* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Download On Representations in Morphology Case, Agreement and Inversion
Ojibwe grammar wikipedia , lookup
Lithuanian grammar wikipedia , lookup
Old Norse morphology wikipedia , lookup
Japanese grammar wikipedia , lookup
French grammar wikipedia , lookup
Zulu grammar wikipedia , lookup
Chinese grammar wikipedia , lookup
Ukrainian grammar wikipedia , lookup
Macedonian grammar wikipedia , lookup
Udmurt grammar wikipedia , lookup
Navajo grammar wikipedia , lookup
Kannada grammar wikipedia , lookup
English clause syntax wikipedia , lookup
Scottish Gaelic grammar wikipedia , lookup
Germanic weak verb wikipedia , lookup
Polish grammar wikipedia , lookup
Old Irish grammar wikipedia , lookup
Portuguese grammar wikipedia , lookup
Modern Hebrew grammar wikipedia , lookup
Germanic strong verb wikipedia , lookup
Swedish grammar wikipedia , lookup
Ancient Greek grammar wikipedia , lookup
Turkish grammar wikipedia , lookup
Sotho verbs wikipedia , lookup
Italian grammar wikipedia , lookup
Russian grammar wikipedia , lookup
Kagoshima verb conjugations wikipedia , lookup
Old English grammar wikipedia , lookup
Latin syntax wikipedia , lookup
Yiddish grammar wikipedia , lookup
Hungarian verbs wikipedia , lookup
Spanish grammar wikipedia , lookup
Icelandic grammar wikipedia , lookup
Lexical semantics wikipedia , lookup
Serbo-Croatian grammar wikipedia , lookup
On Representations in Morphology Case, Agreement and Inversion in Georgian Author(s): Stephen R. Anderson Source: Natural Language & Linguistic Theory, Vol. 2, No. 2 (Jul., 1984), pp. 157-218 Published by: Springer Stable URL: http://www.jstor.org/stable/4047487 Accessed: 26/10/2009 15:57 Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use. Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at http://www.jstor.org/action/showPublisher?publisherCode=springer. Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission. JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected]. Springer is collaborating with JSTOR to digitize, preserve and extend access to Natural Language & Linguistic Theory. http://www.jstor.org STEPHEN R. ANDERSON ON REPRESENTATIONS CASE, AGREEMENT IN MORPHOLOGY AND INVERSION IN GEORGIAN* 0. INTRODUCTION As a fundamentalbuildingblock of linguisticstructure,the morphemeis supposedto be the locus of the phonologicalexpressionof meaningand of grammatical categories. Traditionally, morpheme-basedtheories construedthese categories as being in a more or less one-to-one relationwith discrete substringsof phonologicalstructureor FORMATIVES. If this were an adequateview, it would reduce the 'morphologicalrepresentation'of a word to the sequence of formativescomposingit, where each formativeis (uniquely)associated with some semantic materialor grammaticalcategories as its content. Especiallyin the treatmentof inflection,though, the range of classicalpuzzlesconcerningthe natureof 'morphemes'refusesto dissolve; and closer analysissuggests a ratherdifferent picture from the usual 'beads on a string' view of morphologicalstructure. The observationsof Aronoff (1976) concerning 'morphemes'with no isolable meaning already recall the literatureof the 1940's and 1950's (e.g., Hockett 1947 and the controversiesprovoked by this paper). The currentdevelopment of morphologicaltheories not based on morphemes can be traced to Matthews'(1972) discussion of these traditionalproblems, and his proposalsfor a WORD AND PARADIGM view of inflectional structure.Such a view substitutesan inventory of rules modifyingword * This paper represents a revision of material presented under the same title to the Workshopon LexicalPhonologyand Morphologyat StanfordUniversity,1 March,1983; to the GLOW Colloquium at York, England on 30 March, 1983, and to the Caucasian Colloquiumat the Universityof Hullin July,1983. I have benefitedfromcommentsprovided on these and relatedissues by MichaelHammondand Alan Timberlake.Specialthanksare due to Alice Harris and George Hewitt, as well as Frank Heny and several anonymous refereesfor this journal,for extensive and detailed commentson an earlierversion of this paper. Informationon Georgiancomes from the sources listed in the referencesbelow, as well as fromworkwithTamara'Japaridze at UCLA andclasseswithYolandaMarchevat the Universityof Zurich. Naturally,none of the above should be held in agreementwith my views, or responsiblefor my errorsor failure to take their advice. The present paper was preparedwhile the authorwas supportedby a post-Doctoralfellowshipfrom the American Council of LearnedSocieties; supportfrom the Committeeon Researchat UCLA is also acknowledged. Natural Language and Linguistic Theory 2 (1984) 157-218. 0167-806X/84/0022-0157 ? 1984 by D. Reidel PublishingCompany $06.20 158 STEPHEN R. ANDERSON structure(by affixation,internalchange, etc.) for the usual inventory of morphemes.These rules depend on a representationof the inflectional structureof formswhich is not directlyderivablefrom the constituentsof phonologicalform, however, and whose natureremainsto be elucidated. These questions of representation arise particularly clearly in the inflectionaltheory of Anderson (1977, 1982, 1984a), where a notion of MORPHOSYNTACTIC REPRESENTATION forms the interface between syntactic and phonological structure. This kind of representationcan be expected to have its own properties which are worthy of closer investigation. The representationsin question can be traced to the proposals of Chomsky (1965), who presents the terminal elements of P-Markersas complex symbols or internally unstructuredcomplexes of features. In addition to features of subcategorization(and selection), some of these features represent inflectional properties (e.g., case, agreement, tense, etc.), and it is assumed that rules of grammarwhich are not part of the syntaxper se establisha correspondencebetween these complex symbols and actual sequences of phonologicallyrealizedformatives.Recent work in morphologydevelops this picture further,showing that these symbols have a significantinternalstructurewhich forms the basis of a system of morphologicaloperationsmappingthem onto phonological form. These operations are the morphologicalrules which establish correspondences between morphosyntacticand phonologicalform, and which thus replace (at least in the domain of inflection) a list or lexicon of 'meaningful' grammaticalmorphemes. This paper explores the propertiesof such morphosyntacticrepresentations,the rules that create and modifythem, and the role they play in the grammar,on the basis of an analysisof case marking and agreement in Georgian.The interestof Georgianfor such purposeslies in the fact that the inflectionalmorphologyof this languageis substantiallymore complex than that of more familiarlanguages,and thus providesa better test of the expressive power of a proposed morphological theory. Nonetheless, despite the inherentinterestof the Georgianfacts, the goal of the paperis not simplyto provide an analysisof Georgian,but to clarify the natureof morphosyntacticrepresentations. The structureof the paperis as follows. In section 1, we sketch the basic propertiesof verbalagreementand case markingin Georgian,limitingthe discussionto sentences whose verbs appearin one of the two fundamental tense/aspectseries. We then review the evidence for associatingparticular grammaticalpositions with particularagreement and case markingmor- CASE, AGREEMENT AND INVERSION IN GEORGIAN 159 phology, establishingthe role of the notions of subject, direct object and indirect object in Georgian. Section 2 discusses the INVERSION constructionwhichis found in the thirdtense/aspectseries, and also in the first and second series for a specific set of verbs; the relation of this constructionto those discussedin section 1 is treated.An analysisof inversion proposedby Harris(1981, 1982, 1983) is presented;the explanatoryscope of this analysis is explored, and arguments are presented against the syntactic relation-changingprocess on which it depends. An alternative, purelymorphologicalaccount suggested (butnot arguedfor) by Anderson (1982) is then introduced. The discussion in sections 1 and 2 recapitulatesfacts about Georgian from the literature,as well as Harris'analysis.This duplicationof material available elsewhere seems necessary for the benefit of the majority of readers, who will not be familiar with either the facts or their prior analysis;indulgenceis begged of those readerswho are betterprepared.It is particularlyimportantto stressfrom the outset the crucialrelianceof the present paper on Harris' work: her insights into the structure of the Georgian verbal system are relied on heavily here, and the analysis eventually arrived at bears close similaritiesto hers (though with some essentialdifferences).Indeed, the presentpapercan be regardedboth as a critical response to Harris' work and as an applicationof the theory of inflectionexplored here. Section 3 then goes on to explore the morphology of Georgian agreement in greater detail, on the basis of a discussion of the morphosyntacticrepresentationsthat should be assigned to the various verb classes in the language. It is concludedthatwhen these representationsare adequately explicated, no independent rule of Inversion (even a morphological one) is required,though a morphologicalrestructuringoperation analogousto the relationalrule of 'Unaccusative'plays a role in the account. A comprehensiveset of agreement and case markingrules are proposed which explicitly accommodate a wide range of apparently complex phenomena in a simple and natural way. One result of this analysis is the elimination of arbitrarylexical markers for the various conjugationclasses in the language. Section 4 then provides conclusions and a summaryof results. 1. BASIC MORPHOLOGICAL AND SYNTACTIC CATEGORIES IN GEORGIAN Georgian is well-known for the complexity of its inflectional apparatus, 160 STEPHEN R. ANDERSON and especiallyfor the dependenceof the case-markingsystemon the tense of the verb. There are roughly a dozen tense-aspect forms (called 'screeves'in the recent literature)in whichGeorgianverbs can appear;the exact inventorydependson whetherthe Imperativeis treatedas a separate form, whether the partiallyobsolete conjunctive perfect is included, and whetherone includesthe rare imperfectiveaoristform as a distinct tense. These tense-aspectformscan be divided into threebasic series: Series I or the 'Present(-future)'series, Series II or the 'Aorist'series, and Series III or the 'Perfect' series. Of these, Series I and II are conveniently treated together both because they sharecertainpropertiesand because Series III is demonstrablysecondary to them in every conceivable way: in morphological structure, in semantics and in usage, and also historically. Whether there is also a syntactic difference between Series III and the others will be a matterof considerableinterest below. 1.1 Agreementand case markingin SeriesI and II For the reasons just discussed, we postpone the discussion of Series III until section 2 below, and begin by summarizing the properties of agreementand case markingin the more basic series. 1. 1.1 Agreement Series I and II show the same morphologicalpatternsof agreementwithin the verb. Subjects, (direct) objects, and indirect objects can condition agreementin their clause; however, there are only two positionsin which this agreement material is found. Agreement is marked either 1) as a prefix, preceding the verb root (and any pre-radicalor 'version' vowel), but following a verbal prefix (the latter usually have perfective significance,and are separatedfrom the rest of the verb in citations with the boundary'='); or 2) as a suffix,following the rest of the verb stem. Where more than one agreementprefixmight appearto be motivated,or more than one such suffix,only one of each appearsovertly (see Anderson 1982, 1984 as well as table I below for some of the details of the disjunctiverelationshipsinvolved here). In sketching the agreementsystem at this point, before exemplifyingit, we run the risk of incomprehensibility.The details of this discussion, however, are of limited importance:it is only the overall categories of agreement markersthat need be attended to for what follows. Nonetheless, the details (which are summarized below in table I) must be introducedat some point. CASE, 161 IN GEORGIAN AND INVERSION AGREEMENT TABLE I Basic verbal agreement markers Series: Person/Number lsg 2sg 3sg lpl 2pl 3pl a b (Subject) v (D.O.) m v. 0 -s,-a v-... -t -t -enb Ing0 gvg- 0 (Indirect Object) e u h migi- mg- -t h_a u- gvg- gvigi '-t u- ha -t megeegvege- e- a -t mngaagvaga- *-t a- h, s, x or 0, depending on the following segment. -en, -an, -es, -on, or -nen depending on tense and verb class. Since Georgian is at least prima facie an ergative language (or possibly more accurately, as Harris 1981 argues at length, an 'active' language: see section 1.1.2.2 below), one might raise the question of which NP in a clause is properly called its subject. The syntactic evidence on this point is quite clear, however (cf. section 1.2 below), and confirms a decision to call 'subject' that NP which usually corresponds to the subject in a translation into English or other languages with familiar structure. With this understanding, we can then claim that all and only subjects (of either transitive or intransitive verbs) are marked on the verb with markers from what we can call (with Sanije 1980) the v-series. These appear in the first column of table I: v- marks first person and 0- second person; the suffix -s or -a (depending on the verb's conjugation class) marks third singular, and the third plural is indicated by one of the suffixes -en, -an or -nen (again depending on conjugation class). Examples will appear below in later sections. The direct object is marked on the verb with markers from what Sanije (1980) calls the m-series. These are the prefixes of the second column of table I: m- for first person singular, gv- for first person plural, or g- second person, and 0- for third person. Indirect objects (which may appear with either transitive or intransitive verbs) are marked on the verb in one of four distinct, but related ways. Following the terminology of Aronson (1982a, 1982b), we can refer to these markers as the h-series, the u-series, the e-series and the a-series. Each of these (shown in the third through sixth columns of table I) can be regarded as made up of a marker from the (direct object marking) m-series, followed by 0 (for the h-series) or one of the pre-radical vowels i, e or a (for the u, e, and a series, respectivelvV 162 STEPHEN R. ANDERSON The indirect object markers differ from simple sequences of direct object markerplus preradicalvowel in the thirdperson. The thirdperson h-seriesmarkeris s, h, x or 0 (dependingon the following consonant),and the thirdpersonu-seriesmarkerconsistsin replacingthe pre-radicalvowel i otherwisecharacteristicof this series by u. Evidence that indirectobject markingis distinctfrom direct object marking(plus a pre-radicalvowel), then, consistsof two facts: first,the thirdpersonh-seriesmarkerh-, s- or xhas no parallel in direct object marking; and second, the marker uuniquely marks indirect objects, since a simple combination of third person directobject markingwith preradicalvowel i (markingsome other function,such as the futurestem of medial verbs, subjective version, etc.) yields simplyi ratherthan u. Whichof the four series will markindirectobjects in a particularverb is a lexical property of the verb stem (cf. Aronson 1982b for arguments againstthe traditionalclaim that this differenceis semanticallybased).We represent this propertyin lexical entries by including the (only partially predictable)preradicalvowel in the entry for each verb; since the lexical entriesfor verbs with and withoutindirectobjects must be kept distinctin any event, there is no loss of generalityentailed in doing this. One final element in the person markingsystem is the pluralizingsuffix -t, which appears on a verb (replacing a final third person -s from the v-series, if the form would otherwisehave one) to markpluralityin either first or second person subjects or second person (direct or indirect) objects. The readerwhose head is still above water in this mass of details will note that these are precisely the cases in which a non-third person actant'spluralityis not explicitlyindicatedby the form of its corresponding marker(i.e., firstpluralsubject or second pluralsubject or object). At most one -t appears in a form, even if two would be motivated (in the event of a firstperson pluralsubject combined with second person plural object, for example). Examplesentences and forms cited below will be providedwith glosses in which these elements are identified.The analysisof agreementmarking is largely uncontroversial(though we have simplifiedsome points here, especially with regard to the markingof third person subjects), and the argumentsbelow should be comprehensiblewithoutcommittingthe facts just reviewed to memory. There is one point, however, which requiressome furtherdiscussion.It will be recalled that the Georgian verb provides room for only one prefix and one suffix:what, one may ask, of the case in which three distinctNPs CASE, AGREEMENT AND INVERSION IN GEORGIAN 163 must be overtly agreed with? The problem arises specificallyunder two sets of circumstances:1) where a firstperson subject (markedby v-) and a non-nullobject prefixshouldco-occur; and 2) those cases in which both a direct and an indirect object ought to be markedby prefixes.In the first case, the markerv- simplyfails to appear;this is effected by making the rule which would introduce it disjunctive with respect to the other agreement prefixation rules, as detailed elsewhere (Anderson 1982). Under these circumstancesthe fact that the subject is first person can often be deduced from other aspects of the verb's marking.In the case of co-occurringdirect and indirectobjects, a problemis only presentedwhen the direct object is other than third person, since third person direct objects are markedwith no overt affix. When a non-third person direct object is combined with an indirect object (of any person),the normalagreementprocess is suspendedand an alternativeconstruction(called "OBJECT CAMOUFLAGE" by Harris 1981) is used. This involves replacing the first or second person direct object pronoun by a form consisting of a possessive plus tavi 'head'. These tavi-phrases are normally reflexive in meaning, but under the special circumstances of object camouflage, their anaphoric interpretationis suspended and they serve as simple pronouns. Their relevance to this construction lies in the fact that (although their reference is to first or second person,)they are grammaticallythirdpersonNP, and thus call for 0 direct object agreement; they thereby allow the language to evade the limitationsof the formalapparatusat the disposalof its agreementsystem. Furtherdiscussionof the object camouflageconstructionwill be found in section 3.5.1. 1.1.2 Case marking The agreementsystemsummarizedin table I appliesequallyto both Series I and Series II forms. The case marking of Noun Phrases, however, depends on a number of mutuallyindependent parameters:a) the tense Seriesof the verb of the clause; b) the lexicalclass of this verb; and c) the grammaticalrelation which the NP bears. The overall system of case markingis summarizedbelow in table II (at the end of section 2.1). We discussthe facts of case markingin non-invertedclauseswhose verb bears a tense from Series I or II immediately below; the case marking and agreement properties of the inversion construction which appears with Series III tenses and with certain verbs (those of lexical class IV) in other series as well will be treated in section 2. 164 STEPHEN R. ANDERSON 1.1.2.1 SeriesI. The tenses which make up Series I are the present, the imperfect, the future, the conditional, and two subjunctives. In the majorityof cases, the futurediffersfrom the present,the conditionalfrom the imperfect, and one subjunctive from the other by the addition of a lexically idiosyncraticperfectivizing pre-verb which we write in parentheses when it has this function, followed by the '=' boundary. Subjects of all verbs whose tense is from Series I appear in the NOMINATIVE case, which is markedon a noun either by no suffixor by a final-i (if the stem ends in a consonant).Transitiveverbs in Series I take a directobject whichappearsin the ACCUSATIVE case; this is markedon the noun by the suffix-s. Indirectobjects (regardlessof the transitivityof the verb) appear in the DATIVE case; this is formally identical with the accusative, and is also marked by -s. Traditional grammarstreat the dative/accusativeas a single case on the basis of this formal identity, but for expository convenience we treat this syncretismas homophony between two distinctcategories. The analysisbelow does not depend in any essentialway on a difference between 'dative' and 'accusative',and could easily be reformulatedin more traditionalterms. The points discussedthus far are illustratedin (1)-(3). The sentences in (1) exemplify an intransitive verb in the present tense, marked for its subject by a memberof the v-series. The pronounin (lb) and subsequent examples appearsin parenthesesto indicate that (as in many languages, especiallythose with extensive inflectionalagreementsystems),it normally appears only when emphatic. It should be noted that first and second person pronounsin Georgiando not-varyfor case; case is indicatedin the glosses of these elements only by analogy to that which would be formally markedon a thirdperson NP in the same position. (1) a. b. ivane c'veba John (NOM) lies-down-3SBJ John is lying down, going to bed. vc'vebi (me) I (NOM) 1SBJ-lie-down I am lying down, going to bed. The examples in (2) represent a transitive verb, and thus have an m-series markerto identify their object as well as a v-series markerfor their subject.These sentences are in the futuretense, indicatedby the fact that the verb forms are preceded by the appropriate perfectivizing CASE, AGREEMENT AND INVERSION 165 IN GEORGIAN pre-verb.Note thatthe 3rd personobjects in (2a,b) call for 0 markersfrom the m-series. (2) a. ivane mo- k'lav-s mc'er-s John (NOM) insect-ACC PVB-kill- 3SBJ John will kill the insect. b. (me) ivane-s mo- v- k'lav I (NOM) John- ACC PVB-lSBJ-kill I will kill John. c. ivane mo- g- (sen) k'lav- s John (NOM) you(SG., ACC) PVB-2SGDO-kill- 3SBJ John will kill you. The examples in (3) contain indirect objects. The verb of (3a) is intransitive;since this form takes indirectobjects from the u-series, its IO is markedby u (followingthe preverba= in this future tense form). The verb in (3b, c) also takes u-series markers;since its direct object is third person, this is not overtly markedin the form. ivane-s au- sendeb-a (3) a. es saxl- i this house-NOM John- DAT PVB-3IO-built: 3SBJ This house will be built for John. b. (me) v- u- xat'av deda- s I (NOM) I SBJ-3 IO-paint mother-DA T surat- s picture-A CC (=DAT) I am paintingmother a picture. c. deda c. mi- xat'av-s (me) surat- s mother (NONf) 1 IO-paint- 3 SBJ me(DAT) picture- ACC Motheris paintingme a picture. 1.1.2.2 SeriesII. The tenses whichmake up SeriesII consist of the aorist, the optative, and the imperative(virtuallyalwaysidenticalwith either the aorist or the optative, depending on person). While the case markingof NPs associated with verbs whose tense comes from Series I tenses is a simple and straightforwardexample of a nominative/accusativesystem parallelto that of familiarEuropeanlanguages,that associatedwith verbs whose tense comes from Series II is somewhatmore unusual. 166 STEPHEN R. ANDERSON It is necessaryfirst of all to distinguishbetween two classes of intransitive verbs: MEDIAL verbs such as qeps '(he) barks', goravs '(he) rolls', musaohs '(he) works', lap'arak'obs'(he) speaks', and many others; and NON-MEDLAL intransitivessuch as ic'erebs '(it) gets written', xdeba '(it) happens', elodeba '(he) is waiting for (him, it)'. The class of medials has traditionallybeen treated as simply a set of exceptions to the supposedly normal behavior displayed by non-medial intransitives, but a recent detailed study by Holisky (1981a) makes it clear that the medial verbs constitute a large, productive, class whose behavior is morphologically, syntacticallyand semanticallycoherent (cf. also Harris1981) and no more 'exceptional' than that of the non-medials. Holisky characterizes the medial verbs semanticallyas the set of verbs denoting agentive, atelic activities (cf. also Holisky 1981b). We make no attempt to evaluate this description here, since our concern is to describe morphological and syntactic differencesbetween medial and non-medial intransitivesrather than their semantics. The distinction between medial and non-medial intransitives, and between both of these and transitiveverbs, is a lexical one whose precise representationwill be discussed in section 3. As a purely descriptive convenience, we adopt below the terminology of Harris, according to which transitive verbs are characterizedas belonging to 'class I', nonmedial intransitivesto 'class II', and medial intransitivesto 'class III'. The additional category of 'class IV' or 'indirect' verbs will be discussed in section 2 below. Given this classification,we can say that the subjects of verbs of both class I and class III appear in the ERGATIVE case when their tense is a member of Series II. This case is formallymarkedon nouns by the suffix -M(a). Subjectsof non-medialintransitives(class II), in contrast,appearin the NOMINATIVE (exactlyas with Series I tenses). Since directobjects also appear in the nominative,we would have a system of the ergative type associated with the tenses of Series II if we confined our attentionto the non-medialexamplesof intransitiveverbs. Finally,indirectobjects appear in the DATIVE case in Series II just as in Series I. In light of the existence (and completely non-marginalstatus) of the medial intransitives,however, the correct descriptionof the case marking system in sentences with verbs in Series II tenses is arguably as an ACTIVE system, as proposed by Harris (1981). We ignore here the question of whether the semantic difference between medial and nonmedial verbs is precisely that usually associated in the literaturewith the 'active' systemssometimesdescribedfor AmericanIndianlanguagessuch CASE, AGREEMENT AND INVERSION IN GEORGIAN 167 as Dakota, Choctaw, etc. The label 'active' is used here to mean simply that there are two classes of intransitiveverbs, one whose subjects share formal propertieswith the subjects of transitives,and one whose subjects share formalpropertieswith direct objects. We illustratebelow the patternof case markingin sentences with Series II verbs, as well as the pattern of their agreement (which is essentially identical with that found in Series I examples). (4a, b) contain a nonmedial verb in the aorist; their subjects are therefore in the nominative. The verb of (4c), in contrast,is a medialintransitive,whose subject is thus in the ergative. (4) a. ivane mo- k'vd-a John (NOM) PVB-died-3SBJ John died. b. mo- vk'vdi (me) I (NOM) PVB-lSBJ-died I died. k'at'a-m ik'navl- a cat- ERG meowed-3SBJ The cat meowed. In (5), the verb is transitive.Its subject is thus ergative, and its object nominative. (5) a. ivane-m mc'er-i mo- k'l- a John- ERG insect-NOM PVB-killed-3SBJ John killed the insect. b. (me) ivane mo- vk'ali I (ERG) John (NOM) PVB-1 SBJ-killed I killed John. The sentences in (6) contain indirectobjects, which occur in the dative regardless of the transitivity of the associated verb. In all of these exampfes,the indirectobject is reflectedby a markerfrom the u-series. In (6a, b) the verb is transitive and its subject therefore appears in the ergative, its direct object in the nominative. The verb of (6c) is a non-medial intransitive,while that of (6d) is a medial intransitive;the subject of the former is thus nominative, and the subject of the latter 168 STEPHEN R. ANDERSON ergative. These case markingfacts are the same as in the examplesabove where no IO was present. (6) a. da- vu- xat'e deda- s (me) I(ERG) PVB- 1SBJ-31O-paintedmother-DAT surat- i picture-NOM I painted mother a picture. b. deda- m da- mi- xat'- a (me) mother-JERG PVB-1 I0-painted-3SBJ me(DAT) surat- i picture-NOM Mother paintedme a picture. c. es saxl- i ivane-s a- u- send-a this house-NOM John- DAT PVB-3IO-built-3SBJ This house was built for John. d. megobr- is jayl-ma ivane-s u- qep- a neighbor- GEN dog- ERG John- DAT 3IO-barked-3 SBJ The neighbor'sdog barkedat John. 1.1.3 Summary In the discussionthat follows, it is not the detailed formal expressionof particularcategoriesthat will be of interest,but ratherthe extent to which agreement and case marking are correlated with syntactic categories defined by structuralpositions (or by grammaticalrelations). We have observed that verbal agreement marking distinguishes three structural categories: SUBJECT, DIRECT OBJECT, and INDIRECT OBJECT. These distinctionsare made regardlessof tense or of the transitivityof the verb. While case markingdistinguishesthe same three categories, it does so in different ways depending on tense (Series I vs. Series II) and verb class (transitive, or class I; non-medial intransitive,or class II; and medial intransitive,or class III). 1.2 Syntacticpropertiesassociatedwithparticularterms The morphologicaldifferences between Series I and Series II just sur- CASE, AGREEMENT AND INVERSION IN GEORGIAN 169 veyed naturallylead to the questionof whetherthere exist corresponding syntactic differences. In order to answer this, we must first establish syntactic properties that are associated with given positions in grammatical structure, and then determine how the distribution of these properties is related to morphologicalcharacteristics.With this goal in mind,Harris(1981) providesan extensive surveyof the syntacticstructure of Georgian within the frameworkof RelationalGrammar.She isolates a numberof propertiesof NPs which appearto depend on the grammatical relation they bear within the sentence; since reference to these grammatical relations can plausibly,be seen as part of the descriptionof the properties involved (necessarily,in a frameworklike that of Relational Grammar,but at least implicitlyin other theoriesas well), the propertiesin question can serve as criteria for the identification of grammatical relationsborne by NPs in cases where the analysisis in doubt. Harrisargues that despite the morphologicaldifferences, the syntactic structureof sentences is the same in Series I and II. That is, the same NP which serves as 'subject'in a sentence with a verb in Series I is identified by syntacticcriteriaas the subjectif the verb is put in a SeriesII tense with its concomitant morphological changes; and the same for direct and indirect objects. This conclusion is thus an intra-linguisticanalog of the line of argument in Anderson (1976) that syntactic criteria identify the same NPs as 'subject', 'direct object', etc. in 'ergative/absolutive'and in 'nominative/accusative'languages.' Some of the properties identifiedby Harris are common to subjects, direct objects, and indirectobjects; they thus serve to distinguishthe class of such 'terms' from other, 'non-term' NPs. Terms, for example, are reflected in the verb by agreement (as we have seen above), while non-terms are not. Similarly(and, we will suggest below, closely connected with the facts of agreement),non-emphaticpronounsare generally ' There are two qualifications that must be made to the statement that grammatical relations in one language correspond in a relatively straightforward way to those in another. On the one hand, individual lexical items in one language may show exceptional correspondences with their closest equivalents in some other language. For example, French Mes an_s me manquentis semanticallybut not structurallyparallelto its Englishgloss I miss my friends: the subject of naquer corresponds to the object of miss and the indirect object corresponds to the subject of miss. On the other hand (and more significantly), a small fraction of the world's languages, of which Dyirbal is by far the most celebrated example, show a quite different association between grammatical relations and semantic roles than that characteristic of English and most other languages. The existence of such languages, as pointed out in Anderson (1976), actually increases the significance of the conclusion elsewhere, since it establishes the point that the correspondence between morphology and syntax which 'ergative' languages suggest is actually a possible one, though incorrect in most instances. 170 STEPHEN R. ANDERSON omitted in the position of terms,while non-termpositionssuch as genitive modifiersand objects of post-positionsmust be filled either by full NPs or by overt pronouns. Of more interest for our discussion, however, are propertiesthat distinguishone term position from another.The following sub-sections note some of the syntactic characteristics of the most importantterms, the subject and the direct object. 1.2.1 Subjects An importantdistinguishingpropertyof subjects is the fact that they are the only NPs that can serve as the antecedentof (non-possessive)reflexive expressions. Georgian reflexives are formed from the noun stern tav-, which occurs independently with the sense 'head'. The person of the reflexiveNP is indicated(alwaysif firstor second person,optionallyif third person)by a possessive pronounmodifyingtav-. Such 'tav-reflexives'can occur in direct or indirectobject position, or as objects of postpositions; they are alwaysinterpretedas coreferentialwith the subjectof theirclause. Thus sentence (7) below unambiguouslyindicatesthat the paintingwill be done for the benefit of the painterhimself: (7) tav- istvis mxat'varidaxat'avs vanos painter will-paint Vano-ACC self- GEN-for The painteri will paint Vano, for himself i,*j. An obviously related construction with tavis serves as a possessive reflexive.In the case of possessives,however, the antecedentis not limited -to the subject of the clause, and such cases therefore do not provide evidence for subject-hood. Another systematic property of subjects is revealed in the morphologically derived causative construction.When an intransitiveverb is made causative,the subjectof the relatedsimpleverb appearsas the direct object of the correspondingcausative: (8) avari'sa mama-m mzia father-ERG Mzia-NOM caused-to-exercise Father made Mzia exercise. In causative verbs correspondingto basic transitiveverbs, the indirect object of the causative correspondsto the subject of the basic verb: (9) cecxli daantebina mama-m mzia- s father-ERG Mzia-DAT caused-to-light fire-NOM Father made Mzia light the fire. CASE, AGREEMENT AND INVERSION IN GEORGIAN 171 Unfortunately,as Harris(1981: 291) points out in anotherconnection, the patternof NPs in associationwith causativesshows less than we might like it to. This is because indirect objects of intransitive verbs also correspond to indirect objects of the related causatives. Therefore, the indirect object of the causative in (9) could in theory have that status either because it correspondsto the subject of a transitive,or because it corresponds to the indirect object of an intransitive; and since it is precisely the choice between these two possibilitiesthat is most often in question, causatives have limited value as evidence in this connection. 1.2.2 Direct Objects An interesting property of direct objects is the fact that they undergo raising to subject position when their clause is embedded under advili 'easy' and similarpredicates.This constructionis illustratedin (10) below: (10) sarvali advili-a ninos-tvis sesak'eravad trouserseasy- be Nino-for to-sew Trousersare easy for Nino to sew. (or 'to sew for Nino') The fact that only direct objects, and not either subjects or indirect objects, are availableto be raisedin this constructionis illustratedby the fact that sentence (11) below is ungrammatical,regardlessof whethernino is interpreted as underlying subject of esak'erad 'to sew' or as a benefactive indirect object: (11) *nino advili-a sarvl- is sesak'eravad Nino easy- be trousers-GEN to-sew *Nino is easy to sew trousers(for). Another propertyof direct objects has alreadybeen illustrated:in the causative of a transitiveverb, the direct object correspondsto the direct object of the correspondingbasic verb. Sentence (9) above illustratesthis pattern. Again, however, the value of such facts as evidence is limited: direct objects of causatives may correspond either to direct objects of basic transitives,or to subjects of basic intransitives(cf. (8)). Another propertyof direct objects whichfiguresin Harris'discussionis the fact that certain transitiveverbs show suppletivestem forms, depending on propertiesof their direct objects. Thus, the verb (gads = )a-gdeb 'throw'only occurs with singulardirect objects, while (gada=)qri 'throw' is used with plural direct objects; similarly,the verb (da=)ban 'wash' is used only with personal (human)direct objects, while the corresponding verb (ga=)recxav 'wash' is used with non-personal (non-humanor in- 172 STEPHEN R. ANDERSON animate)direct objects. The relevance of this to the status of particular NPs in the clause is arguedto derive from the apparentlyvalid generalization that if a transitiveverb shows suppletionfor some propertyof one of its arguments,the relevant argumentis alwaysthe direct object (and not the subject). It is by no means clear that these facts are actually relevant to the syntactic analysis of Georgian, however. When we consider intransitive verbs, we find a numberthat show similarsuppletionbased on properties of their subject: thus, (da=)Jdeb 'sit down' is used with singularsubjects, while (da=)sxdeb is used with plural subjects; and c'evs '(he) is lying down' is used with personalsubjects,while devs '(it) is lying down' is used with non-personalsubjects. These facts show that not only direct objects can cause suppletion, but subjects as well. The full generalizationthus appears to be that suppletion is governed by the properties of an intransitivesubject or a transitiveobject. In fact, we suggest, this generalizationis only indirectlyconnected with the syntacticrelationsinvolved: the apparentrelationarisesbecause these are preciselythe positionsoccupied by the NP fillingthe semanticrelation (or '0-role') of THEME in the interpretationof a clause. We suggest that it is the propertiesof themes that are relevant to the choice of suppletive verbs. A similar proposal is made by Hewitt (1983). The importanceof themes in grammaticalstructureis well known (cf., e.g., Anderson 1977b and references there); given the semantic nature of the restrictions involved in suppletion, it is particularlyplausible that they should be controlledby an aspect of semantic,ratherthansyntacticstructure.But in that case, the facts of suppletionare not directlyrelevantto identifyingthe grammaticalrelation a NP bears. 2. THE INVERSION CONSTRUCTION The facts surveyedto this point are quite straightforward,if a bit complex: aside from the point that Series I and Series II tenses differ in their associated case-marking (though not agreement) patterns, the morphologicaland syntacticpropertiesof Georgianare not especiallyunusual in comparisonwith other languages. But the language also illustratesyet anotherpatternof case-markingand agreement,which, while by no means isolated in cross-linguisticterms, is nonetheless remarkable.This is the 'inversion'construction. CASE, AGREEMENT AND INVERSION IN GEORGIAN 173 2.1 Morphologicalpropertiesof 'inverted'forms. Under certain circumstances(to be described immediatelybelow), the apparent subject of transitive verbs, as well as of intransitivesof the 'medial'class, appearsin the dative case, ratherthan the nominative(as is usualwith Series I tenses) or the ergative (as is usualwith Series II). At the same time, instead of being indicated on the verb with markersfrom the v-series (which,it will be recalled,normallymarkssubjectagreement),it is marked with an element from the u-series, or in the case of certain exceptional verbs, with elements from one of the other sets normally employed to mark indirect objects. The subject in this constructionthus behaves morphologicallyin exactly the way indirectobjects behave in the constructionsdiscussedin section 1. . At the same time, under the same circumstances,the apparentdirect object of a transitiveverb appearsin the nominativecase. We can recall that this is the normal case for direct objects with Series II (though not Series I) tenses, but in contrast to these, direct objects in the 'inverted' forms are indicated on the verb by markersfrom the v- (or 'subject') series. A further complication arises from the fact that the verbs in questionare followedby a suffixedform of the verb qopna 'be', agreeingin person with a first or second person direct object. This peculiarity (considered as a part of 'v-series agreement') is also found in agreement with the subjects of present.forms of some non-invertedverbs (especially medials), and so is not isolated within the morphology of Georgian. It does, however, reinforcethe conclusionthat in the 'inverted'forms,direct objects have the morphologicalpropertiesof subjects. As far as the indirect objects of 'inverted'transitiveor medial intransitive verbs are concerned, these can only appear in the form of a post-positionalphrase markedby -tvis. In its other uses in the language, this post-position can generally be glossed 'for', and it overlaps in its semanticswith the range of interpretationsassignedto indirectobjects. In the 'inverted'forms,(notional)indirectobjects are not markedon the verb at all, and their appearanceis limited to such -tvis phrases. Finally, we can note that when non-medial intransitiveverbs appear under the conditions that call for 'inverted' forms of other verbs, no correspondingchanges in case markingor agreementare observed. Their subjects, that is, appearin the nominativeand are markedby the v-series on the verb;.andtheir indirectobjects (if present)appearin the dative and are indicated by one of the appropriateseries of markers. 174 STEPHEN R. ANDERSON There are two sets of circumstancesunderwhich such 'inverted'forms are found. The firstof these is in the final group of tenses, those classified in traditionaldescriptionsas Series III and includingthe perfect, pluperfect, and conjunctive perfect. Although Series III is usually called the 'Perfect Series' in the literature, these tenses are not 'perfects' in a standardsense. They are rathertense formswhose main use is to describe events the speaker has not actually witnessed, but rather inferred (cf. Aronson 1982a, Harris 1981, Sanije 1980, and other references cited in these works for extensive discussion of the semantics of the Georgian Series III tenses). The perfect is also the basic past tense in negative contexts corresponding to the (non-negative) aorist. With non-medial intransitives(whether or not they take an indirect object), as we noted above, there is no inversion.Although in this case the internalformation of the SeriesIII tenses is differentfrom that associatedwith transitivesand medial intransitives, it is nonetheless clear from their syntactic and semantic distributionthat the same set of tenses is involved. Invertedforms are also found in Series I and II formsof a particularset of verbs, mostlyinvolving perception,emotion, ability,or mental attitude. These we referto as constituting'class IV' following the terminology(due to Harris)introducedabove for the otherlexical classes of verbs. The class IV verbs in Georgian(i.e., those that show inversionin all tense series and not only in Series III) are cognate with verbs that often show similar, inverted behavior in other languages on a locally idiosyncratic basis. Compare, for example, archaic English Me thinks, Italian A Giorgio piacciono le sinfonie di Beethoven 'Georgio likes the symphonies of Beethoven', or Russian Emu zaxotelos' rabotat' doma 'He felt like working at home'. In all of these cases the notional subject experiencer shows the morphologicalcharacteristicsof an indirect object. An over-emphasison the partialsemanticcoherence of the set of verbs making up class IV can lead to a mistaken appreciationof the status of inversionin Georgian. Merlan (1982), for example, effectively limits her discussionof Georgian inversion constructionsto a considerationof the class IV examples, and concludes that the cross-linguisticsimilarityjust noted is sufficientto force us to consider the invertedforms as having (at all levels) a syntaxwhich is directlyrevealed by their surfacemorphology. Since her discussion takes none of the relevant syntactic evidence into account, however, and largely ignores the problemof inversion in Series III tenses, this conclusion seems unwarranted.The semantic similarities between class IV verbs in Georgian and those that show some sort of inverted structure in other languages remain simply suggestive (and CASE, AGREEMENT IN GEORGIAN AND INVERSION 175 nothing more) until an adequate and explicit account is available of the relation between syntactic and semantic categories in the languages involved. We have nothing to say here about the semanticcharacteristics of some verbs which make them candidatesfor such 'inverted'syntax in manylanguages;what is importantabout them in Georgianis the fact that their constructionis the same as that associatedwith a much wider set of verbs (i.e., classes I and III as well) under certain systematic conditions (namely,when their tense is one of those in Series III). We now illustrate the properties of the inversion construction. The sentences in (12) below are negatives corresponding to non-negative sentences in the aorist;their tense is thus the 'perfect',a memberof Series III: ar da- (12) a. (me) dedis- tvis mi- xat'av-s I(DAT) not PERF- 11O- paint- 3SBJ mother-for surat- i picture- NOM I didn't paint a picture for mother. b. deda- s ar da- u- xat'av-s cem-tvis mother- DAT not PERF- 31O- paint- 3SBJ me- for surat- i picture-NOM Mother didn't paint a picture for me. The verb in (12) is a basic transitive(class I); its subject thus appearsin the dative and is markedon the verb by what is normallyan indicatorof indirectobjects. The directobject is in the nominative,andshowsconcord with a 'subject' marker,while the indirectobject appearsin a tvis-phrase and is not markedon the verb at all. The same morphologicalproperties can also be observed for the subjects of class III (medial intransitive) verbs, as shown by (13): (13) a. k'at'a- s cat- ar u- a k'navli- DAT not 31O- has meowed- 3 SBJ The cat has not meowed. b. (me) ar mi- k'ivli- a I (DAT) not IO- have screamed- 3 SBJ I didn't scream. 176 STEPHEN R. ANDERSON Note thatin (13), the medialverbs are inflectedwith a 3sg 'subject'marker (regardlessof the person of their actual subject, which correspondsto an 'indirect object' marker).They thus behave (by morphologicalanalogy with class I verbs) as if they had a dummy3sg direct object. This fact will play a role in the analysisbelow. In (14) below, we illustratea verb from class IV, the class that shows 'inversion'in all tenses. These sentences have their verbs in the present tense, which would normallycall for nominativesubjects markedby the v-series, and accusative objects markedby the m-series; instead we find dative subjectsmarkedby the u-series, and nominativeobjects markedby the v-series (plus an enclitic form of the copula, when the object is non-thirdperson): ana (14) a. (me) mi- qvar-s I (DAT) Anna (NOM) 1IO- love- 3SG I love Anna. vb. ana- s u- qvar-var (me) Anna- DAT me (NOM) 1SG- 3IO- love- am Anna loves me. Membershipin class IV, with its associatedinversionin all tenses, is not limited to transitiveverbs. A few intransitivesbelong to this class as well; like the Series III forms of intransitivemedials, these verbs appear to be inflectedfor a dummy3sg direct object. m- jinav- s (15) a. (me) I (DAT) 1IO- sleep- 3 SG I'm sleeping. b. vano- s jinav- s sVano- DAT 3 IO- sleep- 3 SG Vano is sleeping. At this point, we have introducedall of the patternsof case markingand agreementwhich occur in Georgian as a function of verb class and tense Series. A summaryof these patternsis presented in Table II below. 2.2 Harris' analysis of inversion The questionraisedby the inversionconstruction,of course, is whetheror not the morphological properties of such sentences are an accurate CASE, AGREEMENT AND INVERSION IN GEORGIAN 177 TABLE II Case markingand agreementpatterns PatternA: Case-Marking Agreement Subject Direct Object IndirectObject Nominative v-series Accusative m-series Dative h-, etc. series Subject Direct Object IndirectObject Ergative v-series Nominative m-series Dative h-, etc. series Direct Object IndirectObject Nominative v-series (tvis-phrase) (none) PatternB: Case-Marking Agreement PatternC ('inverted'clauses): Subject Case-Marking Agreement Dative h-, etc. series Distributionof PatternsA, B, and C: Series: I II III Verb Class: 1 (Transitive) 2 (Non-MedialIntransitive) 3 (Medial) 4 (Inversionverbs) A A A C B A B C C A C C indicatorof their syntacticstructure.Is it the case, that is, that in a clause whose verb has a tense from Series III, the NP correspondingto the subject of the same clause with a Series I or II tense is in fact a structural indirectobject? An analysisthat did not posit majorstructuraldifferences correlatedwith such tense differences would have an immediateappeal. 2.2.1 The syntacticstructureof inversionclauses Harris(1981) argues that in inversion forms, the dative NP is indeed the syntacticsubjectand the nominativeNP is the directobject. She bases this conclusion on the same constructions that she uses to demonstratethe structuralparallelbetween Series I and Series II (non-inverted)forms.For 178 STEPHEN R. ANDERSON example, in Series III clauses (as in Series I and II), only the notional subject can serve as the antecedentof tav-reflexives.Sentence (16) below, whose verb is in the 'perfect' tense, illustrates this; note the ungrammaticalityof the variant (16b), in which the reflexive is in the position whichappearsmorphologicallyto be an indirectobject and the antecedent in what appearsmorphologicallyto be that of the subject: (16) a. gela- s turme da- u- rc'munebi-a tavisi Gela-DAT evidentlyPVB-3IO-convince- 3 SBJ self's tavi self-NOM Evidently Gela has convinced himself. b. *tavis tav- s da- u- rc'munebi-a gela self s self-DAT PVB-3IO-convince- 3SBJ Gela-NOM (*Evidentlyself has convinced Gela.) Evidence from other sources bearingone way or the other on this issue is not easy to obtain. For instance, when we consider -causatives,the evidence from Series III forms is not immediatelyrelevant, since it is the causative verb itself that is 'inverted'in such structures(as the bearer of tense) ratherthan the basic verb from which it is derived. We might then turn to class IV verbs: since these appear 'inverted'in all tense forms, it might appearthat they have a basicallyinvertedstructure,which ought to be reflected in the form of their causative. a(17) a. vano- m gela- s a sejulVano- ERG Gela- DAT PVB- 31O- caused to hate- 3SBJ nik'o Niko-NOM Vano made Gela hate Nik'o. b. (mat) t'usayi mo- asives they-ERG prisoner-NOMPVB- caused hunger-3PLSBJ They let the prisonergo hungry, they starved the prisoner. Unfortunately,the causativeof a transitiveclass IV verb such as m-jul-s 'I hate him' as in (17a) is of no relevance to the issue at hand. In this structurethe argument(gela) correspondingto the notionalsubject of the basic verb 'hate' does appearas an indirectobject, which is whatwe would CASE, AGREEMENT AND INVERSION IN GEORGIAN 179 expect if it were basically the subject of a transitive;and the argument (nik'o) correspondingto the notional object of 'hate' appearsas a direct object, again as we would expect for a basic direct object. As noted above (andby Harris1981, p. 291), this is exactly the same as the resultwe would predict if the personhated were structurallythe subject of (an intransitive verb) m-jul-s, and the hater the indirect object. In the case of an intransitiveinversionverb such as m-sia 'I am hungry',however, the fact that the NP with morphologicalindirect object propertiesin association with the basic verb appearsas a directobject with the causativeas in (17b) confirmsits notional status as a subject. Turning to raising constructions,there is again no evidence available from Series III tense forms,since the clause from which an object is raised (in, e.g., sentence (10) above) is necessarilynon-finite (i.e., appearsin a form that bears no tense). With regard to class IV verbs, we see that indeed the notionalobject of such a verb can be raised,and no other NP in the clause has this possibility: (18) direkt'ori AZvili-a vanos- tvis sesajuleblad director-NOMeasy- is Vano- for to hate The directoris easy for Vano to hate. Harris (1981: 292) notes, however, that this fact is also equivocal in its bearing, because the verb m-jul-s (like most other inversionverbs) has a related form (se=) i-juleb 'begin to hate someone', a class I verb built on the same root. Since non-finiteforms involve the neutralizationof class differencesamongverbs with the same stem and preverbsuch as this, one could argue that (18) arises from raising the object of the class I verb, rather than the (commoner)class IV verb. Of course, sentence (18) is consistentwith an analysiswhich treatsthe embeddedverb as the class IV form, and raises its direct object - but it does not provide positive evidence for this analysis. Harriscites some additionalfacts in supportof her conclusion,including suppletionfor number,animacy,and tense with invertedverbs, the facts of number agreement, and the marking of 'retired terms'. We suggested above that the facts of suppletion are not directly relevant to the determinationof syntactic structure,since these are related to semantic rather than syntactic relations. The facts of number agreement will be discussed below in section 2.3.4, where they will be seen to reinforcethe conclusion that NPs with the morphologicalpropertiesof indirectobjects in inversionconstructionsare in fact subjects. The facts of 'retiredterm' 180 STEPHEN R. ANDERSON markingwill not be dealt with separatelyhere, though I do not feel they provide substantialadditionalsupportfor proposalsabout the underlying structureof inversionstructures. We conclude, however,that such facts as there are either supportor are at least consistent with the claim that 'inverted' clauses (i.e., those containing Series III tense forms or class IV verbs) have the same syntacticstructureas others. Given the obvious parallelsbetween Series I and II, on the one hand,and SeriesIII on the other, we accept here Harris' claim that the two should be analyzedas having the same basic structure, in particularthe same subject and object NPs. It remains, however, to account for the morphologyof the 'inverted'forms. 2.2.2 A syntacticanalysis of inversion Within the frameworkof Relational Grammar,Harris (1981) provides a resolution of the apparentlycontradictoryaspects of the inversion construction. The essence of that analysis is the following: since NPs in inversion clauses have the same subject and object properties as the correspondingNPs in non-inversionclauses, we can assign all types of clause a common underlying structure (at least as far as grammatical relations are concerned). However, in the presence of the two inversiontriggeringfeatures([classIV] as a propertyof the verb, or [Series III] as a property of its tense), the morphology corresponding to the syntactic subject is that of an indirectobject; and that correspondingto a syntactic direct object is that of a subject. Therefore, a rule can be posited which demotes subjectsto indirectobject; anotherrule then promotesthe direct object to the status of a subject: (19) a. Inversion: Subject-> Indirect Object (triggered by class IV verbs or Series III tense forms) b. Unaccusative: In the absence of a Subject, a Direct Object becomes a Subject The process is split into two rules, Harris argues, because the Unaccusative rule (19b) is necessary independently of the inversion constructionalone. In particular,all clauses with verbs of class II (non-medial intransitives)have underlyingrepresentationscontaining a direct object, but no subject; and this direct object is promotedto subject by the same rule (19b) that functions in the inversion construction.It is beyond the scope of this articleto evaluate this proposal(equivalentto the suggestion of a class of 'unaccusative'verbs in a numberof other languages;cf. e.g. CASE, AGREEMENT AND INVERSION IN GEORGIAN 181 Postal 1977, Perlmutter1983 and referencesthere, Burzio 1981, etc.), but withinGeorgianwe suggest that it does not have independentmotivation beyondthe facts that a) it providesa descriptionof the differencebetween class II andclass III;b) it relatesthe structureof class II to that of passives; and c) it requires no apparatusbeyond that necessary to describe inversion. We returnto these points below. Given the rules in (19), inversion clauses (such as that in (20), containinga verb of class IV) will be given a Relationalstructuresuch as that below: _ (20) s- julebi- a vano- s direkt'ori 3IO- hate- 3SBJ Vano- DAT director Vano hates the director. In this structure,vano is underlyinglysubject and direkt'ordirect object: the former undergoes demotion (from '1' to '3') by rule (19a), with the result that the clause no longer has a subject; and (19b) promotes direkt'or to subject status. The resulting structure provides a suitable basis for morphologicalmarkingof the NPs involved. Harris'relationalanalysisprovides a compact and elegant account of the facts of inversionconstructions,and their relation to other aspects of Georgian syntax.The only controversialaspect of this analysisis the fact that it treats inversion as an essentially syntactic (rather than morphological) phenomenon,by positing a rule which alters syntactic grammatical relations and then treating the morphology as straightforward given the change made by this rule. The reason this is problematicis the following:evidence for an initialstructurein which inversionclauses have the same structure as non-inversion clauses derives from syntactic phenomena. Morphologicalevidence for this structure is either highly problematic in theoretical terms (number agreement; cf. section 2.3.4 below) or of unclear status in the grammar ('retired term marking'). Argumentsfor the (derived) indirect object status of the surface dative NP, on the other hand, are exclusively morphological: they consist essentiallyof the facts of case markingand verb agreement. An analysis which expressedthe morphologicalpropertiesof the constructionwithout requiring (otherwise unmotivated) changes in syntactic structure, and 182 STEPHEN R. ANDERSON which did not lose generalizationscapturedin the syntacticanalysis,would enjoy at least the advantageof conceptual economy. Theories other than that of Relational Grammar typically posit a connection between structuralpositionsand semanticroles such that even in the presence of a syntactic relation-changing inversion rule, the descriptionof Georgian inversionmust involve another, essentiallymorphological component. For example, an analysiswithin the frameworkof Lexical FunctionalGrammar(Bresnan1982) would requirea lexical rule associating a difference in role assignmentwith the morphology of the relevant inversion structures,if the grammaticalrelations in the surface form of inversion structures differ from those in corresponding noninverted forms. On the assumptions of Government/Binding theory (Chomsky 1981), both direct and indirect object are sub-categorized positions, and each must thus be assigned a 6-role. In order to avoid a violationof the 6-criterion,thismeansthat the 6-role normallyassignedto subjects must be re-assigned to indirect object position instead. A Government/Bindinginterpretationof Harris'analysisthus requiresus to posit a lexical rule which (at least) associates this change in 6-role assignment (as well as the absorption of case from the direct object position) with inversion morphology,assumingthat there is movement in such constructions. Note that the syntacticanalysisrequiresthat Inversion(takingthis term to refer to the pair of rules in (19), including Unaccusative) essentially must not interactwith other relation-changingrules. This is because other rules which change relations (e.g., passive, raising) do not apply to the output of inversion, and inversiondoes not apply to their output either. The one exception to this is the formationof causatives,to which inversion can apply; but this process would be described as lexical rather than syntactic by grammariansworking in frameworksother than Relational Grammar,and so cannot be called a clear counterexampleto the claim that inversion is not fed by genuinely syntactic rules. The failure of Inversion to feed other rules follows (in Harris' formulation)from the fact that such other rules have structuralrequirements that are inconsistentwith the output of inversion. The second half of the generalization,however, is describedby Harrisby imposingthe condition that "Inversion[i.e., (19a) above] only applies to initial subjects"(Harris 1981, p. 247). As a result, the formulationof Inversionbecomes global in character (since it must refer simultaneouslyto the fact that a given argument is a subject, and to the fact that it was initially a subject), a consequence which most views of syntaxwould reject. On the other hand, CASE, AGREEMENT AND INVERSION IN GEORGIAN 183 Inversion must precede (and feed) the morphological rules of case marking and agreement. An alternativeanalysisshould attempt to make these interactions follow as far as possible from more general considerationsof the inter-relationof morphologyand syntax. 2.3. Problemswith a syntacticanalysis of inversion In the previoussection, we suggested some reasonsto believe that it might be preferablea priori to describe the facts of the inversion construction within the limits of the morphologicalsystem of Georgian, rather than adoptingan analysisinvolving a change of syntacticstructuresuch as that posited by Harris.In this section we providesome more explicit arguments againstthe syntacticapproach.In each case, the thrustof our observations is that the structureunderlyinginversion constructionsshould be maintained unchanged: either the syntactic rule of Inversion leads to a structurewhich is in some way ill-formed, or there is evidence that the posited underlyingstructure(ratherthanthe outputof syntacticInversion) is appropriate as a derived structure as well. If we assume that no alteration of the underlyingstructure is involved in an account which treatsinversionas a fact about the structureof inflectedwordsratherthan as a fact about the syntax of 'inverted' clauses, these argumentssupport such an analysis. 2.3.1 The well-formednessof the movementinvolved Recall the effect of rule (19a) above, the central part of the syntactic analysis of inversion. This rule demotes subjects (in the presence of an appropriatetrigger) to indirect objects; the result is a structurewith no subject NP. In the case of transitiveverbs, this lacunais immediatelyfilled through the operation of (19b), promoting the former direct object to subject. Intransitiveverbs can undergo inversion too, however - at least those like k'navis 'meows',which belong to class III (the medials).In this case there is no NP availableto occupy subjectposition.WithinRelational Grammar, this would violate the 'Final-1 Law', which stipulates that clauses musthave a surfacesubject;it is thus necessaryto posit a syntactic dummy element which occupies subject position (either directly, or by advancementfrom direct object position). Within the Government/Bindingtheory, inversion structureslike (21) below would violate otherwise valid binding conditions if they involved genuine syntactic movement, since movement from subject to indirect 184 STEPHEN R. ANDERSON object position would leave a subject trace bound only by an NP in indirect object position. Such a trace would not be c-commanded by its antecedent, and thus the structurewould not be well formed. (21) S NP VP NP [e]i ' V NP V 0 u-k'navli-a [kat a-s]i cat-DAT 3 IO- meowed-3SBJ The cat has meowed. (perfect tense) We have representedindirectobject position as a daughterof VP, and treateddirect objects as daughtersof V; nothing hinges on this detail, so long as indirect object position is a part of some projection of V within which the main verb can be subcategorizedfor its complements, while subject position is externalto the maximumprojectionof V. The analysis must reflect the fact that direct and indirect objects are independently subcategorizedby the verb while subjects are not. The ill-formednessof the antecedent-tracerelation in (21) under the bindingtheoryof Chomsky(1981) has no obvious resolution;note that we cannot assumethe trace is simply'covered'by a syntacticdummyelement (such as English there),since regardlessof the morphologicalindication on the verb, this position cannot be filled by an overt pronoun or other element under any circumstances.In contrast, the morphological treatment of inversion involves no movement from subject to indirect object position, and results in no structurethat is problematicfor the binding theory.The subject NP continues to fill the subject position, regardlessof its morphologicalreflection, and no trace in that position must be bound by an element subordinateto it. CASE, AGREEMENT AND INVERSION IN GEORGIAN 185 2.3.2 Non-finiteforms and the natureof verbclasses Verbs which are not marked for tense and agreement must not trigger syntacticinversion,regardlessof whetheror not they contain roots whose finite forms can be inflected as belonging to class IV. In sentence (22) below, we see that the dependentsof sejuleba 'to hate' are markedas they would be in a normal,active nonfinitestructure:tkven is markedwith the post-position mier, as is normal for a transitive subject (n.b.: not an indirectobject), while cemi megobrisis a genitive, the normalform for a direct object dependent of a non-finiteform: (21) tkven mier cemi cem-tvis gaugebaria you by my me-for incomprehensible megobr- is sejuleba friend- GEN hating For me it is incomprehensiblethat you would hate my friend. An adequategrammarmust thus ensurethat infinitiveforms like gejuleba do not trigger inversion,even if they correspondto class IV verbs. This raises the issue of what it means for a verb to belong to 'class IV' (or any other class). The evident generalizationhere (suggested to me by Alice Harris)is that "a non-finiteformhas no class":i.e., that class is a fact aboutfiniteinflection.Therefore,on an analysisthat treatsinversionnot as syntactic structuralchange but ratheras a fact about a verb's agreement morphology,there is nothingfurtherto be said to ensure that inversionis blocked in non-finiteformseven for those verbs which otherwiseundergo it everywhere (i.e., class IV): lack of inversionfollows directly from the fact that such forms have no agreementmorphologyto 'invert'. An analysisthat treats inversionas syntactic,however, does not derive this consequenceso directly.This is because the Inversionrule is said to be triggeredby two conditions:SeriesIII inflection(for verbs of class I or III) and class IV verbs. Harris shows that no reference to verb class is necessary on her account to prevent inversion in Series III forms from applyingto verbs of class II, but the inversionrule must still refer to class IV forms. While it is true that the notion of 'class IV' is only relevant to finite forms, and thus that non-finite forms could never be subject to inversion,it remainsthe case that the referenceto class IV inflectionin the syntactic Inversion rule is in principle independent of the actual agreement morphology itself. As we will see below, the morphological account does not posit any lexical indication of class beyond the 186 STEPHEN R. ANDERSON agreement morphology, and the operation of inversion is a completely predictablesubpartof the developmentof certainfiniteforms (ratherthan being triggered by a reference to those forms in the syntax), so the generalization that non-agreeing forms have no 'class' in the relevant sense is a tautology. 2.3.3 Wordorder Georgian has considerable freedom of surface word order within the clause, but the extent of this freedom has sometimes been exaggerated (e.g. by Hammond1981 who arguesthat the languagehas completelyfree order, suggesting a 'non-configurational'account of its syntax).There is fairly clearly a neutral order in declarative sentences, deviations from which have special stylistic motivations (emphasis, highlighting of new information, etc.). Based on a study of texts by Pochua, Vogt (1971) argues that the neutralword order has the subject in initial position; the direct object is adjacent to the verb (on either side), and the indirect object (if present) is either on the other side of the verb from the direct object or separatedfrom the verb by this latter NP. Schematically,the occurringneutralordersare: S-V-DO-IO, S-DO-V-IO, and S-10-DO-V. This suggests that a structurealong the lines of that posited in (21) above is appropriate,with word order being unspecified internal to the projections of V.2 Permutationsbeyond those of the neutralorderwould then be produced by late stylistic rules, whose operation is irrelevant to the syntax per se. What is importanthere is the fact that the central generalizationabout neutralword order concerns the position of the subject: this is sentence initial.Furthermore,among the subjects which obey this principleare the (basic,or underlying)subjectsof sentences thathave undergoneinversion. This is illustrated below for both types of inversion structure: (23a) containsa verb of class IV, and (23b) has a verb in the pluperfect(a series III tense). 2 The account developed here predicts a fourth possible neutral order: S-IO-V-DO. That this is not cited by Vogt as having the same status as the other three mentioned here shows that some additional principle(s) are at work in determining Georgian word order: hardly a remarkable conclusion. In this case, we might propose that a non-subject immediately preceding the verb is most naturally interpreted as its direct object, if possible. We have no independent evidence for this claim, but it does not seem implausible; and in any event, what is at stake is this section is the ordering property of subjects. CASE, AGREEMENT AND INVERSION IN GEORGIAN 187 tavisi svilebi u- qvar-s (23) a. mama- s father- DAT self's children3IO- love- 3 SBJ Father loves his children. a b. deda- s gogo da- emalmother-DA T girl PVB- 3IO- had hidden- 3SBJ Mother had hidden (pluperfect)the little girl. The fact that the notionalsubject occupies the same positionin (neutral) word order regardlessof whether the constructioninvolves inversion or not follows triviallyif we assume it continues to be a subject (despite its morphologicalreflection).On an analysisinvolvingsyntacticinversion,on the other hand, it is necessary to formulatethe relevant generalizations about word order in an inherently global fashion. Harris (1981, p. 302) thus states that the "'first subject that is a final term' regularlyoccupies the first position"- a formulationwhich is unavoidableif the underlying subject ceases to be a subject in the surface structures of inversion constructions,but which requiresreference simultaneouslyto the surface status of a NP and to its derivationalsource. 2.3.4 Numberagreement The conditions on number agreement in Georgian are somewhat more complexthanthose on person agreement.Essentially,the verb agrees with any first and second person NP in number if it agrees with that NP in person; but verbs agree with third person NP in numberonly if these are subjects. The agreement in some cases is by a markerwhich is syncretic with the personmarker:thus, the thirdpersonpluralv-series markers-en, etc., represent both person and number; as does the first person plural m-series markergv-. Where no such syncreticmarkerexists (i.e., for first or second person v-series markers, and for second person m-series markers),numberagreement is by means of the suffix-t 'plural'. Because third person NPs that are marked on the verb with markers from series other than the v-series are generally non-subjects, nothing normallyappearson the verb to indicatetheirplurality.The one exception is in the inversion construction:here a markerof the u-series (or other 'indirectobject series',3when requiredby a particularverb) in the third 3 Note that the labels on the columnsin Table I referto the grammaticalrelationsborneby. the correspondingNP's in non-invertedformsonly. The essence of the syntacticaccountof 188 STEPHEN R. ANDERSON person may be accompaniedby the pluralmarker-t in agreementwith a 3pl NP: (24) tavisi mama u- qvar-t svileb- s children-DAT self's father-NOM 31O- love- PL The childrenlove their father. Interestingly,a third person plural (direct object) NP marked on the verb with a v-series marker in the inversion construction cannot show numberagreement, and must agree as if it were singular: (25) mama- s tavisi svileb- i uqvar-s/ father- DAT self s children-NOM 3IO- love- 3 SGSBJ/ *u- qvar- en 31O- love- 3PLSBJ The father loves his children. Given the generalization that third person NP can trigger plural agreement if and only if they are subjects4, these facts provide further supportfor the claim that the subject in an inversion constructionis the same as that in a non-inversionstructure. inversion is the claim that these labels are also applicable to (the surface structureof) inverted clauses. On the analysisbeing developed here, this is not the case; but we will continue to refer to v-series agreement as 'subject-agreement',to u-, b-, etc. series agreement as 'indirect-object-agreement',and so on, for want of a better alternative. Context should make clear the extent to which the 'indirect-object'in a reference to 'indirect-object-agreement', etc., shouldbe taken seriously. 4 Hewitt (1983) arguesthat in some cases, the indirectobjects of relative intransitiveverbs can triggerpluralagreement: ra mo- u- vid- at mat what (NOM) PVB- 3IO- came- 3SG- PL 3PL-DAT What came over them? (i) He also notes that,while Harrisdescribesthe verb da-e-k'arg-ain (iia)below as not showing plural agreementbecause muoblebs'parents'is an indirect object, she also suggests that e-k'sg-eb-a-t in (iib) does have plural agreement, because here zmnebs 'verbs' is the subject of an invertedverb: (ii) a. b. da- ebavivi msobl- eb- s k'arg- a parent-PL- DAT PVB-3IO- lost- 3SBJ child The child was lost to the parents. v et evian zmn- eb- s k'argeb-a3SBJ- PL v-having verb- PL- DAT v (NOM) 3IO- loseVerbs in v lose v. CASE, AGREEMENT AND INVERSION IN GEORGIAN 189 The problemis that on a syntactic analysisof inversion,these subjects have ceased to be subjects at the point agreementapplies:Inversionhas demoted them to indirectobjects. The statementof numberagreementfor non-subjectsmust thus refer both to the grammaticalrelationborne by a given NP on the surface, and to the fact that this NP was originally a subject. The implicitly global formulationprovided by Harris (1981, p. 219) states that "[a] thirdperson nominaltriggersNumberAgreement in the verb of which it is a finalterm if [...]it is the firstsubject of that verb that is a final term." In contrast, on a purely morphological account, the fact that third person 'indirectobject' agreementis sensitive to numberonly in inversion constructions (and conversely, third person 'subject' agreement is insensitive to number in exactly the same forms) follows directly and non-globallyfrom the fact that exactly subjectstrigger plural agreement with third person NPs. If such subject agreement material eventually undergoes translation as 'indirect object' markers, it nonetheless can reflect number. On the other hand, whether a non-subject eventually Since the verbs in these sentencesappearto be the same (withthe exceptionof their tense), the claim that the datives in them bear differentrelationsappearsto be an inconsistency. In fact, the phenomenonin questionis ratherwidespread.Tschenkeli(1958, pp. 484-490) discussesit at some length, and cites some apparentlynear-minimalpairs: (iii) a. kurd-i ga- ep'ar- a p'oliciel- eb- s thief- NOM PVB- 3IO- escape- 3SBJ police- PL- DAT The thief escapedfrom the police. (emphasis:the thief managedto get away from the police) b. p'oliciel- eb- s ga- e- p'ar- at kurd-i police- PL- DAT PVB- 3IO-escape- 3SBJ-PL thief- NOM The thief escapedfrom the police. (emphasis:the police are the affectedones, in that they had the misfortuneto have the thief escape from them) According to Tschenkeli, this apparentlyinverted use of relative instransitiveverbs (evidenced by word order and especially by the possibilityof plural agreementwith an 'indirectobject')arisesspecificallywhen the indirectobject is interpretednot simplyas a goal or 'undergoer'of the action described,but as affected by it or an experiencerof it. This suggeststhe resolutionof the apparentinconsistencyin the sentencescited by Harris: for a large class of verbs, two interpretationsare availablewhich differ not in their basic sense, but rather in the way described by Tschenkeli. Since the morphologicalpatterns associated with relative intransitiveverbs and with inverted (class IV) verbs are almost identical,shiftsbetweenthese classesusuallyhave no formalreflectionbeyondthatshownas the differencebetween (iia) and (iib), (iiia) and (iiib). We have to do here then not with an inconsistentdescription,or with optionalnumbermarkingassociatedwith syntacticindirect objects, but ratherwith switchesbetween two inflectionalclasses dependingon whetherthe dative NPs associatedwith certainverbs are interpretedas experiencersubjectsor simplyas affected goals. 190 STEPHEN R. ANDERSON triggers 'subject'or 'object' agreementmarkers,there is no way for it to reflect number in the third person. Naturally, it remains for us to substantiatethis claim with an explicit analysisbelow, but it is clear that essential aspects of the syntactic account (in particular,the need to have syntactic inversiondemote subjects to indirectobjects before agreement applies) will necessitate a global statement of number agreement something that might be avoided if subjects in inversion constructions remainsubjects. 2.4. An alternativeto the syntacticanalysis In the previoussectionswe have seen that while the syntactictreatmentof inversion offered by Harris provides an elegant account of most of the facts, there are some reasons to believe that an alternative might be preferable. In particular,an analysis which did not involve a syntactic movement rule could avoid several specific difficulties, including the necessity to posit global formulationsof certain rules. Further,since the inversionconstructionnecessitatesa lexical rule linkedto the morphology of inversioncategoriesin at least some frameworks,it is worth askinghow much of the work of describingthe inversionconstructioncan be done in this way. On the basis of such considerations, a preliminarymorphological formulationof inversionwas proposedby Anderson(1982) in the context of a generaldiscussionof inflectionalmorphology.That analysiswas based on the theoryof EXTENDED WORD AND PARADIGM morphology,originating in proposalsof Anderson (1977), and ultimatelyof Matthews(1972). As noted in the introductionto the present paper,.the central notion of that view of inflection is the replacementof specific morphemes(in the sense of minimalunits pairing sound with meaning)by rules relating the form of an inflectedwordto its morphosyntacticrepresentation.The latter is taken to be a complex symbol, containing features indicating the categories of inflectionalmorphologythat are representedby the form in question. In these terms, Anderson (1982) proposed that inversion could be formulatedas an operationon morphosyntacticrepresentationsalone, makingno change whatsoeverin the syntacticstructureof sentences. Such an analysisremovesinversionfrom the syntaxin the usualsense, and treats it as strictlyan aspect of the inflectionalmorphologyof words. Essentialto the morphologicaldescriptionof inversionis the notion that morphosyntacticrepresentationshave internal structure, for it is this structurewhich is to be manipulatedby the proposedrule. This proposal CASE, AGREEMENT AND INVERSION IN GEORGIAN 191 was originallymade by Anderson(1977), in the context of a descriptionof the inflectionalmorphologyof the AlgonquianlanguagePotowatomi.Thislanguage, like Georgian, presents a situation in which verbs must agree with more than one NP (i.e., with both subject and object), and in which the two sets of agreementfeaturesmust be kept distinct (since e.g. 'I saw you' and 'you saw me' are inflecteddifferently).It was suggested that the difference between e.g. subject and object agreementfeatureswithin the morphosyntacticrepresentationof a single verb could be representedby a hierarchicalstructure5,in which one of the sets of featuresis treatedas a unitaryblock co-ordinatedwith membersof anotherset, in a way which is clearly recursive. Suppose, for example, we have a language in which verbs must agree (separately) with both subjects and objects; and in which there are generalizationsacross the two sets which lead us to believe it would be inappropriateto treat this by simply duplicatingthe agreement features (distinguishing a feature [+lsg Subject] from another feature [+lsg Object], for example). In that case, we might assume that there are two partsto the agreementrule:one copyingfeaturesfrom the object onto the verb, and one copying featuresfrom the subject. Assumingfor concreteness' sake that the object agreementrule operatesfirst,once it has applied the verb will be characterizedfor agreementfeatures. When the subject agreement rule now comes to apply, its result is determined by the principle that a rule adding features to a complex already specified for those features does so by creating a new layer of (hierarchical)structure within the morphosyntacticrepresentation.The result could thus be pictured as in (26): (26) [Tense, Aspect, etc.; Subject features [Object features]] Features such as tense and aspect are not affected by the 'layering' involved, since these are specifiedonly once and thus do not come under the domain of the convention in question. For furtherdiscussion of the mechanics of this proposal, see Anderson (1982) and references there. With reference to Georgian, we see that up to 'three layers of hierar5 Hierarchicalorganizationis not, of course, the only possible way to reflect formallythe structurewhichmust be attributedto morphosyntacticrepresentations.Orderedn-tuplesof featuresets, for example,would do as well. Any representationwhichkeeps the varioussets of inflectionalfeaturesapart, and allows for the statementof a subordinationconvention, would do as well. We choose hierarchicalstructuresimplyfor concreteness'sake; it is not clear what sort of evidence would allow us to distinguish among the various formal possibilitiesavailable. 192 STEPHEN R. ANDERSON chical structuremust be recognized and created by the agreement rules, since the subject, direct object, and indirect object can condition agreement on the verb. Let us assume that, in this general case, the outermost layer of structure in the morphosyntacticrepresentationof verbs corresponds basically to the features of the subject (at least in non-invertedforms),and thus is used to triggermarkersfrom the v-series. The innermostlayer, in contrast,can be taken to correspondto the direct object, and to triggerm-seriesmarkers;an intermediatelayercorresponds to an indirectobject, and triggersmarkersfrom the h-, u-, e-, or a- series (dependingon the particularverb). We could then treatthe basic morphosyntacticrepresentationsof class I (transitive)verbs as in (27a) below. Drawingon the analogybetween class I and class III (medial intransitive)verbs in their inflection, we could suggest that the latter have the same, two-layeredstructure,except that the innerlayeris necessarilyspecifiedas an 'inflectionaldummy':that is, as third person singular with no reference. This dummy, note, represents merely a fact about the verb's inflection:since it does not correspondto any subcategorizedargument,it does not representa syntacticdependent of the verb. It also does not correspondto any distinguishableaspect of the logical form of medialverbs (thoughinsofaras membershipin this class is semantically predictable, its appearance is deducible from whatever semantic factors are involved in this prediction). (27) a. Class I: [tense/aspect, Subject [Direct Object]] Class III: [tense/aspect, Subject [ 0 (3sg.)] ] Given this structure;we could then treat inversion as a rule operatingon morphosyntacticrepresentations,which has the effect of extracting the innermostlayerof a two-layeredstructure,adjoiningit as a new outermost layer, and leaving an 'inflectionaldummy'as its trace: (28) [ X person/number[ person/number] ] Inversion: -- 3+[ 1 2 1 2 3 [ 0,3sg Condition:X includes "Series III" or "Class IV" Rule (28), of course, has no effect whatsoeveron the syntacticstructureof clauses in which the affected morphosyntacticrepresentationsappear since it appliesonly to the latter.The resultof applyingit to (27a) will be to convert that structureinto a three layered one like (29): (29) [tense/aspect, Direct Object [ Subject [0, 3sg]]] CASE, AGREEMENT AND INVERSION IN GEORGIAN 193 Notice that now the features agreeing with the direct object are in a position to trigger v-series markers, while those associated with the subject are in a position to trigger 'indirectobject' markers- exactly the correct result for the agreementpropertiesof the inversion construction. Rule (28) can thus describe the basic facts of agreement in Georgian inversion structures without necessitating any alteration of syntactic structure.It will be seen that this providesa potentialway of resolvingthe problem which arose on the syntactic analysis.In order to conclude that this account is adequate,however, we must extend it so that its empirical coverage approaches that of Harris' description. We must therefore expandthe range of verb classes covered by the analysis,and also provide a mechanismof (surface)case markingthat is integratedwith it. It is to these tasks that we now turn. 3. Developingthe morphologicalanalysis In the previous section, a preliminaryaccount of inversionwas sketched that treats this phenomenon as morphological rather than syntactic in character. As the basis of that analysis, a schematic view of the morphosyntactic representationof Georgian verbs was presented, and a concrete proposal was made concerning the morphosyntacticrepresentations of class I and class III verbs. In order to give this account enough substanceto make it a seriouscompetitorwith Harris'syntacticanalysisof inversion, there are a numberof areas in which it must be developed: a) the range of verb classes included must be extended to cover all of those that are distinguishedin the traditionalclassification;b) the assumptions made about the operationof agreementmust be translatedinto an explicit set of agreement rules; and c) an account of case marking must be developed that is appropriately integrated with the description of agreement. 3.1 Extendingthe analysis to otherverbclasses The usual classificationsof Georgian verbs (in e.g. Tschenkeli 1958, Aronson 1982a, and other works listed in the References section below) are based on several distinctions. First, transitive verbs (those taking a direct object) with non-invertedstructureare distinguishedfrom others; following Harris, we refer to these as 'class I'. Secondly, among the intransitiveverbs the medials are distinguishedfrom the non-medialson several grounds:in partthese concern the verb-internalmorphology(e.g., formationof the futurestem of medialswith preradicalvowel i- and stem 194 STEPHEN R. ANDERSON formant-eb, ratherthanwith a preverb),but they also include the fact that medials take their subject in the ergative in Series II tenses while non-medials take nominative subjects in all series; and the fact that medials undergo inversion in Series III while non-medials do not. The non-medialsare treated as class II, while the medials are assignedto class III. Finally, the verbs which show inversionin all tense series are treated separatelyas class IV. The analysissuggested in the previous section concerns only the basic forms of verbs from classes I and III. We now extend it to include verbs from these classes with indirectobjects (section 3.1.1), and verbs of class II (section 3.1.2). In section 3.1.3 we propose a refinement of the formulation of Inversion which recognizes the similaritiesbetween inverted verbs and relative verbs of class II. 3.1.1 Relative verbs In addition to the distinctions among basic classes, there is a further parameterwhich cross-classifieswith them: the abilityof a verb to govern syntactically an (agreeing) indirect object. Since the inversion constructiondoes not allow for agreementwith a NP other than the subject and direct object, this possibilitydoes not arise within class IV; but in classes I, II, and III we must distinguishRELATIVE verbs (those taking a syntactic indirect object in addition to their basic argument(s)) from non-relative forms. Frequently, relative and non-relative variants (i.e., formswith and withoutan indirectobject) exist for the same basic verb; as these differ in their inflection(reflectingthe extra argumentpresentin the relative form), they must be distinguishedin the morphology. In the previous section, we introduced the notion of hierarchically structuredrepresentationsof the agreement features of verbs. Using the abbreviationsFsbj, FDO, and FlO to represent the features of agreement with the subject, the direct object, and the indirect object respectively; 'T/A' to indicate the features of tense and aspect; and 0 to indicate a feature complex appropriatefor agreement with a third person singular argumentbut uncorrelatedwith any NP in the sentence, we recapitulate the proposed treatmentof the agreementstructureof class I and class III verbs in (27) as (30) (30) I: [T/A, FSbj [ FDO]] III: [T/A, FSbj [O]] The morphosyntacticrepresentationsof relative forms of these verbs CASE, AGREEMENT AND INVERSION IN GEORGIAN 195 ought to differfrom (30) by containingan additionallayer of structure;on the assumptionsmade above about the operationof the rules spelling out agreement, the layer showing agreementwith an indirect object ought to be the middle one of three. This yields the following schematicrepresentations for relative class I and III verbs: (31) I: [T/A, FSbj[ FIO[ FDO]]] III: [T/A, Fsbj[ FIo [ 0 ] ] ] From the descriptionof inversionin section 2.1, it will be recalled that inverted verbs cannot show agreement with a syntactic indirect object. Instead, their indirect object appears as the object of the postposition -tvis, in a syntacticallyindependentphrase.There are two approacheswe could take to these facts: either we could assume that inversion,as a side effect, converts an indirectobject to such a phrase;or we could assume that inversion is simply not applicable to relative verbs, and that the relevant forms are supplied from the corresponding non-relative verb together with a syntacticallyindependentphrase. Of these two approaches,it is the formerthat is taken in Harris'analysis. As a generalprincipleof RelationalGrammar,when any rule assignssome relation to an argument, any other argument that (previously) was assigned that relation becomes a 'chomeur'. When Harris' rule of Inversion demotes a subject to indirect object, then, any existing indirect object becomes an indirect object chomeur; such arguments are later markedwith the post-position-tvis. Since the morphological account of inversion being developed here does not involve any re-arrangementof syntactic structure, such an approachis not open to us. Instead,we claim simplythat relativeverbs do not as such undergo inversion. In a sentence such as (32a), which has undergone inversion because its tense is from Series III, the -tvis phrase must thus be treated as a strictly external argument of the verb. The verb in such a sentence is not itself relative. (32) a. turme rezo- s u- cukebi- a samajuricem- tvis apparentlyRezo- DAT 310- gave- 3SBJ bracelet me- for ApparentlyRezo gave me a bracelet. b. rezo- m macuk- a samajuri Rezo- EGR 1SGIO- gave- 3SBJ bracelet Rezo gave me a bracelet. 196 STEPHEN R. ANDERSON But if we thus claim that it is a non-relativeverb which appearsinvertedin (32a), how do we account for its apparentsynonymy(apartfrom tense and aspect) with the relativeform in (32b), which is in the aoristand thus does not show inversion? The problem, of course, is that a relative verb assigns one more semantic role than its non-relative counterpart; and yet that role is apparentlypresentin the interpretationof a sentence like (32a), which we claim has a non-relative verb. The resolution of this difficulty which suggests itself is to argue that the role in questionis assigned not directly by the verb, but indirectlyby the post-position-tvis. The situationis thus completely analogous to that in the English passive, where both the Govemment/Bindingtheory and Lexical FunctionalGrammarclaim that the subject's (agent) semantic (or theta) role is assigned indirectlyby the preposition by; a similarmove is argued for by Marantz(1981). The claim that the post-position-tvis can assigna semanticrole whichis alternativelyassigneddirectlyby a verb to its indirectobject is a necessary one in Georgiangrammarin any event. For one thing, quite independent of the inversion construction,it is often (though not always) possible to paraphrasea relative verb by a non-relative form plus a tvis-phrase (Tschenkeli 1958, p. 383): (33) a. me vu- sxam st'umar-s vino- s I 1SBJ-3IO-pour guest- DAT wine-ACC I pour out wine for the guest. asxamst'umrisa-tvis vino- s b. me vI 1SBJ- pour guestfor wine- ACC I pour out wine for the guest. Secondly, such indirectassignmentof semanticroles is also necessaryin connection with oblique dependentsof non-finiteforms: (34) a. vasli viqide masc'avleblis-tvismisacemad apple 1SBJ- bought teacherfor to-give I bought an apple to give to the teacher. b. sacukarijneli- a anzoris-tvismisacemad gift difficult-is Anzor- for to-give Gifts are difficultto give to Anzor. Sentence (34a) involves a purposeclause, and (34b) a raisedobject. Both CASE, AGREEMENT AND INVERSION IN GEORGIAN 197 constructionsinvolve a non-finiteform which cannot govern an indirect object directly; in such a case, the semanticrole of the indirect object is assigned by the post-position -tvis. These sentences are discussed by Harris (1981, p. 173) as examples of the marking of 'retired' indirect objects, but if one does not derive such non-finite verbs from tensed clauses, the alternativeis simply to recognize -tvis as a potential indirect assigner of the semantic roles which can be assigned directly by finite relative verbs to their indirectobjects. We can conclude from these facts that only non-relativeverbs of classes I and III need undergo inversion(ignoringclass IV for the present)in the presence of Series III tenses, but if that is the case, it must be explained how inversionis blocked in the correspondingrelative forms. We suggest that this follows from an independentlynecessary constraint:if rule (28) above appliedto representationssuch as those in (31), the resultwould be a morphosyntacticrepresentationwith four layers, not three - and such a representationcould not be translatedby the rules that supply the overt markersof morphologicalcategories (recall that agreement distinguishes exactly threecategories of agreeing element). In conformitywith this, we proposethat the grammarof Georgianeffectively filtersout any representations with more than three layers of structure: (35) *[ W[X[ Y[z]]]] This effect of the morphologyis independentlymotivatedby the need to block certain other constructions,such as causativesof relative transitive verbs6, and "version-objects"together with indirect objects (cf. Vogt 6 As noted by Harris(1981, pp. 99f.), Vogt (1971, p. 132), and other authors,occasional formsare cited in the Georgiangrammaticalliteraturethat seem to call for agreementwith four NPs, two of which are apparentlyindirectobjects.These are generallynot found in the standardlanguage,and theirexistence as other than abstractgrammaticalpossibilities(e.g., as causativesof causativesof transitiveverbs) is not at all clear. George Hewitt (personal communication)pointsout a particularlyfascinatingexamplefrom a fairystoryfor children appearingin the Georgiannewspaperfor GeorgiansabroadSamsoblo: aramcda aramceg erl'emalixar- eb- s ar bull- PL- ACC not not and not that ram mo- mik'vlevino- t PVB- 1SGIO-let-kill- 2PL Under no circumstancesare you to let the bulls kill that ram for me. As Hewittnotes, "If 4-place verbs are interpretableby childrenfrom fairy-tales,we should perhaps be careful about dismissing them from practical grammarsof the language." Nonetheless,theirmarginalstatusis clear. We have nothingfurtherto say here about these forms. 198 STEPHEN R. ANDERSON 1971, Harris 1981); it will also have the effect of blocking inversion in relative forms of class I and III verbs. 3.1.2 Verbsof class II We turn now to the non-medial intransitiveverbs, or class II. In their non-relative forms, these are maximallysimple in structure:they agree with only one argument (the subject), and show no trace of any purely formal inflectional element such as the dummy 'object' found with the medial intransitives.There is thus no reason to attributeany structureto their morphosyntacticrepresentationbeyond that in (36): (36) [T/A, FSbj] We have remarkedabove that inversionapplies to (non-relative)verbs of classes I and III but fails to applyto class II; but there is no difficultyin explainingthis fact. The structuraldescriptionof rule (28) above calls for a morphosyntacticrepresentationwith (at least) two layersof structure,and if the structurein (36) is assigned to non-relativeclass II verbs, these will fail to meet that requirement.Theirfailureto invertthusfollows from (36). Tuming to the relative forms of class II, we note that these show agreement with both the subject and the indirect object. On the general assumptionswe have made thus far about the relation between layers of structurein a morphosyntacticrepresentationsand classes of agreement elements, we would like to assign these verbs a representationwith three layers of structure,the middle one of which is occupied by the featuresof the indirectobject. This leaves two layers to be filled; we propose that in these verbs, the subject agreementfeatures are markedin the innermost layer, and the outermostlayer is filled by an inflectionaldummy: (37) [T/A, 0 [ FoI [ Fsbj ]]] As with other relative verbs, these fail to undergoinversionin Series III tenses. Representation(37) immediatelyaccounts for that fact, since rule (28) would convert it into a structureviolating (35). This representationexpresses the similaritiesbetween subjects of class II verbs and objects of transitive verbs (capturedin Harris' analysis by treating class II subjects as underlyingobjects, and having them undergo the 'unaccusative' rule). In particular,many class II forms are passives correspondingto transitiveverbs of class I. On the present analysis,what subjects of class II verbs have in common with direct objects of class I verbs is agreementwith an innermostlayer of morphosyntacticstructure, CASE, AGREEMENT AND INVERSION IN GEORGIAN 199 not the underlyinggrammaticalrelation they bear in their clause. The most unusualfeatureof (37) is undoubtedlythe fact that it contains an obligatory inflectional dummy in its outermost structurallayer. We have already suggested the existence of such dummies in the characterizationwe gave of the inflectionalrepresentationsof medial verbs, but in fact they must be more widely distributedthan this. For example, the assumption that the lexical representationsof some (classes of) verbs require an obligatory dummy morphological element is independently necessary to account for verbs like da-u-k'ravs 'he plays it (an instrument)'which are inflectedas if they took an indirectobject but with which no non-dummyindirect object can appear. A numberof other verbs in Georgian show such purely formal inflection for elements which cannot appearamong their arguments(Tschenkeliet al. 1960-74); the notion of inflectionaldummies in morphosyntacticrepresentationprovides an apparatuswhich is necessaryto describe these facts. 3.1.3 A morphological'unaccusative'rule At this point, we have suggested schematic morphosyntacticrepresentations for Georgian verbs of classes I-III, and also for relative forms. A problemarisesin connectionwith the representationsuggestedfor relative verbs of class II, however. If we continue to assumethat the normalverb agreementrule insertsa subject markerin accord with the featureson the outermostlayerof structure(as in all other classesof verbs consideredthus far), this will fail to give the correctresultsfor relativeclass II verbs if their subjectsfeaturesare indeed located on the innermostlayerof a three-layer structure.The obvious alternative,thatof treatingrelativeclass II verbs as having the structure in (38), is unsatisfactorybecause this is the same representationassignedto relativeverbs of class III, and the two sets must be kept distinct for inflectionaland case-markingpurposes. (38) [FSbj [FIO [ 0 ] ] ] The other possibilityis to assumethat relative verbs of class II have the morphosyntacticrepresentation(37) at the time they are inserted into a structure,and that it is this representationthat is relevant to the assignment of case (see section 3.3 below); but that internal to the set of morphologicalrules which develop the actual surfaceform of an inflected verb, these representationsundergo a morphologicalrestructuringso that their subject featuresappearin the correct position. Such a rule would be the morphologicalanalog of Harris''Unaccusative': 200 (39) STEPHEN R. ANDERSON 'Unaccusative' [ X 0 [ Y Z] ] ] 12 3 4 1 4 [3 [0]]] Rule (39) substitutes the content of an innermost layer of a threelayeredstructurefor an inflectionaldummyin the outermostlayer,leaving a dummy as its trace. The effect is to cause specified features in such a representationto triggerv-series affixes,insteadof m-seriesones. Clearly, this rule is directly analogous to the Unaccusative rule posited in Harris' analysis; but with the difference that, since it affects only the morphosyntacticrepresentationof words, it does not entail any movement or other alterationin syntactic structure. But now we can observe that rule (39) performs a change which duplicates part of rule (28) (Inversion):the transferof features from an innermostlayer of structureto the outermostlayer. We can thus remove that part of the change from (28), if the latter rule simply inserts an additional (initially empty) layer of structure in inversion forms. The resulting simplified rule still applies, like the original morphological Inversionrule (28), in that part of the syntaxwhich is responsiblefor the construction of the morphosyntactic representation relevant to case markingand agreement. (40) SeriesIII Inversion: [ 1 1 X[ Y] ] 2 3 0 [ 2 [ 3]]] Rule (40) must apply in the syntax,and feeds rule (39), which appliesto morphosyntacticrepresentationsin the course of the development of inflected word forms in the phonology. The latter rule will thus perform the required restructuringon inversion forms as well as on the relative class II forms for which it was originallymotivated. The Inversionrule (40) operateson the morphosyntacticrepresentation of a verb, and has the consequence that a 'dummy'layer of agreementis presentin the outermostlayer of structurewhen the verb belongs to Class IV or is in a series III tense. We will see in sections 3.2 and 3.3 that this dummy plays an important role in the processes of case marking and agreementin invertedclauses, but there is still somethingunsettlingabout rule (40). As it stands, it compromisessomewhatour claim that Inversion should be treated as a fact about inflectionalmorphology,rather than a syntactic process, since rule (40) crucially applies in the syntax. An account that avoided this result would be preferable,but before develo- CASE, AGREEMENT AND INVERSION IN GEORGIAN 201 ping such an alternative (in section 3.4 below), we must first detail our assumptionsabout the operationof case markingand agreement. 3.2 The descriptionof Georgianverbagreement At this point, we have presented a picture of Georgian inflectional morphology that extends to most of the productive verb classes of the language.It is necessarynow to providea more explicitaccountof the way in which the necessary morphosyntacticrepresentationsunderlyingthis morphologyare developed. We assume,in line with Anderson (1982), that there are two aspects to this issue. On the one hand, a morphosyntacticrepresentationof inflected words is developed within the syntax, without direct reference to the peculiaritiesof individuallexical items that might eventuallybe inflected in accord with it. Within the lexicon, on the other hand, individualitems are providedwith some inflectionalspecificationsthat indicate their local idiosyncrasies.At a minimum,for instance, lexical items in virtually all languages are characterizedas [+Noun], [+Verb], etc. (or perhaps for some other set of features which make the same divisions among major word class). Individualstems within the lexicon may be more narrowlyspecified: thus, for example,the stem thoughtin English is characterizedspecifically as [+Verb, +Past], while think-issimply[+Verb]. We presumethat lexical stems are grouped together into paradigms of items differing only in inflectionalproperties.When a lexical item comes to be inserted into a phrase marker,it is associated with a particularposition; this position is identifiedinflectionallyby the morphosyntacticrepresentationdeveloped for it in the syntax.The stem that is insertedin such a position is then that memberof the relevantparadigmwhich is most specificallycharacterized, consistent with the morphosyntacticrepresentationof the position in question. For concreteness' sake, we assume here that the overall grammarin which this account is embedded follows the lines of the Government/Binding theory, though an analogous picture could be developed without significant alteration in other theories such as that of Lexical Functional Grammar. Consistent with this framework, and with the morphological assumptionsjust outlined (developed in more detail by Anderson (1982)), we posit a set of (syntactic) agreement rules that operate in the syntax to construct an agreement marker in INFL. This markerreflects the propertiesof a particularstructure;since INFL will 202 STEPHEN R. ANDERSON eventually be attached to the main Verb of the clause, an inflected verb which is inserted into such a structure must itself have an inflectional representation which is non-distinct from that in INFL. The lexical representations of individual Georgian verb stems characterize their inflectionalpropertiesin terms of how many layers of structurethey are inflected for, and whether any of these layers are requiredto be null. We assumethat Georgianhas (at least) enough configurationalstructure to allow subject, direct and indirect object to be distinguished,either directly in phrase structure representationsor indirectly in functional structure (on the assumptionsof Bresnan 1982). As we noted above in section 2.3.3, this is not totally uncontroversial;it is contrary to the proposal of Hammond (1981), for example, though consistent with observed word-ordergeneralizations.In any event, it is clear that verbs must be able to be subcategorized for direct and indirect objects independently, and we assume that whatever structural difference this reflects is also visible to the agreement rules. Within this overall picture, we propose that a set of rules operates to copy inflectionallyrelevantfeaturesfrom the argumentsof a verb in order to constructthe representationof agreementin INFL.This representation is initially characterized only for tense and aspect (in terms of an appropriatefeature system, in particularone that recognizes the division of Georgian 'screeves' or tense/aspect categories into three series). We then formulatethe following agreement rules: (41) a. (obligatory)Copy referentialindexand person/numberfeatures from Direct Object or if there is no Direct Object, copy index and features from the Subject b. (optional) Copy referentialindex and person/numberfeatures from Indirect Object c. (optional) Copy referentialindex and person/numberfeatures from Subject d. (optional)Add null reference and 3sg features Presumably,rules such as those in (41) should be formulatedwithin an appropriategeneral notation for agreement processes. In the absence of such a theory, we simply state them in ordinary language, but some general properties of agreement which we clarify below are implicit in them. We assumethat such a set of rules constitutesa scHEMAin that they interact as a unit with other rules, and that there is a significantinternal CASE, AGREEMENT AND INVERSION IN GEORGIAN 203 organizationto the set; though again, in the absence of a comprehensive theory of agreement,it is difficultto substantiatethese assumptionshere. Besides these generalproperties,a numberof remarksare calledfor on the details of this schema. 1) As already remarked in section 2.4, we assume that when an agreement rule adds features and an index to a complex symbol that already contains others, the previous materialis organizedinto an inner layer of structureby general convention (as proposedin Anderson 1977a, 1982). As a result, (41) constructsmulti-layeredstructureswith respect to the agreementfeatures.The tense/aspectfeaturesremainin the outermost layer, since no additionalfeatures from this group are assigned beyond those generatedin the base. Naturally,if (41) were to lead to a structure with four layers,such a representationwould be blocked by (35). Thus, at least one of the four sub-rules in (41) must fail to apply in any given representationif a well-formedstructureis to be obtained. 2) [+Plural]is only copied from non-subjectsin the presence of firstor second person. This could be accomplishedin at least two ways: either a rule might make all third person non-subjects [-Plural]; or else the constraintcould be built into (41) directly.Largelyfor aesthetic reasons, we preferto assumethat non-subjectthirdpersonNPs are all inflectionally [-Plural] by rule (or convention), though of course they may be semantically (and internally)plural. In any event, some version of this generalizationmustbe built into any analysis(as noted in section 2.3.4), since the same plural NP (e.g. bavsv-eb-s 'child-plural-dative')will show plural agreement if it is a subject (in an inversion construction),but singular agreementif a non-subject. 3) Finally, we assume that any of the rules in (41) may introduce an empty referentialindex with 3sg agreement features (i.e., an inflectional dummy) instead of copying. Note that this operation (equivalent to applyingrule (41d) instead)will in general only yield well-formedresultsif the grammaticalposition referred to by the sub-rule in question is not filled; as we will see in the next section, if an NP in an argumentposition does not trigger agreement, it will fail to be assigned case, violating a general requirement.With regard to the subject, either rule (41a) or rule (41c) can effect the agreement necessaryto allow case-assignmentto this position.The possibilityof insertinginflectionaldummiesinsteadof actual agreementmaterialdoes not representa syntacticfact aboutthe language, but rather reflects the fact that a number of different inflectionaltypes exist for the verbs in the lexicon, correspondingto the variousclasses we have discussedabove. 204 STEPHEN R. ANDERSON Though we will not demonstratethis fact in detail here, it should be clear that the rules in (41) allow us to produce morphosyntacticrepresentations correspondingto all of the inflectional classes we have posited above. Exampleswill be providedbelow of lexical entriesthat correspond to each of the possibilitiesallowed within this schema. 3.3 Case marking We mustnow providean accountof the assignmentof case to the NPs that serve as argumentsof Georgian verbs. It will be recalled from sections 1 and 2 that case marking,like agreement,varies from one constructionto another;one of the advantagesof Harris'syntacticanalysisof inversionis the fact that it provides a direct account of the differencesbetween case marking in inversion and non-inversionstructures.We must thus show that the morphologicaldescriptioncan do as well. Implicitin the analysisdeveloped thus far is the claim that case marking is effected by a set of rules which take into account the morphosyntactic representationin INFL, which includes the agreement representation developed by the rules in section 3.2. At least some reference to the content of INFL is a necessary assumption in any event, since case marking of subjects and objects is sensitive to the tense series of the clause. As apart of the process of agreement, not only are inflectional features copied from NP into INFL, but also a relation is established between the referentialindexof such NPs and this element. Let us call this relation 'co-superscripting',and assume that it forms a subcase of the relationof PROPER GOVERNMENT (cf. Chomsky 1981 and related work). Using the ad hoc notation '/pg' to mean 'when properly governed by', we can formulatethe case-markingrules as follows: (42) a. NPi [+Ergative] /pg [+Series II, i [X]] b. NPi [+Dative] /pgI X[ i [ Y ]] ] c. NPi [+Accus.] /pg [-Series II, j [ i]] /pg [-Series II, j [ X[ i]]] /pg [+V] or d. NPi --)[+Nominative] Rule (42a) assigns ergative case to NPs that are co-super-scriptedwith the outer layerof a representationwhich contains(at least) two layers,and which bears a tense from Series II. The rule thus appliesto the subjectsof Class I and Class III verbs in these tenses; note that it does not affect the subject of a class II verb, since either a) the verb is non-relative,and thus has only one layer, or b) the featuresof such a subject NP with a relative CASE, AGREEMENT AND INVERSION IN GEORGIAN 205 class II verb only come to be in the outer layerof structurein the course of the operationof the rulesdeveloping the inflectionalform of verbs within the phonology. At the point the case markingrules apply (essentially,at the end of the syntax),these features are still in the innermostlayer of the morphosyntacticrepresentation. Rule (42b) assignsdative case to a NP which is co-super-scriptedwith the middlelayerof a three-layerrepresentation.This class includesexactly the indirect objects of relative verbs (of any class), and the subjects of inversionverbs (whichcome to occupy the middle layer as a result of the operation of rule (40)). Rule (42c) assignsaccusativecase (syncretic,it will be recalled,with the dative).Thisrule appliesonly in tenses other thanthose of Series II (in fact, only in Series I, thoughthis fact does not have to be separatelynoted), and affects NPs that are co-superscriptedwith a layer of the morphosyntactic representationwhich is not the outermost.This rule furtherrequiresthat the outermost layer be co-superscripted with some argument; this prevents the assignmentof accusative to the subjects of relative class II verbs and the direct objects of 'inverted'verbs. The two sub-casesof this rule could be unified as '/pg [-Series II, j[(X)[i]]]' under some convention to the effect thatwheneveran (X) option is not taken,the vacuous layer of bracketsdisappears. Rule (42d) is the 'elsewhere'case of the schema. It assigns nominative case to NPs that are not assigned some other case, but which are co-superscriptedwithsome layer of morphosyntacticagreementstructure. this includes: a) subjects of Series I verbs, regardlessof their class; b) subjects of non-inverted(i.e., class II) verbs in Series III; c) direct objects of Series II verbs, which cannot be assigned accusative because of the restrictionin rule (42c); and d) direct objects of (inverted)Series III or class IV verbs. In order for this rule to apply correctly, we assume a convention of disjunctiveorderingthat appliesto the schema:withrespect to a given NP, the rulesare appliedin sequence, and once a given rule has applied no furtherrules are examined. This mechanism(or somethinglike it) is a necessarypartof the grammar of Georgian, since the occurring differences of overt (or 'surface')case must be describedin some way. Given it, however, no additionaldevice of 'abstract'Case (cf. Chomsky 1981) is necessary:we need only say that every overt NP must be assigned a surface case. As we remarkedin the preceding section, it is this requirementthat ensuresthat every argument of a verb will be markedin its agreementrepresentation,since if some NP were not so marked,rule (42) could not apply to assign it a surface case, 206 STEPHEN R. ANDERSON and the 'case filter'just alluded to would block the structure.This is one instance (among others; cf. Anderson 1984b for a rather different example) in which the apparatusnecessary to account for the surface distribution of elements in a rich inflectional (case marking) system obviates an appealto the rathermore abstractnotion Chomskymakes use of for English and other languages with minimalovert markingof NPs. Further investigation is clearly warranted of the difference between languages in which the notion of abstractCase is motivated and those in which it can be dispensedwith. 3.4 A refinementof the Inversionrule Returningto the Inversionrule (40), we can note that it operates on the morphosyntacticrepresentation in INFL, and that its consequence is effectively to render 'dummy agreement' in the outermost layer of structure(rule (41d) above) obligatorywhen the verb belongs to Class IV or is in a series III tense. Thus far we have said nothing about the morphosyntacticrepresentations of class IV verbs. We know that they must undergo inversion, which we now see consists in the addition of an outer-layerinflectional dummyto a representationwhich is otherwisethat of a class I or III verb. We could ensureexactly this effect if we were to representClass IV verbs lexically with an inflectionalstructurecontaining a dummy outer layer: [0 [-{-]]] (where'0' indicatesa specifieddummyand '-' indicatesa position whose content is not furtherspecified). A verb with such an inflectional representationcould not be inserted unless (41d) had applied. This is preciselythe same as the representationgiven to relativeverbs of class II (reflectingthe fact that both are inflected precisely with 'subject' and 'indirectobject' markers);the two differ in that relative class II verbs are subcategorizedto take an indirect object, while class IV verbs are subcategorizedfor a direct object. As a consequence of this difference in argumentstructure,and its effect on the operationof rule (41), the same inflectionalmaterialwill correspondto very differentpositionsin syntactic structure. This difference, however, falls out automatically from the system developed above. But now the same treatment can be extended to Series III forms, eliminatingaltogetherthe need for a distinctInversionrule in the syntax. Since the Perfect stem is formallyidiosyncraticto a certainextent, it must in any case be entered in the lexicon. We want the Perfect stem to correspond to an 'inverted' representation:i.e., one with an additional CASE, AGREEMENT AND INVERSION IN GEORGIAN 207 outer-layerinflectionaldummysuperimposedon the representationof a class I or III verb. We can accomplishthis by including,in the lexical rule which forms the Perfect stem, the following systematicrelation between the inflectionalrepresentationsof the Perfect and non-Perfect stem: (43) [+V Xt Y]I 4? [+V, +Series III, 0 [ X [ Y ] ] ] Recall that on the syntactic analysis of inversion, a lexical rule was requiredin order to specify the redistributionof semantic roles between inverted and non-inverted structures (at least on the assumptions of Government/Bindingtheory or of Lexical Functional Grammar).This function of the rule is not necessaryon the analysisdeveloped here, since semanticroles are in general associatedwith a constantsyntacticposition. All we need say is that Perfect stems differ in their inflectionalrepresentation from the other stems of a given verb in the way specifiedin (43). Of course, we must also include the fact that a semanticrole which would be assignedto an indirectobject of a relative verb can be indirectlyassigned by the postposition -tvis in association with the Perfect stem of a correspondingnon-relativeverb. We must also include whatever informationcan be systematizedabout the formal characteristicsof the Perfect stem; for instance, the fact that the Perfect stem involves replacementof any pre-radicalvowel by i- (and the additionof this vowel to stems that otherwisedo not have a pre-radical vowel). As an example, the verb (da=)a+nt+eb 'light (a fire)', with subcategorizationframe[ DO] and inflectionalrepresentation[-[-] ] (i.e., a member of class I), has a Perfect stem da=i+nt+i with the same subcategorization frame, but the inflectional representation [0[-[-]]]. As advocated at the end of section 3.1.3, we have thus eliminated 'inversion'as a distinct rule in either the syntax or the morphology:what remains of it is the fact that 'inversion' structures are inflectionally characterizedby a three-layermorphosyntacticrepresentation,the outer layer of which is an inflectional dummy. This comes about either as a lexical property of the verb as a whole (for 'class IV' verbs), or of its Perfect stem (which is used in all series III tenses). Inversionverbs share this propertywith relative verbs of class II, as well as the applicabilityof rule (39), Unaccusative,in the course of the phonologicaldevelopmentof the surface form of such verbs. 3.5 Residual issues Before concluding, there are some points which require additionaldis- 208 STEPHEN R. ANDERSON cussion. Among these are the phenomenonof object camouflage,the way in which the several partsof the morphologicalanalysisinteractwith each other, the resolution of problems raised by the syntactic analysis, and finally an illustrationof the lexical characteristicsattributedto various sorts of verbs under the present analysis. 3.5.1 ObjectCamouflage We noted in section 1 that there is one set of circumstancesunderwhich the morphologyof verb agreementin Georgian is unable to manifestthe relationbetween a verb and all of its arguments.This problemariseswhen the verb has a non-thirdperson direct object (calling for a non-0 marker from the m-series), and also an indirect object (calling for a non-zero marker from one of the four indirect object series). The facts are illustratedby the sentences in (44): agivi- s (44) a. vano anzor- s dareb- s Vano Anzor- ACC 3IO- compare-3SBJ Givi- DAT Vano is comparing Anzor to Givi. ab.*vano (sen) gdareb- s givi- s Vanoyou 2DO- 3IO- compare-3SBJ Givi- DAT (Vano is comparingyou to Givi.) adareb- s givi- s c. vano sens tav- s Vano your self- ACC 3IO- compare-3SBJ Givi- DAT Vano is comparingyou to Givi. Sentence (44a) shows that the verb (se=) a-dareb'compare'takes both a direct and an indirectobject; but from the ungrammaticalityof (44b) we can see thatwhen the indirectobject is present,the directobject cannotbe second person. Parallel examples would show the same thing for first persondirectobjects. Instead,the constructionthat is used is that in (44c): the expressionsens tavs (ceemstavs for first person singular,and similar forms for first or second person plural), normally a reflexive pronoun, substitutes for the impossible form. Since these reflexives, although referring to second (or first) person are grammaticallythird person, the constraintis not violated. In the courseof her analysis,Harris(1981, pp. 48ff.) proposesto account for these facts by generatingfirstand second person direct objects freely, but then converting them to possessive pronoun+ tavi phrases if the CASE, AGREEMENT AND INVERSION IN GEORGIAN 209 clause contains an indirect object. Within a theory (such as Government/Bindingor Lexical Functional Grammar)that disallows structurebuilding rules, pronominalizationtransformations,and the like, such an analysiswould be excluded. The only availablealternativeis to generate the possessive+ tavi phrasesdirectly, and then account separatelyfor a) the absence of normal first and second person direct objects in the presence of an indirectobject; and b) the non-reflexiveinterpretationof the possessive+ tavi phrase. In fact, the grammarmust generate possessive+ tavi phrases in any event, to wit, the tav-reflexiveswe referredto in earlier sections of this paper. The problem is that these are normallyinterpreted as reflexive anaphors, and must be c-commanded by a coreferentialsubject within their clause. This latter property(on the assumptionsof Chomsky 1981, for example)is a consequenceof their lexical statusas [+Anaphor];but in the object camouflageconstruction(44c), no such antecedent is present (or necessary). What we need to say, then, is that first and second person anaphorsare (exceptionally)[-Anaphor] underthe conditionsof the object camouflage construction. Since possessive+ tavi phrases are not possible with [-Anaphor] interpretationoutside of these circumstances,we propose that a special rule of interpretationis involved: (45) +m possessive+ tavi ]i /pg[ X [ Y [ i]]] .+you Optionally , [-Anaphor] (where X, Y are non-null) [NP Although this rule allows us to generate the correct interpretationfor (44c), it does not yet account for the ungrammaticalityof (44b). The reasonfor the absenceof such sentences is apparent:they directlyreflecta limitation of the expressive possibilities of the morphology of verb agreement, since a non-null m-series marker and an indirect object markerwould compete for the same 'slot' in the morphologyof the verb form. The situationis thus the same as for the filter (35), which prevents verbs inflectedfor more than three arguments:there is simplynot enough formalapparatusavailableto accommodatesuch cases. The grammarthus behaves as if the following filter applied to morphosyntacticrepresentations: (46) +me 210 STEPHEN R. ANDERSON Ideally,we shouldbe able to derive both (35) and (46) as theoremsfrom general constraints on the operation of the rules of morphological expression;but since the purposeof this paper is to examine the syntactic side of agreementmorphologyratherthan its phonological side, we leave open the issue of how this should be done. This subsection is intended simply to show that the facts of object camouflage can be incorporated into our a7nalysiswithout excessive difficulty. It might be objected that we have decomposed a single phenomenon into two separate,complementaryparts: rule (45), and the constraintin (46). This does not seem an implausiblemove, however,since we maintainthat there are indeed two aspects to the phenomenon:a) the impossibility of certain verb forms, and b) a special interpretationof otherwise-reflexiveexpressionswhich allows the missing meanings to be expressed. Only by decomposingthe facts in the above way does it seem possible, for instance, to look for an explanationof the constraintin (46) along the lines suggested in the preceding paragraph. 3.5.2 Interactionof agreement,case marking, and inversion The account presented above involves several distinct components:a) a set of case-markingrules; b) a set of agreement rules; c) a set of lexical representations,includinglexical rules which express systematicrelations among these; and d) a set of ruleswhich formallyrealize the verbal forms correspondingto the morphosyntacticrepresentationsof the inflectional categories of words (in particular,verbal agreement morphology). It is necessaryto say a few wordsaboutthe waysin which these elementsof the analysisinteract. We note firstthat the case markingrules (42) cannot (intrinsically)apply until after the creation of an indexed morpho-syntacticrepresentationin INFLby meansof the Agreementrules(41). Since the only syntacticreflex of 'inversion' is the necessity to apply agreement rule (41d) in order to insert an 'inverted' verb (Class IV or Series III) from the lexicon, the relative orderingof 'Inversion'and Case markingthus follows automatically. Within the phonology, surface forms of words are developed. A preliminarystep in this process is the re-arrangementof morphosyntactic representations performed by Unaccusative (39); this is followed by (morphological)verb agreement marking. Morphological agreement is the schema (still withinthe phonology)which introducesv-series markers in agreement with the outermost layer of morphosyntacticstructure, CASE, AGREEMENT AND INVERSION IN GEORGIAN 211 h-,u-,e- or a-series markersin agreementwith a middle layer (second of three), and m-series markersin agreementwith the innermostof two or three layers. Since the rules of Case markingapply in the syntax,they are naturallyinsensitive to the effects of Unaccusative (which applies in the phonology). 3.5.3 Resolutionof the problemspresentedby a syntacticview It is worth reviewing briefly the advantages presented by the morphological view of inversion, agreement, and related phenomena advocated here in comparison to the syntactic analysis of the same phenomena.A numberof problemsarose on that view, it will be recalled, which related to the central claim of the syntactic inversion analysisthat the surface grammaticalrelationsof inversionclauses were differentfrom the underlying grammaticalrelations. Since the morphological analysis does not posit any such reorganization of grammatical relations in inversion clauses, such problemsas those of improperlybound traces and the statement of word order and rule interactionswithout recourse to global formulationsobviouslydo not arise on this account. Similarly,the fact that non-finite forms of otherwise class IV verbs do not show inversion follows directly from the treatment of inversion as a part of agreement marking(i.e., of finite inflection). The case of number agreement in inversion clauses is a particularly interestingone. Recall that on the syntacticanalysis,a global formulation of this rule was inevitable:agreementcannot operate until after Inversion (and Unaccusative);but at this point it is necessaryto say that thirdperson indirect objects trigger plural agreement only if they were originally subjects. On the morphological analysis, no such global reference is necessary: the Agreement rule (41) copies the morphologicalfeature of number in third person NPs if and only if they are subjects. This can be determined on a strictly local basis, and refers only to the grammatical relations borne by an NP in s-Structure. Subsequently,within the phonological component, Unaccusative (39) may have the effect of relocating subject agreement features which are not on an outermost layer of structure;indeed, the only way the feature [+Plural] can appearin conjunctionwith [3rd Person]on a non-outermost layer is by first being placed in an outer layer by subject agreement,and then having a new outer layer created by dummy-agreement.None of these processes require reference to more than one representationat a time, however. The limited circumstancesin which plural markingcor- 212 STEPHEN R. ANDERSON relates with a thirdperson markerfrom an indirectobject series are thus describedwithoutthe use of global rules;indeed, there is no possible role for globality to play in this phenomenon, since no structurallydistinct syntactic representationsare involved in the analysis. TABLE III Sample lexical entries for verbs of various types Subcateg. Inflection [_ [ [ [ [ [-[-] Class I (transitive verbs): (da=)c'er 'write' da=i+c'er+i [perfect stem] (no=)p'ar+av 'steal s.t. from s.o.' (da=)i+mat'+eb 'add s.t. to s.t.' (da=)i+k'r+av 'play (an instrument)' DO] DO] [+Ser. III 0[-[-D] DO, IO] DO, 10] DO] [-[- [-m [-[- [-m [-[0 [H Class II (non-medial intransitives): (mo=)xd+eb 'happen, occur' (da=)i+c'er+eb 'be written' (0=)e+lod+eb 'wait for s.o., s.t.' (mi=)e+c'er+eb 'be written to s.o.' [ ] [.] [ 10] [ 10] [-] [-] [0 [4-]] [0[-[-] [-] [] [ IO] [_] [ IO] [] [-[0D] [+Ser. III 0 [-[0]]] [-[-[0] [-[0]] Class III (medial intransitives): qep 'bark' i+qepn+i [perfect stem] (da=)i+qep 'bark at s.o.' i+cin+i 'laugh' (da=)cin+i 'laugh at s.o.' c vim 'rain' [-[-[0]D [0 [0]] Class IV ("Indirect" verbs) i+qvar 'love' a+k'l+i 'lack' gon+i 'think, seem [ [ [ jin+av 'sleep' [-] DO] DO] (DO)] [0 [-[-]] [0 [-[-]] [0 [-[-]1 [0 [-[0]] CASE, AGREEMENT AND INVERSION IN GEORGIAN 213 3.5.4 Examplesof lexical representations Recall that in order for a specificverb to be lexicallyinsertedinto a given syntacticstructure,it is necessarythat the structurein which it is to appear a) satisfy its subcategorization requirements, and b) contain a morphosyntacticcomplex symbol that is consistentwith its lexically assigned inflectionalrepresentation.The various classes of verbs distinguishedin the traditionalschema differin respectof these properties;however, there are also differencesamong verbs withinthe same class. Examplesof some lexical specificationsare given in table III above. Note in particularthat althoughwe have only provideda smallnumberof Perfect stem entries,as examples,this does not implythat other verbs do not have such additional stems in theirlexical representations.Quite independentof the questionof such additional stems characterizedfor a particulartense or series of tenses, most verb roots in Georgianare capableof serving as the bases of a large number of somewhat idiosyncratic formations, and a glance at Tschenkeliet al. (1960-74) will quicklyconvince the readerof how much more there is to be said about any Georgian verb. 4. CONCLUSIONS In the precedingsections, we have developed the principalcomponentsof an analysis of verbal agreement, case-marking, inversion, and related phenomena in Georgian. The most importantaspect of this treatment throughout has been the extent to which it is a description of the .morphologyof the language, employing mechanisms necessary for the description of inflectional morphology rather than the mechanisms of syntax. In particular,no alterationsof syntactic structureare posited to account for the inversion construction (or for the difference between Series I and Series II tenses). To the extent to which this account is well-motivated, it establishes our basic descriptive points: 'Inversion'in Georgian is a fact about morphology,not a rule of the syntax. While the morphologicalanalysisof this paper differs in fundamental ways from the syntactic treatmentof Harris(1981, 1982, 1983), there are also many obvious similarities.For example, both analysesmake use of a rule of 'Unaccusative',with somewhatsimilarfunctions.Both differentiate the subjectsof class II and class III verbs by treatingthe formeras having similaritiesto direct objects of other verbs, where the subjects of class III verbs are similarto transitivesubjects.Otherparallelscould be added;the major difference remains the fact that the present analysis locates these 214 STEPHEN R. ANDERSON parts of the grammarof Georgian in the inflectionalmorphologyof the language, while Harrislocates them in syntactic rules and structures. This differenceis hardlysurprising.The theoryof RelationalGrammar, within which her analysis is formulated, cannot be said to contain an explicit theory of morphology, in the sense of a set of systematic mechanismsfor relatingthe inflectionalpropertiesof wordsand their role in a syntactic structure to their surface form. Once we provide a morphological frameworkin which to discuss these issues we see that Harris' insights about the structure of the language are fundamentally correct, and can be maintainedin largerpart, but that they pertainto the morphologyratherthan to the syntax. Beyond the basic descriptiveissue of how to treat inversion, however, the analysisabove justifiesa certain numberof broaderconclusions, not limited to Georgian.We take up some of these below, in increasingorder of their generality. 4.1 Class markers In the analysisdefended here, the several classes of Georgian verbs are differentiatedlexically in terms of independentlymotivated morphosyntactic (or inflectional) representations,rather than by arbitrary'conjugation-class'markers.For example, a 'class IV' diacritic is no longer necessary,since such verbs are uniquelyidentifiedby the combinationof subcategorizationrequirementsand inflectionalpropertiesfor which they are lexicallyspecified.In fact, all of the variousGeorgianverb classes are now uniquelyspecified in this way, and no separate arbitrarymarkersof conjugation class membershipare necessaryat all in the lexicon. On Harris'analysis,she argues (1981, pp. 228ff.) that conjugationclass markersare similarlyunnecessary,though they are used throughouther descriptionas a purelyexpositoryconvenience. The inflectionalpatternof a given verb is supposed to follow from the set of argumentsit takes, together with the locations of these in (underlyingthe surface) syntactic structure.While this is surely true (given her syntactic assumptions)for membershipin classes 1, 2, and 3, the fact that a verb belongs to class 4 is not similarlydeducible from the pattern of its arguments:rather, this is relatedto the fact that it undergoesinversion.The morphologicalproperty of class 4 inflection must thus be accessible to the syntactic rule of Inversion,or else the specificderivationalhistoryof the structurein which a verb appearsmustbe accessible to the rules of inflection.Either of these is a varietyof interactionbetweenword formationprocesses and syntactic CASE, AGREEMENT AND INVERSION IN GEORGIAN 215 structure which a restrictive theory would like to exclude; they are unnecessaryon an analysislike the present one, in which the locality of reference of inflectionalpropertiesis preserved. Naturally, the elimination of arbitrarymarkers of inflectional class membership(other than structurallymotivated aspects of the form of wordsbelongingto such classes) is defendedhere only for Georgian;but it remainsto be seen whether,in other languageswhere arbitrarypartitions of the lexicon are generally posited to account for differences in inflectionalproperties,it may not be possibleto reduce this arbitrarinessin light of independentlymotivated propertiesof lexical entries. For example, in those Romance languages where verbs are traditionallyorganized into arbitraryclasses ('first conjugation', 'second conjugation', etc.), it seems quite likely that representationsof verb stems provided with a thematic vowel in the lexicon could eliminate the need for such an unilluminatingdivision (cf. Platt 1981 for one such attempt). The point is a moderatelysubtle one, since it could be maintainedthat the theme vowels simply act as diacriticson such an analysisof Romance verb classes. It can be argued,however,that the morphologymust contain some principleto insertthese vowels in any event, and that prohibitingany subdivisionof verb stems other thanone which is substantivelymotivated in such a way yields a more restrictive theory than one which allows explicit diacriticsunrelatedto any unitarystructuralcharacteristicof the forms they categorize. Naturally, the validity of this suggestion is impossible to assessclearlyuntil it has been exploredin more concrete detail in actual analyses. 4.2 The natureof agreement The rules above make crucial use not only of agreement in inflectional features between a verb and its arguments,but also of (some sort of) 'co-indexing' relation between the morpho-syntacticrepresentationsin INFL, which eventuallydetermineagreementin verbs, and the NPs they agree with. Such co-indexing is familiarin the case of subject agreement, and in fact plays a crucial role in the Government/Bindingtheory in governing the subject position in finite clauses; it is less obvious for non-subjects. Stowell (1981), however, has proposed that verbs are co-indexed with all of their subcategorizedargumentsin order to yield a unified definition of 'proper government'. In the case of Georgian, one prediction which results from this is that all positions reflected in agreement are automaticallyproperly governed, from which it follows 216 STEPHEN R. ANDERSON (correctly) that such positions may be unfilled phonetically. One could express this in current idiom by saying that Georgian is a generalized 'Pro-drop'language. A further instance in which coindexing between non-subjects and agreementis necessaryis cited in Anderson(1974). In the Abkhaz-Abaza languages, a verb-initial agreement marker /y/ (marking third person plural or neuter singular intransitive subjects or transitive objects) is deleted if and only if the verb is immediatelypreceded in linear order by the NP with which it is co-indexed. In fact, when one considersinflectional systems of even moderatecomplexity,the plausibilityof the position that all agreeing arguments(and not only the subject) are co-indexed in the verb is considerable. The relation of co-indexing that is involved, however, is clearly differentfrom that obtainingbetween freely occurringNPs. In sentences containinga reflexivephrase,it is clear that the agreementelement related to the subject must be kept separate from that relating to the object, despite the fact that the 'reference'of the two is the same. If the subject is first or second person, for example, the reflexive will have different agreement features (namely, third person singular) from those of its antecedent, and the two must be kept separatedin the morpho-syntactic representationof the verb. It is for such reasonsthat we have referredto 'co-superscripting'rather than 'co-indexing' as the relation obtaining between an argument and the morphosyntactic representation of agreementin the clause. 4.3 The natureof morphosyntactic representations Finally,we conclude that analysessuch as that presentedhere validatethe notion of a morphosyntacticrepresentationwith significantinternalstructure, and of rules which create and manipulate such representations withouteffecting other syntactic structure.As long as one confines one's attention to languages with relatively simple inflection, it is possible to sweep inflectional morphology under the rug to a considerable extent, assuming that the formal categories of inflected words bear a rather straightforwardrelation to the syntactic structuresin which they appear. Just as it would be extremelydangerousto generalizeabout the theoryof segmentalphonology on the basis solely of analysesof Chinese (however extensive), though, it is unlikely that an adequate picture of inflectional morphology can be derived from the study of lanauages like English, German, and French in which this aspect of grammaticalstructure is CASE, AGREEMENT AND INVERSION IN GEORGIAN 217 relatively impoverished. A closer study of a language like Georgian suggests richer possibilitiesfor the apparatusdescribing the traditional domain of inflection- possibilitieswhich might be productivelyexplored even in languages like English, where the data are not rich enough to motivate them by themselves. REFERENCES Anderson,StephenR.: 1974, 'On Dis-AgreementRules', LinguisticInquiry5, 445-451. : 1976, 'On the Notion of Subjectin Ergative Languages',in C. Li (ed.), Subject& topic,Academic Press,New York, pp. 1-23. 1977a, 'On the FormalDescriptionof Inflection',in W. A. Beach, S. E. Fox, and S. Philosoph(eds.), Papersfrom the ThirteenthRegional Meetingof the Chicago Linguistic Society,pp. 15-44. 1977b, 'Commentson the Paperby Wasow:the Role of the Themein Grammatical Analysis', in P. W. Culicover, T. Wasow, and A. Akmajian (eds.), Formal Syntax, Academic Press, New York, pp. 361-377. :1982, 'Where'sMorphology?',LinguisticInquiry13, 571-612. 1984a, 'Rules as "Morphemes"in a Theory of Inflection', in D. Rood (ed.), Proceedingsof the 1983 Mid-AmericaLinguisticsConference,Boulder, University of Colorado. 1984b, 'KawakwalaSyntaxand the Govemment/BindingTheory',in E. Cook & D. Gerdts (eds.), Syntax & Semantics 16: The Syntax of Native Amenrican Languages, Academic Press, New York, pp. 21-75. Aronoff,Mark:1976, WordFormationin GenerativeGrammar,MIT Press, Cambridge. Aronson,Howard:1982a, Georgian:A Reading Grammar,Slavica,Columbus,Ohio. : 1982b, 'On the Statusof Version as a GrammaticalCategoryin Georgian',Folia Slavica 5, 66-80. Bresnan,Joan(ed.): 1982, The MentalRepresentation of GrammaticalRelations,MIT Press, Cambridge. Burzio, Luigi: 1981, IntransitiveVerbs and Indian Auxiliaries, unpublishedPh.D. dissertation,MIT, Cambridge. Chomsky,Noam: 1965, Aspectsof the Theoryof Syntax,MIT Press,Cambridge. :1981, Lectureson Governmentand Binding,Foris, Dordrecht. Hammond,Michael: 1981, 'GeorgianVerbal Agreementand AbstractCase', Presentedat LSA AnnualMeeting, New York. Harris,Alice: 1981, GeorgianSyntax,CambridgeUniversityPress, Cambridge. :1982, 'Georgianand the UnaccusativeHypothesis',Language 58, 290-306. 1983, 'Inversion as a Rule of Universal Grammar:Georgian Evidence', in D. Perlmutter(ed.), Studiesin Relational Grammar,vol. II, Universityof Chicago Press, Chicago, pp. 259-291. Hewitt, B. G.: 1983, review of Harris1981, Lingua 59, 247-274. Hockett, Charles:1947, 'Problemsof MorphemicAnalysis',Language 23, 321-343. Holisky,Dee Ann: 198la, Aspectand GeorgianMedial Verbs,CaravanPress,Delmar,NY. 1981b, 'Aspect theoryand GeorgianAspect', in P. Tedeschi and A. Zaenen (eds.), Tense and Aspect (=Syntax & Semanticsvol. 14), Academic Press, New York, pp. 127-144. Jorbenaje,B. A.: 1981, 'Principystanovlenijainversionnyxglagolov v gruzinskomjazyke' [Principlesof the Formationof InversiveVerbs in Georgian],Annual of Ibero-Caucasian Linguistics8, 66-77. 218 STEPHEN R. ANDERSON Marantz, A.: 1981, On the Nature of Grammatical Relations' unpublished Ph.D. dissertation, MIT. Matthews, Peter: 1972, Inflectional Morphology, Cambridge University Press, Cambridge. Merlan, Francesca: 1982, 'Another Look at Georgian "Inversion"', Folia Slavica 5, 294-312. Perlmutter, David: 1983, 'Personal vs. Impersonal Constructions', NLLT 1, 141-200. Platt, D.: 1981, 'Old Provenqal Verb Inflection: The Balance Between Regularity and Irregularity', in T. Thomas-Flinders (ed.), UCLA Occasional Papers 4: Working Papers in Morphology, Dept. of Linguistics, UCLA. Postal, Paul: 1977, 'Antipassive in French', Linguisticae Investigationes 1, 333-374. ganije, Ak'ak'i: 1980, Kartuli gramat'ik'is sapujvlebi [Fundamentals of Georgian Gtammar; republication of 1973 editon], Georgian Academy of Sciences Press, Tbilisi. Stowell, Timothy: 1981, The Origins of Phrase Structure, unpublished Ph.D. dissertation, MIT. Tschenkeli, Kita: 1958, Einfuhrung in die georgische Sprache, Amirani Verlag, Zurich. [with assistance of Yolanda Marchev, et al.]: 1960-1974, Georgisch-Deutsches Worterbuch,Amirani Verlag, Zurich. Vogt, Hans: 1971, Grammaire de la langue georgienne, Universitetsforlaget, Oslo. Dept. of Linguistics University of California Los Angeles, CA 90024 U.S.A.