Harris Georgian Syntax 1981 PDF

In this series
I DAVID CRYSTAL: Prosodic systems and intonation in English·

2 PIETER A.M.SEUREN: Operators and nucleus
3 RODNEY D.HUDDLESTON: The sentence in written English
4 JOHN M.ANDERSON: The grammar of case·
5 M.L.SAMUELS: Linguistic evolution·
6 P.H.MATTHEWS: Inflectional morphology.
7 GILLIAN BROWN: Phonological rules and dialect variation·
8 BRIAN NEWTON: The generative interpretation of dialect·
9 R.M. W.D IX 0 N: The Dyirbal language of North Queensland ·
10 BRUCE L.DERWING: Transformational grammar as a theory of language
acquisition·
II MELISSA BOWERMAN: Early syntactic development·
12 W. SIDNEY ALLEN: Accent and rhythm
13 PETER TRUDGILL: The social differentiation of English in Norwich·
14 ROGER LASS and JOHN M.ANDERSON: Old English phonology
IS RUTH M.KEMPSON: Presupposition and the delimitation of semantics'"
16 JAMES R.HURFORD: The linguistic theory of numerals
17 ROGER LASS: English phonology and phonological theory
18 G.M .AWBERY: The syntax of Welsh
19 R.M.W.DIXON: A grammar of Yidi/l
20 JAMES FOLEY: Foundations of theoretical phonology
21 A.RADFORD: Italian syntax: transformational and relational grammar
22 DIETER WUNDERLICH: Foundations of linguistics·
23 DAVID W.LIGHTFOOT: Principles of diachronic syntax·
24 ANNETTE KARMILOFF-SMITH: Afunctional approach to child language·
25 PER LINELL: Psychological reality in phonology
26 CHRISTINE TANZ: Studies in the acquisition of deietic terms
27 ROGER LASS: On explaining language chonge
28 TORBEN THRANE: Referential-semantic analysis
29 TAMSIN DONALDSON: Ngiyambaa
30 KRISTJAN ARNASON: Quantity in historical phonology
31 JOHN LAVER: The phonetic description of 'Voice quality
32 PETER AUSTIN: A grammar of Diyari, South Australia
33 ALICE C.HARRIS: Georgion syntax
·Issued in hard covers and as a paper-back

GEORGIAN SYNTAX
A study in relational grammar
ALICE C. HARRIS
Research Assistant Professor of Linguistics, Vanderbilt University
CAMBRIDGE UNIVERSITY PRESS

CAMBRIDGE
LONDON NEW YORK NEW ROCHELLE
MELBOURNE SYDNEY
Published by the Press Syndicate of the University of Cambridge
The Pitt Building, Trumpington Street, Cambridge CB2 IRP
32 East 57th Street, New York, NYI0022, USA
296 Beaconsfield Parade, Middle Park, Melbourne 3206, Australia
© Cambridge University Press I98I
First published I98I
Printed in Great Britain by Western Printing Services Ltd, Bristol
British Library Cataloguing in Publication Data
Harris, Alice C.
Georgian syntax. - (Cambridge studies in
linguistics; 33 ISSN 0068-676x)
I. Georgian language
1. Title II. Series
499.996 PK9I 05 80-4I497
ISBN 0 52I 2358 4 7 Univ.' Blbliothek

Passau
/1 2 -j � )
FOR JIM
Foreword
This book addresses a number of central issues in linguistic theory.

The first and most fundamental issue is that of the relevance of the
notions 'subject', 'direct object', and 'indirect object' to syntactic
description. Georgian has been claimed to be a language to which these
notions are not relevant. It therefore provides a particularly good test
of the basic claims of relational grammar. The issue of the relevance of
grammatical relations to Georgian syntax also affects other issues. These
include: (i) the viability of linguistic universals stated in terms of
grammatical relations, and (ii) the extent to which Georgian is different
from other languages.
In this book, Dr Harris brings out generalizations in Georgian gram
mar that can be captured in terms of grammatical relations, but not in
terms of case or word order. She penetrates the complexities of Georgian
morphology to reveal the underlying syntactic generalizations. The re
sult is a striking confirmation of the relevance of grammatical relations
to grammatical description. At the same time, Dr Harris shows that
Georgian has constructions such as Passive, Object Raising, Causative
Clause Union, and others found in better-known languages. Thus,
Georgian is not as different from other languages as has been claimed
in earlier work on Georgian. Indeed, Georgian is shown to have exactly
the constructions and kinds of phenomena that have been claimed in
relational grammar to characterize natural languages and to require
description in terms of grammatical relations.
The analysis of Georgian presented in this book also bears on what
promises to be another central issue in syntactic theory in the 1980s:
the question of whether or not it is necessary to posit more than one
syntactic level. Working in a derivational framework, Dr Harris shows
that there are rules and generalizations in Georgian that refer to distinct
syntactic levels. Most striking here are the rules that refer to the initial
level, those that refer to the final level, and those responsible for
V1l1 Foreword
the marking of 'retired terms' - nominals that bear a term relation

at one level and a non-term relation (chomeur or emeritus) at a
subsequent level. These results constitute a significant challenge to
theories that claim that a single level is sufficient for syntactic descrip
tion.
This book illustrates the kinds of contributions that theory can make
to the understanding of individual languages, and the study of indi
vidual languages to the development of linguistic theory. One of the
principal problems this book addresses is that of the Georgian case
system. Traditional descriptions state that transitive clauses in Georgian
occur in three distinct case patterns. Dr Harris argues convincingly that
Georgian has the Inversion construction, in which the subject is de
moted to indirect object and the direct object promoted to subject.
Once Inversion is recognized in Georgian grammar, the three case pat
terns are reduced to two. Another traditional problem in the Georgian
case system concerns intransitive clauses. Intransitive verbs are divided
into two classes, each of which is associated with a different case pattern.
This raises two questions: (i) What determines the class assignment of
intransitive verbs? (ii) Are there generalizations uniting the case patterns
in intransitive clauses with those in transitive clauses? Dr Harris
analyzes the case patterns of intransitive clauses in terms of the Unac
cusative Hypothesis, under which there are two fundamentally different
types of initially intransitive clauses: one with an initial subject and the
other with an initial direct object. She shows that under this analysis,
the assignment of intransitive verbs to the two classes is not arbitrary,
and that there are indeed generalizations uniting the case patterns in
transitive and intransitive clauses. This book thus shows how theoretical
constructs can illuminate language-particular phenomena such as the
Georgian case system. At the same time, the Georgian data provides
evidence for these theoretical constructs.
In addition to its contributions to linguistic theory and its exemplifi
cation of relational grammar, this book also gives a good picture of what
a portion of the grammar of a morphologically complex language looks
like. It provides clear, refutable analyses of syntactic phenomena in
what initially appears to be a n unusually complex language. These
analyses are in a form that facilitates comparison with other languages
and the study of linguistic universals. Indeed, this work provides one
of the best and most thorough studies yet available in English of the
syntax of a non-Indo-European language. The interplay of theory and
Foreword IX
description that has produced this extensive documentation of Georgian

syntax is a model of syntactic investigation itself.
DAVID M.PERLMUTTER
San Diego, California

12 June, 1980
Contents
Preface xv
Notes on presentation xvu
INTRODUCTION 1
I. Posing the problems 1
2. The approach taken: theoretical framework 4
3. Results of the investigation 8
4. Some necessary preliminaries 12
1 SYNTACTIC TESTS FOR TERMHOOD
I. Tav-Reflexivization
2. Tavis-Reflexivization
3 . Person Agreement
4. Unemphatic Pronoun Drop
5. Summary
2 CASE MARKING IN SERIES I AND II 39

I. The case marking differential 39
2. Evidence for the analysis of case in Series II 41
3 . Case as a test for termhood 45
Appendix: Constituent screeves of Georgian Series 46
3 OBJECT CAMOUFLAGE
I. The facts to b e considered

2. An analysis
3. Interaction with some rules previously considered

4. An alternative analysis
5. Object Camouflage as a test for termhood
4 OBJECT RAISING 53
I. A description of the data in Georgian 53
xii Contents
2. An analysis 54
3. Facts concerning the nominative-nominal 56
4. Facts concerning the infinitive 60
5. Facts concerning the tvis-nominal 62
6. Object Raising as a test of direct-objecthood 63
5 CAUSATIVE CLAUSE UNION 66

I. An analysis of organic causatives 66
2. Initial grammatical relations 70
3. Derived grammatical relations 73
4. The tvis-nominal 81
5. Theoretical implications 85
·6 VERSION: RULES THAT CREATE INDIRECT OBJECTS 87

I. An analysis 88
2. Benefactive Version 89
3. Stative verbs 92
4. Why not generate version objects directly? 93
5. Coreferential Version Object Deletion 95
6. On so-called four-person verbs 99
7. Version as relation-changing rules 1 00
8. Summary 102
7 PASSIVIZATION 103
.
I. The passive from the viewpoint of language universals 1 04
2. Indirect objects in passives 110
3. Theoretical issues 112
4. Interaction with other rules 1 14
5. Conclusion 116
8 INVERSION 1 17
I. The problem: the case marking differential in Series III 1 17
2. The proposal: a rule of inversion 1 18
3 . Preliminary arguments 120
4. Inversion verbs 1 27
5. Additional arguments 132
6 . The form of the rule 1 33
7. Interaction with other rules 1 38
8. Summary 141
Appendix A: Additional arguments for inversion verbs 141
Appendix B : Transitive inversion verbs with no overt suhject 144
Contents XIIl
9 WHY PATTERN A IS NOT REDUCIBLE TO PATTERN B 1 46

I. An analysis of A as a special instance of B 147
2. Arguments against the proposed analysis 1 47
3 . Conclusion 1 50
10 NON-FINITE VERB FORMS 151

I. Masdars : gerundives or 'derived nominals'? 1 52
2. Infinitive or 'future participle in the adverbial case'? 154
3 . The marking of nominals governed b y non-finite verb forms 1 56
4. The nature of the nominals governed by non-finite verb forms 1 60
II RETIRED TERM MARKING 168

I. Motivating retired termhood 168
2. Correlations between grammatical relations and final marking 1 70
3. Alternative analyses 175
4. Implications of this analysis 178
12 TRANSITIVITY 181
I. Analysis 181
2 . Causative Clause Union 1 82
3 . Retired Term Marking 1 84
4. Conclusion and extension 186
5. Transitive and intransitive 1 86
Appendix: Real and apparent exceptions to transitivity 187
13 SYNTHETIC PASSIVES 191

I. Identification of direct, analytic passive and synthetic
passive constructions 191
2. Some syntactic-semantic characteristics 195
3. A proposal that accounts for the differences 197
4. Arguments for the initial grammatical relations proposed
for synthetic passives 199
5. Arguments for the final grammatical relations proposed for
synthetic passives 201
6. Conclusion and extension 202
Appendix: On four 'passives' in Georgian 203
14 REFLEXIVIZATION 205
I. A review of the facts 205
2. A proposal 208
3. An alternative proposal 210
xiv Contents
15 NUMBER AGREEMENT 211

I. Polypersonalism 211
2. First refinement: failure of third person to trigger
Number Agreement 2 13
3. Second refinement : statement on final termhood 2 15
4. Third refinement : first subjects 2 16
5. Last refinement: relational hierarchy 218
6. The applicability of rule (18) 220
7. Alternative analyses of Number Agreement 223
8. Conclusions 226
16 THE NATURE OF THE GEORGIAN VERB CLASSES 228

I. Hypothesis A: The Ergative Hypothesis 229
2. Hypothesis B: The Unaccusative Hypothesis 235
3 . Inversion and the Georgian verb classes 247
4. The semantics of Class 249
5. Comparison and conclusions 252
6. Theoretical implications 257
Appendix A : Sample lists of verb classes 259
Appendix B: Ambivalent exceptions 267
EPILOGUE 275
1. The grammatical relations 'subject', 'direct object' and
'indirect object' 275
2. Simplifications in case marking 275
3 . Retired term marking 277
4. Rule interaction 277
5. Characteristics of grammatical relations 278
Notes
References
Index
Preface
This investigation of Georgian syntax originated as my 1976 disser

tation at Harvard University ; additional chapters were written in the
fall of 1977. The work is based on interviews with native speakers of
Georgian. The interviews were conducted during a twelve-month stay
in Tbilisi in 1974-5 and a short research visit in 1977 and were sup
plemented by work with Georgians in the United States from 1973
to 1978. The research was supported in part by the International
Research and Exchanges Board, by a Sinclair Kennedy Fellowship
from Harvard University, and by the National Science Foundation.
Although I have had occasion in this monograph to question many
assumptions and claims made by traditional Kartvelologists, I view my
work as a continuation, not a contradiction, of theirs. Without the
foundation laid by Chikobava, Shanidze, Topuria, Tschenkeli, Vogt,
,
and many other specialists, this work would have been impossible.
I am deeply indebted to David Perlmutter for ideas about language
universals that helped inspire the research reported here ; his insights
and criticism have improved this work in many ways. I am grateful to
Dee Ann Holisky for her willingness to debate with me any aspect of
the structure of Georgian and for her fine eye for the detail of language.
I wish also to extend a special thanks to my Georgian teacher for her
long-suffering tolerance of my ideas about the structure of her language.
I am grateful to Bernard Comrie for criticism that was extremely
valuable in revising this work. I wish to thank Stephen Anderson for
introduCing me to the problems of ergativity, which led to my working
on Georgian. In addition, Judith Aissen, Winfried Boeder, Jorge
Hankamer, George Hewitt, Susumu Kuno, Paul Postal, Hans Vogt, the
students in my class on Georgian syntax and other members of the
Department of Linguistics at Harvard University have read parts of the
analysis presented here and have given valuable comments. They do
not necessarily agree with the views presented.
XVi Preface
A great many Georgians, in the United States and in Georgia, have

helped me - as informants, as teachers, as colleagues, and as friends. I
am grateful, too, to the libraries of the Georgian Academy of Sciences,
of Tbilisi State University, and of the Linguistics Institute of the
Georgian Academy of Sciences for helping me to acquire research
materials.
Earlier or different versions of three chapters of the present work
have been separately published; these are:
Chapter 8, as 'Inversion as a Rule of Universal Grammar: Georgian
Evidence,' Studies in Relational Grammar, ed. David M. Perlmutter
(to appear).
Chapter II, as 'Marking Former Terms: Georgian Evidence,' Pro
ceedings of the Seventh Annual Meeting of the North Eastern Linguistic
Society, 8 1-98 (1977).
Chapter IS, as 'Number Agreement in Modern Georgian,' The Classifi
cation of Grammatical Categories (International Review of Slavic Linguis
tics 3. 1-2), ed. Bernard Comrie, 75-98 (1978).
Nates ;an presentatioYf
I Transliteration. The following system is used:

Georgian letter Phonetic equivalent Transliteration
'> a a
2> b b
0 g g
� d d
D e, c e
3 v, w, f v
� z z
(» t' t
(') i, I 1
J � �
� 1 1
a m m
I) n n
ro 0 0
3 p p
-a z z
M r r
lJ 8 8
0 t t
::J u u
en p' p
k' k
a y y
�
� q q
s- - s
'a
c c
f)
(3 C(t8) C
d dz 3
xviii Notes on presentation
Georgian letter Phonetic equivalent Transliteration
V 9 Us)
3 x
b x
� j j
h h
J
2 A note on glosses
The morphology of the Georgian verb. is very complex, and there is
no way to escape the use of examples with co mplicated verb forms. In
addition to the lexical meaning of the root, a single verb form may code
the following information:
person of subject
person of direct object
person of indirect object
number of subject
number of direct object
number of indirect object
tense
aspect (complete/incomplete, habitua l/non-habitual)
VOice
mood
direction and orien�ation
cilUsative/non-causative (cf. ch. 5)
version (cf. ch. 6)
etc.
In this work, the gloss of a verb will not i!lclude all this information
for two reasons: (i) A great deal of it is irrelevant to the topic of this
monograph and would simply overwhelm the reader with a mass of
material not necessary for interpreting pa rticlJlar examples. (ii) Although
Georgian is generally agglutinative, the information necessary to in
terpret particular examples cannot always be attributed to any par
ticular morpheme. Therefore, the principle behind my glosses is that
they should give only that information which will enable the reader to
understand the example in the context of the point under discussion.
Since one of the main concerns of this monograph is the correspondence
between case and grammatical relation, case is always dearly marked.
Since case varies with Series and Class of the governing verb, those
are always marked in the verb gloss . The grammatical relation is not
Notes on presentation XIX
marked in the gloss, but it is discussed at length in the text. Particulars

are given below.
2.1 Analytic gloss.In the analytic gloss, which is directly below the
Georgian example, the information carried by one word is hyphenated
together in English; e.g. darera 'he-wrote-it'. Analytic glosses are not
generally given for examples quoted in the text itself.
2.1.1 Nouns. Plurality is indicated by plurality in the English word,
not by a 'PL'; e.g. 3aylebi 'dogs', not 'dOg-PL'.

Case is noted separately in capitals; genitives are indicated separately,
not by an English genitive; e.g. 3aylis 'dOg-GEN', not 'dog's'. An excep
tion to this is the possessive reflexive, tavis-, which is genitive with
secondary case marking; it is glossed 'self 's'.
While nouns and most pronouns have a complete declension in
Georgian, first and second person personal pronouns and a few others
do not. These personal pronouns use a single form for the cases of
greatest concern to us, the ergative, nominative, and dative. For
example, me represents the ergative, nominative, and dative of 'I, me'.
In the examples, a case is always specified for these pronouns, on the
basis of the case used in the comparable sentence for a noun or third
person pronoun.
Positionals (adpositions) in Georgian are all postpositions. Some are
written together with the noun in Georgian and are hyphenated to
the noun in the English gloss; e.g. saxlSi 'house-in' from saxl- 'house'.
A few postpositions are not written with the noun in Georgian; I have
followed the Georgian spelling conventions, both in the Georgian
example and in the English gloss; e.g. 3aylis mier 'dog by'. Postpositions
govern particular cases, but these play no part in the discussion, and so
are generally not included in the glosses.
The Georgian alphabet has no capitals, so proper names are not
capitalized in Georgian. I have followed this spelling convention in the
examples transliterated here, but have capitalized the English gloss so
that the reader can identify those words as proper names; e.g. goca
'Gocha-NoM. '
2.1.2 Verbs. Because case patterns vary with Series and Class of the
verb, and because these characteristics would not otherwise be apparent

to the reader, they have been included in the gloss of every verb. They
are always the last two elements in the analytic gloss. Series is always
indicated with a Roman numeral, in text and gloss alike; Class is always
xx Notes on presentation
indicated with an Arabic numeral. For example, dafera 'he-wrote-it

II- I ; rleba 'he-stays-I-2'. (These elements are not relevant to the rules
discussed in §4 of the Introduction, and therefore are not included
until ch. I. )
Person Agreement is indicated by the appropriate pronouns in the
following order: subject, v erb root, indirect object, direct object ; e.g.
gcem 'I-give-you-it-I - I ' . All terms are indicated in the gloss, though
they may be disguised in Georgian by zero markers or morphophonemic
rules (cf. ch. I , § 3).
Gender is included to aid the reader, although it is not indicated in
Georgian. Third person singular subjects are glossed 'he' except when
a female name is used, or when the subject must be inanimate. Objects
are glossed 'it' except wh en a nimacy is indicated by a noun in the
clause, or when an animate would be more natural.
Number Agreement is indicated in the gloss only when it is clearly
marked in the Georgian.
ma3levs sacukrebs.
he-gives-me-it-I- I gifts-DAT
'He gives gifts to me'.
Although 'gifts' is plural, i t does not trigger Number Agreement ; this
fact is reflected in the 'it' of the analytic gloss of the verb. In the fol
lowing example, number of the indirect object is indicated morpho
logically and is glossed:
gva3levs sacukrebs.
he-gives-us-it-I - I gifts-DAT
'He gives gifts to us.'
Tense is indicated in analytic glosses only by the present, past, and
future forms of the Engli sh verb ; e.g. dafers 'he-will-write-it'. The
simple (habitual) present in English is used in analytic glosses for the
present ; this is used for brevity only and implies no claim as to the
nature of this tense in Georgian .
When it is important that s ome morpheme be identified in the verb
form, as in the discussion s of agreement, the important morphemes
will be isolated by hyphenating or using heavy type. In some instances
the corresponding element in the gloss is in heavy type also ; e.g.
mornfera 'he-wrote-rne-it-I I- I '. Otherwise, parts of the verb will not
be separated by hyphens.
Notes on presentation XXl
2. 1 .3 Adjectives. In Georgian, attributive adjedives do agree with the

nouns they modify, but the case is not fully apparent. For this reason
and because it is not important to the· discussion, no case is indicated
for adjectives that precede their head nouns.
2.2 Final gloss. The final gloss gives the sentence which corresponds
most closely in English, including the tense, which is the appropriate
one in English. In some instances, English has no corresponding syntac
tic rule, and the sentences in which the rule has applied and those in
which it has not have the same gloss.
Final glosses are sometimes omitted from ungrammatical sentences,
since those sentences do not always have a meaning. But if it would not
be clear what the ungrammatical sentences show, a final gloss is given
in parentheses, that corresponds to the intended meaning of the sentence.
2.3 Quotational form. When a noun is referred to in the text, it is

usually quoted in its nominative case form or in its stem form (without
case marking). A verb will be quoted in its masdar (nominalized) form,
unless that is inappropriate, in which case it will be given in a finite
form ; this will be made apparent by the gloss.
3 Diagrams
This work uses the conventions of the 'network' diagramming developed
for relational grammar (Perlmutter & Postal 1 977). A clause and the
elements which constitute it are represented as nodes. Arrows connect
ing the clause node with other nodes indicate that the latter are depen
dents of the clause. Labels on the arrows indicate the grammatical
relations which various elements bear to the clause ; only the central
grammatical relations (cf. P.S below) are indicated in network dia
grams. The predicate relation is labeled 'P' . The subject relation is
marked 'I', the direct object relation '2 ', and the indirect object re
lation ' 3 ' . Benefactives are marked as 'B'. For example, network (I )
mno perangi bavSvi

'shirt' , child'
xxii Notes on presentation
should be read as 'gela is the subject of the matrix clause (or of the verb,
codna "know"); and its direct object is the clause consisting of the predi
cate se�erva "sew", the subject nino, the direct objectperangi "shirt", and
the benefactive bavsvi "child".' This corresponds to the sentence (2).
(2) gelam icis, rom nino �eravs perangs bavsvistvis.

Gela he-knows that Nino she-sews shirt child-for
'Gela knows that Nino is sewing a shirt for the child. '
Levels are indicated as in (3), where Benefactive Version has applied

to make the benefactive an indirect object.
se�erva nino perangi bavIvi
This network indicates, among other things, 'bavsvi is an initial benefac

tive and a final indirect object'.
Retired subjects, direct objects, and indirect objects are marked 'R-I',
'R-2', and 'R-3', respectively. O ther symbols used are introduced where
appropriate.
4 The reader
The monograph is written primarily for syntacticians who know little
or nothing of Georgian. For this reason, material that will be of interest
only to the Georgian specialist is put, as often as possible, into notes or
appendices. The complexities of case marking and verb agreement
are introduced gradually.
5 The dialect described

This monograph is based on work with informants in Georgia and the
United States from 1973 to 1978. The dialect described here is that
s poken in the capital city of Georgia, Tbilisi (Tiflis). In those insta nces
where I have isolated two dialects among my informants, I have made
reference to this fact in the text or in a footnote. No attempt is made to
account for divergences found in dialects outside Tbilisi or in earlier
periods, or for phenomena restricted to written Georgian.
Introduction
I Posing the problems
To many general linguists, Georgian has long seemed an inscrutable

language. The main difficulty has been that the apparent subject of a
given clause is not always in the same case, as it is in most languages.
The sentences in (I) illustrate the problem of the case differential.
(I) (a) glexi tesavs siminds.

peasant-NOM he -sows-it-I-I corn-DAT
'The peasant is sowing corn.'
(b) glexma datesa simindi.
peasant-ERG he-sowed-it-II-I corn-NOM
'The peasant sowed corn.'
(c) glexs dautesavs simindi.
peasant-DAT he -sowed-it-III-I corn-NOM
'The peasant has sown corn.'
The traditional names are used for cases; the names themselves are not
intended as a claim about the real structure of the language. The nomi
na tive case is marked by -i (0 after a vowel), the ergative by -ma (-m
after a vowel), and the dative by -so The three case marking Patterns
possible for a given clause are stated in (2).
(2) Subject Direct Object Indirect Object

Pattern A ERGATIVE NOMINATIVE DATIVE
B NOMINATIVE DATIVE DATIVE
C DATIVE NOMINATIVE tvis-nominal
A second ob vious problem in case marking is that the case marking

Pattern does not vary for all verbs in the same way. The distribution of
Patterns A, B, and C is stated in (3).
WI"'
2 Introduction
(3) Series I II III

Class I B A C
2 B B B
3 B A C
4 C C C
Moreover, in Georgian Pattern A, with ergative subjects, is by no means

limited to transitive verbs. In order to predict the case of a subject in
Georgian, one must know the Series and Class of the governing verb
form. The three Series are groups of tense-aspect-mood categories; these
are listed in the appendix to ch. 2. In this work, the Class of the verb
is defined morphologically; the precise criteria used are given in ch. 16,
Appendix A. It is shown below that the Classes correlate with specific
syntactic and semantic properties.
Looking at (I) again in light of (2-3), and knowing that the governing
verb is a Class I verb, we can see that in (Ia), which is in a Series I
tense, the subject must be in the nominative and the object in the dative.
In (Ib), which is in a Series II tense, the subject is in the ergative, the
object in the nominative. And in ( IC) , we see the Series III variant:
subject in the dative and object in the nominative.
I have been using the labels 'subject' and 'direct object' in the way
they are usually used in the description of more familiar languages. In
fact, it is not at all obvious that these notions are even appropriate ones
for Georgian. This is the primary problem that this monograph
addresses: Are the notions 'subject', 'direct object', and 'indirect object'
relevant to Georgian?
The difficulty in approaching the notion 'subject' in Georgian is that
there is no agreement among the three most obvious criteria for defining
this concept - case, verb agreement, and some intuitive idea of subject.
In general, in order to make a claim about which nominals are subjects,
linguists have had to choose between these criteria on an arbitrary basis.
Shanidze (1973: 195-7) and Tschenkeli (1958: 497) select the intuitive
idea, and would call glexi 'peasant' the subject in all of the sentences
in (1). Vogt (1971: 81) chooses instead Person Agreement; this means
thatglexi is subject in (la-b), and simindi 'corn' in (IC). Chikobava (1968)
and Aronson (1970) define the subject as that nominal which always
triggers Number Agreement in the verb.! Marr and Briere (1931: 244),
writing mainly of Old Georgian, say that the 'nominative indicates the
grammatical object - in fact the real su�ject' in (rb-c). Patsch (1952/53)
I. Posing the problems 3
considers 'subject' an Indo-European notion and therefore confusing

in Georgian. Sommerfelt (1937) suggests that 'subject' is an inappro
priate notion for a language like Georgian.
Each of the above-mentioned analyses of 'subject' in Georgian is
unsatisfactory because the choice of the criterion used is not clearly
motivated. It is not clear, for example, why Chikobava considers Num
ber Agreement to be the most important criterion, while Vogt considers
Person Agreement to be.2
No analysis of the notion 'subject' in Georgian can be adequate if it
is based upon a single criterion. Recent work on language universals
has shown that the following syntactic characteristics are typical of
subjects: they trigger reflexivization, they trigger coreferential deletion,
they trigger verb agreement, they undergo Unemphatic Pronoun Drop,
etc. (cf. Anderson 1976; Keenan 1974, 1976; Postal & Perlmutter 1974).
But no one of these criteria is sufficient to establish the notion 'subject'
for any language. In this monograph, I use, among others, the following
syntactic rules to argue for the notion 'subject':
(a) Tav-Reflexivization (full-pronoun reflexivization)
(b) Tavis-Reflexivization (possessive reflexivization)
(c) Person Agreement
(d) Number Agreement
(e) Unemphatic Pronoun Drop
(f ) Coreferential Version Object Deletion.
Additional phenomena will also be used to argue for the existence

of a subject in Georgian, as well as for a direct object and indirect
object.
Nor could an analysis of the notion 'subject' in Georgian be adequate
if it failed to recognize distinct levels of derivation. A large portion of
this work is devoted to an analysis of the major rules that change gram
matical relations in Georgian. Because Georgian has rules like Passivi
zation, Object Raising, Causative Clause Union, and Inversion, the
notion 'subject' cannot be adequately characterized by reference to a
single level of derivation. 3
This leads us to the second major problem addressed by the present
work: \Vhat is the nature of the rules that change grammatical relations
in Georgian? This question is addressed from several points of view.
From the language-internal viewpoint, individual rules provide
additional criteria for identifying the grammatical relations 'subject',
4 Introduction
'direct object', and 'indirect object'. Further, analyses of these rules

lead to a consideration of rule interaction in Georgian. From the point
of view of language universals, it is hoped that a detailed study of
individual rules in Georgian will contribute to a better understanding
of the ways in which particular rules vary and ways in which they are
the same, across language.
The third major problem considered here is how to account for case
marking, given the complexities represented in (2-3). Even after the
subject, direct object, and indirect object are identified on various
derivational levels, it is not obvious how nominals get marked with
particular cases.
Finally, the discussion of these questions raises another problem:
What is the nature of the verb Classes, which play an important role in
determining case marking? In this work, the membership of the verb
Classes is determined on the basis of morphological criteria (see ch. 16,
Appendix A). 4 The syntactic and semantic nature of the Classes is
investigated in the final chapter.
2 The approach taken: theoretical framework
This study is an exploration of the adequacy of relational grammar for

describing the syntax of a language. The version of relational grammar
used here has been modified to make the work more accessible to those
interested primarily in Georgian. Only those parts of relational gram
mar are discussed here that are most important to this research on
Georgian ; other concepts are introduced in the text as they become
relevant. The theory of relational grammar is discussed in detail in
Johnson ( 1 974, 1 977; to appear), Perlmutter ( 1 978, 1979, to appear a,
b, c, d), Perlmutter & Postal ( 1 974, 1977, to appear a, b, c), Postal
(1977), and other works cited below.
2. 1 Some principles of relational grammar

The central claim of relational grammar is that the processes of human
languages are best described by rules operating, not on strings of
ordered elements, on cases, or on constituent structure trees, but on
grammatical relations. It is held that grammatical relations are the most
appropriate basis for stating generalizations, both for universal rule
types and principles, and for language-particular data. For example,
the rule of Passivization is best described universally as a rule that
2. Theoretical framework 5
promotes direct objects to subjecthood, not as a rule that moves the
second NP from the left all the way to the left, etc. (cf. Perlmutter &
Postal 1 977).
A fundamental principle of this theory is that the structure of a
clause is the set of grammatical relations obtaining between the elements
of a clause. The predicate bears the predicate relation to its clause. Three
types of nominal-clause relations will figure in this work. (i) Terms or
term nominaIs bear a subject relation, a direct object relation, or an
indirect object relation. Subjects and direct objects together constitute
the nuclear terms. (ii) The second type of nominal-clause relations are
oblique relations; among these, only the benefactive relation, instrumental
relation, superessive relation, locative relation, and comitative relation will
be referred to here. (iii) The third type is retirement relations - chomeurs
and emeritus terms. Nominals bearing one of these relations are terms
that become non-terms through the application of some syntactic rule,
as specified by the Chomeur Law (below) or some other condition (cf.
ch. 7, §3 and ch. II, §r). The notion 'non-term' includes nominals
bearing oblique relations and those bearing retirement relations. Predi
cates and nominals are said here to be dependents of the clauses to which
they bear grammatical relations. The norninals bearing grammatical
relations to a clause are said here to be governed by the verb of that
clause. The predicate relation, term relations, and oblique relations are
undefined primitives of the theory ; chomeur relations and emeritus
relations are defined by the theory.
Only the central nominal-clause relations are investigated in this
work. The overlay relations, such as question and topic, are not treated
here. The internal structure of nominals must also be ignored here ;
only the relations of whole nominals to their clauses are investigated.
A number of other elements, such as adverbs, are completely ignored
in order to study in depth the relations named above.
The following hierarchy among nominal-clause grammatical relations
is posited :
(4) subject
direct object
indirect object
non-terms
This hierarchy plays a role in a variety of processes of natural language
(cf. Keenan & Comrie 1977 and other recent works).
6 Introduction
Rules that change grammatical relations are of three types : (i) Revalu
ations include advancements and demotions. An advancement is the
promotion of a nominal dependent of a clause up the hierarchy (4). A
familiar example of this is Passivization, the advancement of a direct
object to subjecthood (Perlmutter & Postal 1974, 1 977, and ch. 7). A
demotion lowers a nominal dependent of a clause on the hierarchy (4).
One demotion is discussed here ; Inversion is the rule that demotes
subjects to indirect-objecthood (cf. Harris (to appear a), Perlmutter (to
appear b), and ch. 8). (ii) Ascensions raise a nominal out of an embedded
clause, making that nominal a dependent of the higher clause. A
familiar example of this is Non-Subject Raising ( Tough-Movement),
the ascension of an object from a sentential subject to subjecthood of
the higher clause (cf. ch. 4). (iii) Clause Union makes all dependents of
an embedded clause dependents of the matrix clause. Clause Union is
discussed here in relation to organic causatives (ch. 5) and inceptives
(ch. 1 6, §2.2.3 and n. I f).
Since the structure of a clause is the set of grammatical relations
holding between the elements of a clause, and since syntactic rules may
act on nominals to change grammatical relations, the theory recognizes
distinctions in derivational levels, or strata. It is found in this work that,
for Georgian, reference to initial and final levels is adequate for the
formulation of most rules. Intermediate levels, however, are also recog
nized. (The schematic representation of levels of derivation is discussed
in the introductory Note on Presentation.) An important principle of
relational grammar is that a particular grammatical relation is not
simply a bundle of properties ; rather, the syntactic rules of a particular
language predict precisely which properties are associated with initial
and which with final termhood (cf. Johnson 1 977). For example, we
will find that in Georgian the derived subject of a passive can trigger
Subject Person Agreement, a property of final subjects, and cannot
trigger Tav-Reflexivization, a property restricted to initial subjects.
The theory of relational grammar defined in the works cited above
puts a number of constraints on each of the types of rules that exist
and thus the types of human language that can exist. Three of these are
central and are stated informally here. Others are introduced in relation
to a specific rule as they become relevant. Some are not relevant to this
work and are not discussed here (cf. works cited above, especially
Perlmutter and Postal (to appear c), for a more detailed discussion of
constraints ).
2. Theoretical framework 7
The Oblique Law requires that if a nominal bears an oblique relationi

at any level, it bears that relationi initially. This rules out the possibility
of any term or non-term nominal changing to an oblique relation, while
permitting the possibility of an oblique being promoted to a term
relation.
The Stratal Uniqueness Law, informally stated, requires that at each
stage of derivation no more than one nominal bear a given term relation
in a clause. That is, at any stage of derivation, a clause may have no
more than one subject, one direct object, and one indirect object (cf.
ch. 5, §4.2 and ch. 6, §6 for additional discussion).
The Chomeur Law, or Chomeur Condition, informally stated requires
that if a nominali assumes the term relation borne by nominal), nominalj
becomes a chomeur. For example, in English the advancement of the
direct object under Passivization puts the initial subject en chomage.
Here the subject chomeur, unlike a subject in English, cannot trigger
Reflexivization, cannot trigger Verb Agreement, cannot trigger Equi,
cannot undergo Subject-to-Subject Raising, etc.
2.2 Why relational grammar?

Georgian is a language with 'free' word order ; that is, the order of
words in a sentence does not directly indicate grammatical relations.
For this reaso n, formulating the rules of Georgian syntax on the basis
of word order would not be as straightforward as for a language like
English. The simplest way to do so would be to formulate rules on the
basis of the most unmarked order, then have 'scrambling' rules apply
in recognition of the artificiality of that order. Similarly, it would be
possible to formulate rules on the basis of cases. But each rule of syntax
would essentially have to incorporate the case marking differential rep
resented in (2-3). Grammatical relations, on the other hand, provide
the basis for a simple formulation of the rules of Georgian syntax and
for the capturing of linguistically significant generalizations about the
syntax and semantics of the language.
A second reason for working within the theory of relational grammar
is that a major concern of this work is universal grammar, in the sense
that it is an investigation of the appropriateness of certain proposed
universals in a language with a typologically unusual structure. A num
ber of universals have been proposed concerning rules like Passivization
(cf. Keenan 1 975, Perlmutter and Postal 1 977, etc.) and causative for
mation (cf. Aissen 1974a, b ; Comrie 1975 ; Perlmutter & Postal 1 974 ;
8 Introduction
and Shibatani 1976). Testing these proposals on Georgian data is one

purpose of this work. Since various languages have various word orders,
universals of rules of this type are most simply stated on the basis of
grammatical relations, not on the basis of movement of constituents.
Relational grammar thus provides the best framework for a comparative
investigation of this sort.
Finally, it is well known in linguistics that a theory of language may
highlight certain problems, and that a theory itself may point to sol
utions. In this instance, relational grammar provides the keys to the
central problem of Georgian linguistics - the case differential sketched
above. In particular, the two rules of Inversion and Unaccusative,
defined within this theory, lead to the realization that case marking in
Georgian is not as complex as it appears to be (cf. §3.3 below and chs.8,
9, 16). Relational grammar also highlights problems in Georgian that
had not previously been addressed. For example, because relational
grammar identifies a class of retired terms, we can confront the problem
of how this class behaves and how it is marked (cf. §3.2.2 below and
ch. II). Thus, the theoretical framework in part shapes the investigation.
3 Results of the investigation
3. I The notions 'subject', 'direct object', and 'indirect object'

It is established in chs. I and 2 that, although the dependents of a
clause may be marked with different cases when the governing verb is
in Series I and II (as illustrated in (la-b)), one nominal has in both
Series a particular set of syntactic characteristics, which are associated
with 'subject'. Another nominal has in both Series a set of characteris
tics which can be associated with 'direct object', and a third has ones
associated with 'indirect object'. In ch. I , various syntactic phenomena
are introduced and associated with particular grammatical relations in
Series 1. In ch. 2, it is shown that the same nominal is involved in the
same phenomena in Series II, even though its case marking is not the
same as in Series 1 .
Chapters 2-8 and 10 establish the simplicity of a grammar in which
syntactic rules refer to grammatical relations. Each of the chapters 4-8
and 10 introduces one or more rules that change grammatical relations.
In each instance it is shown that a nominal that bears a particular gram
matical relation undergoes the rule, regardless of its other syntactic or
morphological characteristics. Thus, while these rules may be �imply
stated on the basis of grammatical relations, in Georgian it would be

impossible to capture the linguistically relevant generalizations on the
basis of case.
3.2 The nature of the rules that change grammatical relations

3.2.1 Inversion. In ch. 8 it is shown that Georgian has a rule of Inver
sion, which demotes the subject to indirect-objecthood. Phenomena
of this type have been discussed under the names 'Flip', 'Psych Shift',
or 'Subject-Object Inversion' for other languages ; but the evidence
supporting a syntactic rule of this type has generally been sparse (cf.
Davison 1969; Harris 1973; G. Lakoff 1970; R. Lakoff 1968; N.
McCawley 1976; Postal 1970, 1971; Rosenbaum 1967; Sridhar 1976a, b;
and Ziv 1976). In Georgian, on the other hand, the evidence is very
strong. In Georgian the rule is not restricted, as in many languages, to
a class of so-called 'affective' verbs. It applies also if triggered by a
particular mood (the evidential) of the governing verb. The construc
tion in this environment provides a kind of evidence rather different
from that found in other languages.
3.2.2 Retired Term Mark-ing. At least six different rules in Georgian

Object Raising, Causative Clause Union, Passivization, Inversion,
Masdar Formation, and Infinitive Formation - create retired terms (cf.
above §2.1). In ch. I I it is shown that the rules that mark retired
terms can be stated in a completely general way on the basis of gram
matical relations, without reference to the rule that creates the particular
retired term.
3.2.3 Rules that refer to both -init-ial and final termhood. Chapters 14-16
establish that the rules of Tav-Reflexivization, Number Agreement, and
two of the rules that assign cases are stated on initial and final termhood;
that is, they are so-called 'global rules'. (Tav-Reflexivization is stated
more simply in ch. I and refined in ch. 14 on the basis of facts con
sidered in intervening chapters.)
3.2.4 Rule interaction. It is found to be unnecessary to impose conven

tional extrinsic ordering on syntactic rules. In general, rule interaction
can be handled simply by stating rules on initial termhood, on final
tcrmhood, with reference to termhood at more than one level (cf.
*3.2.3), or without restriction (cf. Epilogue, §4).
10 Introduction
3.3 Accounting for the case marking differential

In ch. 8 it is shown that a syntactic rule, Inversion, applies just where
Pattern C is found in (3). Thus, Pattern C is shown to be derived from
Pattern B by a regular synchronic rule. In ch. 9 it is established that
no such rule relates Pattern A to Pattern B, and that these two case
Patterns must be assigned by different sets of rules. Finally, ch. 16
shows that the rules of case marking for Series II can be most simply
stated if they take into account both the initial and final grammatical
relations established in this work.
3.4 The nature of the Georgian verb classes

In early chapters verb Classes are treated as arbitrary groupings of
verbs, morphologically defined. It is established that verbs of Classes 1
and 3 have two important syntactic characteristics :
(a) They govern Pattern A in Series II.
(b) They govern Inversion in Series III.
In ch. 16 it is shown that these two characteristics are related to a
single semantic-syntactic trait that distinguishes these verbs from those
of Classes 2 and 4. The fact that the semantic-syntactic nature of each
verb can be predicted on the basis of universals means that Class
membership is not arbitrary, as has been thought ; it is therefore not
necessary to have a lexical listing of Class membership.
3·5 Georgian data from the point of view of universals

Because of its structure, Georgian provides evidence that differs in an
interesting way from that given by other languages. For example, it has
been claimed that Causative Clause Union creates simplex clauses
(Aissen 1974a, b; Comrie 1975, 1976e; Postal & Perlmutter 1974 ; etc.).
In Georgian the evidence for this is particularly strong. First, each
term, not just the subject, triggers Person Agreement. Second, the
nominals in the causative are assigned cases according to (2-3) above.
This is evidence that the rule that forms causatives is not a rule that
assigns particular cases to the initial terms of the embedded clause, but
that it actually makes them dependents of the matrix clause (cf. ch. 5).
Discussion of the nature of nominalizations has generally been based
on selection restrictions, productivity, and nominal-like structure (cf.
Chomsky 1970; Comrie 1976d ; G. Lakoff 1< )70 ; Lees 1 963 ; and
McCawley 1970). Georgian provides a differ · n t k i lld of ' v i d ' 0 .t;
that certain nominalizations are derived from clauses. I n Georgian,

retired terms are marked regularly (cf. §3.2.2). The fact that this
marking is also used in nominalizations enables us to identify the
grammatical relations involved. An analysis that did not derive these
nominalizations from full clauses would be claiming that the generality
of this marking was an accident (cf. chs. IO and I I ) .
3.6 Relational grammar

This work constitutes a powerful argument for relational grammar.
Georgian presents a complex problem in terms of case marking. I have
shown here that the facts of case marking in simple sentences and in the
major complex constructions of the language can be stated in a simple
and elegant fashion on the basis of grammatical relations. A straight
forward statement of the case marking facts cannot be made on the
basis of linear order and dominance relations.
Second, it is shown throughout that Georgian, a non-Indo-European
language whose syntax is not well known outside Georgia, behaves as
predicted by relational grammar. There are some points on which the
wrong prediction or no prediction at all is made by the theory; these
are pointed out in the text. But the fact that the majority of the data
are correctly predicted is a very strong argument for some form of this
theory. Clause Union and Inversion provide particularly good examples
of the fact that the universal core of a rule can be simply stated on the
basis of grammatical relations, but not on the basis of linear order or
case. Case marking rules are particularly good examples of the fact that
language-particular rules may be stated most simply on the basis of
grammatical relations.
Third, among current theories, only relational grammar defines a
class of retired terms. As pointed out above and as shown in detail in
ch. J r, any analysis of Georgian that fails to recognize this class of
l1omina\s will miss a set of significant generalizations about the language.
Thus, any analysis written in terms of a theory that does not define such
a class of nominals will be giving an inadequate account of Georgian.
3 .7 The systematicity of Georgian

Georgian has been characterized by some general linguists as a language
in which every verb is an exception, as a language that is 'totally irregu
lar'. I believe that this monograph, because it shows the consistency in
a large number of syntactic-semantic phenomena, establishes that
12 Introduction
Georgian is systematic. Like any other language, it has irregularities; but

these should not be permitted to obscure the general, regular patterns
of the language.
4. Some necessary preliminaries
In this section several rules and phenomena are introduced. The first
has no theoretical importance to this work, but it is necessary to the
understanding of some of the data presented below. The second is
Question-Word Question Formation; its importance derives from the
fact that it is used as a test of constituency. Suppletion is introduced
in §4.3; it will be the basis for arguments concerning termhood in later
chapters. Finally, a statement is made concerning word order in
Georgian.
4. I Aris-Cliticization
In Georgian, there are pairs of sentences like (5-8), where aris 'he/she/it
is' alternates with its c1itic form, -a.
(5) (a) es arts ceml r;tgni.
this is my book
'This is my book.'
(b) es cemi {:ignia.
book-is
'This is my book.'
(6) (a) sad aris cemi {:igni?
where is my book
'Where is my book?'
(b) sadaa cemi {:igni?
where-is
'Where is my book?'
(7) (a) es {:igni cemi aris. / es (:igni aris cemi.
'This book is mine.'
(b) es {:igni cemia.
'This book is mine.'
(8) (a) gela aris ekimi.
Gela is doctor
'Gela is a doctor.'
(b) gela ekimia.

doctor-is
'Gela is a doctor.'
As we can see by comparing (6) with (5) and (7-8), -a occurs after
vowels, -aa after consonants.
It is most natural for the cliticization rule to apply wherever its input
conditions are met, except when the word aris is emphasized. Because -a
is more natural than aris, most of the examples given in this work
use -a ; it will be glossed as in the (b) examples above.
I have called -a a clitic because it can never bear stress; nor can
there be a pause between the enclitic and the word to which it attaches.
This rule applies to reduce only the third person singular form only
i n the present tense; there are no clitic (i.e. unstressable) forms of the
other persons and numbers in the present or of any person in other
I ·nses. 6
That this rule is clause-bounded is shown in (9). In (a) the -a has

attached to a constituent of its own clause, ekimi 'doctor'. In (*b) and
· c ) it has cliticized to constituents of the higher clause.
(Ij) (a) gelam icis, rom anzori ekimia.

Gela knows that Anzor doctor-is
'Gela knows that Anzor is a doctor.'
(b) *gelam icisaa, rom anzori ekimi.
Gela knows-is that Anzor doctor
('Gela knows that Anzor is a doctor.')
( c) *gelama icis, rom anzori ekimi.
Gela-is knows that Anzor doctor
('Gela knows that Anzor is a doctor.')
The -a most often cliticizes to a predicate nominal or predicate

. l ! ljcctive, hut it may attach to other constituents as well. In (a) of ( 1 0)
I t is on a participle, in (b) on the subject.
( 1 0) ( a ) es noxi gar;mendilia.
this rug cleaned-is
' This rug has been cleaned.'
(b) es noxia gar;mendili.
rug-is cleaned
'This rug has been cleaned.'
14 Introduction
4.2 Question-word Question Formation
In this section I describe the formation of Q-word questions in Georgian.
Because of constraints described in §4.2.3 and 4-2.4, this rule serves as
a test of clause constituency (cf. §4.2.6).
For forming content questions, Georgian uses question pronouns, an
adjective romeli 'which', and a variety of adverbs.
4.2. 1 The position of the Q-word in a simple sentence. In general, the

Q-word in a simple sentence in Georgian occurs in the position immedi
ately preceding the finite verb. Since the position of the verb relative
to other constituents is fairly free, the Q-word may be sentence-initial
or sentence-internal. The sentences of ( I I ) differ only in the relative
order of their constituents.
( I I ) (a) sad (:avida nino ?
where she-went Nino
'Where did Nino go?'
(b) nino sad (:avida?
'Where did Nino go?'
(c) "'sad nino (:avida?
(d) "'(:avida nino sad?
Those sentences are ungrammatical ( "' I I C, d), in which the Q-word
does not immediately precede the verb.
There is one constituent which, if present, must intervene between
the Q-word and the finite verb that governs it ; this is the negative
particle: ar 'not', afar 'no more', nu negative imperative, rodi em
- -
phatic negative, or ver 'cannot' .6 (12) gives variant word orders ; only
(12a) is grammatical, with the order Q-word NEG VERB. Any other order
is ungrammatical.
(12) (a) sad ar ginda (:asvla?
where not you-want going
'Where do you not want to go?'
(b) "'ar sad ginda (:asvla?
(c) "'ar ginda sad (:asvla?
I will refer to the proper position of the Q-word as 'immediately pre
ceding the finite verb', in spite of the intercession of these particles.
4 .2.2 Multiple Q-words. Multiple Q-words, when thcy occur in a

single clause, are all attracted to the verb as describt·J above. They
occur in preverbal position in the same order that the full nominals for
which they substitute occur most naturally in clauses, that is, SUBJECT
UJJJECT. (13a) gives a declarative sentence; (13b) is its grammatical
ounterpart with Q-words as both subject and direct object. (*13c-e)
�ive variant word orders.
( [ 3) (a) nino purs qidulobs.

Nino bread buys
'Nino is buying bread.'
(b) vin ras qidulobs?
who what buys
'Who is buying what?' cf. *Who what buys?
( c) *ras qidulobs vin?
( d) *vin qiduZobs ras ?
cf. Who buys what?
( e) *ras vin qidulobs?
I � b) differs from the ungrammatical examples in that its subject and

/ lhj t Q-words both precede the verb and occur in the order SUBJECT
·
·
""J ECT.
'1 . 2 . � What constituents may be questioned? In Georgian any constituent

I I I . I Y be questioned in a simple sentence. In (14a) the subject is ques
l i PIH:d, in (14b) the direct object, in (14c) the indirect object, in ( I 4d)
t l l object of a postposition, and in ( I 4e) a non-term in the instrumental
.
l .1 '.
(II (a) xvaZ vin midis kalakSi?

tomorrow who he-goes city-in
'Who is going into the city tomorrow?'
(b) bebia ras acukebs sviliSvils?
grandmother what she-gives-him-it grandchild
'What will grandmother give (her) grandchild?'
( ) vis elaparalf-eba
' vano?
who he-talks-to-him Vano
'Who is Vano talking to?'
(d) oi� tan midixar?
'
·
who-at you-go
'Tu whose house arc you going?'
16 Introduction
( e) rit arts lS saxli aienebuli?

what-INST it-is that house built
' What is that house built out of?'
... "1.G 1t. � \ I ' } .,<'1 (L ('
n unmarked order in non-interrogative sentences, all attributives
�
'
{recede the head of the nominal, except the relative clause, which
follows. When a constituent of a nominal is questioned, the whole
nominal precedes the verb. In ( I sa) the possessive precedes its head,
the whole nominal immediately preceding the verb. In (*I Sb) the
nominal-internal order is reversed so that the possessive (the Q-word)
immediately precedes the verb ; in (* I sc) the head follows the verb.
( I S) (a) visi svili xar sena ?7
whose child you-are you
'Whose child are you ?'
(b) *svili visi xar sena ?
( c) *visi xar sen svili?
( 1 6) illustrates similar facts, this time with romeli 'which'. Again, ( I 6a)
differs from (* I6b) and (* I6c) only in word order.
( 16) (a) romeli �igni ginda?
which book you-want
'Which book do you want ?'
(b) *9gni romeli ginda?
(c) *romeli ginda �igni? 8
Thus, for nominals of all kinds, the whole nominal must immediately
precede the verb, the questioned constituent preceding its head noun. 9
4.2.4 A constraint on Question Formation. As in English and other

languages, the question-word remains in its own clause in indirect
questions. ( I7b) illustrates an indirect question with a (direct) question
in the matrix clause also ; ( I 7a) illustrates the same embedded clause
without the matrix question.
( 1 7) (a) marinam icis, lp-eba sad ikneba.
Marina she-knows meeting where it-will-be
'Marina knows where the meeting will be.'
(b) icit, sad ikneba lp-eba?
you-know where will-be meeting
'Do you know where the meeting will bc ?'
In Georgian, Q-words i n direct questions cannot move out of

'r nbcdded clauses (cf. §4.z.5 for the exception to this generalization).
I n ( 1 8) and ( 1 9), the (*b) sentences are ungrammatical attempts to
move the questioned constituent into the matrix clause. The (*c) sen
tences show that it is also ungrammatical to leave the Q-word in the
'mbedded clause.
( I H) (a) (ase) vpikrob, rom nino moigebs.

thus I-think that Nino will-win
'I think that Nino will win.'
(b) *vin pikrob(,) (rom) moigebs ?
who you-think
(,Who do you think will win ?')
(c) *pikrob, (rom) vin moigebs?
('Who do you think will win ?')
I l) ( a) gelam tkva, rom mama favida kalakSi.
Gela said that father went city-in
' Gela said that Father went into the city.'
(b) ''Yfsad tkva gelam, (rom) mama favida?
where
(,Where did Gela say that Father went ?')
(c) *gelam tkva, (rom) mama sad favida.
(,Where did Gela say that Father went ?')
I I)h) is ungrammatical in the intended meaning, but has a grammatical

I t ' : I d i ng, on which the Q-word originated in the higher clause. The
'Iol l l l rnatical meaning, then, is 'Where was Gela when he said that
I ' , I wr left ?' ( 1 8-19) show that in Georgian it is impossible to question
I ('I ! I 1st ituent of an embedded clause with a direct question.
' I ' l l ' mea ni ng intended for (*1 8b) may be expressed by (zoa) or (zob),
( ." (u) ras pikrob ? vin moigebs?

what you-think who will-win
'Who do you think will win ? I What do you think ? Who will
win ?'
(b) rogor pikrob? vin moigebs?
how
'Who do you think will win ?'
� U( JW or it O a rguments to suggest that (2oa) is a single sentence, or

18 Introduction
that (2ob) is. Each has two intonation peaks, one on each Q-word : ras,
. .
Vtn, rogor, vtn.
4.2 5 An exception to the clause-boundedness of Question Formation. When

a clause is embedded in a matrix clause containing an independent
modal verb, a constituent of the embedded clause may be moved out,
contrary to the generalization stated in §4. 2.4. There are three such
independent, finite modals in Georgian : unda 'he wants it', seu3lia 'he
can, he is able', and s{irdeba 'he needs it'. (2 1 ) illustrates the question
ing of a constituent of an embedded clause and the movement of the
questioned constituent to the matrix verb. (2 Ia) is a declarative sen
tence, (2 Ib) a properly formed question, and (*2 IC) a question with
an improperly positioned Q-word.
(21) (a) es minda, rom ninom moigos.
this I-want that Nino will-win
'I want Nino to win.'
(b) vin ginda, rom moigos?
who you-want
'Who do you want to win ?'
(c) *(es) ginda, rom vin moigos?
(21) differs from ( I S) only in the governing verb. (The different form
of the verb of the dependent clause and concomitant change of case is
governed by the matrix modal verb. This should not concern the
reader now ; it is the topic of the main part of this monograph.)
4.2.6 Question Formation as a test of constituency. Keeping in mind the

single exception noted in §4 2.5, we can use Question Formation as a
.
test of constituency in a clause, since it generally applies only within a

clause, as shown in §4.2.4. This test will be used, for example, in ch. 5,
to show that the derived structure of an organic causative is a simple
sentence.
The fact that nominals are treated as inseparable wholes, as shown in
§4.2.3, provides us with a test for the constituency of a nominal. This
will be used, for example, to show in ch. 4 that the tvis-nominal ('for'
nominal) is not a constituent of the chomeur of the sentential subject
in object-raised sentences.
4.3 Suppletion
In Georgian there are certain verbs that arc supplctivl! for ani macy of
t he direct object, number of the direct object, or number of the sub

jl' 't. l O Three of these are described below, since they will form the
1 asis for arguments about grammatical relations in subsequent chap
t · rs . In addition, there are verbs that are suppletive for tense ; one of
t hese is described in §4.3.4, as it will be the basis of an argument in

ch. y
.1 . 3 . I Animacy of direct object. The verb mitana/raqvana 'take' is sup
I detive for the animacy of its direct object, such that the former occurs
I I n l y with inanimates and the latter only with animates, as shown in
..! ) and (23). In the (a) sentences, the direct object is animate, in the
h) sentences, inanimate. (22) contains the form miitana, and is gram

I " " t ical only with an inanimate direct object ; (23) contains raiqvana,
II l I d is grammatical with an animate.
l_) (a) *anzorma gela miitana studkalakSi.

Anzor Gela took student-compound-in
(,Anzor took Gela into the student compound. ')
(b) anzorma �o�a miitana studkalakSi.
J ug
'Anzor took the jug into the student compound.' .
. 3) (<1) anzorma gela raiqvana studkalakSi.
took
'Anzor took Gela into the student compound.'
(b) �' anzorma �o�a raiqvana studkalakSi.
('Anzor took the jug into the student compound.')
' ';::1 )
is grammatical if Gela is a corpse. In spite of this, these two
I I " I I I � do not have the meanings of 'carry' and 'lead', respectively, in
I I I ;I i:-;h. This difference shows up in (24), where the notion 'lead' would
I II I I 1 : 1 p p r op r iate , but where raqvana is grammatical, not m#ana.
I I ) (:I) 'ela mivitane mcxetasi mankanit/ UI'emit.
( ; cla I-took Mtsxeta-in car-with/ ox-cart-with
('1 took Gcla to Mtsxeta by car/by ox cart.')
( h) �ela t;:aviqvane mcxetasi mankanit/uremit.
I-took
. r took Gcla to Mtsxeta by car/by ox cart.'
1.,1 '0\1 'r, (aqvana, not m#ana, is used with the direct object bavsvi
tht child i fl krl or carried :
I l I l d ' , W I H.:tl lcr
20 Introduction
(25) *bavSvi mivitane.

child . I-took
('I took the child.')
(26) bavsvi �aviqvane.
I-took
'I took the child.'
The noun mankana 'car' is a systematic exception - a grammatical
animate - with all verbs that are suppletive for the animacy of their
direct objects.
(27) (a) *mankana mivitane mcxetaSi.
car I-took Mtsxeta-in.')
(b) mankana �aviqvane mcxetaSi.
I-took
' I took the car to Mtsxeta.'
The same suppletion occurs for this verb root with other preverbs,
verbal prefixes which indicate, among other things, differences of
direction ; for example, setana/seqvana 'take in', motana/moqvana 'bring'.
Most of the examples will use the preverbs used in the example sen
tences above.
4.3.2 Number of direct object. The verb gadagdeba/gadaqra 'throw' is

suppletive for the number of its direct object, such that the former
form occurs with singular direct objects, and the latter only with
plurals. These facts are established by (28) and (29), which contain
gadagdeba and gadaqra, respectively. The (a) sentences have singular
direct objects, the (b) sentences plurals.
(28) (a) bavsvma gadaagdo kva.
child he-threw stone
'The child threw the stone.'
(b) *bavsvma gadaagdo kvebi.
stones
('The child threw the stones.')
(29) (a) *bavsvma gadaqara kva.
he-threw
(,The child threw the stone.')
(b) bavsvma gadaqara kvebi.
'The child threw the stones.'
4. 3.3 Number of subJect. The verb Jdoma/sxdoma 'sit' is suppletive for

the number of its subject, such that the former occurs with singular
subjects and the latter only with plurals. ll (30) shows that this is so:
(*30a) has a singular subject, (30b) a plural.
(30) (a) *bavsvi dasxda.
child he-sat-down
(,The child sat down.')
(b) bavsvebi dasxdnen.
children they-sat-down
'The children sat down.'
4 . 3.4 Tense/Class. The verb 'take' discussed in §4.3.2 is actually multiply

m
! ppletive - for animacy of its direct object and for tense. In some
tenses the verb is a Class I verb, in some Class 4. The distribution of
these characteristics and of the roots is summarized in the chart below.
' :inite verb forms are given so that the different roots will show up ; the
roots are in heavy type. All forms express 'I take it/him' in the various
( l ·oses.
Inanimate Object Animate Object

Present mimakvs Class 4 mimqavs Class 4
Aorist mivitane Class I {:aviqvane Class I
, " he fact that the final grammatical relations in the Class 4 verb forms
d i f l'cr from those in the Class I forms will be used as one argument to
1 I10ti vate the rule of Inversion in ch. 8.
, ." Number Agreement

','or many speakers, only animate nominals trigger Number Agreement,
l \ ' 1 1 if the nominal otherwise meets the conditions to be a Number
\ ' 1 " m e n t trigger (ch. IS ) . This can be seen in (32), where the animate
'
1 I ! l i cet in (a) triggers Number Agreement, but the inanimate subject

I I I ( h) fails to do so.
C \ .! (a) (mutebi goraven.

kittens they-roll
' The kittens are rolling.'
( h) bu.rtebi goravs.
balls it-rolls
'The balls are rolling. '
22 Introduction
Although this dichotomy is required by prescriptive norms, the rule is

not always followed by speakers. Most of the examples given here do
follow the rule, and the verb is glossed in the singular when a plural
fails to trigger Number Agreement, as in (32b). Nominals that occur
with a quantifier are in the singular in Georgian and also fail to trigger
Number Agreement, as shown in (33).
(33) (a) sami 1!-nuti goravs.
three kitten it-rolls
'Three kittens are rolling.'
(b) ramdeni 1!-nuti goravs.
how-many
'How many kittens are rolling ?'
4.5 Word order

There is considerable freedom of word order in Georgian. The follow
ing orders are very frequent and natural : S V DO 10, S DO V 10,
and S 1 0 DO V. It is clear that the subject usually precedes other
major constituents, but it is not clear that Georgian is more 'basically'
either OV or VO. A detailed discussion of word order can be found in
Vogt ( 1974) and in works cited there.
I Syntactic tests for termhood
In this chapter I will introduce several tests for termhood (subjecthood,

direct-objecthood, indirect-objecthood).
In Georgian the problem of isolating the notion 'subject' is particu
larly complex. As discussed in the Introduction, § I , an adequate analysis
cannot be based on a single criterion. In this chapter, preliminary tests
for subject are introduced ; additional tests for subjects are defined in
later chapters. An adequate analysis of the notion 'subject' must also
take into consideration all Classes of verbs and all Series (cf. Intro
duction, §I). In this chapter, I deal with Classes 1-3 in Series 1. I n
ch. 2, Series II will b e introduced. Treatment of Series III and Class 4
verbs, both of which govern Case Pattern C, is deferred until ch. 8.
I Tav-Reflexivization
In this section I will examine the rule that governs the occurrence of
the reflexive pronoun tav-. It will be shown that this pronoun may be
coreferent only to the subject of its clause and that it therefore serves as
a test of subjecthood. 1
I.I Morphology
Reflexivization is expressed in Georgian with the pronominal element
tav-. This root also functions as an independent noun meaning 'head'
or 'source' ; it will be glossed here as 'self' when it functions as a
reflexive. The forms tavis- (singular) and taviant- (plural) serve as
possessive reflexives. The first person singular reflexive is cemi tavi
'my self' , the second seni tavi 'your self', and the third person reflexive
is tavisi tavi 'self's self', that is, 'himself, herself, itself', without dis
tinction of gender. 2 The possessive element, tavisi, is deleted from the
third person reflexive under certain circumstances (cf. ch. 6). The
reflexive phrase is grammatically third person and triggers third person
24 Syntactic tests for termhood
agreement. The conditions on the independent occurrence of tavis-f

taviant- are different from those on tav-, and they are described in
§2 below.
1 .2 Subjecthood of the triggn

The reflexive pronoun tavis always coreferent with the subject of its
clause. That the subject may be coreferent with tav- can be seen in
( I) (a) vano irr;munebs tavis tavs. 3

Vano-NOM he-convinces-him-I-I self's self-DAT
'Vano is convincing himself.'
(b) vano elapara1!-eba tavis tavs.
VanO-NOM he-talks-him-I-2 self's self-DAT
'Vano is talking to himself.'
(c) vano pikrobs tavis tavze.
VanO-NOM he-thinks-I-3 self's self-on
'Vano is thinking about himself.'
(2), (3), and (4) show that the reflexive pronoun is not coreferent with
the direct object, the indirect object or the object of a postposition.
(2) mxatvari daxatavs vanos tavistvis.

painter-NOM he-paints-him-I- I Vano-DAT self-for
'The painter! will paint Vanoj for himselfi.'
(3) (a) nino acvenebs p atara gZVts tavis
NinO-NoM she-shows-him-him-I-I little Givi-DAT self's
tavs sar1!-eSi.
self-DAT mirror-in
'Ninoi is showing little Givij herselfi in the mirror.'
(4) vano gzvzze elapara1!-eba tavis tavs.
Vano-NOM Givi-on he-talks-him-I-3 self's self-DAT
'Vanoi is talking to himselfi about Givij .'
,-
In (2) tav- can refer only to mxatvari, not to vanos; the direct object,
vano, cannot trigger tav-. Similarly in (3a), tav- refers only to nino, the
subject, never to givis, the indirect object. The coreference that obtains
between the subject and tavis independent of the relative positions of
the pronoun and antecedent, as can be seen in (3b) and (3c), which
differ from (3a) only in word order.
I.I Tav-Reflexivization 25
(3) (b) m'no p atara givis tavis tavs acvenebs sar�eSi.

(c) sar�esi p atara givis tavis tavs acvenebs nino.
'Ninoi shows little Givij herselfi in the mirror.'
In ( 4) the reflexive pronoun can refer only to the subject, vano, never to
givi, an oblique nominal.
Since the coreference of tav- with the subject has been illustrated for
a Class I verb in ( I a), a Class 2 verb in ( I b), and a Class 3 verb in ( I C),
we can conclude that in general tavis triggered by the subject-nominal. 4
1 .3 Clausemate constraint
The rule of Tav-Reflexivization is governed by a clausemate constraint,
which requires that the pronoun and its antecedent be dependents of
the same clause. Thus in (sa) there is no ambiguity as to the meaning
reference of tav- ; it can refer only to nino, the subject of its own clause.
(s) (a) vano pikrobs, rom nino sa{mels
Vano-NOM he-thinks-I-3 that Nino-NoM food-DAT
amzadebs tavistvis.
she-prepares-it-I-I self-for
'Vanoi thinks that Ninoj is preparing food for herselfj .'
In spite of the fact that vano is also a subject, it cannot be the ante
cedent of tav-, since these two nominals are not dependents of the same
clause. We can see from (Sb) that this is independent of relative word
order, since here tav- precedes nino, yet nevertheless is unambiguously
coreferent to it.
(s) (b) vano pikrobs, rom tavistvis nino sa{mels amzadebs.
'Vanoi thinks that Ninoj is preparing food for herselfj.'
The lack of ambiguity in ( 5) shows that tav- in a postpositional phrase
cannot be triggered by the subject of a higher clause ; the ungram
maticality of (*6) shows that this is not possible, even where identifying
the reflexive pronoun with the embedded subject would produce
nonsense.
(6) *vano icvams axal �ostiums, romelic
Vano-NOM he-puts-on-it-I-I new suit-DAT which-NOM
tavistvisaa se�rili.
self-for-it-is-I-z sewn
(' Vano is putting on the new suit, which was made for him.' )
There i s no sensible grammatical reading for (*6).
The ungrammaticality of (*7) shows that it is equally impossible for

the reflexive pronoun to be the subject of the lower clause, triggered
by the subject of the higher clause.
(7) (a) *vano pikrobs, rom tavlSl tavi tamada

Vano-NoM he-thinks-I-3 that self's self-NoM toastmaster-NOM
ikneba.
he-will-be-I-2
('Vano thinks that he will be toastmaster.')
(b) *rom tavisi tavi tamada ikneba, ase pikrobs vano.
thus
('Vano thinks that he will be toastmaster.')
(*7b) differs from (*7a) only in that in the latter the lower clause is
first, the pronoun thus preceding its antecedent.
Similarly, tav- in the higher clause cannot be coreferential to the
subject of the lower clause :
(8) (a) *tavisi tavi pikrobs, rom vano

self's self-NOM he-thinks-I-3 that VanO-NOM
tamada ikneba.
toastmaster-NOM he-will-be-1-2
(*'Himself thinks that Vano will be toastmaster.')
(b) *rom vano tamada ikneba, ase pikrobs tavisi tavi.
thus
(*8b) differs from (*8a) only in that in the former the embedded clause
is first, and the trigger thus precedes the target of reflexivization. In
(*8) tav- in the matrix clause is triggered by vano, the subject of the
embedded clause.
1 .4 Coreference with dropped pronouns

The nominal with which the tavis coreferent may be omitted in the
final structure of the sentence, if its referent has been established in the
discourse. This is accounted for by the rule of Unemphatic Pronoun
Drop, which is described in §4 below. It is nevertheless clear that tavis
coreferent to the subject, for the person and number of the pronoun
correspond to those of the subject marked in the verb form (cf. §3 on
marking subjects in the verb). This can be seen in the glosses of (9),
which represent the only possible interpretation of these sentences.
I.2 Tavis-Rejlexivization 27
(9) (a) cerns tavs vakeb.

my self-DAT I-praise-him-I- I
' I praise myself.'
(b) sens tavs akeb.
your-sG self-DAT you-praise-him-I- I
'You (sg) praise yourself. '
(c) tavis tavs akebs.
self's self-DAT he-praises-him-I-I
'He praises himself. '
1 . 5 Summary
Since the full pronoun tav- may have as its antecedent only the subject
of its own clause, it provides us with a test for subjecthood within a
single clause.
2 Tavis-Reflexivization 5
In this section, the rule that governs the occurrence of the possessive
reflexive pronoun will be described. The distribution of tavis-ftaviant- is
different from that of the full pronoun tav- ; we must therefore account
for these phenomena in terms of two distinct rules. It will be shown
that tavis-ftaviant- may be coreferent only with terms ; it therefore
serves as a test for termhood.
Any third person term can trigger tavis-ftaviant-. The subject is the
trigger in ( 10), the direct object in ( I I ), and the indirect object in ( 1 2) .
( 1 0) deda bans tavis svils.
mother-NOM she-bathes-him-I-I self's child-DAT
'The motheri is bathing heri child.'
( I I ) mno a3ievs bavsvs tavis dedas.
Nino-NOM she-gives-her-him-I-I child-OAT self 's mother-oAT ·
'Nino; is giving the childj to itsj mother.'
'Ninoi is giving the childj to herl mother.'
( t :.i.) ras acukebs gela ias tavis
what-oAT he-gives-her-it-I-I Gela-NOM la-oAT self's
dabadebisdyeze?
birthday-on
'What does Gelai give laj on herj birthday ?'
'What does Gela! give laj on his! birthday ?'
In ( I I ) and ( 1 2) either term may be the antecedent. Thus, any term may
be coreferential with tavis-.6
A non-term cannot trigger Tavis-Reflexivization, as established by
(13) gela saubrobs giastan tavis fignze.

Gela-NoM he-chats-I-3 Gia-at self's book-on
'Gelai is chatting with Giaj about hisi book.'
( 14) babua J:.maqopilia (misi) sviliSvilit
grandfather-NOM pleased-he-is-I-z his grandchild-INST
tavis korfilsi.
self's wedding-in.
'The grandfatheri is pleased with his! grandsonj at his! wedding.'
In ( 13) tavis- cannot refer to gia, which is the object of a postposition

and not a term. Nor can the non-term instrumental-NP, svilisvilit,
trigger the possessive reflexive in ( 14), even though it is semantically
more natural that way than with the grammatical reading. Thus, we
see that all and only terms can serve as antecedents of tavis-.
3 Person Agreement
In Georgian any one of the terms governed by a verb may trigger

Person Agreement ; subjects trigger Subject Person Agreement, direct
objects trigger Direct Object Person Agreement, and indirect objects
trigger Indirect Object Person Agreement. The term 'Person Agree
ment' will be used to refer to these three rules together, since they
apply under similar circumstances. Each rule serves as a test for specific
termhood.
3 . 1 Description of the basic morphology of Person Agreement

Paradigms ( I S), ( 1 6), and ( 1 7) represent the intersection of the cat
egories of person and number. But these two agreement categories
must be described by distinct rules in Georgian, since the conditions
under which a given nominal triggers Number Agreement are different
from the conditions under which it triggers Person Agreement. The
rules of Person Agreement are simple and can be described now, but
the rule of Number Agreement must take into account various other
factors and will not be described until ch. 1 5. 7
I.3 Person Agreement 29
( I S) Subject markers
singular plural
1 . person v- v-t
2. person " -t
3 . person -s/a/o -en/es/nen, etc.
( 1 6) Direct object markers

I . person m- gv
2. person g g-t
3 . person " "
( 1 7) Indirect object markers

I . person m- gv
2 . person g g-t
3 . person s/h/0- s/h/0-t8
. rhe dash represents the position of the verb stem, which includes the
root and various formants. The alternation between third person sub
ject markers is governed by the tense-aspect and Class of the verb form.
. fhe alternation in third person indirect object markers is phonologi
-ally determined. The latter alternation, first clarified in Shanidze
( [ 9 20), is also described in Tschenkeli ( 1 95 8 : vol. I, 370-2), Vogt
( [ 97 1 : 83), and Shanidze (1973 : 1 84-5) ' For some speakers the third
person indirect object marker is always zero, or has become part of the
vcrb root, thus making (16) and ( 17) nearly identical. Because the use
of the third person indirect object markers is not consistent, I will not
l i se the difference between them to argue that a given nominal is a
d i rect object as opposed to an indirect object, or vice ve rsa. But the

Il larkers in ( 1 6) and (17) will be used to argue that a given nominal is
a n object, without distinguishing on the basis of marking alone between
d i rect and indirect. Because of the similarity between ( 1 6) and ( 1 7) and
t i l - confusion between them, they will often be referred to jointly as
'olJject ll1arkers'. Since I had informants who adhered strictly to the
I'I dcs distinguishing ( 1 6) and ( I7), and others who did not distinguish
t I i - m , my examples will not necessarily be consistent ; but these facts
wi I I never be crucial to any argument.
Bclow, ( 1 8) illustrates the use of subject markers, ( 19) the use of
I i i ,. 'ct object markers, and (20) the use of indirect object markers ; each
( I ' ' u rs with verbs of Class J, 2, and 3 .
( 1 8) (a) namcxvars v-acxob.

pastrY-DAT I-bake-it-I- I
'I am baking pastry.'
(b) namcxvari cxveb-a
pastry-NoM it-bakes-I-2
'The pastry is baking.'
(c) stven?
you-whistle-I-3 (zero marker)
'Are you whistling ?'
(19) (a) gela g-irvevs supraze.
Gela-NoM he-invites-you-I-I table-on
' Gela is inviting you to the banquet.'
(b) mama motxrobas uqveba nmos.
father-NoM story-DAT he-tells-her-it-I-2 Nino-DAT (zero)
'Father tells Nino a story.'
( c) ia ipovis satvales.
la-NoM she-will-find-it-I-3 glasses-DAT (zero marker)
'Ia will find (her) glasses.'
(20) (a) mi-s-cems sasmels stumars.
he-gives-him-it-I-I drink-DAT guest-DAT
'He is giving drink to the guest.'
(b) mama motxrobas g-iqveba.
father-NoM storY-DAT he-tells-you-it-I-2
'Father is telling you a story.'
(c) gela m-pasuxobs.
Gela-NoM he-answers-me-I-3
'Gela is answering me.'
The paradigms ( 1 5-1 7) represent an abstraction of the categories of

person and number from all verb forms. Derived indirect objects show
some superficial variations from ( 17) ; these are described with their
morphology in chs. 6 and 8.
3.2 Conditions on the occurrence of subject and object markers

In Georgian, Person Agreement may be triggered by any term of the
verb. Yet the · reader will not always see the morpheme that marks
agreement with a particular term. This may be due to any one of several
factors. First, the zero markers in ( 1 5-17) may disguise agreement.
I.3 Person Agreement 31
Second, there are morphophonemic rules that govern the co-occurrence

of particular affixes :
(21) + v + -7- 0/- +g+
The morpheme v deletes before the morpheme g.
+s+ -+ 0/+V+ -
The morpheme s or its variant h deletes following the morpheme v.
(22) +s + -7- 0/ - + t +
The morpheme s deletes before the morpheme t .
( 2 1 ) describes the co-occurrence of personal prefixes ; v i s the first per
son subject marker, g the second person object marker, and s or h the
third person indirect object marker. (22) describes the co-occurrence
of the suffixes s, third person subject, and t, a marker of plurality. 9
Third, there are syntactic rules which affect the occurrence of
person markers on final forms of the verb. The first such rule is Tav
Reflexivization, which is described above in § I . The reflexive pronoun
is itself third person, though it may be coreferential to a first or second
person. Being third person, it triggers a zero marker in most instances
(cf. paradigms 1 6-17). This rule thus insures that we never get
sequences like *v-m-, where v is a first person subject marker, and
m is a first person object marker. The second syntactic rule that affects
the occurrence of person markers will not be discussed in detail until
ch. 3, but it is stated here for the reader's convenience.
(23) Object Camouflage
If a clause contains an indirect object, a first or second person
direct object is realized as a possessive pronoun + tavi, where the
possessive reflects the person and number of the input form.
[ Iere, too, the possessive + tavi phrase is third person and triggers
t. hird person agreement. Thus, this syntactic rule insures that we
never get sequences like *g-m-, where g and m are second and first
[lnson object markers, respectively.
Because of the conditions described above, there is not always a one
to-one correspondence between terms and their markers in the verb
form. But every final term triggers Person Agreement, the realization
I . f which is specified by ( 1 5-17), (21-23), Tav-Reflexivization, and the
l o w level rules required to select the forms of third person subject
I n a rkers, depending on plurality, tense-aspect, Class, and morpho
I ( )�ical sub-group of the verb form.
3.3 Implications of Person Agreement

Keeping in mind the conditions (2 1-23) and Tav-Refiexivization, we
can determine which nominal in a clause triggers Subject Person
Agreement, and which ones trigger Object Person Agreement. Subject
Person Agreement then serves as a test for subjecthood : only subjects
trigger the markers in (1 5). Object Person Agreement will not be used
as a test for specific termhood, since the conditions on it vary with the
speaker, but it does provide a test for general objecthood.
4 Unemphatic Pronoun Drop
Compare the sets of sentences (24) and (25). The examples show that
mi(:era/mo(:era 'write' takes an indirect object, and is always interpreted
as having an indirect object even when there is none on the surface.
Da(:era 'write', on the other hand, does not take an (initial) indirect
object, and is interpreted as having a direct object, but not an indirect. 1 0
( 24) (a) (:erils miv(:er 3mas.
letter-DAT I-will-write-him-it-I-I brother-DAT
' I will write a letter to my brother.'
(b) (:erils miv(:er.
' I will write a letter to him.'
(c) miv(:er.
' I will write it to him.'
(25) (a) *(:erils dav(:er 3mas.
I-will-write-it-1- 1
('I will write a letter to my brother.')
(b) (:erils dav(:er.
' I will write a letter.'
( c) dav(:er.
' I will write it.'
The root of both verbs is (:er ; da-, mi-, and mo- are preverbs, the
presence of which indicates completed aspect. The difference between
mi- and mo- is determined by the grammatical person of the indirect
object (cf. ch. 6, example ( 19» ; first and second person require mo-,
third mi-. In addition, the difference between da- on the one hand, and
mi- and mo- on the other, corresponds to the glossed difference between
(24) and (25) ·
To account for the interpretation of the non-overt terms i n (24 25),
I4 Unemphatic Pronoun Drop 33
I propose (i) the lexical specification that mi�era/mo�era takes an obli

gatory initial indirect object, while da�era does not take an initial
indirect object, and (ii) a rule of Unemphatic Pronoun Drop, which
applies only to terms.ll The rule drops first, second, and third person
independent personal pronouns unless they are either emphatic, or
else non-terms. In the following sections I will demonstrate that such
a rule affects only terms. Additional evidence to support a rule of
Unemphatic Pronoun Drop is given in ch. 3, n. 3 .
4. 1 Motivating Unemphatic Pronoun Drop and the proposed inventories

of initial terms
4. 1 . 1 Person Agreement. The rule of Unemphatic Pronoun Drop is
needed to account for the fact that verbs are marked to agree with a
particular person, even when the corresponding nominal is absent
from the surface structure. The verb form v�er 'I write', for example,
will always be associated semantically with a first person subject, not
a second or third. Similarly, when the subject pronoun is stressed, we
necessarily get me v�er 'I write', not ""sen v�er 'you I-write', or ""is v�er
'he I-write' .
The inventories of initial terms proposed above are necessary to
explain the fact that mi�era/mo�era requires Indirect Object Person Agree
ment, while da�era does not. This fact is established in (26-29). (26), in
which mi�era/mo�era occurs with indirect object markers, is grammatical.
( 26) (a) vanom mom�era �erili.
Vano-ERG he-wrote-me-it-II-I letter-NOM
'Vano wrote me a letter.'
(b) vanom mog�era �erili.
he-wrote-you-it-I I I-
'Vano wrote you a letter.'

(c) vanom mis�era �erili.
he-wrote-him-it-II-I
'Vano wrote him a letter.'
(-27), where there are no markers of indirect object agreement, IS
I I ngrammatical.
(27) (a) ""vanom mo�era �erili.

(,Vano wrote me/you a letter.' )
(b) o¥. ·uanom mi�era �erili. 1 2
(' Vano wrote him a letter.')
34 Syntactic tests for termlwod
With the verb dafera, on the other hand, the indirect object agreement
marker must not occur. This is shown by the ungrammaticality of the
examples in (*28), which are parallel to (26), and the grammaticality of
the example in (29), which is parallel to (*27).
(28) (a) *vanom damfera ferili.
('Vano wrote me a letter.')
(b) *vanom dagfera ferili.
('Vano wrote you a letter.')
( c) *vanom dasfera ferili.
(,Vano wrote him a letter.')
(29) vanom dafera ferili.
'Vano wrote a letter.'
These four sets of sentences show that mifera/mofera must take an
indirect object marker and that dafera must not. This can be accounted
for in a straightforward manner if (i) these verbs are assigned the
initial terms claimed for them above, and (ii) Unemphatic Pronoun
Drop drops the pronominal indirect objects in (26) after they have
triggered Indirect Object Person Agreement. Since dafera will not
have an initial indirect object, nothing will trigger the rule, and the
sentences of (*28) will not be produced.
4.1.2 Number Agreement. Although the rule of Number Agreement

has not yet been introduced, we may observe here that first person
terms always trigger the rule (cf. ch. 1 5). Unemphatic Pronoun Drop
is needed to account for the fact that the meaning of the form vfert is
always 'we write', not 'I write' . Similarly, this rule explains the fact
that when the subject is emphatic, it is always the pronoun even 'we'
that occurs with this verb form, not me 'I'.
4.1.3 Tav-Rejlexivization. The rule of Unemphatic Pronoun Drop is

needed to explain why we get (30a), but not (*30b, c). In the first, the
verb agrees with a first person subject and has a first person reflexive.
The two ungrammatical examples also have first person Subject Person
Agreement, but have second and third person reflexives, respectively.
(30) (a) tems tavs vakeb.
my self-DAT I-praise-him-I-I
'I praise myself.'
I.4 Unemphatic Pronoun Drop 35
(b) *fens tavs vakeb.

your self-nAT
(*'1 praise your self.')
(c) *tavis tavs vakeb.
self 's self-nAT
(*'1 praise himself.')
In order to account for the ungrammaticality of (*30b) and (*30c), we
must have initial structures with personal pronouns, a rule of Tav
Reflexivization, and a rule of Unemphatic Pronoun Drop.
4. 1 ·4Tavis-ReJlexivization. In (3 Ia) below, the possessive reflexive

tavis- is triggered by vano ; but in (3 1 b), where the reference of tavis- is
the same, the antecedent is not present.
( 3 1 ) (a) vanos vugzavni tavis (:igns.
Vano-nAT I-send-him-it-I-I self 's book-nAT
' I am sending Vanoi hisi book.'
(b) vugzavni tavis rigns.

'I am sending himi his; book.'
Since tavis- is triggered by a coreferential nominal, a pronoun mas
'him-nAT' must be posited for initial structure, with a rule of Unem
phatic Pronoun Drop to account for the absence of the pronoun in (3 1 b )
'1- . 2 . 1 Terms. (30) illustrates the dropping of the subject pronoun, (3 1 )

the dropping of the indirect object. (32) shows that the direct object
pronoun is also dropped.
(32) cenzs megobars, gelas, icnob? �i, vicnob.
yes I-know-him-I-I
'Do you know my friend, Gela ? Yes, I know him.'
The inclusion of the pronoun mas 'him' would imply a contrast, for
'xample : �i, mas vicnob, mis das �i ara. 'Yes, I know him, but not his
sister . '
4.2.2 Instrumentals. Instrumentals can never b e dropped. (33) IS a

Rtatcment with a nominal in the instrumental case.
(33) xmas cav(:er magnitaponit.
voice-vAT I-will-write-it-I- I tape-recorder-INsT
' 1 will record (my) voice.' Lit : 'I will write my voice with a tape
recorder.'
In reply to the question, es magnitaponia? 'Is this a tape recorder ?',

(34a) is a possible response ; (34b), while grammatical, is unnatural
because it makes no reference to the tape recorder, as indicated in the
translation.
(34) (a) lJi, xmas cav�er amit.
yes it-INST
'Yes, I will write my voice with it.' i.e. 'I will record my voice
with it.'
(b) lJi, xmas cav�er.
'Yes, I will write down my voice.'
This shows that instrumentals cannot be dropped, even when they are
unemphatic pronouns.
4-2.3 Nominals marked with a postposition. (35) illustrates the use of

the complex postposition -tan ertad 'with, together with'.
(35) mcxetali �avedi gelastan ertad.
Mtsxeta-in I-went-II-2 Gela-with
'I with to Mtsxeta with Gela.'
In answer to the question, gelas icnob ? 'Do you know Gela ?', (36a)
would be a natural response ; but (36b), while grammatical, would be
highly unnatural as a response.
(36) (a) diax, �avedi mcxetaSi mastan ertad.
yes him-with
'Yes, I went to Mtsxeta with him.'
(b) diax, �avedi mcxetali.
'Yes, I went to Mtsxeta.'
(36b) shows that the postpositional phrase containing mas cannot be
dropped ; (*36c) shows that it is also impossible to drop the unemphatic
pronoun out of a postpositional phrase, leaving the postposition.
(36) (c) *�avedi mcxetali tan ertad.
(*'1 went to Mtsxeta with.')
4.2.4 Conjoining. In (37), the subject is ia da gela 'Ia and Gela', not
ia, and not gela.
(37) ia da gela �avidnen mcxetali.
la-NOM and Gela-NoM they-went-II-2 Mtsxeta-in
'Gela and Ia went to Mtsxeta.'
I.4 Unemphatic Pronoun Drop 37
Even when the referent is established in discourse, as in answer to a

question, one of these conjoined nouns cannot be dropped. For example,
in answer to the question ia sad aris? 'Where is Ia ?', a possible answer
would be (38a) ; but (*38b) with one conjoined nominal omitted is
totally ungrammatical.
(38) (a) igi da gela (:avidnen mcxetaSi.
she-NOM
'She and Gela went to Mtsxeta.'
(b) *da gela (:avidnen mcxetaSi.
(,She and Gela went to Mtsxeta.') 13
Changing the order of elements will not save (*38b), as (*38c) shows.
(38) (c) *gela da (:avidnen mcxetasi.
('Gela and she went to Mtsxeta.')
The subject of (37), if its referent has been established in discourse,
and if it is unemphatic, can be dropped, as in (39).
(39) (:avidnen mcxetaSi.
'They went to Mtsxeta.'
The sentences above show that while terms may be dropped, a member
of a conjoined set that constitutes a term may not be dropped.
4.2.5 Possessives. The subject of (40) is givis 3ma ' Givi's brother'.
(40) givis 3ma cemi megobaria . .
Givi-GEN brother-NoM my friend-NoM-he-is-I-z
'Givi's brother is my friend.'
r f the referent of givi has been established in discourse, it may b e
pronominalized, but not dropped, as shown in (41). I n this case,
hecause of the necessity of interpreting the relation 'brother' as belong
ing to someone, the meaning of the sentence is entirely different if the
possessive pronoun is dropped ; but (4Ib), lacking the possessive, is not
natural.
(4 1) (a) misi 3ma cemi megobaria.
his
'His brother is my friend.'
(b) ?3ma cemi megobaria.
'My brother is my friend.'
I conclude that although subjects may be dropped, a possessive that is

a constituent of a subject may not be dropped.
4.3 Conclusion
We have seen that the rule of Unemphatic Pronoun Drop applies to
terms generally and not to a variety of nominals that are not terms ; I
conclude, therefore, Unemphatic Pronoun Drop can be used as a test
for termhood, since it applies only to terms. 14
5 Summary
In this chapter I have introduced four processes that provide us with

several tests for termhood. While none of these traits is universally
characteristic of terms, all of them are typical of terms (cf. Anderson
1 976 ; Keenan 1974, 1976 ; and Postal & Perlmutter 1974). The level at
which the nominals identified by these tests are terms has not yet been
specified. This will be considered in later chapters, as we consider the
interaction of these rules with other rules of the grammar.
Chapter 2 will be concerned mainly with a different problem, but a
test for termhood will be a by-product of that analysis also. Chapters 3
and 4 will describe two rules, each of which is a further test for direct
objecthood. In later chapters, the tests discussed here will provide
evidence as to the termhood of particular nominals in more complex
constructions.
2 Case marking in Series I and II
All of the examples in ch. 1 were in tenses which belong to Series I

(d. Appendix, 'Constituent screeves of Georgian Series'). Sentences
with verbs in tenses of Series II exhibit different patterns of case
marking. In ch. 1 I established that the nominative-nominals of
clauses containing verbs of Classes 1-3 in Series I share a set of syn
tactic properties, which are associated with the notion 'subject'. I n
Series I I we might expect to find (i) that the nominative-nominal has
those properties, (ii) that the nominal which corresponds to the intuitive
notion 'subject' has the subject properties, regardless of its case, or
(iii) that the subject properties are divided between those two nominals.
A variety of different approaches have been taken with regard to this
problem (cf. Introduction, §1). In this chapter I will present syntactic
evidence to support (ii) ; that is, I will show that in Series II the nominal
which corresponds to the intuitive notion 'subject' has all of the syn
tactic subject properties established in ch. 1 .
I The case marking differential
The sentences in ( 1 ) contain intransitive verbs in Series I ; those in (2)

contain transitive verbs in Series I. The (a) sentences contain Class I
verbs, the (b) sentences Class 2 verbs, and the ( c) sentences Class 3
verbs. In each, the subject is in the nominative, the direct and indirect
objects are in the dative.
( I ) (a) nino amtknarebs.

NinO-NoM she-yawns-I- 1
'Nino yawns.'
(b) vaxtangi ekimia.
Vaxtang-NoM doctor-NoM-he-is-I-2
'Vaxtang is a doctor.'
(c) vano pikrobs mari1!-aze.
Vano-NOM he-thinks-I-3 Marika-on
'Vano is thinking about Marika.'
(2) (a) nino acvenebs suratebs gias.
Nino-NOM she-shows-him-it-I-I pictures-DAT Gia-DAT
'Nino is showing pictures to Gia.'
(b) mama uqveba motxrobas ninos
father-NOM he-tells-her-it-I-2 story-DAT Nino-DAT
' Father is telling Nino a story.'
( c) kartuli ena sesxulobs sitqvebs
Georgian language-NoM it-borrows-it-I-3 words-DAT
rusulidan.
Russian-from
'The Georgian language borrows words from Russian.'
(3) and (4) give the same sentences in the aorist, which belongs to
Series II. Notice the different cases used here.
(3) (a) ninom daamtknara.
NinO-ERG she-yawned-II-I
'Nino yawned.'
(b) vaxtangi ekimi iqo.
Vaxtang-NoM doctor-NOM he-was-II-2
'Vaxtang was a doctor.'
(c) vanom ipikra mari1!-aze.
VanO-ERG he-thought-II-3 Marika-on
'Vano thought about Marika.'
(4) (a) ninom acvena suratebi gias.
NinO-ERG she-showed-him-it-II-I pictures-NoM Gia-DAT
' Nino showed the pictures to Gia.'
(b) mama mouqva motxrobas ninos
father-NOM he-told-her-it-II-2 story-DAT Nino-DAT
'Father told a story to Nino.'
(c) kartulma enam isesxa sitqvebi
Georgian language-ERG it-borrowed-it-II-3 wordS-NOM
rusulidan.
Russian-from
'The Georgian language has borrowed words from Russian.'
In the sentences which contain Class 2 verbs, the case pattern is the
2.2 The analysis of case in Series II 41
same in Series I I as in Series I, as can be seen by comparing (3-4b)

with ( I-2b). (3a, c) and (4a, c) show that in sentences that contain verbs
of Class I or 3 , on the other hand, the case pattern is different in Series
II from that used in Series I, regardless of the transitivity of the verb.
(The question of whether the transitivity of the verb is relevant to case
marking in Georgian is addressed in chs. 12 and 16.) In §2 below, I
will present syntactic arguments to show that in clauses containing
Class I or 3 verbs in Series II, the ergative-nominal is subject. The
exact correspondences I am arguing for are summarized in (5).
(5 ) Georgian Series II
Subject Direct Object Indirect Object
Class 1 ERGATIVE NOMINATIVE DATIVE
2 NOMINATIVE DATIVE DATIVE
3 ERGATIVE NOMINATIVE DATIVE
2 Evidence for the analysis of case in Series II
In this section I will show that, in spite of the case marking differential,
the subjects of sentences containing Series II verbs correspond to the
intuitive notion of subject and to the subjects of Series I. These argu
ments are based on the tests for termhood established in ch. I . In each
instance I will demonstrate that for each verb Class, the ergative
nominal, the nominative-nominal, and the dative-nominal have all the
properties of the grammatical relation indicated for them in (5).
2. I Tav-Rejiexivization
J t was established in ch. 1 that tav- can be coreferential only to the
�ubject of its clause. In (6) we see that tavis coreferential to the nom
inal bearing the case appropriate to the subject as stated in (5). That is,
fav- is coreferential to the same nominal as in Series I, though its case
is difft:rent in Series II. (6) is parallel to ( I) of ch. I ; only changes in

I I H! tense and concomitant changes in case marking have been made.
(6) (a) vanom dairf:muna tavisi tavi.

VanO-ERG he-convinced-him-I I-I self's Self-NOM
'Vano convinced himself.'
( h) 'vano elapara�a tavis tavs.
Vano-NoM he-talked-him-II-2 self's self-DAT
'Vano talked to himself.'
( c) vanom ipikra tavis tavze.

Va nO-ERG he-thought-II-3 self's self-on
'Vano thought about himself.'
Thus, tavis coreferent to the ergative-nominal in (6a, c) and the
nominative-nominal in (6b). In each instance, the antecedent is the
nominal claimed in (5) to be subject.
The sentences in (7) show that tav- cannot be coreferential to nom
inals other than the subject. The sentences are parallel to (2), (3a), and
(4) , respectively, of ch. I .
(7) (a) mxatvarma daxata vano tavistvis.
painter-ERG he-painted-him-II-I VanO-NOM self-for
'The painter! painted Vanoj for himself!.'
(b) ninom acvena p atara givis tavisi
NinO-ERG she-showed-him-him-II-I little Givi-DAT self's
tavi sar�si.
self-NOM mirror-in
'Ninol showed little Givij herself; in the mirror.'
(c) vano gzvzze elaparaf?a tavis tavs.
VanO-NOM Givi-on he-talked-him-II-2 self's self-DAT
'Vano; talked to himself; about Givij .'
(7a) is particularly important, since it shows that the nominative
nominal, vano, does not trigger Tav-Reflexivization, rather the ergative
nominal does.
The fact that it is the ergative-nominal of Class I and 3 verbs and
the nominative-nominal of Class 2 verbs that trigger Tav-Reflexivization
confirms the analysis of subjects made in ( 5). The fact that other nom
inals do not trigger the rule, as shown in (7), is weaker confirmation of
the analysis of objects in (5).
2.2 Tavis-Reflexivization
In ch. I , §2, it was established that tavis- can be triggered by any
term. (8) shows that tavis- can be coreferential in Series II with those
nominals identified as terms in ( 5) above. In (8a) a subject triggers
tavis-, in (8b) a direct object, and in (8c) an indirect object. (8) 1S
parallel to (10-12) in ch. I , only the tense and cases being different.
(8) (a) dedam dabana tavisi svili.
mother-ERG she-bathed-him-II- I self's child-NOM
'The mother! bathed hert child.'
2.2 The analysis of case in Series II 43
(b) ninom misca bavsvi tavis dedas.

Nino-ERG she-gave-her-it-II-I child-NOM self's mother-DAT
'Nino! gave the childj to itsj mother.'
'Nino! gave the childj to her! mother.'
(c) ra acuka gelam ias tavis
what-NoM he-gave-her-it-II-I Gela-ERG la-DAT self's
dabadebisdyeze?
birthday-on
'What did Gelai give laj on hisI/herj birthday ?'
Though the examples given here illustrate only Class I verbs, they show
that the nominals identified as terms in (5) can trigger Tavis-Reflexivi
zation. This is consistent with the fact that all and only terms can
trigger this rule.
2.3 Subject Person Agreement

r n ch. I I established the following paradigm of person and number
markers for subjects of verbs in Series I :
Subject markers
singular plural
l . person v- v-t
2. person o -t
3. person -slalo -enleslnen, etc.
' I 'h ' same markers indicate the person of the subject in Series II. The
' ''i l l llpies in (9) are parallel to those in ( 1 8) of ch. I ; only the tense and
'il . 's have been changed.
I)) (a) namcxvari gamo-v-asxve.

pastry -NoM I-baked-it-II- I
' I baked pastry.'
(h) Ilomcxz)ari gamocxv-a.
pastry -NoM it-baked-II-2
'The pastry baked.'
t . ) istvine?
you-whistled-II-3 (zero marker)
' Did you whistle ?'
(II c)a) lh . person subject triggers v- ; and in (b) the third person
fi rst
I I l l; ·n t riggers - u . Th e second person subject in (c) triggers the zero
marker. This agreement pattern is constant in Series II, irrespective of

the case of the subject. This confirms the analysis of subject in (5).
2.4 Object Person Agreement

The markers of Object Person Agreement are repeated below :
Direct object markers

singular plural
I . person m- · gv-
2. person g- g-t
3 . person 13 13
Indirect object markers

1 . person m- gv-
2. person g- g-t
3. person S/h/0- s/h/0-t
(10) and ( I I), which repeat (19) and (20) of ch. I in tenses of Series II,
show that the nominals analyzed as objects in (5) are indeed those that
trigger Object Person Agreement.
( 1 0) (a) gelam mo-g-if:via supraze.

Gela-ERG he-invited-you-II-I table-on
'Gela invited you to the banquet.'
(b) mama motxrobas mouqva ninos.
father-NOM story-DAT he-told-her-it-II-2 Nino-DAT
(zero marker)
'Father told Nino a story.'
(c) iam ipova satvale.
la-ERG she-found-it-II-3 glasses':'NOM (zero marker)
'Ia found her glasses. '
( I I) (a) mi-s-ca sasmeli stumars.
he-gave-him-it-II-I drink-NOM guest-DAT
'He gave drink to the guest.'
(b) mama motxrobas mo-g-iqva.
father-NOM story-DAT he-told-you-it-II-2
'Father told you a story.'
(c) gelam m-ipasuxa .
Gela-ERG he-answered-me-II-3
'Gela answered me.'
2.3 A test for termhood 45
The fact that the same nomina Is trigger Obje<,:t Person Agreement in
( 1 0) and ( I I ) as in their Series I counterparts, regardless of the case
marking they bear, shows that these nominals are indeed objects, as
stated in (5).
2 . 5 Unemphatic Pronoun Drop

I t was established in ch. I, §4.2, that only terms can undergo the rule
of Unemphatic Pronoun Drop. The analysis summarized in (5) predicts
t hat the nominals listed there can all undergo this rule. ( 1 2) illustrates
the dropping of a variety of nominals ; in each instance the gloss indi-
'ates that something has been dropped, not merely omitted.
( 1 2) (a) miv�ere �erili.

I-wrote-him-it-II- I letter-NOM
'I wrote him a letter.'
(b) da�era.
he-wrote-it-II-I
'He wrote it.'
(c) dada.
he-stayed-II-2
'He stayed.'
( d) gelam gipasuxa.
Gela-ERG he-answered-you-II-3
' Gela answered you.'
1 11 ( 1 2<1) the ergative-nominal and dative-nominal have been dropped

[ I t ) l l l the surface, in ( I 2b) the ergative-nominal and nominative-nominal
( . f. eh. I , §4), in ( I 2C) the nominative-nominal, and in ( I 2d) the dative
l i l l I l l i nal. In each instance the person -reference is recoverable. Since
•. 1 1 and only the nominals analyzed as terms in (5) can be dropped by
\ 1 1 it! rule, the proposed analysis is again confirmed . .
l ase as a test for termhood
. I ' l l ' subject properties established in ch. I are borne by the same
1 I 1 1 1 1 1 i nai in Series I and II, but the case marking is different in these
I W( I Series. The case pattern used in Series II, moreover, is governed
I . t he Class of the verb form. In ( 1 3) and ( 1 4) I combine the facts
1 I I I I I n arized in (5) above with those established for Series I i n ch. I .
46 Appendix : Constituent screeves
( 13) Subject Direct object Indirect object

( 1 4) Series I II
Class I, 3 B A
2 B B
I have shown that an analysis that made use of case a priori to estab
lish grammatical relations would be inadequate. That is, subject
properties in Georgian are not always related to one particular case ;
nor are direct object properties. An analysis of Georgian based on the
assumption that the nominative-nominal (or the ergative-nominal) was
always subject would be inadequate.
However, now that the case-grammatical relation correspondences
have been established through syntactic tests, case can be used as an
argument for grammatical relations. For example, it can be stated that
the nominative-nominal associated with a Class I verb in Series II
must be its direct object. Thus Case Marking provides a further set of
tests for termhood. It will be used as evidence for final subjecthood,
direct-objecthood, and indirect-objecthood in all of the syntactic con
structions considered in this work.
Appendix : Constituent screeves of Georgian Series
I use the word 'tense' or 'tense-aspect' throughout this work to refer to a

conjugational unit, traditionally called a m��rivi. There are ten or eleven
m��ivi in Georgian, divided among the three Series. Each is a unique collo
cation of tense, aspect, and mood. For example, the present 'tense' is in
Series I and represents present time, incomplete aspect, either continuous
or non-continuous aspect, and indicative mood. The aorist, on the other
hand, is in Series II, and represents past time, complete aspect, and indicative
mood. I refer to the m��ivi as 'screeves' or 'tenses', since this word has no
satisfactory translation in English.
The reader may be curious about the tenses which constitute the different
Series. Therefore, although these will not be essential to the discussion, I
list them here. Series III is introduced in ch. 8.
Constituent tenses of Series I
present
future
Appendix : Constituent screeves 47
imperfect
conditional
present subjunctive
future subjunctive
Constituent tenses of Series II
aorist
optative (second subjunctive)
{;onstituent tenses of Series III
perfect (first evidential)
pluperfect (second evidential)
third subjunctive (third evidential)
The use of these tenses is described in Shanidze (1973 : 21 5-23), Tschenkeli
( 1 958 : chs. 8, 10, 1 2-13, 1 8-19, 39-40), and in other handbooks.
3 Object Camouflage
This chapter describes a rule that determines the surface form of first
and second person direct objects in clauses that also contain indirect
objects. This phenomenon will provide a further test for direct
objecthood and for the presence of an indirect object.
I The facts to be considered
( I) illustrates a simple sentence with third person subject, direct object,

and indirect object ; (a) is in a Series I tense, (b) in a Series II tense.
( I ) (a) vano anzors adarebs gZVtS.
Vano-NoM Anzor-DAT he-compares-him-him-I-I Givi-DAT
'Vano is comparing Anzor to Givi.'l
(b) vanom anzori seadara gzvzs.
VanO-ERG Anzor-NoM he-compared-him-him-II-I Givi-DAT
'Vano compared Anzor to Givi.'
(*2) is exactly parallel to ( I ) , but the second person singular personal

pronoun has been substituted for the third person direct object, and
the corresponding marker of Direct Object Person Agreement has been
added to the verb form.
(2) (a) *vano (sen) gadarebs givis. 2

Vano-NOM YOU-DAT he-compares-him-you-I- I Givi-DAT
('Vano is comparing you to Givi.')
(b) *vanom (sen) segadara gzvzs.
VanO-ERG YOU-NOM he-compared-him-you-II-I Givi-DAT
('Vano compared you to Givi.')
(*2) is ungrammatical whether or not the pronoun len 'you-SG' IS
dropped by Unemphatic Pronoun Drop.

3.I The facts to be considered 49
The sentences of (3) are like (*2) except that the expression seni tavi
'your-sG self' has been substituted for sen 'you-SG'.
(3) (a) vano sens tavs adarebs glVlS.

your self-DAT he-compares-him-him-I-I
'Vano is comparing you to Givi.'
(b) vanom seni tavi seadara glVts.
your self-NoM he-compared-him-him-II-I
'Vano compared you to Givi.'
Notice that the meaning of these grammatical sentences is what (*2)

was meant to express. The verb in (3) agrees with seni tavi as a third
p rson direct object ; that is, it has a zero marker (cf. paradigm ( 1 6) in
· h . I ) . (*4) shows that the verb cannot agree with seni tavi as a second
p ' rson direct obj ect, which would require the marker g.
( � ) ( a ) *vano sens tavs gadarebs glVlS.

he-compares-him-you-I-I
('Vano is comparing you to Givi.')
(b) *vanom seni tavi segadara gzvts.
he-compared-him-you-II-I
('Vano compared you to Givi.')
I II (·5) and (6) we see that the same constraint that governs the occur-
1 l' 1 It' · of the second person also governs the first person ; (*5) and (6)
t l d l' ' I' from (*2) and (3), respectively, only in that the former have .first
1 '1 ' 1 !Ion direct objects instead of second person.
( a) "'vano(m e ) madarebs givis. 2

me-DAT he-compares-me-him-I-I
( ' Va no is comparing me to Givi.')
( I » ·1)anOm ( me) semadara glVts.
me-NOM he-compared-me-him-II- I
( ' Vano compared me to Givi.')
( I.) ( : . ) (1110cems tavs adarebs glVlS.

my sclf-DAT he-compares-him-him-II-I
• Va no is comparing me to Givi.'
I ,) fJ(lIIOm cemi tavi seadara glVts.
my self-NoM he-compared-him-him-II-I
•
I I n o 'ompared me to Givi.'
50 Object Camouflage
The plurals of the first and second persons are also governed by this
constraint, but the relevant examples need not be given here.
It remains to be shown that this rule applies to direct objects only
when there is an indirect object in the clause. (7) and (8) give paradigms
that change for the person of the direct object in Series I and II, respec
tively. s It is clear from these that the simple pronoun forms of the first
and second persons as well as those of the third person occur as direct
objects, as long as there is no indirect object.
(7) (a) (me) mar�munebs.
me-DAT he-convinces-me-I-I
'He is convincing me.'
(b) (Sen) gar�munebs.
YOU-DAT he-convinces-you-I-I
'He is convincing you.'
(c) vanos ar�munebs.
VanO-DAT he-convinces-him-I-I
'He is convincing Vano.'
(8) (a) (me) damar�muna.
me-NOM he-convinced-me-I I-I
'He convinced me.'
(b) (sen) dagar�muna.
YOU-NOM he-convinced-you-II-I
'He convinced you.'
(c) vano daar�muna.
Vano-NOM he-convinced-him-II-I
'He convinced Vano.'
(5-6) and (2-3) show that Object Camouflage must apply to first and
second persons, respectively ; ( I) shows that it does not apply to third
persons. (7-8) show that it operates only if an indirect object is present
in the clause. 4
2 An analysis
I propose to account for the facts discussed above by permitting the

rules that generate initial structures to specify first and second person
direct objects, as for any other grammatical relation, and by including
in the grammar the rule of Object Camouflage.
3.3 interaction 51
(9) Object Camouflage

If a clause contains an indirect object, . a first or second person
direct object in that clause is realized as a possessive pronoun + tavi,
where the possessive reflects the person and number of the input
form.
Such an analysis is needed to account for the fact that first and second
person pronouns, which occur freely as subjects and indirect objects,
and which occur as direct objects when there is no indirect object, do
not occur only when there is an indirect object. Further, this rule is
needed to account for the fact that misi tavi 'his self' is not found in
this situation. 6
3 Interaction with some rules previously considered
3. ( Person Agreement
I t was observed above that the verb agrees with the phrases cemi tavi
II1d seni tavi as third person nominals, not as first and second persons.
This can be accounted for by insuring that Person Agreement apply
\ 0 \ h<.: output of Object Camouflage. In chs. 4-8 I will establish that
( Jhj<.:ct Camouflage applies generally to the output of rules that change

'nl lnmatical relations.
, . .: Unemphatic Pronoun Drop

I I I I h <.: chapters that follow, we will see evidence that Unemphatic
l ' f l l l l ( ) \ l fI Drop, too, is stated on final terms. We can account for the
1 11' 1 t hat camouflaged objects are never dropped by insuring that
I I ' m p hatic Pronoun Drop always applies to the output of Object
"1 I 1 l( )u f l agc. However, the same fact might be handled equally simply
" l i l l i it ing Unemphatic Pronoun Drop to personal pronouns ;6 this
\ l Jl l ld also correctly prevent this rule from applying to reflexive tav-,
I I " . \ ion -words, relative pronouns, indefinite pronouns, negative pro
/ 1 1 1 1 1 1 1 , I;\: .
( '(lSI' Marking
II 1'II 11100 db g ed object has the same case marking as any direct object
1 I \ l l d l an v ', as detcrmined by Series and Class of the governing
I h.
52 Object Camouflage
4 An alternative analysis
The expressions cemi tavi and seni tavi are used both as camouflaged
objects and as reflexive pronouns (cf. ch. I, § I. I) . Because of this, it
might be proposed that Object Camouflage and Tav-Reflexivization
are a unified phenomenon. However, these phenomena cannot be
accounted for with a single syntactic rule, since the rules apply under
different conditions, including the following :
(i) Tav� Reflexivization requires coreference between the trigger and
the target ; Object Camouflage has nothing to do with coreference.
(ii) Tav-Reflexivization applies to any nominal dependent of the clause
which contains its antecedent ; Object Camouflage applies only to
direct objects.
(iii) Object Camouflage applies only to first and second persons ; Tav
Reflexivization has no such constraint.
(iv) Object Camouflage applies only in the presence of an indirect
object in the same clause ; Tav-Reflexivization has no such con
straint.
I conclude that there is no way to unify the rules of Tav-Reflexivization
and Object Camouflage.
5 Object Camouflage as a test for term hood
Because Object Camouflage applies to direct objects in the presence of

indirect objects, it constitutes a test for direct-objecthood and for
indirect-objecthood. In the chapters that follow, Object Camouflage
will be the basis of arguments that the initial direct objects of passive
and inversion constructions are not final direct objects. It will. also be
the basis of an argument that indirect objects created by Causative
Clause Union and by the so-called 'Version' rules do indeed bear this
particular final grammatical relation. Object Camouflage also provides
evidence with respect to final direct and indirect objects in other
syntactic constructions.
4 Object Raising
I n this chapter I will argue for a rule, Object Raising. 1 Analyses of

t'imilar rules in other languages have shown that a variety of restrictions
may be put on this type of rule. In English, for example, any non
flubject may be raised to subject (Postal 1971). Chung ( 1 976b) shows
t hat in Indonesian only initial direct objects that are derived subjects
Inay undergo this rule. In Portuguese, an initial direct object may
u l ldergo the rule, whether or not it has undergone Passivization (Perl
lIlutter, ms.). In Georgian only initial direct objects that have not
- h a ng e d their grammatical relation can undergo this rule. For this
I' 'ason, Object Raising provides us with a test for the direct object in
( ; ·o rgi an .
Berman ( 1974) presents evidence that the for-nominal in
( I ) Tadpoles are easy for children to raise.
( o rigi nates, not as the subject of the embedded clause, but as a dependent
( o f t he matrix clause. She argues that this for-nominal is coreferential
" i t II the embedded subject, which gets deleted by Equi. In §s bel()w,
" l i d i l l ch. I I , I give two arguments to support a similar analysis for
:rorgian. The Georgian arguments are of a type very different from the
/)1 1(' w h i ch Berman presents for English. Chung ( 1 976b) argues for
.1 1 1 ill la!ogous structure in Bahasa Indonesia. The fact that languages
I d i l l' 'rent as English, Georgian, and Bahasa Indonesia offer evidence

1 1 1 1 t h i::; analysis suggests that it is universal, regardless of superficial
1 1 1 1 1 t rai nts of the kind mentioned above.
d cription of the data in Georgian
I l t .l i n adjectives, such as advili 'easy', 3neZi 'hard, difficult', and

l I / (/II/(J 110 ' n ice, pleasant', occur with qopna 'be' in sentence pairs like
54 Object Raising
those in (2-4). In the (a) examples, the subject is a mas dar, or deverbal
noun, together with the nominals it governs. 2 The (b) examples contain
infinitives. Both masdars and infinitives are treated more completely
in ch. 10.
(2) (a) am gvelis mo�vla Jnelia.

this snake-GEN killing-NoM hard-it-is-I-2
'Killing this snake is difficult.'
(b) es gveli Jnelia mosa�lavad.
this snake-NOM to kill
'This snake is hard to kill.'
(3) (a) amanatebis mireba saxlidan sasiamovnoa.

parCeIS-GEN receiving-NoM home-from pleasant-it-is-I-2
'It is pleasant to receive parcels from home.'
(b) amanatebi sasiamovnoa saxlidan misarebad.
parcelS-NoM to-receive
'Parcels are nice to receive from home.'
(4) (a) cemtvis Jnelia, �argi magalitebis mOJebna.

me-for hard-it-is-I-2 good exampleS-GEN finding-NOM
'It is hard for me to find good examples.'
(b) �argi magalitebi Jnelia mosaJebnad cemtvis.
examples-NoM to-find
' Good examples are hard for me to find.'
The a which cliticizes to the adjectives in (2-4) is the enclitic form of

the third person singular present tense aris 'is' ; cf. Introduction, §4. 1 .
For the failure of inanimate subjects to trigger Number Agreement,
see Introduction, §4+ The (b) sentences are analyzed in the next
section.
2 An analysis
Below I will argue that the grammatical relations in object-raised sen

tences are partially represented by (5) and (6)3 and that the rule of
Object Raising can be informally described as (7).
4.2 An analysis 55
(5)
3neli qOPn,�
' hard ''bi
a P
r�\
� ��
It 2\1
m03ebna me �argi magalitebi
, find ' 'I' ' good examples'
(6)
me �argi magalitebi
(See Notes on Presentation on the principles of diagramming used here.)
(7) With matrix verbs of the type, advili aris 'is easy' , the direct object
in a sentential subject may ascend.
This analysis is based on Postal & Perlmutter's proposal for a class

of rules called 'ascensions', defined in the Introduction, §2. I (cf. Perl
I f l uttcr & Postal 1 974 and Perlmutter (to appear c)). The Relational
Succession Law, not yet stated in this monograph, requires that the
:I:lccnding nominal (here �argi magalitebi) assume the grammatical
I' 'lation borne by its host (here ""me �argi magalitebi m03ebna), in this
instance becoming subject (cf. Perlmutter & Postal (to appear b» . The
Chomeur Law (cf. Introduction, §2. I) predicts that the displaced sub
i . ·t, the cmbedded clause, will become a chomeur. Since it is suggested
I h a t these processes are universal, it is not necessary to state them in a
grammar of Georgian. In the grammar, we need only observe that

( : 'orgian has this particular rule, and state the language-particular fact
I h a t only direct objects may undergo it (cf. §6).
56 'Object Raising
3 Facts concerning the nominative-nominal
In this section it is argued that the nominative-nominal must be the

initial direct object of the embedded clause and a final dependent of
the matrix clause, rather than the initial subject.
3 . 1 Suppletion
The first step is to argue that �argi magalitebi 'good examples' of (4b)
is initially the direct object of the embedded clause, as in (5). The
arguments will be based on suppletion of the verb root for number or
animacy of the direct object. The arguments will be of the following
form :
(i) In simple clauses that have undergone no change of grammatical
relations, there is suppletion of verb roots such that one root is used
if the direct obj ect meets a certain criterion (on animacy or plu
rality), an d a different root is used if the direct object fails to meet
this criterion.
(ii) In object-raised sentences, the same suppletion is observed with
respect to the animacy or plurality of the nominative-nominal.
(iii) If the nominative-nominal is, at some stage of derivation, the direct
object of the suppletive verb, the simple sentences and the object
raised sentences can be accounted for by a single rule. If, on the
other hand, the nominal in question is not the direct object of the
suppletive verb, the grammar must contain two rules to account
for these facts.
(iv) The nominative-nominal must therefore be a sometime direct
object of the suppletive verb.
tv) Since it is shown below that the nominative-nominal is final sub
ject of aris 'be', and since it could not simultaneously bear a
grammatical relation to two verbs, it must be at the initial level of
derivation that it is direct object of the suppletive verb.
This analysis requires that the rules of suppletion be stated on initial

terms ; this is confirmed by additional data in later chapters.
3 . 1 . 1 Suppletionfor number of direct object. In Georgian there are several

verbs that are suppletive for the number of the direct object ; that is,
one verb root is used with singular direct objects, and another with
plural direct objects. One such pair is gadagdebaJgadaqra 'throw' ;
4.3 The nominative-nominal 57
gadagdeba (root gd) is used with singulars, while gadaqra (root qr)
occurs only with plurals. Evidence to support this and a complete dis
cussion is given in the Introduction, §4.3.z.
The root alternation in this verb is also governed by the plurality
of the nominative-nominal in object-raised sentences, as shown in (8).
In (*8a) the final subject of 3nelia 'it is hard' is singular. (8b) differs
from (*8a) only in that the final subject of the former is plural, and the
sentence is grammatical.
(8) (a) *es didi kva 3nelia gadasaqrelad.

this big stone-NOM hard-it-is-I-z to-throw
(,This big stone is hard to throw.')
(b) es didi kvebi 3nelia gadasaqrelad.
stoneS-NOM
'These big stones are hard to throw.'
(8) may be compared with (z9) of the Introduction. (*8a) is ungram

matical because it violates the suppletion condition on the number of
didi kva/kvebi.
If (8) has an initial structure in which kva/kvebi 'stone/stones' is the
direct object of the verb 'throw', as proposed in §z, then we can predict
the ungrammaticality of (*8a) with the same rule that predicts the
ungrammaticality of (*z9a) in the Introduction ; otherwise the rule must
be complicated.
3 . I . z Suppletion for animacy of direct object. In Georgian . some verbs

are suppletive for animacy of the direct object. One such verb is
mitana/�aqvana 'take' ; mitana (root tan) is used only with inanimate
direct objects, while �aqvana (root qvan) occurs only with animate
direct objects. This suppletion is established in the Introduction,
§4. 3 . I , where this verb is discussed at length.
The root alternation in this verb is also governed by the animacy of
the nominative-nominal in object-raised sentences, as shown in (9).
Where an animate nominative-nominal is matched with �aqvana, as in
(9b) , the sentence is grammatical ; but where it is paired with mitana,
as in (*9a), the sentence is ungrammatical. Similarly, where an inani
mate nominative-nominal occurs with �aq'vana, as in (*9d), the sentence
is ungrammatical ; but where it occurs with mitana, as in (9c), the result
is grammatical.
58 Object Raising
(9) (a) ""gela advili ikneba studkalakSi misatanad.

Gela-NoM easy it-will-be-I-2 student-compound-in to-take
('Gela will be easy to take into the student compound.')
(b) gela advili ikneba studkalakSi �asaqvanad.
to-take
'Gela will be easy to take into the student compound.'
( c) es �o�a advili ikneba studkalakSi misatanad.
this jug-NOM
'This jug will be easy to take into the student compound.'
(d) ""es �of:.a advili ikneba studkalakSi �asaqvanad.
(,This jug will be easy to take into the student compound.')
(9) may be compared with (22-23) of the Introduction. The comparison

shows that the suppletion that occurs in verbs in simple sentences with
respect to animacy of their direct objects also occurs in infinitives of
object-raised sentences with respect to the animacy of the nominative
nominal.
According to the proposed analysis, where gela and es f:.o�a 'this jug'
are initial direct objects of the verb 'take', the sentences of (9) are
accounted for by the same principle needed to account for suppletion
in simple sentences, such as (22-25) in the Introduction. An analysis
in which gela and es f:.of:.a were not direct objects of 'take' would have
to complicate the suppletion rule unnecessarily.
3.2 Selection restrictions on the matrix subject

In Georgian, concrete nouns cannot normally occur as the subjects of
'be' plus a predicate adjective such as advili, etc. ; this is shown in ( 1 0) .
( 10) (a) ""anzori advilia.

Anzor-NoM easy-it-is-I-2
( 'Anzor is easy.')
(b) ""es s�ami advilia.
this chair-NOM
('This chair is easy.')
( c) es mu!aoba advilia.
work-NOM
'This work is easy.'
In (""lOa) and (""lob) the subject of 'be' +advili is concrete ; in ( I OC) it

is abstract.
4.3 The nominative-nominal 59
But the subject in object-raised sentences may be concrete ; indeed,

it is concrete in all the examples of object-raised sentences given above.
These facts are handled in a natural way under the proposed analysis.
The constraint is stated on initial structures. The derived subjects of
'be' + predicate adjective originate in initial structures as the direct
object of another verb, and hence do not violate the constraint. But a
grammar that did not derive object-raised sentences by a rule that
changed grammatical relations would have to state that predicate
adjectives + 'be' do not allow concrete subjects unless accompanied by
certain kinds of complements. Thus, I conclude that the final subjects
in object-raised sentences are not subjects in initial structures.
3.3 Question Formation

Question Formation insures that the questioned constituent immedi
ately precede the verb that governs it. The rule cannot, however, move
constituents out of a clause ; cf. Introduction, §4-z+ For this reason, it
provides a test for constituency.
The sentences in ( I I) show that the nominative-nominal, if ques
tioned, must immediately precede aris 'is'. In (* I I b, c) the questioned
nominative-nominal does not immediately precede aris.
(I I ) (a) ra aris 3neli mosa3ebnad?
what-NoM it-is-I-z hard to-find
'What is hard to find ?'
(b) *ra 3ne1ia mosa3ebnad?4
(c) *ra mosa3ebnad aris 3ne1i?5
Because a questioned constituent cannot move out of its clause, ra
'what-NoM' must be a dependent of the clause containing aris 'is'. This
is consistent with our analysis, (6), as long as Question Formation
applies to the output of Object Raising. In later chapters, we will see
additional evidence that Question Formation must apply to the output
of relation-changing rules.
3.4 Case Assignment

The verb qopna 'be', like all Class z verbs, governs the nominative case
for its subject in all Series. In the sentences of (Z-4b), es gveli 'this
:make', amanatebi 'parcels', and �argi magalitebi 'good examples' are in
t h e nominative case. This supports the analysis above, where these
nominaIs arc derived subjects of the verb 'be'.
60 Object Raising
3 . 5 Subject Person Agreement

The nominative-nominal triggers Subject Person Agreement (cf. para
digm ( I S), ch. I). ( 1 2) gives a paradigm of an object-raised sentence
changing for the person of the nominative-nominal.
(12) (a) 3neli var dasar�muneblad.
hard I-am-I-2 to-convince
'I am hard to convince.'
(b) 3neli xar dasar�muneblad.
you( SG)-are-1-2
'You(SG) are hard to convince.'
( c) gela aris 3neli dasar�muneblad.
Ge1a-NoM he-is-I-2
'Gela is hard to convince.'
The verb qopna 'be' is irregular, and the subject agreement markers
listed in paradigm ( I S) of ch. 1 are not realized in exactly the same form
in this verb. But the v- of the first person is obvious in the var 'I am'
of ( I 2a). The -s of the third person shows up in the full form aris 'is'.
It was observed in ch. 3 that Person Agreement applies generally to
the output of rules that change grammatical relations. The proposed
analysis predicts therefore that Subject Person Agreement will apply to
the output of Object Raising, treating the raised nominal as subject.
Thus, ( 1 2) confirms the proposed analysis, in particular the final
subjecthood of the nominative-nominal.
3.6 Summary
It has been shown that the nominative-nominal in the (b) sentences of
(2-4) has the properties of the initial direct object of the verb which is
realized as an infinitive, and the properties of a non-initial ( §3.2) sub
ject of aris 'is'. The obvious way to account for these facts is with a rule
that makes the initial direct object of the embedded clause the subject
of the matrix clause.
4 Facts concerning the infinitive
In §4. 1 it is argued that the infinitive, together with the nominals it

governs, constitutes a subject chomeur. In §4.2 it is shown that the
infinitive and the nomina Is it governs are a constituent and that these
nomina Is do not become dependents of the matrix clause.
4.4 The infinitive 61
4. 1 Case Assignment
In ch. 2 I showed that in Series I and II in Georgian, all terms are
marked with one of the following cases : nominative, dative, or ergative.
The infinitive bears none of these cases. This supports the proposed
analysis, according to which the infinitive is a chomeur, not a term of
the matrix clause. Nor could the infinitive be a predicate nominal, since
predicate nominals are marked with the nominative case in all Series.6
4.2 Question Formation

In Question Formation, the Q-word immediately precedes the verb of
the questioned clause, unless the Q-word is part of a larger nominal, in
which case that entire nominal must immediately precede the verb of
the questioned clause �cf. ( 1 5-16) in Introduction). In ( I 3a) the Q-word
vis (in visgan) does not immediately precede the matrix verb aris 'is' .
Instead, the whole nominal containing vis, namely visgan misarebad
'from whom to receive' , precedes aris. This shows that visgan misarebad
is a nominal constituent of the matrix clause. In (*I3b, c) visgan itself
immediately precedes the matrix verb ; the fact that these orders are
ungrammatical shows that visgan is not governed by aris. In (*13 d)
visgan precedes misarebad as in ( I 3a), but the whole fails to precede
aris. The fact that this order is ungrammatical rules out the possibility
that visgan misarebad is an indirect question, comparable to ( 17a) 111
the Introduction.
( I 3) (a) amanatebi visgan misarebad aris sasiamovno?
parcels-NoM who-from to-receive it-is-I-2 pleasant
'From whom are parcels nice to receive ?'
(b) *amanatebi visgan aris sasiamovno misarebad?
(c) *visgan aris sasiamovno amanatebi misayebad?
(d) *amanatebi sasiamovnoa visgan misarebad?
Thus, ( 1 3) shows that visgan must be governed by misarebad and that
7.1isgan misayebad is a nominal governed by aris, as proposed in §2.
(14) gives an example of a similar kind, which shows that the tvis
nominal is governed by misacemad 'to give' .
( 1 4) sacukrebi vistvis misacemad aris 3ne1i?
gifts-NOM who-for to-give it-is-I-2 hard
'To whom is it hard to give gifts ?'
Notice that the interpretation of the tvis-nominal in ( 14) is very different
62 Object Raising
from that of the tvis-nominal in (4b). I assume that the former is the
initial indirect object of micema 'give'. The tvis-nominal of (4b) is dis
cussed below in § 5.
4.3 Summary
In this section I have shown that the infinitive behaves as predicted by
the analysis proposed in §2 : it is a non-term ( §+ I ) nominal governed
by aris ( §4.2) ; the nominals it governs initially are not finally governed
by aris ( §4.2).
5 Facts concerning the tvis-nominal
In this section I will discuss the tvis-nominal which occurs in (4b). I

will argue against the analysis usually given for Object Raising in most
languages, with respect to the tvis-nominal (,for' -nominal). The usual
analysis makes this nominal the initial subject of the clausal subject.
Berman ( 1974) proposed for English that the for-nominal originate in
the matrix clause, be coreferent to the subject of the embedded clause,
and trigger the deletion of the latter by Equi. It is this analysis I wish
to propose for Georgian.
If the tvis-nominal originated as the subject of the embedded verb,
it would nevertheless not have the properties of a superficial subject.
In ch. 1 0 it is shown that the initial subjects of non-finite verb forms do
not have superficial subject properties. Therefore, I will not bother to
show here that this nominal fails to trigger Subject Person Agreement,
fails to undergo Unemphatic Pronoun Drop, and is not assigned a
subject case.
I will, however, show that the tvis-nominal is not governed by the
infinitive. If the tvis-nominal had originated as the subject of the
embedded verb, we would expect it to remain a non-term dependent
of that clause. In chs. 10 and I I it is shown that the initial dependents
of non-finite clauses continue to be dependents of those clauses, even
though they cannot have the properties of superficial terms. Thus, a
demonstration that the tvis-nominal is the dependent of a different
clause will show that it cannot have been the initial subject of the
embedded clause.
In §4.2.4 of the Introduction it was shown that only the questioned
dependents of a questioned clause occur immediately before the verb
4.6 A test for direct-objecthood 63
of that clause. The fact that vistvis 'for whom' immediately precedes
the verb in
( 1 5) sacukrebi vistvis aris 3ne1i misacemad?

gifts-NOM who-for it-is-I-z hard to-give
'For whom is it hard to give gifts ?'
shows that it is governed by the verb, aris 'is'. The English translations
capture the difference in the meanings in Georgian ; ( 1 4) asks toward
whom gift-giving is difficult, ( 1 5) asks who finds gift-giving difficult.
English uses to in the former function, and for or to in the second. In
Georgian tvis is used for both purposes. Both ( 14) and ( 1 5) are unam
biguous in Georgian, because of the constituent structure ; but ( 1 6) is
ambiguous.
( 1 6) sacukrebi 3nelia anzoristvis misacemad.

gifts-NOM hard-it-is-I-z Anzor-for to-give
'Gifts are hard to give to Anzor.'
'Gifts are hard for Anzor to give.'
Since tvis marks nominals in the functions illustrated in both (14) and
( 1 5), ( 1 6) is ambiguous between these two uses. Only in questions does
the word order unambiguously reflect constituent structure.
I have shown here that the tvis-nominal is a dependent of the matrix
clause in derived structure. If it had originated as the subject of the
embedded clause, it should have remained a dependent of the infinitive,
as did visgan and vistvis in ( 1 3) and ( 1 4). In ch. I I an additional argu
ment is given that the tvis-nominal must have originated in the matrix
clause. This argument is based on the fact that this nominal does not
have the properties of a retired subject.
6 Object Raising as a test of direct-objecthood
In Georgian, only the direct object may undergo Object Raising. In

Ci01 7b) an indirect object raises, in (*I8b) the object of a postposition,
and in (*1 9b) the subject of an intransitive ; all the resulting sentences
; I r e ungrammatical. (17d) and ( 1 8d), however, show that the direct
object from those same sentential subjects can raise. The (*b) and (*c)
s ' ntences differ from one another only in that the latter the direct
ohj 'ct has been omitted, as is possible in English.

64 Object Raising
(17) (a) 3nelia anzoristvis salukris micema.

hard-it-is-I-z Anzor-for gift-GEN giving-NoM
'Giving gifts to Anzor is difficult.'
(b) *anzori 3nelia salukris misacemad.
Anzor-NoM to-give
(,Anzor is hard to give gifts to.')
( c) * anzori 3nelia misacemad.
(,Anzor is hard to give to.')
(d) salukari 3nelia anzoristvis misacemad.
gift-NOM
'A gift is hard to give to Anzor.'
(18) (a) advilia, ninostvis sarvlis se�erva.
easy-it-is-I-z Nino-for trouserS-GEN sewing-NoM
'Sewing trousers for Nino is easy.'
(b) *nino advilia sarvlis sesa�ravad.
Nino-NoM to-sew
(,Nino is easy to sew trousers for.')
(c) ,Y,nino advilia sesa�eravad.
('Nino is easy to sew for.')
(d) sarvali advilia ninostvis sesa�eravad.
trouserS-NOM
'Trousers are easy for Nino to sew. '
'Trousers are easy to sew for Nino.'
(See §s above on the ambiguity of (17-18d).)
(19) (a) damtknareba sasiamovnoa golastvis.
yawning pleasant-it-is-I-z Gocha-for
'To Gocha, it is pleasant to yawn./ To Gocha, yawning is
nice.'
(b) *gola sasiamovnoa dasamtknareblad.
Gocha-NOM to-yawn
*'Gocha is pleasant to yawn.'
(19) is included just to show that this rule applies, not to absolutives,
but to direct objects. In (17) and (18), indirect objects and objects of a
postposition have raised like direct objects, but the result is ungram
matical. These data show that this rule applies only to direct objects.
Notice that in English the comparable rule applies to any non-subject.
Because of its restriction in Georgian, this rule will s(;rve as a test for
direct objects.
4.6 A test for direct-objecthood 65
Object Raising is the first of the rules examined that change grammatical
relations. The main purpose in coming chapters will be to examine the
nature of other rules that change grammatical relations. A major con
cern will be the case marking of terms and the marking of retired terms,
both processes differing in Georgian from the equivalent ones in other
languages.
5 Causative Clause Union
Recent studies of causatives have focused on a type of causative for

mation that is important among the languages of the world (Aissen
1974a, b ; Comrie 1975, 1 976c ; Xolodovich 1969 ; Shibatani 1976, etc.).
This type can be characterized as having a complex (two-clause) initial
structure and a simplex (single-clause) final structure. Georgian has a
causative of this type. I will refer to this as the 'organic causative', 1
since Georgian also has a causative with a complex (two-clause) final
structure. The organic causative is illustrated in ( I ) .
( I ) mas(:avlebelma gadaatargmnina gelas a�a�is
teacher-ERG he-caused-translate-him-it-II-1 Gela-DAT Akaki-GEN
leksi.
poem-NOM
'The teacher made Gela translate Akaki's poem.'
I will have nothing to say here about the periphrastic causative.
The Georgian data are important for the analysis of causatives
because of the particularly strong evidence they bring to bear on the
simplex nature of the final structure. Because of the case differential
described in ch. 2, an analysis of Georgian that simply assigned the
object cases to the nominal dependents of the embedded clause, instead
of making them terms of the matrix verb, would be inadequate. Object
Agreement and Object Camouflage offer further evidence of a kind
not found in more familiar languages. The organic causative provides
a particularly strong argument in favor of relational grammar, since the
relevant generalizations cannot be simply stated on the basis of linear
order and dominance or on the basis of case.
I An analysis of organic causatives
In the studies cited above, it is shown that in this type of causative the
following is generally true across languages :
5. 1 An analysis 67
The initial matrix subject is derived subject of the causative.

In the causative of an intransitive, the initial subject of the embedded
verb is realized as direct object of the causative.
In the causative of a transitive,
(A) the initial embedded subject is derived indirect object
the initial embedded direct object is derived direct object
the initial embedded indirect object is a derived non-term
or
(B) the initial embedded subject is a final non-term
the initial embedded direct object is derived direct object
the initial embedded indirect object is derived indirect object.
Georgian exemplifies type (A) . In ( I) above, the initial matrix subject,
mast;avlebeli 'teacher', is the final subject of the organic causative ; the
initial embedded subject, gela, is the indirect object ; and the initial
embedded direct object, a�a�is leksi 'Akaki's poem', is the direct
object. This can be seen immediately in the case pattern and will be
supported by other evidence in §3.
French has causatives of type (B), as well as type (A) . Consider (2)
from Aissen ( 1974a).
(2 ) Jean a fait chanter l'hymne par les gendarmes.
'Jean had the anthem sung by the policemen.'
Aissen shows that in (2) the initial matrix subject, Jean, is also the final
subject of the causative ; the initial embedded direct object, l' hymne,
is the final direct object ; and the initial embedded subject, les gendarmes,
is finally a non-term.
Postal & Perlmutter ( 1 974) argue that the Georgian type is basic
and that the French pattern in ( 2) is to be explained by the application
of Passivization in the embedded clause. They propose that there is a
general rule of Clause Union, partially specified as (3) . 2
(3) Causative Clause Union
The complement clause unites with the matrix clause to form a
single clause in the following way :
a. The initial subject of an embedded intransitive becomes the
direct object of the matrix verb.
b. The initial subject of an embedded transitive becomes the
indirect object of the matrix verb.
c. The initial direct object of the complement clause becomes the

direct object of the matrix verb.
d. The complement verb becomes a retired verb.
e. Other dependents of the complement clause become emeritus
dependents of the united clause.
There are still a number of open questions concerning (3), including
the following :
(i) Is (3b) part of the derivation of type (B) causatives, as claimed by
Perlmutter & Postal ?
Some linguists have argued that the rule of Clause Union itself has a
variant output, and that (3) is not an adequate description for all
languages (cf. Comrie 1 976e ; Cole & Sridhar 1 976).
(ii) What is the nature of the complement verb finally ?
This problem is touched upon in ch. 10.
(iii) Are there language-particular conditions that affect (3e) ?
Georgian, with type (A) organic causatives only, is accurately de
scribed by (3). But while it supports (3), it offers no evidence as to
whether this analysis is adequate for Causative Clause Union in all
languages. Georgian does provide especially strong evidence that an
initial indirect object of the complement clause may behave as a non
term dependent of the united clause. Since an emeritus dependent of
a clause is not a term of that clause, this behavior would follow from its
being emeritus indirect object of the matrix verb under Clause Union.
The evidence that I present here does not distinguish between types of
retired indirect object ; it can be interpreted as evidence that the initial
complement indirect object is emeritus dependent of the matrix verb
as stated in (3e). I will also present evidence for (3a, b, and c) in Georgian.
In Georgian, the matrix verb and the retired verb fuse to form a
single word ; the matrix verb is realized morphologically as the causative
circumfix (cf. n. I) . The same is true in Turkish and Japanese (cf.
Aissen 1 974a, b ; Kuno 1 973), but not in French, where the matrix and
retired verbs are distinct words (cf. Kayne 1975), or German, where
the matrix and retired verbs are not even necessarily contiguous in the
clause (cf. Reis 1 973 ; and Harbert 1977). In Georgian all causatives
are in Class I ; this may be attributed to the membership of the matrix
verb in Class I.
I propose that (3) is a part of the grammar of Georgian. It relates
5.I An analysis 69
initial structures like (5) and (8) to complete networks (6) and �9);
respectively.
(4) direktorma astumra vanos

director-ERG he-caused-visit-him-it-II-I Vano-DAT
lJ,omisia.
commission-NoM
'The director had the commission visit Vano.'
rr�
(5)
CAUSE /.� di�'ektori

, dlrector , / I \ \
stumroba �omisia vano
' visit ' ' commission '
(6)
stumroba �omisia vano
. .
mamam mtmacemtna vardebi dedistvis.
father-ERG he-caused-give-me-it-II-I roses-NOM mother-for
'Father had me give roses to Mother.'
(8)
CAUSE mama
, fathe r '
lIucema me vardebi deda

' g ive ' , I' , roses ' , m o the r '
1111ce111a me vardebi deda
(The principles of diagramming used here are briefly explained in the

Notes on Presentation, §3. Here 'U' marks a retired verb.)
In the next two sections I will show that the initial structure of an
organic causative has as its direct object a clause, and that the relations
that nominals bear to the clause are the same as the relations borne in
a simple clause. I will show that the direct objects of embedded tran
sitives and the subjects of embedded intransitives behave like direct
objects of the united clause under Causative Clause Union, with
respect to all rules stated on final relations. That the subject of an
embedded transitive and the indirect object of an embedded intransi
tive behave like indirect objects will also be shown. In §4 I will show
that the initial indirect object of an embedded transitive, under Clause
Union, behaves like a non-term with respect to all rules stated on final
relations.
2 Initial grammatical relations
Aissen ( I 974a, b) points out that many syntactic rules fail to apply in
the embedded clauses of causatives. She proposes ways of accounting
for this, but there remain problems. The fact that causatives disallow
the application of many rules in complement clauses makes it difficult
to argue for the initial grammatical relations. While the arguments given
here are strong ones, particularly the last, they are not numerous. But
the claim made for the initial relations in Georgian is supported by the
fact that there is evidence of various types for such structures in many
languages t cf. references cited above).
5.2 Initial grammatical relations 71
2.1 Lexical entries

If we do not derive causative sentences from structures like ( 5) and (8),
then we must complicate the lexical entry for each verb in at least the
following ways : (i) In addition to the inventory of initial grammatical
relations for each non-causative verb, we must list an inventory for
each causative verb. The verb gadarqveta 'decide', for example, has a
subject and direct object in its inventory of initial grammatical relations ;
its causative has subject, direct object, and indirect object. The verb
se3inva 'sleep' has only a subject in its inventory ; its causative has a
subject and direct object. The verb tamaloba 'play' has a subject and
optional direct object in its inventory of initial grammatical relations ;
the corresponding causative, however, has a subject, direct object, and
optional indirect object (cf. ch. 1 2, §2). All of these facts about causa
tives are predictable from the analysis given in § I .
(ii) In addition to the selection restrictions o n each non-causative
verb, we would have to list selection restrictions on each causative. The
verb revs 'lies', for example, takes only animate subjects, as shown
in ( 1 0).
(10) (a) gela (:evs taxtze.

Gela-NoM he-lies-I-2 couch-on
'Gela is lying on the couch.'
(b) "'(:igni (:evs taxtze.
book-NOM it-lies-I-2 couch-on
(,The book is lying on the couch.')
. fhe same constraint governs the direct object of the corresponding

causative, as shown in ( I I).
( I I ) (a) mama gelas a(:vens taxtze.

father-NOM Gela-DAT he-causes-lie-him-I- I couch-on
'Father makes Gela lie on the couch.' /'Father lays Gela on the
couch.'
(b) "'mama (:igns a(:vens taxtze.
book-DAT
(, Father lays the book on the couch.')
sc facts are predictable if we derive ( l I a) from a structure like (5),

' I 'll
wh T C the verb ro1a 'lie' is embedded under CAUSE, since the subject
pf revs ' l ies' is thc direct object of a(:vens 'causes to lie'.
72 Causa#ve Clause Union
(iii) The rules that relate syntactic structures to semantic represen

tations would have to be complicated, since, in ( l oa), for example, it is
the subject that lies, but in ( I I a) the direct object.
A complex structure analysis makes the correct generalizations about
the language and predicts the regular relations which exist between
every verb and its causative. These regularities could also be accounted
for by lexical redundancy rules. In the following section we will see,
however, that the facts of Tav-Reflexivization force us to choose the
complex structure analysis over the redundancy rule analysis.
2.2 Tav-Rejlexivization 3
In ch. I, §I I showed that Tav-Reflexivization can be triggered only
by subjects. In an organic causative, tav- can be coreferential either
with the matrix subject, or with the final indirect object in causatives of
transitives, or with the final direct object in causatives of intransitives.
In ( 1 2), the causative of an intransitive, tav- may be coreferential to
ekimma 'doctor' or to 'llano, the final subject and direct object, respec
tively. In ( 1 3), the causative of a transitive, tav- may be coreferential
to genom or rezos, the final subject and indirect object, respectively .
( 1 2) ekimma 'llano alap ara�a tavis tavze
doctor-ERG Vano-NOM he-caused-talk-him-II-I self's self-on
'The doctori got Vanoj to talk about himself! . j . '
(13) genom miatanina rezos rignebi
Geno-ERG he-caused-take-him-it-II-I Rezo-DAT books-NOM
tavistan.
self-at
'Genoi got Rezoj to take the books to his!, j place.'
Only if 'llano (in ( 1 2» and rezo (in ( 1 3» are initial subjects, and Tav
Reflexivization is stated on initial terms, are these sentences consistent
with the facts of Tav-Reflexivization established in ch. I . 4
2.3 Summary
I conclude that causatives like (4) and (7) must have complex initial
structures like (5) and (8). This analysis will account for the lexical
correspondences between causatives and the corresponding non
causatives in a simple and elegant way. This is the only analysis that
can account in a straightforward way for the fact that the final indirect
object can trigger Tav-Reflexivization. The arguments to support the
5.3 Derived grammatical relations 73
initial structure proposed here are not numerous, because, as is known,

causatives generally disallow the application of many syntactic rules in
the embedded clause. The conclusion reached here is consistent with
conclusions other researchers have drawn for other languages based on
similar evidence.
3 Derived grammatical relations

We will find that Georgian gives particularly strong evidence for the
derived relations proposed in § I .
3 . 1 Case Assignment
In this section I will show that the initial embedded direct object and
the initial embedded subject of an intransitive have the case marking of
a direct object, dative in Series I and nominative in Series I I . I will
show that the initial embedded subject of a transitive and the initial
embedded indirect object of an intransitive have the case marking of an
indirect object, regardless of Series. This is very strong evidence that
the case marking of nominals in organic causatives is not arbitrary ;
rather, the nominals actually become direct or indirect objects.
Below I give an inventory of organic causatives in order to show the
case marking of each kind of nominal. Because of the case differential,
each example of an organic causative will be given in a Series I tense
and in a Series II tense. In each instance, the syntax of a causative verb
form will be compared with that of the non-causative corresponding
to it.
In Georgian all organic causatives are Class 1 forms, regardless of
the Class of the corresponding non-causative verb form.
3 . 1 . 1 The causative of an intransitive. Intransitive verbs may be in
Class I , 2, or 3 .
(14) (a) bavsvi i3inebs.
child-NOM he-sleeps-I- I
'The child i s sleeping.'
(b) ubeduri semtxveva xdeba baySi.
unfortunate accident-NOM it-happens-I-2 garden-in
'An unfortunate accident happens in the garden.'
(c) sp artsmeni varjisobs.
athlete-NOM he-exercises-I-3
'The athlete is exercising, training.'
The causatives in ( I S) and ( 16) should be compared with the non

causative verbs of ( 14) ; (14a) contains a Class I verb, (I4b) a Class 2,
and (I4c) a Class 3 verb. In ( 1 4) and ( I S) the verbs are in Series I,
and in ( 1 6) in Series II. In ( 14) the nominative-nominal is the subject
of each verb. In ( I S ) and ( 1 6) this nominal has the case marking of the
direct object : dative in Series I (examples ( I S» , nominative in Series II
(examples (16» . In each causative example, the final subject is marked
nominative in Series I, ergative in Series II.
( I S) (a) deda bavlvs a:;t'nebs.
mother-NoM child-DAT she-causes-sleep-him-I-l
'The mother is getting the child to sleep .'
(b) patara biti axdens ubedur
little bOY-NOM he-causes-happen-it-I-I unfortunate
semtxvevas baySi.
accident-DAT garden-in
'The little boy is causing an unfortunate accident to happen in
the garden.'
(c) mrvrtneU sportsmens avarjiSebs.
coach-NoM athlete-DAT he-causes-exercise-him-I - I
'The coach i s making the athlete exercise./The coach is train
ing the athlete.'
(16) (a) dedam bavlvi daa:;ina.
mother-ERG child-NOM she-caused-sleep-him-II-1
'The mother got the child to sleep.'
(b) patarma bitma moaxdina ubeduri
little bOY-ERG he-caused-happen-it-II- I unfortunate
semtxveva baySi.
accident-NoM garden-in
'The little boy caused an unfortunate accident to happen in
the garden.'
( c) 1nfvrtnelma sportsmeni avarjisa.
coach-ERG athlete-NoM he-caused-exercise-him-II- I
'The coach made the athlete exercise./The coach trained the
athlete.'
Thus, with an intransitive verb of any Class in the embedded clause,
the nominal which corresponds to the subject of the non-causative is,
in the corresponding causative, in the case of the direct object ; that is.
it is in the dative in Series I and the nominative in Series II.
5.3 Derived grammatical relatrom 75
3. 1 .2 The causative of a transitive. While there are transitive verbs in

Class 2, they present problems with respect to identifying grammatical
relations (cf. ch. 1 6, Appendix B). Class I and 3 transitives are illus
trated here. In §4 examples of Class 2 transitives with indirect objects are
given, and we will see that their causatives behave like those of Class I
and 3 verbs.
( 1 7) (a) mziam cecxli daanto.

Mzia-ERG fire-NoM she-lit-it-II-1
'Mzia lit the fire.'
(b) mziam itamasa nardi.
she-played-II-3 backgammon-NOM
'Mzia played backgammon.'
Compare the non-causatives of ( 17), which are in Series II, with the
catlsatives of ( I S) and ( I 9), which are in Series I and II, respectively.
I II each of the non-causatives, the subject is mzia ; in the corresponding
rallsatives, mzia is in the indirect object case, dative in both Series. The
I I i reet objects of the non-causatives, cecxli 'fire' and nardi 'backgammon',
are in the case appropriate to the direct object in each causative : dative
i l l St r ies I {examples ( I S» , nominative in Series II (examples (19». The

'
,"u hject of each causative is in the subject case : nominative in Series I,

(' , gat ive in Series II.
( , Ii) ( a ) mama mzias antebinebs cecxls.

father-NoM Mzia-DAT he-causes-light-her-it-I-I fire-DAT
' Father makes Mzia light the fire.'
(11) mzias vatamaseb nards.
I-cause-play-her-it-I-I backgammon-DAT
'I am getting Mzia to play backgammon.'
( i iI) (a) mamam mzias daantebina cecxli.
father-ERG he-eaused-light-her-it-II-1 fire-NoM
' Father made Mzia light the fire.'
( II) mzias vatamase nardi.
I-caused-play-her-it-II-1 backgammon-NoM
'I got Mzia to play backgammon.'
Thllll, the nominal which corresponds to the subject of the non

ulUlMtin' transitive verb is, in the causative, in the case of the indirect
uh,('('I ; that is, it is dative in both Series. The nominal which
76 Causative Clause Unian
corresponds to the direct object of the non-causative is in the case of

the direct object in the causative construction.
3 . 1 . 3 The causative of an intransitive with indirect object. There is a

group of intransitive predicates that take indirect objects ; e.g. esalmeba
'he greets him', etamaJeba 'he plays with him', elapara�eba 'he talks to
him'. These forms are in Class 2 ; they are illustrated in (20) for Series II.
(20) (a) bavsvi miesalma deidas.
child-NOM he-greeted-her-II-2 aunt-DAT
'The child greeted (his) aunt.'
(b) �omisia estumra vanos.
commission-NoM it-visited-him-II-2 VanOS-DAT
'The commission visited Vano.'
The non-causatives in (20) should be compared with the causatives in
(21) and (22), which are in Series I and II tenses, respectively. The
subject of each of the non-causatives bavsvi 'child' and �omisia 'com
-
mission' - in the causative bears the case of the direct object ; that is,
it is dative in Series I and nominative in Series II. The indirect objects
of (20) also bear the case of the indirect object in causatives, that is, the
dative case in both Series.
(21) (a) deda asalmebs bavsvs deidas.
mother-NOM she-causes-greet-her-him-I-I child-DAT aunt-DAT
'The mother is making the child greet (his) aunt.'
(b) dire�tori �omisias astumrebs
director-NOM commission-DAT he-causes-visit-him-it-I- I
vanos.
VanO-DAT.
'The director is having the commission visit Vano.'
(22) (a) dedam miasalma bavsvi
mother-ERG she-caused-greet-her-him-II-I child-NOM
deidas.
aunt-DAT
'The mother made the child greet (his) aunt.'
(b) dire�torma �omisia astumra
director-ERG commission-NoM he-caused-visit-him-it-II-I
vanos.
VanO-DAT
'The director had the commission visit Vano.'
5.3 Derived grammatical relations 77
On the basis of the data discussed in §3. I . 1 and these data, we can
, I raw the following conclusion about the case pattern in the causative
of intransitives : The subject of the non-causative corresponds to the
direct object of the causative, and the indirect object, if there is one, to
t i t(; indirect object.
An analysis that did not treat the nominals being discussed here as
I -rms of the united clause would have to complicate the case marking
r u les. In particular, such an analysis would have to repeat the facts of

d irect object marking for Class 1 verbs : under causative formation,
i n i tial direct objects are marked with the dative case in Series I and the
nominative in Series II. In addition, such an analysis would claim that
t h ' direct object of a Class 2 verb in Series II is marked with the
lIom inative, a case not otherwise used for the direct object of a Class 2
\' orb (cf. (2-3) in the Introduction). The analysis proposed here con-
id "["s all causative verbs to be Class 1 verbs, and it is therefo re natural
I l l d predictable that their direct objects will be in the nominative in
SU'i(;s II. The Class 1 status of these verbs is not determined by the
, '� I tl . marking they govern, but by their morphology (cf. ch. 16, Appen
d i x !\) ; the case marking is predictable from their status as Class 1

\ nbs.
I . ':' Tav-Rejlexivization
' 1 ''' . of Tav-Reflexivization is governed by a clausemate constraint
I' Ilk
( , r. ·h. § 1 .3) ; tav- has as its antecedent the subject of its own clause.
I,
I h l l i n an organic causative, tav- may be coreferential to a nominal

\\ hivh i s, in initial structure, in another clause. Consider again ( 1 2)
I I I I ( 1 3), which are repeated here for the convenience of the reader.
( I ' {·himma vano alaparalJ-a tavis tavze.

dOl:tor-ERG Vano-NoM he':'caused-talk-him-II-I self 's self-on.
'
'Th(; doctorl got Vanoj to talk about himself;, j .
( I I ) J,:f'lIom miatanina rezos {:ignebi
( ; 'nO-ERG he-caused-take-him-it-II-I Rezo-DAT books-NOM
Illvistan.
·If-at
, ; 'nol got RezoJ to take the books to his;, j place.'
I IH' r 'ad i ng
'The doctor got Vano to talk about the doctor' for ( 1 2) is
' 1 1 i I 'IIt with the clausemate constraint on Tav-Reflexivization only
I kill/II/a 'doctor' and ta?!- arc final dependents of the same clause.
Similarly, in ( 1 3) the reading 'Geno made Rezo take the books to

Geno's place' is consistent only with a simple final structure.
3 .3 Person Agreement
In ch. 1 it was shown that only terms trigger Person Agreement ; In
chs. 3-4 it was established that this rule is stated on final terms.
The initial embedded subject does not trigger Subject Person Agree
ment in organic clauses ; it does trigger Direct Object Person Agreement
or Indirect Object Person Agreement, depending upon whether it is
the initial subject of an intransitive or transitive verb. That this is so
is shown by (23), which illustrates the causative of a transitive. The
(*a) sentences have first person Subject Person Agreement v-, the (b)
sentences, first person Object Person Agreement m- (cf. paradigms in
ch. I , §3 ).
(23) (a) *tamadam vamyera.
toastmaster-ERG I-caused-sing-him-II- I
('The toastmaster made me sing.')
(b) tamadam mamyera.
he-caused-sing-me-II-I
'The toastmaster made me sing.'
(24) (a) *ninom gavaflnendina saxli.
Nino-ERG I-caused-clean-her-it-II-I house-NoM
(,Nino made me clean the house.')
(b) ninom gamar;mendina saxli.
she-caused-clean-me-it-II I -
'Nino made me clean the house.'

The Person Agreement in (23-24) is explained by (a) the fact that
Person Agreement is stated on final terms, and (b) the first person in
(23-24), while being initial subject, is final object, as described in §I
above.
3 .4 Unemphatic Pronoun Drop

I showed in ch. I, §4 that this rule operates only on terms ; in ch. 3 I
suggested that it is stated on final terms. Some examples have already
been given in this chapter which show that those nominals which are
terms on the proposed analysis may be dropped. (7) illustrates the
dropping of the initial subject of an embedded transitive. ( 1 8-19b)
illustrate the dropping of the initial matrix subject. (25) and (26) below
5.3 Derived grammatical relati()TlS 79
"how that the initial direct object and the initial subject of an intransi
I i v . can also drop ; the sentences are used naturally when the reference
or the dropped nominals has already been established in discourse.
( 2 5 ) mamam mimacernina dedistvis.

father-ERG he-caused-give-me-it-II-I mother-for
'Father made me give it to Mother.'
( .;: I) 3ayli atirebs.
dog-NOM he-causes-cry-him-I-I
'The dog makes him cry.'
' ( 'h . fact that an initial matrix subject, an initial embedded subject of
t ransitive or intransitive, and an initial embedded direct object can
I
dropped is consistent with the analysis in §I and with the proposal

,II b '
, h.ll Unemphatic Pronoun Drop is stated on final terms.
, Ii Question Formation
\ hown in §4.2 of the Introduction, Q-words must occur immediately
hl' fc l r ' the verbs which govern them. Consider the sentences below.
• rhty 5how that all the nominals under discussion here must occur
1 1 1 1 1 \ \ 'd iately before the clause-united verb.
, 7 (a) Tin
• migacemina forti dedistvis?
whO-ERG he-caused-give-you-it-II-I cake-NOM mother-for
' Who made you give the cake to Mother ?'
'''vin torti migacemina dedistvis?
"'vin dedistvis migacernina torti?
" ) 1flamam vis miacemina forti dedistvis?
father-ERG who-DAT he-caused-give-him-it-II-I
' Who did Father make give the cake to Mother ?'
lfivis mamam miacernina dedistvis torti?
lfimamam miacemina vis dedistvis torti?
( .) 1IIomam ra migacemina dedistvis?
what-NoM he-caused-give-you-it-II-I
' What did Father make you give to Mother ?'
lfi rn mamam migacemina dedistvis?
- menna migacemina ra dedistvis?
P I C 'POS ·d analysis correctly predicts that the nominative-, ergative-,

I d.l l ivc-nom inals are final dependents of the united clause and will
I 1 \ ) 1 1' i l l i m 'd iatcly precede the causative verb in questions.
3.6 Object Camouflage

This rule provides additional evidence that in final structure organic
causatives are simple clauses. It will be remembered that the rule sub
stitutes a possessive pronoun + tavi for a first or second person direct
object when there is an indirect object in the same clause. In organic
causatives the rule applies to final direct objects if there is in the same
clause a final indirect object (= initial embedded transitive subject or
initial embedded intransitive indirect object). This is shown in (28) ; in
(*28b) the rule has failed to apply.
(28) (a) anzorma gaalan3yvzna vanos
Anzor-ERG he-caused-insult-him-him-II-I Vano-DAT
cemi tavi.
my self-NOM
'Anzor made Vano insult me.'
(b) *anzorma gamalan3yvina vanos (me).
he-caused-insult-him-me-II-I me-NOM
(,Anzor made Vano insult me.')
In ch. 3 I showed that Object Camouflage applies only if there is an
indirect object in the same clause. In (28) the only indirect object is
the derived one, vanos, which is an initial subject ; (28) thus confirms
that vanos is an indirect object. Since the rule has only direct objects
as its targets, it confirms that me is the direct object. It also shows that
Object Camouflage must be stated on the output of Causative Clause
Union, not on its input, since the absence of an initial indirect object
in (28) is inconsequential to the application of the rule. 5
3 .7 Summary
In this section and the preceding one I have argued that the initial
structure of an organic causative must be complex, and that its final
structure must be simplex. It has been shown that the initial structure
of an organic causative has as its direct object a clause in which the
relations that nominals bear are the same as the relations borne in a
simple clause. Furthermore, the direct objects of embedded transitives
and the subjects of embedded intransitives behave like direct objects
under Causative Clause Union, with respect to all rules stated on final
relations. The subject of an embedded transitive and the indirect object
of an embedded intransitive behave like indirect objects under these
conditions.
5.4 The tvis-nominal 81
• The tvis-nominal
In the causative of a transitive, the indirect object of the embedded
· 1<'I u s e is marked with the postposition tvis in all Series. (29a) illustrates
I his for Series I, and (29b) for Series II.
(29) la) mama maceminebs torts dedistvis.
father-NOM he-causes-give-me-it-I-I cake-DAT mother-for
'Father is making me give the cake to Mother.'
(b) mamam mimacemina torti dedistvis.
father-ERG he-caused-give-me-it-II-I cake-NOM mother-for
'Father made me give the cake to Mother.'
M ;rems 'gives', the non-causative corresponding to the verb form of
�( ), is a Class I verb. The causatives of Class 2 verbs show the same
pat tern of case marking ; (3oa) illustrates the causative of a Class 2
\' ' I' b in Series I, (30b) in Series II.
( Jo) (a) anzori vanos daapirebinebs

Anzor-NOM VanO-DAT he-will-cause-promise-him-it-I-I
sabe?d mankanas ninostvis.
printing machine-DAT Nino-for
'.'\nzor will make Vano promise Nino a typewriter.'
( b) anzorma vanos daapirebina sabe{di
AnzOr-ERG he-caused-promise-him-it-I I- I
mankana ninostvis.
machine-NoM
'Anzor made Vano promise Nino a typewriter.'
\ 1 1 .I i I of the sentences of (29) and (30) the initial indirect object of the
, 1 1 1 hrdd .J clause is marked with tvis 'for'. The marking of this nominal
I I I I .1 1 1!\a t i vcs in Series III is discussed in ch. I I , §2+
( I 1 1 1 · rule of Clause Union in §I above, predicts that the initial

•
IIId l l l'rl object of an embedded transitive will be a derived non-term.
I t v is-nominal as a non- term governed by the organic causative

/'I/('
I I {jll('sfion Formation. I have shown previously that Question For
a test for clause constituency, since just those nominals
, I 1 1 I�'d hy :1 verb immediately precede it when they are questioned.
I H OW, th nt the t vis - n om i nal, when questioned, must immediately

I I I \' I l w organ i c c ausa t i ve .
(3 r ) mamam vistvis migacemina torti?
father-ERG who-for he-caused-give-you-it-II-r cake-NoM
'Who did Father make you give the cake to ?'
*vistvis mamam migacemina torti?
*mamam migacemina forti vistvis?
Hence, the tvis-nominal must be governed by the clause-united verb.
4. I .2 Case Assignment. In previous chapters I showed that all final terms

are marked with the nominative, ergative, or dative case in Series I and
II. The tvis-nominal is not marked with one of these cases. It follows
that it is not a final term.
4. r · 3 Unemphatic Pronoun Drop. In chs. r and 3 it was shown that Unem

phatic Pronoun Drop applies to all and only unemphatic final term
pronouns. If the tvis-nominal were a final term, it should undergo this
rule.
(32) shows that the tvis-nominal cannot be dropped without changing
the meaning of the sentence. Compare (32a), where the initial indirect
object is also final indirect object, to (32b), where the initial indirect
object is a final non-term. Only the former may be the target of
Unemphatic Pronoun Drop.
(32) (a) genom momca torti.
Geno-ERG he-gave-me-it-II-r cake-NoM
' Geno gave me a cake.'
(b) mamam genos miacemina forti
father-ERG Geno-DAT he-caused-give-him-it-II-r cake-NoM
cemtvis.
me-for
'Father made Geno give me a cake.'
tc) ?mamam genos miacemina torti.
'Father made Geno give the cake (to someone).'
The fact that this pronoun is not droppable confirms that it is a final
non-term.
4. I .4 Person Agreement. The rules of Person Agreement also provide

evidence that the tvis-nominal in organic causatives is not a final term.
This nominal triggers none of the agreement rules ; this is shown in
54 The tvis-nominal 83
(33), where the person of the tvis-nominal changes, but the verb form
docs not.
(33) (a) mamam forti miacemina cemtvis.
father-ERG cake-NoM he-caused-give-him-it-I I-I me-for
' Father had a cake given to me.'
(b) mamam forti miacemina sentvis.
you-sG-for
' Father had a cake given to you (SG).'
( c) mamam forti miacemina genostvis.
Geno-for
' Father had a cake given to Geno.'
S i nce only final terms trigger Person Agreement (cf. chs. 3 and 4), the
lal:k of agreement in ( 33) is explained by the proposed analysis, accord
i ng to which the initial indirect object of a transitive verb is not a final
I ' .. 01 in the corresponding organic causative.
, . 1 . 5 Summary. Question formation provides a test of clause constitu

'1 1 'y and shows that the tvis-nominal is a final dependent of the simplex
-I .' I ISC, not of the embedded clause in which it originated. Other syn-
1 . 1 'l ie rules also show that the tVzS-nominal is a final non-term, a
I ·t i red term. 6 Additional support for this position may be found in
, h. I I.
\ _ ' On so-called 'jour-person causatives'

Ti oiditional analyses have discussed causatives of transitives in which
I I I ' i n i tia l embedded indirect object is in the dative case (cf. Kiknadze
I I) .n ; Shanidze 1 973 : §43 1 ; Vogt 1 97 1 : §z.7S). (34) illustrates this
e 1 1 1 1 t ruction.
(1\ 'da kals ar migaceminebt teimurazsa. '

1 1 1 1 I woman-DAT not I-cause-give-you-her-I-I Teimuraz-DAT
' . . . and I do not permit you to give the woman to Teimuraz. ' 7
1 " I II ' mbcdded clause, the initial grammatical relations are the foI
l"" I l Ig : subject - tkven 'you' ; direct object - kali 'woman' ; indirect
,hilT I - teimuraz. Each of these is in the dative, though tkven 'you'
I d ro p ped by Unemphatic Pronoun Drop, having triggered Object
I'e I 0 1 1 g r ment i n the organic causative. The existence of such sen
, III I I l lak 's causative formation in Georgian seem u nsystematic and
throws doubt on the accuracy of rule (3), since teimuraz seems to be a

final indirect object, in contrast to the analysis proposed above.
In the remainder of this section, I will make the following points
about examples like (34). (i) They are taken from earlier periods of
Georgian or from non-standard modern dialects, not from the dialect
being described in this work. (ii) There are certain idioms where the
same case pattern is found in the standard modern dialect. (iii) These
idioms are limited to the first and second persons. (iv) In those dialects
where examples like (34) are to be found, one of the dative-nominals is
a retired term, not an indirect object as has been claimed.
Sentences like (34) are unacceptable in Standard Modern (Tbilisi)
Georgian, though they are grammatical both in older Georgian, and
in some modern dialects. Kiknadze, Shanidze, and Vogt cover a broad
period of Georgian and a range of dialects, without always noting the
historical and geographical contexts of various facts. My informants
rejected examples of this type quoted in the works cited, though they
observed that they could understand the sentences.
There are, however, at least two idioms in which this construction
remains in the standard modern dialect. These are sema(mevina 'he;
made me feed himj it' and damalevina 'he; made me give it to himJ to
drink'. 8 These idioms in Modern Georgian, and apparently also in the
construction illustrated in (34), are limited to instances where the
initial embedded subject is first or second person, as shown in (35) and
(*36). (35) contains a grammatical sentence with the idiom for 'cause
feed', where the initial embedded subject is first person. (*36) is
exactly parallel to (35), but the initial embedded subject is third person.
(35) dedam bavSvs papa sema(mevina.

mother-ERG child-DAT gruel-NOM she-caused-feed-me-it-II-I
'The mother made me feed gruel to the child.'
(36) *dedam kals bavsvs papa sea(mevina.

woman-DAT she-caused-feed-her-it-II - I
(,The mother made the woman feed gruel to the child.')
While the double-dative idiom can be used with an initial embedded

first or second person subject, it cannot be used with a third person, as
the ungrammaticality of (*36) shows.
Comrie ( I 976e) discusses this construction in Georgian, based on
the examples quoted by Vogt. Comrie states that there are two indirect
5.5 Theoretical implications 85
objects, a claim he apparently bases on case marking. Notice, however,

that the embedded initial indirect object has no characteristics of final
indirect objects, apart from case. In particular, it cannot trigger Person
Agreement, Number Agreement (cf. ch. I S), or Unemphatic Pronoun
Drop.9 Nor is it productive ; in Standard Modern Georgian it is limited
to a very few idioms, and to first and second person initial embedded
subjects. I suggest that the dative-nominal which is an initial embedded
indirect object is finally a retired indirect object, just like the tvis
nominal in (29). This proposal is not inconsistent with the case marking,
since case marking alone does not define termhood. Postal & Perlmutter
( I 974) have shown that in other languages retired terms often bear the
case marking of the relation they formerly held. 10
Since bavSvs 'child' in (35) and kals 'woman' in (34) have none of the
properties of final indirect objects, save case, and since there is inde
I ,cndent evidence that retired terms in some languages bear the case
marking of terms, I conclude that these nominals, like the tvis-nominal
i n (3 6b), are retired indirect objects. Causative formation in Standard
10dern Georgian is not unsystematic ; rather rule (3) represents the
l:Iystcm, to which there are a very few idiomatic exceptions, such as (35).
5 Theoretical implications
( ; 'orgian provides very strong arguments in favor of an analysis of

ingle-clause causatives on which initial dependents of both clauses
h . 'ome dependents of the matrix clause. An analysis on which initial
( i l-p 'ndents of the complement are not final dependents of the matrix
d ; l l ise would be inadequate in Georgian because (i) the derived term
dl' p 'ndcnts are marked, not with one case, but with whatever case is
' '1 ,propriate to the derived term relation it holds ; (ii) the derived term
dl' p 'nd cnts have the morphological (Person Agreement) characteristics
I l l' f i n al tcrms ; and (iii) the derived terms have the syntactic character
I I in. of final terms (triggers of Tav-Reflexivization, Unemphatic
" I O IlOlln Drop, and Object Camouflage).
( ; ·orgian provides evidence in favor of the formulation of Clause
l I l i on givt:n in (3), in that it shows clearly that the tvis-nominal (initial
1 1 1 1 1 1 ·t lded indirect object) is finally a non-term. It has also been shown
1 1 1 0 1 1 what Rome grammarians have considered a double indirect object
I I(': d ly one derived indirect object and a retired indirect object.
·. l l Ifiativc formation provi des strong evidence in favor of relational
grammar, since the cross-language regularities of causatives with sim

plex derived structures cannot be stated in a simple way on the basis
of linear order and dominance relations. The fact that the theory of
relational grammar makes the correct predictions for Georgian provides
confirmation for this theory.
6 Version: rules tha t create indirect
objects
In Georgian there are syntactic rules that create indirect objects; these
phenomena are traditionally referred to as kceva or 'version'. 1 Three
types are illustrated in ( 1-3) ; in the (b) sentences there is an indirect
object ; in the (a) sentences there is none.
( I) Benefactive Version
(a) gelam self.era axali sarvali fentvis.
Gela-ERG he-sewed-it-II- I new trousers-NOM you-for
' Gela made new trousers for you.'
(b) gelam segiljera axali farvali (fen).
he-sewed-you-it-II-I YOU-DAT
'Gela made new trousers for you.'
(2) Possessive Version
(a) mzia pnends dis pexsacmlebs.
Mzia-NoM she-cleans-it-I-I sister-GEN shoes-DAT
'Mzia is cleaning her sister's shoes.'
(b) mzia Ut;mends das pexsacmlebs.
she-cleans-her-it-I - I sister-DAT
'Mzia is cleaning her sister's shoes.'
( 3) Superessive Version
(a) gelam surati daxata lJ,edelze.
Gela-ERG picture-NoM he-painted-it-II-I wall-on
'Gela painted a picture on the wall.'
(b) gelam surati daaxata Ifedels.
he-painted-it-it-II-I wall-DAT
'GcIa painted a picture on the wall.'
There are the following significant differences between the (a) and (b)
Icntenccs in each pair :
88 Version
(i) In the (b) sentences there is a dative-nominal ; in the (a) sentences

this nominal (henceforth the 'version nominal') is marked with a
postposition (-tvis or -ze) and/or the genitive case.
(ii) In the (b) sentences, but not the (a) sentences, a version marker
has been added to the verb form (i- in ( I ), u- in (2), and a- in (3)).
(iii) In the (b) sentences, but not the (a) sentences, the verb agrees
with the version nominal in person (g- in ( I ), u- in (2), 0 in (3)).
(iv) In the (b) sentences, but not the (a) sentences, the version nominal
is deletable by Unemphatic Pronoun Drop.
In this chapter it is argued that in the (b) sentences of ( 1-3) a version

rule has applied to create an indirect object.
I An analysis
The pair of sentences in ( I) can be explained by a rule of Benefactive

Version (Benefactive Advancement), which advances a benefactive non
term nominal to indirect-objecthood.
(4) Benefactive Version

A benefactive nominal may advance to indirect object.
SIDE EFFECT : Add i- as a prefix to the root of the verb which governs
the version nominal.

( I b) may be represented as network (5).
(5)
gela axali sarvali sen se�erva

, new ) ' trousers ) , you ' ( sew '
Possessive Version, illustrated in (2b), is like Benefactive Version in

that it creates an indirect object and in that the side effect of the rule
is the same. (The morphological difference between ( I b) and (2b) is
the result of Person Agreement, which is discussed in §2. 1 below.)
Possessive Version, however, may belong to a different rule-type, the
6.2 Benefactive Versian 89
ascension (cf. Aissen 1 980 ; Perlmutter, to appear c) ; it will not be

discussed in detail here.
Superessive Version, like Benefactive Version, is an advancement of
an initial non-term nominal to indirect-objecthood. Its side effect is
the prefixation of the version vowel a-. It is also not discussed in detail
here.
The indirect objects created by Possessive Version, Superessive Ver
sion, Locative Version, and Comitative Version behave like those
created by Benefactive Version in all significant respects (cf. Boeder
1968 ; and Harris 1976 : ch. 6) ; they present no known counter examples
to the analysis presented here.
2 Benefactive Version
In this section I show that the analysis proposed in (4-5) makes the
correct predictions with respect to rules previously considered.
2. I Person Agreement
The basic facts of Person Agreement were presented in ch. I, §3. It
was shown in succeeding chapters that Person Agreement is stated on
final terms. The proposed analysis predicts, then, that the derived
indirect object triggers Indirect Object Person Agreement. This is
correct, though not necessarily immediately obvious to the reader who
does not know Georgian, since the markers differ somewhat from those
described in ch. I, for reasons explained below.
Earlier I noted the following paradigm, which represents an inter
section of person and number.
(6) Indirect Object Person Agreement
singular plural
I . person m- gv-
2. person g g-t
3 . person S/h/0- S/h/0-t
(7) illustrates the use of these markers m sentences corresponding
l ( I b).
(7) (a) §e-m-H�era.
'He sewed it for me.'
(b) §e-g-i �era .
-
'He sewed it for you (sc).'

90 Version
(c) se-u-/f-era.
'Hei sewed it for himj .'
The object markers occur together with a 'version vowel' or 'version
marker' , which is introduced by the rule of Benefactive Version, ( 4).
In (7a, b) we can clearly distinguish the first and second person markers. 2
But in (7c) the object marker that is elsewhere realized as S-, h-, or '"
has combined with the i-marker by a synchronic morphophonemic
rule to give u. An i- prefix occurs in Georgian (i) as a marker of Benefac
tive or Possessive Version, (ii) as part of the marker of future, aorist,
etc. for Class 3 verbs (cf. ch. 1 6, §I), (iii) as part of the marker of the
first evidential for Class 1 and 3 verbs, and (iv) as an empty morpheme
with a few verbs. In any one of these functions it may combine with
the third person indirect object marker as the fusional morpheme u-.
(8) +h+i+ � + u +
The morpheme h- and a following morpheme i- combine to u-.
One example of this rule occurs in (9c), where i- functions as part of
the future marker for a Class 3 verb.
(9) (a) miqvirebs.
he-will-yell-me-I-3
' He will yell to me.'
(b) giqvirebs.
he-will-yell-you-I-3
'He will yell to you.'
(c) uqvirebs.
he-will-yell-him-I-3
'He will yell to him.'
Another example may be found in ch. 8, n. 5 illustrating the second
evidential. Thus, in spite of the occurrence of the u- variant in the
third person, we can see that (7) contains regular indirect object markers.
The fact that the dative-nominal in sentences like (I b) triggers Indirect
Object Person Agreement, while the object of the postposition tvis
in sentences like ( I a) does not trigger any Person Agreement, supports
the analysis of the former as an indirect object and the latter as a
(benefactive) non-term.
2.2 Case
In ch. 2 it was shown that the indirect object is marked with the dative
case in both Series I and Series II. In ( lob) merab is in the dative in
6.2 Benefactive Version 91
Series I I ; this is good evidence that merab is the indirect object in that
sentence. In ( l Oa), on the other hand, merab is not in the case of the
indirect object, but in the genitive with a postposition ; this is good
evidence that merab is not the indirect object in that sentence.
( 10) (a) gelam se�era axali sarvali merabisatvis.
Gela-ERG he-sewed-it-II-I new trousers-NOM Merab-for
'Gela made new trousers for Merab.'
(b) gelam seu�era axali sarvali merabs.
he-sewed-him-it-II-I Merab-DAT
'Gela made new trousers for Merab.'
( I I ) corresponds to (10), but is in a Series I tense.
( I I) (a) gela �eravs axal sarvals merabisatvis.
Gela-NOM he-sews-it-I-I new trousers-DAT
'Gela is making new trousers for Merab.'
(b) gela u�eravs axal sarvals merabs.
he-sews-him-it-I-I
'Gela is making new trousers for Merab.'
In ( I I a) merab is the genitive object ,Of the postposition -tvis, in ( I Ib)
it is in the dative. This is consistent with the latter, but not the former,
being an indirect object.
2.3 Unemphatic Pronoun Drop

It has been shown that the rule of Unemphatic Pronoun Drop drops
all and only unemphatic pronouns that are final terms. As indicated
above, sen 'you' can be dropped from ( Ib), but not from ( r a). This is
consistent with the analysis of this nominal as a benefactive non-term
in (a)-type sentences and as an indirect object in (b)-type sentences.

The rule of Object Camouflage replaces a first or second person direct
object with a possessive pronoun + tavi If there is an indirect object in
the clause. ( 1 2) shows that this rule applies when there is a version
indirect object in the clause. In ( I 2a) neither Benefactive Version nor
Object Camouflage has applied. In ( * I 2b), Benefactive Version has
created an indirect object, but Object Camouflage has failed to apply.
In ( r 2c) Benefactive Version and Object Camouflage have both applied.
92 Version
( 1 2) (a) vazam dagxata (Sen) cemtvis.

Vazha-ERG he-painted-you-II-I YOU-NOM me-for.
' Vazha painted you for me.'
(b) *vazam damixata (Sen) (me).
he-painted-me-it-II-I I-DAT
('Vazha painted you for me.')
(c) vazam damixata seni tavi (me).
your self-NoM
'Vazha painted you for me.'
In ( 1 2C) tavi can have its more concrete meaning 'head', in addition to
'self' (cf. ch. I , S I ). For this reason, another expression, such as seni
partreti 'your portrait' is preferred by some speakers for the more
general meaning. But the fact that (*I2b), without Object Camouflage,
is ungrammatical, while ( I 2C), with Object Camouflage, is grammatical
shows that this rule must apply here. Since Object Camouflage applies
only when there is an indirect object in the clause, me 'me' must be an
indirect object.
2.5 Summary
The four phenomena discussed above show that the version nominal,
when it is in the dative case, is the indirect object, which supports the
final grammatical relations proposed in § r . The initial grammatical
relations are discussed in the following section.
3 Stative verbs
Holisky (1 978) shows that true stative verbs in Georgian consistently
do not occur with benefactive nominals, as illustrated in (* 1 3).
(13) (a) *pasuxi vicodi mas�avleblisatvis.

answer-NOM I-knew-it-II-3 teacher-for3
('I knew the answer for the teacher.')
.
(b) *mcioda dedisatvis.
l-cold-I-4 mother-for
('I was cold for my mother.')
The same set of verbs do not occur with verSlOn objects, as (*14)
shows.
6.4 Direct generation 93
( 1 4) (a) *pasuxi mast;avlebels vucodi.

teacher-DAT 1-knew-him-it-II-3
('I knew the answer for the teacher.')
(b) *skeli dedas vuqavi.
fat mother-DAT 1-was-her-II-2
('I was fat for Mother.')
If version objects are derived from non-term benefactive nominals, as
proposed in §I, the ungrammaticality of (*14) follows automatically
from that of (*1 3).
4 Why not generate version objects directly?

Since some verbs have initial indirect objects, it does not seem unreason
able to propose generating all indirect objects in the same way.
Consider the sets of sentences below. They show a correlation between
the presence of a character vowel (here u-), the occurrence of an
indirect object, and the meaning attached to that indirect object. What
is especially interesting is that the pattern for the verb mit;era 'write' is
different from that for the verb dat;era 'write' (cf. also ch. I, §4).
( 1 5) ta) t;erili mist;era 3mas.
letter-NOM he-wrote-him-it-II-I brother-DAT
'He wrote a letter to (his) brother.'
(b) t;erili mist;era 3mistvis.
brother-for
'Hei wrote a letter to himj for his brotherk.'
( c) * t;erili m£ut;era 3mas.
(,He wrote a letter to him for his brother.' f'He wrote a letter
to his brother.')
( 1 6) (a) *t;erili dast;era 3mas.
(,He wrote a letter to (his) brother.')
(b) t;erili dat;era 3mistvis.
he-wrote-it-II- I brother-for
'He wrote a letter for his brother.'
(c) t;erili daut;era 3rnas.
'He wrote a letter for his brother.'
94 Version
I propose to account for the different patterns exhibited by these two

verbs in the following way. Mir;era obligatorily takes an initial indirect
object in what we may loosely describe as the 'to-relation'. Dar;era may
not take an initial indirect object. As in (6) above, third person indirect
objects trigger the sjhj0-prefix, which is realized as s before r;. These
facts account for the (a) sentences of ( 1 5-16). Both mir;era and dar;era
may take a benefactive-nominal ; as non-terms, such nominals do not
trigger agreement. This accounts for the (b) sentences above ; in (I Sb)
the initial indirect object is dropped by Unemphatic Pronoun Drop, as
shown in the gloss. The rule of Benefactive Version may advance the
benefactive-nominal in ( 1 6), where there is no initial indirect object,
but not in ( IS), where there is an initial indirect object. 4 The different
meanings of ( I sa) and ( I 6c) are attributable to their different initial
structures. The difference between the meanings of ( I Sb) and ( I 6b) is
also to be attributed to the fact that the former, but not the latter, has
an initial indirect object.
Consider now a different analysis, on which version objects are not
derived from non-term nominals. Such a grammar would state that
mir;era takes an indirect object obligatorily, while dar;era takes one
optionally. Instead of a rule of Benefactive Version, it will have rules
relating the benefactive relation to the occurrence of i- in the verb
form and the 'to-relation' to the occurrence of 0 in the verb form.
This direct-generation analysis thus requires one rule more than the
analysis proposed here.
In addition, that analysis fails to capture the fact that indirect objects
in the 'to-relation' exist in all languages, while those in the benefactive
relation do not. The fact that the 'to-relation' is morphologically un
marked must be treated as an accident by that grammar. The analysis
proposed here, on the other hand, embodies the claim that the 'to
relation' for indirect objects is basic and unmarked, while the benefactive
relation for indirect objects is derivative and marked.
The importance of this difference is even more obvious when we
consider Superessive Version, illustrated in (3). The direct-generation
analysis, if extended to Superessive Version, would require that the
superessive indirect object in (3b) be associated by rule with the occur
rence of the a-prefix in the verb form. According to the analysis pro
posed here, the superessive indirect object is derived by Superessive
Version, the side effect of which is the insertion of a-. Again, only
the latter analysis captures the fact that indirect objects in the
6.5 Coreferential Version Object Deletion 95
superessive relation do not exist in all languages, are derivative, and are
marked.
5 Coreferential Version Object Deletion
I n (17) the version nominal is coreferential to the subject of its clause.

In (17a), Tav-Reflexivization has applied ; in (*17b) both Benefactive
Version and Tav-Reflexivization have applied, according to the rules
as stated.
( 1 7) (a) gela tavistvis �eravs axal sarvaZs.
Gela-NoM self-for he-sews-it-I-I new trouserS-DAT
'Gela is making new trousers for himself.'
(b) *gela (tavis) tavs u�avs axal sarvals. 5
self's self-DAT he-sews-him-it-I-I
('Gela is making new trousers for himself.')
(c) gela i�eravs axal sarvals. 6
he-sews-self-it-I - I
'Gela is making new trousers for himself.'
In ( 17c) the benefactive nominal is not overt. It is like the examples of
Benefactive Version considered earlier, in that benefaction is included
in the meaning, and the marker of Benefactive Version, i-, is present
in the verb form. ( 1 7c) differs from ( l I b) in the following significant
respects :
A. In (17c) the subject and the benefactive nominal are coreferential ;
in ( I I b) they are not.
B . In ( l Ib) there is third person Indirect Object Person Agreement (cf.
§2. 1 ) ; in (17c) there is no Indirect Object Person Agreement.

c. In ( l Ib) the version object is overt ; in (17C) it is not.
In the remainder of this section, I will discuss the treatment of ( 17)

in the grammar of Georgian.
5 . 1 Analysis
I propose that ( 1 7) be accounted for in the following way. The rule of
Benefactive Version, as proposed in § I , does not refer to coreference. A
rule of Coreferential Version Object Deletion obligatorily deletes
reflexive version objects. (A similar analysis is proposed for Tzotzil, a
Mayan language (Aissen ms.).) There are two obvious ways to state
96 Version
this condition : (i) An object created by the application of a version rule

deletes under the conditions specified. (ii) An initial non-term (of any
verb) that is a derived indirect object deletes under the conditions
specified. I know of no reason to prefer one of these over the other ; the
rule below is stated on the basis of a 'version object', which could be
defined in either of these ways.
( I 8) Coreferential Version Object Deletion
A reflexive version object deletes. 7
5.2 Deletion of indirect objects

Indirect objects coreferential to the subjects of their clauses do not gen
erally delete in Georgian. ( I9) illustrates a transitive verb which takes
an obligatory indirect object. In ( 19a) the indirect object coreferential
to the subject has been reflexivized ; this should be compared with
(*17b). (I9b) is superficially similar to ( 17c), in that the indirect object
does not appear ; but (19b), unlike (17c), must be interpreted with a
non-coreferential indirect object.
(19) (a) �erili miv�ere (tems) tavs.
letter-NOM I-wrote-him-it-II-1 my self-DAT
' I wrote a letter to myself.'
(b) �erili miv�ere.
'I wrote a letter to him. '
(20) and ( 2 I) illustrate other verbs with indirect objects ; they pattern
like (19), not like ( I7).
(20) (a) merabi elaparalJ-a tavis tavs.
Merab-NoM he-talked-him-II-2 self 's self-DAT
'Merab talked to himself.'
(b) merabi elap ara�a.
'Merab talked to him.'
(2I) (a) merabi umreris tavis tavs.
he-sings-him-I-3 self's se1f-DAT
'Merab is singing to himself.'
(b) merabi umreris. 8
'Merab is singing to him.'
(22) shows that derived indirect objects, too, may not delete under
the stated coreferentiality conditions. It contains a causative in which
6.5 Coreferential Version Object Deletion 97
the derived indirect object is coreferential to the final subject. In (22a),

which is comparable to (* 17b), Tav-Reflexivization has applied. In
(22b), which is superficially similar to ( I 7c), there is no overt indirect
object. But in (22b), unlike (17c), the deleted indirect object is in
terpreted as an unspecified (non-coreferential) nominal.
(22) (a) gelam gaafmendina tavis tavs

Gela-ERG he-caused-clean-him-it-II-I self's self-DAT
sasaxle.
palace-NoM.
' Gela forced himself to clean the palace.'
(b) gelam gaafmendina sasaxle. 9
' Gela had the palace cleaned. /Gela had him clean the palace.'
Sentences ( 1 9-22) show that indirect objects coreferential to the sub

jects of their clauses do not necessarily delete in Georgian. Rather, just
those indirect objects that are derived by the application of a version
rule are subject to deletion, as in (17) .
5.3 Other possible analyses

Traditional grammars analyze Georgian mainly on the basis of mor
phology or final relations. Since (17c) is formally quite different from
( I Ib), as pointed out in the introduction to this section, these verb
forms are traditionally considered quite different. The former is called
sataviso kceva 'the version for the self', the latter sasxviso kceva 'the
version for another'( cf. Chikobava 1950 ; Fahnrich 1965 ; Schmidt 1965 ;
Shanidze 1973 ; Tschenkeli 1958 ; and Vogt 1 971).10 The extension of
that approach is that these two form-types are derived by two distinct,
unrelated rules. While I do not know of anyone extending the tra
ditional point of view to a grammar stated in terms of precise rules, I
would like to argue that such an extension would be inadequate.
Consider what each analysis would require to account for the examples
in ( I I) and (17).
The proposed analysis (Grammar A) :

a. Coreferential Version Object Deletion, rule ( 1 8).
b. Rule ( 1 8) is obligatory.
c. Rule ( 1 8) is stated on non-final terms.
Description : The optionality of Benefactive Version accounts for the

98 Version
grammaticality of both ( I Ia) and (l Ib) and of both ( 1 7a) and ( 17c). The
obligatoriness of rule ( 1 8) accounts for the ungrammaticality of (*17b).
Since rule ( 1 8) is stated on non-final grammatical relations, the deleted
version nominal does not trigger rules like Person Agreement, which
are stated on final termhood. This accounts for the difference between
the morphology of (l I b) and that of ( 1 7c) (cf. §2. 1 ).
The extended traditional analysis (Grammar B) :
a. Rule (23) : Coreferential Benefactive Deletion
A reflexive non-term benefactive nominal deletes.
b. Side Effect of rule (23) : Add i- as a prefix to the root of the verb
which governs the deleted nominal.
c. Rule (23) is optional.
d. Rule (23) is stated on initial terms.
e. Constraint (24) : Version may not apply to nominals coreferential
to the subjects of their clauses.
Description : ( I Ia, b) are accounted for as above. Rule (23), unlike
( 1 8) , applies directly to initial structures, to grammatical relations that
have not been changed by Benefactive Version. The optionality of (23)
accounts for the grammaticality of ( 1 7a). Rule (23) and its side effect,
applying to initial terms, account for the grammaticality of ( 1 7c).
(*17b) is derived by the application of Benefactive Version and Tav
Reflexivization, which must be included in any grammar of Georgian.
In order to block (* 17b), constraint (24) is added to the grammar.
Evaluation. Both grammars must include Benefactive Version, a state
ment of its side effect, and a statement of its optionality. Rules ( 1 8)
and (23) are of equal complexity. But the Extended Traditional Analysis
is significantly more complex and fails to capture two important gen
eralizations.
(i) The Extended Traditional Analysis requires the extra state
ment B(b). It fails to capture the generalization that the side effect of
rules (4) and (23) are the same. The proposed analysis, on the other
hand, does not repeat this side effect ; it captures this important gen
eralization by deriving ( I I ) and ( 17) through the application of Benefac
tive Version, with one side effect.
(ii) The Extended Traditional Analysis requires the extra condition
(24) in order to block (*17b). On the analysis proposed here, the non
occurrence of such sentences is an automatic consequence of two
rules that are independently needed, Benefactive Version and Coref-
6.6 So-called four-person verbs 99
erential Version Object Deletion. But on the Extended Traditional

Analysis, an entirely ad hoc condition must be introduced.
It is clear that the analysis proposed here must be preferred over the
extension of the traditional point of view, since it explains the gram
matical status of the sentences of ( 1 7) and establishes the relatedness
between ( l I b) and ( 1 7c).H
The discussion above raises the possibility of a third type of grammar :
Grammar C. Description : No additional rule is needed. Benefactive
Version applies freely, and the version object is dropped by the rule
of Unemphatic Pronoun Drop.
This grammar is obviously simpler than the proposed analysis, but
it is shown below that Grammar C makes the wrong predictions.
Arguments against Grammar C. (i) According to the rule of Unemphatic
Pronoun Drop, a pronoun is dropped only if it is unemphatic; an
emphatic pronoun is retained. But the benefactive indirect object coref
erential with the subject cannot be retained, even as an emphatic
pronoun, as shown in (*17b).
(ii) In no other situation does Unemphatic Pronoun Drop apply to
reflexives (cf. ch. 3, §3·2).
(iii) Nominals dropped by Unemphatic Pronoun Drop trigger Person
Agreement ; that this is so is shown in ch. I, §4. N ominals deleted by
Coreferential Version Object Deletion do not trigger Person Agreement ;
this is shown by example ( 1 7c) above.
I conclude that the analysis proposed in §S. I is the preferred gram
mar and that ( 1 7c) is derived by the application of Benefactive Version
and Coreferential Version Object Deletion.
6 On so-called four-person verbs
Kiknadze ( 1937) and Shanidze ( 1 973) give examples of sentences with

two datives in Series II and three in Series I, stating that such sentences
have two indirect objects. One such sentence is repeated here as
example (25).
(25) ' . . . svilebs qeli dami(ra.'
children-DAT throat-NoM he-cut-me-it II-I
'. • He cut my children's throats.' (Shanidze 1 973 : §402)
•
Sentences of this type are generally not acceptable in the dialect de
scribed here, the modern Tbilisi dialect. But in the dialects where they
100 . Version
are acceptable, such sentences pose a problem for the Stratal Unique
ness Law, the claim that there can be at most one subject, one direct
object, and one indirect object in a clause (cf. Introduction, g2).
The verb da{ra 'cut' does not take an initial indirect object ; I assume
that the two dative-nominals (Svilebs and me 'me') were derived by
successive applications of Possessive Version, from a structure contain
ing the direct object cemi svilebis qel(eb)i 'my children's throat(s)'. When
me becomes an indirect object, svilebs becomes an indirect object
chomeur. 12
The only evidence that could be cited to show that svilebs 'children'
is an indirect object in (25) is the fact that it is in the dative case in
Series II. But this is not sufficient evidence, for in natural languages, a
nominal may become a retired term!, while still bearing the case marking
of a termi (cf. Perlmutter & Postal 1974). This does not occur in
Standard Modern Georgian, but it did occur in Old Georgian (cf.
Harris 1979) and apparently in the dialect represented by (25).
Notice that me 'me', but not svilebs 'children', has the following
additional properties of an indirect object. (i) It triggers Indirect
Object Person Agreement (m-marker). (ii) In the first and second
persons, it triggers Number Agreement.13 (iii) It can be dropped by
Unemphatic Pronoun Drop, as shown by (25). That svilebs cannot be
dropped is shown by the fact that qeli dami{ra means 'he cut my throat',
not 'he cut my someone's throat' (cf. Shanidze 1 973 : §402). I have
been unable to test additional aspects of this construction, as I have
had no informant who spoke this dialect.
Since Svilebs does not have the other characteristics of an indirect
object, I conclude that it is not an indirect object. The principal differ
ence between this dialect and the standard one is that in the non-standard
dialect, some former terms carry the case marking of the term relation
which they last bore. As shown in ch. I I , this is not true in the standard
dialect.
7 Version as relation-changing rules
I have proposed that the Version rules are form�lated on the basis of
the relations benefactive, possessor, and superessive. A theory of gram
mar according to which rules were based on linear order and dominance
relations would be forced to formulate Version on the basis of the post
position or case with which the nominals concerned occurred, or on the
6.7 Relation-changing rules 101
basis of their position with respect to some other constituent. But both
alternatives present enormous difficulties, as I will show below.
Consider, for example, a formulation of Possessive Version on the
basis of the case of the possessor : 'Genitive nominals may become in
direct objects.' (26a) illustrates a grammatical sentence with a genitive ;
the ungrammaticality of (*26b) shows that this nominal cannot be an
indirect object in this sentence.
(26) (a) kvis saxli avasene.

stone-GEN house-NOM I-built-it-II-I
'I built a stone house.'
(b) *saxli avusene kvas.
I-built-it-it-II-I stone-DAT
('I built a stone house.')
As shown by (26), some genitive nominals cannot become indirect

objects ; it is only those that bear the possessor relation that may
undergo this rule.
Similarly, only those tvis-nominals which bear the benefactive
relation can undergo Benefactive Version. (27a) illustrates a non
benefactive tvis-nominal ; (*27b) shows that it may not be an indirect
object.
(27) (a) mot;apeebma unda moamzadon es ga�vetili

pupils-ERG should they-prepare-it-II-I this lesson-NoM
xvalistvis.
tomorrow-for
'The pupils should prepare this lesson for tomorrow.'
(b) *mor;apeebma unda moumzadon es ga�vetili
they-prepare-it-it-II - I
xvals.
tomorrOW-DAT
('The pupils should prepare this lesson for tomorrow.')
Notice that an analysis that referred to the possessor by its position

with respect to a head nominal would also be unable to account for the
ungrammaticality of (*26b), since the unmarked position of kva is
exactly the same as the unmarked position of gela in the grammatical
pair (28).
102 Version
(28) (a) gelas saxli avasene.

GelaS-GEN house-NoM I-built-it-II-I
'I built Gela's house.'
(b) saxli aVuSene gelas.
I-built-him-it-II-I Gela-DAT
'I built Gela's house.'
Since a grammar based on linear order cannot refer in a straight
forward way to the correct set of nominals, it cannot account for the
sentences in a natural and simple manner. Relational grammar, on the
other hand, refers to exactly the set of nominals involved, those that
bear the relations possessor, benefactive, or superessive.
8 Summary
In Georgian there are rules that create indirect objects ; the derived
indirect objects have all of the syntactic characteristics of superficial
indirect objects. A grammar that derives version objects from nominals
bearing non-term relations is simpler than one which generates all
indirect objects directly. A grammar which includes a rule of Corefer
ential Version Object Deletion is simpler than one that deletes non
term benefactives directly. It has been shown that the so-called second
indirect object which is acceptable in some dialects does not have the
properties of an indirect object, aside from the case marking. Thus, the
Stratal Uniqueness Law and the Chomeur Law have been upheld in
this set of data. 1 4
7 Passivization
The (b) sentences below illustrate the Georgian passive construction,

the (a) sentences the corresponding ' direct construction' . !
( I ) (a) vasls miscems mast;avlebels.

apple-DAT he-gives-him-it-I-I teacher-DAT
'He will give an apple to the teacher.'
(b) vasli micemulia mast;avleblistvis.
apple-NoM given-it-is-I-2 teacher-for
'The apple is given to the teacher.'
(2) (a) 3arli u�bens bavsvs.

dog-NOM he-bites-it-I-I child-DAT
'The dog is biting a child.'
(b) bavsvi da�benilia 3arlis mier.
child-NOM bitten-it-is-I-2 dog by
'The child is bitten by a dog.'
The passive is composed of the past passive participle plus an inflected
form of the verb qopna 'be'. In ( I -2b) the auxiliary aris 'is' has cliticized
to the participle as -a (cf. 'Aris-Cliticization' in the Introduction, §4. I ) .
Although i n each instance the full form, aris, could b e used, the enclitic
is more natural. The past passive participle is invariant for person,
number, tense, Series, etc. Like the verb qopna 'be', all passives are
Class 2 verbs and govern Pattern B in all Series ; 2 the passive subject
is therefore always in the nominative case.
It is necessary to show that the construction illustrated in (I-2b) is
really a passive because little attention has been devoted to this con
struction 3 and because a different construction has been called a
'passive'. The latter is discussed in ch. 1 3 (see also appendix to ch. 1 3
and Harris, t o appear b). The indirect object i n the passive poses a
problem ; in ( I b), the nominal which corresponds to the indirect object
104 Passivization
of ( I a) is not treated as an indirect object. It is shown below that this
nominal is a retired indirect object, and theoretical issues relating to it
are discussed.
I The passive from the viewpoint of language universals
In Comrie ( 1 977), Keenan ( 1 975), Perlmutter (1978), and Perlmutter &

Postal ( 1 974, 1 977), it is shown that a clause in which an initial direct
object becomes subject and in which an initial subject is a final non-term
is a passive construction.4 For Georgian, I will argue that the nomi
native-nominals of (I-2b) are initial direct objects, while the mier
nominal of (2b) is the initial subject, and that the nominative-nominals
are final subjects and the mier-nominal is a final non-term. Since the
nominative-nominal is an initial direct object and final subject, and the
mier-nominal is an initial subject and final non-term, this construction
is a passive with the characteristics stated above.
I . I Arguments for initial relations

I . I . I Suppletion for number of direct object. The verb gadagdebajgadaqra
'throw' is suppletive for the number of its direct object, such that the
former is used with singulars and the latter only with plurals (cf. Intro
duction, §4.3.2). This same constraint governs the nominative-nominals
of passives. That this is so is shown in (3-4b) ; (3-4a) give the cor
responding direct constructions.
(3) (a) *bavsvma gadaqara kva panjridan.
child-ERG he-threw-it-II-I stone-NOM window-from
('The child threw a stone from the window.')
(b) *es kva gadaqrilia panjridan.
this stone-NOM thrown-it-is-I-z
(,This stone was thrown from the window.')
(4) (a) bavsvma gadaqara kvebi panjridan.
stoneS-NOM
'The child threw stones from the window.'
(b) es kvebi gadaqrilia panjridan.
'These stones were thrown from the window.'
(*3) is ungrammatical because the initial direct object of the form
gadaqra 'throw' must be plural.
The grammatical status of (*3) and (4) can be accounted for simply,
7.I Language universals 105
if the nominative-nominal of the (b) sentences is treated as the initial

direct object.
I . I .Z Suppletion for animacy of direct object. The verb mitanaffaqvana

'take' is suppletive for the animacy of the direct object, such that the
former is used only with inanimate direct objects, and the latter only
with animates. The constraint that governs the direct objects of the
direct constructions also governs the nominative-nominal of the
passives. (S-6b) below illustrate passives of both verb forms, with the
nominative-nominal gela, a man's name ; (s-6a) give the corresponding
direct forms.
(s) (a) *gelas miitans mcxetali.
Ge1a-DAT he-takes-him-I-I Mtsxeta-in
('He will take Gela to Mtsxeta.')
(b) *gela mitanilia mcxetali.
Gela-NoM taken-he-is-I-2
('Ge1a is taken to Mtsxeta.')
(6) (a) gelas faiqvans mcxetasi.
he-takes-him-I-I
'He will take Gela to Mtsxeta.'
(b) gela faqvanilia mcxetali.
taken-he-is-I-z
'Gela is taken to Mtsxeta.'
(*S) is ungrammatical because an animate nominal occurs with the form
that takes only inanimate direct objects. This is good evidence that in
the (b) sentences gela is the initial direct object.
I . I .3 Tav-Rejlexivization. In ch. I it was shown that tav- can be

coreferential only to the subject of its clause. In ch. 5 it is further
established that this constraint on reflexivization is stated on initial
termhood ; that is, initial subjects may trigger Tav-Reflexivization even
though they are not final subjects.
In passives, tav- may not be coreferential to the nominative-nominal,
as (7) shows.
(7) (a) gelam daxata vano tavistvis.
Gela-ERG he-painted-him-II-I VanO-NOM self-for
'Gelai painted Vanoj for himself1.'
106 Passivization
(b) *vano daxatulia tavistvis.

painted-he-is-I-2
('Vano is painted for himself.')
In (7a), tav- cannot be triggered by vano because the latter is the

direct object.
The ungrammaticality of (*7b) is consistent with the facts of Tav
Reflexivization previously established if (i) the nominative-nominal is
not the initial subject, and (ii) tav- triggers are limited to initial subjects,
as stated in (8).
(8) Only initial subjects trigger Tav-Reflexivization.
Rule (8) makes the correct predictions for (7) and for examples con
sidered in previous chapters, but it fails to account for the ungram
maticality of (*9), where the mier-nominal ( = initial subject) has trig
gered Tav- Reflexivization.
(9) *vano daxatulia cnobili mxatvris mier tavistvis.

famous painter by
('Vano is painted by a famous painteri for himselfi.')
In ch. 14 I will reconsider the formulation of Tav-Reflexivization and

show that (8) is essentially correct, but that an additional clause must
be appended to account for (*9). Nevertheless the ungrammaticality of
(*7b) is evidence that the nominative-nominal of a passive is not its
initial subject.
1.2 Arguments for final relations

1 .2 . 1 Person Agreement. The nominative-nominals ( = initial direct
objects) in p-2b) trigger Subject Person Agreement. This will not be
immediately apparent, since it is the irregular auxiliary qopna 'be' that
registers agreement. The agreement will be more apparent from a
comparison of the following two paradigms. (10) illustrates the verb 'be'
as an independent verb, varying here for the person of its subject.
( I I) shows the verb 'be' as an auxiliary in the passive construction,
varying for the person of the nominative-nominal.
(10) (a) pexsisveli var.

barefoot l-am-I-2
'I am barefoot.'
7.I Language universals 107
(b) pexsisveli xar.

youtSG)-are-I-z
'You(SG) are barefoot.'
(c) pexsilvelia.
he-is-I-z
'He is barefoot.'
( I I ) (a) dar�munebuli var.
convinced I-am-I-z
'I am convinced.'
(b) dar�munebuli xar.
you(SG)-are-I-z
'You(SG) are convinced.'
(c) dar�munebulia.
he-is-I-z
'He is convinced.'
( I O-I I C) are derived from pexSisveli aris and dar�munebuli aris by Aris
Cliticization, which is described in the Introduction, §4. I .
I have shown i n previous chapters that Person Agreement is stated
on final terms. The fact that the nominative-nominal t = initial direct
object) triggers Subject Agreement shows that this nominal is the final
subject.
The mier-nominal fails to trigger any rule of Person Agreement, as
shown by the invariance of the verb in ( I Z) . If the mier-nominal trig
gered Person Agreement, the verb would vary with the person of the
nominal.
( 1 2) (a) saxli gaqidulia lems mier. 5
house-NOM sold-it-is-I-z me by
'The house is sold by me.'
(b) saxli gaqidulia sens mier.
you by
'The house is sold by you.'
(c) saxli gaqidulia mis mier.
him by
'The house is sold by him.'
In ch. I, §3, it was shown that terms trigger Person Agreement ; in
chs. 3-6, evidence was given that Person Agreement is stated on final
terms. The fact that the mier-nominal fails to trigger any Person
Agreement therefore supports the view that it is a final non-term.
108 Passivization ·
1 . 2.2 Unemphatic Pronoun Drop. I showed above that all and only
unemphatic pronominal final terms may be dropped. ( 13) is parallel to
(2b) and differs from it only in that the nominative-nominal has been
pronominalized. (1 3a) has an emphatic pronoun as final subject ; ( 1 3b)
has an unemphatic pronoun as final subject.
(13) (a) is dalf.benilia Jaylis mier.
he-NoM bitten-he-is-I-2 dog by
'He is bitten by a dog.'
(b) dalf.benilia Jaylis mier.
'He is bitten by a dog.'
This confirms that the nominative-nominal is a final term.
The mier-nominal cannot be dropped in the same way. In ( 1 2) we
see the pronoun and postposition remaining. (*14) is parallel, but the
pronoun has been dropped.
(14) *saxli gaqidulia mier.
house-NoM sold-it-is-I-2 by
('The house was sold by him.')
If the entire postpositional phrase were omitted, the resulting sentence
would be grammatical, but the meaning would not be that of ( I2) ;
saxli gaqidulia means 'the house is sold', not 'the house is sold by him' .
Since these pronominal initial terms cannot undergo Unemphatic
Pronoun Drop, and since terms can undergo this rule, these must not
be final terms. This is consistent with the conclusion drawn in earlier
chapters that Unemphatic Pronoun Drop is stated on final terms.
1 .2.3 Case Marking. As noted above, the verb qopna 'be', which is the
auxiliary used with passives, is a Class 2 verb according to the mor
phological criteria that define this Class (cf. ch. 1 6, Appendix A). In
all tense Series, Class 2 verbs govern Case Pattern B, which requires
that the subject be in the nominative case. This is evidence that the
nominative-nominal of passives is the subject.
The mier-nominal is not in a case associated with terms, · but is
marked with a postposition. Since Case Marking is stated on final term
hood (cf. chs. 4-6), and since the mier-nominal is the initial subject, it
must be finally a retired subject.
1 .2.4 Object Camouflage. Object Camouflage acts on direct objects in

the presence of an indirect object ; it applies to the output of rules that
7.r Language universals 109
change grammatical relations (cf. chs. 5 and 6). If I am correct, then,

in claiming that the initial direct object is a derived subject, then, the
target conditions for Object Camouflage are not met in a passive, and
the rule should not apply. In ( i sa) Object Camouflage has correctly
applied in a direct construction. But in the corresponding passive, ( I 6b),
Object Camouflage must not apply, as shown by the ungrammaticality
of (*16a).
( I S) (a) masr;avlebels gaugzavnes cemi tavi.
teacher-DAT they-sent-him-him-II-1 my self-NoM
'They sent me to the teacher.'
(b) *masr;avlebels gam(i)gzavnes (me).
they-sent-him-me-II-I I-NOM
('They sent me to the teacher.')
( 1 6) (a) *cemi tavi gagzavnilia masr;avleblistvisjmasr;avlebels.
my self-NoM sent-he-is-I-2 teacher-for jteacher-DAT
('I am sent to the teacher.')
(b) (me) gagzavnili var masr;avleblistvis.
I-NOM sent l-am-I-2
'I am sent to the teacher.'
The proposed analysis correctly predicts the ungrammaticality of
(* I 6a). If 'I' is the derived subject of the passive sentence, there is no
final direct object ; since Object Camouflage is stated on derived term�
hood, the rule cannot apply in this sentence.
1 . 3 Summary
The nominative-nominal has been shown to be both initial direct
object and final subject, while the mier-nominal is initial subject and
final non-term. On these grounds I conclude that the construction con
sidered here is a passive in the general linguistic use of that notion. The
arguments presented in this section imply no particular analysis beyond
the initial and derived grammatical relations of each nominal involved.
The data given above also confirm conclusions drawn in earlier
chapters with respect to rule interaction and rule statement : Tav
Reflexivization must be stated on initial terms, and the rules of Person
Agreement, Case Marking, and Unemphatic Pronoun Drop on final
terms. §1 .2.4 confirms that Object Camouflage applies to the output
of rules that change grammatical relations.
1 10 Passivization
2 Indirect objects in passives
In ( I a) masfavlebels 'teacher' is an indirect object, initially and finally ;

in ( I b) its grammatical relations are less clear. I n this section I will
discuss the syntactic characteristics of this nominal.
2. I Initial indirect object

In ch. I , §4, I showed that mifera 'write', but not dafera 'write', takes
an initial indirect object. The same constraint that governs the occur
rence of an indirect object in direct forms of these two verbs also
governs the occurrence of the tvis-nominal in passives. The post
position -tvis means 'for' and can occur in that meaning with dafera,
in its direct or passive constructions. While mifera may also have a
t�is-nominal in its clause in the direct construction, if a tvis-nominal
occurs with the passive of this verb, it is interpreted as a 'to' indirect
object. (*ISb) below is grammatical, but not in the meaning glossed
for it.
( 17) (a) ferili misfera 3mas.
'He wrote a letter to his brother.'
(b) *ferili da(s)fera 3mas
he-wrote-(him)-it-II -I
('He wrote a letter to his brother.')
( I S) (a) ferili miferilia 3mistvis.
written-it-is-I -2 brother-for
'The letter is written to his/my brother.'
(b) *ferili daferilia 3mistvis.
written-it-is-I-2
'(The letter is written to his/my brother.')
The last example is grammatical only in the meaning 'The letter is

written for his/my brother'. This is predictable from the restrictions
independently placed on the inventories of initial terms of these two
verbs if the tvis-nominal is the initial indirect object of the passive
sentences.
2.2 Final non-term

The tvis-nominal of the passive construction does not have the charac
teristics of final terms. It can be seen in (I b) that this nominal does not
7.2 Indirect objects 111
trigger Person Agreement. Nor does it bear the case marking of a final
subject, direct object, or indirect object, but rather that of a final non
term (cf. ch. I I ) . This nominal cannot undergo Unemphatic Pronoun
Drop. If the pronoun is dropped, leaving the postposition, the result is
ungrammatical, as in (*I 9a).
( I 9) (a) *vaSli micemulia tvis.

apple-NoM given-it-is-I-2 for
(,The apple is given to him.')
(b) vaSli micemulia.
'The apple is given.'
( I 9b), while grammatical, can only be interpreted as having an unspeci

fied indirect object, not a pronominalized indirect object. Thus, the
tvis-nominal lacks the ability to undergo Unemphatic Pronoun Drop
and the other characteristics of final terms. It must therefore be a final
non-term.
2.3 Variations in the marking of the initial indirect object

The status of mas�avleblistvis 'to the teacher' in ( I b) is confusing for
several reasons. First, there are at least two dialects in Standard Modern
Georgian with respect to such sentences. For some speakers, ( I b) is
grammatical. But for some, Passivization is blocked when there is an
initial indirect object ; these speakers find (I b) unacceptable. Second,
with the passive of certain verbs, the initial indirect object is in the
dative, not marked with tvis.6 We therefore find the following contrasts :
Dialect A
(20) (a) viracam acvena vanos surati.
someone-ERG he-showed-him-it-II-I Vano-DAT picture-NoM
'Someone showed the picture to Vano.'
(b) *surati iqo nacvenebi vanos.
it-was-II-2 shown
('The picture was shown to Vano.')
(c) surati iqo nacvenebi vanostvis.
Vano-for
'The picture was shown to Vano.'
(d) surati iqo nacvenebi.
'The picture was shown.'
1 12 Passivization
Dialect B
(20') (a)
(b)*
(c)*
(d)
Dialects A and B
(zr) (a) es 1!.acebi mivacviet musaobas.
these men-NOM I-accustomed-it-them-II-r work-DAT
'I accustomed these men to work.'
(b) es 1!.acebi arian cveuli mufaobas.
they-are-I-z accustomed
'These men are accustomed to work.'
(c) *es 1!.acebi arian cveuli musaobistvis.
work-for
('These men are accustomed to work.')
In Dialect A the verb 'show' is grammatical in the direct construction;
(zoa) , and in the passive, with or without an initial indirect object,
which is marked with tvzS «(zoe), (zod), respectively). In Dialect B, the
direct construction is grammatical, but the passive is possible only if
it has no initial indirect object (contrast (zo ' d) with (*zo ' b, c)). In both
dialects, the verb 'accustom', unlike 'show', is grammatical with a
dative indirect object (example (z 1 )) .
3 Theoretical issues
With respect to the behavior of the initial subject and direct object,
the Georgian passive is in no way remarkable, as shown in § r . But the ,
behavior of the initial indirect object is puzzling. At issue is the moti
vation of the final non-term status of at least some initial indirect objects.
Some linguists claim that 'spontaneous chomage' is possible in language
(Comrie 1 977 ; Keenan I975). If such rules must indeed be counten
anced, we can simply add a rule to the grammar of Georgian describing
what happens in passives. But Perlmutter and Postal, attempting to
characterize possible variations among languages, have proposed the
Motivated Chomage Law, a constraint on the class of possible gram
mars (Perlmutter 1978 ; Perlmutter & Postal I977, to appear c). This
law, formally stated in Perlmutter & Postal (to appear c), claims in
effect that nominals bear the chomeur relation only under the con-
7.3 Theoretical issues 1 13
ditions specified in the Chomeur Law (cf. Introduction, §2). · This

means that the conditions for the existence of chomeurs are completely
determined by universal principles. In the instance of tvis-nominals
of the Georgian passive, it is not obvious how the indirect object might
be put en chomage. If Perlmutter and Postal's claim is correct, either
the tvis-nominals are not chomeurs or the conditions of their chomage
have yet to be discovered.
Relational grammar defines two classes of initial terms that are final
non-terms - chomeurs and emeritus terms (Perlmutter & Postal, to
appear c). Together these are called 'retired terms'. Chomeurs are
governed by the Chomeur Law and the Motivated Chomage Law, and
their behavior has been studied in some detail. 7 Much less is under
stood about nominals bearing the emeritus relations. They are known
to exist in causatives (cf. ch. 5 and works cited there). I n ch. I I it is
suggested that emeritus terms also exist in nominalizations and infini
tive constructions. If Georgian indicated clearly the difference between
chomeurs and emeritus terms, the tvis-nominal of the passive construc
tion would be an empirical test of the validity of the Motivated Chomage
Law. However, as shown in ch. I I , Georgian marks chomeurs and
emeritus terms in the same way, and they have the same behavior with
respect to the superficial syntactic properties investigated here. It is
therefore impossible to determine at present whether the tvis-nominal
is an indirect object chomeur or an indirect object emeritus. Therefore,
it is impossible to give a complete analysis of the construction. From
§I it is clear that the initial direct object is a final subject and that the
initial subject is a final retired subject. I therefore propose that (2b) is
correctly described by the relational network (22) and that Passivization
in Georgian is at least partially stated as the universal rule (23).
(22)
da�lbel1a 3ayli bavsvi

' bite ' ' dog ' ' child '
(23) Passivization
In the presence of a subject, a direct object may advance to subject.
Neither (22) nor (23) describes the fate of the initial indirect object,
1 14 Passivization
but there is not sufficient evidence for doing so at present, for the
reasons stated above.
4 Interaction with other rules
In this section I will consider the rules we have previously dealt with
and show that this analysis makes the correct predictions in each case.
4. I Object Raising
Object Raising and Passivization can never apply in a single clause.
(24a) is grammatical without Passivization ; but (*24b), where Passivi
zation has applied in the embedded clause, is ungrammatical.
(24) (a) es motxroba advilia fasa1!.itxavad.
this story-NOM easy-it-is-I-2 to-read
'This story is easy to read.'
(b) *es motxroba advilia fa1!.itxuli (sa)qopnad.
read to-be
('This story is easy to be read.')
This is predicted by the proposed analysis. Since the input conditions
to each rule require a direct object, and since the output of each con
tains no direct object, they are mutually exclusive in a clause.
4.2 Causative Clause Union

This analysis correctly predicts that Passivization can apply to the out
put of Causative Clause Union, since the output of that rule contains
a direct object. (2S) exemplifies that situation (cf. example ( I Sb) of
ch. S).
(2S) ubeduri semtxveva moxdenilia patara bi?is mier.
unfortunate accident-NoM caused-it-is-I-2 little boy by
'An unfortunate accident was caused by the little boy.'
Passivization does not, however, apply freely to a causative with three
terms.
While Passivization can apply to the output of Causative Clause
Union, the latter cannot apply to the output of the former. This is not
automatically predicted from our analysis, but the same is true in other
languages, such as French, Turkish, etc. (cf. Aissen 1 974a, b). The
inability of rules to apply in the complement of causatives is discussed
above in ch. 5.
7-4 Interaction with other rules 1 15
4.3 Version
The Version rules cannot apply to the output of Passivization. This may
be due to the fact that the finite verb of passives is qopna 'be', which
cannot take a version object, as shown in (z6) and (z7).
(z6) (a) es vaSli aris sentvis.
this apple-NoM it-it-I-z you-for
'This apple is for you.'
(b) *es va§li garis.
it-is-you-I-z
(,This apple is for you.')
(z7) (a) es aris seni otaxi.
this it-is-I-z your room-NOM
'This is your room.'
(b) *es garis otaxi.
it-is-you-I-z
('This is your room.')
Since qopna never takes Version, we may conjecture that this also
blocks the application of Version to a passive sentence.
Passivization cannot apply to the output of Version. In (z8a), neither
rule has applied. In (z8b), Possessive Version has applied, making
zurab the indirect object. In (*z8c, d), both Possessive Version and
Passivization have applied. Since Version makes zurab the indirect
object, and Passivization causes it to retire (cf. §z and 3) , we would
expect it to be marked with tvis or the dative case, as in (*z8c) and
(*z8d), respectively.
(z8) (a) zurabis das (oraven.
Zurab-GEN sister-DAT they-gossip-her-I-I
'They are gossiping about Zurab's sister.'
(b) das u(oraven zurabs.
they-gossip-him-her-I-I Zurab-DAT
'They are gossiping about Zurab's sister.'
(c) *da ga?orilia zurabistvis
sister-NOM gossiped-she-is-I-z Zurab-for
('Zurab's sister is being gossiped about.')
(d) *da ga(orilia zurabs.
(,Zurab's sister is being gossiped about.')
(e) zurabis da ga(orilia.
'Zurab's sister is being gossiped about.'
116 Passivization
(28e) establishes that Passivization can apply with this verb, as long as
Version does not apply. Thus, (28) documents the inability of Possessive
Version and Passivization to interact.
We cannot determine whether Passivization can apply to the output
of Benefactive Version. Since the initial indirect object in a passive
construction and the non-term benefactive are both marked with tvis,
it is impossible to tell whether or not Benefactive Version has applied
in the derivation of
(29) axali sarvali se�erilia sentvis.
new trouserS-NOM sewn-it-is-I-2 you-for
'New trousers are made for you.'
The inability of Version to apply in passives does not follow auto
matically from any rule previously proposed. 8
5 Conclusion
The construction illustrated in (I -2b) has been shown to be a passive,

according to accepted definitions of passive. Peculiarities of the initial
indirect object in a passive construction have been discussed ; it has
been shown that that nominal is a final non-term, but no proposal has
been made concerning its precise derivation. Finally, the interaction of
Passivization and other rules of the grammar has been discussed.
8 Inversion
In ch. 2 I described the different correlations between grammatical

relations and case markings as they exist in Series I and II. There is
also a Series III, in which a third set of correspondences apparently
holds. This chapter will describe and account for that case marking
differential.
I The problem : the case marking differential in Series ill
The sentences in ( I ) are in the present tense, which belongs to Series I ;

( I a) contains a Class I verb, ( Ib) a Class 2 verb, and ( I C) a Class 3
verb. In each, the subject is in the nominative, the direct and indirect
objects are in the dative.
( I ) (a) rezo samajurs acukebs dedas.
Rezo-NOM bracelet-DAT he-gives-her-it-I-I mother-DAT
'Rezo is giving mother a bracelet.'
tb) gelodebi (me) (Sen).
I-await-you-I-2 I-NOM YOU-DAT
'I am waiting for you.'
( c) deida myeris naninas.
aunt-NOM she-sings-I-3 lullabY-DAT
'Aunt is singing a lullaby.'
(2) gives the same sentences in the so-called perfect or first evidential,
which belongs to Series III. Notice that, for verbs of Classes I and 3,
the cases differ from those of Series I, though the meaning remains
essentially the same. For verbs of Class 2, the subject remains in the
nominative, and the object in the dative.
(2) (a) turme rezos samajuri ucukebia dedistvis.
apparently Rezo-DAT bracelet-NoM he-gave-it-III-I mother-for
'Apparently Rezo gave a bracelet to his mother.'
1 18 Inversion
(b) turme daglodebivar (me) (Sen).

I-awaited-you-III-2 I-NOM YOU-DAT
'Apparently I (have) waited for you.'
(c) turme deidas umyeria nanina.
aunt-DAT she-sang-it-III-3 lullabY-NoM
'Apparently Aunt has sung a lullaby.'
Traditionally, Georgianists have called the construction illustrated by
(2a, c) 'inversion', and this name has been adopted in relational gram
mar. Inversion is illustrated here in the evidential mode, a verb form
used to remove the speaker, in some sense, from the statement.l The
evidential occurs felicitously with the word turme 'apparently' , and
examples from Series III may be translated this way in English even
when turme does not actually occur in the Georgian example.
The correspondences illustrated in (2), as well as those discussed in
earlier chapters for Series I and II, are summarized in (3-4), which
may be compared with ( 1 3-14) of ch. 2.
(3) SUbject Direct Object Indirect Object
C DATIVE NOMINATIVE tvis
(4) Series I II III
Class 1 B A C
2 B B B
3 B A C
In this chapter it is argued that (3-4) represent an analysis that is
unnecessarily complex. In particular, it is shown that Pattern C is
reducible to a special instance of Pattern B by the application of two
rules, Inversion and Unaccusative, which change grammatical relations,
as described below in §2.
Traditional treatments have observed that Object Camouflage does
not apply in Series III tenses ; this has been treated as an idiosyncratic
fact (cf. Shanidze 1973 ; Tschenkeli 1958 ; Vogt 1 971). The failure of
Object Camouflage to apply follows from the analysis presented here,
with no additional device.
2 The proposal : a rule of inversion

I propose that the initial grammatical relations of (2a) are as listed in
(5), while the final grammatical relations are as listed in (6).
8.2 The proposal 119
(5) Initial grammatical relations

rezo subject
samajuri 'bracelet' direct object
deda 'mother' indirect object
(6) Final grammatical relations
rezo indirect object
samajuri subject
deda indirect object chomeur. 2
I propose, further, that the grammatical relations in (5) are related to
those in (6) in the following way. The rule of Inversion applies, making
the subject an indirect object. According to the Chomeur Law, this
makes the initial indirect object a chomeur. The initial direct object
becomes subject by the rule Postal and Perlmutter have called Unac
cusative. In Georgian this rule applies to a direct object at any level
where it does not co-occur with a subject, making the direct object a
subject. This may all be schematized as (7), where the top stratum
represents initial grammatical relations, the second represents gram
matical relations resulting from the application of Inversion, and the
third represents grammatical relations resulting from the application
of Unaccusative.
cukeba l'ezo samaJun deda

' g ive ' , bracelet ' , mother '
Inversion is triggered by a Class I or 3 verb in a Series III form ; the

rule must apply if this trigger condition is met.
Traditionally, some Georgianists have considered the dative-nominal
to be the subject in Series III on the basis of the intuitive notion of
'subject' ; others have considered the nominative-nominal to be the
subject in this construction on the basis of the case and person agree
ment facts. But the question has been argued on the basis of these facts
alone. 3 Only by recognizing distinct levels of derivation and by con
sidering a number of criteria, including syntactic as well as morpho
logical ones, can we give an adequate analysis of the 'subject' in this
construction.
120 Inversion
Most of this chapter is devoted to arguments that in inversi on con:

structions, such as (za, c), the initial and final grammatical relations
are as stated above. The arguments will establish that in inversion the
dative-nominal, which corresponds to the syntactic subject in the
indicative (Series I and II), has the syntactic properties of initial
subjects and the syntactic properties of final indirect objects. Similarly,
the nominative-nominal of inversion will be shown to have the syntactic
properties of initial direct objects and those of final subjects. For
expository reasons, the arguments will not be grouped in this way,
-however. Rather, the first group of arguments ( §3) presents evidence
for both initial and final relations and is based on the simplest type of
inversion only, as illustrated in (z). I will then introduce an inversion
construction that is not, like (z), limited to the evidential mode ( §4) ;
two properties of this type of construction are used as the basis for
additional arguments. In §5, some further arguments, based on both
types of inversion are given. A total of five arguments to support the
proposed initial grammatical relations are presented - in §3.3, 3.4, 4. 1 ,
and 5 . 1 . Five arguments to support the final grammatical relations
proposed are given in §3 . 1 , 3 .z, 4.Z, and 5 . 1 . Additional evidence to
support this rule is given in chs. I I , I S and 1 6.
The arguments in §3-5 concern the initial and derived grammatical
relations only ; a discussion of the specific rules proposed is in §6. The
interaction of Inversion with some other rules of syntax is described
in §7.
3 Preliminary arguments
The form of the argument will be the same for both initial and final
grammatical relations : r will show how one of the phenomena whose
general characteristics in Georgian syntax have already been established
operates in sentences like ( za, c). If the grammatical relations are as I
have proposed, the phenomenon can be treated in a general manner
throughout the grammar. If not, special ad hoc rules will be needed to
account for each of these phenomena in inversion constructions. Final
grammatical relations are discussed first.
Person Agreement
-3 . 1
Paradigm (8) represents the intersection of the categories of person
and number for Subject Person Agreement, first introduced in ch. I.
8.3 Preliminary arguments 121
(8) Subject Person Ag1'eement

singular plural
I . person v--- v-t
2. person o -t
3. person -sJaJo -enJesJnen, etc.
The dash represents the position of the verb stem, which includes the
root and certain formants. The alternations in the third person are
determined by the tense and Class of the verb.
The markers in (8) indicate Subject Person Agreement generally in
Georgian, as shown in ch. I , §3 . I . In subsequent chapters it was shown
that Person Agreement is stated on final termhood (cf. chs. 4, 5, 6,
and 7).
In Series III, it is the nominative-nominal which triggers Subject
Person Agreement. (9 ) gives a paradigm of the singular with the
markers of (8) in heavy type.
(9) (a) mas undoda, (rom) (me) gavegzavne kalakSi.

that I-NOM he-sent-me-III-I city-in
'He wanted to send me into the city.'
(b) mas undoda, (rom) (Sen) gaegzavne kalakSi. (zero
YOU-NOM he-sent-you-III-I marker)
'He wanted to send you into the city.'
(c) mas undoda, (rom) (is) gaegzavna kalaksi.
he-NoM he-sent-him-III- I
'Hei wanted to send himj into the city.'
(9 ) establishes that the nominative-nominal triggers Subject Person

Agreement in Series III. 4 From this and the fact established in earlier
chapters that final subjects trigger S ubject Agreement, we must con
clude that the nominative-nominal in the inversion construction in
Series III is a final subject.
Now let us turn to Indirect Object Person Agreement ; the markers are
given in paradigm ( 1 0) .
( 1 0 ) Indirect Object Person Agreement

singular plural
I . person m- gv
2. person g g-t
3 . person sJhJ0- SJhJ0-t
122 Inversion
Here the alternation m the third person markers is phonologically

determined.
The markers in (10) indicate Indirect Object Agreement generally in
Georgian, as shown in ch. I, §3 . I . In subsequent chapters it was shown
that Indirect Object Person Agreement is stated on final termhood (cf.
chs. 5 and 6).
In Series III, the dative-nominal triggers Indirect Object Person
Agreement. ( I I) presents a paradigm of the singular, with the markers
of ( 1 0) in heavy type.
( I I ) (a) mas undoda, (rom) (me) dametesa simindi.

I-OAT I-sowed-it-III-I corn-NOM
'He wanted me to sow corn.'
(b) mas undoda, (rom) (Sen) dagetesa simindi.
YOU-OAT you-sowed-it-III-I
'He wanted you to sow corn.'
(c) mas undoda, (rom) gelas daetesa simindi.
Gela-oAT he-sowed-it-III-I (zero marker)
'He wanted Gela to sow corn.'
Since final indirect objects trigger Indirect Object Person Agreement

in Georgian, and the dative-nominal of the evidential (Series III)
triggers Indirect Object Person Agreement, as established in ( I I), 5 the
dative-nominal must be a final indirect object.
According to this analysis, as a consequence of the Chomeur Law, the
tvis-nominal is an indirect object chomeur. Since Person Agreement is
stated on final termhood, the tvis-nominal, not being a final term, should
not trigger Person Agreement. I will show now that this prediction is
borne out. In the indicative, illustrated with Series II in ( 1 2), the
indirect object triggers Indirect Object Person Agreement, which is
realized as m- in (a) for the first person, and as g- in (b) for the second
person. By comparing the sentences of ( 1 2) with their counterparts in
the evidential, given in ( 1 3), we can see that the nominal which in the
indicative triggers Indirect Object Agreement, in the evidential triggers
no agreement at all.
( 1 2) (a) rezom macuka samaJurt.

ReZO-ERG he-gave-me-it-II-I bracelet-NoM
'Rezo gave me a bracelet.'
(b) rezom gacuka samajuri.

he-gave-you-it-II-I
'Rezo gave you a bracelet.'
(c) rezom acuka samaJurz.
he-gave-him-it-II-I (zero marker)
'Rezo gave him a bracelet.'
( 1 3) (a) turme rezos ucukebia samajuri cemtvis.

apparently ReZO-DAT he-gave-it-III-I me-for
'Apparently Rezo has given me a bracelet.'
(b) turme rezos ucukebia samajuri sentvis.
you-for
'Apparently Rezo has given you a bracelet.'
(c) turme rezos ucukebia samajuri mistvis.
him-for
'Apparently Rezo has given him a bracelet.'
( 1 3) shows that the tvis-nominal of the evidential (Series III) fails to

trigger any Person Agreement, thus supporting the view that it is not a
final term.

In ch. 3 I showed that Object Camouflage applies obligatorily to first
and second person direct objects in the presence of an indirect object,
as illustrated in ( 14).
(14) (a) *deda (sen) gabarebs mast;avlebels.

mother-NOM YOU-DAT she-renders-him-you-I-I teacher-DAT
(,Mother is turning you over to the teacher.')
(b) deda abarebs mast;avlebels sens tavs.
she-renders-him-him-I-I your self-DAT
'Mother is turning you over to the teacher.'
(* 14a) is ungrammatical because the input conditions for Object

Camouflage are met, but the rule has failed to apply ; in (14b) the
initial direct object has been camouflaged. Both sentences of ( 14) are
in Series I.
In Series III Object Camouflage never applies. ( 1 5) is exactly parallel
to (14), but is in Series III. In ( I sa) Object Camouflage has not applied ;
in (*I Sb) it has.
124 Inversion
( 1 5) (a) turme dedas caubarebixar
apparently mother-DAT she-rendered-you-III-I
mas(:avleblistvis.
teacher-for
'Apparently Mother (has) turned you over to the teacher.'
(b) *turme dedas caubarebia seni tavi
she-rendered-him-III-I your self-NoM
mas(:avleblistvis.
(,Apparently Mother (has) turned you over to the teacher.')
(14) and ( 1 5) show that Object Camouflage must apply in Series I and
must not apply in Series III.
I have shown in previous chapters that Object Camouflage applies
to the output of rules that change grammatical relations (chs. 5, 6, 7).
Inversion and Unaccusative are relation-changing rules. From this it
follows that Object Camouflage applies to the output of these two rules.
But Object Camouflage is a rule that affects direct objects ; there is no
direct object in the output of the obligatory rule of Unaccusative. The
input conditions for Object Camouflage are therefore not met wherever
Inversion and Unaccusative have applied (cf. network (16)).
( 1 6)
cabareba vano sen nzascavlebeli

, render ' ' you ' , t�acher '
Thus, on my analysis, the fact that Object Camouflage never applies

with Series III is accounted for automatically. 6 An analysis on which
the nominals have the relations listed in (5) at all levels of derivation
must treat Series III as an exception to Object Camouflage.
3.3 Tav-Rejlexivization
Tav-Reflexivization is triggered only by subjects (cf. ch. I), and the
subject-trigger condition on this rule is stated on initial termhood
(cf. ch. 5, §2.2, and ch. 7, § I . I .3).
In inversion constructions in Series III, only the dative-nominal can
trigger Tav-Reflexivization. In (17a) the nominative subject of a verb
in Series I triggers tav- ; in (17b) the same nominal is the trigger,

though its case is different in the inversion construction.
( 1 7) (a) gela ir�munebs tavis tavs.
Gela-NoM he-convinces-"him-I- I self's self-DAT
' Gela is convincing himself.'
(b) gelas turme daur�munebia tavlSl tavi.
Gela-DAT apparently he-convinced-him-III-I self's · self-NoM
' Gela apparently (has) convinced himself.'
( c) *tavis tavs turme daur{:munebia gela.
self 's self-DAT Gela-NoM
(,Apparently Gela (has) convinced himself.')
(d) *gela turme daur{:munebia tavis tavs.
(*I7c) shows that the nominative-nominal of the inversion construction
cannot trigger Tav-Reflexivization in the dative-nominal. (*I7d) is
identical to (*I 7c) except for word order ; it establishes that the ungram
maticality of (*I 7c) is not due to word order.
The fact that it is the dative-nominal that can trigger tav- in (17) is
consistent with the fact that only initial subjects can trigger Tav.:..
Reflexivization (cf. ch. 7, S 1 . 1 .3), since on the proposed analysis the
dative-nominal is the initial subject of the inversion construction. An
analysis that claimed the relations of (6) for all levels of derivation would
render this construction an exception to Tav-Reflexivization.
3.4 Suppletion
It has been established that there are verbs in Georgian that are sup
pletive for the number or animacy of their subjects or objects. Each of
those verbs supplies arguments in support of the analysis proposed in
§2 ; two will be discussed here.
3+ I Suppletion for number of the direct object. The verb 'throw' is

suppletive for the number of its direct object ; gadagdeba is used with
singular direct objects, gadaqra with plurals (cf. Introduction, §4. 3.2).
This number condition is governed by the nominative-nominal in
inversion sentences in Series III. The sentences in ( 1 8) have singular
nominative-nominals ; those in ( 1 9) have plurals.
(18) (a) turme bavsvs gadaugdia kva.
child-DAT he-threw-it-III-I stone-NOM
'Apparently the child threw a stone.'
126 Inversion
(b) *turme bavsvs gadauqria kva.

he-threw-it-III-I
('Apparently the child threw a stone.')
(19) (a) *turme bavsvs gadaugdia kvebi.

stones-NOM
('Apparently the child threw stones.')
(b) turme bavsvs gadauqria kvebi.
'Apparently the child threw stones.'
On the analysis proposed above in §z, kva/kvebi 'stone/stones' is the

initial direct object in ( 1 8-19). These sentences are therefore accounted
for with the same condition needed for sentences of Series I and II,
stating that condition on initial termhood (cf. ch. 4, §3. I . 1 and ch. 7,
§ I . I . I ). An analysis on which kva/kvebi were not initial direct objects
would have to complicate the rule governing the suppletion of 'throw',
just when it occurred in a Series III form.
3.4.2 Suppletion for animacy of the direct object. The verb 'bring' is
suppletive for the animacy of its direct object ; moitana 'he brought it'
is used only with inanimate direct objects, moiqvana 'he brought him'
only with animate direct objects. It has been shown in earlier chapters
that this constraint is stated on initial termhood.
The same condition on animacy governs the nominative-nominal in
Series III ; on the proposed analysis, these are initial direct objects.
The direct objects in (20) are inanimate, those in (21) animate.
(20) (a) gelas moutania yvino supraze.

Gela-DAT he-brought-it-III-I wine-NoM table-on
' Gela has apparently brought wine to the party.'
(b) *gelas mouqvania yvino supraze.
he-brought-him-III-I
('Gela has apparently brought wine to the party.')
(21 ) (a) *gelas moutania megobari supraze.

friend-NOM
(,Gela has apparently brought a friend to the party.')
(b) gelas mouqvania megobari supraze. .
'Gela has apparently brought a friend to the party.'
(20) and (21) show that the nominative-nominal of the inversion con-
8-4 Inversion verbs 127
struction must meet the animacy requirements on the direct object of

the suppletive verb 'bring'.
On the proposed analysis, yvino 'wine' and megobari 'friend' are
initial direct objects in (20-2 1). They are therefore accounted for with
the same condition needed to account for suppletion in Series I and II.
An analysis that did not derive (20-2 1) from initial structures in which
yvino and megobari were direct objects would have to complicate the
condition governing suppletion in the verb 'bring', just when it occurred
in the inversion construction.
We have seen that the dative-nominal of the inversion construction
has the characteristics of a final indirect object with respect to the rules
of Indirect Object Agreement and the characteristics of an initial subj ect
with respect to the rules of Tav-Reflexivization. The nominative
nominal of the inversion construction has been shown to have the
characteristics of a final subject with respect to the rule of Subject
Agreement and the characteristics of an initial direct object with respect
to two verbs suppletive for properties of their direct objects. In addition,
we have seen that this nominal does not have the characteristics
of a final direct object with respect to the rule of Object Camouflage,
or the characteristics of an initial subject with respect to the rule of
Tav-Reflexivization. The tvis-nominal of the inversion construction,
while it corresponds to the indirect object of Series I and II, has been
shown to have the characteristics of a final non-term. Thus arguments
have been given for both the initial and final relations of each nominal,
as proposed in §2.
4 Inversion verbs
Georgian has a group of so-called 'affective' predicates, including verbs

like miqvarxar 'I love you', uxaria 'he is pleased, happy' , and segi31ia
'you can'. These verbs have many of the same syntactic properties that
the inversion construction of Series III has. Appendix A to this chapter
provides evidence that sentences like (22) are like inversion in Series
(22) gelas uqvars nino.
Gela-DAT he-Ioves-her-I-4 NinO-NoM
' Gela loves Nino.'
III (e.g. (2a, c)) with respect to all of the applicable rules discussed in
g3. Specifically, it shows that the dative-nominal has the syntactic
128 Inversion
properties of a final indirect object and of an initial subject, while the

nominative-nominal has the syntactic properties of a final subject and
of an initial non-subject.
The construction illustrated in (22) differs from the inversion con
struction of (2a, c) in one important respect : 'affective' verbs like that
in (22) govern this construction not only in Series III, but in all Series.
This means that the alternation illustrated by ( l a, c)!(2a, c) does not
exist for affective verbs. Because of this difference, the question arises,
are these constructions the same ?
I will argue below that sentences like (22) involve Inversion and
U naccusative, the same rules that are involved in the derivations of
(2a and c). I am claiming that the 'experiencer' is the initial subject
and final indirect object, and that the 'stimulus' is the initial direct
object and final subject, as stated in (23) and (24).
(23) Initial grammatical relations

.
gela subject
nino direct object
(24) Final grammatical relations

gela indirect object
nzno subject
When Inversion applies in the derivation of (22), it makes the initial

subject, gela, the indirect object. Unaccusative then makes the initial
direct object, nino, the subject. There are no verbs of this type - hence �
forth ' Class 4' or 'inversion verbs' - with an initial indirect object ; thus
there is never an indirect object chomeur, as there may be with inversion
in the evidential. The derivation of (22) may be represented as ( 2 5).
(25)
q'vareba gela nino
, love ' "
Preliminary arguments that this construction is the same as inversion

iii Series III are given in Appendix A to this chapter. In the remainder
of this section I will give two additional arguments for an Inversion
84 Inversion verbs 129
analysis. The first is based on suppletion and is relevant to inversion

verbs alone. This is an argument against a solution where each nominal
has its final termhood at all levels of derivation. The second argument
is based on case marking ; it is an argument for an Inversion analysis for
both inversion verbs and Series III forms and considers the verbal
system of the language as a whole. This set of facts argues against a
solution where each nominal has its initial termhood at all levels of
derivation. Since these two constructions behave the same with respect
to all syntactic rules that interact with them, I conclude that the deri
vations of both types involve the application of Inversion and Unac
cusative.7 The two types differ in that inversion verbs trigger Inversion
in all finite forms, while with other verbs, the evidential mode (Series
III) alone triggers Inversion.
4. I Suppletion for tense

In §4.3.4 of the Introduction I showed that the verb 'take' is multiply
suppletive in Georgian - for animacy of the direct object, and for tense.
In this section I will argue on the basis of the latter suppletion that
Inversion applies with Class 4 verbs. To simplify the situation, I will
cite only those forms used with inanimate direct objects.
Consider the pair of sentences in (26). The first is in the present tense
(Series I), formed on the root kv ; the second is in the aorist (Series II)
and formed on the root tan. These are in a suppletive relation to one
another ; the first root cannot occur in the aorist, the second cannot
occur in the present. What is of particular interest here is the difference
in case marking.
(26) (a) gelas fignebi ,miakvs sam�itxvelosi.

Gela-DAT bookS-NOM he-takes-it-I-4 reading-room-in
'Gela is taking the books into the reading room.'
(b) gelam fignebi miitana sam�itxvelosi.
Gela-ERG he-took-it-II-I
'Gela took the books into the reading room.'
The verb form in (26a), miakvs, is in Class 4 and governs Pattern C
case marking in all Series. The verb form in (26b), miitana, is in Class I,
which in Series I I governs the case Pattern A, as we saw in ch. 2. The
meaning of the two sentences is the same ; only the tense - and for this
verb, that involves also the root - is different.
If miakvs is not treated as an inversion verb and does not undergo
130 Inversion
the rule of Inversion, the rules that relate syntax to semantic interpret
ation must be complicated to account for the pair of sentences in (26).
On the inversion analysis proposed here, on the other hand, the initial
subject of the inversion verb form is the same as the (initial and final)
subject of the non-inversion verb form. Therefore the semantic in
terpretation of both sentences can naturally be related to their initial
term dependents.
4.2 Case Marking

In §I of this chapter, I gave a summary of the Class-Series-case cor
relations for Georgian. (27) repeats ( 3 ) ; (28) repeats (4), with the
addition of Class 4 (inversion) verbs.

(28) Series I II III

Class I , 3 B A C
2 B B B
4 C C C
(27) and (28) summarize in chart form a traditional description of case

marking in Georgian. 8 It is based on an analysis of the inversion con
struction where the dative-nominal is the subject ; thus (28) indicates
Pattern C across the board for inversion verbs. Similarly, it shows
Pattern C (dative subject) for Class I and 3 verbs, which govern
Inversion in Series III.
However, given the analysis of the inversion construction proposed
above, it is not necessary to have Pattern C at all ; it reduces to a special
instance of Pattern B. According to the proposed analysis, the dative
nominal in an inversion construction is not only the initial subject, but
also the final indirect object. If case marking in Series III is stated on
final termhood, then this nominal will be correctly marked by Pattern B,
the most general pattern in the language. Similarly, according to the
Inversion analysis, the nominative-nominal is not only the initial direct
object, but also the final subject. If case marking in Series III is stated
on final termhood, then this nominal will also be correctly marked by
Pattern B. There is independent evidence that case marking is stated
8-4 Inversion verbs 131
on final termhood (cf. chs. 3, 4, 5 , 6, 7 and 1 6). For (27) and (28),
then, we can substitute (27') and (28') :
(27') Final Subject Final Final
Direct Object Indirect Object
(28') Series I II III
Class I, 3 B A B
2, 4 B B B
This correctly represents case marking in relation to final termhood,
showing that Pattern B is general for Georgian, Class I and 3 verbs in
Series II being an exception to the general rule. The traditional state
ment, (27-28), is misleading in the sense that it represents case marking
in relation to initial termhood, while case marking rules are, in fact,
based on final termhood.9 The facts summarized in (27'-28') show that
if one accepts the final termhood proposed above for the inversion
construction, significant simplifications can be achieved in the rules of
case marking.
The marking of the tvis-nominal is also accounted for naturally on
the proposal outlined above. Since case marking in Series III is based
on final termhood, if the initial indirect object of a sentence like (2a)
is finally an indirect object chomeur as proposed, it should have mark
ing appropriate to a non-term. Tvis is used to mark non-term benefac
tives (cf. ch. 6) and delegatives, as well as certain retired terms (cf.
chs. 5 and 7). The systematic nature of the latter will be established
in ch. I I .
4.3 Summary
Inversion verbs, predicates that trigger Inversion in all finite verbal
categories, and inversion in Series III have been shown to involve the
same two rules - Inversion and Unaccusative. One argument, based
on suppletion, has been offered for the initial grammatical relations
claimed in (23), while a second argument provides support for the final
grammatical relations claimed in (6) for Series III and in (24) for
inversion verbs. This is based on the fact that the Inversion analysis
effects a significant simplification in the case marking rules : once
inversion is incorporated into the grammar of Georgian, the need for
Pattern C disappears.
132 Inversion
5 Additional arguments
5 . 1 Causative Clause Union

In ch. 5 I showed that the subject of an intransitive is realized as the
direct object of the causative of that verb. In the causative of an intransi
tive inversion verb, the initial subject is also realized as the direct object
of the causative. (29a) contains an intransitive inversion verb ; (29b)
contains the corresponding causative.
(29) (a) tusays sioda.
prisoner-nAT he-hungered-I- 4
'The prisoner was hungry.'
(b) (mat) tusayi moaSives. 1 0
theY-ERG prisoner-NOM they-caused-hunger-him-II- I
'They let the prisoner go hungry. jThey starved the prisoner.'
On the Inversion analysis, tusay- is the initial subject of -siv- in both
sentences of (29). Inversion, like other rules that change grammatical
relations, does not apply in the complement clause of a causative (cf.
ch. 5). Since Inversion does not apply to tusay- in (29b), it is treated
like the subjects of other intransitive verbs under Clause Union. It
.
therefore becomes the direct object of the causative, just as other
intransitive subjects do. This is shown by case marking (Pattern A),
Person Agreement, etc. The fact that the experiencer of an intransitive
inversion verb is treated like other intransitive subjects argues against
an analysis of inversion constructions where the experiencer has its
final grammatical relation, indirect object, at all levels of derivation. 11
Clause Union and Inversion also interact in the Series III forms of
organic causatives. I have stated above that all causatives are Class I
verbs, and that all Class I verbs undergo Inversion in Series III.
( 3oa) illustrates a complex causative in Series II ; (30b) shows the
Series III form of the same sentence.
(30) (a) mepem gagar;mendina sasaxle.
king-ERG he-caused-clean-you-it-II-I palace-NOM
'The king made you clean the palace.'
(b) turme mepes sentvis gaur;mendinebia sasaxle. 12
apparently king-DAT you-for he-caused-clean-it-III-I
'Apparently the king (has) made you clean the palace.'
•My analysis makes the correct predictions concerning the application
of Inversion to the output of Causative Clause Union. First, the subject
8.6 The form of the. rule 1 33
of the causative, · mepe 'king', should be the final indirect · object (by
Inversion). Second, the direct object of the causative, sasaxle 'p�lace�,
should be the final subject of the inversion construction (by Unac
cusative). Third, · the indirect object of the causative should be an
indjrect object chomeur (by the Chomeur Condition). Each .0£ these
predictions of derived termhood is borne out in (30b), as shown by
the case marking and agreement facts.
5.2 Summary and Extension

Arguments based on a variety of different syntactic and morphological
phenomena have been given for both the initial and final grammatical
relations claimed, and for Inversion both with inversion verbs and with
Series III forms. I3 In chs. I I and I S it will be shown that the rules .of
Number Agreement and Retired Term Marking cannot be stated
simply without an Inversion analysis. Other analyses cannot account
for the fact that the dative-nominal of inversion has the syntactic
characteristics of initial subjects and final indirect objects, or fo r the
fact that the nominative-nominal of inversion has the syntactic charac
teristics of initial direct objects and of final subjects. I conclude that
we must posit rules Of Inversion and Unaccusative in Georgian, to
relate the initial grammatical relations of (5) and . (23) to the final
grammatical relations of (6) and (24), respectively.
6 The form of the rule
Arguments presented above concern only the initial and final gram
matical relations in the inversion construction, without respect to the
form of the rules irivolved. We must consider in more detail what kind
of . rule or rules relate the initial grammatical relations to the final
ones.
Treatments of inversion phenomena written within the framework of
transformational grammar have generally proposed a rule that would
switch the relative line?r order of the subject and direct object (cf;
Rosenbaum 1 967 ; R. Lakoff 1 968 ; G. Lakoff 1970 ; Postal 1 970 & 1971).
This proposal has the problem · that a single rule cannot be used for
languages with different basic word orders. In addition, it makes the
wrong predictions about the word order of many languages, including
Hindi (Davison 1 969), Kannada (Sridhar 1976a, b), and Georgian,
where the initial subject of iriversion constructions and the subject of
134 Inversion
non-inversion constructions occupy the same posltlOn in unmarked

order. Further, this treatment fails to account for the fact that the
initial subject becomes indirect object, not direct object, as the differ
ence between these is, in many languages, not predictable merely from
the position of the nominal. One could, however, propose a relational
counterpart to this, which would obviate these problems. According to
such a proposal, two processes form a single rule :
A. Inversion
(i) Subject -7 Indirect Object
(ii) Direct Object -7 Subject
As an alternative to this, it might be proposed that Inversion is the

advancement of a direct object to subject, the initial subject becoming
a chomeur. The subject chomeur would then advance to indirect
object. According to this proposal, there are two different rules that
often apply together.
B. Inversion
(i) Direct Object -7 Subject
(ii) Subject Chomeur -7 Indirect Object
A third proposal is the one outlined in § I above and is due to Postal

and Perlmutter. According to this hypothesis, Inversion is a single
process :
C. Inversion
Subject -7 Indirect Object
The initial indirect object is put en chomage, according to the Chomeur

Condition. The direct object advances to subject, by Unaccusative.
Inversion proper is just the process of a subject becoming an indirect
object.
Each of these proposals makes different predictions about the range
of data to be found in natural languages. Proposal A predicts that the
two processes will always be found together in all examples. The
second predicts that each process will be found in some languages
without the other. The third predicts that the rule may be found without
Unaccusative, and that the latter will operate in sentences that lack a
subject for reasons other than the application of Inversion.
In addition, each of these proposals raises theoretical questions. Can
8.6 The form of the rule 1 35
a single rule have two parts, as in A ? (Is there any other rule that has
two parts ?) Can a chomeur advance, as in B ? (Is there any other rule
that advances chomeurs ?) Can a term be demoted, as in c ? None of
these questions can be answered here. Nor is it possible at this time to
test the predictions outlined above against the facts of many natural
languages. We can choose between these three proposals only by taking
into account the properties of many languages. However, Georgian does
shed some light on the problem ; these data too must be taken into
account in determining the correct approach.
All of the hypotheses outlined above can account equally well for
inversion with transitive verbs. In Georgian, Inversion can also apply
with intransitive verbs, as shown by the examples that follow. (3 1) con
tains intransitive non-inversion verbs in the indicative ; (32) gives their
counterparts in the evidential (Series III). (33) contains intransitive
inversion verbs.
(3 1 ) (a) merabi amtknarebs.

Merab-NoM he-yawns-I- I
'Merab is yawning.'
(b) merabi mulaobs.
he-works-I-3
'Merab is working.'
( c) merabi #ris
he-cries-1-3
'Merab is crying.'
(d) merabi ce1!-vavs.
he-dances-I-3
'Merab is dancing.'
(32) (a) merabs turme daumtknarebia.

Merab-DAT apparently he-yawned-III-I
'Merab apparently (has) yawned.'
(b) merabs turme umulavnia.
he-worked-III-3
'Merab apparently (has) worked.'
( c) merabs utiria.
he-cried-III-3
'Merab apparently (has) cried.'
1 36 Inversion
(d) merabs ucelJ-via.

he-danced-III-3
'Merab apparently (has) danced.'
(33) (a) bavSvs sia.

child-DAT he-hungers-I-4
'The child is hungry.'
(b) bavsvs sciva.
he-cold-I-4
'The child finds it cold.jThe child is cold.'
(c) bavsvs uxaria.
he-happy-I-4
'The child is happy, pleased.'
It is clear from (32) that intransitive verbs in the Classes that govern
Inversion in the evidential must undergo the rule, just as transitive
verbs do. Intransitive inversion verbs must also undergo Inversion in
all finite verbal categories.
In sentences of this sort, there appears to be no direct object to
undergo the first component of Inversion according to proposals A and B
above. Both proposals could be altered to accommodate the additional
data. On hypothesis A, we could make the direct object optional. That
is, if there is a direct object, it must undergo its part of the rule ; but
if there is no direct object, the second component of the rule could
still apply. On hypothesis B, part (ii) can apply only if the nominal has
been put en chomage (by the Motivated Chomage Law). A dummy
could be introduced to undergo the first rule, putting the subject en
chomage and thus implementing the application of part (ii).
Hypothesis c, on the other hand, can account for the intransitive
cases with no additional devices. Because it treats Inversion as just the
process of a subject becoming an indirect object, it is able to account
for (3 1-33) in a natural and simple way. Unaccusative will never be
involved in the derivation of sentences like these, since there is no
direct object. (3 1-33), then, would have no derived subject.H
(3 1-33) establish that Inversion applies independently of Unaccus
ative in Georgian. In ch. 1 3 it is shown that Unaccusative also applies
independently of Inversion. Unaccusative differs from Passivization in
that, while the latter applies only at a level of derivation where there is
a subject, the former applies only at a levelof derivation where there
8.6 The form olthe rule 137
is no subject (cf. Perlmutter 1978). This difference may be represented

by the partial networks below.
PASSIVIZATION l'N ACCUSATIVE
ff
I 2
R-I I
x y y
Un accusative thus never creates a chomeur. The fact that Unaccusative

applies only in the absence of a subject correctly prevents its applying
before Inversion with a Class 1 or 3 verb in Series III or with a Class 4
verb.
In Georgian, Un accusative applies obligatorily whenever its input
conditions are met. It is this property that blocks the derivation of a
sentence like ( * 3 4),
(34) *gelas uqvars ninos.
Gcla-DAT he-Ioves-her-I-4 Nino-DAT
('Gela loves Nino.')
where Inversion has correctly demoted gela to indirect object, . but
Unaccusative has failed to advance nino, leaving it as direct object,
marked with the dative according to Pattern B.
Hypothesis c handles the Georgian data more elegantly than the
other two proposals considered because it allows for the fact that Inver
sion applies independently of Unaccusatives, as established in (31-33),
as well as the fact that Unaccusative applies independently of Inversion,
as established in ch. 1 3 . It is for this reason that I have adopted it in
the analysis proposed here. It is proposed that the universal part of
these rules is, informally stated,
(35) Inversion
A subject demotes to indirect object.
(36) Unaccusative
In the absence of a subject, a direct object advances to subject.
Inversion in Georgian is triggered by
(37) (a) Class 4 verbs
(b) Series III forms.
138 Inversion
Trigger condition (b) must be further restricted :

(38) Class 1 and 3 verbs trigger Inversion, Class z does not.
In ch. 1 6 it is shown that trigger condition (38) follows from other
facts and that it is therefore unnecessary to state it in a grammar of
Georgian.
7 The interaction of Inversion with other rules of syntax
I have not yet described the interaction of Inversion with three rules -
Passivization, Object Raising, and Inversion itself. Since there are two
distinct triggers of Inversion, there are two distinct ways in which
Passivization and Object Raising could each interact with Inversion ;
this means a total of five potential syntactic environments for inter
action. In each of these five environments only one of the rules can
apply in a given clause.
7. 1 Passivization and Series III

A passive sentence may, of course, be in the evidential ; but Inversion
does not apply with passives in Series III. (39a) illustrates a passive
in Series I, (39b) its counterpart in Series III.
(39) (a) puli damalulia uJrast.
moneY-NoM hidden-it-is-I-z drawer-in
'The money is hidden in a drawer.'
(b) puli damaluli qopila ujra!i.
hidden it-is-III-z
'The money is apparently hidden in a drawer.'
It is apparent from case marking, agreement, and other verbal mor
phology that Inversion has not applied in (39b). (40) illustrates direct
forms of the verb in (39), 'hide'. (40b) shows that this form of the verb
does trigger Inversion in Series III.
t40) (a) merabma damala puli ujraSi.
Merab-ERG he-hid-it-II-I
'Merab hid money in a drawer.'
(b) turme merabs daumalavs puli ujraSi.
apparently Merab-DAT he-hid-it-III-I
'Apparently Merab hid money in a drawer.'
8.7 Interaction with other rules 1 39
These sentences establish the failure of Inversion to apply to the

output of Passivization.15 Although (39a) has a subject, puli 'money',
this nominal does not undergo Inversion in Series III.
7.2 Passivization and Class 4

Inversion does not apply in clauses containing Class 4 verbs if they
have undergone Passivization. (41) illustrates the passive associated
with an Inversion verb.
(41) direktori se3ulelmlia.
director-NOM hated-he-is-I-2
'The director is hated.'
Case marking, agreement, and other verbal morphology show that
Inversion does not apply to the derived subject, direktori. (42) illustrates
the non-passive form of the verb in (41), showing that it is a Class 4
verb and an Inversion trigger.
(42) vanos S3uis direktori.
Vano-DAT he-hates-him-I-4
'Vano hates the director.'
The examples establish the inability of Inversion, triggered by a Class 4
form, and Passivization to apply in a single clause.
7.3 Object Raising and Series III

When an object-raised sentence is in the evidential, Inversion does not
apply. (43b) illustrates the Series III counterpart of (43a).
(43) (a) 1!-argi magalitebi 3nelia mosa3ebnad.
good examples-NoM hard-it-is-I-2 to-find
'Good examples are hard to find.'
(b) 1!-argi magalitebi 3neli qopila mosa3ebnad.
hard it-is-III-2
'Apparently good examples are hard to find.'
Case marking, agreement, and other verbal morphology show that
Inversion has not applied in (43b). (<<) shows that the verb of (43)
triggers Inversion in Series III, if Object Raising has not applied.
(<<) turme rezos mou3ebnia satvale.
Rezo-DAT he-found-it-III-I glasseS-NOM
'Apparently Rezo (has) found his glasses.'
140 Inversion
7+ Object Raising and Class 4

Inversion does not apply in clauses containing Class 4 verbs, if they
have undergone Object Raising. (45) illustrates an object�raised sen-,
tence ; (42) shows that other forms of this verb undergo Inversion.
(45) direktori advilia . vanostvis sesa3uleblad.

easy-he-is-I-2 Vano-for to-hate
'The director is easy for Vano to hate.'
(45) is a simple object-raised sentence of the kind discussed in ch� 4.

If Inversion had applied to the output of Object Raising, direktori
would be an indirect object. Case marking, agreement, and other
verbal morphology show it instead to be the final subject. ' Thus, Inver-
sion does. not apply to the output of Object Raising.
7.5 Class 4 and Series III

Both Class 4 and Series III trigger Inversion. But a Series III form of
a Class 4 verb undergoes Inversion once, not twice, as shown by (46) ;
(42) gives the corresponding clause in Series 1 .
(46) turme vanos s3ulebia direktori.

VanO-DAT he-hates-him-III-4 director-NOM
'Apparently Vano hates the director.'
The input conditions for Inversion are met initially arid again ' after
Inversion and Unaccusative have applied. (47) represents a theoretically
possible network for such a clause ; (48) represents the clause as it
actually occurs.
3uleba vallo direktori 3uleba vallo direhtori

, hate ' , hate '
The non�occurrence of (*47) has not yet been accounted for.
7.6 Summary
We have seen that Inversion - triggered by a Series III or a Class 4
form - fails to apply to certain structures that meet the input con-
Appendix A : Additional arguments 141
ditions that have been specified thtlS far. In ch. 1 6, §3 it is shown that
this is part of a more general, and perfectly regular, phenomenon.
8 Summary
This chapter has established that the initial grammatical relations of

(5) and (23) are realized as (6) and (24), respectively, and that Inversion
with inversion verbs and in Series III is essentially the same. In
addition, evidence has been adduced to support the form of the rules
of Inversion and Unaccusative proposed here, and the interaction of
Inversion with other rules considered in this work has been described ..
This analysis solves two major problems of Georgian syntax. It
shows why Object Camouflage does not apply in Series III. It shows
that Georgian does not have three distinct case marking patterns, but
only two. At the same time, the proposed analysis raises new questions.
If Pattern C reduces to a special instance of Pattern B, cannot Pattern A
be similarly reduced ? In ch. 9 it is shown that Pattern A does not reduce
to B. Why do Class l and 3 verbs, but not Class 2 verbs, govern Inver
sion in Series III ? This question is answered in ch. 16.
Appendix A : Additional arguments for inversion verbs
This appendix contains data to show that the arguments made in §3 above
are applicable, not only to Inversion in the evidential, but also to Inversion
'
triggere d by inversion verbs. There are no new arguments given here,. only
new data.
A. P E R S O N AGREEMENT
In §3.I, I showed that Person Agreement is stated on final termhood. The

analysis proposed for inversion predicates in §4 therefore makes the pre
diction that the nominative-nominal ( final subject) will trigger Subject
=
Person Agreement. It likewise predicts that the dative-n()minal ( final =
'
indirect object) will trigger Indirect Object Perso� Agreement.
As ( 1 ) shows, it is indeed the nominative-nominal that triggers Subject
Person Agreement. with inversion verbs. The bold-type subject marker in (Ia)
is from Pflradigm (8) above ; the example is in the aorist (Series II). In (Ib),
the. bold-type, agreement marker is from the set of secondary markers of
'
sub,j ect agreement (cf. n. 4) ; the example is in the preseht tense. (Series I). ,
,
142 Appendix A : Additional arguments
( I ) (a) me malinve mome(:onet tkven . 1 6

I-OAT then-immediately I-liked-you(PL)-II-4 YOU(PL)-NOM
' I liked you immediately.'
(b) gelas uqvarvar (me).
Gela-oAT he-Ioves-me-I-4 I-NOM
'Gela loves me.'
The second prediction is also borne out. (2) shows that the dative-nominal
triggers Indirect Object Person Agreement with inversion verbs. The
indirect object markers in heavy type in (2) are from paradigm (10) above ;
the sentences are in the present and aorist tenses.
(2) (a) (me) mom(:ons pelamusi.
I-OAT I-like-it-I-4 pelamushi (a food)-NOM
'I like pelamushi.'
(b) (sen) moge(:ona pelamusi.
YOU-OAT you-liked-it-II-4
'You liked the pelamushi.'
From the fact that Person Agreement is stated entirely on final termhood and
the fact that the nominative-nominal triggers Subject Person Agreement in
this construction and the dative-nominal Indirect Object Person Agreement,
it follows that the former is the final subject and the latter the final indirect
object.
B. OBJECT CAMOUFLAGE
Object Camouflage never ,applies with inversion predicates. Compare (*3a)

and (*3b), where Object Camouflage has applied, with (2a) and (Ib), respec
tively, where the rule has not applied.
(3) (a) * tems tavs mos(:ons pelamuSi.
my self-OAT he-likes-it-I-4 pelamushi-NOM
('I like pelamushi.')
(b) *gelas uqvars temi tavi.
Gela-oAT he-Ioves-him-I-4 my self-NOM
(,Gela loves me.')
In (*3a) Object Camouflage has applied to the dative-nominal, in (*3b) to
the nominative-nominal.
The fact that Object Camouflage never applies with inversion verbs is
correctly predicted by the proposed analysis in the following way. As shown
in ch. 3, Object Camouflage applies only to direct objects. Further, the rule
applies to the output of rules that change grammatical relations. But after the
application of Inversion and Unaccusative, there is no direct object for the
Appendix A : Additional arguments 143
rule to apply to. Thus, the ungrammaticality of (*3) is accounted for with no
special devices.
C. TaV - R E F L E X I V I Z A T I O N
In support of the analysis proposed in §4 above, I will show that tav- can
be coreferential only to the initial subject. It has been shown that tav- can be
triggered only by the initial subject of its clause. In (4a) tavis triggered by
the dative-nominal, temur, confirming that the latter is initial subject. In
(*4b), tavis triggered by the nominative-nominal ( final subject) ; the
=
ungrammaticality of this sentence shows that the final subject is not the
initial subject. (*4c) is identical to (*4b) except that the order of dependents
has been changed, showing that the relative order of pronoun and ante
cedent is not responsible for the ungrammaticality of (*4b).
(4) (a) temurs uqvars tavisi tavi.
Temur-OAT he-Ioves-him-I-4 self's self-NoM
'Temur loves himself.'
(b) *tavis tavs uqvars temuri.
self's self-oAT Temur-NoM
('Temur loves himself.')
(c) *temuri uqvars tavis tavs.
Thus, the dative-nominal of inversion verbs behaves like the dative
nominal ( initial subject) of inversion sentences in the evidential and like
=
the initial subject of other verbs, in that it alone triggers Tav-Reftexivization.

I conclude that the dative-nominal is the initial subject, as claimed in §4 above.
D. SUMMARY
In this appendix I have considered in turn each of the relevant arguments

given in §3, which were based on Inversion in the evidential. The arguments
presented in §3.4 are not applicable to Inversion with Class 4 verbs, but a
different argument based on suppletion is given in §4. I . The data presented
here lend further support to the analysis proposed in §4 for inversion verbs.
Only an analysis of this type can account for both the initial subject and final
indirect object properties of the dative-nominal and for both the initial direct
object and final subject properties of the nominative-nominal.
144 Appendix B : Verbs with no overt subject
Appendix B: Transitive inversion verbs with no overt subject
There is a small group of inversion verbs (Class 4 verbs) which pat,terp. like
meSinia 'I fear' in {r)Y
(1) (a) *meSinia sen / 3ayli /tvitmprinavi.
I-fear-it-I-4 YOU-NOM dOg-NOM airplane-NOM
('I am afraid of you/of the dog/of an airplane.')
(b) mesinia seni /3aylisa /tvitmprinavisa.
YOU-GEN dog-GEN airplane-GEN
'I am afraid of you/of the dog/of an airplane.'
Under certain circumstances,18 verbs of this Class require that their initial
direct objects be· marked with the genitive case, instead of the nominative,
which is usual for inversion verbs�
Consider the following hypothesis for accounting for (1) :
(2) Verbs of the sesineba 'fear' class mark their final subjects with the
genitive case.
I will show below that the genitive-nominal does not have the properties of
a final subject, and that (2) is therefore inadequate.
The first argument is, of course, the case itself. In Georgian, final terms are
never marked with a case other than the nominative, ergative, or dative. This
set of verbs would provide the only counter�example to that generalization.
Second, the genitive-nominal in (1 b) fails to trigger Person Agreement:
When the genitive-nominal is second person, the verb cannot be marked for
a second person (final) subject, as the ungrammaticality of (*3) shows. In (*3)
the verb bears a second person subject affix, but otherwise is identical to (1 b);
(3) *meSinixar seni.
I-fear-you-I-4
('I am afraid of you. ')
Nor does the genitive-nominal trigger Object Person Agreement. For these
reasons, I reject the analysis (2) for this sub-class of verbs.
A second possibility for accounting for sentences like (r) involves making
the genitive-nominal the complement of a head noun, which then gets
deleted or dropped. This would explain why these nominals are in the
genitive, and not some other case. The complement of verbs like 'hear',
'fear', etc. have special case marking in many languages. This hypothesis
fails to explain the fact that in languages like German and Turkish the special
case marking used for the complements of such verbs is not genitive.
A third proposal involves the insertion of a dummy, putting the initial
Appendix B : Verbs with no overt subject 145
direct object en chomage. This insertion might apply before or after Unac
cusative, the dummy being inserted as direct object or subject, respectively.
Other analyses fail to explain why the complement of this sub-class of verbs
is marked with the genitive and not some other case. But on the dummy
insertion analysis, their inarking is predicted from a more general process. In
ch. I I I will show that all direct object chomeurs are marked with the genitive
case. I conclude that the most reasonable way of accounting for sentences
like (I) is by a dummy insertion rule. It is to be hoped that investigations of
other languages will shed more light on the nature of such rules.
9 Why Pattern A is not reducible to
Pattern B
It was shown in ch. 8 that the case marking chart represented here as
(I) is unduly complicated and can be simplified to ( I ').
(I) Series I II III
Class I , 3 B A C
2 B B B
4 C C C
( I ') Class I , 3 B A B
2, 4 B B B
( I) refers to the patterns as stated below in (2), while (I ') refers to the
restatement in (2').
C DATIVE NOMINATIVE tvis-nominal
(2' ) Final Final Final
Subject Direct Object Indirect Object
The simplification in (2') is due to the fact that (2) was based on initial
grammatical relations, which is inappropriate, since case marking in
Series III is stated on final grammatical relations. Once this is realized,
and the rule of Inversion is incorporated in the grammar, C is reduced
to a special instance of B, and need not be included in the case marking
rules at all, as shown in ch. 8, §4.2.
It seems natural, then, to ask, (Is not Pattern A of (2') reducible in a
similar way to a special instance of Pattern B ?' The answer, unfor-
9.3 Arguments against the proposed analysis 147
tunately for the simplicity of our grammar, must be 'no'. The reasons
for this are given below.
J: An analysis of A as a special instance of B
The most natural way of reducing Pattern A to Pattern B would be to

consider that the nominative-nominal in Pattern A is the final subject.
That would require that verbs of Class I and 3 undergo a rule that
advances direct objects to subjecthood, thus putting the initial subject
en ,homage, and giving it special chomeur marking with the so-called
ergative case.
(3) Direct objects advance to subjecthood.
Subject chomeurs are marked with the ergative case.
This is not an unreasonable analysis. 1 It would require verbs of Class
I and 3 to undergo an obligatory change of grammatical relations
triggered by Series II tenses, just as they undergo the change of
grammatical relations called Inversion, which is triggered by Series III
tenses. The change from Series I forms to Series I I is primarily a
change of tense and aspect, while that from Series I or II to Series
III is primarily a change of mood.
According to this analysis, sentence (4b), which is in Series II, differs
from (43.), which is in Series I, in that the former has undergone rule
(3), and the final subject is simindi 'corn'.
(4) (a) glexi tesavs siminds.
farmer-NOM he-sows-it-I-I corn-DAT
'The farmer is sowing corn.'
(b) glexma datesa simindi.
farmer-ERG he-sowed-it-II-I corn-NOM
'The farmer sowed corn.'
Although this seems to be a reasonable analysis, I will show below
that it cannot be right.
z Arguments against the proposed analysis
2.I Person Agreement

It has been shown that final subjects trigger Subject Person Agreement.
According to (3), the nominative-nominal, being final subject, should
trigger this rule. It does not, as shown in (5).
148 Why Pattern A is not reducible to Pattern B
(5) (a) (me) gamgzavnes s�ola1i.
I-NOM they-sent-me-II- I school-in
'They sent me to the school.'
(b) amanati miiyo.
packet-NoM he-received-it-II-I (zero marker)
'He received a package.'
In (5) the nominative-nominal triggers object markers, not subject
markers.
Further, the ergative-nominal, being a chomeur on this analysis,
should not be able to trigger Person Agreement at all ; in ch. I , §3, and
subsequent chapters it was established that final non-terms never
trigger Person Agreement in Georgian. Yet the ergative-nominal trig
gers Subject Person Agreement, as shown in examples (6).
(6) (a) namcxvari gamovacxve (me).
pastry-NoM I-baked-it-II-I I-ERG
'I baked pastry.'
(b) istvene (fen)?
you-whistled-II-3 YOU-ERG (zero marker)
'Did you whistle?'
Direct objects trigger Direct Object Person Agreement, which must
be stated on final terms (ch. I). According to the analysis proposed
above in this chapter, no nominal should trigger this rule, there being
no final direct object, in Series II with Class I or 3 verbs. However,
the nominative-nominal triggers Direct Object Person Agreement, as
shown in ch. 2 and example (5b) above.
The fact that the nominative-nominal triggers Direct Object Person
Agreement and the ergative-nominal triggers Subject Person Agree
ment, when coupled with the fact established in chs. 3-8 that Person
Agreement is stated simply on final termhood, indicates that these
nominals could not possibly bear the grammatical relations assigned to
them by rule (3).

In ch. 3 it was shown that a first or second person direct object is
obligatorily replaced by a tav -nominal if there is an indirect object in
the clause. It has been shown further that this rule is stated on the
output of rules that change grammatical relations. The analysis pro
posed above would thus predict that Object Camouflage never applies
9.2 Arguments against the proposed analysis 149
with Class I or 3 verbs in Series II, since (3) obviates the direct object.
(7) shows that Object Camouflage does apply.
(7) (a) *gelam semadara sen.

Gela-ERG he-compared-me-him-II-I YOU-NOM
('Gela compared you to me.')
(b) gelam semadara seni tavi.
your self-NOM
'Gela compared you to me.'
The fact that the nominative-nominal is tav- shows that it must be the
final direct object. (3) makes incorrect predictions in this respect.
2.3 Unemphatic Pronoun Drop

In ch. I, §4, I showed that terms can undergo Unemphatic Pronoun
Drop, while non-terms do not undergo this rule. In later chapters,
especially 5, 7, and 8, it was shown that Unemphatic Pronoun Drop is
stated on final termhood. In particular, it has been shown several times
that retired terms do not undergo this rule.
According to the analysis proposed in §I, the ergative-nominal is a
subject chomeur ; it therefore should not undergo Unemphatic Pronoun
Drop. But in ch. 2, §2.5 it was established that the ergative-nominals
in clauses containing Class I and 3 verbs in Series II do undergo this
rule. Thus, (3) makes incorrect predictions with respect to the rule of
Unemphatic Pronoun Drop.
2.4 Number Agreement

In ch. 1 5 it is shown that only final terms can trigger Number Agree
ment in Georgian. While the trigger conditions on Number Agreement
are much more complex than this, this single facet of the rule provides
the basis for an additional argument against rule t3).
(8) shows that a plural ergative-nominal in Series II triggers Number
Agreement. The plurality is marked with the fusional suffix -es.
(8) durglebma gaa�etes �arada.

carpenters-ERG they-made-it-II-I bookcase-NOM
'The carpenters made a bookcase.'
According to (3), the ergative-nominal is an initial subject, but not a

final term, and therefore not an eligible trigger for Number Agreement.
150 Why Pattern A is not reducible to Pattern B
Thus, the fact that ergative-nominals trigger Number Agreement is not

consistent with (3) ; it is consistent with their being final subjects.
3 Conclusion
We have seen that the ergative-nominal behaves like a subject with

respect to all rules stated on final terrnhood ; rules that change gram
matical relations provide no evidence relevant to the proposal made in
g I . It is important to note that the ergative-nominal does not lack any
characteristics of subjects - initial or final. Only in case marking does
it differ from the nominative-nominal of Series I ; these two nominals
behave in exactly the same way with respect to all other rules. This is
in marked contrast to Series III, where it is the dative-nominal that
behaves like the nominative-nominal of Series I in many respects.
If (3) were incorporated into the grammar in order to simplify the
rules of case marking, all of the following rules would have to be
complicated : Subject Person Agreement, Direct Object Person Agree
ment, Object Camouflage, Unemphatic Pronoun Drop, and Number
Agreement. I conclude that (3) must be rejected and that a complex
set of case marking rules must be incorporated into the grammar.
lO Non-finite verb forms
In Georgian there are at least three types of non-finite verb forms : the
so-called 'masdar' l or nominalization, the infinitive, and the participle,
including so-called 'future (passive) participles', 'past passive par
ticiples', and 'active participles' (cf. Chikobava 1950 ; Shanidze 1 973 ;
Tschenkeli 1958 ; Vogt 1971 for details). The foci of this chapter are the
masdar and the infinitive, but most of the remarks made in §3-5 apply
equally to the participles. Space prevents my considering them in detail.
Non-finite verb forms are not well understood from the point of view
of linguistic theory or language universals. In particular, I know of no
theory that accounts adequately for the initial and final grammatical
relations borne by the nominals governed by non-finite verb forms.
Nor is there a theory that can predict when a subordinate clause will
be realized as a non-finite construction. Is a verb realized as a non-finite
form because of the application of a relation-changing rule or of some
other rule of syntax ? Or is the occurrence of a non-finite form pre
dictable from universal principles such as those that predict the occur
rence of a chomeur (cf. Introduction, §z) ? These questions will not be
answered here. I hope, however, that the data presented will help to
clarify the nature of non-finite verb forms and contribute to solving
some of those problems.
In this chapter, I wish to make the following points about non-finite
verb forms : ( § I ) So-called 'masdars' are derived nominals, not gerun
dives ; ( §z) forms characterized by the morphology sa-elad are syn
chronically infinitives, not participles ; ( §3) nominals governed by non
finite verb forms are marked in ways that are predictable on the basis
of the initial grammatical relations they bear ; ( §4) although these
nominals bear term relations initially, they are final non-terms. While
I am not proposing a complete analysis of non-finite verb forms in
Georgian, the facts presented here will have to be accounted for by
any analysis that might be proposed in the future.
152 Non-finite verb forms
This chapter and the one that follows are intimately related. On the
basis of material presented in §3 and elsewhere, I will propose in ch. I I
a general system of retired term marking. That, in turn, provides evi
dence for a clausal source for non-finite verb forms. This explains some
of the facts presented in §4 and at the same time represents a new kind
of argument for a clausal source for 'derived nominals', which masdars
are shown to exemplify in § I.
I Masdars : gerundives or 'derived nominals'?
Chomsky ( 1970) observes that gerundives and so-called derived nomi

nals differ in at least three ways. I will show below that according to
each of these criteria, masdars are derived nominals.
First, there are a number of constructions to which Masdar For
mation cannot apply. Compare the sentences of ( I ) with the ungram
matical nominalizations in (2).
(I) (a) �argi magalitebi 3ne1ia mosa3ebnad.
good examples-NoM hard-it-is-I-z to-find
' Good examples are hard to find.'
(b) mepem gaarmendina sasaxle.
king-ERG he-caused-clean-him-it-II-I palace-NOM
'The king had the palace cleaned.'
(2) (a) *�argi magalitebis 3ne1i qopna mosa3ebnad. . .
examples-GEN hard being
('Good examples' being hard to find . . .')
(b) *sasaxlis garmendineba mepis mier. . .
palaCe-GEN cause-cleaning king by
('The king's causing the palace to be cleaned . . . ' )
Second, the relationship between the masdar and the associated
proposition is often idiosyncratic. Compare the verbs and nominaliz
ations of (3) in form and meaning. The formant of the masdar is in
heavy type in each.
(3) dapatizebs 'he will , invite him' dapa/izeba 'invitation'
dalJargavs ' he will lose it' da�argva 'loss'
mospobs 'he will destroy it' mosp oba 'destruction'
stxovs 'he asks him for it' txovna 'request'
unda 'he wants it' ndoma 'desire"
IO.I Masdars 153
icinis 'he laughs' sicili 'laughter'

ibrJvis 'he fights' brJola 'battle'
tquis 'he lies' tquili 'lie'
lJalJanebs 'it clucks' lJalJani 'cluck'
saubrobs 'he converses' 2 saubari 'conversation'
9ava 'he will go' fasvla 'going'
Though there are tendencies toward regular correspondences, the use
of the affixes -a, -oma, -ita, si-ila, etc. is not completely predictable
from other facts about the verb. For example, #ris 'he cries', icinis 'he
laughs', and kris 'it blows' are verbs of the same morphological type.
There is nothing to tell us that their masdars will be formed with the
affixes -iii, si-ili, and -ola, respectively. Further, some roots undergo a
change that is not predictable by regular phonological rule ; for example,
the root cin is reduced to c in sicili 'laughter'.
There is also variation in the meaning. Although in general masdars
have the 'factive' and 'action' meanings, they have additional ones as
well. 3 For example, rt;mena means both 'the act/fact of believing' and
'belief'. Tquili means both 'lying' and 'lie'. Dap atizeba means 'invi
tation' in the abstract sense, but not in the sense of something that can
be held in the hand ; there is a distinct expression for that. Soba means
both 'birth' and 'Christmas' .
Third, masdars have the internal structure of a noun phrase. Masdars
have the following noun-like properties :
(a) They decline.
rfmffla 'belief-NoM', rfmfflis 'belief-GEN',
rfltlenas 'belief-DAT', rfmfflit 'belief-INST', etc.
(b) They have number, when they are concrete and not mass-nouns.
mogoneba 'remembering, memory'
mogonebebi 'memories ' 4
(c) They may have quantifiers and adjectives.
erti txovna 'one request'
salineli tquili 'terrible lie'
(d) They may be possessed.
givis tquili ' Givi's lie'
(e) They may co-occur with positionals (postpositions).
tquilis sesaxeb 'about lying/the lie'
(f) They may have relative clauses.
cxubi, romelic vnaxe 'the fight which I saw'
In addition, masdars occur in the full range of nominal functions :

(4) (a) miqvars saubari.
I-Iove-it-I-4 conversation-NoM
'I love conversation.' I 'I love to talk.'
(b) vilapara1!et kalakis mosp oms sesaxeb.
we-talked-II-3 city-GEN destruction about
'We talked about the destruction of the city.'
Thus, according to the criteria outlined in Chomsky ( 1970), Georgian
masdars belong to the type of nominalization there called a 'derived
nominal'. Ii (They are glossed with English -ing forms in order to
standardize them.)
2 Infinitive or 'future participle in the adverbial case'?

Non-finite verb forms like gasa�eteblad 'to make, do' are traditionally
called 'future participles in the adverbial case' . I n this section I will
argue that, while that name may be descriptive of their origin dia
chronically, it does not accurately represent their use synchronically. I
will argue that they are infinitives. This is important for two reasons.
First, in Georgian we can predict that an infinitive will occur when the
subject of the verb it represents is deleted by Equi. Second, by using
terminology consistently, we can compare constructions across lan
guages ; the Georgian infinitive is comparable to infinitives in other
languages.
True future participles are used as modifiers, as illustrated in (5).
The participle is in the dative case in (sa) because it modifies the head
noun indaurs 'turkey', which is the direct object of the verb in Series I ,
and therefore in the dative. In (Sb) the verb is in Series II, and the
direct object and its modifying participle are accordingly in the nomi
native.
(5) (a) vqidulob dasa�lav indaurs.
I-buy-it-I-3 killing-DAT turkeY-DAT
'I buy a turkey to kill./I buy a turkey for killing.'
(b) viqide dasa�lavi indauri. 6
I-bought-it-II-3 killing-NOM turkey-NOM
'I bought a turkey to kill./I bought a turkey for killing.'
The future participle in this use precedes its head noun and agrees
Io.>(lnfinitives 15S
with it in case. I will show below that the so-called 'future participle in
the adverbial case' does not have these properties.
There is no doubt that the infinitive in Georgian bears the mor
phology of the future participle (sa-el) and that of the adverbial case
(-ad). I suggest, however, that synchronically there is a circumfix
sa-elad, which is probably derived diachronically from the conjoining
of the other two ; I shall argue that forms so constructed are infinitives,
not participles. 7
First, observe that the future participle in the adverbial case occurs
in two of the same uses as the infinitive in English. In (6) it is the
chomeur of the sentential subject in an object-raised sentence. In (7)
it represents a purpose clause.
(6) kargi magalitebi snelia mosasebnad.
good examples-NoM hard-it-is-I-2 to-find
'Good examples are hard to find.'
(7) (:avedi tqe§i datvis mosalJ.lavad. 8
l-went-II-2 woods-in bear-GEN to-kill
'I went into the woods to kill a bear.'
As far as I am aware, these are the only uses of this form in Georgian. 9
It cannot be used in all situations where an infinitive would be used in
English ; for example (*8) is ungrammatical.
(8) *minda (misi) gasalJ.eteblad. 1 o
l-want-it-I-4 it-GEN to-do
('I want to do it.')
Second, notice that the forms in (6-7) do not have the characteristics
of the participle in (5) :
a. In (5) the participle modifies a head noun ; in (6-7) the infinitive
does not.
b. In (5) the participle precedes its head noun ; in (6-7) the infinitive
is the head noun and in (7) is preceded by other words with which it
forms a constituent. 11
c. In (5) the participle shows case concord with its head noun ; the
case of the head noun itself is determined by its grammatical relation
to the governing verb. In (6-7) the form of the infinitive is invariant.
Third, the uses of the infinitive in (6-7) are not covered by any of
the regular uses of the adverbial case. 1 2 That is, it is not predictable on
any independent grounds that the adverbial, rather than some other
case, would be used in (6) or in (7). Therefore, if it is treated as a 'future

participle in the adverbial case', the grammar must add distinct rules
for these uses of that case.
Fourth, this verb form can never have a final subject, though it may
govern other term nominals. That this form can have other initial terms
is shown below in §4. (9) shows that it cannot have a final subject.
(9) (a) *gela ak mO'llida saatis gasa�eteblad cems mier.
Gela-NoM here he-came-II-z clock-GEN to-fix me by.
('Gela came here for me to fix the clock.')
(bj *gelam ak moitana saati cemi fcemtvis
Gela-ERG he-brought-it-II-I clock-NOM me-GEN/me-for
gasa�eteblad.
to-fix
('Gela brought the watch here for me to fix (it).')
The grammaticality of (10) shows that the ungrammaticality of (9) is
due to the fact that the initial subjects have not been deleted, or are not
coreferential to the matrix subjects.
(10) (a) gela ak movida saatis gasa�eteblad.
'Gela came here to fix the clock.'
(b) gela ak moitana saati gasalf.eteblad.
'Gela brought the clock to fix (it).'
An infinitive is often, though not exclusively, used in other languages
when the subject of the verb it represents has been deleted or raised out
of its clause, as in the English equivalents of (6-8).
I conclude that this verb form is synchronically a true infinitive, not
a future participle in the adverbial case. I suggest that the infinitive
occurs in Georgian if and only if the subject of the verb it represents
has been deleted by Equi. Equi applies in Object Raising, as discussed
in ch. 4, and in purpose clauses, as illustrated in (7) and in n. 8. In
many of the situations where Equi applies in English, Unemphatic
Pronoun Drop applies instead in Georgian, as in example (8) and
(vi) of n. 10.
3 The marking of nominals governed by non-finite verb forms

In this section I will note the correspondences between the initial
grammatical relation that a nominal bears to a finite verb and the
IO.3 The marking o
' f nominals 157
marking it receives in a masdar or infinitive construction. Arguments

to support the grammatical relations are given in §4. 1.
3 . 1 Direct object
The nominal that corresponds to the initial direct object of a finite
verb is, in construction with a mas dar or infinitive, in the genitive case.
This is so whether it is the only overt term, as in ( I I ), is with an overt
subject, as in ( 1 2), or is with both an overt subject and indirect object,
as in (1 3)'
( I I ) datvis mof$vla am !qeSi a�r3alulia.
bear-GEN killing-NoM this woods-in forbidden-it-is-I-2
'Killing bears in these woods is forbidden.'
( 12) monadiris mier datvis mof$vla a�r3alulia.

hunter . by
'The killing of bears by hunters is forbidden.'
( 13) mamas unda qvavilebis micema mziastvis

father-DAT he-wants-it-I-4 flowers-GEN giving-NoM Mzia-for
vanQs mier.
Vano by
'Father wants Vano to give flowers to Mzia.'
While (13) is grammatical, such a sentence would rarely occur ; a finite

verb form would be more likely to be used.
Examples (7), (loa), and (i) of n. 8 illustrate an infinitive's direct
object in the genitive case. A direct object cannot occur with the
infinitive that is the chomeur of the sentential subject of an object
raised sentence, since the direct object is raised out of its clause (cf.
ch· 4)·
3.2 Subject of an intransitive

The nominal that corresponds to the subject of the finite form of an
intransitive verb is, with a non-finite verb form, in the genitive, case.
In (14-15) the intransitive subject is the only nominal governed by
the masdar ; in ( 1 6) an indirect object is also present.
(14) tamadis damtknareba supraze uzrdelobaa.

tamada-GEN yawning-NoM table-on rudeness-it-is-I-2
'It is rude for the tamada to yawn at the table.'
158 Non..jinite verb forms
( I S) cxenosms ak sesvla alp'3alulia.
horseman-GEN here entering-NoM forbidden-it-is-I-2
'It is forbidden for horsemen to enter here.'
( 1 6) lemtvis bavsvis salami sasixaruloa.

me-for child-GEN greeting-NoM pleasant-it-is-I-2
.
'It is nice for the child to call me.'
As noted in §2, infinitives do not have overt subjects.
3.3 Subject of a transitive

The nominal that corresponds to the subject of a transitive finite verb
is, with the masdar, marked with the postposition mier 'by', which
governs the genitive case.13 This has been illustrated in ( 1 2) and ( 13),
where the subject co-occurs with a direct object. (17) shows that the
same marking is used when the direct object is not overt.
( 17) mona diris mier mo1!,vla a1!,r3alulia.

hunter by killing-NoM forbidden-it-is-I-2
'It is forbidden for hunters to kill.'
The subject of a transitive verb (cf. ch. 1 3) can never be assigned the
marking appropriate to the subject of an intransitive, the genitive case ;
for this reason (''' 1 8) is ungrammatical in the meaning of (17).
(18) flmonadiris mo1!.vla a1!,r3alulia.

hunter-GEN
('It is forbidden for hunters to kill.')
(flI8) is grammatical, however, in the meaning ' It is forbidden to kill
hunters', where 'hunter' is the initial direct object.
The subject of an intransitive similarly cannot be marked with the
postposition mier, which is appropriate only to the subject of a transi
tive. (flI9) and (fl20) are parallel to (14) and ( I S). with the difference
that their subjects are marked with mier.
( 1 9) fltamadis mier damtknareba supraze uzrdelobaa.

('It is rude for the tamada to yawn at the table.')
(20) flcxenosnis mier ak sesvla a1!,r3alulia.
('It is forbidden for horsemen to enter here.')
As observed above, infinitives do not co-occur with their initial subjects.
IO.3 The marking of nominals 1 59
3.4 Indirect objects

Those nominals that correspond to the indirect objects of finite verbs
are, in construction with non-finite verbs, marked with the postposition
-tvis ; they are not marked with a dative in Standard Modern Georgian
(cf. Harris 1979 on the use of the dative in this function in Old
Georgian). The use of -tvis is illustrated with masdars governing a
variety of nominals in ( 1 3) , ( 1 6), and (21 ). An initial indirect object
with an infinitive is illustrated in (22), (34b), and (42a).
(21) ristvis iqide es qvavilebi? minda deidistvis micema.
l-want-it-I-4 aunt-for giving-NoM
'Why did you buy these flowers ? I want to give them to Aunt.'
(22) es qvavilebi viqide maspin3listvis misacemad.
this flowers-NoM l-bought-it-II-3 host-for to-give
'I bought these flowers to give to the host.'
The indirect object can never be marked with just the genitive case.
(""23), (·24), and (·25) correspond to (21 ), (22), and ( 1 6) , respectively.
In the ungrammatical examples, the genitive case alone marks the
initial indirect object.
(23) ·minda deidis micema.
aunt-GEN
('I want to give them to Aunt.')
(24) ·es qvavilebi viqide maspin3lis misacemad.
host-GEN
('I bought these flowers to give to the host.')
(25) ·cemi salami sasiamovnoa.
me-GEN
(,The greeting to me was pleasant.')
(23) and (25) are possible if the genitive-nominal is interpreted, not as
initial indirect object, but as initial direct object Or intransitive subject ;
that is 'I want to give away Aunt. . .' and 'My greeting was pleasant.'
3.5 Summary
Nominals corresponding to initial direct objects and subjects of intran
sitives are marked with the genitive case when they occur in construc
tion with masdars or infinitives. Similarly, initial transitive subjects
are marked with the postposition mier, and initial indirect objects with
the postposition -t'Vis. This marking is independent of what other nom
inals also occur in construction with the non-finite verb form.
.. The nature of the nominals governed by non-finite verb forms
The nominals governed by masdars and infinitives are shown to be

related to the initial terms of corresponding finite verb forms ( §4. I ) ,
but these nominals do not behave like final terms with respect to the
rules that are tests for final termhood (§4. 2). The nature of these con
structions is not well understood from the point of view of language
universals. An adequate analysis of the Georgian data cannot be made
outside a theory of the universal nature of such constructions a theory
that does not exist at this time.
4.1 Facts concerning initial grammatical relations

4 . 1 . I Suppletion for number of direct object. The verb 'throw' is sup
pletive for the number of its direct object, such that forms of gadagdeba
(root gd) occur with singular objects, while forms of gadaqra (root qr)
occur only with plurals (cf. Introduction, §4. 3.2).
The suppletion that is governed by the number of the direct object
is also governed by the genitive-nominal associated with the masdars
and infinitives of this verb. This is established for the masdar in (26)
and (27). In the (a) examples, the genitive-nominal is singular, in the
(b) examples plural. The sentence containing the masdar form gadagdeba
is grammatical only in example (a), that containing the masdar gadaqra
is good only in (b).
(26) (a) kvis gadagdeba alf,r3alulia.
stone-GEN throwing-NoM forbidden-it-is-I-2
'Throwing a stone is forbidden.'
(b) *kvebis gadagdeba alJr3alulia.
stoneS-GEN
('Throwing stones is forbidden. ')
(27) (a) *kvis gadaqra alf,r3alulia.
throwing-NoM
('Throwing a stone is forbidden. ' )
(b) kvebis gadaqra alf,r3alulia.
'Throwing stones is forbidden.'
The same suppletion is established for infinitives of this verb in (28).
I04 The nature of the nominals 161
In these examples, the preverb ca- 'down' has been substituted for
gada-. As the example shows, the same suppletion holds.
(28) (a) *ak amovedi didi kvis lasaqrelad.
here I-came-up-I1-2 big stone-GEN to-throw
('I came up here to throw a big stone.')
(b) ak amovedi didi kvebis lasaqrelad.
stoneS-GEN
'I came up here to throw big stones.'
An additional example is given as (8) in ch. 4.
An analysis that relates the genitive-nominal to the initial direct
object can account for the grammatical status of (26-28) with the same
rule needed independently to account for sentences with 'throw' that
have undergone no change of grammatical relations.
4. 1 .2 Suppletion for number of subject. As noted in the Introduction, the

verb dajdoma/dasxdoma 'sit' is suppletive for the number of its subject,
such that the first occurs with singular subjects, the second with
plurals. This same constraint holds with respect to the genitive-nominal
of these masdars. (Since these are intransitive verbs, their subject is
in the genitive, cf. §3.2.)
(29) (a) im bavsvis ik dajdoma seu31ebelia.
that child-GEN there sitting-NoM impossible-it-is-I-2
'It is impossible for that child to sit there.'
(b) *im bavsvebis ik dajdoma seu3lebelia.
children-GEN
C It is impossible for those children to sit there.')
(30) (a) *im bavSvis ik dasxdoma seu3lebelia.
sitting-NoM
(,It is impossible for that child to sit there.')
(b) im bavsvebis ik dasxdoma seu31ebelia.
'It is impossible for those children to sit there.'
In the (a) sentences the genitive-nominal is singular, in the (b) sentences
plural. Only when dajdoma occurs with a singular and dasxdoma with
a plural, are the sentences grammatical. This can be accounted for by
the same rule that accounts for the grammaticality of sentences with the
finite verb 'sit', if bavs(eb)i 'child(ren), is the initial subject of dajdoma/
dasxdoma.
4.1.3 Idioms. Idioms retain their idiomatic meanings in non-finite
forms. The collocation xels uIlis (literally 'he disturbs his hand') has
the idiomatic reading 'he distracts/bothers him'. The idiom retains its
meaning in the masdar construction in (31) and in the infinitive of
purpose in (32).
(31) xelis sella magizebs.
hand-GEN disturbing-NoM it-crazes-me-I-I
'Distraction drives me crazy.'
(32) aval gelastan xelis lesaSlelad.
I-will-go-I-2 Gela-to hand-GEN to-disturb
'I will go up to Gela's to bother him.'
The meanings of (3 1-32) can be accounted for simply with the rule
independently needed to account for the finite forms of this idiom, if
xeli 'hand' is analyzed as the initial direct object of the masdar and of
the infinitive.
4.1 .4 Restrictions on the inventory of initial terms. The root rer 'write'
occurs with a variety of preverbs. When it occurs with the preverb da-,
the verb takes an initial subject and initial direct object ; it may have
an indirect object only derivatively. When it occurs with the preverb
mi- or mo-, on the other hand, the verb takes an initial subject, direct
object, and indirect object. This has been discussed in previous chap
ters, especially ch. 1 , §4, and ch. 6, §4.
The same distribution of nominals holds with respect to the masdars
and infinitives corresponding to the finite verb forms, except that, as
noted above, the subject does not appear overtly with an infinitive.
This distribution of initial terms is exemplified for infinitives in (33-H).
(33) (a) avedi cems �abine#i rerilis dasarerad.
I-went-up-II-2 my study-in letter-GEN to-write
'I went up to my study to write a letter.'
(b) avedi lems �binetli rerilis dasarerad uprosistvis
boss-for
'I went up to my study to write a letter for the boss.'
(H) (a) avedi lems �abinetli rerilis misarerad.
to-write
'I went up to my study to write a letter to him.'
IO.4 The nature of the nominals 163
(b) avedi lems l$abine#i {:erilis misacerad uprosistvis.
'I went up to my study to write a letter to the boss.'
The different inventories of initial relations of these two different verbs
result in two differences between (33) and (34). First, (34a) is interpreted
as having an initial indirect object that has been dropped ; this is
reflected in the different translations of (33a) and (34a). Second, the
tvis-nominal of (33b) is interpreted as a benefactive (tvis means 'for'),
since this verb form cannot take an initial indirect object. But the identi
cal tvis-nominal in (34b) is interpreted as an indirect object, since this
form of the verb requires an initial indirect object, which has the
meaning 'to'.
(35-36) show that the same distribution holds for masdars and the
nominals they govern.
(35) (a) cerilis da{,:era. . .
letter-GEN writing-NOM
'The writing of the letter. . .'
(b) Cerilis da{,:era uprosistvis.
. .
boss-for
'The writing of the letter for the boss. . ' .
(36) (a) cerilis mi{,:era . . .

writing-NoM
'The writing of the letter to him . . .'
(b) cerilis micera uprosistvis. . .
'The writing of the letter to the boss. . ' .
The facts established in (33-36) can be accounted for with no additional

statement in the grammar, if each nominal governed by the non-finite
verb forms bears to the verb the same initial grammatical relation it
would bear to the corresponding finite verb form.
4. 1 . 5 Selection restrictions. The verb fevs 'he is lying' requires that its
subject be animate, as shown in (37).
(37) (a) gela fevs taxtze.
Gda-NOM he-lies-I-z couch-on
'Gela is lying on the couch.'
(b) *(:igni fevs taxtze.
book-NOM
(,The book is lying on the couch.')
The same constraint governs the genitive-nominal of the corresponding
masdar, fola, as shown in (38).
(38) (a) gelas fola taxtze. . .
Gela-GEN lying-NoM
'Gela's lying on the couch . . .'
(b) *fignis fola taxtze. . .
book-GEN
(,The book's lying on the couch. . .')
This can be accounted for with the same statement independently
needed to account for (37), if figni 'book' and gela are the initial subjects
of the masdars in (38), as in (37).
4.2 Facts concerning derived grammatical relations

In this section I will show that the nominals governed by non-finite
verb forms are final non-terms.
4.2.1 Case Marking. The nominals governed by masdars and infinitives

are not marked like terms in Modern Georgian. The marking they are
assigned has been given in detail in §3 and may be compared with that
listed in (2 ') of the preceding chapter.
4.2.2 Person Agreement. It has been shown that Person Agreement is

stated on final relations. The nominals governed by non-finite verb
forms differ from those governed by finite verb forms, in that the former
cannot trigger Person Agreement. This is shown for the subject of
masdars in (39) and for the direct object of infinitives in (40), where
the nominal in question varies for person, but the verb form does not
change.
(39) (a) gadafera cems mier. . .
copying-NoM me by
'My copying over. . .'
(b) gadafera sens mier. . .
you
'Your copying over . . .'
(c) gadafera imis mier. . .
him
'His copying over . . . '
(40) (a) cemi megobari movida cems sanaxavad.

my friend-NoM he-came-II-2 me-GEN to-see
'My friend came to see me.'
(b) cemi megobari movida sens sanaxavad.
YOU-GEN
'My friend came to see you.'
(c) cemi megobari movida gelas sanaxavad.
Gela-GEN
'My friend came to see Gela.'
It can be seen in (39) and (40) that there is no variation in the form of
the non-finite verbs for person of subject or object.
4.2.3 Object Camouflage. Direct objects that are first or second person
are camouflaged if there is an indirect object in their clause (ch. 3) ;
this rule applies to the output of rules that change grammatical relations.
Object Camouflage does not apply to nominals governed by. non
finite verb forms, as shown for masdars in (41) and for infinitives in
(42). In the (a) sentences the rule has not applied, in the (*b) sentences
it has.
(41) (a) seni cabareba mast;avleblistvis. . .

YOU-GEN rendering-NoM teacher-for
'Turning you over to the teacher . . .'
(b) *seni tavis cabareba mast;avleblistvis. . .
self-GEN
(,Turning you over to the teacher . . .')
(42) (a) gela movida sens casabareblad

Gela-NoM he-came-II-2 YOU-GEN to-render
mast;avleblistvis. 14
teacher-for
'Gela came to turn you over to the teacher.'
(b) *gela movida seni tavis casabareblad mast;avleblistvis.
YOU-GEN self-GEN
('Gela came to turn you over to the teacher.')
These sentences show that the genitive-nominal and tvis-nominal do

not behave like final direct and indirect objects, respectively, in relation
to the rule of Object Camouflage.
166 Non-finite 'Verb forms
4.2.4 Question Formation. As shown in §4.2.4 of the Introduction, nom

inals governed by an embedded finite verb cannot be questioned, when
Question Formation is governed by the matrix verb. But the dependents
of non-finite verb forms can be questioned, as shown in the examples
below.
(43) risi mo�vla aris a�3aluli?
what-GEN killing-NoM it-is-I-2 forbidden
'What is it forbidden to kill ?'
(44) visi sesvla aris a�r3aluli?
WhO-GEN entering-NoM
'For whom is it forbidden to enter ?'
(45) risi gadasatargmnad movida gela?
what-GEN to-translate he-came-II-2 Ge1a-NoM
'What did Gela come to translate ?'
(46) risi mosa�lavad faxvedi tqeSi.

to-kill you-went-II-2 woods-in
'What did you go into the woods to kill ?'
Additional examples involving infinitives can be found in ch. 4, §4.2.

These sentences illustrate a second difference between final terms and
the nominals governed by non-finite verb forms. When a nominal
governed by a finite form is questioned, the question-word immediately
precedes the verb (cf. Introduction, §4.2). But when a nominal governed
by a masdar or infinitive is questioned, as in (43-46), the Q-word pre
cedes the non-finite verb form, which in turn precedes the finite
verb form. Other orders are not possible, as ('11<47) and ('11<48) show. In
the (a) sentences, the Q-word holds the place that would normally be
held by the Q-word governed by a finite verb ; in the (b) sentences, the
non-finite form fails to fall immediately before the finite verb.
(47) (a) ""risi aris mo�vla a�r3aluli?
""risi aris a�r3aluli mo�vla?
(b) ""risi mo�vla a�r3alulia ?
""a�r3alulia risi mo�vla?
(48) (a) ""risi movida gadasatargmnad gela?

""risi movida gela gadasatargmnad?
(b) ""risi gadasatargmnad gela movida?
In (1I<47b) ArtS-Cliticization has applied. Another grammatical order
for (45) is (49).
(49) gela risi gadasatargmnad movida?
Thus, nominals governed by non-finite verb forms differ from final
terms in two ways with respect to Question Formation.
4.3 Conclusion
The evidence amassed above leads me to conclude that (i) a term! of
a finite verb form corresponds to an initial term! of a masdar and
infinitive, but (ii) the initial term! of a masdar or infinitive differs
superficially from the term! of the corresponding finite verb. The facts
presented here can be handled by an initial clausal source to which
syntactic rules apply or by a lexical analysis with redundancy rules.
In ch. I I I will argue for a clausal source.
I I Retired Term Marking
The theory of relational grammar advanced by Postal and Perlmutter

(Perlmutter & Postal 1974, 1977, to appear c, and elsewhere) defines a
set of retired terms. In preceding chapters we have seen six construoc
tions where retired terms may occur. In this chapter I will propose that
in Georgian there must be a set of general rules that mark retired terms
on the basis of the last term relation they bear, without reference to
the rule that retires them.
1: Motivating retired termhood

Relational grammar defines a class of retired terms. A retired term is a
nominal which bears a term relation in some stratum and which does
not bear any term relation in the final stratum. A retired subject, for
example, is a nominal which bears the subject relation as its last term
relation, and which is a final non-term.
Postal and Perlmutter suggest that there are two types of retired
terms - chomeurs and emeritus terms (Perlmutter & Postal 1 977. to
appear c). The former are relatively well understood (cf. references
cited in ch. 7, n. 7) ; their occurrence is entirely predicted by the
Chomeur Law and the Motivated Chomage Law (cf. discussion in
ch. 7, §3). A term becomes a chomeur if and only if another nominal
assumes the grammatical relation borne by it. Emeritus terms are
relatively poorly understood. In theory, their occurrence, too follows
from universal principles ; but there are problems concerning their
occurrence in causatives (cf. ch. 5, §I and references cited there),
and the possibility of their occurrence elsewhere remains relatively
unstudied.
In chs. 4 to 1 0, a class of retired terms has been identified in Georgian.
They are recognized by the fact that (i) they behave like terms with
respect to phenomena like suppletion and Tav-Reflexivization, which
II.I Motivating retired termhood 169
are stated on initial termhood, and (ii) they behave like non-terms with
respect to rules like Person Agreement and Object Camouflage, which
are stated on the output of rules that change grammatical relations. In
Standard Modern Georgian, final terms and final non-terms are clearly
distinguished by a number of phenomena considered in previous
chapters.
However, while linguistic theory defines two types of retired terms,
Georgian grammar does not distinguish between them in any way I
have discovered. All retired terms behave the same with respect to
rules stated on final grammatical relations. The marking they are
assigned does not distinguish between chomeurs and emeritus terms,
as shown below.
I suggest that only those retired terms are chomeurs, whose occur
rence is predicted by the Chomeur Law and the Motivated Chomage
Law. Which nominals are chomeurs, then, is completely determined by
the analyses given in previous chapters. There are at least two chomeurs
in Georgian. The mier-nominal in the passive is a subject chomeur ; it
is put en chomage by the advancement of the direct object to subjecthood
(cf. ch. 7, §3). The tvis-nominal in the inversion construction is an
indirect object chomeur ; it is made a chomeur by the demotion of the
subject to indirect-objecthood (cf. ch. 8, §2 and 6). In §4.2 I will
argue that the genitive-nominal that occurs with certain inversion verbs
represents a third chomeur, a direct object chomeur (cf. ch. 8,
Appendix B).
Perlmutter and Postal analyze as an indirect object emeritus the
initial indirect object that is, under Causative Clause Union, a final
non-term (cf. ch. 5, § I). I suggest that those retired terms in Georgian
that have not been identified as chomeurs are also emeritus terms. It
is to be hoped that by studying the occurrence of these in Georgian
and other languages linguists can discover the universal principles that
govern their occurrence. In Georgian there are a number of retired
terms that are not chomeurs in the sense defined above. The tvis
nominal in the organic causative is an indirect object emeritus. In the
passive, the initial indirect object is a final non-term.
In Object Raising, the direct object of a sentential subject is raised
to matrix subject; the embedded subject is deleted on identity with the
tvis-nominal of the matrix clause. The entire sentential subject is put
en chomage by the raising of the direct object ; an indirect object depen
dent of that clause is a retired indirect object (cf. ch. 4). There is no
170 Retired Term Marking
absolute way in Georgian to determine what kind of retired term this
represents.
Initial subjects of infinitives of purpose are deleted. As shown in the
preceding chapter, all other nominals bearing initial term relations to
verbs realized as infinitives of purpose or masdars are themselves
realized as retired terms.
In the remainder of this chapter I shall not distinguish between
chomeurs and emeritus terms, but shall refer to both as retired terms.
2 Correlations between grammatical relations and final marking
In this section I will review in turn the various retired grammatical

relations and show that each is consistently marked, regardless of the
construction in which it occurs and regardless of whether it is a chomeur
or an emeritus term. I am claiming that the set of rules that marks
retired terms is sensitive to the last term relation that a nominal holds
and to whether it is a final term or non-term. In each instance, the dis
cussion is limited to specified nominals ; unspecified retired terms do
not appear on the surface.
2.1 Retired direct object

A retired direct object is marked with the genitive case. We have seen
at least two examples of this. ( I ) illustrates an initial direct object in a
masdar, (2) in an infinitive of purpose.
( I) datvis mo�vla am tqeSi a�r3alulia.

bear-GEN killing-NOM this woods-in forbidden-it-is-I-2
'It is forbidden to kill bears in these woods.'
(2) favedi tqesi datvis mosa�lavad.

l-went-I I-2 to-kill
' I went into the woods to kill a bear.'
The genitive case is also used to mark the initial direct object in
a subset of inversion verbs. (3) illustrates such a verb with the initial
direct object tkven 'you(PL)'.
(3) mesmis tkveni.

l-understand-it-I-4 YOU-GEN
' I understand you.'
II.2 Grammatical relations and final marking 171
It is argued in ch. 8, Appendix B that these genitive-nominals are

retired direct objects.
2.2 Retired intransitive subject

A retired intransitive subject is marked with the genitive case. I know
of only one construction in Georgian that has retired intransitive
subjects. (4) illustrates genitive case marking of intransitive initial sub
jects of masdars of verbs of Classes 2 (r;asvla) and 3 (musaoba).
(4) (a) �argia �acebis r;asvla.
good-it-is-I-2 men-GEN going-NoM
'It is good that the men go.'
(b) �argia �acebis musaoba.
working-NoM
'It is good that the men work.'
2.3 Retired transitive subject

A retired transitive subject is marked with the postposItIOn mier,
which governs the genitive case. This occurs in at least two construc
tions.
In passives the initial subject is realized as a mier-nominal. This is
illustrated in (5), where 3ayli ' dog' is the retired subject.
( 5) bavsvi da�benilia 3aylis mier.
child-NOM bitten-he-is-I-2 dog by
'The child is bitten by a dog.'
With masdars, the initial transitive subject is also realized as a mier
nominal, as illustrated by (6).
(6) monadiris mier nadiris mo�vla a�r3alulia.
hunter by game-GEN killing-NoM forbidden-it-is-I-2
'It is forbidden for hunters to kill game.'
Thus, retired transitive subjects are marked generally with mier. 1
2 . 4 Retired indirect object

A retired indirect object is marked with the postposition tvis, which
governs the genitive case. There are at least seven situations in which
this occurs.
In a clause in which Inversion applies, an initial indirect object is
put en chomage by the demotion of the initial subject. This is illustrated

in (7).
(7) rezos turme ucukema samaJurz sentvis.
ReZO-DAT apparently he-gave-it-III- I bracelet-NoM you-for
'Rezo has apparently given you a bracelet.'
In Causative Clause Union, the initial indirect object of an embedded
transitive verb becomes the retired indirect object of the causative verb
and is marked with tvis, as in (8).
(8) mamam mimacemina vardebi dedistvis.
father-ERG he-caused-give-me-it-II-I roses-NoM mother-for
'Father had me give the roses to Mother.'
In the causative of a transitive, the initial embedded subject is the
derived indirect object of the organic causative. In Series I and II,
this is marked with the dative case, like other indirect objects. I n
Series I I I all causatives, being Class I verbs, undergo Inversion ; the
indirect object of the input to Inversion is put en chomage by the
demotion of the initial matrix subject. Like the indirect object chomeur
in (7), this one is marked with tvis. (9a) is the causative of a transitive
in Series I ; (9b) is the Series III counterpart of (9a).
(9) (a) masfavlebeli atargmninebs gelas
teacher-NoM he-causes-translate-him-it-I- I Gela-DAT
alJa1!is lekss.
Akaki-GEN poem-DAT
'The teacher had Gela translate Akaki's poem.'
(b) turme masfavlebels gelastvis gadautargmninema
apparently teacher-DAT Gela-for he-caused-translate-it-III- I
alf.a1!is leksi.
poem-NoM
'Apparently the teacher had Ge1a translate Akaki's poem.'
In both sentences, the embedded clause has the initial subject gela and
the initial direct object a1!a1!is leksi 'Akaki's poem' ; Causative Clause
Union makes the former the derived indirect object. Under Inversion,
the initial matrix subject becomes indirect object, putting gela en
chomage ; the causative direct object, a1!alf.is leksi, becomes subject.
(9) differs from (7) in that the retired indirect object in (9) is an initial
subject. not an initial indirect object, as in (7). Yet the two retired
II.2 Grammatical relations and final marking 173
indirect objects are marked in the same way. This shows that the final
marking of retired terms depends on the last held term grammatical
relation.
Retired indirect objects also occur in passives ; these too are marked
with lvis, as illustrated in (10).
( 1 0) valli micemulia masfavleblistvis.
apple-NoM given-it-is-I-2 teacher-for
'An apple is given to the teacher.'
( I I ) illustrates the marking of the retired indirect object of a masdar
with lvis.
( I I ) vallis micema masfavleblistvis. . .
apple-GEN giving-NoM
'Giving an apple to the teacher . . . '
With infinitives of purpose, too, retired indirect objects are marked
with tvis, as in ( 1 2).
( 12) valli viqide masfavleblistvis misacemad.
apple-NOM I-bought-it-II-3 teacher-for to-give
'I bought an apple to give to the teacher.'
Finally, in object-raised sentences, an initial indirect object of the
embedded clause is marked with lvis. Here the initial indirect object is
a retired dependent of the infinitive.
( 1 3) salukari 3nelia anzoristvis misacemad.
gift-NOM hard-it-is-I-2 Anzor-for to-give
'Gifts are difficult to give to Anzor.'
Thus, tvis is used in a wide variety of constructions to mark retired
indirect objects.
2.5 Summary
The examples above are drawn from chs. 4 to 1 0 ; additional exa�ples
are given in each of the relevant chapters. In each instance, evidence
has already been given that the nominals marked with the genitive case,
or with mier, or with tvis are term nominals that have been retired by
the application of some rule.
In Modern Georgian retired terms are marked regularly according
to the last term grammatical relation they bear. Georgian does not dis
tinguish between chomeurs and emeritus terms by marking : it assigns
retired term marking on the basis of the last term relation held, not on
the basis of the rule that retires the nominal. One can imagine a number
of other principles upon which final non-term marking might be
assigned j in §3 some of these will be explored.
2.6 Overview of the proposed grammar

Rules of the type investigated in this work, rules that change grammati
cal relations, might be viewed as consisting of four parts : (i) the state
ment of the change of grammatical relation itself, (ii) the accompanying
morphological changes in the verb form, (iii) the accompanying changes
in nominal marking, (iv) conditions on the application of the rule. (i) is
generalized by relational grammar, so that the grammar of a language
need only register the fact that it has a particular rule and need not
itself state this part of the rule. Part (iv) of the rule may also be partially
universal, but languages also impose their own constraints. Parts (ii)
and (iii) must be language particular, since languages do not generally
share morphemes ; (ii) and (iii) may, of course, have certain things in
common across langua�es.
The grammar that I believe to be possible for Georgian would
require that rules that change grammatical relations state only the
following : (if) a registration of the fact that the grammar of Georgian
contains a particular rule, which is characterized in universal grammar,
(ii) the accompanying morphological changes in verb forms, and (iv')
conditions on the application of the rule, where necessary. The retire
ment of terms is specified by general principles, as described in § I of
this chapter. Once the retirement of terms has been specified by uni
versal principles, Georgian will mark them with a set of general marking
rules used for all retired terms. These are stated below.
( 14) (a) A retired direct object is marked with the genitive case.
(b) A retired intransitive subject is marked with the genitive case.
(c) A retired transitive subject is marked with the postposition mier.
(d) A retired indirect object is marked with the postposition tvis. 2
The definition given for retired term in §I is assumed in (14), Rules
(I4a) and ( I 4b) can be simplified as (I{ab).
( I{ab) A retired absolutive is marked with the genitive case,
where absolutive is defined universally as a direct object or intransitive
subject.
The generalizations made by ( I 4ab, c, and d) can be made only with
II.3 Alternative analyses 175
a theory which, like relational grammar, specifies a set of retired terms.

Any other approach would require more than these three statements
and would fail to show, for example, why some nominal is marked with
mier rather than with some other case or postposition. This is discussed
in detail in §3 .
3 Alternative analyses
Below I will discuss some other principled bases for marking retired
terms but in each instance I will show that this is not the best analysis
of Georgian.
Without further data, we might assume that at least the following
language types exist :
A. Retired terms are individually marked by the rule that retires them.
B. Retired terms are marked by a general set of rules on the basis of
the last term relation borne by the nominal.

c. Retired terms are marked by a general set of rules on the basis of
the initial term relation borne by the nominal.
D. Retired terms bear the same marking as the term relation they last
bore.
E. Retired terms are marked semantically.
Most analyses - of Georgian, as well as of other languages - have
assumed that principle A is the best to describe the data. Perhaps the
most familiar example of this kind of analysis is the rule of Passivization
usually stated for English ; one feature of it is the marking of the initial
subject with the preposition by. This approach is possible for Georgian,
but it fails to draw the significant generalizations embodied in the
rules (If). One could, for example, make tvis-assignment part of the
rules of Inversion, Causative Clause Union, Masdar Formation, Pass
ivization, Object Raising, and the rule that forms infinitives of purpose.
But that analysis would miss the generalization that all retired indirect
objects are marked with tvis, regardless of which rule retires them.
Data from a few of my Georgian informants suggest that c is a
possible language type. Instead of (Ifc, d), they seem to apply rules
( 1 5c, d).
( 1 5) (c) An initial transitive subject that is a final non-term is marked
with the postposition mier.
(d) An initial indirect object that is a final non-term is marked with
the postposition wis.
The evidence for this is a single type of example (cf. 9b) :

( 16) turme masravlebels gadautargmninebia aly.a�is
apparently teacher-DAT he-caused-translate-it-III-I Akaki-GEN
leksi gelas mier. 3
poem-NOw Gela by
'Apparently the teacher had Gela translate Akaki's poem.'
The nominal in question, gela, is an initial subject, made indirect object
by Causative Clause Union and retired by Inversion. I found the fol
lowing judgements : in Georgia none of my informants accepted ( 1 6),
only (9b). Among my three informants in the United States, who are
older and emigrated from Georgia some time ago, two accepted ( 1 6)
and rejected (9b) ; the third accepted only (9b). I assume that (16) rep
resents a real dialect, and hence possible language type, but that (9b)
represents the standard current in Georgia.
In theory there are other constructions that would also test the
difference between ( I fc, d) and ( I SC, d) : (i) In the passive of the causa
tive of a transitive verb, the initial embedded subject would become the
causative indirect object and be retired by Passivization. But Passiviz
ation applies to causatives only in a very limited way, and never to a
causative with an indirect object (cf. ch. 7). (ii) In a double causative of
a transitive verb, such as 'The boss made me make the secretary write
a letter', the initial subject of the most deeply embedded verb would
become indirect object, then be retired. Shanidze ( 1 973) describes the
occurrence of double causative verb forms, such as arerinebinebs 'he
causes him to cause him to write it'. Some of my informants say that
such forms exist, but none will use them in complete sentences with
case-marked nominals. Further, they reject all reasonable translations
of the English sentence above using the form arerinebinebs. Hence, the
inverted form of a causative is the only construction I know of that will
test the difference between the two hypotheses represented by ( I f)
and ( I S).
There are facts that suggest that type D is relatively frequent (cf.
Perlmutter & Postal 197f; Chung I 976a, b). In this system, a retired
term! is marked with the same case or positional used to mark a final
term!. Such a system was used in Old Georgian for retired indirect
II.3 Alternative analyses 177
objects (Harris 1979) and there are some petrified forms in Modern
Georgian that still mark on this basis (cf. ch. 5 , §4.2 ; ch. 6, §6). But
Standard Modern Georgian does not use this marking system pro
ductively (but cf. ch. 7, §2.3).
The putative type E language would mark former terms, not on the
basis of their termhood, but on the basis of semantics. 4 This would
mean, for example, that agents are marked with mier. This can be
shown to make the wrong predictions for Georgian. Agents are gen
erally limited to animate or personified nominals (Fillmore 1 968). But
( 1 7) illustrates the use of mier with nominals that are neither animate
nor personified.
( 1 7) (a) Nazmnari saxelis mier saxelta martva udur
deverbal noun by nOUnS-GEN governing-NoM Udi
enasi
language-in
'The governing of nouns by deverbal nouns in the Udi
language.'
Title of Panchvidze ( 1960)
(b) . .procesi
. ebiani mravlobitis mier zmnis setanxmebisa.
process-NoM plural by Verb-GEN agreeing-GEN
' . . . the process of the eb-plural agreeing with the verb.'
Imnaishvili (1957 : 664)
(c) . . .inpinitivis mier pirdapiri obiektis martvis
infinitive by direct object-GEN governing-GEN
semtxvevebi . . .
instances-NoM
' . . .instances of the governing of the direct object by the
infinitive. . .'
Chxubianishvili ( 1972 : 78)
There is evidence of another kind, too, that mier is not just an agent
marker. The initial subject of 'affective' (inversion) verbs is an
experiencer, not an agent. Yet this experiencer is marked with mier in
masdars, as shown in ( 1 8).
( 1 8) (a) tkveni megobris davifqeba cems mier didi
your friend-GEN forgetting-NoM me by big
uzrdelobaa.
rudeness-it-is-I-2
'It was very rude of me to forget your friend.'
178 Retzred Term Marking
(b) cemtvis gaugebaria tkven mier cemi

me-for not-understandable-it-is-I-2 you by my
megobris seJuleba.
friend-GEN hating-NoM
'For me it is incomprehensible that you would hate my
friend.'
I conclude from the evidence in ( 17) and (18) that mier does not mark
only agents, but rather marks retired subjects. More generally, I con
clude that the markers of retired terms are assigned, not on the basis of
semantics, but on the basis of grammatical relations. 5
In addition to the five possible language types discussed above, there
are many possible mixed systems ; for example, a language might choose
to mark all chomeurs with one case or positional and mark all emeritus
terms with another case or positional. But to use any mixed system to
describe Standard Modern Georgian would be to miss the generaliz
ations made in §2 above.
It has been shown here that type B above represents one possible
language type. It is hoped that further research in other languages will
reveal more about what types of marking exist and more about the nature
of retired terms, especially about the poorly understood emeritus
relations.
4 Implications of this analysis
4. 1 Object Raising
In ch. 4 I proposed an analysis of Object Raising according to which
the tvis-nominal of (19) originates in the matrix clause, not as the
subject of the embedded clause. At that time I discussed evidence for
this position from other languages, and evidence from Question For
mation in Georgian, which shows that the tvis-nominal is not a con
stituent of the embedded clause.
( 1 9) lJ.argi magalitebi Jnelia gelastvis mosaJebnad.

good examples-NoM hard-it-is-I-2 Gela-for to-find
'Good examples are hard for Gela to find.'
The system of former term marking proposed above gives additional

evidence for the analysis of Object Raising proposed in ch. 4. Retired
transitive subjects are marked with mier ( §2.3), so that if the tvis-
II.4 Implications 179
nominal in ( 1 9) were the retired subject of the embedded clause, it

should be marked by mier. This therefore provides another argument
that this nominal is a dependent of the matrix verb. 6
4.2 Transitive inversion verbs without final subjects

In ch. 8, Appendix B I discussed possible analyses for sentences like
(20), where the nominal that is apparently the initial direct object is not
final subject, as expected for inversion verbs.
(20) mestma 3aylisa.

l-fear-it-I-4 dog-GEN
'I am afraid of the dog.'
I argued there for an analysis involving the insertion of a dummy. vVe

can now see that Retired Term Marking provides additional evidence
in favor of this analysis. If a dummy is inserted, the term it replaces
becomes a chomeur (3ayli 'dog' in (20)). There are two possibilities :
(i) The dummy is inserted as direct object, making 3ayli a direct object
chomeur ; the dummy then undergoes Unaccusative, triggers Subject
Person Agreement (third person singular), but does not show up on
the surface. (ii) The initial direct object, 3ayli, undergoes Unaccusative ;
the dummy is inserted as subject, making 3ayli an intransitive subject
chomeur. Since direct object chomeurs and intransitive subject cho
meurs have the same marking in Modern Georgian (cf. §2. I and 2.2),
there is no way to decide between these two analyses. In either case,
the genitive marking of this nominal constitutes an argument for a
dummy analysis, since such an analysis would not need an additional
rule to mark this nominal with the genitive case. Further, this analysis
explains why the genitive is used in this construction in Georgian
instead of some other case.
4-3 Masdar Formation

The generalizations concerning the marking of retired terms also pro
vide a very strong argument in favor of a clausal source for masdars
(derived nominals). If masdars are derived from full clauses, no further
statement need be made concerning the marking of the dependents of
masdars. But if masdars are lexically inserted as nominals, the grammar
will have to state the generalizations concerning marking of retired
terms for the retired terms created by Object Raising, Clause Union,
Version, Passive, and Inversion ; it will have to state the same set of
generalizations a second time as redundancy rules in the lexicon for
masdars. Thus, a grammar that inserts masdars lexically as nominals
will miss the generalization that the retired terms of masdars are
marked like the retired terms of other verbs.
12 Transitivity
The verbs fers ' he is writing it' (Class 1) and tamaSobs 'he is playing'
(Class 3) are alike in that Series II counterparts of both govern ergative
case subjects, and Series III counterparts of both govern Inversion. In
this chapter I will discuss some syntactic differences between these
two verbs. This short chapter has three purposes : (i) to show that some
verbs have an initial direct object obligatorily, while others have an
initial direct object optionally or not at all, (ii) to relate this difference
in inventories of initial terms to the difference between Classes 1 and 3 ,
and (iii) t o discuss and define the notions 'transitive' and 'intransitive',
which will be important in later chapters.
I Analysis
I suggest that inventories of initial terms are lexically determined, and
that a given verb may govern a particular term relation either obli
gatorily, optionally, or not at all. In Georgian, a 'verb' in this sense
must be understood as a verb root with a specific preverb, if that root
takes preverbs. In ch. 1 it was shown that the verb da(:era 'write' takes
an initial subject and direct object, while the verb mi(:era 'write' takes
an initial indirect object, as well as an initial subject and direct object.
Thus, for the purposes of stating initial term inventories, da(:era and
mifera must be considered different verbs, even though they share the
root fer. l
I propose the following specific requirements for some verbs in
Georgian :
( 1 ) VERB GOVERNS AN INITIAL DIRECT OBJECT

da(:era 'write' obligatorily
mifera 'write' obligatorily
datesva 'sow' obligatorily
182 Transitivity
VERB GOVERNS AN INITIAL DIRECT OBJECT

daxrloba 'drown' obligatorily
gaJroba 'dry' obligatorily
gamocxoba 'bake' obligatorily
tamaH 'play' optionally
laparalli 'talk' optionally
cellva 'dance' optionally
qiqini 'croak' not at all
xitxiti 'giggle' not at all
I am claiming, then, that some verbs have an obligatory initial direct
object (e.g., da+era, daxrcoba), some verbs have an optional direct object
(e.g., tamaSi), and some do not permit an initial direct object (e.g.,
qiqini). II

In causatives, the subject of an intransitive is realized as the direct
object of the causative (ch. 5). The embedded subject of a transitive is
realized as the indirect object of the causative. This provides us with
a test for transitivity of the embedded verb.
With the verb dafera 'write' and others of its type, the initial subject
corresponds to the indirect object of the corresponding organic causative,
regardless of whether the initial direct object of dar;era is overtly
expressed. Example (4) gives the causatives corresponding to the simple
sentences (3).
(3) (a) vanom dar;era.
VanO-ERG he-wrote-it-II-I
'Vano wrote it.'
tb) vanom dar;era r;erili.
letter-NOM
'Vano wrote the letter.'
(4) (a) vanos davar;erine.
Vano-DAT I-caused-write-him-it-II-I
'I got Vano to write it.'
(b) vanos davar;erine r;erili.
'I got Vano to write the letter.'
Case marking and Person Agreement show at a glance that vano is the
I2.2 Causative Clause Union 183
final indirect object in the causatives, (4). (*5) shows that the causative
would be ungrammatical if vano were instead the direct object ; (* 5)
is identical to (4a), except that vano is case-marked as the direct object. 3
(5) *vano davaferine.4
Vano-NOM
('I got Vano to write.')
With the verb tamaSi 'play' and other verbs of its type, on the other
hand, the derived termhood of the initial subject in an organic causative
depends upon whether or not an initial direct object of tamaSi is overtly
expressed. The causatives in (7) correspond to the simple sentences
in (6). In the (a) sentences, no initial direct object of tamasi is overtly
expressed ; in the (b) sentences, the initial direct object sami partia
'three rounds' is expressed.
(6) (a) vanom itamaIa.
Vano-ERG he-played-II-3
'Vano played.'
(b) vitamaset sami partia.
we-played-II-3 three rounds-NoM
'We played three rounds.'
(7) (a) vano vatamase.
Vano-NoM I-caused-play-him-II-I
'I got Vano to play.'
(b) bebiam (even) gvatamasa
grandmother-ERG we-DAT she-caused-play-us-it-II-I
sami p artia.
'Grandmother got us to play three rounds.'
Case marking and Person Agreement reveal that here vano is the final
direct object in (7a), that 'us' is the final indirect object in (7b), and
that sami partia is the direct object. (8) shows that it would be ungram
matical for vano to be instead the final indirect object of the causative
in (7a).
(8) *vanos vatamaIe.3
VanO-DAT I-caused-play-him-II-I
('I got Vano to play.')
The simple sentences in (3) and (6) are apparently alike ; the causa
tives in (4b) and (7b) are apparently alike. But the causatives in (4a)
184 Transitivity
and (7a) are different. The initial subject of dOfera 'write' is the final
indirect object of the corresponding organic causative, regardless of
whether an initial direct object of dafera is overtly expressed. The
initial subject of tamali 'play', on the other hand, is the final indirect
object of the corresponding causative if an initial direct object of tamaSi
is overtly expressed - (7b) ; the initial subject is the final direct object
of the causative if no initial direct object of tamasi is expressed - (7a).
This difference between dar;era and tamasi can be handled most
simply by (i) assigning to these verbs initial direct objects as specified
in ( 1 ) , and (ii) recognizing that Unemphatic Pronoun Drop does not
apply to nominals which are optional in the inventory of initial terms. 5
Retired subjects of transitive verbs are marked with the postposition

mier, while retire<;l subjects of intransitive verbs are marked with the
genitive case (ch. I I ) . The marking given to retired terms therefore
provides an additional test for transitivity.
With the verbs dar;era 'write' and miyeba 'receive', the retired subject
is marked with mier, regardless of whether a direct object is overtly
expressed. In the nominalizations of (9), the retired direct object is
expressed ; in ( 1 0) it is not. In both, mier is required.
(9) (a) vanos mier ferilis dar;era . . .
Vano by letter-GEN writing
'Vano' s writing (of) the letter. . .'
(b) glexis mier simindis datesva . . .
peasant by COrn-GEN sowing
'The peasant's sowing (of) corn. . . '
(c) rezos daxrcoba vanos mier . . .
Rezo-GEN drowning Vano by
'Vano's drowning (of) Rezo. . .'
(d) bi{is mier {ur?elis gaSroba. . .
boy by crockerY-GEN drying
'The boy's drying (of) the crockery . . .'
( 1 0) (a) vanos mier dafera . . .
'Vano's writing [something] . . .'
(b) glexis mier datesva . . .
'The peasant's sowing [something] . . . '
I2.3 Retired Term Marking 185
(c) vanos mier daxrcoba . . .

'Vano's drowning [someone] . . .'
(d) bitis mier gahoba. . .
'The boy's drying [something/someone] . . . '
With the verbs tamaSi 'play' and lap araly,i 'speak', on the other hand,
the marking of the retired subject depends upon the overt expression
of a direct object. If the initial direct object is overtly expressed, as in
( I I), the retired subject is marked with mier. If the retired direct object
is not expressed, as in ( 1 2), the retired subject is marked with the
genitive case.
( I I ) (a) givis mier pexburtis tamasi. . .
Givi by soCCer-GEN playing
'Givi's playing soccer . . . '
(b) bavSvis mier sisulelis lap araly,i. . .
child by nonsense-GEN speaking
'The child's talking nonsense . . . '
( 1 2) (a) givis tamaSi. . .
Givi-GEN
'Givi's playing. . . '
(b) bavSvis lap araly,i inglisurad. . .
child-GEN English-ADV
'The child's speaking English . . .'
The marking noted above is not optional. (*13) shows that the
retired subject of da(:era may not be marked with the genitive instead
of the postposition mier. (*13) differs from (IOa) only in this respect.
( 1 3) *vanos da(:era . . .
VanO-GEN
('Vano's writing . . . ')
Similarly, (* 14) shows that the retired subject of tamasi may not
optionally be marked with mier when there is no expressed initial
direct object.
( 14) *givis mier tamaSi. . .
Givi by
('Givi's playing . . . ')
These sentences show that da(:era, but not tamasi, is treated as a transi
tive verb by the rules of Retired Term Marking, regardless of whether an
186 Transitivity
initial direct object is overtly expressed. These facts can be handled in

a simple way according to the proposal made in §2 above.
4 Conclusion and extension
In §2 and 3 I have argued that some verbs have initial direct objects
obligatorily and some optionally. Verbs like qiqini 'croak' and xitxiti
'giggle' never occur with a direct object, and there is no reason to
believe that they ever permit one.
It is important to observe that da{:ers 'he will write it' is typical of
Class I verbs in that it takes an obligatory initial direct object. Tamasobs
'he plays' and qiqinebs 'he croaks' are typical of Class 3 in that they take
an initial direct object optionally or not at all. It is not the case, however,
that all Class I verb forms take an obligatory direct object. Nor is it
the case that all Class 3 verb forms take a direct object optionally or
not at all. Nevertheless, this characteristic is widespread enough for us
to state the following generalization : Most Class I verb forms take an
obligatory initial direct object ; most Class 3 verb forms take an initial
direct object optionally or not at all.
5 Transitive and intransitive
In this work, a transitive verb is understood to be one which has a

subject and a direct object ; an intransitive is one which does not satisfy
those criteria. The notions 'subject' and 'direct object' in Georgian are
clarified in chs. 1-8.
The transitivity of a verb form must be stated with respect to a
particular level of derivation. Certain verb forms ( or clauses) may be
transitive initially but not finally. For example, bavsvi da�benilia 3aylis
mier 'the child is bitten by a dog', is a final intransitive (subject, but no
direct object) derived by Passivization from an initial transitive (cf.
ch. 7). Similarly, givis uqvars nino 'Givi (DAT) loves Nino (NOM) ' is a
final intransitive (subject, but no direct object) derived by Inversion
and Unaccusative from an initially transitive structure (cf. ch. 8) .
Logically, a clause could be intransitive initially and transitive finally.
However, I know of no rules in Georgian that create the grammatical
relations that would be an example of such a possibility.
As shown in this chapter, the verb da{:era always has an initial direct
object. The verb tamasi, on the other hand, is transitive initially (and
Appendix : Real and apparent exceptions 187
finally) in tamasobs burts 'he is playing ball', but intransitive (initially

and finally) in tamasobs 'he is playing'.
Appendix : Real and apparent exceptions to transitivity
This appendix gives examples of verbs that are real or apparent exceptions
to the generalization that Class 1 verbs are transitive and Class 3 intransitive.
§I deals with intransitive and apparently intransitive Class 1 verbs ; §2 dis
cusses transitive Class 3 verbs.
A. SPECIAL CLASS I VERBS
One sub-class of Class 1 verbs never takes a direct object ; this group includes
sei{q#a 'he peeked', daaboqina 'he burped', daaxvela 'he coughed', and
gakusla 'he hurried away'.
Another small group of verbs take a direct object optionally. When they
have a direct object they are Class 1 verbs ; when they lack a direct object,
they are irregularly members of Class 2. For example, seagina 'he cursed him'
(Class I), but seigina 'he cursed' (Class 2) ; �bens 'he bites it' (Class I), i�bineba
'he bites' (Class 2).
A handful of verbs occur with a direct or indirect object, but not with
both. For example, the case marking in the sentences below shows that for
this verb, persons are indirect objects, while inanimates are direct objects. 6
(I) (a) *vano gavige.
VanO-NOM I-heard-him-II-I
(b) vanos (*vano) gavuge.
VanO-DAT I-heard-him-it-II-I
'I heard, understood Vano.'
(2) vanos ambavi /*ambavs gavige.
VanO-GEN news-NoM/news-DAT I-heard-it-II-I
'I heard, understood Vano's news.'
A verb with similar characteristics is dare�va 'ring' , although for this verb
the indirect object is derived by Locative Version. 7 For both of these verbs,
additional examples show that the inanimate behaves as direct object and
the animate as indirect object with respect to Passivization (3-4), Inversion
(5-6), Retired Term Marking (7-8) and Causative Clause Union (9).
(3) zari dare�ilia.
bell-NoM rung-it-is-I-2
'The bell was rung.'
188 Appendix : Real and apparent exceptions
(4) "'deda dare�ilia.

Mother-NOM
(,Mother was rung (up).')
(5) vanos turme daure�ia zari.
Vano-DAT apparently he-rang-it- III-I bell-NoM
'Vano apparently rang the bell.'
(6) vanos turme daurekia dedistvis /"'deda.
mother-for mother-NoM
'Vano apparently rang (up) his mother.'
(7) vanos mier zaris dare�a
Vano by bell-GEN ringing
'Vano's ringing the bell'
(8) vanos mier dedistvis /"'dedis dare�va
mother-for mother-GEN
'Vano's ringing (up) his mother'
(9) vanom damare�ina dedastvis.
VanO-ERG he-caused-ring-me-it-II-I mother-for
'Vano made me ring (up) Mother.'
Examples (8) and (9) show in addition that even though no direct object
occurs, the verb is treated as a transitive with respect to Retired Term
Marking and Causative Clause Union. Dare�va and gageba thus differ from
dar;era and other regular Class I verbs only in that the direct object is not
expressed when there is an indirect object.
The Class I verb secodva 'sin' is similar to dare�va and gageba except that
it never has an expressed direct object in Modern Georgian. In Old Georgian
it had both a direct and an indirect object.
Another small group of Class I verbs takes an initial direct object which
is idiosyncratically demoted to indirect object. The examples below show
that the objects of �cna 'kiss' and miba3va 'imitate' are case marked as
indirect objects.
(10) dedam a�oca svils.
mother-ERG she-kissed-him-II-I child-DAT
'The mother kissed her child. '
( I I) maimunma miba3a bavfvis mokmedebas.
monkeY-ERG he-imitated-it-II-I child-GEN action-DAT
'The monkey imitated the child's action.'
The initial direct-objecthood of these objects is supported by their behavior
with respect to Masdar Formation (12), Object Raising ( 1 3), and Inversion (14).
Appendix : Real and apparent exceptions 189
(12) advilia seni xelis miba3Va.

easy-it-is-I-2 your hand-GEN imitating-NoM
'Imitating your handwriting is easy.'
(13) seni xeli advilia misaba3avad.
your hand-NoM easy-it-is-I-2 to imitate
'Your handwriting is easy to imitate.'
(14) bavSvs ar u�ocnia deida.
child-DAT NEG he-kissed-her- I1I-1 aunt-NOM
'The child didn't kiss his aunt.'
It appears that speakers are somewhat confused about the grammatical
relations borne by the objects of the verbs of this small sub-class. First ,
passives of verbs of this class are avoided, but not considered ungrammatical.
The use of retired terms is also avoided. Second, there is more than usual
variation in grammaticality judgements among speakers, and among various
verbs of this group. This confusion on the part of speakers is not surprising
once we recognize that direct object demotion applies to the objects of only
a tiny group of verbs. Other verbs in this class are S3lia 'he overcame it' and
miaypia 'he achieved, reached it'.
B. SPECIAL CLASS 3 VERBS
One subclass of Class 3 verbs takes an obligatory initial direct object ; this
group includes ipovna 'he found it', isova 'he got it', iqida 'he bought it',
itxova 'he borrowed it', isesxa 'he borrowed it'. The initial direct objects of
these verbs behave regularly with respect to case marking, agreement,
Passivization (16), Inversion (17), Retired Term Marking (18), and all
other rules.
( 1 5) ipove satvale?
you-found-it-I1-3 glasses-NoM
'Have you found your glasses ?'
(16) cemi satvale sanagve qutsi napovni iqo.
my glasses-NoM trash can-in found it-was-I1-2
'My glasses were found in the trash can.'
(17) turme gipovnia satvale.
apparently you-found-it-I1I-3 glasses-NoM
'Apparently you have found your glasses.'
( I 8) satvalis povna
glasSeS-GEN finding
'Finding the glasses'
190 Appendix : Real and apparent exceptions
This class of verbs behaves like Class I verbs syntactically, but like Class 3
verbs :Plorphologically. As discussed in Holisky ( 198oa), these verbs are
intrinsically perfective in aspect and are characterized by the suffix -ulob in
Series 1.
Another small group of Class 3 verbs also take an obligatory initial direct
object. Examples are upasuxa 'he answered it to him' and uqvira 'he yelled
it to h im'. The examples below show that upasuxa behaves like mi(:era, and
that its initial objects are regular with respect to case marking and agree
ment ( 19), Passivization (20), Retired Term Marking (21) and (22), and
Inversion (23) and (24) ·
( 1 9) vupasuxe galt:vetili mas(:avlebels.
I - answered-him-it-II-3 lesson-NoM teacher-DAT
'1 answered the lesson to the teacher.'
(20) ga lt:vetili napasuxebia.

answered-it-is-I-2
'The lesson is answered.'
(21) (ens mier galt:vetilis pasuxi8
y OU by leSSOn-GEN answering
'Your answering the lesson'
(22) mos(:avleblistvis pasuxi
teacher-for
',A.nswering the teacher'
(23) vanos scorad upasuxia galrvetili f"'galt:vetilistvis.
Vano-DAT correctly he-answered-it-III-3 lesson-NoM lesson-for
'Vano apparently answered the lesson correctly.'
(24) vanos s(:orad upasuxia mas(:avleblistvis /""mas(:avlebeli.
teacher-for teacher-NoM
'Vano apparently answered the teacher correctly.'
This appendix contains facts relating to verbs which are exceptions or
apparent exceptions to the generalizations made in this chapter. It is import
ant to keep in mind, however, that the vast majority of Class 1 verbs behave
like dafera and mi(:era, taking an obligatory initial direct object. The vast
majority of Class 3 verbs behave either like tamaSi, taking an optional initial
direct object, or like qiqini, taking no direct object. Examples of regular
verbs are listed in ch. 1 6, Appendix A.
13 Syn thetic passives
In ch. 12 it was shown that some verbs, such as d{1fera 'write' and
daxrcoba 'drown' , have an obligatory initial direct object ; even when
the direct object is not overtly expressed, it must be posited as an initial
relation because of the meaning and because of the syntax of organic
causatives and the marking of retired terms. In this chapter it will be
shown that a subset of such verbs, while they have an initial direct
object obligatorily, take an initial subject optionally.
When verbs of this type lack an initial subject, their (obligatory)
initial direct object becomes the subject. This is illustrated in (Ib).
(I) (a) vano axrcobs rezos.
Vano-NOM he-drowns-him-I- I Rezo-DAT
'Vano is drowning Rezo.'
(b) rezo ixrcoba.
Rezo-NoM he-drowns-I-2
' Rezo is drowning.'
Verb forms like that in (I b) have traditionally been called 'passives' by
Georgianists. I have argued elsewhere (Harris, to appear b) that forms
of this type must be sharply distinguished from true passives, those of
the type discussed in ch. 7. However, because of the traditional name
and the similarities between the two, I will refer to forms like that in
( Ib) as 'synthetic passives' ; true passives, those discussed in ch. 7,
will henceforth be referred to as 'analytic passives' .
The purposes of this chapter are to distinguish between the direct
construction, the analytic passive, and the synthetic passive in Georgian
and to relate them to those notions in general linguistics.
I Identification of direct, analytic passive, and synthetic passive

constructions
The notions 'direct construction', 'analytic passive' and 'synthetic
passive' cannot be defined independently of the initial and final
192 Synthetic passives
grammatical relations that characterize them and the rules that derive
them. These are discussed in §3. In this section, the three construc
tions are illustrated and delimited.
I . 1 Examples
All three constructions are illustrated in (2-6). Additional examples of
direct forms, analytic passives, and synthetic passives are given in
ch. 16, Appendix A as group A of Class I , and groups D and E of Class 2,
respectively. In each example below, the direct (a) and (b) sentences
have undergone no rule that changes grammatical relations. In the (b)
sentences, the subject and direct object are coreferential; in the (a) sen
tences they are not. The (c) sentences are synthetic passives, and the
(d) sentences analytic passives.
(2) (a) vano axrlobs rezos.

VanO-NOM he-drowns-him-I-I ReZO-DAT
'Vano is drowning Rezo.'
(b) vano ixrlobs tavs.
he-drowns-self-him-I-I self-DAT
'Vano is drowning himself.'
(c) rezo ixrloba.
Rezo-NOM he-drowns-I-2
' Rezo is drowning.'
(d) rezo damxrlvalia vanos mier.
drowned-he-is-I-2 Vano by
' Rezo is drowned by Vano./Rezo has been drowned by Vano.'
(3) (a) vanom gamozarda 3ma.

VanO-ERG he-raised-him-II-I brother-NOM
'Vano brought up his brother.'
(b) vanom gamozarda (tavid) tavi
self's self-NOM
'Vano brought himself up.'
(c) vano gamoizarda.
Vano-NoM he-raised-II-2
'Vano grew up.'
(d) 3ma gamozrdilia vanos mier.
brother-NOM raised-he-is-I-2 Vano by
'Vano'sj brother was brought up by him!.'
I3.I Identification 193
(4) (a) p atroni atbobs otaxs.

landlord-NoM he-warms-it-I- r room-DAT
'The landlord heats the room.'
( c) otaxi tbeba.
room-NOM it-warms-I-2
'The room is warming up.'
(d) otaxi gamtbaria p atronis mier
heated-it-is-I-2 landlord by
'The room is heated by the landlord.'
(5) (a) bitma {ur{eli gaasro.
bOY-ERG crockery-NoM he-dried-it-II-r
'The boy dried the crockery.'
(c) {urteli gasra.
it-dried-II-2
'The crockery dried.'
(d) tur{eli gamsralia bi{is mier.
dried-it-is-I-2 boy by
'The crockery is dried/has been dried by the boy.'
(6) (a) deda xatap urs acxobs.
mother-NOM cheese-bread-DAT she-bakes-it-I-r
'Mother is baking cheese-bread.'
(c) xa{apuri cxveba.
cheese-bread-NoM it-bakes-I-z
'The cheese-bread is baking.'
(d) xatap uri gamomcxvaria dedis mier.
baked-it-is-I-z mother by
'The cheese-bread is baked/has been baked by mother.'
The English glosses have been selected to correspond precisely to the
Georgian. (za) would be interpreted as 'Vano is drowning someone
else', even if the direct object were unspecified ; this form of the verb
is used only in this way. (zb) contains a verb form used only where the
subject and object are coreferential. Here Tav-Reflexivization, Possess
ive Version, and Coreferential Version Object Deletion have applied. 1
(zb) means unambiguously that Vano is committing suicide by drown
ing ; it cannot be used to describe an accidental drowning. (zc) is used
when the action is accidental. (zd) states that Rezo has drowned through
Vano's direct intervention. Parallel meanings are associated with the
sentences in (3-6). (3b) is normally used only if Vano has no family.
1.2 Morphological identification

In Georgian, the forms to which the labels 'direct', 'analytic passive',
and 'synthetic passive' apply can be identified at least in part by
morphological criteria. Analytic passives are formed from a past passive
participle (cf. Chikobava 1950 ; Shanidze 1973 ; Tschenkeli 1958 ; and
Vogt 1971) and the auxiliary qopna 'be' ; the form of these is discussed
in somewhat more detail in ch. 7, in the appendix to this chapter, and
in Harris (to appear b).
Synthetic p assives are a subset of the forms characterized in the
present tense by one of the circumfixes i-ebi, e-ebi, or 0-ebi : the
use of these different circumfixes is discussed in Chikobava ( 19 50),
Shanidze (1973), TschenkeIi (1958), Vogt (1971), and in many other
works. Here I consider as synthetic passives only those forms which are
related to direct forms, in the sense that they are formed from the same
abstract verb. Verb forms which share the morphological character
istics of the synthetic passive, but which are not related to direct forms
in this sense are discussed in ch. 16. These include forms like ikneba
'he will be', xdeba 'he becomes', and vardeba 'he is falling'. Synthetic
passives must also be separated from certain verb forms that share their
morphology, but which do not share all of the syntactic�semantic
characteristics which are discussed below. 2 Thus, the claims made in
this chapter for synthetic passives are limited to the examples given
here and the additional forms listed as group E of Class 2 in ch. 1 6,
Appendix A.
'Direct construction' is used here only to contrast with 'analytic
passive' and with 'synthetic passive', and it relates only to verbs which
make such a distinction. Verbs like qopna 'be' and mulaoba 'work' do
not participate in these distinctions. In this sense, 'direct' designates
only constructions which are initially and finally transitive. 3
1 . 3 Identification in general linguistics

The three constructions discussed here correspond to voice distinc
tions. 4 The direct construction corresponds to the active, and the
analytic passive to the passive ; the synthetic passive partially corre
sponds to the middle voice. The traditional names are eschewed here
for several reasons. First, the labels 'middle' and 'passive' have been
used in traditional Georgian linguistics in a way that does not cor
respond to their use in general linguistics (cf. ch. 1 6, §1 and Harris (to
appear b)). To ap p ly the same labels differently here would therefore
I3.2 Some characteristics 1 95
be confusing to Georgianists. Second, the label 'active' has been used

to refer to many different things in linguistics, and it is used for a
different purpose in ch. 16. Third, the synthetic passive as defined here
corresponds to only a sub-class of the middle voice of Indo-European
(cf. Benveniste 1950 ; Gonda 1960 ; Kurylowicz 1 964 ; and many other
works). 5
'Direct form', 'analytic passive', and 'synthetic passive' are used here
just for different forms of a single abstract verb (cf. ch. 12, n. I). Verbs
which have no voice distinctions cannot be said to belong to any one of
these three types.
2 Some syntactic-semantic characteristics
In this section I will establish some syntactic and semantic distinctions

between the direct, analytic passive, and synthetic passive constructions.
2 . 1 Agent is final subject of direct form

In a recent paper, Holisky shows that certain universal characteristics
of agentive subjects are associated with specific syntactic properties in
Georgian (Holisky 1978, section III). For example, adverbs expressing
the notion 'intentionally, deliberately' occur with sentences where the
semantic agent is final subject. Similarly, the command imperative
may be used only in sentences where the semantic agent is the final
subject. These properties serve to show one syntactic difference be
tween the direct form on the one hand, and the analytic and synthetic
passives on the other ; they show at the same time that only in the direct
form is the agent the final subject. (7) shows that direct forms, but not
analytic passives or synthetic p assives, co-occur with the adverb ganzrax
'intentionally'. 6 (8) shows that the direct construction, but not the
analytic or synthetic passive, may be in the imperative.
(7) (a) vanom ganzrax daaxrco rezo.
VanO-ERG intentionally he-drowned-him-II-I Rezo-NoM
'Vano drowned Rezo on purpose.'
(b) vanom ganzrax daixrco tavi.
he-drowned-self-him-II-I self-NOM
' Vano drowned himself on purpose.'
(c) *rezo ganzrax daixrco.
he-drowned-I I-2
(,Rezo drowned on purpose.')
(d) ?*rezo ganzrax damxrcvali iqo.

drowned-he-is-II-z
(,Rezo was drowned on purpose.')
(8) (a) vano! rezo daaxrCve!

Vano-vOc ReZO-NOM you-drown-him-IMP-II-I
'Vano, drown Rezo !'
(b) vano! tavi daixrcve!
self-NoM you-drown-self-him-IMP-II-I
'Vano, drown yourself !'
(c) *rezo! ixrcve!
Rezo-voc you-drown-IMP-II-z
(,Rezo, drown !')
(d) *rezo! damxrevali iqavi!
drowned you-be-IMP-II-z
('Rezo, be drowned !')
(7) and (8) show two of the ways that the direct construction differs
syntactically from the analytic and synthetic passives. These differences
are due to the single fact that only in the direct construction is the final
subject an agent.
z.z Passive has an initial subject

One syntactic property that distinguishes analytic from synthetic pass
ives is that the former, but not the latter, co-occur freely with a mier
nominal. This difference can be seen by comparing (*9) with (z-6d).
(9) (a) *rezo daixrco vanos mier

Rezo-NoM he-drowned-II-z Vano by
(*'Rezo drowned by Vano.')
(b) *3ma gamoizarda vanos mier.
brother-NoM he-raised-II-z Vano by
(*'Vano'si brother grew up by himi.')
(c) *otaxi tbeba fatranis mier.
room-NOM it-warms-I-z landlord by
(""'The room is warming up by the landlord.')
(d) *turteli gaSra bi(is mier.
crockerY-NoM it-dried-II-z boy by
(*'The crockery dried by the boy.')
I3.3 A proposal 197
(e) *xa(apuri cxveba dedis mier.

cheese-bread-NoM it-bakes-I-2 mother by
(*'The cheese-bread bakes by mother.')
The sentences of (2-6c) are identical to those of (*9), except that the
former lack the mier-nominal. The grammaticality of (2-6c) shows that
the ungrammaticality of (*9) is caused by the presence of the mier
nominal.
The sentences of (2-6d) could, of course, occur without an expressed
mier-nominal ; in that case the nominal is unspecified (cf. ch. I , n. 14) .
The sentences cited above show that the analytic passive differs from
the synthetic passive in that the former may have a mier-nominal.
In this section I have pointed out some of the ways in which these
three constructions differ semantically and syntactically.
3 A proposal that accounts for the difi'erences
Analyses of direct constructions and analytic passives have been dis

cussed in ch. 7 and other previous chapters. I propose that synthetic
passives are characterized by having an initial direct object and no
initial subject. ( Some may also take an initial indirect object.) The
initial grammatical relations of all three types are compared in ( 1 0) ,
where indirect objects are not considered.
( 10) Initial grammatical relations

(a) Direct (b) Analytic ( c) Synthetic
passive passive
rr\ «(\ ('\

daxrc- vano
' drown '
rezo daxrc- vano rezo daxrc- rezo
(Any of the nominals may, of course, be unspecified, in any of the

forms.) I propose further that synthetic passives are characterized by a
final subject. Initial and final levels are represented in ( I I ) .
( I I ) Initial andfinal grammatical relations
(a) Direct (b) Analytic (c) Synthetic
passIve passive
11\
daxre- 'vano rezo daxrc- 'vano rezo daxrl- rezo
Analytie passives are derived by Passivization ; in eh. 7 it was shown

that this rule may apply optionally when there are a subjeet and a
direct object at a single level of derivation. Synthetic passives do not
have these characteristics. First, ( I O- I I C) does not meet the input con
ditions for Passivization, since there is no initial subject. Second, the
rule which makes the initial direct object of a synthetic passive its sub
ject applies obligatorily wherever there exists a direct object and no
subject. That is, there are no grammatical sentences in Georgian in
which there is a final direct object and no final subject. The rule which
derives synthetic passives is not Passivization, but Unaccusative, the
rule which also applies in the derivation of transitive inversion verbs
(cf. ch. 8 , §6). Unaccusative, it will be remembered, is an obligatory
rule which applies in the absence of a subject to promote a direct object
to subjecthood. This proposed analysis of Georgian synthetic passives
may be related to a general phenomenon in languages, referred to by
Postal and Perlmutter as the 'Unaccusative Hypothesis' (cf. Perlmutter
1978 and Perlmutter & Postal (to appear a)).
This analysis accounts automatically for the characteristics observed
in §2 above. Holisky's ( 1 978) examples show that ganzrax 'intentionally'
may occur with sentences in which the agent is the final subject. Accord
ing to the present analysis, synthetic passives have no semantic agent
(initial subject), while in analytic passives the agent is finally a subject
chomeur.7 This analysis thus accounts for the fact that (7a -b) are
grammatical, while (*7c-d) are ungrammatical. Similarly, Holisky
shows that imperatives may occur only if the agent is final subject.
According to the analyses presented in ( I I ) , this condition is met only
in (a), the direct form. This accounts for the grammaticality judgements
indicated in (8).
In ch. I I it was shown that mier marks retired subjects of transitive
I34 Initial grammatical relations 199
verbs. According to ( 1 1), the analytic passive, but not the synthetic
passive, has such a nominal. This accounts for the fact that analytic
passives occur with a mier-nominal, as in (2�6d), while synthetic passives
do not, as shown in (*9).
Additional arguments to support the initial and final grammatical
relations proposed in ( 1 I) are presented in §4 and 5 , respectively.
4 Arguments for the initial grammatical relations proposed for

synthetic passives
The arguments given in this section support the proposed initial gram
matical relation, direct object. They argue against an analysis on which
rezo in (2C) is subject at all levels of derivation. s
4. 1 Suppletion
Shanidze observes that the verb 'kill' is suppletive for the number of
its direct object, such that mo/f-vla is used for the singular and daxoca
for the plural (Shanidze 1973 : 504). He cites the examples ( 1 2) ; the
glosses are added.
( I 2) (a) mgeli mov/f-ali.
wolf-NOM I-killed-it-II- I
'I killed the wolf.'
(b) mglebi davxoce.
wolveS-NOM I-killed-it-II- I 9
'I killed the wolves.'
From his exposition we can infer that this alternation is obligatory.
In the corresponding synthetic passive, the same suppletion 1S
governed by the nominative-nominal. Shanidze cites these examples :
(13) (a) sma momi/f-vda . 1 0
brother-NOM he-died-me-II-2
'My brother died (on me).'
(b) smebi damexoca.
brothers-NoM he-died-me-II-2
'My brothers died (on me).'
If the nominative-nominal of the synthetic passive is its initial direct
object, and if suppletion is stated on initial termhood as proposed in
previous chapters, we can account for the suppletion in ( 1 3) with the

same rule independently required for ( 1 2). This argues against an
analysis on which the nominative-nominal is subject at all levels,
requiring distinct rules to account for ( 1 2) and ( 1 3).
4.2 Preverb alternation

Schmidt ( 1957) observes that one of the uses of the preverb da- in
Georgian is to indicate plurality of the 'goal'.l1 He cites a large number
of verbs which regularly use the preverb ga- or some other preverb
with singular 'goals' and da- with plural 'goals'. He shows that for
transitive verbs of this type, the use of da- is triggered by a plural
(direct) object ; while for intransitive forms derived from the same root,
the same preverb is triggered by a plural subject. The forms below are
from Tschenkeli (1960-73) ; the glosses are added.
(14) Direct Form Synthetic Passive

Singular gamovzrdi (mas) (is) gamoizrdeba
'I will raise him' ' He will grow up'
Plural davzrdi (mat) (isini) daizrdebian
'I will raise them' 'They will grow up'
We must add a rule to the grammar to account for the use of the preverb
da- with plural direct objects. The same rule will account for the use
of da- with plural subjects of intransitives if (i) final subjects (nomina
tive-nominals) of synthetic passives are initial direct objects, and (ii)
the da- rule is stated on initial termhood. If the nominative-nominal of
the synthetic passive were not its initial direct object, two syntactic
rules would have to be stated, missing the relevant generalization.
4.3 Restrictions on the inventory of initial terms

In ch. 1 2 it was shown that certain verbs, such as daxrCoba 'drown' and
gaSroba 'dry', have an obligatory direct object in their initial term inven
tory. In the synthetic passive, only the nominative-nominal is obligatory,
as shown by (2-6c). On the analysis proposed in §3, rezo is the initial
direct object in (2C). The sentence therefore satisfies the requirement
that verbs of this type have an initial direct object. An analysis that
recognized only the subjecthood of this nominal would render the
synthetic passive an exception to the generalization that this verb
requires an initial direct object.
I3-5 Final grammatical relations 201
5 Arguments for the final grammatical relations proposed for

synthetic passives
5 . I Person Agreement
The markers of Subject Person Agreement, Direct Object Person
Agreement, and Indirect Object Person Agreement were introduced
in ch. I . In subsequent chapters it was shown that Person Agreement is
stated on final termhood.
The nominative-nominal of the synthetic passive triggers Subject
Person Agreement, as shown in ( 1 5), where the subject markers are
in heavy type.
( 1 5 ) (a) sakartvelosi davibade.
Georgia-in I-born-II-2
'I was born in Georgia.'
(b) sakartvelo!i daibade.
you-born-II-2 (zero marker)
'You were born in Georgia.'
(c) sakartvelosi daibada.
he-born-II-2
'He was born in Georgia.'
The nominative-nominal triggers Subject Person Agreement, not Object
Person Agreement. This supports the final grammatical relations pro
posed above ,
5.2 Case marking

Synthetic passives can be identified as Class 2 verbs by the morpho
logical criteria stated in ch. 1 6, Appendix A. The following facts of
case marking, among others, have been established in the preceding
chapters :
(i) Class 2 verbs govern Pattern B in all Series (chs. I , 2, and 8).
(ii) Pattern B marks the subject with the nominative case.
(iii) Case marking is sensitive to final termhood (chs. 4-8, 19).
From these facts it follows that the nominative-nominal of the synthetic
passive is the final subject, as proposed above.

Final first and second person direct objects obligatorily undergo Object
Camouflage in the presence of an indirect object (ch. 3). Object Camou

flage does not apply with synthetic passives, as shown in ( 1 6).
( 16) (a) daebade (sen) f�vian kals.
you-born-her-II-2 YOU-NOM clever woman-DAT
'You were born to a clever woman.'
(b) *seni tavi daebada {�vian kals.
your self-NoM he-born-her-II-2
('You were born to a clever woman.')
( 17) (a) avejit savse otaxsi dave�arge mSoblebs.
furniture-INsT full room-in I-lost-him-II- 1 parents-DAT
'In the room full of furniture, I got lost to my parents.'
(b) *avejit savse otaxsi lemi tavi daelJarga msoblebs.
my self-NOM he-lost-him-II-I
('In the room full of furniture, I got lost to my parents.')
If 'you' were the final direct object in ( 1 6), being a second person
nominal in the presence of the indirect object (IJvian kals 'clever woman',
it would obligatorily undergo Object Camouflage. ( 1 6) shows that the
synthetic passive must not undergo this rule. This argues against an
analysis on which the nominative-nominal is the direct object at all
levels.
In this section, it has been shown that the nominative-nominal has
the characteristics of a final subject with respect to three rules stated on
final termhood Person Agreement, Case Marking, and Object Camou
flage. Most of the syntactic rules discussed in earlier chapters do not
interact with this construction.12
6 Conclusion an d extension
Sentences like (2-6c), which may be called 'synthetic passives', are

characterized by an initial direct object which is the final subject. The
synthetic passive, which differs from the analytic (or true) passive
morphologically, also differs from it with respect to initial inventory of
terms and deriving rule. The analytic passive is derived by Passivi
zation, the synthetic by Unaccusative, the same rule which applies in
the derivation of transitive inversion forms.
I n ch. 1 2 it was established that some verbs, such as dafera 'write'
and daxrloba 'drown', take an obligatory initial direct object. In this
chapter we have seen that some of these same verbs, such as those in
Appendix: On four 'passives' 203
(2-6), take an initial subject optionally. From the arguments presented

in these two chapters, we can conclude that verbs like those illustrated
in (2-6) have an obligatory initial direct object and an optional initial
subject. Direct forms of these verbs are all Class I verbs and are typical
of that Class. It is not the case, however, that all Class I verbs are direct
forms, as defined here. Synthetic passives are all Class 2 forms, though
the converse is not true (cf. n. 2).
Appendix: On four 'passives' in Georgian
In addition to the synthetic and analytic passives, two other constructions in

Georgian have bcen called 'passive'. These are illustrated below.
(I) ikna-Passive
saxli ikna seyebili �itlad.
house-NOM it-was-II-2 painted red
'The house got painted red.'
(2) Zustandspassiv
m/veri aparia iatakze.
dust-NOM it-spreads-on-I-2 floor-on
'Dust is spread on the floor.'
The purpose of this appendix is briefly to compare the characteristics of

these four constructions, so that the reader is more fully aware of the sys
tematic nature of the verbal complex of the language. The claims made
cannot be fully justified here.
The four 'passives' may be characterized as in table 1 .
Table I
A B C D
Analytic Analytic Synthetic
(iqo-) Passive (ikna-) Passive Zustandspassiv Passive
Initial I & 2 Initial I & 2 Initial 2 Initial 2
Stative Non-stative Stative Non-stative
The analytic iqo-passive (discussed in ch. 7 and in this chapter) and the
analytic ikna-passive have the same initial inventories of terms ; both are
derived by Passivization.13 The Zustandspassiv and synthetic passive, on the
other hand, have initial direct objects, but no initial subjects ; they are derived
by Unaccusative.
The two analytic passives are alike formally, except in the auxiliary they
use. The A-type above uses the verb qopna 'be', while the B-type uses the
204 Appendix: On four 'passives'
verb 'become'. The latter is generally used only in the aorist. The ikna
passive has not been discussed at length here because it is used only in the
written language.
The iqo-passive differs from the ikna-passive in that the former is stative
and the latter non-stative. This is shown by the fact that the ikna-, but not
the iqo-passive, can co-occur with adverbs like sam saatSi 'in three hours',
which limit the duration of the action, as illustrated in (3).14
(3) (a) saxli sam saatli ikna Ieyebili.
house-NOM three hour-in it-was-II-z painted
'The house was painted in three hours.'
(b) *saxli sam saatSi iqo seyebili.
it-was-I I-z
('The house was painted in three hours.')
Note that saxli iqo seyebili 'the house was painted' is grammatical without
the limiting phrase sam saatIi.
The synthetic passive and the Zustandspassiv also differ from one another
in stativity, as stated above.
Finally, the analytic (iqo- and ikna-) passive is necessarily perfective. It
cannot be used to express the dynamic and necessarily imperfective passive of
English, such as the house is being painted. The participle, which constitutes
a part of the analytic passive, must always occur with a preverb, which is in
general an indication of perfectivity in Georgian. The synthetic passive, on
the other hand, may be either perfective or imperfective ; that is, it has forms
both with and without a preverb.
14 Rejlexivization
In this chapter I will show that Tav-Reflexivization in Georgian can

be stated simply and in such a way that there are few exceptions or
problems. The rule, first stated in ch. 1 and refined in later chapters,
will here be stated on the basis of grammatical relations and levels of
derivation, but without extrinsic rule ordering.
I A review of the facts
The pronoun tav- must be coreferential with the subject of its clause
(cf. ch. I). The Tav-Reflexivization rule is stated on initial terms ; that
is, the rule can be triggered only by initial subjects (cf. ch. 5, §z.z ;
ch. 7, §I. 1.3 ; and ch. 8, §3.3). But some facts have been left unaccounted
for. In object-raised sentences, there cannot be coreference between
the derived subject and the matrix tvis-nominal. (*la, b) show that
coreference is not possible between these two nominals if either is
reflexivized ; (*IC, d) show that co reference is also impossible if neither
nominal is reflexivized. (*IC, d) are grammatical on the non-coreferent
reading.
(I ) (a) *( tavisi) tavi ;;nelia gelastvis gasatavisupleblad.
self's self-NOM hard-it-is-I-z Gela-for to-free
cf. *'Himself is hard for Gela to set free.'
(b) *gela ;;nelia tavistvis gasatavisupleblad.
Gela-NoM self-for
cf. *'Gela is hard for himself to set free.'
(c) *is1 ;;nelia gelastvisi gasatavisupleblad.
he-NOM
cf. *'Het is hard for Gelat to set free.'
(d) * Gela t ;;nelia mistvis! gasatavisupleblad.
him-for
cf. ·'Gelal is hard for him! to set free.'
206 Rejlexivization
The grammaticality of (2), where the derived subject and the matrix
tvi>-nominal are not coreferent, shows that the ungrammaticality of (*1)
must be due to coreferentiality.
(2) merabi 3nelia gelastvis gasatavisupleblad.

Merab-NOM
'Merab is hard for Gela to set free.'
In analytic passives, there cannot be coreference between the derived

subject and the mier-nominal initial subject). In (*3a, b), one of
the coreferent nominals is reflexivized ; in (*3c, d), neither is.
(3) (a) *vano iqo mo,!!luli tavis mier

Vano-NOM he-was-II-2 killed self by
cf. *'Vano was killed by himself.'
(b) *tavi iqo mo,!!luli vanos mier.
self-NoM Vano by
cf. *'Himself was killed by Vano.'
(c) *vanoi iqo mo,!!luli misi mier.
him
cf. *'Vanoi was killed by himi.'
(d) *is1 iqo mo,!!luli vanOSi mier.
he-NoM
cf. *'Hel was killed by Vano t.'
But the active sentence (4) is fully grammatical, even though there is
coreference.
(4) vanom moi,!!la tavi.

VanO-ERG he-killed-him-II-I self-NoM
'Vano committed suicide.'
(5) shows that there is no constraint against passivizing this verb.
(5) datvi iqo mo,!!luli vanos mier.

bear-NoM it-was-II-2 killed Vano by.
'The bear was killed by Vano.'
Therefore, the ungrammaticality of (*3) must be the result of the

coreference between these two nominals in a passive.
The mier-nominal of a passive cannot trigger Tav-Reflexivization, as
(*6) shows.1
I4.I A reviefJJ of the facts 207
(6) (a) *es pgnt naqidia ninos mier tavistvis.

this book-NOM bought-it-is-I-z Nino by self-for
(,This book was bought by Nino for herself.')
(b) ""'es figni naqidia tavistvis ninos mier.
('This book was bought by Nino for herself.')
In (""'6) the initial subject is coreferential with tav-. Yet in the corre
sponding direct construction, (7), the initial subject can trigger Tav
Reflexivization.
( 7) ninom tavistvis iqida es figni.

Nino-ERG she-bought-it-II-3
'Nino bought this book for herself.'
Example (8) shows that this verb is compatible with Passivization.
(8) es figni naqidia ninos mier gelastvis.

Gela-for
'This book was bought by Nino for Gela.'
The ungrammaticality of (*6) must therefore arise from the mier

nominal being the antecedent of tav-.
The derived subject is also unable to trigger Tav-Reflexivization in
passives, as established by the ungrammaticality of (*9).
(9) ""'vano daxatulia (cnobili mxatvris mier) tavistvis.

VanO-NoM painted-he-is-I-z famous painter by self-for
(,Vano was painted (by a famous painter) for himself.')
(""'9) has no grammatical reading, whether or not the mier-nominal is

included. Yet without tav-, the sentence would be grammatical.
( 1 0) vano daxatulia (cnobili mxatvris mier).

'Vano was painted (by a famous painter).'
Thus, the ungrammaticality of (""'9) must be due to the fact that the
derived subject triggers Tav-Reflexivization.
In ch. 6 we saw that either the initial embedded subject or the initial
matrix subject can trigger tav- in an organic causative. This accounts for
the ambiguity of ( I I), where vano and ekimi 'doctor' are the initial
embedded subject (= final direct object) and initial matrix subject,
respectively.
208 Reftexivization
( I I) ekimma alaparalf.a vano tavis tavze.

doctor-ERG he-caused-talk-him-II-I Vano-NOM self's self-on
'The doctorl got Vanoj to talk about himself it j .'
In ch. 8 we saw that under Inversion only the initial subject, not the
final subject, can trigger Tav-Reflexivization, as in (12-1 3). pz) illu
strates inversion in Series III, (13) inversion with an inversion verb. In
the (a) sentences the reflexive pronoun is coreferent to the initial subject
final indirect object). In the (*b) and (*c) sentences it is coreferential
with the final subject initial direct object).
lIZ) (a) turme gelas daurrmunebia tavisi tavi.
apparently Gela-oAT he-convinced-him-III- I self's self-NOM
'Apparently Gela has convinced himself.'
(b) *turme tavis tavs daurfmunebia gela.
self's Self-OAT Gela-NoM
('Apparently Gela has convinced himself.')
(c) *turme gela daurfmunebia tavis tavs.
(13) (a) vanos uqvars tavisi tavi.
Vano-oAT he-Ioves-him-I-4 self's self-NOM
'Vano loves himself.'
(b) "'tavis tavs uqvars vano.
self's self-OAT Vano-NOM
('Vano loves himself.')
(c) *vano uqvars tavis tavs.
While I have accounted for the grammatical status of many of the
sentences in (1-13), some have been left unaccounted for. \Ve must
refine the rule of Tav-Reflexivization stated above or add constraints to
the grammar. The former course is taken in the next section.
2 A proposal
I propose the following restatement of Tav-Reflexivization to account
for all the facts enumerated above :
(14) The antecedent of tav- must be
a. the initial subject of a clause of which tavis a dependent,
and
b. a final term.
I4.2 A proposal 209
I will show below how ( 1 4-) correctly predicts the facts of Tav
Reflexivization.
In object-raised sentences, the initial embedded subject is deleted
on coreference to the matrix tvis-nominal (cf. ch. 4- and ch. I I , §4-.I ).
Since it is deleted, it is not a final term, and therefore cannot be an
antecedent of tav-.2 The derived subject cannot trigger tav-, as shown
in (""Ib), since it is not an initial subject, which (14-) requires an ante
cedent to be.
(14-) also correctly predicts the ungrammaticality of (""2a), and (""9)
where the derived subject of a passive triggers tav-. The initial subject
of a passive cannot be the antecedent, as shown in (""2b) and (""6), since
it is not a final term.
( 14-) permits both readings of (I I). The pronoun tavis an initial
dependent of the embedded clause and final dependent of the matrix
clause. Both of the nominals ekimi 'doctor' and vanQ satisfy the trigger
conditions in ( 1 4-), since both are initial subjects of clauses of which
tavis a dependent, and both are final terms subject and direct object,
respectively.
At the same time, the rule also accounts for the Inversion data. In
(I2a) and ( 1 3a) the dative-nominal is the initial subject and a final
term (indirect object). In (""ub, c) and ( "" 1 3b , c) the nominative
nominals, gela and vano, cannot trigger tav-, since they are derived
subjects, and are not initial subjects.
I have not yet accounted for the ungrammaticality of (""IC, d) and
(""3C, d). We can see from the grammaticality of (4-) that there is no
initial-structure constraint blocking (""3). The grammaticality of ( 1 5)
shows that (""IC, d) also break no constraint on initial structures ; ( I S)
is derived from the same initial structure as (""I) (cf. ch. 1 0).
(IS) tavis tavisupleba 3ne1ia gelastvis.
Self-GEN freeing-NoM hard-it-is-I-2 Gela-for
'It is hard for Gela to set himself free.'
We can therefore not account for the ungrammaticality of (""IC, d) and
(""3c, d) in terms of an initial-structure constraint. We can account for
them correctly with a convention, ( 1 6).
(16) Coreference involving an initial subject that is a clausemate of the
nominal coreferential to it must be resolved by reflexivization. 3
Examples (""IC, d ) and (""3C, d ) break this convention. The grammatical
sentences above, (4-), (7), (I I), \ I 2a), and ( 13a), do not violate ( 1 6).
210 Reflexivization
Coreference not satisfying the conditions of (16) may be resolved

with ordinary personal pronominalization, as in ( 17a). Here the
coreferential nominals are not clausemates at any level, and there IS
therefore no violation of (16).
(17) (a) vano pikrobs, rom (is) didi sp ortsmeni
Vano-NoM he-thinks-I-3 that he-NoM big athlete
ikneba.
will-be-I-2
'Vano thinks that he will be a great athlete.'
(b) ""vano pikrobs, rom (tavisi) tavi didi sportsmeni ikneba.
self's self-NoM
('Vano thinks that he will be a great athlete.')
Thus, there is no general convention that all coreference must be
resolved by reflexivization.
3 An alternative proposal
Postal ( 1971 ) proposes to account for sentences in English similar to
the ones we have been discussing in Georgian with a Crossover Con
straint. He proposes that coreferent nominals cannot cross one another
in a derivation. 4
A crossover approach makes sense only with respect to ordered
strings. If we adopt a linear order analysis, instead of the relational
analysis proposed in this work, the Crossover Constraint will correctly
predict the ungrammaticality of the object-raised sentences (""l a, b)
and the passive sentences (""3a, b). But it could not account for the
grammaticality of the inversion sentences (I 2a) and ( 13a). In order to
state that under Inversion the initial subject becomes the final indirect
object and that the direct object becomes the final subject, any gram
mar in which rules operate on structures in which constituents are
linearly ordered must have a rule in which these two nominals cross one
another in the course of derivation. The Crossover Constraint would
then predict incorrectly that ( I 2a) and ( 13a) are ungrammatical.
In order to save Crossover, we might propose to account for inversion
sentences without a crossing rule. But it would greatly complicate the
statement of the rules of Case Marking, Subject Person Agreement,
Indirect Object Person Agreement, and Object Camouflage in such a
grammar not to have these nominals cross in Inversion.
I conclude that a Crossover approach is inadequate for Georgian
because it misses the generalizations captured simply in (14) and (16).
15 Number Agreement
In this chapter I will consider the rule of Number Agreement and the
conditions under which it applies.1 It will be shown that certain facts
can be accounted for best with a rule sensitive to both initial and final
grammatical relations. The primary purpose of the chapter is to provide
an example of the relevance of initial grammatical relations to agree
ment. Until recently it has generally been assumed that all agreement
rules are stated on final grammatical relations. This chapter contributes
an example of a process in which initial grammatical relations are one
of several determining factors in agreement.
The approach will be this : I will give data and make a generalization
that describes it, then give more data and revise the generalization to
account for the additional facts. Several increments of this sort will be
necessary. It is first established that any term grammatical relation can
trigger Number Agreement in Modern Georgian. Then it is shown
that third person nominals trigger Number Agreement only if they are
subjects. Third, it is final termhood that is relevant to the trigger con
dition among third person nominals. Fourth, it is shown that among
final third person terms, that nominal which is the first subject triggers
Number Agreement. This correctly excludes initial subjects that are
not final terms, includes derived subjects of passivization, and includes
the initial subject of the inversion construction. As a last refinement, it
is shown that this condition holds unless the third person nominal is
outranked (on the relational hierarchy) by a first or second person
nominal in final structure. It will be shown that this rule accounts for
a wide variety of constructions in Modern Georgian. In �7 some
alternative analyses will be considered and shown to be inadequate.
I Polypersonalism
In Georgian, any term grammatical relation can trigger Number

212 Number Agreement
Agreement. The plural morphemes themselves are either markers of

subject plurality or of object plurality, as summarized in charts ( I),
(2) , and (3). These paradigms show an intersection of the categories
of person and number, since for fusional morphemes these cannot be
separated. (la-3) repeat ( 1 5-17) of ch. I ; ( I b) is first introduced in
ch. 8, n. 4, (ii).
( I ) Subject markers
(a) Primary
singular plural
I . person v-- v-t
2. person e -t
3 . person -sla lo -enlesl nen, etc.
(b) Secondary
I . person (v)-var (v)-vart
2. person -xar -xart
3 . person -a Is -als
(2 ) Direct object markers
singular plural
I . person m- gv-
2. person g- g-t
3. person e e
(3) Indirect object markers

I . person m- gv
2. person g g-t
3. person slh l e- s/hfe-t2
As shown in previous chapters, the primary subject markers are used
for most verb forms, the secondary for a few. The agglutinating plural
affix, -t, overlaps (1 -3 ) . But those persons (first and third) that have
some fusional morphology in their paradigms distinguish clearly between
subject and object markers ; for example, plurality in a first person
subject is marked by the agglutinating -t, but plurality in a first person
object is marked by the fusional affix gv-.
A term, if it is singular, triggers singular agreement ; if it is plural, it
triggers plural agreement. But singular agreement is realized as " in
some persons, namely where it is marked with an agglutinative affix,
e.g., in first and second person subjects. The 0-marker can also occur
for other reasons, as shown in some of the examples below. For this
IS.2 Failure of third person trigger 213
reason, examples of number agreement in this chapter will always

illustrate plural agreement, which necessarily has an overt marker. 3
The examples in (5) show that Number Agreement can be triggered
by any term relation. In (sa) the plural morpheme is triggered by the
subject, in ( Sb) by the direct object, in (sc) by the indirect object. Each
sentence in (5) corresponds to a sentence in (4), with the difference
that in (5) the appropriate term has been pluralized for contrast with
(4). Throughout, markers of agreement are set off by hyphens and
plural markers are in heavy type.
(4) (a) imulav-a.
he-worked-II-3
'He worked.'
(b) g-icnob-s.
he-knows-you-I-I
' He knows you(SG).'
(c) m-alukeb-s am figns.
he-gives-me-it-I-I this book-DAT
'He is giving me this book.'
(5) (a) imusav-es.
they-worked-II-3
'They worked.'
(b) g-icnob-t. 4
he-knows-YOU(PL)-I-I
'He knows YOU(PL).'
(c) gv-acukeb-s am figns.
he-gives-us-it- 1- I
'He is giving us this book.'
From (5) we see that any term can trigger Number Agreement.
2First refinement: failure of third person to trigger Number

Agreement
As the examples in (6) show, there are instances in which Number
Agreement fails to apply. When Number Agreement does not apply, a
singular form occurs. There are no plurality markers in the verbs of (6).
(6) (a) knob-s cems studentebs.
he-knows-it-I-I my students-DAT
'He knows my students.'
(b) alukeb-s tavis (igns lems studentebs.
he-gives-him-it-I-I self's book-DAT my students-DAT
'He is giving his book to my students.'
Although it was shown in (Sb) that direct objects trigger Number
Agreement, in (6a) the direct object fails to do so. Although it was
shown in (sc) that indirect objects trigger Number Agreement, in (6b)
an indirect object fails to trigger this rule. I) The objects that trigger
agreement in (S) are first and second person ; those that fail to trigger
agreement in (6) are third person. On the basis of this and similar data,
I conclude that
(7) (a) First and second person nominals trigger Number Agreement
in the verbs of which they are terms.
(b) Third person subjects trigger Number Agreement in the verb
which governs them, but third person objects do not.
Part (b) of this statement is supported by (sa) and (6). The first clause
of (7) is supported by (Sb, c), which show that second person direct
objects and first person indirect objects trigger Number Agreement.
(8) confirms that second person subjects trigger Number Agreement,
(9) that second person indirect objects trigger this rule. ( 10) confirms
the same for first person subjects and ( I I ) for first person direct objects.
In each example, the (a) sentence is singular, and the (b) sentence
differs from it only in having the appropriate term in the plural.
(8) (a) imulave?
you-worked-II-3 (zero marker)
'Did youlSG) work ? '
(b) imulave-t?
YOu(PL)-worked-II-3
'Did you(PL) work ?'
(9) (a) g-alukeb-s am figns.
he-gives-you-it-I-I this book-DAT
'He is giving you(SG) this book.'
(b) g-alukeb-t am figns.
he-gives-you(PL)-it-I-1
'He is giving YOU(PL) this book.'
( 10) (a) v-imuSave.
I-worked-II-3
'I worked.'
I5.3 Statement on final termhood 215
(b) v-imuJave-t.
vve-vvorked-II-3
'We vvorked.'
( I I) (a) m-icnob?
you-knovv-me-I-l
'Do you knovv me ?'
(b) gv-icnob?
you-knovv-us-I- I
'Do you knovv us ?'
3 Second refinement : statement on final termhood
In the examples above, the sentences have undergone no rule that vvould
change the grammatical relations in them. When vve look at analytic
or synthetic passives or at object-raised sentences, vve find that (7)
must be stated vvith respect to final termhood.
Analytic passives are derived by Passivization. vvhich promotes
direct objects to subjecthood ; the nominal vvhich bore the subject
relation becomes a chomeur, according to the Chomeur Condition (cf.
chs. 7 and 13). In ( 12). the verb agrees vvith the derived subject, not
vvith the initial subject.
(12) Analytic Passives

(a) kurdebi da?erili ari-an policielis mier.
thieves-NoM caught they-are-I-2 policeman by
'The thieves are caught by the policeman.'
(b) da?erili v-ar-t.
vve-are-I-2
'We are caught.'
(c) kurdi da?erili-a policielebis mier
thief-NoM caught-he-is-I-z policemen by
'The thief is caught by the policemen.'
The verb agrees in number vvith its third person final subject. The
derived subject vvas the initial direct object and, being third person,
could not trigger Number Agreement according to (7) ; only in its
derived capacity as subject could the third person nominal trigger
Number Agreement according to (7). Further, in ( 12C), the initial sub
ject, policieleb- 'policemen' , fails to trigger Number Agreement ; only
in final structure, where this nominal is a chomeur, is this fact consistent

with (7).
Synthetic passives are also derived by promoting a direct object to
subjecthood (cf. ch. 1 3). (13) shows that in synthetic passives, too, it is
the final subject that triggers Number Agreement.
(13) Synthetic passives
(a) bifebi daixrc-nen.
bOYS-NOM they-drowned-II-z
'The boys drowned.'
(b) da-�'-ixrce-t.
vve-drowned-II-z
'We drowned.'
Since the nominal in ( 13a) is third person, according to (7) it must

have been in its capacity as derived subject that it triggered Number
Agreement.
Object-raised sentences are derived by promoting the direct object
of a sentential subject to subjecthood of the matrix clause . ( 1 4) shows
that the verb agrees with the derived subject.
( 14) Object-raised sentences
( a) bavSvebi advili ari-an dasaferad.
children-NoM easy they-are-I-z to-catch.
'The children are easy to catch.'
(b) advili v-ar-t dasa(erad.
vve-are-I-z
'We are easy to catch.'
As in the other examples, third person final subjects trigger Number
Agreement. From these facts, I conclude that (7) is stated with respect
to final termhood.
(7 ' ) (a) First and second person nominals tl igger Number Agreement
in the verb of which they are final terms.
(b) Third person final subjects trigger Number Agreement, but
third person final objects do not.
4 Third refinement : first subjects
With (7') in mind, consider the sentences of ( I S), which are in the
evidential mode.
IS4 First SIlbjects 217
( I S) (a) turme studentebs gamo-u-gzavni-a-t gela.

apparently students-DAT they-sent-him-III-I Gela-NoM6
'Apparently the students (have) sent Gela.'
(b) turme gelas gamo-u-gzavni-a studentebi.
Gela-DAT he-sent-him-III- I students-NOM
'Apparently Gela (has) sent the students.'
These sentences reveal the need for a further revision of (7b'). ( 1 5)
shows that in the inversion construction, third person final subjects do
not trigger Number Agreement, but third person initial subjects do.
Yet ( 1 2-14) showed clearly that in three other constructions, third
person final subjects trigger Number Agreement, while third person
initial subjects do not. Further, it has been shown in example (6b) that
third person final indirect objects do not, in general, trigger Number
Agreement ; yet in ( 1 5) the Number Agreement trigger is a third person
final indirect object.
To understand the apparently anomalous position of ( 1 5), contrast
these sentences with (16). In (I S ) and (16) the final grammatical
relations are the same : there is a final subject, bavsvi 'child', and a final
indirect object, mSoblebs 'parents'. Yet in ( 1 5) the final indirect object
triggers Number Agreement, while in (16) the final subject does so.
(16) (a) msoblebs dae�arg-a bavsvi.
parents-DAT he-Iost-him-I1-2 child-NOM
'The child was lost to the parents./The parents' child was lost.'
(b) msoblebs dae�arg-nen bavsvebi.
they-Iost-him-I1-2 children-NoM
'The children were lost to the parents./The parents' children
were lost.'
The final indirect objects in ( 1 5) differ from those in (16) in one import
ant respect : the former are initial subjects, while the latter are not.
Inversion constructions are like object-raised sentences and passives
in that the initial direct object, if there is one, becomes subject. Inver
sion constructions differ from the others in that their initial subjects are
final terms (indirect objects). In analytic passives and object-raised
sentences, the initial subject is a chomeur, a final non-term. In syn
thetic passives, there is no initial subject. This difference between
inversion sentences and the others lead us to a further revision of {7b').
(7b") A third person nominal triggers Number Agreement if it is the
first subject that is a final term.
In sentences that undergo no change in grammatical relations, the

initial subject is the first subject that is a final term, and it triggers
Number Agreement in all persons, as stated in (7a')/(7b "). In analytic
passives, the initial subject is finally a chomeur, not a term. The first
subject that is a final term is the derived subject; (7b") correctly states
that it will trigger Number Agreement. In synthetic passives, there is
no initial subject ; the first subject that is a final term is the derived
subject. (7b ") makes the correct statements for this construction too.
Similarly, in object-raised sentences, the object which is raised to sub
ject is the first subject that is a final term, the sentential subject being a
chomeur. In inversion constructions, the initial subject is a final indirect
object, a final term; therefore it is the first subject that is a final term.
Thus (7b ") captures a generalization that accounts for all of the data
considered thus far.
(7b") is complex, but it is not possible to account for all of the sen
tences of Modern Georgian with a simple rule of Number Agreement.
The constraint on which third person nominals trigger Number Agree
ment is partially similar to the constraint on which nominals trigger
Tav-Reflexivization (cf. ch. 14).7
5 Last refinement: relational hierarchy
The sentences of (17) show the need for a further condition on the state
ment of Number Agreement. ( 17) is analogous to ( I S). In ( I S) both
nominals are third person ; in ( I7) the initial direct object ( = final
subject) is second person.
(17) (a) turme s!uden!ebs gamo-u-gzavni-xar (sen).

apparently students-DAT he-sent-you(sG)-III-I yOU(SG)-NOM
'Apparently the students (have) sent you(SG).'
(b) turme students gamo-u-gzavni-xar-t (tkven) .
student-DAT he-sent-you(PL)-III-1 YOU(PL)-NOM
'Apparently the student (has) sent you(PL).'
(17) differs from (IS ) only in the person of the final subject initial
,
direct object) and in Number Agreement. In ( 17) student(eb)s 'student(s)
does not trigger Number Agreement as it does in ( I S). In (17) the final
subject does trigger Number Agreement, while in ( I S) it did not. In
fact, if there is a first or second person term in a clause, a third person
ISo5 Relational hierarchy 219
nominal in that clause triggers Number Agreement only if it is subject

at all levels of derivation. This generalization is expressed in a different
form in ( I S), which incorporates (7a') and a further-revised version
of (7b").
( I S) (a) First and second person nominals trigger Number Agreement
in the verb of which they are final terms.
(b) A third person nominal triggers Number Agreement in the
verb of which it is a final term, if
(i) It is the first subject of that verb that is a final term,
and
(ii) It is not outranked finally by a first or second person nom
inal in its clause. 8
Here 'outranking' refers to the hierarchy of grammatical relations,
where subject ranks highest : subject-direct object-indirect object
non-terms . In ( I sa) the third person nominal student(eb)s meets both
criteria of ( I Sb). As the initial subject, it is the first subject that is a
final term. There is no first or second person nominal that could out
rank it in its clause. In (17), on the other hand, this same nominal is
outranked finally by the second person nominal, senjtkven 'you'. 8enl
tkven is the final subject and therefore higher on the relational hierarchy
than student(eb)s, the final indirect object.
Rule ( I S) accounts for the trigger of Number Agreement in all sen
tence types considered here. In a sentence which undergoes no change
in grammatical relations, an initial third person subject is also the final
subject and thus (i) is the first subject that is a final term, and (ii) is
not outranked by any nominal finally. In a passive, the initial subject
is not a final term. A third person direct object is (i) the first subject
that is a final term and, as subject, is (ii) not outranked by any nominal
finally. Thus, ( I S) makes the correct statements for the selection of the
Number Agreement triggers for each of these constructions ; the reader
can confirm that it also makes the correct statements for synthetic
passives and for object-raised sentences.
By ( I S), Number Agreement may be triggered by more than one
term, but only if each satisfies either ( I Sa) or both parts of ( ISb).
Rule ( I S) could be stated in other ways, but only a complex rule referring
to both initial and final grammatical relations can capture all of the
linguistically significant generalizations and can resolve the apparent
contradictions in the nature of Number Agreement in Georgian. II
6 The applicability of rule (18)
It was shown above that the rule proposed here as ( IS) accounts for a
wide variety of constructions, including simple sentences (clauses in
which no change of grammatical relations takes place), analytic and
synthetic passives, object-raised sentences, and inversion constructions.
In this section, I will discuss some additional constructions - causatives
and version - which completes the list of major syntactic constructions
of Modern Georgian. In each instance I will show that ( I S) correctly
accounts for Number Agreement. I conclude that ( I S) accounts for
Number Agreement generally in Modern Georgian.
6.1 Causatives
If an embedded clause is intransitive, its subject becomes the direct
object of the corresponding causative ; (19) illustrates this type.
(19) (a) ekimma gv-alap ara�-a (even).
doctor-ERG he-caused-talk-us-II-I we-NOM
'The doctor got us to talk.'
(b) ekimma alapara�-a bi{ebi.
he-caused-talk-him-II-I bOYS-NOM
'The doctor got the boys to talk.'
The initial grammatical relations in (19) are established in ch. 5.
(i) Ekimma 'doctor' is the initial subject of the matrix clause. (ii) The
predicate of this clause is 'cause'. (iii) Its direct object is the clause con
taining the verb lap ara�- 'talk'. (iv) even is the initial subject of that
clause in ( I9a), and bi?ebi 'boys' is in ( I9b).
According to (ISa), the first person final direct object in (I9a) should
trigger Number Agreement. It does, and this is reflected in the first
person plural direct object marker gv-. According to ( ISb), ekimma
' doctor' in ( I9a, b) should trigger Number Agreement, since it is the
final subject of the causative and is (i) the first subject of that verb
that is a final term, and (ii) not outranked finally by any nominal. The
number of this nominal is reflected in the -a suffix, which is a marker
of third person singular subjects. Bi?ebi 'boys' in (I9b) does not satisfy
criterion (I 8b-i) and is therefore not a Number Agreement trigger. It
does not satisfy (I8b-i) because, although it is a final term of the
causative, it is never its subject. Although it is the initial subject of
lap ara�- 'talk', it is not a final term of that verb. As predicted by ( I 8b),
the number of bi{ebi is not reflected in the verb form.
I5.6 Applicability of rule (r8) 221
The causative of a transitive is illustrated in (20).

(20) (a) mepem ga-gv-afmendin-a sasaxle (even).
king-ERG he-caused-clean-us-it-II-I palace-NOM we-DAT
'The king made us clean the palace.'
(b) mepem gaafmendin-a monebs sasaxle.
he-caused-clean-him-it-I I-I slaves-DAT
'The king made the slaves clean the palace. '
The initial matrix subject, mepem 'king', is the final subject of the
causative verb ga-afmendin- 'cause to clean'. The initial direct object of
ga-t;mend- 'clean', the nominal sasaxle 'palace', is the final direct object
of the causative verb. The initial subjects of ga-t;mend 'clean', the
nominals even 'we' and monebs 'slaves', are the final indirect objects of
the causatives.
According to (1 8a), the first person final indirect object of (2oa),
being a final term of the causative, should trigger Number Agreement.
Its plurality is indeed reflected in the first person plural indirect object
marker, gv-. According to ( I Sb). the third person final subject of (2oa, b)
should trigger Number Agreement, as in (19)' Its singular number is
reflected in the third person singular subject suffix -a. The third person
final indirect object of (2ob), monebs 'slaves', is not a qualified Number
Agreement trigger according to (ISb-i). Although it is the initial subject
of the verb ga-t;mend- 'clean', it is not a final term of this verb, rather
of the causative verb, ga-acmendin- 'cause to clean'. Although it is a
final term of the causative, it is never its subject. As predicted, its
plurality is not reflected in the verb form. The third person final
direct object, sasaxle 'palace', does not satisfy ( I Sb-i) ; its number is
not reflected in the verb form.
Thus, ( I S) accurately accounts for Number Agreement in causatives
in Modern Georgian.
6.2 Version
Benefactive Version promotes a benefactive-nominal to indirect-object
hood. Number Agreement in this construction is illustrated in (21).
(21) (a) jadosanma (tkven) ga-g-i'!J.et-a-t jadokruli
witch-ERG yOU(PL)-DAT she-made-you(PL)-it-II-I magic
sar!J.e?
mirror-NoM
'Did the witch make a magic mirror for yOU(PL) ?'
(b) jadosanma mepis �acebs ga-u-�et-a jadokruli

king-GEN men-DAT she-made-him-it-II-I
sar�e.
'The witch made a magic mirror for the king's men.'
The final indirect objects of (21), tkven 'yOU(PL)' and mepis �acebs
'king's men' , are initially non-terms.
( 1 8a) correctly predicts that the second person final indirect object
in (2Ia) will trigger Number Agreement. Its plurality is reflected in
the -t suffix in the verb form. ( 18b-i) correctly predicts that the third
person final indirect object in (2Ib) will not trigger Number Agreement.
Its person is reflected in the third person marker -u-, but its number is
not reflected in the verb form.
(18) thus makes the correct statement of Number Agreement triggers
for benefactive version, as well as for other version constructions in
Modern Georgian.

Object Camouflage effects the substitution of tav- and a possessive
pronoun for the former direct object. The possessive pronoun cannot
trigger Number Agreement, since it is not a term. This is shown in (22),
where the initial direct object (represented by the possessive pronoun)
is singular in (a) and plural in (b).
(22) (a) ga-g-acn-o cemi tavi.
he-introduced-you-him-II-I my self-NOM
'He introduced me to you(SG).'
(b) ga-g-acn-o cveni tavi.
our
'He introduced us to you(SG).'
The head of the final direct object is always singular ; that is cveni ta'vi
literally 'our self', not ""cveni tavebi 'our selves'. Since the marker of
singularity is zero in third person objects, there is no overt indication
of agreement.
6-4 Conclusion : invariance

It has been established in this section and the preceding sections that
rule (18) correctly accounts for a wide variety of constructions in
Modern Georgian. The generalization stated in (18) is invariant from
one Class to another and from one Series to another. The Series III
IS.7 Alternative analyses 223
inversion forms are discussed in §4 and 5. In saying that Number

Agreement does not vary from one Series to another, I mean that,
stated on the basis of grammatical relations, as in ( 1 8), the same gen
eralization is valid in Series I, II, and III. The fact that the corre
spondence between grammatical relations and the facts of Number
Agreement is invariant supports the proposed analysis.
7 Some alternative analyses of Number Agreement
7. 1 A case-based analysis
It might be proposed that the determination of the triggers of Number
Agreement should be stated, not on grammatical relations, but on cases.
Because of the peculiarly complex system of case marking in Modern
Georgian, a solution of this kind would be particularly complicated.
Stating Number Agreement on cases would entail incorporating the
case marking differential « 1 '-2') of ch. 9) into the Number Agreement
rules. It would require making at least the statements of (2 3 ) .
(2 3) (a) First or second person nominative-, dative-, or ergative
nominals trigger Number Agreement.
(b) Third person nominative-nominals trigger Number Agree
ment in Series I, verbs of Class I , 2, or 3 .
(c) Third person nominative-nominals trigger Number Agree
ment in Series II or III, verbs of Class 2.
(d) Third person ergative-nominals trigger Number Agreement
in Series II, verbs of Class I or 3 .
(e) Third person dative-nominals trigger Number Agreement in
Class 4 verbs in all Series and in Class I or 3 verbs in Series
III, just in case the nominative-nominal is also third person.
This analysis fails to draw any relevant generalizations. (23 ) appears
to be a random set of unrelated facts, while ( 1 8) presents general prin
ciples for determining Number Agreement triggers. Thus, ( 1 8),
although complex, is more general than a case-based analysis.
7.2 A morphology-based analysis

To account for the failure of third person objects to trigger Number
Agreement, an analysis might be proposed according to which a zero
marker was triggered by third person plural objects. In order to
compare this hypothesis with that proposed in this chapter we must con
sider some additional aspects of Georgian morphology. With the con
vention that object agreement suffixes attach to the right of subject
agreement suffixes,10 the relevant morphophonemic rules may be
stated as (24) and (25).
(24) +s+ --+ 0/ _ +t+

The morpheme s deletes before the morpheme t.
(25) + t + --+ 0![PLURAL] _
The morpheme t deletes after any plural marker.
The morpheme t in both rules is a plural marker (cf. paradigms ( 1-3 /3'» .
Rule (25) accounts for the non-occurrence of the following sequences,
where 'V' represents the verb stem : *V-en-t, ""'V-an-t, *V-es-t, ""'V-nen-t,
and ""'V-t-t. None of these sequences is ruled out by a phonological
rule.
The morphology-based analysis would require the following state
ments to account for simple sentences : (1-2), (3') to replace (3) of the
proposed grammar, and the morphophonemic rules (24-25).
(3') Indirect object markers (cf. (3»

singular plural
I . person m- gv
2. person g g-t
3 . person s/h/0- s/h/0-
To account for (IS) this analysis needs, in addition, rules (26-27).

(26) The third person plural object suffix is not 0, but rather -t, if and
only if the sentence has undergone Inversion and the final subject
is third person.
(27) The third person plural subject marker is not -en, -es, etc., but
rather -a or -s, just in case the clause has undergone Inversion and
the final object is third person.
Rules (26) and (27) are ad hoc, and this analysis fails to show that
Number Agreement is based in part on non-final grammatical relations.
The hypothesis proposed in this chapter, on the other hand, needs
statements ( 1-3), (24-25), and (18). While ( 1 8) is complex, it draws
linguistically significant generalizations and avoids ad hoc rules.
I5.7 Alternative analyses 225
7.3 A rule- or construction-based analyst's

Traditional analyses assume ( 1-3) and (24-25), but replace (18) with
statements equivalent to those in (28).
(28) I In the inversion construction:

A. All first and second person final terms trigger Number
Agreement.
B. Third person initial subjects trigger Number Agreement, if
there is no first or second person final term in the clause.
II In all other constructions :
A. All first and second person final terms trigger Number
Agreement.
B. Third person final subjects trigger Number Agreement.
(28) could instead be formulated in terms of the application of rules ;

that is, 'I If Inversion has applied, . . . II If Inversion has not applied, . . . '
It is immediately clear that (28) is missing a generalization, since
statement A is repeated in parts I and II. Aside from this, (28) fails to
relate third person initial subjects in IB to third person final subjects in lIB.
( 1 8), on the other hand, does relate these through reference to the notion,
'first subject'. Thus ( 1 8) is to be preferred to a solution which gives
different rules for inversion and non-inversion constructions.
7.4 No-Inversion Analysis

Finally, the complexity introduced by the inversion construction in §4
might be taken as an indication that the inversion analysis is incorrect.
Since dative-nominals trigger Number Agreement in ( I S), as final sub
jects do elsewhere, it might be proposed that the dative-nominal is the
final subject. ll Like the other grammars considered, this one would
need statements ( 1-3) and (24-25). In addition, it would need the
following statements to account for (15-17) :
(29) First and second person nominals trigger Number Agreement in

the verb of which they are final terms ( = I 8a).
(30) Third person nominals trigger Number Agreement in the verb of

which they are final subjects.
(3 1 ) Third person nominals do not trigger Number Agreement in the
verb of which they are final subjects if:
(a) (i) the verb is in Series III

and
the verb is in Class 1 or 3
or
(ii) the verb is in Class 4 (in any Series)
and
(b) there is a first or second person nominal in the clause.
(29-3 1) would replace (18) of the grammar espoused here. While (29)
and (30) are comparable to (1 8a) and ( I8h-i), respectively, (3 1) is a
great deal more complex than ( I 8b-ii). The complexity of (3 I), especially
of part (a), is directly attributable to the fact that this grammar lacks
Inversion.
A No-Inversion Analysis would be unable to account for the fact
that the markers of Number Agreement are selected entirely on the
basis of the final grammatical relations stated by the rules of Inversion
and Unaccusative tcf. n. 8). That is, the final indirect object s!uden!ebs
'students' in ( I sa) triggers an object plural marker -t from set (3), not
a subject plural marker from set (I). The No-Inversion Analysis outlined
above would have to have an additional ad hoc rule to account for this
fact.
In addition, the No-Inversion Analysis would require giving up the
economy in case marking effected by Inversion and would require
complications in the rules of Subject Person Agreement, Indirect
Object Person Agreement, Object Camouflage, Tav-Reflexivization,
and suppletion of various types, as set out in ch. 8. Thus, while ( 1 8)
is complex, a grammar that lacked Inversion would be still more
complex.
Unlike the alternative analyses discussed in this section, rule ( 1 8)
states a linguistically significant generalization. It avoids the entirely
ad hoc rules (23), (26-27), (28), and (3 1) required by the other hypoth
eses. While (18) is a more complex rule than is usually needed in most
languages for the statement of agreement phenomena, it presents an
accurate account of the facts in the most succinct and general way
possible.
8 Conclusions
Research on grammatical relations has revealed a hierarchy - subject,

direct object, indirect object, other grammatical relations - such that
I5.8 Conclusions 227
subject is highest. The relevance of this hierarchy has been established

in the accessibility of nominals to certain rules, including relativization
(Keenan & Comrie 1977). It is also thought that the relational hierarchy
affects verb agreement in the following way: If a language has agreement
with direct objects, it will also have agreement with at least some sub
jects. If a language has agreement with indirect objects, it will have
agreement with at least some subjects and direct objects. This chapter
establishes a further relevance of the relational hierarchy in agreement
phenomena. §5 shows that whether a particular nominal triggers agree
ment may be determined by whether it is outranked on the relational
hierarchy by certain types of nominals (in Modern Georgian, by first
or second person nominals). On the basis of our present understanding
of the relational hierarchy and facts like those presented in ( 1 5) and
( 17), we can hypothesize that the presence of a nominal bearing a gram
matical relation lower on the relational hierarchy would never be relevant
to the selection of an agreement trigger. In Modern Georgian, the
presence of a (first or second person) nominal bearing a grammatical
relation higher on the relational hierarchy is relevant, as expressed in
( 18b-ii) ; but we may hypothesize that the converse of this would never
occur in a natural language.
In §4, it is established that in Modern Georgian the determination
of the triggers of Number Agreement is based on both final and non
final grammatical relations. The determination of the triggers of Person
Agreement, on the other hand, is based on final grammatical relations
alone (cf. chs. 1-8, 10, 1 3). It is also interesting that in Old Georgian,
the determination of Number Agreement triggers was based on final
grammatical relations alone (cf. Aronson 1976 ; Chikobava 1941 ;
Shanidze 1 976 ; Zorrell 1 930). The fact that non-final grammatical
relations are relevant to agreement in Modern Georgian is of interest
to universal grammar, since it has been widely assumed that only final
termhood is relevant in these phenomena.
16 The nature of the Georgian
ver b classes
In previous chapters we have examined several aspects of the distri

bution of cases in Georgian. In chs. 2 and 8, it was established that there
are two case marking Patterns in Modern Georgian, Patterns A and B ;
in ch. 9, it was shown that Pattern A cannot synchronically be reduced
to Pattern B. I have shown that the Patterns are correlated with gram
matical relations in the following way:
( 1) (a) Distribution of cases in terms of Patterns
Subject Direct object Indirect object
The distribution of cases has been precisely stated in terms of the
Class and Series of the governing verb form :
(b) Distribution of cases in terms of Series and Class
Series I II I II
Class 1 B A B
2 B B B
3 B A B
4 B B B,
but it has not yet been determined what kinds of verbs constitute each
Class. I have not yet stated what characteristic of a verb form deter
mines which Pattern it will govern. Thus, I must still address the fol
lowing problem within the context of both general linguistics and
Georgian grammar: How can it be predicted which verbs govern Pattern A
in Series II, and which Pattern B?
In this chapter I will present in detail two solutions to this central
problem of Georgian syntax. The first hypothesis ( §I) represents the
approach taken by most Georgianists today ; they analyze the case
I6.I The Ergative Hypothesis 229
marking system of Series I I as ergative. The second hypothesis ( §2)

represents a new proposal based on the initial and final termhood of the
nominal dependents, rather than their final termhood alone. This
analysis draws on the important notion of initial intransitive direct
objects (defined below). In §3 and 4, I will discuss Inversion and the
semantics of Class, respectively. In §5, I will compare the two hypoth
eses, showing that the analysis proposed here is simpler with respect to
case marking, Inversion, and predicting the Class membership of pro
ductive derivational types introduced in §4 and in earlier chapters. §6
is a brief discussion of the theoretical contributions of these aspects of
Georgian to the study of language universals. Sample lists of each
Class of verbs are included for reference, as Appendix A to this chapter.
Appendix B deals with a few types that are irregularly ambivalent with
respect to Class membership, discussing some verbs that are exceptions
to both hypotheses.
Traditional analyses of Georgian have observed that Class 2 verbs
do not undergo Inversion in Series III ; this has been treated as an
idiosyncratic fact. The failure of Class 2 verbs to undergo Inversion
follows automatically from the analysis proposed here, with no
additional ad hoc device.
I Hypothesis A : The Ergative Hypothesis

Nearly all of the recent full-scale analyses of Georgian have claimed that the
language has ergative case marking in Series II : Chikobava ( 1 948, 1 950,
1961, 1 968), Klimov ( 1973), Nebieridze (to appear), Schmidt ( 1 973),
Shanidze (1973), Tschenkeli ( 1 958), Vogt ( 1 971), etc. A prominent
exception is the work of Aronson, who specifically argues that Georgian
is not ergative (d. especially Aronson 1 970)' The tradition of the
ergative analysis is embodied in the use of the traditional name,
'ergative', for the case that marks subjects in Pattern A.
1 . 1 What is ergative case marking?

Anderson (1976) states that 'In [ergative] languages, the morpho
logical category to which the subject NP of an intransitive verb belongs
is shared not with the NP we expect to be subject of a transitive verb,
but rather with the NP we expect to be object of that verb'. In terms of
case marking, in an ergative system, the case which marks the subject
of an intransitive verb is shared, not with the nominal which is the
230 The nature of the Georgian verb classes
subject of a transitive verb, but with the nominal which is the object
of that verb. In an ergative system of case marking, one case marks the
subject of an intransitive verb and the direct object of a transitive verb,
while another case marks the subject of a transitive verb. Similar
definitions are given in other recent discussions of general ergativity
(Catford 1976 ; Chikobava 1 961 ; Comrie 1 973, 1 978 ; Dixon 1979 ;
Johnson (to appear), Klimov 1973 ; Kurylowicz 1946 ; Sapir 1 9 1 7 ;
Silverstein 1976 ; and Troubetskoj 1939). According to all of these
definitions, an ergative system is predicated upon a division of verbs
into transitive and intransitive. The terms 'transitive' and 'intransitive'
are used here in their general linguistic meaning, which is discussed
in ch. 12, §s.
1.2 What are the implications of the Ergative Hypothesis for Georgian?
We have established the existence of two case marking Patterns in Mod
ern Georgian, as shown in ( I a). A large amount of evidence has been
amassed to support the correlations between the cases and the gram
matical relations stated in ( I a) ; this evidence is presented particularly
in chs. 2-9 and 1 3-16. In particular, these data show that nominals
marked with the ergative in Pattern A and with the nominative in
Pattern B are subjects, while those marked with the nominative in
Pattern A and with the dative in Pattern B are direct objects.
The ergative analysis of Georgian Series II identifies the 'ergative'
marked subject with transitive verbs and the 'nominative' -marked sub.
ject with intransitive verbs. According to the Ergative Hypothesis, in
Series II, transitives govern Pattern A and intransitives govern Pattern
B. We have seen the following distribution of Patterns A and B with
respect to different Classes in Series II.
(2) Distribution of Patterns in Series II
Class 1 Pattern A
Class 2 Pattern B
Class 3 Pattern A
Class 4 Pattern B
Grammarians writing within the framework of the Ergative Hypothesis
have made the following statements about these verb Classes :
(3) Class 1 is transitive and governs Pattern A.
Class 2 is intransitive and governs Pattern B .
Class 3 is neither transitive nor intransitive and governs Pattern A.l
1 . 3 Insights of this analysis

Historically, the Ergative Hypothesis was proposed for Series II of
Georgian as an attempt to distinguish it from the obviously very different
systems found in most European languages. In the context of attempts
to force all languages into the mold of the accusative construction, it was
essential to recognize that case marking in Series II of Georgian is not
accusative ; and so it was grouped with the ergative type. It was observed
that most Class 1 verbs are final transitives, most Class 2 verbs final
intransitives, and that a large number of verbs in Georgian fall therefore
into the pattern,
Subject Direct Object
Transitive ERGATIVE NOMINATIVE
Intransitive NOMINATIVE
which defines an ergative case marking system. For example, there are
a great many final transitive-intransitive pairs, like those listed in (4).
(4) Transitive Intransitive

svams 'he drinks it' ismeba 'it is drinkable'
�lavs 'he kills him' �vdeba 'he dies'
zrdis 'he raises him' izrdeba 'he grows up'
xerxavs 'he saws it' ixerxeba 'it saws easily ; it is being sawn'
lexavs 'he breaks it' lqdeba 'it breaks' (cf. ch. 1 3)
All of the transitive verbs in (4) govern Pattern A in Series II, and all
of the intransitive ones Pattern B. In addition to pairs like these, verb
forms like mivida 'he went', movida 'he came', iqo 'he was', and darla
'he stayed, he remained', are intransitive and govern Pattern B. Further,
as established in ch. 8, inverted verbs are finally intransitive, with Pat
tern B marking. Thus, the Ergative Hypothesis makes the correct
predictions for a large number of verbs in Georgian.
1 .4 Inadequacies of this analysis

A glance at (3) will show that Class 3 verbs present a problem for the
ergative analysis. Class 3 contains a few verbs that may take a direct
object, such as tamaSi 'play' and lap ara�i 'talk' (cf. ch. 1 2) ; most verbs
in this Class, however, never take a direct object: qiqini 'croak', xitxiti
'giggle'. In order to include Class 3 verbs in the system of case marking,
traditional grammarians have given some rather interesting apologiae
for this Class. In this section, those explanations will be examined.
(i) The first such explanation is embodied in (3), namely that Class 3
verbs are neither transitive nor intransitive. In the sense defined in
ch. 1 2 for these notions, this claim has no apparent meaning. In the
work of linguists who make this claim about Georgian, the terms
'transitive' and 'intransitive' have lost the meanings that they have in
general linguistics and have become the names of particular verb
Classes, regardless of the syntactic properties of the verbs. Thus, 'tran
sitive' has become the name of Class I , and 'intransitive' the name of
Class 2. This nomenclature has led to calling Class 3 verbs 'neutral'
and stating that they are neither transitive nor intransitive. Clearly this
statement is meaningless if the notions 'transitive' and 'intransitive'
are also to be used with reference to languages other than Georgian.
(ii) A second approach is to play down the importance of this large
class of counterexamples by labeling it 'irregular'. Indeed, the Class has
even been defined as the set of irregular verbs (cf. Holisky 1980a for a
history of the analyses of this Class). The implication seems to be that
if a group of verbs is morphologically irregular, it cannot be expected
to be regular from the point of view of the case Pattern it governs. The
label 'irregular' seems to have been applied on the grounds that verbs
of Class 3 build their future/aorist fonns2 with a formant different from
that used by verbs of Classes I and 2. This Class is, however, regular
in the sense that its members regularly use one particular formant to
build the tenses in question. (5) lists the formants regularly used to
form these tenses by verbs of Class I and 2 on the one hand, and of
Class 3 on the other.
(5) Classes I, 2 Class 3
Future Preverb- i-eb
Conditional Preverb-(0)d-i i-eb-(0)d-i
Future Subjunctive Preverb-(o)d-e i-eb-(0)d-e3
Aorist Preverb-e/i i-e
Optative Preverb-ofa t--o
In (5), dashes indicate the position of the verb root. The preverbs used
by Classes I and 2 are chosen from a set which includes mi-, mo-, a-,
amo-, ca-, lamo-, da-, ga-, gamo-, fa-, famo-, etc. and are lexically
determined. In the aorist and optative, respectively, most of the Class I
and 2 verb forms use the -e and -0 suffixes ; some, however, use -i and -a,
respectively. All of the Class 3 verbs on the other hand, use -e and -0
regularly. In order to illustrate the striking regularity of this Class, I
have listed the verbs in the future in Appendix A to this chapter. While
there exist irregularities in this Class as in every other, the overwhelm
ing majority of verbs in Class 3 show a consistent morphology. These
verbs have 'irregular' morphology, then, only in the sense that their
morphology is different from that of Class I and 2 verbs. 4-
The 'irregularity' of this Class has also been understood to mean
that it is arbitrary, that is, that the verbs which constitute the Class are
an arbitrary set. S Holisky (I98oa) shows that Class 3 is semantically
regular, as well, constituting the Class of dynamic verbs that are
inherently of durative aspect (cf. n. 4). Even a glance at the representa
tive list of Class 3 verbs in Appendix A will reveal that the verbs which
constitute Class 3 are not an arbitrary set.
In earlier chapters we have seen that Class 3 verbs exhibit a high
degree of syntactic regularity. In particular, all verbs which satisfy the
morphological definition of Class 3 (cf. Appendix A), also have the fol
lowing syntactic characteristics : (a) they trigger Inversion in Series III,
and (b) they govern ergative case subjects (Pattern A) in Series II.
This Class of verbs is strikingly regular, then, in its morphology,
syntax, and semantics. Labeling these verbs 'irregular' in no way
accounts for the case Pattern they govern in Series II. Labeling them
in this way simply disguises the fact that linguists have not been able
to account for the case Patterns governed by a large and systematic
Class of verbs in Georgian.
(iii) A third explanation that has been offered for the case marking
of Class 3 verbs in Series II is that these verbs ' do not have their own
Series II forms' and that the Series II forms used are 'borrowed from
the corresponding' Class I verbs. (This entails that the future forms
are also 'borrowed', since futures and aorists are similar morpho
logically for all regular verbs in Georgian ; cf. (5) above.) Thus, for
example, the aorist of vbatonob 'I reign' vibatone 'I reigned' - is said
to be 'borrowed' from the Class I transitive verb : gavibatone cemi tavi
'I made myself ruler'. Since the form is borrowed, it is implied, so is
the case Pattern. Thus, an intransitive verb, vbatonob ' I reign', has an
ergative case subject in Series II because the morphology of the verb
is 'borrowed' from a transitive, which naturally governs an ergative
subject.6 Thus, by claiming that these verbs have 'no S eries II forms
of their own' , linguists have tried to explain the fact that these intransi
tive verbs govern an ergative case subject in an ergative system of case
marking.
There are at least two problems with such an explanation. The first
is that this theory does not explain why syntactic characteristics sho,uld
be 'borrowed' along with morphological ones. In a language with an
ergative case system, case marking is determined by the transitivity of
the verbs, rather than by the origin of their morphology. If a language
has ergative case marking, it will apply the principles of that system to
all verbs alike, whether or not they are derived or borrowed. Until the
borrowing of morphological formants can be shown to be related sys
tematically to the borrowing of syntactic characteristics, this explanation
for the use of the ergative case by intransitive verbs is completely ad hoc.
A second fault of such a hypothesis is that it offers no explanation of
why some verbs (Class 3) should 'borrow' their morphology while
others (Class I , 2) do not. Even if we accept the notion that 'borrowed'
morphology entails 'borrowed' syntax, we still have no answer as to
why Class 3 is different from Class 2.7
To summarize, neither the notion 'neither transitive nor intransitive',
the label of 'irregular', nor the concept that Series II forms of Class 3
verbs are 'borrowed' offers a satisfactory explanation of the systematic
use of Pattern A by many intransitive verbs in Series II. We must
recognize that there is no way adequately to incorporate Class 3 verbs
into the ergative analysis of Georgian.
There are other problems for the claims made in (3). Intransitive
verbs in Class I include daaboqina 'he burped' , daamtknara 'he yawned',
and daacxi�va 'he sneezed'. The transitive verbs in Class 2 are dis
cussed in Appendix B to this chapter. While the existence of intransi
tives in Class I and transitives in Class 2 is contrary to the claims of (3),
it does not pose a serious problem for the Ergative Hypothesis ; these
verbs are few in number and may be treated as simple exceptions. The
existence of the large, systematic Class 3, however, shows the real
inadequacy of the ergative analysis for Georgian.
At the beginning of this chapter I stated a problem for general
linguistics and Georgian grammar, as it has traditionally been viewed
by Georgian specialists. Looking at the distribution of cases in (2), we
see that it is appropriate to pose this question in a different way as well :
Problems for general linguistics and Georgian grammar

1 . How can we capture a generalization uniting Class I with Class 3
in Series II ?
2. How can we predict which verbs govern Pattern A in Series II ?
I6.2 The Unaccusative Hypothesis 235
:a Hypothesis B: The Unaccusative Hypothesis

It will be shown that the notions 'active' and 'inactive' correlate with
the morpho-syntactic divisions of Classes, such that Classes I and 3
constitute the set of active verbs, while Classes 2 and 4 constitute the
set of inactive verbs. On this basis the case marking rules of Georgian
can be stated.
2. ILanguage universals
Sapir ( 1 9 17) was perhaps the first to make a clear typological division
of languages on the basis of case marking systems. He observed the
existence of five distinct types. Of particular interest to us here is the
clear distinction he made between ergative (transitive vs. intransitive)
and active vs. inactive systems, deploring the usual inclusion of these
two types under one rubric. He represents these two case systems
schematically as in (8), which also includes his representation of an
accusative language for comparison ; the names given to the types of
case systems are not used by Sapir (op. cit.). 8
(8) Object Subject Subject Type of
transitive intransitive transitive case system
Inactive Active
A A B Ergative
A A B B Active/Inactive
A B B Accusative
Although Sapir does not go into detail about the active/inactive type,
we may infer something about its nature from Dakhota, the language
he cites as its archetype. It has often been said that Dakhota and the
other Siouan languages distinguish between stative verbs and dynamic
verbs. In Lakhota, a Dakhota dialect, sentences like those below suggest
that wa marks first person singular in dynamic verbs, while ma marks
the same person and number in statives.
wakte' 'I kill it'.
thawa'sose 'I spit.'
mahli'ske 'I am tall'.
However, examples like
mahi'xpaye 'I fall down.' (Examples from Van Valin ( 1977»
show that ma is used not only by statives but also by intransitive
dynamic verbs that are non-agentive, involuntary, or non-controllable.
Recent treatments of the syntax of Dakhota and other Siouan languages

have observed that the wa/ma dichotomy is not dynamic/stative, but
controllable/non-controllable (Matthews 1 965 : 63 ; Van Valin 1 977 : 1 0).
A close examination of the verb classes in other languages reveals that
a controllable vs. non-controllable distinction is not uncommon (e.g.
Chafe 1970). Since the languages cited by Sapir distinguish controllable,
voluntary, or agentive verbs on the one hand from non-controllable,
involuntary, or non-agentive on the other, we may assume that it is
this dichotomy he intended ; it is this distinction, not a dynamic/stative
distinction, that is here referred to as 'active/inactive'.
Postal and Perlmutter observe that there is evidence that the final
subject of 'inactive' verbs is initially not a subject, but a direct object
(cf. Perlmutter 1978 ; and Perlmutter & Postal, to appear a). They have
proposed a theory of language which would recognize at least the fol
lowing initial term inventories :
(9) Initial term inventories
Subject and direct object (transitive)
Subject (intransitive)
Direct object (intransitive)
The initial intransitive direct object becomes subject, as will be shown
below.
Initial term inventories are determined by the semantics of the indi
vidual verb. Initial subjects include agents, cognizers, and other con
trolling nominals, as well as experiencers. Initial direct objects include
patients and stimuli. Examples of verbs that take initial intransitive
subjects include 'work', 'fight', 'talk', and 'growl'. Examples of verbs
that take initial intransitive direct objects include 'exist', 'fall', 'remain',
and 'grow up'.
Verbs which have an initial intransitive subject correspond to Sapir's
'active intransitives' ; verbs which have an initial intransitive direct
object correspond to his 'inactive intransitives'. Thus, languages in
which transitive subjects are marked with one case, and intransitive
subjects and transitive direct objects are marked with another, have
'ergative' case marking. Languages in which final subjects are marked
with one case, and final direct objects with another, have 'accusative'
case marking. And languages in which initial subjects are marked with
one case, and initial direct objects with another, have what Sapir
termed 'active/inactive' case marking.
Notice in (8) that there is only a relatively small difference between

the ergative and the active/inactive systems ; this difference is the
marking assigned to the subjects of active intransitive verbs, verbs like
'work', 'play', • fight' , 'growl', and 'talk'. This is exactly the type of
verb that constitutes Class 3 in Georgian, the Class that causes diffi
culties for the ergative hypothesis (cf. §I .4). According to Sapir, the
subject of active intransitive verbs in an active/inactive system is marked
with the same case as that used for the subject of a transitive verb. We
shall see in §2.2 that this provides the key to case marking in Georgian.
Postal and Perlmutter have further observed that the distinction in
(9) is used also as the basis for other processes in various languages. For
example, in Mohawk it is a necessary, but not sufficient, condition on
incorporation that the incorporated noun be the (final) subject of an
inactive verb or direct object of a transitive verb (those nomina Is
marked with case 'A' in the active/inactive system in Sapir's schema (8»
(Postal, personal communication). Perlmutter (to appear d) shows that
this distinction accounts for a number of distinct phenomena in Italian,
including the selection of auxiliaries and the rule of Partitive Ascension.
The schema in (9) accounts for the fact that the same intransitive verbs
which take case A in (8) also may allow incorporation in Mohawk and
take the auxiliary essere in Italian.
2.2 Specific proposal for Georgian

I suggest that case marking in Series I I is of the active/inactive type. Ii
In general, verbs that take an initial intransitive subject are in Class 3
in Georgian, while those that take an initial intransitive direct object are
in Class 2. The initial term inventory of each verb form is determined
by its semantics, as specified in §2. I . Notice that the verbs listed above
as exemplifying initial intransitive direct objects all belong to Class 2 in
Georgian, while those illustrating initial intransitive subjects all belong
to Class 3 in Georgian. In the discussion below, I will show in detail
how each derivational type fits into a precise schema characterizing the
Classes in terms of initial and final termhood. The reader is referred to
Appendix A of this chapter for representative lists of verbs of the
derivational types constituting each Class. Capital letters in parentheses
in the text below refer to sub-lists under each Class in Appendix A.
2.2. I Class I . Class I contains two derivational types : basic verb forms
(A)lO and organic causatives (B).l1 Dafers 'he will write it' is a basic
238 The natur e of the Georgian verb classes
Class 1 verb (direct construction) with initial subject and direct object,
as established in ch. 13. The structure of this form is represented in
network (10).
(10)
dar;era dar;ers.
' write ' ' he ' , it ' ' He will write it. '
This Class includes a few verbs which take an initial intransitive sub
ject ; one is represented in ( I I ) .
(I I)
gakusvla gakusla.
' gallop away ' ' he ' ' He galloped away. '
Daaferina 'he caused him to write it' is the causative of a transitive ; its
derivation is represented in ( 1 2) (cf. ch. 5).
( 12)
daar;erina.
dar;era ' He made him
' write ' ' he ' , it ' write it. '
As observed in ch. 12, most Class 1 verbs are transitive. All Class 1
verbs are characterized by an initial subject, as illustrated in networks
( 1 0--12).
2.2.2 Class 3. Class 3 also contains two derivational types : basic verb
forms (M) and denominals (N) . Tamasobs 'he plays' is an example of the
former. As established in ch. 12, it takes an optional initial direct
object ; ( 13) and (14) give the two constructions possible for this verb
form.
tamasi burt itamasa burti.

' play ' ' he ' ' ball ' , He played ball. '
tamaH itamasa.
' he ' ' He played. '
The construction in ( 1 3) is identical to that in (10), and (14) to ( I I).

Class 3 verbs which never take a direct object, like iqiqina 'it croaked',
have only constructions like ( 14),
Verbs like gmirobs 'he is taking on the characteristics of a hero, he is
acting brave' are formed from nouns like gmiri 'hero'. They have a
structure similar to ( 1 4).
Class 3 verbs are characterized by an initial subject. Class 3 differs
from Class I only in that the latter are mostly transitive (initially and
finally), while the former mostly take a direct object optionally or not
at all (initially and finally).
2.2.3 Class 2. Class 2 contains the following derivational types : basic

verb forms (c), analytic passives (n), synthetic passives (E), and incep
tives (F) . The ikna-passive (G) and Zustandspassiv (H) are discussed
briefly in the appendix to ch. 13 ; derivationally, they do not differ
from types n and E, respectively.
Network ( I S) represents an analytic passive (d. ch. 7) , and (16) a
synthetic passive (ch. 1 3).
dar;era dar;erilia imis mier.

, write ' ' he ' , it' ' It is written by him. '
(16)
daxrcoba daixrco.
' drown' ' he ' ' He drowned. '
Passivization is responsible for the change in grammatical relations

represented in ( I S). Unaccusative is responsible for the change in ( 1 6).
We know from preceding chapters (especially cbs. I and 2) that in
the clause is iqo enatmecnieri 'he was a linguist', is 'he' is a final subject;
but we have no evidence concerning its initial termhood. I suggest that
this verb, as well as other basic Class 2 verbs, is characterized by an
initial intransitive direct object. The set of basic Class 2 verbs (c) is
consonant with Sapir's 'intransitive inactives', with the set of intransi
tive verbs that may allow incorporation in Mohawk, and with the set of
intransitive verbs that take the auxiliary essere in Italian. Semantically
the term nominal associated with these verbs (the final subject) is always
non-controlling, a patient. I suggest further that the initial direct object,
is 'he' in the example above, becomes subject, and that this operation
is performed by the rule Unaccusative (cf. chs. 8 and 1 3 for the charac
teristics of this rule). Unaccusative, it will be remembered, applies in
the absence of a subject, and promotes direct objects to subjecthood. In
Georgian, Unaccusative is obligatory; there are no verbs such that their
initial term inventory includes a direct object, but no subject, and the
direct object fails to undergo Unaccusative.12 This clause is represented
in ( 17),13
enatmecnieri qopna is iqo enatmecnieri.

, linguist ' , be ' ' he ' ' He was a linguist. '
Inceptives in Georgian are discussed in Holisky ( I 98ob). Inceptives

may be formed on two types of predicates in Georgian: adjectives or
verbs. Each is represented below. The predicate COME ABOUT is a basic
Class 2 verb form (C).
I6.2 The Unaccusati've Hypothesis 241
tetri tetrdeba.
' white ' , it ' ' It becomes white.'
tirili a#rdeba.
, '
cry ' he ' ' He begins to cry. '
Clause Union applies in both ( 1 8) and ( 1 9). As discussed in ch. 5, it is

predictable that direct objects and intransitive subjects of the embedded
clause become direct objects of the united clause, as shown in ( 1 8) and
( 1 9). Since there is no subject at the output level of Clause Union,
Unaccusative must apply, advancing the direct object to subject. I4
Class 2 verbs like darla 'he remained', iqo 'he was', and gaxda 'he
became' differ from Class 3 verbs like imepa 'he reigned', itamaSa 'he
played', and iqiqina 'it croaked' in the following way. The Class 3
intransitives have initial subjects, while the Class 2 intransitives have
initial direct objects which become final subjects by the application of
some rule (Passivization or Unaccusative).
Notice that synthetic passives (E) and basic Class 2 verbs (c) do not
differ derivationally (cf. (16) and ( 1 7» , morphologically, or syntac
tically. They differ in only one respect. Synthetic passives are formed
on verb roots which also form the base for basic Class I verbs ; this is
not true of basic Class 2 verbs. I5 Thus (c) and (E) constitute a natural
sub-class of Class 2 verbs.
2.24 Class 4. Class 4 contains three derivational types : basic Class 4

verbs (p), desideratives (Q), and the type discussed in ch. 8, Appendix
242 The nature of the Georgian 'Verb classes
B (R) . Basic Class 4 verbs are mostly initial transitives, but there are a
few intransitives. These are represented in (20) and (21), respectively.
(20)
qvareba miqvarxar.
' ' you ' 'I love you. '
, love 'I'
�
H
civeba mciva .
' feel cold ' 'I ' 'I feel cold. '
The first change in grammatical relations is due to the application of

Inversion, the second «20) only) to Unaccusative.
Desideratives may be formed on verbs or on nouns. No specific
analysis will be proposed here, but it is clear that these verb forms
involve Inversion. An example of a desiderative formed on a verb is
mece�veba 'I feel like dancing, I want to dance', where m- is the first
person singular indirect object agreement marker, e-eb is the circumfix
for deriving desideratives, ce�v is the root 'dance' , and -a is the third
person singular subject agreement marker. An example of a desiderative
formed on a noun is ar meyvineba 'I don't feel like any wine', where ar
is the negative, and the root 'wine' is yvino.
Class 4 verbs are like Class 1 verbs in that they have an initial subject
and most have an initial direct object. They are like Class 2 verbs in that
their initial direct object, if they have one, is advanced to subject.
Class 4 differs from both Class 1 and 2 in that it triggers Inversion in
all Series.
I can now state my solution to Problem I posed at the end of the
preceding section.
Generalization 1
In Series II, Class I and 3 verbs are those verbs which have an
initial subject that is also their final subject.
By referring to networks ( 1 0-14), which characterize transitive and

r6.3 The Unaccusative Hypothesis 243
intransitive verbs of Classes 1 and 3, we can see that their initial sub
jects are also their final subjects. Class 2 verbs, on the other hand,
either have no initial subject (networks ( 16-19» , or their initial subject
is a final chomeur (passive network ( I S» . Verbs of Class 4 have an
initial subject ; but having obligatorily undergone Inversion, their
initial subject is not a final subject, but a final indirect object (networks
(20-21» . A solution to Problem 2 will be given in the following section.
Most of the derivations discussed here have been justified in detail
in earlier chapters. The only new claims presented here are (a) initial
termhood governed by basic Class 2 verbs (network (1 7» , (b) the
derivation of the inceptive (networks (18- 1 9» , and (c) the generaliz
ations concerning the nature of each Class. The first two of these new
claims are supported by the additional evidence which is presented in
§3 and 4. The generalizations made in this section follow logically from
all of the derivations which have been individually justified.
2.3 The Term Case Marking rules of Georgian

The framework outlined in §2.2 above enables us to make the following
generalization, which corresponds to Problem 2 stated above in § I .4 :
Generalization 2
In Series II, verbs with an initial subject that is also final subject
govern Case Pattern A ; other verbs govern Pattern B.
As shown above, the set of verbs with an initial subject that is also final
subject exactly corresponds to the set of Class I and 3 verbs. Thus, the
distinction between initial intransitive subjects and initial intransitive
direct objects, which distinguishes Class 2 from Class 3, enables us to
account correctly and succinctly for the facts summarized in (2) above.
Now it is possible to state the term case marking rules of Georgian.
The rules for Series I and III are simple and, like those of most
languages of the world, are based on final grammatical relations alone :
I If the verb is in Series I or III,
a. the final subject is marked with the nominative case.
b. the final objects are marked with the dative case.
It was shown in earlier chapters, especially chs. 8 and 9, that case
marking in Series I and III is based entirely on final grammatical
relations. Initial subjects that undergo no change, subjects derived by
Passivization, subjects derived by Object Raising, subjects derived by
Unaccusative - all are marked with the nominative case in Series I and
III. All final objects - direct or indirect - are marked dative, whether
they are initial objects that have undergone no change, were derived by
Inversion, were derived by a Version rule, or were derived by Clause
Union. Thus, the two parts of Rule I correctly account for the case
marking of all term nominals in Series I and III.
For Series II, we must formulate a rule that marks with the 'ergative'
case only the subjects of Class I and 3 verbs ; these are the verbs whose
initial subject is also final subject :
II If the verb is in Series II,
a. an initial subject which is also a final subject is marked with
the ergative case.
This insures that retired subjects, subjects demoted by Inversion, and
subjects whose grammatical relation is changed by Clause Union will
not be marked with the 'ergative' case.
The nominative case marks the following nominals in Series I I :
initial direct objects that undergo no change in grammatical relation

initial direct objects that advance to subject by Passivization
initial direct objects that advance to subject by Unaccusative
initial direct objects that become the subjects of object-raised con
structions
initial direct objects that become the subjects of inceptives (tetrdeba
and semiqvardeba types)
initial direct objects that become the direct objects of organic causa
tives
initial subjects that become the direct objects, then the subjects, of
inceptives (atirdeba type)
initial intransitive subjects that become the direct objects of organic
causatives.
(Analyses of all the constructions referred to are found in preceding
chapters, except inceptives, which are discussed in n. 14 and in Holisky
( I 98ob).) A rule that includes all these nominals will be complex, and
might be formulated in several ways ; the simplest is

b. a direct object that is a final nuclear term is marked with the
nominative case.
lIb correctly marks the nominals listed above and prevents retired
direct objects from being marked with the nominative case.
In Series II the dative case marks indirect objects that have undergone
no change, as well as indirect objects derived by Version or Inversion :
c. a final indirect object is marked with the dative case.
Unlike rules IIa and lIb, rule IIC need not refer to initial grammatical
relations. Yet because of the rule inventory of Georgian, there is no
overlap between the rules as stated. These five term case marking rules
account for cases for all of the constructions of Georgian syntax.
These five rules could be written in a variety of other ways, some of
them as succinct as those given here. 1 6 But the generalizations embodied
in them must be preserved by any valid grammar of Georgian.
The five rules stated above do not refer to verb Classes as ( I b) does.
This means that with respect to the rules of case marking, it is unnecessary
to list Classes for each verb form in the lexicon. Rather, we can predict
case marking on the basis of rules I and II, together with the semantics
of the verb. Thus, rules I and II effect an important simplification in
the grammar, in terms of not requiring that Class be listed in the
lexicon.
2.4 Residual problems

While the proposal made in §2.2 makes the correct predictions for the
vast majority of verbs in Georgian, there remain a few verbs which it
does not correctly account for. Both types of exception exist : seman
tically active verbs that are in Class 2 and govern Pattern B in Series II,
and semantically inactive verbs that are in Class 3 and govern Pattern A
in Series II. Three important points must be made about these excep
tions : (a) All of them occur in synchronically non-productive morpho
logical subclasses. (b) Some of them cause confusion for native speakers.
(c) Some or all of them are regularized in the non-literary dialects.
Holisky ( I980a) analyzes the morphological subclasses of Class 3
(medial) verbs, showing the formation process for each, distinguishing
productive from non-productive types, and investigating the semantic
characteristics of the subjects that are possible for each type. She shows
that every Class 3 verb that is an exception to my proposal is in a frozen,
non-productive subclass ; most of the Class 3 verbs in the frozen sub
classes and all those in productive subclasses are active verbs, in the
sense of §2. 1 . Her data are an exhaustive list of the hundreds of Class 3
verbs, which she culled from standard dictionaries and tested with
informants. Similarly, the semantically active verbs of Class 2 occur in
closed subclasses ; these include subclasses J and K (discussed in Appen
dix B to this chapter) and subclass L (discussed briefly in Harris (to
appear b) and Harris 1976). The productive subclasses of Class 2 -
inceptives, analytic passives, synthetic passives, object-raised con
structions are, without exception, inactive in the sense of §2. I .
A second important point about these exceptions is that some cause
confusion among native speakers. (22) gives two variants for the verb
�etva 'drip' an inactive verb formally in Class 3 . In the (a) sentence,
Pattern A is used, in the (b) sentence, Pattern B.
(22) (a) macivridan fqalma i�eta.
refrigerator-from water-ERG it-dripped-II-3
'Water dripped from the refrigerator.'
(b) macivridan fqali i�eta.
water-NOM
'Water dripped from the refrigerator.'
Both Patterns are used for this verb, though (a) is considered more
literary and is preferred by prescriptive normsY Case marking, com
plex as it is in Georgian, does not generally cause any difficulty for the
native speaker. The fact that it does cause a problem with this verb
confirms its status as an exception, rather than a counterexample.
Examples of confusion of case marking with active verbs formally in
Class 2 are discussed in more detail in Appendix B.
Finally, although these frozen forms are retained as exceptions in the
literary language of the metropolis, many or all of them are regularized
in a number of dialects, including Gurian and Imerian (western
Georgia), the mountain dialects (north central Georgia), and Pereidnian
(Iran). 18 The examples most frequently cited in the literature on these
dialects are of the type �acma favida 'the man-ERG went' and �acma
adga 'the man-ERG got up ' , where the subjects are in the ergative, as
would be expected for active verbs ; in the dialect of Tbilisi, these and
several other very common verbs are irregular, taking Pattern B in
Series II. It is well known in linguistics that a literary dialect tends to
preserve irregularities, particularly with very common words.
I6.3 Inversion 247
3 Inversion and the Georgian verb classes
In earlier chapters it was noted that the Class of the governing verb
determines two distinct syntactic phenomena case marking in Series
II and Inversion in Series III. In the preceding section it was shown
that the rules of case marking do not need to refer to the notion of
' Class'. Rather, case marking can be predicted on the basis of the initial
and final termhood of nominals, which is, in turn, determined by the
semantics of the verb. If it were possible also to predict which verbs
undergo Inversion in Series III without reference to Classes, then the
notion 'Class' would not be needed in the syntax. In this section, I shall
show that this is possible.
In ch. 8 it was shown that Inversion applies under two conditions :
(23) Inversion is triggered by
(a) Class 4 (inversion) verbs
(b) Series III (evidential) forms.
(23b) must be further restricted by (24).
(24) Class I and 3 verbs trigger Inversion,
Class 2 does not.
Chapter 8, §7 describes five environments in which Inversion does not
apply in a clause with some syntactic rule. Whether triggered by Class 4
or Series III forms, Inversion fails to apply to the output of Passiviz
ation, Object Raising, or Inversion and Unaccusative. Each of the
impossible derivations discussed there involves the application of Inver
sion to a derived subject. If clauses of the types represented by networks
( 1 6-19) of the present chapter are put into Series III, they too fail to
undergo Inversion. In each instance, Inversion fails to apply to a non
initial subject, just as was the case with the examples discussed in ch. 8.
In fact, Inversion never applies to a non-initial subject ; further, every
clause to which Inversion fails to apply (if triggered by (23» contains a
non-initial subject. I therefore propose to replace the ad hoc statement
(24) with
(24') The Initial Subject Constraint on Inversion
Inversion applies only to initial subjects. 1 9
In the framework proposed here, (24') prevents Inversion from applying
to analytic passives (network ( I S)), synthetic passives (network ( 1 6» ,
basic Class 2 verbs (network ( 1 7» , inceptives of both sorts ( 1 8-19),
basic inversion verbs (20-2 1), as well as with types G, H, Q, and R,

which could not be discussed in detail above. In each of those deriva
tional types, there is, according to my analysis, no initial subject which
could undergo Inversion under the conditions imposed by (24'). It is
entirely possible that (24') is a part of the rule of Inversion universally.
In the absence of sufficient evidence, I propose it as a language-partic
ular constraint only.
Let us consider in greater detail some specific examples. Suppose that
example ( 1 7) is in Series III, is enatmecnieri qopila 'he is apparently a
linguist'. There is an initial direct object, but no initial subject. Inver
sion cannot apply, because its input conditions are not met (there is no
subject). The input conditions for Unaccusative are met, and this rule
obligatorily applies, making the direct object the subject. Now the
input conditions for Inversion are met, since there is a subject. But the
Initial Subject Constraint correctly prevents Inversion from applying
to this non-initial subject. It remains subject, as shown by the case
marking, verb agreement, etc.
Let us take as an example a verb root that occurs in both Class 1
and Class 2 forms. Consider (25) as an initial structure.
(25)
da�era ameri�eli �igll-

' write ' ' American ' , book '
If no rule changes the grammatical relations, dafera will be realized as

a transitive direct Class 1 verb. Then in Series III it will undergo
Inversion as predicted, as well as Unaccusative.
(26)
da�era ameri�eli �ign-
amerilf.els dauferia es figni.

American-oAT he-wrote-it-III- I this book-NOM
'Apparently an American wrote this book.'
I64 The semantics of Class 249
If, on the other hand, Passivization applies to (25), darera will be

realized as an analytic passive ( Class 2 verb). In Series III, it will not
be able to undergo Inversion, because this is blocked by the Initial
Subject Constraint. The subject which satisfies the input conditions for
Inversion (the derived subject) is not an initial subject. The resulting
sentence will be (27), exactly as predicted.
dOf:era omerikeli f:iglli
es rigni qopila dacerili amerikelis mier.

this book-NOM it-was-III-2 written American by
'This book was apparently written by an American.'
Thus, on the analysis proposed here for Class 2 verbs (crucially involv
ing the notion of initial intransitive direct objects), the correct pre
dictions are made for Series III by the rules of Inversion, U naccusative,
Passivization, (23) and the Initial Subject Constraint, all of which are
independently necessary.
The proposal made in §2 enables us to effect three simplifications in
the grammar : (i) the rules of case marking can be stated in a general
way, (ii) a general statement can be made of which verbs govern
Inversion in the evidential, and (iii) because neither case marking rules
nor Inversion need refer to verb Class, the lexicon can be simplified. 20
In addition, the Initial Subject Constraint on Inversion, in the gram
mar proposed here, explains the otherwise arbitrary fact that Class 2
verbs do not undergo Inversion in Series III. These economies and this
explanation can be made only with the analysis presented in §2.
1- The semantics of Class
In this section I will discuss two productive pairs of verb forms, where
one member of the pair belongs to Class 2 and the other to Class 3 . The
purpose of this section is to show that in each instance the Unaccusative
Hypothesis makes the correct predictions concerning Class membership.
Or, differently stated, this hypothesis makes the correct predictions
concerning case marking and Inversion. Other productive formations
have already been discussed in detail elsewhere, including the analytic
and synthetic passives. The two types discussed here particularly stand
out because of the fact that in each pair both members are intransitive
(finally and initially). This means that the Ergative Hypothesis will
make the wrong predictions concerning the case marking and Inversion
for one member of each pair. The fact that these formations are pro
ductive emphasizes the inadequacy of treating one member ( Class 3
form) of each pair as an exception, as the Ergative Hypothesis does.
Holisky ( 1978) points out the difference between pairs like tamarma
imepa 'Tamar reigned' and tamari mepe iqo 'Tamar was (a/the) monarch'.
Both predicates are based on the root mepe 'monarch'. The first is
deliberate, active, controlled by the subject ; the second is stative, non
deliberate, and out of the control of the (final) subject. On the basis of
the semantics of the verb, we know that the first takes an initial subject ;
the second takes an initial direct object, as qopna 'be' always does. The
final subject of the first is its initial subject, and is correctly marked
ergative by rule IIa. The final subject of the second is its initial direct
object, and is correctly marked nominative by lIb. Derivations are
represented in (28) and (29).
(28)
tamarma imepa.
mepoba tamar tamar-ERG she-reigned
' reign' ' Tamar reigned .'
tamari iqo mepe.

mepe qopna tamar tamar-NOM she-was monarch-NOM
( monarch be ' , Tamar was the monarch . '
The Unaccusative Hypothesis also makes the correct predictions

concerning Inversion in Series III. The Initial Subject Constraint
correctly predicts that tamar can undergo Inversion in (28), but not in
(29)' These predictions are borne out in (30).
I6.4 The semantics of Class 251
(30) (a) tamars umepia.

Tamar-DAT she-reigned-III-3
'Apparently Tamar reigned.'
(b) tamari mepe qopila.
Tamar-NOM monarch-NOM she-was-III-2
'Tamar was apparently the monarch.'
(c) *tamars mepe qopila.
(,Tamar was apparently the monarch.')
The initial subject in (30a), but not the derived subject in (30b), under
goes Inversion.
Thus, the Unaccusative Hypothesis makes the correct predictions
concerning the case marking and Inversion of each member of this
type of pair, without reference to Class, but rather on the basis of
initial grammatical relations, which are predictable on the basis of the
semantics of the verb. Since both predicates are intransitive, the
Ergative Hypothesis cannot distinguish between them, except by treat
ing them as exceptions and referring to a lexical listing of Class. Without
referring to Class, the Ergative Hypothesis would incorrectly predict
that both verb types govern nominative case subjects in Series II and
fail to undergo Inversion in Series III.
A second productive type is illustrated in (3 1 ) and (32). (31) is for
mally an inceptive ; it is inactive, and non-controllable.
(3 1) (a) vano ayiyinda.

Vano-NoM he-come-about-croon-II-z
'Vano began to croon/sing in a low voice.'
(b) laidani ayiyinda.
teakettle-NoM
'The teakettle began to croon (Le. whistle).'
The grammaticality of (3 I b) shows that the final subject of this verb is

not semantically a controller, since an inanimate could not control this
action. By the same token, (32) shows that the form of the verb illus
trated there is controllable, since the inanimate subject is not gram
matical.
(32) (a) vanom (atytyma.

VanO-ERG he-crooned-II-3
'Vano began to croon.'
(32) (b) *laidanmaj*laidani cairiyina.

teakettle-ERGjNOM
('The teakettle began to croon.')
The difference between (3 1) and (32) cannot be adequately glossed in
English. 2 1 (31) and (32) represent non-controllable and controllable
forms of the same verb.
As a semantic patient, the final subject of the verb in (31) must be an
initial direct object. As a semantic agent, the final subject of the verb
in (32) must be an initial subject. The final subject of (3 1) is correctly
marked nominative by lIb. The final subject of (32) is correctly marked
ergative by IIa.
Further, the Initial Subject Constraint correctly predicts that the
subject in (32). being an initial subject, will undergo Inversion in
Series III, while that in (3 1 ) will not. That this is correct is shown in
(33). where the Class 3 verb, but not the Class 2 verb, has undergone
Inversion, as indicated by case marking and agreement.
(33) (a) vano ayiyinebula.
Vano-NoM he-come-about-croon-111-2
'Vano apparently began to croon.'
(b) *vanos ayiyinebula.
Vano-DAT
' Vano apparently began to croon.'
(c) vanos cauytytma
Vano-DAT he-crooned-III-3
'Vano apparently began to croon.'
The Unaccusative Hypothesis makes the correct predictions in each
instance on the basis of initial and final termhood, which is predictable
from the semantics of the verb and its dependents. The Ergative Hypoth
esis is unable to distinguish between these two intransitives without
reference to a lexical listing of Class. 22
Each of these pairs represents two productive derivational types. In
each type there is a correlation between the semantics of the final
subject and the syntax.
5 Comparison and conclusions
In §S. I I shall summarize the most important aspects of the two

theories considered in detail above. Comparing the grammars required
I6.5 Comparison and conclusions 253
of the two will reveal that the Unaccusative Hypothesis is more general,
and therefore superior. In §5 . 2 I will briefly consider and dismiss the
proposal that Georgian rather exhibits a low 'degree of ergativity'.
5 . 1 Hypotheses A and B
5 . 1 . 1 Case Marking. Hypothesis A ( The Ergative Hypothesis). According
to the Ergative Hypothesis, in Series II the case marking of nominals
governed by verbs of Classes I and 2 is predictable on the basis of the
transitivity of the verb. The grammar must, therefore, include the
following rules of case marking.
n If the verb is in Series II.
a'; the subjects of transitive verbs are marked with the ergative
case.
n If the verb is in Series II,
b'. the subjects of intransitive verbs and the direct objects of
transitive verbs are marked with the nominative case.
In addition, it will include rules Ia and b and rule nc, all stated above
in S2.3. These rules ( na ' , b') are inaccurate for the regular systematic
group of Class 3 verbs, which are mostly intransitive, but do not fit the
ergative generalization. For them, we must include another rule,
n If the verb is a Class 3 verb in Series II,
d. the subject is marked with the ergative case and the direct
object with the nominative.
Few Class 3 verbs have direct objects, but the last clause must be
included for those that do.
Hypothesis B (The Unaccusative Hypothesis). According to the pro
posal introduced here, in Series II, the case marking of nominals
governed by verbs of all Classes is predictable on the basis of initial
and final termhood. The transitivity of the verb and the secondary
grammatical relations, ergative and absolutive, play no part in term
case marking. The grammar must include rules I and II (five rules).
Conclusion. When we compare the statements (rules) which must be
included in each grammar, we find that Hypothesis B is significantly
more general. Hypothesis A is unable to include Class 3 in the general
ization that covers the rest of the grammar ; this is reflected in the
inclusion of rule lId, which is unnecessary in Hypothesis B. There are
exceptions to both proposals, including the verbs discussed in Appendix

B. But while each has exceptions, Hypothesis B accounts for a signifi
cantly larger group of verbs. In this way it gives a general account of
term case marking throughout the language.
5 . 1 .2 Inversion. Hypothesis A . I n Hypothesis A, no attempt is made to

explain why Inversion fails to apply with Class 2 verbs in Series III.
Proponents of the Ergative Hypothesis make statement (23) and add
condition (34).
(34) Transitive verbs trigger Inversion ; intransitive verbs do not.
Since Class 3 verbs are syntactically intransitive, but do not trigger

Inversion, proponents of this theory must add,
(35) Class 3 verbs trigger Inversion,
which is treated as the statement of an irregularity.
Class 4 verbs must also be prevented from undergoing Inversion

twice (cf. §3 and ch. 8, §7). But if (34) is interpreted as referring to
transitivity at the level of input to the rule, an inversion verb will auto
matically be prevented from undergoing Inversion twice, since it will
be intransitive after the first application of the rule. Hypothesis A
therefore requires no additional statement for this purpose.
Hypothesis B. As stated above in §3, Hypothesis B requires rule (23)
and the Initial Subject Constraint o n Inversion. The latter accounts in
a general way for both the fact that Class 2 does not undergo Inversion
and the fact that Class 4 verbs do not undergo that process twice in
Series III.
Conclusion. Hypothesis B requires fewer rules than Hypothesis A. In
addition, Hypothesis B makes a linguistically significant generalization
that is not made by A.
5 . 1 . 3 Predictive pO'll.Jer. Hypothesis A. As pointed out in §4, Hypothesis A

is unable to predict the difference between the members of the two
types of productive pairs considered there : imepa 'he reigned' /mepe iqo
'he was monarch', and 9aiyiyina 'he deliberately began to croon' /ayiyinda
'he non-deliberately began to croon'. Since both members of each pair
are intransitive, Hypothesis A predicts incorrectly that both members
will govern the nominative case subject in Series II and will fail to
undergo Inversion in Series III.
I6,s Comparison and conclusions 255
Hypothesis B. On the basis of the semantics of the verb, it is pre

dictable on Hypothesis B that the final subject of the first member of
each pair will also be its initial subject and will therefore be marked with
the ergative case in Series II and undergo Inversion in Series III.
Similarly, the semantics of the verb make it clear that the second mem
ber of each pair has an initial direct object, which is therefore marked
with the nominative case in Series I I and fails to undergo Inversion in
Series III.
5. 1.4 Reference to Classes. Hypothesis A. Since rules lId and (35) refer
specifically to Class 3 , each verb of this Class must be indicated in the
lexicon. I n addition, exceptions must be indicated.
Hypothesis B. There are also exceptions to this theory and they must
be listed in the lexicon. Class 3 is not referred to as such by any rule,
and need not be indicated in the lexicon. Indeed, there is no need to
make any special indication of the syntax of regular verbs discussed
above.
5 . 1 . 5 Interaction with other rules of grammar. There is no significant

difference between Hypothesis A and B with respect to the formulation
of any rule discussed in earlier chapters. The difference between them
is irrelevant to most rules of the grammar. Of the rules that have not
already been discussed here, only Tav-Reflexivization, Number Agree
ment, and Causative Clause Union refer - directly or indirectly - to
specific initial termhood. The last of these treats intransitive subjects
and transitive and intransitive direct objects in the same way ; therefore
the new proposal does not practically affect the application of this rule.
The reader can confirm that in earlier chapters, when any generalization
was based on transitive vs. il1transitive verbs, both types of intransitive
were taken into consideration.
The Unaccusative Hypothesis does not alter the applicability of the
statement of Number Agreement, since the same nominal is the 'first
subject that is a final term' on both analyses. Number Agreement in the
synthetic passive i s discussed in ch. I S , and Number Agreement in
other Class 2 verbs operates the same way.
Tav-Reflexivization is the only rule whose formulation could require
alteration as a consequence of the analysis proposed here. I have not
been able to find examples that would test this rule with Class 2 verbs.
Further research could reveal a need to alter the statement of Tav
Reflexivization to
The antecedent of tav- must be

a. the first subject of a clause of which tavis a dependent, and
b. a final term,
where 'initial subject' has been changed to 'first subject'. This formu
lation, if found to be correct, would not be more or less complex than
that given in ch. 14 (cf. ch. I S , §4), and therefore would not affect the
choice between Hypotheses A and B.
5. 1 .6 Conclusion. The grammar required by the Ergative Hypothesis is

more complex than that required by the Unaccusative Hypothesis in
the treatment of case marking, in the treatment of Inversion, and with
respect to lexical listing of Class. Hypothesis A fails to capture the
linguistically significant generalizations which are embodied in the
approach of Hypothesis B. Hypothesis A is unable to make correct
predictions concerning the syntactic behavior of at least two productive
derivational types. Hypothesis B must be adopted.
5.2 Ergativity in Georgian

Because it is more general, we must prefer the Unaccusative Hypothesis
over the Ergative Hypothesis. This means that case marking in Series
II is not ergative, but belongs to the type Sapir described as active vs.
inactive.
Nevertheless, some might suggest that Georgian is not a non-ergative
language, but rather exhibits a low 'degree of ergativity'. In considering
the usefulness of this idea in the description of Georgian, there are two
points that must be made. First, labeling a language 'ergative' or stating
that it has a high or low 'degree of ergativity' is an imprecise description
and provides no insight into the nature of the language. A particular
rule can be ergative, in the sense that it refers to transitive subjects on
the one hand, or to direct objects and intransitive subjects on the other
(cf. § I . I ). Case marking and verb agreement are often cited as ergative
rules in this sense, but other rules may also be ergative. The notion
'ergative language', on the other hand, has not been used in a consistent
way. It has been used to refer to languages with ergative case marking
only (e.g., Udi), languages with ergative verb agreement only (e.g.,
Abxaz) , languages with both ergative case marking and ergative verb
agreement (e.g., Avar), and languages with neither ergative case marking
nor ergative verb agreement (e.g., Georgian). In order to give an accur
ate description of a language, it is better to avoid vague labels like
I6.6 Theoretical implications 257
'ergative language' and use notions that can be precisely defined, such
as 'ergative rule'.
Second, Georgian has, to the best of my knowledge, only two rules
that might be considered ergative. Retired Term Marking is truly
ergative ; retired ergatives are marked one way (mier) and retired
absolutives another �genitive case). Causative Clause Union is also
ergative, in the sense that it treats ergatives one way, making them
matrix indirect objects, and absolutives another way, making them
matrix direct objects. 23 Interestingly, this type of causative occurs in
languages that apparently have no other ergative rule, such as French,
Turkish, and Japanese (cf. ch. 5, § 1). As shown above, neither the term
case marking rules, nor the agreement rules have anything to do with
the secondary grammatical relations ergative and absolutive. 24
Term case marking in Georgian might be mistaken for a 'partially
ergative' rule because of the fact that marking in an active/inactive
system differs from that in an ergative system only with respect to
initial intransitive subjects. On the other hand, marking in an active/
inactive system differs from that in an accusative system only with
respect to initial intransitive direct objects.
Series II of Georgian belongs to a case marking type that was recog
nized at least as early as 1 917. Sapir deplored the 'inclusion under one
rubric of transitive versus intransitive, and active versus inactive' (Sapir
1 9 1 7 : 85) ; yet sixty years later, the same fallacy is perpetuated in
analyses of Georgian, and it is labeled an 'ergative language' . To main
tain that Georgian is ergative or even 'partially ergative' would be to
ignore the need for accurate description and to do violence to the
integrity of the active/inactive system as a distinct type.
6 Theoretical implications
The Georgian data presented in this and earlier chapters are important
for the study of universal aspects of active type rules. These data bring
particularly strong evidence to bear on three interrelated questions :
(a) What level of derivation is relevant to the active/inactive distinction ?
(b) Can active type rules be stated directly on semantic notions, rather
than on syntactic relations ? (c) Why is active distinguished from inactive
in intransitive verbs, but not in transitive verbs, in (8) ? The evidence
from Georgian in each respect is summarized briefly below.
The many rules that change grammatical relations in Georgian make
it clear that both initial and final levels of derivation must be taken into
account in the following way :
(i) Initial grammatical relations are determined by semantic
notions 'agent', 'experiencer', 'patient', etc. (cf. §2. I).
(ii) Syntactic rules may change grammatical relations.
(iii) Case marking in Series I and Inversion in Series III are
sensitive to grammatical relations at more than one level (cf.
§2·3 and 3).
Case marking in Series II cannot be stated directly on semantic
notions, without the intervention of the syntactic relations, as stated
above. Although agents in sentences like (10-14) are marked with the
ergative case, it is not true that all and only agents are so marked. For
example, the initial subject of a passive is a semantic agent but is not
marked with the ergative. On the other hand, the subject of (34) below
is non-agentive but is marked with the ergative.
Georgian also provides data that show why Sapir did not distinguish
active from inactive in transitive verbs. Like intransitives, transitive
verbs may be semantically active or inactive. The initial subjects of
active transitives are agents, those of inactive transitives are experi
encers. If inactive transitives undergo no change in grammatical re
lations, they will have a final subject and direct object, as in (34).
(34) (a) kartulma da�arga 'vin' da 'ra' �tegoriata. . .
Georgian-ERG it-lost-it-II-1 who and what categorieS-GEN
farmoeba.
formation-NOM
'Georgian lost the formation of the human and neuter cat
egories.'
(Example from text, Chikobava 1 940 : 1 6.)
(b)
kartul
' G eo rgi an '
r;armoeba �ategoria
' formation' ' catego ry '
If, on the other hand, Inversion applies to an inactive transitive, it is a

final intransitive, as in (35).
(35) (a) evian zmnebs v elJ,argebat . . . . 25

having-v verbs-OAT V-NOM they-lose-it-I-4
'Verbs in v lose v.'
(Example from text, Shanidze 1973 : 445.)
(b)
dakargva zmna v
' lose ' ' verb '
Because case marking and agreement are sensItive to grammatical

relations, not to the semantic categories agent and experiencer, the
inactive transitive in (34) does not differ in these respects from the
active final transitive in ( I O). Similarly the inactive initial transitive that
is a final intransitive in (35) does not differ in any apparent way from
other inactive final intransitives discussed in §2.2.3. Thus (8) dis
tinguishes only three types because inactive transitives coincide in the
relevant respects with either active transitives or with inactive intran
sitives.
Appendix A: Sample lists of verb classes
The division of verbs into Classes I through 4- corresponds approximately to

divisions made by traditional grammarians. However, because different
criteria are used, the divisions differ slightly from traditional ones. The
justification for this departure from tradition is the high degree of coincidence
of morphological , syntactic, and semantic criteria in the system adopted here.
Traditional labels are not used here because they are imprecise. For example,
Class I verbs are traditionally called 'transitive verbs' (cf. § I above) ; yet this
Class includes some verbs which never take direct objects, such as daamt
knarebs 'he will yawn'. Therefore a neutral name, ' Class I', is used here.
Morphological criteria are taken here as the point of departure for defining
Classes.26 Morphologically, each Class can be characterized in the following
way :
260 Appendix A : Sample lists of verb classes
CLASS I: a. The future/aorist tenses are formed with a preverb.

b. I n the future tense, the suffix -s marks third person singular
subjects, the suffix -en third person plural subjects.
c. In the aorist tense, the third person plural subject is marked
by the suffix -es.
CLASS z: a. The future/aorist tenses are formed with a preverb or the
character vowel e-.
b. In the future tense, the suffix -a marks third person singular

subjects, the suffix -an third person plural subjects.
by the suffix -nen.
CLASS 3 : a. The future/aorist tenses are formed with the circumfix i - (eb) .
b. In the future tense, the suffix -s marks third person singular

subjects, the suffix -en third person plural subjects.
by the suffix -es.
CLASS 4 : a. The future/aorist tenses are formed with the character vowel e-.
b. In the future tense, the suffix -a marks third person singular
subjects.
c. In the aorist , the third person plural final subject is marked
as a singular (cf. ch. 15).
Characteristic (a) sets apart Class 3. Characteristic (b) separates Classes I
and 3 from z and 4. Characteristic ( c) separates Classes I, 3 from z and from 4.

For a few irregular verbs, one criterion or another may be inapplicable ;
in this case, the others will define the Class. For example, the verb qopna 'be'
forms its future irregularly by root suppletion, so criterion (a) is inapplicable.
But according to (b) and (c), this verb is in Class z;
ikneba 'he will be'

iknebian 'they will be'
iqvnen 'they were'.
Syntactically, we may observe the following correlations (cf. chs. 2 and 8) :

d. Verbs which fall into Class I or 3 by the above criteria govern
Pattern A in Series I I.
e. Verbs which fall into Class I or 3 by the above criteria trigger
Inversion in Series I I I ; verbs which fall into Class 2 by the
above criteria do not trigger Inversion in Series III.
f. Verbs which fall into Class 4 by the above criteria trigger
Inversion in all Series. 27
The very few partial exceptions to these syntactic correlations are discussed
in Appendix B and §2.4 above.
We may add one further syntactic correlation :
g. Verbs which fall into Class I or 2 according to the above

criteria take an initial direct object obligatorily ; verbs which
fall into Class 3 according to the above criteria take an initial
direct object optionally or not at all.
The lists following show that (g), unlike (d-f), is a tendency, not an absolute
correlation.
Some of the semantic correlates of Class are discussed in Holisky (I98oa).
It must be noted that these are Classes of derivational forms, not of verb
roots ; a particular root may be represented in more than one derivational
type. The root fer, (A) dat;ers 'he
for example, occurs as a basic Class I verb
will write it', a causative (B) daat;erinebs 'he will cause him to write it', an
analytic passive (D) dat;erili ikneba 'it will be written', an ikna-passive dat;erili
ikna 'it got written', a Zustandspassiv (H) efereba 'it will stand written'. The
root cur 'swim' occurs as a basic Class I verb (A) gacuravs 'he will swim it',
and as a basic Class 3 verb (M) icuravebs 'he will swim'. Ce'f!,v 'dance' occurs
as a causative (B) ace'f!,vebs 'he will cause him to dance', as an inceptive (F)
ace�vdeba 'he will begin to dance', as a basic Class 3 verb (M) ice'f!,vebs 'he
will dance', and as a desiderative (Q) ece�veba 'he feels like dancing'. Tetri
'white' occurs as a causative (B) gaatetrebs 'he will make it white', and as an
inceptive (F) gatetrdeba 'it will become white'. Each of these forms represents
a regular, productive derivational category. Without exception, all causatives
(B) belong to Class I ; all synthetic passives (E), analytic passives (D), incep
tives (F), ikna passives (G), and Zustandspassivs (H) to Class 2 ; and all desider
-
atives (Q) to Class 4. Class 3 contains all the members of the -ob denominals
(N) discussed in Holisky (I 98oa).
The lists are presented here to illustrate the verb types that constitute each
Class and to illustrate the verbs which constitute each derivational type
referred to in §2.2. The verbs selected are typical; they illustrate the general
izations made in the text above. At the same time, the lists contain examples
of every kind of exception to the generalizations, both to the claims of tra
ditional analyses and to the proposals made here.
S A M P L E L I S T OF C L A S S I V E R B S
A. gaIlis 'he will spread it (out),

gaatbobs 'he will heat, warm it'
gamoacxobs 'he will bake it'
gazrdis 'he will raise him, grow it'
dairqebs 'he will begin it'

gaxsnis 'he will open it up'
daxuravs ' he will close, shut it (up)'
aantebs 'he will light it , turn it on'
gaadnobs 'he will melt it'
daylis 'he will tire him out'
gaxdis 'it will reduce him (in weight)'
la/y.elavs 'he will lock it'
daxevs 'he will rip it'
la3iravs 'he will sink it'
moxarsavs 'he will cook it'
garqvels 'he will break it off'
,
gaoJrobs 'he will dry it (off/out)
modrely,s 'he will bend it'
eaakrobs 'he will put it (light) out'
farmosobs 'he will engender it'
sobs 'she will bear/bears it'
dabadebs 'she will bear it'
dafers 'he will write it'

dap-is 'he will wound him'
garecxavs 'he will wash it'
gadatargmnis 'he will translate it'
da/y.argavs 'he will lose it'
dagvis 'he will sweep it (out)'
daxarjavs 'he will spend it, use it up'
datesavs 'he will sow it'
se/y.eravs 'he will sew it'
,
gacuravs 'he will swim it (e.g., river)
daus1Vens 'he will whistle it'
miscems 'he will give it to him'

aCukebs 'he will give it to him as a gift'
misrers 'he will write it to him'
areevs 'he will select, prefer this over that'
daamtknarebs 'he will yawn'

amoaxvelebs 'he will cough'
B. daaferinebs 'he will cause him to write it'
gaoJlevinebs 'he will cause him to spread it (out)'
daa{revinebs 'he will cause him to wound him'

gaatbobinebs 'he will cause him to heat, warm it'
miaceminebs 'he will cause him to give it'
acelJ.vebs 'he will cause him to dance'

amyerebs 'he will cause him to sing'
atamalebs 'he will cause him to play'
varjiIebs 'he will train him, cause him to exercise'
amepebs 'he will cause him to reign'

amusavebs 'he will cause him to work'
amecadinebs 'he will teach him, cause him to study'
gaatetrebs 'he will make it white'

gaamravlebs 'he will increase it, cause it to increase'
gaa3virebs 'he will make it more expensive'
gaagr3elebs 'he will make it longer'
S A M P LE L I S T OF C L A S S Z V E R B S
c. ikneba 'he will be'

elodeba 'he will wait for him'
ga!qdeba 'it will break'
feesabameba 'it will correspond to s.t.'
seesa!qviseba 'it will agree with s.t. , suit s.t.'
gaipotleba 'it will leaf'
lavardeba 'he will fall'
darleba 'he will remain'
moxdeba 'it will happen'
D. gallili ikneba 'it will be spread (out)'
gamtbari ikneba 'it will be heated, warmed'
gamomcxvari ikneba 'it will be baked'
gazrdili ikneba 'he will be raised'
dat;qebuli ikneba 'it will be begun, started'
gaxsnili ikneba 'it will be opened up'
daxuruli ikneba 'it will be closed, shut (up)'
antebuli ikneba 'it will be lit , turned on'
gamdnari ikneba 'it will be melted'
daylili ikneba 'he will be tired out'
,
gamxdari ikneba 'he will be reduced (in weight)
laly,etili ikneba 'it will be locked'
daxeuli ikneba 'it will be ripped'
laJiruli ikneba 'it will be sunk'

moxarIuft' ikneba 'it will be cooked'
gafqve#li ikneba 'it will will be broken off'
gamfrali ikneba 'it will be dried (off/out)'
modre�i1i ikneba 'it will be bent'
lakrobili ikneba 'it will be put out (light)'
farmofobili ikneba 'it will be engendered'
fobili ikneba 'it will be borne'
dabadebuli ikneba 'it will be borne'
daferili ikneba 'it will be written'

da{rili ikneba 'he will be wounded'
garecxili ikneba 'it will be washed'
gadatargmnili ikneba 'it will be translated'
da�arguli ikneba 'it will be lost'
dagvili ikneba 'it will be swept (out),
daxarjuli ikneba 'it will be spent, used up'
E. gaifleba 'it will spread (out) ,
gatbeba 'it will heat, warm up'
gamocxveba 'it will bake'
gaizrdeba 'it ",ill grow (up)'
daifqeba 'it will begin'
gaixsneba 'it will open (up)'
daixureba 'it will close, shut (up)'
ainteba 'it will light up, turn on'
gadneba 'it will melt'
daiyleba 'he will tire out'
gaxdeba 'he will reduce (in weight)'
lai�eteba 'it will lock'
daixeva 'it will rip'
laiJireba 'it will sink'
moixarfeba 'it will cook'
gai(:qviteba 'it will break off '
gafreba 'it will dry (out/off) ,
moidri�eba 'it will bend'
cakreba ' it (light) will go out'
farmoiJveba 'it will originate'
iSveba 'it will be/is born'
daibadeba 'it will be born'
F. gatetrdeba 'it will become white'
gamravldeba 'it will become plentiful/multiply'
ga3virdeba 'it will become expensive'

,
gagrJeldeba 'it will become long(er)
gamepdeba 'he will become king'
acely;vdeba 'he will begin to dance'

amyerdeba 'he will begin to sing'
atamaIdeba 'he will begin to play'
G. daferili ikna 'it got written'
da{rili ikna 'he got wounded'
leyebili ikna 'it got painted'
H. e{:ereba 'it will stand written, be in a state of having been written'
exateba 'it will stand painted, be in a state of having been painted'
usxia 'it will be poured with respect to him, he will have a full glass of wine'
J. etamaleba 'he will play with him'

elaparalJ,eba 'he will talk to him'
eCxubeba 'he will fight with him'
estumreba 'he will visit (with) him'
K. mouqveba 'he will tell it to him'
sepirdeba 'he will promise it to him'
L. ilJ,bineba 'he bites'
ilanJyeba 'he abuses, curses'
igineba 'he curses, swears'
S A M P L E L I ST O F C L A S S 3 VERBS
M. icely;vebs ' h e will dance'

imyerebs ' he will sing'
itamaIebs 'he will play'
ivarjilebs 'he will exercise, train'
ilxubebs 'he will quarrel'
ibrJvis 'he will fight'
isuntkebs 'he will breathe'
itirebs 'he will cry'
ixitxitebs 'he will giggle'
ilJ,ivlebs 'he will scream'
iqvirebs 'he will yell'
ibuzyunebs 'he will grumble'
ilaparalJ,ebs 'he will talk'
isaubrebs 'he will converse'
ilaqbebs 'he will chatter'
iqbedebs 'he will chatter'

istvens 'he will whistle'
icinebs 'he will laugh'
ioxrebs 'he will sigh'
i1!.vnesebs 'he will groan'
ixvnefebs 'he will moan'
ixvrinebs 'he will snore'
iprutunebs 'he will snort'
ibzuvlebs 'he will hum, buzz'
iqiqinebs 'he will croak'
izuzunebs 'he will hum, buzz'
iyuyunebs 'he will coo'
iciguravebs 'he will skate'
icuravebs 'he will swim'
irbens 'he will run'
igorebs 'he will roll'
icocebs 'he will crawl'
isrialebs 'he will slide'
i{rialebs 'he will turn'
im03ravebs 'he will move about in one place'
ipikrebs 'he will think'
upasuxebs 'he will answer him'
N. imepebs 'he will reign'

imusavebs 'he will work'
imecadinebs 'he will study'
iqaraulebs 'he will guard it'
ibavSvebs 'he will behave childishly'
ibebiavebs 'she will do midwifery'
igmirebs 'he will behave bravely'
ilotebs 'he will carouse'
imgzavrebs 'he will travel'
imorigevebs 'he will be on duty as a doorman, concierge'
iparisevlebs 'he will flatter, play the hypocrite'
kartvelobs 'he behaves like a Georgian'
o. i1!.as1!.asebs 'it will glisten'

i1!.rialebs 'it will shine'
iprialebs 'it will shine'
iduyebs 'it will boil'
i�etebs 'it will drip'
itkrialebs 'it will gush'
ilxrialebs 'it will jingle'
SAMPLE LIST OF CLASS 4 VERBS
P. uqvars 'he loves him'

S3uls 'he hates him'
mos�ons 'he likes it'
avi�qdeba 'he forgets it'
axsovs 'he remembers it'
esmis 'he hears, understands it'
egemeba 'it tastes . . . to him, he tastes it'
eqnoseba 'it will smell . . . to him, he will smell it'
akvs 'he has it'
hqavs 'he has him'
seu3lia 'he can, is able to . . .'
unda 'he wants it, wants to . . .'
spirs 'he needs it, needs to . . .'
u{irs 'he finds it difficult'
aklia 'he finds it lacking, necessary'
u/f.virs 'he finds it surprising, it surprises him'
surs 'he wants it, wishes for it'
st/f.iva 'it hurts him'
uxaria 'he finds it pleasing, he is pleased, happy'
hgonia 'he finds it . . . , it seems . . . to him'
esinia 'he is afraid of it'
rcxvenia 'he is embarrassed about it, he finds it embarrassing'
enatreba 'he misses him'
enaneba 'he regrets it'
Mia 'he is hungry'
sfquria 'he is thirsty'
sciva 'he is cold'
scxela 'he is hot'
Q. ece/f.veba 'he feels like dancing'
emyereba 'he feels like singing'
eyvineba 'he feels like having wine'
e3ineba 'he feels sleepy'
esxvapereba 'it appears different to him'
R. esinia 'he is afraid of it'
furs 'he is shy of it'
esmis 'he understands it'
268 Appendix B: Ambivalent exceptions
Appendix B : Ambivalent exceptions
The characteristics of Classes are of three types : morphological, syntactic,

and semantic (cf. Appendix A). These three types of characteristics generally
fall together, such that if a verb belongs to Class n according to morpho
logical criteria, it will have also the syntactic characteristics of Class n. This
is true of both the regular verbs and the few exceptions to the generalizations
stated in §2 to 4 of this chapter. In this appendix, I will deal with three types
of ambivalent verbs, giving the properties of their ambivalence and showing
that their classification by normative grammarians does not always reflect the
properties of the verbs, as they are actually used.
It is not at all typical for verbs in Georgian to be ambivalent about the
case Pattern they govern or about undergoing Inversion. The only ambivalent
verbs I know of belong to one of the very small non-productive groups dis
cussed here, or to the type discussed briefly in §2.4.
Ambivalent verbs in Georgian have the following properties : (i) The
Unaccusative Hypothesis predicts that they have the initial and final term
inventories characteristic of Class X, but they belong formally (morpho
logically) to Class Y. (ii) Syntactically, they may have the characteristics
(case Pattern, Inversion)of either Class X or Class Y. The ambivalence on the
part of these exceptional verbs seems therefore to reflect on the one hand an
adherence to the Class of which they are formally members, and on the
other hand the behavior that would be regular for verbs of that semantic/
syntactic type, as predicted by my analysis. Not surprisingly, normative
grammar classifies them as belonging only to the Class which they fit formally.
Three types are discussed below.
I T R A N S I T IVE V E R B S IN CLASS 2
This type has the following ambivalent properties : (i) My analysis predicts
that they would belong to Class I , since they have an initial subject and
direct object, the initial subject being also the final subject. But they have the
morphological characteristics of Class 2.28 (ii) Syntactically, they behave
either like Class I or Class 2 verbs. This group includes the following verbs :
mouqva 'he told it to him', seelf-itxa 'he asked him it', sehpirda 'he promised
it to him', seevedra 'he begged him for it' , and seexveca 'he asked him for it'.
An example of this group in Series I is provided in ( I ) .
( I ) mama motxrobas uqveba ojaxs.

father-NOM story-DAT he-tells-them-it- I-2 familY-DAT
'Father is telling a story to the family.'
We can see that these verbs are transitive through their behavior under
various rules. Under Causative Clause Union, for example, the nominals
behave as we would predict for an embedded subject, direct object, and
indirect object, respectively ; that is, the embedded subject is the indirect
object of the causative, the embedded direct object the direct object of the
causative, and the embedded indirect object is the retired indirect object of
the causative (cf. ch. 5) :
INITIAL Matrix 1 Embedded 1 Embedded 2 Embedded 3
CAUSATIVE 1 3 2 Retired 3
(2) illustrates the causative of ( I ) in Series II. Recall that all organic causatives
are in Class 1 and thus govern Pattern A in Series II. (2) shows that in the
causative of qola 'tell', each of the terms is case marked as expected for a
causative according to the chart above.
(2) ninom mamas moaqola motxroba ojaxisatvis.

NinO-ERG father-DAT she-caused-tell-him-it-II-1 story-NoM family-for
'Nino got father to tell a story to the family.'
In (2),motxroba 'story' is clearly marked as the direct object of the causative,

and mama 'father' as its indirect object. This confirms that the former is the
direct object and the latter the initial subject of the verb moqola 'tell'.
A second argument that these verbs have a subject and direct object comes
from the conventions of marking retired terms (cf. ch. I I ) . In masdars and
other constructions that have retired terms, retired subjects of transitives are
marked with mier. Retired direct objects are marked with the genitive case.
Masdars of verbs of this group show clearly that the verbs are transitive.
(3) . . . be{edis sepireba vanos mier . . .

ring-GEN promising-NoM Vano by
'Vano's promising (of) a ring . . .'
The fact that the nominal vano is marked with mier in (3) shows that it is
the retired subject of a transitive verb. The fact that be{edi 'ring' is marked
with the genitive in (3) supports the view that it is the retired direct object.
A further argument that these are transitive verbs comes from Passiviz
ation. In ch. 7 it was shown that Passivization is a rule that promotes a direct
object to subject. The fact that moqola 'tell' can undergo Passivization sup
ports the view that it has a direct object initially and is a transitive verb. The
passive of qola is illustrated in (4) .
(4) motxroba iqo moqolili myelvarebit.
story-NoM it-was-II-2 told excitement-INST
<The story was told with excitement.'
2 70 Appendix B : Ambivalent exceptions
Not aU of the verbs of this group can passivize, however. For example, there
is no Georgian sentence corresponding to ' ?\Vater was begged for' . 2 9
Thus, causatives, masdars, and passives provide evidence that these verbs
are transitive. My analysis predicts that their initial subject would undergo
Inversion in Series III, and that in Series II the initial subject, being final
subject, would be marked ergative. Yet these verbs unexpectedly have the
morphology of Class 2. Sentences (s) and (6) show that uqveba 'he tells it to
him' may have the syntactic characteristics of either Class I or 2 .
(s) Class I
(a) mamam ojaxs motxroba mouqva.

father-ERG family-nAT story-NOM he-told-them-it-II
'Father told the story to the family.'
(b) turme bits saxeli (se)u!ptxia gogostvt.'s.
apparently boy-nAT name-NOM he-asked-it-III girl-for
'Apparently the boy (has) asked the girl her name.'
(6) Class 2
(a) mama ojaxs motxrobas mouqva.
father-NOM story-nAT
'Father told the story to the family.'
(b) turme bip saxels sely.itxia gogos.
boy-NOM name-nAT he-asked-her-it- III girl-nAT
'Apparently the boy (has) asked the girl her name.'
In (sa) Pattern A is used, and in (6a) Pattern B. In (Sb) the verb has under
gone Inversion, in (6b) it has not. The difference in the verb forms in (S-6b)
corresponds to the different morphology required for verbs that have under
gone Inversion and verbs that have not.
I have shown that the verbs of this group are ambivalent in the following
way : (i) On semantic grounds, they should belong to Class I , but they have
the morphological characteristics of Class 2. (ii) Syntactically, they may
behave like Class I or 2 verbs ambivalently.
::>. A C T I V E I NT R A N S I T I V E S I N C L A S S 2
These verbs are ambivalent in the following way : (i) my analysis predicts
that they all have an initial subject, some an optional initial direct object ;
.
that is, they are Class 3 verbs. Morphologically, they have some character-
istics of Class 2 and some of Class 3. (ii) With respect to case marking in
Series II, these verbs may have the characteristics of Class 2 or 3, under cir
cumstances that are specified below. With respect to Inversion, they are
strictly Class 3 verbs. Verbs of this group include elapara�eba 'he is talking
Appendix B : Ambivalent exceptions 271
to him', etamaseba 'he is playing with him', estumreba 'he is visiting him',
esaubreba 'he is conversing with him' , and others.
Each of the verbs in this group corresponds to a regular Class 3 verb :
(7) Class 3 : ob form Class 2 ? : e-ebi form

laparalJ,obs 'he is talking' elaparalJ,eba 'he is talking to him'
tamasobs 'he is playing' etamaseba 'he is playing with him'
stumrobs 'he is visiting' estumreba 'he is visiting him'
saubrobs 'he is conversing' esaubreba 'he is conversing with him'
Those verbs which, like tamasobs, take an optional initial direct object in the
ob form (cf. ch. 1 2), may also take an optional initial direct object in the
e-ebi form. The e-ebi type differs from its ob counterpart only in that the
former, but not the latter, have an indirect object. so Based on the semantics
of the verb, we would posit four possible inventories of final relations :
(8) (a) (b)
tamas tamas- burt

' play ' ' he ' ' he ' ' baU '
(c) (d)
tamas- b13asvil tamas- burt- bi3asvil-

' he ' ' cousin ' ' he '
Based on the semantics of the verb, we would predict that each of the above
types represents a Class 3 verb. However, the first two examples are expressed
by ob forms, and the last two by e-ebi forms. It is not clear why the differ
ence between having or not having a final indirect object should mean a
difference in Class, especially as it does not elsewhere in the grammar
(cf. ch. 6).
Although normative gr ammars indicate that e-ebi forms belong to Class 2,
that does not tell the whole story. Let us consider each of the morphological
characteristics of Class listed in Appendix A, with reference to these verb
forms. The first (a) is not applicable, since these forms have no future distinct
from their present and form their aorists differently from any of the types
listed in Appendix A. According to characteristic (b), these verbs are strictly
in Class z, since they use -a/-an as the third person markers in the future
( = present) :
(9) elap aralJeba

(a)
(b) * eIapara.kebs
} ' he W1'11 talk to h1m
"
With respect to criterion (c), on the other hand, my informants accepted -es
(typical of Class 3) or -nen (typical of Class z) equally:
( 1 0) elaparalJes
(a)
(b) e Iapara.1men
}' they t. alked to h1m
" .
In addition to the morphological characteristics listed in Appendix A, the

use of the morpheme -od- or -d- distinguishes roughly between Class z and 3
respectively (cf. n. z6). While the ob forms cited above use -d-, the e-ebi
forms use only -od-, which is characteristic of Class z :
( I I) (a) *velapara�bdi } ' I was talking'.

(b) velaparalJebodi
Thus, the morphological characteristics are ambivalent, but mostly indicate

membership in Class z.
Syntactically, the balance swings the other way, With respect to case
marking, either a nominative or ergative subject is possible in Series II, as
shown by ( IZ),31
(IZ) (a) mas�avlebeli dianas esaubra kartulad.

teacher-NoM Dianne-DAT he-conversed-her-II-z Georgian-ADv
' The teacher conversed with Dianne in Georgian.'
(b) mas�avlebelma dianas esaubra kartulad.
teacher-ERG
'The teacher conversed with Dianne in Georgian.'
For some speakers, the ergative is used as subject case only when there is a
direct object, and the nominative only when there is not :
( 1 3) (a) gela melaparalJa.

Gela-NoM he-talked-me-II-z
'Gela talked to me.'
(b) *gela melaparalJa bevrsjbevri.
lot-DAT/NOM
(,Gela talked a lot to me.')
(14) (a) ·gelam melapara�a.

Gela-ERG
( ' Gela talked to me.')
(b) gelam melaparaka bevri.
'Gela talked to me a lot.'
In ( 1 3) the subject, gela, is in the nominative in Series II ; in (14) it is in the

ergative. In the (a) sentences, there is no direct object ; in the (b) sentences
bevr- 'a lot, much' occurs, which has the characteristics of a direct object in
Georgian. To summarize, there are at least three types of speakers: (i) ones
for whom the e-ebi forms may have their subject either in the nominative
or ergative in Series II, regardless of the presence of a direct object, (ii)
speakers who use the ergative with e-ebi forms when there is a direct object
and the nominative when there is not, (iii) speakers who follow the literary
norm in using only the nominative case for the subject of e-ebi forms in
Series II.
With respect to Inversion, the e-ebi forms behave strictly like Class 3
verbs, as shown in ( I S). In (I sa) Inversion has applied, in (· I Sb) it has not.
l I S) (a) gelas ulapara�ia iastan.

Gela-DAT he-talked-III-3 la-with
' Gela talked with Ia. '
(b) ·gela laparalJebulalslapara�ebia ias.32
Gela-NoM he-talked-her-III-z Ia-DAT
( ' Gela talked with Ia.')
Thus, syntactically, the e-ebi forms show more characteristics of Class 3

than of Class 2.
To summarize, this group of verbs is ambivalent both morphologically
and syntactically, showing some characteristics of Class 2 and some of
Class 3. My theory predicts that they belong to Class 3 . I have no explanation
for their Class-2-like behavior.
3 ARSBBOBS
The verb arsebobs 'it exists' is ambivalent in the following way: (i) On the
basis of the semantics of the verb, my theory predicts that it belongs to
Class 2. Morphologically, it is a member of Class 3. (ii) Syntactically, it
behaves like a verb of Class 2 or 3 ambivalently, under circumstances that
are stated below. Not surprisingly, normative grammars classify it, according
to its morphology, in Class 3 only.
(16) shows that arseboba 'exist' may trigger Inversion only if a location is
stated. Triggering Inversion is a characteristic of Class 3 verbs, not of
Class 2 verbs.
( 1 6) (a) ""devebs ar uarsebniat.

devis-DAT not they-exist-III- ?
(,Devis do not exist. ')
(b) devebs ar uarsebniat betaniasi.
Betania-in
'There are no devis in Betania.'
(Devis are mythological creatures ; Betania is a wood near Tbilisi.) In both
sentences of (16), Inversion has applied. This shows that in (I6b), the verb
has the characteristics of a Class 3 verb, but not in (""I 6a). (17) is the gram
matical counterpart of (""I 6a) ; here Inversion has not applied. This shows
that here the verb has the syntactic characteristics of a Class :2 verb.
(17) devebi ar arsebulan.33
devis-NoM they-exist- I II-:2
'Devis do not exist (in general).'
In a general statement of existence, beyond the control of the final subject,
arseboba behaves like a Class :2 verb. On the other hand, in a statement of
location, where the subject is in control (of its own location), arseboba is used
as a Class 3 verb, with the syntactic characteristic of that Class.
4 S U M M A R Y AND C O N C L U S I O N S
The purpose o f this appendix has been t o show that the strict classifications
of normative grammars and dictionaries do not always reflect the true use
of the verbs in the speech of present-day dwellers of Tbilisi. In each of the
examples cited above, my theory predicts that a particular verb belongs to
one Class, and normative grammars classify it as belonging to another Class.
The truth in each instance is somewhere in between.
Epilogue
In this epilogue, I will try to draw together related results that have
been discussed in separate chapters. In addition, I will give very brief
summaries of the most important conclusions reported here.
I The grammatical relations 'subject', 'direct object' and 'in

direct object'
Some linguists have expressed doubts that these notions play a role in
the syntax of a language like Georgian, where the apparent subject may
be marked with any one of three cases and may trigger either of two
sets of agreement markers. I have shown that a large number of rules
refer to these syntactic notions, regardless of the case of the nominal.
In particular, the rules of Tav-Reflexivization, Subject Person Agree
ment, Causative Clause Union, Inversion, Retired Term Marking, and
Number Agreement refer crucially to the grammatical relation subject.
The rules of Direct Object Person Agreement, Object Camouflage,
Object Raising, Causative Clause Union, Passivization, Retired Term
Marking, and Unaccusative are among the rules that refer to the gram
matical relation direct object. And the rules of Indirect Object Person
Agreement, Object Camouflage, Causative Clause Union, the several
Version rules, and Retired Term Marking refer to the grammatical
relation indirect object. For Georgian, none of these rules can be
stated in a simple way on the basis of case marking or linear order.
z Simplifications in case marking
In the Introduction and early chapters of this work, it was shown that
case marking appears to be very complex in Georgian. Three Patterns
were identified,
276 Epilogue
( I) Subject Direct Object Indirect Object

and their distribution was stated in terms of Series of tense-aspect

categories and Classes of finite verb forms :
(2) Series I II III

Class I B A C
2 B B B
3 B A C
4- C C C
In chs. 8 and 1 6 it is shown that significant simplifications can be

effected with two rules, Inversion and Unaccusative, whose inclusion
in the grammar is independently justified. It has been shown that
Inversion applies in Series III with Class I and 3 verbs and with Class 4-
verbs in all Series. This rule changes grammatical relations such that
Pattern B accounts for marking on final termhood in the inversion con
struction. This permits the simplification (1 '-2').
( I ') Initial Initial

Initial Subject Direct Object Indirect Object
(2') Series I II III

Class 1 & 3 B A B
2 & 4- B B B
The second simplification is effected by recognizing initial structures

with intransitive direct objects (direct objects not accompanied by
subjects) which advance to subjecthood. This permits ( 1 '-2') to be
simplified in two areas. First, the notion 'Class' need not be referred to
by the grammar. Second, the rules for term case marking can be stated
simply as
I If the verb is in Series I or III,

a. the final subject is marked with the nominative case.
b. the final objects are xnarked with the dative case.
Retired Term Marking 277
II If the verb is in Series I I,

a. an initial subject which is also a final subject is marked with
the ergative case.
b. a direct object which is a final nuclear term is marked
with the nominative case.
c. a final indirect object is marked with the dative case.
At the same time, these rules show that the term case marking rules in
Georgian are not ergative, in the generally accepted meaning of that
notion.
A retired term! is a nominal which bears the i-relation as its last term
relation, and which is a final non-term. At least six constructions in
Georgian involve retired terms : object-raised clauses, organic caus
atives, passives, inversion clauses, masdars, and infinitives of purpose.
In ch. II it is shown that all of the retired relations involved in these
disparate constructions can be accounted for with three simple rules :
I. A retired absolutive (= direct object or intransitive subject) is

marked with the genitive case.
II. A retired ergative (= transitive subject) is marked with the

postposition mier.
III. A retired indirect object is marked with the postposition tvis.
4 Rule interaction
In the investigation reported here, I have not assumed conventional

extrinsic rule ordering. Rather, I have attempted to make the minimum
specification of rule interaction that would allow all grammatical sen
tences while blocking all ungrammatical ones. In addition, I have not
assumed any rule interaction that has been shown to exist in other
languages, but have tried to state only what is supported by the data
from Georgian. The statements of rule interaction can be made most
simply on the basis of initial, final, and first termhood. This leads to the
typology below.
278 Epilogue
Rule Types in Georgian

I. Rules stated on initial termhood.
II. Unrestricted rules.
III. Rules stated on final termhood.
IV. Rules referring to termhood at more than one level (global rules).
Rules of the first type include the various rules of suppletion (intro
duced in §4.3 of the Introduction and shown in later chapters to be
stated on initial termhood). Type II includes Object Raising, Version,
Unaccusative, etc. ; these may apply whenever their input conditions
are met.
Type I I I includes the three rules of Person Agreement, some of the
rules of term case marking (d. above, §2), Unemphatic Pronoun Drop,
and Question Formation. These rules apply to the output of rules that
change grammatical relations.
Rules of the last type include Tav-Reflexivization, Number Agree
ment, and two of the rules of term case marking (na and lIb). The first
of these refers to initial subjects that are final terms, the second to first
subjects that are final terms. The third includes a rule referring to initial
subjects that are final subjects and a rule referring to direct objects that
are final terms.
There is one rule that does not fit neatly into this typology. Object
Camouflage applies to the output of rules that change grammatical
relations ; these rules must not apply to its output. But the rules of
Person Agreement and Unemphatic Pronoun Drop apply to the output
of Object Camouflage.
Evidence to support the conclusions stated here is given above in
various chapters.
5 Characteristics of grammatical relations
The investigation reported here takes into account a large body of facts
and shows the systematicity and consistency in apparently disparate
rules. The data assembled here do not support the notion of a 'squish'
of grammatical relations (cf. Keenan (I976) and Johnson ( 1 977) for two
sides of this problem). The system analyzed here supports instead the
idea of a distinction between levels of derivation (or initial and final
termhood) and rules which take these into account, in something like
the manner summarized in §4. Taken as a whole, the system worked out
Characteristics of grammatical relations 279
here shows that nominals in a particular syntactic construction do not

arbitrarily partake of some characteristics of one grammatical relation
and other random characteristics of another grammatical relation. For
example, an inversion nominal does not merely have some properties
of a subject and other properties of an indirect object. Rather, on the
basis of facts like those summarized in the preceding section, it is
absolutely predictable which subject and which indirect object charac
teristics a nominal will have.
Notes
Introduction
1
A problem with Chikobava's and Aronson's analysis is noted in ch. I S , n. I I .
2 Johnson ( 1977) discusses the difficulties of arbitrarily selecting criteria which would
define the notion 'subject'.
3 Some traditional works do recognize levels of derivation to a certain extent. They
distinguish, for example, between 'real subject' and 'grammatical subject' or

'morphological subject'. One example of this is Chikobava (1968 : especially 1 3 6 ff).
, Class I corresponds to the traditional group of gardamavali or aktiuri zmnebi. Class 2
corresponds to the group traditionally called vnebiti, gardauvali, and some of those
called medi�iun·. Class 3 contains most of the verbs traditionally called saIualo
zmnebi. Class 4 is usually called inversiuli zmnebi. The problems associated with the
traditional terminology are discussed in detail in chs. 1 2 and 1 6 and in Harris (to
appear b).
6 Diachronically diticization played a much more extensive role in the derivation of
finite verb forms.

8 The word unda 'should, would, might, must' when it occurs with the optative also
must intervene in this order : Q-word, NEG, undo, VERB. It is not clear whether undo
is synchronically a verb (auxiliary or modal) or a particle in this use.
7 'Emphatic -a' is added to personal pronouns in sentence-final position, as in len-a,
and under certain other circumstances.
s Both (.I6b) and (*I 6c) can have a grammatical reading. Each requires a heavy
pause to separate figni from the rest of the sentence, and is equivalent to left- and
right-dislocation in English, in that the isolated element represents a fore- or after
thought with respect to the sentence as a whole.
D Relative clauses are an exception to this generalization. When the head of a relative
clause is questioned , the relative clause is necessarily separated from its head, as
shown in
vis knob, vine mulaobs bibliotel!ali ?

who you-know whO-RilL works library-in
,
'Who do you know that works in a library?
·vis, vine muIaobs bibliotel!ali, icnob ?
·vine muIaobs bibliotel!ali, vis icnob ?
10
It might alternatively be proposed that the suppletion described in §4.3.I and 4.3.2
is conditioned by the animacy/number of the semantic patient, rather than that of
the direct object. These positions are not distinct, however, since in chs. 4, 5 , 7.
and 8 it is established that it is the initial direct object that governs suppletion, and
in ch. 16 it is shown that all semantic patients are initial direct objects.
Notes to Chapter I 281
11 Although prescriptive norms call for the use of jdoma only with singular subjects,
in fact it is used Vl<ith both singulars and plurals. Sxdoma, however, is used only
with plurals, as my examples show. Cf. Shanidze (1 973 : 5 01-2) ; Tschenkeli (1958 :
z86-7) ; Tschenkeli (1960-73).
Chapter :l
1 The dialect described here and referred to throughout this monograph is the more
restrictive of two dialects I identified with respect to Tav-Reflexivization. Both are
described in Harris (ms.).
2 These same forms occur as emphatic pronouns and as camouflaged objects (cf.
ch. 3 for a complete treatment of the latter).

3 For some speakers, the form arfmunebs is preferred here. The meaning and gram
matical relations are the same as in (Ia).

4 The sentences below show that it is really the subject that is triggering Tav
Reflexivization, and not the nominal with which the speaker empathizes (cf. Kuno
1975 ; and Kuno & Kaburaki 1 977).
(i) geZas sma eZaparalj,a mas tavis tavze.
Gela-GEN brother-NOM he-talked-him-II-z him-DAT self's self-on
'Gela'sJ brother! talked to himJ about himsel6.'
(ii) akamde ara'Vin ar elaparaf!;eboda ninos tavis tavze.
until-now no one-NOM not he-talked-her-I-z Nino-DAT self's self-on
'So far, no one! has talked to Ninoj about himself!.'
If empathy, rather than subjecthood, controlled Tav-Reflexivization, tav- should
refer to geZa in (i), since the double occurrence of this nominal in this way has
established the speaker's empathy with geZa, not with geZas 3mas 'Gela's brother'.
Similarly, in (ii) the spesker would not empathize with ara'Vin 'no one' ; yet this
triggers tav-.
5 See Vogt (197z) for a discussion of this phenomenon in Old Georgian.
a For many speakers, there is an additional constraint that tavis- cannot occur in the
subject. For these speakers, both (iii) and (iv) are impossible.
(iii) "'!viZs bans tavisi deda.
child-DAT she-bathes-him-I-l self's mother-NOM
('His! mother, bathes the child!.')
(iv) "'tavisi deda bans !viZs.
Some speakers have additional constraints on this rule.
Tavis- may precede its antecedent (cf. (29-30) of ch. 8 of Harris 1976), just as
tav- may (cf. example (3a-C». It is more natural for the reflexive to follow its
antecedent.
7 The rules of Person Agreement must introduce, not specific morphemes, but
abstract elements : first person subject, second person subject, etc. Rules that
specify the precise form of subject markers must be sensitive to grammatical
relations, person, number, tense-aspect, Class, and morphological sub-group of
the verb. A detailed proposal for agreement rules that make direct reference to
abstract features of this type is given in Hale (1973).
8 The t of the third person plural indirect object marker is usually not triggered (cf.
ch. IS); it is stated here on the basis of examples like ( I s a) of ch. I S.
Q The morpheme t (plurality marker) deletes under conditions that are specified in
ch. I S .
282 Notes to Chapter 3
10 Dafera can take a derived indirect object, as shown in ch. 6. This indirect object
differs from that used with mifera/mofera in meaning and in the morphology it
requires in the verb form.
11 The notion of dropped unemphatic pronouns is similar to Vogt's 'catalysts' (Vogt
1971) and to the practice common among Georgianists of listing term pronouns
after verbs, e.g. fers is mas mas 'writes he-NoM he-DAT he-DAT'.
12 (!lZ7b) would not be ungrammatical in those dialects where the third person indirect
object marker is always zero.
13 (+38b) has a grammatical reading in English, where and is a clausal conj unction ;
the Georgian example has no grammatical reading.
14 Unemphatic dropped pronouns are distinct from unspecified nominals. The former
are used when speaker and hearer know the identity of the referent of the nominal.
The latter occur when either the speaker or hearer or both do not know the identity
of the nominaL They occur in sentences like (v).
(v) amboben, rom sakarroelo lamazia.

they-say-it-I-I that Georgia-NoM pretty-it-is-I-z
'They say that Georgia is pretty.'
Like its English counterpart, (v) may in context mean either that some particular
people (referent identified) say this, or that in general it is said (referent not ident
ified). In the latter case the nominal is unspecified ; in the former, it is an unemphatic
anaphoric pronoun.
Chapter 3
1 Sentences of this type in Series I are disambiguated by word order.

2 The (a) example, which is in Series I, is grammatical in the interpretation where
givi is direct object. This is because, as we have seen in chs. 1 and z, the two objects
are both marked with the dative in Series 1. The (b) example, on the other hand,
has no grammatical reading. The (b) example is in Series II, where the direct and
indirect object are case marked differently.
3 If §edareba 'compare' occurs in a clause in which the indirect object is not realized
on the surface, the clause is interpreted as having had its initial indirect object
dropped by Unemphatic Pronoun Drop. This is shown by the fact that (+i) without
Object Camouflage is ungrammatical; while (ii), where Object Camouflage has
applied, is gr ammatical.
(i) !lvano (me) gadarebs.

Vano-NoM I-DAT he-compares-you-him-I-I
('Vano is comparing me to you.')
(ii) vano lems toos gadarebs.
my self-DAT
'Vano is comparing me to you.'
In (i-ii) the dropped indirect object triggers Object Camouflage obligatorily. This
constitutes additional evidence for a rule of Unemphatic Pronoun Drop, since
without such a rule the ungrammaticality of (+i) would be unexplained.
, The person of the indirect object does not affect the rule in the dialect reported
here. In (Z-3) and (5-6), the indirect object is third person ; in (iii) below, it is
first person, and in (iv) second.
Notes to Chapter 4 283
(iii) (a) ·vanom (sen) semada1'a (me).
YOU-NOM he-compared-me-him- I I - r me-DAT
('Vano compared you to me.')
(b) vanom seni tavi semadara (me).
'Vano compared you to me.'
(iv) (a) ·vanom (me) segadara (sen).
me-NOM he-compared-you-him-II-r you-DAT
(b) vanom cemi tooi segadara (Sen).
'Vano compared me to you.'
(1) shows that this rule does not apply to third person direct objects with a third
person indirect object. (v), which differs from (1b) only in the person of the indirect
object, shows that the third person direct object is unaffected, regardless of the
person of the indirect object.
(v) vanom anzori segadara (sen).
'Vano compared Anzor to you.'
There exists another dialect in which Object Camouflage applies only if the
clause contains an indirect object of the first or second person. That dialect is
described in Boeder ( 1 968 : § I .6) and in Harris ( 1 976 : Appendix to ch. 3). I found
that dialect only among speakers not native to Tbilisi.
5 The phrase tooisi tooi 'self's self' occurs, but only as a reflexive (cf. ch. r, § I ) .
Therefore,
vanom tavisi tam semadara (me).

VanO-ERG self's self-NOM he-compared-me-him-I I - I I-DAT
means, not 'Vano! compared himj to me', but 'Vano compared himself to me'.
6 'Personal pronouns' must here be understood to include es, igi, and is, which also
function as demonstrative pronouns.
Chapter 4
1 I found one older speaker who did not have Object Raising at all. All of my younger
informants did have the rule, though object-raised constructions do not occur
frequently.
2 I have glossed the masdars with English -ing forms ; this is not intended as a claim
that the two are the same syntactically.
3 A refinement of this analysis is made in ch. 1 6.
, See Introduction, §4.1 on Aris-Cliticization. An optional form of ( I I ) has aris cliti
cized to ra-
raa Jneli mosaJebnad?
5 While ( I I ) is related to (4b), (i) is related to (43).

(i) rlS! mOJebna aris Jneli?
what-GEN finding-NoM it-it-l-:a hard
'What is the finding of hard ?'
The genitive in (i) cannot occur with the infinitive, as predicted by the analysis
in §:a:
(ii) ·ris(i) mosaJebnad aris Jneli ?

what-GEN to-find
('What is it hard to find ?')
e (iii-v) establish that predicate nominals are marked with the nominative case in
Series I-III, respectively.
(iii) ekimia.
doctor-NOM-he-is-I-z
'He is a doctor.'
(iv) ekimi iqo.
he-was-II-z
'He was a doctor.'
(v) ekimi qopila.
he-was-III-z
'He must have been a doctor.'
Chapter 5
1 The morphology of organic causatives is further described in Getsadze et ai. ( r 969) ;

Schmidt ( 1 966) ; Shanidze ( 1 973 : §4Z3) ; Taqaishvili ( 1 974) ; Tschenkeli ( 1 95 8 :
vol. I , ch. 30) ; and Vogt ( 1 97 r : §z.70-z.73).
2 (3) is an oversimplification in that it states the effect of Clause Union only on
causative constructions. A more general formulation would encompass other
instances of Clause Union as well. See Aissen & Perlmutter ( 1 976), Frantz (r976),
and page Z41 for examples of Clause Union in constructions other than the
causative.
3 The argument given here is of the same structure as that given for Japanese in
Kuno ( I 973) and McCawley (r97z).

4 The following sentence shows that a nominal that is an initial, but not final, subject
can be the target of Tav-Reflexivization :
(i) vanom alaparaka taVISJ tavi sxvaze.
VanO-ERG he-caused-talk-him-II-r self's self-NOM other-on.
'Vano made himself talk about someone else.'
fi The fact that Object Camouflage applies to the output of all rules that change
grammatical relations correctly predicts that this rule and Object Raising cannot
apply in a derivation. (ii) bears out this prediction.
(ii) ·seni tavi 3ne1ia givistan sesadareblad.

your self-NOM hard-he-is-I-z Givi-with to-compare
(,You are difficult to compare with Givi.')
(iii) (sen) 3neli xar givistan sesadareblad.
YOU-NOM hard you-sG-are-I-z
'You are difficult to compare with Givi.'
After Object Raising applies, there is no direct object ; the input conditions for
Object Camouflage are therefore not met, and the rule cannot apply.
8 That Causative Clause Union can apply interative1y is shown by (iv), which is the
causative of the causative sentence (r I).
(iv) bebia mamas acveJlinebs gelas

grandmother-NoM father-DAT she-causes-cause-lie-him-him-I-r Gela-DAT
tax!ze.
couch-on
'The grandmother makes the father let Gela lie on the couch.'
But speakers dislike a double causative if it creates a retired tenn ; this happens, of
course, if the lowest clause of the would�be double causative has two or more
terms. For this reason, (v b), the causative of (v a), is unacceptable.
(v) (a) vanom daaferina mdivans ferili.
VanO-ERG he-caused-write-him-it-II-1 secretary-DAT letter-NOM
(b) .uprosma vanos {daafer�nebina }

'Vano had the secretary write a letter.'
ferili
daaferzna
bOSS-ERG Vano-DAT he-caused-cause-write-him-it-I I-1 letter-NOM
mdivnis mier/tvis.
secretary by for
('The boss made Vano have the secretary write a letter.')
The double causative fonn daaferinebina is said to exist 'in principle', but informants
will not use it in a sentence that contains all case-marked nominals (cf. Shanidze
1 973 ; and Vogt 1 97 1 : §2.79).
7 This example is quoted in Vogt ( 1 97 1 : §2.7S), from Chikobava ( 1950).
8 Some speakers do not accept these forms as double causatives at all. For such
speakers, Iemaimevina means 'he fed it to me'; that is, it is a single causative. Those
speakers I interviewed who did accept this fonn as a double causative considered
it ambiguously a single causative. (See also n. 6.)
9 The comparison of (vi) with (35) shows that this nominal does not trigger Person
Agreement or Number Agreement.
(vi) dedam tkven papa Iema{mevina.
mother-ERG yOU(PL)-DAT gruel-NOM she-caused-feed-me-it-II-1
'The mother made me feed gruel to YOU(PL).'
The fact that changing the initial embedded indirect object to second person plural
causes no change in the verb shows that this nominal fails to trigger both agreement
rules. It can also be seen in Shanidze's and Vogt's additional examples that in other
dialects this nominal fails to trigger the agreement rules.
If tkven 'YOU(PL)' is omitted from (vi), the meaning is changed :
(vii) dedam papa Iema{mevina.
'The mother fed me gruel.' (Cf. n. 8.)
This shows that the nominal cannot undergo Unemphatic Pronoun Drop.
10
In Old Georgian, retired indirect objects in all constructions were regularly marked
with the dative, as in these idioms (Harris 1 979). See ch. 1 1 for marking of retired
indirect objects in Modem Georgian.
Chapter 6
1 There is confusion in the traditional tenninology. I use the word version only with
reference to rules that create indirect objects. The marker that they trigger in the
verb I call 'version vowel' or 'version marker'. It is important to distinguish this
from other uses of the so-called 'character vowel', or vowel prefix which occurs
immediately before the verb root. Character vowels are used as (a) markers of
version, (b) parts of the productive markers of synthetic passives, such as icvleba 'it
changes' (cf. ch. 1 3), (c) parts of the markers of certain tenses including future and
aorist, for Class 3 verbs, such as imuIava 'he worked', (d) parts of the markers of
causatives, such as antebinebs 'he causes him to light it', (e) parts of the markers of
286 Notes to Ckapter 6
Series III forms for verbs of Class I and 3, such as dauferia 'apparently he has
written it' (cf. ch. 8), and (f) morphologically empty markers which are attached to
finite forms of some verbs, e.g. ifqebs 'he begins it', icinis 'he is laughing'.
2 The same is true in the plural :
(i) fe-gv-i-1!.era.
'He sewed it for us.'
(ii) fe-g-i-1!.era-t.
'He sewed it for YOU(PL).'
(iii) fe-u-1!era.
'He sewed it for them.'
(iii) is not distinct from (7c) ; that is, plurality is not marked in the third person.
This is explained in ch. I S.
s Examples (*'13) and (*'14) are respectively (*'34) and ("36) from Holisky (1978) ;
transliteration and glosses have been changed to correspond to the conventions
adopted here. The verb in (*'I3-14a) is highly irregular and not easily assigned to
a Class.
4 In Harris (1976) I proposed a general constraint on Version, such that it cannot
create a chomeur. While this constraint is supported by many examples, there are
exceptions to it. In (iv), gela has been put en chomage by the advancement o f 'me',
as shown in (v). Gela is marked with the dative case in (iv a) and with a positional
in (Iv b). (As discussed in §6, indirect object chomeurs may be in the dative case
in non-standard dialects.) Both sentences are marginal, but each is accepted by
some speakers. The grammaticality of these examples depends on the version
nominal being first or second person (cf. ch. 5, §4.2 for a similar situation).
(iv) (a) mdivanma rmmtfera ferili gelas.

secretary-IlRG he-wrote-me-it-II-I letter-NOM Gela-DAT
'The secretary wrote a letter to Gela for me.'
(b) mdivanma mimifera ferili gelastan.
Gela-at
'The secretary wrote a letter to Gela for me.'
(v)
ferili gek
wr
ite , secretary '
' '
'letter ' ' me '
While there are exceptions like (iv), some form o f this constraint must b e included
in the grammar to account for the non-occurrence of (I sC) and many other examples.
fi The counterpart of (*'I7b), with the version vowel i- instead of u- is also ungram
matical :
(vi) *'gela (lovis) lovs i1!.eravs axal farvals.

('Gela is making new trousers for himself. ')
6 There is no specific morpheme that corresponds to the gloss 'self'. The meaning
'self' is indicated in Georgian by the version marker i- without a marker of Person
Agreement.
7 It can also happen that the version nominal, the subject of its clause, and the
direct object are all coreferential. In such a case, the version nominal and direct
object are reflexivized, as in (vii).
(vii) �lavs tavis tavs.

he-kills-him-I-1 self's self-DAT
'He kills himself.'
Version may optionally apply, with Coreferential Version Object Deletion, giving
(viii) i�lavs tavs.

he-kills-self-him-I-1 self-DAT
'He commits suicide.'
The obligatoriness of Coreferential Version Object Deletion blocks (ix).
(ix) -iklavs tavs tavs.

('He commits suicide.')
Thus, the rules already proposed correctly account for these phenofl,1ena.
8 There is no form -mivi"ere ('I wrote a letter to myself') corresponding to (19).
While ilapara�a and imyera, corresponding to (20) and (:21) exist as forms, they
do not mean 'he talked to himself', 'he sang to himself', but simply 'he talked' and
'he sang', respectively.
9 There is no -gairmendina.
10 Among the analyses known to me, only Boeder ( 1 968) and Nebieridze (1976) have
recognized that synchronically sataviso kreva is a special instance of Benefactive or
Possessive Version and that the i- is a marker of Benefactive or Possessive Version,
whether or not there is coreference between the subject and version nominal.
1 1 This discussion has been limited to Benefactive Version for the sake of simplicity.
In fact, Coreferential Version Object Deletion also applies to indirect objects

created by Possessive Version. When the latter is included, there are two additional
arguments against the Extended Traditional Analysis :
A. Rule (18) is stated with respect to a 'reflexive nominal'. It could be stated
instead with respect to the coreferential clausemate of the subject, but to formulate
it in that way would unnecessarily repeat the conditions on full pronoun reflexi
vization (Tav-Reflexivization). Rule (23), if revised to include Possessive Version,
could not make the same economy. Rule (18) refers crucially to an indirect object,
but the revised version of (23) would refer to a possessor or benefactive. We have
seen in ch. I that the conditions on possessive reflexivization ( Tavis-Reflexivization)
are different from those on full pronoun reflexivization ; in particular, pqssessive
reflexivization can be triggered by any term. Therefore, to state the revised version
of (23) with respect to a 'reflexive possessor' would permit a possessor that was
coreferential with the direct object to delete. This would predict that (x b) was
related to (x a) ; but (x b) is ungrammatical in the intended meaning.
(x) (a) avadmqopi s�udenti gavgzavne tavis ekimtan.
ill student-NoM I-sent-him-II-I self's doctor-at
'I sent the ill studenti to his! doctor.'
(b) -avadmqopi s�udenti gavigzavne ekimtan.
('I sent the ill student; to hisi doctor.')
To state the revised version of (23) with 'reflexive' therefore would constitute the
false claim that (x b) and similar sentences express coreferentiality. Therefore (23)
must be restated as (xi), which is more complex in this sense than ( I S) and consti
tutes an argument against the Extended Traditional Analysis.
(xi) Coreferential Benefactive/Possessor Deletion

A benefactive or possessor that is coreferential to the subject of its clause
deletes.
B. Both grammars must contain a condition that Possessive Version, which is
otherwise optional, is obligatory if the relation whole-part obtains with respect to

the possessor-possessed (cf. Harris 1976 : ch. 6, §z.z). This condition blocks the
derivation of (*xii a) from an initial structure like that necessary for (xii b).
(xii) (a) *v(u)ban bavSvis xels.

I-wash.(him}-it-I.I child-GEN hand-DAT
('I am washing the child's hands.')
(b) vuban xels bavSvs.
I-wash·him-it-I-I child-DAT
'I am washing the child's hands.'
« xii a) has a grammatical reading from another source, corresponding to 'I am

washing his child's hands'.}
The Extended Traditional Analysis must include, in addition to this condition
on Possessive Version, the same condition on the revised deletion rule, (xi). Such a
condition is needed to block the derivation of (*xiii a), since deletion must be an
optional rule, as shown above.
(xiii) (a) ·v(i)ban lems xels.
l-wash·(self)-it·I-1 my hand-DAT
('I am washing my hands.')
(b) viban xels.
'J am washing my hands.'
Thus, the Extended Traditional Analysis requires an additional constraint on

deletion ; this duplicates the constraint independently required for Possessive Ver
sion. The analysis proposed here, on the other hand, requires no additional con
straint on deletion. (*xiii a) will be blocked by the constraint on Possessive Version
and the independently necessary statement that ( IS) is obligatory.
1 2 In the Thilisi dialect this promotion of me 'me' would be blocked by the constraint
discussed in n. 4.
13 This can be seen in other examples, such as (xiv) ; here gv- is the first person plural
marker.
(xiv) ' • rom migvifero
• . ferili.'
that you-write-us-it-II-I letter-NOM
' . . . that you may write a letter for us.'
This is quoted in Shanidze ( 1973 : §4oz), from Vazha Pshavela, an author who wrote
in the Pshav dialect.
1 4 Seiter (1979), however, presents data that pose a serious challenge to both Laws.
Chapter 7
1 Direct construction is used in this work instead of active, in order to avoid the mul
tiple ambiguity of the latter.
2 Passivization is an optional rule. If a clause undergoes the rule, the verb it contains
is a Class 2 verb. If the same clause does not undergo the rule, the verb it contains
is a member of another Class.
3 The formation of the past passive participle, but not its use in the passive con
struction, is described in Chikobava ( 1 950), Shanidze (I 973), Tschenkeli ( 1958),

and Vogt ( 1 97 1 ) .
4 These scholars d o not agree about how a passive is derived. However, this dis
agreement is irrelevant to the question of whether (1-2b) are passives, since they
have properties which all of the works cited consider characteristic of passives.
5 Both fems and cem (.fem and .fen, etc.) are used with mieT; cf. Tanamedrove kartuli
sa/itera,uro enis normebi ( 1 970), pp. 1 1 7-8.

8 I am grateful to Hans Vogt for calling this fact to my attention and for supplying
me with several examples from literature.
7 The notion 'chomeur' was first introduced by Perlmutter and Postal (cf. Perlmutter
& Postal 1 974). Some particular studies of their occurrence with various advance
ment rules are reported, for example, in Bell ( 1974. 1 976), Chung ( 1 976a),
Perlmutter (1 978), Perlmutter & Postal ( 1 977), Postal ( 1977), and Sheintuch (1 976).
The ability of chomeurs to trigger agreement is discussed in Allen & Frantz ( [ 978)
and Lawler ( I 977).
8 I have no data concerning the interaction of Superessive Version and Passivization.
Chapter 8 and Appendices

1 Peikrishvili ( 1 974), a recent work on the evidential, lists, among other uses,
a. The evidential may express an action which was not witnessed directly by the
speaker.
b. The evidential may express an action which is retained as a memory in the
mind of the speaker.
c. The evidential may express the simple negative of a past action (the corre
sponding positive being in a non-evidential past).
d. The evidential may express the result of an action rather than the action itself.
e. The evidential may express the presumption on the part of the speaker that
the action took place.
2 The general analysis of Inversion which follows is due to Postal and Perlmutter.
This analysis is discussed for other languages in Perlmutter (t<) appear b) and in
Harris (to appear a).
a Aronson ( 1 976) gives a more detailed summary of various positions that have been
taken on this issue. See also individual works, Chikobava ( 1 950, 1 968), Deeters
(1 930), Lafon (1 963), Marr & Briere ( 1 93 1 ), Shanidze ( 1961 , 1 963. 1973), Tschenkeli
( 1958), and Vogt ( 1971).
, The examples in (9) are from the second evidential or 'pluperfect'. The use of the
subject markers in the third evidential ('third subjunctive') is the same:
(i) (a) gavegzavno kalakIi.
he-sent-me-III-I city-in
'May he send me into the city.'
(b) gaegzavno kalakIi.
he-sent-you-III-I (zero marker)
'May he send you into the city.'
(c) gaegzavnos kalakIi.
he-sent-him-III-I
'May he send him into the city.'
(The third evidential is restricted in use in colloquial Standard Modern Georgian.)
The markers used in the first evidential (or 'perfect') are different from those in (i)
and (9) :
(ii) (a) ga(v)ugzavnivar kalakSi.
he-sent-me-I I I - I
'Apparently h e (has) sent me into the city.'
(b) gaugzmmixar kalakli.
he-sent-you-III-I
'Apparently he (has) sent you into the dty.'
(c) gaugzavnia kalakli.
he-sent-bim-III-I
'Apparently he (has) sent him into the city.'
The suffixes in heavy type in (ii) are taken from the verb 'be ' : var 'I am', xar 'you
are', -a the unemphatic enclitic form of aris 'he/she/it is'. Synchronically, these
represent a secondary set of subject markers, used (a) here in the first evidential,
(b) with some of the verbs discussed in §4, and (c) with a few other verbs, such as
mivdi-var 'I am going', vdga-var 'I am standing', v/iri-var 'I am crying', etc. In
all instances this secondary set of agreement markers, like that in (8), indicates the
person and number of the subject. Thus, in spite of the use of a secondary set of
markers, it is clear that in (ii), as in (9) and (i), the nominative-nominal triggers
Subject Person Agreement.
Ii The examples in ( I I ) are from the second evidential. The use of indirect object
agreement markers in the third evidential is the same :
(iii) (a) dametesos simindi.
I-sowed-it-III-I corn-NOM
'May I sow com.'
(b) dagetesos simindi.
you-sowed-it-I II-I
'l\1ay you sow com.'
(c) daetesos simindi
be-sowed-it-III-I (zero marker)
'May he sow corn.'
The markers used in the first evidential are similar, but in the third person we find
a variant of that noted in ( 1 0) ;
(iv) (a) damitesavs simindi.
I-sowed-it-III-I
'Apparently I sowed corn.'
(b) dagitesavs simindi.
you-sowed-it-I I I- I
'Apparently you sowed corn.'
(c) dautesavs simindi.
be-sowed-it-III-I
'Apparently he sowed corn.'
Notice that in the first and second persons there is a morpheme i, which is part of
the stem formant of the first evidential ; this morpheme immediately precedes the
verb root and immediately follows the indirect object markers. It is quite generally
the case in Georgian that the third person indirect object agreement marker (5/h/o)
combines with the formant i, the two morphemes together being realized as u. The i
formant occurs in a variety of functions, which include its being (a) part of the marker
of the first evidential, (b) part of the marker of future and aorist tense groups for one
class of verbs, (c) the marker of promotion of a benefactive or possessive to indirect
obj ect (cf. ch. 6, §Z.I), and (d) an empty morpheme that occurs in a few verbs. In
all these functions, the formant i combines synchronically with a third person
indirect object marker as 11.. Examples (7) and (9) of ch. 6 illustrate functions (c) and
(d) respectively. Thus, in spite of the occurrence of the 11. variant in the third person,
we can see that the first evidential in (iv) contains regular object markers, just as
(II ) and (iii) do.
6 The fact that Object Camouflage never applies in Series I I I with Class I and 3
verbs, which trigger Inversion, is accounted for by the fact that the input conditions
are not met, as stated. Obj ect Camouflage never applies in Series I I I with Class z
verbs either, and these verbs do not trigger Inversion. However, there is no Class z
verb that meets the input conditions for Object Camouflage, and therefore the rule
never applies with a Class z verb in any Series. To understand why no Class 2 verb
satisfies the input conditions for Object Camouflage, it is necessary to recognize
that the Class z verbs which take a direct and indirect object are a tiny, irregular
set (cf. ch. 16, Appendix B). They are all verbs of 'saying' and therefore do not
take first or second person direct objects:
mama motxrooas 11.qveba ninos.

father-NOM story-DAT he-tells-her-it- I-z Nino-DAT
'Father is telling a story to Nino.'
IJI;mama me/cems tavs 11.qveba ninos.
I my self-DAT
(IJI;'Father is telling me to Nino.')
Without a first or second person direct object, the input conditions for O bj ect Camou
flage are not met. Thus, the proposal made above accounts for the failure of Object
Camouflage to apply in Series I I I , with Class z verbs as well as with Class I and 3
verbs.
7 In both types, Unaccusative applies only if its input conditions are satisfied, of
course. This is discussed below in §6.
B Not all traditional descriptions of Georgian take the position summarized in (27-28).
Shanidze (1 963), for example, takes the view that the relations indicated in (Z4) are
the only grammatical relations. From this position, case marking could be simplified
as in (Z7'-28'), but the facts observed in §4. 1 , 5 . 1 , and Appendix A cannot easily
be accounted for.
9 This statement is refined in ch. 16.
10
Alternative forms for some speakers are moamIives and mOQmIies; the meaning and
grammatical relations are the same as in (29b).
11 In the causative of a transitive inversion verb, the experiencer is the derived indirect
obj ect, and the stimulus is the derived direct object. This shows nothing, however,
since the derived indirect-objecthood of the experiencer could arise either from its
being the subject of a transitive or from its being an indirect object. Similarly, the
final direct-objecthood of the stimulus could arise either from its being a direct
object or from its being an intransitive subject. Since these are precisely the pairs of
relations that are in question, the causative of a transitive inversion verb sheds no
light on the problem.
12
There is a systematic ambiguity in examples of this type, which derives from the
fact that tvis not only marks retired indirect objects, but also marks benefactives.
Thus (30b) could also be glossed 'Apparently the king has had the palace cleaned
292 Notes to Chapters 9-IO
for you,' with the initial subject of 'clean' left unspecified. The alternative word
order, tUNne mepes gaupnendinebia sasaxle Ient'Vis is also, for many speakers, accept
able and ambiguous.
13 Example (45) below appears to provide additional support for the initial direct
objecthood of the unaccusative nominal, since it undergoes Object Raising, which

is restricted to direct objects. However, as discussed in Harris (1976 : 249-50), it
might be argued that the infinitive in (45) is derived, not from the Class 4 verb in
(42), but from a Class 1 verb with the same root. If the latter were correct, it would
not be a counter example to any part of the analysis presented here. But because of
this ambiguity in the fonn, (45) and similar examples do not provide strong support
for my analysis, and are not discussed in detail from that point of view.
14 This differs slightly from Perlmutter and Postal's analyses of similar data in other
languages, cf. Perlmutter (to appear b) and Perlmutter & Postal (to appear c).
15 The inability of Passivization to apply to the output of Inversion, on the other hand,
is accounted for automatically by the fact that the output of Inversion (or Inversion
and Unaccusative) fails to meet the input conditions on Passivization. Passivization,
as stated above in §6, may only apply in clauses containing a subject and direct
object. The output of Inversion (only) contains no subject. The output of Inversion
and Unaccusative contains no direct object. Therefore, it is correctly predicted that
Passivization cannot apply to the output of Inversion. The ungmmmatical sentences
that would result cannot easily be illustrated, since the fonns are lacking.
The same is true of the other interactions discussed in §7.2-7.5, as the reader
can confinn for himself.
16 This example is from Tschenk�li (1958 : vol. II, 268, no. 3).
17 At least the following verbs require a genitive under certain circumstances : mjera 'I
believe it', mesmis 'I hear, understand it', melinia 'I fear it', mrfams 'I believe in it',
mrcxvenia 'I am ashamed, shy of it', mexatreba 'I am shy of it', mSurs 'I envy it'.
1 8 It is not clear under exactly what circumstances the complement must be genitive.
In (x) both animate and inanimate nominals must be in the genitive, but cf.
mjera Ieni fasen.

I-believe-it-I-4 YOU-GEN YOU-NOM
mjera faambavisaf amhavi.

newS-GEN newS-NOM
'I believe the news.'
Chapter 9
1 Analyses similar to this have been proposed by several scholars, including Marr
(1925) and Schuchardt (1895). Chikobava (I96r) considers in detail this analysis of
Georgian and several others and argues eloquently in traditional terms that the
ergative-nominal in Series II is both the 'real' (initial) subject and the 'morpho
logical' (final) subject.
Chapter 10
1 The name 'masdar' is used here as it is traditionally in Georgian linguistics; masdars

are also called sacqisebi 'beginnings'. These deverbal forms do not have all of the
characteristics of the masdar in Arabic and related languages.
2 Shanidze (r973 : 565) suggests that "a�ni and saubari are not masdars, but are used
Notes to Chapter IO 293
as masdars ; his distinction seems to be based on the derivational morphology. See
also Holisky ( 1980a : 1 27-9).
8 Lees ( 1 963) characterizes these meanings as 'the fact that SUB] VERB' and 'the way
that SUB] VERB' ; both meanings also occur in English.
4. This example is from Vogt ( 1971 : 243).
5 In ch. I I , I will argue that these nominalizations have a clausal source, which
Chomsky ( 1970) claims does not exist for 'derived nominals'. This is consistent with
the suggestion made in Comrie ( 1 976d) that it is inappropriate to make a strict
bifurcation between noun-like and sentence-like nominalizations.
Schmidt ( 1 967) considers the question of whether the masdar is a Verbal
abstraktum or infinitive, deciding in favor of the latter. He, too, notes that masdars
decline, but feels that this is not a good way of distinguishing nominals from
infinitives (p. 160). In spite of examples like those in (3), he considers that masdars
are morphologically relatively uniform, and in this respect like infinitives (p. 1 58).
See Shanidze ( 1973 : 560-5) for more exhaustive lists of masdars from each of the
types represented in (3).
e The examples in (5) are from Shanidze ( 1 973 : 568). I have confirmed them as
Standard Modern Georgian with an informant and have added glosses.
7 In Old Georgian the infinitive had the shape of a 'masdar in the adverbial case'
(cf. Cbxubianishvili 1972; Matrirosovi 1 955).
8 In ch. 4 I showed that the tvis-nominal of object-raised sentences (cf. for-nominal
in English) is not the initial embedded subject, but is coreferential with the initial
embedded subject, and that the latter gets deleted. I assume that the initial subject
in (7) is also deleted by Equi. In (7) the initial subject of the embedded clause is
coreferential with the initial and final subject of the matrix clause. It may be
coreferential with a final subject that is not an initial subject, as these passives show :
(i) • rae • agretve kartulSi

• mizezis garemoebis gamosaxafavad
which-NOM also Georgian-in reaSon-GEN adverb-GEN to-depict
aris gamoqenebuli.
it-is-I-2 used.
' which is also used to express the adverb of purpose in Georgian.'
• • •
Imnaishvili (1 957 : 723)

(ii) mivlinebuli viqavi udur enaze samusaod.
sent (on a job) l-was-II-2 Udi language-on to-work
'I was sent to work on the Udi language.' Panchvidze ( 1937 : 295)
9 Some speakers find the infinitive acceptable in (iii), but most of my informants
prefer (iv), with a true participle.
(iii) bevri makvs gasalJeteblad.

much-NOM l-have-it-I-4 to-do
'I have a lot to do.'
(iv) bevri makvs gasa�etebeli.

doing-NOM
'I have a lot to do.'
The infinitive in (v) is also generally considered grammatical.
(v) bevria gasa�eteblad.

much-it-is-I-2 to-do
'There is a lot to do.'
294 Nates ta Chapter IO
10 (vi) and (vii are grammatical variants of (·8).
(vi) minda gaval$eto
I-want-it-I-4 I-do-it-II-I
'I want to do it.'
(vii) minda misi ga1$eteba.
it-GEN doing-NOM
' I want to do it.'
(On the occurrence of these two types, cf. Piitsch ( 1 965).)
11
While other orders are acceptable, these are definitely the unmarked ones.
12 The use of this case in Modem Georgian is quite restricted. In general, it is used
for the predicate nominal of a verb of 'being' when the subject of that verb is not
in the clause. For example, when the subject of 'be' undergoes Subject-to-Object
Raising, the predicate nominal is in the adverbial. (The verb 'be' obligatorily
deletes.)
(viii) (a) nino relis, rom gela f�vjania.
Nino-NOM she-considers-it-I-I that Gela-NOM clever-he-is-I-z
'Nino considers that Gela is clever.'
(b) nino tvlis gelas f�vianad.
Gela-DAT clever-ADv
'Nino considers Gela clever.'
Similarly, in (ix a) the (final) subject is dropped by Unemphatic Pronoun Drop,
while in (ix b) it is not present at all.
(ix I (a) masfavlebeli aris.
teacher-NOM he-is-I-z
'He is a teacher.'
(b) masravleblad qopna . . .
teacher-ADv being-NOM
'Being a teacher . . . . '
Other uses of the adverbial case are mostly translatable in English with 'as' -nominal :
(x) muSaobs masravleblad.
he-works-I-3 teacher-ADV
'He works as a teacher.'
(xi) kvrivad darla.
widow-ADV he-remained-II-z
'He remained as a widower.'
Compare (xi) with (xii).
(xii) kvrivi darla.
widOW-NOM
'He remained a widower.'
1
3 The nominal which corresponds to the subject of a finite transitive verb form is,
with a participle, typically marked with mier, as shown in (xiv) and (xv).
(xiv) a. !ipneris mier dadasturebul parmas • . .
A. Schiefner by confirmed form-DAT

'. . . (paid attention) to the form confirmed by A. Schiefner. . . '
Panchvidze ( 1 94z: 69)
Notes to Chapter II 295
(xv) even . •va1M(:mebdit
. a. Sipneris mier mocemuls da lven mier
we-NOM we-verified-it-I-I given and us by
vartasenSi segrooebul leksi�$.
Vartasen-in collected lexicon-DAT
'We . . . verified the lexicon given by A. Schiefner and (that) collected by us
in Vartashen.'
Panchvidze ( 1 937 : 3 1 5)
However, at least some native speakers also accept phrases like (xvi), where the
nominal corresponding to the initial subject is marked with the genitive case, not
with wer.
(xvi) im �acis da{:erili rigni . • •
that man-GEN written book-NOM

'A/The book written by that man . . . '
Such marking of a transitive subject is totally unacceptable with masdars :

(xvii) ·im �acis (rignis) da{:era. . .
book-GEN writing-NOM
(,The writing (of the book) by that man . . .')
14 The genitive (or possessive) of each first and second person personal pronoun has
several forms ; two forms of the second person singular occur in (4Ia) and (4Za)
(cf. ch. 7, n. 5).
Chapter II
1 Retired transitive subjects of participles are also usually marked with mier (cf.
ch. 10, n. 1 3), but a detailed analysis of participles has not been proposed here.
There is a second postposition that is sometimes used in place of mier. (i) illus
trates this use of the postposition -gan 'from', apparently marking a retired transitive
subject.
(i) es vasli micemulia bavlvisagan.
this apple-NOM given-it-is-I-z child-from
'This apple is given by ( ?from) the child.'
The use of -gan can be predicted from the fact that the nominal is a retired subject,
together with the fact that the governing verb is micema 'give'. This verb permitll
the use of -gan to mark its retired subjects generally.
In tills matter there is considerable variation among speakers, which I will
characterize in terms of dialects. In one dialect, -gan is acceptable only with a very
limited number of verbs, including micema of (i). Speakers of this dialect find (ii)
unacceptable.
(ii) gelasagan (:erilis dtJfera . • .
Gela-from letter-GEN writing

'The writing of the letter by ( ?from) Gela. . . '
It is probable that in this dialect -gan has only the meaning 'from' and does not
mark retired subjects, even in (i). In a second dialect, -gan is used more widely,
but it is still restricted to a particular set of verbs. Speakers of this dialect find (ii)
acceptable. In a third dialect, -gan is sometimes preferred to mier : (iii) is preferred
over (5).
296 Notes to Chapter II
(iii) bavfvi dalt.benilia 3aylisagan.
child-NOM bitten-it-is-I-z dog-from
'The child is bitten by the dog.'
In this dialect -gan clearly functions as a marker of retired transitive subjects, as

well as in the meaning 'from'. I have reported the first of these dialects in this work.
as it seems to represent the majority speech.
2 While the relations between specific termhood and the markers -tvis, mieT, and the
bare genitive case have not previously been observed, Latsabidze ( 1 975) suggests
that genitives are transformationally related to terms. I consider that postpositions
function like cases in marking grammatical relations and consider it a coincidence
that mieT and tvis both govern the genitive case. Latsabidze, on the other hand, dis
counts the importance of mieT and tvis and observes that all three term relations
are marked with the genitive as an alternative to the usual term cases.
Since I first proposed the correlations (14) in Harris (1 976, 1977), some Georgian
linguists have suggested to me that other postpositions also mark retired terms (cf.
n. I). In particular, it has been suggested in personal communication that meiveobit.
saIualebit, and sesrulebit (said to be literary) may mark retired transitive subjects.
and that the suffixes -euli and -admi may mark retired direct objects. Additional
field work has shown that these mark ordinary non-terms (obliques), but do not
serve the purely grsmmatical function of marking retired terms. The evidence con
cerning all of these cannot be given here, but the first is taken as an example. In
the passive (iv), either mier or meiveobit may occur, suggesting that they function in
the same way.
(iv) vaTdi gazrdilia vanos

{ mier
.
meiveobtt.
rose-NOM grown-it-is-I-z Vano
'The rose is grown [ ?by] Vano.'
But in the intransitive (v), vano is not the initial subject; mier is unacceptable.
{
while meiveobit is acceptable.
"'mier
(v) vardi gaizarda vanos
mesveobj't.
v
it-grew-II-z
'The rose grew because of Vano./The rose grew through the intervention of
Vano.'
(The nature of constructions like (v) is discussed further in ch. 13.) I conclude that
meiveobit means 'through the intervention of' rather than having a purely gram
matical function. It is interesting that the same meaning attached to mier in Old
Georgian. though today it is not used that way in the standard dialect (cf. Harris
1979),
S Because I had not included sentences like (9)/(16) in my 1974-75 field work in
Georgia, in Harris (1 976, 1977) I reported only (r6), on the basis of work with only
one informant in the United States. Based on ( 16) and other examples of that type,
I concluded that marking was as in ( I S). Subsequent field work in Georgia in 1977
revealed that the dialect originally reported is not widespread.
4 One fact seems to support this possibility : a very few verbs govern a particular post
position other than tvis or mier for marking a specific retirement relation. For
example, miYfeva 'reach' governs -mde 'until' fpr its retired object : miueva hal
akamde 'reaching the city'.
Notes to Chapter IZ 297
£ The relationship between grammatical relations and semantic relations is explored
further in ch. 16.
• Mier in Standard Modern Georgian is used only to mark retired subjects. Tvis, on
the other hand, while it marks retired indirect objects, also marks other non-terms
in the general meaning 'for'. Thus, the fact that the nominal in question in the
object-raised sentences is marked with tvis does not indicate that it is a retired
indirect object.
Chapter l:2 and Appendix

1 The label 'verb' is used in this work with the systematic ambiguity that is usual in
the use of that word : it may refer to the concrete inflected occurrence of a verb,
or to the abstract notion 'verb'. In this chapter, I am claiming that some verbs, in
the latter sense, have an obligatory initial direct object. That is, the verb dafera
'write' has an obligatory initial direct object, whether it is a direct form (dafeTa 'he
wrote if), a passive (da�eTilia 'it is written'), a causative (daar:erina 'he made him
write if), etc., and whether it is in Series I, II, or III. Whatever the surface realiz
ation of this verb, it has an initial direct object.
There are a few instances where two verbs, in this sense, are formed on one root,
and where one verb takes a preverb and the other does not. In this case, too, their
initial term inventories may differ. For example, ga{OTVa in (i) takes an obligatory
initial direct object, while {oraoba in (ii) does not permit an initial direct object.
(i) (len) m{orav (me) dedacemtan.

YOU-NOM you-gossip-me-I-r I-DAT my-mother-to
'You gossip to my mother about me.'
(ii) (len) {oraob lemze dedacemtan.

you-gossip-I-3 me-on
'You gossip to my mother about me.'
Ga{oTva and {oraoba belong to different morphological Classes and must be con
sidered different verbs in the sense discussed here.
S Verbs of the last type have corresponding organic causatives which have a direct
object. But in this instance, it is not the embedded verb, but the matrix verb,
CAUSll, which has an initial direct object.
a In examples (*'5) and (*'8), the markers of Person Agreement do not differ from
those in (4a) and (7a), respectively, because the marker of third person singular
direct objects is the same as the marker of third person singular indirect objects in
this phonological environment.
, (5) has a grammatical reading, 'I had Vano written', where Vano is the direct object
as the word written.
6 Nominals which are obligatory in the initial inventory of terms can be dropped by
Unemphatic Pronoun Drop. Nominals which are not in the initial inventory of
term nominals can also be dropped if they are final terms ; for example, benefactives
that advance to indirect object may be dropped, daUfeTa 'he wrote it for him'. I
know of no counterexamples to the statement made in §z.
8 Examples (I a), (2), (10), (r r), and (30a) of ch. I, ( r ) of ch. 3, (9) and (22) of ch. 8,
and (2) and (3 ) of ch. 13, as well as other examples in this work show that it is not
typical for verbs in Georgian that animacy plays a role in grammatical relations.
7 I assume here that the indirect object is an initial locative and advances to indirect
298 Notes to Chapter I3
object by a rule of Locative Version, similar in essential respects to Benefactive
Version (cf. ch. 6). I make this assumption for the following reasons :
A. The meaning of this indirect object differs consistently from that of initial
indirect objects.
B. The locative-nominal may occur either as an indirect object, in the dative
case and triggering Indirect Object Agreement, or as a non-term, with the
postposition -tan and not triggering agreement:
(i) vanrml dare{ia dedastan.
VanO-ERG he-rang-it-II-I mother-at
'Vano rang his mother's .'
(ii) vanom dauTe1!a dedas.

he-rang-her-it-I I - I mother-DAT
'Vano rang his mother.'
c. When the locative-nominal has advanced to indirect object, as in (ii), it

triggers the version vowel i- (cf. (8) of ch. 6); when it has not advanced to
indirect object, as in (i), no version marker occurs.
D. The dative-nominal in (iO, but not the locative in (i), is deletable by Unem-
phatic Pronoun Drop.
Since Locative Version applies optionally, there is a parallel set of sentences in

which the locative has not been advanced and occurs as dedastan 'at mother's'.
Parallel to (6), (8), and (9) are
(iii) 'llanos turme dauTe{iia dedastan.
(iv) 'llanos mier dedastan darekva,
and
(v) vanom damare�ina dedastan.
Although Locative Version is infrequent, it does apply with other verbs, such as
mibera'lls 'it is blowing on me' (cf. beravs 'it is blowing') and movur;re mas 'I made it
there on time' (cf. mOVafTe mastan 'I made it there on time').
8 Pasuxi is one of the verbs that allows -gan to replace mier (cf. ch. I I, n. I). Pasuxi
is not only a masdar, but a concrete noun ; therefore, seni pasuxi 'your answer' is also
grammatical. These two uses of pasuXi may be distinguished by co-occurrence with
adjectives or adverbs ; sr;orad 'correctly' may co-occur with the masdar, but only
SfOTi 'correct' with the concrete noun :
sengan ga�etili sfOTad pasuxi
·seni sfOTad pasuxi
seni sfori pasuxi.
Chapter I3 and Appendix

1 Possessive Version is not obligatory, but usual, in this environment (cf. introduction
to ch. 6).
2 Within the set of verbs characterized morphologically by the circumfixes i-ebi,
e--ebi, or o-ebi, at least four types must be distinguished ; they do not correspond
to the distribution of the circumfixes :
A. Synthetic passives are illustrated and characterized in the body of this chapter.
B. Potentials, like synthetic passives, are related to direct forms, in the sense that
they are formed from the same abstract verb. For example, es xe ixerxeba 'this
Notes to Chapter I3 299
tree/wood cuts well', a potential, is related to viyac xerxavs am xes 'someone
is cutting this wood/tree', a direct form. Potentials may be distinguished from
other verb forms with the same morphology by the fact that they have no
forms outside Series I.
As far as I am aware, all of the characteristics discussed in this chapter with
respect to synthetic passives apply equally to potentials. I have excluded them
because of lack of data.
c. Certain other verb forms which share the morphology of synthetic passives
do not share some of the other characteristics discussed in §2. This group
includes icereba 'it is written', iyeba 'it is received', and ifreba 'it is cut'.
D. Forms like vardeba 'he falls', as stated above, share the morphology of syn
thetic passives, but may be distinguished from them by the fact that they do
not have corresponding direct forms.
Traditionally these sets of verbs have been described together, since they share
morphology. The arguments adduced in this chapter do not necessarily apply to
types B-D. In ch. 16 the similarity of these four types and others of Class 2 is
explored.
There are also several verb forms which are morphologically like the synthetic
passive, but which are not regularly derived. These include verb forms like igineba
'he curses', imaleba 'he hides'. Some of the problems presented by these irregular
verb forms are discussed in Harris (to appear b) and Harris (1976), Appendix C.
3 This definition rules out the inversion construction. This could be considered as a
fourth contrasting construction, but it has already been discussed in detail in ch. 8.
;1 Notice that animacy plays no role in distinguishing between the analytic and the
synthetic passive ; both may have animate or inanimate initial direct objects. These
possibilities are illustrated in the examples of § 1.
;; Some Indo-European middles correspond to the forms I refer to as 'synthetic
passives', others to some of the other forms which share that morphology (cf. n. 2).
Still other Indo-European middles correspond to the Georgian Benefactive or
Possessive Version with Coreferential Version Object Deletion, as observed in
Schmidt (1965).
While most Georgianists have called synthetic passives and the other forms
described in footnote 2 'passives', Blake ( 1 932 : 234) observes that they are 'in
reality middle'.
6 According to Holisky, these diagnostics test for the 'presence of an agent' ( 1 978 :
I SO, 1 52). There is no doubt that the adverbs of intention and the command
imperative are impossible in sentences lacking an agent, but neither are they possible
in all sentences containing an agent. They may in general occur in sentences with
semantic agents only if the agent is a final subject. As an exception to this, ganzrax
'intentionally' may marginally occur with analytic passives if the initial subject is
specified and ganzrax is associated with it by word order, as (·i) and (ii) show ; the
word order of (.i) is the unmarked order.
0) "'rur{eli ganzrax gamfralia.

crockery-NoM intentionally dried-it-is-I-2
('The crockery is intentionally dried.')
(ii) ?furfeli gamfralia bifis mier ganzrax.

boy by
'The crockery is dried by the boy intentionally.'
Holisky (1978) gives additional tests, but I have not been able to apply all of
these to synthetic and analytic passives in work with informants.
7 This assumes that all semantic agents are initial subjects. It is shown in ch. 16 that
this assumption is correct.
s Lomtatidze (I95z) and Machavariani (1959) argue that those synthetic passives
characterized by r!I-ebi in the present tense (illustrated by (4-6) in the present
chapter) are 'dynamic intransitives', not passives as most grammarians have said. I
believe that this is correct, but that the same applies, not only to those forms marked
by 0-ebi, but also to those synthetic passives marked by i-ebi and e-ebi (cf.
group E of Appendix A to ch. 1 6 for additional examples). From the fact that they
are not true passives, Lomtatidze and Machavariani draw the conclusion that their
final subjects are not initial direct objects. It is argued in this section that that
conclusion is unfounded and incorrect.
9 The direct object here does not trigger Number Agreement (cf. ch. 1 5).
10
The verb root I}:al alternates with I}:vd, and both are derived from an original root
'*I}:v (Topuria 1940 : 53 5 ; Kavtaradze 1954 : 3 19, footnote z).
11 Preverbs are not agreement markers, nor are they used with all verbs. In addition
to the function described here, they may indicate direction, tense, and aspect (cf.
Chikobava 1950 ; Shanidze 1973 : Tschenkeli 1 958 : and Vogt 1 971).
12 The extent to which the synthetic passive interacts with other syntactic rules is
stated briefly below.
Masdars and infinitives correspond to abstract verbs (cf. ch. lZ, n. I), not to par
ticular finite verb forms. Thus, a given masdar or infinitive does not correspond to
any specific voice form. A given masdar or infinitive may have the inventory of
initial terms which characterizes the direct construction or that which characterizes
the synthetic passive, but this difference is not reflected in the verb form.
Causatives offer no evidence concerning the nature of the synthetic passive. Since
both an intransitive subject and a direct object are realized as the direct object of
the causative (cf. ch. 5), the derived grammatical relations provide no evidence
concerning initial grammatical relations in this instance.
Passivization and Unaccusative cannot apply in a single clause, since the input
conditions for both cannot be met in the same clause.
Version applies freely in the synthetic passive. Pairs like (iii), where Possessive
Version has applied, might be construed as the basis for an additional argument to
(iii) (a) mixurav$ I}:ars.

he-closes-me-it-I-I door-DAT
'He is closing my door.'
(b) mexureba I}:ari.
it-closes-me-I-z door-NOM
'My door is closing.'
support the initial grammatical relations proposed in §3. In the direct construction,
(iii a), the version object possesses the direct object; in the synthetic passive, (iii b),
it possesses the final subject. However, the constraints on exactly which nominals
may be interpreted as possessed by the version object are still poorly understood.
No argument can be based on sentences like (iii) until this fundamental property
of Possessive Version is determined.
Case marking and Inversion as evidence for the analysis proposed here are con
sidered at length in ch. 16.
13 In the ikna-passive, like the iqo-passive, the initial indirect object is realized as a
final non-term , as shown by (iv) (cf. ch. 7, §z).
Notes to Chapters I4-I5 301
(iv) es vardi dedastvis ikna micemuli.
this rose-NOM mother-for it-was-II-z given
'This rose was given to Mother.'
The retired indirect object, deda 'mother', is marked with tvis.
14 The use of such phrases as tests of stativity in Georgian is established in Holisky
(1978).
Chapter I4
1 This generalization is supported by many additional examples, such as (i).

(i) lilies perangi lovnilia genos mier tavistvis moslt-ovsi.
this shirt-NoM gotten-it-is-I-z Geno by self-for Moscow-in
('This shirt was gotten by Geno for himself in Moscow.')
There is, however, at least one exception to the generalization. In (ii), the initial sub
ject, which is a final non-term, triggers Tav-Reflexivization. I have no explanation
for the grammaticality of this sentence.
(ii) bevri sisulele iqo nalaparalt-e7.;i vanos mier tavisi tavis sesaxeb.
much nonsense-NOM it-was-II-z spoken Vano by self's self about
'A great deal of nonsense was spoken by Vano about himself.'
Z Pronouns dropped by Unemphatic Pronoun Drop, on the other hand, function all
final terms with respect to Causative Clause Union, Retired Term Marking (cf.
ch. 12), Person Agreement (cf. ch. I, §3), Number Agreement (cf. ch. I S), and
Tav-Reflexivization. Nominals deleted in the course of Object Raising do not func
tion as terms in this way (cf. ch. 4, §S). This difference is accounted for here by
stating Unemphatic Pronoun Drop on final terrnhood (cf. Epilogue, §4).
3 It is intended that ( I 6) be interpreted to mean that reflexivization must apply if
the coreferential nominals are clausemates at any level of derivation ; cf. ( I I ) , where
coreferentiality is resolved by reflexivization whether it occurs in initial or final
structure.
4 Postal himself no longer supports a crossover analysis for English (Postal 1971 : v).
Chapter IS
1 I have already stated in the Introduction one constraint on Number Agreement : in

some dialects all inanimates trigger singular agreement.
2 The third person plural indirect object agreement marker will not show up in the
first examples considered ; it is stated here on the basis of examples like (l sa). The
third person plural direct object agreement marker is never triggered, so '0' is a
construct.
3 When two plural terms co-occur in a clause, the plural agreement marker of one is
deleted by the rule given below as (Zs). For this reason, examples with two plural
terms are avoided here.
4 In this example, the subject agreement marker is deleted by rule (Z4), given below.
5 In spite of examples like those in (6), which show that third person objects regularly
fail to trigger Number Agreement, one finds rare examples of object Number
Agrecment in the literature. Whether such examples evcr occur in thc Standard
Modern Georgian of Tbilisi I cannot tell. Chikobava ( 1 941 : §n) cites cxamples of
third person indirect object Number Agreement from eastern (non-standard)
dialects.
8 The reader may need to be reminded here that the order of elements in the analytic
gloss of the verb forms reflects the English word order, with the initial subject first,
then the verb, then the initial direct object. The gloss directly reflects the presence
of a plural morpheme in this way: In ( I sa), t indicates plurality of the third person
initial subject, not of the third person initial direct object. This is indicated in the
analytic gloss by 'they' in preverbal position. In ( I S b), the fact that there is no plural
marker is indicated by the use of singular pronouns in the analytic verb gloss.
7 The notion 'first subject that is a final term' also plays a role in the statement of
unmarked word order in IVIodern Georgian. It is this nominal that regularly occupies
the first position (cf. Introduction, §4.S).
8 ( 18) accounts for the selection of the Number Agreement triggers, but it does not
account for the selection of the marker of Number Agreement. While the trigger is
selected on the basis of final and non-final termhood, the actual marker is chosen
on the basis of the final termhood of the trigger. Thus, in ( 1 2.-14) final subjects
trigger markers from (I), while in (IS) the final indirect object triggers a marker
from (3).
9 The dialect described here is that described also by Tschenkeli (1958), Vogt (1971)
and by Georgian linguists. However, most of my informants found (i) at least
marginally acceptable ; one preferred it to ( 17a).
(i) turme studentebs gamQugzavnixart (sen).
apparently students-DAT they-sent-you-lII-I YOU(SG)-NOM
'Apparently the students (have) sent you.'
Here a third person initial subject triggers Number Agreement, even when a second
person term is in the clause. That is, the dialect represented by speakers who prefer
(i) to (17a) has a simplified version of the rule ( 1 8), which lacks clause (b--ii)
altogether. This dialect apparently does not differ in other respects from the one
described in this chapter.
10
Plural suffixes do not co-occur (cf. rule (2.5» . In general, object suffixes do follow
subject suffixes, as in example ( Isa).
11
This is essentially the analysis proposed in Aronson ( 1970) and in Chikobava ( 1 968).
Aronson defines the 'grammatical subject of transitive verbs as that member of the
predication which can be marked by the verb for number in the third person'
(1970: 2.94). From the discussion it is clear that he considers the inversion construc
tion transitive. According to his definition, the dative-nominals of ( I S) are gram
matical subjects. But, by the same token, those of ( 1 7) are not grammatical subjects,
since they cannot be marked for number. But ( I S) and ( 1 7) differ only in the
person of the nominative-nominals and in Number Agreement.
Chapter 16 and Appendices

1 Vogt's statement of this is among the clearest : 'Les verbes neutres [Class 3] ont ete
definis comme des verbes qui ne sont ni transitifs ni intransitifs' (Vogt 1971 : 1 3 3 ;
also cf. 87--9). Similar analyses are made by other authors.
Class 4 is not included in many parts of my discussion of the ergative analysis.
This is because I am trying to represent accurately the consensus of opinion on
ergativity in Georgian ; but there is no unity in the treatment of Class 4 among
proponents oftbe ergative hypothesis (cf. Aronson 1 976 comparing various analyses).
2 The term 'future/aorist' is used here and throughout this chapter to denote the set
of tenses listed in (5), all of which share some formant, for each verb Class.
3 In the Conditional and Future Subjunctive, -d- is used generally by Class I . -od- by
Class 2; -d- is used by Class 3 verbs, except those with the thematic suffix -i, which
use -ad- instead.
4 While most Georgianists have labelled Class 3 'irregular', both Nozadze ( 1974) and
Holisky ( 1980a) emphasize the regularity of this group. Nozadze (1974: 30ff.) also
traces the historical development of the morphology that sets Class 3 apart from
other verbs. Holisky shows that the difference is due to the fact that the preverb in
Georgian is associated with punctual aspect, and this category is inapplicable to
medial verbs because they are atelic.
5 An example of this is the statement below. Emphasis is added .
. . . there are a number of intransitive verbs that exceptionally require an ergative

subject, though this class of intransitive verbs is essentially arbitrary.
(Comrie 1976b)
8 This particular example is taken from Tschenkeli (1958 : vol. I, 293-4) ; it is typical
of many other authors as well. See also Tschenkeli ( 1958 : 299, no. 2).
7 There are also serious problems with the details of such a theory of borrowing.
First, most Class 3 verbs may not take a preverb in the future/aorist, while the
corresponding Class 1 verb (causative) requires a preverb in the future/aorist. This
is the case with vibatane (*gavibatane) 'I reigned' and gavibatone ( ?*vibatone)
cemi tam 'I made myself ruler'. Second, in the Class 1 form (causative) the i- results
from Possessive Version and Coreferential Version Object Deletion and depends
upon the identity of subject and object. In the Class 3 verb, on the other hand,
there is no object at any level of derivation.
Nozadze ( 1974: 30ff.) shows that borrowing of forms is also an untenable hypoth
esis diachronically.
8 Fillmore incorporates the active/inactive type in his case grammar (cf. Fillmore
1968 : 53-4). Klimov (1973) also discusses this type ; curiously, he nevertheless
analyzes Georgian as having ergative case marking, not 'active/inactive'. He states
that Class 3 verbs represent a relic of an earlier active/inactive system (P.50) or an
anticipation of a future accusative system (P.1 89). This analysis apparently arises
from his theory that languages can develop from an active/inactive system to an
ergative system, or from ergative to accusative, but not the reverse. See Comrie
(1976b) for an enlightening discussion.
S Although it has not previously been recognized that case marking Series II is an
active/inactive rule in Sapir's sense, three Georgian linguists have observed that
the use of the ergative or nominative as the subject case in Series II depends upon
whether the verb is active or inactive (Jajanidze 1970 ; Jorbenadze 1975 : 219ff. ;
Topuria 1923 : 1 I 5 and 120, and 1954: no. 7).
0
1 The notion 'basic' will not be defined ; d. lists of examples in Appendix. 'Basic
Class I verbs' may be taken to be Class 1 verbs that are not causative, etc.
11
The type atetrebs 'he makes it white' from the adjective tetri' white' is also a causa
tive. The type xerxavs 'he is sawing it' (xerxi 'a/the saw') may represent an ad
ditional derivational type. In any case, its term relations are identical to those in (10).
1 2 Perlmutter & Postal (to appear c) propose a Final 1 Law, requiring every clause
to have a final subject. If this law is correct, it obviates the necessity of making a
language-particular statement that Unaccusative is obligatory.
13 Not only noun subjects, but also sentential subjects of basic Class 2 verbs are
initial direct objects. This includes sentential subjects of qopna in sentences in which
Object Raising applies. This analysis is summarized in the network below (cf. (6)
of ch. 4).
' h ard ' ' be '
mo;ebna me �argi magalitebi

' find' '1' "good examples �
14 Inceptives may also be based on inversion predicates, as in (i).

(i)
qr-areba semiq'Z·arlieba.
" love' 'I' ' he ' • I will fal l i n love with hill!. ,
The analysis presented here is supported by the following facts. The initial
grammatical relations represented in the dependent clauses of ( 1 8), ( 1 9), and (i),
are the same as those which the unmarked form of each verb takes in an independent
clause. The initial grammatical relation of the embedded clause to the abstract
predicate COME ABOUT is the same as that of a non-sentential dependent to the
concrete verb axda, moxda, gaxda 'it came about, happened, became'. The final
grammatical relations are consistent with Person and Number Agreement and case
marking (cf. (I» . In each of the three structures (1 8), ( 1 9), and (i), the initial gram
matical relations are correctly related to the final grammatical relations by the rules
of Clause Union and Unaccusative, withno special apparatus. (Cf. Aissen & Perlmutter
1976 and Frantz 1976 on Clause Union in constructions other than causatives.)
Finally, to the best of my knowledge, this analysis violates no rule or principle
of Georgian or universal grammar.
15 Basic Class 2 verbs (c) may correspond to causatives (B), which are always Class I
verbs, e.g., daarlina 'he left him, he caused him to remain', Such causatives do not
provide an argument to distinguish between initial intransitive subjects and initial
intransitive direct objects, since Causative Clause Union treats direct objects and
intransitive subjects alike.
IS
I have generalized the rules for Series I and III (rule I) but the rules for Series I I
apply equally well for Series III, as the reader can verify for himself.
17 One of my informants made a revealing remark about example (22). "''hen pre
sented with the alternative of (22a) or (22b), he replied that he knew that (22a) was
right (normative), but that it sounded strange, as though the water were dripping
on purpose, as though the water were active (ak/iuri).
18 Although the use of the ergative case with verbs that irregularly govern Pattern B
in the literary dialect has been much discussed in the literature on Georgian dialects,
only a few linguists have observed that this phenomenon is limited to semantically
active verbs that are formally in Class 2 (Jajanidze 1 970 : 258 ; Topuria 1 923 : 1 20,
and 1954 : 455). Topuria, in the works cited, also shows that the use of this case is
not due to the influence of Mengrelian, but that it is a process of regularizing
(unipi�acia).
19 'With respect to complex structures, (24') must be interpreted to mean that Inversion
may apply in a clause only to the initial subject of that clause. In clause union and
raising constructions, Inversion applies only to initial subjects of matrix clauses. In
sentences that are finally multi-clausal, Inversion may apply in an embedded clause
to the initial subject of that clause. For example, in the saying
ar iei, vardi ue�lod aravis mou�repia ?

NEG you-know-it-I-I rose-NOM without-thorns no-one-DAT he-gathered-it-III-1
'Don't you know that no one has ever picked a rose without thorns?'
Inversion applies in the dependent clause only (because of the negative), demoting
the initial subject of that clause (aravis 'no one') and marking the verb of that
clause with Series I I I morphology.
20 Nor does the grammar need to refer to Class 4. (23a) can be restated on the basis
of semantics as
(Z3a') verbs having experiencers.
21 In (3 1) the meaning 'begin' is due to the inceptive form, which is represented

morphologically by the -d suffix. In (3Z) it is due to the use of the preverb ca, which
adds the notion 'begin' to some verbs. As shown by these examples, (31) is non
controllable, and (3Z) controllable ; this is not easily captured in the English gloss.
22 Since these facts have not been discussed in literature on Georgian, it seems appro
priate to give additional examples. The sentences given below are parallel to
(31-32), in the same order.
(ii) vano aml(erda.

Vano-NoM he-come-about-sing-II-z
'Vano began to sing.'
caldani aml(erda.
teakettle-NoM
'The teakettle began to sing.'
vanom iml(era.
VanO-ERG he-sang-II-3
'Vano sang.'
"'caidanma/"'caidani iml(era.
teakettle-ERG/NOM
('The teakettle sang.')
(iii) lelt-vi atalt-da.
PUPP Y-NOM he-come-about-shake-II-z
'The puppy suddenly shook/jerked (e.g. in his sleep).'
saxli selo�da.
house-NoM
'The house suddenly shook/jerked.'
lelt-vrno. #olt-a.
puppy-ERG he-shook-II-3
'The puppy shook.'
"'saxlma/"'saxli ito�.
hOuSe-ERG/NoM
(,The house shook.')
The sentences in (iv) and (v) are parallel to (33), in the same order.
(iv) vano amyerebula.

Vano-NoM he-come-about-sing-III-z
'Vano apparently began to sing.'
"'vanos amyerebula.
VanO-DAT
('Vano apparently began to sing.')
vanos umyeria.
he-sang-I 1I-3
'Vano apparently sang.'
(v) turme le�i ato�ebula.
PUPPY -NOM he-come-about-shake-III-z
'Apparently the puppy suddenly shook.'
"'turme lelt-vs ato�ebula.
puppy-DAT
('Apparently the puppy suddenly shook.')
turme le�s uto�nia.
he-shook-III- 3
'Apparently the puppy shook.'
23 Some linguists prefer to limit the notion 'ergative rule' to rules that apply in a
single clause or to rules of inflection. But multi-clausal ergative rules are recognized
in some important discussions of ergativity (for example, Dixon 197z).
24 Surprisingly, it has been suggested that verb agreement in Series III of Georgian
is ergative. This is based on the claim that the 'intransitive subject' and the 'direct
object' trigger the same (bold type) agreement marker in sentences like (vi) and (vii).
(vi) me vqopilvar.
I-NOM I-am-I1I-z
'I am.'
(vii) me gavgzavnivar.
I-NOM he-sent-me-III-l
'He (has) sent me.'
In ch. 8 it was shown that the markers of Person Agreement here reflect the final
grammatical relations, 'I' in both sentences being the final subject. Numerous
arguments for this analysis have been amassed in chs. 8, I S and in §3 of this chapter.
85 In (3Sa) the initial subject ( final indirect object) triggers the plural marker -t,
as predicted by the rule of Number Agreement formulated in ch. I S . I am grateful
to Dee Ann Holisky for bringing this example to my attention.
26 These characteristics were selected instead of some other possibilities because of
(a) their high degree of mutual consistency, and (b) their high degree of correlation
with syntactic characteristics (d-f). Other morphological properties include
1. the use of -d- or -od- for the formation of the conditional/imperfect.
2. the use of -a or -s as a third person singular marker in the present tense.
These properties are less regularly correlated with the syntax and semantics than
are those noted in the text.
27 The verb codna 'know' is an absolute exception. Unlike any other verb in Modern
Georgian, it uses Pattern A in some tenses of Series I ; in others it undergoes
Inversion.
28 Of the three morphological characteristics listed in Appendix A, only character
istic (c) (third person plural suffix in the aorist) was tested with informants. Infor
mants consistently rejected the Class I suffix -es for all four verbs tested. In addition,
they accepted only the imperfect formant -od-, which is used for Class 2, not
Class I, verbs.
29 .fqali iqo sevedrebuli.

water-NOM it-was-II-2 begged
('Water was begged for.')
30 I assume here that the indirect object is an initial comitative and advances to
indirect object by a rule of Comitative Version, essentially similar to Benefactive
Version (cf. ch. 6). I make this assumption for the following reasons :
I. The meaning of this indirect object differs consistently from that of initial
indirect objects, such as those with the verbs cems 'he gives it to him' or
mifeTa 'he wrote it to him'.
2. The comitative-nominal may occur either as an indirect object, in the dative
case and triggering Indirect Object Agreement, or as a non-term, with the
postposition -tan and not triggering agreement:
(viii) gela kiSpobs an:;:OTtan

Gela-NoM he-competes-I-3 Anzor-with
' Gela competes with Anzor.'
(ix) gela ekiSpeba anzors.

he-competes-with-him-I-z Anzor-DAT
However, the assumption that the final indirect object is an initial comitative, not
an initial indirect object, is not crucial to the remarks that follow.
81 Non:p.ative grammars deprecate this use of the ergative (Tschenkeli I9S8: vol. 1, 430).
32 The ungrammaticality of (·ISb) is sometimes 'explained' by saying that the form
does not exist in Series III. Regular Class 2 verbs with indirect objects do have
Series III forms of exactly this type, e.g., dasfqebia 'it began for him'. If it is claimed
that (·ISb) is ungrammatical because the form does not exist, the non-existence of
the form must still be explained. I believe that the form does not exist because the
failure of Inversion to apply to the initial subject (gela in (-I Sb» results in ungram
maticality .
33 Since normative grammars do not recognize that arseboba fails to trigger Inversion
308 Notes to Chapter I 6
when its meaning is 'exist (generally)', it seems imperative to quote examples from
literary Georgian :
(x) mis gverdit unda arsebuliqo is mr:l$rivic.
it-GEN side-INST MODAL it-existed-III-z that screeve-NoM-too
'That screeve apparently must have existed beside this.'
(xi) . . .romelic rogorc cans, martlac arsebula 3vel kartuW.
which-NOM as it-seems truly it-existed-III-z Old Georgian.
'. • •which, as it appears, really did exist in Old Georgian. '
Kiknadze 1 96 7 : 186-7
In neither example has Inversion applied. Notice that (xi) is particularly similar to
(I6b) ; but the verb here is not controllable and does not undergo Inversion.
R eferences
Abbreviations
BSL Bulletin de la Societe de Linguistique

CLS Papers from the (n :th) Regional Meeting of the Chicago Linguistic
Society
IJAL International Journal of American Linguistics
IKE Iberiul-Kavkasiuri Enatmecniereba
KESS Kartvelur Enata S!rukturis Sa�itxebi
LI Linguistic Inquiry
LSA Linguistic Society of America
NELS Proceedings of the (n :th) Annual Meeting of the North-Eastern
Linguistic Society
NJL(NTS)Norwegian Journal of Linguistics
Proc. BLS Proceedings of the (n :th) Annual Meeting of the Berkeley Linguis
tics Society
SLS Studies in the Linguistic Sciences, University of Illinois
Aissen, J. ( I 974a) The Syntax of Causative Constructions. Dissertation,

Harvard University.
- (1974b) 'Verb Raising'. LI 5 : 325-66.
- (1980) 'Possessor Ascension in Tzotzil'. In L. Martin (ed.) Papers in Mayan
Linguistics. Columbia, Mo. : Lucas.
(ms.) 'Valence and Coreference'. An earlier version was presented at the
fifth annual meeting of the Berkeley Linguistics Society, 1979, under the
title 'Voice and Coreference'.
Aissen, J. & Perlmutter, D. M. (1976) 'Clause Reduction in Spanish'. In
Proc. BLS 2.
Allen, B. J. & Frantz, D. G. (1978) 'Verb Agreement in Southern Tiwa'. In
Proc. BLS 4.
Anderson, S. R. (1976) 'On the Notion of Subject in Ergative Languages'.
In C. Li (ed.) Subject and Topic. New York : Academic Press.
310 References
Aronson, Howard I. (1970) 'Towards a Semantic Analysis of Case and Subject

in Georgian'. Lingua 25 : 291-30L
- (1976) ' Grammatical Subject in Old Georgian' . Bedi Kartlisa 34 : 220-3 1 .
Bell, S . J. ( 1974) 'Two Consequences of Advancement Rules in Cebuano'.
NELS 5 : 257-64.
(1976) Cebuano Subjects in Two Frameworks. Dissertation, Massachusetts
Institute of Technology.
Benveniste, E. (1950) 'Actif et moyen dans Ie verbe'. Journal de Psychologie
43 : 1 1 9-27; reprinted in Readings in Linguistics II, E. P. Hamp, F. W.
Householder, and R. Austerlitz (eds.) (1966). Chicago : University of
Chicago Press.
Berman, A. (1974) Adjectives and Adjective Complement Constructions in
English. Report No. NSF-29 to the NSF, Formal Linguistics, Susumu
Kuno, principal investigator.
Blake, R. P. (1932) 'Khanmeti Palimpsest Fragments of the Old Georgian
Version of Jeremiah'. Harvard Theological Review 25 : 225--72.
Boeder, W. (1968) ' Uber die Versionen des Georgischen Verbs', Folia
Linguistica 2 : 82-152 ; with unpublished Nachtrag.
Catford, J. C. (1 976) 'Ergativity in Caucasian Languages'. NELS 6 : 37-48.
Chafe, W. L. (1970) A Semantically Based Sketch of Onondaga. Memoire 25
of the International Journal of American Linguistics.
Chikobava, A. (1940) ' Mesame piris subiektis u3velesi ni�ani kartvelur
eneb�i' (The ancient marker of the third person subject in the KartveIian
languages). Eni�is Moambe 5-6 : 1 3-42.
- (1941) 'Martivi <;inadadebis evoluciis 3iritadi tendenciebi kartulsi' (Basic
tendencies of the evolution of the simple sentence in Georgian). Sakartvelos
SSR Mecn. AI!ad. Moambe 2 : 197-202, 561-8 ; reprinted 'as an appendix
to Chikobava ( 1 968).
- ( 1 948) Ergatiuli ",onstrukciis problema iberiullJavl;t.asiur enebSi I (The prob
lem of the ergative construction in the Ibero-Caueasian languages). TbiIisi :
SakartveIos SSR Meen. A�ad. Gamomeemloba.
(1950) 'Kartuli erus zogadi enatmecnieruli daxasiateba' (A general linguis
tic characterization of the Georgian language). Kartuli enis ganmartebiti
leksi"'oni (An explanatory dictionary of the Georgian language). Tbilisi :
Sakartvelos SSR Meen. A�ad. Gamomcemloba.
- (1961) Ergatiuli i!-onsrrukcit's problema iberiul",avlJasiur enebSi II (The prob
lem of the ergative construction in the Ibero-Caucasian languages).
TbiIisi : Sakartvelos SSR Mecn. A�ad. Gamomeemloba.
- ( 1 968) Marjivi �inadadebis problema kartulSi I (The problem of the simple
sentence in Georgian). Second edition, Tbilisi : Gamomeemloba 'Mec
niereba'.
Chomsky, N. (1970) 'Remarks on Nominalization'. In R. A. Jacobs and
References 311
P. S . Rosenbaum (eds.) Readings in English Transformational Grammar.

Waltham : Ginn.
Chung, S. (1 976a) 'An Object Creating Rule in Bahasa Indonesia'. LI
7 : 41-87.
- (1 976b) 'On the Subject of Two Passives in Indonesian'. In C. Li (ed.)
Subject and Topic. New York : Academic Press.
Chxubianishvili , D . (1972) Inpini/ivis s�itxisatvis Jvel kartulSi (On the
question of the infinitive in Old Georgian). Tbilisi : Gamomcemloba
'Mecniereba' .
Cole, P. & Sridhar, S. N. ( 1976) ' Clause Union and Relational Grammar :
Evidence from Hebrew and Kannada'. SLS 6 : 2 1 6-27.
Comrie, B. ( 1 973) 'The Ergative : Variations on a Theme'. Lingua 32 : 239-53.
- ( 1975) 'Causatives and Universal Grammar'. Transactions of the Philo
logical Society 1 974 : 1-32.
(1976a) Aspect. Cambridge : Cambridge University Press.
( 1 976b) Review of Klimov (1973). Lingua 3 9 : 252-60.
- (1 976c) Review of Xolodovic ( 1 969). Language 52 : 479-88.
- ( 1976d) 'The Syntax of Action Nominals : A Cross-Language Study'.
Lingua 40 : 1 77-201.
(1 976e) 'The Syntax of Causative Constructions : Cross-Language Simi
larities and Divergences'. In M. Shibatani (ed.) The Grammar of Causative
Constructions. New York : Academic Press.
- ( 1977) 'In Defense of Spontaneous Demotion : The Impersonal Passive'.
In P. Cole and J. Sadock (eds.) Grammatical Relations. New York : Aca
demic Press.
- (1978) 'Ergativity'. I n W. P. Lehmann (ed.) Syntactic Typology : Studies in
the Phenomenology of Language. Austin : University of Texas Press.
Davison, A. ( 1 969) ' Reflexivization and Movement Rules in Relation to a
Class of Hindi Psychological Predieates'. CLS 5 : 37-5 1 .
Decters, G. ( 1 930) Das khartvelische Verbum. Leipzig : Kommissionsverlag
von Markert und Petters.
Dixon, R. M. W. ( 1 972) The Dyirbal Language of North Queensland. Cam
bridge : Cambridge University Press.
- (1 979) 'Ergativity' . Language 55 : 59-138.
Fahnrich, H. ( 1 965) 'Die Funktionen des Charaktervokals i im georgischen
Verb '. Wissenschaftliche Zeitschrift der Friedrich-Schiller- Universitiit Jena
(Gesellschafts- und Sprachwissenschaftliche Reihe) 1 4 : 1 5 3-7.
Fillmore , C. J. (1968) 'The Case for Case'. In E. Bach and R. T. Harms
(eds.) Universals in Linguistic Theory. New York : Holt, Rinehart and
Winston.
Frantz, D. G. ( 1 976) 'Equi- Subject Clause Union'. Proc. BLS 2.
Getsadze , L 0., Nedjalkov, V. P., & Xolodovich , A. A. ( 1 969) 'Morfologiceskij
312 References
kauzativ v gruzinskom jazyke'. In A. A. Xolodovich (ed.) Tipologija

kauzativnyx konstrukcij : morfologiceskij kauzativ. Leningrad : Izdatel'stvo
'Nauka' .
Gonda, J. (1 960) 'Reflections on the Indoeuropean Medium'. Lingua 9 : 30-
67, 1 75-93·
Hale, K. (1973) 'Person Marking in Walbiri'. In S. R. Anderson and P.
Kiparsky (eds.) A Festschrift for Morris Halle. New York : Holt, Rinehart
and Winston.
Harbert, W. ( 1 977) 'Clause Union and German Accusative Plus Infinitive
Constructions'. In P. Cole and J. M. Sadock (eds.) Grammatical Relations.
New York : Academic Press.
Harris, A. (1973) 'On Psychological Predicates in Middle English'. LSA
annual meeting.
- ( 1976) Grammatical Relations in Modern Georgian. Dissertation, Harvard
University.
- (1977) 'Marking Former Terms : Georgian Evidence'. NELS 7 : 8 1-98.
- (1979) 'Retired Term Marking in Old Georgian'. The Elements : A Para-
session on Linguistic Units and Levels, CLS.
(ms.) 'Uls:ukcevitobis transpormaciuli analizi tanamedrove kartulSi' (A
transformational analysis of Reflexivization in Modern Georgian). Unpub
lished.
- (to appear a) 'Case Marking, Verb Agreement, and Inversion in Udi'. In
D. M. Perlmutter (ed.) Studies in Relational Grammar.
- (to appear b) 'Vnebiti kartulsi' (The Passive in Georgian). Macne.
Holisky, D. A. (1978) 'Stative Verbs in Georgian and Elsewhere'. In B.
Comrie (ed.) The Classification of Grammatical Categories (International
Review of Sla'vic Linguistics 3 . 1 -2).
- (I 980a) A Contrihution to the Semantics of Aspect : Georgian l\l1edial Verbs.
Dissertation, University of Chicago.
- (1980b) 'On Derived Inceptives in Georgian'. In B. Comrie (ed.) Studies in
the Languages of the USSR (International Re'view of Slavic Linguistics 5 . 1-2).
Irnnaishvili, V. (1957) Saxelta bruneba da brunvata punkciebi Jvel kartulSi (The
declension of nouns and the function of cases in Old Georgian). Tbilisi :
Stalininis Saxelobis Tbilisis Saxelm<;ipo Universitetis Gamomcemloba.
(1977) 'I>-vlav sereuli ls:avsirebitis sesaxeb 3vel kartulsi (Once more on
the mixed optative in Old Georgian).' Tbilisis Saxelm�ipo Universitetis
Sromebi 1 87 : 1 3-24.
Jajanidze, P. (1970) 'Sintaksuri movlenebi gurulsi' (Syntactic phenomena in
Gurian). Kutaisis Saxelm(:ipo pedagogiuri Ins!itu/is Sromebi 33 : 249-62.
Johnson, D. E. ( 1974) 'On the Role of Grammatical Relations in Linguistic
Theory'. CLS IO : 269-83.
- (1977) 'On Keenan's Definition of "Subject Of" '. LI 8 : 673-9 1 .
References 313
- (to appear) 'Ergativity in Universal Grammar'. In D. M. Perlmutter (ed.)

Studies in Relational Grammar.
J orbenadze, B. A. (1975) Zmnis gvaris pormata �armoebisa da punkciis saJt.itxebi
kartulSi (Questions of the formation and function of verbal voice forms in
Georgian). Tbilisi : Tbilisis Universitetis Gamomcemloba.
Kavtaradze, I. ( 1 954) Zmnis 3iritadi lJa!egoriebis is!oriistvis 3vel kartulfi (On
the history of the basic verbal categories in Old Georgian). Tbilisi :
Sakartvelos SSR Mecn. �ad. Gamomcemloba.
Kayne, R. S. ( 1 975) French Syntax: The Transformational Cycle. Cambridge,
Mass. : MIT Press.
Keenan, E. L. ( 1974) 'The Functional Principle : Generalizing the Notion
"Subject Of" ' . CLS 1 0 : 298-3 1 0.
- (1975) 'Some Universals of Passive in Relational Grammar'. CLS I I :
340-52•
(1976) 'Towards a Universal Definition of "Subject" '. In C. Li (ed.)
Subject and Topic. New York: Academic Press.
Keenan, E. L. & Comrie, B. (1977) 'Noun Phrase Accessibility and Uni
versal Grammar'. LI 8 : 63--99.
Kiknadze, L. ( 1937) ' Zmnis otxpirianobisatvis kartulsi' (On four-person
verbs in Georgian). �omunis!uri Ayzrdistvis: 8--9.
- ( 1967) ' Sereuli xolmeobitis m��rivi 3vel kartulsi' (The mixed habitual
screeve in Old Georgian). Orion;: Akaki Sanizes. Tbilisi : Tb ilisis
Universitetis Gamomcemloba.
Klimov, G. A. ( 1 973) Olerk obSlej teorii ergativnosti. Moscow : Izdatel'stvo
'Nauka'.
Kuno, S. ( 1 973) The Structure of the Japanese Language. Cambridge, Mass. :
MIT Press.
- (1975) 'Three Perspectives in the Functional Approach to Syntax'. Papers
from the CLS Parasession on Functionalism : 276-337.
Kuno, S. & Kaburaki, E. ( 1 977) 'Empathy and Syntax'. LI 8 : 627-72.
Kurylowicz, J. (1946) ' Ergativnost' i stadialnost' v jazyke'. E'squisses linguis
tiques. Warsaw, 1960. Translated from the Russian by Peter Culicover as
'Ergativeness and the StadiaI Theory of Linguistic Development'. In The
Study of 1l1an 2 : 1-21.
( 1964) The Inflectional Categories of Indo-European. Heidelberg : Carl
Winter Universitatsverlag.
Latsabidze, L. (1975) ' Natesaobiti brunvis pormata semantil5:uri da sintaksuri
�avsiri zmnastan tanamedrove kartuIsi' (The semantic and syntactic
relation of genitive case forms to the verb in Modern Georgian). 1l1acne
4 : 1 50-8.
Lafon, R. (1963) 'Notes explicatives a Chanidze' (cf. Shanidze 1 963). BSL
58 : 27-40 •
314 References
Lakoff, G. ( 1970) Irregularity in Syntax. New York : Holt, Rinehart and

Winston.
Lakoff, R . (1 968) Abstract Syntax and Latin Complementation. Cambridge,
Mass : MIT Press.
Lawler, J. M. ( 1977) 'A Agrees with B in Achenese : A Problem for Re
lational Grammar'. In P. Cole and J. M. Sadock (eds.) Grammatical
Relations. New York : Academic Press.
Lees, R. B. (1 963) The Grammar of English Nominalization. Indiana Uni
versity Research Center i n Anthropology, Folklore, and Linguistics.
Lomtatidze, K. ( 1 95z) ' Tbeba tipis zmnata istoriisatvis kartulsi' (On the
history of tbeba-type verbs in Georgian). IKE 4 : 75-81.
Machavariani, G .( 1 959) ' ' 'Unisno vnebiti" kartvelur enebsi' (The 'marker
less passive' in the Kartvelian languages). KESS I : 101-z9.
Marr, N. ( 1 9z5) Grammatika drevneliteraturnogo gruzinskogo jazyka
(Materialy po jafetileskomu jazykoznaniju, IZ). Leningrad.
Marr, N. & Briere ( 193 1) La langue georgienne. Paris : Libraire de Paris,
Firmin-Didot.
Martirosovi, A.
(1955) 'Masdaruli lj:onslrukciis genezisisatvis 3vel kartulsi'
(On the origin of the masdar construction in Old Georgian). IKE
7 : 4-3-59·
Matthews, G. H. ( 1965) llidatsa Syntax (Papers on Formal Linguistics 3).
The Hague : Mouton.
McCawley, J. D . ( 1 970) 'Where Do Noun Phrases Come From ?' I n R. A.
Jacobs and P. S. Rosenbaum (eds.) Readings in English Transformational
Grammar. Waltham : Ginn.
McCawley, N. A. ( 1 97z) ' On the Treatment of Japanese Passives'. CLS
8 : 259-70.
( 1 976) 'From OE/ME " Impersonal" to "Personal" Constructions :
What is a " Subject-less" S ?'. Diachronic Syntax, CLS parasession
volume.
Nebieridze, G. ( 1 976) ' Kcevis lj:ategoria kartulsi : Transpormaciuli analizi'
(The category of version in Georgian : a transformational analysis). Macne
4 : 1 32-45.
(to appear) 'Problema ergativnoj konstrukcii v kartvel'skix jazykax'. G.
Axvledianisadmi MiJVnili �rebuli (A Collection Offered to G. Axvlediani).
Nozadze, L. (1974) 'Medioakliv zmnata <;armoebis zogi salj:itxi kartulsi'
(Some questions on the formation of medioactive verbs in Georgian).
IKE 1 9 : 25-5 1.
Panchvidze, V. ( 1 937) 'Uduri ena da misi lj:iloebi' (The Udi language and its
dialects). Enim/ps Moambe 2 : 295-3 1 6.
- ( 1 942) 'Micemit-subiektiani zmnebi udur enasi' (Dative-subject verbs in
the Udi language). Enim�is Moambe IZ : 5 1-72.
References 315
Patsch, G . ( 1 952/53) 'Die georgische Aoristkonstruktion'. Wissenschaftliche

Zeitschrift der Humboldt� Universitiit Berlin (Gesellschafts- und sprachwissen
schaftliche Reihe) 1 : 5-13.
(1965) 'Die konjunctionslose Verbindung zweier Verbalbegriffe im Alt- und
Neugeorgischen'. Bedi Kartlisa 1 9-20 : I I9-28.
Peikrishvili, U. ( 1974) 'Turmeobitebis mni§vneloba da gamoqeneba axal
kartul§i' (The meaning and use of the evidential in Modern Georgian).
Kartuli Enis Strukturis SalJitxebi 4: 5 3-70.
Perlmutter, D. M. ( 1 978) 'Impersonal Passives and the Unaccusative Hypoth
esis'. Proc. BLS 4.
- (1 979) 'Working IS and Inversion in Italian, Japanese, and Quechua'.
Prot. BLS 5.
- (ms.) Untitled paper on the rule of Object Raising in Portuguese.
- (to appear a) 'Relational Grammar'. In E. Moravcsik and J. Wirth (eds.)
Current Approaches to Syntax. New York: Academic Press.
(to appear b) 'Evidence for Inversion in Russian and Kannada'. In D. M .
Perlmutter (ed.) Studies in Relational Grammar.
- (to appear c) 'Possessor Ascension and Some Relational Laws'. In D. M .
(to appear d) ' Multiattachment and the Unaccusative Hypothesis : The
Perfect Auxiliary in Italian'. In D. M. Perlmutter (ed.) Studies in Relational
Grammar.
Perlmutter, D. M. & Postal, P. M. (1974) Course in relational grammar
given at LSA Summer Institute.
( 1 977) 'Toward a Universal Characterization of Passivization'. Proc.
BLS 3.
- (to appear a) 'The I�Advancement Exclusiveness Law'. In D. M. Perl
mutter (ed.) Studies in Relational Grammar.
- (to appear b) 'The Relational Succession Law'. In D. M . Perlmutter (ed.)
Studies in Relational Grammar.
- (to appear c) 'Some Proposed Laws of Basic Clause Structure'. I n D. M.
Postal, P. M. (1 970) 'On the Surface Verb "Remind" '. LI 1 : 37-120 ; re
printed in C. J. Fillmore and D. T. Langendoen (eds.) Studies in Linguistic
Semantics, 1 97 1 . New York : Holt, Rinehart and Winston.
- (1971) Crossover Phenomena. New York : Holt, Rinehart and Winston.
( 1 977) 'Antipassive in French'. NELS 7 : 275-3 1 3 .
Reis, M . ( 1 973) ' I s There a Rule o f Subject-to-Object Raising i n German?'
CLS 9 : 5 1 9-29.
Rosenbaum, P. S. (1967) The Grammar of English Predicate Complement
Constructions. Cambridge, Mass. : MIT Press.
Sapir, E. (1917) Review of Uhlenbeck (1916), Het Passieve Karakter van het
316 Referellces
Verbum Transitivum of van het Verbum Actionis in Talell van Noord

Amerika (The Passive Character of the Transitive Verb or of the Active
Verb in Languages of North America). IJAL 1 : 82-6.
Schmidt, K. H. (1957) 'Eine siidkaukasische Aktionsart'. Miincheller Studien
zur Sprachwissenschaft 1 0 : 9-24.
(1965) 'Indogermanisches Medium und Sataviso im Georgischen'. Bedi
Kartlisa 19-20 : 129-35.
(1966) 'Zur Syntax des Kausativums im Georgischen und in indogermani
schen Sprachen'. Bedi Kartlisa 21-22: 121-7.
- (1967) 'Die Stellung des Verbalnomens in den Kartvelsprachen : Historisch
vergleichend und typologisch Betrachtet'. Bedi Kartlisa 23-24 : 1 53-60.
- (1973) 'Transitive and Intransitive'. In G. Redard (ed.) 1ndogermanische
und allgemeine Sprachwissenschaft. Wiesbaden.
Schuchardt, H. (1895) Vber dell passiven Charakter des Transitivs in den
kaukasischen Sprachell (Sitzungsberichte der Akademie der Wissenschaft,
Wien, Philologisch-historische Classe, 133, I).
Seiter, W. J. (1979) 'Instrumental Advancement in Niuean'. L1 1 O : 595-621 .
Shanidze, A. ( 1920) Subiektun' prepiksi meore pirisa da obiekturi prepiksi
mesame pirisa kartul zmnebSi (The subject prefix of the second person and
the object prefix of the third person in the Georgian verb). Tbilisi: Stamba
'Saxalxo Sakme' ; reprinted in Shanidze (1957).
- (1922-3) 'U3velesi kartuli tekstebis a¥mocenis garno' (On the appearance
of ancient Georgian texts). Tpilisis Universite#s 11,1oambe 2 ; reprinted in
Shanidze (1957).
- (1926) 'Kartuli enis zmnis sakcevi' (Version in the Georgian verb). Tpilisis
Universitetis Moambe 6 : 3 1 2-38.
- ( 1927) 'Xanmeti Mravaltavi'. Tpilisis Universitetis Moamhe 7 : 98-1 52.
- (1957) Kartuli enis strukturisa da istoriis sa1!itxebi (Questions of the struc-
ture and history of the Georgian language). Tbilisi : Tbilisis Stalinis
Saxelobis Saxelmc:;ipo Universitetis Gamomcemloba.
- (1961) 'Gramati�uli subiekti zogiert gardauval zmnastan kartulSi' (The
grammatical subject of certain intransitive verbs in Georgian). 3veli
Kartuli Enis ]f:.atedris Sromebi 7 : 207-28.
- (M. A. Chanidze) (1963) ' Le sujct grammatical de quelques verbes intran
sitifs en Georgien'. Translated by Rene Lafon, BSL 58 : 1-27.
- (1973) Kartuli ellis gramat�is sapU3vlebi I (Fundamentals of the grammar
of the Georgian language). Second edition, Tbilisi : Tbilisis Universilelis
Gamomcemloba.
(1976) 3veli kartuli enis gramati1!a (A grammar of the Old Georgian
language). Tbilisi : Tbilisis Universilelis Gamomcemloba.
Sheintuch, G. (1976) 'On the Syntactic Motivation for a Category " Chomeur"
in Relational Grammar'. SLS 6 : 49-56.
References 317
Shibatani, M. (1976) (ed.) The Grammar of Causative Constructions. New

York : Academic Press.
Silverstein, M. ( 1976) 'Hierarchy of Features and Ergativity'. In R. M. W.
Dixon (ed.) Grammatical Categories in Australian Languages. Canberra:
Australian Institute of Aboriginal Studies; New Jersey : Humanities
Press.
Sommerfelt, A. (1 937) 'Sur la notion du sujet en georgien'. In Melanges
offerts a J. van Ginneken. Paris : Klincksieck, 1 83-5.
Sridhar, S. N. (1976a) 'Dative Subjects'. CLS I 2 : 582-93.
- (1976b) 'Dative Subjects, Rule Government and Relational Grammar'.
SLS 6 : 130-51.
Tanamedrove kartuli saliferaturo enis normebi (The norms of the literary
Modern Georgian language) (1970). Tbilisi : Gamomcemloba 'Mecniereba'.
Taqaishvili, A. (1 974) '�auzativis <;armoebis zogi sa�itxi kartulsi' (Some
questions of the formation of the causative in Georgian). Kartvelur enata
s!rukluris sa�itxebi IV : 5-34.
Topuria, V. (1923) 'Sintaksuri analogiis erti semtxveva kartulsi dialektebis
mixedvit' (One instance of syntactic analogy in Georgian dialects). Cveni
Mecniereba: 1 1 3-21 .
- (1940) ' Kartvelur enata sirqva<;armoebidan II' (From word formation in
the Kartvelian languages, II). Enim�is Moambe 5-6 : 533-40.
- (1954) ' Gramati�ul movlenata ertgvarovani procesi kartvelur enebsi' (One
type of process of grammatical phenomena in the Kartvelian languages).
IKE 6 :445-55.
Trubetskoj, N. S. (1 939) 'Le rapport entre Ie determine, Ie determinant et Ie
defini'. Melanges de linguistique offerts a Charles Bally. Geneva.
Tschenkeli, K. (1958) Einfiihrung in die georgische Sprache, HI. Zurich:
Amirani.
- (1960-73) Georgisch-Deutsches Worterbuch. Zurich : Amirani.
Van Valin, R. D., Jr. (1977) Aspects of Lakhota Syntax. Dissertation, Uni
versity of California at Berkeley.
Vogt, H. ( 1 938) 'Esquisse d'une grammaire du georgien moderne'. NTS
9 : 5-1 14.
- ( 1950) 'Un aspect du probleme actif-passif dans Ie verbe'. Grammaire et
psychologie, special issue of Journal de psychologie normale et pathologique
43 : 130-8.
- (1971) Grammaire de la langue georgienne. Oslo : Universitetsforlaget.
- (1972) 'Remarques sur Ie pronom possessif reflechi du vieux georgien'
NJL(NTS) 2 6 : 91-7.
- (1974) 'L'ordre des mots en georgien moderne'. Bedi Kartlisa, 32: 48-56.
Xolodovich, A. A. (1969) (ed.) Tipologija kauzativnyx konstrukcij: morfo
logiceskij kauzativ . Leningrad : Izdatel'stvo 'Nauka'.
318 References
Ziv, Y. (1976) 'On the Diachronic Relevance of the Promotion to Subject

Hierarchy'. SLS 6 : I95-2 I 5 .
ZorreIl, F. (1930) Grammatik zur altgeorgischen Bibelilbersetzung. Rome:
Pontificium Institutum Biblicum.
Index
Notes referred to here are cited by chapter ; notes for all chapters begin on
p. 280. Small capitals refer to sub-classes listed in Appendix A of ch. 1 6, beginning
on p. 259. Entries in bold type indicate pages where a definition, explanation, or
statement of a rule can be found.
-a, see Aris-Cliticization ascensions, 6, S5 ; see also Obj ect Raising,

absolutive relation, 64, x74, 257, 277 Possessor Ascension, and Subject-to
Abxaz, 256 Object Raising
active, see active/inactive case marking aspect, 1 90, 204, 2 3 3 , n. 4 (ch. 1 6)
and stativefdynamic ; (voice) see direct Avar, 256
construction
active/inactive case marking, z35-7, Bell, S. J., n. 7 (ch. 7)
237-59 passim benefactive relation, xxi-xxii, 5, 88-I 0 I
adjectives, 58, 1 5 3 , n. 8 (ch. 1 2) ; gloss of, passim, 1 3 1 , 1 63
L...i ; in Object Raising, 53-4 Benefactive Version, see under Version
advancements, 6 ; see also Passivization, Benveniste, E., 195
Unaccusative, and Version Berman, A., 5 3 , 62
adverbial case, 1 54-6, n. 12 (ch. 10) Blake, R. P., n. S (ch. 1 3)
'affective' verbs, see inversion verbs Boeder, \V., n. 4 (ch. 3), 89, n. 1 0
agent, 1 77, 1 9 S-6, 198, 236, 252, 258, (ch. 6)
259 Briere, M., 2, n. 3 (ch. 8)
agreement, of modifier, xxi, 1 54-5
agreement, of verb, :xx, 3, see also case marking
Number Agreement and Person active/inactive, z35-7, 237-59 passim
Agreement of adjectives, xxi
Aissen, J., 7, 10, 66, 67, 68, n. 2 (ch. 5), as a basis for rule formulation, 7, 66,
70, 89, 95, 1 14, n. 1 4 (ch. 1 6) 101, 223-4, 275
Allen, B. J., n. 7 (ch. 7) as a characteristic of termhood, z,
ambiguity, n. 1 2 (ch. 8), n. 6 (ch. I I) ; 45--0, 146-50, 1 89, 1 90 ; with
see also word order, disambiguation by Causative Clause Union, 73-7, 82,
analytic passive, see under passives 85, 132, 1 33 . 1 8 2 ; with Inversion,
Anderson, S. R., 3, 3 8, 229 1 30-1 , 1 3 8-40 passim, 1#, 210, 248 ;
animacy : and agency, 177, 251 ; in with Object Camouflage, 5 I ; with
glosses, :xx ; and Number Agreement, Obj ect Raising, 59, 6 1 , 62; with
21-2 ; and selection of direct object, Passivization. 103, 108, I I I ; with
7 1 , 163-4 ; and treatment of direct Unaccusative, 201 ; with Version,
object, n. 1 8 (ch. 8), 1 87, n. 4 (ch. 90-1
1 3 ) ; see also under suppletion differential, x-z, 4, 8, 1 0, 39-4x, 66,
Arabic, n. I (ch. 1 0) 73, I I 7, 130-x, 146, 276
Aris-Cliticization, xz-x3, 54, n. 4 (ch. 4), Patterns , I
103, 107, 1 67 of predicate nominals, 6 1
Aronson, H. 1., 2, n. 3 (ch. 8), 227, n. rules, 9 , I I , 77, z43-5, 276-7
I I (ch. 1 5 ), n. I (ch. 16), 229 in Series I & I I, 39-41 ; I I , 228-47.
320 Index
249-58 passim, 270, 272 ; III, I I 7- see also under exceptions ; for Class 3 ,
I I 8 , 130-- 1 see medial verbs ; for Class 4, see
universals, 235-7, 256-9 inversion verbs
see also Retired Term Marking clause-boundedness, see under conditions
cases : gloss of, xix ; names of, I , 229 ; on rules
see also case marking and individual clause union, 6 ; see also Causative Clause
cases Union and Inceptive Clause Union
Catford, J. C., 230 clitic, see A ris-Cliticization
Causative Clause Union, 18, 66-86, 238, Cole, P., 68
244, 26 1 ; as an ergative rule, 255, 25 7 ; comitative relation, 5
with Inversion, 1 32-3 , 1 76 ; with Comitative Version, see under Version
Number Agreement, 220-- 1 ; with Comrie, B., 5, 7, 10, 66, 68, 84, 104,
Passivization, 1 14 ; 176 ; reflexivization 1 1 2, n. 5 (ch. 10), 227, 230, n. 5 & 8
with, 72, 77-8, 207-8, 209 ; retired (ch. 1 6)
terms, 9, 169, 172, 176 ; as a test of concrete/abstract, 58-9
transitivity, 1 82-4, 1 87-8, 269 ; conditions on rules, 52, 1 74
universals, 6, 7-8, I I , 66-8, 70 clause-boundedness: in Aris
causatives, B ; double, n. 6 & 8 (ch. 5), Cliticization, 1 3 ; in Question
84, 1 76 ; 'four-person', 83-5 Formation, 16-18, 59, 61-2, 62-3,
Chafe, W. L., 236 79, 8 1 - 2 ; in reflexivization, 25--6,
character vowel, n. I (ch. 6), 260, n. 7 77-8, n. I I (ch. 6)
(ch. 16) Initial Subject Constraint, 247-9,
Chikobava, A., 2, 3 , n. 3 (Intro. ), n. 7 250-2
(ch. 5), 97, n. 3 (ch. 7), n. 3 (ch. 8), lexical, 53-4, n. 1 2 (ch. 10), 1 8 8-9
1 5 1 , 1 94, n. I I (ch. 1 3), n. 5 & I I obligatoriness, n. I I (ch. 6), 95, 97, 98
(ch. 1 5 ), 227, 229, 230, 258 n. 2 (ch. 7), 1 1 9, 1 3 7 , n. I (ch. 1 3 ) ,
chomeur, 5, 7, 1 1 2-14, 1 5 1 , 168-70, 1 78 ; 1 98, 240
of Inversion, 1 1 9, 1 22, 1 28 , 134-45 relating to animacy, 2 1 -2, n. 1 8
passim, 1 79 ; of Object Raising, 1 8 , 55, (ch. 8)
60--2, 1 5 5 ; o f Passivization, 1 1 3 , 198, relating to Class and Series, I 1 9 , 1 3 7-
215, 2 1 6 , 218, 243 ; spurious chomage 1 3 8 , 247
of ergative-nominal, 1 47-9 ; of relating to level of derivation, 106,
Version, n. 4 (ch. 6), 100 ; see also 208- 10, 2 1 6ff., 243-5, 278
retired terms relating to person, 48-50, 52. 84-5,
Chomeur Law (Chomeur Condition), 5, n. 4 (ch. 6), 2 1 3ff.
7, 55, 1 1 3, 1 1 9, 122, 133, 1 34, 1 68, 1 69, relating to pronouns, 5 1 , 99
215 relating to termhood, 24-5, 27-8, 29-
Chomsky, N., 10, 1 5 2, 154 32, 3 5-8, n. 6 (ch. 1), 49- 50, I I I ,
Chung, S. , 53, n. 7 (ch. 7), 176 1 14, 1 84, n . 1 2 (ch. 1 3), 2 1 1ff.
Chxubianishvili, D., n. 7 (ch. 10), 177 see also coreference
Class, xix, 2, 4 conjoined nominals, 36-7
as a basis for rule formulation, 242-3, constituency, 66, 1 5 5 ; Question
245-55 passim Formation as a test of, 18, 59, 6 1 -2,
criteria for classification, 259--61 62-3, 79, 8 1 - 2 ; Tav-Reflexivization as
of derived constructions, 68, 73, 103, a test of, 25--6, 77-8
20 1 , 203, 237-42 constraints
invariant treatment of, 8 I , 1 7 1 , 222 Crossover, 210
morphological correlates of, 90 on derivations, n . 6 (ch. I ) , n . 4 & 6
semantic correlates of, 233, 236 , 249- (ch. 5), n. 4 & 1 2 (ch. 6), 1 1 6, 1 76
252, 258, 26 1 on grammars, 7 ; Chomeur Law, 5, 5 5 ,
syntactic correlates of, 249-55 passim, 1 1 3, 1 1 9, 1 22, 1 3 3 , 1 34, 1 68, 1 69,
260-- 1 ; case marking, 39-41, 59, 2 1 5 ; Final I Law, n. 12 (ch. 1 6 ) ;
147-50, 231-4, 23 7-45 ; Inversion, Motivated Chomage Law, 112- 1 4,
117ff. , 1 27-45 passim, 247-9 ; 1 3 6 , 1 68, 169 ; Relational
transitivity, 1 8 1 ,
186 , 1 87-90, 231- Succession Law, 5 5 ; Stratal
232, 237-42 Uniqueness Law, 100
Index 321
see also conditions on rules in Inversion, 1 1 7-35 passim, 144-5 ,
coreference 170-1, 208
Coreferential Version Object Deletion, and Number Agreement, 21 3ff., 220
3, 95-9, 1 9 3 , n. 5 (ch. 1 3 ) and Object Camouflage, 48-5 2, 1 23-4,
with dropped pronouns, 26-7 142
Equi : with infinitives of purpose, n. 8 and Obj ect Demotion, 1 88-9
(ch. 10), 1 5 6 ; with Object Raising, and Object Raising, 56-8, 60, 63-4,
5 3 , 62 I I4, 1 69, 205
lack of in Object Camouflage, 52 and preverb alternation, 200
without reflexivization, 209-10 and Possessive Version, 100
Tav-Reflexivization, 3 , 9; formulation in questions, 1 5
of, 205-10, 2 1 8, 255-6 ; in a simple retired, 1 69, 170-1, 174, 277 ; in
sentence, 23-'7 , 3 1 , 34-5 , 41 - 2, 95 - 9 Inversion, 144-5 , 1 79 ; with non
passim, 1 92, 193 ; as a test for finite verb forms, 1 5 7 , 1 5 9, 1 60- 1 ,
constituency, 77-8 ; as a test for 1 62-7 passim
subjecthood, 72, 105-6, 1 24-5, 143 and transitivity, 1 8 1 -90, 236-7
Tavis-Reflexivization, 3 , 27-8, 3 5 , and Unaccusative (synthetic passives),
42-3 1 1 8- 1 9 , 1 3 6-7, 197-200, 203, 240,
Coreferential Version Object Deletion, 24 1 , 244, 248
see under coreference Unemphatic Pronoun Drop of, 3 5 , 79
Crossover, 2 1 0 see also under suppletion
Direct Object Person Agreement, see
Dakhota, 2 3 5 , 236 Person Agreement
dative case, 1-2, 4 1 , 46, 1 5 9 , 243-5 ; in discourse, n. 4 (ch. I ) , 32-8, 79
causatives, 74-7 passim, 83 ; in Dixon, R. M. W., 230, n. 23 (ch. 1 6 )
Inversion, I 1 7-18, 1 30-1 ; in passives, double causatives, see under causatives
I I I ; in Version, 88, 99-100 double version, see under Version
Davison, A., 9, 1 3 3 dynamic, see stative/dynamic
Deeters, G., n . 3 (ch. 8)
delegative relation, 1 3 1 emeritus relations, 5, 68, 1 1 3, 168-70,
demotions, 6 ; of direct object to indirect 178 ; see also retired terms
object, 1 88-9 ; see also Inversion en chomage, see chomeur
denominals, 238-9, N English, 7, 1 6 , n. 8 (Intro.), 5 3 , 62, 6 3 ,
dependence, 5, see also constituency 64, 1 5 5 , 1 5 6 , n. 3 (ch. 10), 1 7 5 , 204,
derivation, see levels of derivation n. 4 (ch. 14)
desiderative, 24 1 , 26 1 , Q Equi, see under coreference
diagramming, xxi-xxii, 70 ; see also ergative case, 1-2, 41-6 passim, 147-50
networks passim, 244, 250 ; in causatives, 74-7
dialectal variation, see under variation passim ; irregular use of, 270, 272,
dialect described, xxii 273 ; name, 229 ; see also ergativity
direct construction, 103, 104- 1 2 passim, ergativity, 229-34, 235-7 passim, 250-2
1 9 1-9, 248 passim, 253-7, 277, n. 24 (ch. 1 6)
direct object, 5, 275 evidential, 9, 1 1 7-27, 1 28-41 passim, 1 72,
agreement markers, 29 (see also 1 8 1 , 247-9, 250-2, 254, 270
Number Agreement and Person exceptions : to agency tests, n. 6 (ch.
Agreement) 1 3 ) ; to Class membership, 1 87-90,
in analytic passives, 104-6, 108-9, 1 1 4, 245-6, 254, 255, 26 1 , 268-'74, J, K, L,
169, 206, 269-70 0 ; to the clause-boundedness of
case, 3 9-46, 73-6, 243-4, 250-2, 270 Question Formation, 1 8 ; to constraints
in causatives, 67-80, 1 1 4, 1 8 2-4, 269 on chomeurs, n. 4 (ch. 6) ; to the
coreference, 24, 27-8, 42-3 Ergative Hypothesis, 23 1-4, 250, 25 1 ,
diagramming, xxi 255 ; to generalizations about
identifying, 2-4, 8, 27-38 passim, 45- transitivity and Class, 1 87-90, 261 ;
46, 52, 63-5, 147-9 passim idiomatic, 84-5 ; to marking of retired
intransitive, 236-7, 239-41, 243-57 indirect objects, 84-5, 1 1 1 - 1 2 ; to
passim marking of retired transitive subjects,
322 Index
n. 1 3 (ch. 10), n. I (ch. I I) ; to ImnaishviIi, V., 177, n. 8 (ch. 10)
Number Agreement, n. 5 (ch. I S ) ; to impersonal inversion, see under inversion
Tav-Reflexivization, n. I (ch. 14) ; to verbs
word order in questions, n. 9 (Intro.) Inceptive Clause Union, 241 , 244
experiencer, 128, 1 32, 1 77, 236, n. 20 inceptives. 6, 240-1, 247, 2 5 1 , 2 6 1 , F
(ch. 16), 258, 259 inchoatives, see inceptives and Inceptive
Clause Union
Fahnrich, H . , 97 indirect object, s, 1 87-8, 275
'fonr-person verb forms', 83-5, 99-100, agreement markers, 29 (see also
177 Number Agreement and Person
Fillmore, C. J., 177, n. 8 (ch. 1 6 ) Agreement)
Final I Law, n. 1 2 (ch. 1 6 ) in analytic passives, 1 03-4, 1 I0-14,
Flip, see Inversion I I 5, 173, n. 1 3 (ch . 1 3 )
Frantz, D. G., n. 2 (ch. 5), n. 7 (ch . 7), case, 3 9-46, 73-6, 90- 1 , 243, 245
n. 14 (ch. 1 6) in causatives, 67-70, 75-85 passim,
Fraser, B., 10 1 69, 172, 1 82-4
French, 67, 68, I I4, 257 coreference, 24, 27, 42-3 , 95ff.
future-aorist forms. 232, 260 diagramming, xxi
future participle, see under participles identifying, 3-4, 8, 27-38 passim, 45-
46, 5 2
-gan. n. 1 (ch. I I), n. 8 (ch. 1 2) in Inversion, I I7-25, 1 27-36, 140,
gender, xx ; lack of, 23 1 4 1 , 1 69, 1 7 1 -2, 208
genitive case : distribution of, 170- 1 , 174, and Number Agreement, 2 1 3 - 1 5 ,
257; with Inversion, 1 44-5 , 179 ; with 2 1 7-1 9 , 221
non-finite verb forms, 1 5 7-9. 1 60-5 and Object Camouflage, 50-2
passim ; for possessor, 101. n. 8 (ch. and Object Demotion, 1 88-9
12) ; as a test of termhood, 1 84-5 . 269, and Object Raising, 63-4, 173
277 in questions, 1 5
German, 144 retired, 169. 171-3, 174, 277 ; in
Getsadze, I. 0., n. 1 (ch. 5) causatives, 8 1 -5, in Inversion, I I 9,
'global' rules, 9, 208, 2 1 9, 244, 278 122, 1 28 , 1 3 1 ; with non-finite verb
gloss, xviii-xxi, n. 2 (ch. 4), n. 6 (ch. 6), forms, 1 5 9, 1 62-7 passim, 1 73 ; in
193, n. 6 (ch. IS) ; of -a, 1 3 ; of turme, passives, I I O-12, n. 1 3 (ch. 1 3 ) ; in
II8 Version, 99-100, n. 4 (ch. 6)
Gonda, J., 1 9 5 in synthetic passives, 197
governance, 5, see also constituency Unemphatic Pronoun Drop of, 3 2-4,
grammatical relations, xxi, 5. 7, 8-9 ; as a 79
basis for rule formulation, 175-8, 205, in Version, 87-102, 1 I 5 , n. 7 (ch. 1 2),
275 ; see also individual relations and n. 30 (ch. 16)
inventory of initial grammatical Indirect Object Person Agreement, see
relations Person Agreement
Indo-European, 1 95
Hale, K., n. 7 (ch. I ) Indonesian, 53
Harbert, "V., 68 infinitives, 54, 60-2, 63, l S I , 154-6,
Harris, A. C., 6, 9. n. 4 (Intro.), n. I 1 57-67 . 170, 173
(ch. I), n. 4 (ch. 3), n. 10 (ch. 5), 89, Initial Subject Constraint, 247-4), 250-2
100, n. 4 & II (ch. 6), n. 2 & 1 3 (ch. instrumental relation, s, I S , 28, 3 5-6
8), 103 . 1 59, 177, n. 2 & 3 (ch. I I ), intransitives. 186. 1 87-<) ; active, 238-9 ;
1 9 1 , 1 94, n. 2 (ch. 1 3), 246 case marking governed by, 3 9-46,
Hindi. 1 3 3 1 30- 1 , 23 5-46 ; causatives of, 69-70,
Holisky, D . A . , 92. n. 2 (ch. 10). 1 90, 73-4. 76-7, 220; in the ergative
1 9 5 . 1 98 , n. 14 (ch. 1 3 ) , 232, 2 3 3 , analysis, 230-4; inactive, 2 3 9-42 ;
240. 244, n. 4 & 2 5 (ch. 1 6), 250. 261 inversion of, 1 3 5-6, 242, 254; Retired
Term Marking with, 1 7 1 , 1 74, 1 84-6 ;
idioms, 84-5 , 162 see also transitivity
ikna-passive, see under passives inventory of initial grammatical relations,
Index 323
32-5, 50, 54-5, 69, 7 1 , 88, 104. I I 9 ; 186 ; see also networks and under
and derivational sub-class, 1 81-90. rules, formulation of
1 9 1 , 196-203 passim, 237-45, 247-5 1 lexical entries. 245-55 passim ; see also
passim, 268-7 1 passim: and indirect inventory of initial grammatical
objecthood, 93-5. 100, I lO, 1 62-3 ; relations and selection restrictions
with non-finite verb forms. 1 60ff., n. locative relation. 5. n. 7 (ch. 12)
12 (ch. 1 3 ) ; and semantics, 236 Locative Version, see under Version
Inversion, 9. I I , 2 1 , II7-45, 225-6 ; and Lomtatidze, K., n. 8 (ch. 13)
case marking. 8, I30-I. 244. 245. 276 ;
with irregular sub-classes. 1 87-90 Machavariani. G., n. 8 (ch. 1 3)
passim. 270, 273 ; Number Agreement Marr. N., 2, n. 3 (ch. 8), n. I (ch. 9)
with, 217-1 9 ; refiexivization with. Martirosovi, A., n . 7 (ch. 10)
105-6. 143. 208, 209 ; retired terms masdar, 54. 15I-4, 1 57-67, 170, 1 7 1 ,
with. 169. 1 7 1 -2, 176; rule type, 6 ; 173. 177-8, 179-80
transitivity. 186 ; and the Unaccusative Masdar Formation, 9, 1 52, 1 79-80, 1 88-
Hypothesis, 242, 247-9. 250-2 passim, 189
254. 259 ; universals. 1 34-7. 248; see Matthews, G. H., 236
also inversion verbs McCawley. J . D., 10
inversion verbs (Class 4), 9. 127-33, McCawley, N. A., 9. n. 3 (ch. 5)
141-4 ; and the condition on Inversion, meaning: of indirect objects. 87-9. 94-5,
139. 140, 248 ; impersonal, 144-5. 1 69 , 1 I 0, n. 7 (ch. 12), 1 62-3. n. 30 (ch.
170-1, I79, 241-2 ; inceptives of, n. 1 4 16); with Inversion. I I 7, 129-30 ;
(ch. 1 6) ; intransitive. 13 5-7; Retired with non-finite verb forms, 152-3,
Term Marking with, 1 77-8 1 62-3 ; with Unaccusative, 193 ; see
Italian, 237, 240 also ambiguity and semantics and
iqo-passive, see under passives individual semantic notions
medial verbs (Class 3), M, N. 0 ; case
Jajanidze, P., n. 9 & 18 (ch. 16) marking governed by, 39-46, 1 30-1 ;
Japanese, 68, n. 3 (ch. 5). 257 clause structures governed by, 238-9 ;
Johnson, D . E., 4. 6, n. 2 (Intro.), 230, in the ergative analysis, 230, 23 1-4,
278 253-4; as an Inversion trigger, 1 3 5-6 ;
Jorbenadze, B. A., n. 9 (ch. 16) irregular. 245-6, 270-4 ; morphology,
90. 260; transitivity of. 18 1-6. 189-90 ;
Kaburaki, E., n. 4 (ch. I) see also intransitives
Kannada, 1 3 3 middle verbs, see medial verbs
Kavtaradze. L. n . 1 0 (ch. 13) middle voice, 195
Kayne. R. S .• 68 mier : assignment of, I74, 175, 1 77-8,
Keenan, E. L . • 3, 5 , 7 . 38. 104. 1 I 2. n. 8 (ch. rz), 257, 277 ; -norninals in
227, 278 coreference, 206-7 ; with non-finite
Kiknadze. L. 83. 84, 99, n. 33 (ch. 1 6) verb forms, 1 58, 1 7 1 ; in passives, 107-
Klimov. G. A., 229. 230, n. 8 (ch. 1 6) 108. 1 69 , 1 7 1 ; as a test of transitivity,
Kuno. S., 68. n. 4 (ch. I), n. 3 (ch. 5) 1 84-5, 1 96 -7, 1 98-9, 269
Kurylowicz. J., 195. 230 Mohawk, 237. 240
mood. 46, 1 47 ; see also evidential
Lafon. R.. n. 3 (ch. 8) morphology, see case marking, Number
Lakhota. 235 Agreement, Person Agreement. and
Lakoff, G. • 9, 10, 1 3 3 under verbs
Lakoff, R., 9 , 1 3 3 Motivated Chomage Law, I I 2-14, 1 36,
Latsabidze, L . , n. 2 (ch. I I ) 168. 169
Lawler, J. M . , n. 7 (ch. 7)
Lees, R. B., 10, n. 3 (ch. 10) Nebieridze. G., n. 10 (ch. 6), 229
levels of derivation, 6, 38; and activeJ negative, n. 1 (ch. 8)
inactive rules, 236, 257-9 ; and the networks : Causative Clause Union. 69-
application of rules, n. 3 (ch. 14), 278 ; 70; Class I, 23 8 ; Class 2, 239-4 1 , n .
and grammatical relations, 3 ; and 1 3 & 1 4 (ch. 1 6) ; Class 3 , 2 3 9 ; Class
retired terms. 1 6 8 ; and transitivity, 4. 242 ; direct construction, 1 97-8,
324 Index
24B, 250 ; Inversion, I I 9, 124, 1 28, order : of rules, see rule interaction ; of
140, 248 ; Object Raising, 5 5 , n. 13 words, see word order
(ch. 16); Passivization, Il3, 137, 1 97-
1 98, 249 ; Unaccusative, I I 9, 124, 1 28 ,
140, 1 97-8, 241-2, 248, 250; Version, 88, Panchvidze, V., n. 8 & 1 3 (ch. 10), 177
n. 4 (ch. 6), 27 1 participles: future (passive), 1 5 1 , 1 54 ;
nominal, internal structure of, 5, 1 5 3 'future participle in the adverbial
nominalization, ro-I I ; see also masdar case', I 54ff. ; past (passive), 1 3 , 1 03 , 1 5 1 ,
nominative case, 1-2, 41, 42, 46, 1 47-9, n . 1 3 (ch. 10), 1 94, 204; present
231, 243-5, 250 ; in analytic passives, (active) participle, 1 5 1
103, 104 ; in causatives, 74-'1 passim ; passives : analytic (ikna-), :203-4, 239,
in Inversion, I I 7-1 8, 130-1 ; irregular, 261 , G; analytic (iqo-), I03-I6, 1 9 1 -9,
272, 273 ; in object raised construc 203-4, 215ff. , 239-40, 261, D ;
tions, 56-9 ; in synthetic passives, 20 I synthetic, I9I-204, 2I6ff., 239-40,
non-finite verb forms, 15 1-67; 247, 26 1 ; E ; Zustandspassiv, 203-4,
dependents of, 62, 1 56-67 ; see also 239, 261, H ; see also Passivization and
infinitives, masdars, and participles Unaccusalive
non-terms, 5; see individual relations Passivization, I03-I6, 203, 2 1 5 ; with
nouns, gloss of, xix ; see also case Causative Clause Union, I I 4, 1 76 ;
marking with Inversion, 1 3 8-9, 247, 249 ; in the
Nozadze, L., n. 4 & 7 echo 16) overall verb system, 198, 203, 240,
nuclear terms, 5, 244. 277 ; see also 241, 243, 244; reflexivization with,
subject and direct object 105-{), 206-7, 209 ; retired terms with,
number, 103. 1 5 3 ; gloss of, xix, xx; and 9, 1 69, 1 7 1 , 173, 175 ; as a test for
preverb alternation, 200 ; see also direct object, 1 87-90 passim, 269-7° ;
Number Agreement and under transitivity, 1 8 6 ; universals of, 4-5 , 6,
suppletion 7, I04ff. , I I 2-I4; see also passives,
Number Agreement, 9, 34, 2H-27, 255 ; analytic
animacy condition, 21-2, 5 4 ; gloss of, past participle, see under participles
xx ; test for termhood, 3 , 85, 100, patient, n. 10 (lntro.), 236, 252, 258
149-50 ; see also Person Agreement Patsch, G., 2, n. 10 (ch. 10)
Patterns, I ; see also case marking
object, 1 5 ; see also direct object and PeikrishviIi, U., n. I (ch. 8)
indirect object perfect, see evidential
Object Camouflage perfective, see aspect
agreement, 3 I , 222 Perlmutter, D. M., xxi, 3 , 4, 5, 6, 7, 1 0,
applicability of, 48-52, 80, 91-2 3 8, 5 3 , 5 5 , 67, 68, 85, 89, 1 00, 104,
failure to apply : with final intransitives, I I 2, I I 9, 1 34, 1 3 7 . n . 14 (ch . 8), 1 6 8 ,
108-9, ! I 8, 123-4, 142-3, 201-2, 1 69, 1 7 6 , 1 98, 236, 237, n. 1 2 & 1 4
2 1 0 ; to retired terms, 1 65 , 1 69 (ch. 1 6)
Object Demotion (2-to-3 Retreat), 1 88-9 person, xix, 48-52 passim, n. 4 (ch. 6),
Object Raising, 6, 1 8 , 53-65, n. 1 3 (ch. 2 I3ff. ; in idioms, 84-5
16) ; case, 243, 244 ; and infinitives, Person Agreement, 3. 1 89, 1 90 ; with
ISS. 1 5 6 ; with Inversion, 1 3 9-40, 247 ; analytic passives, 106-7, I I I , 1 38,
with Number Agreement, 2 I 6ff. ; 139; with Causative Clause Union, 78,
with Object Camouflage, n. 5 (ch. 82-3, 85, 132, 133, 1 8 2 ; gloss of, xx ;
5 ; with Passivization, I I4 ; with Inversion, 121-3. 1 32, 1 3 3 , 141-
reflexivization, 205-6, 2°9; retired 142, 144, 1 79, 210, 226, 248, 252; lack
terms, 9, 169, 173, 178-9 ; as a test of, 62, 82-3, 85, 95, 100, 122, 144,
for terms, 188-9 ; universals, 5 3 , 55, 62 1 64-5, 1 6 9 ; with Object Camouflage,
Oblique Law, 7 5 1 ; with Object Raising, 60, 62, 139,
obliques, 5, 24, 28, 3 6 ; see also individual 140; principles of, 28-32 ; in Series II ,
oblique relations 43-5, 147-8 ; with synthetic passives,
Old Georgian, n. 5 (ch. I), n. 10 (ch. 201 ; with Unemphatic Pronoun Drop,
5), 100, 1 59, n. 7 (ch. 10), 176, n. 2 33 ; with Version, 88, 89-90, 94, 95,
echo I I ), 188 , 227 99, 100; see also Number Agreement
Index 325
plural, see number and Number retired terms, S, 8, I I, I l 3 ; 168-'70, 189,
Agreement 269, 277 ; diagramming of, xxii ;
polypersonalism, 28-3 1 , 2 I I- 13 ; see also marking of, 168-80, 244-5 ; see also
Person Agreement chomeur and emeritus relations and
Portuguese, 53 under direct object, indirect object,
possession, 1 5 3 ; gloss of, xix; see also and subject
possessor relation and Version, retired verbs, 68
Possessive ; reflexive, see Tavis revaluations, 6 ; fOT advancements, see
Reflexivization under coreference Passivization, Unaccusative, and
Possessive Version, see under Version Version ; for demotions, see I nversion
Possessor Ascension, 88-1) and Object Demotion
possessor relation, 37-8, 100, 101, 102, Rosenbaum, P. S., 9, 1 3 3
n. I I (ch. 6) n. 8 (ch. 12) rule interaction, 4 , 9 , 3 I , 5 I , 72-85 . n.
Postal, P, M. , xxi, 3, 4, 5, 6, 7, 9 , 10 , 3 8, 4 & 7 (ch. 6), r I 4-I 6, 1 23-4, 138-41 ,
53. 55 , 67. 68, 85. 100 , 104, 1 I 2, 1 I 9, n. 8 (ch. 10), 202, n . 1 2 (ch. 1 3), 20S-
133 . 1 34. n. 14 (ch. 8), 168 , 169 . 176, 209, 215-18, 247-9, 250-2, 2SS, 277-
198, 236, 237, n. 12 (ch. 1 6) 278 ; see also individual rules and levels
postpositions, I S . 24, 28, 36, 63-4, 88, of derivation
1 5 3 ; gloss of, xix; see also -gan, mier, rule order, 9, 205, 277 ; see also rule
-tan, -tvis interaction
potential, n. 2 (ch. 1 3 ) rules
predicate nominals, 1 3 , 58, 6 1 alignment of, 257-9
predicate relation, xxi, 5 formulation of: basis for. 6-9 passim,
present participle, see under participles 66, 101, 175-8, 205, 223-4, 257-8,
preverbs, 20, 32, 1 8 1 , n. I (ch. 1 2), 200, 275 ; language-specific problems, 52,
204, 232, n. 4 & 7 97-102. 210; reference to more than
(ch. 1 6) one level of derivation, 9, 205, 208-
pronouns : camouflaged, S I-2 ; gloss of, 209, 2I6ff., 244-5 , 278 ; universal
xix ; interrogative, 14, S I ; lack of case problems, 67-8, 85-6, I I2-I4, 1 33-
distinction in first and second persons, 1 3 8 , 17S-8
xix ; lack of gender, 2 3 ; reflexive, 23- types of, 3-4, 6, 9, 88-9, 174. 2 78
24, 27, 3 1 , S I ; personal, S I , n. S (ch. 7), see also conditions on rules, rule
n. 14 (ch. 1 0) ; see also Unemphatic Pro interaction, and individual rules
noun Drop
Psych Movement, see I nversion Sapir, E., 230, 23S, 236, 237, 2S6, 257 .
purpose clauses, I S S , I S6, 1 62, 170, 1 73 258
Schmidt, K. H., n. I (ch. 5), 97, n. 5
Question Formation, 14-18, 59, 6 1-2, (ch. 10), n. S (ch. 1 3 ) . 200, 229
62-3, 79, 81-2, 166-7, indirect Schuchardt, H . • n. I (ch. 9)
questions, 1 6, 6 1 screeve, 2. 2 1 , 29. 46-7, 147
Seiter, W. J.. n. 14 (ch. 6)
raising, see Object Raising and Subject- selection restrictions : animate/inanimate,
to-Object Raising 7 1 , 1 63-4; concrete/abstract, 58-9
redundancy rules, 1 80 semantics: as a basis for rule formula
reflexivization, see coreference tion. 1 75 . 177-8, 2S7-8 ; and inventory
Reis, M., 68 of initial grammatical relations, 236 ;
relational grammar, 4-8, I I , 66, 102, representation, 72 ; see also meaning
1 I 2-I4. 1 I 8, 1 68-9, I 74-S, 23 6ff. and individual semantic notions
relational hierarchy, S, 2 1 9, 227 Series, xix, 2, 46-7. 103 , 222-3 , 243-5 ;
Relational Succession Law, 55 I & II , 39-47. 74-6, 9 1 , 99 ; II , 146-
relative clauses, n. 9 (I ntro.). I S 3 I S O, 228-47. 249-54, 270 ; I I I , 1 17-
Retired Term Marking, 9 , I S 2 , 168-80 ; 127, 128-41 passim, 172 . 1 8 1 , 247-9.
as an ergative rule, 174, 257. 277 ; as a 250-2, 254, 270
test of transitivity, 184-6, 1 87-90 Shanidze, A .• 2, n. I I (I ntro.), 29, 47,
passim, 269 ; universals, I I 2- I 3 , 168-9, 83. 84, n. I & 6 (ch. 5). 97, 99. 100,
J75-8 n. 3 (ch. 7). J I 8 , n. 3 & 8 (ch. 8).
326 Inde�
1 5 1, n. 2, 5 , & 6 (ch. 10), 1 76, 1 94, Subject Person Agreement. see Person
199-200, n. I I (ch. 13). 227. 229, 259 Agreement
Sheintuch, G., n. 7 (ch. 7) Subject-to-Object Raising, n. 1 2 (ch. 10)
Shibatani. M . • 8, 66 superessive relation, 5, 1 00, 102
Silverstein, M. 230 Superessive Version, see under Version
Siouan languages. 235. 236 suppletion, 1 8-21 . 168; for animacy of
Sommerfelt. A., 3 direct object, If)-20. 5 7-8, 105, 1 2 6-7 ;
spontaneous chomage, I I 2 for number of direct object. 20. 56-7,
Sridhar, S. N., 9, 68, 1 3 3 104-5, 1 25-6. 1 60-1, 1 99-200 ; for
stative/dynamic, 92-3, 204, 233. 2 3 5 , 250 number of subject, 21, 1 6 1 ; for tense,
stimulus, 128, 236 21, 1 2 9-30
strata, see levels of derivation synthetic passive, see under passives
Stratal Uniqueness Law, 7, 100
subject, 5. 275 -tan. n. 4 (ch. 6), n. 7 (ch. 1 2), n. 30 ;
agreement markers, 29; see also (ch. 1 6)
Number Agreement and Person Tanamedrove kartuli saliteraturo enis
Agreement normebi, n. 5 (ch. 7)
in analytic passives, 1 04-9, 139, 1 69, Taqaishvili, A., n . I (ch. 5)
r 7 1 , 206, 269-70 tav-, see pronouns, reflexive and
with Aris-Cliticization, 1 3 pronouns, camouflaged
case, 39-46, 62, 74-7. 243-4, 250-2, tavis-/taviant-, see pronouns, reflexive
270 Tavis-Reflexivization, see under
in causatives, 67-9, 182-4, 207-8, 220, coreference
269 Tav-Reflexivization, see under
and Coreferential Version Object coreference
Deletion, 95-9 passim tense, xx, xxi. 103 ; -aspect categories, 2,
diagramming, xxi 2 1 , 29, 46-7, 129-30, 1 47 ; sequence
first, 256, 278 of, 18
identifying, 2-4, 8, 23-38 . 45-6, I I9, term relations, 5, 23-38, 2I I-1 3 , 275 ;
147-50 constraints, 208-9. 2 16-1 8 ; see also
intransitive : of causatives, 70, 73-4, direct object, indirect object, inventory
79, 1 8 2-6, 257 ; rctired, 1 5 7-8, 1 59, of initial grammatical relations, and
1 6 1 , 1 7 1, 1 74, 179 subject
in Inversion, 1 1 7-21, 124-36. 1 40-5 Topuria, V., n. 10 (ch. 1 3), n. 9 & 1 8
passim, 169, 177. 1 79, 208, 247-8, (ch. 16)
270 Tough-Movement, see Object Raising
and Number Agreement, 2 13ff. transitive inversion verbs with no overt
in object raised constructions, 58-60, subject. I44-5, 169. 1 70 - 1 . 179. 241-2
62, 63-4, 1 40, r 69 transitives. 186, 1 89-90, 194. 248, 258 ;
in questions, 1 5 case marking governed by. 3 9-46,
retired, 169, 171, 174, 1 78, 277 ; with 1 30-1 , 23 5-46 ; causative of, 69-70,
non-finite verb forms, 1 5 7-8, 16 1-7 75-6, 8 1-5, 1 82-3, 221 ; inactive, 2 5 8 ;
passim, 1 77. 1 84-6, 269 ; in object inversion of, I I7-31 , 1 3 5-6, 242, 2 5 8 j
raised constructions, 5 5 , 6 1 , 63 ; in Retired Term l\1arking with, 1 70-1 ,
passives, 1 07-8 174, 1 84-6 ; see also transitivity
in synthetic passives, 1 97-202 transitivity, 181-90, 230, 236, 253, 257-
and Tav-Reflexivization, 24-5, 41-2, 259, 26 1 , 268-70 ; and Retired Term
72, 105-6, 124-5. 205- 10 Marking, 1 7 1 ; and Term Case
and Tavis-Reflexivization, 27-8, 42-3 Marking, 39-4 1 , 229-40, 249-53 ,
transitive : of causative, 70, 75, 79. 257-9; see also intransitives and
1 82-4 ; retired, 1 58, 1 7 1 , 174, 178, transitives
1 79. 1 84-6 transliteration, xvii-xviii
and transitivity, 41. 182-6. 236 Trubetskoj . N. S., 230
Unemphatic Pronoun Drop of. 3 5, Tschenkeli, K., 2, n. I I (Intro.), 29, 47.
78-9 n. I (ch. 5). 97, n. 3 (ch. 7), 1 I 8, n.
see also under suppletion 3 & 16 (ch. 8), 1 5 1 , 1 94, 200, n. I I
Index 327
(ch. 1 3), n. 9 (ch. 1 5), 229, n. 6 & 3 1 governed by lexical items, n. 1 8 (ch.
(ch. 1 6) 8) ; in marking retired terms, I I I
Turkish, 68, 1 14, 1 44, 257 I I 2, n. I & 4 (ch. 1 1) ; in marking
turme, l I S terms, 189, 246, 268-74
-mis in meaning of nominalizations, 1 5 3
assignment of, 171-3, 174, 176, 277 i n rules, 5 3 , 68
with obliques ; in object raised verbs
constructions, 1 8 , 61-2, 62-3, 1 69, gloss of, xix-xx
178-9. 205-6 ; other, 90, 1 0 1 , 1 3 1 morphology of, n. I (ch. 6), n. 5 (ch.
with retired indirect objects : in 8), 1 74 ; analytic passives, 103 ;
causatives, 81-5, 1 6 9 ; with non causatives, n. I (ch. 5) ; Class, 190,
finite verb forms, 1 5 9, 162-3, 1 6 5 ; 2 32-4, 25�o, 268, 270-2 ;
in passives, 1 10-12, 1 1 3, I I 6, n. 1 3 infinitives, 1 5 5 ; inversion, 1 3 9-40 ;
(ch. 1 3 ) ; in Series I I I , I , 1 22-3, Number Agreement, 223-4 ;
1 30, 1 3 1 synthetic passives, 1 9 4 ; version, 88,
Tzotzil, 95 90, n. 6 (ch. 6), 97
see also Class, inventory of initial
Udi, 256 grammatical relations, Number
Unaccusative ; with intransitive direct Agreement, Person Agreement,
objects, 1 9 1-204, 2 1 6, 240, 24 1 , 244, preverbs, selection restrictions,
248; with Inversion, 1 1 9. 1 3 3 . 1 3 4-7, Series, tense, and transitivity
1 79, 1 86, 242, 247 ; Passivization Version, 87-102, I I 5-1 6, 221-2, 244,
compared with, 1 36--7 , 197-8. 203 ; 245 ; Benefactive, xxii, 87ff., I I 6, n. 5
see also passives, synthetic (ch. 1 3 ) ; Comitative, n. 30 (ch. I 6) ;
Unaccusative Hypothesis, 235-46, 249- double, 99-100; Locative, 1 8 7 ;
256, 268 Possessive, 87-9, n . 1 1 (ch. 6), 100- 1 ,
Unemphatic Pronoun Drop ; conditions I I 5 , 1 93 , n . 5 & 1 2 (ch. 1 3 ) ;
on, 5 1 , 99, 1 84 ; failure to apply to Superessive, 87-9
retired terms, 62, 82, 88, 9 1 , 108, I I I ; version marker. n. I (ch. 6), 88. 89. 90.
principles, 32-8, 1 56, n. 2 (ch. 1 4) ; n. 7 (ch. 1 2)
with terms, 3 , 26-7, 45, 78-9, 85, 88, version object (version nominal), 88, 92,
9 1 , 100, 1 49, n. 7 (ch. 1 2) 95, 96, 1 0 1
universals, 7-8, I o-I I , 1 74-5 ; of version vowel, see version marker
agreement, 227 ; of case marking, 235- Vogt, H . , 2, 3, 22, 29. n. 5 & I I (ch. I ) .
237, 256-9 ; of causatives, 66--8 , 70 ; of 83 , 84, n. I & 6 (ch . 5), 97, n. 3 & 6
indirect objects, 94 ; of Inversion, 1 34- (ch. 7), 1 18, n. 3 (ch. 8). l S I , n. 4
1 37, 248 ; of Object Raising, 5 3 , 5 5 , (ch. 10). 1 94 , n. I I (ch. 13), n. 9
62; of non-finite verb forms, l S I , 1 60 ; (ch. 1 5), 229. n. I (ch. 1 6)
of passives, 104ff., I I 2- I 4 ; of Retired voice, 194; see also direct construction
Term :Marking, 1 75-8 ; of retirement and passives
relations, 1 1 2- 1 3 , 1 68-9 ; of rule
alignment, 257-9 ; of subjects, 3 ; of word order. 7-8, 1 54, n. 6 (ch. 1 3 ) ;
voice distinctions, 1 98 disambiguation by, n. I (ch. 3), n. 1 2
unspecified nominals, n. 1 4 (ch. I ) , I I I , (ch. 8) ; with respect to inversion
170, 1 93 , 1 9 7 constructions, 1 3 3 - 4 ; in questions,
14-16, 1 8 , 6 1 , 63. 1 66-7 ; with respect
Van Valin, R. D., Jr, 2 3 5 to reflexivization, 24-5, n. 6 (ch. I ), 1 2 5 ,
variation
1 43 ; unmarked, 1 3 , 22. 1 0 1 , I 3 3-4,
dialectal, n. I (ch. I ), n. 4 (ch. 3),
1 55 , n. 7 (ch. I S )
n. I (ch. 4), 1 8 9 ; in marking retired
indirect objects, 84-5, n. 4 (ch. 6),
99-100, 1 1 1 -12, 1 76 ; in marking Xolodovich, A. A .• 66
retired subjects, n. 1 3 (ch. 10), n. I
& 3 (ch. I I ) ; in verb agreement Ziv, Y., 9
29. n. 5 & 9 (ch. 1 5 ) ; in marking Zorrell, F., 227
subjects, 246, 272-3 Zustandspassiv, see under passives
View publication stats

Harris Georgian Syntax 1981 PDF

Uploaded by

Copyright:

Available Formats

Harris Georgian Syntax 1981 PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Harris Georgian Syntax 1981 PDF

Uploaded by

Copyright:

Available Formats

In this series

I DAVID CRYSTAL: Prosodic systems and intonation in English·

4 JOHN M.ANDERSON: The grammar of case·

5 M.L.SAMUELS: Linguistic evolution·

6 P.H.MATTHEWS: Inflectional morphology.

7 GILLIAN BROWN: Phonological rules and dialect variation·

8 BRIAN NEWTON: The generative interpretation of dialect·

9 R.M. W.D IX 0 N: The Dyirbal language of North Queensland ·

10 BRUCE L.DERWING: Transformational grammar as a theory of language

12 W. SIDNEY ALLEN: Accent and rhythm

13 PETER TRUDGILL: The social differentiation of English in Norwich·

14 ROGER LASS and JOHN M.ANDERSON: Old English phonology

IS RUTH M.KEMPSON: Presupposition and the delimitation of semantics'"

16 JAMES R.HURFORD: The linguistic theory of numerals

17 ROGER LASS: English phonology and phonological theory

18 G.M .AWBERY: The syntax of Welsh

19 R.M.W.DIXON: A grammar of Yidi/l

20 JAMES FOLEY: Foundations of theoretical phonology

21 A.RADFORD: Italian syntax: transformational and relational grammar

22 DIETER WUNDERLICH: Foundations of linguistics·

23 DAVID W.LIGHTFOOT: Principles of diachronic syntax·

24 ANNETTE KARMILOFF-SMITH: Afunctional approach to child language·

25 PER LINELL: Psychological reality in phonology

26 CHRISTINE TANZ: Studies in the acquisition of deietic terms

27 ROGER LASS: On explaining language chonge

28 TORBEN THRANE: Referential-semantic analysis

29 TAMSIN DONALDSON: Ngiyambaa

30 KRISTJAN ARNASON: Quantity in historical phonology

31 JOHN LAVER: The phonetic description of 'Voice quality

32 PETER AUSTIN: A grammar of Diyari, South Australia

33 ALICE C.HARRIS: Georgion syntax

·Issued in hard covers and as a paper-back

A study in relational grammar

CAMBRIDGE UNIVERSITY PRESS

LONDON NEW YORK NEW ROCHELLE

© Cambridge University Press I98I

First published I98I

Printed in Great Britain by Western Printing Services Ltd, Bristol

British Library Cataloguing in Publication Data

ISBN 0 52I 2358 4 7 Univ.' Blbliothek

This book addresses a number of central issues in linguistic theory.

the marking of 'retired terms' - nominals that bear a term relation

description that has produced this extensive documentation of Georgian

San Diego, California

1 SYNTACTIC TESTS FOR TERMHOOD

2 CASE MARKING IN SERIES I AND II 39

I. The facts to b e considered

3. Interaction with some rules previously considered

5 CAUSATIVE CLAUSE UNION 66

·6 VERSION: RULES THAT CREATE INDIRECT OBJECTS 87

9 WHY PATTERN A IS NOT REDUCIBLE TO PATTERN B 1 46

10 NON-FINITE VERB FORMS 151

II RETIRED TERM MARKING 168

13 SYNTHETIC PASSIVES 191

15 NUMBER AGREEMENT 211

16 THE NATURE OF THE GEORGIAN VERB CLASSES 228

This investigation of Georgian syntax originated as my 1976 disser­

A great many Georgians, in the United States and in Georgia, have

I Transliteration. The following system is used:

This investigation of Georgian syntax originated as my 1976 disser

indicated with an Arabic numeral. For example, dafera 'he-wrote-it

This network indicates, among other things, 'bavsvi is an initial benefac

3.2.2 Retired Term Mark-ing. At least six different rules in Georgian

3.2.4 Rule interaction. It is found to be unnecessary to impose conven

t he direct object, number of the direct object, or number of the sub

h) sentences, inanimate. (22) contains the form miitana, and is gram