMPERF-N º always have a headache.' cf. ma-ΙΑ sardl zA-l-I Ahe I-DAT cold.F become-PERF-F be º have a cold.' b. mAzI acAnak badll zA-l-I I-GEN.F suddenly transfer.F become-PERF-F º suddenly got transferred.' cf. tyA-ΙΑ mulgA zA-l-A he-DAT son.M become-PERF-M 'He had a son.' c. mAza ti-cyA-war prem Ahe I-GEN.N she-GEN-on love.N be º love her.' cf. ml ti-cyA-war prem kar-t-o I.NOM.M she-GEN-on love do-PRES-M º love her.' In view of the fact that this type of construction is widely attested among areally and genetically distinct languages around the world (see later), it is not entirely clear whether its spread throughout the Indian subcontinent is due to an areal factor. Still, it is likely that there have been both borrowing and mutual reinforcing among genetically different languages. For example, the occurrence of dative subject constructions among Tibeto-Burman (e.g., Newari) and Munda languages (e.g., Mundari) is probably due to borrowing, since they are not uniformly seen throughout these groups of languages. What makes this question difficult to answer—in addition to a wide distribution of analogous constructions elsewhere—is that the predicate types involved form well definable sets, indicating that dative subject constructions (simply referred to as 'dative constructions' hereafter) are semantically motivated, hence may occur independently of outside influence. The relevant predicate types are those expressing: (13) a. Possession/Existence (e.g., 11) b. Psychological states (e.g., 2, 7)

314 · Masayoshi Shibatani and Prashant Pardeshi c. Physiological states (e.g., 1, 6) d. Visual/auditory perceptions, including the notion of 'appearance/ seeming' (e.g., 5) e. Modal states of necessity and wanting including the notion of obligation ('must') (e.g., 3, 9,10) f. Modal states of potentiality, including ability and the notion of permission ('may') (e.g., 4,8). The extent of the spread of dative constructions and their analogues is reflected in the wide variety of languages covered in past studies. These range from South Asian languages (e.g., Verma 1976 and Verma & Mohanan 1990) to Georgian (e.g., Harris 1984), and from Japanese (Kuno 1973) to Italian (e.g., Perlmutter 1984) and Quechua (e.g., Jake 1985). Most of the studies share the basic assumption that these constructions with two nominal arguments are transitive (or at least at some level of representation, such as in Relational Grammar treatments). We argue in this paper that this assumption is incorrect and that dative constructions are structurally and relationally distinct from both canonical transitive constructions and straightforward intransitive constructions. We advance a novel analysis that treats dative constructions as variants of double subject constructions widely attested among Asian languages such as Japanese, Chinese, and Indonesian.2

• 2. Dative Nominals as Subjects Many languages exhibiting dative constructions allow alternate coding of similar situations. These include both South Asian and non-South Asian languages. There are thus doublets of the following type, where both canonical transitive coding patterns and dative coding patterns are seen: (14) Sinhala (GaiV 1990:17) a. mamA ee wacAne kiwwa. I.NOM that word say.PAST º said that word.' b. maTA ee wacAne kiyAwuna. I.DAT that word say.P.PAST Ί blurted that word out.' (15) Bengali (Onishi, forthcoming) a. tumi bondhu-ke Teliphon kor-ech-o? 2ORD.SG.NOM friend-OBJ telephone do-PERF-PRES.2ORD 'Have you made a telephone call to your friend?' b. toma-r bondhu-ke Teliphon kOr-a 2ORD.SG-GEN friend-OBJ telephone do-VN

Dative Subject Constructions m 315 ho-ech-el become-PERF-PRES.3ORD 'Have you finished your telephone call to your friend?' While the parallel coding patterns shown above may be construed as evidence for considering dative constructions to be transitive, there is even more compelling evidence supporting the view that the dative nominals correspond to nominative subjects of canonical transitive clauses. Consider word order first. While many languages show flexible word order, it is normally possible to identify an unmarked one. Take transitive and ditransitive sentences (16)(17) and dative subject sentence (18) in Marathi for example. In the former, the order NOM-DAT(-NOM-)PRED (16)a-(17)a is unmarked, whereas in the latter, DAT-NOM-PRED (18)a is unmarked. A similar observation can be made in other languages as well. Thus, word order indicates that the dative subject indeed occurs in subject position (sentence-initially in Marathi), in contradistinction to a dative marked object. (16) a. rAm dararoj shAm-lA rnAr-t-o. Ram.NOM everyday Sham-DAT3 beat-PRES-M 'Ram beats Sham everyday/ b. IshAm-lA rAm dararoj mAr-t-o. Sham-DAT Ram.NOM everyday beat-PRES-M 'Ram beats Sham everyday/ (17) a. rAm sitA-lA sangit shikaw-t-o. Ram.NOM Sita-DAT music.NOM teach-PRES-M 'Ram teaches music to Sita.' b. ?sitA-lA rAm sangit shikaw-t-o. Sita-DAT Ram.NOM music.NOM teach-PRES-M 'Ram teaches music to Sita.' (18) a. rAm-lA sitA AwaD-t-e Ram-DAT Sita.NOM like-PRES-F 'Ram likes Sita.' b. ?sitA rAm-ΙΑ AwaD-t-e Sita.NOM Ram-DAT like-PRES-F 'Ram likes Sita.' Next, consider reflexive binding with the Marathi form swatahA 'self. That only subjects can bind this reflexive form can be seen in the following example, where the ordinary indirect object dative nominal cannot be the antecedent. (19) ml rAm-ΙΑ swatahA-d Topi ghAt-l-I I.NOM Ram-DAT self-of cap.F puton-PERF-F º put on Ram my/*his cap.'

316 · Masayoshi Shibatani and Prashant Pardeshi

In the dative construction of the following type, however, it is the dative nominal, rather than the nominative NP, which binds the reflexive form, as indicated in the translation. (20) rAm-ΙΑ sitA swatahA-cyA bahlNI pekshA jAsta Ram-DAT Sita.NOM self-of sister than more AwaD-t-e like-PRES-N 'Ram likes Sita more than his (Ram's/* Sita's) own sister/ Though different languages (as well as specific constructions within a single language) differ in the extent to which the dative nominals exhibit subject properties (see later), the situation above is also paralleled by other Indie and Dravidian languages, as well as those outside these language families (see Shibatani, forthcoming). Kachru et al. (1976) summarizes the behavioral subject properties of different types of subject in selected Indie languages as follows: (21) Behavioral properties of Hindi-Urdu, Kashimiri, and Panjabi subjects (Kachru et al. 1976: 94) Rule Reflexivization Equi Conjunction reduction Raising

Controller

Accessible

SI, ST, S DAT, S OBL, SP SI, ST, S DAT, S OBL, SP SI, ST, S DAT, S OBL SI,ST

SI, ST, S DAT SI,ST SI,ST

(SI = intransitive subject, ST = transitive subject, S DAT = dative subject, S OBL ·· oblique subjects, SP = derived subject of the passive)

• 3. Past Analyses Past analyses of dative constructions have tended to assimilate observations of the kind above to the canonical transitive construction, whereby the dative and the nominative (unmarked) NP are respectively treated as subject and object. In the South Asian context, both Gair (1990) and Masica (1991) are quite explicit about this approach showing parallel structures of canonical transitive and dative constructions, as below: (22) Sinhala (Gair 1990:25) a. mamA ee wacAne kiwwa (canonical transitive construction; Gair's 6a). I.NOM that word say.PASTA º said that word.'

Dative Subject Constructions m 317

N" mamA NOMINATIVE T

N" ee wacAne (ACCUSATIVE) 'that word'

kiyAwA'say'

b. maTA ee wacAne kiyAwuna (dative construction; Gair's 6b). I.DAT that word say.PASTP º blurted that word out.'

N" ee wacAne (ACCUSATIVE) 'that word' (23) Hindi (Masica 1991: 363) a. raam caay piitaa hai. Ram.NOM tea drink AUX 'Ram drinks tea.' Transitive S

NP1

VP NP 2

Vb

318 · Masayoshi Shibatani and Prashant Pardeshi

b. raam-ko bhuukh lagii thii Ram-DAT hunger strike AUX 'Ram was hungry.' Dative-Experiencer S NP1 -f ko

VP NP2

Vb

Studies in South Asian languages in general have little to say about the status of the nominative (unmarked) NPs involved in these constructions; they are simply assumed to be objects. This is largely because in the ergative construction of these languages the patient argument assumes the unmarked form. Once the relevant structure is determined, the task is to account for the case-marking pattern. The question is why, if they are transitive in structure, dative constructions exhibit the DAT-NOM-PRED case pattern and not the canonical NOM-ACC-PRED pattern. The past analyses concentrated their efforts on accounting for dative marking, largely ignoring unmarked (nominative) nominals. In doing so, they tended to resort to the thematic relations borne by the dative subject. A general impression based on predicate types—see (13)—leads one to associate dative marking with the thematic role of experiencer. That is, the subject of a dative construction is marked dative because it bears the experiencer role, as opposed to the agentive role of the subject of a canonical transitive construction. However, there is no consistent correspondence between dative marking and the experiencer role; not all experiencers are marked dative, and not all datives are experiencers, as seen below: (24) Marathi a. ml ti-cyA~war prem kar-t-o I.NOM.M she-GEN-on love do-PRES-M º love her.' b. tyA-ΙΑ mulgA zA-l-A he-DAT son.M become-PERF-M 'He had a son.' Other proposals such as Mohanan & Mohanan (1990), who analyze dative subjects as bearing the goal role, and Pandharipande (1990), who proposes a locative solution, are equally problematic, since many dative subjects are hard

Dative Subject Constructions m 319 to construe in term of these roles; e.g., (4), (5), (8). Whereas the marker used for the goal and the locative role is often identical to the dative marker across languages and while this fact calls for some explanation, these proposals are not quite sufficient in handling a large number of similar constructions with a variety of markings which appear to correspond to the dative subject, e.g.: (25) Taroo-ga Hanako-ga suki da Taro-NOM Hanako-NOM like COP Taro likes Hanako' (Japanese). (26) nay-ka nuktay-ka mwusep-ta. I-NOM wolf-NOM afraid-IND º am afraid of the wolf (Korean; Lee, p.c.). (27) mage oluwe kaekkumak tiyenAwa. I.GEN head.LOC ache.INDEF be-INANIMATE.PRES º have a headache' (Sinhala; Kumara Henadeerage, p.c.). (28) bacce se shiishaa TuuT gayaa. child INST mirror break went/PASS 'The child (inadvertently) broke the mirror' (Hindi; Kachru 1990:60). (29) lamAya-atin kooppe biNduna. child-INST cup break.PAST.P 'The child (inadvertently) broke the cup' (Sinhala; Wijayawardhana, et al 1995:107). (30) ma-baata sisaa phuT-yo. I-ABL glass break-PERF 'The glass broke (and I inadvertently happened to be its cause)' (Nepali; Madhav Pokharel, p.c.). (31) mara-thii jAngAl-ma ekla nAhi jAv-a-y. ABL jungle-in alone not go-PASS-PAST Ί couldn't go into the jungle alone' (Gujarati; Lambert 1971:169). (32) rAm-cyA-hAt-Un cukun zopDI cirD-l-I Ram-GEN-hand-ABL by mistake hut.F crush-PERF-F ge-l-1 go-PERF-F 'The hut got crushed by mistake at the hands of Ram' (Marathi). (33) asil par apne puure parivaarkii jimmevaarii hai. Asit on self whole family of responsibility be Asit is responsible for his whole family' (Kachru 1990:60). (34) nuca-ta-ca uma-ta nana-wa-n-mi I-ACC-TOP head-ACC hurMOBJ-3-WIT 'My head hurts' (WIT = witnessed) (Imbabura Quechua; Jake 1985: 196). (35) Nup snl yob alkjon ay-a-k. him boil big armpit form-3SG-PAST ¢ large boil has formed in his armpit' (Kalam; Pawley et al forthcoming: 12).

320 · Masayoshi Shibatani and Prashant Pardeshi

(36) d-agavu no-k~i-e me-pain PROG-fasten-3SG-IND Tain is gripping me; i.e., I am feeling pain' (Yagaria; Renck 1975:145). Relational Grammar (Perlmutter 1984; Harris 1984; Jake 1985, etc.) handles the problems of dative constructions in terms of inversion,' which entails a relational network in which 1,2, and 3 represent the subject, object, and indirect object respectively. Marathi sentence (37a) would receive an analysis such as that in (37b): (37) a. rAm-lA sitA AwaD-t-e Ram-DAT Sita.NOM.F like-PRES-F 'Ram likes Sita.' b. 1 2 P (initial stratum) 3 l p (final stratum) Ram Sita AwaDUnder this analysis, the surface dative nominal is said to behave like a subject because it is a subject in the initial stratum of the relational network. The surface nominative NP is so marked because it is a final subject. While these explanations seem straightforward, we shall argue below that the basic assumption of Relational Grammar and other treatments of dative constructions as transitive are incorrect and lack both the empirical support and analytical sophistication required to clearly articulate a number of relevant points.

• 4. Transitive and Intransitive Predications Upon closer inspection, the transitive analysis of dative constructions and their variants finds no empirical support. First, consider the relevant predicates: typically intransitive verbal complexes involving verbal heads like 'come,' 'become,' 'go,' 'be/exist'. Once again, consider the alternate nominative/dative (genitive) patterns we saw in (14) and (15). Similar examples are provided below for ease of reference: (38) Malayalam (Jayaseelan 1990:281) a. fiaan Ninn-e sneehik'k'-uNNu I-NOM you-ACC love-PRES º love you.' b. enik'k'A Νίηη-οοάΛ sneeham aanA I-DAT you-2DAT love.NOM be-PRES º love you.'

Dative Subject Constructions · 321

(39) Bengali (Onishi, forthcoming) a. ami toma-ke khub pOchondo kor-i 1SG.NOM 2ORDSG-OBJ very liking do-PRES.l º like you very much' (according to my own criteria). b. ama-r toma-ke khub pOchondo 1SG-GEN 2ORDSG-OBJ very liking hO-y become-PRES.3ORD Ί like you very much' (according to some [socially] set criteria). As seen above, the dative/genitive versions have intransitive heads with the basic meaning of 'be' and 'become,' while their nominative counterparts contain transitive verb phrases. Kachru (1990) has made a similar observation for Hindi. She provides the following tables to illustrate her point that while 'active' verbs entail a canonical coding pattern, stative and change-of-state verbs call for one encoding dative. (40) Hindi (Kachru 1990: 68) State gussaa honaa 'be angry' pasand honaa 'like' yaad honaa 'remember' bhuukh honaa "be hungry' buxaar honaa 'have a fever' vishvaas honaa 'believe' pataa honaa 'know'

Change-of-state gussaa aanaa pasand aanaa dikhaaii denaa 'become visible' yaad aanaa bhuukh lagnaa buxaar aanaa pataa calnaa

Active gussaa karnaa pasand karnaa dekhnaa yaad karnaa vishvaas karnaa pataa karnaa

Predicates calling for dative coding involve stativizing honaa and inchoative aanaa 'come,' both of which we take to be intransitive, while active predicates typically involve a transitivizing do-verb (karnaa), as in many other languages including Korean. Thus, there is lexical evidence that dative constructions and their variants are really intransitive, with verb (verbal complex) predicating over the nominative NP. In other words, canonical transitive and dative construction differ in their predication patterns, as shown below: (41) a. Canonical transitive construction [NP-NOM NP-ACC PRED] SUBJ b. Dative construction [NP-DAT/GEN NP-NOM PRED] SUBJ

322 · Masayoshi Shibatani and Prashant Pardeshi

The analysis proposed above for the dative construction, while consistent with the intransitive verb forms involved, may strike one as problematic in view of the following, which show what appear to be direct objects.4 (42) Sinhala (Gair 1990:19) a. marnA aswAyawA elewwa. I.NOM horse.ACC chase.PAST 'I chased the horse.' b. maTA aswAyawA elewunaa. I.DAT horse.ACC chase.RPAST º involuntarily (on impulse) chased the horse.' (43) Sinhala (Gair 1990:19) a. mamA lamAyaTA baenna. I.NOM child.DAT scold Ί scolded the child.' b. maTA lamAyaTA baenuna. I.DAT child-DAT scold.P.PAST º (involuntarily/on impulse) scolded the child.' (44) Bengali (Onishi, forthcoming) a. tumi bondhu-ke Teliphon kor-ech-o 2ORD.SG.NOM friend-OBJ telephone do-PERF-PRES.2ORD 'Have you made a telephone call to your friend?' b. toma-r bondhu-ke Teliphon 2ORD.SG-GEN friend-OBJ telephone kOr-ho-ech-e do-VN become-PERF-PRES.3ORD 'Have you finished your telephone call to your friend?' (45) Bengali (Onishi, forthcoming) a. ami toma-ke khub pOchondo kor-i 1SG.NOM 2ORDSG-OBJ very liking do-PRES.l º like you very much' (according to my own criteria). b. ama-r toma-ke khub pOchondo 1SG-GEN 2ORDSG-OBJ very liking hO-y become-PRES.3ORD º like you very much' (according to some [socially] set criteria). Prima facie, the dative/genitive sentences in the (b) versions appear to countenance the same objects as corresponding canonical transitive sentences (a), supporting the transitive analysis. Upon closer inspection, however, it is revealed that the object-marked nominals of the (b) sentences are not entirely parallel to those found in their transitive counterparts. First, take Sinhala forms (42b) and (43b). These sentences contain the so-called P-forms of the

Dative Subject Constructions m 323

relevant verbs. Examination of various P-forms in Sinhala indicates that these were originally something like passive and/or middle forms. As is well known, Indie passives allow impersonal forms in which the original objects are retained. The situation is similar to the German impersonal passive in the following, where the dative object remains in the accusative form. A comparable contemporary Marathi example is also given, along with a personal passive construction. (46) a. Dem Kind wurde geholfen (German impersonal passive) the.DAT child became help.PP Lit. 'There was helping of the child/The child was helped.' b. tyA mulA-lA mAr-NyAl A-l-e (Marathi impersonal passive) that boy-DAT kill-PTCPL come-PERF-N That boy was killed.' c. to mulgA mAr-l-A ge-l-A (Marathi personal passive) that boy.NOM kill-PERF-M go-PERF-M 'That boy was killed/That boy died.' In modern Sinhala, P-forms are used as intransitive verbs without passive force. As such there is no outside agent. (47) a. aswAyawA mxruna. horse.ACC kill.P.PAST The horse died' (Kumara Henadeerage, p.c.). b. minihawA tamaNge pokune w&tuna. man.ACC self.GEN pond fall.P.PAST The fellow fell in his own pond' (Gair 1990:18). Even with the non-passive readings as shown in (47a) and (47b), many Sinhala speakers retain accusative object marking on animate nominals, although there are apparently some who opt for nominative. Our claim with regard to expressions like (42b)-(43b) is that what appears to be an object is not the object of a transitive clause; rather it is an object-marked subject like the ones in (47). The same analysis can be applied to the Bengali genitive forms (44b) and (45b). The predicates here involve the -hO 'become' ending, which 'deagentivizes' a transitive clause while still retaining original objects. The retention of object marking is expected if this process is considered to be (impersonal) passivization, as in Bengali traditional grammar (see Onishi, forthcoming). We assume that a similar explanation can be extended to Hindi and other languages that exhibit object-marked nominals where nominative forms are expected. The grammatical status of such nominals might differ from a language to another. However, they may either be objects of impersonal clauses or object-marked intransitive subjects, as in Sinhala.

324 · Masayoshi Shibatani and Prashant Pardeshi

Involvement of an impersonal structure in the dative construction is seen elsewhere. Quechua is one of those languages where this is quite clear. (48) nuca-ta-ca uma-ta nana-wa-n-mi I-ACC-TOP head-ACC hurt-lOBJ-3-WIT 'My head hurts/(lit.) It hurts (me) the head with respect to me' (WIT = witnessed) (Imbabura Quechua; Jake 1985:196). The use of the impersonal structures is one of the factors that need to be integrated in a successful account of dative constructions and their variants. This, and the intransitive nature of the relevant predicates—as well as that of the predicate semantic types enumerated in (13)—must be all tied together in order to show the important semantic contrast seen between dative constructions and their canonical transitive counterparts. The dative and the canonical coding pattern contrast in meaning along the controllability parameter. Our claim is that the canonical transitive and dative constructions (and their variants) represent different conceptualization patterns. The subject of a canonical transitive construction is an argument of the verb and typically represents an agent in control of the event expressed. In other words, the canonical transitive construction codes an event as a controllable situation. Whether the event is actually carried out as such is a different matter. What does matter is that a speaker conceptualizes a situation as something controllable by the agent. Dative constructions, on the other hand, represent states of affairs that are conceived as uncontrollable. This is evident from the predicate types involved in the construction: most typically physiological and psychological states over which one has no control. Some languages (e.g., Marathi) extend the list of relevant predicates to include certain verbs of mental activity such as forgetting, remembering, and knowing, as well as verbs expressing other uncontrolled events, e.g., losing an article of possession. Impersonal constructions code situations brought about by some unknown entity, hence are also beyond one's control. Frequent use of passive structures is similarly motivated. Verbs such as 'become' and 'come' also express the spontaneous occurrence of an event. The use of basic and derived intransitive verbs of both basic and derived types—especially those expressing the semantic types delineated in (13)—is thus tied closely to the conceptualization pattern represented by dative constructions and their variants. The grammatical correlates of the contrast between controllable and noncontrollable coding patterns were recognized in one of the earliest studies of dative constructions. Sridhar (1976) makes this point by comparing the possibility of embedding a transitive and a dative structure in a Kannada control construction: (49) a. ShiilaLige aapareeshan aayitu (dative) Sheela.DAT operation became 'Sheela had an operation.'

Dative Subject Constructions m 325

b. *avaru ShiilaLigei [0i aapareeshan galu] HeeLidaru they Sheela.DAT operation become told They told Sheela to have an operation.' (50) a. Shiilaa apareeshan maaDisikonDaln (reflexive-transitive) Sheela.NOM operation had done.REFL 'Sheela had an operation [done to herself].' b. avaru ShiilaLigej [0, aapareeshan maaDisikoLLalu] heeLidaru they Sheela.DAT operation to have done.REFL told They told Sheela to have an operation [done to herself].' In our analysis of dative constructions, all these factors are brought together in a way that sharply contrasts with the past transitive analyses. By recognizing the basic structure of dative constructions to be intransitive (see [41]), the possibility opens up that the dative nominal lies outside the scope of lexical predication, i.e., it is not a direct argument of the lexical predicate. This possibility (pursued here) provides a natural means of distinguishing a dative nominal from the agentive nominal of a transitive construction, where it is a direct argument of the lexical predicate. In the analysis proposed here, the controllability contrast discussed above finds structural expression in the distinction of whether or not the relevant nominal is a direct argument of the lexical predicate. Before detailing this analysis, we first review the subject properties of nominative NPs in the dative construction, which we claim to be the subject of intransitive predication.

• 5. The NP-NOM as a Grammatical Subject: [NP-DAT NP-NOM PRED] In European and other languages, case marking and agreement provide a clear means for the identification of grammatical relations. Our analysis, which considers the nominative NP of the dative construction to be subject, is consistent with both case marking and agreement patterns in agreement languages, as clearly shown in the following examples from German, Russian and Modern Hebrew. (51) a. Mir geff llt dieses Buch (German) I.DAT like.SGthis book.SG.NOM º like this book', b. Mir gefallen diese B cher I.DAT like.PL these books.PL.NOM º like these books.' (52) a. Mne nravitsja kinga (Russian) I.DAT like.3SG.REFL book.SG.NOM º like the book.'

326 · Masayoshi Shibatani and Prashant Pardeshi

b. Mne nravjatsja kniff, I.DAT like.PRES.3PL.REFL book.PL.NOM like books.' (53) a. Le Moshe haya sefer (Modern Hebrew; Anne Hartenstein, p.c.) DAT be.3SG.M.PAST book.M.SG 'Moshe has a book.' b. Le Moshe hayu shlosha sfanm DAT be.3PL.M/F.PAST three book.M.PL 'Moshe has three books.' The grammatical status of the nominative NP has not been debated a great deal in the context of South Asian languages. Once the dative nominal is taken as the subject of the transitive structure, the nominative NP is automatically assumed to be object. The fact that this object occurs in an unmarked (i.e., nominative) form and that it triggers verb agreement has not been of much concern. In the ergative construction, which occurs in many Indie languages in the so-called perfective tense, the object assumes the nominative form and controls verb agreement. Since verb agreement appears to be generally triggered by an unmarked (i.e., nominative) NP irrespective of its grammatical status, agreement does not seem to provide conclusive evidence for the identification of the grammatical relation of either the dative or the nominative NP. Despite this general situation, there still seem to be some agreement facts that point to the subject status of the nominative NP in the South Asian dative construction. Nepali is interesting in that the ergative NP of a transitive clause agrees with the verb, indicating that agreement in this language operates in an accusative fashion regardless of the morphological ergativity found in the perfective tense. (54) haamro choraa aaja aa-yo^ our son today come-3SG.M.PERF Our son came today' (Nepali; Clark 1963:17). (55) tyo paanii lyaauna Dhaaraa-maa ga-u 3SG water fetch Dhaaraa-at go-3SG.F.PERF 'She went to Dhaaraa to fetch water' (Clark 1963:59). (56) raam-le nayφφ lugaa laa-yol Ram-ERG new clothes wear-3SG.M.PERF 'Has Ram put on the new clothes?' (Clark 1963:20) The above situation—especially the agreement pattern in (56)—contrasts with a number of other major Indie languages whose agreement systems are sensitive to case marking in that a case-marked NP does not trigger agreement. As can be seen in (56), a case-marked nominal does not block

Dative Subject Constructions m 327

agreement; agreement is triggered by a subject whether it is unmarked or case-marked. From this it is expected that the dative NP of the dative construction in Nepali would trigger agreement if it were the subject of the clause. This is not the case, however; as agreement takes place between the nominative (unmarked) NP and the predicate, indicating that the subject of the dative construction is the one in the nominative case. (57) keTaa-laaii keT-i raamr-i boy-DAT girl-F beautiful-F 'The boy likes the girl' (Madhav (58) keTi-laaii keT-o raamr-o girl-DAT boy-M beautiful-M 'The girl likes the boy' (Madhav

laag-ii_ feel-F Pokharel, p.c.). laag-yo^ feel-M Pokharel, p.c.).

In a similar vein, Bengali honorific marking points out the subject status of the nominative NP in a dative construction which involves genitive marking. Both nominative intransitive and transitive subjects in canonical constructions trigger honorific marking on the verb. (59) a. tini

has-ch-en

(Onishi, forthcoming)

3HON.SG.NOM laugh-IMPERF-PRES.2/3HON

'(S)he (HON) is laughing (HON)/ b. tini ciTHi-Ta poR-ben. 3HON.SG.NOM letter-DEF read-FUT.2/3HON '(S)he (HON) will read (HON) the letter.' Bengali marks subjects with locative case if they have indefinite or generic reference. Subjects so marked behave almost identically with nominative subjects; in the following example, the locative marked subject thus triggers honorific marking on the verb, just like regular nominative subjects (Onishi, forthcoming). (60) kalke chilo robibar Onek-e-i baRi-te chilen one.day. be.PAST Sunday manyhouse- be + PAST + removed 3ORD LOC-EMPH LOG 2/3HON 'Yesterday was Sunday. (Therefore) many (HON) were (HON) at home.' A case-marked subject, in other words, triggers honorific marking on the verb. The genitive counterpart of the dative subject, however, does not trigger subject-honorific marking; rather, it is the nominative NP that does this, as shown below: (61) a.

tini has-ch-en 3HON.SG.NOM laugh-IMPERF-PRES.2/3HON 'S/he (HON) is laughing (HON).'

328 · Masayoshi Shibatani and Prashant Pardeshi b. t -r

h si

pa-cch-e

3HON.SG-GEN laugh get-IMPERF-PRES.3ORD 'S/he (HON) feels like laughing (ORD).' (62) a. ama-r baba achen 1SG-GEN father be/have.PRES.2/3HON º (ORD) have a father.' b. t -r du-Ti chele ache 3HON.SG-GEN two son be/have.PRES.3ORD 'He (HON) has two sons/ A similar honorification pattern is seen in Nepali: (63) wahaan-ko batchaa cha (Sujeet Pradhan, p.c.) he.HON-GEN child be/exist 'He (HON) has a child.' (64) #wahaan-ko batchaa hunuhuncha he.HON-GEN child be/exist.HON 'He has a child.' (65) wahaan-ko buwaa hunuhuncha he.HON-GEN father be/exist-HON 'He (HON) has a father (i.e., his father is still alive).' Existential predicates in Sinhala—like Japanese—impose an animacy selectional restriction which operates between the nominative NP and the predicate, not between the dative NP and the predicate, as seen below: (66) a. maTA

duwek

innAwa

(Sinhala; Kumara Henadeerage, p.c.) I.DAT daughter.INDEF be-ANIMATE-PRES º have a daughter.' b. *maTA duwek tiyenAwa (ibid., p.c.) I.DAT daughter.INDEF be-INANIMATE-PRES º have a daughter.' (67) a. maTA potak tiyenAwa (ibid., p.c.) I.DAT booklNDEF be-INANIMATE-PRES 'I have a book.' b. *maTA potak innAwa (ibid., p.c.) I.DAT book.INDEF be-ANIMATE-PRES º have a book.'

A similar selectional restriction is seen in Gujarati with respect to verbs of pleasing or liking. When the pleasing thing is food, bhave ehe is used, but when it is non-food, gAm- is used;

Dative Subject Constructions m 329

(68) Ramesh-ne pen gAm-y-i (Mistry 1976:249) Ramesh.OBJ pen.F like-PAST.F 'Ramesh liked the pen.' (69) apne gujAratl khorak bhave ehe? (Lambert 1971:53) you.OBJ Gujarati food be liked be 'Do you like Gujarati food?' Finally, some nominative NP's show the behavioral subject property of reflexive binding in Marathi. Attributing the following sentence to Kashi Wali, Pandharipande (1990:165) agrees with Wali that the nominative form to 'he,' rather than the dative nominal, binds the reflexive. (70) ma-ΙΑ to swatahA-cyA gharAt dislA I-DAT he.NOM self-of house.in see.PERF.3M º saw him in his/*my house.' All in all, there is a great deal of evidence that argues for the subject status of the nominative NP. This comes into direct conflict with the discussion in Section 2, however, where we saw evidence pointing to the subject status of the dative nominal. The dilemma can be easily resolved in a Relational Grammar analysis, in which some subject properties are attributed to the initial subject status of the dative nominal, some to the final subject status of the nominative NP (see 37b). However, the distribution of subject properties over these two kinds of nominal is not as uniform and consistent as Relational Grammar would predict. Dative constructions turn out to be not entirely uniform, and a more finely-tuned analysis of them is required. This may not be easily accommodated in a framework like Relational Grammar. Before discussing this issue, however, we address another issue that holds the key to all the essential problems surrounding dative constructions and their variants.

• 6. Elliptical Nature of Intransitive Predication Besides the syntactic dilemma noted above, there is a yet more serious one that has not been dealt with in past analyses of dative constructions. This has to do with the fact that while the relevant predicates are intransitive in nature, they appear to require two arguments, a dative and a nominative NP. Perhaps the strongest reason for past analyses to consider dative constructions as transitive comes from the obligatory nature of these two arguments (see Kuno 1973). (It still doesn't seem to have bothered the proponents of the transitive analysis that the relevant predicates are irftransitive.) Thus, the following expressions without dative/genitive nominals are considered to be elliptical.

330 · Masayoshi Shibatani and Prashant Pardeshi

(71) Malayalam (Mohanan & Mohanan 1990:54) a. kuTTik'k'A wayarA nirannu child.DAT stomach.NOM fill (intr.).PAST 'The child was full.' b. wayarA nirannu stomach.NOM fill(intr.).PAST '(Someone) was full.' (72) Hindi a. use gussaa aayaa he.DAT anger came 'He became angry.' b. gussaa aayaa anger came '(Someone) became angry.' (73) (Sinhala) a. maTA oluwe kxkkumak tiyenAwa I.DAT head.LOC ache.INDEF exist.INANIMATE º have a headache.' b. oluwe kaekkumak tiyenAwa head.LOC ache.INDEF exist.INANIMATE '(Someone) has a headache.' (74) (Marathi) a. malA don bhAu Ahe-t I.DAT two brothers.NOM be-PL º have two brothers.' b. don bhAu Ahe-t two brothers.NOM be-PL '(Someone) has two brothers.' Notice that all the (b) sentences are syntactically complete; they all contain intransitive verbs and the subjects in the nominative case that show agreement with the verb satisfy the selectional restriction where applicable. The crux of problem is why these syntactically complete sentences are felt to be elliptical. In order to understand this issue, we turn to double subject constructions, whose structure and semantics are better understood.

• 7. Double Subject Constructions A fair number of Asian languages exhibit double subject constructions in which there are two nominative subjects or two unmarked subjects if the language, for instance, Chinese, does not have case marking.

Dative Subject Constructions m 331 (75) Japanese a. Zoo-ga

hana-ga

nagai (koto)

elephant-NOM nose-NOM long (that) '(that) an elephant has a long nose/trunk' b.

elephant-NOM LARGE SUBJ

NP2

PRED

hana-ga

nagai

nose-NOM SMALL SUBJ

long

Here nagai 'long' unambiguously represents a one-place intransitive expression predicating over the NP hana-ga 'nose-NOM'. Indeed, sentence (75a) says that the nose (i.e., the trunk) is long, not the elephant. The second nominative NP (NP2) is thus the subject of the predicate nagai 'long'. On the other hand, zoo-ga 'elephant-NOM' is also a subject in the sense that the entire sentence is about an elephant. It is understood to mean that an elephant is such that the state of affairs of a nose's being long obtains true with respect to it. In other words, the clause hana-ga nagai 'nose is long'is predicating over the NP zoo-ga 'elephant-NOM,' giving rise to a double subject, double predication structure. For ease of reference, we shall call the internal subject (NP2) 'a small subject', and the external one (NPi) 'a large subject', as indicated in (75b). 82, which predicates over the large subject, is called a 'predicate clause' or 'clausal predicate'. The interesting thing about double subject constructions is that the predicate clause (S2) cannot stand by itself. Thus, (76) below is decidedly odd. (76) Hana-galwa nagai nose-NOM/TOP long ¢ nose is long.' As in the English translation, there is nothing syntactically wrong with this sentence. It is the truth-value of the sentence that is being questioned, since the sentence makes a universal claim that a nose is long, which is not in fact true. Compare this sentence with the following, which is more acceptable. Generally speaking, everyone agrees that it is universally true that a flower is beautiful. (77) Hana-ga/wa uiukusii flower-NOM/TOP beautiful ¢ flower is beautiful.'

332 · Masayoshi Shibatani and Prashant Pardeshi

There are basically three types of sentence that relate to the point being made. Some describe what is not universally true (76), while others describe what is generally accepted as expressing a universal truth (77). Finally, there are sentences between these two, i.e., those which can be contested with regard to their truth, and which can specify the domain of their application. Compare the following patterns in English. Sentence (78a) states a universal truth, hence is odd to contextualize as in (78b); the sentence is perhaps only possible when uttered by a dissenter of the Flat Earth Society. (79a), on the other hand, can be contested, hence can be 'personalized,' as in (79b). (78) a. b. (79) a. b.

The earth is round. ?To me, the earth is round. Fish tastes good. To me, fish tastes good.

Predicate clauses in double subject constructions typically express those states of affairs that are not universally true; accordingly, their domain of application must be limited in one way or another. One simple way of achieving this is in terms of'narrowing down' the referent of the subject nominal into a specific entity, turning a generic reference as in (76) to a more restricted one, as in (80a). (80) a. Zoo-no hana-ga nagai elephant-of nose-NOM long 'The elephant's nose/trunk is long.' b. Zoo-ga hana-ga nagai (= 75a) elephant-NOM nose-NOM long An elephant has a long nose/trunk.' Some languages have an additional means of delimiting a universal statement, by means of couching the expression in the double subject construction, as in (80b). There, the large subject provides a domain to which the truth of the predicate clause is limited. Thus, the literal meaning of sentence (80b) is something like: An elephant is a domain in which the proposition "a nose is long" is true.' Another way of looking at the situation is to think of the relationship between the large subject and the predicate clause in terms of dependency. That is, the truth of the state of affairs expressed in the predicate clause is dependent upon (the domain provided by) the large subject. This notion of dependency figures prominently in our understanding of the structure of the dative construction and its variants. Before we turn to the dative construction, let us make sure that the large subject of the double subject construction behaves like a subject syntactically as well.

Dative Subject Constructions · 333 In Japanese, the subject honorification process simply attaches the prefix oorgo-, when adjectives and adjectival nominals are involved, as in (81) below: (81) Hata-sensei-ga wakai/o-wakai Hata-prof-NOM young/HON-young Trof. Hata is young.' The large subject triggers the same honorification process, as a comparison of the behavior of the possessor nominal and the large subject nominal reveals. (82) a. Hata-sensei-no migi-me-ga Hata-prof-GEN right-eye-NOM Trof. Hata's right eye is bad/ b. Hata-sensei-ga migi-me-ga Hata-prof-NOM right-eye-NOM Trof. Hata has a bad right eye.'

warui/?o~warui bad/HON-bad warui/o-warui bad/HON-bad

The large subject, just like a regular subject, raises to the main clause object position under a predicate such asomou 'to think' orminasu 'to consider.' (83) a. Hata-sensei-wa [Ken-ga totemo baka da] to Hata-prof.-TOP Ken-NOM very stupid COP that omot-te i-ru think-CONJ be-PRES Trof. Hata thinks that Ken is very stupid.' b. Hata-sensei-wa Ken-o [totemo baka da} to Hata-prof.-TOP Ken-ACC very stupid COP that omotte i-ru think-CONJ t be-PRES Trof. Hata considers Ken to be very stupid.' (84) a. Hata-sensei-wa [Ken-ga totemo atama-ga warui] Hata-prof.-TOP Ken-NOM very head-NOM bad to omot-te i-ru that think-CONJ be-PRES Trofessor Hata thinks that Ken has a very bad brain.' b. Hata-sensei-wa Ken-o [totemo atama-ga warui] to Hata-prof.-TOP Ken-ACC very head-NOM bad that omot-te i-ru think-CONJ be-PRES Trofessor Hata considers Ken to have a very bad brain.' Thus, there is evidence for the syntactic subject status of the large subject, in keeping with the term 'double subject construction'.

334 · Masayoshi Shibatani and Prashant Pardeshi

• 8. Dative Constructions as Double Subject Constructions We shall now advance the claim that dative constructions are in fact variants of double subject constructions. Our claim boils down to this: The intransitive predication involved in the dative construction typically expresses a state of affairs that cannot be considered universally true, i.e., a cognitive state whose realization is dependent upon a particular cognizer. The dative nominal provides both a domain and a cognizer just as the large subject of the double subject construction does. Consider again those dative sentences that appear to be elliptical without dative nominals. (85) Malayalam (Mohanan & Mohanan 1990:54) a. kuTTik'k'A wayarA nirannu child.DAT stomach.NOM fill (intr.).PAST 'The child was full/ b. wayarA nirannu stomach.NOM fill(intr.).PAST '(Someone) was full.' (86) Hindi a. use gussaa aayaa he.DAT anger came 'He became angry/ b. gussaa aayaa anger came '(Someone) became angry.' (87) Sinhala a. maTA oluwe kxkkumak tiyenAwa I.DAT head.LOC ache.INDEF exist.INANIMATE º have a headache.' b. oluwe kxkkumak tiyenAwa head.LOC ache.INDEF exist.INANIMATE '(Someone, likely to be T) has a headache.' (88) Marathi a. malA don bhAu Ahe-t I.DAT two brothers.NOM be-PL º have two brothers.' b. don bhAu Ahe-t two brothers.NOM be-PL '(Someone) has two brothers.' (89) (Marathi) a. rAm-lA sitA AwaD-t-e Ram-DAT Sita.F like-PRES-F 'Ram likes Sita.'

Dative Subject Constructions · 335

b. sitA AwaD-t-e sitA like-PRES-F '(Someone) likes Sita.' In (85b), it is not true (unfortunately) that a stomach was full. This state of affairs obtains only within a specific individual's domain, which the dative nominal provides in (85a). (86b) and (87b) are both semantically odd, because anger comes (i.e., arises) and a headache exists only when there is someone who recognizes these conditions. When nobody feels anger, there is none. Likewise, a physiological state obtains only when someone recognizes it. In other words, mental and physiological states cannot occur apart from a cognizer, and that is why (86b) and (87b) are felt to be elliptical. Dative nominals supply such a cognizer. Possession and existential expressions are similar, as these notions are defined in relation to a possessor or a location. Brothers, for example, cannot exist in a vacuum (88a). They must be located either physically (as in an existential sentence such as 'My brothers are in the room') or in relation to another person (as in the possessive sentence [88b]). The Marathi verbAwaD- 'like' in (89) is similar to the Japanese adjectival nominal suki da 'like', German gefallen, Spanish gustar, and Russian nravitsja, all with a similar meaning. These verbs are often translated as 'like' or 'pleasing,' but they are not really transitive verbs. They are intransitive but not quite like the English adjectives 'likable' and 'pleasing,' which can be used in describing the quality of an object. The sentence Mary is likable is comparable to Mary is tall and Mary is intelligent. The emotive predicates in Marathi and Japanese are like psychological verbs such as happy and angry in that they must occur together with a cognizer. The difference is that the emotive predicates in question ascribe a state to another entity, who/which causes the emotions of liking, hatred, etc., in the mind of a cognizer, the only one through whom the emotive state materializes. In these languages then, someone is likable only when someone else feels so. The same analysis of dative nominals above also obtains with regard to oblique constructions, as in the following: (90) Oblique constructions a. bacce se shiishaa TuuT gayaa (Hindi; Kachru 1990: 60) child INST mirror break went/PASS 'The child (inadvertently) broke the mirror.' b. ma-baata sisaa phuT-yo (Nepali; Madhav Pokharel, p.c.) I-ABL glass break-PERF 'The glass broke (and inadvertently I happened to be the cause of

it).-

In these constructions, the initial NP (marked instrumental or ablative) provides a domain to which the occurrence of a particular state of affairs can be attributed.

336 · Masayoshi Shibatani and Prashant Pardeshi

The proposed analysis of the dative construction as an instantiation of a double subject construction involves positing the following structure: (91) Dative subject constructions S,

NPi rAm-lA Ram-DAT

S2 NP2 sitA Sita.NOM.F

PRED AwaD-t-e (Marathi) like-PRES-F 'Ram likes Sita.'

(92) Variant constructions a. [NP-DAT[(NP-ACC/DAT)PRED]] maTA aswAwawA elewunna (Sinhala; Gair 1990:19) I.DAT horseACCchase.P.PAST 'I involuntarily (on impulse) chased the horse.' maTA lamAyaTA baenuna (Sinhala; Gair 1990:19) I.DAT child.DAT scold.P.PAST º (involuntarily/on impulse) scolded the child.' maTA danAgsehuna (Sinhala; Wijayawardhana et αϊ 1995: 124) I.DAT kneel.RPAST º kneeled/My knees gave way' (e.g., on seeing the terrible accident). b. [NP-GEN [(NP-NOM/ACC) PRED]] ama-r baba achen (Bengali; Onishi, forthcoming) ISG-GEN father.NOM be.have.PRES.2/3HON º have a father.' ama-r toma-ke bhalo lag-e (Bengali; Onishi, forthcoming) ISG-GEN 2ORD.SG-OBJ good be attached-PRES.3ORD º feel good about you.' t -r h si pa-cch-e (Bengali; Onishi, forthcoming) 3HON.SG-GEN laugh get-IMPERF-PRES.3ORD 'S/he (HON) feels like laughing.' c. [NP-OBL[NP-NOMPRED]] bacce se shiishaa TuuT gayaa (Hindi; Kachru 1990: 60) child INST mirror break went/PASS 'The child (inadvertently) broke the mirror.' ma-baata sisaa phuT-yo (Nepali; Madhav Pokharel, p.c.) I-ABL glass break-PERF 'The glass broke (and inadvertently I happened to be the cause of it).'

Dative Subject Constructions m 337

One may wonder about the status of non-nominative case-marked large subjects, normally unmarked or in the nominative case. However, nonnominative case-marked large subjects (in combination with the nominative case) do occur, as seen in the following Japanese sentences. (93) [kono heya-kara-ga (huzi—san-ga yoku mieru)] (Japanese) this room-from-NOM Fuji-Mt-NOM well visible 'It is from this room that Mt. Fuji is very visible.' (94) [Hanako-to-ga (itiban benkyoo-ga si-nikui)] H-with-NOM most study-NOM do-difficult 'It is with Hanako that studying is most difficult to do.' (95) \Tookyoo-made-ga (kuroo-ga ooi)] T-upto-NOM trouble-NOM many 'It is up to Tokyo that there are many troubles.' As the representations in (89) and (90) indicate, large subjects are not direct arguments of lexical predicates; instead they are subjects of clausal predicates. Predication of this kind requires a general constructional meaning that binds the two elements together. The notion of dependency discussed above (see Section 7) is a minimal meaning relationship that all double subject constructions must satisfy. In addition, dative constructions provide a clue as to how the large subject contributes to this dependency relationship. Case markers of the large subject NP indicate the degree of dependency between the large subject and the predicate clause, and sometimes provide a clue as to how the large subject is relevant to the state of affairs expressed in the predicate clause. In summary, our analysis provides a framework in which the question of differential marking on the large subject can be meaningfully pursued. The answer to this question requires a better understanding of the notion of dependency, especially the degree of dependency.

• 9. Degree of Dependency and the Case Marking of the Large Subject In this paper we claim that differential case marking on the large subject reflects different degrees of dependency between the large subject and the predicate clause. While admittedly it is not easy to determine how differential case marking on the large subject correlates with the notion of dependency in some cases, e.g., between dative and genitive marking, the Sinhala pattern below may provide a first approximation as to what we are attempting to elaborate. That inalienable possession in this language requires dative marking may indicate that the dative case signals a higher degree of dependency than the genitive; cf. (96a) and (96b).

338 · Masayoshi Shibatani and Prashant Pardeshi

(96) Sinhala: DAT > GEN > INST5 a. maTAfill mage lokuoluwak tiyenAwa (Kumara, p.c.) I.DAT/l.GEN big head-INDEF exists º have a big head.' b. maTA/mage loku kaekkumak tiyenAwa (ibid., p.c.) I.DAT/l.GEN big estate exists º have a big estate.' c. maTA eyaa gaenA matak unaa (Wijayawardhana etal. 1995:127) I.DAT he about remember.PAST.P º remembered him.' d. lamAyaTA naeTenAyva (ibid.: 1995:123) child.DAT dance.PRES.P 'The child is dancing willy-nilly (e.g., because, with the music, she cannot help it.) e. maa-atin ballaa maeruna (ibid.: 1995:116) I-INST dog kill.PAST.P º accidentally killed the dog.' f. ballaa maeruna (Kumara, p.c.) dog kill.PASTP. The dog died.' Deletion of the dative/genitive phrases in (96a)-(96d) results in elliptical sentences, indicating that dative and genitive are used to mark high dependency. The instrumental phrase in (96e), on the other hand, can be omitted; the resulting sentence in (96f) is a full sentence able to stand by itself. This indicates that the dependency relation between the instrumentally marked large subject and the predicate clause is low. The same point can be made with regard to the difference between a dative marked large subject and the ablative one in Nepali. (97) Nepali (Madhav Pokharel, p.c.) a. keTaa-laaii keT-i raamr-i laag-u boy-DAT girl-FEM beautiful-FEM feel-FEM 'The boy likes the girl.' b. keTi raamr-i laag-u (elliptical) girl-FEM beautiful-FEM feel-FEM '(Someone) likes the girl.' (98) a. ma-baata sisaa phuT-yo I-ABL glass break-PERF 'The glass broke (and inadvertently I happened to be the cause of it).'

Dative Subject Constructions m 339

b. sisaaphuT-yo (non-elliptical) glass break-PERF 'The glass broke.' Dative and genitive case are two favored patterns of marking the large subject in South Asian languages. In its prototypical use as a marker of an indirect object, the dative case marks an indirectly affected participant in an event. This property of marking an indirect entity seems to be a major contributing factor in the use of the dative case in dative constructions.6 In languages like Japanese and Korean, nominative case competes with the dative in marking the large subject, again reflecting the degree of dependency or relevance, with nominative case marking a more highly involved individual (see Shibatani 2000). As alluded to above (see footnote 1), genitive marking on a large subject seems to go back to the original possessive construction. Retention of genitive marking on the large subject is a reasonable way of indicating its relevance to the state of affairs described in the predicate clause, as a possessor is naturally involved in an event which affects his body part or possession. In Newari, however, there seems to be a movement toward shedding the genitive marking on the large subject, turning the genitive double subject construction into a nominative double subject construction of the Japanese type (see [99b] below). When a first or second person possessor is involved, it may not retain the genitive marking on the large subject, while with third persons genitive marking must be retained. (99) a. ji-gu dhaten chon syaa (Newan; Sujeet Pradhan, p.c.) I-GEN truly head.NOM hurt 4 1 truly hurt in the head.' b. ji dhaten chon syaa I.NOM truly head.NOM hurt º truly hurt in the head.' In their basic use, oblique markers like instrumental and ablative indicate more remotely relevant entities than those marked by the dative and the genitive case. In the passive and causative, they mark a passive and a causee agent, which are conceptualized and grammaticalized as a non-central or secondary—albeit essential—figure in the described event. Marking the large subject of the double subject construction by an oblique marker as a means of expressing an entity secondarily relevant to the main prediction also seems well motivated.

• 10. Distribution of Subject Properties The degree of dependency between the large subject and its predicate clause has an important grammatical consequence; namely, differences in the

340 · Masayoshi Shibatani and Prashant Pardeshi

distribution of subject properties. As we saw earlier, both the dative nominal and the nominative NP of a dative construction exhibit subject properties, but no precise formulation of how they are distributed over these two nominal elements has been attempted in the past. Our analysis posits two subjects—a large subject and a small subject—and provides a basis for a possible formulation in terms of the degree of dependency discussed in the preceding section. There are general areal, typological considerations that must be addressed in thinking about this problem. That is, a certain group of languages assign only a very small number of subject properties to the large subject. This appears to be the case with the group of so-called Standard Average European languages, including German, Dutch and French. There is a high typological pressure among these languages to align the distribution of subject properties with the nominative argument, so that a uniformity in the alignment of morphology and syntax is achieved. In these languages, large subjects do not seem to exhibit phenomena pointing unequivocally to their syntactic subjecthood (see Haspelmath forthcoming). In other languages—where there does not seem be a strong requirement for uniform morphology-syntax alignment—the case-marked large subject exhibits a fair number of subject properties. In such a situation, the higher the dependency between the large subject and its predicate clause, the greater the number of subject properties the large subject exhibits. In other words, a large subject of higher relevance syntactically dominates over a smaller one. On the other hand, the small subject asserts its subject status more strongly when the dependency relation is low. Earlier it was pointed out that of dative and oblique (instrumental/ablative) marking, the former correlates with a higher degree of dependency between the large subject and the predicate clause. This would predict that a dative marked large subject exhibits a greater number of subject properties than an obliquely marked one. Corroborating evidence is found in the earlier work on Indie languages by Kachru et al. (1976), who noted that different kinds of subjects in Hindi and some other Indie languages show a different degree of subjecthood, as summarized below: (100) Degree of subjecthood among Indie languages (Kachru et al. 1976: 94) SI ST < S DAT < S OBL < SP7 (SI = intransitive subject, ST = transitive subject, S DAT = dative subject, S OBL = oblique subjects, SP = derived subject of the passive) The degree of dependency we have been discussing is a highly subtle matter. We expect this subtlety to be reflected in the grammar such that even those constructions involving the same marking on the large subject may exhibit subtle but crucial differences in their patterning. This would suggest that hierarchical representations like the ones in (96) and (100) are gross representations

Dative Subject Constructions m 341

of the overall picture. In reality, similarly marked large subjects may have different syntactic properties, and an obliquely marked large subject may show a greater degree of subjecthood than a dative marked subject, depending on the type of construction and the grammatical phenomena being examined. Thus, individual constructions must be carefully examined with respect to specific grammatical phenomena in assessing their properties. To appreciate this, consider again the following reflexive binding patterns in Marathi: (101) rAm-lA sitA swatahA-cyA bahlNI pekshA jAsta Ram-DAT Sita.NOM self-of sister than more AwaD-t-e like-PRES-N 'Ram likes Sita more than his (Ram's/*Sita's) sister.' (102) ma-ΙΑ to swatahA-cyA gharAt dis-l-A. I-DAT he.NOM self-of house.in see-PERF-3M º saw him in his/*my house' (Pandharipande 1990:165). Among native speakers of Marathi there is no disagreement that the dative nominal (i.e., the large subject) in (101) exclusively binds the reflexive form. In (102), on the other hand, the nominative NP (i.e., the small subject) controls reflexive binding. (A few speakers find the sentence to be ambiguous, but the preferred reading for them is the one in which the small subject is construed as the antecedent.) This clear difference in the interpretation of reflexive binding indicates that dative constructions are not uniform in their interpretation. Our claim is that the degree of dependency between the large subject and the predicate is a factor contributing to this. The verb roots AwaD- 'like' and dis'see/visible' used in the examples above differ in their usage. While the former can be used only in a dependent construction, the latter has an independent usage. Compare the following; (103) sitA AwaD-t-e Sita.NOM like-PRES-N '(Somone) likes Sita.' (104) ith-un bAg dis-t-e here-from garden.NOM be visible-PRES-N The garden is visible from here/ Sentence (103) is clearly elliptical—it must be made dependent upon a large subject, which functions as a cognizer experiencing the emotion of liking Sita. This is not the case with the verb root dis- 'see/visible,' as used in (104), which functions as a complete sentence describing the nature of the subject— namely, that the garden is in a visible state. Our claim is that this difference is reflected in (101) and (102). In (101), the large subject has dominance over the small one with regard to reflexive binding, as it has a greater contribution to

342 · Masayoshi Shibatani and Prashant Pardeshi

the realization of the state of affairs expressed in the predicate clause. On the other hand, the state of affairs expressed in the predicate clause, cf. (102) is less dependent upon the large subject: something can be in a visible state independently from an actual viewer. Because of this relative independence, the small subject—the subject of the predicate clause—has greater autonomy and asserts its subjecthood with respect to reflexive binding. The discussion above has dealt with two extreme cases of dependency between the large and the small subject. If it has any validity, we can expect an intermediate degree of dependency in this relationship. Agentive passives— passive clauses with an overtly expressed agentive phrase—may show just such an intermediate situation. In order to see this, it will be useful to briefly look at the general characteristics of Indie passives. In a highly useful survey of what he calls experiential constructions in South Asian languages, Hook (1990) notes that the morphosyntatic properties of dative constructions resemble those found in passive or (negative) potential constructions of the following type: (105) rAm-kaDUn noTis wAc-l-I ge-l-I (Marathi) Ram-ABL notice.ENOM read-PERF-F go-PERF-F 'The notice was read by Ram/Ram could read the notice.' Here we extend Hook's analysis and claim that Indie agentive passives are really variants of dative constructions. Indie passives are peculiar in that their agentive phrases exhibit subject properties unlike adjunctive phrases in other languages such as Japanese and English. Pandharipande (1981) summarizes the subject properties of the Marathi passive agent and compares them with the basic subject, as well as with the subject of a passive clause, as follows: (106) Subject properties of the passive-agent in Marathi; Pandharipande (1981: 86)

Reflexivization Conjunction reduction (a) undergoes (b) controls Equi-NP deletion (a) undergoes (b) controls Raising Participialization 1 Participialization 2

Basic subject

Passive-agent

Passive-subject

+

+

+

+ +

+

+ + + + +

+ -

+ +

By claiming that agentive passives in Indie languages are variants of dative (or double subject) constructions, we are essentially proposing an analysis in

Dative Subject Constructions m 343

which the agentive phrase occurs in the large subject position of the double subject construction, as below: (107) (Marathi)

rAm-kaDUn Ram-ABL

NP noTis notice.ENOM

VP wAc-l~I ge-l-I. read-PERF-F go-PERF-F

'The notice was read by Ram/Ram could read the notice.' The proposed analysis naturally explains why in these languages the passive agentive phrase occurs in sentence initial position, along with many subject properties. The parallelism between the passive agentive phrase and the dative subject is naturally captured. It also explains the fundamental difference between the agentive phrase of Indie passives, and that of the passive clauses in other languages, where the agent is coded as an adjunct.8 If the current situation in Gujarati is any indication, modern Indie passives do/did not permit an overt expression of an agent, as in the case of contemporary Gujarati.9 When an 'agent' is expressed in Gujarati, the sentence is interpreted as a (negative) potential expression, as is the case with intransitive passive forms with an overt 'agent' in Hindi and other languages. (108) Gujarati passive and impersonal-type expressions (Lambert 1971:167) a. Em yukti kArtδ bAdha cor pAkaD-a-ya this means by all thieves.NOM catch-PASS-PAST 'By this means all the thieves were caught.' b. 'ahiN gujarati bolay ehe (Cardona 1965:117) here Gujarati speak.PASS be 'Gujarati is spoken here.' c. a rAst-e sTeSan-e jAvay ehe (Lambert 1971:168) this street-by station-to go.PASS be One goes to the station by this road/ (109) With expressed 'agents' a. rnara-thii jAngAl-mδ ekla nAhi jAvay (ibid.: 169) I-ABL jungle-in alone not go.PASS.PAST 'I couldn't go into the jungle alone.' b. tEna-thii evii kADAvii dAva piivaii he-ABL such bitter medicine drink.PASS.PAST nAhi (Ibid.:169) not 'He couldn't drink (take) such bitter medicine.'

344 · Masayoshi Shibatani and Prashant Pardeshi

We take it that the overt agentive phrase in the passive clause in modern Indie languages other than Gujarati is the result of an extension of the agent of the (negative) potential construction of the above type. Recall that potentials are typically expressed in the form of dative constructions in some languages. Our claim is that the agentive passive in modern Indie languages is based on such a potential construction. In this connection, Japanese shows an interesting and instructive development of the potential and the passive construction. The following sentence without an overt agent is ambiguous between a potential and a passive reading. (110) korera-no kodomo-tati-wa osie-rare-na-katta these-of child-PL-TOP teach-PASS-NEG-PAST (i) These children were not taught.Y(ii)These children couldn't be taught.' When the agent is expressed in initial position (i.e., the large subject position of the double subject construction), a potential interpretation obtains, whereas a passive one results when the agent is expressed sentence internally.10 (111) a. Ken-ni-wa korera-no kodomo-tati-ga Ken-DAT-TOP these of child-PL-NOM osie-rare-na-katta teach-PASS-NEG-PAST 'Ken couldn't teach these children.' b. korera-no kodomo-tati-wa Ken-ni osie-rare-na-katta these-of child-PL-Top Ken-DAT teach-PASS-NEG-PAST These children weren't taught by Ken.' To summarize, while Japanese potential and passive constructions diverge structurally—the former is a dative (i.e., double subject) construction, the latter is not—the potential and the agentive passive in modern Indie languages are built on the same syntactic template, namely the double subject structure. The subject properties of the agentive phrase of the Indie passive identified by Pandharipande (1981) and others derive from this structural property. The distribution of subject properties in agentive passive clauses (see [106]) indicates that the properties of the large subject (the agentive phrase) and the small subject (the subject of a passive clause) split in a balanced manner. This indicates that passive clauses of this type show a middle degree of dependency between the large subject (the agentive phrase) and the predicate clause (the passive clause). It is possible for a passive clause to stand by itself without an agentive phrase. Passive clauses without an agentive phrase are autonomous; i.e., they are not elliptical. They do imply the existence of an agent, however. Once this agent is overtly expressed in large subject position, the realization of the event expressed in the passive (predicate) clause is made entirely

Dative Subject Constructions m 345

dependent on it—the event wouldn't have occurred without this agent. This schizophrenic character of passive clauses is reflected in the interpretation of reflexive binding phenomenon. Without an overt agent, the passive subject controls the reflexive. In the example below, patient retains the dative marking, and the verb is in the neuter form, indicating that the sentence is impersonal. Yet, the dative marked patient behaves like a subject with respect to reflexive binding. This shows that in Marathi too impersonal passives are becoming more like intransitive clauses with a dative marked subject— although semantically they still retain a passive force11 (cf. earlier discussion on Sinhala P-forms in Section 4). (112) fyA mulA-lA swatAhA-cyA kholl-t mAr-l-e ge-l-e that boy-DAT self-GEN room-in kill-PERF-N go-PERF-N That boy was killed in self's room.' The situation is more problematic when an agent is overtly expressed. Recall that in the case of (101) and (102), native speakers' reactions were fairly unanimous and decisive. However, they have a harder time agreeing on the following Marathi agentive passive: (113) rAm-kaDUn tyA mulA-lA swatahA-cyA kholl-t mAr-l-e Ram-ABL that boy-ACC self-of room-in kill-PERF-N ge-l-e go-PERF-N 'The boy was killed in self's room by Ram.' While some native speakers understand this sentence to mean that the room in which the child was killed is Ram's (the large subject), others report that the room is the child's (the small subject). While we expect this kind of sentence to be ambiguous between these two readings, native speakers seem to force themselves to one or the other. Whatever the actual situation turns out to be, when a larger number of native speakers are consulted, our initial survey supports the claim that dative constructions and their variants show a subtle variation along the dependency parameter. • NOTES 1. The fact that an adverb may intervene between the genitive form and the following nominal (as in this example) indicates that they do not form a constituent. The fact that the genitive and the semantic head agree in gender in Marathi indicates that these constructions historically arose from the genitive construction. 2. A similar analysis is suggested in Lindholm (1976), though it is is not fully developed and the extent of its application is not made clear. Among the past researchers in the field, Pandharipande (1990) explicitly rejects a transitive analysis and suggests that dative constructions are intransitive, at least in Marathi. The analysis proposed here shares the basic insights

346 · Masayoshi Shibatani and Prashant Pardeshi of these two works; however, it is independently developed from the double subject construction in Japanese. 3. As in other Indie languages both animate direct and animate indirect objects are marked by LA, while inanimate objects are unmarked. In this paper, we gloss -LA as DAT and unmarked nominals as ÍÏÌ. 4. The grammatical role of the dative/genitive nominal in the above representation will be made clear in Section 8. 5. Kachru (1990: 66-7) suggests the order of GEN> DAT> LOC/INST for Hindi. 6. Recall the Dativ (in)commodi and the Dativ ethicus in European languages. 7. Here, the sign' '; i.e., the dative subject displays a greater number of subject properties than the oblique subject. 8. The proposed analysis of the passive clause also makes more plausible a scenario for the passive-to-ergative change in Indie languages and elsewhere. 9. Since the Gujarati passives we are examining involve synthetic (older?) forms, the comparison drawn here between them and the analytic passives (using the auxiliaries 'go' and 'come') in other languages may not be entirely appropriate. 10. A potential interpretation is possible with (11 Ib), but the preferred position of the potential 'agent' is in initial position as in (Ilia). 11. When the patient occurs in the (unmarked) nominative form, the passive form is clearly usable as a non-passive intransitive sentence with a spontaneous meaning. The following Marathi sentence is therefore ambiguous: to mulgA mAr-l-A ge-l-A that boy.NOM kill-PERF-M go-PERF-M "The boy was killed/The boy died.'

• REFERENCES Abbi, A. 1990. Experiential constructions and the 'subjecthood* of the experiencer NPs in South Asian Languages. In Experiencer subjects in South Asian languages, ed. by M.K. Verma and K.P. Mohanan, 253-68. Aikhenwald, A., R.M.W. Dixon and M. Onishi (eds.) Forthcoming. Non-canonical marking of subjects and objects. Amsterdam: John Benjamine. Cardona, G. 1965. A Gujarati reference grammar. Philadelphia: The University of Pennsylvania Press. . 1976. Subject in Sanskrit. The notion of subject in South Asian languages, ed. by M.K. Verma, 1-39. Clark, T.W. 1963. Introduction to Nepali. Cambridge: W. Heffner & Sons. Gair, J.W 1990. Subjects, case and INFL in Sinhala. Experiencer subjects in South Asian languages, ed. by M.K. Verma and K.P. Mohanan, 13-42. Harris, A. 1984. Inversion as a rule of universal grammar: Georgian evidence. Studies in relational grammar 2, ed. by D. Perlmutter and C. Rosen, 251-91. Haspelraath, M. Forthcoming. Non-canonical marking of core arguments in European languages in Non-canonical marking of subjects and objects, ed. by A. Aikhenwald et al. Hook, P.E. 1990. Experience« in South Asian languages: a gallery. Experiencer subjects in South Asian languages, ed. by M.K. Verma and K.P. Mohanan, 319-34. Jake, J.L. 1985. Grammatical relations and Imbabura Quechua. New York: Garland Publishing. Jayaseelan, KA. 1990. The dative subject construction and the pro-drop parameter. Experiencer subjects in South Asian languages, ed. by M.K. Verma and K.P. Mohanan, 269-83. Kachru, Y. 1990. Experiencer and other oblique subjects in Hindi. Experiencer subjects in South Asian languages, ed. by M.K. Verma and K.P. Mohanan, 59-75.

Dative Subject Constructions m 347 Kachru, Υ., B.B. Kachru and T.K. Bhatia. 1976. The notion 'subject', a note on Hindi-Urdu, Kashimiri and Panjabi. The notion of subject in South Asian languages, ed. by M.K. Verma, 79108. Kuno, S. 1973. The structure of the Japanese language. Cambridge, MA: MIT Press. Klaiman, M.H. 1980. Bengali dative subjects. Lingua 51. 275-95. Lambert, H.M. 1971. Gujarati language course. Cambridge: Cambridge University Press. Lindholm, J.M. 1976. Nested case relations and the subject in Tamil. The notion of subject in South Asian languages, ed. by M.K. Verma, 152-82. Masica, C.P. 1991. The Indo-Aryan languages. Cambridge: Cambridge University Press. Mistry, PJ. 1976. Subject in Gujarati: an examination of verb-agreement phenomenon. The notion of subject in South Asian languages, ed. by M.K. Verma, 240-69. Mohanan, K.P. and T. Mohanan. 1990. Dative subjects in Malayalam: Semantic information in syntax. Experiencer subjects in South Asian languages, ed. by M.K. Verma and K.P. Mohanan, 4358. Onishi, M. forthcoming. Non-canonically marked S/A in Bengali. Non-canonical marking of subjects and objects, ed. by A. Aikhenwald et al. Pandharipande, R. 1981. Syntax and semantics of the passive construction in selected South Asian languages. Urbana: University of Illinois Ph.D. Thesis. . 1990. Experiencer (dative) NPs in Marathi. Expenencer subjects in South Asian languages, ed. by M.K. Verma and K.P. Mohanan, 161-80. Pawley, A., S.P. Gi, I.S. Majnep and J. Kias. Forthcoming. Hunger acts on me: the grammar and semantics of bodily and mental process expressions. Kalam Grammatical Analysis in morphology, ed. by V. de Guzman and Byron Bender, 153-85. Honolulu: University Hawaii Press. Perlmutter, D. 1984. Working 1's and inversion in Italian, Japanese and Quechua. Studies in relational grammar 2, ed. by D. Perlmutter and C. Rosen, 292-330. Perlmutter, D. and C. Rosen (eds.) 1984. Studies in relational grammar 2. Chicago: The University of Chicago Press. Renck, G.L. \915.AgrammarofYagaria. Canberra: Department of Linguistics, Research School of Pacific Studies, The Australian National University. Shibatani, M. 2000. Non-canonical constructions in Japanese. Non-canonical marking of subjects and objects, ed. by A. Aikhenwald et al. . Forthcoming. Dative subject constructions twenty-two years later. Sridhar. S.N. 1976. Dative subject, rule government, and relational grammar. Studies in the Linguistic Sciences. 6 (1). 130-51. Department of Linguistics, University of Illinois, Urbana-Champaign. Verma. M.K. (ed.) 1976. The notion of subject in South Asian languages. Madison: South Asian Studies, University of Wisconsin-Madison. Verma, M.K. and K.P. Mohanan (eds.) 1990. Experiencer subjects in South Asian languages. Stanford: The Center for the Study of Language and Information, Stanford University. Wijayawardhana, G.D., D. Wickramasinghe and T. Bynon. 1995. Passive-related constructions in colloquial Sinhala. Subject, voice and ergativity, ed. by D.C. Bennett, T. Bynon and G.B. Hewitt, 105-41. London: School of Oriental and African Studies, University of London.

Morphological Diversity and Morphological Borrowing in South Asia1 ι RAJENDRA SINGH ι This paper argues (a) that a proper understanding of morphological diversity in South Asia, and hence of morphological typology in general, can be reached only when realistic morphological profiles of South Asian and other languages are constructed and (6) that treatments of morphological borrowing in general (and of such matters on the subcontinent in particular), need to (i) make more distinctions than they generally do and (ii) to see things differently. The morphological profile of Modern Hindi (cf. Singh and Agnihotri 1997) and re-analyses of some cases of morphological borrowing, including such cases as Kurux and Kharia, analyzed in Abbi (1998), from India are provided as examples.

• 1. Introduction I shall attempt in this paper to argue (a) that a proper understanding of morphological diversity in South Asia (and of morphological typology in general) can be reached only when realistic morphological profiles of South Asian (and other) languages are constructed and (b) that scholars who take up matters of morphological borrowing on the Indian subcontinent, particularly the ones that tend to see morphological convergence everywhere, need to make more distinctions than they generally do. Needless to say, I shall also examine the validity of the 'ideal' morphological typology that operates with terms such as agglutinating and flectional, particularly in the context of contact and convergence. In this, I shall follow Sapir (1921) and raise questions like the following: 'Do borrowing patterns of languages differ according to the (ideal) type they belong to?' If not, we are,

350 · Rajendra Singh

unfortunately, left with the Sapir-Bazell question: what is the content of this taxonomy? It needs a serious response, and I shall suggest that only realistic profiles can help answer that question. The fact that the profiles that can answer questions like the Sapir-Bazell question are based on what I shall call whole word morphology must, in my view, count as evidence against what Janda (1983) calls morphemology (or, equivalently, as evidence for whole word morphology [cf. Ford, Singh, and Martohardjono 1997]). At appropriate places in the text, I shall also discuss theoretical consequences and implications of the profile and re-analyses I intend to present.

• 2. Towards a Word-based Approach to Morphological Typology: An Illustration The purpose of this section is to present what I shall call the morphological profile of Modem Hindi. The profile to be presented is based on the description of Hindi morphology provided in R. Singh and R.K. Agnihotri (1997). The description itself is based on the assumptions and premises outlined and motivated in Ford and Singh (1991) and Ford, Singh, and Martohardjono (1997). As all descriptions are theory-dependent, it will be necessary to present the theoretical sign-posts that have guided the description this morphological profile is based on, and I shall do so, but without any attempt to justify them here. My main interest is to invite the reader to assess the profile to be presented as a possible tool for the construction of what for the lack of a better term I shall refer to as Realistic Morphological Typology. • 2.1. The Theory According to the theory of morphology developed by Ford and myself, all that needs to be said about word structure in any language (of any type whatsoever) can and must be said by instantiations of the schema in (1) below. We refer to these instantiations as W(ord) F(ormation) S(trategies) because as generalizations drawn from known particular facts, they can be activated in the production and understanding of new words (cf. Ford and Singh 1991 and Ford, Singh, and Martohardjono 1997).2 WFS's must be formulated as generally as possible, but, and this is crucial, only as generally as the facts of the matter permit. (1) /X/ a
/X'/b

where: a. /X/a and /X Yb are words and X and X' are abbreviations of the forms of classes of words belonging to categories a and b (with which specific words belonging to the right category can be unified or on to which they can be mapped); b. 'represents (all the) form-related differences between /X/ and /XV;

Morphological Diversity and Borrowing m 351 c. a and b are categories that may be represented as feature-bundles; d. the