The Yearbook of South Asian Languages and Linguistics: 2002 9783110245271

Citation preview

EDITORIAL BOARD CHIEF EDITOR: Rajendra Singh, D£pt. de Linguistique, Universit6 de Montreal, C.P. 6128 Succ. Centre-ville, Montreal H3C 3J7, Canada. Voice: (514) 343-2113; Fax: (514) 343-2284; e-mail: [email protected] ASSOCIATE EDITORS: Probal Dasgupta, Centre for Applied Linguistics, University of Hyderabad, Hyderabad 500046, India. Voice: (91-40) 258-500; Fax: (91-40) 258-120; e-mail: [email protected] REGIONAL EDITORS: R.K. Agnihotri, University of Delhi, India; Tej K. Bhatia, Syracuse University, NY, USA; W.S. Kanmatillake, University of Kellaniya, Sri Lanka; Baber S.A. 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 Osnabrück, Germany; Anjum P. Saleemi, The National University of Pakistan, Lahore, Pakistan; Udaya Narayana Singh, Central Institute of Indian Languages, Mysore, India; E. Tiffou, Universit6 de Montroal, Canada; Yogendra P. Yadava, 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.E Konrad Koerner, University of Ottawa, Canada; Bh. Krishnamurti, Hyderabad, India; Jayant K. Lele, Queen's University, Kingston, Canada; Marvin Minsky, MIT, Cambridge, MA, USA; Pieter Muysken, Universiteit van Amsterdam, The Netherlands; N.S. Prabhu, Bangalore, India. EDITORIAL ASSISTANT: Honore Kamany, Universito de Montroal 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.

-h The Yearbook of South Asian Languages and Linguistics 2002 Chief Editor RAJENDRA SINGH

Available From ·

SERIALS SUBSCRIPTION SERVICE (INDIA)

SUBSCRIPTION AGENT · BOOK SELLERS PUBLISHERS · STOCKISTS OF BACK-VOLUMES 4, B.S.S. HALL, KHYBER PASS MARKET, CIVIL LINES DELHI-1 10054 (INDIA), PH. '.3811659 , E-mail ; wlaKOMtyam. . . FAX : 91-11-3812076, BRANCH OFFICE : 4630/24, AN8ARI ROAD, DARYA QANJ NEW DELHM 10002 (INDIA). PH.: 3246225

Sage Publications New Delhi · Thousand Oaks · London

Copyright © Rajendra Singh, 2002 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 2002 by Sage Publications India Pvt Ltd M-32 Market, Greater Kailash-I New Delhi 110 048 Sage Publications Ine yÄi Sage Publications Ltd 2455 Teller Road 1^1 6 Bonhill Street Thousand Oaks, California 91320 \Sr London EC2A 4PU Published by Tejeshwar Singh for Sage Publications India Pvt Ltd, typeset by Asian Telelinks, New Delhi and printed at Chaman Enterprises, Delhi.

ISSN: 0971-9539

ISBN:0-7619-9694-X(US-HB) 81-7829-162-2 (India-HB)

>

V

BIBLIOTHEK

LEIPZIG

' £r\~ liVV $~ - (f 'p / ( ? * . ' '

Two, Sage Production Team: Abantika Chatterjee, Shweta Vachani, Sushanta Gayen and Santosh Rawat

Contents Commentum Editoris A. INVITED CONTRIBUTIONS Sociolinguistic Theory and Practice: The Indian Counterpoint R.K. AGNIHOTRI Agreement Features and Projections of TENSE and ASPECT ALICE DAVISON Sanskrit as She has been Misanalyzed Prosodically RICHARD D. JANDA and BRIAN D. JOSEPH B.

C.

OPEN SUBMISSIONS The Nepali Converbs: A Holistic Approach JOHN PETERSON Three Levels of Lexical Codification ANITA RAVANAM Syntax Learnability: The Problem that Won't Go Away ANJUM P. SALEEMI On the Modal Meanings of the Subjunctive in Hindi GHANSHYAM SHARMA .

7

11 27 59

93 135 157 177

REGIONAL REPORTS, REVIEWS AND ABSTRACTS Regional Reports 1. Europe JOHN PETERSON 2. India PROBAL DASGUPTA Reviews Anvita Abbi (ed.), Languages of Tribal and Indigenous Peoples of India: The Ethnic Space by ALICE ANUGRAHAM Suresh Canagarajah, Resisting Linguistic Imperialism in English Teaching by OTTO IKOME

201 217

233 237

6 · Contents Tariq Rahman, Language, Education and Culture by ETIENNE TIFFOU W.C. Ritchie and T.K. Bhatia (eds), Handbook of Child Language Acquisition and Handbook of Second Language Acquisition by OTTO IKOME Udaya Narayana Singh (ed.), The Second Turn: Papers on Literary Translation by PAUL ST. PIERRE D.

241 243 248

DIALOGUE Minimal Look-Ahead TANMOY BHATTACHARYA The Problem that Won't Go Away: A Response to Anjum Saleemi TERESA SATTERRELD Against Afghanistanism: A Note on the Morphology of Indian English RAJENDRA SINGH

269

About the Chief Editor Notes on Contributors

275 277

253 263

Commentum Editoris We are happy to put this the fifth volume of The Yearbook of South Asian Languages and Linguistics in your hands. It contains invited contributions on the state of South Asian Sociolinguistics (Agnihotri), the agreement features and projections of tense and aspect in Hindi-Urdu (Davison), and the representational games that have been played with Sanskrit by some contemporary formalists (Joseph and Janda). It is our sincere hope that Agnihotri's and Joseph and Janda's provocative contributions will encourage South Asianists to take stock of things and do what they think is needed, including, of course, setting the record straight, if necessary. The Open Submissions section contains Ravanam's insightful attempt to restore the linguistics in sociolinguistics, Saleemi's careful study of learnability, and Sharma's elegant analysis of modals in Hindi. Although we did receive a few submissions from Ph.D. students, we regret that none could be included in this volume. We hope some of these potential contributors will come back after making the rather extensive revisions they were asked to make before their papers could be reconsidered. The Reports subsection of Reports, Reviews and Abstracts is rather thin this year, mostly because quite a few of our Regional Editors have recently changed their affiliations—U.N. Singh, for example, has moved from Hyderabad to Mysore and Anjum P. Saleemi from Denmark to Pakistan. What is included, however, is interesting indeed—Peterson's fully documented report on Europe and Associate Editor Probal Dasgupta's personal assessment of the state of linguistics in South Asia, India, in particular. The Review subsection offers the usual fare—expert reviews of a handful of recent important publications. As usual, we conclude with the Dialogue section, which contains Bhattacharya's invitation to examine what he argues is quite possibly a fallacy

8 · Contents in the formulation of an important principle in current generative syntax and Singh's invitation to revisit the morphology of Indian English. Rajendra Singh Probal Dasgupta

Invited Contributions

forR.S.

Sociolinguistic Theory and Practice: The Indian Counterpoint1 ι R.K. AGNIHOTRI This paper briefly examines different theoretical proposals in sociolinguistic enquiry. It tries to show their limitations as they do not give us any useful insights into the nature and structure of relationship that may obtain between language and society, in particular the ways in which language may connive with other attributes of power in society in perpetuating exploitation. Several studies done in the Indian multilingual context constitute a counterpoint to the quantitative, deficit, ethnographic and social psychological approaches to the study of language. More than anybody else a sociolinguist is responsible to suggest at least some hints towards the understanding and resolution of Plato's Problem and Orwell's Problem.

• 1. Introduction If we examine the linguistic scene since Saussure, we notice that though there has been significant progress in the formal areas of phonology, morphology and syntax, the world of sociolinguistic enquiry, based largely on the sociological theory of consensus (cf. Williams 1992), has remained rather stagnant and repetitive. It is very difficult to say whether we have made any significant progress in our understanding of the complex relationship between language and society since Bloomfield (1933) and Weinreich (1953). There is no doubt that there was a spurt of vigorous sociolinguistic research during the period 1965-85 and several models of sociolinguistic analysis including variationism, sociology and social psychology of language, ethnography of communication, elaborated and restricted codes and social networks were formulated. In fact,

12 · R.K. Agnihotn even before Bloomfield and Weinreich, we have the very important work done, among others, by the early dialectologists such as Wenker, Gillieron, Grierson and Hans Kurath in the late 19th and early 20th centuries. We also have the extremely bold hypothesis of linguistic relativity put forward by Sapir (1929) and Whorf (1940/1956). However, we do not witness any major theoretical breakthroughs in the analysis of the relationship between language and society. It is almost with a sense of frustration that we notice that most sociolinguistic work is neither sensitive sociology nor rigorous linguistics. The essential sociolinguistic questions concerning the role of language in the consolidation of social power and exploitation on the one hand and the assertion of human dignity and freedom on the other do not even form a part of the sociolinguistic agenda. Nor do we see any striking consequences of the intervention of sociolinguistic theory or practice in such applied areas as literacy, language teaching, education or language planning. Singh's (1996) Lectures Against Sociolinguistics actually constitute a telling commentary on the futility of a greater part of our sociolinguistic enterprise. Unfortunately, there has not been any powerful theoretical voice from the Indian subcontinent either. However, the sociolinguistic work done in India has consistently illustrated the inadequacies of the various models located essentially in monolingual and monocultural settings. It has constituted a counterpoint to the mainstream sociolinguistic research and has in a sense forced scholars to reexamine their proposals.

• 2. Variability Paradigm In his chapter on the 'Speech Communities', Bloomfield (1933: 42-56) explored the issues of the formation of speech communities, language standardization, bilingualism, language maintenance and language shift and language variation. Weinreich (1953) explored the issues centering around language contact in great detail. Weinreich, Labov and Herzog (1968) argued that grammars obtain in speech communities and that an understanding of language will have to be based on a description of its orderly differentiation in a speech community. It is not easy to identify the ways in which we have gone beyond their formulations in these areas. In terms of sociolinguistic research methodology, die work of Labov (1966) and Trudgill (1974) among others has certainly made significant improvements over the methods used by the early dialectologists. It is because of their efforts that we realize the significance of the vernacular/casual speech for sociolinguistic analysis and we owe the contemporary techniques of data elicitation (including conceptualization of systematic stylistic variation and resolution of the Observer's Paradox) and data analysis to them. However, Labov was aware that purely correlational studies of the social and linguistic variables will lead us nowhere and he therefore

Sociolinguistic Theory and Practice: The Indian Counterpoint m 13

turned his attention to structural issues trying to trace hints of diachronic change in synchronic linguistic analysis across different generations in a given speech community. But so far as social norms were concerned, Labov assumed a kind of inevitability and neutrality about them. Labov (1966: 64) quotes Barber (1957) at some length to suggest that social stratification is the product of 'normal workings of society* and that differentiated linguistic forms are ranked in status by 'general agreement' among the members of a speech community. We know for sure that this is not true. Labov (2001: 503) says, 'a certain linkage between social dimensions and linguistic traits is established as the result of an arbitrary and accidental concatenation in history'. It is always the variants used by the powerful that are ranked high in a society; the variants used by the lower classes are invariably stigmatized in spite of the certificate of systematicity they receive from the liberal linguist. The sociolinguist's faith in the theory of social consensus is almost unethical. Despite his caution, Labov's pioneering work in Sociolinguistic methodology set in motion a series of correlational studies across the world including India; studies that delighted in statistical sophistication but said nothing substantial either about language or about society or their complex relationship with each other. But as we used this approach in traditionally multilingual societies and minority migrant communities, we realized the inadequacy of the theoretical assumptions underlying it. For example, my study (1979) of the Punjabi Sikh migrants in Yorkshire in England showed that the Sikh children were being penalized for doing exactly what was expected of them, namely assimilate the native patterns of behavior. They had no access to the prestigious R.P. variants and were likely to be discriminated against in the university education and job market for speaking like Yorkshire people. Not only that, their emergent Punjabi-English mixed code defied any analysis in terms of standard linguistic paradigms. Though it was a marker of their identity and an important means of intra-group communication, it remained an object of ridicule among both the host and migrant community. Socio-cultural accommodation and assimilation was thus not simply a matter of acquiring a high frequency of the diagnostic variants of the host community's speech. Similarly, Mukherjee's (1980) and Satyanath's (1982) studies have shown that Punjabis, Bengalis and Kannadigas react very differently to the overwhelming pressures of the predominantly Hindi Delhi majority society. Hindu Punjabis assimilate maximally and Bengalis minimally; Kannadigas assimilate considerably allowing Hindi even into their homes but maintain Kannada for a variety of cultural and religious activities. It was clear that the behavior of these communities could not be analyzed independent of the pressures of the host society and their own histories, attitudes, stereotypes and future aspirations. These studies also brought out the limitations of Fishman's (1972,1978) model that required association of distinct languages with different domains and wanted us to

14 · R.K. Agnihotri

understand the complexities of the sociology of language purely in terms of consensual domain configurations.

• 3. Ethnographic Approaches Before we turn to the Indian sociolinguistic scene, we may briefly examine some other theoretical approaches to the analysis of the relationship between language and society. Bernstein's (1971-75) deficit hypothesis proposed that linguistic behavior could be seen in terms of elaborated and restricted codes, the former being associated with the middle class and the latter with the lower class. He felt that the deficit between the elaborated and restricted codes may be closely associated with cognitive deficits. He also argued that it is primarily the social structure that determines linguistic behavior which in turn reproduces social structure. This was a hopelessly fatalistic situation for the redressal of which Bernstein could only propose remedial teaching. This proposal seemed to insult human speech and intelligence and appeared to most of us to ignore the real issues that should be addressed by a sociolinguistic enquiry. The roots of this proposal of course lie in the Sapir-Whorf hypothesis which was indeed bold enough to sustain empirical investigation into modern times (see for example, Gumperz and Levinson 1996). Whorf (1940/1956) argued that we dissect nature along lines laid down by our native languages and that different grammars create different worlds. In spite of Berlin and Kay's (1969) demonstration to the contrary, suggesting that basic colors have a language independent saliency, Whorfian hypothesis has fascinated many scholars. In his recent work, Gumperz (1996) has engaged in the analysis of communicative practices in which meaning is constantly being negotiated and ideology, in terms of overt and covert views about language and language use, plays a dominant role. As contemporary urban societies become increasingly complex both linguistically and culturally, there is very little that we can assume to be socio-culturally shared in a given communicative encounter. In fact, the encounter itself constitutes a matrix for creating shared spaces: "distinct communicative practices arise through communicative co-operation in different networks of relationship' (Gumperz 1996:363). Yet to a sociolinguist interested in the issues of social oppression and individual creativity, this is not particularly helpful. However, the early work of Gumperz and his colleagues seemed particularly relevant to multilingual contexts. For example, Gumperz and Wilson (1971) analyzed the extremely complex sociolinguistic contact in Kupwar where in addition to distinct religious and cultural groups, Indo-Aryan Marathi and Urdu and Dravidian Kannada and Telugu were being simultaneously used and were often mixed in a variety of complex ways. It was interesting to note that Dravidian Kannada and Indo-Aryan Marathi had

Sodolinguistic Theory and Practice: The Indian Counterpoint m 15

become almost identical in their deep structure and syntax while retaining the identity of their individual lexical and grammatical formatives. It appeared that widely different languages could develop a common syntax and lexify according to the needs of specific situations. It was also clear that local varieties could converge while the standard varieties could continue to diverge. I think the Kupwar study was ample demonstration of the fact that processes of pidginization and creolization are not unique to pidgin situations but are active in all human interactions. It was not clear though how the local power structures interacted with the processes of convergence. In fact, Gumperz's subsequent work (e.g., 1982a, 1982b) focused on breakdowns in communication often suggesting that the onus of understanding the misunderstood person rested squarely with the person who was misunderstood; the dimensions of power and human negotiation of meaning were completely overlooked in this discourse'analysis. Singh (1996: 97-98) points out three major serious sociological problems with such an approach: first, culture is seen as a static object; second, miscommunication may actually be in the mind of the analyst rather than in actual fact which may be characterized by negotiation; finally, such an analysis reduces the culture of other people to 'ethnicity'. Since most ethnographic proposals were located in monolingual situations or at best in monolingual, monocultural approaches to multilingual and pluricultural situations, it is not surprising that this approach was rarely pursued in India. The fluidity of multilingual situations has been at the heart of the work of Le Page (1968), Le Page and Tabouret-Keller (1985) and their colleagues. Le Page believes that an individual creates his systems of verbal behavior in terms of the groups she wishes to be identified with depending upon her ability to identify these groups, her motives, opportunities she gets and her ability to learn. These patterns are then projected onto the social screen, some in the process get focused while others get diffused. Every human encounter is in essence a contact situation subject to the processes of pidginization and creolization. According to Le Page, there must always be a number of theoreticallypossible outcomes to contact situations and a process of selection among those possibilities which is socially as well as linguistically mediated. The precise selection made in the creation of a new phonology, grammar and lexicon will depend upon the set of individual constraints, for example, the nature of our repertoire and motivation, interacting with the general internal economy of linguistic systems; the precise nature of the evolution of a communal phonology, grammar and lexicon will depend upon what the individuals themselves have created, and upon the social set of constraints, for example, density of interaction and common interests, again interacting through the general internal economy of linguistic systems (Le Page 1984: 115). Since it is located in fluid multilingual societies, the work of Le Page and his colleagues has attracted several Indian scholars (see for example, Agnihotri 1979/1987;

16 · R.K. Agnihotn Mukherjee 1980/1996; Sahgal 1983; Satyanath 1982 among others). Yet these formulations appear to exist in a world of social and linguistic neutrality. The hierarchical power structures that are constitutive of any communicative encounter do not seem to impinge on the creation of systems of verbal behavior; nor do we see any hints of the 'discernable shape and appreciable architecture' of human linguistic knowledge (Singh 2001) and it is not very clear what we mean by the economy of linguistic systems. As Singh (2001) argues, the discomfort of the sociolinguist, particularly of the ones working in multilingual societies, with the Chomskyan enterprise is understandable. Their scepticism is indeed justified but their retreat from examining the overall architecture of language is not.

• 4. Social Psychological Perspectives Another theoretical perspective that has had considerable influence in India is the one offered by Gardner and Lambert (1972) and Gardner (1985). In this social psychological model, language learning is essentially a matter of imposing 'elements of another culture into one's own lifespace' (Gardner 1979: 193). If you are integratively rather than instrumentally oriented and wish to become a part of the target language community, you are likely to have higher levels of proficiency in the target language. Even though Gardner (1988) knew that this was not the whole story, he noted that, other things being constant, those who are integratively oriented 'will probably be motivated to learn the other language, will probably have positive attitudes toward the other community, will probably view the language learning situations positively, will probably seek opportunities to practise the language, etc....and therefore, will probably be more successful in learning the second language than individuals not so motivated' (ibid.: 105). The way 'probably' has been used above actually suggests 'definitely1. Unfortunately, that is not the case. As elsewhere in the world, this hypothesis set in motion a series of studies in India. Several studies (Agnihotri et al. 1988; Khanna 1983; Khanna and Agnihotri 1982; Mathur 1991; Sahgal 1983, 1992; Virmani 1991 among others) conducted in India showed that success in second language learning was not so much a matter of attitudes and motivation as of overwhelming social pressures. Even when major social variables like schooling and socio-economic status were held constant, what turned out to be important were variables like exposure to the target language and patterns of language use, variables that in no way can be held constant. Let me just give you a couple of examples. We (Agnihotri et al. 1988) worked with a sizeable sample of 356 first year undergraduates of the University of Delhi. We examined their proficiency in the use of tenses in English in relation to 19 different variables such as gender,

Sociolinguistic Theory and Practice: The Indian Counterpoint · 17

socio-economic status, schooling, exposure to English, patterns of language use, attitudes and motivation and claimed control over English. The results of correlational and regression analysis showed that schooling was the most significant predictor of achievement followed by exposure to English and patterns of language use. Our initial reaction was that it is the setting, i.e., native vs non-native that may make the difference; we tried to argue that social psychological variables will be important where the native target language community is present. We were again proved wrong in Khanna, Verma, Agnihotri and Sinha (1998) in which we examined the proficiency in English of 133 UK adult immigrants from different parts of the world. Once again, the variables that turned out to be important included claimed control over English, patterns of language use, exposure to English, use of English in the family, etc. It was also clear that the acquisition of English could not be meaningfully examined independent of learners' commitment to their heritage languages. Mathur (1991) showed that proficiency in learning German as a foreign language in Delhi depended more on education and exposure to German than on individual differences in attitudes and motivation. In fact, the work of Khanna (1983), Au (1988) and Agnihotri and Khanna (1997) clearly showed that the conclusions drawn by Gardner and Lambert about the role of attitudinal and motivational variables in second language learning are actually not borne out by the correlational and factor analysis in Gardner and Lambert (1972). As I said at the beginning, the Indian studies, though not proposing any serious alternative sociolinguistic framework had clearly underscored the inadequacy of Lambert's social psychological model.

• 5. Mixed Codes In fact, all these approaches, be it the variability hypothesis or the deficit hypothesis, social psychological perspectives on language or the ethnography of communication, take the existence of 'a language' and the societal consensus about norms for granted. Language is seen first and foremost as a system, a pairing of a set of syntactic rules with a lexicon. The dimension of sociopolitical power that is always instrumental in the eventual construction and use of this system is consistently ignored. The behavior of a complex multilingual and multicultural speech community is then analyzed in terms of this system. Variants or utterances that do not conform to these systems are interpreted as deviations. We need to appreciate that there are certain Universal Constraints on language and within those constraints, languages are always in the process of making through individual creativity and societal struggles. There are invariably small elite groups that manipulate, in addition to other things, languages to maintain a status quo that is defined by exploitation;

18 · R.K. Agnihotri

there are also masses who often create new idioms to voice their protests and assert their freedom. With some notable exceptions (e.g., Avrorin 1977; Fairclough 1992; Kress and Hodge 1979; Lee 1992 among others), the work on the association of language with social power and political protest has remained rather limited. In any community, there are large areas of overlap of sociolinguistic competence among individual members; there are also areas of instant pidginization where languages effortlessly flow into each other and create new patterns of behavior and competence. Such phenomenon is seen in sharp relief in multilingual societies. We should look towards them for new models of sociolinguistic analysis. The severe limitations of working with the concepts of 4a language' and 'a speech community' become apparent as we try to apply them to pidgin, Creole and mixed languages. Whose languages are these? Which speech community do they belong to? What is the underlying system? When Labov (1971) was confronted with mind-boggling alternation of two or more languages, he said: 'so far no one has been able to show that such rapid alternation is governed by any systematic rules or constraints, and we must therefore describe it as the irregular mixture of two distinct systems' (ibid.: 457) and again, 'when actual conversation is examined, it usually appears that oscillation between the two (languages) takes place many times within a sentence, and to explain all variation as code-switching becomes increasingly artificial. It is equivalent to asserting that the speaker switches systems within a single utterance many times, unpredictably, without apparent motivation or conditioning' (ibid.: 461). When you start with the concept of a sociolinguistic system as a given, rather than with the conviction that human beings create language as they talk to each other in a matrix provided by a Universal Grammar on the one hand and the socio-political reality on the other, you are bound to be pushed to such a position. Effortless mixed code discourse, which is a normal way of talking in many multilingual societies, has made many linguists extremely uncomfortable. Their search for a system has become increasingly relentless, that is, they believe that at any given point of time a speaker must talk in one system which is already given. If they cannot fix it in a system, they begin to doubt the proficiency levels of speakers; you are said to mix languages because you cannot hold a sustained discourse in any one of them. This is indeed surprising because many of the mixed code users I know are actually nearly Bloomfieldian fluent bilinguals. The effortless normal discourse of bilinguals is not seen as a system in its own right but is subjected to segmentation from the point of view of the norms of individual language systems. In an almost arbitrary way, chunks are assigned to different languages involved in the mixed code discourse and a search for the 'matrix language' and the 'embedded language' starts; the hypothesis being that every constituent must belong to 'a language' (see for example, Di Sciullo et al. 1986; Kachru 1975; Pfaff 1979; Sankoff

Sociolinguistic Theory and Practice: The Indian Counterpoint m 19

and Poplack 1981; Sankoffet al. 1991 among others). Kachru (1975) proposed 'the determiner constraint' and 'the conjunction constraint' and Pfaff (1979) and Poplack (1981) proposed the 'equivalence constraint' and the 'free morpheme constraint'. The essence of these constraints is to suggest that switching can only occur in those areas of grammar that are shared by both the languages and that switching will occur at full-word boundaries only. According to the equivalence constraint, switches will tend to occur at points of discourse where juxtaposition of elements from the two languages does not violate a syntactic rule of either language. But literature on mixed codes is replete with umpteen examples that violate these constraints vigorously. The free morpheme constraint forbids mixed code users to inflect words of one language with the morphology of another, something they effortlessly do all the time. The equivalence constraint checks the mixing at points where the surface structure of the two participating languages is dissimilar; once again a phenomenon we often find in say Kashmiri-English mixed code, where the V-2 Kashmiri is mixed with the SVO English. Once again, the Indian counterpoint has indeed been telling (see for example, Agnihotri 1979,1998b; Kak 1995; Kakand Agnihotri 1996). In these studies, we not only demonstrated the frequent violations of the mixed code constraints but also showed that the mixed code sentences are perfectly acceptable to their users. We need to point out that constraints on mixed codes have been proposed by scholars who are far from being native speakers of mixed codes. This is just another example of the tyranny of monolingualism on sociolinguistic theory and practice. It is not being claimed that there are no rules and no constraints; it is just that our extant conceptual machinery of 'a language' with its attendant codified grammar and a dictionary may not be the most optimal to investigate human linguistic behavior. We need to train ourselves to notice irregularities. Paul Davies (1987) talks of a new approach in physical sciences according to which 'complexity and irregularity are seen as the norm and smooth curves the exception; in the traditional approach one regards complex systems as complicated collection of simple systems;...the new approach treats complex or irregular systems as primary in their own right' (ibid.: 22).

• 6. Empirical Validation If we wish that our sociolinguistic theory should be continuously enriched and refined, we need to seek constant empirical validation for it. If it is wellconceived, it will certainly change our understanding of the role of language in the processes of socio-political exploitation on the one hand and mass movements for freedom on the other. It should also change the face of our projects in the areas of literacy, education and language planning. Literacy

20 · R.K. Agnihotri

projects across the world are based on a minimalistic agenda; they insult the multilingual competence of the learner and have no place for the wealth of her prior knowledge. Literacy is invariably conceptualized as the skills to count, read the alphabet and write one's name. If literacy programs are conceived in a sociolinguistic framework, we may be able to appreciate the historical fact that successful literacy programs have generally been embedded in socio-political struggles; we may also be able to free ourselves from the 'banking' concept of education (Freire 1985) and enter into an authentic dialogue with the learners. Both in literacy programs and education, we have done everything in our power to undermine the multilingual and cognitive potential of learners destroying local literacies in the process (cf. Agnihotri 1994). As Illich (1981: 30) points out, 'we first shamelessly spend enormous amounts of money to standardize languages at the cost of ethnic languages telling our children and students how to speak and then spend token amounts to teach the counterfiets of ethnic languages as academic subjects'. The disasters of language planning—be it in India, Africa or the United States—are there for everyone to see. One is not sure whether the linguistic division of states, the three language formula or the continuation of English as an associate official language in India were sound sociolinguistic decisions. A more realistic language planning will perhaps start at the grassroots rather than be imposed from above. It is time that sociolinguistic theory moves towards adulthood (Agnihotri 1998a).

• 7. India as a Multilingual Country India is indeed a paradigm example of a complex system with its attendant charms and conflicts. Here several religions, cultures and languages coexist, now in peace, now in conflict. We notice systems that have continued unchanged since centuries and we witness new systems in the making everyday. People speak over 1600 different languages which belong to four different language families; the constitution of India mentions 18 major regional languages, an official language (Hindi) and an associate official language, that is, English (notice that we do not have a national language); school education is conducted in 47 different languages, radio uses 71 languages and print media is available in 87 different languages. We use one script for writing several languages, for example, Devnagari for Sanskrit, Hindi, Marathi, Nepali, etc., and write the same language in several scripts, for example, Sanskrit is written in Devnagari, Grantha, Malayalam, Telugu, Bhoti, Sharda, Bengali, Marathi, etc. And yet there is overwhelming evidence that India constitutes a linguistic, sociolinguistic and cultural area. Any sound sociolinguistic theory must come to terms with this multilingual and pluricultural phenomenon. As

Sociolinguistic Theory and Practice: The Indian Counterpoint · 21

Pandit (1972) pointed out, languages here are not barriers to communication; they facilitate it and effortlessly flow into each other. Pandit (1969) felt that bilingualism should be seen as a special frame for the observation of language change where besides the proposed models of stratified equilibrium, we may also understand the dynamics of multilingual situations. In India, we have on the one hand (for example) the powerful tradition of Bhakti poets who are even today understood and appreciated across large tracts of multilingual areas; we also have on the other hand a history of social protests and suicides associated with language policies. Whenever, a voice is raised for a separate cultural and territorial identity, language invariably forms an integral part ofthat voice. In such a situation you should not be surprised if you find that the establishment subsumes several different languages under a single symbolic rubric, for example, Hindi; nor should you be surprised if a single language such as Hindustani comes to be known as two distinct languages like Hindi and Urdu. You will notice that the three recently added languages to the Indian Constitution were relatively unknown. You will also notice that in the case of Santali, today four different scripts are competing for national recognition. As I have argued throughout this paper, India has made telling counterpoints to most theories in sociolinguistics. Alternative voices have also started emerging from here. May I mention some of them: Dasgupta (1993); Gupta (2000); Gupta and Aggarwal (1998); Heugh et al. (1995); Lele and Singh (1989); Ramakrishna (1997); Singh (1995, 1996, 1998); Singh et al. (1997).

• 8. The Challenge This paper makes a rather depressing reading. There are no easy solutions. But we do need to take the challenge head on. If we agree that the Indian studies constitute a counterpoint to most contemporary sociolinguistic theory, we need to work towards a new sociolinguistic theory which would at least try to do the following: (a) redefine the concept of 4a language' in such a way that it captures the fluidity inherent in multilingual situations; human linguistic competence is seen as essentially multilingual rather than monolingual; the normal mixed code data is not seen as pathological; (b) redefine the concept of a speech community in such a way that it no longer remains constrained by the limits of geographical or territorial encapsulation, a single speech code, frequency of interaction and an almost completely shared matrix of socio-cultural norms; this I think is extremely important in the modern context; as in the case of language, we need a more fluid concept of a speech community;

22 · R.K. Agnikotri

(c) respect the laws of Universal Grammar that constraint all speech and constantly try to understand what Chomsky calls 'Plato's Problem'; (d) examine the ways in which the socially powerful use of language to exploit the masses and the ways in which masses sometimes use language to assert their freedom; how is consent manufactured and how does what Chomsky calls Orwell's Problem comes about; (e) keep space for individual identity and creativity (in a non-Chomskyan extended sense); (/) examine human interaction as negotiation of meaning and power rather than as what Singh calls 'communicational asynchronies'. • NOTE 1. This is a revised version of the paper presented at the *Indology in University Curricula' workshop held at the Saint Petersburgh University, Saint Petersburg!), Russia during 21-23 September 2000.

• REFERENCES Agnihotri, R.K. 1979. Processes of assimilation: A sociolinguistic study of Sikh children in leeds. University of York, UK: D.Phil thesis (also published as Crisis of Identity. New Delhi: Bahri Publications, 1987). . 1994. Campaign-based literacy programmes: The case of the Ambedkar Nagar experiment. Language and Education 8 (1&2). 47-56. . 1998a. Sociolinguistics: Awaiting adulthood, ed. by R.S. Gupta and K.S. Aggarwal, 1998. 17-29. -. 1998b. Mixed codes and their acceptability, ed. by R.K. Agnihotri, A.L. Khanna and I. Sachdev, 1998. Agnihotri, R.K. and A.L. Khanna. 1997. The social psychological perspective on second language learning: A critique. Grammar, Language, and Society, ed. by R. Singh, 325-42. New Delhi: Sage. Agnihotri, R.K., A.L. Khanna and A. Mukherjee. 1988. Tense in Indian English: A sociolinguistic study. New Delhi: ICSSR and Bahri Publications. Au, S.Y. 1988. A critical appraisal of Gardner's social-psychological theory of second language learning. Language Learning 38(1). 75-100. Avrorin, V. 1977. On the subject matter of social linguistics. Theoretical aspects of linguistics, ed. by J. Gerasemova. Moscow: USSR Academy of Science. Barber, B. 1957. Social stratification. New York: Harcourt, Brace and Co. Bernstein, B. 1971-75. Class, codes and control, Vols 1-3. London: Routledge and Kegan Paul. Berlin, B. and P. Kay. 1969. Basic colour terms: Their universality and evolution. Berkeley: University of California Press. Bloomfield, L. 1933. Language. London: George Allen and Unwin. Davies, P. 1987. The cosmic blueprint. London: Unwin Hyman.

Sociolinguistic Theory and Practice: The Indian Counterpoint · 23 Di Sciullo, A.M., P. Muysken and R. Singh. 1986. Government and code-mixing. Journal of Linguistics, 22. 1-24. Fairclough, N. (ed). 1992. Critical language awareness. London: Longman. Fishman, J. 1972. The sociology of language: An interdisciplinary social science approach to language in society. Rowley, MA: Newbury House. (ed.). 1978. Advances in the study of societal multilingualism. The Hague: Mouton. Freire, P. 1985. The politics of education: Culture, power and liberation. New York: Bergin and Garvey (Tr. by Donaldo Macedo). Gardner, R.C. 1979. Social psychological aspects of second language acquisition. Language and social psychology, ed. by H. GUes and R. St. Clair. Oxford: Basil Blackwell. 193-220. . 1985. Social psychology and second language learning: The role of attitudes and motivation. London: Edward Arnold. . 1988. The socio-educational model of second language learning: Assumptions, findings and issues. Language Learning 38(1). 101-26. Gardner, R.C. and W. Lambert. 1972. Attitudes and motivation in second language learning. Rowley, MA: Newbury House. Gumperz, JJ. 1964. Hindi-Punjabi code-switching in Delhi. Proceedings of the 9* International Congress of Linguists, ed. by H.G. Lunt. The Hague: Mouton. 1115-24. . 1982a. Discourse strategies. Cambridge: Cambridge University Press. . 1982b. Language and social identity. Cambridge: Cambridge University Press. . 1996. The Social Matrix: Culture, Praxis, and Discourse. Rethinking Linguistic Relativity, ed. by JJ. Gumperz and S.C. Levinson. Cambridge: Cambridge University Press. Gumperz, JJ. and R. Wilson. 1971. Convergence and creolization: A case study from the Indo-Aryan and Dravidian border in India. Pidginization and creolization of languages, ed. by D. Hymes. London: Cambridge University Press. 151-68. Gumperz, JJ. and S.C. Levinson (eds). 1996. Rethinking linguistic relativity. Cambridge: Cambridge University Press. Gupta, R.S. (ed.). 2000. Directions in Indian sociolinguistics. Shimla: Indian Institute of Advanced Studies. Gupta, R.S. and K.S. Aggarwal (eds). 1998. Studies in Indian sociolinguistics. New Delhi: Creative. Heugh, K., A. Siegruhn and P. Pluddemann (eds). 1995. Multilingual education for South Africa. Johannesburg: Heinemann. Hymes, D. (ed.). 1971. Pidginization and creolaization of languages. London: Cambridge University Press. Illich, I. 1981. Taught mother tongue and vernacular tongue. Multilingualism and mother tongue education, ed. by D.P. Pattanayak, 1-39. Delhi: Oxford University Press. Kachru, B.B. 1975. Lexical innovation in South Asian English. International Journal of the Sociology of Language 4. 55-74. Kak, A.A. 1995. Acceptability of mixed code Kashmiri-English sentences: A sociolinguistic survey. University of Delhi, Delhi: M.Phil dissertation. Kak, A.A. and R.K. Agnihotri. 1996. Acceptability of mixed code sentences. CIEFL Bulletin 8(2). 5-28. Khanna, A.L. 1983. A Study of some learner variables in learning English as a second language. University of Delhi, Delhi: Ph.D thesis. Khanna, A.L. and R.K. Agnihotri. 1982. Language achievement and some social psychological variables. CIEFL Bulletin 18. 41-51. Khanna, A.L., M.K. Verma, R.K. Agnihotri and S.K. Sinha. 1998. Adult ESOL learners in Britain: A cross-cultural study. Clevedon, UK: Multilingual Matters.

24 · R.K. Agnihotri Khubchandani, L.M. (ed.). 1988. Language in a plural society. Delhi: Motilal Banarsidass and HAS. Kress, G.R. and R.V. Hodge. 1979. Language as ideology. London: Routledge and Kegan Paul. Labov, W. 1966. The social stratification of English in New York City. Washington, DC: Center for Applied Linguistics. . 1971. The notion of 'system' in Creole languages. Pidginizjation and creolization of languages, ed. by D. Hymes, 447-72. London: Cambridge University Press. . 2001. Principles of Linguistic Change: Social Factors, Vol. 2. Maiden, MA: Blackwell. Lee, D. 1992. Competing discourses: Perspective and ideology in language. London: Longman. Lele, J. and R. Singh. 1989. Language and society: Steps towards an integrated theory. Leiden: EJ Brill. Le Page, R.B. 1968. Problems of description in multilingual communities. Transactions of the Philological Society. 189-212. . 1984. Review of 'The social context of creolization1, ed. by Ellen Woolford and William Washabaugh, 1983, Karoma, Ann Arbor. Journal of the Australian Linguistic Society 4( 1). 109-16. Le Page, R.B. and A. Tabouret-Keller. 1985. Acts of identity: Creole-based approaches to language and ethnicity. Cambridge: Cambridge University Press. Mathur, C. 1991. The role of attitudes and motivation in foreign language learning: The case of German in India. University of Delhi, Delhi: M.Phil dissertation. Mukherjee, A. 1980/1996. Language maintenance and language shift: Punjabis and Bengalis in Delhi. New Delhi: Bahri Publications. Pandit, P.B. 1969. Parameters of speech variation in an Indian community. Language and Society in India, ed. by A. Poddar, 207-29. Shimla: HAS. . 1972. India as a sociolinguistic area. Pune: University of Poona Press. . 1988. Towards a grammar of variation. Language in a plural society, ed. by L.M. Khubchandani, 40-49. Delhi: Motilal Banarasidass and HAS. Pfaff, C.W. 1979. Constraints on language mixing: Intra-sentential code-switching and borrowing in Spanish/English. Language 55. 291-318. Poplack, S. 1981. Syntactic structure and social function of code-switching. Latin discourse and communicative behaviour, ed. by R. Duran. New Jersey: Ablex. 169-84. Ramakrishna, S. (ed.). 1997. Translation and multilingualism: Post-colonial contexts. Delhi: Pencraft International. Sahgal, A. 1983. A sociolinguistic study of the spoken English of the Delhi £lite. University of Delhi, Delhi: M.Phil dissertation. . 1992. Bilingualism and scholastic achievement. University of Delhi, Delhi: Ph.D. dissertation. Sankoff, D. and S. Poplack. 1981. A formal grammar for code-switching. Papers in Linguistics 14. 13-^6. Sankoff, D., S. Poplack and S. Vanniarajan. 1991. The empirical study of code-switching. Papers for the Symposium on Code-Switching in Bilingual Studies: Theory, significance and perspectives, Vol. 1, 181-206. Barcelona: ESF. Sapir, E. 1929. The status of linguistics as a science. Language 5. 207-14. Satyanath, T.S. 1982. Kannadigas in Delhi: A sociolinguistic study. University of Delhi, Delhi: M.Phil dissertation. Singh, R. 1995. Linguistic theory, language contact, and modern hindustani: The three sides of a linguistic story. New York: Peter Lang. . 1996. Lectures against sociolinguistics: New York: Peter Lang (also published by Munshiram Manoharlal, Delhi, 1998).

Sociolinguistic Theory and Practice: The Indian Counterpoint · 25 Singh, R. (ed.). 1998. The native speaker: Multilingual perspectives. New Delhi: Sage. . 2001. Chasing butterflies in a multilingual garden: Grounding even linguistic form in multilinguality. The R.N. Srivastava Memorial Lecture, 12 January 2001. University of Delhi: Delhi. Singh, R., P. Dasgupta and J.K. Lele (eds). 1997. Explorations in Indian sociolinguistics. New Delhi: Sage. Trudgill, P. 1974. The social differentiation of English in Norwich. London: Cambridge University Press. Virmani, K. 1991. Communicative writing skills in English: A socio-psychological perspective. University of Delhi, Delhi: M.Phil dissertation. Weinreich, U. 1953. Languages in contact. The Hague: Mouton. Weinreich, U., W. Labov and M. Herzog. 1968. Empirical foundations for a theory of language change. Directions for Historical Linguistics, ed. by W. Lehamann and Y. Malkiel, 97-195. Austin: University of Texas Press. Williams, G. 1992. Sociolinguistics: A sociological critique. London: Routledge. Whorf, B.L. 1940/1956. Language, thought and reality: Selected writings of Benjamin Lee Whorf, edited by J.B. Carroll. Cambridge, MA: MIT Press.

Agreement Features and Projections of TENSE and ASPECT ALICE DAVISON ·* This paper compares tense, aspect and agreement in two Indie languages, standard Hindi/ Urdu and Kurmali, an 'Eastern Hindi* language spoken in Orissa. The languages have similar syntactic structures, similar case uses, including ergative marking of transitive subjects, and the same tenses and aspect. The most striking difference is in the nature and placement of agreement features on verbs. Hindi/Urdu distinguishes non-finite number-gender, associated with aspect, from person-number gender, associated with tense. This distribution is universal according to Giorgi and Pianesi (1997). Kurmali has only person-number (gender) agreement and two distinct agreement relations, one with the local subject and the other pragmatically determined. The paper explores the consequences of the difference of agreement systems. Kurmali has distinct finite forms for past and present tense, with person-number agreement. Hindi/Urdu has bare perfective and imperfective aspect forms with non-finite agreement, which I propose gets a tense interpretation by unselective binding from the context. Nonfinite agreement on a lower ASPECT projection allows object agreement in Hindi/Urdu, which is impossible in Kurmali, where agreement is only on the higher TENSE. Aspect agreement is also associated with split ergative case, unlike Kurmali, which has ergative subject marking in all finite clauses. These associations appear to define the differences among many Indie languages.

• 1. Introduction Overt morphological agreement is not well understood beyond straightforward description. It is not clear what necessary function agreement *I am grateful to Roumyana Slabakova for calling my attention to Giorgi and Pianesi 1997, to the Reading Group for helpful discussion, and especially to Paula Kempchinsky for detailed helpful comments on an earlier version of this paper.

28 · Alice Davison

morphology has, if some languages do without it, while others have very elaborate systems of agreement. Normally agreement morphology has no semantic content, and it normally occurs as a component of tense and aspect inflection. Yet agreement is closely bound up with case licensing, subject grammatical function, and subject properties such as the ability to control null subjects in other clauses (Ura 2000). My goal in this paper is first, to show how the two languages differ in the position and feature composition of agreement, then to define the consequences for clause structure. The crucial difference lies in whether a finite clause must project TENSE. In Hindi/Urdu, finite clauses may consist of a projection of ASPECT alone (licensing ergative subject case), suffixed with only the non-finite morphology which is typical of aspect. In Kurmali, TENSE and finite agreement morphology is obligatory in all finite clauses, but ASPECT has no agreement morphology. If agreement is linked in Indie with subject case licensing, then its presence or absence has consequences for the possible positions of subjects. Furthermore, if TENSE inflection may be omitted, then its Reichenbachian semantic contribution is also omitted, entailing differences in the interpretation of ASPECT alone or in combination with TENSE. Conversely, if finite agreement is projected, then the sentence receives a tense interpretation. The differences observed in these two closely matched languages are found in broad outline in the whole Indie family (Masica 1991).

• 2. Two Agreement Patterns Standard Hindi/Urdu has an agreement pattern which is very common in Indie languages, with multiple agreement forms all agreeing with the same antecedent, which must be nominative, as in (la, b, c). (1) Single agreement antecedent (Hindi/Urdu) a. (vee/pro) abhii aa rah -ee hoo-Ngee 3pl-nom now come prog.-pf be-fut 3mpl mpl 3mpl

[Subject agreement]

'He will be coming just now; he's right on his way' (McGregor 1995:29). b. ghiisuu-nee doo seer puuriyaaN maNgaa-yiiN [Object agreement] Ghisu-erg 2 seer puri-pl order-pf-fpl 3ms 3fpl Ohisu ordered two seers of puns' (Barz and Yadav 1992: 96).

Agreement Features and Projections of TENSE and ASPECT · 29 c. Long-distance object agreement maaNi-nee baacooNj-koo [PROj apnii^ mother-Erg children-Dat self's 3fs 3mpl kitaab-eeN paRh-nee] diiN book-Pl-Nom read-Inf-Obl give-Pf 3fpl (3)fp 'Mother allowed the childrenj [PROj to read self'Si/j books]' (Davison 1999a, 2001). Another pattern is found in eastern Hindi languages. Kurmali, the subject of a very clear and detailed study by P. Mahto (1989), is like standard Hindi/ Urdu in having more than one agreement morpheme, but each agreement morpheme must have a different antecedent. The first one agrees with the local subject, while the second agrees with some other constituent, not necessarily an argument or even in the same local clause. The agreement antecedent need not be nominative, and in some cases must not be nominative. (2) Multiple agreement antecedents in Kurmali a. okar* gilaas-ti' he-Gen glass-def.fem

bhääg -1 -e* - ij break-past-3s -3s

'His glass broke' (subject agreement in AGR1, possessor in AGR2) (Mahto 1989: 26). b. torm betaa-tää-y* (a) your son-def-Erg okhar* betii -ti -kej (b) their daughter-def-Dat (c) 'Your son beat their daughter' (d) (Mahto 1989:68-69).

maar-1 -e'-ij [Subject-dir object] beat-past 3s 3s maar-i-de-1 -e1 -ink [Subj-Obj poss] beat Prt-Past 3s 3pl *maar-l -ej -ik [Subj-Obj poss] beat- Past 3s 3pl de particle req. *maar-l -ol -um [Subj-subj poss] beat-past 3s 2s

c. Agreement across finite clause boundary raam-e* jaan -oj -un k -laa [je gopaal-e* Ram-Erg know-3s-2pl-Pres that Gopal-Erg tohre-ke* maar -le as -h -o* -un k ] you-pl-Dat beat-Pit Pres-be-3s-2pl

30 · Alice Davison

'Ram knows [that Gopal has beaten you-pi]' (Mahto 1989: 99). Matrix: Agrl=local subject, AGR2 = embedded clause dativemarked object Emb.Cl Agrl=local subject, AGR2 = embedded clause dativemarked object An overview of the major Indie languages classified according to a number of agreement patterns is given in Klaiman (1987). Masica (1991: 259-321) provides an exhaustive overview of tense, aspect and agreement in Indie languages, separating the languages which combine agreement features with participle (aspect) forms as well as tense forms (1) from the language which combine agreement features only with finite tense suffixes (2). Languages of type 1 include Hindi/Urdu, Punjabi, Kashmiri, Sindhi, Gujarati, Marathi, and Nepali while those of type 2 include Oriya, Bengali, Assamese, Maithili, (Yadav 1998), Magahi (Verma 1991) and Bhojpuri, as well as Kurmali (Mahto 1989). An example of type 1 is given from Gujarati: (3) a. G.te-thii huM anumaan karuM chuM ke... this-from I-nom inference do-Is be-pres-ls [V+PN AGR, be-PRESENT-PN] (Subject agreement) 'From this I make the inference that...' (Masica 2000: 136). b. G. em-Ne maarii paase eviimaagaNii kar-ii ke ... 3s-Erg my side such request do-pf-fs that [Perfective+N/GAGR] (Object agreement) 'He asked me [that...]' (Masica 2000: 143). The second pattern is found in various versions in the 'eastern Hindi' languages Maithili, Magahi and Bhojpuri. Magahi has many possible combinations of agreement (Verma 1991), two of which are shown in (4). (4) a. ham to(h)raa dekh-1- i - ï I you (hon) see-past-ls-2pl º saw you (honorific)' (Verma 1991: 132).

Agreement Features and Projections of TENSE and ASPECT · 31

b. ham unkaa dekh-1 - i- o I he (hon.) see-past-ls-2pl I saw him (hon.) + hon. reference to addressee (Verma 1991: 138). Like Kurmali, Magahi has separate agreement for subject and direct object, as well as the possibility of references to the addressee. • 2.1. Agreement and INFL: Tense/Aspect Agreement in person, number and gender is associated with the inflectional form of tense and aspect suffixes. An account of agreement in the Indie languages which show agreement is necessarily based on an account of tense and aspect in these languages. In this paper, I investigate two distinct patterns. In Pattern I, illustrated by Hindi/Urdu, agreement is very closely linked to both tense and aspect. For most forms, the boundary between inflection and agreement is not clear. Agreement is syncretic with both tense (5a, b) and aspect (5c, d). (5) a. hai 4be-present-3s' b. hoo 4be-present-2s'

c. paRh-aa 'read-perfective-ins' d. paRh-ii 'read-perfective-fs'

In Pattern II, on the other hand, agreement is analytic, very clearly delimited from tense (6a, b, c). (6) a. bhääg-1 - ii 4break-past-3sf' b. bhääg-1-e-i 4break-past-3s-3s c. bhääg-1 - e - in 4break-past-3s-3pl (Mahto 1989: 26-27). Agreement in Pattern II does not occur with aspect suffixes, only on tense forms. The antecedents of each agreement suffix are distinct. Agreement in Pattern I languages shows a very clear separation into verbal or finite agreement, and adjectival agreement on aspectual forms. The formal difference is expressed in the category of agreement feature. Giorgi and Pianesi (1997) relate morphological categories of person, number and gender to semantic/syntactic categories of TENSE and ASPECT. TENSE is a verbal category, associated with finite agreement (person/ number). It has to be instantiated on a V, which may be an auxiliary (ibid.: 1997: 29). ASPECT is an adjectival category, with number gender features. TENSE and ASPECT are each associated with a corresponding agreement cluster:

32 · Alice Davison (7) a. TENSE -AGRI (Person/number) +V -N (Verbal) [Finite tense] b. ASPECT -AGR2 (Number/gender) +V + N (Adjectival) (Giorgi and Pianesi 1997: 40). The distinction in (7) corresponds very closely to verbal morphology in Hindi/ Urdu and many other languages. Aspect typically has adjectival agreement of number and gender, while tense is associated with finite person number agreement. The tense distinctions found in these Indie languages are: (8) a. Tense: [Finite]: past, present and future [Non-finite: infinitive]. b. Aspect: (Finite and non-finite clauses) perfective, imperfective, progressive. c. Auxiliary verbs: copula 'be', progressive HU rahaa; other lexical auxiliaries. Tense and aspect combine, yielding the present, past and future versions of perfective, imperfective and progressive aspect in finite clauses. The perfective without a tense auxiliary is used for simple past, while the imperfective alone is used for 'stative' present. Non-finite clause may project tense alone (infinitive), aspect alone (perfective participle 'having V-ed'), or some combination of aspect and non-finite tense (infinitive of or perfective participle of 'be')· Kurmali has distinct present and past tenses, as well as aspectual auxiliaries and suffixes. What is absent in this language is inflected adjectival participles. These facts will be illustrated in sections below.

• 2.2. Clause Structure and the Position of AGR Both languages have the common structure shown in (9), with both TENSE and ASPECT.1 (9)

Agreement Features and Projections of TENSE and ASPECT · 33

TENSE requires some V projection, with may be a copula auxiliary or an aspectual verb. Adding AGR1 and AGR2, the structure I propose for Hindi/ Urdu is (10): (10) Hindi/Urdu TP

AuxVP AspP

Tense [Present/past/future] -»-AGRI: Person number

V-aux (Copula form of hoo-naa 'be' ]

\ Asp'

VP

Asp [Perfective/imperfective suffixes; progressive aux verb] +AGR2: Number gender

(11) doolaRkee meerii kitaabeeN paRh two boys(mpl) my books (fpl) read rah-ee haiN prog-mpl be-pres-3p-pl (Number-gender) (person-number) AGR2 AGR1 Two boys are reading my books.' The two agreement suffixes (bold) refer to a single nominative antecedent (bold). The exact mechanism for checking agreement features for subjects is likely to involve movement from a subject position within VP through the ASPECT and TENSE positions (cf. Chomsky 1995; Speas 1990), but I will not address that issue here, nor the more complex issue of how object agreement is checked. Ura (2000) proposes that the object agreement is checked across phrasal boundaries without movement upward. The same basic syntactic structure underlies sentences in Kurmali, but the distribution of agreement suffixes is different.

34 · Alice Davison (12) a.

TENSE-AGR1 (-AGR2)

b. tor"1 betaa-t -y' (a) your son-def-Erg okhar* betii -ti -ke* (b) their daughter-def-Dat

maar-1 -ej -ij [Subject-dir object] beat-past 3s 3s maar-i-de-1 -e1 -ink [Subj-Obj poss] beat Prt-Past 3s3pl

¾ÏÉÉà son beat their daughter' (Mahto 1989:68-69). Agreement with the local subject is obligatory on AGR1, immediately following TENSE. AGR2 immediately follows AGR1, and is optional. AGR2 may refer to the dative-marked direct object, or to the possessor of the direct object. AGR1 and the local subject are in a specify-head relationship, explaining the obligatory nature of coindexing. The Outer' AGR2 refers to both arguments and non-arguments. It may be fused to TENSE in such a way that it has no relation to the subject, but only to some outer specifier at Logical Form/ after Spellout. The conditions on agreement are sharply different. Hindi/Urdu has just one antecedent for all agreement markers, while Kurmali agreement allows two distinct antecedents, and no argument may control both agreement markers. Hindi/Urdu permits only nominative antecedents, which Kurmali varies: the subject marker AGR1 may have a nominative (or ergative) antecedent, but AGR2 may not have a nominative antecedent. Hindi/Urdu agreement refers only to arguments, while Kurmali AGR2 may refer to a possessor. In Hindi/Urdu, an agreement relation may not hold over a finite clause boundary, while in Kurmali, AGR2 on the matrix verb may refer to the subject or other constituent of an embedded finite clause. In addition, there is a morphology distinction of TENSE between the languages corresponding to the differences of the position of AGR1 and AGR2 in the two languages. AGR2 in Hindi/Urdu is fused with ASPECT, and in Hindi/Urdu the ASPECT projection may occur alone in finite clauses. The

Agreement Featuresand Projections of TENSE and ASPECT · 35

perfective alone has the meaning of the simple neutral past, without reference to a specific time of speaking or other time in the past. The imperfective alone, without a tense auxiliary, has the force of a simple present tense. The inclusion of a tense auxiliary introduces additional temporal reference, as I will note next. In Kurmali, it is possible to have aspectual auxiliaries or suffixes combined with tense, but no aspectual participles occur alone. Instead there are specific tense forms: (13) a. HU Simple aspect form Aspect + tense aux. (a) ga-yaa 'go-perf.; went1 (b) ga-yaa hai 'go-perf is; has gone' (c) jaan-taa 'know-impf.; knows' (d) jaan-taa hai 'know-impf is; knows' b. K. Tense morphemes (a) -1'past' (b) -naaMaa 'present' (c) -b,/ -T 'future' (Mahto 1989: 33-35). The agreement suffixes in the two languages are distributed differently on tense and aspect, and in the next section I will base an explanation of the tense/aspect differences on the evidence provided by the position of AGR.

• 3. Tense and Aspect in Indie Languages This next section discusses in more semantic detail the tense and aspect morphology which is associated with AGR in Hindi/Urdu and Kurmali. First, I summarize a current Reichenbachian view of the semantics of tense and aspect, and following the proposal in Giorgi and Pianesi (1997), I relate the semantic information to specific syntactic projections in the clause structure. Second, I show that the tense/aspect categories in Hindi/Urdu and Kurmali are very similar, with the difference that Kurmali has simple past and present tense suffixes, while Hindi/Urdu uses perfective and imperfective aspect suffixes for these tense meanings. I argue that this difference argues for the fusion of AGR 1 (person-number and AGR2 [number-gender] in TENSE in Kurmali. Agreement is therefore associated only with finite tense, which has independent reference to time. AGR2 is not associated with ASPECT. In Hindi/ Urdu, AGR2 (number-gender) is projected independently of tense, in association with ASPECT, which does not have independent reference to time, and is therefore non-finite. Finally, I address the problem of how perfective and imperfective aspect alone can have finite tense interpretations.

36 · Alice Davison

• 3.1. Tense and Aspect—An Overview Tenses are basically links between the proposition expressed in the sentence and the discourse context when the sentence is uttered. Tenses, present, past and future, link the sentence to speech time, S. The verb introduces reference to an event or state. Sentence aspect relates the boundaries of the event or state to some other time frame; events are completed or open, relative to that time frame (Smith 1997; Swart 1998). Linguistic accounts of tense and aspect make use not only of the time of the event and the time of utterance but also a reference time, an innovation due to Reichenbach 1947, now a standard part of linguistic accounts of tense and aspect (14): (14) a. E = Event time (of the event referred to by V) b. S = Speech time c. R = Reference time Aspect is viewed as a temporal relation by Giorgi and Pianesi (1997: 26-30, 37-45) and Swart (1998). It is a relation between the event time in the verb's argument structure E and a reference time R. The reference time also has a relation to speech time S. These relations are: (15) a. precedence b. overlap

(S R, R (S,R), (R,S).

E)

Tense and aspect combinations are composed of two temporal relations among the three times, representing TENSE 1 and TENSE 2. For example, the past tense (a), the present perfect (b), the future (c) and future perfect (d) (Giorgi and Pianesi 1997: 29, 43). (16) a. b. c. d. e. f. g.

past present perfect future future perfect simple present proximate future distant future

ASPECT TENSE (E,R) combined with (R S)=E,R_S (E_R) (S,R) =E_S,R (R,E) (S R)=S_R,E (E_R) (S_R) (R,E) (S,R) =S,R,E (R_E) (S,R) =S,R_E (R R) (S_R) = S_R E

The difference between the past and future lies in the ordering of the speech time S and the reference time R. The difference between the simple tenses and the perfect lies in the fact that the reference time does not overlap the event time, but precedes or follows it. The simple present can be broadly

Agreement Features and Projections of TENSE and ASPECT · 37

described as the overlap of all three times, but a fuller account is necessarily more complex, as languages differ in possible interpretations of the present (Giorgi and Pianesi 1997: Chapter 4). The clause structure in (17) shows the position of these semantic categories in the syntactic projections within a clause:

(17)

TP1 Ã

TENSE (Reference time, Speech time) ASPECT VP OBJECT

V (Aux)

ASPECT (Reference time, Event time) V (Event argument: P [x,y,e])

I will assume this structure underlies the clauses in both Hindi/Urdu and Kurmali. It is a pattern which is similar to the ones found in European languages (cf. Giorgi and Pianesi 1997: Chapters 2 and 3); the Indie languages differ only in word order.

• 4. Tense and Aspect in Hindi/Urdu and Kurmali In this section, I compare Hindi/Urdu with Kurmali. I begin with compounds of tense and aspect. An auxiliary verb may be projected between tense and aspect. The present perfect forms correspond exactly, except that in Kurmali, the perfect suffix -al is invariant. (18) a. HU.vee abhii aa-ee haiN [Present perfect; E_R,S] 3pl now come-PF-mpl be-pres-3pl They/he(hon) have come just now.' b. K. okhre1 [PRO to-ke1 dekhe] aaw -al aa -h -th' -uj [Pres. perf] they-nom you-acc see-inf come-pf pres-be-3pl-2s 3pl 2s They have come [PRO to see you] [pf for past]' Mahlo 1989.

38 · Alice Davison The future form is a simple tense form unspecified for aspect: (19) a. HU vee abhii aa-eeN-gee [Future, S,R_E] 3pl now come-fut-3pl-mpl 'They/he (hon) will come just now.' b. K. oy' tokej kitaab-taa-0 aan-i de- -t- -oj -uj he-nom you-dat book-def-acc bring fut. 3s 2s 3s 2s 3s-neut 4

He wiU bring the book to/for you' (Mahto 1989: 58).

The verb combines directly with the future marker (19b), with no sentence aspect expressed, though in Hindi/Urdu, it is possible to combine aspect with the future, as in (1) given earlier. The progressive aspect is projected as an auxiliary verb.2 Since tense does not combine directly with aspect, tense requires an auxiliary verb as well. (20) a. HU.vee abhii aa rahee hooNgee [Future progressive] 3pl now comeprog-mpl be-fut-3pl-mpl 'He (hon)/they should be coming just now.' b. HU siitaa kitaab paRh rahii thii [Past progressive] Sita book read prog-fs be-past-fs 'Sita was reading a book.* (21) a. K. sitää-y' kitaab-taa-0 paDh-e laagal ra-h—ii1 [Past progressive] Sita-Erg book-Def read-Pit prog Past-be-3sf 3sf 'Sita was reading a book' (Mahto 1989: 50). b. K. sitää-y1 kitaab-taa-0 paDh -te ra-h- -ii* [Past perfect progressive] Sita-Erg book-Def read-perf Past-be-3sf 3sf prog 'Sita had been reading a book' (Mahto 1989: 51). Up to this point, the two languages are similar. They differ in two ways. The 'simple past' in Hindi/Urdu consists of a verb with perfective aspect without any tense and auxiliary (22). If there were a tense + auxiliary, the interpretation would be as in (18a), the present perfect.

Agreement Features and Projections of TENSE and ASPECT · 39 (22)

a. HU vee samay-par aa-ee [Neutral past; no reference time] 3pl time-on come-pf-mpl 'They came on time.' b. K. torJ betaa-taa-y sinemaa-0 dekh -1 -os ~éÑ [Past, E,R _S] your son-def-erg film-ace see-past-3s-2s 2s 3s ¾ÏÉÉà son saw the film' (Mahto 1989: 49).

Kurmali has a past tense suffix -/- which is distinct from the perfective -al (18b). The other difference is illustrated in (23). In Hindi/Urdu, generic and Stative sentences have a combination of imperfective aspect with optional tense: (23) a. HU(aam-toor-par) vee samay-par generally 3pl time-on aa-tee (haiN) come-impf -mpl be-pres-3pl

[Imperfective/ Present imperfective]

'They usually come on time (habitual, generic).' b. HU aap [is- baat-koo aasaan] you this matter-dat easy 3pl samajh-tee haiN consider-impf be-pres-pl

[Present imperfective]

'You consider [this matter easy]' (stative). c. HU vee aa nahiiN sak-tee / sak-ee [Auxiliary verbs] 3pl come not able-impf-mpl / able-pf-mpl They are/were not able to come' (stative). (24) K. tohre4 [kes-taa-ke1 aasaan] you-pl-nom case-def-dat easy 2pl 3s-neut bujh -ihaa' -kj -laa consider -2pl -3s -pres

[present; E, R, S]

'You consider [this case easy]' (Mahto 1989: 90).

40 · Alice Davison

In Kurmali, there is a present tense suffix -laa, which attaches directly to a main verb+AGR. Curiously, the present attaches to the right of the AGR, V-AGR-Pres, unlike the past and future which attach to the V, yielding the order V-Tense-AGR. As we see from the comparison of the sentences (23b) and (24), the use of the imperfective with stative meaning in Hindi/Urdu corresponds to the use of present tense in Kurmali. In some cases a comparison of identical or very similar sentence meanings is possible using the common story text used in the Linguistic Survey of India (the Kurmali text in Grierson [1903/1968] can be compared with the corresponding standard Hindi texts [Grierson 1916/1968]). In Hindi/Urdu, there is no simple present tense, comparable to the form in Kurmali in (24). The imperfective form of the main verb, with or without a tense copula, is equivalent to an unbounded present, appropriate with stative verbs like samajh-naa 'consider' above (23b). The imperfective with event verbs can also have a proximate future meaning, as in (25). (25) a. HU maiN uth-taa huuN aur [PRO baap-kee paaas jaa-kar] I rise-impf am and father -to go-having kah-uuNgaa ki... say-fut-lsm I will get up and having gone to my father, I will say that... (Grierson 1916/1968: 101).

[Imperfective, proximate future interpretation]

b. K. maay [baapek Thaai jai-ke] kah-am ... I-(erg) father's near go-PRT say-fut.ls

[Future]

'Having gone to my father's house, I will say... (Grierson 1903/1968:152). Hindi/Urdu uses the imperfective present to express the proximate future in relation to the speech time, as opposed to the more distant future expressed by the future morphology in (23a). The Kurmali version uses the future morphology for the distal future reference, and expresses the preceding event in relation to the future event, not the speech time (also possible in HU). The imperfective INHU with or without the present tense does not allow the 'continuous present' reading found in Italian and other languages (26a); for that meaning, the progressive is required in Hindi/Urdu (27c). (26) a. (Italian) mangi-o unamela [V+PN-AGR=Simple present] eat Is an apple (Giorgi and Pianesi 1997: 153).

Agreement Features and Projections of TENSE and ASPECT · 41

º eat an apple: I am eating one at this moment; I generally eat an apple (everyday)' (Continuous interpretation; generic/habitual interpretation) b. HU. maiN seeb khaa-taa huuN [Present imperfective] I apple eat-impf -ms be-pres-1 s º eat apple(s); * I am at this moment eating an apple' (Generic/habitual interpretation only) c. HU maiN seeb khaa rahaa huuN [Present progressive] I apple eat prog-ms be-pres-Iss º am (at this moment) eating an apple.1 (27) a. K. maay eetnaa din tar muniSeek-leekhee khaaT-ahaN [Present, I so-many days your servant -like labor-pres-ls E,R,S] º have been laboring for so many days as your servant' (Lit. º work for so many days', with extended present meaning) (Grierson 1903/ 1968: 153-54). b. K. kiinaa-laay etak naac baajna hekak re? what-for this much dance play be-pres-3s Part

[Present, E, R, S]

For what is so much singing and dancing (going on) (continuous present meaning) (Grierson 1903/1968: 153). It appears from these examples that the present in Kurmali does have a continuous interpretation for the present tense on non-stative main verbs (activities). The preceding Sections 2-4 have focused on the surface differences of agreement and TENSE/ASPECT in Hindi/Urdu and Kurmali, as well as some underlying similarities. In Section 2, it has been shown that Kurmali has multiple agreement markers associated with tense, with different antecedents for the two agreement morphemes. Hindi/Urdu allows only one nominative antecedent for all the agreement morphology in the clause. In Section 3,1 proposed a structure for the clause which is the same for both languages, with differences in where the agreement morphology is placed. In Hindi/Urdu, number-gender agreement is fused with ASPECT, and person-number agreement is fused with TENSE. In Kurmali, all kinds of agreement are affixed to TENSE. In Section 4,1 related agreement to a Reichenbachian (partial) account of TENSE and ASPECT. Following Giorgi and Pianesi 1997, number-gender agreement

42 · Alice Davison

is associated with ASPECT, which semantically relates an event time and a reference time, while person-number agreement is associated with TENSE, which relates reference time with speech time. Tense+aspect combinations in Hindi/Urdu and Kurmali are very similar, with two striking differences. Hindi/ Urdu represent neutral past and present meanings with aspectual participles alone, while Kurmali has distinct past and present tense morphemes.

• 5. Positions of AGR and Tense/Aspect-Consequences for Interpretation In this section, I will propose an explanation for the differences between Hindi/ Urdu and Kurmali shown in examples (22)-(27). Kurmali requires a finite tense in all sentences types, while Hindi/Urdu has independent clauses with only a perfective or imperfective participle, expressing past and present meanings respectively. The explanation rests on the position(s) of AGR, and the feature content of AGR, whether PN(fmite) or NG(adjectival). If ASPECT is associated with AGR, and AGR is typically associated with nominative case. Then the case of DP in the subject position (or specifier of ASPECT) is licensed, as in (9, 17). The interpretation of ASPECT alone determines the temporal interpretation of the sentence. This is a somewhat surprising outcome, given the semantic vacuousness of AGR. But the distinction which I draw between Hindi/Urdu and Kurmali reflects a very broad distinction among the Indie languages. • 5.1. How can Aspect Phrases Count as Finite Clauses in HU? It is clear that clauses in HU with only an aspect marker count as finite clauses. Ergative subject case marking is possible only in finite clauses (28a, b/29a, b) (see discussion in Davison 1999a, 2001). But sentences like (29c, d) both require ergative subject marking, though they lack an overt tense copula with finite agreement (person-number): (28) Non-finite embedded clauses (Hindi/Urdu) a. mujhee [us-kaa/*us-nee aisii filmeeN dekkh-naa] I-dat 3s-gen/*3s-erg such films see-inf pasand nahiiN lag-taa liking not striking-impf 4

1 don't like [him/her to see such films].'

Agreement Features and Projections of TENSE and ASPECT · 43

b. pitaa-koo [kuttaa *kuttee-nee zamiin-meeN father-Dat dog -Nom/*dog-erg ground-in gaDDhee khood-aa (huaa)] jaan paR-aa holes-Norn dig-pfbe-pf seem-pf 4

It appeared to father that the dog had dug holes in the ground.' (Ms agreement with kuttaa3)

(29) Finite clause (Hindi/Urdu) a. laRkee /*laRkooN-nee apnii-apnii kitaabeeN boys-mpl-nom boys-mpl-erg self's-self's books-fpl paRh- eeNgee read-fut-3pl-mpl 'The boys will each read their own books.' b. laRkooN-nee apnii-apnii kitaabeeN boys-mpl-erg self's- self's books-fpl paRhiiN read-pf-fpl

(Object agreement ergative subject)

'The boys (each) read their books.' Finite embedded clause (Hindi/Urdu) c. pitaa-koo jaan paR-aa [ki kuttee-nee/*kuttaa zamiin-meeN father-dat seem-pf that dog-erg dog-nom ground-in gaDDee khood-aa (hai)] holes dig-pf is 4

It seemed to father [that the dog dug/has dug holes in the ground].'

The ergative is possible only in finite perfective clauses, but the tense does not have to be overtly expressed (29a-c, also Ib-c). Kurmali has an ergative marker -e which is suffixed to transitive subjects, but it is found in nonperfective as well as pefective clauses (30): (30) a. Past progressive [progressive aspect = -e + lagaal] (Kurmali) sitää-y' kitaab-taa-0 paDh -e laagal ra-h- -ii4 Sita-Erg book-Def read-Prt prog Past-be-3sf no agreement agreement 4

Sita was reading a book' (Mahto 1989: 50).

44 · Alice Davison

b. Past perfect (Kurmali) sitää-y4 kitaab-taa-0 paDh -le ra -h -i4 Sitaa-erg book-def-acc read pf past be 3sf 'Sita had read the book' (Mahto 1989: 51). In Hindi/Urdu, perfective aspect is necessary for ergative case marking, but Kurmali has ergative subjects in finite clauses, whether perfective or not. In this section, I propose an answer to the question of how the perfect alone (3la) counts as the equivalent of the simple past (31b), though lacking tense inflection (31c). (31) a. Perfect aspect E_R b. Simple past: tense E, R S c. Present perfect E_R, S In (32), the simple past a is contrasted with combinations of the perfect and past, present and future tense—b, c, d: (32) a. b. c. d. e.

past present perfect past perfect future perfect proximate future

Aspect (E,R) combined with (E_R) (E_R) (E_R) (R_E)

Tense Combination (R S)=E,R_S (S, R) =E_S, R (R_S) =E_R_S (S_ R) (ambiguous)4 (S, R) = S, R_E

(Giorgi and Pianesi 1997: 27-28). One way of assimilating (29a) to (29b) is to say that S is introduced pragmatically, from the discourse context, and overlaps with R introduced by Perfect (E R). If that were the case, then the perfect alone (E R) plus S overlapping with R should have the same interpretation as the present perfect (31c, 32b). Another possibility is to say that the perfective alone (3la) is lexically reanalyzed as the past (31b), a solution I reject here but will reserve for other Indie languages. My reason is that ergative case in HU is licensed by the perfective alone (where it counts as past) and also the perfective in combination with tense (where it has aspectual meaning). The lexical reanalysis proposal loses a generalization about split ergativity in HU, that it is licensed by perfective aspect. This objection does not hold for languages which have no ergative case, or have only non-split ergative case, as in Kurmali. The answer may lie in an observation by McGregor (1995: 27), about the differences between the following sentences in temporal location:

Agreement Features and Projections of TENSE and ASPECT · 45

(33) a. maiN kal aa-yaa (perfect alone) 'Neutral' past I yesterday come-pf.ms 4

I came yesterday' (McGregor 1995: 27).

b. maiN kal aa-yaa huuN (Present perfect, 30b) I yesterday come-pf.ms be-pres.ls º came yesterday; I have come yesterday' (McGregor 1995: 27). c. maiN kal aa-yaa thaa (Past perfect, 32c) I yesterday come-pf.ms be-pastms º came yesterday'; I had come yesterday' (McGregor 1995: 27). d. abhii aa-ee hooNgee (Future perfect, 32d) now-imph come-pf be-fut.3plm 'He will have come just now; he must have just come' (McGregor 1995: 29). Perfective aspect indicates completion of an event before a reference time. McGregor observes that wherever there is an expressed tense, the tense established a temporal location for the reference point. The perfective alone (33a) is 'neutral with respect to context and time of action' (McGregor 1995: 27). The sentence in (33c) indicates completion relative to a time in the past, and the present or future tenses in (33b, d) make a similar contribution of an ordering of the reference time (R) and speech time (S). The 'neutral perfective' is very close in meaning to the 'neutral' past in that the event is not obligatorily related to the speech time.5 The solution I propose for HU is that the perfect alone with a past tense interpretation, is somewhat different in meaning from both the past (E, R S) and the present perfect (E R, S), or other tensed perfective sentences. If we take the morphology in (33a) literally, it says about sentences like (28c) and (33a), that their verbal inflection lacks a link between reference time and speech time. The temporal adverb kal 'yesterday' in (33a) does establish a broad link between reference time and speech time, but it is an option and not part of verbal inflection. The conclusion to draw about the perfective alone is that is not exactly identical to simple past tense, which lexically introduces a precedence relation between reference time and speech time (3 Ib). The reference time R is not linked at all to S. We may suppose that it is like an indefnite NP, an unbound variable which receives a contextually determined interpretation by Existential Closure (Diesing 1992; Heim 1982).

46 · Alice Davison

The account proposed here (31 a, 34) provides the most transparent view of the perfective as simple past.6 (34) a. E_R perfective b. E R = unselective discourse binding by 5, no specific reference time c. E,R _S [Past tense] Perfective aspect alone uses a discourse-specified R reference time in place of the S introduced by past tense. The difference between the imperfective alone and with a tense copula is less striking semantically, but there are contexts, such as the counterfactual condition, in which a realise tensed copula is not possible. A similar interpretation may then be provided for the imperfective alone, without specific temporal reference introduced by an auxiliary. It is used for stative predicates which have no specific link to speech time, and it also has a conditional or non-factual interpretation in irrealis contexts, that is, negative sentences and contrary-to-fact conditionals (McGregor 1995:20,137). Sentences of this sort make no specific (extensional) reference to the discourse situation and speech time. (A full account of imperfective and progressive temporal meanings is beyond the scope of this paper; see Giorgi and Pianesi [1997] for a discussion of the complexities in the nature of these forms.) We are led to an account of finite clauses in HU in which ASPECT alone licenses ergative subjects in HU (29a-c ), but TENSE does not have to be overtly projected in every sentence. That means not every clause is a TP, headed by TENSE. Some may consist solely of a projection headed by aspect. Both TP and AP license subjects with structural case, nominative or ergative. In addition both heads require some sort of subject, whether it has nominative/ ergative case (29a-c), or lexical dative case. By contrast, Kurmali always requires a TENSE suffix with agreement suffixes. A sentence without TENSE consisting solely of a projection for ASPECT is not possible. Instead, there are simple past and present tense affixes, combined with person-number (gender) agreement. The difference between Hindi/Urdu and Kurmali is found elsewhere in the Indie languages. • 5.2. Generalizations about Tense, Aspect and Agreement in Indie Languages Within the Indie languages which have overt agreement morphology, Hindi/ Urdu and Kurmali represent extreme points in the range of variation. This variation has been exhaustively analyzed and categorized in Masica (1991: 259-315), along with the distribution of ergative case marking (Masica 1991: 263-64). Masica's comparison makes use of the basic schema (35) and a classification of concord (36):

Agreement Features and Projections of TENSE and ASPECT · 47 (35) V + Asp + (c) + Tense Marker + (c) (c) = concord (Masica 1990: 258). (36)

Concord: I = primary personal concord added to V with tense meanings II = secondary personal concord added to participle forms AC = Adjectival concord [Number-gender] PC = Person/number concord (Masica 1990: 265).

The basic schema (35) has the same structure as the tree structure proposed for HU and Kurmali in (10). Adjectival concord corresponds to the non-finite number-gender agreement associated with ASPECT, noted in previous sections. The classes of person/number concord I, II and PC correspond to finite person-number agreement. I will make the assumption, following Giorgi and Pianesi (1997), that the overt presence of person-number concord indicates the projection of TENSE, even when not overtly indicated, or where the difference between tense and aspect is not clear, as in the use of aspect forms in Hindi/Urdu for neutral present or past tense discussed in the last section. With this information, it is possible to say whether Hindi/Urdu and Kurmali are idiosyncratic for case and inflectional morphology, or if they represent more general patterns. Furthermore, we can test the 'underspecification' account given in the previous section for bare aspect forms with tense interpretation, against a reanalysis account. One of the deciding factors in favor of underspecification was that it allows the generalization that perfective aspect is required for split ergative case. The first question, then, is which languages have only adjectival concord on verbs marked for aspect. Languages which allow perfective aspect + adjectival concord alone with a neutral perfective/past meaning are Hindi/ Urdu (293), Punjabi (295), Kashmiri (297), Gujarati (301), and Marathi (303-4). Those without adjectival concord use secondary person-number concord on directly on the verb + past marker: Oriya (308), Bengali (310), Assamese (311). And to these we should add Maithili (R. Yadav 1996; Y. Yadav 1998) and Kurmali. Languages with adjectival concord allow the imperfective alone with present habitual/counterfactual meaning. These are Hindi/Urdu (292), Punjabi (294) Gujarati (301) Marathi (303). Those without adjectival concord require a tense auxiliary with person/number suffixes with the imperfective, or place personnumber agreement suffixes directly on the V stem. The V+AGR combination has a tense interpretation, as in the Italian example (26a). Assamese, Bengali and Oriya have this combination. Maithili (R. Yadav 1996) and Kurmali have a present tense suffix combined with person-number agreement. Though the uses of perfective and imperfective aspect + non-finite agreement are not exactly identical across all Indie languages, there is a clear correlation between

48 · Alice Davison

the use of adjectival concord and the use of a bare aspectual form. If finite verbal concord is required, then the combination with V has a tense interpretation, which may be explicit, or implied by the presence of finite agreement. The next question asks if there is a correlation between the use of a split ergative case marker and V+perfective+adjectival concord. The languages with ergative subject marking required in transitive perfective (finite) clauses are Kashmiri, Nepali, Punjabi, Lahnda, Hindi/Urdu, Gujarati and Marathi (Masica 1991: 342-46). Those which have no ergative marking at all are Bengali, Oriya, Maithili, Magahi, Sinhala, while Kurmali (30a, b) Shina and Assamese have ergative subject marking in all tenses (ibid). Languages which have Adjectival concord are Kashmiri, Nepali, Punjabi, Lahnda, Hindi/Urdu, Gujarati and Marathi. (Masica 1991: 263-64). Those which do not are Assamese, Kurmali, Maithili, Bhojpuri, Bengali and Oriya. There seems to be a quite clear correlation between split ergativity, adjectival concord and the possible absence of overt tense inflection. In summary: (37) a. Split ergative languages require perfective aspect, which combines with number-gender (adjectival) agreement features (Hindi, examples [Ib, c] Gujarati, examples [3a, b]). b. Languages with adjectival concord allow the perfective and imperfective to occur without overt tense, and a neutral perfective and imperfective (habitual, generic, sometime progressive) meaning. These are the languages with E, R relations, but underspecified reference to S added by the context. (31) (Hindi, examples [22a, 23a, b, 25a, 26b, 33a, note 5(i)]). c. Those without adjectival concord require person-number (finite) agreement and overt tense in all instances (Kurmali [2,18b, 22b, 34, 25b, 27a, b]). d. Summary Ergative case

Hindi

Kurmali

Split (perfective)

Non-split

Number gender concord Aspect

Tense (optional)

Gender-number concord Tense

Tense (obligatory)

Present (generic)

Imperfective alone Tense suffix alone, distinct from progressive

Agreement Features and Projections of TENSE and ASPECT · 49

Past tense

Perfective alone

Tense suffix distinct from perf.

Interestingly, both Sindhi (310) and Nepali (314) have ergative markers with somewhat different conditions than the languages noted here. Nepali requires ergative case in perfective transitive clauses, but allows ergative case as a kind of topic marker (Wallace 1985). The perfective is combined with secondary person markers when not combined with a tense copula. Its status seems to be ambiguous—it is sometimes a aspect marker, sometimes a past tense marker. In Sindhi, the ergative is not distinct from a passive agent marker. When the perfective aspect is used alone without a tense copula, it is combined with secondary person markers. As in Nepali, the perfective has dual interpretation as an aspect marker (in combination with overt tense) and a past tense marker combined with person-number marking. Languages without split ergative marking do not use adjectival concord on the aspectual forms of verbs, and combine tense forms with person-number marking. I will propose that in these languages, P-N concord adds the R, S relation to the E, R relation expressed by the aspect marker, and the combination is lexically reanalyzed as past tense E, R S. In Sindhi and Nepali, the addition of R, S relation is an option for V+PERF+P/N, in the absence of the tense copula. Agreement features on verbal inflection notoriously have no semantic content (Chomsky 1995: Chapter 4). But the presence of agreement features in specific combinations universally seems to follow a strict pattern (cf. Giorgi and Pianesi 1997), so that person-number agreement is associated with (finite) tense (with indices and an ordering for R, S), and number-gender agreement is associated with non-finite agreement and indices and an ordering for E and R indices. Tense, then, need not be overt if characteristic person-number agreement features are present (as is the case in Italian [26a]). Further, if only non-finite agreement is expressed, this fact is an indication that tense is not projected, and the sentence is underspecified. I also have proposed that some verbal forms with only aspect and number-gender agreement are underspecified for the S index, and gain a full interpretation for the R index from contextual information. These sentences in Hindi-Urdu license overt nominative or ergative subjects just as other sentences do with both aspect and tense. In some other languages which do not have the same kind of ergative case licensing as HU, perfective combines with person-number (finite) agreement, and is reanalyzed lexically as past tense. In the next section, I discuss how ASPECT phrases alone license subjects in Hindi/Urdu, and why languages like Kurmali do not allow this possibility, but instead require tense inflection and agreement.

50 · Alice Davison

• 5.3. Subject Licensing and the Role of Functional Categories in Hindi/Urdu and Kurmali For subjects to be licensed in finite clauses, they must be in a VP-external position, which also licenses their structural case (nominative or ergative). In the theoretical terms I am assuming in this paper, this means that a feature of TENSE or ASPECT forces the subject to move from VP to the Specifier position of TENSE (or ASPECT). The NP subject's position, independently of its case, is the effect of the Extended Projection Principle, realized as a feature on TENSE/ASPECT requiring a NP to be in the Specifier in the overt syntax; in other words, the EPP category feature is strong in both Kurmali and Hindi/ Urdu. In addition, a structural case must be licensed on the subjects which do not have a lexical (inherent) case already assigned within VP). The 'unmarked' value universally seems to be that nominative case is checked by the same NP which also checks agreement features, and this cluster of features is a property of TENSE/ASPECT (some variation in this correlation is explored in Ura [2000]). I will assume that in both Kurmali and Hindi/Urdu, these features are found on functional heads having AGR properties, and the features are weak, since they do not force overt movement. By AGR, I mean finite and non-finite agreement in Hindi/Urdu, and the first agreement position in the Kurmali tensed verb. The second agreement position has quite different properties from the other agreement. Both Kurmali and Hindi/Urdu allow non-nominative subjects (38-41): (38) Kurmali dative experiencer a. to -kei okhar* [Theme has AGR1, Exp has AGR2] you-s-Dat their betaa-taa pasand aa-h-o 1 -iP son-def like Pres-be- 3s 2s 'You like their son' (Mahto 1989: 77). b. *to -ke* okhar* beTaa-taa pasand aa -h -s1 -ink you-s-Dat their son-def like Pres-be- 2s 3pls 'You like their son' (Mahto 1989: 77). c. torj ghoRaa' aa -h -o'-uJ you-Gen horse Pres -be -3s -2s

[Possessed NP has Agreement I]

'You have a horse' (Mahto 1989: 73).

Agreement Features and Projections of TENSE and ASPECT · 51

(39) Hindi/Urdu dative experiencer a. tumheeN un-kaa beeTaa pasand aa-taa 2s-dat 3pl-gen son-nom liked come-impf-ms hai/ *haiN/*hoo be-pres-3s/3pl/2pl 4

You like their son.'

b. beeTee-koo. apnee^. pitaa-see Dar nahiiN hai son-dar self's father-from fear-nom not is The son is not afraid of self's father.' In both languages, the dative experiencer does not control agreement, though it may have other subject properties, such as the ability to bind a subject oriented reflexive (39b) (this property is documented for Bengali (Sengupta 1999: 293). Instead, a nominative NP determines the correct PNG features for all the agreement in HU, and the first agreement position in Kurmali. Note that the dative NP controls the second agreement position in (38a). Since the dative of the experiencer is a consequence of the verb—it is associated with specific lexical verbs—its case is checked within VP (see Ura 2000), and the AGR with which it is coindexed in (38a) does not have any case-checking function. In fact, the second AGR position in Kurmali may not check nominative case: (40) Kurmali nominative/dative direct objects a. K. raam'-e aamMaa (Subject agreement only, AGR 1) Ram-erg mango-def. khaa -1 - aak1 eat-past-3s (3p, m/f, without AGR2) 'Ram ate the mango' (Mahto 1989: 64). b. K. *raanV-e aamMaa (Subject AGR1, nom. object AGR2) Ram-erg mango-def. khaa -1 - e j -ij eat-past-3s (3p, m/f, without AGR2) 'Ram ate the mango' (Mahto 1989: 64).

52 · Alice Daoison

c. K.raam'-e aanP-ta-ke (Subject AGR1, nom. object AGR2) Ram-erg mango-def.-dat khaa -l - e' -V eat-past-3s (3p, m/f, without AGR2) 'Ram ate the mango' (Mahto· 1989: 64). Agreement 2 may be omitted (40a), or the direct object may be marked with dative case and coindexed with Agreement 2 (40c). But a nominative direct object may not be coindexed with Agreement 2 (40b). In this respect, Kurmali differs sharply from HU (41). (41) HU nominative and dative direct objects a. baccee-nee doo kitaabeeN paRh-iiN /paRh-ii haiN child-erg two book-fpl-nom read-perf-fpl / read-perf-f be-pres-3pl 'The little boy read/has read two books.' b. baccooN-nee in doo kitaabooN-koo child-pl-erg these two book-fpl-dat paRh-aa / paRh-aa hai read-perf-ms / read-perf-ms be-pres-3s 'The little boys read/have read these two books.' In HU, a nominative direct object (or predicate N) controls agreement (4la). If both subject and direct object have (structural) case postpositions, the default 3ms features are found, and no nominative case needs to be checked (41b) (cf. Ura 2000). The explanation which I propose for bare aspectual sentences in HindiUrdu is that in HU ASPECT has an EPP category feature, and also structural case licensing features. So ASPECT can attract a subject to check the nominal category feature (EPP), project a specifier position, check the structural case of its specifier, and at the same time check non-finite agreement features (number-gender). Hence sentences in Hindi/Urdu which have no overt tense, but do have aspect, are well-formed. Of course, TENSE can also be projected, in which case it attracts the subject and checks case. In Kurmali, ASPECT does not have agreement features which could check structural case. So ASPECT alone cannot attract a subject, project a specifier position and check its structural case. (It remains open to further investigation whether ASPECT may have a lexically case subject on its own, or if ASPECT can license an overt subject in a non-finite context, without AGR1. In other words, does the EPP feature entail AGR1 in this language?)

Agreement Features and Projections of TENSE and ASPECT · 53

To conclude this section, I want to speculate about the nature of Agreement 2 in Kurmali. It does not have a structural case-checking role, and it does not have to be coindexed with local arguments. It may be coindexed with a local, non-argument genitive, or a non local argument of a subordinate clause. It has a pragmatic effect in single clauses: it can mark a NP whose referent is 'affected' by the event: (42) Kurmali a. okarj gilaas-ti1 bhääg -l -ii* he-Gen glass-def.fem break-past-3sf 'His glass broke' (subject agreement in AGR1) (Mahto 1989: 26). b. okarj gilaas-ti1 bhääg -1 -es - ij he-Gen glass-def.fem break-past-3s -3s 'His glass broke' (subject agreement in AGR1, possessor in AGR2) (Mahto 1989: 26). The second version of the sentence conveys that the possessor-referent was affected (42b), while the first version is neutral (42b). Agreement 2 is associated either with TENSE, or possibly with some higher functional projection such as COMP or FOCUS. The fact that Agreement 2 blocks gender agreement on AGR1 suggests that Agreement 2 is closely associated with AGR1 on TENSE. In any case, Agreement 2 is not lower than AGR1 in Kurmali. In Hindi/Urdu, the non-finite agreement set of features seems to be independent of finite agreement, and also occurs in embedded non-finite contexts. Agreement is not found in non-finite contexts in Kurmali, such as the infinitive verb in (43): (43) Kurmali raam-e' to-kej [PRO mohan-ke maar-e] kah -1 -o' -uj Ram-erg you-dat Mohan-datbeat-inf say-past 3s 2s 4

Ram asked you [PRO to beat Mohan]' (Mahto 1989: 85).

(44) Hindi/Urdu un-koo [PRO saaikal calaa -nil] aa -tii hai 3p-dat bicycle ride-inf-f come-impf-f be-pres-3s 3mpl 3fs They know [(how) PRO to ride a bicycle].'

54 · Alice Davison

An option in Hindi/Urdu (44) is for the embedded infinitive to take on the NG features of nominative argument, and the features on the infinitive trigger object agreement in the matrix clause. Non-finite agreement features are possible on both the non-finite tense suffix in (44), as well as on the aspectual forms discussed in previous sections of this paper. In contrast, both agreement suffixes in Kurmali are associated only with finite tense.

• 6. Conclusion This paper has considered two languages within the Indie family which are very closely matched in syntactic structure, case marking and categories for tense and aspect. They are head final, have ergative subjects in transitive clauses, and dative case on direct objects as well as experiences. They have past, present and future tense, as well as perfective, imperfective and progressive aspect. The chief formal difference between these languages is in the nature of agreement suffixes and where they are projected. In Hindi/Urdu, non-finite (number-gender) features are fused with ASPECT, which projects an aspect phrase, able to license a subject with either nominative or ergative case. Finite person-number features are fused with TENSE, which projects a tense phrase, with a nominative or ergative subject. In Kurmali, ASPECT may be projected, but it has no case-licensing features, and so may not form a well-formed clause without a TENSE projection. In Kurmali, TENSE has an obligatory set of person, number (gender) agreement features which license nominative or ergative case. The ergative case in Kurmali is not dependent on specific aspect forms, as it is in Hindi/Urdu. Agreement suffixes on the verb make no direct contribution to the interpretation of the sentence, yet they can serve as overt forms which are surrogates for TENSE. In Italian, Giorgi and Pianesi (1997) have noted that present tense interpretation is associated with a verb stem combined with finite personnumber features. The same V+AGR combination is found as present tense in Assamese, Bengali and Oriya. I conclude that these languages indicate TENSE indirectly. For the same reason, I conclude that TENSE is absent when a sentence has only non-finite agreement on an aspectual participle. Hindi/Urdu uses perfective aspect forms alone for a neutral past, and the imperfective alone for a neutral or counterfactual present. In these sentences, I propose that the TENSE interpretation is underspecified. The E and R times are introduced by ASPECT, with ordering (perfective) or overlap (imperfective), and the R reference time is unselectively bound by the discourse context, through a process on the analog of existential closure, which gives an index to indefinite NPs. The alternative is to assume lexical reanalysis of perfective as past (which may reflect actual historical development in the eastern Indie languages). Unselective binding allows us to keep

Agreement Features and Projections of TENSE and ASPECT · 55

the generalization that ergative case in HU is uniformly licensed by perfective aspect, whether or not tense is also projected. Otherwise, we would have to say that perfective aspect licenses the ergative case in combination with a tense copula, but it is licensed by a past tense with non-finite aspect and agreement morphology when the tense copula is not projected. The study of Indie languages by Masica (1991) allows us to see whether the differences between Hindi/Urdu and Kurmali are part of a larger pattern, or are random idiosyncrasies. Languages like HU with split ergative case require non-finite morphology on the perfective aspect form, and allow aspectual forms to be used alone as equivalents of simple past and present. Languages without split ergative case either lack ergative case entirely, or have ergative case for all transitive subjects (Kurmali), and also require finite agreement morphology. Languages of an intermediate sort, such as Nepali, Sindhi and Marathi, have a less clear distinction between ergative and nonergative subjects, and also give perfective forms an ambiguous analysis. In combination with a tense copula, they are aspectual, and have adjectival agreement. Alone, they have person as well as number and gender markings, in other words, they have acquired finite agreement. If finite agreement indicates TENSE, then the perfective suffix must be reanalyzed as past tense. Agreement features may lack meaning in themselves, but they acquire a surrogate semantic value because of the selection properties of TENSE and ASPECT. TENSE allows only person-number features, while ASPECT allows only number-gender features, not person. In the ambiguous or indeterminate instances here of what may be tense or may be aspect, the nature of agreement suffixes seems to be a reliable surface indication of the form and its semantic interpretation. • NOTES 1. In both Hindi/Urdu and Kurmali, ASPECT may iterate. TENSE does not iterate. 2. Broadly speaking, the progressive aspect focuses on the internal stages of an event, excluding the initial and final endpoints (Smith 1997), much like the imperfective aspect in some languages. The semantics of the progressive aspect are, however, rather complex (see a summary and discussion in Giorgi and Pianesi 1997). I do not give an analysis in terms of E, R and S for this reason. But morphologically the progressive aspect consists of an aspect auxiliary plus a tense copula in both Kurmali and Hindi. 3. For most speakers consulted, the option of the ergative on the embedded subject is ungrammatical. For some, however, it is an option: (i) pitaa-koo [kuttee-nee/ /amnn-meeN gaDDhee khood-ee (huee) lag-tee/jaan paR-ee father-Dat dog-ms/erg ground-in hole-Nom-mpl dig-Pf-mpl seem-Impf-mpl 'It seemed lo father that the dog had dug holes in the ground.' (Mpl agreement with gaDDhee) (ct. Mahajan 1990).

56 · Alice Davison For these speakers it may be that it is sufficient to have tense copula verb huee, in nonfinite participle form, to license the ergative, or that the matrix tense is the licensing factor. I leave this issue open. 4. Comrie (1981: 70-71) points out that this combination has three possible interpretations, because there is no ordering established between S and event time E (see also Giorgi and Pianesi 1997: 28). 5. The present perfect in Hindi/Urdu is very clearly specified for 'present relevance', while the perfect alone simply locates the event prior to some reference time R, the default identification being the speech time. The following minimal pair (Rashmi Gupta, p.c.) illustrates the difference: (i) mujhee pataa hai [ki tum-nee coorii nahiiN kii] [Perfect alone] I-dat information is that you-erg theft not do-pf-fs º know that you didn't steal anything (simple fact, end of discussion).1 (ii) mujhee pataa hai [ki tum-nee coorii nahiiN kii hai] [Present perfect] I-dat information is that you-erg theft not do-pf-fs be-pres-3s º know that you didn't steal anything (but there is still a problem)/ The perfect alone is used in reporting a simple fact, which is not related to something at the present moment. The present perfect connects the sentence to a continuing discourse. 6. There are two alternatives to the unselective binding account of R, both of which require a lexical reanalysis of the perfective morpheme. The first involves the pragmatic introduction of S combined with R. R is reanalyzed as overlapping with E and preceding S: (i) a. E R perfective b. E R, S add S, identical to present perfect c. E, R S Move R to E, preceding S The second possibility is to say that a null tense is projected, of the form R S. It combines with the perfective. The perfective is reanalyzed as E, R: (ii) a. E_R + R__S (null) b. E, R__S Reanalyzed by shift of R to E Both involve stipulations of additions or null tense, which is not motivated in the absence of any triggers, such as finite morphology (see discussion Sections 5.1 and 5.2).

• REFERENCES Barz, Richard and Yogendra Yadav. 1993. An introduction to Hindi and Urdu. Delhi: Munshiram Manoharlal. Chomsky, Noam. 1995. The Minimalist program. Cambridge, MA: MIT Press. . 1981. Tense. Cambridge: Cambridge University Press. Davison, Alice. 1999a. Ergativity, functional and formal issues. Functionalism and formalism in linguistics, Vol. I (SLCS 41), ed. by Edith Moravesik and Michael Noonan, 177-208. Amsterdam: John Benjamins, 177-208.

Agreement Features and Projections of TENSE and ASPECT · 57 Davison, Alice. 1999b. Lexical anaphors and pronouns in Hindi/Urdu. Lexical anaphors and pronouns in selected South Asian languages, ed. by Barbara Lust, Kashi Wall, James W. Gair and K.V. Subbarao, 397-470. Berlin: Mouton de Gruyter. . 2001. Ergative case licensing, agreement and reflexive binding. Linguistic structure and language dynamics in South Asia, ed. by An vita Abbi, R.S. Gupta and Ayesha Kidwai. Delhi: Motilal Banarsidass. Diesing, Molly. 1992. Indefinites. Cambridge, MA: MIT Press. Giorgi, Alessandra and F. Pianesi. (1997). Tense and Aspect. Oxford: Oxford University Press. Grierson, Sir George. 1903/1968. Linguistic Survey of India, Vol. V The indo-Aryan family, eastern group, pan II. Delhi: Motilal Banarsidass. . 1916/1968. Unguistic Survey of India, Vol. IX, Parti. Indo-Aryan, central group, western Hindi and panjabi. Delhi: Motilal Banarsidass. Heim, Irene. 1982. The semantics of definite and indefinite noun phrases. University of Massachusetts dissertation. Hornstein, Norbert 1993. As time goes by. Cambridge, MA: MIT Press. Klaiman, Miriam. 1987. Mechanism of ergativity in South Asia. Lingua 71. 61-107. Mahajan, Anoop. 1990. The A/A-bar distinction. MIT unpublished dissertation. Mahto, Panchanan. 1989. On the nature of empty pronominals. Hyderabad: Central Institute of English and Foreign Languages unpublished dissertation. Masica, Colin P. 1991. The Indo-Aryan languages. Cambridge University Press. McGregor, R.M. 1995. Outline of Hindi grammar (Third edition). Oxford: Oxford University Press. Sengupta, Gautam. 1999. Lexical anaphors and pronouns in Bangla. Lexical anaphors and pronouns in selected South Asian languages, ed. by Barbara Lust, Kashi Wali, James W. Gair and K.V. Subbarao, 277-332. Berlin: Mouton de Gruyter. Smith, Carlota. 1997. The parameter of aspect (2nd Edition). Dordrecht: Klu wer Academic Publishers. Speas, Margaret 1990. Phrase structure in natural languages. Dordrecht: Klu wer Academic Publishers. Swart, Henriette. 1998. Aspect shift and coercion. NLLT 16(2). 347-85 Ura, Hiroyuki. 2000. Checking theory and grammatical functions in Universal Grammar. Oxford: Oxford University Press. Verma, M.K. 1991 Exploring the parameters of agreement: The case of Magahi. Language Sciences 13(2). 125-43. Wallace, William. 1985. Subjects and subjecthood in Nepali. University of Illinois. Unpublished dissertation. Yadav, Ramawatar. 1996. A reference grammar of Maithili. Berlin: Mouton de Gruyter. Yadav, Yogendra. 1998. Issues in Maithili syntax: A Government-Binding approach. Munich: LINCOM Europa.

Sanskrit as She has been Misanalyzed Prosodically • RICHARD D. JANDA and BRIAN D. JOSEPH ·

The various alternations between aspirated and unaspirated segments in Sanskrit phonology that have come to be discussed under the rubric of Grassmann's Law and Bartholomae's Law and their interaction have long been a stumbling block as well as a proving ground for various phonological theories. In this paper we give a detailed critique of analyses of these phenomena within a 'Late Classical' autosegmental framework (circa mid-to-late 1980s) which have not been challenged in the literature even with the major changes in theory heralded by Feature Geometry and Optimality Theory. While updating our critique to encompass these further advances in phonological theory, we find no purely phonological account to be satisfactory, and thus argue that the key to understanding these Sanskrit aspiration alternations lies in recognizing the essentially morpholexical nature of the relevant phenomena. Accordingly, we provide a sketch of an account within a Process Morphology framework, in the spirit of the analysis given by Pänini nearly 2500 years ago.*

• 1. Introduction The title of this paper deserves some explanation. It is inspired by a brief anonymous note published in the Indian periodical, Indian Linguistics (1981: 81) *An earlier version of this paper was presented at the 9th South Asian Language Analysis Roundtable (SALA 9) at Cornell University, June 5, 1987. We are especially grateful to Hans Hock and David Odden, for helpful suggestions and comments regarding this paper, and to Raj Singh, for encouraging us to reopen our consideration of this still quite timely topic.

60 · Richard D. Janda and Brian D. ]oseph and entitled 'Sanskrit as she is spoke at MIT Press'. In it, attention was drawn— with playful intent, we believe—to a textbook in historical linguistics (Jeffers and Lehiste 1979) in which some example sentences were constructed using Sanskrit verbs but English noun-forms; these sentences thus did not represent Sanskrit in any real sense, being at odds with the actual facts of the language, e.g., regarding case-marking.1 In this paper, focusing not on MIT Press but rather on the MIT linguistic tradition, we question some generative linguistic descriptions and analyses of Sanskrit which we feel are at odds with the actual facts of the language. In particular, we subject to severe criticism—all of it, we believe, justified— two treatments of Sanskrit aspiration-related alternations. These alternations derive historically from the interaction of Grassmann's Law2 with other sound changes, including Bartholomews Law (discussed at some length later), and are often seen as involving an 'aspiration throwback' effect; compare, e.g., bodh-ati '(s)he knows' and bhot-sya-ti '(s)he will know'. The two analyses on which we focus, Borowsky and Mester 1983 (hereafter B&M) and Kaye and Lowenstamm 1986 (hereafter K&L), adopt an aspiration throwback approach (as have virtually all works since Sag [1976] and Schindler [1976]; cf. Joseph and Janda [1988a], Janda and Joseph [1989]), but most importantly they treat the phenomenon as being purely phonological in nature. By contrast, we provide an alternative morphological analysis, already foreshadowed in spirit by Sag (1976) and consistent (as Sag himself explicitly notes) with the essentially Process-Morphological treatment of Pänini from 2500 years ago. Given that the two accounts in question appeared in the mid-1980s and therefore predate many theoretical developments that have characterized phonological theory in the intervening 16 years—not only Optimality Theory but also Feature Geometry, for instance—it might be thought that, in discussing these, we are beating a dead horse into the ground (so to speak). However, this is in fact not the case; we are not beating a dead horse but defending a dead language. In the first place, with only one exception—Broe (1991), who does not discuss either B&M or K&L3—we do not know of any more recent published treatments focusing primarily or even in passing, on the Sanskrit facts mentioned earlier. Instead, the recent literature on aspiration in Sanskrit is made up of surprisingly brief references on the order of the following taken from Kenstowicz (1993): [-»•spread gl[ottis]]... may delink from one position and surface elsewhere (e.g., Grassmann's Law in Sanskrit) (Kenstowicz 1993: 493)

Sanskrit as She has been Misanalyzed Prosodically m 61 [an] unlicensed laryngeal feature ... may delete or relink at some adjacent position (e.g., the aspiration throwback of Grassmann's Law) (ibid.: 495). Such short-cuts (to phrase matters generously) show a cavalier disregard for both the terminology and the content associated with discussions of Grassmann's Law in the specialist literature within Indo-European, Indo-Iranian, and Indie historical linguistics—indeed, but even within generative phonology itself. The source of such an error is not far to seek: many articles have discussed the two distinct but related phenomena of Grassmann's Law and aspiration throwback in the same breath, as it were. Moreover, within a single widely cited volume, the Handbook of Phonological Theory, edited by John Goldsmith (1995), Sanskrit phonological and morphological phenomena have been misleadingly presented. For instance, Steriade (1995: 149) states it as a known fact that Only one aspirated stop is allowed within a given root in Indo-European, Sanskrit, and Greek'. While deaspirate roots are admittedly rare, and possibly late, in the Sanskrit tradition, they do exist in the language; examples include Vdhrakh- 'become dry', Vdhrägh- 'be able', and Vbharbh- 'injure'.4 Such forms are found as well occasionally even in Greek (e.g., thuphlos 'blind', a dialect variant of the more widely attested tuphlos).5 Further, McCarthy and Prince (1995: 365, n. 8) suggest, albeit somewhat tentatively, that 'apparently, complex onsets... in Sanskrit... [are absent from] all affixes,... not just reduplicative ones'. Appearances, however, can be deceiving, given that there are complex onsets not only in (somewhat lexical, i.e., content) prefixes such as pra- 'before' and prati- 'against, in return, towards' but also in such grammatical suffixes as the 2PL. non-active ending -dhve, tbe gerund ending -tvä, etc.—all with complex onsets of obstruent plus sonorant. On the other hand, the admittedly brief further statements that relate directly or indirectly to Sanskrit aspiration in a number of other papers in Goldsmith (1995) are accurate. Three of them bear only minimally on the matter at hand: Clements and Hume (1995: 269), Kiparsky (1995: 661), and Ohala (1995: 714). A fourth, moreover—Blevins (1995)—is not only right on target as regards the facts but furthermore absolutely consistent with the major thesis defended next: namely that Sanskrit aspiration alternations demand a treatment within morphology (and not just via some 'morphologically conditioned phonological rule', whatever that might be—see Janda [1987]). That is, Blevins explicitly recognizes that'... [constraints on aspiration in Sanskrit hold within the morphological stem' (Blevins 1995: 235, n. 8).6

62 · Richard D. Janda and Brian D. Joseph Before we proceed to the details of the analyses and of our counter-analysis, a word on notation and theoretical orientation is in order. Both B&M and K&L operate within what might be called a 'Late Classical' autosegmental framework, one heavily influenced by McCarthy (1979,1981). An aspiration feature, represented as H, constitutes an autosegment which is (or can be) linked to a C-position on a skeletal template (sometimes unsystematically represented as a direct association with a phonemic-melodic element such as /b/ or /p/) or to another autosegment (we illustrate their practice more explicitly next). However, in the same year as K&L's paper, Clements (1985) appeared— a highly influential paper that did much to establish the current geometrical approach to feature structure. Later still, McCarthy and Prince (1986) pioneered the less segmental and more mora-, syllable-, and foot-based approach known as Prosodic Morphology. While we briefly sketch next what the feature geometric equivalent of B&M and K&L's earlier autosegmental analyses would be, we retain their original notation for most purposes, mainly for ease of exposition, but also because it seemed inappropriate to claim to have refuted an analysis which was recast entirely by us rather than by the original authors. In parallel fashion, B&M and K&L obviously predate by nearly a decade the rise and spread of Optimality Theory (OT; see, e.g., Prince and Smolensky [1993]). Although in this case we provide some discussion of our analysis in OT terms, we again—for precisely the reasons just mentioned—eschew the exercise of converting their analysis into an Optimality Theoretic one. Furthermore, we would like to call attention here to the fact that OT is a much more congenial home for Process Morphological analyses than was the pre-OT scene, since, e.g., it gives no particular credit for—and sees no theoretical advantage to—the positing of floating autosegments which require intermediate derivational steps. Nevertheless, none of our arguments hinge on these analytical and expositional decisions. Let us note here, though, that there is an unfortunate tradition, in generative accounts of Sanskrit, of understating the range of facts that need to be taken into consideration in accounts of particular phenomena (on reduplication, for example, see Janda and Joseph [1986]).7 Not wishing to fall into this practice, we begin our exposition by pointing put, as most researchers in fact do realize, that actually there is more allomorphy in the aspiration throwback roots than just budh-lbhut-\ there is also an allomorph bud-. We illustrate all these more fully in (1), continuing to use the root budh- as our examplar: (1) 3SG.PRES.INDIC.ACT bodh-ati 4(s)he knows' 1SG.AOR.INDIC.MID a-bhut-s-i º knew'

Sanskrit as She has been Misanalyzed Prosodically m 63 3SG.FUT.INDIC.ACT PST.PASS.PARTCPL

bhot-syati '(s)he will know' bud-dha- '(having been) known'

Descriptively, these alternations involve the transfer of aspiration between the two consonants in the root, though aspiration never occurs overtly on both, as well as the transfer of aspiration to a non-root segment. The relevant allomorphs of such roots—i.e., the actual forms that an adequate grammar of Sanskrit must somehow account for, are listed in (2)— with the form of the vocalic nucleus abstracted to V: (2) bVdh- ~ bhVt- ~ bVd-. As is easily imaginable from our introductory discussion, alternations such as these have long been a stumbling block, as well as a proving ground, for phonological theories—within generative phonology, starting as early as Kiparsky (1965) and Zwicky (1965). Thus it is no surprise that, just barely after the dust had settled from reconsiderations of this controversy within the framework of 'Late Classical' (or 'Middle High') generative phonology, in the early-to-mid 1970s (represented, for example, by Hoard [1975]; Phelps [1975]; Sag [1974]; Schindler [1976]), a recasting of the problem in the nonlinear framework of autosegmental phonology was soon undertaken. Stemberger (1980) was the first such treatment, followed in elaborated form by B&M, with further refinements offered soon thereafter by K&L (we have already mentioned the revisions in notation, but not content, later made available by Feature Geometry, Prosodic Morphology, and OT). These alternations would seem, at first glance, to be exactly the sort of phenomenon that autosegments were invented for, so to speak, inasmuch as there is an element, aspiration, that appears in different but nearby locations. Thus, the two main autosegmental treatments, B&M and K&L, along with the further analytical steps that they entail are the focus of our attention here, because we feel that, in spite of some initially attractive features,' their analysis proves, on close inspection, simply to be untenable. Despite the fact that these accounts were proposed several years ago, the bases of the autosegmental analysis have not, to the best of our knowledge, been challenged in the literature, nor have many crucial details on which both B&M and K&L err been brought to light, all this despite the fact that certain aspects of the analysis are beset from the start by serious problems. Thus, here—besides clearing up several errors of fact in these treatments—we offer a rebuttal to the key features of the autosegmental phonological analysis of the Sanskrit aspiration alternations.

64 · Richard D. Janda and Brian D. Joseph

• 2. The Borowsky and Mester/Kaye and Lowenstamm Analyses We first give an overview of B&M's and K&L's autosegmenlal analyses, for the benefit of readers who (through no fault of their own) may not be intimately acquainted with them. Following what has become standard terminology in discussions of these alternations (as alluded to above in our introduction), we refer to the presence of initial aspiration in a root in the absence of root-final aspiration—as e.g., in the future stem bhot-sya-, from (1)—as the 'aspiration throwback' effect. B&M and K&L attempt to explain the aspiration throwback effect in purely autosegmental phonological terms, and to account at the same time for the absence of the aspiration throwback effect in other related forms—as, e.g., in the past passive participle buddha-, also from (1). The basic approach taken is to treat aspiration throwback and related phenomena as an inherently phono· logically conditioned set of processes; the fundamental aspects of their analysis are summarized in what follows. First, the autosegmentalists assume that the aspiration which is found with the voiced stops (e.g., dh) is to be distinguished from the aspiration that is found with the voiceless stops (e.g., th). This distinction between what can be called 'voiceless aspiration' and 'voiced aspiration' can be encoded in Sanskrit grammar in a number of ways; the approach suggested by B&M is to assign the same feature (their '[+H]') to one place (one autosegmental level) in the internal structure of the feature complex for voiced stops but to a different place for voiceless stops. Moreover, they assume that only voiced stops can be associated with the type of aspiration that is found with voiced stops. K&L achieve this result by linking aspiration (H) to a voicing autosegment (their 'Z'), thus essentially differentiating between voiced aspiration and voiceless aspiration, in the sense that voiceless aspiration is neither autosegmentally realized nor linked to a voicing autosegment, as is voiced aspiration. Thus for B&M, the Sanskrit bilabial voiced aspirate stop bh (so described also in traditional terms) bh is represented as in (4a); for K&L, the relevant representation is as in (4b), for which one updated feature geometric version would be (4c): (4) a.

H I /b/

b.

H I Z I /p/

c.

root / \ laryngeal oral cavity / \ / \ / \ C-place \ / \ I \ [spread] [voice] [labial] [-continuant]

Sanskrit as She has been Misanalyzed Prosodically · 65 Second, a further claim emerges, in B&M's and K&L's view, from their decision to represent the voiced aspirates (e.g., dh) with autosegmental aspiration—i.e., as an element on a different phonological level (or tier) from the segmental melody—and from their assumption that only voiced stops can be bearers of 'voiced aspiration'. In particular, when devoicing occurs somewhere in a root (as with the root-final element in the future bhotsyati), the aspiration autosegment must be disassociated from the devoiced stop, and this newly 'liberated' aspiration autosegment must reassociate with an acceptable bearer—in this case, the root-initial element, which is acceptable because it is a voiced stop.8 A sample derivation is given in (5) for the future bhotsyati (with some non-essential details ignored): (5)

H H H I ROOT-FINAL REASSO- / CVC+CCV==> CVC + C C V - = = > C V C + C C V I I I I I I DEVOICING I I I I I ICIATIONI I I I I I I bVd

si a-

bVt

si a-

bVt

sia-

The difference between their representation of voiced aspirates versus voiceless aspirates is utilized by B&M to account for the difference between the aspiration throwback effect in bhotsyati and what they claim to be the lack of such aspiration throwback in a root which ends in a voiceless aspirate and has a voiceless stop elsewhere in the root as the potential bearer of the aspiration autosegment. Such a root is Vprach- 'ask',9 which shows the future stem praksya-, and not *phraksya-. Thus, there is no aspiration throwback effect with Vprach-. This fact is explained, according to B&M, as resulting from the fact that the root-final voiceless aspirate does not have the [+H] aspiration autosegment, aspiration on voiceless stops instead being a feature of the segmental matrix. Therefore, ch actually would not have an [+H] autosegment that could be liberated and so be in need of reassociation. Thus, segmental deaspiration would occur, but no throwback would be possible, inasmuch as it depends on autosegmental aspiration.10 K&L are silent on this matter, but given their general approach, it can be assumed that they would basically follow B&M here. In feature-geometric terms, one can simply write the relevant delinking/relinking rules for Sanskrit to operate on a glottal/ laryngeal node only if it dominates both [spread gl(ottis)] and [voice], rather than [spread gl] alone. Third, something must be done in an autosegmental analysis—or any framework, for that matter—with the -DDH- clusters that arise in forms such as the past passive participle buddha—i.e., the Bartholomae's Law phenomenon. B&M analyze the -DDh- cluster in buddha· (and those similar ones in

66 · Richard D. Janda and Brian D. Joseph other forms) as being aspirated throughout: i.e., as being phonetically [budhdha-]. Bartholomae's Law (BRTHLAW) is thus treated as a progressive voicing assimilation process with no deaspiration, but rather with aspiration spreading. This step then obviates the need for aspiration throwback since in their terms there is no deaspiration of the root final segment, and thus no liberated autosegmental aspiration in need of reassociation. This aspect of the analysis is shown in (6). (6)

H H I BRTH I \ CVC + C V - =====> CVC + C V I I I I I L A W I I I I I bud t abudda(= [budhdha-], graphic )

K&L, on the other hand, utilize a notion of syllabic prominence, a device originally employed mainly for accentual phenomena, and assume that aspiration, like accent in some languages, can reassociate only with non-recessive (i.e., strong) positions in a syllable—e.g., the syllable onset, but not the syllable coda. Thus for them, the cluster -ddh- need not be aspirated on both segments but instead need only be assumed to have the first element in one syllable as the coda and the second element in another syllable as an onset; only the second segment in the cluster would then be 'strong' and thus capable of bearing aspiration. Conveniently, Kenstrowicz (1993: 160-61) provides an explicit featuregeometric formulation of Bartholomae's Law, which we cite at length next: A common form of reduction is the suppression of laryngeal distinctions in the coda of the syllable. For example, many languages oppose plain, aspirated, and voiced stops [p, b, ph] in syllable onsets but limit their coda to just [p].... Other laryngeal features sometimes display similar behavior. A possible example is Bartholomae's Law in the development of Sanskrit, where an Indo-European [bh + t] cluster is realized as [b + dh], with the apparent transfer of voicing and aspiration from the coda [bh] to the following onset [t].... Under the earlier theory in which features form an unorganized bundle, this sound change requires a simultaneous modification of the coefficients of both [spread gl] and [voiced] at two successive positions in the cluster (22a). (22)

a.

[+spread gl] [-spread gl] —> [-spread gl] [-»-spread gl] [ -i-voiced ] [ -voiced ] [ +voiced ] [ +voiced ] [bh] [t] [b] [dh]

Sanskrit as She has been Misanalyzed Prosodically · 67

b.

[+cons] I Pharyngeal I / Glottal / \ [+voiced] [+spread gl] [bh] [+cons] I Pharyngeal I Glottal

[b]

\

\

[+cons] / \ Pharyngeal Oral I *Coronal

—>

[t]

[+cons] / \ Pharyngeal Oral I Glottal Coronal / \ / \ [+voiced][+spread gl] [dh]

—>

But, given the Articulator Model, it can be described as the delinking and subsequent relinking of the glottal articulator. We note that Kenstowicz too, like us, analyses the cluster that results from this rule as having only final aspiration, not aspiration on both members. Fourth and finally, as has been frequently observed (so Sag [1974, 1976]; Schindler [1976] and others), reduplication syllables in Sanskrit generally do not tolerate aspiration.11 B&M block the reassociation of a floating aspiration autosegment with a reduplication syllable via a brute-force stipulation—e.g., through prespecification of reduplication template as tolerating only nonaspirated consonants—and thus 'explain', after a fashion, the absence of aspiration in (most) reduplicative affixes. K&L, on the other hand, have nothing to say about this particular deaspiration phenomenon.12 As noted here, the B&M approach has not been seriously challenged in the literature, though K&L offered some early refinements. We therefore here present a rebuttal of their autosegmental arguments in favor or purely phonological analyses for Sanskrit aspiration throwback. In particular, we furnish counter-evidence to each of the main points of B&M's analysis as summarized earlier, and of K&L's where appropriate, especially where the two analyses differ. Taken together, these counter-arguments have the effect of demonstrating that the autosegmental approach to aspiration throwback (and any other possible recasting thereof in a similar vein) is not only poorly motivated in the absolute but also distinctly inferior to a morphological account. More generally, we argue that the basic problem with B&M/K&L-type approaches

68 · Richard D. ]anda and Brian D. Joseph

is that they take a lexically restricted process (about which more later)—one with various morphological conditions on it (e.g., that aspiration is absent from reduplication in most morpholexical categories)—and try to turn it into pure phonology. Finally, we present the outlines of an analysis, framed within a Process Morphology model, in which not everything that affects the phonological shape of a morpheme must be carried out in the phonological component or by phonological rules.

• 3. Contra Autosegmental Accounts of Aspiration Throwback We turn now to a point-by-point consideration of B&M's and K&L's analyses—though, for expository reasons, we do so in a different order from the presentation of their arguments given earlier. • 3.1. Regarding Voiced Aspiration as an Autosegment First, with regard to the claim that voiced aspiration should be treated as an autosegment (or alternatively, as a unitarily de-/re-linking feature geometric node), it must be admitted that this certainly is an initially plausible analysis. Nevertheless, there is no independent evidence for such a treatment, however attractive it might be on theoretical grounds—especially given that both of B&M's arguments in support of this view turn out to be nonarguments. K&L offer no additional arguments; they instead merely assume B&M's position. B&M point to alternations and/or alternatives involving voiced aspirates and the Sanskrit consonant /h/—see (7)—as evidence that aspiration is an autosegment, since they reason (Borowsky and Mester 1983:53) that in these, 'the [segmental] melody may be deleted [but] the aspiration is unaffected': (7) a. hita- 'having been put' « dhitvä 'having put' (from Vdhä- 'put') b. grbhnäti» grhnäti '(s)he seizes' (from Vgrabh-/Vgrah- 'seize') Such facts might seem to give an autosegmental treatment a clear advantage over alternatives, since earlier, segmental, approaches which treat [aspiration] as just one element in a feature matrix appear to require considerable power in paring down such a matrix from bh or dh to just h. That is, though perhaps not impossible, such a tactic of differential feature-matrix shakedown (as it were) is difficult to reconcile with usual assumptions about the writing of rules and about reference to features in such rules. Still, while autosegmental aspiration might initially seem to simplify such matters, the situation is far more complicated than B&M let on.

Sanskrit as She has been Misanalyzed Prosodically · 69

In particular, the decision to derive, in forms such as those in (7), the Sanskrit aspirate h (usually described as a voiced—probably murmured—glottal continuant; cf. Whitney 1889: §65 for discussion) by conversion from an obstruent (even an aspirated one) runs into the same problems encountered by featuregeometrical (nee autosegmental) treatments of the so-called 'aspiration' of/s/ in Andalusian as well as many New World varieties of Spanish. In both of these, the crucial requirement for an elegant account is that the conversion in question be effected by a single unitary localized operation on feature structure; in the words of Kenstowicz (1993: 159): 'many researchers have viewed such sound changes as... [s] to [h]... as a suppression of the supraglottal articulation ... [but] the status of [continuant] in these changes is controversial'. Since Sanskrit dh and bh are [-continuant], whereas h has been treated as [+continuant], Kenstowicz's remarks about Spanish apply here as well: i.e., it may take two or more operations to carve dh/bh down to A, thus potentially nullifying any other advantages of the autosegmental approach. B&M's second argument has to do with the fact that, as noted earlier, aspiration virtually never occurs in reduplicated syllables, i.e., with the fact that, in their terms, the segmental melody can be copied in reduplication independently of aspiration, as in (8): (8) a. Vdhä- 'put' —> present stem da-dhab. Vbhi- 'fear' —> aorist stem bi-bhay-aHowever, B&M themselves recognize that there is a need for prespecification of the reduplicated syllable (what might now be recast as melodic overwriting, cf. McCarthy and Prince 1995: 340, 365, n. 12), at least for the change of velars to palatals as a regular part of reduplication; see (9): (9) a. Vkr- 'make' —> perfect stem ca-karb. Vgä- 'go' —> present stem //-gain actuality, then, a further prespecification (or overwriting) for [-aspirate] is hardly an added burden on the grammar of Sanskrit, since some prespecification is already independently necessary, and hence independently motivated, for the velar-to-palatal part of reduplication. Curiously, B&M also realize (Borowsky and Mester 1983: 62, n. 5) that accepting prespecification vitiates their 'argument from reduplication for the independence of the autosegmental H from the segmental melody', but they add—somewhat plaintively, it seems to us—that 'this is not devastating since... there is enough independent support for our analysis'. In view of our overall remarks here, however, this last statement of theirs is seriously open to question.

70 · Richard D. Janda and Brian D. Joseph • 3.2. On the Unity of Voiced Aspiration and Voiceless Aspiration The next major part of the autosegmental analysis is the claim that, in phonological representations, voiced aspiration is to be distinguished from voiceless aspiration. Here again, we argue that the autosegmental analysis is on shaky ground.13 First, as concerns the phonetics of the aspiration in the so-called 'voiced aspirates' versus that in the voiceless aspirates, it is not entirely clear just how different the two types are. It is of course true—as both Allen (1953) and Bare (1980) have pointed out in their surveys of the statements made regarding these sounds in the präüsäkhyas (the native Sanskrit grammatical treatises dealing with phonetics)—that there are some differences in the descriptions of these two sorts of sounds. Allen observes that 'the voiced aspirates are considered as more fully voiced than the non-aspirates, and the voiceless aspirates more fully breathed than the non-aspirates'. It is also true that writings by some phoneticians—e.g., Ladefoged (1975)—have claimed that the socalled 'voiced aspirates' are actually breathy voiced, or murmured, consonants. However, Allen (1953) himself ends up adopting the view that the differences between the two types of stops are not significant with regard to aspiration, and he speaks offen, dh, etc., as legitimate voiced aspirated stops. Bare's consensus overview of the various präüsäkhyas reaches a similar position. Moreover, despite Ladefoged's early (1975) view, it is significant that in subsequent works he later recanted (so to speak) (cf. Ladefoged 1982, 1993: 144-46). In any case, there are studies, such as that of Benguerel and Bhatia (1980), which show that it is indeed not improper to speak of 'voiced aspirates' (even if the term is phonetically a bit misleading). This is especially so if— following Ladefoged (1976)—one defines aspiration in a somewhat general way as 'a period after the release of a stricture and before the start of regular voicing in which the vocal cords are further apart than they are in regularly voiced sounds', or—like Dixit (1979)—as 'glottal friction produced with or without pulsing while the glottis is narrowly or widely open and the supraglottal vocal tract is unobstructed'. Thus the phonetic evidence is somewhat ambiguous, but it leans towards favoring a unified treatment of voiced and voiceless aspiration.14 Second, from a phonological standpoint, there is no reason to distinguish th (etc.) from dh (etc.) vis-ä-vis aspiration, because they show parallel phonological behavior—synchronically and diachronically—when they undergo various processes which involve aspiration. For example, both voiceless aspirates and voiced aspirates deaspirate when reduplicated, as shown in (10): (10)

a. Vbhid- 'split' —> perfect stem bi-bhed-, desiderative stem bi-bhit-sab. Vphar- 'scatter' —> intensive stem par-pharac. Vphal- 'burst' —> perfect stem pa~phal-

Sanskrit as She has been Misanalyzed Prosodically m 71 And, further relevant forms bearing on the matter at hand are shown in (11): (11) a. samidh- 'wood, fuel' —> samit (NOM.SG) b. tristubh- 'a Vedic metre' —> tristup (NOM.SG) c. kaprth- 'penis' —> kaprt (NOM.SG)

Here, both voiceless and voiced aspirates can be seen to deaspirate in wordfinal or root-final position, though examples with voiceless aspirates seem to be restricted to a single lexical item, kaprth- 'penis'.15One other deaspiration context in which the voiced and voiceless aspirates behave alike comes from the automatic doubling of a consonant after r (and in some other contexts; see Vaux 1992). Thus, e.g., arka- 'light' is realized as [arkka-], and when an aspirate, whether voiced or voiceless, is so doubled, aspiration is found only on the final (rightmost) member of the doubling; e.g., we find artha- 'goal' —> [arttha-] (not *[arththa-]), dirgha- 'long' —> [dirggha-] (not *[dirghgha-]), etc. Third, from a diachronic standpoint, both voiceless aspirates and voiced aspirates in nonreduplicated syllables not only deaspirated via Grassmann's Law but also triggered deaspiration via Grassmann's Law; the two examples shown in (12) attest to their parallel historical behavior in this regard: (12) a. vidatha- 'distribution' < *vi-dh-atha- (cf. v/ + Vdhä- 'distribute') b. kumbha- 'pot' < *khumbha- (cf. Avestan xumba-) Fourth, the conclusion that there is a close relationship between the voiceless and the voiced aspirates is suggested by the fact that a voiceless aspirate becomes a voiced aspirate when it is in a position where it can undergo voicing assimilation—in particular, in a Bartholomae's Law context: one involving a preceding voiced aspirate, where progressive voicing assimilation is the norm.16 Thus, the second person singular middle-voice past ending -thäs, when added to the root Vlabh- 'take' in an s-stem aorist (past) form, surfaces with [dh]: hence, alabdhäs 'you took' from /a-labh-s-thas/.17 Thus, changing the voicing feature of a voiceless aspirate yields a voiced aspirate, and this suggests that the two types differ only in voicing, not in the nature of their aspiration. Given all this evidence, it seems best to admit that from the standpoint of their phonological behavior, the voiced aspirates and the voiceless aspirates show a parallel type of aspiration. There is thus little to be gained from the arbitrary splitting that B&M and K&L impose on the phenomenon of aspiration in all processes and contexts—for instance into an autosegment for voiced aspiration and something different for voiceless aspiration. In fact, it would seem that a complication must be introduced into the grammar of Sanskrit

72 · Richard D. Janda and Brian D. Joseph under such an approach to aspiration, since there would be a need for one rule to deaspirate voiceless aspirates in word-final position (as with kaprth-) and a different rule to deaspirate voiced aspirates in the same position (as with samidh-). Moreover, the one piece of evidence that B&M give for a major (representational) difference between voiced and voiceless aspiration turns out to be non-evidence. This evidence has to do with the differential behavior regarding the aspiration throwback effect evinced by the root Vprach- 'ask' and its final voiceless aspirate, on the one hand, versus a root such as Vbudh- 'know' and its final voiced aspirate, on the other. While it is true that Vprach- forms a future stem praksya- and not *phraksya-, it is not at all clear that the future stem has to be derived from a root form with a voiceless aspirate. That is, while Vprach- does behave in some ways like a full-fledged root, with derivatives such as pfcchä- 'question', there also exist apparently related derivatives that do not have a voiceless aspirate: e.g., prasna- 'question'. The occurrence of forms such as prafna-, apparently from a root form Vpras- is significant, since a future stem praksya- is exactly what would be expected from the root form Vpras-. Diachronically, too, it is clear that the -cA- in Vprach- is the result of a special formative that characterizes present stems.18 Thus, the future stem associated with Vprach- is actually irrelevant to the question of aspiration throwback and its (non-)relation to the voiceless aspirates.19 There are two other forms, however, which are somewhat like praksya· in not showing aspiration throwback onto a voiceless stop and thus need further comment. For instance, the stem kaprth- 'penis', given earlier in (11), loses its stem-final aspiration in, e.g., the nominative singular (kaprt), but that aspiration does not reappear on either of the other voiceless stops in the word (i.e., we find neither *khaprt nor *kaphrt). Just as in the situation with Vprach-, then, what B&M treat as 'voiceless aspiration' is at issue here. The other form, however, offers a slightly different problem. The root Vtrh- 'crush' forms a future with the suffix -sya-, and the latter expectedly occasions deaspiration of the root final voiced aspirate A, yet there is no accompanying aspiration throwback: the future stem is tarksya- not ^harksya-.20 These forms admittedly do suggest that there is some difference between voiced aspiration and voiceless aspiration, a situation that is consistent with B&M's treatment. On the other hand, given the morpholexical solution that we favor here—one in which participation in aspiration throwback is not an automatic property of a root having a final voiced aspirate, even when the root-initial stop is voiced (cf. the discussion of gutsa- below)—we simply have to say that, for whatever reason (most likely analogical interference from unaspirated forms of the root21), the root Vtrh- simply failed to get on the aspiration throwback 'list'. In a sense, the occurrence of such exceptions, even if they might appear to have a phonological basis (e.g., if one claims a systematic basis to the absence

Sanskrit as She has been Misanalyzed Prosodically m 73 of B&M's so-called 'voiced aspiration' with voiceless stops), is also consistent with the morpholexical approach advocated here. It should be noted that if there were no phonological-behavioral evidence uniting the voiced and voiceless aspirates, then the absence of *tharksya- and *khaprt/kaphrt would probably favor B&M's approach. But, dh and th do in fact show parallel behavior—and do so in several ways, as discussed earlier—and that behavior is a sufficient counter, we submit, to this one possible strike against a nonphonological account. • 3.3. On Bartholomae's Law Forms and -DDh- Clusters To continue with details of the B&M/K&L analysis: we noted that, in order to account for the failure of aspiration throwback to occur in the participle buddha- and other forms like it which have undergone Bartholomae's Law, B&M claim that a cluster such as -ddh· is in fact aspirated throughout the cluster, and they hold that 'the written form reflects orthographic convention and not phonetic fact'. A number of related points need to be made here, however. First, the Sanskrit writing system notes many details of phonetic realization, but not necessarily consistently; thus, one could say that the burden of proof is on those who claim that some aspect of the written form is merely an orthographic convention—i.e., on B&M—rather than on one who would take the spelling at face value. Second, there are spellings with doubled aspirated consonants, especially—but not necessarily exclusively—in expressive forms like Vedic akhkhal'i-krtya 'shouting with joy; making (the noise) akhkhala\ or jajhjhatir 'hissing',22 and these show that such spellings were available. The fact that they were not consistently used for Bartholomae's Law clusters as in buddha- then becomes somewhat problematic for B&M's position, in which -ddh- is viewed as aspirated throughout the cluster. In the interests of fairness, though, it must be admitted that until a thorough examination of Sanskrit spelling practices is available, perhaps the safest thing to say is that the orthographic issue raised by B&M is moot, and favors neither one position nor the other. At the very least, though, the interpretation of the -ddh- clusters is not obviously simply a matter of spelling convention, as B&M suggest. The K&L approach to the treatment of the -DDh- clusters, on the other hand, runs into a factual problem. While their decision to extend from their original use for accentual matters the notions recessive versus non-recessive portions of a syllable to matters involving the spread of a feature such as aspiration can be recast in terms of licensing, (cf., e.g., Goldsmith [1990]; Kenstowicz [1993: 285-91]; and Steriade [1995: 158-65], among many others), the more important issue is the fact that voiced aspirated stops can indeed occur in syllable-final position.23 For example, in the word budhna- 'bottom',

74 · Richard D. Janda and Brian D. Joseph

the syllabification is [budhSna-], to jtidge from the description of syllabification prescribed in thepräüfäkhyas as summarized by Allen (1953: 82): 'the general rule is that an intervocalic consonant, as also an absolute initial or initial group, belongs with the following vowel, but that the first consonant of a medial group belongs with the preceding vowel'. Hence, budhna- has the voiced aspirate in a recessive (syllable-final) position.24 In any case, though, ordering (or ranking) of the Bartholomae's Law effect ahead of any final deaspiration process (or constraint) would be sufficient to prevent its apparent reassociation, as indicated in (13): H H I \ ROOT-FINAL \ C V C - - C V - => C V C + C V L A W I I I I I DEASP I I I I I

b u d t a -

budda-

budda-

With such an ordering (or ranking), there is no need for the apparent reassociation of H (or [spread gl(ottis)]) with a root-initial segment, since it is already associated with a segment after the operation of Bartholomae's Law.25 • 3.4. Devoicing and Revoicing Problems The autosegemental analysis depends on its (voiced) aspiration autosegment being liberated' (delinked), once its host segment is devoiced and thus being in need of finding a new host with which to associate (relink). It is thus a curious fact about this analysis that the segment from which aspiration is delinked can in fact surface as voiced, if a voiced segment follows it. That is, there exist more than the allomorphs given in (2) and repeated here for convenience of reference: (2) bVdh- ~ bhVt- ~ bVdNamely, there also occurs the allomorph bhud-, which is found in various voicing environments where deaspiration is nonetheless called for—e.g., before a voiced obstruent, as with the instrumental plural suffix -bhis (thus, bhud-bhis) or post-lexically when the nominative singular is followed by a voiced segment (e.g., ...bhud ## a...).26 What is problematic about this fact for the B&M/K&L analyses is its implication that the final segment of a relevant root could actually be an appropriate bearer of the voiced aspiration (eventually—i.e., later on—in a derivational approach); the absence of aspiration from that position must be stipulated, therefore, and its reappearance on a root-initial segment is not an automatic consequence of delinking.

Sanskrit as She has been Misanalyzed Prosodically m 75 • 3.5. Dealing with Exceptionality A fifth major problem faced by autosegmental analysts of Sanskrit aspiration is that there is no natural way in their framework to account for the exceptional behavior of certain words—for instance, words that, under certain assumptions, can be analyzed as failing to show expected aspiration throwback effects. These forms include the lexical item grtsa- 'wise', which is at least diachronically related to the root Vgrdh- 'be greedy' and thus might be expected to show aspiration throwback since Vgrdh- is historically a deaspirate. In this case, though, it turns out that Vgrdh- happens not to occur in any of the relevant forms—except for grtsa—that could show aspiration throwback and so this root may very well have been relexicalized as being only finally aspirated. Moreover, the synchronic derivation of grtsa· 'wise' from Vgrdh- 'be greedy' is not a transparent one. There are other forms which are like grtsa-, in being potentially irrelevant perhaps equally irrelevant for the matter at hand, but which are still worth noting. These include grap-sa- 'tuft, bunch' which is historically connected with the root Vgrbh- 'take, seize' (see Mayrhofer 1956 et seqq.: s.w.); if the connection was strong enough (i.e., sufficiently motivated from a semantic standpoint) to be available to speakers synchronically at any point throughout the history of Sanskrit, then grap-sa-, with no aspiration throwback even though presumably from /grbh-sa-/, would be another form demonstrating the non-phonologicality of this process. The same can be said ofdrap-sa- 'drop', from a Proto-Indo-European root *dhrebh-, as long as at least some speakers linked this word with the root Vdrbh- 'string together'— a connection which, though not necessarily compelling, is certainly permitted by the semantics of the forms involved (and possibly reinforced by the rhyme with the semantically parallel grap-sa-) but which could only be folketymological in nature, given what is known about the Indo-European sources of this noun and the historically unrelated verbal root (see Mayrhofer 1956 et seqq.: s.vv). For each such form, their relevance to assessing the phonological nature of aspiration throwback depends on the strength of what are most likely to be at best rather tenuous synchronic derivations. A second problematic type of form is like grtsa-/grapsa-/drapsa- except that the synchronic derivation is quite solidly indicated. In particular, there is gut-sa- 'bunch' 27 —undoubtedly a hyper-Sanskritization of guccha- 'bunch, bundle'—which was treated as if Prakrit (it may well be a Dravidian loanword) and 'Sanskritized' by inverse Overapplication' of the regular correspondence of Sanskrit -ts- to Prakrit -cr/z-, which is seen, e.g., in the linkage of Sanskrit. matsara- (cf. Vmad- 'be glad; intoxicate') with Prakrit, macchara- 'cheerful; intoxicating' (see Mayrhofer [1956 et seqq.: S.V.] and Lee [1986]). Interestingly, some native grammarians (opining admittedly rather late in the Sanskrit tradition, in the Unudi-Sutra) folk-etymologically connected gui-sa- with a

76 · Richard D. Janda and Brian D. Joseph root Vgudh- 'tie' that was 'created' (i.e., posited) by the grammarians in order to explain this and other forms: e.g., gudhera- 'protecting1, itself a hyperSanskritization of guhera- 'protector' from the root Vguh- 'hide, cover'. This native analysis ofgut-sa- as derived from Vgudh- was proposed even though the supposedly derived noun does not show aspiration throwback. This has a direct bearing on the phonologicality of aspiration throwback, since if it were phonological, and triggered solely by the deaspiration of a voiced aspirated stop before -s-, then deriving the noun from /gudh-sa-/ would be expected to trigger throwback, thus leading to *ghutsa-. The fact that the absence of aspiration throwback in gut-sa- did not prevent the relevant grammarians from making a connection with Vgudh- indicates that—at that time, at least— aspiration throwback was not part of the (automatic) phonology of Sanskrit. Moreover, within the morpholexical approach that we are advocating here, one would simply have to say that Vgudh- was not on the list of aspiration throwback triggers; on this approach, complete consistency based on phonological form is not expected. Third, there are Vedic variants such as duksa-/dhuksa—aorist stems from Vduh- 'milk'—where aspiration throwback reveals itself to be optional. Whatever the diachronic explanation for such variation might be,28 it is clear that a purely phonological solution to aspiration throwback—one in which the throwback effect is automatic, given certain conditions, as the autosegmentalists would claim—cannot easily incorporate either the optionality of the processes involved or their lexical restrictedness. Fourth, phonologically similar, and in one case even identical, suffixes show differential behavior with regard to aspiration throwback, a situation which cannot be accommodated easily into a purely phonological account. For instance, the 2SG active imperative ending -dhi does not trigger aspiration throwback (cf., e.g., dug-dhi '[you] milk!' from Vduh-) whereas the 2PL middle imperative ending -dhvam does (cf, e.g., dhug-dhvam '[you all] milk [for yourselves]!', also from Vduh-). It is hard to see how the difference between /dh/ and /dhv/ could be associated with the absence or presence of a triggering of aspiration throwback, yet such would presumably have to be the case in a purely phonological account. Even more telling is the pair Schindler (1976: 634) notes: namely a derived abstract noun in -ßíá-, where aspiration throwback occurs with the root Vduh-, i.e., pratidhuktva- 'the property of being called fresh milk (prati-duh-)\ versus the gerund form with the suffix -tvawhere Bartholomae's Law operates and there is no aspiration throwback, i.e., (prati-)dug-dhv 'having milked'. The exceptionality here comes from similar or identical phonological material behaving differently in terms of triggering (or not) aspiration throwback; from a wider perspective, differential behavior for phonetically identical forms shows that the phenomenon in question cannot be purely phonological. The only way to distinguish the -rv- which triggers

Sanskrit as She has been Misanalyzed Prosodically m 77 aspiration throwback from the -iv- which undergoes Bartholomae's Law and thus occasions no aspiration throwback is in terms of precisely which suffix they belong to—i.e., a morphological and lexical (morpholexical) fact, and so clearly a non-phonological property. A fifth and final problem involving exceptionality is posed by the weak present stem of Vdhä- 'put', which behaves before obstruents as it if were underlyingly /dhad-/ (e.g., in the 3SG middle form dhat-te) rather than the expected /dadh-/ (which is found before vowels and resonants); that is, one would expect a 3SG middle present **daddhe (as if from /dadh-te/), so that this root constitutes an exception to Bartholomae's Law and thus seemingly shows aspiration throwback unexpectedly (along with devoicing/deaspiration of the stem final dh).29 Presumably, the variant stem-form her, most likely the result of a restructuring of the stem due to analogy, would have to be listed as a separate allomorph in the lexicon. Admittedly, though, this root is exceptional in anyone's treatment (see Schindler [1976] for discussion of these forms). • 3.6. The General Problem More generally, though, it seems that the basic problem with the B&M/K&L approach is that it takes a lexically restricted number of processes involving aspiration (recall the exceptional forms discussed earlier)—ones which are subject to various morphological conditions (such as the near-total absence of aspiration from reduplication)—and tries to turn everything into a matter of pure phonology. We suggest that the first step towards a solution to the aspiration throwback problem (and related effects) must be to recognize the inherently morphological character of these phenomena. Once we begin to incorporate exceptional morpholexical elements into our solution, there is less and less to be gained from a purely phonological solution. Accordingly, we propose, as did Pänini and others (e.g., Sag and Schindler) before us, that the key to understanding these aspiration phenomena lies in treating them as morphological in character, even if they manipulate some elements of sound structure.

• 4. Towards an Account within a Process Morphology Model Such a morphological solution to a deceptively phonological-looking problem is easily accommodated within a Process Morphology approach.30 In this morphological framework, a distinction is made between, on the one hand, the set of operations that are employed for morphological marking and, on the other hand, the actual morphological rules which are built up from

78 · Richard D. Janda and Brian D. Joseph combinations of these processes employed for the purpose of marking particular categories. All of these, however, count as part of the morphology of the language. For example, in German, umlaut occurs not only alone but also quite (in fact most) frequently in tandem with one or more of various affixes. It is thus an operation that, along with various affixations, is used for the morphological marking of such categories as diminutives and plurals: e.g., some plurals are marked only by umlaut, some by umlaut and suffixation, and some (actually, many) only by suffixation. Applying this view to Sanskrit, we can say, for example, that voicing assimilation is just one among the several operations that occur in the marking of past passive participles by—among other things—the suffix /-ta-/. There can of course be some phonological conditions on the realization of this operation—e.g., regressive assimilation as the default case but progressive assimilation if the root-final consonant is a voiced aspirate—but, within this framework, such conditions do not automatically make a generalization count as a phonological rule (or constraint) per se. Similarly, palatalization and deaspiration are among the operations that can accompany the marking of a category such as [PERFECT] via reduplication. As for where and when aspiration throwback occurs in Sanskrit, it is important to recognize, as should be clear from the discussion up to this point, that not only is the number of roots that participate in this phenomenon rather limited (a total of 13 in all), but the full range of Sanskrit morphological categories showing aspiration throwback is itself somewhat limited, too, though these categories are far from being a totally disparate set. For verbs, the relevant categories cluster around certain morphosyntactic features mainly involving tense/aspect (future, aorist, and desiderative) and person/number (especially second person, though once also third, and frequently plural number, though also singular). For nouns, the principal morphological concentration of aspiration throwback is to be found in the oblique cases (Instrumentals, datives, ablatives, and locative, though also an isolated nominative, accusative, and/ or vocative) and in the non-singular numbers (dual and plural, though occasionally also the singular). An enumeration of these categories is given in (14) for verbs and in (15) for nouns: (14) Verbal Aspiration Throwback Categories a. verbs marked for one type of PUT stem: that with the suffix -/sya/-, (but not in futures with the suffix -/isya/-) b. verbs marked for one type of DESID stem: that with the suffix -/sa/-, (but not in futures with the suffix -/isa/-)

Sanskrit as She has been Misanalyzed Prosodically · 79

c. verbs marked for either of two types of AOR stems: i. that formed by any of the -/s/-initial suffixes -/s/- or -/sa/- or -/sis/- (but not by the suffix -/is/-) ii. that formed without any aorist-suffix at all—but only in such 'root' aorists' second- and third-person singular, where these person/number combinations are marked by the respective underlying suffixes -/s/ and -l\l d. various 2SG and PL forms, plus one 3SG form: i. 'primary' 2SG.ACT -/si/ (as for the PRES.IND), plus 'secondary' 2SG.ACT -/s/ (e.g., for the root aorists mentioned in c.ii), 'primary' 2SG.MID -/se/, and imperative 2SG.MID -/sva/ ii. middle 2PL -/dhve/ and -/dhvam/ iii. 'secondary' 3SG.ACT -/t/ (as for the root aorist above) (15) Nominal Aspiration Throwback Categories: (mainly) case-forms of root nouns (but cf. [d] for a derivational category) a. M/F.NOM.SG. (-/s/ > 0—though possibly just -0), plus NTR.NOM/ACC.SG. (-0) and M/F/NTR.VOC.SG. (-0) b. the '-/bh/...'-cases: e.g., the INST.PL (-/bhis/), plus the DAT/ ABL.PL (-/bhyas/) and the INST/DAT/ABL.DU (-/bhyarn/), all of which trigger external (post-lexical) sandhi c. LOC.PL (-/su/), also an external-sandhi trigger d. -/tva/- abstract noun forming suffix. With regard to aspiration throwback outside of bare-root contexts, we thus prefer to analyze it as one of the operations which accompanies certain types of affixation—namely, those processes which add the affixes noted in (14) and (15) to a particular set of roots: in particular, those lexically marked (On the list', so to speak) as being able to undergo the throwback. The realizational rules for such categories might take the form exemplified in (16) for the vocative singular of nouns and in (17) for the 2PL.MIDDLE.PRESENT of verbs (the latter additionally marked with the suffix -dhve), where the first line represents the relevant morphosyntactic features, the second line (in slashes) shows the schematic phonological shape of the root, and the fourth line indicates the changes effected in that phonological string in such a way as to help generate the surface allomorph which realizes the relevant set of

80 · Richard D. Janda and Brian D. Joseph features (although both further morphological rules and automatic phonological processes may later apply before the required surface form is obtained): (16)

[+NOUN,+VOC., /([C, -continuant, + voiced]) [-»-segment], ([C, -»-voiced, -i-aspirated])/ 1 2 3 —> /[I, -»-aspirated] 237

(17)

[-»-VERB, -»-2, +PL, +PRES, +IND, +MID] /([C, ]) [-»-segment], ([C, ])/ 1 2 3 —> /[I, ] 23-1- dhvel

In (17), the angled brackets indicate that while suffixation of ~-dhve always marks the relevant morphosyntactic feature combination, aspiration throwback depends on certain conditions being met. The parentheses in the same rule allow for both vowel-initial and vowel-final root shapes (where throwback is obviously impossible but -JAve is still suffixed. Moreover, we would posit what can be called 'meta-redundancy-rules', or 'meta-templates' (cf. Janda and Joseph 1991, 1999) which serve to parse the various realizational rules and express generalizations across them— thereby expressing the unity of, for example, all the instances of suffix-initial -/s/ shared by the verbal 2SG rules respectively for ACT primary -/si/, secondary -/s/, MID primary -/se/, and IMPER -/sva/ (perhaps also relatable to the further verbal rules for PUT -/sya/-, DESID -/sa/-, and AOR -/s/, -/sa/-, or -/sis/-, as well as the toot-nominal rules for M/F.NOM.SG -/s/ and LOC.PL -/su/.31 Similar redundancy-stating metarules and the 'constellations' they express (see note 31) likewise allow groupings across the various aspirationthrowback-containing processes for-/bh/...- suffixation, -/dhv/...- suffixation, and the remaining, suffixless aspiration throwback forms. Finally, we would combine (i) lexical listing for idiosyncratic forms with (ii) lexicalcorrespondence (redundancy) rules (cf. Janda 1987, and references there, on the available variety of such devices) in order to handle exceptional morpheme combinations such as those seen in the various allomorphs of Vdhä- which fail to trigger Bartholomae's Law but do show aspiration throwback: e.g., /dhad+te/(3 SG PRES IND M|D } = [dhatte] (whereby /dhä/ and /dhad/—plus /dadh/— alternate). On the subject of lexical listing, we raise here the possibility that aspiration throwback in Sanskrit may have been totally lexicalized and hence that it would only have been a redundancy-expressing correspondence statement. On this approach, Sanskrit speakers would simply have memorized for each of the 13 aspiration throwback roots those particular forms, both nominal and

Sanskrit as She has been Misanalyzed Prosodically m 81 verbal, which show the throwback effect. The number of relevant forms would not have been huge and so would not have been a large burden on a speaker's memory. We are emboldened to raise this possibility because a similarly morpholexicalized process in another language has recently been analyzed as completely lexicalized, and the dephonologized operation in question shows up in even far more forms than does aspiration throwback in Sanskrit. At issue here is umlaut in Modern High German, which Fery (1996) and some others have treated as essentially only a lexical phenomenon. For them, the vowel fronting which constitutes Modern High German umlaut is preferably to be analyzed via massive lexical listings which pair an umlauted stem together with one or more affixes (or with a meaning other than that of a bare stem). Here, the only rules involved are redundancy statements which parse such complex listed forms and relate their parts to those of other complex words (and to the component morphemes and/or formative processes mentioned in the redundancy rules themselves). The evidence for this approach in Modern High German is rather strong: even in cases where umlaut is regular (exceptionless), it is rarely productive. In short, umlaut in Modern High German seems to be dying—though it is doing so via petrification, and thus through lexicalization is leaving behind widespread traces of its former lively presence. But it is dying at different rates in different categories and this provides a somewhat morbid but very real confirmation of its morpholexical nature. Janda (1998a) discusses this completely lexicalized approach to umlaut but concludes that, in the event that an analysis of umlaut other than one with lexical listings should be preferable, then one making use of the alignment constraints of Optimality Theory is surely preferable to any of various nonlinear, nee autosegmental, approaches still maintained by some. Thus, e.g., Janda (1998b) exploits the fact that there is no longer any need for floating autosegments in an analysis of Modern High German umlaut that makes use of OT's constraints on, e.g., alignment and faithfulness, which allow abandonment of derivations and their associated intermediate steps. The tableaux for such a morpholexical OT account of umlaut can consist simply of a set of ALIGN constraints which require the alignment of (the left edge of) a particular suffix with (the right edge of) a root whose last vowel is [+front] (alternatively, [-back]). These constraints are ranked higher than FAITHFULNESS for umlauttriggering suffixes but lower than the latter constraint for non-umlaut-triggering suffixes. Here, still, the morpholexical nature of Modern High German vowelfronting remains unmistakable. The option of adopting such an analysis for Sanskrit aspiration throwback is readily available. For the Sanskrit case, we would likewise require a set of ALIGN constraints, but here the relevant generalizations require the alignment of (the left edge of) a particular suffix with (the right edge of) a root whose first consonant is [-»-voice, + aspirated]. These constraints are ranked higher

82 · Richard D. Janda and Brian D. Joseph than FAITHFULNESS for aspiration-throwback-triggering suffixes but lower than the latter constraint for non-aspiration-throwback-triggering suffixes. One example each of such constraint rankings is given next in (18); we note that, for ease of exposition, these tableaux have been considerably simplified: relevant constraints other than those just mentioned have been omitted and we therefore have likewise factored out the deaspiration which occurs in all aspirationthrowback contexts. Nevertheless, we believe that the flavor of an alignmentbased OT account of Sanskrit aspiration throwback comes through from these tableaux. (18) Sanskrit Aspiration Throwback Tableaux AUGN (lst-C=voiced-asp. root, right, FAITHFULNESS /dugh-Kihve/ 'You all [ATB-triggering] suffix -dhve, left) milked' 4 («•mlk' + 2.pl.mid.') *! [dug+dhve] * â* [dhug+dhve] /dugh+dhi/ 'Milk! [2.sg.acL]' FAITHFULNESS (« 'milk' -f- '2.sg.imper.act') «**

[dug-Klhi] [dhug+dhi]

ALIGN (lst-C=voiced-asp. root, right, [ATB-triggering] suffix -dhi, left) *

*!

Given that OT allows, for the same phenomenon, both in essence purely morpholexical treatments like that just sketched, and purely phonological accounts relying entirely on syllable structure and its consequences for phonological configurations, it is clear that OT in no way resolves the seemingly eternal battle between phonologists and morphologists over the best analysis of alternations like Sanskrit aspiration throwback. Our own view is that OT (especially because it contains the function GEN, which generates essentially all possible analyses and related configurations of sound) is rather like Pandora's Box; it strikes us as including literally the full gamut of possible analyses, ranging from the very best to the very worst. Nevertheless, since OT, to repeat, offers no rewards for floating autosegments or intermediate derivational steps (even via universal association conventions), we welcome the opportunity it has given morphologists to reclaim Sanskrit aspiration alternations and other morphological property, so to speak, that once seemed permanently expropriated by phonologists. Both an OT morphological account and a Process Morphological account (as sketched above) agree in their overall orientation and in their major aspects. On the other hand, in saying, we do not thereby wish to downplay the fundamental insight of Process Morphology.

Sanskrit as She has been Misanalyzed Prosodically · 83

which, true to its name, reconceptualizes all non-root 'morphemes' as processes—or rules—rather than as 'things' (only roots and built-up stems are 'things'). While the present paper cannot pursue this idea to any appreciable length, the OT revolution's effect of leading scholars to rework rule-based analyses in terms of constraint-based ones (with constraints which are often not identical but sometimes quite similar to earlier analysts' rules) raises the intriguing possibility that an OT-oriented recasting of Process Morphology could conceptualize every (non-root) 'morpheme' as a constraint.32 Most importantly, though, we would emphasize the following: the mere fact that morphological operations manipulate sounds does not make them phonological in nature. After all, morphology like that above must be realized through sounds, and so, ipso facto, even highly lexically specific operations— such as aspiration throwback—are best treated as basically morphological in nature although they affect sounds. In short, then, we retain the well-supported basic insight of Sag and Schindler—and Pänini before them—that the Sanskrit aspiration throwback effects are basically a matter of morphology, and so we conclude that the attempts of autosegmentalists to deal with this purely in phonological terms are poorly motivated.33 While B&M and especially K&L have shown how autosegmental phonology would describe the facts in question (and have thereby hinted at the sorts of alternative analyses that its various successor theories might be tempted to propose), they have not shown that there is anything to be gained from such an account. Indeed, as we have demonstrated at length here, their attempts simply cannot get around the fact that, in their analyses, problems abound. • NOTES 1. Such a practice is of course entirely defensible for pedagogical purposes, which after all is the guiding principle behind a textbook; too much detail in the presentation of examples requires additional explanation that can distract the student from the point being made. 2. 'Grassmann's Law' is the name given to the sound change(s) by which the first of two Proto-Indo-European (PIE) aspirated stops in successive syllables became deaspirated: i.e., schematically, Ch...Ch —> C...Ch. The results of this change are found in both Greek and Indie (and possibly also in Tocharian), and it seems to have taken place independently in each of these branches. The label 'Grassmann's Law' is also often applied to the synchronic rule(s) found in these branches as the aftermath of the earlier sound changes. See Collinge 1985 for general discussion and for references both to the considerable early literature and to more recent works on this phenomenon. 3. Vaux (1992) discusses Bartholomae's Law to some extent, but he does so in the context of his main focus, namely gemination in Sanskrit. 4. For each of these, monoaspirate alternant forms are found, i.e., Vdräkh-, Vdragh-, and Vbharb- all occur. 5. Steriade's statement contains a potential pitfall for the uninitiated with regard to IndoEuropean, in that she here seems to be assuming the often-criticized 'Glottalic Theory' of

84 · Richard D. Janda and Brian D. Joseph

6. 7.

8.

9.

10.

11.

12.

13.

14.

Gamkrelidze and Ivanov (1972) (see Collinge [ 1985] for additional references), for whom G rassmann's Law is to be treated as a rule of allophonic distribution in PIE rather than a sound change that took place on the way from PIE to Greek and to Indie respectively. See Joseph and Wallace (1994) for a critique of this view. We are assuming that by 'stem' here she essentially means 'root'. Contrast Marantz (1982)'s mention of the Sanskrit reduplication rule with the enumeration in Janda and Joseph (1986) of some 10 or more distinct reduplication templates and hence rules. In a feature-geometric analysis, the association of devoicing and deaspiration follows automatically from the fact that [spread (gl[ottis])j and [voice] (also known as [-»-slack vocal cords]) are sisters both dominated by a glottal (or laryngeal) node whose deletion removes both of them simultaneously. It is important to point out that Vprach- is essentially the only root for which a claim concerning aspiration throwback onto a voiceless stop could be tested, since all other roots of the shape T/DVTh- (where T = voiceless stop and D = voiced stop) are so-called set roots. As a result, they show insertion of the vowel -i- in root-final position before most suffixes; thus, with set roots, there is never final devoicing/deaspiration to trigger aspiration throwback. Even if the aspiration of the root-final ch were somehow represented autosegmentally, the liberated autosegment could not reassociate onto the root-initial pt since p—being a voiceless stop—is not a possible bearer of the aspiration autosegment. See, however, Janda and Joseph (1986) for further discussion and especially for some details regarding reduplicative formations that do show aspiration in the reduplicative prefix. As discussed by Kenstowicz (1993: 631-33), McCarthy and Prince (1986) replace templatic prespecification with melodic overwriting (due to cases like English cancer-shmancer versus Oedipus-Shmoedipus where there is no constant template and hence apparently nothing to prespecify). See also the discussion in Joseph and Janda (1989). Some of these arguments are anticipated in Ejerhed (1981), though stated in somewhat different terms, and in any case, we disagree with her conclusion that the aspirated stops of Sanskrit are to be treated as clusters of plain stops plus /h/ (as Zwicky [1965] did for the voiceless aspirates). In particular, the cluster analysis, as Ejerhed herself realizes, complicates the statement of certain cluster simplification rules and runs counter to the behavior of the aspirates in Sanskrit poetic metres. Furthermore, while she counts the reduplicative behavior of aspirates as favoring a cluster analysis—recall, as above, that there is generally no aspiration in reduplication—we note that there are some exceptional cases in which aspiration occurs in reduplication (e.g., Vedic bhari-bhr-, intensive stem of Vbhar- 'carry'), there are none involving clusters; similarly, while she treats word-final deaspiration as just a case of final cluster reduction, there again are exceptions with nonaspirates (e.g., Vedic (VS) ürk, NOM.SG of or/- 'nourishment*) but none with aspirates. There is a rather substantial literature on the phonetic nature of 'voiced aspirates' in modern Indo-Aryan languages (we note Yadav [1984] and Dixit [1987], for instance), and a considerable literature that has built productively on the assumption that voiced and voiceless aspirates are phonetically similar in relevant ways (we note Ahmed and Agrawal [1969], for instance, on perception confusion experiments with Hindi consonants in which the aspirates behave alike), though we acknowledge that the results have sometimes been contradictory as to the appropriateness of the traditional characterization of these sounds. For instance, Schiefer (1987, 1988) and Davis (1987) come to opposing conclusions as to whether or not the term 'aspiration* is appropriate in the description of

Sanskrit as She has been Misanalyzed Prosodically m 85 these Hindi sounds, and this despite access to native speakers and state-of-the-art instrumentation that obviously is not available for the sounds of ancient Sanskrit. Although we do not claim to be anywhere close to settling the controversies surrounding the phonetics of these sounds, we lean towards considering the voiced and voiceless aspirates as being indeed phonetically similar as that matches their phonological behavior in Sanskrit, a domain in which we do have the relevant evidence. Still, by way of suggesting why a judgment of phonetic similarity might be called for, we add here spectrograms (provided through the courtesy of Dr Shyam Agrawal of CEERI, in India, when he was a visitor at Ohio State University in 1987), of Hindi /t th d dh/, as they invite the interpretation of that each so-called aspirate indeed show similar period of 'glottal noise ... following the release of oral closure' (as Dixit [1987] describes the situation):

15. More generally, there is deaspiration before any obstruent, as Sanskrit has no clusters of the type [ghd] or [kKt] or [k h s] or the like. The reasons are varied but are to be sought (at least in part) in the diachrony that such clusters could have had. For instance, clusters with a voiced aspirate as leftmost member would have triggered the progressive assimilation of Bartholomae's Law (thus giving [ghdh] and ultimately [gdh]). Voiceless aspirates derive, historically speaking, primarily from combinations of voiceless unaspirated stops with a Proto-Indo-European 'laryngeal' consonant (usually symbolized as *H, but with no intended phonetic interpretation), especially the so-called 'second* or 'á-coloring' laryngeal; between consonants (thus, in clusters of, e.g., *-kHt-), these laryngeal consonants ultimately yielded (though most likely not directly so) a vowel (é in Indo-Iranian), so that *kHt clusters would have given *kit, and thus no consonant cluster. There are a few relevant forms—e.g., the Atharva Vedic 3SG present form grnatti 4 (s)he ties', presumably from the root Vgrath- 'tie'—but following the suggestion of Schindler

86 · Richard D. Janda and Brian D. Joseph

16.

17. 18.

19. 20.

21.

22.

23.

24.

25. 26.

(1976), we ignore it, even though it apparently shows root-final deaspiration (without throwback). Schindler feels that, since it is a nonce-form, unparalleled in the paradigm of this verb, it is likely to have been modelled on the form krnatti *(s)he spins', from the semantically related root Vkrt-. See Schindler (1976) for further discussion. If a voiceless aspirate is preceded by an unaspirated voiced stop (e.g., in a sequence /-g + th-/, the relevant assimilation rule in Sanskrit would be regressive, and so would yield [kth] as the outcome. Bartholomae's Law applied through/across -s- in these clusters, and the -s- was later deleted; see Schindler [1976] for discussion. That is to say, the present stem prccha- derives from Proto-Indo-European *prk'-sk'e-, with *-sk'e- as present-stem formative (found also in gaccha-, from *gwm-sk'e-, present stem to Vgam- 'go'. In synchronic terms, one would have to posit suppletion only in root forms from which grammatical stems are built, e.g., Vprcch- for the present and VpraSfor the future. Here, we again are ignoring AV grnatti—see note 15. There actually are two other forms of Vtrh- besides the future that are relevant here: the s-aorist and the desiderative formed with the suffix -sa-. Interestingly, however, all of these forms are ones that are not directly attested in texts but instead are cited by native grammarians (who sometimes constructed forms to fill out paradigms). One way of interpreting such forms with a root shape lark· is that, for the grammarians, this root simply was not on the list of aspiration throwback roots, just as in a similar way, the root they created for gutsa-, namely Vgudh, was not on the list and thus occasioned no aspiration throwback. Similar considerations hold for Vkrudh- 'be angry', for which the grammarians give forms with a deaspirated final such as future stem krotsya-; it too would not have been On the list' of aspiration throwback forms. This is essentially the way that Schindler (1976) deals with occasional nonce forms in which aspiration throwback does not occur (e.g., Vedic nominative singular mitradruk 'hostile to Mitra', versus expected (and actually attested) mitradhruk, from the root Vdruh'be hostile'). That is, Schindler argues that such examples show the sporadic effects of analogical influence from forms, elsewhere in the paradigm which lack initial aspiration, i.e., in cases where no deaspiration of the root-final stop would occur, such as the genitive singular -druh-as. The fact that these forms are onomatopoetic and/or expressive does not vitiate their value in showing that spellings with doubled aspirates were available, presumably when the phonetics actually warranted such a graphic representation. Many thanks to Hans Henrich Hock for bringing this form and the syllabification evidence to our attention. These facts about the syllabification of forms such as budhna- are also fatal for the analysis given in Broe (1991), which crucially depends on the counterfactual assumption that aspiration is not licensed in syllable codas. K&L were presumably working from the assumption that, since DhN- could occur in word-initial position—e.g., ghnanfi 'they strike'—then such a sequence must always be syllable-initial. However, as noted above, the prätifäkhyas are quite clear about the syllabification of medial clusters in Sanskrit. This observation and argument is due to Peter Lasersohn. The instrumental ending is considered in some treatments to behave like a separate word (hence the native grammatical tradition of calling these "pada endings—i.e., 'word' endings), even though it never occurs alone as a separate word. On such a view, the two contexts noted here would be collapsible as post-lexical, but only with a considerable degree of abstractness in the analysis.

Sanskrit as She has been Misanalyzed Prosodically · 87 27. The semantic parallel between this word and grapsa- is striking, and it may not be fortuitous if some sort of sound symbolism is responsible for linking at least the forms seen here with meanings that involve coalescence and a shared phonological pattern ofSTOP(r)-V-STOP-sa-. 28. The evidence from isolated forms such as drapsa-, grapsa-, and grtsa- (all mentioned above in the text) shows that the absence of aspiration throwback when -s- was the deaspirating trigger was the original development, so that dhuksa-, with aspiration thrown back, is the innovative form (presumably analogically based on forms in which aspiration throwback was not linked to deaspiration before -s-). The limited but detectable productivity evinced by aspiration throwback in this context—within which aspiration spread to a small number of additional roots—need not be taken to indicate that the process is phonological in nature. As we argue more fully below, morphological rules in a process morphology approach can likewise show productivity of varying degrees. 29. There is also the Vedic nonce-form dhak-tam (2DU.AOR.INJUNC.ACT of Vdagh4 reach'), which also shows aspiration throwback but displays exceptional behavior with regard to Bartholomae's Law (since one would expect **dagdham). We follow Schindler (1976: 635) in considering it to be modeled analogically on the 2SG.AOR.INJUNC.ACT form dhak, 'abetted by the occurrence of varktam in the same line [in the Rig Veda in which it occurs]'. 30. On this approach to morphology, see Anderson (1982,1992); Beard (1995); Janda (1983, 1987); Matthews (1972); and Stump (2001) among others. 31. In this way, the aspiration throwback phenomenon constitutes what we have elsewhere called a '(morphological) constellation', i.e., a group of elements which share at least one characteristic property of form but are distinguished by individual idiosyncrasies (of either form or of function) that prevent their being collapsed with one another. See Janda and Joseph (1986, 1989, and 1999); and Joseph and Janda (1988b) for discussion and further examples. 32. What we have in mind here is quite different from the approach of Colston (1996) (and even from its inverse); rather it is more in the direction of the non-morphemic, wordbased approach to morphology advocated by Singh and Agnihotri (1997). 33. To be fair to the autosegmentalists—and as indicated by our own discussion of (16) and (17), with its reference to 'automatic phonology'—we readily admit that there can be some aspects of deaspiration phenomena in Sanskrit which are not a matter of morphology. As Hans Henrich Hock has reminded us (see also Section 2.2 and references there), deaspiration automatically takes place when a consonant is 'doubled' (geminated), e.g., after r, as in /artha-/ 'goal' —> [arttha-], /dirgha-/ 'long' —> [diggha-]. Since the process in question is automatic and exceptionless, we see no conflict here between the deaspiration in gemination and the morphologically (even morpholexically) restricted deaspiration which remained in the aftermath of Grassmann's Law (for instance, in reduplication), or with another aspiration alternation—i.e., aspiration throwback—being morphological in nature. In Sanskrit, some aspects of the treatment of aspiration phenomena in general may indeed be automatic and thus better accounted for phonologically, but we submit that it is not the case that all such phenomena are best handled in that way.

• REFERENCES Ahmed, R. and S.S. Agarwal. 1969. Significant Features in the Perception of Hindi Consonants. JASA 48: 758-63. Allen, W.S. 1953. Plwnciics in ancient India. London: Oxford University Press.

88 · Richard D. Janda and Brian D. Joseph Anderson, S. 1982. Where's morphology? Linguistic Inquiry 13(4). 571-612. . 1992. A-morphous morphology. Cambridge: Cambridge University Press. Bare, J. 1980. Phonetics and phonology in Pänini. The System of Features Implicit in the Astädhyäyi. University of Michigan: Ph.D. dissertation. Beard, R. 1995. Lexeme-morpheme base morphology: A general theory of inflection and word formation. Albany: State University of New York Press. Benguerel, A. and T. Bhatia. 1980. Hindi stop consonants: An acoustic and fiberscopic study. Phonetica 37. 134-48. Blevins, J. 1995. The syllable in phonological theory. Handbook of phonological theory, ed. by J. Goldsmith, 1995. 206-44. Borowsky, T. and A. Mester. 1983. Aspiration to roots. Remarks on the Sanskrit diaspirates. Papers from the nineteenth regional meeting of the Chicago Linguistic Society, 52-63. Broe, M. 1991. Paradox lost: A non-derivational approach to Grassmann's Law. Declarative Perspectives on Phonology (Edinburgh working papers in cognitive science, Vol. 7) ed. by S. Bird, 1-19. Edinburgh: Centre for Cognitive Science, University of Edinburgh. Clements, G.N. 1985. The geometry of phonological features. Phonology 2. 225-52. Clements, G.N. and E. Hume. 1995. The internal organization of speech sounds. Handbook of phonological theory, ed. by J. Goldsmith, 1995. 245-306. Collinge, N. 1985. The laws of Indo-European. Amsterdam: John Benjamins Publishing Co. Davis, K. 1987. Phonetic cues to the Hindi voiced aspirates. Paper presented at the 9th South Asian language analysis roundtable (SALA 9) at Cornell University, 5 June 1987. Dixit, P. 1979. Aspiration: What is it and how is it produced? Journal of the Acoustical Society of America 65(1). 23. . 1987. In defense of the phonetic adequacy of the traditional term 4voiced aspirated*. Proceedings of the Xlth international congress of phonetic sciences (Tallinn) Vol. 1. 145-48. Ejerhed, Eva. 1981. The analysis of aspiration in Sanskrit phonology. Nordic Journal of Linguistics 4(2). 139-59. Fe*ry, C. 1996. The trochaic ideal. University of Tübingen, ms. Gamkrelidze, T. and V. Ivanov. 1972. LingvistiCeskaja tipologija i rekonstrukcija sistemy indoevropejskix smyCnyx. Conference on comparative-historical grammar of the IndoEuropean languages, ed. by S. BernStejn et al., 15-18. Moscow: AN SSSR. Goldsmith, J. 1990. Autosegmental and metrical phonology. Oxford: Blackwell. (ed.). 1995. Handbook of phonological theory. Oxford: Blackwell. Golston, C. 1996. Direct optimality theory: Representation as pure markedness. Language 72.713^8. Hoard, J., C. Cogen, H. Thompson, G. Thurgood, K. Whistler and J. Wright 1975. On incorporating Grassmann's Law into Sanskrit phonology. Proceedings of the first annual meeting of the Berkeley Linguistics Society, ed. by Cogen et al., 207-20. Berkeley: Department of Linguistics, University of California. Janda, R., J.E Richardson, M. Marks and A. Chukerman. 1983. 'Morphemes' aren't something that grows on trees: Morphology as more the phonology than the syntax of words. Papers from the Parasession on the interplay of phonology, morphology and syntax, ed. by Richardson et al., 79-95. Chicago: Chicago Linguistic Society. . 1987. On the motivation for an evolutionary typology of sound-structural rules. UCLA: Ph.D. dissertation. . 1998a. German umlaut: Morpholexical all the way down from OHG to NHG (Two Stützpunkte for romance metaphony). Rivista di Linguistica 10(1). 165-234. . 1998b. More than 2 points on umlaut in Modern Standard High German. Paper presented at annual meeting of the Linguistic Society of America (New York), January 1998.

Sanskrit as She has been Misanalyzed Prosodically m 89 Janda, R. and B. Joseph. 1986. One rule or many? Sanskrit reduplication as fragmented affixation. Studies on language change, Ohio State University Working Papers in Linguistics Vol. 34, ed. by B. Joseph, 84-107 (earlier version in ESCOL 1985. Papers from the second eastern states conference on linguistics. 1986, 103-19]. Columbus: The Ohio State University Department of Linguistics. . 1989. In Further Defense of a Non Phonological Account of Sanskrit Root-Initial Aspiration Alternations. Proceedings of the Fifth Eastern States Conference on Linguistics, edited by J. Powers and K. De Jong. Columbus: The Ohio State University Department of Linguistics. 246-60. . 1991. Meta-templates and the underlying (dis-)unity of reduplication in Sanskrit. ESCOL '91. Proceedings of the eighth eastern states conference on linguistics, ed. by G. Westphal, B.Ao and H.R. Chae. Columbus: The Ohio State University Department of Linguistics, 160-73. . 1999. The modern Greek negator ìç(õ)(-) as a morphological constellation. Greek linguistics: Proceedings of the 3rd international conference on Greek linguistics, ed. by G. Babiniotis, 341-51. Athens: Elinika Gramata. Joseph, B. and R. Janda. 1988a. On the unity of Sanskrit aspiration. Discussion papers for the sixth international phonology meeting and third international morphology meeting Vol. 1: Phonology. Wiener Linguistische Gazette 6. 29-31. . 1988b. The how and why of diachronic morphologization and demorphologization. Theoretical morphology: Approaches in modern linguistics, ed. by M. Hammond and M. Noonan, 193-210. San Diego, CA: Academic Press. Joseph, B. and R. Wallace. 1994. Proto-Indo-European voiced aspirates in Italic: A test for the glottalic theory. Historische Sprachforschung 107. 244-61. Kaye, J. and J. Lowenstamm. 1986. A non-linear treatment of Grassmann's Law. Proceedings ofNELS 16. 7985, ed. by S. Berman, J. Choe and J. McDonough, 220-33. Amherst: GLSA, Department of Linguistics, University of Massachusetts. Kenstowicz, M. 1993. Phonology in generative grammar. Oxford: B lackwell. Kiparsky, P. 1965. Phonological change. Cambridge, MA: MIT Ph.D. dissertation. . 1995. The phonological basis of sound change. Handbook of Phonological Theory, ed. by J. Goldsmith, 1995, 640-70. Ladefoged, P. 1975/1982/1993. A course in phonetics (lst/2nd/3rd editions). New York: Harcourt Brace. . 1976. The Stops of Owerri Igbo. Studies in African linguistics 6. 147-63. Lee, G. 1986. Diglossia in Ancient India. Studies on language change. Ohio State University working papers in linguistics 34 (1986): 151-64. Matthews, P. 1972. Inflectional morphology: A theoretical study based on aspects of Latin verb conjugation. Cambridge: Cambridge University Press. Marantz, A. 1982. Re reduplication. Linguistic Inquiry 13. 435-82. Mayrhofer, M. 1956 et seqq. Kurzgefasstes etymologisches W rterbuch des Altindischen. Heidelberg: Carl Winter. McCarthy, J. 1979. Formal problems in Semitic phonology and morphology. Cambridge, MA: MIT Ph.D. dissertation [published, 1985, by Garland Press, NY]. . 1981. A prosodic theory of noncatenative morphology. Linguistic inquiry 12. 373-418. McCarthy, J. and A. Prince. 1986. Prosodic morphology. University of Massachusetts/ Brandeis University, ms. . 1995. Prosodic morphology. Handbook of phonological theory, ed. by J. Goldsmith, 1995, 318-66. Ohala, J. 1995. Experimental phonology. Handbook of phonological theory, ed. by J. Goldsmith, 1995,713-22.

90 · Richard D. Janda and Brian D. Joseph Phelps, E. 1975. Sanskrit Diaspirates. Linguistic Inquiry 6. 447-64. Prince, A. and P. Smolensky. 1993. Optimality theory: Constraint interaction in generative grammar. Boulder: Rutgers University/University of Colorado, ms. Sag, I. 1974. The Grassmann's Law ordering pseudo-paradox. Linguistic Inquiry 5. 591-607. . 1976. Pseudosolutions to the pseudoparadox: Sanskrit diaspirates revisited. Linguistic Inquiry 1. 609-22. Schiefer, L. 1987. Perceptual and acoustic correlates of Hindi breathy voiced stops. Paper presented at the 9th South Asian language analysis roundlable (SALA 9) at Cornell University, 5 June 1987. . 1988. * Voiced aspirated* or 'breathy voiced* and the case for articulatory phonology. Paper presented at the sixth international phonology, Krems (Austria), July 1988. Schindler, J. 1976. Diachronie and synchronic remarks on Bartholomae's and Grassmann's Law. Linguistic Inquiry 7. 622-37. Singh, R. and R. Agnihotri. 1997. Hindi morphology: A word-based description. Delhi: Motilal Banarsidass. Stemberger, J. 1980. Another look at Grass mann's Law. Glossa 14. 113-35. Stump, G. 2001. Inflectional morphology: A theory of paradigm. Cambridge: Cambridge University Press. Steriade, D. 1995. Underspecification and markedness. Handbook of phonological theory, ed. by J. Goldsmith, 114-74. Vaux, B. 1992. Gemination and syllabic integrity in Sanskrit. Journal of Indo-European Studies 20(3-^). 283-303. Whitney, W. 1889. Sanskrit grammar. Cambridge: Harvard University Press. Yadav, R. 1984. Voicing and aspiration in Maithili: A fiberoptic and acoustic study. Indian Linguistics 45(1-4). 1-30. Zwicky, A. 1965. Topics in Sanskrit phonology. Cambridge, MA: MIT Ph.D. dissertation.

Open Submissions

The Nepali Converbs: A Holistic Approach* ι JOHN PETERSON

The object of the present study is to examine the form, meaning and scopal behavior of the various converbs in Nepali, the national language of the Kingdom of Nepal. As the title suggests, the converbs are described from a number of different perspectives, for example, morphosy n tactic, semantic and discourse-pragmatic, and a new classification of these forms on the basis of these factors is proposed.

• 1. Introduction In recent years, an increasing number of publications has appeared on the somewhat ellusive category of 'converbs', which are generally considered to be a cross-linguistically valid category, although it has proven immensely * This study is based largely on an analysis of the data which I gathered during two stays in Kathmandu, the capital of Nepal. I would like to take this opportunity to express my gratitude to the Swiss National Science Foundation for a generous grant which made the first trip possible. I would also like to thank Mr Bishnu Bhandari, Mr Deb Kumar Chhetri, Mr Madan Giri, Mr Dipak Pahari, Dr Manes Raj Pant, Miss Mimamsa Pant and Mr Ram Raj Lohani for their kind help and sheer endless patience in answering my questions and explaining the intricacies of their native language to me. Additional data from written works are also included (cf. Primary Sources in the literature). A final word of thanks goes to Balthasar Bickel, Ellen Brandner, Karen Ebert, Claudia Gerstner-Link, Johanna Mattissen and Ulrike Mosel for their insightful comments and suggestions on earlier versions of this paper, and to Yogendra Yadava for all his help in Nepal. Of course, I alone am responsible for all errors and misconceptions which may have found their way into the paper.

94 · John Peterson

difficult to pin down just what the defining characteristics of this category in fact are. Converbs have also been studied in detail in South Asian languages for many years mainly under the names Conjunctive participles' (sequential converbs) and 'imperfective' or 'present participles' (imperfective converbs) and many insights have been gained on their behavior and meaning. One such study is devoted to the sequential convert) in Hindi and proposes a largely pragmatic approach to the topic (Davison 1981). Such a view has also been supported in a recent review (Bickel 1998b). Also worth mentioning here is Bickel (1993), which takes a close look at these factors for subordination in general in Belhare (Tibeto-Burman, Nepal). It is in this spirit that the present study was conducted. After presenting the relevant forms and their functions in Nepali, this paper will offer a new classification of the Nepali converbs on the basis of their morphosyntactic, semantic and pragmatic properties. The study is organized as follows: Section 2 presents a brief overview of the Nepali verb system. In Section 3,1 provide a working definition for the concept of 'converbs', based primarily on Nedjalkov (1995). Section 4 provides an overview of the various forms and functions of the Nepali converbs. In Section 5,1 present a detailed look at the scopal properties of various operators when a converbal form connects two different cores or clauses. In order to better understand how each of the converbs fits into a system of oppositions, I take a detailed look in Section 6 at what I believe are the two defining characteristics of the Nepali sequential converbs—their use to describe two discrete parts of what is perceived to be one single, larger situation (Section 6.1) and perfective aspect (Section 6.2), at least when the con verb and the main predicate refer to two different actions. On the basis of this analysis, I then contrast this category with the remaining two converbal categories and propose a new classification of the Nepali converbs in Section 7. Section 8 consists of the conclusions.

• 2. A Brief Overview of the Nepali Verb System1 Nepali, which belongs to the Indo-Aryan branch of Indo-European, is a predominantly verb-final language with relatively free word order. In the indicative, the intransitive core argument (S) appears in the absolutive case. The marking of the transitive agent (A) and the transitive patient (O) is not quite as straight-forward: In the simple past, the perfective, the perfect and with nonfinite verbs, A is obligatorily marked by the ergative, and optionally in other categories as well to mark an unexpected agent (cf. Bickel 1998a). Otherwise A appears in the absolutive. Ο appears in the absolutive, unless it is human, in which case it is marked by the objective marker -laL Thus, this

The Nepali Converts: A Holistic Approach m 95 often produces a tripartite marking system, with A, S and O all receiving different marking. Regardless of the marking, the verb always agrees in the indicative with A and S. Hence, I will refer to these two core arguments in the following as the 'subject' of the clause. Diagram 1 gives a partial representation of the Nepali aspectual system. Diagram 1: The Nepali Aspectual System (Simplified) Perfective Imperfective

II-E-II

Continuous //-Mdai-//

Noncontinuous //-MDA-//

The perfective marker//-E-// is realized as -e- or -/- before consonants and ~ybefore vowels and diphthongs. There is a further distinction in the imperfective between the continuous and the noncontinuous. The noncontinuous imperfective is marked by //-MDA//, realized as -0- after consonants, -HI- (nasalization) after diphthongs and -nin conjunction with vowels. However, some speakers/writers prefer the 'full' forms -da-, after verbal stems which end in a consonant, and -mda-, found elsewhere.2 The continuous imperfective is formed by the focused form of the fullform of the imperfective marker, //-Mdai-//, which has the same distribution as the full-forms of the noncontinuous marker.3 • 2.1. Nonfinite Forms Many nonfmite forms, such as the converbs, which we will be dealing with shortly, have the three-way aspectual split shown in Diagram 1. With other nonfmite verb forms there is only a binary aspectual split between perfective and imperfective. Two examples are the so-called 'infinitival' participle in -ne and the 'perfect' participle in -eko which, as Genetti (1992: 416ff) notes, are imperfective and perfective participial endings, respectively. I analyze -ne as -n-e (-IPFV-ATT) and -eko as-e-ko (-PFV-ATT).4 Some examples (adapted from Genetti 1992: 416f): (1) mai-le bihδ gar-e-ko belδ (Genetti's example [34]) IS-ERG wedding do-PFV-ATT time The time when I married' (2) pδtan jδ-n-e bδto (Genetti's example [35]) Patan go-IPFV-ATT road The road which goes to Patan'

96 · John Peterson

m 2.2. Finite Forms With finite verb forms the situation is slightly more complex: When the perfective marker -e is followed by the past-tense marker, the category is also perfective. For example, the past perfective ga-e-thyo (go-PFV-PT.3S.NF) 4he went'. On the other hand, the interpretation of the category in which the perfective marker directly combines with person-marking in a portmanteau morpheme (cf. Wallace 1982: 78ff and Peterson 1999: 342f) is an aspectually neutral simple past. For example, ga-eqi (go-SPT.lS) Ί went'.5 With non-negated finite forms, the imperfective shows a continuous/ noncontinuous distinction. With negated finite forms, this distinction is neutralized. Finally, sentential negation is achieved by what Payne (1985: 226ff) refers to as 'morphological' negation, marked by -na-, if we extend his definition to include not only derivational but also inflectional morphology. With most finite verb forms, this marking occurs AFTER aspect marking and BEFORE tense and person-marking. With some finite forms, however, and with all nonfinite forms, na- is prefixed to the verbal root. Schematically, we can represent the maximal possible structure of the finite verb in Nepali as in Table 1. For completeness, other categories which will appear in the examples (e.g., passive and actionality marking) have been added here. The names refer to the positions or 'slots' of the markers which denote the corresponding categories on the finite verb. The first -/- represents one of two Fugenelements which are found exclusively in the so-called 'compoundverb' constructions (see Sections 4.2 and 4.3), which generally denote actionality. The second -/- is found only in negation. Negation is located at either of the two positions indicated, but never at both simultaneously. Finally, gender, person, number, mood and honorific status (HON) are usually represented by a portmanteau morpheme. Table 1: A Schematic Representation of the Nepali Finite Verb NEC-VERB STEM-CAUS-/-ACTIONALITY-PASS-ASPECT-/-NEG-TENSE-GENDER/ PERS/NUM/MOOD/HON

• 3. Converbs Haspelmath (1995: 3) defines the converb as 'a nonfinite verb form whose main function is to mark adverbial subordination. Another way of putting it is that converbs are verbal adverbs, just like participles are verbal adjectives' (emphasis in original).

The Nepali Converbs: A Holistic Approach m 97 For Nedjalkov, nonfiniteness is not an essential criterion. He writes: At first approximation, we can define a convert) as a verb form which depends syntactically on another verb form, but is not its syntactic actant, i.e., does not realize its semantic valencies. Thus, a canonical (i.e., noncombined) convert) can occupy (1) the position of an adjunct, i.e., an adverbial, but cannot occupy the positions; (2) of the only predicate of a simple sentence (without additional auxiliary elements); (3) of nominal attributes; (4) of a clausal actant (i.e., it cannot depend on verbs such as begin, order, etc.); (5) of a nominal actant (i.e., it does not occur in subject and object position) ...(1995: 97). This serves to differentiate the (canonical) convert) from the following other canonical forms: finite forms (predicate of a simple sentence), participles (nominal attributes), infinitives (clausal actants) and gerunds (nominal actants). Although fmiteness is of central importance in the discussion of the crosslinguistic validity of converbs in general (cf., e.g., Ebert 1993a), we can leave this discussion aside here as all converbs in Nepali are clearly nonfmite in form (i.e., they lack both tense and person-marking) and function (i.e., they cannot serve as the main verb of a sentence). Finally, I will further restrict this definition of converbs, at least for the present discussion, so that the convert) consists of the ENTIRE VERB FORM. Combined with the first half of Nedjalkov's second criterion above this means that converbs are restricted to the main (nonfmite) verb of a SUBORDINATE CORE or CLAUSE ONLY. Thus, unlike Nedjalkov's point (2) which allows for converbal forms in the main predicate in the presence of 'additional auxiliary elements', that is, in periphrastic forms, I will exclude this possibility: If a verb form contains a form which is homophonous with the converbal marker and also contains additional auxiliary elements, I will consistently refer to the marker which is homophonous with the converbal marker as a Fugenelement in the present study. For example, consider the suffix -ι in the verb form gar-i-di-eqi. (do-0BEN-PT.1S) Ί did [something for someone]'. This form is homophonous with one of the sequential converbal markers. However, as it is used verbinternally, i.e., gar-i-di-etfi. is a single verb form, it will not be considered a converbal marker in this paper but rather a Fugenelement which derives from, and is of course homophonous with, a converbal marker (cf. the discussion of the 'compound-verb' constructions in Sections 4.2 and 4.3). That is, for present purposes I will not consider these forms converbal markers. Converbal markers in this study are only those which mark the ENTIRE PREDICATE as being a subordinate verb form.6

98 · John Peterson

There is still one formal aspect which must be discussed here: Nonfinite verb forms in Nepali are often followed by some kind of marker (suffix, postposition, conjunction, etc.) which serves to connect them both semantically and syntactically to the rest of the sentence. One example in Nepali is verbs marked by the perfective suffix -e but unmarked for person and tense (i.e., they are nonfinite). A verb base marked for perfective aspect by the suffix -e can appear without a following suffix or focal particle and serves to express the protasis of a probable, hypothetical conditional sentence. Hence, it is referred to as a 'conditional con verb' by Ebert (1993a: 102). The same form can also be followed by a number of postpositions, conjunctions and pragmatic particles. Some examples are found in Table 2. Table 2: Some Nonfinite Uses of the Nepali Perfective Marker -e Verb Stem + -e + pani la pachi -...'Samma ra

'also* =$ topic marker 'after' 'until' (NEC-... as.long.as) 'and*

conditional-concessive clause counterfacrual clause temporal clause temporal clause 'sequential convert)' (cf. Table 3)

Indeed, not only -e but also the imperfective converbs in //-Mdai// and //- Mdδ// (cf . Section 4.2) can be used in conjunction with a number of different postpositions or pragmatic particles. Considering all of these forms to be converbs would result in a whole plethora of con verbal forms in the language but, as all of these are based on the suffixes -e, //-Mdai// and //-Mdδ//, this hardly seems desirable, as it would be missing a vital point, namely that all of these 'converbs' are in fact based on the SAME THREE FORMS. And yet there is a general consensus that the form ending in -e-ra (-PFV-and) is a convert). In this work I will only consider the following verbal forms to be converbs: A verbal stem, followed either by TAM-marking or a morphologically opaque ending, which meets the criteria proposed by Nedjalkov (1995: 97), with the further restrictions as discussed earlier. If this form is followed by a further subordinating marker (postposition, conjunction, etc.), the verbal form minus this subordinating marker will be considered the converbal form. Hence, in the examples gar-e-pachi (do-PFV-after) 'after having done' and na-gar-e-samma (NEG-do-PFV-as.long.as) 'until doing', only the form (na·) gar-e will be considered the convert), to which various postpositions may be added. There is still one problem with this approach. As noted earlier, there is a general consensus that the suffix -e-ra (-PFV-and) is a converbal suffix, despite its morphological transparency. In spite of what was said earlier, there is in fact good reason for treating this suffix as a converbal marker.

The Nepali Converbs: A Holistic Approach · 99

First of all, as will be shown in Section 6.2, the meaning of the form in -era is not always interpretable as a combination of perfective (-e-) and 'and' (ra). Hence, it is often most appropriate to treat this as a simple converbal suffix (i.e., -era). Perhaps the most important argument favoring the interpretation of -era as a simple converbal ending is the fact that it can ALWAYS be replaced by the converbal ending -Γ, with which it freely alternates. It is only logical that, if -/" is considered a converbal ending, which seems to be universally the case, then the ending -era, which fulfills virtually the same function, must be one as well.7 Note that considering a verbal form which ends in -Fto be a convert) is consistent with the definition of a con verb we have chosen here, as it fulfills the criteria discussed earlier and is also not further analyzable morphologically. Hence, any other form which fulfills the same function as this must also be a convert). In summary: All verb forms which meet the criteria proposed by Nedjalkov as discussed here will be considered converbs if their morphology is purely 'verbal' (i.e., ΤΑΜ marking as well as those forms whose ending serves no other function). Those forms which are composed of such a verb form plus some kind of subordinating marker will only be included under the concept 'convert)' in this paper if they freely alternate with other forms which are clearly converbal endings. Where a converbal form according to this definition is followed by a subordinating marker, only the form without the subordinating marker will be considered the convert).

• 4. The Nepali Converbs Generally speaking, using the semantic classification proposed by Nedjalkov (1995: 106-10), all Nepali converbs can be considered as belonging to the contextual type, although the sequential and the imperfective converbs partially overlap with the narrative use. This is illustrated in Table 3. Table 3: An Overview of the Nepali Converbs Contextual

Narrative

Perfective (-e) Imperfective (-//-Itfda//, -//-tfdai//) Sequential (-era, º, -Tkana)

Table 3 lists the Nepali converbs in more-or-less standard fashion. Although the names for the individual categories can vary considerably, it seems that this three-way opposition is generally at least implicitly assumed, if the perfective converb is taken into consideration at all.

100 · John Peterson

The problem with this classification, however, is that it tells us virtually nothing about the types of oppositions which exist between the various categories. In fact, it suggests that there is merely a simple, three-way opposition between the categories8 and it also covers over many similarities, such as the aspectual similarities between the perfective and the sequential categories, as well as certain functional similarities between the sequential and the imperfective categories, as I will show in the following pages. In order to understand these similarities and differences, a closer look at the functions and meanings of the individual categories is first needed. After discussing these converbs in more detail—especially the so-called 'sequential' converbs—in the following sections, I will provide a new classification in Section 7 which takes these morphosyntactic, semantic and pragmatic factors into consideration. • 4.1. The Perfective Convert) This form, referred to by Matthews (1992) as the 'second perfect participle', functions as a conditional convert) when it is not followed by a subordinating marker. Consider the form na-gar-e in the following example: (3) δphu-le jόthocόlo na-gar-e REFL-ERG washing of pots and utensils NEG-do-PFV bhok-ai bas-nu-par-0-cha. hunger-FOC remain-INF-OBL-NCT-NPT.3S 'If he doesn't wash up himself, he'll have to remain hungry' (GM: 69). At first glance, this seems to be an example of what Nedjalkov (1995: 126ff) refers to as a 'futurate' convert): A con verb which only appears in sentences whose main predicate is in the future, the imperative, the subjunctive or a similar category. In these environments we consistently find a conditional interpretation. However, when this convert) is followed by the additive focal particle pani 'also', this restriction to a 'futurate' category no longer holds. In those cases where the main predicate does appear in a 'futurate' category, what results is a conditional-concessive, as in the following example: (4) kehl rnsδpat di-na par-e pani ma taiyδr chu. some credit give-INF OBL-PFV also IS ready COP:NPT.1S 'Even if I have to give [you] credit, I am ready [to do so]' (LS: 12).

The Nepali Converbs: A Holistic Approach m 101 If however the main predicate appears in a category where the status of the main clause as factual may not be called into question, for example, the simple past, the interpretation is that of a simple concessive. (5) a. bδhun-le mδlMδi jhόto mantra brahmin-ERG gardner-OBJ false mantra sik-a-e-ko bha-e9 pani learn-CAUS-PFV-ATT AUX-PFValso b. mδli-le cokho man-le pόjδ gar-yo gardner-ERG pure mind-INST worship do-SPT.3S.NF ra ό svarga pug-yo. and 3S heaven reach-SPT.3S.NF 'Even though the Brahman had taught the gardner a false mantra, the gardner worshipped with a pure mind and he reached heaven' (TO: 143). As mentioned here, the same converbal form also appears with various postpositions, such aspachi 'after': (6) a. kδt-n-e manche hatiyδr li-era aghi sar-e-pachi cut-IPFV-ATT man weapon take-SEQ forth move-PFV-after b. thito-le δphό-lδi kδt-na δu-n-e iad-ERG REFL-OBJ cut-INF come-IPFV-ATT manche-ko mukh-ma her-yo. man-GEN face-LOC look-SPT.3S.NF 'After the executioner (= 'the man who cuts') took his weapon and came forward, the lad looked into the face of the man who was coming to behead (= 'cut') him(self)' (TD: 56f). As there is no generally agreed-upon designation for this convert), which is certainly NOT a participle from a synchronic point of view, as it can never be used in attributive function, I will refer to it for now as the perfective converb, on the basis of its formal marking. • 4.2. The Imperfective Converbs The imperfective converbs are marked by forms which are based on the imperfective marker //-Mda//. These forms are referred to as 'imperfect participles' by Matthews (1992).

102 · John Peterson

I analyze the two imperfective converbal endings as the continuous ending //-Mdai// and the noncontinuous ending //-Mdδ//. The two are in the same kind of relationship to each other as are the two imperfective aspect markers in the finite categories discussed in Section 2, i.e., the continuous converbal marker //-Mdai// is the focused form of the noncontinuous converbal marker //-Mdδ//, which itself is the oblique (and plural) form of the imperfective participle marker //-Mdo//. The form in //-Mdo// was undoubtedly once quite common in Nepali10 but nowadays it is restricted to its use in periphrastic categories such as the progressive mirative and has lexicalized in a few expressions such as cäkhlägdo 'interesting, tasty', from cäkh 'taste, interest' and the emotive verb lag- (cf. Matthews 1992: 226f). In fact, all of the imperfective markers, both in the finite categories as well as in the converbs, originally derive from this form, the modern-day reflex of the Old Indo-Aryan 'present participle' ending -ant or the cognate Middle Indo-Aryan form in -anta·. Slater (1994: 154f) notes that the con verb marked by //-Mdai// always implies complete simultaneity, which is compatible with my interpretation of this form as the continuous imperfective converbal marker. An example: (7) ü susel-dai bδhira nisk-yo re. 3S whistle-CNT out emerge-SPT.3S.NF EVID 'They say that he went out whistling.' On the other hand, Slater notes on the same page and elsewhere (1994: 142f) that the use of the converbal form in //-Mdδ// does not require total simultaneity, which is compatible with my interpretation of this form as the noncontinuous converbal marker. In fact, examples can be found that show that simultaneity may not be present at all with the noncontinuous form, as in the following example, where the noncontinuous converb is used sequentially: (8) jnδncandra-kδ yastδ kurδ sun-dä ghar-ko mδlik-le J.-GEN.P such.P speech.? hear-NCT home-GEN owner-ERG un-lai δdar-pδrvak bas-δl-δ bhan-yo... 3S-OBJ respect-full sit.down-CAUS-SEQ say-SPT.3S.NF On hearing Jnδncandra's words, the owner of the house respectfully seated him and said...' (LS: 70). The difference between the use of the noncontinuous converb in this function and the sequential converb, which we will be dealing with in the next section, appears to be that the action denoted by sun-dä 'hearing' is portrayed here as

The Nepali Converbs: A Holistic Approach m 103 being extended in time, although it is also clearly bounded in the present example (cf., also Peterson 1999: 359). Other interpretations, such as adverbial of manner in (9), may result as well. Also, as (9) and (12) show, both imperfective converbs may also appear together when the convert) is reduplicated. On this use of //-Mdδ// and //-IVfdai// occurring together, cf. Slater (1994: 156-60).n Both imperfective converbal forms may also be followed by postpositions/suffixes ([10] and [II]),12 or an additive focal particle (pani in [12]), which further specify their appropriate interpretation. (9) yas-ai gar-dä-gar-dai gvδmge-le sab-ai mδsu thus-FOC do-NCNT-do-CNT G.-ERG all-FOCmeat kukkur-lai khvδ-i-siddhyδ-yo. dog-OBJ feed-0-TEL-SPT.3S.NF 'By doing thus, Gvδmge fed all the meat to the dog' (LS: 6). (10)... pahile lδt-le hän-dä-samma ta gaumthali kehi first kick-INST beat-NCNT-as.long.as TOPG. anything bol-e-ki thi-i-na... say-PFV-ATT.F AUX-PFV.F-NEG.3S13 '... up to [him] kicking [her] the first time, Gaumthali didn't say anything../ (GM: 65). (11) khasro ciurδ capäu-mdä ra nun-ko dhiko coarse parched.ricehusks chew-NCNT and salt-GEN lump tok-tai-mä dosro gundδ-ko pani deharδ bite-CNT-LOC second rascal-GEN also allotted.time bit-yo. pass-SPT.3S.NF * While chewing the coarse ricehusks and biting the lump of salt, the second rascal's allotted time also passed' (BA: 5If). (12) yastl rδmri tarum svδsni ghar-ma such:Fgood: F young:F wife:Fhome-LOC cha-mdä cha-mdai pani tyo mahajan COP-NCNT COP-CNTalso that merchant barδbar randtbδji gar-da-thyo. constantly visiting.prostitutes do-NCNT-PT.3S.NF 4

Although he had such a good young wife at home, that merchant was constantly visiting prostitutes' (BA: 49).

104 · John Peterson

The use of the imperfective con verbs in clause-chaining is seldom but possible, as in the following elicited example: (13) bδto-mδ susel-dai saharjä-mdai ra δphno kam-ko road-LOC whistle-CNT city go-CNT and own work-GEN bδremδ soc-dai mai-le un-lai bhet-em about think-CNT IS-ERG 3S.MGH-OBJ meet-SPT.lS 'While whistling on the road, going to the city and thinking about my own work, I met him/her.' Finally, the continuous converb in //-Mdai// is homophonous with the Fugenelement in a kind of compound verb. In this construction, the lexical verb (or ) is marked by the form //-Mdai// and is followed by one of the three markers äu-, ja- or raha-. These are generally referred to as ' V2' or 'vector verbs', since they derive from lexical verbs. As lexical verbs they have the meanings 'come', 'go' and 'remain', respectively. Although VI and V2 are written separately in this construction, they are best considered two components of one single verb. An example (adapted from Pokharel 1999: 194, ex. 14): (14) ahile-samma ma pathäu-mdai a-e-ko chu now-as.long.as IS Vl:teach-0 V2:VEN-PFV-ATT AUX:NPT.LS have been teaching till now.' The use of the V2 in this construction together with the lexical verb marked by //-Mdai// denotes both that the action is continuous over a period of time as well as the fact that the action in this case has continued up to the speechact situation. Following Bickel (1996), I will refer to it as a ventive marker. However, as this use of the ending //-Mdai// as a Fugenelement does not correspond to the definition of 'converb' that I have chosen to use, since it is internal to the main predicate of the sentence, this construction will not be dealt with further. • 4.3. The Sequential Converbs The sequential converts, traditionally referred to as 'conjunctive participles', are marked by either -era, which as we have seen derives from -e-ra 'PFVand', or the opaque older form - , which is especially common in the written language. There is also the focused form in -ikana, where - - is the older converbal ending and -kana is the former objective postposition. However, as

The Nepali Converbs: A Holistic Approach m 105 -kana is no longer a productive postposition, I will simply gloss both 'suffixes' together as 'SEQ.FOC. The form in -Ikana is considered 'more emphatic* than the other two (cf. Matthews 1992: 116) and, as with negation in Nepali (and elsewhere) in general, the focused form is especially common when the converbal clause is negated. As we shall see in the following pages, the designation of these converbs as 'sequential' is not without problems. However, for the moment I will retain this designation until their function has been dealt with in more detail. In Section 7, an alternative classification will be proposed. The sequential converbs have three main uses: • In connecting two cores or clauses, both of which denote different actions. This is probably the most common use of the converbs and may be considered their 'core' function. As an example, consider the form li-era in (6a) earlier. Although one seldom finds more than one sequential convert) used in one sentence, the sequential converbs also allow for narrative-chaining. The largest number of examples I have been able to locate to date in a non-elicited text has four sequential converbs in -f (in 15b, c [twice] and d) in addition to the main predicate (in 15e): (15) a. δphu-sanga bha-e-kδ varadδ ra moksadδ REFL-with COP-PFV-ATT:P Varadδ and Moksadδ bhan-n-e din sakti say-IPFV-ATT two goddess b. phόlcokl ra dhyδnacco giri-mδ sthδpnδ garPhόlcoki and Dhyδnacco mountain-LOC placement do-SEQ c. phόlcoki-bδta eklδkh phόl-harό lyδPhόlcoki-ABL 100,000 flower-P bring-SEQ jyoti-lδi carh-δlight-OBJ proceed-CAUS-SEQ d. blc-ko bhδg-lδi δphno khadga-le kδtmiddle-GEN part-OBJ own sword-INST cut-SEQ e. nepδl khδldo-ko pδni jammai bδhira Nepal valley-GEN water completely:FOC out pathδu-n-e vicδr gar-e. send-IPFV-ATT thought do-SPT.3S.NF.HGH

106 · John Peterson

'He (HGH) considered placing the two goddesses called Varada and Moksadδ, who were with him, on the mountains Phulcokl and Dhyδnacco, bringing 100,000 flowers from Phulcokl and offering them up to the light, and cutting the middle part [of the mountain] with his sword and sending all the water of the Nepal valley out [of the valley]' (LM: 40). Unlike the perfective and imperfective converbs, the sequential converbs may not be followed by a postposition, although they can be followed by either an additive or restrictive focal particle: (16) timl-le bhan-era mδtrai yo kam gar-em 2S-ERG say-SEQ only this work do-SPT.lS did this only because you told me' (adapted from Bickel 1998b: 394). We also find the sequential con verb used in the so-called 'tail-head linkage', also referred to as 'recapitulation' (cf., e.g., Grain 1992:43ff). Here, the finite verb of one sentence is repeated in the next sentence, where it is marked as a sequential converb. An example: (17) a. kisδn ... rδti bhδt khδ- ... sut-yo. farmer night dinner eat-SEQ go.to.sleep-SPT.3S.NF b. ekchin sut-eral4...ό ...kh[e]t-mδ her-na δ-yo. 1 momentsleep-SEQ3S field-LOC look-INFcome-SPT.3S.NF 'The farmer... ate dinner in the evening and went to sleep. After he had slept for a moment,... he ... went to the field to take a look' (TD: 461f). • To form adverbials from verbs. Here, there is only one action denoted by the sentence and the 'converb' merely functions to further modify this action as an adverbial of manner. This will be dealt with in detail in Section 6.2. A few examples: (18) hos gar-era 'carefully' kara-era 'loudly' khelcare do-SEQ shout-SEQ play-SEQ 'playfully' hδms'happily' laugh-SEQ

The Nepali Converbs: A Holistic Approach m 107

• The form -f is also homophonous with the verb-binding element found in what is generally referred to as the 'compound-verb construction'.15 An example: (19) mai-le gar-i-di-em IS-ERG Vl:do-0-V2:BEN-SPT.lS *I did [it] (for someone else's benefit).' Here, VI is the lexical verb (in this case gar- 'do') while V2 is one of approximately 25 markers which generally denote Aktionsart and may also convey some additional meaning as well, such as 'benefactive', etc.16 The V2 di- in example (19), which is homophonous with the lexical verb meaning 'give', denotes that the action was performed on someone else's behalf, although, depending on the meaning of the lexical verb or predicate, it can also simply express telicity (cf. Pokharel 1999: 204).l7 The term 'VI' in this construction is somewhat misleading, as there can be more than one lexical verb used. In these cases, only the last 'VI' is written together with the 'V2\ An example with three lexical verbs: (20) tara, sab-ai-le kh -ϊ pi-ϊ sut-i-sak-e-pachi but all-FOC-ERG eat-0 drink-0 sleep-0-TEL-PFV-after pani tini-har kur gar-dai thi-e. also 3-P talk do-CNT AUX:PT-3P 'But even after all had finished eating, drinking and had already fallen asleep, they [= the other two] were [still] talking' (VK: 62). Also, as Pokharel (1999: 197, ex. 31 and 202) notes, more than one 'V2' can be used in this construction, hence the expression 4 V2' is also somewhat inappropriate. An example for the use of two ' V2s' from my own data: (21) un-le kican saph gar-i-di-i-hal-in. 3S.MGH-ERG kitchen clean do-0-BEN-0-IMMED-SPT.3S.F.MGH 'She (MGH) cleaned the kitchen for me right away.' Note that ΤΑΜ marking occurs only once in this construction, after the final 4 V2'. Verbs marked by a *V2' can also appear as sequential converbs (22), perfective converbs (cf. sut-i-sak-e in [20]) or imperfective converbs (23):

108 · John Peterson

(22) bhδt khδ-i-sak-era ό sut-na ga-yo. dinner eat-0-TEL-SEQ 3S sleep-INF go-SPT.3S.NF 'Having finished eating dinner, he went to sleep.' (23) tapδim-le khδ-i-di-mdδ 2s.HGH-ERG eat-0-BEN-CNT merδ bδbu-le pδ-e. my.HGH father-ERG get-SPT.3S.NF.HGH 'By your eating this [for him], my father (HGH) [in heaven] received it' (LS: 26). A Fugenelement with the form -era or -ikana is not found. Although this verb-binding form is homophonous with, and diachronically related to, the con verbal suffix in - , like the use of//-Mdai// in (14) this use in the formation of a complex verb does not constitute a con verb according to the definition given in Section 3 and will not be further discussed. Also, examples such as (22), where the verb form contains both this Fugenelement and the sequential converbal marker, clearly demonstrates the different status of these two markers.

• 5. Operator Scope Having presented the most important morphosyntactic properties of the various converbs, we will now take a look at their scopal properties. In the following, I will follow the terminology of Role and Reference Grammar (RRG) in discussing operator scope and nexus relations (cf. Van Valin and LaPolla 1997 [especially Chapters 2 and 8]). Thus, core-level operators are directionals, deontic modality and internal negation, while clause-level operators include epistemic modality and external negation ('status'), tense, evidentials and illocutionary force. A 'core' is defined as a predicate and its core arguments, while a 'clause' is a core and its periphery.18 I will procede as follows. In Section 5.1, junctures involving the perfective con verb will be dealt with. Section 5.2 is dedicated to scope in junctures involving sequential converbs. We will begin with same-subject (or 'core-level') junctures involving the sequential converbs. This will be followed by a discussion of operator scope in different-subject (or 'clause-level') junctures involving the sequential converbs.19 However, as tense is somewhat more complex, it will be dealt with separately after different-subject junctures. Section 5.3

The Nepali Converbs: A Holistic Approach m 109 then takes a brief look at scope involving the imperfective converbs. The findings are summarized in Section 5.4. • 5.1. Junctures with the Perfective Converb Junctures of this type only allow one of the cores or clauses to be in focus at a given time. That is, both cores/clauses may never be in focus simultaneously. This potential scopal pattern holds in all cases in which the perfective convert) is used, whether or not there is subject-identity (i.e., core-level vs. clauselevel junctures) or whether or not the perfective convert) is followed by a postposition. An example (for convenience, that part of the sentence which is within the scope of the respective operator will be underlined in the translation): (24) phalphul tach-e-pachi nam-lal di-oni? fruit peel-PFV-after little.girl-OBJ give-INJ.lS 4

Should I give the child the fruit after I peel it (i.e., not before)?' 'After I peel the fruit, should I give it to the child?'

However, it cannot mean the following: *'Should I peel the fruit and, after that, should I give it to the child?' As we shall see in the next section, this is quite different from the potential focus domain of the sequential converbs. • 5.2. Junctures with the Sequential Converbs Let us now take a look at scopal patterns in junctures with the sequential converbs.

5.2.2. Same-subject or 'Core-level' Junctures When both cores have the same subject, the scope of core- and clause-level operators is ambiguous. It may extend over either core or both simultaneously. Hence, depending on the situation, (25) can have any one of the following three interpretations: (25) ma ghar ga-era bhδt khδ-na bhyδu-m-chu. IS home go-SEQdinnereat-INFbe.able-NCT-NPT.IS a. 'After I go home, I can eat dinner/ b. *I can go home to eat dinner.' c. can go home and eat dinner.'

110 · John Peterson

Which interpretation is intended depends on how much information is considered pragmatically presupposed or accomodated when the speaker utters it.20 For example, if the speaker has just announced that s/he is going home, only the main clause can be considered to be in focus. Hence, the scope of deontic modality is restricted to the main clause (a). On the other hand, if the speaker has announced that s/he will eat dinner—hence, the main clause is presupposed—the question can only be understood as informing the addressee that the speaker can first go home (b). The third possibility denotes that both cores .are in focus. Here, the speaker is announcing two new pieces of information with the meaning in (c) intended—i.e., s/he is announcing that s/he can go home AND eat dinner. This holds for the core-level operator obligation as well and is also applicable to the following clause-level operators. Consider the following example. Here, the scope of epistemic modality (in this case, the mirative -kha-e-cha, formally the 'nonpast perfective')21 is ambiguous and only a knowledge of the speech-act situation can allow one to decide which of the three following paraphrases is most appropriate: (26) yas-le ghar ga-era bhδt khδ-e-cha. 3S-ERG home go-SEQ dinner eat-PFV-NPT.3S.NF knew he went home, but now I see he ate dinner/ knew that he was going to eat, but now I see that he ate at home/ *I just discovered: He went home and ate dinner/ The same is also true of second-hand or hearsay evidential (denoted by re): (27) tyas-le ghar ga-era bhδt khδ-yo re 3S-ERG home go-SEQ dinner eat-SPT.3S.NF EVID 'After he went home, they tell me that he ate/ 'They tell me that he ate aLhoms (and not somewhere else)/ They say that he went home and ate/ An example for illocutionary force (compare with [24]): (28) phalphul tδch-era nδm-lδl di-όip? fruit peel-SEQ little.girl-OBJ give-INJ.lS 'After I peel the fruit, should I give it to the child?* 'Should I peel the fruit, before I give it to the child?' 'Should I oeel the fruit and give it to the child?'

The Nepali Converbs: A Holistic Approach mill

That is, the scope of illocutionary force—in this case, the interrogative—is ambiguous here. Note that the scope of illocutionary force is necessarily identical with that of other operators, here deontic modality. In summary: Operator scope with the sequential convert) in same-subject junctures can be described as follows: Both the converbal core and the main core may be within the scope of all core- and clause-level operators, or one may be within this scope while the other is outside it. The only requirement is that at least one of the two be inside this scope. Despite this possible ambiguity, there is nonetheless an unmarked preference for operator scope to extend to both cores. When my informants were initially presented with examples such as (25) but were given no context, they virtually always (with very few exceptions) extended operator scope to include both cores. Nonetheless, this default interpretation is, as we have seen, easily overriden by other factors, given the appropriate context. 5.2.2. Different-subject

or 'Clause-lever Junctures

While the scope of core-level operators does not extend to both clauses in different-subject junctures,22 the scope of clause-level operators can. Consider the following examples: (29) rδm-le bhan-era us-le tyo kitδb padh-yo re. Ram-ERG say-SEQ 3S-ERG that book read-SPT.3S.NF EVID 'Ihear that he read that book because Ram told him to (and not because he wanted to).' *I hear that he read that book, since Ram told him to (i.e., instead of watching television).' *I hear that Ram told him to and that he read that book/ The same also holds for illocutionary force. Thus the following example also has three possible interpretations: (30) rδm-le bhan-era us-le tyo kitδb padh-yo ki? Rδm-ERG say-SEQ 3S-ERG that book read-SPT.3S.NF Q 4

Did he read that book because Ram told him to (and not because he wanted to)?' 'Since Ram told him to, did he read that book?' 'Did he read that book and did Ram tell him to?'

112 m John Peterson Hence, the potential scope of clause-level operators in clause-level junctures is the same as both core- and clause-level operators in core-level junctures, which is to be expected (cf. Van Valin and LaPolla 1997: Chapter 8). 5.2.3. Tense The clause-level operator 'tense' occupies a special position in sentences containing a sequential convert). This is because the action denoted by the sequential con verb, when used to join two clauses or cores, must precede that of the main verb or rather, its initial point must precede the action denoted by the main verb. Thus, we are concerned here merely with the question of whether the action denoted by the converb may refer to a past action or situation when the main verb refers to either a present or future action or situation, which is clearly possible. The following example may have either of two possible interpretations, where the underlined segment refers to the scope of the nonpast: (31) pδtlo lugδ lagδ-era iskφl-mδ jδ-mdai-cha. old clothes put.on-SEQ school-LOC go-CNT-NPT.3S 'Wearing (i.e., "having put on") old clothes, s/he is going to school/ *S/he's going to put on old clothes and go to school/ Here as well, the correct interpretation can only be gleaned from context, as it is necessary to know which information is considered presupposed or accomodated. • 5.3. Scope with the Imperfective Converbs After having taken a detailed look at scope in junctures involving the perfective and sequential converbs, let us now take a brief look at scope involving the imperfective converbs. When the imperfective converbs are not followed by a postposition,23 we find the same scopal pattern as with the sequential converbs. For example, let us return to example (7), repeated here. Note that (7) is an example of a samesubject (i.e., core-level) juncture and also allows for three possible interpretations. (7) ό susel-dai bδhira nisk-yo re. 3S whistle-CNT out emerge-SPT.3S.NF EVID 'They say that he went out while he was whistling (and didn't come in).' They say that he was whistling when he went out (and not singing).' They sav he went out and was whistling.'

The Nepali Converbs: A Holistic Approach · 113

Again, which interpretation is intended depends on which information may be considered presupposed or accomodated at the time of utterance. In the following example, we have a different-subject (i.e., clause-level) juncture and, just as with the sequential converbs, we again find three possible interpretations: (32) rδme sahar jδ-mdδ came ghar-ma bas-yo re. R. city go-NCNT C. home-LOC stay-SPT.3S.NF EVID They say Came stayed at home while Rδme went to the city.' 'They say Came stayed at home (i.e.. not somewhere else) while Rδme went to the city.' 'They say that Came stayed home and that Rδme went to the city (at the same time).' • 5.4. Summary of Scopal Properties With the exception of tense, the scopal patterns of the converbs can be summarized as follows: • With the perfective converb (in -e), either core/clause may be within the scope of all core- and clause-level operators, but never both simultaneously. • With the sequential and imperfective converbs on the other hand, at least when the imperfective converbs are not followed by a postposition, all core- and clause-level operators may have scope over either or both cores in core-level junctures and all clause-level operators may have scope over either or both clauses in clause-level junctures. Let us now take a closer look at the exact function of the converbs, beginning with the sequential converbs (Section 6). In Section 7, this will be contrasted with the function of the perfective converb and the imperfective converbs. This will then allow us to find a new designation for these various forms which better highlights their functions in a system of semantic-pragmatic oppositions.

• 6. The Meaning of the Nepali Sequential Converb As we saw in the last section, the Nepali sequential converbs are PRAGMATICALLY UNDERSPECIFIED - i.e., together with the main core they may be part of the focus in what Lambrecht (1994) terms a predicate-focus (i.e., topic-comment)

114 · John Peterson

construction, they may serve as the (pragmatically presupposed or accomodated) framework for the sentence, or they may themselves be the focus while the main core represents presupposed or accomodated information. Also, as can be seen from the translations of (25)—(32), the sequential converts are SEMANTICALLY UNDERSPECIFIED or 'vague'—i.e., while cores or clauses marked by the sequential converb may translate as causal, temporal or conditional clauses, etc., they are not marked as such in Nepali nor is this semantic interpretation necessarily intended. This can easily be observed when trying to find an appropriate paraphrase in Nepali for the converb-subordinated clause. Even in a sentence which clearly seems to bear a causal relation to the rest of the sentence, some informants initially rejected any paraphrase in Nepali involving the postposition le 'because'. That is, there is a certain resistance on the part of some informants to accept these paraphrases as meaning the same thing. This has also been noticed by other authors, such as Nedjalkov (1995: 100), who writes that 'constructions with converbs and conjunctions are by no means always interchangeable in languages that have both converbs and conjunctions.' What then could be the features for which the sequential converbs in Nepali are POSITIVELY marked? I believe the following quote from Davison, citing an apparently unpublished paper by Staneslow (1980) on the use of the sequential converb in Hindi, provides the answer: Staneslow (1980) notes that the perfective aspect affix is used in the verbal element of sentences that indicate transitions from one episode to another, in narrative. When the perfective aspect is demanded by the meaning of the proposition, but the sentence is not intended by the writer to indicate a transition in the narrative, the perfective meaning is expressed by -kar [i.e., the sequential converb ending in Hindi, JP] (Davison 1981: 120). Following Davison, I will argue here that the sequential converbs in -era, etc., in Nepali are positively marked in exactly these same two respects: 1. perfective aspect (which is denoted in the form -era by the perfective suffix -e) and 2. the action/event denoted by the converb is considered to be an integral part of a larger situation, of which both the subordinated core or clause and the main predicate are a part. This is indicated in the form in -era by the conjunction ra 'and'. These two characteristics have in fact been noted in some form for the sequential converbs in a number of other languages of the area—for example in Davison's article on Hindi, but refer also to Masica (1991: 399ff), where it is noted that these forms in many languages are often described by researchers as denoting some kind of 'natural relevance' between two different 'clauses' although, as Grain (1992: 126) in her study on Nepali notes, 'the nature of the

The Nepali Converbs: A Holistic Approach · 115

connection is not specified'. I will now deal with both of these criteria separately. • 6.1. Discrete Parts of a Whole Just what does it mean for two or more actions to be seen as integral parts of a larger situation? To begin with, consider the verb 'bring' in English. This clearly consists of two actions: 'take (something)' and 'come (with it)'. This combination of discrete actions has been lexicalized in English and many other languages, but not fully in Nepali. Consider the closest Nepali equivalents for 'bring': lyäu-, leräu- and li-era äu-, all three of which represent various stages of lexicalization of the combination of the converbal form of//- 'take' and the lexical stem äu- 'come'. The first form is the most lexicalized and employs the older form of the sequential convert), -f. Its form can be reconstructed as *//-f aw-, 'having taken, come' (cf. Turner 1931: 563). The second form is still completely transparent and is a slightly more lexicalized version of the third form, li-era äu- 'having taken, come'. In all three cases we have an action which is seen to consist of two INTEGRALLY RELATED parts, i.e., taking something and coming with it. However, no one particular semantic relation is expressed (or intended) between the two verb forms— they are simply two parts of a whole. Nonetheless, the first part of the action here necessarily PRECEDES the second part. In cases such as 'bring' and its functional equivalents in Nepali, it is quite easy to argue for what Kφnig (1995: 66) refers to in a somewhat different context as 'perceptual unity'. But this of course only accounts for a minimal number of the situations in which the sequential converb is used in Nepali. Now consider the following example: (33) lau bis rupiyδm li-era yi gδi-harό-ko jyδn PART 20 rupees take-SEQ these cow-P-GEN life chod-i-de-u. leave-0-BEN-IMP.2S 4

Here! Take these 20 rupees and spare the lives of these cows' (LS: 12).

This sentence was uttered by a pious Brahman to a Muslim who intended to slaughter the cattle and sell their meat. The Brahman was so disgusted by this thought that he bought the cattle on the spot to save their lives. The two clauses are intimately connected to one situation—by taking the 20 rupees, the Muslim will sell the cattle to the Brahman, thereby sparing their lives. The Brahman is not telling the Muslim to do two different and unrelated

116 m John Peterson actions but two parts of one and the same. Note also that, if the main verb appeared in the past, the example above could simply be an answer to the question 'What happened?': (34) bis rupiyam li-era yi gδi-harό-ko jyδn 20 rupees take-SEQ these cow-P-GEN life chod-i-di-yo. leave-0-BEN-SPT.3S.NF 4

He took the 20 rupees and spared the cattle's lives.'

or

'By taking the 20 rupees, he spared the cattle's lives.' or even 'He took the 20 rupees, thereby sparing the cattle's lives.' This is not necessarily true of subordination by means of a postposition or conjunction. For example, the sentence (35) My nephew was born after I moved to Switzerland. simply places two completely unrelated events into a temporal relation to each other. As all actions take place in time, they may of course be compared with each other temporally. However, there is "no need for them to be related to each other in any other way, as in (35). Note that the Nepali equivalent of (35) with the subordinate clause containing a converb in -era does produce a 'grammatical' sentence: (36) ma svitjarlyδnd ga-era merobhδnij janm-yo. IS Switzerland go-SEQ my nephew be.born-SPT.3S.NF However, (36) has a very different meaning than the one intended, namely 'Because I went to Switzerland, my nephew was born.' Instead, an appropriate translation of (35) would make use of the perfective converb in -e, followed by the postposition pachi 'after', i.e., ga-e-pachi instead of ga-era in (36). As has long been known, the conjunction and in English, when used in clause-level conjunctions, is underspecified with respect to the semantic relation between the two clauses (cf., for example, Davison [1981: 123], who cites an earlier work by Schmerling). As shown in the following example, this is also true of Nepali ra:

The Nepali Converbs: A Holistic Approach m 117 (37) δja today ra and

yasto such yasari thus

ke kam par-i-δ-yo what? work fall-0-VEN-SPT.3S dagur-daiδ-yau? run-0 VEN-SPT.2S

'Whatever happened today to make you come running here?' (LS: 3). (Lit.: 'What work happened [= "fell"] today and you came running?') It is precisely this vagueness which is expressed by the suffix ra in the sequential converb. Any language will of course have an upper limit to the number of concepts which it lexicalizes, although we often encounter situations where we view two or more of these lexicalized actions/situations merely as parts of a larger, integral whole. For various phases of this situation, lexicalized concepts will undoubtedly exist, just not for the entire situation. What I am claiming here is that the main function of the sequential con verbs in Nepali is merely to indicate this relationship, but not to express the semantic relation between the various phases, when no lexicalized concept exists for this situation in the language. This 'integrating' element is clearly missing from the meaning of the 'perfective' converb in -e. As already noted, this interpretation is in fact not new but has been noticed for several languages with 'clause-chaining'. Refer to Masica (1991:400), who quotes other studies which claim that the converb is said to connect closely related 'clauses' while Grain (1992: 126) speaks of a closer connection between 'predicates'. Both of these terms are of course fully justified. Here terminology can become somewhat confusing: As we have seen, the con verbal clause can modify 'adsententially' or 'adpredicately', depending on whether its discourse-pragmatic status is different or the same as that of the main clause, respectively. Finally, as we shall see in the next section, the converb can also modify adverbally (in the most literal sense). This is made explicit in Grain (1992: 128ff), where it is shown that Nepali converbs cover almost the entire range of Grain's 'Predicate Connectedness Continuum'. Here, the tightest junctures are referred to as being typical of 'serial verbs', a term I have chosen not to use due to the many meanings which it is given (cf. Ebert 1993a: 85), while the looser junctures are typical of 'clause chaining'.24 • 6.2. The Perfective Status of the Sequential Converb The second feature, [-»-perfective], is obvious enough when the verbal lexeme marked by the suffix -era, etc., represents a discrete part of the overall situation, such as 'take' in the lexical decomposition of 'bring' given earlier. Indeed,

118 m John Peterson this is always true of the sequential converbs when they refer to a different action than does the main clause. Now consider the following examples (adapted from Bickel, 1998b: 395): (38) ek chin u-sanga kurδ gar-era au-m-chu one moment 3S-with talk do-SEQ come-NCT-NPT.lS will talk to him for a moment and then I will come.' (39) himd-era au-m-chu walk-SEQ come-NCT-NPT.lS will come by foot.' In (38), the perfective status of the first clause is obvious, as the situation denoted by the predicate kurä gar- 'talk' (lit.: 'do talk') is necessarily bounded in the present example (cf. ek chin 'for a moment'). In (39) we have a very different use of the convert), in which the convert? modifies the main verb as an adverbial of manner. In fact, (39) can be analyzed as [+perfective]. It should be noted that the lexeme himd- is one of a large number of verbs in Nepali which are ambiguous between two inherent aspectual classes, one of which is what Sasse (1991) refers to as inchoativ-stativ, which I will refer to here as 'ingressive', while the other is either total-stativ or Aktion. In the present example, hitfid- is ambiguous between an ingressive and an activity interpretation. Consider the following definition of himd-, taken from Pradhan (1998:967): 'to set off; to move; to start; to w&lkprasthän garnu; godä cälera paidalai ek thautpdekhi arko fhauqimäjänu... [to depart; to go from one place to another, having set out on foot]'. I therefore suggest that in (39) we do not find an exception to the feature [+perfective] but rather a regular example of this. Thus, in order to 'come on foot', I must first 'set out on foot'. There are, however, many cases where this approach does not appear to be a sufficient explanation, such as the following: hos gar-era 'carefully' khel'playfully' hδms- 'happily' care do-SEQ play-SEQ laugh-SEQ Some examples: (40) yo poko dherai gahrumgo cha. this package very heavy COP:NPT.3S hos gar-era bok-a hai. care do-SEQ carry-IMP.2S PART

The Nepali Converts: A Holistic Approach · 119

This package is very heavy. Be careful how you carry it, now1 (Matthews 1992: 117). (41) tyas-sanga kati din mai-le häms3S-with how.many day IS-ERG laugh-SEQ khelbit-a-em. play-SEQ pass.of.time-CAUS-SPT.lS 4

How many days I spent happily and playfully with her' (VK: 59).

A quite common pattern is when the verb marked by the converbal ending belongs to the same semantic class as the main verb. Generally, the convert) is more 'specific' than the somewhat generic main (finite) verb, although the converb is not necessarily a hyponym of the main verb. Some examples: karδ-era shout-SEQ kamkalδ mournful

kurδ gar-*speak loudly' dub-era mar- 'die by drowning' talk dodrown-SEQ diesabda gar-era ru- 'cry loudly' sound do-SEQ cry-

(42) karä-era kurä gar-da tim-harό kaile kaile δphno shout-SEQ talk do-NCT 3-P sometimes own tebul-mδ phyδmk-0-the. table-LOC throw-NCT-SPT.3P

hat hand

Talking loudly, they sometimes pounded their hands on the table' (VK: 61). (43) ... yastδ kangδli such pitiful dub-era drown-SEQ

mora-ki joi bha-era bas-nu-bhanda corpse-GEN.F wife COP-SEQ live-INF-than mar-nu jati die-INF good

'... it would be better to drown than live as the wife of such a pitiful corpse [i.e., scoundrel]!' (GM: 65). (44) gaumthali kamkala sabda gar-era ru-na lag-L G. mournful sound do-SEQ cry-INF begin-SPT.3S.F 'Gaumthali began to cry with loud sobs' (GM: 65). Further complications arise from the fact that the sequential converbal forms of a number of verbs have lexicalized or further grammaticalized, such as

120 · John Peterson

barhi 'more1 from barh- 'grow, increase', bhanera 'quotative marker; that, in order that' from bhan- 'say' and perhaps even the objective marker /άί from lau- 'take' or l gi 'for' from lag- 'apply, affect'.25 I believe it is precisely the typical actional ambiguity described above (cf. himd-), as well as the relatively large number of ingressive verbs of the language, which have led to the further development of the sequential converb in Nepali to a general strategy for deriving adverbs from verbs. This can be explained as follows. Although there are as yet no exact figures for the number of common predicates belonging to the ingressive class or which are ambiguous between it and another class,26 the relatively high frequency of predicates in one of these two categories is quite striking and has been noticed by other authors (e.g., Ebert 1995: 191). In the following, I will provide a few examples of such verbs. As work here is still in progress, these must be considered tentative, although even the preliminary data I have strongly suggest the classification of the following predicates as either ingressive or ambiguous between the ingressive class and another class: • Presumably ALL experiential predicates which make use of the light verb lag-, which as a lexical verb means 'apply', e.g., bhok lag- 'become hungry' hunger apply(45) ma-lal bhok lag-yo. 1S-OBJ hunger apply-SPT.3S.NF Ί am/was hungry.' • • • • •

dukh—'hurt, start hurting' /»mrf—'walk, set out (on foot)' hu—'be, become' j n—'know, find out' pan!par— 'rain, start raining' water fall• sut—'sleep, go to sleep' • ubhi—'stand, stand up' • has—'sit, sit down'

There is also a non-productive process which seems to involve only intransitive ingressive predicates or those which are ambiguous between this class

The Nepali Converbs: A Holistic Approach m 121 and another class. In this process, a transitive verb is usually derived by means of a change in vowel quality or quantity from the intransitive verb, often accompanied by a change in the final consonant. This group of intransitive verbs is the Nepali class of unaccusatives. Normally, it is possible to increase the valency of a verb in Nepali by means of the causative marker -aw, added directly to the verb stem, e.g., gar- 'do', gar-au- 'get done, have someone do'. This process is almost completely regular and shows very few irregularities.27 However, in the non-productive process alluded to above, the two verbs are perhaps best considered separate lexemes, although the morphological relation between the two lexemes has a high degree of predictability to it. Table 4 presents those lexemes belonging to this group which I have so far been able to gather. Although I have not yet been able to test each of the following lexemes to determine its exact actional status, the following definitions, based on those given in various lexica, strongly suggest that a change-ofstate interpretation is at least possible for the intransitive member of each pair. Table 4: Intransitive/Transitive Pairs of Verbs Class I: u —> o / ä chut- 'leave' (itr.) khul- Open' (itr.) mur- 'turn back' (itr.) phut- 'burst, break' (itr.) tut- * break, be broken'

chod- / chad- 'leave' (tr.) khol- Open' (tr.) mor- 'turn' (tr.) phor- / phod- 'break, smash' tor- 'break' (tr.)

Class II: a / 0 -» ä bajra- 'collide /strike / fall violently' bal- 'burn, catch fire' bigra- 'spoil' (itr.) cal- 'move' (itr.) gad- 'be buried, sink into' gal- 'melt, dissolve' (itr.) jhar- 'fall, descend' mar- 'die' orla- / orli- 'come down, descend' pagla- 'melt' (itr.) pagra- Of milk to flow (from the udder)' par- 'fall; happen' ukla- 'climb up, ascend' umla- 'come to a boil' (NOT 'boil') urla- / urli- 'be tossed up, surge up' utra- / utri- 'descend, dismount, get off of

bajar- 'knock together' bδl- 'burn' (tr.) bigar- 'spoil' (tr.) cδl- 'move' (tr.) gad- 'bury' gδl- 'melt, dissolve' (tr.) jhar- 'bring down, put down' mar- 'hit, kill' oral- 'cause to descend, bring down, lower' pagδl- 'melt' (tr.) pagδr- 'cause milk to flow' par- 'drop; cause' ukδl- 'raise, lift' umδl- 'bring to a boil' ural- 'toss up, instigate' utδr- 'cause to descend, take off (clothes)'

This intuition is further confirmed by data on the use of a number of these verbs, including some which at first glance do not seem to be of the ingressive

122 · John Peterson

type. Consider the following verbs in the perfect, all of which have (or can have) a progressive interpretation: (46) thun-mδ dόdhajhai pagr-e-ko cha teat-LOC milk still flow-PFV-ATT AUX:NPT.3S.NF 'The milk is still flowing into the teat.' (47) ό ajhai rόkh-mδ ukl-e-ko cha 3S still tree-LOCclimb-PFV-ATTAUX:NPT.3S.NF 'He is still climbing the tree.' (48) pan! ajhai uml-e-ko cha water still come.to.a.boil-PFV-ATTAUX:NPT.3S.NF 'The water is still boiling/ Nepali also has a 'progressive' category which is formed by the compoundverb construction with the V2 raha-, which as a lexical verb means 'remain'. The entire predicate then appears in the perfect. For (46) and (47) there are progressive alternatives (although there are also slight differences in meaning between the two), while there is no progressive form for umla·. Thus: (49) thun-mδ dόdh pagr-i-rah-e-ko cha teat-LOC milk flow-0-PROG-PFV-ATT AUX:NPT.3S.NF 'The milk is flowing into the teat.' (50) ό rόkh-mδ ukl-i-rah-e-ko cha 3S tree-LOCclimb-0-PROG-PFV-ATT AUX:NPT.3S.NF 'He is climbing the tree.' (51) *pam uml-i-rah-e-ko cha water come.to.a.boil-0-PROG-PFV-ATT AUX:NPT.3S.NF 'The water is boiling.' This strongly suggests that pagra- and ukla- are ambiguous between the ingressive and activity types, while umla- is unambiguously of the ingressive type only. One last example:

The Nepali Converbs: A Holistic Approach m 123 (52) ό mar-e-ko cha / φ mar-yo 3S die-PFV-ATTAUX:NPT.3S.NF 3S die-SPT.3S.NF 'He has died. / He died. / He is dead.1 Thus, unlike its English counterpart, mar- in Nepali is a purely ingressive verb. To say 'Someone is dead1, either the perfect or the past is used. Also, the progressive is not allowed here: (53) *u mar-i-rah-e-ko cha 3S die-0-PROG-PFV-ATT AUX:NPT.3S.NF 'He is dying.1 To express the English meaning, a different construction must be used: (54) ό mar-na lag-e-ko cha 3S die-INFbegin-PFV-ATTAUX:NPT.3S.NF 'He has begun to die. / He is dying/ Thus, although not all of the lexical verbs given in Table 4 have been studied in detail yet, there is nevertheless strong evidence for assuming that a possible interpretation of the intransitive member as ingressive is essential. Even if it should turn out that not ALL of these predicates are of the ingressive type or ambiguous with this class, the preceding data should suffice at least to demonstrate the relatively high frequency of this class. A fuller discussion of this and related phenomena will be presented in Peterson (forthcoming).28 Now let us return to the adverbial use of the con verbs. Consider the following sentence: (55) mai-le bas-era eutδ kitδb path-em IS-ERG sit.down-SEQ l.NCLbookread-SPT.lS sat down and read a book1 / 4 I read a book sitting.1 With a verb such as has- 'sit (down) 1 in a converbal clause, we would have the following situation: A change of state into an 'activity1 (sitting) has taken place which is followed by another activity. However, unless an explicit interval of time is mentioned (or a goal with verbs of movement), the normal interpretation of the sentence is that the resultant state (i.e., 'sitting 1 ) still holds while the second action takes place. Hence, the converbal form of these

124 · John Peterson

verbs also allows for a simultaneous interpretation between a resultant state and an activity. In a language like Nepali which has a large number of such verbs, it seems likely that this interpretation can become productive and eventually even be used for verbs which are neither of the ingressive type nor ambiguous with it. This would then account for the adverbial uses found in (40M44)· In this analysis, the use of the 'sequential con verbal' endings as a general means of deriving adverbs from verbs must be considered a further development of their use, where these adverbial markers coexist with the homophonous converbal endings. Viewed this way, this adverbial use should not be considered a function of the 'sequential' converbs, just as the use of the Fugenelements in the compound-verb constructions, described in Sections 4.2 and 4.3 earlier, are not instances of converbs. The Fugenelements there, and the adverbial markers here, both represent FURTHER DEVELOPMENTS of the use of converbs. Their use in these environments has simply further grammaticalized to fulfill a very different function. The same comments of course apply to postpositions, conjunctions, etc., which have evolved from converbs. If this restriction is accepted, this would then allow us to consider the Nepali 'sequential converbs' to be consistently marked as [+perfective] when they refer to an action which is distinct from that of the main predicate. Similar comments would most likely also hold for the 'conjunctive participles' in Hindi and other Indo-Aryan languages as well. The adverbial use of forms which are homophonous with the perfective converbal forms would then have to be considered a further development, similar to the Fugenelements in the compound-verb constructions.

• 7. A New Classification of the Nepali Converbs As we saw in the last section, there are good reasons for considering the 'sequential' converbs to be 'perfective' converbs, although their use is clearly distinguishable from that of the 'perfective' convert) in -e. Since the ending -era is by far the most commonly used form of this particular category, a comparison of its form and meaning with the 'perfective' converbal ending will undoubtedly be the best guide to the proper designation of these converbs. Note that both -e and -era include the perfective marker •e, while only the 'sequential' form is further marked by ra 'and'. This morphological difference is also reflected in the semantics of the two categories: While the 'perfective' convert) in -e can be used for a number of semantic relations where the two clauses are not in any way considered to be integral parts of a whole, the 'sequential' converb in -era is confined to cases where

The Nepali Converbs: A Holistic Approach m 125 both clauses are considered integral parts of a whole (cf. [35] and [36] given earlier). Since the 'sequential' con verbs can also be analyzed as perfective con verbs, I would suggest also referring to the so-called 'sequential' converbs as perfective, with the further designation 'integrating' to denote the close relationship between the two cores or clauses. Thus what have hitherto been referred to as the 'sequential' converbs (in -f, -ikana and -era) would be 'integrating perfective converbs', while what has so far been termed the 'perfective' converb (in ~e) would properly be termed the 'non-integrating perfective convert)'. These new designations would then capture both the similarities as well as the differences between these two converbal categories which are obscured by the terms 'perfective' and 'sequential', although these terms are of course accurate. This classification is also supported by the fact that one of the integrating perfective converbs, -Γ, is homophonous with, and diachronically related to, the Fugenelement in the compound-verb construction, discussed in Section 4.3. Since in the compound-verb construction both V1 and V2 refer to different aspects of one and the same action/situation, it is easy to see why the verbbinding element is homophonous only with an integrating perfective converb and not with the non-integrating perfective converb. The same comments also apply to the endings used to form adverbials from verbs. These are homophonous with, and of course diachronically related to, the integrating perfective converbs and again only one single action is involved. Not surprisingly, only endings homophonous with the integrating converbs, never with the non-integrating form, are found here. Finally, the imperfective converbs are underspecified with respect to their integrational status, i.e., they may be used with or without a postposition which specifies the nature of their relation to the main clause. With that, we can now summarize the most salient properties of the Nepali converbs together as in Table 5. Table 5: The Nepali Converbs and Their Use with Postpositions Designation Perfective Imperfective

Non-integrating Integrating Continuous Nonconlinuous

Form

Possible Use with Postpositions

-e -ϊ, -ikana, -era //-Mdai// //-Md //

yes no yes (?) (cf. note 12) yes

In this analysis, it is not surprising that the integrating perfective converbs cannot take a postposition which specifies the interpropositional semantic relation as these forms already denote a close relationship to another core or

126 · John Peterson

clause. Since the non-integrating perfective convert) does not have this connotation, it is to be expected that the exact interpropositional semantic relation can be overtly specified, as is the case. Recall also that the distinction in the imperfective category is a purely aspectual one, while the distinction in the perfective category pertains to the 'integrational' function of the converbs. Hence, the three-way aspectual division illustrated for the verb system as a whole in Section 2 is also found in the converbs, while the perfective converbs also show a further, non-aspectual distinction.

• 8. Conclusions and Outlook As I have shown in the preceding pages, the Nepali converbs should be viewed in a 'holistic' manner, which includes morphology, syntax, semantics, pragmatics and even the lexicon. Any view which does not take all of these areas into account will necessarily be distorted and give at best only a partial view. Also, the intricate relations between these different areas make the Nepali converb somewhat more than just the sum of its parts: • In terms of morphology, the converbs in Nepali are nonfmite and lack tense, person and, in two cases (i.e., - , -Ikana), aspect marking. • Syntactically they are dependent, subordinate verb forms. • Their semantics can be rather complex: Although no interpropositional semantic relation is necessarily overtly stated, it can still be unambiguous in certain situations, although it is generally vague. • This goes hand-in-hand with the discourse-pragmatic status of the two cores/clauses: Depending on the situation and what, if anything, may be considered presupposed or accomodated information, either clause— and often both clauses—may be within the scope of the various operators. • The lexicon is involved to the extent that the integrating perfective converbs—and most likely the imperfective converbs when not followed by a postposition—are employed to denote part of a larger situation which is viewed as a whole but for which no lexicalized concept exists. An interesting, but extremely complex issue is that of areal influence on Nepali. It is interesting to note that a number of languages in the region employ a TAM-marked verb form followed either by a conjunction which is (at least etymologically) related to a word for 'and' or a comitative marker, two closely related concepts. Refer, for example, to the Classical Newari (Tibeto-Burman) sequential converb in -ä-vä '-ASP-COM'.29

The Nepali Converbs: A Holistic Approach m 127 The history of the Nepali forms is relatively certain: The integrating perfective converb ending -Γ almost certainly derives from the Old Indo-Aryan form in -ya, itself a 'sequential' converb marker, unfortunately generally referred to as the 'absolutive' or 'gerund'.30 As the form in -Γ was slowly becoming predominantly associated with the compound-verb construction in spoken Nepali, and as the erstwhile perfect participle in ~e (or rather, its earlier form -ya) was developing into the perfective marker, the form in -e-ra (-PFV-and) made its first appearance and it is tempting in the extreme to attribute this to areal contact.31 In this respect it is interesting to note that the first documented account of -era that I have been able to locate is from 1766 (Wallace 1985: 103, ex. 59a), by which time the Nepali- and Newari-speaking communities were in close contact to each other. Before this date, only the form in -ι / -i is found. However, without further detailed research, it is impossible to determine whether or not this is due to borrowing and if so, from which language. Abbreviations 0 - Fugenelement ABL - ablative ASP - aspect marker ATT - attributivizer AUX - auxiliary BEN - benefactive CAUS - causative CNT - continuous COM - comitative COP - copula ERG - ergative EVID - evidential F - feminine FOC - focus particle GEN - genitive HGH - high-grade honorific IMMED - immediacy marker IMP - imperative

INF - infinitive INJ - injunctive INST - instrumental IPFV - imperfective LOC - locative MGH - middle-grade honorific MIA - Middle IndoAryan NCL - numeral classifier NCNT - noncontinuous NHG - negation NF - non-feminine NOM - nominative NUM - number marking NPT - nonpast OBJ - objective OBL - obligation ΟΙΑ - Old Indo-Aryan Ρ - plural

PART - discourse particle PASS - passive PERS - person marking PFV - perfective PROG- prgoressive PT - past Q - interrogative marker REFL - reflexive pronoun S - singular SEQ - sequential converb SPT - simple past TEL - telicity marker TOP - topic marker VI - lexical stem in compound-verb construction V2 - actionality (etc.) marking in compoundverb construction VEN - ventive

NOTES I have profited greatly from countless discussions on the Nepali verb system with Balthasar Bickel and have incorporated many of his ideas into the following discussion, such as

128 · John Peterson referring to the two tenses in the imperfective as 'past' and 'nonpast', as opposed to their more common interpretation as 'past' and 'present', respectively. However, the remaining interpretations are my own and do not necessarily reflect his views. 2. Although there seem to be no clear-cut rules as to which contexts the full and short forms are used in, the data (but not the discussion!) in Pokharel (1996) suggest that this choice may be determined at least to some extent by regional factors. On the other hand, Matthews (1992: 228) notes that the full form is 'almost entirely restricted to the written language, being particularly common in newspaper Nepali*. 3. The focused form is in reality the imperfective marker combined with one of the many pragmatic particles of Nepali. In this case it is a focus particle which attaches directly to the word it modifies and has either the form - or -/ , depending on the form of the word it follows. Its use with the imperfective marker may be likened to an intensive marker and undoubtedly goes back to a time when the present-day imperfective marker //- Mda-// was still a participial marker. Hence, the participle at that time was a word in its own right and could thus be marked by this particle. 4. I will refer to the so-called 'perfect* participle as the 'perfective' participle. Although this form is used with the auxiliary-copula to form the perfect, and although (as example [ 1 ] shows) it can be used attributively to denote a past action, the following two examples should suffice to show that this form is not necessarily connected to past time, whether absolute or relative past: (a) pyδri! bhet-e-ko badδ din! beloved:F meet-PFV-ATT great day 'Beloved, the great day when we [will] meet' (referring to a future day when the man, who is planning to go to Tibet but who has not yet left, will return from Tibet to his wife in Kathmandu) (LD: 1280(b) tirni dhokδ banda gar-e-ko belδ-mδ kahδm hu-n-e-chau? 2S door close do-PFV-ATT time-LOC where COP-IPFV-ATT-NPT.2S 'Where will you be when [they] close the door?' (said of the door of a youth hostel which is locked early in the evening to a person who will be locked out.) This is further complicated by the fact that not all speakers accept sentences such as a and b as grammatical. This seems to be due to regional variations and will be dealt with elsewhere (Peterson, forthcoming). A note on the two 'attributivizers' is perhaps also necessary. I will gloss both of these forms as , although the two have a somewhat different status. While -ko is homophonous with, and diachronically related to, the genitive marker, -e (not related to the perfective marker!) is a 'pure' attributivizer and is frequently used to place nouns in attributive function. For example the name of one of the stories in Divas (1975/76) is müs-e bhat, 'the younger brother (bhat) who is a mouse (mφs)' or simply 'mousy younger brother'. 5. The analysis given here is somewhat different from what I argued in Peterson (1999). Since then, following up on suggestions by Balthasar Bickel and Karen Ebert (p.c.) that the 'simple past' is in fact aspectually neutral, I have been able to gather data which clearly show that this finite category is indeed not a perfective category but must be considered a simple past tense. However, as argued in Peterson (1999: 3420, the same marker //-E-// is

The Nepali Converbs: A Holistic Approach m 129

6.

7.

8. 9. 10. 11. 12. 13. 14.

15.

16.

17.

clearly used to mark this category as well as categories where it functions as a perfective marker, such as the perfective participle (cf., e.g., Genetti 1992:416f) or the past perfective. The term Fugenelement for this marker is borrowed from B. Bickel (p.c.). His original suggestion was only with respect to the marker -/-, which is homophonous with the sequential converbal marker (cf. Section 4.3), not for the marker //- Mdai//, which is homophonous with the continuous imperfective converbal marker (cf. Section 4.2). As I consider these two markers to have similar functions in these constructions, I have opted to use this term for both markers, which may not have been his intention. The difference between the two is mostly one of register: While -era is favored in the spoken language (at least in eastern Nepal, B. Bickel, p.c.), -Γ is especially common in written Nepali. Note that this only applies to the 'purely converbal' uses of -Γ, not to the homophonous form used in the so-called compound-verb construction (cf. Section 4.3). To my knowledge, only Slater (1994) distinguishes between the two imperfective converbs. The predicate in example (5) is properly speaking sik eko bhae, which consists of the perfective participle, marked by the compound suffix -e-ko (-PFV-ATT), plus the perfective converbal form of the auxiliary: bha-e (AUX-PFV). Judging from the data for the Jumli dialect of Nepali in Pokharel (1996), //- Mdo// and its cognate forms may still be quite common in some dialects of Nepali. Note that example (9) would seem to violate the rule given by Slater that this construction never allows for temporal coextensivity (1994: 159f)· Clearly, more research is needed on this somewhat ellusive construction. As Slater (1994: 153) notes that what I refer to as the continuous converb in //-Mdai// may not be followed by a postposition, it may be the case that it can in fact only be followed by a postposition when used in conjunction with the noncontinuous converb. Occasionally, the feminine form in -/ is used with female human beings instead of the nonfeminine form of the perfective marker -e. (cf. Genetti [1993] for a detailed discussion of verb agreement in Nepali.) The lexeme sut- is ambiguous between the ingressive readings 4fall asleep', 'go to sleep', 'go to bed', etc., and the presumably 'activity' reading 4sleep'. The first occurrence of sut- in example (17) clearly has the meaning 'go to bed/go to sleep'. That the second occurrence involves the activity reading of the lexeme and not the ingressive interpretation is clearly indicated by the overt specification of the amount of time, i.e., ek chin *a moment'. Vowel length is not phonemic in Nepali: The choice of -Γ- or -/- in writing is largely due to convention and does not always reflect actual pronunciation. As I am using a transliteration of the standard written language, the vowel length will be given here as it is encountered in the texts. Pokharel (1999: 192) lists 18 V2s which can be used in this construction, while my own data, gathered from a corpus of Nepali literature from various sources, as well as the discussion in Matthews (1992: 237ff), gives a total of 20 V2s. As Pokharel's data lists five V2s which were not found in the other data I used, this puts the total of more or less productive V2s which can appear in this construction at at least 25. Interestingly, data I gathered on an additional trip to Nepal show that the use of the V2 didoes not imply volitionality on the part of the subject. In this respect, the use of the V2 dicontrasts sharply with the use of the cognate V2 de- in Hindi, which apparently always implies volitionality (cf. Kachru [198la: 186ff]). Since 'benefactive' implies 'volitional', this is probably a misnomer. Perhaps the term 'affected object', used by Neukom (forthcoming) for a different construction in Santali (Munda, Austroasiatic), would be more appropriate, if we restrict this in Nepali to 'positively affected objects'.

130 · John Peterson 18. Due to the intricacies involved and the fact that much basic work still needs to be done on aspect and actionality in Nepali, I will not go into nuclear-level junctures here in any detail but will concentrate on core and clause-level junctures. 19. Nepali sentences whose 'cores' have different subjects are treated as not containing any common arguments. Hence, this type of juncture will be referred to as a 4clause-leveF juncture, even in cases where the one 'core' is understood to contain an object which is simultaneously the subject of the main clause, as in (29) and (30). 20. I will use the term 'pragmatically accomodated' here as it is used in Lambrecht (1994: 65-73). Accomodated information is information which is not presupposed but which the hearer is willing to accept as presupposed or backgrounded for the following asserted information. It would seem to have what Givon (1987) refers to as an intermediary presuppositional status and is generally also 'cataphorically grounded*. 21. For a discussion of mirativity in Nepali, cf. Michailovsky (1996) and Peterson (forthcoming). Note that Michailovsky refers to this category as TinferentieF, as it is also used to mark inference through results. For a broader discussion of mirativity in general, cf. DeLancey (1997). 22. In fact, I have not been able to find examples of sentences with clause-level junctures and the main verb marked for deontic modality which native speakers accepted as correct. For example, a sentence such as the following was considered incorrect. *rδm-le bhan-era us-le tyo kitδb padh-na sak-0-cha R.-ERG say-SEQ 3S-ERG that book read-INF be.able-NCNT-NPT.3S.NF 'Because Ram said so, he can read that book'

23.

24. 25.

26. 27.

Instead, the convert) (bhan-era) here was replaced by another verb form plus a conjunction/ postposition, e.g., the perfective participle form plus a postposition -bhan-e-ko-le (sayPFV-ATT-because) 'because he said so', or by a conditional clause. Apparently, the interpropositional semantic relation between the two clauses here is considered too explicit for a convert) to be used (cf. Section 6). For clauses which are subordinated by means of a postposition or conjunction, one would expect to find the same scopal pattern for the imperfective converbs as with the perfective convert), since the interpropositional semantic relation is clearly spelled out. Unfortunately, the data at my disposal are not sufficient to answer this question, hence this topic will not be dealt with further here. Note that what Crain refers to as 'Framing' and 'Association' (cf. Grain 1992: 58f, 72ff) seem to fall under adverbial modification here, while typical 'clause chaining' corresponds to what I am referring to here as adsentential and adpredicate modification. Kachru (1981b) also provides a large amount of data which call Davison's (1981) analysis of the Hindi 'conjunctive participle' or sequential convert) as [-1-perfective] into question. Even taking lexicalization into account, such as thatha-kar (explode-SEQ) 'loudly', from thathä- 'explode* which, as Kachru notes (1981b: 48, n. 4) 'is no longer an independent verb in Hindi-Urdu', there still seems to be a large amount of data against an interpretation of the 'conjunctive participle* in Hindi as [+perfective]. Many of Kachru's comments concern this adverbial use of the convert) and hence apply to Nepali as well. Work in this area is currently in progress (Peterson, forthcoming). There is also a small number of intransitives with lexical causatives, such as ja- /pathäu*go / send', etc., which need not be dealt with here. The only genuine examples of exceptions that I am aware of are the following two small groups:

The Nepali Converbs: A Holistic Approach · 131 Group I: khas-al- 'drop' from khas- 'fall' bas-al- 'set, settle' from has- 'sit, sit down* nikäl- 'take/bring out' from niska- 'go/come out, emerge' Group : khuväu- / khväu- 'feed' from khä- 'eat' piläu- /piväu- 'give to drink' from piu- 'drink'.

28.

29. 30. 31.

The intransitive members of Group I show strong similarities to the class of unaccusatives, although the intransitive members of these pairs very often appear to be agentive. This, however, would also seem to be true of certain 'unaccusatives' such as ukla- 'climb, ascend' in Table 4. Research in this area is currently in progress (cf. Peterson, forthcoming). The preceding discussion is a greatly simplified exposition of these verbs, especially concerning their use with the adverb ajha(i) 'still*. As it is merely intended to demonstrate the large number of verbs which belong to this class, we will not go into further detail here on this topic. A detailed treatment will be presented in Peterson (forthcoming). The exact status of J0rgensen's (1941: 67ff) so-called *A6' form is unclear. However, whatever its exact status may be, it certainly belongs to the realm of tense/aspect, and is possibly a perfective marker. Another strong piece of evidence for this suggested phonological development is that of the homophonous passive morph - - in Nepali from the (homophonous) corresponding form in -ya-. Cf. Wallace (1982) on the development of the perfect in Nepali and Bybee and Dahl (1989) on the development of a perfect to a perfective in general. These developments, as well as the emergence of the mirative in Nepali, are discussed in detail in Peterson (forthcoming).

• REFERENCES Primary Sources [BA] Adhikari, Bodhavikram. 1939. Barδline poiko tδlim. Modern Literary Nepali, ed. by Michael J. Hutt, 1997,49-54. [LD] Devakotä, Laksmiprasäd. 1935. Munδ-Madan (excerpts). Modern Literary Nepali, ed. by Michael J. Hutt, 1997, 128-51. [TD] Divas, Tulasi (ed.). 1975/76. Nepali lokkathä. Kδthmδndauip: Nepal Rδjakiya PrajnaPratisthan. Ne.Saip. 2032. [GM] Mainäfi, Guruprasäd. c. 1938. Parδlko ago. Modern Literary Nepali, ed. by Michael J. Hutt, 1997, 64-74. [LM] Munarnkarmi, Lfläbhakta. 1966. Hajär var$aaghidekhiko aitihäsik kathä-sarpgraha, Bhaktapur, Kδthmδndό: Lila Bradarsa. [LS] Sijäpati, Lalitajanga (ed.). 1996/97. Dantyakathä-mälä. Pulcok: Sδjhδ PrakδSan. Ne.Sam. 2053. Secondary Sources Bickel, Balthasar. 1993. Belhare subordination and the theory of topic. Studies in clause linkage. Papers from the first Kola-Zürich workshop, ed. by Karen H. Ebert, 1993b, 23-55. . 1996. Tense-aspect-mood in Nepali: A preliminary synopsis. University of Zόrich, ms.

132 · John Peterson Bickel, Balthasar. 1998a. Indexability effects in Himalayan languages: Clearing the ground for a typology. Paper held at the European cooperation project on Himalayan languages, Heidelberg, 3-6 June 1998. . 1998b. Review of Martin Haspelmath and Ekkehard Kφnig (eds), Converbs in crosslinguistic perspective: Structure and meaning of adverbial verb forms—adverbial participles, gerunds. Linguistic Typology 2. 381-97. Bybee, Joan L. and Osten Dahl. 1989. The creation of tense and aspect systems in the languages of the world. Studies in Language 13(1). 51-103. Grain, Laura Diane. 1992. Clause chaining in Nepali discourse. University of California: Los Angeles, Ph.D. dissertation. Davison, Alice. 1981. Syntactic and semantic indeterminacy resolved: A mostly pragmatic analysis of the Hindi conjunctive participle. Radical pragmatics, ed. by Peter Cole, 10128. New York/London/Toronto/Sydney/San Francisco: Academic Press. DeLancey, Scott 1997. Mirativity: The grammatical marking of unexpected information. Linguistic Typology 1. 33-52. Ebert, Karen H. 1993a. Kiranti subordination in the South Asian areal context. Studies in clause linkage. Papers from the first Köln-Zürich workshop, ed. by K. Ebert, 1993b, 83-110. (ed.). 1993b. Studies in clause linkage. Papers from the first Köln-Zürich workshop. Zurich: Seminar fόr Allgemeine Sprachwissenschaft, Universitδt Zόrich. (Arbeiten des Seminars fόr Allgemeine Sprachwissenschaft [ASAS], 12). . 1995. Ambiguous perfect-progressive forms across languages. Temporal reference, aspect and act tonality. Vol. 2: Typological perspectives, ed by Bertinetto, Pier Marco, Valentina Bianchi, Osten Dahl and Mario Squartini, 185-203. Torino: Rosenberg & Sellier. Genetti, Carol. 1992. Semantic and grammatical categories of relative clause morphology in the languages of Nepal. Studies in Language 16(2). 405-27. . 1993. Variation in agreement in the Nepali finite verb. South Asian Language Review 3(2). 90-104. (ed.). 1994. Aspects of Nepali grammar. Santa Barbara: Department of Linguistics, University of California (Santa Barbara Papers in Linguistics, 6). Givon, T. 1987. Beyond foreground and background. Coherence and grounding in discourse, ed. by Rόssel S. Tomlin, 175-88. Amsterdam/Philadelphia: Benjamins (Typological Studies in Language, 11). Haspelmath, Martin. 1995. The convert) as a cross-linguistically valid category. Converbs in cross-linguistic perspective. Structure and meaning of adverbial verb forms—adverbial participles, gerunds, ed. by Martin Haspelmath and Ekkehard Kφnig, 1-55. Haspelmath, Martin and Ekkehard König (eds). 1995. Converbs in cross-linguistic perspective. Structure and meaning of adverbial verb forms—adverbial participles, gerunds. Berlin/New York: Mouton de Gruyter. Hutt, Michael J. 1997. Modern literary Nepali. An introductory reader. Delhi: Oxford University Press (SOAS Studies on South Asia). J0rgensen, Hans. 1941. grammar of the Classical Newärl K0benhavn: Munksgaard. (Det Kgl. Danske Videnskabernes Selskab, Historisk-filologiske Meddelelser. XXVII, 3). Kachru, Yamuna. 198la. Transitivity and volitionality in Hindi-Urdu. Studies in the Linguistic Sciences 11(2). 181-93. . 1981b. On the syntax, semantics and pragmatics of the conjunctive participle in HindiUrdu. Studies in the Linguistic Sciences 11(2). 35-49. König, Ekkehard. 1995. The meaning of con verb constructions. Converbs in cross-linguistic perspective. Structure and meaning of adverbial verb forms—adverbial participles, gerunds, ed. by Martin Haspelmath and Ekkehard Kφnig, 57-95.

The Nepali Converbs: A Holistic Approach · 133 Lambrecht, Knud. 1994. Information structure and sentence form. Topic, focus, and the mental representations of discourse referents. Cambridge: Cambridge University Press (Cambridge Studies in Linguistics, 71). Masica, Colin P. 1991. The Indo-Aryan languages. Cambridge: Cambridge University Press. (Cambridge Language Surveys). Matthews, David. 1992. course in Nepali. London: School of Oriental and African Studies, University of London. Michailovsky, Boyd. 1996. L'Infe*rentiel du nopali. L'£nonciation m diatis e. ed. by Zlatka Guentchova, 109-23. Louvain/Paris: iditions Peeters. Nedjalkov, Vladimir P. 1995. Some typological parameters of converbs. Converbs in crosslinguistic perspective. Structure and meaning of adverbial verb forms—adverbial participles, gerunds, ed. by Martin Haspelmath and Ekkehard Kφnig, 97-136. Neukom, Lukas. Forthcoming. Santali. Mόnchen: LINCOM Europa (Languages of the World/ Materials). Payne, John R. 1985. Negation. Language typology and syntactic description. Volume I: Clause Structure, ed. by Timothy Shopen, 197-242. Cambridge et al.: Cambridge University Press. Peterson, John. 1999. The Nepali subordinated verb. Topics in Nepalese linguistics, ed. by Yogendra P. Yadava and Warren W. Glover, 337-70. . Forthcoming. Evidential, inferentials and mirativity in Nepali. Person and evidence in Himalayan languages: Contributions of the European Cooperation Project on Himalayan Languages. Special Edition of the Linguistics of the Tibeto-Burman Area, ed. by Balthasar Bickel. -. Forthcoming. Nepali predicates and the interplay between transitivity, actionality and volitionality (working title). Pokharel, Balkrishna. 1996. Magar impact on Nepali: A case of simple present. Journal of Nepalese Studies 1/2. 203-9. Pokharel, Madhav P. 1999. Compound verbs in Nepali. Topics in Nepalese linguistics, ed. by Yogendra P. Yadava and Warren W. Glover, 185-208. Pradhan, Babulall. 1998. Ratna's Nepali English Nepali dictionary. Varanasi: Trimurti Prakashan. Sasse, Hans-Jürgen. 1991. Aspekttheorie. Aspektsysteme, ed. by Hans-Jόrgen Sasse. Kφln: Institut fόr Sprachwissenschaft, Universitδt zu Kφln. (Arbeitspapier Nr. 14 [Neue Folge]). Slater, Keith W. 1994. On differentiating daa and day. Aspects of Nepali grammar, Carol Genetti, 132-65. Turner, Ralph Lilley. 1931. A comparative and etymological dictionary of the Nepali language. New Delhi: Allied Publishers. Van Valin, Robert D., Jr. and Randy J. LaPolla. 1997. Syntax. Structure, meaning and function. Cambridge: Cambridge University Press. (Cambridge Textbooks in Linguistics). Wallace, William D. 1982. The evolution of ergative syntax in Nepali. Studies in the Linguistic Sciences 12(2). 147-211. . 1985. Subjects and subjecthood in Nepali: An analysis of Nepali clause structure and its challenges to relational grammar and government & binding. University of Illinois at Urbana-Champaign: Ph.D. dissertation. Yadava, Yogendra P. and Warren W. Glover (eds). 1999. Topics in Nepalese linguistics. Kathmandu: Royal Nepal Academy.

Three Levels of Lexical Codification ANITA RAVANAM ι

'Difficult word* is a sociolinguistic term in the sense that some people of a speech community understand certain words while some do not. This inequality in the comprehension of words is implicit in the labels of difficult and easy. Can difficult word also be a grammatical term? This study proposes to throw light on this unanswered question. Grammatical practice at present assumes all the words of a language to be equally easy for its speakers. I shall argue here that the levels of lexical difficulty of the words of a language need to be studied not only sociolinguistically, but grammatically as well.

• 1. Focusing on Lexical Difficulty Difficulty is a phenomenon that varies across speakers. Some people may find certain words difficult to understand while for some others, they may be very easy. But what happens if virtually all members of a certain speech community find certain words too difficult to decode? One might try a sociolinguistic approach. But surely that becomes inappropriate if the whole community experiences the same problems. Hence my proposal to work towards a grammatical representation of relative difficulty levels, distinguishing words harder to understand from words easier to understand. This study tries to formulate as a general issue the question of what counts as a difficult word. My proposal is that this notion needs to be characterized within generative linguistics, as a formal concept, instead of being left at the level of sociological considerations that might have been enough if it were only a matter of some section of the speech community. In particular, this

136 · Anita Ravanam

study argues for a theory of lexical entries (the Differentiated Theory of the Lexicon, DETL) that begins to provide a specifically generative analysis of the notion 'difficult word'. Generative grammar—in the inclusive sense of the 'grammar', encompassing both generalizations and the lexicon containing word-particularizations— has consistently worked with a zero sociolinguistics. This was made explicit in Aspects (Chomsky 1965:4), but was arguably already present in structuralism or even in the Neogrammarian School of historical linguistics that gave rise to structuralism. My point is that grammar as a field has surely become strong enough to drop that abstract and oversimplified idealization and move closer to the relevant realities. I take it that initial assumptions in generative grammar were in fact slightly revised versions of structuralist ideas, and that only the mature formulations that are now being improved upon are to be taken seriously as generative. In the context of this search for truly generative solutions, it becomes possible for core linguists to stop assuming zero sociolinguistics as a base line. In this context, the problem that the present paper takes up for study is that of representing lexical difficulty. In this paper, I take as my empirical point of departure the story of the modernization of Bangla. This success of the Chalit Bhasha movement—for the acceptance of speech-like (chalit or current) written language which would displace the archaic (sadhu or pure) written norm—is considered to be one of the more visible aspects of the cultural modernization of Bengal in the 20th century. I wish to first address some issues that appear at this level of our understanding of this success. Given the purposes of this study, it is of course inappropriate to dwell on the particular (historical or sociolinguistic) details of which chalit words or locutions or constructions replaced which sadhu forms as part of the code shift. My purpose, instead, is to formulate and address the cognitive question of why educated users of chalit Bangla (CB) find certain Sanskritic words relatively accessible and certain ones more opaque, to this day. This contemporary fact is surely related to the undebatable historical fact that Modern Bangla took a major turn with the rise of CB. At this juncture, I am definitely trying also to address readers whose primary concern is with the history and sociolinguistics of that major turn. But I wish to view the problem cognitively. It is clear to all observers that the movement for the general adoption of CB as the written Bangla standard was effectively proposing to move the cognitive system encoded in the Bangla lexis away from its traditional support system in Sanskrit. At one level, the purpose was to reassociate it with the new base available in English. But this was not the professed intention, for the CB proponents were nationalists. They could not and did not suggest a wholesale borrowing of English terms into the Bangla lexis. They continued the long-established practice of calquing or loan translation

Three Levels of Lexical Codification · 137 which involved Sanskrit-derived coinages (or revivals from the Sanskrit corpus) becoming modern Bangla equivalents for items drawn from the technical vocabulary of contemporary English. These modernists thus ended up as sponsors of a Sanskrit-based term planning enterprise, and ironically reversed their own process of making the standard written language democratic and easy to use for the lower classes who were at last attaining literacy and aspiring to social equality. My earlier work (Ravanam 1995) focused on the details of this contradictory modernization process. The argument of the present study tries to use the material from this sociolinguistic story in the new context of issues in core linguistics. The specific context here is the proposal that generative grammar should take seriously the task of formally representing the lexical difficulty aspect of the individual native speaker's knowledge of his or her language. The rest of this study unpacks this argument in the following manner. Section 2 argues that word formation strategies (WFS) morphology is suitable for the task at hand. As I see it, the task is to explain why, although many technical terms were coined for Bangla during the modernist period, only some terms have gained currency in the ordinary prose of the community while others have not gained such currency. Section 2 shows that my application of the theory is a natural continuation of the WFS morphology itself. The particular examples that I have considered in Section 3 juxtapose the sociolinguistic context of term planning with morphological theory based on formation strategies. In Section 4,1 spell out the easy word vs difficult word contrast in terms of generative linguistics. My implementation is based on a particular theory of lexical entries called the Differentiated Entry Theory of the Lexicon (DETL). It is presented as an alternative to the current default theory, which one may call the Uniform Entry Theory of the Lexicon (UETL).

• 2. The Three Models The morphology I have found most suitable for the purpose of this study is the morphological theory proposed by Ford, Singh and Martohardjono (1997). At the heart of this theory lies the WFS (word formation strategy), defined as follows. Consider the WFS (1): (1) [ where (i) X and X' are words (ii) a and b are morphological categories

138 · Anita Ravanam (iii) 4-> indicates an equivalence relation (a bidirectional implication) (iv) X7 is a semantic function of X (v) ' indicates a formal difference between the two elements of the relation of morphological relation (vi) ' can be null if a # b For example, the morphological relation between the two English words dog and dogs can be represented as follows: IX] noun, singular ~

z noun, plural

Why should WFS morphology be especially appropriate for the study of lexical difficulty? One possible line of response to this question is that it simply is the best of the models available. Ford, Singh and Martohardjono (1997) and related writings already provide a defense of WFS morphology at that level. I omit that purely morphological answer to the question here. Instead, I approach the question from a perspective that considers models of morphology in terms of sociocultural assumptions associated with them. I am not claiming that persons who espouse these theories in morphology are logically compelled to hold particular social views corresponding to them. My point is that the logic of each morphological theory invites, or makes natural, the adoption of certain sociocultural attitudes. To focus on these, I shall consider the Lex (S), the mental lexicon of an individual speaker S; and I shall also bear in mind Diet (L), the social dictionary culturally associated with the speech community that uses language L. With these concepts in mind, I shall now successively outline three lexical matrices. A lexical matrix, or a model of lexical codification, is an idea that I am using to make it easier to focus on grammatical and sociolinguistic concepts simultaneously. My application of the WFS model of morphology to the lexical difficulty domain works through the third of the three LMs (lexical matrices) considered in the following exposition. My argument for the WFS model and against the other theories of morphology is embedded in this discussion. The notion of a LM is best viewed as an embodiment of the social arrangements whereby the lexicon of the entire speech community SC (L), in a given time and place, defines the relation between the social dictionary, Diet (L), and the mental lexicon, Lex (S), for any arbitrary speaker S within SC (L). For ease of exposition, I shall mimic the standard three-period reading of linguistic advance. I thus begin with (a) traditional grammar (and its extension into historical grammar), continue with (b) structuralism, and I end at (c) generative grammar in its mature form. Correspondingly I postulate three LMs— (a) the traditionalist lexical matrix (TLM), (b) the modernist lexical matrix (MLM), and (c) the postmodernist lexical matrix (PLM).

Three Levels of Lexical Codification · 139 I propose, then, a model that initially identifies these LMs as follows. The TLM is what has been inherited by nearly all literate societies. The MLM has driven linguistic stability and changed patterns (often through language planning) in many speech communities over the last few centuries. Finally, the PLM that seems to be emerging today in societies familiar to us and is waiting for careful investigation. At the same tentative level, I shall approach the three LMs in terms of a question that arises from the notion of a word formation strategy. The operative question for me will be: Which actor is supposedly pursuing the strategies that enable word formation? In TLM, nominally SC (L) as a whole (in practice, some controlling ilite) is the sole strategist who alone can apply WFSs to form new words out of old. This LM works nominally with Diet (L), in practice with the largely unified Lex (S)s shared by most or all olite speakers (this unification is arranged by an authoritarian educational system), and denies or ignores the ability of a non-£lite speaker S to use WFSs on his or her Lex (S). Under TLM, an arbitrary speaker S is expected to acquire a Lex (S) that copies (from society into S's mind) as much of Diet (L) as S's 'merit' will allow. S must leave WFSs—and the composition of Diet (L) in general—to SC (L), that is, to the 61ite of this speech community. Now, the WFS model in linguistics is committed to the view that morphology is embodied as WFSs which apply in S's mind to sets of entries in Lex (S) and not to some Diet (L) controlled by the 61ite of SC (L). Therefore WFS morphology can only treat TLM as a social pathology and must treat its inbuilt conception about word formation as ideology-bound hypotheses that a science of language must identify and criticize. My second lexical matrix, MLM with the initial M for modernist, reflects the worldwide Enlightenment process that ushered in the Modern period. Under the sponsorship of the social forces of modernization, and moving towards the articulation called Modernism that develops some potentialities of the modern period but limits other, MLM realigns the relations between morphology, the social Diet (L), and the mental Lex (S). In MLM, morphology's word formation strategies are attributed to the typical speaker S as strategist and visualized as operations on the typical Lex (S). Some divine, royal, or other authoritarian sanction used to allow an SC (L) under the TLM system to make some Diet (L) count as sacred (and to sidetrack the morphology from Lex (S) to Diet (L). That traditional sanction is either suspended socially, or removed theoretically from sociolinguistic consideration. This leaves open the question of the relevance for linguistics of some minimal Diet (L) (containing only the minimal set of words) shared by the entire set of speakers or some maximal Diet (L) listing all the words known by speakers, even words known only to some speakers. Only the traditional sanction favoring a particular Diet (L) and the claim that morphology

140 · Anita Ravanam

pertains to that social object are canceled. MLM does not cancel the reality of Diet (L) level objects in society or their relevance to linguistics. MLM's image of a typical Lex (S) of a typical S is a loaded image. It carries with it the social implication that some undeclared elite of SC (L) will build a hidden special Diet (L) on the basis of the secular knowledge claims that drive modernization. It is further implied that, through the educational, literary and media systems, this elite will sell this Diet (L) to most speakers in SC (L) as their typical Lex (S). If such processes continue in society, the normal style of functioning of any elite guarantees that such a modernist elite will inherit the traditionalist elite's dictionary-bound habit of taking the language seriously and ignoring the individual speaker. This amounts to reducing the language to its existing words and texts, a set of products, rather than a potentiality, in the mind of every speaker S, for new creative speech appropriate to all future occasions. In other words, this amounts to a pregenerative, structuralist view of language. Such a view applies economy criteria to a nonhuman (or mind-neutral) mapping between sounds and meanings. If economy bypasses the speaker, it arrives at the morpheme as the basic unit of grammar. Our third and final lexical matrix, PLM with P for postmodernist, reflects the now emerging learner-focused realities of social resistance to (and correction of) the authoritarian teacher-centered excesses of the Enlightenment process. In PLM, WFSs are defined on the individual speaker S's actual Lex (S). There is not even a hidden role for Diet (L) in the linguistics of word formation. As a consequence, WFSs are viewed as active mental strategies of reception and production by the speaker-hearer. They are not disguised versions of rules that list what is available in Diet (L), or of redundancy rules that structurally economize the information content of such a list. The mental economy of the speaker treats the word as its basic unit. Correspondingly, in serious generative morphological work, successful analyses rest on the word and not the morpheme. This LM corresponds to the social reality that, after the modernist system has been in power for some time in a particular society, voices emerge to represent the needs of individuals that go unsatisfied under modernism. Subsequently, new social forces compel modernization to become a servant, not a master, of the individuals who are supposed to be beneficiaries of modernization. This is the postmodernization process. The goal which linguistics shares with literature, presumably, includes a better understanding of the sociolinguistic and linguistic facts we are presenting in terms of the PLM. This matrix, in our conceptualization given earlier, allows the individual speaker S to come into his (or her) own as the site of generative processes in linguistic knowledge and use. S may well not belong to the 61ite of SC (L). Thus, the mental object Lex (S) for a speaker of Bangla, for instance,

Three Levels of Lexical Codification m 141 may have little to do with the Sanskrit-laden sectors of MaxDict (L)—a social object to which PLM gives no pride of place. But PLM makes morphological processes, WFSs, accountable only to Lex (S) and formulable only as strategies for a mind1 to use. Thus, WFS morphology, coupled with appropriate assumptions about the way speakers use strategies, leads to the prediction that only some of the technical terms coined on a Sanskritic base by the term planners for Bangla are likely to become part of the general lexical currency of Bangla. Other terms, which could be understood only if one were to dissect them into constituent roots and affixes as one is taught to by the prescriptive pedagogic grammar of Sanskrit—terms that the WFSs of the Lex (S) of any odd speaker of Bangla (outside the trained £lite) do not enable S to make sense of—are predicted to be still-born creations. It is clear that the generative morphologies employing rules (rather than, as in the WFS model, strategies) are ambiguous between a rule-bound structuralism (associated with MLM) and a truly generative attention to mental abilities and actions of speaker S, associated with PLM. I suggest that generative grammar, being committed to the individual's point of view, should select PLM as its default sociolinguistic baseline, among the three models of lexical codification considered in this section.

• 3. The Third Model and Word Formation Strategy Morphology In support of the claim that the third or PLM model attains the highest degree of adequacy, Section 3 presents some concrete analyses. I shall look at some failures and some successes among the technical terms proposed by the modernist planners to enrich the corpus of the ascendant chalit Bangla written code. Let me first quickly present some relevant institutional facts. The University of Calcutta set up a technical terminology committee in 1930. Its products were first made available in a set of vocabularies. These vocabularies were arranged subjectwise in the form of English to Bangla glossaries. These are now readily accessible in the terminological appendix of the widely used desk dictionary Calontikaa (Basu 1956) (but the final section ofthat appendix is drawn from the post-Independence work of an official terminology committee established by the Government of West Bengal—preserving continuity with the work of the University of Calcutta's committee). Individuals had already been coining terms for modern discourse in the decades leading up to 1930, and many of the proposals were already in circulation among the intellectual elite. Modernist authors around 1930 were already using them in small quantities. The use of these terms obviously grew after 1930. But I am not

142 · Anita Ravanam studying the history of this use as a separate issue. My interest is in the modernist logic that leads to the coinage of certain types of terms, and the processes that make some subtypes relatively acceptable to the public but not other subtypes. Taking the official stand of the educated public at face value, the practical assumption made by the term planners seems to have been that educated Bengalis could be expected to know their Sanskrit. The idea was that their schooling enabled them to use the prescriptive grammatical code to penetrate the root-affix structure of any arbitrary new coinage from the tatsama vocabulary system. If this was true, they would, for example, easily understand that shamaahartaa must mean 'collector'. For this new coinage attaches the agent suffix taa 'performer of action specified' to a base assembled from the prefixes sham 'in one place' and aa 'consequences of act flowing towards actor' plus the Vhri 'taking or removing resources'. This base is also seen in the existing (but highly 'Sanskritic' or 'erudite') word shamaahaar 'aggregation'. If one agrees with the basic assumptions of the enterprise of coining Sanskritic terms for modern Bangla, then one sees the existence of this word as a factor that reinforces the ability of the educated Bengali to decode the new term shamaahartaa. As a matter of social fact, however, the educated Bengali public of the early 20th century had only a precarious familiarity with Sanskrit. Furthermore, the public seemed relieved when the language was removed from the school curriculum. There has never been any popular clamor for its reintroduction. Nationalist compulsions may have led to a public posture defending the community's right to retain Sanskrit in the curriculum as something in the nature of a cultural counterweight to English as long as the latter was associated with the colonial masters. Once the British left India, these compulsions disappeared. At that juncture even the overt posture of the educated public began to reflect what had throughout been the psychological reality on the ground— the fact that the school-educated Bengali public never had acquired a true Bangla-Sanskrit bilingualism. Next to this social or sociolinguistic fact, it is important to place the relevant cognitive or core linguistic (specifically, morphological) facts about the recognition of a new word in its proper perspective. In the absence of special familiarity with morpheme constructs acquired through training in Sanskrit grammatical precepts that help decode unfamiliar tatsama (Sanskritic) creations—and we have just seen that educated Bengalis were unable to use whatever special knowledge the system tried to impart to them—a Bangla speaker S will try to understand new words on the basis of word formation strategies (WFSs) that extend the sound-meaning association patterns of the existing words in S's Bangla lexicon Lex (S). Consequently, S cannot

Three Levels of Lexical Codification m 143 understand shamaahartaa Collector' on the basis of the normal principles whereby viable new words can be recognized without special training. I turn now to some comparisons between modernist coinages that have gained, and terms that have failed to gain, common public currency. This is where one can see the WFS model in action. • 3.1. /shamaahartaa/ 'collector' vs /porimaapok/ 'surveyor' Why do speakers of Bangla seem to find the relation between /porimaap/ 'survey measurements' and /porimaapok/ 'surveyor' more transparent than the committee-imagined relation between /shamaahaar/ 'collection' and the term /shamaahartaa/ 'collector' proposed by the terminological planners? The answer lies in the word formation strategy presented next and the examples that follow:

(2) (3) a. b. c. d. e.

paatxh onubaad aabishkaar bicaar procaar

f. naash g. pwtirodh

'reading' 'translation' 'discovery' 'judgement' 'propaganda, spread' 'destruction' 'resistance'

a', b'. c'. d'. e'.

paatxhok onubaadok aabishkarok bicaarok procaarok

f. naashok g'· protirodhok

'reader' 'translator' 'discoverer' 'judge' 'propagator, disseminator' 'destroyer' 'resister'

Here we find a consistent relation of covariation of expression and content between the left-hand set (3) a-g and the right-hand set (3) a'-g'. In sharp contrast, there is no way that a Bangla speaker can relate /shamaahaar/ 'collection' to the proposed form /shamaahartaa/, which is therefore impossible to understand on the basis of existing words in Bangla. For this supposed relation to be visible to the speaker, a WFS such as (4) would have to receive support in the lexical patterns of the language: (4) *[Xaar] ActionNoun «-»[Xartaa] AgemNoun Does it receive such support? At least there are indeed action nouns ending in /aar/ such as (5): (5) a. aahaar b. bicaar c. beabohaar

Beating' 'judgement' 'use, behavior'

144 · Anita Ravanam

d. bishodgaar e. shanghaar

'spouting poison out, vituperation' 'destruction'

However, corresponding agent nouns ending in /artaa/ in general do not exist. None of the conceivable counterpart forms listed in (6) is attested: (6) *aahartaa *bicartaa *beabohartaa *bishodgartaa *shanghartaa

'eater' 'judge' 'user' *' vituperator' 'destroyer'

In the other direction, the /taa/ element that prescriptive Sanskrit grammar treats as an affix and that provides the basis for the term planners' proposal /shamaahartaa/-does occur at the end of some agent nouns, such as those listed on the left-hand side in (7), which can be paired with action nouns like the ones listed on the right-hand side in (7). This fact may initially look promising: (7) a. b. c. d.

racoyitaa nirmaataa kretaa hantaa

'writer' 'builder' 'buyer' 'killer'

a', b'. c'. d'.

raconaa nirmaan krae hanon

'writing' 'construction' 'purchase' 'killing'

But, in fact, these examples distinctively show that no single pattern of form-content covariation emerges to give psychologically real support in Bangla to anything like a /taa/ affix even in a stretched sense of this term. Thus, it does not come as especially important news that list (7) can be extended to one case that would have fitted the pattern in (4): (7) e. aabishkaar 'discovery'

e'. aabishkartaa 'discoverer'

Bangla evidently has two words for 'discoverer', /aabishkaarok/ given above at (3c') and /aabishkartaa/ given now (7e'). But there is a difference. (3c') forms part of a pattern formalizable as a word formation strategy. In contrast, (7e') stands alone exactly like (7a', b', c', d') and does not lend support to any WFS. A single pair like (7e, e') cannot be adduced as a basis for a WFS like (4). This is one reason that the analogy of (7e, e') does not make it possible to interpret /shamaahartaa/ as 'collector'. Another, equally pertinent reason is that /shamaahaar/ has the sense of 'collection' as a set, not as an action of 'collecting', and thus does not really count as an action noun.

Three Levels of Lexical Codification m 145 • 3.2. /taarxito/ 'electrical' vs /boiddutik/ 'electrical' The competition here is between the morphological support base for the pairing of (8a, a') and the base for that of (9a, a'): (8) a. torxit

'electricity'

a', taarxito

'electrical'

(9) a. biddut 'lightning, electricity' a', boiddutik 'electrical' One factor is the relatively erudite character of /torxit/ compared to its more popular synonym /biddut/. Nonetheless, this would not make /taarxito/ quite as inaccessible to the new word interpretation system as it in fact turns out to be. Let us look carefully at the WFS situation to understand the problem better. In both cases, to use the terms of prescriptive Sanskrit grammar, there is a vriddhi process applying within the word, mapping the vowel of the first syllable into /aa/ in (8a') and into /oi/ in (9a'>. The affixes with which this vriddhi is associated are final /o/ in (8a') and final /ik/ in (9a'). If Sanskrit grammar were the issue, (8a') and (9a') would seem equally complex, and there should be no difference in acceptability. However, the WFS difference is much greater. For (8a, a') to be supported by a WFS, the relevant strategy would need the interdigitated constant shown in (10), whereas the WFS that needs to exist to provide support to pair (9a, a') would involve the interdigitated constant shown in (11): (10) WFS format for (8a, a'): [CoX] NNoun [CaaXo]Adjective A1 . L

J

L

J

(11) WFS format for (9a, a'): [CiX] NNoun [CoiXik] Adjective A1. 1

J

l

J

What do existing Bangla words say about the respective viability of these WFS formats? They vote clearly for the claim that (11) is indeed a true WFS, as we see when we extend list (9): (9) b. c. d. e f.

bishae biggean iiihaash pishaac binaash

'matter' : 'science' : 'history' : 'brute' : 'destruction':

b'. c'. d'. e'. f.

boishoyik boiggeanik oitihaashik poishaacik boinaashik

'material' 'scientific' 'historical' 'brutal' 'destruction-related'

146 · Anita Ravanam g. din h. bidesh

'day' : g', doinik 'foreign country': h'. boideshik

'daily' 'foreign'

No extension of list (8) seems possible. After a prolonged search, I have failed to turn up further examples of what would have been strategy (10). Thus, there is no support for the Sanskrit-inspired proposal (8a') in Bangla. In the face of the plentiful support available for (9a'), it is thus clear why its rival (8a') never stood a chance. • 3.2. /udbaayi/ 'volatile' vs /udbaastu/ 'refugee' These are not in competition, but merely in contrast. It seems fair to say that speakers who have learnt chemistry in Bangla have assimilated the term /udbaayi/ in the technical sense of 'volatile'. But the general prose registers would not permit any author to imagine using /udbaayi raajniti/ for 'volatile polities', for instance. In the general context, and for the many speakers who have not done high school chemistry in Bangla, the word simply does not exist. In contrast, an apparently similar word such as /udbaastu/ for 'refugee' is available for a wide range of uses and users. Why does this difference in usability exist? One finds several words that seem to support WFS (12), where the adjective /utX/ formed from the noun /X/ means 'moving out of or away from X if X is a location, moving/stretching out one's X if X is a body part', as seen in list (13), where the t-to-d shift is automatic: (12) [X] Noun [utX]AdJcclive (13) a. b. c. d. e.

kendro pinjar baahu kantxho baastu

'center' 'cage' 'arm' 'throat, neck' 'homestead'

: : : : :

utkendro utpinjar udbaahu utkantxho udbaastu

'eccentric' 'released from cage' 'with arms stretched out' 'anxious' 'dislodged from home'

Since the noun /udbaastu/ 'refugee' emerges as an extension based on the adjective /udbaastu/ 'dislodged from home' that receives so much support, it faces little difficulty in finding acceptance as a general term in the language. But what about /udbaayi/? It could in principle be related to /shamobaayi/ 'collective'. But the latter in fact has to do with /shamobaae/ 'collection, cooperative (said of a society)', whereas there is no */udbaae/ corresponding to /udbaayi/. In fact one would have needed to establish a relationship between /udbaayi/ and /baayu/ 'air', but there is no morphological basis for this. For no WFS otherwise links any word of the /Xu/ type to any word that looks like

Three Levels of Lexical Codification m 147 /utXi/ or even just /Xi/. Since none of the possible connections actually works, the consequence is that /udbaayi/ is simply an isolated form. Thus, no other word in the language can help a speaker to make sense of /udbaayi/ unless it has been specially taught. The word /udbaayi/ 'volatile' thus ends up being much less accessible than /udbaastu/ 'refugee'. • 3.4. /ashshaaxr/ 'amnesia' vs /srixtibhrangsho/ 'loss of memory' The term /ashshaaxr/ must have seemed to the terminological committee to be relatable to /srixti/ 'memory', /shaaxrok/ 'memory-stimulating', /shaxron/ 'remembering'. The difficulties that make /shaaxr/ unusable carry over to /ashshaaxr/. In contrast, /srixtibhrangsho/ is available for 'loss of memory' quite straightforwardly: (14) a. srixti

'memory'

b. srixtibhrangsho

'loss of memory'

(15) a. jaati

'caste'

b. jaatibhrangsho

'fall from caste'

(16) a. buddhi 'intelligence' b. buddhibhrangsho

'sudden stupidity'

Evidently these forms instantiate a WFS such as (17), with X denoting an abstract quality or state and [Xbhrangsho] denoting its loss: (17)[X] N o u n ^[Xbhrangsho] N ( ; u n Against such viable competition, /ashshaaxr/ did not stand a chance. • 3.5. /apocaar/ 'corruption' vs /durniti/ 'corruption' I have found no instance of authors actually using the committee's proposal /apocaar/ for 'corruption'. In our times, the standard term for 'corruption' is /durniti/. But this term cannot be particularly old. Calontikaa glosses /durniti/ as /ashodaacoron, kuniti/ and glosses /kuniti/ as /ashodaacoron, durniti/, where all three words mean 'unprincipled or unethical behavior'. A term for corruption has to have a narrower semantic range than this. Calontikaa seems to reflect usage in a period when the language did not specify such a range. To begin with, even the word /apocaar/—while it does occur, as the equivalent for the English word 'corruption', in the appendix that makes available the labors of the terminologists—is defined in the body of the dictionary in a non-committal way, as /abihito aacoron/ 'inappropriate conduct', hardly a

148 · Anita Ravanam

specification of corruption. At a more detailed level, if we take up examples of corruption, it becomes important to note that Calontikaa has no term for 'nepotism': the term /shajontoshon/, now standard, is not recorded under /shajon/, under /toshon/, or separately. It describes 'bribery' (under /ghush/ and /utkoc/, two conventional terms for it) as an /aboidho puroshkaar/ or 'illicit reward'. But/aboidho/ 'illicit' is a general term of moral disapproval, used in Calontikaa also to describe, say, /beabhicaar/ 'fornication', glossed as /stripurusher aboidho sangsargo/, 'illicit contact between men and women', which exemplifies immorality but not 'corruption' in the strict sense. I presume that /durniti/ must have arisen as a subsequent coinage, possibly a creation of the mass media, and taken on the specific sense of 'corruption'. The term /apocaar/ in this sense had never taken root and therefore was not even available as a rival when the public had to decide whether to accept /durniti/. If there had been competition, /durniti/ would still have beaten /apocaar/. The well-established term /shuniti/ for 'good behavior' pairs with /durniti/ in the manner of the other pairs listed under (18) and formalized as a WFS in (19) where /shuX/ is a positive and /durX/ a corresponding negative term: (18) a. b. c. d. e. f.

shudin shujon shugandho shunaam shubuddhi shubodh

'good times' 'good man' 'fragrance' 'reputation' 'good intention' 'easy to understand1

g. shumoti 'good intention' h. shulokkhon 'good sign'

a', b'. c'. d'. e'. f.

durdin durjon durgandho durnaam durbuddhi durbodh

'bad times' 'bad man' 'stench' 'notoriety' 'bad intention' 'hard to understand' durmoti 'bad intention' g'· h'. durlokkhon 'ill-omen'

(19)[shuX] Noun ^[durX] Noun This situation makes the /durniti-shuniti/ pair easy for a speaker of Bangla to understand and remember. In contrast, there is no support for a basically new item like /apocaar/, which forms part of no Bangla set, even if those who already know their Sanskrit know this word as a term of moral disapproval. Things would obviously have been different if the Bangla set shown in (20) could have included /apocaar/: (20) a. b. c. d.

baad maan karmo ceshtxaa

'spoken intervention': 'honor' : 'deed' : 'attempt' :

a', apobaad b'. apomaan c'. apokarmo d'. apoceshtxaa

'slander' 'insult' 'misdeed' 'heinous attempt'

Three Levels of Lexical Codification m 149 e. debotaa 'deity' f. proyog 'use, usage' g. jash 'fame'

e'. apodebotaa' ghost or demon' f. apoproyog 'misuse, wrong' g'. apojash 'notoriety'

But /apocaar/ is unsupported by the pairings in (20) since no word /caar/ bearing an action or behavior related sense exists (the word /caar/ 'four' is clearly irrelevant to this discussion). In the other direction, /apocaar/ finds no home in (21) either: (21) a. b. c. d. e.

aacaar bicaar upocaar procaar shancaar

'custom, practice, pickle' 'judgement' 'ritual, object needed for ritual' 'propaganda, dissemination' 'flow'

These words have no identifiable meaning component in common. They provide no basis for any WFS that would have helped interpret /apocaar/. The details I have just surveyed suffice to make it clear that it is quite impossible to connect the candidate expression /apocaar/ to a supporting lexical environment in Bangla, whatever its support levels in Sanskrit. It is these factors that help us to understand the currency of /durniti/ and the failure of /apocaar/ to become current. • 3.6. /doho/ 'dairy' vs /gobboshaalaa/ 'dairy' The ground realities are that most speakers of Bangla would resort to an English loan and call a dairy a /dxeyaari/, neither a /doho/ nor a /gobboshaalaa/. However, of the two, /gobboshaalaa/ if used by others would be interpreted correctly as an attempt to refer to a dairy, whereas /doho/ would look opaque and not be understood at all. My task here is to discuss this difference between /doho/ and /gobboshaalaa/ with respect to intelligibility. The relevant word set is clearly (22): (22) a. b. c. d. e. f. g. h.

ashsho go hosti joggox paatxh otithi shoshsho nritto

'horse' 'cow' 'elephant' 'sacrifice' 'study' 'guest' 'grains' 'dance'

a', ashshoshaalaa b'. goshaalaa c'. hostishaalaa d\ joggoxshaalaa e'. paatxhshaalaa f . otithishaalaa g\ shoshshoshaalaa h\ nritto shaalaa

'stable for horses' 'stable for cattle' 'stable for elephant' 'place for sacrifice' 'school' 'guesthouse' 'granary' 'dancing-hall'

150 · Anita Ravanam This set implies the WFS (23), with the specifications that /X/ is normally a tatsama word and /Xshaalaa/ denotes a place for /X/: (23) [X] N o u n [Xshaalaa] Νουη Given (23) and the existence of the word /gobbo/ 'milk or milk product' it follows that the speaker needs no special training to understand /gobboshaalaa/ as a tatsama word formed along the lines of (22). However, this is a WFS that insists on the tatsama nature of /X/ and is thus distinctly confined to the learned stratum of the vocabulary. This excessive localization within the lexicon weakens the levels of potential lexical support for it in Bangla. On the other side of the equation, its competitor /dxeyaari/ gains strength from the general modern Indian preference for English loans over indigenous coinages as far as the names of objects and constructions are concerned. Modern Indian languages consistently prefer to borrow English terms for typewriters, screwdrivers, refineries, schools, and the like. This preference is especially strong in the case of relatively short words such as /dxeyaari/. Hence the demise of both the totally opaque /doho/ and the relatively viable /gobboshaala/ in favor of the foreign loan /dxeyaari/. Why was /doho/ proposed at all? It was supposed to strike a Sanskriteducated speaker of Bangla as related to /dohon/ 'milking' and /dugdho/ 'milk'. But, as we have seen throughout this study, such associations are in fact not clear to the mind of a Bangla speaker after the disappearance of Sanskrit from the education system. Thus, /doho/ is uninterpretable without special training. And 'dairy' is not the kind of concept for which one expects to have to be taught a special technical term. • 3.7. /nidhaar/ 'assessment' vs /mullaayon/ 'assessment' The word /mullaayon/ is now the standard term for 'assessment', but appears to have emerged after Calontikaa, which does not record it at all. The appendix records /nidhaar/ for 'assessment' in the Public Services section—proposed by the West Bengal government's post-Independence terminological committee and not by that of the University of Calcutta. But /nidhaar/ has never gained any currency in the general discourse. This pair calls for a WFS analysis. The winning candidate /mullaayon/ receives support from word set (24): (24) a. komol

'soft' :

a', komolaayon

b. shobuj

'green':

b'. shobujaayon

'annealing (chemistry)' 'greening, afforestation'

Three Levels of Lexical Codification m 151 c. rup

'form'

d. e. f. g.

'real, reality1: 'character' : 'industry' : 'Prakrit' :

baastob coritro shilpo praakrito

: c'. rupaayon d'. e'. f. g'.

'embodiment, fulfilment' baastobaayon 'implementation' coritraayon 'character acting' shilpaayon 'industrialization' praakritaayon 'prakritization'

The process terms on the right-hand side of (24) are seen as analogous to /mullaayon/ 'assessment, evaluation'. The WFS may be formulated as (25): (25) [X] NO™. Adject «-»(Xaayon] ê^^ Not only the term /mullaayon/ itself, but also the support-giving words /shobujaayon, baastobaayon, coritraayon, shilpaayon/ are new enough to have been omitted by Calontikaa. One observes shops in several towns in Bengal with names like /bastraayon/ from /bastro/ 'clothes', /dipaayon/ from /dip/ 'lamp, light', and so on. Even though these particular /Xaayon/ words have no precise meaning beyond 'some complicated activity that has to do with /X/', this seems to be enough to launch them as commercially viable names for shops engaged in the activities so designated. This is an indication of the lexical support level for WFS (25). In contrast, the committee-proposed /nidhaar/ for 'assessment', which appeals only to the speaker's ability to segment the word into Sanskrit morphemes, takes on the required meaning 'assessment' only if this meaning is taught to the user. The main body of Calontikaa does not record the word as a traditional Bangla word in any earlier sense. Even in Sanskrit the word is marginal—it is not recorded by Apte (1970)! No wonder, then, that it did not take off in modern Bangla. It is clear from the cases considered throughout Section 3 that the success of the word formation strategy analyses of the facts resides in their using the word, and not the morpheme, as the operative minimal unit of grammatical inquiry.

• 4. Non-uniform Lexical Entries The present section broadens the level of argumentation. I take up here the more general issue of how to characterize linguistically the contrast between an easy word and a difficult word. I spell this out formally in terms a particular theory of lexical entries. This theory translates the contrast between easy and difficult words into a technical distinction for the core cases. Note that Section 3 considered 'extremely difficult' or unintelligible words also (the stillborn

152 · Anita Ravanam

creations by the term planners); they are not the issue here. The essential contrast is between easy and relatively difficult items in the lexicon. The theory I am adapting to my purposes—called the Differentiated Entry Theory of the Lexicon (DETL)—can be contrasted with the currently prevalent Uniform Entry Theory of the Lexicon (UETL). I intend only to use the formal devices of DETL here, and not to accept the context in which it was first proposed by Dasgupta in his K.C. Dutta Memorial Lectures (1997). In other words, I do not commit myself to his psycholinguistic analogy with bilingualism, or to his pragmatic notion of paradigmatic speech act embedding. Furthermore, I do not commit myself even to his morphosyntactic suggestion that difficult words and difficult sentences share some identifiable formal feature. I am introducing the DETL in the new context of the lexical matrix theory that I proposed in Section 2. In the following exposition, I first present the core features of DETL itself and then embed it in the lexical matrix theory of Section 2. Intuitively, DETL differentiates light, uncoded, unmarked entries in a lexicon from heavy, coded, marked entries. It claims that the UETL, which is the default account, is mistaken in assuming that speaker S has equally ready and uncluttered access to all parts of S's linguistic knowledge (LK). Formally, DETL elaborates these intuitions by making two claims. One claim, working out the idea of heaviness, is that certain relatively heavy lexical entries, such as the Bangla word /barnaadxdxho/, are associated with their interpretation through a mediating gloss that uses relatively light words and with which the heavy entry has a relation of asymmetric synonymy. The second claim is about codedness. I shall come back to it presently. The theory DETL itself is only a framework and does not impose particular solutions for this or that problem. But, simply to give an example of what types of solutions are available for debate, I shall now suggest a particular description of the heavy item /barnaadxdxho/ in Bangla. For this word I suggest the gloss /naanaa range, rongin/ 'colorful (as opposed to black or white) with several hues'. Readers should feel free to replace this with their preferred alternative gloss if any. This particular gloss is not the point. My point is that the lexicon treats the light word like /rongin/ 'colorful' as less heavy than, as therefore capable of elucidating the meaning of, a heavy word /barnaadxdxho/ 'varicolored'. This formal relation between /barnaadxdxho/ and /rongin/ can be called asymmetric synonymy. The word /rongin/ 'colorful' is light enough to be an elucidator for /barnaadxdxho/ 'varicolored'. But /barnaadxdxho/ is too heavy to count as an elucidating synonym for /rongin/. For argument's sake, we may regard them semantically as being entirely synonymous with each other. But in lexical reality, only /barnaadxdxho/ has a synonym usable for the task of glossing it, /rongin/. In the other direction, /rongin/ has no such usable

Three Levels of Lexical Codification m 153 synonym, certainly not /barnaadxdxho/. For /rongin/ is a basic word in the Bangla lexicon. An imaginary speaker unfamiliar with /rongin/ cannot be helped by the use of other Bangla words. No lighter synonyms exist for this basic word of the language. I have furnished presenting the first formal claim made by DETL—the claim that the heaviness of heavy items involves asymmetric synonymy in the glossing relation. DETL's second formal claim concerns their codedness. A lexical item is code flagged as exemplifying a particular type of lexical difficulty. I adopt the DETL proposal that, formally, every lexically difficult item carries a code flag specifying its difficulty type. I suggest that the item /barnaadxdxho/ in the Bangla lexicon bears the code flag (Erudite Sanskritic). Thanks to this flag, it requires glossing. The gloss proposed earlier, /naanaa range rongin/ interprets it in terms of lighter words. I can now point out that these lexical items /naanaa/, /rang/, and /rongin/ carry no code flag. They are basic Bangla words, and therefore are not coded. So far I have presented certain aspects of the original version of DETL. Now notice that the contrast between the non-coded or basic kernel and the various coded sectors within the Bangla lexicon enables me to embed such an account within the theory of lexical matrices. I claim under the MLM regime the typical Bangla speaker S's lexicon Lex (S) is manipulated. The elite, using its power over the educational system, smuggles many actually erudite Sanskritic items into the uncoded kernel of the lexicon. They do this by successfully persuading each educated person to develop a complex and believe that some typical educated speaker does in fact find these items easy to understand. In contrast, PLM works with the actual Lex (S) of each S, which the elite's power cannot reach. In other words, an item like /barnaadxdxho/ falsely counts as basic in the elite-shaped imaginary typical Bangla speaker's Lex (S) under MLM. But PLM makes it possible to describe it correctly, as carrying the code flag (Erudite Sanskritic) for most speakers, barring a few highly Sanskrit-trained persons for whom it has become reclassified as a basic Bangla word. DETL does not totally link the first claim about glosses to the second claim about flags. Every lexically difficult item carries a code flag. I am only arguing that some lexically difficult items can have a satisfactory gloss that completely explains the word and eliminates its difficulty within the lexicon itself. Finally, I would like to consider a few advantages of the DETL account of lexical entries at the level of completing the transition from the structuralist view to the generative view of language in sectors that have not received the serious attention of most linguists. The main advantage is that a DETL-based theory of language can distinguish mediated from unmediated interpretation of expressions. It is also capable of distinguishing degrees of mediation.

154 · Anita Ravanam

The interpretation of kernel words and of expressions composed entirely of kernel items is direct or unmediated. Coded words and expressions featuring them have their interpretations mediated by glosses. Mediation is a matter of degree. If certain coded words serve as mediators for other coded words (either as glosses or as part of a multi-word gloss), then clearly these other coded words involve more mediation. These distinctions between the coded and the uncoded, and between the less coded and the more coded, are obvious in a DETL perspective. Thus the perspective makes linguists capable of quantifying degrees of coded distance from truly spontaneous and untutored use of language. This capability is relevant in the contemporary context of completing the generative project. Why does generative study need to be able to distinguish direct knowledge of some aspects of linguistic structure from indirect knowledge of other parts? Generative linguistics wishes to describe explicitly what native speakers acquire on the basis of Universal Grammar directly triggered by exposure to primary data. In so doing, a generative linguist needs to be able to distinguish such direct linguistic knowledge from indirect linguistic knowledge. The indirect material is taught to a speaker by society's cultural and educational systems. These indirect parts of one's knowledge carry the mark of having entered one's mind in the past through the mediation of teaching. In structuralist linguistics and its continuations in early generative grammars, all use of language was regarded as equally uncoded. This made all uses equally coded in effect. At this juncture, it may help to clarify that structuralism cannot sponsor DETL. Why? Because a Saussurean signifier is supposed to have a direct signified in linguistic form. To this directly corresponds a piece of extra-linguistic substance. So one should not have to go through any lexical mediation. In other words, DETL is an embodiment of the generative decision not to let any word, any linguistic sign, pretend to be a sovereign reality. Generative grammar takes seriously the dependencies between words—syntagmatic as in syntax and paradigmatic as in the serious generative study of the lexicon. It is DETL that brings out the latent capacity of structuralism's successors, namely sociolinguistics and generative grammar, to come out of the prison of the code. Then one can locate the base of a language in a truly open space, reformalizing codes in terms of how some words depend on other words. And these codes become the locus of lexical difficulty, which now has a characterization central to the format of the generative description of every natural language. This paper extends to sociolinguistics the range of descriptive success of the WFS approach. As in other cases, here too the success of WFS's resides in their being really based in words and not morphemes. I do not doubt that scholars who retain their faith in morphemes will propose to sweep all the data under the rug of the idea that some affixes are more 'productive' than

Three Levels of Lexical Codification · 155 others. But the trends which such scholars formalize in terms of productivity have long been known to reflect the interplay of extralinguistic factors. This paper offers an unpacking of some of these real factors, which turn out on scrutiny to operate in terms of words. Those who wish to seriously redescribe our facts in terms of a morphemic approach are welcome to try. No compelling arguments have ever been provided for treating 'productivity' as a grammatical feature. It is to be viewed as the extent to which all patterns are acceptable to speakers, measured along a continuum of several parameters, such as phonological, transparency, semantic opacity, and salience in terms of memory. At the same time it is to be borne in mind that all these linguistic properties must be calculated on a par with extralinguistic factors such as origin, user prestige, role and associated semantic phenomena. The present paper tries to use the theme of lexical difficulty as a key that helps open some of these extralinguistic doors while remaining wedded to specific linguistic assumptions. This approach will need to be tried out elsewhere to see if it holds up. • NOTE 1. On the issue of whether this mind must be individual, see Bourdieu (1993) for a sociological conceptualization of the transition from rules to strategies, of which this study has tried to clarify some linguistic aspects. The problem of strategies clearly has wider ramifications and goes beyond any simple dichotomy between individual and social, Bourdieu argues.

• REFERENCES Apte, Vaman Shivram. 1970. The Student's Sanskrit-English Dictionary. 2nd Edition. Delhi: Motilal Banarsidass. Basu, Rajasekhar (ed.). 1956. Calontikaa. Calcutta: M.C. Sarkar. Bourdieu, P. 1993. From rules of strategies. In other words: Essays towards a reflexive sociology, ed. by P. Bourdieu, 59-75. Tr. Matthew Adamson. Cambridge: Polity Press. Chomsky, N. 1965. Aspects of the theory of syntax. Cambridge: MIT Press. Dasgupta, Probal. 1997. The Presence of English in India at the Crossroads. K.C. Dutta Memorial Lectures, Department of History, Dibrugarh University, Assam, 9-11 November 1997. Ford, Alan, Rajendra Singh and Gita Martohardjono. 1997. Pace Panini: Towards a wordbased theory of morphology. New York: Peter Lang. Ravanam, Anita. 1995. Rival elites and controversy of styles: A comparative study of Bangla and Telugu. Hyderabad, University of Hyderabad: M.Phil, dissertation.

Syntax Learnability: The Problem that Won't Go Away* ANJUM P. SALEEMI Even to speak of Peter as having the I-language L is a severe simplification; the state of any person 's faculty of language is some jumble of systems that is no more likely to yield theoretical understanding than most other complex phenomena in the world. Peter is said to be multilingual when the differences among his languages happen to interest us for one or another reason; from another point of view, everyone is multiply multilingual (Chomsky 2000α: 169).

This paper outlines an approach to solving the learnability problem that takes parametersetting to be graded rather than an on/off process. It is argued that such graded or fuzzy approach is motivated by both linguistic and learnability considerations. The linguistic data relevant to the argument mostly pertain to the split Case/agreement system of Hindi-Urdu. It appears that a natural language as an object of learning may not be as parametrically consistent or uniform as is normally assumed under idealizations attendant upon the switch-setting model of parameter setting, as even an ideal monolingual native speaker must be shown to be able to learn a target language that may by syntactically split, perhaps along many dimensions. However, simple, binary parameters do not appear to be adequate to handle the complexity of the task in hand. What is, therefore, needed is a dynamic, adaptive model of learning syntactic variation that allows the inductive learner to generalize very conservatively, on a constructionby-construction basis, that is to say, only to the extent that particular syntactic constructions have been exhibited in the experiential input. This should be compatible with empirical data of both descriptive and developmental kinds as well as ensure learnability.

*An earlier version of this paper was presented at the 1996 Linguistics Association of Great Britain meeting at the University of Wales, Cardiff. I thankfully acknowledge the helpful comments I received from Stefano Bertolo, Janet Fodor, Dick Hudson, Andrew Spencer and many others.

158 · Anjum P. Saleemi

• 1. Introduction This paper attempts to address the problem mentioned in the title by adopting a novel approach. I argue that the problem in question will not go away unless: (a) We are willing to give up the assumption that languages, even when construed within an internalist perspective as I-languages in the sense of Chomsky (1986, and elsewhere), are coherent natural objects; (b) It is acknowledged that, as a corollary of the foregoing point, the distinction between monolingualism and multilingualism is just a matter of degree; (c) There is a recognition of the fact in research practice that the tension between descriptive adequacy and explanatory adequacy (leaving aside feasibility for the present purposes), in other words between particular languages and Universal Grammar (UG) or the initial state of the language acquisition device, is unresolvable as it is particular languages (constrained by any innately determined universal basis) which have to be learned by the child, and they are typically an outcome of an intricate interaction among psychological, social and historical factors; (d) A principled distinction is made between the human language learning system and UG, such that the latter is considered to constitute preexperience knowledge, whereas the former is held to embody a complex of many learning algorithms which can be employed as and when needed, as they are essentially equipped to handle incoming experiential data on a case-by-case basis; (e) Parameter fixation is treated not as an either/or binary mechanism but as a process whereby the choices made by the learner, given the partial and local effects (e.g., as in the case of dative movement in English) of grammatical subsystems underlying the epiphenomena of an ambient language, are graded somewhat in the manner of multi-valued or fuzzy logic, with the result that the options are selected gradually to make it unnecessary in most cases for the learner to subsequently unlearn nonlocal and massive overgeneralizations. (The dative alternation, once again, is a case in point; see, Mazurkewich and White [1984] and other relevant work.) While the quotation from Chomsky at the outset seems to acknowledge the above points explicitly or implicitly, in practice the view stated does not generally figure as the basis of most learnability research; e.g., see Bertolo (2001) for a state-of-the-art view, and the sources cited therein for most previous significant work in the area. The wisdom behind this practice, one should recall, is Chomsky's (1965) dominant view, which is contrary to that expressed

Syntax Learnability: The Problem that Won't Go Away · 159

in the foregoing quotation, that in order to achieve explanatory depth, linguistic analysis at the level of description has to give way to an idealized scenario built around a native speaker who is part of a monolingual homogeneous speech community. The problem inherent in the two views deepens further once one realizes that even a monolingual system may not be uniform, being more like a multilingual system in several respects. This means that, for the idealization to be truly fruitful, one might wish to simply and refine it by requiring that the object of investigation be a system that is parametrically consistent, or, in other words, is able to capture the details of a given target language in a way that can be summarized in the form a set of parametric choices that apply to the target language as a whole.1 In what follows these issues are investigated with reference to the wellknown split in the Case and agreement patterns of Hindi-Urdu, which is in part dependent on the aspectual system of the language in question.

• 2. Describing the Problem A significant problem for linguistic theory, which might be termed THE CARDINALITY PROBLEM, is that of determining the exact membership of the set of 'syntactically distinct' natural languages (Jain et al. 1999; Osherson et al. 1984, 1986). For the sake of the task in hand let us assume that any two languages are distinct from each other—modulo general lexical variation—if their grammars do not generate the same set of structural descriptions, or SDs. The consequent learnability problem is that of delimiting the variation space, such that it can be effectively searched by means of a memoryless enumeration procedure without access to negative input. Since it is supposed to permit only a finite and presumably small number of grammars, a syntactic model like the Principles-and-Parameters theory (Chomsky 1981; Chomsky and Lasnik 1993) that imposes non-trivial constraints on linguistic variation should be free from the learnability problem. However, it has been demonstrated that in effect the problem for this model, at least in the worst case, could be nearly as hard as it was for the standard theory (Clark 1992; Gibson and Wexler 1994). Most of the arguments are based on the potential ambiguity of input in respect of the parameter(s) that might need to be set; a related problem is due to the assumption that a cluster of syntactic variables may hinge on a single parameter, thereby making various syntactic properties, and indeed various parameters, so interdependent on each other that independent selection could become very uncertain or difficult. Independent selection of variables can of course be ensured by delinking them, and by making the child learner more sensitive to particular constructions. However, the current dogma in the generative paradigm is against construction-specific learning.

160 · Anjum P. Saleemi

and against too many independent parameters, since that would lead to a fragmentation of the theory (Safir 1987). While delineating the historical developments leading up to the advent of the Minimalist Program (Chomsky 1995 and subsequent work), Chomsky states: The research program sought to show that the complexity and variety are only apparent, and that the two kinds of rules [namely, phrase-structure rules and transformational rules: APS] can be reduced to simpler form. A 'perfect' solution to the problem of variety of phrase-structure rules would be to eliminate them entirely in favor of the irreducible operation that takes two objects already formed and attaches one to the other, forming a larger object with just the properties of the target of attachment: the operation we call Merge. Recent work indicates that this goal may be attainable (Chomsky 2000a: 13). Although the motivation underlying arguments against construction-specific learning is understandable, it does rest on a simplistic, over-idealized view of parameter setting, the so-called 'switch-box' model: to quote Chomsky yet again: We can think of the initial state of the faculty of language as a fixed network connected to a switch box; the network is constituted of the principles of language, while the switches are the options to be determined by experience. When the switches are set one way, we have Swahili; when they are set another way, we have Japanese. Each possible human language is identified as a particular setting of the switches—a setting of parameters, in technical terminology. If the research program succeeds, we should be literally able to deduce Swahili from one choice of settings, Japanese from another, and so on through the languages that humans can acquire. The empirical conditions of language acquisition require that the switches can be set on the basis of very limited information that is available to the child. Notice that small changes in switch settings can lead to great apparent variety in output, as the effects proliferate throughout the system. These are the general properties of language that any genuine theory must capture somehow (Chomsky 2000a: 8). One major problem with this view is that it considers the object of language acquisition to be a uniform, internally consistent entity. In spite of the assumption that acquisition via switch-throwing should make the Cardinality Problem trivial, that indeed is not the case. As we are well aware, in the worst case a UG containing 10 binary parametric choices yields 102 possible combinations, a formidably large learning space indeed! (See Bertolo 2001; Clark 1992;

Syntax Learnability: The Problem that Won't Go Away · 161

Gibson and Wexler 1994). In principle, constraining this parameter space should be possible, but, as pointed out earlier, that course of action is bedeviled by the unavoidable problem of how to make triggering unambiguous. I suggest that the triggering theory of parameter setting (or T-theory, to use a term adopted in Culicover 1995b) leads to such problems because it is committed to the view that construction-specific learning is not desirable. However, it seems that the latter view is exactly what is required to get rid of these problems. In particular, it does away with the notion of the object of learning being a unified countable entity, and therefore is able to dispose of the Cardinality Problem. The crucial implication is that a particular language, whether perceived extensionally or intensionally, is not a well-defined concept or a natural category, as the reverse idealization to multiple competence effectively destroys this possibility. Another problem with the switch-setting model is that it is not compatible with requirements of either descriptive adequacy or explanatory adequacy. It cannot account for the former because there is more often than not sufficient (but clearly not unlimited) inconsistency in any linguistic system that cannot be captured by across-the-board/all-or-nothing learning choices. It fails to achieve the latter since, as Chomsky remarks, every individual speaker is 'multiply multilingual', and also because the idealization to a monolingual picture cannot do justice to developmental data, and explain the intricate passage from the initial state of UG to the complex state of competence finally attained. It should be mentioned at the outset that a graded theory of acquisition of the kind advocated in this paper (somewhat akin to what Culicover [1995b] terms an adaptive or Α-theory) assumes that there is no ontological distinction to be made between the parser that develops in the child learner's mind and a grammatical representational system per se (although one presumes under certain circumstances it probably is legitimate to treat them distinctly). If learning a language L means building up a parser P for it, as I assume is the case (in part following suggestions in Culicover [1995a, 1995b]), then anything like parameter-setting must be preceded by an increasingly correct analyses of constructions exhibited at the experiential end of the acquisition process, AND eventually followed by a reconstruction of these (possibly in procedural or some other sort of associative memory) in a form in which they can be exploited online by the child learner in any context requiring either further analysis of the input or creative language use. If the acquisition process is idealized in such a way that a grammar G of L that is learned and the parser P corresponding to it are distinct, i.e., G * P, then the outcome of the assumption would be unnecessarily complex on grounds of parsimony as well as empirical adequacy, considering that if L is parametrically inconsistent (in a partial

162 · Anjum P. Saleemi sense) then neither G nor P can be parametrically consistent. An obvious consequence of this observation in regard to G, if it is correct, is that the learner's system would be unable to handle all the incoming data, and would instead endlessly go on setting and re-setting parameters for the language as a whole, whereas P by its very nature should be sensitive to letting the parametric effects spread through the language only to the extent that they have been exhibited as being relevant to some set of data from L. It is suggested that, given the redundancy of maintaining both P and G and the consequent lack of empirical and methodological dividends, something must give: either P or G, or (what seems equivalently, or perhaps more, logical) the distinction between the two. Gibson and Wexler (1994) and Bertolo (1995) consider a variety of issues bearing on the problem within a framework that in general assumes parameters to be binary and language-wide, and thus linguistic variation to be intrinsically simple, although its deductive consequences are known to be far from simple. Some other research (e.g., Saleemi [1992] on null subjects, Manzini and Wexler [1987] on binding) has taken a different view of variation, and multi-valued parameters have been proposed to deal with the problem of ostensibly graduated linguistic variation (however, see Hermon [1994] for an opposite account of binding, and Jaeggli and Safir [1989] for a similar perspective on null subjects; also see Law [1993] for a insightful survey of the cross-linguistic diversity). The type of solution that remains the most favored, implicit in Kayne (1994) and Chomsky (1995), relies as much as possible on more articulated syntactic proposals purported to radically limit the apparent range of variation, presumably reducing, at the same time, the burden on the language learning system of the child, interestingly, with similar implications for the developmental psycholinguist's ability to construct theories of acquisition (see Osherson and Weinstein 1995). Thus Chomsky (1995) suggests that the computational system of derivations (i.e., syntax) might be virtually uniform across the range of human languages, and that any variation that exists is either severely constrained or grounded in the morphology and the lexicon. The logic is that if syntax is almost uniform in this sense, then it should not have to be learned by the human child and the problem of learnability for syntax should disappear, at least in principle. In the same vein, Kayne (1994) outlines a radical proposal that underlyingly all human languages might have the same S VO word order, and that the attested variation can be explained by invoking movement rules. Since languages DO vary syntactically and at least some of this variation appears to be only partially systematic and thus not merely an epiphenomenal mixture attributable to interaction among parameters or their interaction with intrinsic constraints, a major contention of this paper is that an impoverished

SyntaxL·arnability:The Problem that Won't Go Away · 163

view of variation cannot do justice to the richness of linguistic diversity. Further, by removing any variation from one component of language (e.g., syntax) one can only succeed in relocating the problem to another component (e.g., the lexicon), without actually resolving it. A few variable syntactic features of UG, largely related to Case and agreement, are examined in the pages that follow by considering some data from Hindi-Urdu and Punjabi. These data suggest that neither the language-internal nor cross-linguistic typological differences among languages can be determined merely by fixing a small number of binary, language-wide parameters. Further, it appears that children's syntax develops strictly incrementally, rather than in a fashion compatible with a generalized on/off parameter-setting model (Culicover 1995a and b). Given the nature of linguistic variation (see Section 3) and the incremental pattern of syntactic development (Section 4), it might be the case that what is straightforwardly needed is either a 'cognitively uglier' conception of UG, or a greater amount of reliance on the child's language learning system, or perhaps even both, a set of issues that is dealt with in Sections 5-7.

• 3. Case and Agreement System An interesting case of both cross-linguistic and m/ra-language variation is embodied in the Case/agreement system. The following simplified crosslinguistic variation matrix will serve as our starting point. (1) a. Nominative-accusative (English). b. Ergative-absolutive (parts of Hindi-Urdu, Punjabi, etc.). The system of Case/agreement in English, which is uniformly nominative, is exemplified in (2). (2) a. / hit him with a book. b. He likes me a great deal. Some representative data from the more complex Hindi-Urdu (HU) system follow in (3). These are confined to two-argument constructions for the sake of simplicity. (3) a. LaRkee nee chiTHii likhii thii. boy ERG letter-SF write-PERF-SF be-PAST-SF2 The boy had written a letter/

164 · Anjum P. Saleemi

b. LaRkaa chiTHii likh chukaa thaa. boy letter-SF- write PERF-3SM be-PAST-SM 'The boy had written a letter/ c. LaRkaa chiTHii likh rahaa thaa. boy letter-SF write PROG-3SM be-PAST-SM The boy was writing a letter.' d. LaRkaa billii koo maartaa rahtaa thaa. boy cat ACC beat-HAB PROG-HAB-3SM be-PAST-SM 'The boy used to keep beating the cat.' As is clear from these data, the HU Case/agreement system, contingent upon the presence or absence of the perfective aspect and the choice of the aspectual forms, is split between an ergative subsystem, represented by (3a)— let us refer to this part of HU as HU*8—and a nominative-accusative subsystem, illustrated in (3bM3d), which may be termed HUNom. If the perfective aspect is manifested as an inflection on the main verb, then HUErg is the output of the system. If, on other hand, the perfective aspect is manifested as an auxiliary (like chukaa) independently of the main verb (3b), or is altogether absent (3c-3d), then the result is HUNom. (For further detail, see Saleemi [1994a]; see also Mahajan [1993], for an essentially Kayne-type typological perspective.) This partitions the relevant parts of HU in the following manner, where HUP = perfective HUNom, and HUN = non-perfective HUNom. (Note that a language is conceptualized here as a set of SDs, not as a set of strings.) (4) a. HU 2 HU** u HUNom b. HUNom 3 HUP u HUN It is shown by (4b) that HUNom is far from uniform, and consists of perfective (HUP) as well as non-perfective (HU N ) constructions. The ergative pattern of HU*8 is not very uniform either. In fact, the definition provided for it in the foregoing discussion is inadequate, as it does not cover at least one type of non-perfective construction that exhibits an ergative pattern of Case/agreement: an overt Case particle, the ergative nee or the dative koo (depending on dialectal/idiolectal variation), must appear on the subject of sentences containing embedded non-finite clauses of the kind illustrated in (5), yielding an essentially ergative pattern and creating a split within HUErg

Syntax Learnability: The Problem that Won't Go Away · 165 shown in (6) (where HUP' = perfective HU^*, and HUN = non-perfective HUEr*). (5) a. LaRkee nee [chiTHii likhnii] thii. boy ERG letter-SF write-INF-SF be-PAST-SF 'The boy was to write a letter/ b. LaRkee koo [chiTHii likhnii] thii. boy DAT letter-SF write-INF-SF be-PAST-SF 'The boy was to write a letter.' (6) HU** o HUP u HUN We have examined HU only in respect of its Case/agreement system. So what has so far been loosely referred to as HU is, in fact, a part of HU as a whole; let us call the former HUCA. We have shown that HUCA consists of the union of at least four sets of SDs, or sublanguages, identified in (7) by the symbols already used for them. (7) HUCA 2 HUP u HUN u HUP u HUN What is depicted in (7) is what the child learning the Case/agreement system of HU must (minimally) learn, namely, four interlinked sublanguages. Each of these consists of a set of closely related SDs or construction types, whose constitution in part rests on the choice of some crucial lexical items (e.g., the aspectual auxiliaries chuk, rah). Even further construction-specific variation is attested in the various dialects of HU and in other closely related South Asian languages. For instance, the ergative subset of Punjabi, in general similar to HUEr8, does not allow the ergative pattern with the first and second person pronominal subjects, as the contrasts in (8) clearly demonstrate. (8) a. MeiN chiTHii likhii sii. PRON-1S letter-SF write-PERF-SF be-PAST Ί had written a letter.' b. *MeiN nee chiTHii likhii sii.

166 · Anjum P. Saleemi

c. TuuN chiTHii likhii sii. PRON-2S letter-SF)write-PERF-SF be-PAST 4

You had written a letter.'

d. *TuuN nee chiTHii likhii sii. e. Us nee chiTHii likhii sii. PRON-3S f. *Us chiTHii likhii sii.3 The data here present a picture of variation that is by no means random or unsystematic, but hardly uniform or consistent, which means that the human language faculty tolerates this sort of lack of homogeneity within a linguistic system. Now let us have a cursory look at some acquisitional evidence, which presents a pattern of development that reflects (among other things) the child's inherent ability to learn any heterogeneous linguistic systems.

• 4. The Incremental Emergence of Child Syntax At least part of linguistic variation, it appears, may not necessarily be epiphenomenal, and could, in fact, be just what it appears to be, i.e., non-deductive variation within limits imposed by UG. As stated before, this is consistent with the usual pattern of language development. The developmental data seem to suggest that the child actually does not always fix parameters in a holistic fashion. Choices made by the child spread from SD to SD, even if the principle underlying the variation is theoretically unifiable. The following English (9) and Urdu (10) acquisition data illustrate the child's typical mode of gradual progress from short to long utterances, and from simple to complex structures. (9) a. Wiping. Gone. b. Baby talking. Reading book. Want paint. c. Me show mommy. Wayne sitting on gate. Her gone school.

Syntax Learnability: The Problem that Won't Go Away · 167 d. Want baby talking. Want press that. Want lady open it. e. These are red flowers. It was a shopping bag. Daddy gives me lots of money. f. I want to sit up there. I want to buy this boat. g. / think he ΊΙ play with the truck. I don't know where my car is. (10) a. gaaRii vehicle b. gaaRii aaii vehicle-SFcome-3SF The vehicle came/ c. gaaRii aa gaii vehicle-SF come PERF-3SF The vehicle has come/ d. abbu aa gae haiN daddy comePERF-3MH be-PRES-3H 4

Daddy has come/

e. abbu nee chaae pii In hai. Daddy ERG tea-F drink PERF-SFbe-PRES-S "Daddy has had tea/ f. meiN keh rahaa huuN meiN nee skool PRON-1S say PROG-SMbe-PRES-lS PRON-1S ERG school bus par baiTHnaa hai. bus on sit-INF be-PRES-S Ί am saying that I want to sit in the school bus/

168 · Anjum P. Saleemi

These examples, the set (10) selected from the naturalistic study of a child's acquisition of Urdu (Saleemi 1994b, 1995) and the set (10) containing very typical English acquisition data culled from various sources, suggest the undoubtedly and inevitably step-by-step nature of the emergence of structure, from one type of SD to another one, or from a simple to a complex SD of the same kind. Various maturational proposals notwithstanding (see Bertolo [ 1995] for a recent formalization of the various possibilities), the case for the incremental setting of parameters appears to be equally, if not more, plausible.

• 5. The Structure of Universal Grammar Is most syntactic variation of the type illustrated in the foregoing sections really construction-specific, or is it instead primarily reliant on lexicalmorphological variation and the 'rich deductive structure' of UG, as Chomsky (1981, 1986) claims (see also Atkinson 1992)? It is clearly possible to argue that such variation is indeed epiphenomenal, resulting from the interaction of lexical-morphological variation with the basic principles of syntax, or of one parameter with some other parameters or constraints, a viewpoint that rests on the (methodological) assumption that UG or the human language faculty is rigidly structured in a deterministic, deductive fashion. Let us label this view UGD, where the superscript D stands for 'deductive' or 'deterministic'. A natural corollary of UGD is that it should be learnable by a related minimal learning theory (LTM). However, we know that even some apparently very simple aspects of natural language variation may not be altogether free from potentially serious learnability problems (see Gibson and Wexler 1994; also Berwick 1994; Niyogi and Berwick 1994). Further, the deductive paradigm implies that much of the syntactic diversity is only deceptive, and is reducible to very few variable determinants of syntactic variation. However, extensive demonstration that that is indeed the case has not been forthcoming, at least not in terms of cross-linguistic, or even adequate intra-language, evidence. Therefore the sort of heterogeneity that has been sketched out earlier (note that we have considered only a couple of related languages so far, and very partially at that) is very likely to pose a very serious challenge to both UGD and LTM. Even if it turns out that linguistic variation is largely dependent upon simple, binary choices, it is by no means necessary that a language, basically defined as the output of an individual mental system, should always adopt one of the two choices uniformly in an across-the-board fashion. If languages can actually be split, non-uniform systems, then it would make a lot of sense for the child learner to be really conservative (that is, gradual) in the process of acquisition,

Syntax Learnability: The Problem that Won't Go Away · 169

and to impose any particular value of a parameter on the whole of a given target language only if he/she has encountered sufficient data to that effect. Data of the kind illustrated earlier may be taken to indicate that UG, though systematic enough, is not as neat and tidy as many of us would like it to be. Let us suppose, at any rate, if only for the sake of argument, that UG is indeed somewhat 'ugly' and 'fuzzy'. Then a richer learning system, compatible with what is known about the child's learning ability and the typical time-frame of language acquisition, should help solve the syntactic learnability problem. I shall talk about the possible shape of such a system in the next section. First let us try to conceptualize the idea of a 'fuzzy Universal Grammar', that, if along the correct lines, will necessitate a richer learning theory (LTR) that could replace LTM. If linguistic facts do actually compel us to conceptualize linguistic variation, as captured by parameters, to be gradient rather than of the simple binary (11) or multi-valued (12) type, at least in some cases, then it will be necessary to move towards an even 'looser' view of variation than that incorporated in a multi-valued parameter. Under this view the values for a FUZZY PARAMETER Pf may be a set like (13), obviously adapted from the fuzzy set theory. This view will make it possible to capture variation in a more elaborate manner, building upon the two extremes of the simple binary system presented in (11), but going well beyond it in its range of application. (Π) {0,1}. (12) {v1, v2, v 3 ,... V}. (13) {0,0.1,0.2,0.3,... 1}. The numeral 0 in (13) indicates a (possibly non-default or marked) value of a given Pf that is meant for a whole language L (e.g., none of the SDs in L is nominative-accusative), whereas the numeral 1 indicates that Pf is set at the diametrically opposite (possibly default) value, again for the whole of L (i.e., all the SDs are nominative-accusative). Any decimal fraction in between the two extreme values will suggest a partial setting at the nominative-accusative value applicable to an increasingly large subset of L, to be specified in relation to the size of the decimal in some precise manner. To conclude this paragraph, the fuzziness in (13) is a reflection of the size of the set of SDs {SDi, SD2, SD3,... SDn} in a language that are affected by the setting of a given Pt, such that one of the two extreme values (0 or 1) may apply to some subset of the language L, and not necessarily to L as a whole. Thus a fuzzy parameter P» may possibly have opposite values assigned to two distinct subsets

170 · Anjum P. Saleemi

A = {SDi, ... , SDm} and B = {SDj, ... , SDn} of the total number of SDs a language L contains, i.e., A u B c L. This implies that there are perhaps as many as three types of parameters, namely: (14) a. Binary; b. Multi-valued; c. Fuzzy. Needless to say, parameters of the type of Pf can deal better with the wellattested problem of partial systematicity in the syntax of many natural languages that have been relatively intensively investigated, and would therefore be descriptively more adequate in such cases. The foregoing suggestion should also contribute new tools to the explanatory armory of linguistic theory. Admittedly, it will remain largely speculative as long as a lot of convincing empirical evidence in its support is not made available. However, considering the amazing diversity of natural languages, many of which have been studied only insufficiently or not at all, a reasonable hunch is that a fuzzy theory of UG may not be too far wide of the eventual empirical mark.

• 6. The Structure of the Language Learning Theory As stated earlier, a major assumption in the orthodox generative paradigm of language acquisition is that the syntactic characteristics of an ambient language can be fixed in one fell swoop. Let us say that this follows from a learning principle, implicit in much of the research and in many parts of the foregoing discussion, that may be termed the comprehensive triggering algorithm (CTA) and defined as in (15). (15) Comprehensive Triggering Algorithm After examining the triggering sequence SEQT of n positive examples from a language L relevant to a parameter P, assume PVJ, the value of P exhibited by SEQT, to be for all the SDs included in L. It should be quite obvious that, in spite of being methodologically attractive, the CTA does not mesh with the developmental scenario. On the other hand, a relatively more 'intelligent' and 'richer' view of the child learner, incorporating restrictive access to certain more powerful learning strategies, may have a better chance of yielding the required results. A crucial aspect of the potential candidate for such as an enriched learning system that has just been outlined

Syntax Learnability: The Problem that Won't Go Away · 171 is that it is driven by the types of SDs characteristic of a language. These are deduced from the input strings, presumably on the basis of semantic, pragmatic and frequency factors, and stored as part of the current grammar of the child. Where simple parameter setting does not work, a certain amount of construction-specific learning may turn out to be necessary (see also Saleemi 1994a and b). The CTA may be undesirable in part because it can be incorrectly implemented, leading to over- or incorrect generalization, a familiar problem in the learnability literature. If the CTA has to be abandoned, at least in its strong form, it can be replaced with what may be called the partial or incremental triggering algorithm (PITA), stated here in a very general form. (16) The Partial or Incremental Triggering Algorithm A triggering sequence SEQT leads to partial or incremental adoption of generalizations (parameter settings and the like) only in respect of those particular construction types (SDs) that have been actually observed in the data from the ambient language. The PITA implies that the child learner typically generalizes cautiously, in a step-by-step fashion, not in one fell swoop. (See Culicover [1995a and b, 1999] for a comparable view.) An advantage of the PITA is that it might render maturational devices like Bertolo's MTLA (maturational triggering learning algorithm, adapted from Gibson and Wexler's TLA, i.e., triggering learning algorithm) redundant, relying on a partly input-driven rather than a largely deterministically driven learning system. Note that constructionspecific learning embodied in the PITA is compatible with the usually accepted view of the child as a conservative inductive learner, which sharply contrasts with the conception of the child as a biologically pre-determined learning system. That is why the PITA does not insist that the child automatically extend a parameter setting to all types of simple and complex constructions. Moreover, it is equally, if not more, consistent with the incremental nature of language development among children.

• 7. In Search of the Human Language Learning System Is there a specific human language learning system (HLLS) deployed by the child learner? What does it look like? Is it indeed more powerful than is claimed under the set of orthodox assumptions espoused under LTM? If the answer to this last question is yes, then the effectiveness of HLLS as a set of learning procedures will correspondingly reduce the burden of linguistic theory to

172 · Anjum P. Saleemi

explain away linguistic variation, in my view one of the desirable courses to pursue. It may be useful to point out that many of the reasons for adopting LTM are methodological, not empirical. The strongest minimal view of acquisition and learnability, primarily dictated by the methodology emanating from the logical problem of language acquisition, requires that there be NO learning whatsoever (conscious or unconscious) involved in the process of child language acquisition. Though posing intriguing intellectual problems and therefore heuristically very useful, this view has to buckle under the empirical pressure of the developmental problem of language acquisition when it comes to explaining the actual course of language development. The next major line of defence for LTM is to assume that maturational mechanisms can explain the developmental problem, without having to appeal to any type of learning (Bertolo 1995; Borer and Wexler 1987). This view must also be eventually given up, as acquisitional transition seems to be considerably more gradual than conceived within the maturational framework and its allied linguistic assumptions. I think the generative approach to natural language is beset with an apparent foundational paradox: it claims first language acquisition and learnability arguments to be central to the task of formulating linguistic theory, but at the same time it regards acquisition/learning to be undesirable. It is usually asserted that a language (in the internalistic sense) is hardly an acquired entity, and that it comes into being because the human language faculty biologically grows and matures, with the role of input being merely that of triggering. This account of innatist epistemology may be termed biological nativism (BN), and may be considered to be a particularly strong version of the thesis of strong nativism (SN) (Osherson et al. 1984), which postulates that the number of realizable human language systems is finite. Clearly, several versions of SN are possible, of which BN is perhaps the strongest. The orthodox theory of parameters is a crucial aspect of BN, since a parameter is conceived not just as a particular aspect of surface variation but a variable concretely and deeply rooted in the language faculty and moreover one which is responsible for many facets of language acquisition and development. The arguments presented in this paper suggest that the empirical evidence (developmental as well as structural) does not at all compel one to choose BN over one of the possible weaker versions of nativism. It appears that, following the behaviorist debacle, the generative pendulum swung completely to the opposite direction, in the sense that the behaviorists assumed a sweeping theory of learning that was too general to be even remotely viable in the acquisition of the extremely rich systems of knowledge characteristic of human beings, whereas the generative theory has got caught on the other horn of the dilemma by claiming that the knowledge systems in the mind are so articulated prior to

Syntax Learnability: The Problem that Won't Go Away · 173

any experience that the role of any learning must be minimal, if not entirely absent. It is perhaps time now to redress the balance in the light of empirical findings, if we want the problem of syntax learning in particular, and language acquisition in general, to go away—eventually, that is. The likely measures along these lines may necessitate modification(s) in the standard assumptions about (a) linguistic structure, (b) the child's learning system, or (c) both.

• 8. Concluding Remarks Some caveats, explanations and reiterations are in order here. First, although I have relied largely on Hindi-Urdu data, it should not be too hard to find similar examples of absence of total systematicity in many languages. One gets the impression that inconsistencies of the kind which have been under discussion in the foregoing pages are a dime a dozen in the variegated world of languages. Second, a general caveat but one that is crucial to the validity and soundness of the argument sketched out in the foregoing pages: Nothing that I have said implies that languages are just a jumble of uncontrollably arbitrary bits and pieces. One could claim that languages are hopelessly fragmented only on pain of gross overstatement. Languages appear to be an optimal combination of order and arbitrariness (see Pinker [1999] on this and related issues) as if the mind does not like either too much order or too much arbitrariness. In particular, it seems to balk at rules that apply language-wide. It is reasonable to assume, in fact, that if languages were entirely and incoherently arbitrary (whether due to almost random external factors such as historical contingencies or purely mind-internal, psychological factors), they would not be learnable at all, though they would still in principle be memorizable, given absurdly huge memory resources. Whatever the provenance of a particular instance of arbitrariness, it goes without saying that the mind is indeed a rule-making and rule-imposing device, that also has considerable, but not unlimited, capacity for memorization. It should be stressed again that the implications for the cardinality problem are radically different under the switch-box and the graded/mixed models of variation and learnability. On the latter view a particular language is not a well-defined notion, and therefore cannot be an object of learning in deep, psychological terms; an obvious consequence of such a view is that the concept of cardinality loses its relevance and therefore ceases to be a legitimate problem. To sum up, Chomsky's statement quoted at the very outset of this paper is obviously a far cry from his well-known methodological strategy that relies

174 · Anjum P. Saleemi

upon an idealization requiring simplification of linguistic data to the competence of a hypothetical native speaker belonging to a homogeneous monolingual community. It is easy to see the attractive reasoning underlying such a strategy, as putatively it can facilitate our quest for explanatory adequacy. (It is worth reminding the reader that the classic view of explanatory adequacy is based upon an empirically-adequate description that is learnable; see Chomsky [1965], and elsewhere.) However, the problem is that even in the context of the individual psychology of a native speaker acquiring a language, the system that emerges and is deployed in use is often likely to be resolutely recalcitrant to the idealization in question. This is so because particular languages, though consistent enough to be identifiable as distinct objects in common sense and to some extent technical terms, are, strictly speaking, rather heterogeneous and inconsistent entities. It has not been the intention of this paper to suggest by any means that the traditional generative paradigm should be abandoned altogether, as the arguments presented herein are merely suggestive.4 Our modest goal has been to point out that there are other potentially viable avenues that should not be left unexplored. In the end, it is hoped that this paper has managed to drive home at least one point forcefully enough, namely, that the battle between descriptive and explanatory adequacy is far from over. • NOTES 1. I am of course leaving aside the issue of an even greater amount of inconsistency exhibited in the lexicons of particular languages. 2. The abbreviations used in the glosses are explained below. S = singular F = feminine M = masculine 1/2/3 = I*1/ 2nd/3rd person PRON = pronoun

NOM = nominative ACC = accusative ERG = ergative DAT = dative H = honorific

PERF = perfective aspect PROG = progressive aspect HAB = habitual aspect PRES = present tense INF = infinitive

3. Note that (8f) will be considered well-formed in some dialects of Punjabi. 4. As remarked by Richard Hudson (personal communication), construction-specific variation (or learning) may be further broken down into word-specific variation, as in word grammar, where words are associated with treelets, which in turn combine to give rise to more complex phrasal and clausal constructions. Though somewhat less obviously, this sort of approach is also seems to be of a piece (at least in spirit) with the Minimalist Program (Chomsky 1995, 2000b).

• REFERENCES Atkinson, Martin. 1992. Children's syntax: An introduction to principles-and-parameters theory. Oxford: Blackwell.

Syntax Learnability: The Problem that Won't Go Away · 175 Bertolo, Stefano. 1995. Maturation and learnability in parametric systems. Language Acquisition 4(4). 277-318. (ed.). 2001. Language acquisition and learnability. Cambridge: Cambridge University Press. Berwick, Robert. 1994. Cartesian computation and linguistics in a current context. Current Issues in Mathematical Linguistics, ed. by C. Martin-Vide. Amsterdam: Elsevier. Borer, Hagit and Kenneth Wexler. 1987. The maturation of syntax. Parameter setting, ed. by Tom Roeper and Edwin Williams. Chomsky, Noam. 1965. Aspects of the theory of syntax. Cambridge, MA: MIT Press. . 1981. Lectures in Government and Binding. Dordrecht: Foris. . 1986. Knowledge of language: Its nature, origin and use. New York: Praeger. . 1995. The Minimalist Program. Cambridge, MA: MIT Press. . 2000a. New horizons in the study of language and mind. Cambridge: Cambridge University Press. . 2000b. The architecture of language, ed. by Nirmalangshu Mukherjee and B. N. Patnaik. New York: Oxford University Press. Chomsky, Noam and Howard Lasnik. 1993. The theory of principles and parameters. Syntax: An international handbook of contemporary research, Vol. 1, ed. by J. Jacobs, A.V. Stechow, W. Sternefeld and T. Vennemann. Berlin: Walter de Gruyter. Clark, Robin. 1992. The selection of syntactic knowledge. Language Acquisition 2(2). 83-149. Culicover, Peter. 1995a. Adaptive learning and concrete minimalism. Ohio State University, ms. . 1995b. Adaptive learning and concrete minimalism. Proceedings of GALA. Unpublished. . 1999. Syntactic nuts: Hard cases, syntactic theory, and language acquisition. Oxford: Oxford University Press. Gibson, Edward and Kenneth Wexler. 1994. Triggers. Linguistic inquiry 25(3). 407-54. Hermon, Gabriella. 1994. Long-distance reflexives in UG: Theoretical approaches and predictions for acquisition. Syntactic Theory and First Language Acquisition: CrossLinguistic Perspectives, Vol. 2: Binding, Dependencies, and Learnability, ed. by Barbara Lust, G. Hermon and J. Komtilt. Hillsdale, NJ: Erlbaum. Jaeggli, Oswaldo and Ken Safir. 1989. The null subject parameter and parametric theory. The Null Subject Parameter, ed. by Oswaldo Jaeggli and Ken Safir. Dordrecht: Kluwer. Jain, Sanjay, Daniel Osherson, James S. Royer, Arun Sharma. 1999 (2nd Edition). Systems that learn: An introduction to learning theory. Cambridge, MA: MIT Press. Kayne, Richard. 1994. The antisymmetry of syntax. Cambridge, MA: MIT Press. Law, Paul. 1993. On null subjects and null arguments. Canadian Journal of Linguistics 38. 1-41. Mahajan, Anoop. 1993. The ergativity parameter: have-be alternation, word order and split ergativity. Proceedings of N ELS 1993, University of Massachusetts. Manzini, Rita and Kenneth Wexler. 1987. Parameters, binding theory, and learnability. Linguistic Inquiry 18. 413-44. Mazurkewich, Irene and Lydia White. 1984. The acquisition of the dative alternation: Unlearning overgenerelizations. Cognition 16. 261-83. Niyogi, Partha and Robert Berwick. 1994. Learning from triggers. Center for Biological and Computational Learning, MIT, ms. Osherson, Daniel and Scott Weinstein. 1995. On the study of first language acquisition. Journal of Mathematical Psychology 39. 129—45. Osherson, Daniel, Michael Stob and Scott Weinstein. 1984. Learning theory and natural language. Cognition 1. 1-28.

176 · Anjwn P. Saleemi Osherson, Daniel, Michael Stob and Scott Weinstein. 1986. Systems that learn: An introduction to learning theory for cognitive and computer scientists. Cambridge, MA: MIT Press. Pinker, Steven. 1999. Words and rules: The ingredients of language. London: Weidenfeld and Nicolson. Roeper, Tom and Edwin Williams (eds). 1987. Parameter setting. Dordrecht: Reidel. Safir, Ken. 1987. Comments on Wexler and Manzini. Parameter setting, ed. by Tom Roeper and Edwin Williams. Saleemi, Anjum P. 1992. Universal Grammar and language learnability (Cambridge Studies in Linguistics 61). Cambridge: Cambridge University Press. . 1994a. If that's the Case, I don't agree: Case, agreement and phrase structure in HindiUrdu. Papers from the Fifteenth South Asian Language Analysis Conference 1993, ed. by A. Davison and F. Smith. Iowa City: South Asian Sudies Program, University of Iowa. Pp. 277-95. . 1994b. Derivational constraints in early Urdu syntax. Paper presented at the Eighteenth Boston University Conference on Language Development. . 1995. On the acquisition of split ergativity: some evidence from Urdu. The Proceedings of the Twenty-Sixth Annual Child Language Research Forum, ed. by Eve Clark. Stanford: CSLI. Unpublished.

On the Modal Meanings of the Subjunctive in Hindi GHANSHYAM SHARMA

The paper deals with the modal meanings of the Hindi subjunctive in the light of an overall picture of the Hindi verbal predicate. It argues that modal meanings are the speaker's meaning which may or may not be grammaticalized, and therefore it is difficult to provide a full picture of modal meanings of the subjunctive in syntactic terms only. Furthermore, the paper establishes that the Hindi subjunctive is employed by the speaker to express both epistemic and deontic possibilities rather than epistemic and deontic necessities.

• 1. Introduction In the present paper we make an attempt to analyze various kinds of modal meanings expressed through the subjunctive verbal predicates in Hindi.1 In doing so, we will be concerned mainly with the epistemic and deontic aspects of the subjunctive forms of the Hindi verb and will not investigate in any detail their possible syntactic characterizations. In order to establish modal meanings of the subjunctive verbal predicate, we shall try to see it in the light of an overall picture of other verbal predicates which express either epistemic necessity and possibility, or deontic necessity and possibility. The paper seeks to establish that the modal meanings in natural languages are expressed at different levels of the utterance and therefore can be studied from different perspectives, that is, from the point of view of the speaker, from the point of view of the person referred to, or, in the case of reported speech, from the

178 · Ghanshyam Sharma

point of view of the speaker referred to in the utterance. The present study, however, concerns modality expressed at the level of sentence only and does not take into consideration modality which is related to the persons mentioned or referred to in the utterance. In accordance with the functions of the subjunctive in many languages of the world, the Hindi subjunctive is used by the speaker to express meanings relating to modal possibilities rather than to express modal necessities which are generally conveyed by the Hindi indicatives. Furthermore, the paper claims that in Hindi this dual distinction between possibility and necessity can be found at the level of deontic modality as well. A speaker can, for example, ask his addressee to carry out an action ÷ either necessarily or possibly. In the former case, the speaker issues a command through an imperative, asking the addressee to carry out the task necessarily, whereas, in the latter, the speaker gives the addressee the possibility of noncompliance with the command. Such meanings are articulated by polite commands made with the subjunctive.

• 2. What do We Mean by Modality? The term modality has been used in different senses in linguistics to indicate those different kinds of additional semantic elements which are attached by the speaker to a sentence or utterance by different means in different languages of the world. At times, the term is used to indicate any grammatical elements which do not fit into any known categories, even to indicate sentence types such as interrogative,2 etc., whereas sometimes it is used in a very technical and limited sense, taking into consideration grammatical categories which express epistemic and deontic modalities only. The term modality is used differently also with respect to its scope; at times it is related to an agent referred to by the hearer and at times it is considered in relation to the sentence as a whole. Keeping in mind this diversity of uses of the term 'modality', Van der Auwera and Plungian (1998) presents a classification of different types of modalities according to which the necessity and possibility elements can be found both at sentence level (participant-external level) as well as at the level of the participants (participant-internal level) of the discourse.3 It is this diversity of the meaning of the term modality that is responsible for its having been neglected in linguistic research for a long time. Though the origin of modality-based research can be traced back to grammarians such as Jespersen4 (1924) and logicians like Von Wright5 (1951) and Rescher (1968), it is Lyons (1977) and Palmer (2001), however, who constitute a systematic beginning of research on modality, especially in linguistics. In the present study, by modality we mean those semantic or grammatical elements of a sentence or

On the Modal Meanings of the Subjunctive in Hindi m 179

utterance — similar to aspect, tense, etc. —which indicate the way in which the speaker is committed to the truth of the proposition contained in a sentence (or utterance). The tense is thought to indicate the time reference of the action in relation to the time of utterance and the aspect indicates the internal temporal constituency of the action;6 modality is concerned primarily with the way in which the speaker commits himself to the addressee about the truthfulness of the action(s) or any state of affairs reported in the sentence. In other words, at times the speaker is in a secure position of asserting the states of affairs described in the sentence since he knows them, at other times, he cannot with certainty share with the addressee the information about the action(s) or state(s) of affairs contained in the sentence. In such cases he merely informs the addressee of the possibility that they may or might take place or might have been taking place. Furthermore, the speaker, at times, does not know about the states of affairs, but rather believes that they must necessarily be taking place or will certainly take place. The veridicality of the proposition, therefore, is the sole factor in recognizing the modality of the sentence. We will keep to this view throughout the present paper and will be interested in analyzing sentence level modality only, leaving the topic of participant-internal modality for further studies. • 2.1. Different Layers of Modality in Utterances As stated earlier, modality should be studied at least at three levels: at the level of the speaker, at the level of subject of the utterance and at the level of the person mentioned. Theoretically, these levels could be even more than three, but it will suffice to say that it is not sufficient to provide syntactic or semantic explanations for the modality operators just at the level of sentence. Sentences do not always show all the layers of the modality at their surface level and hence no theory can be provided to account for all the semantic elements of modality. We propose to make a distinction between these different layers or tiers of modality and will be representing them with brackets: [Layer 1], [Layer 2], and [Layer 3], etc. To elaborate this topic, it would be useful to consider a simple proposition John has left Venice: Uycr2

Uycr3

[I know that [l believe that

Ijiycr3

Uycr3 Uycr3

[I know that [I know that

Uycr2 Uyer2

[I know that believe that

[Mary says that [Mary says that

^"'[John has left Venice.]] [John has left Venice.]]

Uycrl

^"'[John has left Venice.]]] [John has left Venice.]]]

Uycrl

Laycr2

Uycrl

Uycr2

Uycrl

[Mary knows that [Mary believes that

[John has left Venice.]]] [John has left Venice.]]]

180 · Ghanshyam Sharma U cr3

y [l believe that ^^[Mary knows that [I believe that ^^[Mary believes that

Uycr3

u crl

y [John has left Venice.]]] ^«é[John has left Venice.]]]

Uyer3

[I know that uyer2[Mary is unhappy that ^âÃÀ[John has left Venice.]]] [I believe that ^^[Mary is unhappy that uyerl[John has left Venice.]]]

Uycr3

Now, to check the validity of any theory which wants to give an account for subjunctive-indicative divide, for example, one has to see at which layer realisirrealis contexts are attached to the proposition. It is possible that a sentence has only one layer and modality elements are attached to the proposition at that layer. It is also possible that when a sentence contains different layers the modality items are attached to one of the layers. Generally speaking, if a sentence contains more than one layers—whether hidden of evident—it is the outermost layer (the second or the third) which is likely to contain modality elements, never the innermost (the first) one.

• 3. The Overall Structure of Modality at the Level of Verb in Hindi Unlike languages such as English, where modality is mainly expressed through modals, Hindi has a verbal system which is organized to express different modalities in conjunction with tenses and aspects. Some Hindi moods, for example, are coded in the auxiliary verb which carries a tense as well, while others are coded in the main verb without an auxiliary and hence without any time reference. There is not, however, a clear-cut morphological system which can be thought to be related directly to the mood system in Hindi. Bearing in mind this complexity of the modality system in Hindi, Agha (1998: 123-26) maintains that moods in Hindi should be considered as a derivative category cluster of either different minimal mood categories or as a category in combination with tense and other elements of the verb. In considering the morphological structure of the verb, he further claims that there are only three minimal mood distinctions in Urdu-Hindi,7 namely, (a) [+/- imperative]; (b) [+/potential]; and (c) [+/- prospective], but we believe that, although this proposal sounds valid on the morphological level, it would be illogical not to recognize the Hindi modality system in its full detail, irrespective of whether all the Hindi moods are coded morphologically or not. For a full picture of the modality system in Hindi,8 we provide, below, an overall formally-organized structure of the Hindi verbal predicate.9

C/3

4_»

Ï

§· ^

Aspect III Perfective

|| ^> ;*.

^

R *-

*

c

| t l j i ^ - .s s t !^s *C 1 jc= ^

^

S

S

—^ , icd

^> ö>

^dS?

*«f

^f £ C

^ ?

** icd =3 -C * c3 PQ >

c

^ « * ~ «S â 3? o-S

l

58

1i!!! oil?

! ted ^

• -B ¢ ffll?

«S

D

I -" M ^)

I -S _Q

.£c*a T 2.

als -fc «cd g

00

i l &s CO

< DC

O

i! g, 3 CA cd

S: á-ß &

.

Ss '

s? *^ S "

It I C/5

!

-1« «

M

^«,

l ä A' 52 '«

ï

- j s 311

co cd

5 · -q « 's; g ^"3 9

g*· o»l

II

8 -g

>>. .»N *;

ßË