Typological Variation in the Ergative

Morphology of Indo-Aryan Languages�

Ashwini Deo [adeo@stanford.edu]

Devyani Sharma [dsharma@stanford.edu]

Stanford University

August 2002

Abstract

While New Indo-Aryan languages are a common example of morpho-logical ergativity, the range of variation in ergative marking and agreementamong these languages has not been examined in detail. The goals of thispaper are twofold. We �rst present a detailed typology of ergative markingand agreement in Indo-Aryan languages, demonstrating that a progressiveloss of ergative marking has occurred to varying degrees diachronically indi�erent systems. This process is manifested in two distinct strategiesof markedness reduction: loss of overt subject marking in the nominaldomain and loss of marked agreement in the verbal domain. Second, wepresent a formal account of this typological range within the framework ofOptimality Theory (OT; Prince and Smolensky 1993) by developing con-straints on case-marking (Aissen 1999; Woolford 2000) and agreement interms of universal subhierarchies of markedness. An extension of this anal-ysis to dialectal variation in one language, Marat.hi, demonstrates that thedialectal typology parallels the cross-linguistic typology within the rangepermitted by the parent language (Old Marat.hi). This dialectal typologyalso furnishes more evidence of intermediate systems predicted by the OTanalysis. Finally, the synchronic typological analysis of languages anddialects is correlated with the diachronic stages of loss of ergativity.

�This work is based in part on work supported by the National Science Foundation underGrant No. BCS-9818077. We wish to thank Judith Aissen, Joan Bresnan, Paul Kiparsky,Hanjung Lee and Peter Sells for their insightful comments and suggestions. Earlier versionsof this paper were presented at the parasession on South Asian languages at the WesternConference on Linguistics (WECOL, Fresno, October 2000) and the Stanford/UCSC JointWorkshop on Optimal Typology (Stanford, December 2000). We are also indebted to theaudiences of both sessions for valuable input. Any remaining errors are our own.

1

Introduction

New Indo-Aryan (NIA) languages are commonly presented as an example ofmorphological ergativity. A less commonly noted fact is that ergative markingand agreement patterns are not uniform across these languages. The overtmorphological expression of ergative case marking occurs to varying degreesin their nominal paradigms; variation is also observable in the ways in whichagreement morphology cross-references arguments. Additionally, the languageswhich share ergative case-marking patterns do not necessarily share agreementmarking patterns, resulting in an intricate cross-classi�cation across systems.

This study �rst presents the range of variation in case and agreement mark-ing in this language family. We derive the distinct systems of ergative casemarking and agreement in di�erent NIA languages (Hindi, Nepali, Gujarati,Marat.hi, and Bengali) as consequences of strategies to reduce the morpho-syntactic markedness of the ergative paradigm | both within the nominal andverbal domains. Following this, we discuss dialectal variation within one of theselanguages, Marat.hi. The dialect data strengthens the hypothesis that dialectalvariation mirrors cross-linguistic variation (Bresnan and Deo 2001), but withinthe typological constraints set out by the parent language. In other words, thedialectal data indicates that a directionality is inherent in progressive loss ofsubject-marking. This is evidenced by the fact that no Marat.hi dialect revertsto a more di�erentiated subject-marking system than Old Marat.hi; the dialectsonly vary in terms of greater reduction of subject-marking than the parent sys-tem. Finally, we relate the current typological range to the diachronic pictureof change in the Indo-Aryan system by locating each language at a di�erentstage along a diachronic trajectory that reduces the markedness of marked con-structions (in this case, the ergative clause). Our analysis shows that the rangeevident across all these systems does not re ect a set of idiosyncratic trajec-tories of change, but rather derives from the in uence of universal markednessconstraints on subject marking and agreement.

The analysis is framed in Optimality Theory (OT; Prince and Smolensky1993) and employs language-particular rankings of universal constraints (Princeand Smolensky 1993; Aissen 1999) which allow an interaction of nominal mark-ing, verbal in ection, and universal markedness to derive the distinct systemsfrom the historically prior Middle Indo-Aryan (MIA) system. One trajectoryfor reducing the markedness of the original MIA ergative clause is a gradual lossof overt marking of agentive, perfect subjects. Hindi, Gujarati, Marat.hi, Pun-jabi, and Bengali are treated as successive diachronic stages in the the loss ofsubject marking, achieved by the demotion of universal markedness constraints.Another strategy, found in Nepali, is to maintain subject marking but insteadreduce the markedness of the verb agreement system.

In x1, we lay out the historical foundations of morphological ergativity inIndo-Aryan languages, examining the marked features of the MIA ergative con-struction. Following this, in x2 the �ve main types of subject-marking and agree-ment systems are presented, and brie y summarized in a cross-classi�cation(x3). We present our OT analysis in x4 and x5. This analysis is then extended to

2

Marat.hi dialectal variation in x6{x8 to demonstrate the similiarities in the typo-logical distribution of systems within a language and across languages. Finally,in x9, we relate our typological analysis to the diachronic stages in Indo-Aryanlanguages and summarize our �ndings.

1 The Ergative Construction in Indo-Aryan Lan-

guages

1.1 Ergativity and the Domain of the Present Study

The shift from an accusative-type to an ergative-type morphology in a languageis generally connected to a passive construction which gets reanalyzed as anactive one (Plank 1979; Dixon 1994; Peterson 1998). Standardly, ergativity hasbeen conceived as of as follows:

(1) a. A grammatical pattern or process shows ergative alignment if it iden-ti�es intransitive subjects (Si )and transitive direct objects (dO)as op-posed to transitive subjects (St ).

b. It shows accusative alignment if it identi�es Si and St as opposed todO. (Plank 1979)

Split-ergativity refers to the occurrence of the ergative type of grouping ofarguments in certain (usually) syntactic-semantic con�gurations, namely withhigh animacy objects, or in the perfect aspect. The type of ergative-split foundin the New Indo-Aryan language is aspect-based. Transitive clauses in the per-fect aspect and the past tense1 show ergative morphology. The subjects ofintransitive clauses and direct objects of transitive clauses are marked nomina-tive.2

In this paper, we restrict ourselves to the patterns of variation found intransitive, perfect clauses. Indo-Aryan languages do not show much variation inthe subject marking and agreement patterns of subjects of intransitive, perfectclauses, or in the marking of intransitive and transitive subjects of non-perfectclauses. In these cases, the highest argument is marked nominative and is cross-referenced by the verb as the highest, nominative argument. It is cruciallyin the ergative clause that morphological and syntactic prominence are notaligned, leading to a marked construction. We will argue that the variationin Indo-Aryan languages shows a typology of possible strategies to render thisconstruction unmarked.

1The past tense in the modern languages derives historically from the perfect.2This is a simpli�cation, as the case-marking facts are much more complex, depending on

both the properties of events (volitional vs. nonvolitional, agent-controlled vs. non-agent-controlled) and the properties of arguments (animacy and de�niteness) (Butt 2000; Mohanan1994). But these facts do hold unambiguously in canonical transitive and intransitive clauses.

3

1.2 The Chronology of the Indo-Aryan Language Family

Hindi, Nepali, Gujarati, Marat.hi and Bengali, the languages discussed in thispaper, are Indo-Aryan languages and arise from a common Old Indo-Aryanancestor, Sanskrit (or a related dialect). The language typology that we arelooking at is that of historically related languages, which moreover inheritedmorphological ergativity from Middle Indo-Aryan dialectal variants of Sanskrit.The synchronic patterns of variation that our account analyzes must thereforebe regarded as the result of either retention or innovation in the domain ofergative morphology within each of the daughter languages.

The table in (2) gives an overview of the periods and languages associatedwith the periods of NIA, MIA, and Old Indo Aryan (OIA).3 It re ects thegenerally accepted classi�cation and chronological position of the Indo-Aryanlanguages. The three periods are characterised by distinctions in the perfectconstruction that bear on the typology presented in this paper. We discussthese diachronic stages in greater detail in x9.

(2) Chronological Progression of the Indo-Aryan Languages

Period Linguistic stage TermUntil 600 BC Vedic600BC-200AD Epic and Classical Sanskrit Old Indo-Aryan (OIA)200 BC-700AD P�al�� and Pr�akrit languages700-1100AD Apabhram�sa Middle Indo-Aryan (MIA)1000AD- present Nepali, Bengali, Marat.hi, Hindi New Indo-Aryan (NIA)

1.3 The Emergence of the Ergative Clause

Aspect-based split ergativity in Indo-Aryan (IA) languages has been presentedas a classic case of the passive to ergative reanalysis seen cross-linguistically(Dixon 1994). A concise summary of the process may be outlined in the followingway: Old Indo-Aryan (OIA) did not have an active, ergative construction. Theoriginal construction that gave rise to the ergative clause in the NIA languageswas, in OIA, a passive, periphrastic perfect construction, which involved the useof a non-�nite form of the verb (a deverbal adjective in the perfect aspect).4 Thisconstruction was one of the multiple devices in OIA of expressing the perfectaspect and the past tense. The rich tense-aspect system of OIA, underwent aradical process of simpli�cation in MIA (Pischel 1981). Most in ectional formssuch as the aorist and the in ectional perfect, were lost, and by late MIA,

3It must be noted that none of these languages can be regarded as directly descendingfrom an earlier stage. For example, it has been suggested that Classical Sanskrit does notdirectly descend from Vedic, but another dialect, contemporaneous to the Vedic language. Thehypothesis that Sanskrit was the source of Prakrit (and Pali) has also been proved untenable(Pischel 1981).

4This morphological form is also referred to as the past passive participle or the -ta par-ticiple with an Indo-European ancestor in -to (Hock 1986; Klaiman 1978).

4

even the non-perfect in ectional past disappeared.5 The periphrastic passiveconstruction survived and became the only means of expressing the past tenseand the perfect aspect. This loss of the in ectional system has often beencited as a reason for the increase in the frequency and the scope of this passiveconstruction, which in turn led to the unmarking of the marked passive voice ofthis clause, and resulted in an active, ergative clause in late MIA (Hock 1986;Bubenik 1998).

1.4 Markedness of the Ergative Construction

As a result of this diachronic path, NIA languages show morphological erga-tivity based on the aspect split. They also represent an example of one of themain diachronic explanations for ergative patterning in languages { a shift froman accusative to an ergative system due to reanalysis of a passive structure asan active, ergative clause (Dixon 1994; Butt 2000). The agent, or the logicalsubject, which is an oblique grammatical function in the passive construction, isreinterpreted as the grammatical subject, while retaining its oblique morphol-ogy. The patient-like role, which is the passive subject, loses its grammaticalsubjecthood, and is mapped onto the object function. In the former stage,intransitive clauses may or may not passivize, depending on language-speci�cproperties. However, in the latter, ergative stage, intransitive clauses are active,with no overt marking on the sole verbal argument.

The ergative construction in the Middle Indo-Aryan period had the followingergative properties:

(3) a. The agent, marked in the instrumental case, showed subject properties.

b. The object of the transitive and the subject of intransitive clausesshowed nominative case marking.

c. The verb, based on the earlier passive participle, showed gender andnumber agreement with the nominative object.

d. In intransitive clauses, the verb agreed in number and gender with thesole argument of the clause.

From the point of view of markedness theory (Battistella 1990), the reanal-ysis of the passive clause to an active one, can be understood as the emergenceof the unmarked. The formalization of this process is beyond the scope ofthis paper. Informally, however, this process can be understood as the over-riding of faithfulness constraints on syntactic voice, and discourse prominence,by markedness constraints on the morpho-syntactic expression of agents andpatients. In other words, while the passive construction violates the constraintthat agents must be realized as subjects in order to satisfy voice conditions,

5Traditional grammarians do provide instances of the in ectional perfect and the aoristduring this period, but they remain as isolated, unanalyzed forms for a few verbs like �aha-`say' and ak�ashi -`do'.

5

its reanalysis as an active (ergative) construction violates voice conditions inorder to express agents as subjects. However, while this constraint reranking ismotivated by markedness reduction, it is crucial to note that that the resultingactive, ergative clause is also morpho-syntactically marked in several respects.

In this discussion, we treat the surface realization of marked and unmarkedvalues in terms of zero and non-zero morphemes (Greenberg 1966). Here,this applies to overt case and agreement marking. Indo-Aryan is essentially anominative-accusative case-marking language.6 While cross-linguistically, mor-phological ergativity in the perfect aspect (and past tense) is a commonly at-tested phenomenon, this construction may be considered marked in terms ofmorphological structure.

First, assuming a prominence scale of subject > object > non-corefunction (Aissen 1999), the MIA ergative construction is marked in that theleast marked function (subject) is expressed by a morphologically more markedcase (ergative), while the more marked function (object)is expressed in the un-marked (nominative) case.

Second, agreement generally indexes the least marked grammatical function,and subject agreement is the most commonly attested pattern; however, in theergative construction, agreement is with the object.

A third marked feature of the construction is the lack of object marking.OIA nominative and accusative cases became syncretized in MIA leaving anabsolutive case for both subject and object marking (Bubenik 1998). The overtmarking on objects which has developed in NIA languages is based on factorsof de�niteness and animacy, the presence of which has been cross-linguisticallyattested (Silver 1976). This pattern of object marking developed in the non-perfect aspects and spread to the ergative construction in the perfect. ModernNIA languages, therefore, have a complex system of object case-marking whichis dependent on factors of de�niteness and animacy (Cf. Di�erential ObjectMarking (DOM) in Aissen (2000), for Hindi). All the NIA languages in thetypology under consideration here show more or less the same patterns of DOM.Thus, modern NIA has developed variable case marking for direct objects intransitive, perfect clauses, allowing both nominative and accusative markedobjects. An OT analysis as in Aissen (2000) can be applied to this type ofmarking, and we assume this is yet another strategy used by NIA languages toreduce the markedness of the ergative clause. However, we do not discuss thisany further in our paper, because typologically this strategy does not show muchvariation in Indo-Aryan languages, and thus has little bearing on the typologyof NIA languages that we develop.

This paper focuses on the typological variation in ergative marking on sub-jects and agreement patterns in �ve types of Indo-Aryan languages and treatsthem, as in the case of the development of DOM, as strategies for approachinga more unmarked case and agreement system. We show that the reduction of

6There have been arguments that Indo-Aryan cannot be considered as a simple nominative-accusative case-marking system, since there is evidence that it employs a complex system ofcase-marking that is closely inter-linked with verbal semantics (Mohanan 1994; Butt 2000).

6

markedness of the ergative construction in NIA follows two basic strategies (notincluding DOM, discussed above):

(4) a. making the ergative subject paradigm less marked.

b. reducing the markedness of the agreement pattern.7

Thus, the reduction of markedness will be looked at with respect to twoaspects of the perfect clause: the morphological marking on the subject and thecross-referencing of arguments with agreement.

2 The Typology of Variation in NIA

In this section, we present the patterns of ergative marking and agreementfound in several New Indo-Aryan languages: Hindi, Nepali, Gujarati, Marat.hi,Punjabi, and Bengali. As Marat.hi and Punjabi show essentially the same caseand agreement pattern, we represent the two patterns with examples from onlyone of them: Marat.hi. This pattern will be referred to as the Marat.hi/Punjabipattern. The array of data in this section illustrates two crucial points regardingthe typology of these languages. First, languages have reduced the original MIApattern of perfect subject marking to di�ering degrees but by following the samepaths; the languages are in fact presented in order of decreasing overt subjectmarking. Second, the agreement pattern of each language is related to, butcrucially not fully determined, by the subject-marking pattern. We will returnto these points after the presentation of each ergative system.

2.1 Hindi

In Hindi, the perfect subject is morphologically marked with the ergative post-positional clitic in all persons and numbers. The ergative post-positional cliticin Hindi is -ne. In Table (5) and all of the paradigm tables that follow, the formsthat are in boldface represent overt marking in the perfect subject paradigm.

7Throughout this paper, we look at agreement as a device that indexes any grammaticalproperties of NPs on the verb. The languages represented here show variation in the speci�cgrammatical properties of the NPs that are indexed by the verb. For example, Hindi, Marat.hi,Punjabi, and Gujarati show gender and number agreement with the object in ergative clause.Nepali shows person and number agreement, while Bengali has only person agreement on theverb. It is important to note that those languages which have object agreement do not haveperson-based agreement, while those languages which have subject-agreement show agreementbased on person.

7

(5)number

aspect personsingular plural

Non-perf 1 ma~i hamPerf 1 mai-ne ham-neNon-perf 2 tum �apPerf 2 tum-ne �ap-neNon-perf 3 vah, yah ve, yePerf 3 is-ne, us-ne inho-ne, unho-ne

Agreement in Hindi is governed by the following rule, fromMohanan (1994:105):

(6) The verb agrees with the highest arg[ument] associated with the nom[inative]case.

The data in (7) shows the agreement facts. The arguments that the verbagrees with and the agreement morphemes are in boldface. (7a) shows the non-perfect clause, in which the verb agrees with the nominative subject. In (7b),theverb agrees with the nominative object, because it is the highest nominativeargument. The verb may not agree with the ergative marked subject. Theverb in (7c), on the other hand, shows default masculine singular agreementwhen the object is accusative. Agreement is blocked because both argumentsare case-marked.

(7) a. s��t�a r�am-ko p��t.-t�� hai

S��t�a-fem-nom R�am- masc-acc hit-pres-fem-sg aux-3rd-sg

`Sita hits Ram.'

b. r�am-ne chid. iy�a dekh-��R�am- masc-erg bird-fem-nom see-perf-fem-sg`Ram saw a sparrow.'

c. s��t�a-ne r�adh�a-ko p��t.-�aS��t�a-fem-erg R�adh�a- fem-acc hit-perf-masc-sg

`Sita hit Radha.'

2.2 Nepali

Nepali, like Hindi, has ergative marking on the subject in all three persons. Theparadigm of Nepali subject marking is given in Table (8).

(8)number

aspect personsingular plural

Non-perf 1 ma h�am��Perf 1 mai-le h�am��-leNon-perf 2 ta tim��Perf 2 tai-le tim��-leNon-perf 3 �u un��Perf 3 usu-le un-le

8

Nepali di�ers from other languages in the present typology in the fact thatovert case-marking on the subject does not block subject agreement. The tran-sitive verb in Nepali agrees in person and number with the subject. In (9),we compare an intransitive clause with a nominative subject, and a transitiveclause with an ergative subject.

In the examples, we can see that the verb takes the same ending both in thecase of the nominative subject and of the ergative subject. The object in (9b),in spite of being nominative, does not trigger agreement. In Nepali, therefore,it is the subjecthood of an argument that triggers or blocks agreement and notovert case-marking (or absence thereof), as in the other languages presentedhere.

(9) a. ma bas-enI-nom sit-pst-1-sg

`I sat.'

b. mai-le mero lug�a dho-enI-erg my clothes-nom wash-pst-1-sg`I washed my clothes.'

2.3 Gujarati

Gujarati also shows subject marking in all three persons, like Hindi and Nepali.The paradigm for Gujarati subject marking is shown in Table (10).

(10)number

aspect personsingular plural

Non-perf 1 h~u amePerf 1 me~n ameNon-perf 2 tu tamePerf 2 te~n tameNon-perf 3 te te-oPerf 3 ten. e temn. e

The notable feature of the Gujarati paradigm in (10) is the morphologicalsyncretism of the nominative and the ergative in �rst and second person subject,but only in the case of plural subjects. In other words, the loss of subjectmarking has only occured in subparts of the person (�rst and second) andnumber (plural) paradigms.

The other point at which Gujarati di�ers from most other IA languages isin its object agreement pattern. In Gujarati, the verb agrees in number andgender with the object, irrespective of whether it is in the nominative or theaccusative, as seen in (11). In (11b) the object, R�aj is case marked, but theverb still agrees with it.

(11) a. Seeta-e k�agal v�ac-yoS��t�a-fem-sg-erg letter-masc-sg-nom read-pst-masc-sg

`Seeta read the letter.'

9

b. S��t�a-e r�aj-ne pajav-yoS��t�a-fem-sg- erg R�aj -masc-sg-acc harass-pst-masc-sg`Seeta harassed R�aj.'

(adapted from Mistry 1997)

In this kind of language then, case marking doesn't block agreement withthe object, but it does block agreement with the subject. This contrasts withHindi, which shows default agreement if there is no nominative argument in theclause.

2.4 Marat.hi

Marat.hi and Punjabi, though geographically separated, share the same systemof ergative subject marking and take the reduction of subject marking slightlyfurther than Gujarati. Where Gujarati has reduced subject-marking in a subsetof �rst and second person subjects (only those which are plural), all �rst andsecond person perfect subjects in Marat.hi and Punjabi are syncretized with thenominative forms. Furthermore, it is important to note that Old Marat.hi wasin fact like Gujarati, in that �rst and second person plural subjects did not bearergative case, while ergative case was overt in the singular. This suggests thatthe progressive loss of ergative subject marking proceeds systematically throughthe paradigms of person and number.

The paradigm for Marat.hi non-perfect and perfect subjects is shown in Table(12).

(12)number

aspect personsingular plural

Non-perf 1 m�� �amh��Perf 1 m�� �amh��Non-perf 2 t�u tumh��Perf 2 t�u tumh��Non-perf 3 to/t��/te tePerf 3 ty�a-ne, ti-ne ty�a-n��

The agreement facts for both Marat.hi and Punjabi are not self-evident fromthis paradigm. One might expect the verb to agree with the subject when itis unmarked in the local person (and apparently nominative) and not to agreewhen the subject is overtly marked. However, in spite of overt morphologicalsyncretism with the nominative case, �rst and second person perfect subjectsin Marat.hi/Punjabi do not agree with the verb.

This is shown with the Marat.hi example in (13). In (13a), the subject isa non-perfect nominative subject and the verb agrees with it. In the perfectclause in (13b), the verb agrees with the nominative object, as in Hindi, eventhough the �rst-person subject m�� does not show overt case marking. In (13c),agreement with the object is blocked because of overt accusative marking onthe object, and so default neuter agreement marks the verb.

10

(13) a. m�� s��t�a-l�a bagha-toI-masc-nom S��t�a-fem-acc see-pres-masc-sg

`I see Sita.'

b. m�� ek chimn.�� baghit-l��I-masc-erg one sparrow-fem-nom see-pres-fem-sg`I saw a sparrow.'

c. m�� s��t�a-l�a baghit-laI-masc-erg S��t�a-fem-acc see-perf-neut-sg

`I saw Sita.'

2.5 Bengali

Whereas the Marat.hi/Punjabi pattern constitutes an intermediate stage of lossof subject-marking, Modern Bengali has lost subject marking altogether. OldBengali had an ergative construction in the perfect aspect (Chatterjee, 1926,1970:947-8), which showed properties similar to the MIA ergative clause. Mod-ern Bengali, however, has lost this pattern, and shows the same kind of subjectcase-marking for its non-perfect and perfect subjects. The Bengali paradigmfor perfect and non-perfect subjects is shown in (14).

(14)number

aspect personsingular plural

Non-perf 1 �ami �amr�aPerf 1 �ami �amr�aNon-perf 2 tui, tum�� tor�aPerf 2 tui, tum�� tor�aNon-perf 3 o or�aPerf 3 o or�a

The data in (15) captures the agreement facts for Bengali. In both (15b)and (15c), the verb agrees with the subject in person, just as it does in thenon-perfect clause in (15c). The verb thus maintains a nominative-accusativepattern of case and agreement marking in all its tenses and aspects.

(15) a. �am�� s��t�a-ke dekh-ch��I-nom S��t�a-acc see-1-sg-pres

`I see Sita.'

b. �am�� s��t�a-ke dekh-l�amI-nom S��t�a-acc see-1-sg-past

`I saw Sita.'

c. anu s��t�a-ke dekh-loAnu-fem-nom S��t�a-acc see-3-sg-past

`Anu saw Sita.'

11

What is crucially di�erent in Bengali, when compared with Marat.hi andPunjabi (which show loss of ergative marking in the �rst and second persons),is that the complete loss of ergative marking has triggered subject agreement inBengali.

3 Cross-Classi�cation of the Indo-Aryan Lan-

guages

Summarizing the data that we have examined so far, it can be seen that thelanguages presented can be classi�ed in di�erent ways according to their subjectmarking and agreement patterns.

3.1 Subject Marking

Table (16) shows how these languages may be cross-classi�ed according to sub-ject marking. Hindi and Nepali pattern in the same way with regard to theirsubject marking, with overt ergative case in all three persons.

In the Marat.hi/Punjabi pattern, there is no overt marking on �rst and secondperson ergative subjects, and ergative case is present in terms of abstract casefeatures alone.

Gujarati also shows a restriction of subject marking in �rst and second per-son ergative subjects, but restricts this within the domain of number. In Gu-jarati, only plural perfect subjects in the �rst and second person have becomesyncretized with the nominative forms.

Finally, Bengali has no morphological or abstract case on its perfect sub-jects, but marks them nominative, patterning in a third way. Thus, the tabledemonstrates several stages of loss of subject marking proceeding in a single di-rection. There are no systems in this typology where only �rst or second personperfect subjects are marked, but third person subjects are not.

(16) Overt subject marking in perfect constructions:

subject hindi nepali gujarati marat.hi/punjabi bengali1st sg

p p p ; ;1st pl

p p ; ; ;2nd sg

p p p ; ;2nd pl

p p ; ; ;3rd sg

p p p p ;3rd pl

p p p p ;

3.2 Agreement marking typology

Table (17) shows the classi�cation of languages according to agreement markingon the verb. The �rst two rows indicate whether subject agreement occurs inperfect clauses in a given language, and the next two rows show the types of

12

object agreement. Hindi and Marat.hi/Punjabi agree with the nominative ob-ject, showing default agreement otherwise. Gujarati extends object agreementto accusative objects. Nepali and Bengali pattern together in that they bothshow agreement with the highest grammatical function { the ergative subjectin the case of Nepali, and the nominative subject in the case of Bengali.

There are two important points to observe in the typology of agreement here.First, the range of agreement types is wider than is often supposed for Indo-Aryan languages and cannot always be described as default agreement with thenominative argument. Second, we never �nd a system where the verb agreeswith the non-nominative object rather than the nominative subject.

(17) Agreement in perfect constructions:

agreement hindi marat.hi/punjabi gujarati nepali bengaliAgrS (nom) ; ; ; ; pAgrS (non-nom) ; ; ; p ;AgrO (non-nom) ; ; p ; ;AgrO (nom)

p p p ; ;

It should be clear from these two tables that the groupings of subject-marking types and agreement types do not overlap exactly. The languageswhich group together with respect to subject case marking in Table (16) arenot necessarily the ones showing similar agreement marking properties. Forexample, Hindi and Marat.hi/Punjabi don't share subject marking systems inTable (16), but they do share the the nominative object agreement patternin Table (17). Conversely, Gujarati groups with Hindi in terms of subject casemarking, but not in terms of object agreement. Finally, Nepali and Bengali havedi�erent subject marking patterns, but the same subject agreement pattern.

While many of these patterns in Indo-Aryan languages have been noted inprevious research, a synthesis of these systems into this broad, yet structured,typology has not been adequately made in the literature. Treating these diversesystems as all pertaining essentially to the reduction of markedness of the erga-tive construction requires any theory of Indo-Aryan languages to allow, on theone hand, such diverse phenomena as agreement over case-marking and lack ofagreement with null-marking, while, on the other hand, still ruling out a fewunattested systems. In the analysis that follows, in x4, we present a set of con-straints that restricts the typology where necessary, but permits a wide enoughrange of systems to emerge as is called for by the diversity in the data.

4 Subject case constraints

This section presents the Optimality Theoretic constraints on subject-marking.In Optimality Theory, grammars are represented by language-particular rank-ings of universal constraints. Constraints generally fall into two classes: faith-fulness constraints, which relate a feature in the input to one in the output, andmarkedness constraints, which place restrictions on possible output structures.

13

Candidate outputs are evaluated for a given input according to these rankedconstraints. We �rst present two faithfulness constraints we will assume, andthen present the markedness constraints. For the latter, we draw on Aissen's(1999) implementation of harmonic alignment and constraint conjunction insyntax, based on Prince and Smolensky (1993). Alignment constraints are de-rived from the alignment of various universal hierarchies.8 Since such constraintalignments are derived from universal scales, they cannot be mutually rerankedwithin one constraint subhierarchy. The prohibition on mutual reranking ofconstraint alignments within a universal subhierarchy is crucial to the analysis.

4.1 Faithfulness constraints

The faithfulness constraints we adopt in this analysis relate to the realizationof abstract case. Our analysis is primarily an account of the paths availablefor reducing the morphological marking of subjects. However, a purely morpho-logical account is complicated by the data from Marat.hi and, later, the dialectof Gowari as well. These systems provide evidence that morphological and ab-stract case do not always pattern identically, but may show partial mismatches.Local person perfect subjects in Marat.hi are morphologically nominative-like,yet behave like ergative subjects.

The evidence for this is twofold. First, as was pointed out in (13), ergativesubjects that do not bear case in the Gujarati and Marat.hi systems do nottrigger subject agreement in the way that Bengali null-marked subjects do.Second, null ergative and nominative subjects behave di�erently with respectto the case of their modi�ers. This is demonstrated in (18).

(18) a. m�� ek �amb�a kh�a-ll�aI-erg one mango-nom eat-perf-3.sg`I ate a mango.'

b. vedy�a ash�a m�� ek �amb�a kh�a-ll�afoolish-obl like-obl I-erg one mango-nom eat-perf-3.sg`Foolish me ate a mango.'

c. ved�� ash�� m�� ek �amb�a kh�a-tefoolish-nom like-nom I-nom one mango-nom eat-pres-1.sg`Foolish me eats a mango.'

(Marat.hi)

In (18b), the adjectival modi�er of the perfect, transitive subject, whichagrees with the head noun, occurs in the oblique case. Modi�ers in Marat.hi andGujarati are marked in the oblique case when they occur with non-nominativenominal heads. By contrast, the same modi�er takes the nominative form when

8Essentially, this operation takes a binary structural scale (e.g. a grammatical functionscale) and aligns each member of a second scale (e.g. an animacy scale) with the �rst.

14

modifying a non-perfect subject as in (18c). Furthermore, the verb does notagree with the null ergative subject in (18b) either.

These data indicate that abstract and morphological case cannot be assumedto always coincide. Woolford (2001) also notes the need to distinguish betweennull ergative and nominative in languages like Marat.hi. However, she exclusivelyrestricts her case analysis to abstract case, and thus does not deal with di�erencesin morphological expression. She concludes her discussion of Marat.hi sayingthat \these subject person splits do not involve an alternation between di�erentCases, but only whether or not the ergative abstract Case is morphologicallyrealized. The present paper [Woolford 2001] is limited to dealing with situationsinvolving choices among di�erent abstract Cases" (2001:534-5).

However, in the case of the Marat.hi dialect of Gowari, discussed later, we �ndthat the morphologically null local subjects in perfect clauses may in fact showabstract nominative-like behavior in terms of agreement. The fact that bothalternatives are available and are closely related is not addressed by Woolford.

Woolford directs the reader to Aissen's work for an explanation of Marat.hi,as Aissen deals with morphological realization of case. As we will show in ouranalysis, Aissen's (1999) constraints do in fact provide an explanation of thenull-marking of certain subjects. However, as her analysis applies only to mor-phological case, it has the converse problem of Woolford's constraints. WhileWoolford's account does not explain the morphological variation in Marat.hi sub-jects, Aissen's account fails to explain the non-nominative agreement behaviorof these subjects. In the present data, we need to distinguish between true nom-inative subjects (Bengali) and surface nominative subjects (Marat.hi) in perfectclauses. The data, therefore, poses a central problem for both Woolford's andAissen's interpretations of constraints on case, as neither of their analyses treatsthe relation between abstract and morphological case in the speci�c instanceswhen the two do not coincide.

To start with, we assume a basic faithfulness constraint that maps mor-phological features to abstract features. Rather than limiting the discussion toeither abstract (as Woolford does) or morphological (as Aissen does) case, weassume that there is a direct mapping from morphology to abstract features.

We adopt a standard Lexical-Functional Grammar (LFG) model of mappingfrom the surface structure to the semantic/grammatical information structure.In its OT interpretation, this is expressed in terms of a relation between an un-derspeci�ed input and various possible candidate outputs, containing pairings ofsurface representations (constituent-structure) and abstract grammatical repre-sentations (functional-structure). Figure 1, from Bresnan (2000:26), shows thismapping. Three of many possible candidates are shown in the example.

At this level, faithfulness is construed as an identity between the surfacerealization of morphological structure and the corresponding abstract informa-tional content of the clause. Following Kuhn (2001), Lee (2002a) discusses indetail the need for this mapping, treating it more speci�cally as an output-output correspondence between the c-structure and the f-structure of a givencandidate. The faithfulness constraint we assume in our analysis for the purposeof ensuring this mapping is given in (19):

15

Figure 1: Architecture of input-output relations in ot-lfg (From Bresnan2000:26).

GF [. . .]TNS . . .PRED . . .GF [. . .]

IP1

NP2 VP3

V4 NP5

GF [. . .]TNS . . .PRED . . .GF [. . .]

1,2,3,4< >

IP1

NP2 I'3

I4 VP5

NP6

GF [. . .]TNS . . .PRED . . .GF [. . .]

1,3,4,5< >

VP1

NP2 NP3 VP4

GF [. . .]TNS . . .PRED . . .GF [. . .]

1,4< >

INPUT CANDIDATES

16

(19) cs-fs: Case in the f-structure of an argument must beidentical with the case-marking in the c-structure of the argument

When this constraint is highest ranked, it selects candidates which haveidentical morphological and abstract case. In other words, if a clause has asubject with overt ergative marking, then the constraint will select a clausewith an abstract ergative subject in its f-structure, as opposed to an abstractnominative, for example. This is analogous to constructive case as developed inNordlinger (1998), as the morphology drives the abstract case realization. Theconstraint is not symmetric: c-structure (surface morphological) case informsf-structure (abstract) case, but not vice versa.

The second faithfulness constraint we adopt for the present discussion isWoolford's (2000) constraint, which marks perfect subjects as abstract ergatives.

(20) faith-lexperf : A lexically-speci�ed inherent Case licensing featuremust be checked in perfect clauses

When highly ranked, this constraint requires abstract ergative case to beassociated with perfect subjects, independent of the loss of overt morphologi-cal marking. This constraint con icts with the basic c-structure-to-f-structuremapping constraint, as it imposes an abstract ergative regardless of the mor-phology.

4.2 Markedness Constraints

The typological range in the tables in (16) and (17) | with partly independentsubject and agreement patterns | lends itself to an analysis which draws onuniversal markedness hierarchies. In this section, we present universal subject-marking case constraints and their speci�c rankings.9 All the markedness con-straints here apply strictly to the morphological expression of case.

The hierarchy of subject-marking constraints used here was proposed in Ais-sen (1999) for a range of case phenomena. In (21), the two universal scales ofgrammatical function and person rank are listed in the �rst column. The ty-pological markedness reversal between subjects and objects noted by Comrie(1989) and Battistella (1990) among others, namely that what is most markedfor subjects is least marked for objects, is captured by direct and inverse align-ments of subject and object respectively with the person hierarchy. This isshown in the second column. These harmonic alignments state, for instance,that it is more harmonic for a subject to be associated with �rst person thanthird person. Finally, the universal subhierarchies of actual constraints areshown in the third column. These are derived by pre�xing the \Avoid" op-erator (*) to each alignment and stating the ranking in terms of decreasing

9We assume that a correspondence of argument-structure and functional-structure estab-lishes the grammatical functions of arguments in the input. The focus here is mainly onthe association of morphological marking with grammatical functions, not the determinationof grammatical functions themselves, which will be assumed to be independently ensuredthrough the type of argument-function correspondences proposed in Lexical Mapping Theory(LMT; Bresnan & Kanerva, 1989) and further developed for OT in Bresnan (2001).

17

markedness. Most importantly, the ordering of these constraints relative to oneanother is universal.10

(21) universal scales harmonic alignment constraint alignmentSubject > Object Su/loc � Su/3 *su/3 � *su/loclocal (1st,2nd) > 3rd Oj/3 � Oj/loc *oj/loc � *oj/3

Aissen conjoins these constraints with the constraint *�c , resulting in arequirement to mark these arguments with some case form. This captures theidea that marked con�gurations of features should be morphologically marked.

The ranking in (22) basically states that 3rd person subjects are universallymore marked than 1st and 2nd person subjects. Each constraint can only besatis�ed by overt case-marking.

(22) *su/3 & *�c � *su/local & *�c (Aissen 1999:673)

As our data is speci�c to the domain of the perfect aspect, we conjoin Aissen'sconstraints with a constraint on perfect subjects.11

(23) *su/3 & *su/perf & *�c � *su/loc & *su/perf & *�c

This is a necessary contextual restriction to perfect contexts. The highestconstraint in (23) states that a subject occurring in the perfect context andsimultaneously being 3rd person must be overtly marked. The universally lessmarked constraint requires this of local person perfect subjects. As all of thepresent discussion applies to this domain of perfect clauses, we omit the speci�-cation of perfect from the description of these constraints. In all other respects,these constraints are identical to Aissen's (1999).

(24) *strucc: Avoid (case speci�cation) structure(Prince & Smolensky 1993:25; Aissen 1999)

Finally, the constraint in (24) penalizes any morphological structure; Aissen(1999) employs the constraint *strucc to speci�cally penalize case morphology,which is our use here. *strucc serves as an economy constraint.

Using just the three constraints in (23) and (24), we can begin to accountfor changes in the various IA subject-marking systems. In (25) we list thepossible re-rankings of the three constraints, along with the systems in whichthese rankings are found.

(25) � *strucc (Bengali: No subjs marked)

su/3 & *�c � *strucc (Marat.hi/Punjabi: Only 3p subjs marked)

su/loc & *�c � *strucc (Nepali, Hindi, MIA: All subjs marked)

10The corresponding object marking constraints are not addressed in this paper (see Aissen(2000) for a discussion of this constraint subhierarchy).

11See Sharma (2001) for a discussion motivating constraints on overt-marking of argumentsbased on perfectivity.

18

In (25), *struc is progressively promoted above the constraints requiringsubject marking | partially in Marat.hi/Punjabi and completely in Bengali |allowing the universal avoidance of overt subject marking to emerge.12

The tableaus in (26)-(30) demonstrate these rankings. The examples showonly the relevant, partial inputs for transitive, perfect clauses for clarity ofpresentation. Each example in (26)-(30) contrasts two di�erent inputs, to showwhich candidate gets selected according to the person feature in the input andthe language particular rankings. In the �rst two tableaus, the candidates (a),(b), (c), and (d) are evaluated according to the ranked constraints.

(26) Nepali, Hindi, MIA

*su/3&*�c

*su/loc&*�c

*struc

input: Subj(loc)☞a. S-erg �

b. S-� �!☞c. S-erg �

d. S-� �!

In (26), *struc is ranked below both subject constraints, resulting in markedsubjects always satisfying one of the higher ranked constraints. In (27), on theother hand, *struc dominates both markedness constraints, so the morphologi-cally unmarked candidates are selected as optimal. Since there is no discrepancybetween morphological and abstract case in Nepali, Hindi, and Bengali, we donot include the faithfulness constraints in the tableaus for the (26) and (27).

12Gujarati is not included in this tabulation, although it falls in the same intermediategroup as Marat.hi and Punjabi. It is excluded only because its subject-marking pattern issensitive to both person and number and therefore cannot be exclusively accounted for by thearrangement of the person constraints in (25).

19

(27) Bengali

*struc

*su/3&*�c

*su/loc&*�c

input: Subj(loc)a. S-erg �!

☞b. S-� �input: Subj(3rd)

c. S-erg �!☞d. S-� �

In Marat.hi, shown in (28), however, the role of the faithfulness constraintsis apparent, as is the intermediate ranking of *struc. In terms of markednessconstraints, *struc intervenes between the two subject person constraints. 3rdperson subjects must be case-marked to satisfy the highest constraint, eventhough they violate *struc, and so candidate (d) is chosen over (f). However,since the lower-ranked subject constraint is below *struc, null-marked 1st and2nd person subjects are preferred, so candidate (c) is chosen over (a).

The candidates (b) and (f) show how the two faithfulness constraints interactwith the others. Without Woolford's constraint, which lexically requires anabstract ergative case, a nominative 1st and 2nd person subject would win;however, this choice would not derive the facts presented earlier regarding thenon-nominative behavior of these null-marked subjects. The constraint cs-fsis included to show that it is violated in favour of satisfying faith-lexperf .The constraints therefore predict that it is also typologically possible for theorder of the two faithfulness constraints to be switched. Later, in the discussionof Marat.hi dialects, we will see that the dialect of Gowari instantiates thisprediction.

20

(28) Marat.hi, Punjabi

faith-lexperf

*su/3&*�c

*struc

*su/loc&*�c

cs-fs

input: Subj(loc)a. S-erg(overt) �!b. S-nom(�) �! �

☞c. S-erg(�) � �input: Subj(3rd)

☞d. S-erg(overt) �e. S-nom(�) �! �f. S-erg(�) �! �

Finally, in Gujarati the ranking is the same as in Marat.hi and Punjabi,except for an additional, independent phenomenon of number-sensitivity insubject-marking. The marking of subjects is restricted according to both per-son and number. First and second person plural subjects undergo syncretismwith the nominative. This does not in fact pose a problem for a person-basedanalysis of subject-marking reduction, as Gujarati clearly shows the same per-son hiearchy sensitivity as Marat.hi and Punjabi. It simply adds the additionaldimension of number to this process.

The morphological markedness of the plural over the singular number isobservable cross-linguistically in various domains (Greenberg 1966:28-9). Theplural is more susceptible to syncretism or neutralization in marked contextsthan the singular. In order to account for the speci�c syncretism here, weprovisionally use the constraint in (29).

(29) * syncret/sg Avoid syncretism of case-marking in the singular number.

This constraint penalizes morphological syncretism between distinct cases(such as the nominative and the ergative) in the singular, thus capturing thecross-linguistic observation that syncretism is more common in the plural con-text, as well as accounting for the subject-marking properties in some Marat.hidialects later in this paper.

21

(30) Gujarati

faith-lexperf

*syncret/sg

*su/3&*�c

*struc

*su/loc&*�c

cs-fs

input: Subj(loc-pl)a. S-erg(overt) �!b. S-nom(�) �! �

☞c. S-erg(�) � �input: Subj(3rd-pl)

☞d. S-erg(overt) �e. S-nom(�) �! �f. S-erg(�) �! �

input: Subj(loc-sg)☞a. S-erg(overt) �

b. S-nom(�) �! �c. S-erg(�) �! � �

input: Subj(3rd-sg)

☞d. S-erg(overt) �e. S-nom(�) �! �f. S-erg(�) �! � �

The three basic patterns of constraint rankings shown in the three examplesdiscussed above give us the three broad sets of language types from (16). As*su/3 and *su/loc are constraints within a universal subhierarchy, they arenever mutually reranked; their ranking only varies in relation to *struc. Underour analysis, their progressive demotion below *struc represents the systematicelimination of overt marking on subjects.

The next section presents an analysis of the agreement patterns in theselanguages, which, as was shown earlier, are not always exact complements ofovert case marking.

5 Agreement Constraints

As the data earlier showed, agreement cannot be captured as a direct defaultwhich occurs only when case is absent. Nepali allows agreement across case-marked subjects, and Gujarati allows agreement with case-marked objects. Theconstraints deriving agreement patterns, need to be correspondingly nuanced tocapture these patterns of variation. The constraints we develop for agreementare given in (31), (32) and (33).

(31) express agr: A predicate agrees with some argument

The constraint in (31) is a type of faithfulness constraint requiring agree-ment of some sort. The crucial di�erence in choosing to formulate the constraintas express agr rather than as a markedness constraint such as *express

22

agr (formulated like *struc) is that the formulation in (31) favours agree-ment of some sort, while a markedness constraint would favour non-agreement.Cross-linguistically, case marking performs a discriminant function amongst ar-guments, often signalling a marked situation, as noted by Dixon (1994), amongothers. We treat agreement, on the other hand, as a prominence relation withthe least marked argument. In other words, case and agreement do not performidentical functions; case is avoided except under marked circumstances, such as3rd person subjects, while agreement is a default, occurring in unmarked con-texts such as subjects over objects. We also distinguish explicit number, genderor person agreement from default agreement here. The constraint in (31) isnot satis�ed by default agreement, which takes the form of masculine or neutersingular in ection in these languages.

Note that since default agreement is treated as non-agreement in our anal-ysis, it is assumed to occur in order to satisfy an independent requirementfor �niteness marking. A constraint on �niteness marking is necessary to en-sure some overt (default case) marking to distinguish �nite from non�nite verbstems. Since this requirement holds identically for all the languages in ques-tion, we exclude this constraint from our discussion. As a result, candidateswith completely unin ected verb forms are not considered here. Candidateswith default agreement, however, may be optimal for certain inputs and defaultagreement does interact with full agreement. The examples that follow show thecircumstances under which default agreement is preferred over full agreement.

The constraint express agr is used in Bresnan (2001) only for subjectagreement. In order to generalise her constraint systematically, we align itwith the relational hierarchy to permit agreement with more than just subjects,but still to constrain the relative markedness of each type of agreement. Theresulting constraint alignments are given in (32).

(32) *non-core gf/agr � *oj/agr � *su/agr

The universally least marked agreement pattern, according to this constrainthierarchy, is subject agreement. Object agreement is more marked and agree-ment with non-core grammatical functions is the most marked. This highestconstraint is left out of the discussion, as it is never violated in the data here;for example, we never �nd agreement with dative subjects or objects (atleastwithin the Indo-Aryan typology).

So far, the agreement constraints simply derive agreement based on gram-matical function. However, the typology of agreement systems in (17) showedthat overt case-marking sometimes blocks agreement across the board, but inother instances does not. In other words, the data we have presented showsthat agreement does in fact occur with NPs bearing certain overt cases.

Thus, agreement is not only sensitive to grammatical function but also to thecase that a particular grammatical function bears. This requires two dimensionsof agreement relations: one with grammatical function, as above, and one withcase.

To derive the latter, we assume a similar alignment of agreement with case,as shown in (33). In her cross-linguistic study of the interaction of case-marking

23

with faithfulness constraints, Woolford (2000) proposes the universal hierarchyof *ergative,*dative � *accusative � *nominative, which we followhere to derive case-sensitive agreement.

(33) *erg/agr � *acc/agr � *nom/agr

We restrict this hierarchy to abstract case as Woolford does. In fact, thisgives the correct results, as the constraint *erg/agr correctly prevents agree-ment with 1st and 2nd person subjects in Marat.hi/Punjabi and Gujarati, re-gardless of whether their morphological case is overt.

These two universal subhierarchies of agreement constraints in (32) and (33)interact to derive the observed language types. In the examples that follow,case selection is assumed to be ensured by the constraints presented in thelast section. We therefore only include candidates with the correct subjectcase, and the examples are restricted to examining agreement alternations.13

Each example shows three di�erent types of clausal inputs | perfect clauseswith speci�c (therefore accusative) objects, perfect clauses with non-speci�c(therefore nominative) objects, and non-perfective clauses | to show how theconstraints interact to derive agreement for di�erent clause types.

First, we turn to Hindi, Marat.hi and Punjabi. This group allows agreementwith either subject or object, as long as its case is nominative.

(34) Hindi, Marat.hi, Punjabi

*erg/agr

*acc/agr

expagr

*oj/agr

*su/agr

*nom/agr

input: S O(spec) V(perf)a. S-erg O-acc V-Sagr �! �b. S-erg O-acc V-Oagr �! �

☞c. S-erg O-acc V-default �input: S O(nonspec) V(perf)

d. S-erg O-nom V-Sagr �! �☞e. S-erg O-nom V-Oagr � �

f. S-erg O-nom V-default �!input: S O V(nonperf)

☞g. S-nom O-nom V-Sagr � �h. S-nom O-nom V-Oagr �!i. S-nom O-nom V-default �!

In (34), the �rst input requires case on both subject and object. In thissituation, default agreement wins out of candidates (a), (b), and (c), becauseagreement with either argument would violate the restriction on agreement withergative or accusative. When the object is not marked accusative, as in (d),(e), and (f), object agreement is preferred to a violation of the higher-ranked

13So, for example, all ergative-marked subject candidates for perfect inputs in Bengali wouldbe ruled out by the rankings that we saw in (27).

24

express agr. Finally, if neither subject nor object is case-marked, as in (g), (h),and (i), then subject agreement is ideal because object agreement is universallymore marked.

Turning to Gujarati, the only di�erence between the Hindi-type group in(34) and the ranking for Gujarati in (35) is the promotion of the faithfulnessconstraint express agr above *acc/agr in Gujarati.

(35) Gujarati

*erg/agr

expagr

*acc/agr

*oj/agr

*su/agr

*nom/agr

input: S O(spec) V(perf)a. S-erg O-acc V-Sagr �! �

☞b. S-erg O-acc V-Oagr � �c. S-erg O-acc V-default �!

input: S O(nonspec) V(perf)d. S-erg O-nom V-Sagr �! �

☞e. S-erg O-nom V-Oagr � �f. S-erg O-nom V-default �!

input: S O V(nonperf)☞g. S-nom O-nom V-Sagr � �

h. S-nom O-nom V-Oagr �! �i. S-nom O-nom V-default �!

This reranking only a�ects the �rst input in (35). In the Hindi group, thisinput resulted in default agreement since both arguments were case-marked.In Gujarati, because express agr is higher ranked, agreement with the ac-cusative is less bad than default agreement and so candidate (b) wins. The �rstinput listed in the Gujarati tableau is therefore the clause-type that requiresour inventory of agreement constraints to permit agreement with a case-markedargument.

The other two types of inputs are una�ected by the reranking, and still selectthe highest nominative argument present for agreement.

Finally, Nepali also contradicts the nominative agreement pattern of theHindi group by allowing agreement with the ergative. This violates the highestconstraint in (33). But, this is still preferable to agreement with any type ofobject, hence the ranking shown in (36).

25

(36) Nepali, (Bengali)

expagr

*oj/agr

*su/agr

*erg/agr

*acc/agr

*nom/agr

input: S O(spec) V(perf)☞a. S-erg O-acc V-Sagr � �

b. S-erg O-acc V-Oagr �! �c. S-erg O-acc V-default �!

input: S O(nonspec) V(perf)☞d. S-erg O-nom V-Sagr � �

e. S-erg O-nom V-Oagr �! �f. S-erg O-nom V-default �!

input: S O V(nonperf)

☞g. S-nom O-nom V-Sagr � �h. S-nom O-nom V-Oagr �! �i. S-nom O-nom V-default �!

Bengali falls in the same category as Nepali in terms of agreement but hasno subject marking at all, so although the ranking in (36) is identical for Bengaliagreement, candidates with nominative subjects would be the winners.

The only di�erence here between the Nepali/Bengali group and the othergroups is that none of the case constraints on agreement outrank the gram-matical function subhierarchy. One way of looking at this is that since overtsubject case-marking in Nepali has not been reduced, the language has optedfor an alternative path of markedness reduction by focusing on the markednessof object agreement.

Before turning to dialectal variation, let us �rst recapitulate the points cov-ered by the analysis thus far. At the outset of this discussion, we showed thatthe MIA ergative, perfect construction had several universally marked features,such as case-marking on the subject, and verb agreement with the object. Aformal model of two general strategies of reducing markedness along universalhierarchies has been provided for the range of NIA systems. The formal group-ings of languages according to our analysis correspond to the cross-classi�cationof the data in (16) and (17).

The analysis shows that the paths along which each ranking has movedessentially represents some trajectory of markedness reduction. In terms of sub-ject marking, the gradual promotion of *struc above case marking constraintsgives rise to unmarked subjects in Bengali and Marat.hi. Those languages whichretain subject-marking in all three persons { Hindi, Gujarati, and Nepali {show markedness changes in agreement instead. The promotion of the faithful-ness constraint express agr combined with universal hierarchies of agreementtypes leads to the emergence of unmarked patterns of either subject agreementor nominative agreement. In Hindi, agreement is restricted to the least markedcase, the nominative; in Nepali, to the least marked grammatical function, thesubject.

26

6 The Typology of Variation in Marat.hi Dialects

Thus far, we have discussed the typology of variation in case marking and agree-ment seen in the Indo-Aryan languages, which had a common ancestor in MiddleIndo-Aryan. In this section, we demonstrate that one of the languages describedabove, Marat.hi, re ects the same typological variation within its dialects. How-ever, two distinctions arise between the dialect typology and the language ty-pology. First, certain micro-level typological predictions that are not found inthe language data are attested in dialectal data.

Second, an outer bound is evident in the range of dialectal variation inMarat.hi. The dialects of Marat.hi discussed here derive from their historicalancestor, Old Marat.hi (1000-1400 AD). The variation observed in these dialectsturns out to be constrained by the typological features of Old Marat.hi (OM). Bycontrast, the typological features of the Indo-Aryan languages already discussedabove are constrained by the historical system from which they are derived(MIA).

In terms of agreement marking patterns, OM and MIA are identical; the verbagrees with the highest nominative argument. The crucial di�erence betweenthe MIA and OM system is in the nominal paradigm of the ergative subject.The ergative (instrumental case marking) in the MIA system occurs throughoutthe nominal case paradigm and is distinct from the nominative in all cells of theparadigm. In OM, there is a syncretism between the ergative and the nominativemarking of pronouns in the plural of the �rst and second person pronouns, justas in modern Gujarati. These morphologically nominative forms maintainedabstract ergative-like behavior, as evidenced by oblique adjectival modi�cationand agreement.

What the following data demonstrate, and what our analysis also predicts,is that the syncretism in the nominal paradigm of OM never reverts to a moredi�erentiated or more MIA-like paradigm in any of its daughter dialects. Inother words, variation occurs within the typological space a�orded by the par-ent language with respect to the nominal paradigm. The dialectal variation inthe nominal ergative paradigm is a result of the systematic promotion of the*struc constraint over what its base position has been in OM. A fully distinc-tive subject-marking system like Hindi is not observed in any dialect of Marat.hi.This suggests that, even historically, the changes in these dialects can be ex-plained by assuming a uni-directional promotion of the *struc constraint. Inthe sections that follow, we examine the data from six dialects of Marat.hi.

6.1 The Dialect Data

A note on the nature of the dialectal data that we use is in order. Most analysesof modern Marat.hi deal with the standard variety of the language, which is adialectal variety spoken in Pun. e. The data here is not based on independent�eldwork, but is culled from Volume VII of the Linguistic Survey of India (Gri-erson 1905), which is a volume devoted to the dialectal varieties of Marat.hi.The data in this survey is of three kinds:

27

� Skeletal grammars (1 page) of selected dialects, with full nominal andverbal paradigms in some cases.

� A set of sample words and sentences (averaging 300 per dialect) for eachdialect, which have been elicited from respondents.

� A sample of continuous text in each collected dialect, which in most casesis the translation of `The Parable of the Prodigal Son'.

The data available from this survey is sparse, and not always systematic. Ithas not been possible to obtain full paradigms of the verbal in ections in thedomains discussed in this paper. However, the claims we make are based oneither Grierson's observations for a particular dialect, or on the crucial dataavailable in the limited textual material. The claims about nominal and verbalmarking are based on the data in the sample texts, which yields a reasonablyaccurate picture of the in ectional paradigms under discussion.

This section discusses the dialectal data and points to the correspondencesbetween these Marat.hi dialects and the Indo-Aryan languages that we discussedin x2 and x3.

6.2 Warhad.hi Brahman. i, Akola

This dialect is discussed �rst because it appears to be the most conservativewhen compared to the OM system of ergative morphology. This pattern isidentical to the OM perfect paradigm, both in terms of subject marking andagreement.

In Warhad.hi Brahman. i, the morphological distinction between the ergativeand the nominative case, is articulated in the following way in transitive, perfectclauses. The transitive, perfect subject is marked ergative in all three personsin the singular. However, in the plural, there is overt, ergative marking only inthe third person. First and second person transitive subjects are identical tosubjects in non-perfect clauses. This is illustrated in (37).

(37)number

aspect personsingular plural

Non-perf 1 mi �amiPerf 1 my�a �amiNon-perf 2 tu tumiPerf 2 ty�a tumiNon-perf 3 to tePerf 3 ty�a-na ty�a-ni

Historically, the OM ergative paradigm showed syncretism in the same person-number con�gurations (ergative subjects in the plural, local person), whichsuggests that Warhad.hi Brahman. i has retained the archaic pattern while otherlanguages have innovated in this domain.

In terms of the abstract pattern (i.e. not actual morphological exponence)this subject marking pattern is identical to the Gujarati synchronic subject

28

marking pattern and may be derived by the same ranking of subject markingconstraints as in Gujarati. The agreement facts of this dialect are, howeverdi�erent from Gujarati, but identical to the OM pattern. The verb agrees withthe nominative object, whether or not the subject shows overt ergative markingin the transitive, perfect clauses. If the object is marked accusative becauseof de�niteness and animacy, the verb gets a default neuter agreement. This isillustrated in (38).

(38) a. my�a dev�a-chy�a kahy�a-baher tum-chy�aI-erg-sg God-gen-sg against you-gen-sg

samor p�ap kela �ahe

in front of sin -neu-nom-sg do-neu-3-sg-past be-3-sg-pres

`I have committed a sin against God and in front of you.'

b. ty�a-na ty�a-l�a dukar ch�ar�ay-l�a �aply�ahe-erg-sg he-acc-sg pig-nom-pl graze-inf self's

v�avr-�at dh�ad-la.�eld-loc-sg send-neu-3-sg-past`He sent him in the �eld to graze pigs.'

In (38a), the verb agrees with the neuter sin and in (38b), the verb showsdefault neuter singular agreement, because the object is marked accusative.Agreement in Warhad.hi Brahman. i patterns exactly like agreement in standardMarat.hi (the Pun. e dialect), discussed in (13).

6.3 K~okan. i, Savantwad. i

The paradigm for K~okan. i perfect and non-perfect transitive subjects is struc-turally identical to the Warhad.hi Brahman. i paradigm, although it is based ondi�erent morphological material. This paradigm is illustrated in (39).

(39)number

aspect personsingular plural

Non-perf 1 hav amiPerf 1 haven amiNon-perf 2 tu tumiPerf 2 tuven tumiNon-perf 3 to tePerf 3 t�a-nen ta-ni

The K~okan. i subject paradigm may also be derived from the exact rankingof constraints that derive the Gujarati and the Warhad.hi Brahman. i paradigm.The K~okan. i agreement pattern is distinct from the Warhad.hi Brahman. i pattern,and identical to the Gujarati pattern. In K~okan. i, the verb agrees with thenominative object in transitive, perfect clauses. But, unlike the other Marat.hidialects, the verb does not show default agreement when the object is markedaccusative. The verb, in spite of overt marking on the object, agrees with it.

29

In this kind of dialect then, case marking does not block agreement with theobject, but it does block subject agreement, exactly like Gujarati. The followingexamples illustrate this agreement pattern.

(40) a. Haven tujhe utar

I-erg-sg you-gen-sg word-neu-nom-sgmod. -le nabreak-neu-3-sg-past neg

`I never broke your word.'(disobeyed you.)

b. t�a-ne t�a-k�a dukre ch�ar�akhe-erg-sg he-acc-sg pig-nom-pl graze-inf

shet-�at dh�ad-lo�eld-loc-sg send-mas-3-sg-past

`He sent him in the �eld to graze pigs.'

In (40a), the verb agrees with the nominative object, similar to other dialectslike the Pun. e variety, and Warhad.hi Brahman. i. In (40b), the verb agrees withthe accusative object, rather than showing default agreement. Other dialectswhich show agreement patterns as well as ergative marking similar to K~okan. iinclude Chitp�avani and Kud.�al.i dialects.

The pattern for K~okan. i agreement, as we said earlier, is identical to theagreement pattern in Gujarati, and may be derived with the same ranking ofagreement constraints.

6.4 Standard Marat.hi, Pun. e

This is the dialect of Marat.hi that has been the subject of most linguistic liter-ature on Marat.hi, and it is also the dialect discussed in the earlier section. Inthe Pun. e dialect, �rst and second person subjects in perfect clauses do not re-ceive any overt ergative marking and are morphologically similar to non-perfectnominative subjects. In the third person, however, subjects are marked withan overt case clitic. The paradigm for subject marking is identical to the onepresented for Marat.hi in (12), repeated here for convenience.

(41)number

aspect personsingular plural

Non-perf 1 m�� �amh��Perf 1 m�� �amh��Non-perf 2 t�u tumh��Perf 2 t�u tumh��Non-perf 3 to tePerf 3 ty�a-ne ty�a-n��

As seen from (41), the Pun. e dialect does not have overt ergative marking inthe �rst and second person. The verb, however, shows agreement only with thenominative object. The verb exhibits default neuter singular agreement whenthe object is marked accusative based on factors of de�niteness or animacy.Examples are given in (42) repeated from (13).

(42) a. m�� s��t�a-l�a bagha-toI-masc-nom S��t�a -fem-acc see -pres-masc-sg

`I see S��t�a.'

30

b. m�� ek chimn.�� baghit-l��I-masc-erg one sparrow-fem-nom see -pres-fem-sg`I saw a sparrow.'

c. m�� s��t�a-l�a baghit-laI-masc-erg S��t�a -fem-acc see -perf-neut-sg

`I saw S��t�a.'

In (42b), the verb agrees with the nominative object, although the subjectm��does not show overt case marking. In (42c), agreement with the object is blockedbecause of overt accusative marking on the object. This points out the fact thatin spite of overt morphological syncretism with the nominative case, �rst andsecond person subjects in the Pun. e dialect bear ergative features, as opposed tothe system in Dharwari (discussed below), where they are nominative, triggeringagreement.

6.5 Marhet.hi

Marhet.hi has an identical nominal case marking pattern as the Pun. e dialect, inthat overt ergative case is present only in the third person. The paradigm fortransitive perfect and non-perfect subjects in Marhet.hi is illustrated in (43).

(43)number

aspect personsingular plural

Non-perf 1 m�� �amh��Perf 1 m�� �amh��Non-perf 2 t�u tum��Perf 2 t�u tum��Non-perf 3 to tePerf 3 ty�a-ni ty�a-nni

The agreement pattern, however, di�ers from the Pun.e dialect in that theverb agrees uniformly with the subject. The relevant examples are given in (44).

(44) a. m�� ��shwar�a-che iruddh �anikh tudzy�a-s�amneI-erg-mas-sg God-gen-sg against and you-in front of

p�ap ke-lo

sin-neu-nom-sg do-mas-1-sg-past

`I committed a sin against God and in front of you.'

b. tum-che b�ap�a-ne ch�angla bhojanyour father--mas-erg-sg good feast-neu-nom-sg

ke-l�a �ahe

do-mas-3-sg-past be-3-sg-pres

`Your father has organized a good feast.'

31

In (44a), the verb agrees with the �rst person pronominal which is a non-overt ergative. In (44b.), the verb agrees with the overtly ergative marked thirdperson NP. This illustrates that the verb shows a uniform subject markingpattern regardless of overt case marking on the subject. This distinguishesMarhet.hi from the Pun. e dialect where non-overt ergative features on the subjectprevent agreement with the subject.

This pattern of agreement is identical to the agreement pattern in Nepali,where ergative subjects do not block agreement. This pattern in Marhet.hi canbe derived from the same constraint ranking for agreement as in Nepali, whereagr-sub is ranked higher than agr-nom.

6.6 Gowari, Bhand. ara

Gowari has a nominal in ectional paradigm that is identical to the Pun.e dialect.First and second person subjects in transitive, perfect clauses are identical tonon-perfect subjects, with no overt ergative marking. There is overt marking inthe third person. The in ectional paradigm for Gowar�i is given below in (45).

(45)number

aspect personsingular plural

Non-perf 1 m�� �ami��Perf 1 m�� ��m��Non-perf 2 t�u tum��Perf 2 t�u tum��Non-perf 3 to tePerf 3 ty�a-n ty�a-nn��

The nominal paradigm for Gowari is identical to the paradigm of the Pun. e di-alect, formally. However, morphologically similar transitive, perfect subjects inboth these dialects trigger distinct agreement patterns. In the Pun.e dialect, themorphologically zero-marked transitive perfect subject in the �rst and secondperson does not trigger agreement, suggesting that it bears ergative features.In Gowari, on the other hand, the perfect subject in the �rst and second persontriggers agreement, suggesting that it is indeed nominative, and identical mor-phologically as well as in terms of case features to its non-perfect counterpart.The data in (46) illustrates this.

(46) a. m�� dev�a-javal. tudjyaa-s�amne p�ap ke-lo.I-nom-sg God-near you-in front of sin-neu-sg do-mas-1-sg-past`I committed a sin near God and in front of you.'

b. mag ty�a-n baap�a-l�a uttar di-lan.then he-erg-sg father-dat-sg answer-neu-nom-sg give-neu-3-sg-pastThen he gave an answer to his father.'

In (46a), the verb agrees with the �rst person subject which is unmarkedfor case. In (46b), however, the verb agrees with the nominative object, ratherthan the subject, because the subject is marked with ergative case.

32

6.7 Dharwari, Dharwar

Dharwari does not show any morphological distinction between perfect andnon-perfect transitive subjects. There is no ergative marking on the transitive,perfect subject in any person. The Dharwari paradigm is shown in (47). Whilethe pattern for full NPs is not illustrated in the data available, it must be notedthat in all languages full NPs pattern exactly like the third person pronouns, interms of ergative marking and agreement.

(47)number

aspect personsingular plural

Non-perf 1 m�� �amh��Perf 1 m�� �amh��Non-perf 2 t�u tum��Perf 2 t�u tum��Non-perf 3 to ty�aniPerf 3 to ty�ani

Dharwari shows agreement with the subject in person in perfect, transitiveclauses, thus maintaining a nominative-accusative pattern of case and agreementmarking in all its tenses and aspects. Examples are given in (48).

(48) a. m�� tudzy�a-samor �an. i parlok�a-cheI-nom-sg you-in front of and heaven-gen-sg

viruddha p�ap kelo.against sin-mas-nom-sg do-mas-1-sg-past

`I committed a sin in front of you and against the heavens.'

b. ty�ach�a b�ap �apl�� zindag��his father-mas-nom-sg self's property-fem-nom-sg

vibh�ag kar�un dil�a.division-mas-nom-sg make-gerund give-mas-3-sg-past.

`His father, having divided his property, gave it (to them).

In (48), the subject and the object are marked nominative in both clauses.The verb agrees with the subject, as in other non-perfect tenses and aspects.Dharwari shows a uniform pattern of agreement with the least marked case andfunction in transitive, perfect clauses. Formally, the nominal paradigm of theDharwari dialect di�ers minimally from the Pun. e dialect, in that there is overtmarking only in the third person. Nevertheless, there is an important syntacticdi�erence between the case features borne by the zero-case marked transitive,perfect, �rst and second person subjects in these dialects. First and secondperson perfect subjects in Dharwari trigger agreement, as is seen in (15), whilein the Pun.e dialect, the verb shows agreement with the nominative object.

Another dialect showing similar nominal marking and agreement patternsto Dharwari, is Hal.bi, a dialect spoken in the Bastar region. It is interestingto see that the nominal in ectional paradigm, inspite of being based on very

33

di�erent morphology patterns the same way. The nominal paradigm for Hal.bitransitive subjects in perfect and non-perfect clauses is given below in (49).

(49)number

aspect personsingular plural

Non-perf 1 mui hamiPerf 1 mui hamiNon-perf 2 tui tum��Perf 2 tui tum��Non-perf 5 hun hun-manPerf 3 hun hun-man

The pattern for Dharwari and Hal.bi dialects, as we have described it, isidentical to the pattern for Bengali. Like Bengali, these dialects do not markergativity in the perfect aspect. This absence of morphological ergativity, more-over, must be understood as an innovation, because Old Marat.hi, the immediateancestor of these dialects did exhibit ergative morphology. This innovation then,like Bengali, involves the reduction of the markedness of the ergative clause bythe demotion of markedness constraints with respect to *struc. The transi-tive, perfect clause, similar to the transitive, imperfect clause does not havean ergative subject argument. Agreement with the subject does not violateany high-ranking constraint on agreement with the nominative argument, thusleading to uniform agreement with subject across the paradigm.

7 Cross-Classi�cation of the Marat.hi Dialects

The dialect data discussed indicates that, while each dialect of Marat.hi doesnot show a perfect correspondence with one of the Indo-Aryan languages wediscussed, in terms of both subject marking and agreement, there are correspon-dences with respect to these patterns taken separately. These correspondencesare shown in the following table. While there is no a priori reason for de�ningthe agreement and subject marking patterns of the dialects in terms of thoseof the languages under discussion, we have chosen to express these in that waybecause the range of subject marking patterns in the languages is wider thanthat of the Marat.hi dialects.

(50)pattern

dialectsubject marking agreement

Warhad.hi Brahman. i Gujarati Marat.hi

K~okan. i Gujarati Gujarati

Pun.e Marat.hi Marat.hi

Gowari Marat.hi Marat.hi

marhet.hi Marat.hi Nepali

Dharwari Bengali Bengali

From the data, it can be seen that the dialects discussed may be classi�edin di�erent ways according to their agreement and subject marking patterns.

34

7.1 Subject Marking

The subject-marking patterns are summarized in the table in (51). Dharwari hasno ergative case on transitive, perfect subjects. Marat.hi, Gowari, and Marhet.hipattern in the same way with regard to their subject marking, with overt ergativecase in only in the third person.14 Warhad.hi Brahman. i and K~okan. i patternsimilarly in that they have a more articulated morphological distinction betweenthe ergative and the nominative case through the paradigm. However, they showdi�ering patterns in agreement marking.

The main di�erence from the typological range earlier in (16) is the lackof a system that marks the ergative subject overtly in all three persons andnumbers, as in Hindi, Nepali, and MIA. The most articulated subject-markingsystem is the Gujarati-like system of Warhad.hi Brahman. i and K~okan. i, and sincethis was also the system of OM, it imposes the outer limit of subject-markingpossibilities for the dialects.

(51)subj mkg dharwari pun.e gowari marhet.hi wb k~okan. i1-sg ; ; ; ; p p1-pl ; ; ; ; ; ;2-sg ; ; ; ; p p2-pl ; ; ; ; ; ;3-sg ; p p p p p3-pl ; p p p p p

7.2 Agreement Marking Typology

Those dialects which group together with respect to subject marking patternsare not necessarily the ones showing similar agreement marking properties. Thisis summarized in the table in (52).

(52)agreement dharwari gowari marhet.hi pun.e war.brah. k~okan. iAgrS (nom)

p p p ; ; ;AgrS (non-nom) ; ; p ; ; ;AgrO (nom) ; p ; p p pAgrO (non-nom) ; ; ; ; ; p

Note again that case-marking and agreement patterns do not classify uni-formly across all dialects, but rather cross-cut between dialects. The mostimportant addition to the earlier range in (17) is that of Gowari. While itssubject-marking system is identical to the Pun. e or standard Marat.hi dialect,its agreement system indicates that the null-marked local person subjects areactually treated as true nominatives and thus show agreement.

14The precise features of the unmarked �rst and second person will be discussed later.

35

8 OT Analysis of Dialectal Variation

Using the constraints presented in x4 and x5, we can account for the parallelpatterns of variation in subject-marking patterns within Marat.hi dialects. In(53) we show the ranking of these constraints, along with the dialect system inwhich they are actually instantiated. The possibilities that are not found withinMarat.hi dialects have been instantiated by other languages.

(53) � *struc (Dharwari, Hal.bi)

*Su/3 & *�c � *struc (Pun. e, Gowari, Marhet.hi)

*syncret/sg � *struc (K~okan. i, Warhad. hi Brahman. i)

*Su/loc & *�c � *struc (Hindi)

The constraints on perfect subjects may not be mututally reranked as theyare based on universal hierarchies, but may only be reranked with reference to*struc and * syncret/sg.15 The ranking in (53) illustrates that markednessconstraints on subject marking are progressively demoted below *struc { par-tially in K~okan. i and Standard Marat.hi and completely in Dharwari { allowingthe universal avoidance of overt subject marking to emerge.

8.1 Subject-marking in Marat.hi Dialects

As the table in (50) showed, many of the dialect systems here are identical tounrelated language systems seen earlier; for these cases, we do not reproducetableaus but rather direct the reader to the rankings previously shown.

In terms of subject-marking, the Dharwari and Hal.bi dialects are identical tothe Bengali system; Pun. e and Marhet.hi are identical to the standard Marat.hisystem; and K~okan. i and Warhad.hi Brahman. i are identical to Gujarati. Thetableau in (54), however, has not been seen earlier as none of the languagessurveyed showed this system. In Gowari, the order of the two faithfulness con-straints is switched, with cs-fs being high-ranked and faith-lexperf beinglower. As a result, Gowari does not retain abstract ergative case aside fromwhen it is driven by overt morphology.

15The only other system that a reranking of this constraint set predicts is a system in which*syncret/sg is ranked above *struc and *Su/3 & *Su/perf & *�c . This will generate acase-marking pattern which marks only singular subjects and not plural ones. We have notcome across any example of this type of ergative marking in our research on these dialectsand languages.

36

(54) Gowari

cs-fs

*su/3&*�c

*struc

*su/loc&*�c

faith-lexperf

input: Subj(loc)a. S-erg(overt) �!

☞b. S-nom(�) � �c. S-erg(�) �! �

input: Subj(3rd)☞d. S-erg(overt) �

e. S-nom(�) �! �f. S-erg(�) �! �

8.2 Agreement in Marat.hi Dialects

Again, much of the dialectal agreement rankings are identical to earlier systemsin the linguistic typology. Dharwari and Hal.bi pattern like the Bengali andNepali agreement systems that were seen in (36); they have the simplest agree-ment system with no ergative subject case to interact with subject agreement.16

The ranking shared by the Pun. e and Warhad.hi Brahman. i dialects are the sameas were seen in (34) for Standard Marat.hi. These dialects allow agreementwith either subject or object, as long as its case is nominative. The tableauis repeated in (55) for purposes of comparison with the Gowari tableau thatfollows.

16Although we do not treat the issue of dative experiencer subjects here, Dharwari doeshave dative subjects, which do not permit agreement. Thus, the complete set of agreementconstraints for Dharwari would in fact require reference to cases and would therefore moreclosely resemble (55).

37

(55) Pun. e and Warhad.hi Brahman. i

*erg/agr

*acc/agr

expagr

*oj/agr

*su/agr

*nom/agr

input: S O(spec) V(perf)a. S-erg O-acc V-Sagr �! �b. S-erg O-acc V-Oagr �! �

☞c. S-erg O-acc V-�agr �input: S O(nonspec) V(perf)

d. S-erg O-nom V-Sagr �! �☞e. S-erg O-nom V-Oagr � �

f. S-erg O-nom V-�agr �!input: S O V(nonperf)

☞g. S-nom O-nom V-Sagr � �h. S-nom O-nom V-Oagr �!i. S-nom O-nom V-�agr �!

While the ranking of agreement constraints for Gowari is the same as that forPun.e and Warhad.hi Brahman. i, the input is di�erently derived. Unlike the Pun. edialect, unmarked local, perfect subjects in Gowari are nominative and triggeragreement. The input for Gowari clauses, thus di�ers substantively from thePun.e dialect. The e�ect of this on agreement is shown in (56). (The di�erencesare noted in the input speci�cations.)

(56) Gowari

*erg/agr

*acc/agr

expagr

*oj/agr

*su/agr

*nom/agr

input: S(3rd) O(spec) V(perf)a. S-erg O-acc V-Sagr �! �b. S-erg O-acc V-Oagr �! �

☞c. S-erg O-acc V-�agr �input: S(3rd) O(nonspec) V(perf)

d. S-erg O-nom V-Sagr �! �☞e. S-erg O-nom V-Oagr � �

f. S-erg O-nom V-�agr �!input:S(loc)O V(perf)/S O V(nonperf)

☞g. S-nom O-nom V-Sagr � �h. S-nom O-nom V-Oagr �!i. S-nom O-nom V-�agr �!

The crucial di�erence between the Pun. e system in (55) and Gowari in (56)is that true nominative subjects may be inputs in the Gowari perfect clause,if these subjects are local. These subjects then trigger agreement exactly likenon-perfect nominative subjects.

38

The pattern of nominative agreement in (55) and (56) can be described asa simple contrast between nominative and non-nominative arguments, withoutappealing to �ner case distinctions. However, as discussed in x6.3, K~okan. i allowsobject agreement in contexts where both the subject and the object are overtlycase marked. Ergative agreement never occurs, but accusative agreement ispermitted, just as in Gujarati. Here, subjects are not preferred over objectsfor agreement, as might be predicted by the ranking in (32). Rather, �neragreement distinctions are made for di�erent cases in K~okan. i. Thus, the onlydi�erence between agreement in the Pun. e and K~okan. i dialects is the demotionof *acc/agr in K~okan. i. The result is identical to Gujarati in (35) and is notrepeated here.

Finally, Marhet.hi also contradicts the nominative agreement pattern of thePun.e dialect prototype by allowing agreement with the ergative. This violatesthe highest constraint in (33), but is still preferable to agreement with anyobject, resulting in the same ranking as Nepali and Bengali in (36).

To summarize, a majority of the Marat.hi dialect data shows an exact mir-roring of the systems that were seen in the earlier typology of NIA languages.Two main di�erences arose in the typology of Marat.hi dialects as comparedwith that of the NIA languages.

a. No dialect of Marat.hi exists that has a completely distinctive paradigmof overt ergative subject marking in the perfect clause, as Hindi does, forinstance. This is because the parent language, OM, had already reachedan early stage of markedness reduction in the ergative clause. OM had asyncretism between the ergative and the nominative marking of �rst andsecond person plural pronouns; this syncretism is never `undone' in a laterdialect, to give a more di�erentiated system.

b. Gowari �lls a typological gap that is predicted by our earlier analysis,namely morphologically-driven nominative case for local subjects (or, inother words, loss of abstract ergative along with loss of the morphologicalcase-marking).

9 The Diachronic Perspective

The language and the dialectal data discussed above reveal almost uniformpatterns of what we term `markedness reduction.' The emerging typology ofthe linguistic systems of the languages examined, is constrained and convergeson relatively transparent strategies for reducing markedness of paradigms. Webelieve that these strategies are also interesting from a diachronic perspective,in that they provide empirical evidence for directionality of linguistic change.It is beyond the scope of this paper to account for the diachronic trajectory ofeach of these languages in a formal model.

The typology of variation in the dialectal data, for the most part mirrorsthe variation typology found in the language data, especially in the agreementpatterns. However, in the ergative nominal paradigm, Marat.hi dialects fail to

39

show the same range of variation. Speci�cally, a particular constraint ranking,viz. that in which *struc is ranked below all the markedness constraints (theranking deriving Hindi and Nepali perfect subject marking) is not found in thedialects we discuss. Moreover, this ranking and the resultant paradigm (withovert ergative marking on all members of the nominal paradigm) is not foundin any of the dialectal data that Grierson (1905) discusses.

The gap in this factorial typology would appear to be accidental if we failto notice a crucial fact. All the dialects of Marat.hi, it would be reasonable toassume, emerge from (some variant of) Old Marat.hi. The nominal paradigmof Old Marat.hi does not have overt marking on all ergative subjects. There issyncretism in ergative marking in the plural pronouns of the �rst and secondperson. None of the dialects of Marat.hi alter the constraint ranking that derivesthe Old Marat.hi paradigm so as to reduce this sycretism in number. In otherwords, no dialect of Marat.hi changes to become more like Hindi or Nepali; or nodialect of Marat.hi demotes *struc below the markedness constraint, *Su/loc& *�c . In the empirical data that we see and possibly diachronically, the*struc constraint progressively demotes the markedness constraints on nomi-nal paradigms. There is no corresponding dialect with promotion of markednessconstructions over the *struc constraint. This points to a progressive linguisticchange in one direction towards an unmarked nominal paradigm. This unidi-rectional change in constraint ranking is intuitive historically, but can only belooked at as an arbitrary set of events in OT, since OT precludes any inherentdirectionality in constraint ranking and organization.

Historically, the earliest ancestor of the modern ergative construction (MIA)was morphologically similar to the modern Hindi construction with ergativemarking in the entire person-number nominal paradigm. The Old and MiddleMarat.hi construction (1200-1800 AD) showed a pattern similar to the K~okan. i-Warhad.hi Brahman. i pattern, where *struc demotes the constraint on localperfect subjects. The Pun. e dialect represents a further stage where *strucdemotes the *syncret/sg constraint. Dharwari and Hal.bi represent a laterstage in the dialect, where *struc has demoted all the markedness constraints.

Three stages along this path of change can de�nitely be reconstructed withinthe diachronic path of the standard (Pun. e) dialect of Marat.hi. The fourth stage,total loss of ergativity is to be seen in Dharwari, and can probably be analyzedas a further stage arising from the same trajectory, that the standard dialecthas not yet undergone.

The following �gure is the same as (25) in x4, but is expressed as a diachronicpath rather than a synchronic, typological distribution. It also indicates the roleof the change in syncretism patterns within the broader system.

40

(57) � *struc (Dharwari, Hal.bi)

*Su/3 & *�c � *struc (Standard Marat.hi (Pun. e) (1800-now)

*syncret/sg � *struc (Old Marat.hi (1000-1800 AD)

*Su/loc & *�c � *struc (Middle Indo-Aryan (600-800 AD)

This ranking of constraints, considered as a diachronic path, suggests theexpectation that periods of categorical constraint ranking are interspersed withperiods of variable constraint ranking. This expectation is borne out by theolder data for the Pun.e dialect shown in (58).

The data that was given for the Pun. e dialect in x6.4 was not taken fromGrierson (1905) but rather was data from the contemporary dialect. Grierson'ssample grammar and text, by contrast, show that the nominal case-markingpattern in the Pun. e dialect was di�erent from the contemporary variety thatwas given in x6.4. At the time of Grierson's writing (1905), the Pun. e dialecthad overt marking on singular, local person transitive, perfect subjects. Theseovertly marked subjects were in variation with the non-overtly marked subjects.The variable paradigm of Marat.hi (Grierson 1905) is given below.

(58)number

aspect personsingular plural

Non-perf 1 m�� �amh��Perf 1 m��, my�a �amh��Non-perf 2 t�u tum��Perf 2 t�u, tv�a tum��Non-perf 3 to tePerf 3 ty�a-ni ty�a-nni

In the modern Pun. e dialect, unlike the system above, only the syncretized(forms identical to the non-perfect nominative subject) are in use. This suggeststhat a competition between forms in standard Marat.hi has given way to the lossof the more marked form, and the generalization of the syncretized form. Whilewe do not attempt to model it here, this variation and consequent loss of themore marked form can be expressed formally using Stochastic Optimality The-ory. Recent work in this stochastic version of Optimality Theory (see Boersmaand Hayes 2001, Asudeh 2001, Bresnan and Deo 2001, Lee 2001b, among oth-ers) has formalized various aspects of optionality, variable constraint ranking,and diachronic predictions.

Finally, it is interesting that the ranking which yields the typological vari-ation in nominal paradigms also yields a reconstructible diachronic path alongwhich a language developed. Dialectal data has been often used in internalreconstruction, together with historical data, in order to determine paths ofchange in speci�c domains of language change. The dialectal data we examined

41

in this paper also sheds some light on the process of change from the ergativeclause to an active, non-ergative one. Crucially, it points to the relation betweentypology and linguistic change, that may be explored fruitfully within an OTframework, though it is not within the domain of this paper to do so.

10 Conclusion

The ergative construction in modern NIA has been too often analyzed as ahomogenous construction of the classic sort; the variety of morphological vari-ation under this rubric has been for the most part disregarded. To attribute toIndo-Aryan languages an across-the-board ergativity of this kind is to ignore alarger range of data that, according to us, points to the emergence of unmarkedcase and agreement systems. In this paper, we have brought out the charac-teristic patterns of variation within the nominal and verbal ergative paradigmsin a range of Indo-Aryan languages, and derived them from a universal andfunctionally motivated set of constraints. An important insight of this paperis the partial independence of case-marking and agreement systems in many ofthe languages discussed. Deriving nominal and verbal paradigms with indepen-dent sets of constraints, rather than treating agreement as a corollary of case,appears to be the most intuitive way of dealing with these data. The mor-phological facts of Marat.hi dialects strengthened evidence for the observationthat inter-dialectal variation mirrors typological variation, and is derivable bythe same set of constraints. Finally, the limits on variation found in the NIAtypology and more clearly in the Marat.hi dialect data suggests a uniform direc-tionality in markedness reduction, and allows a consideration of the data froma diachronic perspective. This suggests the intuitive consequence that the sameset of constraints can provide a diachronic account of the linguistic changes andsynchronic di�erences in nominal and verbal ergative morphology.

References

Aissen, Judith (1999). Markedness and Subject Choice in Optimality Theory.Natural Language and Linguistic Theory , 17 , 673{711.

Aissen, Judith (2000). Di�erential Object Marking: Iconicity vs. Econ-omy. Ms. University of California, Santa Cruz. Available athttp://ling.ucsc.edu/~aissen/ms.html.

Asudeh, Ash (2001). Linking, Optionality, and Ambiguity in Marathi. In PeterSells (ed.), Formal and Empirical Issues in Optimality-Theoretic Syntax .Stanford, California: CSLI Publications.

Battistella, Edwin L. (1990). Markedness: The Evaluative Superstructure ofLanguage. Albany: State University of New York Press.

42

Beames, John (1966). A Comparative Grammar of Modern Indo-Aryan Lan-guages of india. Munshiram Manoharlal, Delhi.

Bhatia, Tej K. (1993). Punjabi . New York: Routledge.

Bloch, Jules (1965). Indo-Aryan from the Vedas to Modern Times . Paris:Adrien-Maisonneuve. Transl. by Alfred Master.

Boersma, Paul, and John Hayes (2001). Empirical Tests of the Gradual LearningAlgorithm. Linguistic Inquiry , 32 , 45.

Bresnan, Joan (2000). Explaining morphosyntactic competition. In Mark Baltinand Chris Collins (eds.), Handbook of Contemporary Syntactic Theory .Oxford: Blackwell Publishers.

Bresnan, Joan (2001). Optimal syntax. In J. Dekkers, F. van der Leeuw,and J. van de Weijer (eds.), Optimality Theory: Phonology, Syntax andAcquisition. Oxford: Oxford University Press.

Bresnan, Joan, and Ashwini Deo (2001). Grammatical Constraintson Variation: `Be' in the Survey of English Dialects and(Sochastic) Optimality Theory. Stanford University.http://www-lfg.stanford.edu/bresnan/download.html.

Bresnan, Joan, and Jonni Kanerva (1989). Locative Inversion in Chichewa: ACase Study of Factorization in Grammar. Linguistic Inquiry , 20 , 1{50.

Bubenik, Vit (1998). Historical Syntax of Late Middle Indo Aryan(Apabhramsa). Amsterdam: John Bejamins Publishing Co.

Butt, Miriam (2000). A Reexamination of the Accusative to Ergative Shift inIndo-Aryan. Paper presented at LFG 2000 Conference, Berkeley, Cali-fornia.

Bybee, Joan (1994). The Grammaticization of Zero. In William Pagliuca (ed.),Perspectives on Grammaticalization, (pp. 235{254). Amsterdam: JohnBenjamins Publishing Co.

Bybee, Joan, J. R. Perkins, andW. Pagliuca (1994). The Evolution of Grammar:Tense, Aspect, and Modality in the Languages of the World . Chicago:University of Chicago Press.

Chatterji, Suniti Kumar (1970). The Origin and Development of the BengaliLanguage. George Allen and Unwin Ltd. 1st edition [1926].

Comrie, Bernard (1978a). Ergativity. In Winfred P. Lehmann (ed.), SyntacticTypology: Studies in the Phenomenology of Language, (pp. 329{394).Hassocks, Sussex: The Harvester Press.

Comrie, Bernard (1978b). Genitive-accusative in Slavic: the rules and theirmotivation. IRSL, 3 , 27{43.

43

Comrie, Bernard (1989). Language Universals and Linguistic Typology . Oxford:Basil Blackwell, 2nd edn.

Dixon, R. M. W. (1994). Ergativity . Cambridge: Cambridge University Press.

Giv�on, Talmy (1979). On Understanding Grammar . New York: AcademicPress.

Greenberg, Joseph (1966). Language Universals . The Hague: Mouton & Co.

Grierson, G. A. (1905). Linguistic Survey of India VII: Indo-Aryan Family,Southern Group (Marathi), vol. 7. Delhi: Motilal Banarsidass.

Hock, Hans Heinrich (1986). `P-Oriented' Constructions in Sanskrit. In Bh.Krishnamurti (ed.), South Asian Languages: Structure,Convergence andDiglossia. Delhi: Motilal Banarsidass.

Hook, Peter Edwin (1992). On identifying the Conceptual Restructuring ofPassive as Ergative in Indo-Aryan. In Madhav Deshpande and SarojaBhate (eds.), Paninian Studies (Prof. S. D. Joshi Felicitation Volume),no. 4, (pp. 177{199). University of Michigan: Centre for South and SouthEast Asian Studies.

Klaiman, M. H. (1978). Arguments Against a Passive Origin of the IA Ergative.In Karol Todrys Donka Farkas, Wesley Jacobsen (ed.), Papers from thefourteenth regional meeting , (pp. 204{216). Chicago Linguistics Society.

Kuhn, Jonas (2001). Formal and Computational Aspects of Optimality-theoreticSyntax. Doctoral Dissertation, IMS, University of Stuttgart.

Lee, Hanjung (2002a). A Lexicalist OT Approach to Case. Ms. Presented atStanford University, April, 2002.

Lee, Hanjung (2002b). Optimization in Argument Expression and Interpreta-tion: A Uni�ed Approach. Doctoral Dissertation. Stanford University.

Masica, Colin P. (1991). The Indo-Aryan Languages . Cambridge LanguageSurveys. Cambridge, MA: Cambridge University Press.

Master, Alfred (1964). A Grammar of Old Marathi . Oxford: Clarendon Press.

Mistry, P. J. (1997). Objecthood and Speci�city in Gujarati. In Lyle CampbellJane Hill, P.J. Mistry (ed.), The Life if Language: Papers in Honour ofWilliam Bright . Mouton de Gruyter.

Mohanan, Tara (1994). Argument structure in Hindi . Stanford: CSLI Publica-tions.

Nordlinger, Rachel (1998). Constructive case: Dependent-Marking Noncon�gu-rationality in Australia. Stanford: CSLI.

44

Peterson, John M. (1998). Grammatical Relations in Pali and the Emergenceof Ergativity in Indo-Aryan. M�unchen: Lincom Europa.

Plank, Frans (1979). Ergativity, Syntactic Typology, and Universal Grammar:Some past and present viewpoints. In Ergativity: Towards a Theory ofGrammatical Relations . New York: Academic Press.

Prince, Alan, and Paul Smolensky (1993). Optimality Theory: constraint inter-action in generative grammar . To appear in: Linguistic Inquiry Mono-graph Series. Cambridge, MA: MIT Press.

Sharma, Devyani (2001). Kashmiri case clitics and person hierarchy e�ects. InPeter Sells (ed.), Formal and Empirical Issues in Optimality-TheoreticSyntax . Stanford, California: CSLI Publications.

Sharma, Devyani (to appear). Kashmiri case clitics and person hierarchy e�ects.In Peter Sells (ed.), Formal and Empirical Issues in Optimality-TheoreticSyntax . Stanford, California: CSLI Publications.

Smolensky, Paul (1995). On the Internal Structure of the ConstraintComponent Con of UG. ROA-86-0000. Rutgers Optimality Archive.http://ruccs.rutgers.edu/roa.html.

Stump, Gregory (1983). The elimination of ergative patterns of case-markingand verbal agreement in Modern Indic Languages. Ohio State WorkingPapers in Linguistics , 27 .

Wallace, William D. (1982). The evolution of ergative syntax in Nepali. Studiesin the Linguistic Sciences , 12 , 147{209. University of Illinois, Urbana.

Woolford, Ellen (2000). Case Patterns. In G. Legendre, J. Grimshaw, , andS. Vikner (eds.), Optimality-theoretic Syntax . Cambridge, Mass.: MITPress.

45