Current issues in linguistic theory 71 William ogrady categories and case the sentence structure of korean john benjamins publishing company (1991)

While working on this project I benefitted from the assistance of a number of people. First of all, I would like to express my deep gratitude to Sook Whan Cho, whose work on the acquisition of Korean first got me interested in that language and who subsequently encouraged me to work on the problem of case marking. Special thanks are also due to YoungSeok Choi and SungOck Shin, who served as my principal informants and who provided invaluable comments on an earlier version of this manuscript. I am also very grateful for the help of D.J. Lee, my first Korean teacher, who read and commented on different versions of the manuscript, noting many problems and providing invaluable help with the Korean data. His assistance and insights helped me find my way out of many deadends. Don Frantz also read the manuscript in its entirety and made many valuable suggestions, for which I am very grateful. Additional important data and comments were provided by Younghee Na, KeonSoo Lee, Yutaka Sato, Kazue Kanno, GyungRan Kim, Hakneung Joo, EungDo Cook, Joe Ree, Konrad Koerner and two anonymous referees, to whom I express my thanks. Needless to say, any remaining errors are my responsibility. The production of the final version of the manuscript was due in large part to the efforts of Betty Lewis, Susan Langley and Kazue Kanno. Without their help and hard work, this project could not have been completed.

CATEGORIES AND CASE

AMSTERDAM STUDIES IN THE THEORY AND HISTORY OF LINGUISTIC SCIENCE

General Editor

E.F KONRAD KOERNER (University of Ottawa)

Series IV - CURRENT ISSUES IN LINGUISTIC THEORY

Advisory Editorial Board

Henning Andersen (Los Angeles); Raimo Anttila (Los Angeles) Thomas V Gamkrelidze (Tbilisi); Hans-Heinrich Lieb (Berlin)

J Peter Maher (Chicago); Ernst Pulgram (Ann Arbor, Mich.)

E Wyn Roberts (Vancouver, B.C.); Danny Steinberg (Tokyo)

Volume 71

William O'Grady

Categories and Case The Sentence Structure of Korean

CATEGORIES AND CASE THE SENTENCE STRUCTURE OF KOREAN

WILLIAM O ' G R A D Y

University of Calgary

&

University of Hawaii at Manoa

JOHN BENJAMINS PUBLISHING COMPANY

AMSTERDAM/PHILADELPHIA

1991

Library of Congress Cataloging-in-Publication Data

O'Grady, William D (William Delaney),

1952-Categories and case : the sentence structure of Korean / William O'Grady

p cm - (Amsterdam studies in the theory and history of linguistic science

Series IV, Current issues in linguistic theory, ISSN 0304-0763; v 71)

Includes bibliographical references and index

1 Korean language ~ Morphology 2 Korean language ~ Case grammar 3 Korean

language - Categorical grammar I Title II Series

PL919.035 1991

495.7'5 dc20 90-42137 ISBN 90 272 3569 7 (Eur.)/l-55619-127-8 (US)(alk paper) CIP

© Copyright 1991 - John Benjamins B.V

No part of this book may be reproduced in any form, by print, photoprint, microfilm, or

any other means, without written permission from the publisher

Contents

Acknowledgements vi

List of Abbreviations vii

The Yale System of Romanization viii

1 The Problem 1

2 The Grammar 13

3 Grammatical Relations and Thematic Roles 25

4 The Case System 33

5 Passivization and Dative Advancement 47

14 Some Special Challenges 225

15 The Acquisition Problem 249

16 Concluding Remarks 269

Notes 275 References 283 Index 293

Acknowledgements

While working on this project I benefitted from the assistance of a number of people First of all, I would like to express my deep gratitude to Sook Whan Cho, whose work on the acquisition of Korean first got me interested in that language and who subsequently encouraged me to work on the problem of case marking

Special thanks are also due to Young-Seok Choi and Sung-Ock Shin, who served as my principal informants and who provided invaluable comments on an earlier version of this manuscript I am also very grateful for the help of D.-J Lee,

my first Korean teacher, who read and commented on different versions of the manuscript, noting many problems and providing invaluable help with the Korean data His assistance and insights helped me find my way out of many deadends Don Frantz also read the manuscript in its entirety and made many valuable suggestions, for which I am very grateful Additional important data and comments were provided

by Younghee Na, Keon-Soo Lee, Yutaka Sato, Kazue Kanno, Gyung-Ran Kim, Hak-neung Joo, Eung-Do Cook, Joe Ree, Konrad Koerner and two anonymous referees, to whom I express my thanks Needless to say, any remaining errors are

my responsibility

The production of the final version of the manuscript was due in large part to the efforts of Betty Lewis, Susan Langley and Kazue Kanno Without their help and hard work, this project could not have been completed

The Yale System of Romanization

(' marks a tense consonant)

Hangul Yale Basic Phonemic

***1***

The Problem

The study of morphological case constitutes one of the oldest and most puzzling research problems in the study of language From the time of Panini in ancient India and the Stoics in ancient Greece, the role of case affixes in sentence formation has represented an ongoing challenge for linguistic analysis If there is one sentiment shared by all those who have examined this problem, it is that case has a very fundamental grammatical function, as evidenced by both its ubiquity in human language and by its association with such basic syntactic notions as subject and direct object What still remains to be discovered after several centuries of inquiry is the precise nature of the information encoded by case categories and the motivation for this particular use of a grammar's resources This book seeks to uncover the principles governing the use of case suffixes in Korean, a language spoken by over sixty million people on the Korean peninsula as well as in parts of China and the Soviet Union (not to mention immigrants to the United States and Canada)

To the speaker of English who is exposed to Korean for the first time, at least two properties of sentence structure come immediately to the fore First, the Korean verb is heavily agglutinating, frequently consisting of half a dozen or more morphemes For example, the verb form used in a sentence such as 'The deposed

king was caught ' has the structure depicted in (1) (Sohn 1987)

(2) a Ai-ka pap-ul mek-ess-ta

child-N food-Ac eat-Pst-Decl

'The child ate the food.'

2 CATEGORIES AND CASE

b Pap-ul ai-ka mek-ess-ta

food-Ac child-N eat-Pst-Decl

'The child ate the food.'

This 'free word order' option is made possible in part by the existence of the case particles whose precise properties constitute the central concern of this book

Before proceeding, it is necessary to comment briefly on the presentation of the example sentences used throughout this book Following the common practice in linguistic work on Korean, I use the Yale system of romanization (Martin 1954, 1963), with a few minor modifications.1 Yale romanization establishes a virtual one-to-one correspondance between Roman letters (including some digraphs) and the

symbols of Hangul, the Korean alphabet (see the table at the beginning of this book)

For those unfamiliar with Korean, it is perhaps worth mentioning that because

Hangul is a morphophonemic orthography, neither it nor the Yale system of

romanization provides a straightforward phonetic record of Korean speech

The Korean verbs used in the vast majority of example sentences in this book

are in the so-called 'plain' form, consisting of a root, a tense marker, (-ass , -ess or -ss for the past, 0 for the non-past), the 'processive mood' marker -(nu)n (for non-

stative verbs in the non-past), and the declarative suffix -m , which can also be used without tense suffixes to indicate a verb's neutral or 'dictionary' form Since tense and mood markers are irrelevant to the issues that concern me here, I will consistently ignore them in the glosses I give for the Korean examples In order to further facilitiate the analysis of example sentences by nonspeakers of Korean, I will use English proper names as subjects and direct objects where feasible In this, I follow a practice common in the literature on Korean

The Case Suffixes of Korean

In the pages that follow I will be primarily concerned with the function and

distribution of three case particles, the nominative suffix (-ka after stems ending in a vowel, -i elsewhere), the accusative (-lul after stems ending in a vowel, -ul elsewhere), and the genitive (-uy ) A fourth particle, the suffix -kkeyse, which

replaces the nominative in certain patterns where the nominal refers to someone of high social status and which is often considered to be the 'high form' of the nominative ending, will not be considered here The following sentences exemplify the use of Korean case suffixes with NPs bearing a variety of thematic roles

THE PROBLEM

a NOMINATIVE SUFIX MARKING AN ACTOR

Kay-ka John-ul mwul-ess-ta

dog-N John-Ac bite

'The dog bit John.'

b NOMINATIVE SUFFIX MARKING A PATIENT

John-i mwul-li-ess-ta

John-N bite-Pass

'John was bitten.'

C NOMINATIVE SUFFIX MARKING A POSSESSOR

John-i son-i aphu-ta

John-N hand-N hurt

'John's hand hurts.'

a ACCUSATIVE SUFFIX MARKING A PATIENT

John-i chayk-ul ilk-ess-ta

John-N book-Ac read

'John read the book.'

b ACCUSATIVE SUFFIX MARKING A RECIPIENT

Nay-ka John-ul chayk-ul cwu-ess-ta

I-N John-Ac book-Ac give

'I gave John a book.'

C ACCUSATIVE SUFFIX MARKING AN ACTOR

John-i Sue-lul hakkyo-ey ka-key hay-ss-ta

John-N Sue-Ac school-to go-Comp do 'John let/made Sue go to school.'

d ACCUSATIVE SUFFIX MARKING A POSSESSOR

Kay-ka Mary-lul son-ul mwul-ess-ta

dog-N Mary-Ac hand-Ac bite

'The dog bit Mary's hand

a GENITIVE SUFFIX MARKING A POSSESSOR

John-uy chayk

John-Gen book

'John's book'

4 CATEGORIES AND CASE

b GENITIVE SUFIX MARKING AN ACTOR AND A PATIENT

John-uy caki-uy piphan

John-Gen self-Gen criticism

'John's criticism of himself

In choosing to focus on these particles, I take the position that they are qualitiatively different from two other classes of postnominal morphemes in Korean: (a) the class

of 'postpositions' that includes -eyse 'in, at, on', -eyuyhay(se) 'by', -(u)lo 'to, by means of, -hako 'with', and so on; and (b) the class of 'delimiters' (Yang 1972) that includes- 'only', -to 'also',-pafc&ey 'nothing but',-Mad 'even', and so on

It is relatively easy to distinguish between case markers and delimiters since the latter carry quantificational information and do not make the structural distinctions typically associated with case In fact, as the following sentences show, the same delimiter can appear with either a subject or a direct object (for some general discussion, see I Lee 1977)

(4) a -MAN 'only'USED WITH A SUBJECT

John-man holangi-lul po-ass-ta

John-only tiger-Ac see

'Only John saw a tiger.'

b -MAN 'only' USED WITH A DIRECT OBJECT

John-i holangi-man po-ass-ta

John-N tiger-only see

'John saw only a tiger.'

(5) a -TO 'also'USED WITH A SUBJECT

John-to Sue-lul po-ass-ta

John-too Sue-Ac see

'John too saw Sue.'

b -TO 'also'USED WITH A DIRECT OBJECT

John-i Sue-to po-ass-ta

John-N Sue-too see

'John saw Sue too.'

The major difference between case markers and postpositions is categorial: whereas the former elements mark NPs, the latter head PPs A simple distributional test distinguishes case markers from postpositions in Korean As the following

dog-N-only John-Ac bite

'Only the dog bit John.'

b *Chayk-ul-un John-i ilk-ess-ta

book-Ac-T John-N read

'John read the BOOK.'

 *John-uy-man cha

John-Gen-only car

'John and only John's car'

(7) POSTPOSITIONS PRECEDING OTHER PARTICLES

a Hakkyo-eyse-nun John-i kongpwu ha-n-ta

school -at -T John-N study do

'John is studying at school.'

b Cha-lo-man salam-i ka-n-ta

car-by-only man-N go

'The man travels only by car.'

A second characteristic distinguishing case suffixes from postpositions is their optionality in at least certain patterns in the spoken language, given the approporiate discourse conditions and shared background knowledge (see, for example, Lee and Thompson 1987)

(8) a OPTIONAL NOMINATIVE SUFFIX

Mas(-i) iss-ta

taste(-N) exist

'(It) is tasty.'

b OPTIONAL ACCUSATIVE SUFFIX

Haksayng-i chayk(-ul) ilk-ess-ta

student-N book(-Ac) read

'The student read a book.'

6 CATEGORIES AND CASE

 OPTIONAL GENITIVE SUFFIX

(9) John-i hakkyo-*(eyse) kongpwu ha-n-ta

John-N school (at) study do

'John is studying at school.'

(10) Chayk-i John-*(eyuyhayse) phal-li-iss-ta

book-N John (by) sell-Pass

'The book was sold by John.'

The Status of the Dative Marker

A long-standing problem in Korean grammar has to do with the status of the

so-called 'dative' particle -eykey, which is sometimes classified as a postposition and

sometimes as a case marker (see, for example, Hong 1985) Because the use of

Korean -eykey extends well beyond the expression of goals and recipients, the term

'dative' is technically somewhat inappropriate Nonetheless, for the sake of continuity and ease of exposition, I will use this term throughout this book to label

any occurrence of the particle -eykey

Distributionally, -eykey patterns like a postposition rather than a case marker Thus, we find sentences such as those in (1), in which -eykey may co-occur with a

topic marker and a delimiter, respectively

(1) a -EYKEY WITH A TOPIC MARKER

John-eykey -nun nay-ka chayk-ul cwu-ess-ta

John -Dat -T I-N book-Ac give

'As for John, I gave a book to (him).'

b -EYKEY WITH A DELIMITER

Kay-eykey-man John-i mwul-li-ess-ta

dog-Dat-only John-N bite-Pass

'John was bitten by only the dog.'

THE PROBLEM 7

Moreover, as the following sentence shows, -eykey differs from case suffixes and

resembles postpositions in being obligatory.3

(2) John-i Sue-*(eykey) malhay-ss-ta

John-N Sue(-Dat) speak

'John spoke to Sue.'

Another difference between genuine case suffixes and -eykey is noted by

Kuh (1987), who observes that the dative particle differs from the nominative and accusative markers in being able to occur inside a coordinate structure In this, it resembles postpositions

(3) a NOMINATIVE SUFFIX APPEARING OUTSIDE COORDINATE STRUCTURE

[Joe-kwaMary]-ka yenay-lul ha-n-ta

Joe-andMary -N love-Ac do

'Joe and Mary are in love.'

b NOMINATIVE SUFFIX APPEARING INSIDE COORDINATE STRUCTURE

*[Joe-ka -kwa Mary]-ka yenay-lul ha-n-ta

Joe-N -and Mary -N love-Ac do

(4) a ACCUSATIVE SUFFIX APPEARING OUTSIDE COORDINATE STRUCUTRE

Chelswu-ka [emeni-wa apeci]-/w/ mosi-ess-ta

Chelswu-N mother-and father -Ac support

'Chelswu supported his mother and father.'

b ACCUSATIVE SUFFIX APPEARING INSIDE COORDINATE STRUCUTRE

*Chelswu-ka [emeni-lul -wa apeci]-lul mosi-ess-ta

Chelswu-N mother--and father-Ac support

(5) a GENITIVE SUFFIX APPEARING OUTSIDE COORDINATE STRUCTURE

[John-kwa Mary]-uy chayk

John-and Mary-Gen book

'John and Mary's book'

b GENITIVE SUFFIX APPEARING INSIDE COORDINATE STRUCTURE

*[John-uy -kwa Mary]-uy chayk

John-Gen-and Mary -Gen book

8 CATEGORIES AND CASE

(6) a DATIVE PARTICLE APPEARING OUTSIDE COORDINATE STRUCTURE

John-i [Mary-wa Sue]-eykey senmwul-ul cwu-ess-ta

John-N Mary-and Sue -Dat present-Ac give

'John gave a present to Mary and Sue.'

b DATIVE PARTICLE APEARING INSIDE COORDINATE STRUCTURE

John-i [Mary-eykey -wa Sue]-eykey senmwul-ul cwu-ess-ta

John-N -Dat-and Sue -Dat present-Ac give

Because of these facts, I will consider the dative particle to be a postposition rather than a case marker in Korean Nonetheless, as I will endeavor to show in chapter 4, the dative has a special status among postpositions in Korean and interacts in an important way with case markers

The Status of the Topic Marker

Falling between the system of case markers and the system of delimiters is the suffix

-(n)un, which marks discourse notions such as 'topic' or 'contrast' (the latter notion

being especially salient when the suffix is attached to a non-subject; see Ree 1969 and

Na 1986:34ff for extensive discussion)

(1) TOPIC MARKER ASSOCIATED WITH A SUBJECT

John-un ttena-ss-ta

John-T leave

'John left.'

(2) TOPIC MARKER ASSOCIATED WITH A DIRECT OBJECT

Chayk-un Sue-ka cohaha-n-ta

book-T Sue-N like

'Sue likes BOOKS (not TV).'

(3) TOPIC MARKER ASSOCIATED WITH A SUBJECT AND A DIRECT OBJECT

Bruce-nun Ann-un cohaha-n-ta (Na 1986:14)

Bruce-T Ann-T like

'Bruce likes ANN (not Sue).'

Since -(n)un is attached to an NP associated with a particular discourse function and

does not distinguish among grammatical relations such as subject and direct object, it

is quite unlike bona fide case markers In this it resembles delimiters, which are

THE PROBLEM 9

primarily concerned with the expression of pragmatic and discourse-based notions and do not even discriminate between subjects and direct objects (see (4) and (5) in the earlier section on case suffixes)

These facts notwithstanding, the status of the topic suffix remains

problematic For example, -(n)un resembles case suffixes in not occurring inside

coordinate structures and in not being able to precede delimiters (e.g Ahn 1988) Moreover, as I will note in chapter 4, many (if not all) NPs bearing the topic marker occupy a particular structural position (informally, 'sister' of S) To the extent that the topic marker encodes structural information of this sort, its function is easier to reconcile with that of case suffixes, raising interesting questions about its place in the overall sentence-building system of Korean I will return to this matter in chapter 4

Theoretical Objectives

In proposing the theory of case marking that constitutes the central concern of this book, I will attempt to meet two objectives First, I will try to show that each case suffix encodes a single corresponding linguistic relation, thereby satisfying what

I will call the Uniqueness Criterion Second, I will attempt to show that the relations

encoded by the case markers of Korean are manifested at a single level of syntactic

representation (surface structure), consistent with what I will call the Monostratality

Criterion

The theoretical interest of the Uniqueness Criterion and the Monostratality Criterion stems from the fact that they make very strong claims about the nature of case marking and are hence very difficult to maintain As an illustration of the type of challenges facing the Uniqueness Criterion, we need only consider two of the many patterns of accusative case marking found in Korean

(1) Nay-ka John ul/ chayk-ul cwu-ess-ta

I-N John-Ac book-Ac give

'I gave John the book.'

(2) Mary-ka ku namwu-lul kaci-lul cal-lass-ta

-N the tree-Ac branch-Ac cut

'Mary cut the tree's branches.'

Semantically, the four accusative-marked NPs in these structures have nothing in

common: whereas chayk 'book' and kaci 'branch' denote themes or patients, John names a recipient and namwu 'tree' can have a possessive interpretation Moreover,

while the four NPs do share a structural property, it seems not to be relevant to case

10 CATEGORIES AND CASE

marking Thus, even though the accusative-marked NPs in (1) and (2) all occur

within VP, other NPs appearing in this position do not receive the -ul suffix For example, the NP John in sentence (3), which is virtually synonymous with (1), is

one of the verb's subcategorized complements (and hence part of the VP), yet it appears with the dative postposition rather than the accusative case

(3) Nay-ka John-eykey chayk-ul cwu-ess-ta

I-N John-Dat book-Ac give

'I gave the book to John.'

Comparable problems arise in the analysis of the nominative case As the sentences in (4) and (5) (repeated from above) help show, this suffix is not associated with any single obvious syntactic or semantic property

(4) Kay-ka John-ul mwul-ess-ta

dog-N John-Ac bite

'The dog bit John.'

(5) John-i son-i aphu-ta

John-N hand-N hurt

'John's hand hurts.'

While the nominative in (4) marks an agentive subject, neither of its counterparts in (5) marks an agent and at most one marks a subject (on the assumption that there is only one subject per clause) This raises an obvious challenge for any analysis that seeks to satisify the Uniqueness Criterion

Challenges to the Monostratality Criterion are also easy to find, as the work cited in subsequent chapters will illustrate These difficulties notwithstanding, a guiding principle of my analysis will be that there is a single level of syntactic representation in human language and that the properties encoded by Korean case can

be characterized soley in terms of that level

Concluding Remarks

I will end this chapter with a brief discussion of two methodological issues

that have arisen in the course of my research First, while I have tried to consider a broad and representative range of case marking phenomena in Korean, no work of this kind can be comprehensive and I have been forced to place some limits on the scope of my inquiry The single most important phenomenon which I have chosen to

THE PROBLEM 11

ignore has to do with the precise conditions under which case suffixes may be omitted and the possible semantic effects of their omission (e.g Ahn 1988:11) This decision was motivated in large part by the fact, alluded to above, that a variety of nonsyntactic factors (pertaining to discourse and pragmatics) help regulate this phenomenon Since the vast majority of this book focusses on the role of sentence structure (syntax) in the selection of case suffixes, this omission seemed appropriate

A second and more perplexing methodological problem has to do with the data upon which my analysis is based Virtually all syntactic analysis done within a theoretical framework in the last two decades has relied for its data on sentences constructed by linguists rather than on a corpus of spontaneous utterances I too have followed this practice on the assumption that its advantages (see, e.g., Newmeyer 1983:48ff) outweigh any disadvantages Nonetheless, at least one special problem has arisen

For reasons that I do not understand, there is an unusual amount of disagreement among Korean-speaking linguists over the status of sentences involving

a variety of grammatical phenomena (e.g pronoun interpretation, quantifier scope, honorific agreement, case marking, and so on) This fact is widely recognized by people working in the field and is a frequent source of frustration for many In an attempt to circumvent this problem, I have made a special effort to avoid controversial data To begin, I have tried to draw as many examples as possible from independent sources, especially articles published in refereed journals and dissertations Judgments from sentences that I had to construct were initially checked with two native speakers of Korean (both from the Kyengsondo region) who were graduate students at the University of Hawaii The manuscript was subsequently read by an older speaker of Korean from the Seoul area (an instructor at the University of Hawaii), whose comments helped eliminate judgments that might be considered uncharacteristic of Korean speakers in general Additional comments on the judgments in the first six chapters were provided by a recent Ph.D graduate, also from the Seoul area All of these individuals had training as linguists

Next, the manuscript was read by two anonymous referees, one of whom was a native speaker of Korean and a senior scholar in the field The other, although not a native speaker, made use of a consultant who was Their comments and criticisms were taken into account by including sentences questionable in a survey of six Korean-speaking linguists (four graduate students and two professors) at the University of Hawaii With the exception of a small number of sentences inserted during the final revision process (and identified as such), the judgments used here reflect this exhaustive filtering process While there is no such thing as irreproachable data, I believe that the judgments upon which I have relied are at least as dependable and representative as those used by linguists working on English and other European languages

12 CATEGORIES AND CASE

The remainder of this book presents a theory of case marking for Korean that seeks to satisfy both the Uniqueness Criterion (by associating each case suffix with a unique structural property) and the Monostratality Criterion (by generating only one level of sentence structure) In developing this analysis, I will be proceeding as follows In the next chapter, I will outline a theory of sentence formation for Korean based on the combinatorial mechanisms of a 'categorial grammar' In chapter 3,1 will outline a theory of grammatical relations that follows in a straightforward way from this combinatorial system Chapter 4 returns to the problem of case marking, illustrating how simple case marking conventions can be formulated in terms of the syntactic representations formed by the categorial grammar that I propose for Korean Subsequent chapters will be concerned with the extension of this simple system to include a broad range of phenomena including passivization, possessor ascension, dative advancement, causativization, subject-to-object raising and quantifier float, among others

***2***

The Grammar

The syntactic framework within which I will develop my analysis of Korean case makes use of a relatively simply version of categorial grammar The origin of categorial grammar can be traced to work during the 1930's by the Polish logician Kazimierz Ajdukiewicz However, with the exception of seminal work by Bar-Hillel (1953), the relevance of categorial grammar to the analysis of natural language remained largely unexplored until the 1970's, when it was used by Richard Montague

as part of his attempt to provide a rigorous formal treatment of the syntax and semantics of natural language Subsequent work in 'Montague grammar' and related theories (see references below) has led to impressive descriptive successes and has created an increased interest in the potential of categorial grammar for linguistic analysis (see, for example, Oehrle, Bach and Wheeler 1988)

A typical categorial grammar consists of a lexicon and a set of combinatorial operations (e.g Steedman 1985) Categories that are defined in terms of the type of element with which they combine (an input, or 'argument') and the type of element

that results (an output) are called functors All other categories are taken to be basic For example, the category of 'intransitive verb' (e.g fall ) is usually treated as a functor that applies to an NP to give an S (as in Sue fell ) The NP and S categories,

in contrast, are generally analyzed as basic

Standard categorial grammar typically includes the definitions in (1), among others 1 (The 'slash notation' for functor categories is read from right to left, so that S/NP indicates a function that applies to an NP to give an S.)

(1)

BASIC CATEGORIES:

symbol category name examples

NP noun phrase Kim , ku chayk 'the book', say cha 'new car'

S sentence Harry-ka Mary-lul po-ass-ta "Harry saw Mary'

14 CATEGORIES AND CASE

FUNCTOR CATEGORIES:

symbol category name function examples

IV intransitive verb, S/NP talli-ta 'run'

intransitive verb phrase Mary-lulpo-ta 'see Mary'

TV transitive verb IV/NP po-ta 'see', ilk-ta 'read' Adv VP adverb IV/IV ppalli 'quickly'

Classical categorial grammar contains a single combinatorial operation, called 'functional application', which allows a functor category (TV, IV or Adv) to combine with an argument The structures in (2) and (3) provide simple examples of syntactic representations formed by functional application in accordance with the definitions given in (1)

In (2), the TV po-ass-ta 'saw' combines with the NP Mary to form an IV; this category then combines with the NP Sue to give an S, in accordance with the definitions given in (1) Likewise, in (3), the adverb ppalli 'quickly' combines with the IV talli-n-ta 'run' to give another IV,ppalli talli-n-ta 'runs quickly', which then combines with the NP Sue to give an S

No notational distinction is made between lexical and phrasal categories in (2)

and (3), where both talli-n-ta 'run' and ppalli talli-n-ta 'runs quickly' bear the IV

label Although this notational practice is commonplace in standard categorial grammar, I will differentiate between lexical and phrasal variants of N and IV in accordance with the following policy

(4) A category of type X bears the label XP when:

(a) it is the maximal projection of X, or

(b) it is formed by a combinatorial operation

This practice will yield representations such as the following

THE GRAMMAR 1 5

(5)

In (5), ku chayk-ul ilk-ess-ta 'read the book' is an IVP by virtue of the fact that it is formed by a combinatorial operation on ilk-ess-ta 'read' and ku chayk 'book'; the phrase ppalli ku chayk-ul ilk-ess-ta 'read the book quickly' qualifies for the IVP label

both as the maximal projection of IV and as the product of a combinatorial operation

Similarly, Harry counts as an NP by virtue of its being the maximal projection of a nominal category while ku chayk 'the book' is an instance of an NP formed by a

combinatorial operation on two smaller elements These notational practices notwithstanding, I will use the term 'IV category' to mean either an IV or an IVP, 'N category' to mean either an N or an NP, and so on

1 6 CATEGORIES AND CASE

(1)

As (1) indicates, this sentence has a variety of structural properties: it consists of a subject NP and a VP, the latter phrase contains a verb, a PP and an NP, the verb occurs sentence-finally, and so on While it is a relatively simple matter to describe these properties (traditional phrase structure rules such as S —> NP VP will do nicely), it is much more difficult to relate them to more general principles of human language Why, for example, must there be a subject? Why does the verb occur sentence-finally? And so on

One well known attempt to answer this type of question is developed in the Government and Binding (GB) theory of Chomsky (1981, 1986) Within this theory, the properties of the VP in (1) follow from a general principle (of 'X-bar theory'), which requires that each lexical category or 'head' (including V) 'project upward' to

a phrasal category of the same type and that subcategorized complements be sisters of the head Similarly, the occurrence of V at the end of VP rather than the beginning (as

in English) follows from the fact that a general word order parameter has the value 'head-final' in Korean This also accounts for the fact that Korean has postpositions rather than prepositions and prénommai rather than postnominal modifiers Finally, the fact that this sentence must contain a subject stems not from a stipulation that an S must consist of an NP and a VP, but rather from a general principle that 'predicate categories' (such as VP) must be 'completed' by an NP argument (Rothstein 1985) There is, of course, no reason why the attempt to deduce particular structural properties from general principles should be restricted to GB theory To the extent that this research goal is desirable (as I believe it is), it can and should be pursued within other theoretical frameworks as well In what follows I will therefore seek to develop a set of principles from which the configurational and categorial properties of structures such as (2) and (3), repeated here, can be deduced within my version of categorial grammar

THE GRAMMAR 1 7

A key component of my system is a proposal about the nature of the

categories that make up syntactic representations In earlier work (e.g O'Grady

1987a), I adopted the view that each of the major lexical categories (noun,

verb/adjective, and adverb) is associated with a distinct predicate type Adverbs, for

example, are taken to name predicates applying to the denotation of a verb while

verbs name predicates applying to one or more thematic role-bearing NP arguments

Such categories are said to be dependent in that they must combine with separate

linguistically expressed arguments (verbs in the case of adverbs and NPs in the case

of verbs) Thus, we say that adverbs 'depend' on verbal categories and that verbal

categories 'depend' on NPs Of course, this is just another way of capturing the

intuition that is expressed in classical categorial grammar by treating adverbs and

verbs as functors Hence, as a category of the type IV/IV, an adverb must combine

with a verb in accordance with its dependency requirements Similarly, as categories

of the type IV/NP and S/NP, respectively, transitive verbs and intransitive verbs must

combine with NP arguments

In contrast with adverbs and verbs, nominais (common nouns and NPs)

characteristically do not name predicates that apply to a separate linguistically

expressed argument In a sentence such as Harry-ka ttena-ss-ta 'Harry left', for

example, the nominal does not name an argument-taking predicate Rather, it names

the entity of which the property denoted by the IV is predicated The same is true of

common nouns such as cha 'car' in (4)

(4) Cha-ka pissa-ta

car-N expensive

'The car is expensive.'

This suggests that nominais should be considered instances of an independent (or

'basic') syntactic category that does not name an argument-taking function.2 In the

words of Rothstein (1985:7), NPs may be 'closed internally' without requiring

saturation by an argument (as IVs and other dependent categories do)

18 CATEGORIES AND CASE

(5) Nominal (N) : no dependencies

Intransitive Verb (TV) : a category depending on one nominal

Transitive Verb (TV) : a category depending on two nominais

Adverb (Adv) : a category depending on a verbal category (e.g an IV)

An obvious problem for this proposal is that certain types of nominais appear

to have arguments, contrary to what (5) suggests Thus, the noun destruction appears able to take a theme argument (e.g., destruction of the city or the city's destruction ) just as the verb destroy does Crucially, however, the 'arguments' of a

nominalization occur with a genitive suffix or a preposition - morphemes that are used elsewhere in the language to mark modifiers rather than arguments Moreover, as commonly noted, nominalizations have a 'thing-type' interpretation that would be unusual for an argument-taking predicate A similar phenomenon occurs in Korean,

as we will see in chapter 14 For the time being, I will assume that Ns do not take pure arguments and that their traditional treatment as non-functor categories is correct

If these proposals are correct, then sentences such as Ai-ka chayk-ul ilk-ess-ta 'The child read the book' and Cha-ka pissa-ta 'The car is expensive' will have

semantic representations that contain the information depicted in (6) and (7), respectively (These meaning representations, which use the notational devices of Lexical-Functional Grammar (e.g Bresnan 1982) modified along the lines of Pinker (1984:69), reflect the 'psycho-semantic'tradition within linguistics (e.g Jackendoff

et al.) rather than the logico-semantic tradition (e.g Montague) more often associated

with work in categorial grammar PRED ( 'predicate') is used here to mean

'sememe' or 'semantic content' and hence does not always have an argument structure; where present, arguments are placed inside parentheses Italicized small capitals are used as 'shorthand' for the meaning of the corresponding form.)

(6) Ai-ka chayk-ul ilk-ess-ta

THE GRAMMAR 19

(7) Cha-ka pissa-ta

In (6), ilk 'read' is represented as a predicate that takes two arguments, an actor and

a theme The former argument corresponds to ai 'child' and the latter to chayk 'book' Since neither of these elements names an argument-taking predicate, they

have no argument structure of their own The semantic representation in (7) differs

from (6) in being built around pissa 'expensive' an element that names a predicate

taking a single (theme) argument

The correspondences between predicate types and syntactic categories in (6)

and (7) is as expected Thus, the predicate that takes two thematic arguments (i.e ilk

'read') is encoded as a TV in syntactic structure and the predicate that takes one such

argument (i.e pissa 'expensive') is an IV Similarly, the predicates that are not argument-taking functions (ai 'child', chayk 'book' and cha 'car') are NPs These

correspondences are consistent with the categorial definitions in (5) above, repeated here

(5) Nominal (N) : no dependencies

Intransitive Verb (IV) : a category depending on one nominal

Transitive Verb (TV) : a category depending on two nominais

Adverb (Adv) : a category depending on a verbal category (e.g an IV)

The examples just considered were deliberately chosen for their simplicity It cannot be concluded from these examples that a word's category can be predicted entirely from its meaning (at least given our current understanding of semantic content) If such predictions could be made, active and passive verbs would have to belong to the same category (presumably TV) since both imply an actor and a theme argument In fact, information about the manner in which a category's arguments are realized is also relevant to the determination of its category type Hence, the definition

of verbal categories is sensitive only to arguments that are realized as NPs (i.e., entity-denoting arguments that can combine directly with the verb or its 'projection')

Thus, verbs such as noh-ta 'put', which take a locative PP argument in addition to an actor and a theme, belong to the same TV category as ilk-ta 'read' (consistent with

the long-standing view in many theoretical and descriptive traditions) If the type of argument relevant to category determination were not restricted in this way, there would be as many syntactic categories as there are subcategorization frames (say, two dozen) While there is no a priori reason why this could not be so, syntactic

20 CATEGORIES AND CASE

processes that are sensitive to syntactic categories in human language (e.g agreement, case marking, passivization, and the like) seem to exploit a much more limited set of contrasts (essentially IV vs TV and perhaps TTV, as we shall see) For more on this, see chapters 4 and 15

The major difference between my definitions and those of traditional categorial grammar (see (1) above) is simply that mine refer only to a functor category's input, making no mention of the category that results from combining it with another element Thus, (5) requires an adverb to combine with an IV but does not stipulate that the resulting category must be another IV As I will explain shortly, this information can be predicted by a general principle implicit in most work on categorial grammar

Taking this system of categories as our starting point, we can now derive the properties of sentence structure from a small number of general principles To see how this works, let us assume that the combinatorial operation responsible for sentence formation is simply 'Combine elements' A first principle constraining the formation of syntactic representations can now be stated as follows

(8) THE DEPENDENCY REQUIREMENT

Every combinatorial operation must satisfy a dependency

According to (8), two elements can be combined only if one satisfies the dependency requirements of the other (i.e one element is a functor and the other is its 'argument'.) Thus, an adverb should be able to combine with a verb (the category type on which it depends), but not with a nominal since neither of these elements depends on the other This seems to be correct

(9) ADVERB COMBINING WITH A VERB

A second constraint on sentence building can be stated in the following way

(10) THE COMPLETENESS REQUIREMENT

All dependencies must be satisfied

THE GRAMMAR 21

The intuition underlying (10) is simply that each element in a well-formed sentence must have its dependencies satisfied (by combination with a category of the appropriate type).3 Thus, the dependency of an IV on an NP is satisfied by combining it with such a category, as in (11)

Finally, we need a way to determine the category of the phrases formed by combining smaller elements In particular, how are we to know (short of simple stipulation) that the phrase formed by combining a TV with an NP, for example, is an

IV rather than some other category? I take the position that the properties of subsentential phrases follows from the principle in (12)

(12) THE INHERITANCE PRINCIPLE

A phrase inherits the unsatisfied dependency of a component element

The intuition underlying (12) is that when two categories are combined, one dependency is satisfied (in accordance with the Dependency Requirement) However,

in order to ensure compliance with the Completeness Requirement, all dependencies must eventually be satisfied The Inheritance Principle helps ensure this result by requiring that a dependency left unsatisfied after a combinatorial operation be 'inherited' by the resulting phrase so that it can be satisfied by a subsequent combinatorial operation Thus, if an adverb combines with an intransitive verb, for example, the dependency properties of the former category are satisfied but those of the latter category are not since it still has not combined with an NP According to the convention I have proposed, the phrase formed by combining the adverb and the intransitive verb inherits the unsatisfied NP dependency and is therefore interpreted as

an IV(P) (i.e., by definition, a category depending on one NP)

22 CATEGORIES AND CASE

(13)

Since ppalli ket-ta 'walk quickly' is an IV category, it should be able to combine with

a nominal (in accordance with the definition of IVs) This is correct

(14) Harry-ka [ivp ppalli ket-nun-ta]

Harry-N quickly walk

'Harry is walking quickly.'

Now consider what happens when a transitive verb such as mek-ta 'eat'

combines with a nominal Since TVs depend on two nominal arguments, this initial combinatorial operation will satisfy only one of the verb's dependency requirements The resulting phrase will inherit the unsatisfied NP dependency and should therefore have the categorial properties of an IV

(15)

As expected, phrases of this type can combine with nominais, as (16) illustrates

(16) Ai-ka [Ivp pap-ul mek-nun-ta]

child-N rice-Ac eat

'The child is eating rice.'

The Inheritance Principle also makes the right predictions about phrases

formed by combining 'determiner' elements such as etten 'a/some/which', ku 'the/that', motun 'all/every', and so on with a noun

THE GRAMMAR 23

(17)

Since the dependency properties of the determiner in (17) have been satisfied and since nouns are independent (i.e., do not take three arguments), the resulting phrase will exhibit no dependencies - by definition a nominal category (see (5) above) Consider now the status of the category S, as exemplified in (18)

(18)

Since phrases such as (18) contain no unsatisfied dependencies (the dependency requirement of the IVP having been satisfied by combination with the NP), they should have the status of an independent category - the syntactic class hitherto reserved for nominais This is not an entirely counterintuitive result in that NPs and

Ss do share many distributional properties For example, both can serve as direct objects

(19) a DIRECT OBJECT NP

John-i [NP ku somwun]-ul mit-nun-ta

John-N the rumor-Ac believe

'John believes the rumor.'

b DIRECT OBJECT S

John-i [s Mary-ka ca-n-ta]-ko mit-nun-ta

John-N Mary-N sleep -Comp believe

'John believes that Mary is sleeping.'

Moreover, the morpheme -ki suffices to convert an S into an NP-like category that

can not only appear in NP positions such as subject but also take the appropriate case suffix

24 CATEGORIES AND CASE

(20) [[John-i chayk-ul ilk]-ka ] -ka elyep-ta

John-N book-Ac read-Nmnlzr-N difficult

'John's reading the book is difficult.'

Of course, there are also obvious differences, not the least of which is that NPs typically name individuals while Ss typically encode propositions with truth values It seems plausible to attribute this latter fact to the composition of Ss: their formation involves predicating an attribute of an argument We can therefore distinguish the two independent categories by giving S the characterization in (21)

(21) S is the category formed by combining an IV predicate with an NP

Grammatical Relations and Thematic Roles

The system of sentence formation outlined in the last chapter builds syntactic

representations such as (1) and (2) (I assume that a general word order convention

ensures V-final order in Korean; I consider the problem of Scrambling' in chapter 12

below.)

(1) (2)

These syntactic representations provide the basis for several crucial distinctions

among grammatical relations A first distinction has to do with the contrast between

terms (i.e subjects and objects) on the one hand and non-terms or obliques on the

other.1 The former class of elements can be defined as follows (Notice that I am

using the word 'term' here in the sense of Relational Grammar to refer to subjects

and objects and not in the sense of Montague Grammar in which a 'term phrase' is

roughly equivalent to an NP.)

(3) TERMS

Elements that combine with a verbal category to give a category of a new

type

According to this definition, chayk 'book' counts as a term in (2) since it combines

with a TV to give an IVP Likewise, the NP Harry functions as a term in both (1)

and (2), where it combines with an IVP to give an S In contrast, ppalli 'quickly'

2 6 CATEGORIES AND CASE

cannot be a term since it combines with an IV to give a category of the same type (The notational distinction between IV and IVP does not reflect a change in category

type ) I will refer to adverbial elements and other modifiers as nonterms , defining

them as follows

(4) NONTERMS

Elements that combine with a category to give a category of the same type Consider now the problem of distinguishing between subject and direct object terms I will take the position (adopted in a slighly different form by Dowty 1982) that the required contrast reflects the order in which NPs are incorporated into a sentence structure that is built from the bottom up In particular, I claim that subject and direct object terms have the combinatorial properties outlined in (5) (I consider the status of 'indirect objects' in chapter 4.)

(5) THE DIRECT OBJECT: the first term (of two or more)

THE SUBJECT: the last term

The proposed definitions say nothing about which of a verb's semantic arguments (e.g agent or patient) should be encoded as subject - an issue that I will consider shortly Rather, they simply identify two classes of term NPs with respect to the order in which they are incorporated into sentence structure, with the first term corresponding to what has traditionally been called a direct object and the last term to

the subject In a structure such as (6) ((2) above), Harry is therefore the subject and chayk 'book' the direct object

(6)

The definitions in (5) have a number of consequences, including the prediction of facts which must be stated as separate generalizations in theories such as Relational Grammar, which takes grammatical relations to be undefined primes Two

GRAMMATICAL RELATIONS & THEMATIC ROLES 27

such generalizations are the Final 1 [Subject] Law and the Stratal Uniqueness Law

(Perlmutter and Postal 1983), paraphrased below

(7) THE FINAL 1 LAW

All clauses must have a subject in surface structure

(8) T H E STRATAL UNIQUENESS L A W

There can be no more than one term of a particular type at any level of

syntactic representation

Although I will be assuming that these generalizations are correct, they do not have

the status of independent laws in the theory I am proposing Rather they follow as

theorems from the relational definitions I have put forward Since clauses (Ss) are

formed by combining an IVP with an NP and since, by definition, this NP is the

subject (last term), it follows that all clauses must have a subject This gives the Final

1 Law Moreover, since only one NP at a time is incorporated into syntactic structure,

it also follows that there will be no more than one first term (direct object) and one

last term (subject) in any clause This is the Stratal Uniqueness Law, which applies

here to the sole level of syntactic representation permitted by categorial grammar

-surface structure

THEMATIC R O L E S

It is clear that more than just the traditional subject - direct object contrast is

needed to characterize the distribution of NPs in Korean and other languages As

examples such as the following show, there must also be a way of ensuring that the

appropriate NP bears each term relation

(1) a Haksayng-i chayk-ul ilk-ess-ta

student-N book-Ac read

'The student read the book.'

b *Chayk-i haksayng-ul ilk-ess-ta

book-N student-Ac read

'The book read the student.''I will refer to the problem of associating a verb's arguments (actor, theme, etc.) with the appropriate combinatorial positions (e.g subject vs direct object) as the 'mapping

problem' The first step toward resolving the mapping problem involves making a

2 8 CATEGORIES AND CASE

distinction between the two NP arguments in sentences like (1) I will use traditional thematic role labels for this purpose

Two classes of thematic roles are most often associated with term NPs in

Korean The first such class is made up of what I will call actor roles, which are

assigned to the active entity in a particular action or event With Foley and Van Valin (1984), I intend this class to include not only volitional agents (cf (2a)), but also certain instruments, cognizers and experiencers, as exemplified below

(2) SOME INSTANCES OF THE ACTOR ROLE

a Ku salam-i kay-lul chi-ess-ta [agent]

the man-N dog-Ac hit

'The man hit the dog.'

b Tol-i chang-ul kkay-ss-ta [instrument]

rock-N window-Ac smash

'The rock smashed the window.'

 Ku salam-i taytap-ul a-n-ta [cognizer]

the man-N answer-Ac know

'The man knows the answer.'

d Kay-ka soli-lul tul-ess-ta [experiencer]

dog-N noise-Ac hear

'The dog heard the noise.'

The italicized NPs in these sentences all have in common the fact that their referent plays the more active role in the action or event denoted by the predicate phrase

A second class of thematic roles consists of patient and theme type roles, which are associated with the less active participant in an event In ordinary transitive clauses, this type of role is typically associated with an NP whose referent undergoes movement or change (see, for example, Gruber (1965) and Jackendoff (1972)) The italicized NPs in the following sentences have a role belonging to the theme or patient class

(3) SOME INSTANCES OF THE THEME ROLE

a Ku salam-i kay-lul chi-ess-ta

the man-N dog-Ac hit

'The man hit the dog.'

GRAMMATICAL RELATIONS & THEMATIC ROLES 2 9

b Harry-ka kong-ul tenci-ess-ta

Harry-N ball-Ac throw

'Harry threw the ball.'

c Harry-ka Sue-eykey chayk-ul phal-ass-ta

Harry-N Sue-Dat book-Ac sell

'Harry sold the book to Sue.'

d Ku maswulsa-ka salam-ul kaykwuli-lo mantul-ess-ta

the magician-N man-Ac frog into make

'The magician made the man into a frog.'

The direct object in (3a) is a prototypical patient while those in (3b) - (3d) all exemplify the traditional theme role Notice that while the movement associated with the theme role is concrete in (3b), it is more abstract in (c) and (d), involving changes

in possession and identity rather than physical position

The above examples all involve transitive verbs In the case of intransitive verbs, the sole nominal argument can bear either the theme role or the actor role,

depending on the semantic class of the verb Thus, cwuk-ta 'die' assigns a theme role to its sole argument while talli-ta 'run' takes an actor-type argument

(4) INTRANSITIVE VERB WITH THEME SUBJECT

The former type of verb is often called 'unaccusative' and the latter 'unergative'

In what follows, I will assume that the lexicon includes information about a verb's status with respect to transitivity (transitive vs intransitive) as well as the thematic role class of its arguments The lexical entry of a typical transitive verb such

as ilk-ta 'read' would therefore include the information in (6) Subscripts are used to

indicate the class to which the individual thematic roles associated with each NP

belong; a = actor class, t = theme class

(6) ilk-ta 'read': TV - NPa NPt

30 CATEGORIES AND CASE

Although (6) indicates that ilk-ta takes two NP terms, one with an actor-type

role and the other with a theme-type role, it does not specify which of these is to be encoded as subject and which as direct object, Yet, as we have seen, only one option

is permitted Thus, if ilk-ta has the actor term ai 'child* and the theme term chayk

'book', the corresponding sentence must be (7a) and not (7b),

(7) a Ai-ka chayk-ul ilk-ess-ta

child-N book-Ac read

'The child read the book.'

b *Chayk-i ai-lul ilk-ess-ta

book-N child-Ac read

'The book read the child.'

Obviously, we could simply indicate in the verb's lexical entry that the actor is to be subject and the theme direct object (in active clauses) This is the practice adopted in

Lexical-Functional Grammar, for example, where the lexical entry for ilk-ta 'read*

would include the following information (e.g Bresnan 1982)

(9) ilk-ta 'read' - [actor, theme]

Following O'Grady (1987a), I take the position that the device outlined in (10) has more explanatory potential.2

(10) THE MAPPING GRID FOR TERMS

theme

actor

The Mapping Grid provides a representation of the relative 'combinatorial order' to be employed when a verb takes more than one term Reading from the top down, (10)

GRAMMATICAL RELATIONS & THEMATIC ROLES 3 1

stipulates that when there are two terms, the first (the direct object) is to be associated with a theme-type role and the last (subject) with an actor-type role Sentence (7a)

above is consistent with the Mapping Grid since the first term (chayk 'book') can be interpreted as theme and the last term [ai 'child') can be interpreted as actor The

relevant structure is depicted in (11), with subscripts indicating the thematic role classes of the two NPs

(11)

Now reconsider the unacceptable (7b)

(7) b.*Chayk-i ai-lul ilk-ess-ta

book-N child-Ac read

'The book read the child.'

Here ai 'child' is encoded as first term and hence must be interpreted as the theme in accordance with the Mapping Grid while chayk 'book' functions as last term and

hence must be interpreted as actor This creates an obvious semantic anomaly, which

in turn accounts for the unacceptability of the sentence

Of course, where the verb is intransitive (i.e has only one term), the question

of relative combinatorial order does not arise and either an actor or a theme can function as subject regardless of its thematic role class

(12) INTRANSITIVE VERB WITH ACTOR SUBJECT