thematic inversion in georgian (blevins)

34
Thematic inversion in Georgian * James P. Blevins University of Cambridge Abstract. Inversion constructions in Georgian appear to invert canonical case and agree- ment patterns. However, the morphological properties of inversion constructions can be subsumed under the default patterns of Georgian if inversion is treated as a case of thematic demotion in which a logical subject is realized as an indirect object. The analysis presented in this paper suggests how a correspondence-based model of argument structure can cap- ture and extend traditional accounts of inversion, while bringing out the organization of the Georgian conjugational system and clarifying the theoretical status of logical subjects. . Introduction Georgian is well known for the construction in (b), in which canonical patterns of case and agreement marking appear to be ‘inverted’. The clause in (a) exhibits the Future pattern, in which the subject st’udent’i occurs in the nominative, and is indexed by the subject agreement sux -s. The indirect object amxanags occurs in what is traditionally termed the ‘dative’ case, and is indexed by the indirect object agreement prefix s-. The direct object c’erils is also dative (there is no distinct accusative case in Georgian), but is not indexed on the verb. The Perfect clause in (b) inverts the marking of arguments. The logical subject, st’udent’s, is dative and co-occurs with the prefix u-, which is the form of the Perfect ‘version vowel’ i- that encodes the selection of an indirect object. The logical indirect object c’erili occurs in the nominative, and is indexed by the subject agreement sux -a. The logical indirect object amxanagis is unindexed and occurs in the genitive case governed by the postposition tvis. () Case and agreement inversion in Georgian (Tschenkéli, , ) a. St’udent’i student. misc’ers .write. amxanags comrade. c’erils. letter. ‘The student will write a letter to his comrade.’ b. St’udent’s student. miuc’eris .write. amxanagis-tvis comrade.-for c’erili. letter. * I am grateful to Farrell Ackerman, Alice Harris, John Moore and two anonymous NLLT reviewers for numerous comments, corrections and suggestions for improvements. Example glosses identify relevant properties of forms, and suppress preverbs, version vowels and other morphological details. Case is indicated by the labels ‘nominative’ ‘dative’, ‘ ergative’ and ‘genitive’. Agreement glosses are composed of the person values , , and the abbreviations ‘subject’, ‘direct object’ and ‘indirect object’. © Kluwer Academic Publishers. Printed in the Netherlands. tig.tex; 7/05/2007; 16:18; p.1

Upload: ricardo-ventosinos

Post on 20-Feb-2015

56 views

Category:

Documents


2 download

TRANSCRIPT

Page 1: Thematic Inversion in Georgian (Blevins)

Thematic inversion in Georgian *

James P. BlevinsUniversity of Cambridge

Abstract. Inversion constructions in Georgian appear to invert canonical case and agree-ment patterns. However, the morphological properties of inversion constructions can besubsumed under the default patterns ofGeorgian if inversion is treated as a case of thematicdemotion in which a logical subject is realized as an indirect object. The analysis presentedin this paper suggests how a correspondence-based model of argument structure can cap-ture and extend traditional accounts of inversion, while bringing out the organization ofthe Georgian conjugational system and clarifying the theoretical status of logical subjects.

. Introduction

Georgian is well known for the construction in (b), in whichcanonical patterns of case and agreement marking appear to be ‘inverted’.The clause in (a) exhibits the Future pattern, inwhich the subject st’udent’ioccurs in the nominative, and is indexed by the subject agreement sux-s. The indirect object amxanags occurs in what is traditionally termedthe ‘dative’ case, and is indexed by the indirect object agreement prefix s-.The direct object c’erils is also dative (there is no distinct accusative casein Georgian), but is not indexed on the verb. The Perfect clause in (b)inverts the marking of arguments. The logical subject, st’udent’s,is dative and co-occurs with the prefix u-, which is the form of the Perfect‘version vowel’ i- that encodes the selection of an indirect object. The logicalindirect object c’erili occurs in the nominative, and is indexed by the subjectagreement sux -a. The logical indirect object amxanagis is unindexed andoccurs in the genitive case governed by the postposition tvis.

() Case and agreement inversion inGeorgian (Tschenkéli, , )a. St’udent’i

student.misc’ers.write.

amxanagscomrade.

c’erils.letter.

‘The student will write a letter to his comrade.’b. St’udent’s

student.miuc’eris.write.

amxanagis-tviscomrade.-for

c’erili.letter.

* I amgrateful to Farrell Ackerman,AliceHarris, JohnMoore and two anonymousNLLTreviewers for numerous comments, corrections and suggestions for improvements.

Example glosses identify relevant properties of forms, and suppress preverbs, versionvowels and other morphological details. Case is indicated by the labels ‘nominative’ ‘dative’, ‘ ergative’ and ‘genitive’. Agreement glosses are composed of the personvalues , , and the abbreviations ‘subject’, ‘direct object’ and ‘indirect object’.

© Kluwer Academic Publishers. Printed in the Netherlands.

tig.tex; 7/05/2007; 16:18; p.1

Page 2: Thematic Inversion in Georgian (Blevins)

James P Blevins

‘The student has (apparently) written a letter to his comrade.’

Much of the theoretical interest in inversion derives from the fact thatthe logical subject st’udent’s in (b) exhibits the same case and agreementproperties as the indirect object amxanags in (a), while the logical objectc’erili in (b) patterns with the subject in (a). Analyses of this correspon-dence adopt one of two basic strategies. A traditional approach, which isoutlined in Tschenkéli () and developed in greater detail in Harris() and Tuite (), treats the logical subject in an inversion con-struction as a surface indirect object. On this view, inversion involves anoncanonical mapping between ‘logical’ (or ‘initial’) argument structureand ‘surface’ grammatical functions or relations. Inversion disrupts the de-fault, prominence-preserving, association between logical and surface ar-guments by expressing the logical subject as an indirect object. Case andagreement patterns, along with other distinctive properties of inversionconstructions, are then attributed, directly or indirectly, to this mapping.

The alternative proposed inAnderson (, ) andAronson ()expresses a fundamentally dierent intuition by locating inversion in therelation between surface grammatical functions and inflectional proper-ties. On this view, inversion represents a ‘quirky’ valence pattern in whichthe surface subject is marked dative and indexed by agreement markersthat are normally associated with indirect objects, and the object occurs inthe nominative and is indexed by agreement markers that otherwise en-code properties of the syntactic subject. The principal appeal of this accountcomes from its thematic uniformity, since logical subjects are realized assurface subjects in inverted as well as in direct constructions. Yet thematicuniformity complicates the morphosyntactic system, by introducing a newvalence pattern with highly idiosyncratic case and agreement properties.

It might appear that these alternatives merely stipulate the idiosyn-cratic properties of inversion constructions at dierent grammatical lev-els. Whereas the first analysis assigns a construction-specific grammati-cal function, the second specifies exceptional case and agreement features.However, dierences between these accounts emerge when their eectson the Georgian grammatical system as a whole are taken into account.

.. T

An analysis that treats the logical subject as an indirect object accountsimmediately for the case and agreement properties of the dative and nom-inative arguments in inversion constructions. This analysis assimilates themorphosyntactic properties of the Perfect construction in (b) to the Fu-ture pattern in (a) by maintaining a uniform association between sur-

tig.tex; 7/05/2007; 16:18; p.2

Page 3: Thematic Inversion in Georgian (Blevins)

Thematic inversion in Georgian

face grammatical functions and case and agreement features. In both con-structions, nominative arguments are syntactic subjects, indexed by sub-ject agreement markers, while dative arguments are syntactic objects orindirect objects, indexed – where possible – by object agreement markers.

Treating the logical subject as an indirect object also accounts for theform of the other arguments in inversion constructions. In constructionswith a logical object, the ‘demotion’ of the logical subject induces the ‘pro-motion’ of the object. The logical object c’rili ‘letter’ in the ditransitive in(b) and its counterpart surati ‘picture’ in the transitive in (b) both surfacewith the case and agreement properties appropriate to a surface subject.

() Future and Perfect transitives (Tschenkéli, , )a. Mxat’vari

painter.daxat’avs.paint.

surats.picture.

‘The painter will paint a picture.’b. Mxat’vars

painter.dauxat’avs.paint.

surati.picture.

‘The painter (apparently) has painted a picture.’

Treating the dative argument as an indirect object also suggests an ex-planation for the selective detransitivizing eect of inversion. As examples() and () show, inversion does not invariably reduce the number of gram-matical functions that are directly governed by a verb. In the transitivepattern in (), the logical subject and object are expressed as verbal depen-dents in both the Future and Perfect constructions, with the characteristic‘inversion’ of case and agreement properties. In (), the logical subject ofthe ‘absolute’ intransitive ’ ‘dance’ is realized as nominative Merabiin the Present construction and as dativeMerabs in the Perfect.

() Present and Perfect ‘absolute’ intransitives (Harris, , f.)a. Merabi

Merab.cek’vavs.dance.

‘Merab is dancing.’b. Merabs

Merab.ucek’via..dance.

‘Merab apparently (has) danced.’

The analyses of Georgian in Wunderlich () and Joppen-Hellwig () adopt asimilarly direct mapping between grammatical functions and morphological properties.

Following traditional descriptions, the term ‘transitive’ is reserved for verbs that gov-ern a direct object. Intransitive verbs are divided into two classes: ‘absolute’ verbs, whichjust govern a subject, and ‘relative’ verbs, which also select an indirect object.

tig.tex; 7/05/2007; 16:18; p.3

Page 4: Thematic Inversion in Georgian (Blevins)

James P Blevins

Yet a logical indirect object cannot be realized as a direct verbal depen-dent in inversion constructions. In the ditransitive Future clause in (a)and in the intransitive Present clause in (a), the logical indirect objectsamxanags and p’ropesors are realized as dative verbal dependents. But inthe corresponding Perfect clauses, both elements are realized as postposi-tional objects. In (b), amxanagis occurs as the genitive object of tvis ‘for’,whereas in (b), p’ropesor occurs as the dative object of tan ‘to’.

() Present and Perfect ‘relative’ intransitives (Tschenkéli, , )a. St’udent’i

student.elap’arakeba.speak.

p’ropesors.professor.

‘The student is speaking with the professor.’b. St’udent’s

student.ulap’arakia.speak.

p’ropesor-tan.professor.-to

‘The student has (apparently) spoken with the professor.’

The evident generalization is that logical indirect objects cannot surface asindirect objects in inversion constructions. This restriction cannot be at-tributed to any general case or agreement constraints, given that Georgianfreely allowsmultiple dative arguments in uninverted clauses such as (a).

Is Harris () argues, a traditional analysis oers a natural account ofthis restriction, given a principle – such as Stratal Uniqueness (Perlmutterand Postal, ) – which independently allows at most one surface indi-rect object. If the logical subject is realized as a grammatical indirect objectin inversion constructions, it follows that no other argument will also beable to surface as an indirect object. The logical indirect object must thenbe expressed as a more oblique dependent, which in Georgian takes theform of an object of a postposition such as tvis or tan.

Furthermore, a traditional analysis oers a principled basis for deter-miningwhich verbs will participate in inversion. If inversion targets logicalsubjects, it should not apply to verbs that lack logical subjects. There isa clear expectation that ‘unaccusative’ verbs, in the sense of Perlmutter() should not occur in inversion constructions. Georgian grammarsconventionally recognize a class of ‘passive’ verbs, containing passives, in-choatives and other types of unaccusatives. Example (a) illustrates onetype of true passive, corresponding to the active transitive in (a).

() Future and Perfect ‘absolute’ passives (Tschenkéli, , /)a. Surati

painting.ixat’eba.paint.

‘The picture will be painted.’b. Surati

painting.uk’vealready

daxat’uliq’opaint.

tig.tex; 7/05/2007; 16:18; p.4

Page 5: Thematic Inversion in Georgian (Blevins)

Thematic inversion in Georgian

‘The picture (apparently) has been painted.’

Passive verbs have full conjugational paradigms, which include eviden-tial Perfect forms such as daxat’uliq’o in (b). However, as Harris ()observes, Perfect constructions containing unaccusative verbs do not ex-hibit inversion. In both the Present (a) and the Future (b), the logicalobject surati ‘picture’ surfaces as a nominative grammatical subject. Unac-cusatives with indirect objects show a similar pattern. As in (), the logicalobject masc’avlebeli ‘teacher’ surfaces as a nominative grammatical sub-ject in the Present and Perfect constructions in (). Moreover, unlike thePerfect constructions in (b) and (b), the logical indirect object moc’apes‘pupil’ is able to surface as a grammatical indirect object in (b).

() Present and Perfect ‘relative’ unaccusatives (Tschenkéli, , )a. Masc’avlebeli

teacher.udzavrdeba.angers.

moc’apes.pupil.

‘The teacher grows angry at the pupil.’b. Masc’avlebeli

teacher.gasdzavrebia.angers.

moc’apes.pupil.

‘The teacher (apparently) has grown angry at the pupil.’

Voice alternations in other verb classes confirm the role of logical sub-jects. Georgian contains a class of ‘indirect’ or ‘aective’ verbs, which ex-hibit inversion throughout their entire conjugational paradigm. In the Presentconstruction in (a), the logical subject of sduls ‘hates’ is realized as a dativeindirect object and the logical object as a nominative surface subject. Yet asHarris (, ) notes, the passive counterparts of indirect verbs do notexhibit inversion. Instead, as (b) illustrates, the logical objects of indirectpassives surface as grammatical subjects, just as with other unaccusatives.

() Present and Passive ‘indirect’ verbs (Harris, , )a. Vanos

Vano.sdzuls.hate.

direkt’ori.director.

‘Vano hates the director.’b. Direkt’ori

teacher.šedzulebulia.hate.

‘The director is hated.’

.. F

In short, treating inversion as a non-canonical mapping between logicalarguments and surface grammatical functions oers a natural account for

tig.tex; 7/05/2007; 16:18; p.5

Page 6: Thematic Inversion in Georgian (Blevins)

James P Blevins

restrictions on inversion and for the surface form of inversion construc-tions. Few of these properties follow from an account in which inversionrepresents a quirky surface valence pattern. Classifying logical subjects asdative-marked subjects provides no basis for the selective detransitivizingeect of inversion. As the examples in Section . show, a logical indirectobject can surface as a dative argument in a ditransitive constructionwith adative direct object, or in an inversion construction with a nominative sub-ject. It is only when the logical subject is realized with the case and agree-ment properties of an indirect object that the logical indirect object mustbe expressed as a postpositional object. By targeting logical subjects, in-version also patterns strikinglywith passivization, which exhibits the samerestriction, apparently universally (Perlmutter and Postal, ). A quirkyvalence analysis cannot capture this pattern in terms of surface grammati-cal functions – since unaccusative and unergative verbs both have surfacesubjects – but must instead refer to whatever syntactic or semantic prop-erties are taken to distinguish logical subjects. Hence, while a quirky va-lence analysis greatly disrupts the association between grammatical func-tions and case and agreement properties, the analysis cannot be formulatedsolely with reference to surface arguments.

... Relational demotionThe principal challenge for a traditional account lies in characterizing thelevels of ‘logical’ and ‘grammatical’ argument structure and in specifyingthe principles that associate these levels. The account inHarris () con-strues logical and grammatical arguments in terms of the and grammatical relations proposed within Relational Grammar (RG). Logicalarguments are represented at the initial stratum of a relational network,where a logical subject bears a ‘’ (subject) relation, a logical direct objectbears an initial ‘’ (object) relation and a logical indirect bears an initial ‘’(indirect object) relation. The relations assigned in the final stratum of anetwork represent the surface argument structure that mainly conditionscase and agreement patterns. Harris formulates the inversion rule as aprocess that demotes a subject to an indirect object, and attributes the pro-motion of the object to an independent rule of Unaccusative Advancement:

According to this hypothesis, Inversion is a single process:

c. Inversion Subject → Indirect Object.

The initial indirect object is put en chômage, according to the ChômeurCondition. The direct object advances to subject, by Unaccusative. In-

Harris () provides additional support for a traditional analysis, based on phe-nomena such as ‘Object Camouflage’, which is discussed in more detail in Section ...

tig.tex; 7/05/2007; 16:18; p.6

Page 7: Thematic Inversion in Georgian (Blevins)

Thematic inversion in Georgian

version proper is just the process of a subject becoming an indirectobject. (Harris, , )

To account for the fact that unaccusative verbs do not undergo inversion,Harris restricts the application of inversion to initial subjects.

() The Initial Subject Constraint on Inversion (Harris, , )Inversion applies only to initial subjects

The resulting analysis captures the insights of traditional accounts andextends these accounts by oering general explanations for patterns of ad-vancement, and for restrictions on the application of the inversion rule. Theprincipal objection to an analysis based on relational demotion takes issuewith the claim that inversion “reflects an actual change in syntactic struc-ture” (Anderson, , ). However, it is important to recognize thatthis objection is directedmore at a derivational of RG thanat the specific treatment of inversion inHarris (). It is true that RGwasdeveloped as a derivational syntactic model and that Harris () adoptsthis perspective. But, at the same time, it must be acknowledged that RGanalyses tend to make fairly minimal reference to constituency, word or-der or other features of syntactic ‘arrangement’, in the sense of Bloomfield(). Hence many of the patterns of promotion and demotion withinRG analyses can be reinterpreted as lexical alternations, in which stratacorrespond to dierent aspects of the argument structure of a predicate.

The analysis of inversion inHarris () provides a clear case in point,since the relational networks in this analysis contain almost no informa-tion about the form or arrangement of syntactic dependents. Instead, therelational networks proposed in Harris ()mainly express alternationsin the patterns of government or argument associatedwith verbsbelonging to dierent lexical classes and inflectional series. The analysis ofthese alternations do not depend in any fundamental way on the idea thatinversion “reflects an actual change in syntactic structure”.

... Thematic demotionMore generally, the central insights of a traditional analysis are not tiedin any way to a derivational perspective, and are entirely compatible withcorrespondence-based approaches to argument structure. One could refor-mulate the analysis of inversion within a correspondence-based version ofRG, along the lines suggested in Blevins (). However, many of theprerequisites of this analysis are already present in themodels of argumentstructure developed in lexicalist models such as Lexical Functional Gram-mar (LFG) and Head-driven Phrase Structure Grammar (HPSG). A ruleof inversion can be formulated in any approach that distinguishes ‘logical’and ‘grammatical’ levels of argument structure. However, Lexical Mapping

tig.tex; 7/05/2007; 16:18; p.7

Page 8: Thematic Inversion in Georgian (Blevins)

James P Blevins

Theory (LMT) oers a particularly transparent framework for representinglevels and the principles that mediate between them in Georgian.

Logical argument structure is usually characterized in terms of pred-icates and participant roles, though the treatment of roles varies acrossapproaches. Following in the tradition of Fillmore () and Jackendo(), the earliest LMT accounts (Bresnan and Kanerva, ; BresnanandMoshi, ; Bresnan and Zaenen, ) tend to treat roles as discreteobjects that are organized into a hierarchy, as in ().

() Partial thematic hierarchyagent ≻ benefactive ≻ goal ≻ patient

Dowty () provides the basis for a group of later accounts (Ackerman,; Zaenen, ; Alsina, ; Ackerman andMoore, ), in whichparticipant roles are regarded as clusters of Proto-Agent and Proto-Patientproperties. These accounts introduce “an opposition between just two roles… Proto-Agent and Proto-Patient” (Dowty, , ) and correlate therank of a role with number of Proto-Agent properties that it contains.

In principle, thematic inversion can be formulated in terms of discreteroles or cluster concepts. However, proto-role approaches provide an es-pecially straightforward treatment of a logical subject as the highest rank-ing Proto-Agent role in an argument structure (corresponding to Alsina’s() notion of an ‘external argument’, which in turn derives from the dis-tinguished ‘external’ argument in Williams ()). A proto-role accountcan associate a transitive verb, such as daxat’avs ‘paint’ in (a), with anargument structure ‘⟨Ag Pat⟩’, containing an initial Proto-Agent role ‘Ag ’,and associate an unaccusative verb, such as udzavrdeba ‘grow angry’ in (a)with an argument structure ‘⟨Pat Pat⟩’ containing only Proto-Patient roles.An inversion rule that realizes logical subjects as grammatical indirect ob-jects will then apply to the first argument structure, but not the second,capturing the traditional contrast between these classes of predicates.

The second prerequisite of a thematic inversion rule is an appropriatenotion of ‘surface indirect object’. The initial formulations of LFG in Bres-nan () distinguish the core grammatical functions (), (), and (). A given clause may contain at most one of each func-tion, as a consequence of the Uniqueness condition, which stipulates that“in a given f-structure, a particular attribute may have at most one value”(Kaplan and Bresnan, , ). Of these functions, the ‘secondary objectfunction’ is the closest counterpart of a traditional indirect object.

To describe natural classes of grammatical functions, LMT accounts de-fine the corresponding functions , , θ and θ in terms of afeature [±r(estricted)] that distinguishes ‘thematically restricted’ functionsand a feature [±o(bjective)] that characterizes ‘objective’ functions.() LMT feature analyses (Bresnan and Kanerva, , )

tig.tex; 7/05/2007; 16:18; p.8

Page 9: Thematic Inversion in Georgian (Blevins)

Thematic inversion in Georgian

[−r−o

[+r−o

]

[−r+o

[+r+o

]From the present perspective, the main eect of the revision in ()

is the replacement of the unitary grammatical functions and by“θ and θ [which] represent families of relations indexed by semanticroles, with the θ subscript representing the semantic role associated withthe argument” (Dalrymple, , f.). Indirect objects in Georgian couldbe classified as thematically restricted θ functions. But this provides nomeans of expressing the fact that inversion constructions allow exactly oneindirect object, because thematically restricted objects with dierent re-strictions count as dierent functions. As a result, neither the Uniquenesscondition of Kaplan and Bresnan () nor the subsequent Function-Argument Biuniqueness condition of Bresnan andKanerva () restrictsthe number of ‘indirect’ or ‘secondary’ objects, provided that they havedierent thematic restrictions. In particular, there is no obvious way toprevent an inversion construction in which the logical subject is realizedby an agent and the logical indirect object is realized by an goal.

The problemdoes not depend on the choice of thematic restriction. Sub-stituting proto-roles for discrete roles does not make any dierence either,since the logical subject will correspond to a Proto-Agent role that is real-ized by an indirect object, while the logical indirect object will correspondto a Proto-Patient. Clearly, the same diculties arise if indirect objects areanalyzed as thematically restricted obliques, or sometimes as thematicallyrestricted objects and other times as thematically restricted obliques. It isbecause θ represents a ‘family of functions’ rather than a single func-tion that there is no straightforward way to capture the generalization thatinversion constructions allow a single indirect/secondary object.

The simplest alternative can simply require the uniqueness of each ofthe four distinct feature specifications in (), so that oblique and sec-ondary object functions are unique in a lexical form. The property [+r]can still be taken to characterize thematically restrictable functions. Butthe thematic subscripts can be omitted where irrelevant, with the eectthat θ and θ reduce, for present purposes, to and . With therehabilitation of indirect/secondary objects as ‘first class objects’ in LMT,the thematic inversion rule can be formulated as in ().

() Thematic Inversion RuleAg|

tig.tex; 7/05/2007; 16:18; p.9

Page 10: Thematic Inversion in Georgian (Blevins)

James P Blevins

The simplified a(rgument)-structures in () then distinguish the de-fault pattern of argument realization associated with the Future construc-tion in (a) from the inverted pattern associated with the Perfect in (b).

() Future and Perfect transitive a-structuresa. daxat’avs ⟨ Ag Pat ⟩

| |

b. dauxat’avs ⟨ Ag Pat ⟩| |

In the a-structure in (a), Proto-Agent role is associated with the function and the Proto-Patient with the , reflecting the general patternsof argument selection in Dowty (). In (b), the inversion rule re-alizes the Proto-Agent as an indirect object. The demotion of the logicalsubject induces the advancement of the Proto-Patient, when the principlethat associates it with the highest available grammatical function maps itto the function, rather than – as in (a) – to the function.

.. O

The body of the paper now shows how a correspondence-based modelcan capture and extend the insights of traditional analyses. The contri-bution of this type of model lies partly in the factorization that it oers.Defining ‘logical’ argument structure in thematic terms expresses the cen-tral role that semantic properties play in guiding argument realization.The level of ‘surface’ grammatical functions is likewise the natural locusof syntactic and morphosyntactic phenomena, such as agreement and casegovernment. The division of grammatical labour between levels within ana-structure is largely attributable to intrinsic dierences between the typesof elements that comprise each level, rather than to constraints that stip-ulate which kinds of phenomena are conditioned by elements at dierentlevels or strata. In addition, there is a transparent interface between thegrammatical functions in an a-structure and the syntactic analyses of anLFG (or HPSG) grammar. This permits a more detailed treatment of caseand agreement patterns than is developed in traditional or RG accounts.

. Inversion in Georgian

An overview of the conjugational system of Georgian will help to clarifyhow inversion constructions fit into the larger morphosyntactic patternsof the language. The system is defined in large part by four

Argument selection in LMT accounts is usually mediated by the assignment ofspecifications for [±o] and [±r] to roles, as discussed in Section . below.

tig.tex; 7/05/2007; 16:18; p.10

Page 11: Thematic Inversion in Georgian (Blevins)

Thematic inversion in Georgian

(Aronson, ) or (Harris, ) and three basic inflectional -. Although the conjugations are true form classes, defined in terms ofFuture and Aorist principal parts (Harris, , ), it is more informa-tive to refer to conjugations in terms of the associated valence classes in(). The first conjugation consists mainly of transitive verbs. The second‘passive’ conjugation containsmostly unaccusative intransitives, which areoften derived from first conjugation transitives. The third ‘medial’ conju-gation mainly contains unergative intransitives. The fourth ‘indirect’ con-jugation consists of verbs that exhibit thematic inversion in all series.() Georgian verb classes (cf. Tschenkéli , ; Harris, , )

Conjugation Valence Class Example

Transitive daxat’avs ‘s/he will paint it’ Unaccusative darčeba ‘s/he will remain’ Unergative icek’vebs ‘s/he will dance’ Indirect mosc’ons ‘s/he likes it’

The three inflectional series contain sets of paradigms, traditionally termed. It is againmoremnemonic to refer to these series by their tense/aspectproperties than by the usual Roman numerals I, II and III. The Present se-ries (Series I) consists of subseries that contain present and future screeves.TheAorist series (Series II) contains aorist screeves, and the Perfect/Evidentialseries (Series III) contains inverted screeves, which are residually perfectin form but usually evidential in meaning. The basic organization of seriesand screeves is summarized in ().

() Classification of Georgian screeves (Aronson, , )

Series ‘Tense’ Nonpast Past Subjunctive

I Present Present Indicative Imperfect Conjunctive PresentI Future Future Indicative Conditional Conjunctive FutureII Aorist Aorist Indicative OptativeIII Perfect Present perfect Pluperfect

.. V

Each tense/aspect series is associated with distinctive patterns of agree-ment and case government. The patterns exhibited by transitive and unerga-tive verbs in the Aorist and Present series are shown in ().

This classification departs from Aronson () in substituting ‘subjunctive’ for hislabel ‘modal’ and in treating the present perfect as a nonpast rather than as a past screeve.

tig.tex; 7/05/2007; 16:18; p.11

Page 12: Thematic Inversion in Georgian (Blevins)

James P Blevins

() Direct unergative patterns (Tschenkéli, , )

a. St’udent’mastudent.

misc’era.write.

amxanagscomrade.

c’erili.letter.

‘The student wrote a letter to his comrade.’b. St’udent’i

student.misc’ers.write.

amxanagscomrade.

c’erils.letter.

‘The student is writing a letter to his comrade.’

Example (a) illustrates pattern ‘A’, which is characteristic of the Aoristseries. In this pattern, the grammatical subject occurs in the ergative, theobject occurs in the nominative, and the indirect object occurs in the dative.Example (b), repeated from (a), exhibits pattern ‘B’, which is associatedwith the Present series. In this pattern, the grammatical subject is nom-inative, and the object and indirect object are both dative. Both patternsexhibit a ‘direct’ (i.e. non-inverted) correspondence between logical argu-ments and surface grammatical functions, and a similarly direct relationbetween surface functions and case patterns, as summarized in ().() Direct valence patterns in Georgian

Subject Object Indirect Object

A B

One should not read too much into the traditional case labels ‘erga-tive’ and ‘dative’ in (). As Tschenkéli (, ) remarks, “the so-calledergative” is “also named the active or narrative”. Hence the use of the caselabel ‘ergative’ does not imply that ergative arguments enter into an erga-tive/absolutive opposition (Anderson, ), rather than an active/inactiveopposition (Harris, ; Harris, ). Tschenkéli’s use of the term ‘da-tive/accusative’ for the case that is often glossed simply as ‘dative’ likewiseunderscores the fact that there is no distinct accusative form in Georgian,so that the dative may mark direct as well as indirect objects.

Unaccusative verbs exhibit a slightly dierent association with the pat-terns in (). In the present series, unaccusative verbs, like their unergativecounterparts, follow pattern B. This is shown by the Present form ec’erebain (b), which selects a nominative subject and a dative indirect object.In the Aorist series, unaccusatives dier markedly from their unergativecounterparts. Whereas unergative (and transitive) verbs follow pattern Ain the Aorist series, unaccusatives again follow pattern B. Hence the Aoristformmiec’era in (a) selects a nominative subject, which diers from theergative subject selected by misc’era in (a) and thereby neutralizes thecontrast with the subject of the Present unaccusative in (b).

tig.tex; 7/05/2007; 16:18; p.12

Page 13: Thematic Inversion in Georgian (Blevins)

Thematic inversion in Georgian

() Direct unaccusative patterns (Tschenkéli, , )a. C’erili

letter.miec’era.written.

p’ropesors.professor.

‘The letter was written to the professor.’b. C’erili

letter.ec’ereba.written.

p’ropesors.professor.

‘The letter is (being) written to the professor.’

Unaccusatives also conform to pattern B in the Perfect series, as illus-trated by the examples in (), which are repeated from (b) and (b).

() Perfect ‘absolute’ and ‘relative’ passives (Tschenkéli, , )a. Surati

painting.uk’vealready

daxat’uliq’opaint.

‘The picture (apparently) has been painted.’b. Masc’avlebeli

teacher.gasdzavrebia.angers.

moc’apes.pupil.

‘The teacher (apparently) has grown angry at the pupil.’

Although Perfect forms of unaccusative verbs carry the same evidentialmeaning as unaccusatives, they govern nominative subjects and dativeindirect objects. Hence in series, unaccusatives conform to pattern B.

In direct constructions, agreement, like case, is conditioned by surfacegrammatical functions. To avoid prejudging the analysis of inversion con-structions , agreement markers are described in () using the form-basedterms in Tuite () and Anderson (). The v-and m-sets are eachidentified by their sg marker and the h-set by its sg marker.

() Basic agreement marker sets

v-set m-set h-set

sg v— m— m—pl v— t gv— gv—sg — g— g—pl — t g—t g—tsg — s,a,o — h/s—pl — an/en,es,nen — h/s—

The dashes in () represent verb stems, commas separate morphologically condi-tioned alternants, and slashes separate phonologically conditioned variants.

tig.tex; 7/05/2007; 16:18; p.13

Page 14: Thematic Inversion in Georgian (Blevins)

James P Blevins

As summarized in (), v-set markers index ergative subjects in patternA, and the nominative subjects in pattern B. The m-set markers index thenominative object in pattern A and the dative direct object in pattern B.The h-set markers index dative indirect objects in both patterns.

() Direct case and agreement correlations

v-set m-set h-set

A B

.. T I

Inversion arises in two places in the conjugational system of Georgian.In the Perfect series of transitive and unergative verbs, logical subjectsoccur in the dative and condition h-set agreement. The logical objects oftransitive verbs occur in the nominative and condition v-set agreement.Logical indirect objects cannot be realized as direct verbal arguments andmust be expressed by postpositional phrases. Each of these properties isillustrated in (a), repeated from (b). The class of ‘indirect’ verbs, whichexpress psychological states and modal meanings, also exhibit inversion.Yet, unlike transitive and unergative verbs, indirect verbs exhibit inversionin all series, as illustrated by the inverted Present construction in (b).

() Inversion constructions (Tschenkéli, , /)a. St’udent’s

student.miuc’eris.write.

amxanagis-tviscomrade.-for

c’erili.letter.

‘The student has (apparently) written a letter to his comrade.’b. Mamas

Father.uq’varsio.love.

švili.child.

‘The father loves the child.’

Inversion constructions clearly disrupt the canonical association betweenlogical arguments and morphosyntactic case and agreement properties.The analytical choice concerns the locus of this inversion; whether it oc-curs in the mapping from logical to surface argument structure, or in theassociation between surface arguments and morphosyntactic processes.

tig.tex; 7/05/2007; 16:18; p.14

Page 15: Thematic Inversion in Georgian (Blevins)

Thematic inversion in Georgian

... Morphosyntactic inversionOn the second of these alternatives, inversion exhibits a ‘quirky’ valencepattern ‘C’, in which “the grammatical subject ... is in the dative case” and“the object is in the nominative” (Aronson, , ). This analysismain-tains a uniformmapping between logical arguments and grammatical func-tions, since the logical subject is consistently realized as a surface subjectand the logical object as a grammatical direct object. However, as ()shows, pattern C disrupts the relation between surface argument structureand morphosyntactic properties in a highly pattern-specific way.

() Quirky valence (Aronson , ; Anderson , )

Subject Object Indirect Object

A v-set m-set h-setB v-set m-set h-setC h-set v-set —

Logical Subject Logical Object Logical Indirect Object

Dative subjects that trigger h-set agreement are unique to pattern C, asare nominative direct objects that trigger v-set agreement. The recognitionof a new valence pattern also oers no principled explanation for the factthat the logical indirect object cannot be realized as a grammatical indirectobject in pattern C. If the dative argument is treated as subject or, at anyrate, as something other than an indirect object, there is no evident moti-vation for the demotion of the logical indirect object, given that pattern Bpermits multiple datives. Because each argument in pattern C diers fromcorresponding arguments in A and B, there is no way to assimilate C toeither of the other patterns. Hence,a quirky valence analysis of inversionintroduces a third basic pattern into the morphological system in ().

() Correlations between patterns, series and conjugations

Transitive Unaccusative Unergative Indirect

Present B B B CAorist A B A CPerfect C B C C

In short, quirky valence maintains thematic uniformity at the cost ofrecognizing a third morphological pattern that is distinct in fundamentalrespects from patterns A and B. It is a matter of execution whether thesedierences are expressed directly, by pattern-specific rules, or indirectly,for example, by means of disjunctive case and agreement rules that apply

tig.tex; 7/05/2007; 16:18; p.15

Page 16: Thematic Inversion in Georgian (Blevins)

James P Blevins

to structured morphosyntactic representations with expedient empty lay-ers, along the lines suggested by Anderson (, .). In either case,the postulation of morphosyntactic inversion significantly complicates thedescription of the case and agreement system of Georgian.

... Thematic inversionA traditional analysis proceeds from the observation that there are at mosttwo distinct patterns in Georgian. Correlating case andagreement, as in () identifies the case-agreement correspondence in pat-tern C as a subcase of the pattern B correspondence. Pattern C can then befully assimilated to pattern B if, as in a traditional account, the nominativeis treated as a grammatical subject and the dative as an indirect object.

() Traditional grammatical analyses (Tschenkéli, ; Harris, )

Subject Object Indirect Object

v-set m-set h-setA B (C) —

The analysis in () highlights the close relation between grammaticalfunctions and case and agreement properties established by a traditionalaccount. Ergative case is associated with subjects in pattern A, and dativecase is associated with indirect objects in general. Nominative is associ-ated with the highest available ‘nuclear term’ (Perlmutter and Postal, ,) or with [–r] functions in LFG terms. The relation between agreementmarkers and surface grammatical functions is even more direct. The v-setmarkers index grammatical subjects, the m-set markers index direct ob-jects and the h-set markers index indirect objects. This correspondence isindeed so direct that one can dispense entirely with the designations ‘v-set’,‘m-set’ and ‘h-set’ markers in favour of the more informative terms ‘sub-ject’, ‘object’ and ‘indirect object’ markers, which are adopted in Tschenkéli(), Harris (), Aronson () and Hewitt ().

More generally, a traditional account brings out the morpho-thematicbasis of patterns that appear arbitrary on a quirky valence account. Onthe analysis in (), the distinction between patterns B and C are purelymorpho-thematic. In direct constructions conforming to pattern B, the log-ical subject is realized as a surface subject. In the inverted constructionsthat are assigned to pattern C, the logical subject is realized as an indirectobject. Given this contrast, the other dierences between patterns B and Cfollow from the general morphosyntactic conventions of Georgian.

tig.tex; 7/05/2007; 16:18; p.16

Page 17: Thematic Inversion in Georgian (Blevins)

Thematic inversion in Georgian

Assimilating pattern C to pattern B also identifies themorpho-thematicbasis of the contrast between patterns A and B. As () shows, the alterna-tion between pattern A and the general pattern B is confined to the Aoristseries. Moreover, like the variation between patterns B and C, the alterna-tion between A and B correlates with the realization of logical subjects.

() Revised correspondences between series and classes

Transitive Unaccusative Unergative Indirect

Present B B B BAorist A B A BPerfect B B B B

Harris (, ) expresses the correlation between argument structureand case marking in the Aorist series associating ergative case with gram-matical subjects that correspond to initial (or, in present terms, logical)subjects. Transitive and unergative verbs both select logical subjects thatsurface as grammatical subjects, so these subjects aremarked ergative. Un-accusative verbs lack logical subjects altogether, and the logical subjects ofindirect verbs are realized as indirect objects. So the grammatical subjectsof unaccusative and indirect verbs occur in the nominative, in accordancewith the default case marking conventions in Georgian.

... SummaryThe preceding discussion identifies the motivation for a traditional anal-ysis that treats the dative nominal in an inversion construction as an in-direct object. On this account, there is one set of general case and agree-ment conventions, correlated with surface grammatical functions, and twotypes of thematically-conditioned deviations. There is, in particular, oneset of basic agreement conventions and a default pattern of case marking.The resulting description simplifies individual morphosyntactic patternsin Georgian while bringing out the organization of the whole system.

The main prerequisite of this type of analysis is a distinction betweenlogical/initial and grammatical/final levels of lexical argument structure.Inversion can then be regarded as a thematically-conditioned deviationfrom default patterns of argument realization. The alternation between di-rect and inverted patterns can be characterized entirely with respect to theargument structure of individual predicates. Hence, there is nothing in-

The fact that a lexical analysis of inversion does not make any essential reference to‘features of syntactic arrangement’ also allays concerns about whether Tschenkeli’s ()examples accurately reflect the general word order conventions of Georgian. As noted byone reviewer, Vogt (, f.) identifies the dominant word order as verb final.

tig.tex; 7/05/2007; 16:18; p.17

Page 18: Thematic Inversion in Georgian (Blevins)

James P Blevins

trinsically syntactic about the process of inversion, and no sense inwhich itdepends on any notion of “syntactic restructuring” (Anderson, , ).

.. T L M T

For the reasons outlined in Section .., Lexical Mapping Theory (LMT)provides a useful point of departure for a correspondence-based analy-sis of inversion. Logical argument structure can be described in terms ofProto-Agent and Proto-Patient roles, and grammatical argument structurein terms of the functions , , and . The ordering of proto-roles can be taken to reflect the number of Proto-Agent properties (Dowty,; Zaenen, ) or the most “heavily weighted” properties (Acker-man, , ) or even the distribution of distinguished properties. Onany of these alternatives, the Proto-Agent will occur initially, and a Proto-Patient that corresponds to the recipient will precede a Proto-Patient thatcorresponds to a theme. Argument realization then involves the mappingfrom proto-roles to grammatical functions. In standard LMTmodels, map-pings are guided by the assignment of intrinsic and default features toroles. For example, the features assigned in () below reflect the intrinsicrole classifications proposed in Bresnan and Kanerva (, ) and Bres-nan and Zaenen (, ), apart from [+r] in (), which conforms to thedefault in Bresnan and Kanerva (, ). For themost part, these featureassignments will simply be adopted below. Proto-Patients that correspondto themes are appropriately marked as unrestricted ([−r]). Proto-Patientroles that correspond to recipients must be marked as restricted ([+r]) toensure that they are realized as functions in direct constructions andthat they cannot be realized as direct objects in inversion constructions.Marking these roles intrinsically as [+r], rather than as [+o], also allowsthem to alternate with obliques, as illustrated in (b) below.

The principles in () establish a default mapping between roles andfunctions. The notation ‘θ̂’ in (a) “designates themost prominent seman-tic role of a predicate” (Bresnan, , ). The markedness of grammat-ical functions in the hierarchy in (b) reflects the number of ‘unmarked’negative values for [±r] and [±o]. The is the least marked function,because it has only negative values, is the most marked because ithas only positive values, and and have an equivalent intermediatestatus because they each contain one negative and one positive value.

() Mapping Principles (Bresnan and Zaenen, , )a. Subject roles:

i. θ̂[−o ] is mapped onto ; otherwise:

ii. θ[−r ] is mapped onto .

tig.tex; 7/05/2007; 16:18; p.18

Page 19: Thematic Inversion in Georgian (Blevins)

Thematic inversion in Georgian

b. Other roles are mapped onto the lowest compatible functionin the markedness hierarchy: ≻ , ≻ .

The features andmapping principles determine the association betweenroles and functions in (a) and (b), corresponding to (a) and (a). Inboth a-structures, the highest role is a Proto-Agent, which is mapped ontothe function by (a). The Proto-Patient roles are mapped onto thelowest compatible function, in accordance with (b). The [–r] roles aremapped onto functions, while the [+r] role in (b) is mapped ontothe function.

() Default argument realization in transitive a-structuresa. daxat’avs ⟨ Ag Pat ⟩

[–o] [–r]

b. misc’ers ⟨ Ag Pat Pat ⟩[–o] [+r] [–r]

In general, the role-functionmappings in () preserve thematic promi-nence. The one systematic exception involves logical indirect objects, whichare relatively prominent thematically but realized as highly marked syn-tactic functions. This mismatch captures the intuition underlying the no-tion of a ‘– retreat’ in RG (Perlmutter and Postal, , ) while lo-cating the alternation in the mapping from logical to grammatical argu-ments. More generally, the role-function associations in () highlightthe close correspondence between a-structures and bistratal relational net-works. Like relational networks, a-structures represent a relatively purerepresentation of valence, with no information about syntactic propertiessuch as word order, morphological properties such as verb form, or evenmorphosyntactic properties such as case and agreement. The correspon-dence between a-structures and bistratal networks clarifies why it is rela-tively straightforward to recast bistratal RG analyses in LMT terms.

The use of features to guide argument realization is among the mosttheory-internal aspects of standard LMT accounts. Other approaches, suchas Alsina () and Ackerman andMoore (), dispense with featuresin favour of more explicit argument realization rules. Even within standardLMT accounts, mediating features mainly constrain the mappings that canbe established by the default conventions in (). The features associatedwith a given role or proto-role cannot be interpreted as restricting its pos-sible realizations in general, given that nearly all of the standard featureassignments are violated by familiar alternations. Proto-Agents cannot beregarded as intrinsically non-objective, given that they may be realized asdirect or indirect objects in causative constructions (Alsina, ; Acker-man and Moore, ). Proto-Patient roles interpreted as themes are notintrinsically unrestricted, given that they may be realized as thematicallyrestricted objects or obliques in many antipassive constructions (Davies,

tig.tex; 7/05/2007; 16:18; p.19

Page 20: Thematic Inversion in Georgian (Blevins)

James P Blevins

). Demotions in general will tend to associate roleswith functions thatare not compatible with the features assigned to mediate default argumentrealization in LMT.

... InversionGeorgian inversion is a canonical demotion, as it associates themost promi-nent logical argument, the Proto-Agent, with the lowest ranking grammat-ical function, the . Thismapping is established directly by the inversionrule in (), repeated below in (). For themost part, a rule that targets aninitial [–o] role will have the same eect because the same properties thatdistinguish a Proto-Agent from a Proto-Patient also determine whetherthe initial role in an a-structure is lexically marked [–o] or [–r].

() Inversion RuleAg|

The eects of the inversion rule are illustrated in the inverted a-structuresin (), which correspond to the examples in (b) and (b). This rule asso-ciates the Proto-Agent roles in () with indirect/secondary object func-tions, thereby preempting the first subject mapping in (a). The secondprinciple in (a) links the [–r] Proto-Patient onto the function, in-ducing the LMT version of Unaccusative Advancement.

() Inverted argument realization in transitive and ditransitive a-structuresa. dauxat’avs ⟨ Ag Pat ⟩

[–o] [–r]

b. miuc’ers ⟨ Ag Pat Pat ⟩[–o] [+r] [–r]

The demotion of the logical indirect object in (b) follows from the factthat inversion maps the Proto-Agent onto the function, together withthe uniqueness principle in (), which ensures that there can be at mostone . However, the of the demoted Proto-Patient depends onthe classification of postpositional dependents and other types of tradi-tional obliques. The most straightforward account – pending a more gen-eral treatment of chômeurs and retired terms in LMT –maps the demotedProto-Patient onto an function, as in (b).

() Function-Argument Biuniqueness (Bresnan, , )Each a-structure role must be associated with a unique function,and conversely.

The contrast between the Present a-structures in () and the Perfectcounterparts in () shows how inversion disrupts the default realization

tig.tex; 7/05/2007; 16:18; p.20

Page 21: Thematic Inversion in Georgian (Blevins)

Thematic inversion in Georgian

of each of the logical arguments. The demotion of the logical subject toan indirect object induces the advancement of the logical object to gram-matical subject and forces the logical indirect object to be realized as anoptional postpositional oblique. The a-structures in (), corresponding tothe Present and Perfect unergatives in (), exhibit a parallel pattern. Theclassification of the Proto-Patient as a restricted role again ensures that itis realized as an indirect object, not as an object in the direct a-structure in(a), and as an oblique, not as a subject, in the a-structure in (b).() Direct and inverted argument realization in unergative a-structures

a. elap’arakeba ⟨ Ag Pat ⟩[–o] [+r]

b. ulap’arakia ⟨ Ag Pat ⟩[–o] [+r]

The inverted a-structures in () and (b) explicate the detransitiviz-ing eect of inversion. Following the demotion of the logical subject, thelogical object must advance, to be realized as the grammatical subject. Butthe logical indirect object can neither advance, to be realized as the gram-matical object, nor surface as an indirect object. Hence an inverted verbgoverns one fewer grammatical functions than a direct counterpart. In unerga-tive verbs, inversion leads to a surface a-structure with an indirect objectand an optional postpositional dependent. However, the a-structures ofPerfect unergative verbs contain no grammatical subject, in clear violationof the Final Law of RG and the Subject Condition of LFG.

Indirect verbs exhibit a similar pattern of inverted argument realization,but in all series, suggesting that an inverted argument structure is lexicallyassociatedwith this class of predicates. As the Present form in (a) shows,logically transitive indirect verbs pattern with Perfect forms of transitiveverbs. The Present form in (b) likewise shows that logically intransitiveindirect verbs pattern with Perfect forms of unergatives.() Inversion in the Present series of indirect verbs (Harris, , /)

a. VanosVano.

sdzuls.hate.

direkt’ori.director.

‘Vano hates the director.’b. Bavšvs

child.uxaria..be-happy..

‘The child is happy.’In the a-structure of sduls ‘hates’ in (a), the logical subject is realized as asurface indirect object and the logical object as a grammatical subject, justas in the a-structure for dauxat’avs in (). The a-structure of intransitiveuxaria ‘is happy’ in (b) contains just a surface indirect object and, likethe a-structures of Perfect unergatives, lacks a grammatical subject.

tig.tex; 7/05/2007; 16:18; p.21

Page 22: Thematic Inversion in Georgian (Blevins)

James P Blevins

() Argument realization in a-structures of Present indirect verbsa. sduls ⟨ Ag Pat ⟩

[–o] [–r]

b. uxaria ⟨ Ag ⟩[–r]

Indirect verbs bring out the central dierence between the use of dis-crete roles such as agent, patient, goal, etc., and the use of cluster conceptssuch as Proto-Agent and Proto-Patient. An inversion rule that targets agentroles would not normally be expected to apply to the logical subjects ofverbs like ‘be happy’ or ‘love’, which are not canonical agents. In contrast, arule that targets Proto-Agents would be expected to apply to indirect verbsbecause the logical subjects of indirect verbs will qualify as Proto-Agentson nearly all standard definitions. Although these roles do not exercisevolitional control over the event or state denoted by an indirect verb, theyare associated with other Proto-Agent properties, such as sentience andexistence independent of the event. Hence, the logical subjects of indirectverbs will be characterized as Proto-Agents, irrespective of whether Proto-Agents are defined as participant roles (i) with more Proto-Agent proper-ties than Proto-Patient properties (Dowty, ), (ii) with more “heavilyweighted” Proto-Agent than Proto-Patient properties (Ackerman, ),or even (iii) with a distinguished Proto-Agent property, such as sentience.

... UnaccusativityTreating inversion as a process that targets logical subjects oers an ex-planation for the fact that unaccusative verbs do not exhibit inversion.The ‘objectlike’ character of the arguments of unaccusatives can be ex-pressed in terms of proto-roles, discrete roles, or in terms of the [–r] fea-tures assigned to patient-like roles in LMT accounts. However, followingDowty (), the argument structure of unaccusatives are distinguishedbelow by the fact that they contain only Proto-Patient roles. Given thatthe inversion rule in () demotes Proto-Agents, it will fail to apply tounaccusatives, so that argument realization in Perfect unaccusatives willexhibit the same default patterns as in the corresponding Present con-structions. The a-structures in () first illustrate the pattern in ‘absolute’unaccusatives, exemplified by () above, which do not select an indirectobject. In the a-structure of Future ixat’eba ‘will paint’ in (a), the Proto-

An analysis based on discrete roles can express the same distinction by introducing a‘cognizer’ role between the agent and the next-highest role on the thematic hierarchy. Theinversion rule can then apply to all roles above this thematic threshold, for example, to allroles that aremore prominent than benefactive in the hierarchy in (). On both alternatives,the inversion rule targets roles that are associated with all Proto-Agent properties exceptvolitionality, though an analysis based on discrete roles expresses this generalization byreifying a partial bundle of proto-properties as a new participant role.

tig.tex; 7/05/2007; 16:18; p.22

Page 23: Thematic Inversion in Georgian (Blevins)

Thematic inversion in Georgian

Patient role is realized as the grammatical subject, due again to the LMTcounterpart of Unaccusative Advancement. Because inversion applies onlyto verbs that select a Proto-Agent, the Proto-Patient is realized as thesurface subject in the a-structure of Perfect daxat’uliq’o in (b).

() Argument realization in absolute unaccusativesa. ixat’eba ⟨ Pat ⟩

[–r]

b. daxat’uliq’o ⟨ Pat ⟩[–r]

The a-structures in () exhibit a similar pattern in ‘relative’ unaccusatives,exemplified by () above, which govern an indirect object. The a-structureof Present udzavrdeba ‘grows angry’ in (a) conforms to the mappingprinciples in (), which realize the most prominent Proto-Patient roleas the subject and realize the restricted Proto-Patient role as an indirectobject. Because inversion again does not apply, argument realization in thea-structure of Perfect gasdzavrebia in (b) preserves the default pattern.

() Argument realization in relative unaccusativesa. udzavrdeba ⟨ Pat Pat ⟩

[–r] [+r]

b. gasdzavrebia ⟨ Pat Pat ⟩[–r] [+r]

An inversion rule that targets logical subjects will also capture the al-ternation between indirect verbs and their passive counterparts. The a-structure of Present sduls ‘hates’ in (a) contains a Proto-Agent thatmeetsthe prerequisites of the inversion rule. But since the a-structure of the cor-responding passive šedzulebulia in (b) contains only a Proto-Patient, itcannot undergo inversion. Instead, like other unaccusative predicates, pas-sives of indirect verbs exhibit the default pattern of argument realization.

() Argument realization in a-structures of Present indirect verbsa. sduls ⟨ Ag Pat ⟩

[–o] [–r]

b. šedzulebulia ⟨ Pat ⟩[–r]

.. V

Isolating inversion in the mapping from logical to surface arguments pro-vides a stable grammatical locus for morphosyntactic properties. As dis-cussed in Section ., a thematic treatment of inversion highlights thesimple opposition between the valence patterns A and B in ().

tig.tex; 7/05/2007; 16:18; p.23

Page 24: Thematic Inversion in Georgian (Blevins)

James P Blevins

() Valence patterns (Tschenkéli, ; Harris, )

v-set m-set h-setA B

Themorphosyntactic properties that define the valence patterns in () de-rive in turn from general case and agreement principles, which are largelykeyed to surface grammatical functions. The case conventions are summa-rized in Section .. and the basic agreement properties in Section ...

... CaseThe case alternations in () reflect the interaction of two patterns. Bydefault, nominative case is associated with the highest unrestricted gram-matical function (subject or object) in an a-structure.However, this generalpattern is disrupted by more specific constraints on the case of subjectsin the Aorist series and objects in the Present series. The interactions aredescribed below by a general nominative constraint, (a), which is over-ridden by more function-specific constraints on the ergative and dative.The constraint in (b) expresses the insight that the case traditionallytermed ‘ergative’marks final subjects that correspond to an initial subject inthe Aorist series (Harris , ). This is precisely the type of bistratalrelation that can be expressed by the thematic and grammatical levels of anLMT a-structure. The constraint in (c) ensures that nominative is onlyassigned to the highest unrestricted function by associating dative withdirect objects that occur in an a-structure with a nominative subject. Theconstraint in (d) represents the fact that indirect objects always occur inthe dative, irrespective of the marking of other arguments.

() Case Constraintsa. Nominative is associated with an unrestricted function.b. Ergative is associated with a that realizes a Proto-Agent

in the Aorist series.c. Dative is associated with an in an a-structure with a nom-

inative .d. Dative is associated with an .

The case patterns described by these principles are summarized in ()and (), in which the unergative verb ‘to dance’ is represented by the sgPresent form cek’vavs and the ditransitive ‘to write’ by sg misc’ers.

() Case marking in unergative and transitive verbs

tig.tex; 7/05/2007; 16:18; p.24

Page 25: Thematic Inversion in Georgian (Blevins)

Thematic inversion in Georgian

DirectPresentAorist

InvertedPresent

a. cek’vavs ⟨ Ag ⟩[–o]

b. misc’ers ⟨ Ag Pat Pat ⟩[–o] [+r] [–r] —

The unergative pattern is illustrated in (a). In the Present series, thenominative subject conforms to the default in (a). In the Aorist series,the ergative subject falls under (b), which associates a Proto-Agentivesubject with ergative case. The dative indirect object in the Perfect seriesconforms to (d), which associates all indirect objects with the dative. Thecase of grammatical subjects in the ditransitive construction in (b) dic-tates the case of direct objects. If the subject is ergative, the object is nomi-native, in accordance with (a), whereas a nominative subject creates theconditions under which (b) requires a dative object.

The unaccusative a-structures in () preserve a default association be-tween case and grammatical functions in all series. The passive ‘be painted’,represented by the Future form ixat’eba in (a), uniformly governs thenominative because its sole thematic role is a Proto-Patient, which cannotbe realized by an ergative subject in the Aorist series, nor as a dative indi-rect object in the Perfect. The inchoative ‘grow angry’, represented by thePresent udzavrdeba in (b), similarly governs a nominative subject anddative indirect object in all series.

() Uniform case marking in unaccusative verbs

DirectPresentAoristPerfect

a. ixat’eba ⟨ Pat ⟩[–r]

b. udzavrdeba ⟨ Pat Pat ⟩[–r] [+r]

The association of ‘ergative’ case with subjects of intransitives indicates that it doesnot participate in any kind of standard ergative/absolutive opposition. Instead, as Har-ris (, ) argues at length, the ergative marking of subjects in the Aorist seriesidentifies the active members in a case opposition of the active/inactive type.

tig.tex; 7/05/2007; 16:18; p.25

Page 26: Thematic Inversion in Georgian (Blevins)

James P Blevins

... AgreementThere is an even more transparent relation between agreement propertiesand surface grammatical functions. As shown in Section .., v-set mark-ers encode subject properties,m-set markers encode object properties andh-set markers encode the properties of indirect objects. The correlationbetween agreement properties and grammatical functions can be clarifiedfurther ifm-setmarkers are associatedwith objective functions, as in (c).

() Realization of agreement propertiesa. v-set markers realize agreement properties of a .b. h-set markers realize agreement properties of an .c. m-set markers realize agreement properties of an [+o] func-

tion.

The agreement principles in () eliminate much of the redundancybetween them-set and h-setmarkers in () and bring outmore clearly thestructure of the agreement system. As the revised chart in () indicates,there is only one set of indirect object prefixes, h/s, and no distinctive directobject markers. The apparent homophony between the m-set and h-setmarkers in () reflects the fact that each of the homophonous ‘pairs’ infact realize the agreement properties of the most thematically prominentobjective function in an a-structure. The m-set markers realize the prop-erties of indirect object, if one is present in an a-structure, otherwise theproperties of a direct object, as Tschenkéli (, ) stresses when heremarks that “in verbs with objects only a single objective person markercan appear, which serves exclusively to express the indirect object”.

() Revised agreement marker sets

[–o]

sg v— m—pl v— t gv—sg g—pl — t g—tsg — s,a,o h/s—pl — an/en,es,nen h/s—

The description in () summarizes only the basic agreement patternsin Georgian, and does not distinguish the realization of person from num-ber features or address complications due to animacy (Harris, ; Tuite,) or ‘competition’ between markers (Anderson, ; Stump, ).These refinements aremostly orthogonal to the association between agree-ment and argument structure, though it is worth clarifying one source of

tig.tex; 7/05/2007; 16:18; p.26

Page 27: Thematic Inversion in Georgian (Blevins)

Thematic inversion in Georgian

systematic ambiguity. As indicated by the position of markers relative tothe stem ‘ —’ in (), there are two main agreement ‘slots’ in the Georgianverb: a pre-radical slot that predominantly marks object properties and apost-radical slot that predominantly marks subject agreement properties.The lack of a pre-radical marker identifies third person objects or indirectobjects of transitive verbs and correlates with the absence of an object withintransitives. The lack of a post-radical marker identifies verbs with sec-ond person subjects, irrespective of transitivity. Hence, subjectless verbs– such as the Perfect forms of unergatives – do not occur without a post-radical marker, but instead take a formally third person singular marker.Although this pattern is sometimes described in terms of agreement witha ‘dummy’ element (Anderson, , ), it reflects a relatively familiarcase of ambiguity, in which third singular forms alternate between encod-ing the properties of a third person agreement controller and encoding the of an agreement relation (Jakobson, ; Blevins, ).

... Object CamouflageThe observation that agreement is triggered by themost thematically promi-nent objective function also suggests an account of the phenomenon of‘Object Camouflage’ described in Harris (, §). In Georgian, a di-transitive verb cannot govern an indirect object together with a first orsecond person direct object. In order to express this combination, the firstor second person objectmust be represented as a third person object, whichcontains a first or second person pronoun and (the appropriateform of) the noun ‘self’. This phenomenon is illustrated in ().

() Object Camouflage in Georgian (Harris, , )a. *Deda

mother.(šen)you.

gabarebs.renders.

masc’avlebels.teacher.

(‘Mother is turning you over to the teacher.’)b. Deda

mother.abarebs.renders.

masc’avlebelsteacher.

šensyour

tavs.self.

‘Mother is turning you over to the teacher.’

Examples such as (a) above show that a ditransitive verb can governtwo third person objects. Example (a) suggests that a second persondirect object cannot be indexed on a ditransitive verb by the agreementmarker g-, irrespective of whether the object pronoun, here šen, is overtlyexpressed or not. Example (b) indicates that the intended meaning of(a) can be conveyed by ‘camouflaging’ the second person object as apossessive pronoun within the third person object šens tavs. Because thirdperson direct objects are not overtly indexed on a verb, the form adarebsis compatible with the two third person objects in (b).

tig.tex; 7/05/2007; 16:18; p.27

Page 28: Thematic Inversion in Georgian (Blevins)

James P Blevins

Camouflaging first and second person objects as third person formsprovides an intriguing solution to a problem created by the interactionof morphosyntactic principles and morphological structure in Georgian.The most general principle is that the agreement marking of a finite verbin Georgian must be compatible with any grammatical function directlygoverned by the verb. A more language-specific principle is that agree-ment with a more thematically prominent indirect object takes priorityover agreementwith a less syntacticallymarked direct object. The need toassign priority arises because ditransitive verbs in Georgian have more ar-guments than agreement ‘slots’. A camouflaging solution to this mismatchexploits the fact that third person direct objects are not indexed on the verb.

The interaction of Object Camouflage and inversion also lends a mea-sure of support to an analysis of the type outlined above. As Harris (,) notes, “Object Camouflage never applies with inversion predicates”irrespective of whether the inversion is triggered by Perfect forms or withindirect verbs. The pattern in inversion verbs is illustrated in ().

() Lack of Camouflage with Indirect Verbs (Harris, , )a. (me)

me.momc’ons.like.

pelamuši.pelamushi.

‘I like pelamushi (a food).’b. čems

mytavsself.

mosc’ons.like.

pelamuši.pelamushi.

(‘I like pelamushi’.)

The lack of Object Camouflage in () requires no special explanation.In (a), the first person indirect object me is indexed by the first personm-set markerm-, while the subject pelamuši is indexed by the third personv-setmarker -s. There is no direct object to index on the verb – and, indeed,there cannot be, given the detransitivizing eect of inversion. Since there isno mismatch between arguments and agreement slots, there is motivationfor camouflaging a complement of a ditransitive verb as a third person formin an inversion construction. Hence the camouflaged alternative in (b) isunacceptable. The fact that nominative arguments in inversion construc-tions do not undergo Object Camouflage in the presence of an agentiveindirect object also supports the traditional view that the nominatives arenot direct objects, but subjects, as their form and concord would suggest.

No special constraints are needed to regulate the interaction of inver-sion and Object Camouflage. Given the selective detransitivizing eect ofinversion, inversion constructions cannot contain more direct verbal argu-ments than agreement slots. As in earlier treatments of valence patterns or This priority is also reflected in the fact that (b) is acceptable if second person šen

is interpreted as the indirect object andmasc’avlebels as the direct object.

tig.tex; 7/05/2007; 16:18; p.28

Page 29: Thematic Inversion in Georgian (Blevins)

Thematic inversion in Georgian

case and agreement properties, a correspondence-based analysis capturesthe intuition underlying traditional analyses, while preserving a simpli-fied description of the morphological system, which constitutes one of theprimary benefits of a traditional account.

. Conclusions

The main conclusions of this paper have been elaborated in the preced-ing discussion. The central claim is that inversion arises in the mappingbetween thematic roles and grammatical relations, not in the associationbetween grammatical relations and case and agreement processes. Inver-sion, in short, is not a ‘quirky’ valence pattern, but a non-canonicalmappingof a prominent thematic role to a highly marked grammatical function.A correspondence-based model helps to clarify the main objection to aquirky valence analysis. There is no principled reason why Georgian couldnot have dative subjects in inversion constructions. But there is no mo-tivation for a dative subject analysis, given that any grammatical gener-alizations that refer to dative subjects can be expressed in terms of dativeProto-Agents, without themorphosyntactic disruption that dative subjectswould cause. From a correspondence-based perspective, dative subjectsare a symptom of ‘overloading’ the level surface grammatical analysis. Byforcing generalizations that apply to Proto-Agents onto surface subjects,a quirky valence account produces a more heterogeneous class of subjectsand complicatesmorphosyntactic patterns, with no compensating benefits.

At the same time, inversion highlights issues of relevance to LMT, toother correspondence-basedmodels of argument structure, and to approachesto valence in general. One set of issues concerns the treatment of grammat-ical relations. Any frameworkmust be able to distinguish indirect/secondaryobjects, however they choose to classify these elements. AsGeorgian shows,it must also be possible to ensure that at most one such element occurs inthe argument structure of a verb in at least some languages. In addition,inversion constructions based on intransitive verbs confirm that subject-legislating constraints have no place in general theories of argument orclause structure. A Perfect form such as ucek’via ‘danced’ in (b) governsjust a surface indirect object, in clear violation of the Final Law of RG(Perlmutter and Postal, , ), the Subject Condition of LFG (Bres-nan and Kanerva, , ) and counterparts in other approaches. Theexistence of subjectless inversion constructions merely reinforce the ev-idence presented by Babby () and Blevins (), among others, thatsubject-legislating constraints reflect what is at best a typological bias, andat worst a projection of the extreme subject-prominence of English.

tig.tex; 7/05/2007; 16:18; p.29

Page 30: Thematic Inversion in Georgian (Blevins)

James P Blevins

A second set of issues concern the status of logical subjects. The presentaccount followsDowty () in distinguishing just Proto-Agent andProto-Patient roles, rather than the larger inventories recognized by accountsthat follow in the Case Grammar tradition. However, for present purposes,the dierences between these alternatives are inessential. Any general-izations that apply to Proto-Agents (or to actor ‘macroroles’ in the senseof van Valin ()) can be recast as applying to roles above a certainthreshold in a thematic hierarchy. In either case, it is properties or en-tailments that distinguish the roles that are classified as logical subjectsfrom those that are not. However, the notion of ‘logical subject’ relevantfor inversion cannot be defined as “the most prominent semantic role ofa predicator” (Bresnan, , ). The unergative cek’vavs ‘dance’ in ()and the unaccusative ixat’eba ‘be painted’ in () are both intransitive verbs,associated with a single thematic role. This role is the only, and hencemostprominent, role in the respective a-structures. Yet inversion treats the roleassociated with unergative cek’vavs, but not the role associated with un-accusative ixat’eba, as a logical subject. The traditional notion of logicalsubject may reconstructed in terms of the content of roles or proto-roles,or in terms of features (such as [–o] or [–r]) that correlate with these prop-erties. But logical subjects cannot be defined in terms of relative positionin an a-structure.

As in the case of subject-legislating constraints, the Georgian evidenceconfirms a point that was already reasonably well established. Inversiondiscriminates between unaccusative and unergative verbs inmuch the sameway that passivization does inmost, if not all, languages. The RG literaturepresents an extended argument in support of the claim that ‘no impersonalPassive clause in any language can be based on an unaccusative predicate’(Perlmutter and Postal, , ). This pattern cannot be characterized interms of any notion of logical subject based purely on prominence withinan a-structure. Yet this is essentially the analysis assumed by LMT ac-counts that treat the passive as an “operation…which suppresses the high-est thematic role in the lexical argument structure” (Bresnan and Kanerva,, f.). An analysis of the passive that suppresses the highest the-matic role of a predicate allows the passivization of passives and othertypes of unaccusative predicates. The idea that this consequence is de-sirable rests on the belief that passives of unaccusatives are attested. Itseems fairly clear that this belief is unfounded, as putative examples of‘unaccusative passives’ can be shown to be impersonal or evidential con-structions (Blevins, ;Maling, ). Hence the fact that unaccusativeverbs do not undergo inversion in Georgian reinforces the case for distin-guishing the notions ‘logical subject’ and ‘highest thematic role’.

The present account also clarifies the status of grammatical featureswithin LMT. The fact that a logical subject may be realized as an indi-

tig.tex; 7/05/2007; 16:18; p.30

Page 31: Thematic Inversion in Georgian (Blevins)

Thematic inversion in Georgian

rect object shows that logical subjects cannot be characterized as intrin-sically ‘nonobjective’ functions, as suggested in a number of early LMTaccounts (Bresnan and Kanerva, ; Bresnan andMoshi, ; Bresnanand Zaenen, ). Instead, the [–o] features assigned to logical subjectsin these types of accounts serve to inhibit default mapping conventionsfrom realizing logical subjects as objects. In inversion constructions, as incausatives (Alsina, ; Ackerman and Moore, ) and other types ofgrammatico-thematic ‘demotions’, logical subjects are realized as direct orindirect objects. Hence the features assigned to a role in an LMT analysismay restrict the default realizations of that role but do not determine thespace of possible in which the role may participate.

References

Ackerman, F.: , ‘Complex Predicates and Morpholexical Relatedness: Locative Inver-sion inHungarian’. In: I. A. Sag andA. Szabolcsi (eds.): Lexical matters. Stanford: CSLI,pp. –.

Ackerman, F. and J. Moore: , ‘Syntagmatic and Paradigmatic Dimensions of CauseeEncodings’. Linguistics and Philosophy , –.

Ackerman, F. and J. Moore: , Proto-properties and Grammatical Encoding: A Corre-spondence Theory of Argument Selection. CSLI Publications.

Alsina, A.: , The Role of Argument Structure in Grammar: Evidence from Romance.Stanford: CSLI Publications.

Alsina, A.: , ‘A Theory of Complex Predicates: Evidence from Causatives in Bantu andRomance’. In: A. Alsina, P. Sells, and J. Bresnan (eds.): Complex Predicates. Stanford:CSLI Publications, pp. –.

Anderson, S. R.: , ‘On Representations in Morphology: Case Marking, Agreementand Inversion in Georgian’. Natural Language and Linguistic Theory , –.

Anderson, S. R.: ,A-MorphousMorphology. Cambridge: CambridgeUniversity Press.Aronson, H. I.: , Georgian: A Reading Grammar. Columbus, OH: Slavica.Babby, L. H.: , ‘Subjectlessness, External Subcategorization and the Projection Prin-

ciple’. Zbornik za Filologiju i Linguvistiku , –. Reprinted in Journal of SlavicLinguistics , –, .

Blevins, J. P.: , ‘Markedness and Agreement’. Transactions of the Philological Society(), –.

Blevins, J. P.: , ‘Passives and Impersonals’. Journal of Linguistics , –.Bloomfield, L.: , Language. Chicago: University of Chicago Press.Bresnan, J.: , The Mental Representation of Grammatical Relations. MIT Press.Bresnan, J.: , Lexical-Functional Syntax. Oxford: Blackwell.Bresnan, J. and J. M. Kanerva: , ‘Locative Inversion in Chicheŵa: A Case Study in

Factorization in Grammar’. Linguistic Inquiry , –.Bresnan, J. and L. Moshi: , ‘Object Asymmetries in Comparative Bantu Syntax’.

Linguistic Inquiry , –.Bresnan, J. and A. Zaenen: , ‘Deep Unaccusativity in LFG’. In: K. Dziwirek, P. Farrell,

and E. Mejías-Bikandi (eds.): Grammatical Relations: A Cross-Theoretical Perspective.Stanford: CSLI Publications, pp. –.

Dalrymple, M.: , Lexical Functional Grammar, Vol. of Syntax and Semantics.Academic Press.

tig.tex; 7/05/2007; 16:18; p.31

Page 32: Thematic Inversion in Georgian (Blevins)

James P Blevins

Davies, W. D.: , ‘Antipassive: Choctaw Evidence for a Universal Characterization’.In: D. M. Perlmutter and C. Rosen (eds.): Studies in Relational Grammar . Chicago:University of Chicago Press, pp. –.

Dowty, D.: , ‘Thematic Proto-Roles and Argument Selection’. Language , –.Fillmore, C.: , ‘The Case for Case’. In: E. Bach and R. T. Harms (eds.): Universals in

Linguistic Theory. New York: Holt, Rinehart and Winston, pp. –.Harris, A. C.: , Georgian Syntax: A Study in Relational Grammar. Cambridge:

Cambridge University Press.Harris, A. C.: , ‘Inversion as a Rule of Universal Grammar: Georgian Evidence’. In:

D. M. Perlmutter and C. Rosen (eds.): Studies in Relational Grammar . Chicago:University of Chicago Press, pp. –.

Harris, A. C.: ,Diachronic Syntax: The Karvelian Case, Syntax and Semantics . NewYork: Academic Press.

Hewitt, B. G.: , Georgian: A Structural Reference Grammar, London Oriental andAfrican Languages Library. Amsterdam: John Benjamins.

Jackendo, R.: , Semantic Interpretation in Generative Grammar. Cambridge: MITPress.

Jakobson, R.: , ‘Zur Structur des russischen Verbums’. Charisteria Guilelmo Math-esio Quinquagenario pp. –. Reprinted in E. P. Hamp, F. W. Householder andR. Austerlitz (eds.): Readings in Linguistics II. Chicago, University of Chicago Press,–.

Joppen-Hellwig, S.: , Verbklassen und Argumentlinking. Tuübingen: Niemeyer.Kaplan, R. M. and J. Bresnan: , ‘Lexical-functional Grammar: A formal system

for grammatical representation’. In: J. Bresnan (ed.): The Mental Representation ofGrammatical Relations. Cambridge: MIT Press, pp. –.

Maling, J.: , ‘Unpassives of unaccusatives’. Ms, Brandeis University.Perlmutter, D. M.: , ‘Impersonal Passives and the Unaccusative Hypothesis’. In:

Proceedings of the Fourth Annual Meeting of the Berkeley Linguistics Society.Perlmutter, D. M. (ed.): , Studies in Relational Grammar . University of Chicago

Press.Perlmutter, D. M. and P. M. Postal: , ‘Toward a Universal Characterization of Pas-

sivization’. In: Proceedings of the Third Annual Meeting of the Berkeley LinguisticsSociety. Reprinted in Perlmutter (),–.

Perlmutter, D.M. and P.M. Postal: , ‘Some proposed laws of basic clause structure’. In:D. M. Perlmutter (ed.): Studies in Relational Grammar . University of Chicago Press,pp. –.

Perlmutter, D. M. and P. M. Postal: , ‘The -Advancement Exclusiveness Law’. In:D. M. Perlmutter and C. G. Rosen (eds.): Studies in Relational Grammar . Chicago:University of Chicago Press, pp. –.

Stump, G. T.: , Inflectional Morphology: A Theory of Paradigm Structure. Cambridge:Cambridge University Press.

Tschenkéli, K.: , Einführung in die georgische Sprache. Zurich: Amirani Verlag.Tuite, K.: , Kartvelian Morphosyntax, Vol. of Lincom Studies in Causasian Linguis-

tics. Munich: Lincom Europa.vanValin, R.D.: , ‘Semantic Parameters of Split Intransitivity’. Language, –.Vogt, H.: , Grammaire de la langue georgienne. Oslo: Universitetsforlaget.Williams, E.: , ‘Predication’. Linguistic Inquiry , –.Wunderlich, D.: , ‘A Minimalist Model of Inflectional Morphology’. In: C. Wilder,

H.-M. Gärtner, and M. Bierwisch (eds.): The Role of Economy Principles in LinguisticTheory. Berlin: Akademie Verlag, pp. –.

tig.tex; 7/05/2007; 16:18; p.32

Page 33: Thematic Inversion in Georgian (Blevins)

Thematic inversion in Georgian

Zaenen, A.: , ‘Unaccusativity in Dutch: An Integrated Approach’. In: J. Puste-jovsky (ed.): Semantics and the Lexicon. Dordrecht: Kluwer Academic Publishers, pp.–.

tig.tex; 7/05/2007; 16:18; p.33

Page 34: Thematic Inversion in Georgian (Blevins)

tig.tex; 7/05/2007; 16:18; p.34