University of Pennsylvania ScholarlyCommons

IRCS Technical Reports Series Institute for Research in Cognitive Science

May 1996

Null vs. Overt Subjects in Turkish Discourse: A Centering Analysis

Ümit Deniz Turan University of Pennsylvania

Follow this and additional works at: https://repository.upenn.edu/ircs_reports

Turan, Ümit Deniz, "Null vs. Overt Subjects in Turkish Discourse: A Centering Analysis" (1996). IRCS Technical Reports Series. 95. https://repository.upenn.edu/ircs_reports/95

University of Pennsylvania Institute for Research in Cognitive Science Technical Report No. IRCS-96-13.

This paper is posted at ScholarlyCommons. https://repository.upenn.edu/ircs_reports/95 For more information, please contact [email protected]. Null vs. Overt Subjects in Turkish Discourse: A Centering Analysis

Abstract The purpose of this study is to explore an aspect of discourse coherence which involves anaphoric relations between utterances with special emphasis on subjects in Turkish. Based on an analysis of published narratives, three complementary and interrelated questions are addressed concerning discourse anaphora: 1. Which expressions are available for subsequent definite efr erence? 2. What factors determine the most salient entity in Turkish among a set of potential antecedents for subsequent definite efr erence? 3. What are the functions of a particular referential expression (null vs. overt pronouns vs. full NPs), depending on appropriate discourse conditions?

An exploration regarding question 1. indicates that, while some NPs evoke discourse entities, other NPs do not. These two types of NPs represent referential and nonreferential expressions and they can function as antecedents for definite and indefinite nonspecific anaphora, respectively. The distinction between null and overt pronouns in Turkish is that only the former can be in an anaphoric relationship with a nonreferential antecedent. Overt pronouns, on the other hand, are sensitive to referent identity, they must have the same referent with their antecedents. In other words, overt pronouns are strictly coreferential, while null pronouns are not constrained in this way.

The rest of the study investigates answers to questions 2. and 3. in instances where null and definite subjects alternate as definite anaphors. Centering Theory provides a cognitively plausible and computationally tractable framework for such an analysis with its precise rules which rely on linguistic knowledge constraining inferencing. As formulated in Centering Theory, each utterance contains a set of potential antecedents for reference in the subsequent utterance, i.e. a set of forward-looking centers (Cfs),that are ranked on the basis of their salience. The most salience entity in the Cf-list, the preferred center (Cp) is the entity that is predicted to become the backward-looking center in the subsequent utterance. The singleton backward-looking center (Cb) is taken to be the topic of the current utterance, i.e. the entity at the center of attention. Centering transitions, which model the dynamic attentional state in a discourse segment, are obtained by analyzing each adjacent pair of utterances. The functions of referential expressions in subject position are determined on the basis of Centering transitions. The results show that Turkish subject types pattern neatly and categorically when these transitions are taken into account.

A brief discussion of language-specific and universal aspects of discourse anaphora is also included in the study.

Comments University of Pennsylvania Institute for Research in Cognitive Science Technical Report No. IRCS-96-13.

This thesis or dissertation is available at ScholarlyCommons: https://repository.upenn.edu/ircs_reports/95 Institute for Research in Cognitive Science

Null Vs. Overt Subjects in Turkish Discourse: A Centering Analysis (Ph.D. Dissertation)

Umit Deniz Turan

University of Pennsylvania 3401 Walnut Street, Suite 400C Philadelphia, PA 19104-6228

May 1996

Site of the NSF Science and Technology Center for Research in Cognitive Science

IRCS Report 96--13

NULL VS OVERT SUBJECTS IN TURKISH DISCOURSE

A CENTERING ANALYSIS

Umit Deniz Turan

ADISSERTATION

in

Linguistics

Presented to the Faculties of the UniversityofPennsylvania in Partial Fulllment of the

Requirements for the Degree of Do ctor of Philosophy

Ellen Prince Sup ervisor of Dissertation Graduate Group Chair

c

Copyright

by

Umit Deniz Turan

ACKNOWLEDGEMENTS

I am honored to thank Ellen F Prince mymentor for her constant encouragement invaluable

comments and her time during the preparation of this dissertation She has b een very supp ortive

in all myyears as a graduate studentatPenn and has contributed to my training as linguist

throughout her courses and in our dissertation club meetings with her constructive and chal

lenging criticism I am indebted to Sabine Iatridou who has provided invaluable insightinher

eld of exp ertise and continuous encouragement I would also like to thank Aravind Joshi for his

contribution to Centering Theory and to the basis of this dissertation

Ihave learned ab out linguistics as a science from my professors all of whom are devoted to

academic excellence and b elieve in rigorous education Iwould like to thank AnthonyKroch

Gillian Sanko Donald Ringe William Lab ov Susan Steele Jack Ho eksema George Cardona

and Naoki Fukui all of whom broadened myknowledge ab out linguistics and have b een very

supp ortive throughout myyears as a student Jack Ho eksema once said to his students dont b e

a partisan of any theory I will rememb er that for the rest of mylife

I cannot thank my parents Sevinc and Turgut Turan enough who have provided me b oth

emotional and nancial supp ort Without them I could not have nished my do ctoral degree I

am very lucky to have a sister like Secil who has b een so sweet and has continuously comforted

me during our phone calls and has encouraged me to carry out my studies My brothers Asm

an had to deal with all the hassles of running from one place to another to provide what and Tark

I needed from home in all these years when I was awayTarkan cheered me up with humor that

is so much a part of his p ersonality

Iowe sp ecial thanks to Gul DurmusogluKose a wonderful friend and colleague I am very glad

to know her Joanna Lab ov has provided me a shoulder to cry on she has shared my happiness

and sorrow since the day I met her I would also like to thank fellow students and friends at

Penn who unselshly gave me their time listening to my concerns consoling me and discussing

my research and providing me intellectual stimulation Beryl Homan Beth Ann Ho ckey Sharon

Cote Young Suk Lee Barbara Di Eugenio Breck Baldwin Beatrice Santorini Julie Auger Enric

Vallduv and Alexis Dimitriakis among many others

I greatly appreciate the help of Susannah Page and her familyTeresa Lab ov Joanna Lab ov

Sarah Lab ov Jessie Lab ov and Jamison Brigitte Lab ov myyoungest friend at the age of who

have b ecome my family in Philadelphia They provided a very nice atmosphere and made me feel

at home iii

Iwould liketothankZulal Balpnar Ilknur Kecik and my other colleagues at Anadolu

University for their help Celia Kerslake and Steven Colvin provided intellectual insightand

discussed the Turkish data with me Alan Leson ran to help me when my computer brokeatthe

time I was making the nal corrections

This research is partially funded by a grant from Anadolu University iv

ABSTRACT

NULL vs OVERT SUBJECTS IN TURKISH DISCOURSE

A CENTERING APPROACH

Author Umit Deniz Turan

Sup ervisor Ellen F Prince

The purp ose of this study is to explore an asp ect of discourse coherence whichinvolves

anaphoric relations b etween utterances with sp ecial emphasis on sub jects in Turkish Based on

an analysis of published narratives three complementary and interrelated questions are addressed

concerning discourse anaphora

Which expressions are available for subsequent denite reference

What factors determine the most saliententityinTurkish among a set of p otential an

tecedents for subsequent denite reference

What are the functions of a particular referential expression null vs overt pronouns vs

full NPs dep ending on appropriate discourse conditions

An exploration regarding question indicates that while some NPs evoke discourse entities

other NPs do not These twotyp es of NPs represent referential and nonreferential expressions

and they can function as antecedents for denite and indenite nonsp ecic anaphora resp ectively

The distinction b etween null and overt pronouns in Turkish is that only the former can b e in an

anaphoric relationship with a nonreferential antecedent Overt pronouns on the other hand are

sensitive to referent identity they must have the same referent with their antecedents In other

words overt pronouns are strictly coreferential while null pronouns are not constrained in this

way

ull and The rest of the study investigates answers to questions and in instances where n

denite sub jects alternate as denite anaphors Centering Theory provides a cognitively plausible

and computationally tractable framework for such an analysis with its precise rules which rely on

linguistic knowledge constraining inferencing As formulated in Centering Theoryeach utterance

contains a set of p otential antecedents for reference in the subsequent utterance ie asetof

forwardlo oking centers Cfs that are ranked on the basis of their salience The most salience

entity in the Cflist the preferred center Cp is the entity that is predicted to b ecome the

backwardlo oking center in the subsequent utterance The singleton backwardlo oking center

Cb is taken to b e the topic of the current utterance ie the entity at the center of attention

Centering transitions which mo del the dynamic attentional state in a discourse segment are

obtained by analyzing each adjacent pair of utterances The functions of referential expressions

in sub ject p osition are determined on the basis of Centering transitions The results show that

Turkish sub ject typ es pattern neatly and categorically when these transitions are taken into

account v

A brief discussion of languagesp ecic and universal asp ects of discourse anaphora is also

included in the study vi

Contents

Acknowledgements iii

Abstract v

INTRODUCTION

Purp ose of the Study

Terminology

Utterance Discourse and Discourse Structure

Discourse mo del and discourse entities

Assumed Familiarity of Discourse Entities

Null and Overt Pronouns in Turkish

Simple Clauses

Sub ordinating Clauses

Possessive NPs

The Typ ology and Content of Pronouns

Related ResearchonTurkish Pronouns

ErguvanlTaylan

Enc

Kerslake

Ruhi

Tn and Akman

Null Sub jects and Clausal Reference

Organization of the Study

Evoking Discourse Entities Referentiality and Pronominalization in Turkish

Background of the Problem

Some Related Research on English

Karttunen vii

Webb er

Sidner

Heim

Referentiality

Related ResearchonTurkish

Referential Expressions

Nonreferential Expressions

Referential Contexts and Discourse Entities in Turkish

Predicative Nominals

Situationally Evoked Entities

Accusative Case

Generics

Negation

YesNo Questions

Noun Incorp oration

Bare Ob jects Noun Incorp oration and Referentiality

Arguments for Incorp oration

Arguments against Incorp oration

NI is Optional

Shortterm Discourse Entities

Bound Variables and Pronouns of Laziness

Conclusion

CENTERING THEORY AND OTHER APPROACHES TO DISCOURSE

ANAPHORA

Intro duction

Centering Theory

Backwardlo oking centers and Sentence Topics

Fo cusing as a Precursor to Centering Theory

Other Approaches to Discourse Anaphora

Hobbs

Givon

Ariel

Fox

Kuno

Ha jicova viii

Conclusion

RANKING FORWARDLOOKING CENTERS IN TURKISH

Background Assumptions

The Role of Surface Position

Rambow

Gordon Grosz and Gilliom

The Role of Grammatical Relations

The Center Promotion Rule for Turkish

Psychological Verbs

Empathy

Thematic Hierarchy

Related ResearchinFo cusing

Related ResearchinCentering

Brennan

Cote

Possessive Constructions

Discourse Point of View and Segmentation

Sub jective and Ob jectivePoints of View

Point of View Shifts and Centering Predictions

Related Work In Centering

Problems in the Analysis of NaturallyOccurring Data

Intervening Utterances and Segmentation

Discourse Segmentation and Pronominalization

Conclusions

THE FUNCTIONS OF SUBJECTS IN TURKISH DISCOURSE

The Data

A Preliminary Overview of the Functions of Sub jects

Sub jects in a Continue Transition

The Functions of Null Sub jects in a Continue Transition

Overt Pronouns in a Continue Transition

The Functions of Full NPs in a Continue Transition

Retaining the Center

The Functions of Full NP Sub jects in Retain Transitions

The Functions of Imp ersonal Sub jects in Retain Transitions ix

Shifting the Center

The Functions of Full NP Sub jects in RoughShift Transitions

TwoTyp es of Smo othShifts

Null Sub jects in Shiftsub ject

Full NPs and Overt Pronouns in Shiftob j

Null Sub jects in ShiftOb j

More Evidence for twotyp es of Smo othShift

Conclusion

ACROSSLINGUISTIC STUDY OF CENTERING RULES

Intro duction

The Center Promotion Rule across Languages

The Discourse Point of View as a Universal rule

The Functions of Sub jects across Languages

Prop ertysharing Constraint and Zero Topic Assignment in Japanese

Summary

SUMMARY AND CONCLUSIONS

Bibliography x

Chapter

INTRODUCTION

One asp ect of understanding language is relating new information to the information that has b een

intro duced in the previous discourse Languages make use of various devices to link information

between the utterances to form a coherent discourse One such link is accomplished through the

use of anaphoric reference

Anaphoric reference has constituted a ma jor research topic not only in linguistics but also in

various subdiscipli nes of cognitive science in psychology philosophy and articial intelligence

due to its signicance in understanding memory discourse structure syntactic and semantic inter

pretation and its role in natural language pro cessing Various approaches have b een suggested to

analyze how anaphoric reference works throughout the discourse and across adjacent utterances

within the lo cal discourse cf the references in Chapter and Grosz Sidner Grosz

Joshi and Weinstein among others

In spite of this diverse research on English only a few studies have b een conducted on

anaphoric reference in Turkish Some of these studies investigate pronominalization rules in iso

lated sentences rather than discourse ErguvanlTaylan some others explore what is sub ject

to deletion and pronominalization in the subsequent discourse Kerslake Ruhi others

emphasize the role of context in resolving anaphoric reference across sentences Tn and Akman

The rules and generalizations concerning pronominalization in Turkish discourse are yet

to b e analyzed

Purp ose of the Study

As a relatively unexplored research area in Turkish discourse anaphora deserves an investigation

that will shed light on an understanding of pronominalization facts in Turkish discourse The

purp ose of this study is to reveal the characteristics of anaphoric relations in Turkish at the level

of the lo cal discourse segment ie in adjacent utterances

In order to achieve this ob jective three complementary and interacting phenomena concern

ing discourse anaphora will b e explored One of these phenomena involves determining which

expressions can p otentially serveasantecedents for subsequent reference This concerns distin

guishing b etween referential and nonreferential expressions A set of referential expressions can b e

p otential antecedents for subsequent denite anaphora Among this set of antecedents a partic

ular one is the most privileged ie the most salientantecedent that is predicted to b e mentioned

anaphorically in the subsequent utterance This study will explore the factors that determine how

this salience is assigned The third ob jective of this research is to sp ecify the functions of null

and overt pronouns and full NPs in sub ject p osition and to determine their appropriate discourse

conditions

This study will prop ose a set of wellformedness rules at the level of lo cal discourse It will

b e argued that these rules are universal based on a crosslinguistic analysis of languages studied

in the previous literature These rules governing the use of referential expressions are assumed to

b e part of pragmatic and discourse comp etence

Centering Theory provides the theoretical framework for an analysis of these reference relations

and constraints in the lo cal discourse in a cognitively plausible and computationally tractable

manner Centering Theory and some other approaches of discourse anaphora will b e discussed in

Chapter

The following sections intro duce the underlying theoretical background and the terminology

to b e used in this study In addition the typ es of null arguments in Turkish will b e presented

and the research studies on Turkish pronouns will b e reviewed

Terminology

Utterance Discourse and Discourse Structure

The term utterance is considered to b e an expression uttered or written by a particular sp eaker or

writer at a particular time and at a particular lo cation Utterances thus contrast with sentences

which are p ossible abstract constructs not situated in time and space Utterance form is not yet

obvious but is taken to corresp ond to a tensed clause in this study

A discourse consists of two or more sets of utterances that are coherently linked A discourse

es an initiating conversation participant ICP and an may b e written or sp oken and it involv

other conversation participant OCP Grosz and Sidner An ICP is the sp eaker or the

writer who starts the discourse The OCP on the other hand is the addressee ie the hearer or

the reader ICP and OCP will b e referred to as the sp eaker and hearer in this study including

the writer and the reader resp ectively

Grosz and Linde have shown that there exists a highly constrained discourse struc

ture in taskoriented dialogues In Groszs work a owchart of the task was predictable for the

order of mention of related tasks In addition the choice of p ossible antecedents for pronouns

in the discourse reected the structure of the task as well One imp ortant conclusion from these

data was that pronouns could not b e used to refer to entities in already completed subtasks

These studies and related subsequent research Grosz and Sidner Polanyi have

indicated that discourse has a hierarchical structure that is constructed and pro cessed incremen

tally Polanyi Grosz and Sidner or in an online fashion as psycholinguistic theories

of discourse have shown cf Oakhill Garham and Vonk among others Incremental or

online discourse pro duction and pro cessing means that neither of the participants has the dis

course as a total structured unit prior to its pro duction but rather they formulate and pro cess it

on a clausebyclause basis Grosz and Sidner state that

the structure evolves as the discourse do es By the same token the discourse

participants mental state correlates of the structure are not prebuilt neither par

ticipantmayhave a complete mo del of discourse in mind until the discourse is

completed

Grosz and Sidner prop ose that discourse structure consists of three distinct but interact

ing comp onents Linguistic structure intentional structure and attentional state which capture

various asp ects of discourse structure

Linguistic structure contains a sequence of utterances that form discourse segments Discourse

segments can emb ed subsegments and they can also b e emb edded reecting the intentional struc

ture to b e discussed b elow An utterance has a particular role in a discourse segment Although

two adjacent utterances can b e in the same segment they can also b e in dierent segments In

addition since discourse has a hierarchical structure two nonconsecutive utterances can also b e

in the same segment Factors that determine the b oundaries of discourse segments are still under

investigation and little agreement has b een reached on where the segment b oundaries are see also

Chapter Linguistic cues convey information ab out discourse structure and discourse structure

constrains the interpretation of expressions For that reason linguistic expressions provide clues

for discourse segment b oundaries such as the explicit use of cue phrases rstly to start with

etc intonation changes in tense and asp ect system Linguistic structure also constrains the use

of pronouns whic h are restricted across segment b oundaries

Intentional structure captures the purp oses conveyed byspeakers in a discourse Every dis

course and discourse segment has a purp ose analogous to Gricean utterancelevel intentions Even

though a sp eaker mayhave more than one aim in involving a discourse a story mayentertain its

listeners as well as describ e a narrative Grosz and Sidner distinguish one of these purp oses to

b e foundational to the discourse The underlying intentions in discourse and discourse segments

are called Discourse Purp ose DP and Discourse Segment Purp ose DSP resp ectivelyInten

tional structure plays a signicant role in segmenting the discourse A discourse segmentmaybe

considered to corresp ond to a set of utterances whichhave a single DSP

The third comp onent of discourse structure is the attentional state The attentional state

is the comp onent that keeps track of what the sp eakers are talking ab out by using pronouns

and other referential expressions The attentional state mo dels the fo cus or center of attention

of the sp eaker and hearer as the discourse progresses This is the comp onent which Centering

and Fo cusing theories see Chapter address as they mo del sp eakers attention in keeping track

of anaphoric relationships across utterances Grosz Sidner Grosz Joshi and

Weinstein The attentional state mo dels sp eakers attention however it is

only one part of cognitive state crucially not the equivalent of it The latter includes sp eaker

b eliefs world knowledge commonsense inferencing etc much more than the attentional state

The attentional state is a prop erty of the discourse itself and not that of sp eakers

The attentional state is a dynamic comp onent it receives and keeps track of information and

it up dates the center of attention as new information comes into the discourse The attentional

state is mo delled by a set of fo cus spaces and by a set of transition rules or delete fo cus spaces

A fo cus space contains a set of saliententities that the sp eaker and hearer are attending to ie

discussing or pro cessing information

These three constituents of discourse structure provide the information of how a certain utter

ance ts into the rest of the discourse the reason why it is said and what it means The attentional

state furthermore supplies a device for using the information in the other two constituents All

of the comp onents are at work as the discourse pro ceeds

This study concentrates on mo delling the attentional state at the level of the lo cal discourse

segment However since all these three comp onents are interrelated relevant discussions of the

other comp onents will b e included when they interact with the attentional state For example

Discourse Segment Purp oses of the intentional structure aect pronominalization and linguistic

structure Furthermore discourse segments constrain the use of pronouns as well

Discourse mo del and discourse entities

In the ongoing discourse the sp eaker and hearer are exp ected to share and synthesize understand

er a similar discourse mo del Webb er whichinvolves a mental representation which the sp eak

wants to communicate to the hearer The discourse can b e considered a set of tacit instructions to

the hearer for synthesizing a discourse mo del whichmatches the sp eakers as closely as p ossible

A discourse mo del contains a set of discourse entities a term intro duced byWebb er

corresp onding to the discourse referents of Karttunen evoked into the discourse as well as

their prop erties and their relations with other entities

Webb ers motivation to use the term entityistokeep the term refer as a distinct technical

term Sp eakers refer by using discourse entities Discourses on the other hand evoke or access

discourse entities A discourse evokes an entity in the rst mention then it accesses Webb er

or cosp ecies Sidner or realizes Grosz Joshi and Weinstein its

corresp onding entity in subsequent reference

Discourse entities maybeevoked by the discourse via explicit linguistic mention otherwise

entities can b e inferred within the discourse mo del due to universal or particular knowledge of

entities and relations holding among them Entities may also b e situationally evoked Prince

a see b elow

Once a discourse entityisevoked it may b e realized by denite anaphoric expressions as null

or overt pronouns and full NPs In order to b e able to use a pronoun the sp eaker should b elieve

that an entity is already in the hearers discourse mo del and at the centerfo cus of attention The

choice among the expressions represents what sort of information the sp eaker and hearer havein

their discourse mo dels as they sp eak The typ e of referential expression chosen signals the degree

of fo cus or salience which its antecedent is assumed to have b een assigned by the sp eaker This

choice furthermore indicates the nature of the discourse relationship which the current utterance

holds with the previous one in which a reference to the same entitywas made Such salience

identies the p otential antecedent of a pronoun

As we shall see in Chapter not all NPs evoke discourse entities cf Karttunen Webb er

Kamp Heim In English For example the dummy surface sub ject it in

it seems to me that John is clever do es not evokeanentity NPs whichdonotevoke discourse

entities will not b e accessible for denite pronouns

Some newly intro duced NPs can b e unsp ecied in terms of their evoking discourse entities as

discussed in Chapter For example the scop e of negation can b e ambiguous ie either the

existence of the entity is denied or the actioneventinvolving it may b e negated Only in the

latter case can a discourse evokeanentity A sp eakers use of denite reference in such instances

conrms and consolidates the existence of an entity within hisher discourse mo del Furthermore

the use of an anaphor may actually trigger the intro duction of a discourse entityinto the discourse

mo del in the sequence of utterances I was lo oking for a car and I found one the anaphor one

okes an entity that can b e discussed in the subsequent discourse ev

Assumed Familiarity of Discourse Entities

A sp eaker makes certain assumptions when she conveys a message to the hearer concerning

whether the hearer is already familiar with the entity or not These assumptions determine the

choice of a particular expression Prince a provides a taxonomyofsuch assumed familiarity

of discourse entities When an entityisintro duced into the discourse it is new ie it creates a

new le in Heims terms A new entity can b e Brandnew if it is totally unfamiliar to the hearer

For example as a guy I know Otherwise the sp eaker can assume the hearer is already familiar

with the entity but it is not at hisher discourse mo del at the time of utterance For example

Noam Chomsky is familiar to a linguist ie Hearerold but unused unless explicitly mentioned

and it b ecomes discourse new up on mention

Brandnew entities on the other hand are b oth Hearernew and Discoursenew and they are

of twotyp es Anchored Brandnew entities are linked to a known entityFor example in a car

Isaw the brandnew entity a car is linked to the sp eaker by the use of rstp erson singular

pronoun Unanchored Brandnew entities are not linked in this way for example a car

Textually evoked entities are those that are already present in the discourse mo del and situ

ationally evoked entities represent what is already salient in the extralinguistic context including

the sp eaker and the hearer

Inferrables are the entities that are linked to an already evoked entity via logical relationships

For example the intro duction of an entity as a house makes the do or an available discourse

entity This is b ecause p eople can infer via their world knowledge that houses have do ors

Null and Overt Pronouns in Turkish

Turkish is a pro drop language with null sub jects that are lo cally identied via agreement mor

phemes on verbs Turkish also allows denite and indenite null ob jects although they are not

lo cally identied in the same way with agreement markers or ob ject clitics

Null and overt sub jects in Turkish o ccur b oth in main and sub ordinate clauses and b oth

necessarily agree with the numb er and p erson morphology on the verb Thirdp erson singular

agreement morpheme is null and thirdp erson plural morpheme is optional Gender is not marked

in Turkish

Turkish has the following set of overt pronouns

ben I biz we

sen you singular informal siz you plural formal singular

o hesheit onlar they

These pronouns have a zero nominative case in matrix sub ject p osition They can b e overtly

assigned accusative dative ablative in the ob ject p osition dep ending on a structural and inherent

caseassigning verb

The thirdp erson plural morpheme is optional when the sub ject is overt Compare a to b

Sentence a has a thirdp erson singular sub ject and the verb has a null thirdp erson singular

agreement morpheme The sub ject in b is overt and it has thirdp erson plural features the

plural agreement morpheme on the verb is optional On the other hand the sub ject in c is null

and the thirdp erson plural morpheme must b e overt

a Ali bugun Ankaraya gitti

Ali to dayAnkaraDAT goPAST

Ali wenttoAnkara to day

b Ali ile Meral bugun Ankaraya gitti

Ali with Meral to day AnkaraDATgoPAST

Ali and Meral wenttoAnkara to day

c Ali ile Meral bugun Ankaraya gittiler

goPASTPLU

d Ankaraya gittiler

AnkaraDATgoPASTPLU

They went to Ankara

Null sub jects can o ccur in the following contexts which will b e discussed in the subsections

b elow

Simple Clauses

Sub ordinating Clauses

PossessiveNPs

Simple Clauses

Null sub jects can o ccur in simple clauses as in b

a Ben dun bir kitap aldm

I yesterday one b o ok buyPASTSG

Yesterday I b ought a book

b Dun bir kitap aldm

Sub ordinating Clauses

The following inectional suxes are attached to the verb stem in Turkish to form a sub ordinate

clause where the capitalized letters show that the vowel undergo es Vowel Harmony

a DIk

b EcEk

c mE

d mEk

e Is

The sub ject of a sub ordinate clause is assigned genitivecase which agrees in p erson and number

with the p ossessive sux attached to the sub ordinate verb as in The agreementbetween the

sub ject and the verb of a sub ordinate clause is analogous to the agreement in p ossessive NPs where

the two nouns are assigned genitive and p ossessive agreement see example b elow In addition

a sub ordinate clause likeanNP is assigned structural or inherentcaseby the matrix verb It has

b een suggested that sub ordinate clauses in Turkish are nominals however Kural argues

against these suggestions He claims that sub ordinate clauses haveverbal prop erties

Kural s arguments that sub ordinate clauses formed with a morpheme suchasIShaveverbal prop erties

are as follows

a In a sub ordinate clause a real noun cannot replace the putatively nominalized verb as in the example b elow

A Ahmetin bu arabay als

AhmetGENSG this carACC buyIS POSSSG

the fact that Ahmet b ought this car

B Ahmetin bu arabay alm

AhmetGEN this carACC buyingPOSSSG

Ahmets buying this car

b The sub ordinate verb unlike a noun can have one or more of the following verbal morphology negation

passive causative

oldur u s olum oldur um

dieCAUSIS death dieCAUSNOM

c The sub ordinate verb but not a noun can b e mo died by a frequency adverb

C Ahmetin arada bir kosusu

AhmetGEN once in a while runISPOSSSG

the fact that Ahmet runs once in a while

Ali Ahmetin okula gittigini soyledi

Ali AhmetGEN scho olDAT goNOMPOSSSGACC sayPAST

Ali said that Ahmet wenttoscho ol

The thirdp erson plural morpheme is also optional in sub ordinate clauses The plural mor

pheme may b e unsp ecied as in a Note that a is identical to a but the sub ordinate sub ject

in the former is singular while the one in the latter is plural a and b can alternate the plural

morpheme may b e absent as in a or presentasinb

a Ali Ahmetle Meralin okula gittigini soyledi

Ali AhmetWITH MeralGEN scho olDAT goNOMPOSSSGACC sayPAST

Ali said that Ahmet and Meral wenttoscho ol

b Ali Ahmetle Meralin okula gittiklerini soyledi

Ali AhmetWITH MeralGEN scho olDAT goNOMPOSSPLUACC sayPAST

Ali said that Ahmet and Meral wenttoscho ol

The verb in sub ordinate clauses lacks agreement when the twoadverbial morphemes ken

while and ince when are attached to form sub ordinating clauses as in In these cases the

sub ordinate sub ject do es not carry overt case morphology

The sub ject of the sub ordinate clause must b e null when coindexed with the matrix sub ject

as in a and a The overt sub ject in a sub ordinate clause is necessarily disjoint in reference with

the higher sub ject as in b and b Kornlt ErguvanlTaylan

a Orhan calsrken muzik dinler

i i

Orhan workWHEN music listenAOR

Orhan listens to music when he works

b Orhan o calsrken muzik dinler

i

ik

Orhan he workWHEN music listenAOR

a Ahmet tatile gidecegini soyledi

i i

Ahmet vacationDAT goFUTNOMACC sayPAST

Ahmet said that he was going for a vacation

b Ahmet onun tatile gidecegini soyledi

i

ik

Ahmet heGEN vacationDAT goFUTNOMACC sayPAST

On the other hand either an overt or a null sub ject can b e coindexed with the ob ject of the

matrix clause

Ahmet Aliye onun cok begendigi bir kitab ald

k i

ik

Ahmet AliDAT heGEN very likeRELPOSSSG one b o okACC buyPAST

Ahmet b ought Ali a b o ok that he lik ed a lot

D Ahmetin arada bir kosumu

AhmetGEN once in awhile runningPOSSSG

Ahmets running once in a while

Anull sub ordinate sub ject is ambiguous in coreference b etween the matrix sub ject and the

ob ject if they are b oth available antecedents as in

Ahmet Aliye cok b egendigi bir kitab ald

k i

ik

Ahmet AliDATvery likeNOMACC one b o okACC buyPAST

Ahmet b ought Ali a b o ok whichheliked

PossessiveNPs

Turkish has the following p ossessive pronouns and they agree in numb er and p erson with the

p ossessed noun as NGEN NPOSS The examples are as follows in all these cases the sub ject

of the p ossessiveNPcanbenull

b enim evim bizim evimiz

IGENSG housePOSSSG weGENPLU housePOSSPLU

my house our house

senin evin sizin eviniz

youGENSG housePOSSSG youGENPLU housePOSSPLU

your house your house

onun evi onlarn evleri

sheGENSG housePOSSSG theyGENPLU housePOSSPLU

hisher house their houses

The sub ject of a p ossessiveNPmust b e null when it is coreferential with the matrix sub ject

as in a if the p ossessive is informationally fo cused the logophoric pronoun kendi ownself is

used as in b The o vert pronoun is necessarily disjoint in reference as in c

a Ahmet annesini sever

i i

Ahmet motherPOSSSGACC loveAOR

Ahmet loves his mother

b Ahmet kendi annesini sever

i i

self

c Ahmet onun annesini sever

i

ik

heGENSG

Turkish has null imp ersonal pronouns as in Imp ersonal pronouns can never b e overt

alternativelyanovert pronoun can never have an imp ersonal reading Imp ersonal pronouns must

have a thirdp erson plural interpretation the plural numb er agreement on the verb cannot b e

optional in this case Imp ersonal null pronouns are further discussed in chapter

Ahmeti ameliyat ettiler

AhmetACC op eration makePASTPLU

They have op erated on Ahmet

Turkish do es not haveany expletiveambient pronouns as in the weather verbs eg it is

raining it is hot and existentials in English there is

a Yagmur yagyor

rain rainPROG

It is raining

b O yagyor

It

a Dun cok scakt

Yesterdayvery hotCOPPAST

It was very hot yesterday

b Dun o cok scakt

it

a Buzdolabnda biraz sut var

RefrigeratorLOC some milk exist

There is some milk in the refrigerator

In the following we turn our attention to null ob jects in Turkish The ob ject of a matrix

clause can b e null as in B and b

A Gazeteyi gordun mu

i

newspap erACC seePASTSG Quest

Haveyou seen the newspap er

B gormedim

i

seeNEGPASTSG

I didnt see it

a Mektubu yazdm

letterACC writePASTSG

Ive written the letter

b daktilo eder misin

i

typ e doPRES QUESTSG

Could you typ e it

Kornlt shows that ob jects in a sub ordinating clause cannot b e null intrasententiallyas

in

Ho camzi ogrencilerinin onu cok sevdiklerini iddia etti

i i

teacherPOSSPLU studentGEN heACC very likePLUNOMACC claim makePAST

Our teacher claimed that his students like him very much

However an ob ject of a sub ordinate clause can b e null if the corresp onding entity is within

the previous discourse as in b

a Pastalari tazeymis

cakePLU freshhearsay

The cakes are supp osed to b e fresh

b Pasteneden getiren co cuk soyledi

i

bakeryABL bringREL kid sayPAST

The kid who brought them from the bakery said so

Unlikeverbs some p ostp ositions agree with their ob jects in which case the ob ject can b e

null as in and b

Televizyonu yanna brak

TelevisionACC nextPOSSSGDATleave

Leave the television byyour side

a Baharda agaci cicek act

springLOC tree ower blo omPAST

The tree blo omed in spring

b Altnda oturduk

i

b eneathPOSSSGLOC sitPASTPLU

We sat b eneath it

If the p ostp osition do es not carry agreement morphology the ob ject has to b e overt as in

Ona gore bu is biran once cozumlenmeli

heDAT according this issue as so on as p ossible resolvePASSMUST

According to him this issue has to b e resolved as so on as p ossible

The Typ ology and ContentofPronouns

In the following section the typ ology and contentofnull pronouns as empty categories will b e

discussed briey There are four typ es of expressions assumed in Binding theory Chomsky

a anaphor pronominal

b anaphor pronominal

c anaphor pronominal

d anaphor pronominal

An example of a is anaphors ie reexives and recipro cals as in himself and each other

resp ectively Reexives and recipro cals cannot b e free but they must b e b ound within the binding

domain by a lo cal sub ject to b e dened b elow See Kornlt for an account of anaphors in

Turkish Overt pronouns are instances of b b oth in English and Turkish Denite descriptions

such as Mary constitute the typ e in d The category in c is the socalled control PRO the

sub ject of nontensed clauses in English and Turkish PRO either has an imp ersonal interpretation

as in a or it is controlled by the sub ject as in b or by the ob ject as in c

a PROcalsmak gerekli

workINF necessary

It is necessary PROtowork

b Ali PRO dans etmek istiyor

Ali dance doINF wantPROG

Ali wants PRO to dance

c Aliye PRO gitmesini soyledim

AliDAT goINFSGACC sayPASTSG

I told Ali PRO to go

Empty categories have b een explored on the basis of of the following typ ology shown in

The NPtrace is the typ e of trace left after the movement of for example the ob ject to form a

passive or as in the raising of the sub ject in seemconstructions Variables are traces b ound from

a nonargument A p osition likeWhwords In addition in English there are overt pronouns

functioning as variables b ound byaquantier It will b e shown in Chapter that overt pronouns

in Turkish cannot function as b ound variables

OVERT EMPTY

ana pron each other himself NPtrace

ana pron hethey him etc pro

ana pron PRO

ana pron MaryAli variables

Each of these expressions are in complementary distribution in the grammar of a language

Reexives and recipro cals pronouns and full NPs are sub ject to Principles A B and C of Binding

theory Chomsky presentedin

A An anaphor is b ound in its Governing Category GC

B A pronominal is free in its GC

C A Rexpression is free

Governing Category GC Y is a GC for Y i X is the minimal category containing

Y a governor of Y and a sub ject accessible to Y

Government Agoverns B xBABwhere

a AXo or is coindexed with B

b where Y is a maximal pro jection if Y dominates B then Y dominates A

c A constituentcommands B

The accessible sub ject in Turkish George and Kornlt Kornlt is dened as

follows

a the AGReement element in nite clauses

b the syntactic sub ject ie NPSorNP NP elsewhere

Kornlt shows that null sub jects are pure pronominal pro in Turkish She shows that

null sub jects on the one hand and reexives and recipro cals on the other are in complementary

distribution in accordance with the Principles of Binding theory According to Kornlt if

pro were an anaphor it would o ccur in exactly the same p ositions as anaphors Null sub ject pro

is also in complementary distribution with PRO b ecause the latter can only o ccur in nontensed

clauses UnlikePRO the content of pro is identied with the numb er and p erson features on the

verb morphology within a tensed sentence Kornlt also discusses the parallel b ehavior of

pro in Turkish and English pronouns

The content and typ ology of null ob jects in Turkish havenotyet b een investigated Huang

was the rst to investigate and devise a typ ology of null ob jects in Chinese Huang suggests

that languages must b e parameterized in two resp ects null sub ject languages and zero topic

languages Huang agrees with the previous work done byTaraldsen and Jaeggli in that

null sub jects across languages are pronouns ie instances of pro identied byarich agreement

system on the verb morphology This rich agreement provides the recoverability of the null

sub ject and sanctions its existence On the other hand null ob jects lack rich agreement and are

not lo cally identiable They are not pronouns but rather variables b ound by an unexpressed

discourse topic which has moved to a non argument A p osition

Huang shows that empty sub jects but not ob jects in Chinese b ehavelike pronouns in English

tences from Huang in terms of Principle B of Binding theory Compare the following sen

a John said that he knew Bill

b John said that Bill knew him

He and him in a and b resp ectively are pronouns and they are free in their GC the former

is ambiguous it can b e coreferential with the matrix sub ject John or it can b e coreferential with

another entityevoked in the previous discourse The pronoun him in b is free in its GC and

it cannot b e interpreted as Bill but can b e referentially dep endent on John Huang shows that

the null sub ject is p ossible in Chinese in the corresp onding sentence in a but that this typ e of

intrasentential reference cannot b e construed with an empty ob ject as in b The null ob ject can

have its antecedent only within the discourse which is a zero discourse topic in the A p osition in

the sentence The sub jectob ject asymmetry Huang argues stems from the fact that these two

empty categories have dierenttyp ologies

Huangs prop osal is supp orted by Camp os who shows that null indenite ob jects in

Spanish b ehavelike empty op erators rather than like pro He presents diagnostics demonstrating

that indenite ob ject drop in Spanish ob eys the constraints on Abar movement Since Abar

movementinrelative clauses and Whislands involves an op erator the ob ject cannot drop This

is b ecause the empty op erator that is supp osed to bind an empty ob ject cannot co o ccur with

another op erator In other words the complementizer cannot b e simultaneously lled with a Wh

word and another op erator

Rizzi and Farrell argue against such an analysis concerning null generic ob jects in

Italian and null ob jects in Brazilian Portuguese resp ectively Rizzi shows that an empty category

in ob ject p osition can co o ccur with a Wh word in COMP In Italian

a Quale musica riconcilia con se stessi

with oneself Whichmusic reconciles

b Non so che cosa le sue parole p ossano idurre a PRO concludere

I dont know what his words can lead to conclude

Farrell claims that null ob jects in Brazilian Portuguese are instances of pro and that

the analysis provided byvariable analysis makes incorrect predictions Even though the strong

argument for the variableop erator analysis is the limited distribution of zero ob jects Farrell

claims that this phenomenon go es b eyond the level of sentence grammar Farrell states that

is ungrammatical not b ecause of the reasons given by the variable analysis ie the emb edded

ob ject a putativevariable is Ab ound by the matrix sub ject which is ruled out Farrell admits

that the sentence cannot b e ruled out by a proanalysis

O Jono sab e que o Pedro viu

Jknows that P saw

However Farrell claims that the argumentmadebyvariable analysis supp orters is not a valid

argument since matrix sub jects cannot b e antecedents of null ob jects extrasententially either as

in c

a O Jono falou que a Julia esteve no Rio a smana passada

J said that Julia was in Rio last week

b Mas ele nem viu

But he didnt even see

c Mas ela nem viu

But she didnt even see

Hence a condition b eyond the level of sentence grammar is resp onsible for contrasts of this

typ e Huangs variable argument fails to account for these kinds of constructions By contrast a

null ob ject within a relative clause can take a matrix argumentasitsantecedent

Aquela casa nunca foi pintada p elo cara que comprou de mim

That house was nev er painted by the guy that b oughtfromme

This sentence should also b e mistakenly ruled out by Condition C in the variable analysis

However if the empty categories in the example ab ove are treated as pro s no problem arises

They o ccur in p ositions where overt pronouns are allowed

One prop erty of pronouns as opp osed to reexives recipro cals and variables is that they are

not b ound Due to this the following sentence displays ambiguity

O gato vai comer a rato se vo ce deixar ali

The cat will eat the rat if you leave there

If the empty category in the ab ove sentence were a variable b ound by an op erator it would

not b e ambiguous In op erator constructions such as parasitic gaps adjectival complements

comparatives and tough movement the emptyvariable is strictly b ound so that it cannot

manifest any kind of ambiguity

Unlike empty op erators pronouns may precede their antecedents when the antecedentisin

the matrix clause as in the following

Before he came home John did gro cery shopping

Likeovert pronouns empty pronouns can also precede their antecedents This b ehavior of

null ob jects cannot b e accounted for by an op erator analysis

This study will not concentrate on ob jects However a few initial observations will b e made

here concerning the typologyofnull ob jects Huangs analysis do es not app ear to b e maintained in

Turkish First of all empty ob jects coreferential with sub jects are strongly disfavored in Turkish

across sentences Turan b has found in an analysis of ob jects based on Centering theory in

naturallyo ccurring data among tokens when ob jects were overt of the time when they

were coreferential with a sub ject across sentences whereas in tokens they were null of the

time when they were coreferential with other ob jects It is true that null ob jects coreferential with

sub jects are not categorically ruled out in Turkish discourse and they were null of the time

Nevertheless these p ercentages indicate that there are also constrains at the level of discourse on

null ob jects coindexed with sub jects across sentences

Kornlt shows that relative clause formation in Turkish involves an op erator as in

yinTurkish while it is overtly English The dierence however is that the op erator is empt

expressed as a relative pronoun in English

e okula gidenOpi COMP adam

i i

scho olDAT goREL man

the man who go es to scho ol

An ob ject can b e null in Turkish in a relative clause given as in example ab ove If two

op erators are not p ossible at COMP there cannot b e a zero topic op erator at the COMP simul

taneously with the relative clause op erator As a result a null ob ject cannot b e b ound and the

sentence should b e ungrammatical but it is not

Moreover null ob jects like their overt counterparts displayambiguity unlikevariables

Eger orada brakrsan kedi fareyi yer

if there leaveCONDSG cat ratACC eatPRES

If you leave it there the cat will eat the rat

The null ob ject can b e interpreted as the cat the rat or any other entitycheese etc from

the previous discourse

Like pronouns null sub jects can precede or follow their antecedents in certain contexts as in

the following

a Eger gorursen Can yemege davet et

if seeCONDSG JohnACC dinnerDATinvitation make

If you see him invite John for dinner

b Eger Can gorursen yemege davet et

If JohnACC seeCONDSG dinnerDATinvitation make

If you see John invite him for dinner

The null ob ject in a is ambiguous it can b e interpreted as someone intro duced in the

previous discourse or John The one in b is can b e interpreted as John or any other referent

These facts cannot b e explained byavariable analysis in Turkish

Related ResearchonTurkish Pronouns

ErguvanlTaylan

ErguvanlTaylan has investigated the distribution of overt versus null pronouns in Turkish

within a sentence and across sentence b oundaries She prop oses three ways in which the corefer

ence can b e expressed only by the obligatory use of null pronouns only by the obligatory use of

overt pronouns and by the optional use of either overt or null pronouns

Null pronouns are obligatory when the matrix sub ject and the genitive pronoun in the ob ject

p osition are coindexed An overt genitive pronoun forces disjoint reference whether the antecedent

precedes or follows the pronoun as shown in

a Erol onun kars icin herseyi yapar

i i

ik

Erol heGEN wifePOSSSG for everything doAOR

i i i

Erol do es everything for his wife

b Onun kars icin Erol herseyi yapar

i i

ik

heGEN wifePOSSSG for Erol everything doAOR

i i i

Erol do es everything for his wife

The p osition of the antecedent matters when the genitive pronoun can b e coindexed with

either the sub ject or the ob ject The null genitive sub ject in a is ambiguous it can either b e

coindexed with Ahmet or Erol In b and c however it must b e coindexed with the sub ject

and the ob ject resp ectively dep ending up on the precedence

a Ahmet Erola karsn sordu

i k

ik

Ahmet ErolDAT wifePOSSACC askPAST

Ahmet asked Erol ab out his wife

b Ahmet karsn Erola sordu

i k

ik

Ahmet wifePOSSACC ErolDAT askPAST

Ahmet asked Erol ab out his wife

i k

ik

c Erola karsn Ahmet sordu

k i

ik

ErguvanlTaylan also shows that a sub ordinate sub ject coindexed with a matrix sub ject must

be null regardless of the surface order The overt pronoun in the sub ject p osition of an emb edded

clause is necessarily disjoint in reference with the matrix sub ject as in the following examples

cited in ErguvanlTaylan

a Erol bana onun toplantya gelmiyecegini soyledi

i i

ik

Erol IDAT heGEN meetingDAT comeNEGNOMACC sayPAST

Erol told me that he wouldnt come to the meeting

i i

b Onun toplantya gelmiyecegini Erol bana soyledi

i i

ik

heGEN meetingDAT comeNEGNOMACC Erol IDATsayPAST

Erol told me that he wont come to the meeting

i

ik

On the other hand any sub ordinate nonsub ject pronoun coindexed with an argumentinthe

matrix clause must b e expressed byanovert pronoun as exemplied b elow

aldg kzlar onu severler a Ahmetin ise

i i i

AhmetGEN workDATtakeORELPOSSS girlPLU heACC likeAOR

The girls that Ahmet hired like him

b ise aldg kzlar Ahmeti onu severler

i i i i

workDAT takeORELPOSSS girlPLU AhmetACCheACC likeAOR

According to ErguvanlTaylan if the sub ject is not emphasized or contrasted a null sub ject

in the sub ordinate clause is p ossible b ecause the sub ject can b e identied through the agreement

morpheme on the verb However since there is no ob jectverb agreementinTurkish the ob ject

will not b e lo cally identied and hence has to b e overt as in ab

In conjoined clauses unlike sub ordinate clauses b oth null and overt ob jects are p ossible

a Erol Nazan her aksam isinden alr

i

Erol NazanACC every night workPOSSSGABL takeAOR

ve onu y emege gotur ur

i i

and sheACC dinner bringAOR

Erol picks up Nazan from her work every evening and takes her to dinner

b Erol Nazann evine gitmis

i

Erol NazanGEN housePOSSSGDATgoPAST

ama ona hediyeyi vermeden geri donmus

i i

but sheDAT giftACC giveNEGABL back returnPAST

Erol went to Nazans house but came back without giving her the gift

ErguvanlTaylan emphasizes the role of discourse context in pronoun resolution For example

in and the null and overt pronouns in the B sentences are interpreted as coreferential with

the sub ject and ob ject of the previous sentences resp ectively whichconvey information ab out

these two individuals If this information is not supplied in the previous discourse the null

pronoun in B and the overt pronoun in B are interpreted intrasententially

A Erol yemege gelecek miydi

i

Erol dinnerDAT comeFUT QUESTPAST

Was Erol going to come dinner

B Nazan bana onun yedide gelecegini soylemisti

i i

Nazan IDAT heGEN sevenLOC comeNOMFUTSACC tellPASTPAST

Nazan told me that he would come at seven

A Erolu bu sirkette kim sever

i

ErolACC this companyLOC who loveAOR

Who loves Erol at this company

B Ahmetin ise aldg kzlar onu severler

i i

AhmetGEN workDAT takeOREL girlPLU heACC loveAOR

The girls who Ahmet hired likehim

ErguvanlTaylan indicates that ccommand relations correctly predict that the matrix sub ject

cannot b e coindexed with an NP in the sub ordinate clause as in

O bana Erolun gelecegini soyledi

i i

CC sayPAST he IDAT ErolGEN comeFUTPOSSSGA

Hei told me that Erol is coming

i

However ccommand cannot account for the sentences b elow where the sub ordinate ob ject

do es not ccommand the ob ject in the matrix clause and yet the pronoun is used to refer to its

nonccommanding antecedent

Ben Erolun kitabn kayb ettigimi ona soylemedim

i i

I ErolGEN b o okPOSSSACC loseNOMSGACC heDATsayNEGPASTSG

I didnt tell him that Ive lost Erols b o ok

i i

In this case the surface p osition of the antecedent plays an imp ortant role in the coreference

relationship If this precedence do es not hold the two NPs will not b e coindexed

Ben ona Erolun kitabn kayb ettigimi soylemedim

i i

I heDAT ErolGEN b o okPOSSSACC loseNOMSGACC sayNEGPASTSG

I didnt tell him that Ive lost Erols book

i i

ErguvanlTaylan concludes that it would b e preferable to account for pronominalization facts

at the sentence level as well as the discourse level by a single rule rather than separate rules

Enc

Enc prop oses that null and overt sub jects in Turkish are not in free variation but rather the

c hoice b etween the two is determined by the contextual rules governing the use of sentences For

example a sp eaker can utter a when she and the addressee have not b een sp eaking for a while

and when the sp eaker initiates the conversation On the other hand b would b e appropriate

as an answer to the question Why are putting your coat on

a Ben carsya gidiyorum

I marketDAT goPROGSG

I am going to the market

b carsya gidiyorum

marketDAT goPROGSG

I am going to the market

In the former case Enc suggests that the overt pronoun in a establishes a topic of discourse

while in b the question in the context establishes the topic ie why the p erson is putting on

hisher coat The utterance in b is a comment on the previously established topic

Enc assumes that topic is a primitive but p oints out the diculties of dening such a primitive

Informally the topic is dened as what the discourse is ab out Encemploys the notion of topic

akin to the notion prop osed by Keenan and Schiein cited in Enc as the prop osition

that the question of immediate concern presupp oses A topic is a prop osition that states what

the discourse is concerned ab out and the center of topic Encs terminology is what the sentence

is concerned ab out Prop ositions as someb o dys thinking ab out so and so can b e a topic

Enc suggests that overt pronouns in Turkish are used to shift the topic of discourse For

example in the two consecutive utterances in ab the function of the overt pronoun is to shift

the topic from Alis trip to Ankara to his recentabsentmined b ehavior

a Ali yarn Ankaraya gidiyor

i

Ali tomorrow AnkaraDAT goPROG

Ali is going to Ankara tomorrow

unlerde cok dalgn b O bu g

i

he this dayPLULOC very absentminded

Hes b een very absentminded lately

Enc prop oses that any of the six pronouns can b e used to change topic She shows that overt

sub jects are also used to contrast the referents of two NPs as in the following

Arabay Ahmet ykamad Ben ykadm

CarACC Ahmet washNEGPAST I washPAST

Ahmet didnt wash the car I did

a Bu havada kimse top oynamaz

this weatherLOC nob o dy ball playNEGPRES

Nobody will playballinthisweather

b Ben oynarm

I playAORSG

Ill play

In the example ab ove the assertion in a is denied by giving a counterexample with an overt

pronoun

The two dierent functions of overt pronouns according to Enc changing the topic and

indicating contrast are related b ecause the change of topic involves another way of contrast

When a sp eaker uses a pronoun to change a topic she states that they are not talking ab out a

particular topic any more thus contrasting the old and the new topic

However note that dening the discourse topic as conveying prop ositions is not a welldened

or wellundersto o d notion and an analysis cannot b e based on such a fuzzy notion of topics

There are other cues as well that signal that wehavemoved to a dierent discourse segment

For example in the intention of the sp eaker changes from Alis trip to his absentmindedness

and the change in segmentation is also signalled by the change in tense in b The use of overt

pronouns signals a new discourse segment which is usually but not always correlated with a shift

in topic In fact as will b e discussed later in the analysis of the functions of sub jects there are

instances in the data where a sp eaker signals a discourse segment b oundary with an overt sub ject

even though she continues to talk ab out the same entity

Kerslake

Kerslake presents a brief survey of what she calls NP deletion and its relation to agreement

and pronominalization Kerslake identies four distinct typ es of NP deletion in Turkish as follows

Deletion in co ordinate structures under conditions of structural identity

Equi NP deletion

Prodrop where the pro is identied by agreement marking

Zero Anaphora no such agreement iden ties the content of the empty category

According to Kerslake an identical element in the second conjunct must b e deleted as in the

examples b elow

Ayse sark soyler ve piyano calar

i i

Ayse song singPRES and piano playPRES

Ayse sings and plays the piano

Mehmet arabay tamir etti Mustafa temizledi

i i

Mehmet carACC repair makePAST Mustafa cleanPAST

Mehmet repaired and Mustafa cleaned the car

EquiNP deletion in Kerslakes terminologyinvolves control PRO and some instances of

small pro For example is a case of EquiNP deletion according to Kerslake

Ahmet Fransaya gitmek istiyor

Ahmet FranceDAT goINF wantPROG

Ahmet wants to go to France

Likewise Kerslake prop oses that the following empty category is deleted in accordance with

Equi NP deletion This is b ecause the adverbial morpheme ince when in the sub ordinate verb

do es not carry sub jectverb agreement morphology

o gelince hepimiz sofraya oturduk

she comeWHEN allPLU tableDATsitPASTPLU

When she came we all sat down at the dinner table

Furthermore Kerslake writes

There is one other construction in Turkish to which a rule of Equi NP deletion seems

to apply This is the typ e of relative clause in which the sub ject of the participle is

coreferential with the head NP and undergo es obligatory deletion Since the participle

in this typ e of relative clause cannot have p ossessive marking there is no imp ediment

to including this particular deletion pro cess in the Equi category

The example is as follows

bu kitab yazan adam

i i

this b o okACC writeREL man

the man who wrote this b o ok

Kerslakes motivation is to make a distinction b etween those empty categories that are iden

tied with agreement and those that are not However there app ears to b e a problem in this

typ e of classication All of the empty categories in these examples are of dierenttyp es and

they should b e treated dierently The empty category in is a PROcontrolled by the sub ject

the one in is a pro which apparently can alternate with an overt pronoun unlikePRO On

the other hand as Kornlt has shown the empty category in is a variable b ound bythe

empty relative clause op erator and it can never alternate with a pronoun

ts out that in Turkish null sub jects Kerslakes category ab oveinvolves pro drop Kerslake p oin

and overt pronouns corresp ond to unstressed pronouns versus stressed pronouns resp ectivelyin

English She argues that sub jects should b e overt when they receive stress fo cus For example

in the sub jects cannot b e null b ecause they need to receivecontrastive stress

a Onlar evde kaldlar b en sinemaya gittim

they homeDAT stayPASTPLU I cinemaDATgoPASTSg

They stayed at home and I went to the movies

b evde kaldlar sinemaya gittim

She states her observation as follows

The distribution of sub ject prodrop in Turkish corresp onds approximately to that of

unstressed sub ject pronouns in English

The distribution of overt sub ject pronouns in Turkish corresp onds approximately to

that of stressed pronouns in English

Kerslake

Zero anaphora according to Kerslake involves null ob jects b ecause the content of ob jects is

not identied by agreement morphology on the verb Kerslakeshows that in Turkish as indicated

in the following example deletion of the ob ject is the norm and an overt ob ject in b and c

will render the discourse incoherent

aSu kitab bitirince bana verir misin

i i

that b o ok nishWHEN IDAT give QUESTSG

When you nish that b o ok would you give it to me

b Zeynep cok methetmisti

i

Zeynep very praisePASTPAST

Zeynep had praised it highly

c Ben de okumak istiyorum

i

I to o readINF wantPROGSg

Iwant to read it to o

In the rest of the pap er Kerslake explores pro drop and zero anaphora in two pub

lished short stories in Turkish within the framework of Givons accessibility hierarchy for a

discussion of Givon see Chapter Kerslake found that slightly more than half of the sentences

in her stories havenoovert sub jects The ratio of deleted ob jects are very low analysis indicates

that null ob jects o ccur less frequently than null sub jects in these two short stories Kerslakepro

p oses that a sub ject is null if its corresp onding entity is immediately accessible which is dened

as b eing the sub ject of a previous sentence or as having just b een intro duced into the discourse

Ruhi

Ruhi inv estigates the distribution of null and full NP sub jects in Turkish texts Her study

also examines whether paragraph b oundaries blo ck the use of null anaphora As data she uses

narratives written by rstyear university students who are nativespeakers of Turkish She

obtained her data by asking the students to rewrite from memory a story byaTurkish writer in

order to obtain texts of similar typ e and comparable length

Ruhi states that the null sub ject is mainly reserved for the protagonist She observes that

the shift from null to overt pronoun or full NP takes place when nonprotagonist characters

are actively involved She notes that full NPs representing the protagonist o ccur at episo de

b oundaries However she states that this is a tendency rather than a general rule Ruhi observes

that null pronouns are preferred if the sub jects are coreferential in the preceding clause and the

episo de b oundary clause She concludes that the role of a paragraph as a barrier for null pronouns

is not clear

Ruhi investigates anaphora at a global discourse level Her study contributes to an understand

ing of discourse anaphora Nevertheless b ecause she do es not employ a theory in whichtoembed

her ndings her results are very general and impressionistic An analysis of discourse anaphora

based on more precise concepts is preferable if we are to understand discourse comp etence and

how sp eakers and hearers pro duce and pro cess discourse anaphora

Tn and Akman

Tn and Akman prop ose a computational framework for anaphora resolution in Turkish

based on the version of situation semantics prop osed byFenstad et al In this framework

the informational content of an utterance is determined not only by its linguistic form ie its

phonology morphology and syntax but also byanumberofcontextual factors

A simple sentence has a situation schema and a fact schema A situation schema is a the

oretical notion for summing up information from linguistic form that is relevant for semantic

interpretation It is derived through an extension to the traditional format of Lexical Functional

Grammar LFG of Kaplan and Bresnan A situation schema also b ears some similarity

to a Logical Form A simple declarative sentence can b e regarded to have a semantic predicate

relating a numb er of actors playing various roles Agent Theme etc A situation schema is

anchored to a relation with arguments and lo cation and also a p olarity of the fact

A fact schema is a means for conveying partial or complete information either from the

situation schema or world knowledge For a sentence like Erol dun evlendi Erol got married

yesterday a fact schema will convey information like he got married to someone and that

it o ccurred in the past Since wehave sp ecic knowledge ab out the world further information

will come from the sub ject individual that he is male then he is a gro om that he is married

toawoman and she is the bride etc Further from world knowledge it will b e inferred that he

has a wife and that he is her husband All these pieces of information are kept in a pool for

further use during anaphoric pro cessing Then an inference mechanism can use the discourse

biguous pronouns in the text Thus if a sentence like Ahmet Erola context to disambiguate am

karsn sordu Ahmet asked Erol ab out his wife is preceded by Erol got married yesterday

information from the fact schemata kept in the p o ol will resolve the ambiguous genitive pronoun

his

This algorithm is very ambitious since Tn and Akman intend to build a unied theory of

anaphora based on the information from researchonsyntax semantics discourse context and

world knowledge

Null Sub jects and Clausal Reference

Iatridou and Embickshow that pro in Greek and some other pro drop languages eg Bulgar

ian Italian Catalan cannot b e anaphorically linked to a clause while the thirdp erson pronoun

it in English can This distinction b etween Greek and English is observed b oth intersententially

as in and and intrasententiallyasinand

A o Kostas ine panda argop orimenos

the Kostas is always delayed

b Pragmatika Ke pro epise ton patera tu na to agorasi aftokinito

Indeed And pro convinced the father MOD buy him a car

In tended reading It convinced his father to get him a car

A John is always late

BIknowAnd it convinced his father to get him a car

An ftasume arga pro tha tromaksi tin Maria

i i

if we arrive late pro FUT scare the MaryACC

If we arrive late it will make Mary think that we dont likeher

i i

An expletive sub ject pro can app ear in related environments but in this case it is not coref

erential with the clause as in and

An o Kostas argisi pro tha ine dropi

If the Kostas is late pro FUT b e shameful

pro fanike agenes pu ftasame arga

pro seemed imp olite that we arrived late

Iatridou and Embick show that pro in Italian Catalan and Bulgarian b ehave exactly the way

it do es in Greek Their observation app ears to hold for Turkish as well For example the expletive

sub ject in cannot b e interpreted as anaphorically linked to the previous clause In fact ayp

shame seems to function as a sub ject and renders another sub ject null or overt irrelevant In

addition a thematic pro cannot b e coreferential with a clausal antecedent as in

ger geckalrsan ayp olur E

if late remain shame happ enPRES

If you come late it will b e rude

Eger geckalrsan pro baban araba almaya ikna edecek

i i

if late stayCondSG pro fatherPossAcc car buyInfDat convince do

If you are late itll convince your father to buy a car

Iatridou and Embick discuss two p ossible approaches to account for this phenomenon the

rst is what they call the ontological approach and the second the grammatical approach

According to the ontological approach pro cannot refer to whatever the antecedent refers to

In this instance one can claim that pro cannot refer to events However Iatridou and Embick

show that this approach is not sucient for examples like through In an event in which

one often arrives late the pronoun it cannot b e used to refer to a single event it must refer to

a collection of suchevents Secondly this approach is inadequate b ecause it can b e anaphorically

linked to clauses under negation in English

If we dont return by pm it will convince the b oss that

i i

Discarding the rst approach b ecause of these observations and discussions of ambiguitybe

tween events and facts Iatridou and Embick suggest that the grammatical approachisonthe

right track They prop ose the following

pro cannot have a CPIP as linguistic antecedent

pro can have only a nominal antecedent

Note that this extends to Turkish In pro can only b e interpreted as coreferential with the

sub ject of the sentence but not with the IP

a Mehmet problemi cozdu

i k

Mehmet problemACC solvePAST

Mehmet solved the problem

b pro ogretmenini sasrtt

ik

teacherPOSSSG surpriseCAUSPAST

pro surprised the teacher

However the clause itself can b e a sub ject when it is nominalized

Mehmetin problemi cozmesi ogretmenini sasrtt

MehmetGEN problemACC solveINFPOSSSG

The fact that Mehmet solved the problem surprised his teacher

According to Iatridou and Embick the reason why pro cannot have a CPIP antecedentis

that pro is licensed by Phifeatures numb er p erson gender or consists of Phifeatures itself

However an IPCP lacks such features as sho wn in

a I saw John leave

IP i

Isaw it to o

b I saw John leave and Mary arrive

IP i IP k

I saw them to o

i
k

If IP had Phifeatures the antecedents in b would b e plural in numb er and the plural pro

noun could b e interpreted as coindexed with the two IPs in Consequently the incompatibility

of pro in an anaphoric relationship with an IPCP follows

Turkish disallows b oth overt and null sub jectsob jects with Phifeatures to b e anaphorically

linked to IPCP In neither the overt nor the null sub ject can b e coindexed with the clause

Eger geckalrsan opro anneni endiselendiri r

i i i

if late remainSG itshehepro motherPOSSSGACC worryPRES

If you stay out late it will worry your mother

A deictic pronoun in the same p osition is grammatical

Eger geckalrsan bu anneni endiselendi rir

i i

if late remainSG this motherPOSSSGACC worryPRES

If you sta y out late that will worry your mother

The plural numb er feature for the thirdp erson is optional in Turkish as has b een previously

mentioned and it may b e homophonous with the thirdp erson singular Thus sub jectverb

agreement do es not constitute a go o d test to check whether IPCP has Phifeatures that are in

agreementwithverbal morphology Instead the features of the deictic pronoun can b e used

A deictic pronoun as wehave seen in can b e in an anaphoric relationship with a CPIt

has b oth a singular bu this and plural bunlar these form dep ending on the numb er feature

of its antecedent If IPCP lacks numb er features the deictic pronoun do es not agree with its

antecedentinnumb er Consider

a Enasyon artt

i

ination increasePAST

Ination has increased

b Issiz says fazla

k

jobless number much

The numb er of unemployed p eople is high

c Bu bunlar hukumeti onlemler almaya zorluyor

i
k i
k

thisthese government precautionPLU takeINFDAT forcePROG

ThisThese forces the governmenttotake precautions

The deictic pronoun in c can either b e singular or plural when it is in an anaphoric rela

wo IPs in a and b If IP lacks numb er a plural pronoun should not b e tionship with the t

acceptable

Iatridou and Embickinvestigate cases where the verb gets plural morphology in agreement

with a sub ject that consists of two conjoined CPs They cite McCloskey who suggests that

this agreement is p ossible when the conjoined prop ositions are contradictory or incompatible

otherwise the verb has to have singular morphology One example is shown in

a That the President will b e reelected and that he will b e imp eached are equally likely

at this p oint

b That the shares are overvalued and that the decline is in order isare widely b elieved

in Wall Street

According to Iatridou and Embick the verb agrees with the IPCP sub ject when the IPCP

is prop erly individuated as in

Principle of Individuation Clauses trigger plurality only if their contents are prop erly

individuated

Lexical items like b oth and equally can b e used to favor individuation of states or events

In a plural deictic is not p ossible b ecause the principle of individuation do es not applyand

a deictic pronoun unsp ecied for numb er is used since the one that is marked for plural is not

p ossible

a Alinin calstgn Ahmetin televizyon seyrettigini gordum

AliGEN workNomPossAcc AhmetGEN television watchNomPossAcc sawSG

I saw that Ali was working and Ahmet was watching TV

b Bubunlar b eni sasrttmad

thisthese IACC surpriseCAUSNEGPAST

This didnt surprise me

To sum up wehave seen that pro in Turkish as in Greek Italian Catalan and Bulgarian

cannot haveanIPCPantecedent A thirdp erson overt sub ject is also prohibited from b eing

anaphorically related to the IPCP antecedent a deictic pronoun must b e used instead Wehave

seen that Iatridou and Embicks analysis app ears to account for Turkish facts as well IPCP do es

not seem to have PhiFeatures and the deictic pronoun do es not agree unless the clauses are prop

erly individuated Future research on the prohibition of overt pronoun and pro in an anaphoric

relationship with a clausal antecedentinTurkish will hop efully extend these observations

Organization of the Study

The rest of the study is organized as follows Chapter examines which expressions are available

for subsequent denite reference Some related research on English and Turkish is reviewed

and discussed Chapter presents the theoretical framework for the analysis in this study ie

Centering Theory and some other approaches to discourse anaphora Chapter investigates the

factors that determine salience assigned to discourse entities and what is predicted to b e discussed

anaphorically in the subsequent discourse Chapter explores the discourse functions of null vs

overt sub jects and full NPs in Turkish Chapter is devoted to a crosslinguistic examination

of the languages studied in the Centering literature and their ndings are compared with the

ndings in Chapter Finally Chapter summarizes the study and discusses the conclusions

Chapter

Evoking Discourse Entities

Referentiali ty and

Pronominalization in Turkish

One imp ortant task in the study of anaphoric reference is to sp ecify what a discourse p otentially

makes available for subsequent reference This chapter investigates the way in which nominal

expressions evoke discourse entities for denite reference in Turkish

There are cases where overt pronouns and null arguments o ccur in complementary distribution

with resp ect to the referentiality or nonreferentiality status of their antecedents Lo cating the

contexts where overt pronouns are prohibited as opp osed to those where they are allowed is of

critical imp ortance This is b ecause we need to categorize the data according to the availability

of an option b etween overt and null pronouns Thus there are two levels of distinctions made

between overt pronouns and null arguments One is that there are cases where no alternation

between the twoisavailable On the other hand overt vs null pronouns can b e selected dep ending

on the salience assigned to the antecedents and Centering rules In the analysis of these two

complementary tasks of anaphoric reference Centering Theory is devised to determine the salience

and the choice of an appropriate referential expression when sp eakers ha vesuchachoice Centering

do es not prop ose an analysis for evoking discourse entities which enable them to b e referred to

anaphorically by denite pronouns and full NPs but rather is concerned with tracking anaphors

once they are evoked in the discourse mo del

The concern in this chapter is on the form and content of the NPs that rst evokeentities in

a discourse mo del The establishment of a discourse entityisvalidated by a sp eakers reference

to that entity with a denite anaphor which conrms and consolidates its existence This is the

approachtaken in Karttunen and Webb er as well

The chapter is organized as follows After the background of the problem is discussed in Section

and the related research on English is presented in Section the notion of referentialityisgiven

in Section and the relevant researchonTurkish is reviewed Section presents the linguistic

devices that mark referential vs nonreferential expressions in Turkish In Section short term

discourse entities are discussed Section is concerned with the dierenttyp es of pronouns and

nally in Section the ndings in the chapter are summarized

Background of the Problem

NPs that do or do not evoke discourse entities have not b een studied in detail in Turkish Dede

and Tura explore referentialityinTurkish but they do not discuss referential expressions

in relation to future denite reference The studies done by Dede and Tura will b e reviewed in

section Enc has studied sp ecicityinTurkish and her work will b e discussed in section

Erguvanl who works on the function of word order in Turkish has made a brief re

mark that the use of an overt pronoun is prohibited when the antecedent is a nonreferential

expression Ob jects that are not overtly assigned accusative case or mo died by bir one

have b een claimed to represent nonreferential expressions Erguvanl and the other refer

ences cited in this chapter Erguvanl claims that numb er and deniteness are irrelevant for

nonreferential expressions Therefore an NP that do es not represent a referential expression is

not assigned accusative case in the ob ject p osition and is not mo died by the indenite numeral

bir one the devices to mark deniteness and indeniteness resp ectively cf Tura Dede

Accusative case marking is overt when the ob ject receives structural case otherwise the

ob ject receives an abstract null case morpheme under the government of the verb or the caseless

ob ject is assumed to have undergone incorp oration with the verb which will b e discussed later



Accusative casemarking has b een claimed to b e a deniteness or sp ecicit ymarker for ob jects

in Turkish Based on these observations Erguvanl denes bir resim a pictureas referen tial in

a and resim picture as nonreferential in c Erguvanl claims that an overtobjectindisnot

allowed b ecause its antecedent resim picture is nonreferential

Deniteness is also determined by the p osition of the NP in a sentence Denite NPs app ear either in sentence

initial or p ostverbal p ositions while indenites o ccur in the immediately preverbal p osition

Nominative case is phonologica ll y null in Turkish All other cases ie dative ablative genitive p ossessive are

marked byovert morphemes

a Ali kac gundur bir resim yapyordu

i

Ali howmanydays one picture doPROGPAST

Ali was painting a picture for manydays

b Bugun onu bitirdi

i i

to day itACC nishPASTSG

He nished it to day

c Ali kac gundur resim yapyordu

i

Ali how manydays picture doPROGPAST

Ali was picturepainting for manydays

d Bugun onu bitirdi

i i

to day itACC nishPASTSG

He nished to day

In a the expression denoted by the NP bir resim a picture is a referential indenite since it is

preceded by the indenite numeral bir one Therefore an accusative casemarked overt pronoun

onu it can alternate with a zero pronoun to access this entity On the other hand resim picture

in c is a nonreferential expression Hence the infelicityofanovert pronoun in d follows It is

plausible to assume that if a nonreferential expression is intro duced not to highlight the ob ject

that it denotes but rather to mo dify the verb it cannot stand as an indep enden tentityand

therefore cannot b e accessed byanovert pronoun Erguvanl app ears to b e making the correct

prediction at rst glance Consider he naturally o ccurring discourse in

a Annem para verdi

motherPOSSSG money givePASTSG

My mother wanted to givemoney

b adam almad

guy takeNEGPASTSG

the guy didnt take it

b O cok az bir parayd

it very little one moneyCOPPAST

It was a small amountofmoney

Ab dullatif Minkari Bir Cerrahn Anlar

In a para money is a bare NPFollowing Erguvanls claim it represents a nonreferential

expression and the use of an overt pronoun to access it will b e infelicitous as in b and b

However Erguvanls analysis do es not extend to all bare ob jects and her data app ear to

misrepresent the facts First of all nativespeakers have stated that in a unless they wanted

to stress the numb er they would use an NP without the indenite numeral bir one to refer to

a discrete entity This may b e due to Turkish employing two strategies to mark indeniteness

contrary to Erguvanl Dede and Tura ie a null morpheme and an indenite

numeral bir one corresp onding to the English unstressed aan and stressed one resp ectively

Conclusive evidence for this however has to await future research

The second problem is that the nativespeakers have also stated that an overt pronoun to

access the bare noun resim picture in d is felicitous contrary to Erguvanls judgement This

observation is veried in the naturallyo ccurring data where sp eakers refer backtoentities rep

resented by bare NPs using overt pronouns Consider the example b elow

a Balk tutacaktk mutlaka

sh catchFUTPASTPLU absolutely

We were certainly going to catch sh

b Tuttuk da

catchPASTPLU to o

And we did to o

c Ben istavrit yakaladm

i

I mackerel catchPASTSG

I caught some mackerel

d Avladgm ilk balkt onlar

i

HuntRelSG rst shPAST they

They were the rst sh I ever caught

Gulderen Bilgili Bir GeceYolculugu

Note that istavrit mackerel in c is not mo died byany indenite mo dier as biraz some



or bir one yet it is accessed b yanovert sub ject pronoun onlar they in d

Nevertheless Erguvanl has identied and p ointed out an imp ortant observation in Turkish

discourse Since her study concentrated on word order her remarks on prohibition of overt

pronouns in nonreferential contexts were limited to brief remarks and they need to b e elab orated

on with a study on pronouns This need constitutes the motivation b ehind this chapter With

this in mind wenow turn our attention to some related research on English

Some Related Research on English

Karttunen in his pioneering research study distinguishes indenite NPs that can serveas

antecedents for denite pronouns from those that cannot lab eling the former discourse referents

Later work was done byWebb er and Heim In the following their research

will b e reviewed

Karttunen

Karttunen sp ecies the contexts that provide future reference for a novel individual intro duced

into the discourse For example the utterance Bill is not a linguist Karttunen is not



Note that balk sh is a countnouninTurkish and it can b e mo died by bir one

ab out two individuals it is ab out Bill It is out of question that a future reference will contain

the linguist which Bill is not However a car in the following example from Karttunen

may b e followed by information ab out that sp ecic car

a Bill has a car

b It is black

NPs that intro duce individual s into the discourse for subsequent reference such as a car in a

are coined as discourse referents by Karttunen Karttunen distinguishes discourse referents from

NPs under the scop e of negation and those that app ear as complements of mo dal verbs or some

main verbs like exp ect try plan intend The latter do not establish discourse referents On the

other hand a class of verbs which Karttunen calls implicatives as manage venture remember

see t imply the existence of their complement unless they are negated Thus an indenite NP

in an armative sentence establishes a discourse referent if the prop osition represented bythe

sentence is asserted implied or presupp osed by the sp eaker to b e true The complements of a

group of intensional verbs like need promise want do not intro duce discourse referents even if

the sentence is not negated as in

a I needed a car

b It was a Mustang

The lifespan of discourse referents dep end on whether they are asserted to exist in non

mo dal or mo dal contexts While longterm referents can last throughout the discourse short

term discourse referents in a group of mo dal contexts can only serveasantecedents for denite

anaphors provided that they sustain the same mo dal contexts that they o ccur in These will b e

discussed later in some detail

Webb er

Webb er shows that the form and content of NPs that evoke discourse entities contribute to the

interpretation of denite pronouns and full NPs

As has b een previously stated Webb ers term discourse entities corresp onds to Karttunens

discourse referents Discourse entities may b e individuals sets events or actions and they are

the main constructs of Webb ers notion of discourse mo del which corresp onds to Lyons universe

of discourse Heims le cards and Kamps discourse representations Discourse entities are

conceptual coatho oks on which descriptions are hanged A discourse mo del containsasetof

entities evoked into a discourse and makes available the interpretation of denite reference or

memb erset relations as in oneanaphora in referential contexts as in John has b oughtve

candies and he gave me one

Webb er like Sidner and Grosz Joshi and Weinstein distinguishes the notion of

reference from evoke or access Sp eakers refer to ob jects but discourses evokeoraccessentities A

sp eaker refers to an ob ject p erson etc which corresp onds to a discourse entity The rst mention

evokes an entity in the discourse mo del and the subsequentmentions access the corresp onding

discourse entity

Webb er constructs complex semantic representations including quantication over a set of

individuals In Webb ers theory initial descriptions IDs the rst mention of an entityare

sp ecial since they involve information ab out an entity that can b e assumed to b e shared bythe

sp eaker and the hearer alike Thus the sp eaker assumes it is an inference that the hearer can

make at least initially A crucial asp ect of Webb ers theory is that IDs can b e left unsp ecied

This may b e due to the ambiguity of the scop e of negation as in Bill either didnt marry a

blonde but married someone else or there is no such blonde at all In the same vein a has two

interpretations there is a woman Bill is involved with and she is in jail or there is not sucha

woman Both of these interpretations are available initially and they are disambiguated in the

subsequent discourse

a Bill didnt marry a blonde

b She had red hair

a Bill didnt marry

b She is blonde

b She is in prison

Webb ers rules which apply successively are presented b elow for the sake of completeness

In her theoryevery clause is rst syntactically parsed This creates an input to he semantic

interpretationLevel representation Level representation supplies semantic information like

p ossible scop e assignment of quantiers selectional restrictions the argument structure At

el this level ellipsed verbs and scop e assignment remain unresolved IDs are formed at Lev

representation

Webb er suggests that in order to form appropriate IDs the following requirements must b e

fullled Denite and indenite NPs and singular and plural NPs should b e distinguished Fur

thermore for each mo dier in a plural NP it should b e determined whether it conveys information

for the whole set denoted by the NP or whether ab out the individual setmemb ers For example

in an utterance like Bruce gaveWendy ve dollars the NP ve dollars can either evoke a set or

individual entities and the entity is accessible for the individual reading as in one of them was

counterfeit or for the set reading as in it was more than he gave Sue Any VP ellipses should b e

resolved prior to ID formation Quantier scop e assignmentmust b e determined

Sidner

Sidner concentrates on tracking anaphora in discourse rather than evoking entities and her theory

will b e discussed in detail in Chapter Here the relevant part of her study will b e presented

briey Sidners cognitive elements corresp ond to Karttunens discourse referents and

Webb ers discourse entities Sidner likeWebb er keepsthetermtoreferfortheactofaspeaker

p erformed by using the language Thus words themselves do not refer but they sp ecify or co

sp ecify a cognitive element

In the following example the NP green apples sp ecies a cognitive element in the hearers

mind The sp eaker uses the information in a cognitive representation A cognitive element is called

the sp ecication of green apples Cognitive elements are present in the memories of sp eakers and

hearers and are related to other cognitive elements For example the pronoun they in sp ecies

the same cognitive element green apples do es Since b oth NPs b ear the same relation to the

representation of apples they cosp ecify that cognitive element

a I think green apples taste b est

b And they make the b est co oking apples to o

Sidner states that sp ecications represent the ob jects referred to they b ear a wellstructured

corresp ondence to the ob jects in the real world For phrases suchasSanta Claus where there is

no world ob ject a sp ecication represents the mental schema to which are attached the prop erties

normally asso ciated with the imaginary p erson

Sidner shows that pronouns do not always have to cosp ecify cognitive elements For example

in the pronoun they is used to refer to the whole class of bikes while a monster Harley

mentions only some particular Harley Even if a phrase and a pronoun do not cosp ecify

the sp ecication of the phrase may b e used to generate that of a pronoun

a My neighb or has a monster Harley

b They are really huge but gasecient bikes

As a result of this observation Sidner prop oses that the pronoun interpretation is not simply

nding out the antecedent but some additional pro cess must generate a sp ecication for the

anaphor from the related phrase a monster Harley

Heim

Heim develops a le change semantics akin to Kamps Discourse Representation Theory

DRT to account for deniteness and indeniteness Only an indenite NP that intro duces a

discourse referent in Karttunens sense can op en a le card in the hearers mind A denite NP

on the other hand has to b e familiar and the corresp onding le card must b e in the domain

of an already existing le Heim revives the traditional approach to indeniteness whichwas

prop osed by Chistophersen and Jesp ersen later termed as familiarit ytheorybyHawkins

Familiarity theory basically states that A denite is used to refer to something already familiar

and an indenite is used to intro duce a new referent

Every le card is assigned a referential index with a numb er and it mediates b etween language

and world A hearer has a stack of zero le cards in hisher mind and as so on as she hears a

discourse as in a metaphorical le clerk starts writing the entries on the les as shown b elow

a A woman was bitten byadog

b She hit it

c It jump ed over the fence

For every indenite the hearer starts a new le card and for every denite the existing le

card is up dated A novel denite as in inferrables such as a bus the driver where the driver is

intro duced for the rst time into the discourse mo del via the mention of a bus is accommo dated

by a set of accommo dation rules mapping it onto the existing le card

Sentences are generated on various levels of representations one of which is the Logical Form

Each logical form is assigned a le change p otential Heims system also includes an assignmentof

truth conditions to les If a le represents facts ie meets the truth conditions then a sequence

of individuals satises the le a a satises F i a is a woman a is a dog and a bit a

An imaginary le clerk then changes the les as every prop osition p determines a le change from

F to F FpF which reads on the result of up dating F on account if p is F This le change

is ltered out by the wellformedness conditions such as disjoint reference

The domain of a le F is the set that contains every index of some card in the le For File

for example Dom F which means that in File there are two cards numb ered and

A novel NP is not yet in the domain of any le and a new le has to b e created if it establishes

a discourse referent

Heim uniformly distinguishes NPs that do not evoke discourse entities from those that do not

Karttunen has shown that nonsp ecic indenites do not intro duce discourse entities if they are

within the scop e of negative op erators quantiers yesno questions or when they are comple

ments of intensional verbs likewant and need Heim suggests that intensional verbs negation

conditional alikeinvolve op erators which takescopeover an indenite like other op erators ie

genericity questions and quantiers

Heim explains Karttunens examples cited in and b elow as follows both

want and know in volve universal quantication over a certain set of p ossible worlds in one

case the set of worlds that conform to the sub jects desires in the other case the set of worlds

which are compatible with the sub jects knowledge The worlds that conform to someones desires

often do not include the actual world but the worlds that conform to someones knowledge always

do that is what makes know factive Heim accounts for the acceptability of pronouns in b

by accommo dating a card in Johns knowledgeworld

a John wants to catchash

b Can you see it here

a John knew that Mary had a car

b But he had never seen it

Therefore the lifespan of a discourse entity starts with a cards addition to the le and ends

with that cards elimination from the le Heim

Referentiality

Related ResearchonTurkish

Dede and Tura b oth investigate referentiality and deniteness in Turkish Both stud

ies pursue the same approach but they complementeach other in that Dede examines verbal

sentences while Tura analyzes existential sentences Hence the following discussion of Dede

incorp orates b oth studies

Dede denes referentialityfollowing Givn as the sp eakers intent to refer to

or mean a nominal expression to have a nonempty reference ie to exist within a particular

universe of discourse Deniteness on the other hand is dened after Chafe as b eing at

the consciousness of the addressee at the time of utterance Based on these denitions Dede es

tablishes six statuses of sub jects and ob jects in Turkish concerning referentiality and deniteness

i denitereferential

ii denitenonreferential

iii indenitereferential

iv indenitenonreferential

v nondenitereferential

vi nondenitenonreferential

Sub ject NPs that are used with p ossessive morphemes and those that are mo died by deictic

terms have referential interpretation in nonmo dal contexts ie in contexts out of the scop e of

negation future generic mo dal verbs etc Word order plays a signicant role in determining

deniteness Indenite or nondenite NPs o ccur in preverbal p osition An NP in sentence ini

tial p osition is always denite unless it is stressed Dede prop oses that stress on NPs triggers

nondenite nonreferential status For example in B Dede suggests that if NP

saat clo ck is stressed the identity of the sub ject NPs is irrelevantIn other words the sub ject

NPs are incorp orated into the verbs However when the stress is on the verb the NP has a

referential denite interpretation

A Bu ses ne Ne oluyor

this sound what what happ enPROG

What is this sound What is happ ening

B Saat calyor

clo ck strikePROG

The clo ck is striking

Dede shows that nondenite NPs are those that are not marked for number in the

preverbal p osition asco cuk child in

Yerde co cuk yatyordu

groundLOC child liePROGPAST

On the ground child was lying

Bare ob jects have b een claimed in the literature to b e incorp orated and nonreferential see the

references cited in this chapter on Noun Incorp oration Dede shows that not all bare ob jects are

incorp orated This is in agreement with the ndings in this study as we shall see later NPs that

are not marked for numb er and deniteness are nondenite but some can b e referential although

Dede do es not distinguish b etween those that are referential and those that are not

Dede suggests that mo dal op erators as future p ossibility negation conditional question

probability genericity are triggers of nonreferentiality Accusativecaseisamarker of referential

ity in nonmo dal contexts Accusative casemarked ob jects in generic sentences are denite but

nonreferential

Referential Expressions

In this study a referential expression is describ ed as an NP that evokes a discourse entityina

discourse mo del Since entities are ascrib ed the prop erties and relationships predicated of

them in the text Webb er a discrete referential expression intro duced into discourse

can always b e talked ab out anaphorically with a denite pronoun or a denite NP

A referential expression asserts or presupp oses the existence of an individual or set of indi

viduals that could b e describ ed by the expression E within the discourse mo del In other words

when a sp eaker uses a noun phrase with expression E it makes sense to ask Which X is E Ap

p elt The physical existence of an ob ject is not necessary for conceptual entities may

also b e referential Hence fairy is a referential expression in a discourse of fairy tales Thus the

physical existence is a sucient but not a necessary condition for an expression to b e referential

tion on the particular use of an expression The referentiality dep ends on the sp eakers inten

Prince gives a discourse analysis of existential presupp osition In her analysis following

some work of Karttunen all denite NPs are marked as carrying Potential Sentence Presup

p ositions PSP Every PSP corresp onds to a Stated Assumption SA at the level of discourse In

addition to Stated Assumptions each individual has a set of Tacit Assumptions corresp onding to

NPs that denote humans and other entities eg Americans Turks Henry Smith etc along with

a set of prop ositions Fires may b e dangerous etc Prince Tacit assumptions only

exist in the cognition of individuals and hence they vary according to each individual According

to Prince these Tacit Assumptions enable p eople to sp eak ab out entities they may think do not

exist as in a sp eech ab out Go d b etween an atheist and a b eliever in Go d

Hence there is an anaphoric relation among Stated Assumptions of twospeakers SA is

attributed to sp eaker A and SA is attributed to sp eaker B B can utter SA where SA SA

without fear of it b eing attributed to him Prince Stated Assumptions in the discourse

mo dels of twospeakers thus allow them to talk ab out entities anaphorically The hearer will build

a discourse mo del where the entities in SA are intro duced by the sp eaker through the use of

referential expressions Otherwise Prince notes it will not b e p ossible to distinguish b etween

the comments of a psychotic and a therapist The existence presupp osition of an expression then

includes a set of Stated Assumptions that certain entities are accepted to exist in the discourse

mo dels of sp eakers These Stated Assumptions can b e ab out ob jects in the actual or p ossible

worlds

The existence presupp osition covers b oth attributive and referential expressions in the sense

of Donnellan Donnellan describ es this distinction as follows

Aspeaker who uses a denite description attributively in an assertion states that

something ab out who ever is the soandso A sp eaker who uses a denite description

referentially in an assertion on the other hand uses the description to enable his

audience to pick out whom or what he is talking ab out and states something ab out

the p erson or thing In the rst case the denite description might b e said to o ccur

essentially for the sp eaker wishes to assert something ab out whatever or who ever ts

that description but in the referential use the denite description is merely one to ol

for doing a certain job calling attention to a p erson or thing and in general any

other device for doing the same job another description or a name would do as well

In the attributive use the attribute of b eing the soandso is all imp ortant while it is

not in the referential use

Referential and attributive expressions approximately corresp ond to Value Loaded VL and Value

Free VF referential expressions cf Barwise and Perry Grosz Joshi and Weinstein

In a sentence like John is a do ctor the sp eaker refers to a particular individual named John

through a referential expression If the reference is successful the hearer will correctly identify

the individual On the other hand someone mightsay The strongest man in the world can

lift p ounds without identifying anyone particular in the world It may also b e the case

that the sp eaker has a particular p erson like Vladimir Jones in his mind though this additional

information do es not make himher exp ect that the hearer will necessarily pick up the referent

Likewise when someone utters the athlete who will win the race tomorrow she may not exp ect

the hearer to pick up the reference to a particular p erson like John since the hearer could not

p ossibly know who will t the description Nevertheless it is still reasonable to ask which runner

will win the race tomorrow for the expression presupp oses that there will b e such an individual

While John in our example of John is a do ctor is a referential expression the strongest man

in the world and the athlete who will win the race tomorrow are attributive in the sense of

Donnellan

The following examples cited in Dede indicate the referential ValueLoaded and

attributiveValueFree readings of the expression Cumhurbaskan President Unlike the ap

proach taken in this study Dede claims that attributive uses are nonreferential and she gives

the examples to illustrate referential versus nonreferential uses in a and b resp ectively This

study prop oses that b oth expressions are referential although not in Donnellans sense In a

there is a particular p erson that administers the Presidential Oce Mr Demirel in In b

on the other hand there is not a particular p erson who is the president it refers to any p erson

who is elected for the p osition Hence Cumhurbaskan President in a is a referential VL

and the one in b is an attributive VF expression

a Cumhurbaskan bugun Amerikadan dondu

president to day AmericaABL returnPASTSG

The President returned from the States to day

b Cumhurbaskan yedi ylda bir secilir

president seven yearLOC one electPASS AOR

The President is elected every seven years

Unlike Donnellans distinction the term referential in this study covers b oth referential and

attributive expressions Webb er Chapter pp states that the referentialattributive

distinction is irrelevantinevoking discourse entities In either case there is one and only one

individual discourse entity in the sp eakers discourse mo del whether she knows anything more

ab out that individual or not That is all that matters for denite reference A denite pronoun is

used to refer to a unique discourse entity indep endentofhowmany descriptions it satises just

as long as the given description is enough to make it unique

The distinction however is relevant in the subsequent reference Consider cited in Grosz

Joshi and Weinstein

a The vice president of the United States is also the president of the Senate

b Historically he is the presidents key man in negotiations with Congress

c As Ambassador to China he handled manytricky negotiations so he is well prepared

for this job

Both of the pronouns in b and c are third p erson singular The one in b expresses

no single individual vice president whereas in c the p erson that is referred to is George Bush

who was the vice president at the time This can b e accounted for by observing that the evoking

NP the vice president of the United States in a contributes to b oth VF and VL interpre

tations Grosz Joshi and Weinstein suggest that it is also p ossible to move the reference from

aVFinterpretation to a VL interpretation in the next utterance or vice versa They suggest

that although the transition b etween b to c may sound unnatural to some sp eakers similar

sequences are p ossible Nonetheless strong constraints are imp osed on the kinds of transitions

If a given utterance forces either a VL or a VF reading then only that interpretation b ecomes

available in the subsequent utterance This constraint is also observed in Turkish discourse as in

Sp eakers cannot pro ceed with the other reading once they have committed themselves to a

certain reading

a Cumhurbaskan her hafta basbakanla goru sur

president every week prime ministerWITH meetAOR

The President meets the Prime Minister every week

b Ama bu hafta goru semedi

but this week meetNEGPAST

However she couldnt meet him her this week

c Her ay bakanlarla toplant duzenler

every month ministerPLUWITH meeting arrangeAOR

She arranges a meeting with the ministers every month

In a b oth VF and VL readings are available In b however the VL reading is enforced

on the partial information given in a This imp oses the same reading in c whichmighthave

b een ambiguous at the discourse initial p osition

The discussion so far has shown that a discourse entity can b e unsp ecied with resp ect to VL

and VF readings As the sp eaker incrementally builds the discourse she builds and maintains

a discourse mo del in which she attempts to direct the hearer via the reference The pronominal

anaphora can then b e conceived as a set of tacit instructions to the hearer for synthesizing a

discourse mo del which equates as closely as p ossible to that of the sp eaker This observation

supp orts the ndings of Situation Semantics Theory Barwise and Perry Tn and Akman

and the references therein that utterances convey partial information

Nonreferential Expressions

y and cannot A nonreferential expression do es not create a new le card or evoke a discourse entit

b e discussed anaphorically by denite reference in the subsequent discourse It is imp ossible to

ask a question likeWhich X is E for a nonreferential expression E

Referentiality vs nonreferentialityinTurkish is marked byvarious linguistic devices Tura

Dede Erguvanl These factors are discussed b elow

Referential Contexts and Discourse Entities in Turkish

In the following section referential and nonreferential contexts and their roles in evoking discourse

entities will b e discussed It will b e shown that only referential expressions can b ecome discourse

entities and can b e talked ab out subsequently using denite reference

Predicative Nominals

An indenite predicative nominal will not evoke a discourse entity Kuno Webb er

App elt as exemplied in a

talking ab out the comp etition in the exams

a Her sey sonsuz bir yars

i

every thing endless one racecomp etition

Everything is an endless comp etition

b Ben sevmem yarslaronlar

i

I likeNEGSG comp etitionPLUACC

I dont like comp etitions

c Birileri mutlaka cignenecektir yarslarda

onePLU denitely stepPASSIVEFUT comp etitionPLULOC

Some p eople will denitely b e stepp ed on in comp etitions

Gulderen Bilgili Bir GeceYolculugu

In a yars comp etition is a predicative nominal Note that one cannot ask Which com

p etition dont you like to single out the entityItdoesnotevoke a discourse entity that can b e

accessed by a denite pronoun It has to b e mentioned by a full NP in its second mention in b

and even in its third mention in c

Situationally Evoked Entities

A discourse entity can either b e textually evoked or it can b e situationally evoked by p ointing

at some entity in the p erceptual eld Prince a If a cat enters the ro om its app earance

makes it a saliententity and enables the sp eakers to refer to it Likewise the mo dication of a

oked entities Possessive marked NPs noun by demonstratives as in is used for situationally ev

also evokeentities unless the existence of the entity is denied within the scop e of negation

In example su ogrenci that student is situationally evoked

Su ogrenci seni b ekliyor

that studentyouACC waitPROGSG

That studentiswaiting for you

Dede

AccusativeCase

Erguvanl and Dede have emphasized the function of accusative case as a marker of

b oth deniteness and referentiality

Enc claims that accusative case in fact marks sp ecicity She devises a theory of sp eci

city which is closely related to Heims theory of deniteness According to Enc deniteness

and sp ecicity are related phenomena b ecause b oth require linking to previously evoked entities

The dierence b etween denites and sp ecics in Encs framework is the relationship with their

antecedents The relationship b etween the antecedent and the denite NPs names pronouns

and denite descriptions is that of an identity relationship which she calls a strong antecedent

Sp ecic NPs on the other hand involve the inclusion relation which she calls a weak antecedent

Identity of referents also entails inclusion Enc assumes that indenites may b e sp ecic or non

sp ecic while partitives are necessarily sp ecic

Enc argues that ob ject NPs in Turkish unlike in English are never ambiguous with

resp ect to sp ecicity Sp ecic NPs always receive accusative case while nonsp ecic NPs do not

carry case morphology as exemplied in cited in Enc

a Odama birkacco cuk girdi

ro omSGPOSSDATseveral child enterPAST

Several children entered myroom

b Iki kz tanyordum

two girlACC knowPROGPASTSG

I knew two of the girls

b Iki kz tanyordum

two girl knowPROGPASTSG

I knew two girls

If b is uttered after a the two girls are included in the set of children evoked On the

other hand b is ab out two girls who are not in the set of children in a They are novel in

the sense of Heim and thus the NP intro duces new discourse entities and a new le card has to

b e created

Enc prop oses that accusativecasemarked ob jects as b are semantically interpreted as par

titives This in terpretation places a constraint on the structure of discourse in addition to the

constraint placed by the deniteness of the NP In Encs theory of sp ecicity NPs b ear referential

indices as in Heims theoryEnc ensures that all denites are sp ecic by prop osing that all NPs

carry a pair of indices the rst represents the referent of the NP and the second represents a

previously mentioned discourse entity that it may b e related to through the identity or inclusion

relationships The indices themselves b ear a deniteness feature The deniteness feature of the

second index determines the sp ecicity of the NP by constraining the relation of the entitytoa

previously evoked entityFor example in ab ove although the NP in b carries an indenite

ness index it is partitive The familiarity condition requires that it has to b e dep endentona

previously established le Since it is required that iki kz two girls is a subset of birkacco cuk

several children the former NP has a sp ecic interpretation

Encs analysis ensures that every denite NP is sp ecic Since the denite NP has to b e

dep endent on another le card its second index will assign it a sp ecic index Enc

writes The analysis prop osed here predicts that there will b e no nonsp ecic denite NPs As a

consequence all denites in Turkish are predicted to carry accusative case in the ob ject p osition

On the other hand the NP in b carries an index marked for indeniteness it is not de

p endent on a previously evoked entity and the NP is totally novel in the discourse Therefore

the NPs in b and b are b oth indenites but the nonsp ecic indenite in b is more novel

than the sp ecic indenite in b This analysis assumes that sp ecics like denites must have

their antecedents within the previous discourse while nonsp ecic indenites must b e completely

novel A completely novel NP must not haveanantecedent within the prior discourse whereas a

sp ecic indenite has to have one If the indenite marked NP is a subset of a previously evoked

entity it will b e sp ecic If it is not linked to another entity it will b e nonsp ecic

However Encs prop osal that NPs in Turkish are never ambiguous with resp ect to sp ecicity

and that every sp ecic NP must carry an accusative case morpheme app ears to b e to o strong

For example in ab cited in Tn and Akman the NP siyah bir araba a black car is

not overtly accusative casemarked and yet it is ambiguous b etween sp ecic and nonsp ecic



interpretations at the discourse initial p osition The use of an overt pronoun in the subsequent

utterance disambiguates the reading as sp ecic Since this sp ecic NP has a wider scop e than the

op erator of the intensional verb it can create a le card for subsequent denite reference

a Ahmet siyah bir araba aryordu

i

Ahmet black one car seekPROGPAST

Ahmet was lo oking for a black car

b Bir sure sonra onu buldu

i

a while later itACC ndPAST

After a while he found it

c Ama bisikletini hala bulamad

but bicyclePOSSSG still ndABILNEGPAST

But he still couldnt nd his bicycle

Compare with The same NP in do es not evoke a discourse entity b ecause it is

nonsp ecic and is within the scop e of an intensional op erator along the lines of Heim

a Ahmet siyah bir araba aryordu

i

Ahmet black one car seekPROGPAST

Ahmet was lo oking for a black car

b Ama oyle bir araba bulamad

but such one car ndNEGPAST

But he couldnt nd onesuch a car

Furthermore Dede shows that some NPs in generic sentences are optionally accusative

casemarked in the ob ject p osition in Turkish Consider cited Dede and



However in a fo otnote Tn and Akman write that the discourse is more coherent if the NP is accusativecase

marked The native sp eakers I consulted rep orted that is more coherent without accusative case

a co cuklar cukulata sever

childPLU cho colate likeAOR

Children likecho colate

bCo cuklarcukulata y sever

cho colateACC

Children likecho colate

Bizim evde cay her zaman Aytul yapar

our houseLOC teaACC all time Aytul makeAOR

Aytul always makes tea in our family

Erguvanl prop oses that animacy plays an imp ortant role in the assignment of ac

cusative case in generic sentences NPs that denote animates havetobemarked by the plural

morpheme ler and the accusative case at all times while nonanimates can b e bare as in the

following examples Erguvanl

a Ben insanlar severim

I humanPLUACC likeAORSG

I like p eoplehuman b eings

b Ben insan severim

c Ben insanlar severim

In b and c singular insan p erson and plural insanlar p eople are not grammatical the

NP has to b e plural and overtly accusative casemarked

On the other hand inanimate entities can app ear as bare noun phrases in the ob ject p osition

They can app ear without an overt indeniteness marker ie the indenite numeral bir one and

deniteness markers like accusative case p ossessive morphemes and demonstratives Erguvanl

a Ben elma severim

eAORSG I apple lik

I like apples

b Ben elmay severim

appleACC

I like the apple

c Ben elmalar severim

To sum up accusative case marking functions as deniteness and sp ecicitymarker In generic

sentences it has to b e present when the entity denoted is animate or rather human while it can

b e optional in a generic sentence when the ob ject is inanimate It app ears that accusative case

has a function to uniquely identify and distinguish one entity to the exclusion of others

Generics

NPs in generic sentences are considered to b e nonreferential expressions by Erguvanl and

Dede However an accusative casemarked NP in a generic sentence can serveasanan

tecedent for a denite pronoun provided that the antecedent is uniquely identied and distin

guished from all other ob jects For example in the sp eaker can use an overt pronoun to refer



to the generic ob ject However an overt pronoun can access the entity only when it exhausts

the whole category denoted by the NP

A Ben sut u severim

I milkACC likeAORSg

Ilike milk

B Ben onu sevmem

I itACC likeNEGAOR

I dont like it

cay tercih ederim

teaACC preference makeAORSg

I prefer tea

Note that in B the overt denite pronoun is acceptable while in b only null indenite

anaphora can b e used This is b ecause the milk the sp eaker drinks in b is only a subset of the

category and thus the overt pronoun cannot exhaust the whole kind Hence the infelicityofthe

overt pronoun follows

a Ben sut u severim

I milkACC likeAORSg

I like milk

onu her gece icerim b

itACC every night drinkAORSG

I drink some every night

When someone says The butcher sells go o d meat as in a meat denotes the category rather

than distinguishing some particular amount of meat Thus it cannot function as a discourse entity

and cannot b e accessed by a denite pronoun while a null argument is acceptable as indicated

in b

a Kasap iyi et satar

butcher go o d meat sellAOR

The butcher sells go o d meat

b Ben hep onueti ondan alrm

I always itmeat heABL buyAORSG

I always buy it from him



Note the list reading in this sentence imp oses proso dic prominence on the ob ject ie phonological fo cus

If the generic op erator quanties over the denotation of an NPanovert pronoun cannot b e

used to denote a subset of that category but a null pronoun is felicitous If an overt pronoun is

used it must b e completely identical to its referent and cannot b e a subset of it

Negation

Indenite nonsp ecic NPs within the scop e of a negative op erator cannot evoke discourse entities

since negation creates a referentially opaque context Tura writes Negation is the

strongest op erator that aects statuses of indenite NPs in existential sentences It imp oses the

condition that there are no identiable referents for the NPs in its scop e This observation is

exemplied in

Bahcede kop ek yok Korkma

gardenLOC dog notexist afraidb eNEG

There is no dog in the garden Dont b e afraid

Tura

Example in b also shows the referentially opaque context created by the negation op erator

a A Kzla bulusacagm bu aksam

GirlWITH meetFUTSG this night

I will meet my girl friend tonight

bSemsiy em de yok ustelik

umbrella to o not exist b esides

And b esides I dont even haveanumbrella

c B Ben sana onu bulurum

I youDAT ndPRESSG

Ill nd you one

Ozdemir Basargan Gurb et Sofras

In b semsiyeumbrella is a nonreferential expression and although a n ull pronoun can

b e used as in c an overt pronoun is prohibited The null pronoun functions as an indenite

anaphora analogous to nonsp ecic oneanaphora in English as obvious in the translation

Compare with In the utterance b the sp eaker takes it for granted that the water

existed while the existence is denied in a Consequently the utterance in b is not felicitous

b ecause of the nonreferential NP in the negativecontext in a while b is felicitous

a Musluktan bir damla su akmyordu

faucetABL one drop water owNEGPROGPAST

Not even a few drops of water were owing from the faucet

b O da bulankt

it to o muddyCOPPAST

And it was muddy

a Musluktan bir damla su akyordu

faucetABL one drop water owPROGPAST

A few drops of water were owing from the faucet

b O da bulankt

it to o muddyCOPPAST

And it was muddy

However negation do es not create a referentially opaque context if an adverb like artk any

more which presupp oses that the referent existed in the past is used Due to this presupp osition

in a ev house can evokeanentity that can b e accessed by the sub ject and ob ject denite

pronouns o it in b and onu it in b resp ectively

a Artk Alinin evi yok

any more AliGEN housePOSSSg notexist

Ali do esnt haveahouseany more

b O gecen hafta satld

it last week sellPASSPAST

It was sold last week

b Onu gecen hafta satt

itACC last week sellPAST

He sold it last week

In addition as has b een previously mentioned Webb er shows that the scop e of negation

can be potentially ambiguous in English This ambiguity is also observed in Turkish For example

in either the existence of a Swede or Alis marrying her is denied The ambiguity is resolved

in the subsequent discourse in b the reading Alis not marrying is forced On the other hand

when b is uttered the nonexistence of the entity is conrmed Note that since in this reading

the entit y is not evoked in a the NP in b should b e novel

a Ali bir Isvecli ile evlenmedi

i

Ali one Swedish with marryNEGPAST

Ali didnt marry a Swedish

b Onunla birlikte oturuyor

i

sheWITH together livePROG

He lives with her

b Aslndaoyle bir Isvecli yok

In fact such one Swede existnot

In fact no suchSwede exists

To sum up nonsp ecic NPs within the scop e of negation do not evoke discourse entities

Nevertheless the scop e of negation can b e ambiguous and in these cases a single reading may not

be available at rst In subsequent discourse the entitymay b e conrmed by the use of denite

reference or denied and thus the ambiguity will b e resolved

Such partial information is not only due to scop e ambiguities As has b een previously dis

cussed referential expressions maybeambiguous in terms of their Value Free and Value Loaded

readings and NPs can b e ambiguous in terms of sp ecicity and incorp oration

YesNo Questions

Indenite nonsp ecic NPs within the scop e of yesno questions do not evokeentities

A Ali yeni bir araba m ald

Ali new one car QUEST buyPAST

Did Ali buy a new car

B Hayr evini sattktan sonra onu alacak

No housePOSSSGACC sellNOMABL after itACC buyFUT

No hell buy one when he sells his house

In A araba car is a nonreferential expression As a result only indenite null anaphora is

allowed as in B

Noun Incorp oration

It has b een previously stated that denite direct ob jects in Turkish are overtly accusative case

marked while indenite ob jects do not haveovert case markings Indenite ob jects are mo died

bynumeral bir one and indenite quantiers baz biraz some bir kac a few and these ob jects

are not overtly accusative casemarked unless they are sp ecic see the discussion of Enc

ab ove There is another set of ob jects which are neither overtly accusative casemarked nor

mo died by the indenite numeral or any other mo dier These ob jects will b e referred to as



bare ob jects and it will b e investigated whether they can b e discourse entities

Ob jects that are not casemarked b ehave dieren tly than those that are casemarked One

such dierence is that the former are strictly adjacent to the verb in the preverbal p osition while



There is a set of bare ob jects in Turkish that is excluded in the discussion One typ e is the nominal part

of idioms These parts can b e represented with a zero across sentences However since idioms are listed in the

lexicon and the verb and its complementhave lost their original meaning the ob ject will never b e an indep endent

expression

Furthermore languages which allow ob ject drop like Mandarin Chinese and Turkish do not allow a zero indirect

ob ject unless it has an antecedent within the previous discourse These languages avoid zero ob jects in discourse

initial p ositions by using either dummy or cognate ob jects eg yaz yazmak literally to write writings Allerton

These dummy or cognate ob jects are bare as in the cognate ob ject yemek fo o d in yemek yemek to eat

are not indep endent constituents These ob jects can also b e mentioned bynull ob jects in the subsequent discourse

but can never b e accessed with overt pronouns

The bare ob jects of p eriphrastic verbs like etmek to do never evoke discrete entities The p eriphrastic verbs form

averbal complex with a nominal argument mostly b orrowed from foreign languages and listed in the lexicon as a

single unit Since Turkish as other languages resists adapting any foreign verbal morphology it either b orrows the

noun or nominali zes the foreign verbtoformaverbal complex with a native p eriphrastic verb Some examples are

as in ameliyat etmek lit to do an op eration to op erate on someone and telefon etmek lit to do telephone to

telephone The nominal part of suchverbs can b e zero across utterances in a discourse but can never b e accessed

with overt ob jects

the latter can scramble to various p ositions in a sentence Erguvanl Erku Knecht

among others Ob jects that are not overtly casemarked cannot o ccur in sentence initial

or p ostverbal p ositions but are strictly anchored to the preverbal slot They cannot b e separated

from the governing verb byanNPorbyanadverbial For example in while the accusative

casemarked kitab the b o ok can b e separated from the verb byanadverb the bare ob ject should

b e anchored to the preverbal p osition except one particular set of particles de to o bile even

and yesno question particle mI

a Ali kitab yavasyavas okuyor

Ali b o okACC slow slow readPROG

Ali is reading the b o ok slowly

b Ali yavasyavas kitap okuyor

Ali slow slow b o ok readPROG

Ali is reading some b o oks slowly

c Ali kitap yavasyavas okuyor

d Ali kitap da okuyor

Ali b o ok to o readPROG

Ali is reading some b o oks also as well as some other things

or e Ali kitap m okuy

Ali b o ok QUEST readPROG

Is it a b o oks that Ali is reading

The immobility of bare ob jects has led some researchers to argue that they undergo a pro cess

of Noun Incorp oration NI where the ob ject constitutes a semantic unit together with its verb

Nilsson Knecht and the references therein Kuribayashi a b for various

syntactic treatments of NI in Turkish In NI the incorp orated bare ob ject mo dies the verb and

loses its indep endent semantic status The deniteness and the numb er of the item denoted are

irrelevant Erguvanl Homan Due to this these typ es are called nondenite Dede

or categorial Nilsson

Bare Ob jects Noun Incorp oration and Referentiality

Nilsson Knecht and Kuribiyashi a b have prop osed that since a bare ob ject

loses its semantic indep endence it will always b e nonreferential If so incorp orated bare ob jects

will not evoke discourse entities However it do es not app ear to b e the case that all bare ob jects

are incorp orated and lose their semantic indep endenc e Nilsson writes

the choice b etween nonmarking and marking ieby accusative case in the expression

read the Koran Kuran Kuran oku could not b e related in the usual way to the

contrast b etween nonreferring and referring since Koran stands for something

unique and is therefore in some sense always a referring term like prop er nouns

italics ADT

Furthermore Dede classies two categories of bare ob jects dep ending on whether

they are referential vs nonreferential expressions She shows that NPs that are not mo died by

indenite numerals or quantiers o ccur in contexts where identiabilityoftheentity is irrelevant

There are two cases where the identiability of the referentisirrelevant

i The referent is identiable by the sp eaker but he do es not feel it necessary for the

hearer to pick it out That is the sp eaker is not interested in the identiabili ty of the

entityinvolved he is interested in conveying the class memb ership of the referent

Dede exemplies this by the following example

Bugun Aliye corap aldm

to day AliDATsock buyPASTSG

Today I b oughtsocks for Ali

ii The referent constitutes an integral part of the meaning of the verb That is it is

incorp orated into the verb

Aytul kitap okuyor

Aytul b o ok readPROGSG

Aytul is reading a b o ok

Dede argues that in the sp eakers intention is not to make hisher hearer identify the

entity but to mo dify the action of reading and therefore the noun undergo es incorp oration

Dede argues that NPs of class memb ership typ e in i ab ove can b e referential or nonreferential

expressions dep ending on the context whereas the incorp oration typ e in ii will never b e refer

ential For the sakeofconvenience Dedes former class will b e called classmemb ership and the

latter incorp orated in the discussion b elow

Dede do es not explicitly state that the use of denite reference is only allowed if the antecedent

is a referential expression Nevertheless in cited in Dede Dede denes corap so cks

elbise dress and ayakkab sho e as referential expressions As evidence she suggests that they

can b e accessed bythequantier hepsi all in b Dede

a Dun Aliye corap Pnara elbise Aytule ayakkab aldm

Yesterday AliDAT so cks PinarDAT dress AytulD AT sho e buyPASTSG

Yesterday I b oughtsocks for Ali a dress dresses for Pinar a pair of sho es for Aytul

b Hepsini de begendiler

AllACC to o likePASTPLU

They liked all of them

Dede claims the same expressions will b e nonreferential under the scop e of future tense and

hence these NPs will not serveasantecedents for subsequent reference Dede

a Yarn Aliye corap Pnara elbise Aytule ayakkab alacagm

TomorrowAliDAT so cks PinarDAT dress AytulD AT sho e buyFUTSG

Tomorrow I will buy so cks for Ali a dresses for Pinar sho es for Aytul

b Hepsini de begendiler

AllACC to o likePASTPLU

They liked all of them

According to Dede b is infelicitous b ecause at the time of utterance the refer

ents of the NPs are not yet existent for the sp eaker However Dede do es not mention cases as in

where the reference is acceptable Thus if one accepts that the accessibility of an expression

with a pronoun is an indicator of its referential status then these expressions are referential

a Yarn Aliye corap Pnara elbise Aytule ayakkab alacagm

TomorrowAliDAT so cks PinarDAT dress AytulD AT sho e buyFUTSG

Tomorrow I will buy so cks for Ali a dresses for Pinar sho es for Aytul

b Umarm hepsinionlar begendiler

hop ePRESSG AllACCtheyACC likePASTPLU

I hop e theyll like them all

w that reference should b e made in the same context If The contrast b etween and sho

tense is considered to b e a mo dal op erator the reference to the entity should b e made in the same

mo dal context Alternatively if tensemo dalityistaken to b e a discourse anaphor as suggested

byWebb er and the references therein it is conceivable that these twotyp es discourse

anaphors interact and that tenseasp ect constrains the use of pronominal anaphora

Dede argues that ob jects that have undergone incorp oration will always b e nonreferential as

she exemplies in a Dede

a Aytul kitap okuyor cok ilginc

Aytul b o ok readPROGSG very interesting

Aytul is reading a b o oks Thats very interesting

b Aytul bir kitap okuyor cok ilginc

Aytul book readPROGSG very interesting

Aytul is reading a b o ok Its very interesting

According to Dede in a the ob ject is incorp orated forming a unitary constituent with the

verb and thus has lost its indep endence Dede prop oses that in a what is interesting is Aytuls

b o okreading rather than the b o ok she is reading In b on the other hand the NP represents a

discrete entity that is not incorp orated into the verb note that it is mo died by bir one hence

the sp eaker can refer to it in the subsequent discourse

However it will b e shown b elow that the bare ob ject in a is not necessarily incorp orated

For one thing native sp eaker judgments are subtle as far as a and b are concerned Some

informants rep orted that b oth a and b are ambiguous b etween the two readings Aytuls

reading is interesting and the b o ok she is reading is interesting Yet some other informants

have rep orted just the opp osite of Dedes judgments These observations lead one to think that

kitap b o ok in a is ambiguous The bare ob jects in b oth a and a can either b e referential

or nonreferential b ecause NI is optional in Turkish When kitap okumak to read a b o ok is

incorp orated the noun cannot evokeanentity but when NI do es not apply the ob ject evokes an

entity Only in the latter case can a sp eaker utter cok ilginc Its very interesting with the null

sub ject pronoun it felicitously accessing kitap b o ok as a discrete entity after an utterance like

Aytul kitap okuyor Aytul is reading a b o oks

Below b oth evidence and counterevidence for NI will b e presented I will b e prop ose that NI

is optional in Turkish

Arguments for Incorp oration

One piece of evidence for incorp oration is as has b een previously stated no NP or adverb can

intervene b etween the verb and its incorp orated ob ject Secondly an incorp orated noun cannot

head a relative clause which implies that the noun is not an indep endent constituent In a

yemek fo o d is a bare noun the one in b is mo died by bir one Note that only the latter can

head a relative clause Erguvanl

a Nazan hazm zor olan yemek pisiriyor

Nazan digestionPOSSSG hard b eREL meal co okPROG

bNazan hazm zor olan bir yemek pisiriyor

Nazan digestionPOSSSG hard b eREL one meal co okPROG

Nazan is co oking a fo o d dish that is hard to digest

The third evidence for incorp oration comes from stress assignment In lexical comp ounds

whether verbs or nouns p eak stress is assigned to the primarily stressed syllable in the rst

element of the comp ound Stress cannot b e shifted to the head of the comp ound The following

examples are cited in Knecht and attributed to Swift

b oyUNbag necktie b oyun neck bag tie

EV kaps house do or ev house kap do or

Knecht prop oses that primary sentence stress cannot b e assigned to the verb in the case of

NI In the capitalized syllable shows the stressed syllable

A Murat o dasnda mekTUP yazyor degil mi

Murat ro omPOSSLOC letter writePROG not QUEST

Murat is writing a letter in his ro om isnt he

B Hayr yazmyor

no writeNEGPROG

No he isnt

MekTUP okuyor Mektup oKUyor

letter readPROG

He is reading a letter

Arguments against Incorp oration

On the other hand there is also counterevidence for NI in Turkish As has b een p ointed out

ab ove certain typ es of particles ie de to o bile even can intervene b etween the verb and the

so called incorp orated noun

Secondly in naturallyo ccurring data there are instances where a bare ob ject functions as the

head of a relative clause as in and

Krk yasnda guzel bir erkegin kandramayacag kadn m vardr

FortyyearPossLo c handsome one man seduceNEGRELACC woman QUEST exist

Are there any woman who a handsome fortyyear old man cant seduce

Sait Faik Butun Eserleri

In the bare noun kadn woman is the head of the relative clause If kadn kandrmak to

woman seduce is an incorp orated construction it should not b e p ossible to relativize it Likewise

op ek gormek to dogsee is incorp orated the bare ob ject should not head the relative clause if k

in

Siz hic sesi kslms kop ek gordun uz mu

you ever voicePOSSSg lowerPASSPASTREL dog seePASTPLU QUEST

Haveyou ever seen a dog that has lost its voice

Thirdly as Erguvanl suggests NI should detransitivize the verb b ecause the direct

ob ject has putatively formed a complex predicate with its verb If NI really detransitivizes a verb

one would exp ect that this verb should not passivize and yet incorp orated verbs do passivize in

Turkish In although oy vermek to vote give is incorp orated it can b e passivized

Onlara oy verilir

TheyDAT vote givePASSAOR

Lit Votes can b e given to them

However as Erguvanl also states this cannot b e a strong argumentforTurkish b ecause

Turkish allows passivization of intransitiveverbs as well as transitiveverbs as in the following

Gecen hafta daga gidildi

last week mountainDAT goPASSPAST

Lit Last week it was gone to the mountainthe mountain was gone to

Erguvanl

Girilmez

EnterPASSNEGAOR

Lit It is not enteredDo not enter

Knecht

Passivization is not a strong argument against incorp oration since it is conceivable that the

incorp orated complex verb is passivized as though they are intransitiveverbs as in and

The fourth argument against incorp oration however is stronger As Erguvanl notes an

adverb cannot intervene b etween an ob ject and its verb even when the ob ject is a referential

expression as in a and b Thus the adjacency requirement is not necessarily limited to the

cases of incorp orated bare ob jects Consider

a Murat aceleyle bir kitap okuyor

Murat hurriedly one b o ok readPROG

Murat is hurriedly reading a b o ok

b Murat bir kitap aceleyle okuyor

one book hurriedly

c Murat bir kitab aceleyle okuyor

one b o okACC hurriedly

Murat is reading a certain b o ok hurriedly

Erguvanl

In b an adverb cannot intervene b etween the verb and its direct ob ject if the ob ject NP is not

ertly accusative casemarked NP in c allows the intervention overtly casemarked Note that ov

of the adverb The ob ject in b is a referential expression not a bare noun

In conclusion there is b oth evidence and counterevidence for NI in Turkish

NI is Optional

The evidence and counterevidence for NI discussed ab ove and the indeterminacy as to when bare

ob jects represent referential expressions supp ort the theory that some bare ob jects in Turkish can

optionally undergo a pro cess of incorp oration If incorp oration applies the bare ob jects cannot

b ehave as discrete entities and indep endent constituents They b ecome the part of the complex

predicate and thus they cannot serveasantecedents for denite pronouns However when they

are not incorp orated they can evoke discourse entities Consequently a sentence at the discourse

initial p osition can convey partial information as to whether a bare ob ject is incorp orated or not

Consider

Aytul kitap okuyor

Aytul b o ok readPROG

Aytul is reading a b o ok

The ob ject kitap b o ok in will either b e incorp orated or a discrete constituent If it is

incorp orated we will have a le card containing information that she is b o okreading but the

b o ok will not evoke a separate card as shown in a if it is not incorp orated the ob ject will

create a separate le card as in b and this le card can b e up dated in the subsequent discourse

Bo ok

Aytul

readAytul b o ok

She is b o okreading

b a

SimilarlyPortereld and Srivastav observe that in Hindi a phrase corresp onding

to read b o ok is ambiguous b etween referential and incorp orated readings

Shortterm Discourse Entities

So far it has b een assumed that once a referential expression is intro duced it is available through

out the discourse but cf example ab ove However not all entities can p ersist throughout a

discourse segment Consider and

a Eger Cann co cugu olursa

i

If JohnGEN childPOSSSG b ePRESCOND

If John has a child

b onu Adanada buyutecek

i

sheACC AdanaLOC growCAUSFUT

he will raise himher in Adana

c Sonra da onu yatl okula gonderecek

i

Then to o sheACC b oarding scho olDAT sendFUT

And then he will send himher to b oarding scho ol

c Onu seviyor mu

sheACC lovePROG QUEST

Do es heCan love himher

a Ahmetin arabas oldugunu var sayalm

AhmetGEN carPOSSSG b eNOMACC exist countPLU

Lets assume that Ahmet has a car

b Onu havaalannda brakms olabilir

itACC airp ortLOC leavePAST b ePOSSIBPRES

He mayhave left it at the airp ort

c Onu gorebiliyor musun

itACC seeABILPROG QUESTSG

Can you see it

Examples and create hyp othetical worlds in which the referents denoted by the NPs

are assumed to exist Thus the NPs evokeentities which can b e accessed byovert pronouns

However they have a limited lifespan and they cannot serveasantecedents for pronouns in

c and c Karttunen prop oses that f ictitious individuals may b e referred to

anaphorically as long as the prop er ctitious mo de is sustained but when the illusion is broken

they cease to exist

In Heims theory a p ermanententity is either free in the text or b ound bya

quantier whose scop e encompasses the entire text As has b een stated Heim prop oses that

conditionals generics mo dals and attitude nonfactive ie intensional verbs suchaswant

all involve op erators Op erators trigger existential closure inside their scop e Thus NPs within

the scop e of op erators will not b e p ermanententities ie they will b e aliveun til the end of the

scop e of the mo dal or verb but not b eyond For instance when someone says If John buys a

b o ok he must b e reading it the pronoun it is p ossible b ecause it is within the scop e of an

op erator This can explain why the denite reference in b is p ossible but the one in c is not

However note that the pronoun c is allowed in spite of the fact that it is b eyond the scop e

of the op erator The reason Heim suggests is that pronouns as in c and b are acceptable

b ecause they are under the same typ e of op erator as in the initial sentences Heim prop oses an

accommo dation analysis In a p ossible world created by supp ositions and conditionals the facts

do not include the actual world Thus the reference to the actual world is not p ossible One

has to accommo date a card describing entities in a p ossible world Therefore the reference is

wellformed to the extent that b oth denite pronouns and their antecedents are used to refer to

an entity within the same or an equivalent p ossible world

Rob erts devises a formal theory of mo dal sub ordination within the framework of Dis

course Representation Theory DRT to account for shortterm discourse entities Mo o d in

Rob ertss analysis is a feature of sentence use not a grammatical feature of verbs it relates

to the sp eakers commitment to the truth of a sentence in the actual world If a sp eaker indicates

by conventional means that a sentence is interpreted as true in the actual worlditisutteredinthe

factual mo o d If a hyp othetical assumption is made or if some question arises ab out the actual

truth of a sentence then it is uttered in nonfactual mo o d Some means of indicating nonfactual

mo o d are expressions like supp ose that conditionals adverbials like probably supp osedly etc

The sp eaker who makes a hyp othetical suggestion do es not commit himselfherself to the truth in

the actual world Rob erts uses Kratzers theory of mo dality where necessarily p ossiblywould

etc are mo dal op erators A sentence like Ella might lift that refrigerator can b e relativized to

the set of p ossible worlds in which facts ab out human strength gravity etc are true and in that

world Ella can lift the refrigerator These p ossible worlds are sub ordinated under the Discourse

Representation Structures DRS of Kamp

Following Heim Rob erts assumes that conditionals involve an op erator symb olized bythe

square necessity op erator in the DRS ab ove Both of the clauses in the conditional are in non

factual mo o d and sub ordinated Since there is no mo dal in b sp eakers are not directed to

interpret the sentence in the nonfactual mo o d Thus the two utterances in the discourse are

at dierentlevels of DRS one sub ordinated while the other is not As a result the anaphoric

relation in is imp ossible In on the other hand b oth utterances in nonfactual mo o d are

sub ordinated and the anaphoric relation is felicitous

a If John b ought a b o ok hell b e reading it bynow

b Its a murder mystery

a If John b ought a b o ok hell b e reading it bynow

b Itll b e a murder mystery

To sum up not all entities can endure throughout a discourse segment However an entity

can b e accessed by denite pronouns so long as b oth are uttered in the same mo o d

Bound Variables and Pronouns of Laziness

This chapter has argued that an overt pronoun should not b e in an anaphoric relationship with

an NP unless b oth denote an identical referent There are twotyp es of intersentential pronomi

nalization phenomena in English where denite pronouns can b e in a dierenttyp e of anaphoric

relationship ie they can function as b ound variables and pronouns of laziness as will b e sho wn

b elow The suggestions made so far predict that overt pronouns in Turkish cannot o ccur in these

cases and this section investigates whether this prediction is b orne out

Consider where the denite pronouns are anaphorically linked to the quantied NPs This

typ e of link cannot b e that of coreference b ecause for example in a there is not a single

universal mother that every man loves but rather every man has p otentially a dierent mother

whom he loves By the same token neither the quantied NPs nor the pronouns in ac denote a

unique identiable male individual The pronouns in ac are not coreferential with the resp ective

quantied NPs but they are variables b ound within the scop e of a ccommanding quantier and

vary among the range of quantied individual s

a Every man loves his mother

b Nob o dy thinks he should join the party

c Everyone decided that the whisky is go o d for him

Note that overt pronouns in Turkish cannot b e in an anaphoric relationship with a quantied

NPasshown in the a sentences in while they can access a referential expression in the

b sentences Since ob jects in emb edded sentences cannot b e null when coindexed with their

matrix sub jects in Turkish cf Kornlt Erguvanl Chapter a logophoric pronoun

homophonous with reexive anaphor kendi or kendisi himselfherself is used as a b ound variable

a Herkes onu kendisini kovan mudurden nefret ediyor

i i i

everyb o dy sheACChimherselfACC reREL managerABL hatred doPROG

Everyb o dy hates the manager who red himher

b Ali onu kendisini kovan mudurden nefret ediyor

i i i

Ali heACChimselfACC reREL managerABL hatred doPROG

Ali hates the manager who red him

a Her pilot onu kendisini kovalayan ucaga atesetti

i i i

Every pilot sheACChimselfACC chaseREL planeDAT re doPAST

ery pilot red at the plane that chased himher Ev

b Ayse onu kendisini kovalayan ucaga atesetti

i i

Ayse sheACCherselfACC chaseREL planeDATre doPAST

Ayse red at the plane that chased her

a Hic kimse onukendisini elestirenleri sevmez

No p erson sheACCselfACC criticizeRELPLUACC likeNEGAOR

No one likes those who criticize himher

b Canan onukendisini elestirenleri sevmez

Canan sheACCselfACC criticizeRELPLUACC likeNEGAOR

Canan do esnt like those who criticize her

The distinction observed b etween a and b sentences in ab ovevanishes when the

pronouns are emb edded sub jects and p ossessive pronouns which cannot b e overt even if they are

coreferential with the higher sub ject as in the b sentences in Note that the p ostp osition

yanna next to him in ab is marked for agreement with its ob ject and its ob ject cannot b e

overt either

a Herkes onun k endi annesini sever

i i i

Everyb o dy sheGENself motherPOSSSGACC loveAOR

Everyb o dy loves hisher own mother

b Ali onun kendi annesini sever

i i i

Ali heGENself motherPOSSSGACC loveAOR

Ali loves his own mother

b Herkes onun kendisinin cok akll oldugunu sanyor

i i i

Everyone sheGENSGselfGENSG very clever b eNOMACC thinkPROG

Everyone thinks that she is very clever

b Ali onun kendisinin cok akll oldugunu sanyor

i i i

Ali heGENSGselfGENSG very clever b eNOMACC thinkPROG

Ali thinks that he is very clever

a Her adam onun kendi yanna bir masa koydu

i i i

Everyeach man heGENSself nextPOSSSGDAT one table putPAST

Everyeach man put a table next to him

b Ali onun kendi yanna bir masa koydu

i i i

Ali heGENSself nextPOSSSGDAT one table putPAST

Ali put a table next to him

tier Wehave seen that overt pronouns in Turkish cannot b e b ound within the scop e of a quan

Similar observations are made in other prodrop languages in Spanish Montalb etti Spanish

and Catalan Rigau Chinese Liejiong among others Montalb etti

shows that an overt pronoun in Spanish cannot b e construed as a b ound variable while a null

pronoun can Overt Pronoun Constraint OPC unless overt pronouns are in nonadjacent clauses

cf Montalb etti Chapter for details

Below we turn our attention to pronouns of laziness a term coined byGeach also known

as paychecksentences after Karttunens famous example given in

The man who gavehispaycheck to his wife was wiser than the man who gave it to his

mistress

The pronoun it in cannot b e construed as b ound b ecause its antecedent his paycheckis

not a quantied NP It cannot b e in a referent identity relationship with his paycheck either

b ecause the sp eaker intends to refer to a dierentpaycheckby it the second mans paycheck

Heim suggests that pronouns of laziness in English can b e explained by accom

mo dation For example in the denite pronoun it do es not have a corresp onding le card

already in the discourse mo del The hearer up on hearing it must create an auxiliary le card

with an entry of the Bible in this ro om

a Every motel ro om has a copy of the Bible in it

b In this ro om it was hidden under a pile of TV Guides

According to Heim from a psychological p oint of view accommo dation should

go through smo othly when it is sanctioned by a parallel precedent Heim argues that accommo

dation is more acceptable when it adds no new information to the le ie b oth the existing le

and the accommo dated card added to it have identical satisfaction sets There is an already exist

ing le for the Bible every motel ro om has The satisfaction conditions for the accommo dated

card are the same as for this already existing le card



Example shows that pronouns of laziness can b e null pronouns but overt pronouns are

ungrammatical

Maasn karsna veren adam

SalaryPOSSSgACC wifePOSSSgDATgiveREL man

The man who gives his salary to his wife

onu metresine veren adamdan daha aklldr

itACC mistressPOSSDAT giveREL manABL more smartCOPAOR

is smarter than the man who gives it to his mistress

In maas the salary in the second clause is the salary of the man who gives to his mistress

while the salary in the rst clause is the salary of the clever man Since the two salaries are

dierent they have dierent referents Since the pronoun and its antecedent are not coreferential

an overt pronoun is illicit in The same explanation holds for the example in Alis money

and Ahmets money do not denote the identical money and an overt pronoun is prohibited in

b

a Ali parasn bankaya yatrd

Ali moneyPOSSSgACC bankDAT dep ositPAST

Ali dep osited his money in the bank

b Ahmet onu evde saklyor

Ahmet itACC homeLOC keepPROG

Ahmet keeps it at home

To sum up since overt pronouns have to b e coreferential they cannot function as b ound

variables or pronouns of laziness in Turkish

Conclusion

This chapter investigated how discourse entities are evoked into the discourse mo del for subse

quent denite reference in Turkish Wehaveseencontexts where overt and null pronouns are

in complementary distribution an overt pronoun has to pick up the same referent denoted by

its corresp onding entity ie it has to uniquely identify and exhaust the maximal set mass or

category denoted Null pronouns on the other hand can have nonreferential antecedents and

can function as nonsp ecic indenite anaphora or as pronouns of laziness

It is observed that an initial expression can provide partial information with resp ect to its

value Value LoadedValue Free reading with resp ect to the referential versus nonreferential

statuses In addition wehave seen that bare ob jects can b e ambiguous as to whether or not

they have undergone incorp oration with the verb Sp eakers disambiguate an initial reading in the

subsequent discourse either by using denite reference or bydenying the existence of a referent



Note that null ob jects as pronouns of laziness in Turkish are sanctioned if there is a semantic link eg an

entailment relationship b etween the verbs of the two clauses cf Turan a

Thus initial mention of an expression mayconvey partial information ab out the sp eakers dis

course mo del which is up dated in subsequent utterances via a set of tacit instructions that leads

the hearer to devise a mo del that matches that of the sp eakers as much as p ossible

This chapter has lo cated the contexts where null pronouns cannot alternate with overt pro

nouns The rest of this study is devoted to denite anaphora where null and overt sub jects can

alternate as denite referential expressions

Chapter

CENTERING THEORY AND

OTHER APPROACHES TO

DISCOURSE ANAPHORA

Intro duction

In the previous chapter wesawhowanentitycanbeintro duced into the discourse mo del and then

can b e anaphorically referred to via a zero or overt pronoun in the subsequent discourse in Turkish

In the remaining chapters we will investigate anaphoric p ossibilitie s after an entity has b een

evoked and denite reference has b een sanctioned This involves an analysis of the interpretation

of denite full NPs and pronouns Centering Theory provides the theoretical framework for this

analysis The purp ose of this chapter is to intro duce the rules and constraints of Centering Theory

and to discuss the related researchdoneinFo cusing Theory In addition other approaches to

discourse anaphora Hobbs Fox Givon Ariel Kuno and Ha jicova are briey reviewed and

discussed

Centering Theory

Centering Theory has b een formulated by Grosz Joshi and Weinstein as a com

putationally tractable algorithm for resolving anaphora lo cally ie within a discourse segment

Centering Theory mo dels attentional state with resp ect to discourse entities during the incremen

tal construction of the discourse mo del Attentional state as was discussed in Chapter is a

part of sp eakers cognitive the cognitive system has a richer structure than attentional state and

includes at least world knowledge b eliefs desires and intentions Grosz and Sidner

Such a cognitiv e system is capable of complex inferencing Attempting to mo del this complex

inference system for pronoun resolution is not manageable at this stage However Joshi and Kuhn

and Joshi and Weinstein suggest that the inferencing required for discourse pro cess

ing and for anaphora resolution can b e constrained via centering notions These suggestions are

elab orated in Grosz Joshi and Weinstein

Centers of attention are discourse entities that serve to link an utterance to other utterances a

single sentence in isolation do es not haveacenterofattention Besides the same sentence uttered

in dierent discourse situations has dierentcenters Thus centers are not words phrases or

syntactic forms but rather they are semantic ob jects and discourse constructs that can b e realized

linguistically

The term realize intro duces a relation that relates centers to linguistic expressions As stated

in Grosz Joshi and Weinstein

A center c where c is a semantic interpretation is realized in utterance U if either c

is an element of the situation describ ed by the utterance U or c is directly realized by

some part of U

There are two basic typ es of centers Forwardlo oking centers and Backwardlo oking centers

Forwardlo oking centers Cf are a set of p otential antecedents for subsequent reference that are

evoked or realized in an utterance and they are partially ordered and listed in the Cflist on the

basis of salience assigned to them by a sp eaker EachForwardlo oking center in an utterance must

b e realized in that utterance One of these Forwardlo oking centers the preferred center Cp is

the most saliententity that is predicted to b e the b est candidate that will b e under discussion

in the next utterance Brennan Friedman and Pollard This formulation restricts a set of

p otential antecedents down to one

The Backwardlo oking center Cb in an utterance is the preferred center of the previous

utterance and it is the entity that the utterance is mostly centrally concerned ab out There is

precisely one Backwardlo oking center p er utterance and it serves as an anaphoric link to the

preceding adjacent utterance Discourse initial utterances do not haveBackwardlo oking centers

These constraints can b e summarized as follows

For each utterance U in a discourse segment UiUn

There is precisely one Backwardlo oking center Cb

Every entity in the Forwardlo oking center list CfUn must b e realized in utterance Un

The Backwardlo oking center CbUn is the highest ranked entity of the Cflist of the

immediately preceding utterance Un

Constraint states that every entity in the Cflist of an utterance must b e present in that

utterance A Cflist includes discourse entities that are realized with null and overt pronouns

full NPs and also with inferrable entities Prince a Walker Iida and Cote Prince and

Walker though centers can b e extended to prop ositions Nakatani Not all NPs in a

sentence contribute to the Cflist only those that evoke discourse entities can b e listed

Constraint predicts that the Cb of the next utterance is exp ected to b e the highest ranked

entity the preferred center in the Cflist of the current utterance

Centering Theory also has a set of transitions from one utterance to the next whichare

determined bytwo factors whether a particular Cb is also the Cp and whether or not the Cb of

the two consecutive utterances is the same ie whether the center of attention is the same entity

Centering transitions are represented in the following table Brennan Friedman and Pollard

and these transitions will b e exemplied in after Centering rules are presented

CbUnCbUn CbUnCbUn

CbUnCpUn Continue Smo othShift

CbUnCpUn Retain RoughShift

Centering Theory also p osits the following set of rules

If there is a pronoun in an utterance Un the Cb is also realized as a pronoun

The transitions are ordered with resp ect to pro cessing complexity as follows

Continue Retain Smo othShift RoughShift

Rule do es not preclude the existence of another pronoun in the utterance provided that the

Cb is realized with a pronoun This is not a hard rule but the violation is costly on the part of

the hearer which requires additional pro cessing complexity

The use of a full NP instead of a pronoun to realize a centered entitymay servevarious

functions other than merely referring and linking the two utterances with a referential link For

example in the mangy old b east as opp osed to a pronoun either conveys some additional and

new information ab out the entity Grosz Joshi and Weinstein or it represents a new

discourse segment Kameyama Grosz Gordon and Gilliom

a I to ok my dog to the vet the other day

b The mangy old b east

Grosz Joshi and Weinstein

Otherwise the use of full NPs in order to realize a centered entitywould yield an incoher

ent discourse Grosz Joshi and Weinstein Psychological research has also shown that

centers realized as pronouns are read faster than the full NPs see DZmura DZmura and

Tannenhaus Grosz Gordon and Gilliom

As far as Rule is concerned Grosz Joshi and Weinstein state that a sp eaker

maycho ose to CONTINUE discussing the same entity

mayRETAIN the same entity but also intro duce a new one

may SHIFT the attention to a new entity

Brennan Friedman and Pollard suggest that there are twotyp es of SHIFT Shifting

and Shifting which are later termed Smo oth Shift and Rough Shift resp ectiv elyinWalker Iida

and Cote

As has b een previously noted it is presumed that the transitions are not pro cessed with equal

ease We assume Continue is pro cessed with the greatest ease while Retain requires more eort

than Continue but less eort than Smo oth Shift which in turn is easier to pro cess than the

Rough Shift Brennan Friedman and Pollard Prince and Walker

The transitions are illustrated by the following examples Note that initial utterances haveno

Cbs The center is instantiated in the b utterances in

a Mike wanted to go the seaside

i

Cf Mike seaside

b He called Mary

i k

Cb Mike

Cf Mike Mary

c He asked her whether she wanted to join him

i i

Cb He Mike

Cf Mike Mary

CbUn CbUn

CbUn CpUn

CONTINUE

a Mike wanted to go the seaside

i

Cf Mike seaside

b He called Mary

i k

Cb He Mike

Cf Mike Mary

c Mary got angry at him

k i

Cb him Mike

Cf MaryMike

CbUn Cb Un

CbUn CpUn

RETAIN

a Mike wanted to go the seaside

i

Cf Mike seaside

b He called Mary

i k

Cb He Mike

Cf Mike Mary

c She has always enjoyed going to the seaside

k

Cb she Mary

Cf Mary

CbUn CbUn

CbUn CpUn

SMOOTHSHIFT

a Mike wanted to go the seaside

i

Cf Mike seaside

b He called Mary

i k

Cb He Mike

Cf Mike Mary

c Mark had called her b efore

k

Cb her Mary

Cf Mark Mary

CbUn Cb Un

CbUn CpUn

ROUGHSHIFT

Centering Theory is a part of linguistic comp etence involved in b oth the pro duction and

the resolution of discourse anaphora Kameyema Centering rules are constraints on

determining the antecedent and cho osing among various referential expressions in continuing to

refer to an entity or shifting to another one Linguistic comp etence concerning Centering rules

is somewhat dierent from Chomskys comp etence notion which considers sentences as absolute

theoretical constructs Centering rules are not absolutes but rather tendencies whichmaybe

overridden by pragmatic factors see Di Eugenio Walker Iida and Cote

Brennan among others

Backwardlo oking centers and Sentence Topics

In this section the relationship b etween Backwardlo oking centers and sentence topics will b e

briey discussed

Sentence topic has b een used in the literature to refer to various statuses of NPs Prague

scho ol linguists have prop osed that topic is what the sp eaker wants to talk ab out Thus topic is

in contrast with the linguistic notion of fo cus which adds new information to the topic We shall

return b elow to the discussion of fo cus when Fo cusing Theory and Ha jicovas pronoun resolution

algorithm are discussed As we shall see some linguists eg Givon use the term topic to refer

to discourse entities in general

The notion of Backwardlo oking center corresp onds more or less to Reinharts notion of

sentence topic According to Reinhart the topic is what the sentence is ab out Reinhart shows

the diculties of dening topic on the basis of sub jectho o d linear order or old information

Topics app ear in sub ject p osition in the unmarked case since it is easier to use an utterance

when weintend its sub ject to b e a topic Reinhart However not all sub jects are topics

The following example cited in Reinhart illustrate that the notion of topic or Cb cannot b e

dened by the test of sub jectho o d alone Note that although and are identical topic is Max

in b while it is Rosa in b

a Who did Max see yesterday

b Max saw Rosa

a Did anyone see Rosa

b Max saw Rosa

As wesaw in the preceding section Centering Theory has dened Backwardlo oking center as

the highest ranked entity in the previous utterance that is realized in the current utterance We

also saw that if anything is pronominalized the Backwardlo oking center must

Fo cusing as a Precursor to Centering Theory

Centering is a theory that has develop ed in part from Fo cusing Theory p ostulated by Grosz

and extended by Sidner Fo cusing is an algorithm formulated to account for the use and

interpretation of referential expressions with resp ect to attentional state

First however a warning is in order The term fo cus in Fo cusing Theory and in natural

language pro cessing in general contrasts with the notion of fo cus in the linguistics literature In

Fo cusing Theory fo cus is a status of referential expressions in the discourse mo del analogous

to centers in Centering Theory An entity in fo cus is the most activated entity ie the one

in consciousness or in the shortterm memory that the sp eakers are paying attention to In

the linguistics literature on the other hand the term fo cus has b een used to dene various

linguistic phenomena including the one enco ding a phonetically prominent element andor the

one enco ding and adding new information to the sentence Chomsky Chafe Vallduv

among others Vallduv denes linguistic fo cus as the part of the sentence that

enco des information the most dominant and contrarytoexp ectation part of the sentence In

Princes b analysis of fo cus fo cus is seen as a relation b etween the op en prop osition

which contains old ie predictable presupp osed information and what lls the variable in the

op en prop osition conveying the new information in that context

Thus the term fo cus has b een used to refer to distinct phenomena The term fo cus in Fo

cusing Theory do es not havemuch in common with the use of fo cus in linguistics but even more

confusingly it is analogous to the term topic whichcontrasts with fo cus in the linguistics liter

ature In this chapter the term fo cus is used in the sense intended in Fo cusing Theory unless

stated otherwise With this in mind wecannow turn our attention to Fo cusing Theory

Sp eakers consciously or unconsciously fo cus their attention on certain concepts or ob jects

throughout the discourse Fo cusing is an algorithm that keeps track of what the discourse is

ab out Attentional state includes a set of fo cus spaces and a set of transition rules mayadd

or delete these fo cus spaces Grosz identies twotyp es of fo cus global and immediate

Immediate fo cus is a lo cal phenomenon that describ es how a sp eakers fo cus of attention may

remain constant or shift in two successive utterances Global fo cus on the other hand mo dels the

sp eakers attention throughout the whole discourse Fo cusing only a limited numb er of discourse

entities at a time limits the search of the p ossible antecedents for a particular pronoun Discourse

entities that have b een fo cused are said to b e in the fo cus space Fo cus space is dierentfroma

discourse mo del although it is related The fo cus stack at a certain p oint in discourse contains

those entities relevant to the interpretation of the current utterance which includes a part of

discourse mo del A fo cus space is op en if the entity has b een mentioned recently that is if the

entity is still in fo cus Entities that have b een previously mentioned are in the fo cus space but

they are stacked All the entities are stacked according to their relative salience in the discourse

t in which they app eared segmen

Sidner p ostulates three mechanisms to resolve anaphoric expressions A current discourse

fo cus an ordered list of p otential fo ci for the next utterance and a fo cus stack that keeps the

current fo cus to b e returned to later within the discourse The list of p otential fo ci corresp onds to

the Cflist in Centering The Fo cusing algorithm p osits twotyp es of fo cus Actor Fo cus AF and

Discourse Fo cus DF in order to handle multiple pronouns in a single utterance Sidner

prop oses that the DF is the main fo cus that the sp eaker would tend to talk ab out anaphorically

Sidner states fo cusing takes the discourse fo cus as primary the discourse b eing what

the sp eaker is talking ab out so far while the actor fo cus is the lo cus of information ab out actions

in the discourse In this sense the Discourse Fo cus roughly corresp onds to Cb but is not identical

with it As has b een mentioned unliketwo fo ci ie AF and DF Centering p osits exactly one

Cb p er utterance Fo cusing like Centering is an algorithm restricting complex inferencing and

p osits a set of cognitively plausible and computationally tractable rules for pronoun resolution

Other Approaches to Discourse Anaphora

In this section some other approaches to discourse anaphora will b e reviewed and discussed

Hobbs

Hobbs develops a naive algorithm for nding the antecedents of pronouns in English He

incorp orates the wellundersto o d syntactic constraints of coreference and noncoreference relations

A surface parse tree divides the sentence into sub ject verb ob jects etc The algorithm starts

searching for antecedents to pronouns in the following way It b egins at the rst NP or S no de

immediately dominating the pronoun and searches all the branches b elow this no de in a leftto

right fashion If the pronoun cannot nd its antecedent in the S no de it searches for it in the

previous sentence in the text in order of recency the most recent rst in a lefttoright manner

When an NP no de is encountered it is prop osed as the antecedent Hobbs combines this with

selectional restrictions for example in a sentence He moved it the pronoun it must b e used

to refer to a moveable ob ject However Hobbs p ointsoutthatsuch an algorithm cannot detect

when the antecedent selection fails He then suggests a semantic approach on which he later

develops his coherence relations Hobbs semantic approach has four semantic op erations

a Detecting intersentence connectives which includes determining patterns of contrast cause

violated exp ectation temp oral succession paraphrase parallel and example For example

acontrast pattern b etween the previous and currentsentence is observed where the pred

icates of S and S are contradictory or lie at opp osite ends of some scale one pair of

corresp onding arguments of S and S are identical or the other pairs of corresp onding

arguments are similar

b Predicate interpretation includes making inferences by searching the lexicon For example

for the predicate increase in a sentence like The price of coee has increased one must

lo cate a scale with an asso ciated real or vertical orientation Thus one can infer that an

ward motion o ccurs in this scale up

c Knitting helps resolving more implicit anaphoric relationships as b etween inferrables and

their antecedents For example consider

a The b oywalked into the bank

b Moments later he was seen on its ro of

In the bank is represented by the entityXandtheitofitsby X The entire chain of

inference forces us to assume that X is a building The interpretation of ro of demands X

to b e a building to o Then the corresp onding arguments of X and X are merged As a

result the antecedent is identied as the bank

d Identifying entities which is an op eration that seeks to identify unidentied entities For a

denite NP the previous discourse is searched until the corresp onding entity is found for

an indenite a new entityisintro duced

Hobbs suggests that these four op erations recognize the structure and relationships b etween

sentences that are implicit in a discourse

Hobbs determines the coherence relations in a discourse and assumes that pronoun resolution

will followasabypro duct of these relations The role of inference in pronoun resolution is

undeniable However the present study follows another direction where inferencing is constrained

by a set of precise rules and only if these rules are overridden by pragmatic requirements do other

factors such as inferencing and coherence relations b ecome useful devices

Givon

Givon constructs a topic continuity device to accountforhow referential expressions are used

in discourse Givon denes topics as the individuals ab out whom the fate aairs doings trials

and tribulations are discussed in the discourse Topics are therefore most commonly

nominal Potentially he claims all nominal arguments are topics in that sense Givons notion

of topic do es not corresp ond to Reinharts notion of topic whichintuitively corresp onds more or

less to Cb but rather seems to corresp ond to Webb ers discourse entityGivon basically treats

Agent sub jects as topics The following topic accession hierarchy predicts the likeliho o d of the

topics b eing continued in the discourse Givon

AGENT DATIVEBENEFACTIVE PATIENT LOC INSTR MANNER

If the simple clause has an Agent argument then that will b e the sub ject

If the simple clause has no Agent but has a Dative Benefactive argument then that will

b e the sub ject

If the simple clause has no Agent nor DativeBenefactive but has a Patient argument then

that will b e the sub ject

According to Givon the main b ehavioral manifestation of imp ortant topics in the scale ab ove

is their continuity in discourse Continuity is observed by the recurrence or the frequency of

o ccurrence Topic continuity is measured by the following parameters

Index of referential distance or lo okback Measures the distance in terms of the number

of clauses b etween the currentmention of a topic and its previous mention Values range

from clause most continuous to an arbitrary numb er of clauses least continuous

Referential distance reects the idea that topics with a higher value of referential distance

are harder to identify

Index of p ersistence or decay Measures how long the current topic continues to b e mentioned

in the subsequent discourse Values again are expressed with the numb er of clauses from

to the numb er of the clauses in the discourse The idea b ehind decay is that more imp ortant

topics will p ersist longer

Index of p otential interference ambiguity Measures to what degree there are comp eting

antecedents within the discourse

Correlations are then established b etween linguistic forms and their degree of continuityin

terms of these three metrics The most continuous topics then will b e mentioned by zero anaphora

andor unstressed pronouns Less accessible or discontinuous topics will b e expressed by stressed

pronouns full descriptions cleft fo cus constructions in that order The least accessible topics will

b e expressed by referential indenite NPs Givon then explains this with the Iconicity Principle

The more descriptive surprising discontinuous or hard to pro cess a topic is the more co ding

material must b e assigned to it a Givons approach is devised to account for denite

anaphora as is Centering TheoryHowever it do es not incorp orate systematically dened precise

rules for a framework desired for a robust and explicit analysis of pronominalization Unlike

Centering Theory which allows transitions only across adjacent utterances within a lo cal discourse

segment Givon attempts to analyze the totality of discourse While analyzing pronominal rules

at the global discourse level will illuminate our understanding of discourse coherence Givon treats

discourse as a linear construct without anyinternal structure However since Grosz it has

b een known that discourse has a hierarchical structure that one discourse segmentcanembed a

subsegment and that what matters for pronoun resolution is the structure of discourse and not

how many clauses there are b etween the pronoun and its antecedent

Ariel

Ariels approach is similar to that of Givons Ariel provides a theory of Accessibili ty

where information which is not automatically accessible is stored as general knowledge Recent

linguistic material is more readily accessible than other familiar entities The form of referential

expressions serve as guidelines for retrieval

Ariel like Givon discusses the eect of distance in cho osing among the referential expressions

The general picture that emerges from counts restricted to anaphoric references is that pronouns

are predominantly used when the distances are short anaphoric demonstrativ es are used in cases

of intermediate distance and denite descriptions mostly refer backtoantecedents outside the

sentence they o ccur in when their antecedents are not even close by Ariel Ariel

again likeGivon do es not takeinto account hierarchical discourse structure or the dierent

degrees assigned to set of entities

Fox

Fox addresses the relationship b etween anaphor interpretation and discourse structure bas

ing her pronoun resolution approach on rhetorical structures as develop ed by Mann and Thomp

son She suggests that it is necessary to parse the discourse into its comp onent units based on

rhetorical structures b efore the relationship b etween anaphoric patterning and discourse struc

ture are understo o d In a rhetorical structure the smallest unit is a prop osition A prop osition

basically corresp onds to a clause However relative clauses and complement clauses in complex

sentences constitute single prop ositions together with their main clauses The reason Fox argues

is that relative and complement clauses are governed by principles of grammar rather than by

those of discourse However she treats sub ordinate adjunct clauses separately Prop ositions are

organized by rhetorical structures Rstructures An Rstructure containsanucleus the main

goals of the author and by adjuncts which provide supplemental information in the discourse

Anucleus consists of three higher level Rstructures issue list and narrate An issue structure

presents a claim and may include background and elab oration as adjuncts A list structure con

tains an unlimited numb er of prop ositions and no adjuncts A narrate structure describ es a set

of temp orally ordered nuclei Adjuncts might add information ab out reason condition purp ose

etc whichmay b e signaled by conjunctions like b ecause if in order to resp ectively Only en

tities within the active Rstructure can b e pronominalized otherwise a full NP must b e used

An active Rstructure means either that the previous prop osition contains the entity or in the

case of intervening prop ositions that the discourse segment should b e emb edded in a higher level

Rstructure In the latter case whichFox calls return p op the entity can b e p opp ed over bya

pronoun If the intervening RStructure is to o complex a return p op by a pronoun is imp ossible

and the entitymust b e reintro duced by a full NP

The approachtaken in this dissertation diers from that of Foxs in several ways First of

all discourse segmentation prior to resolution of pronouns is not necessarily required This may

merely b e a metho dological dierence but the purp ose of this study is to account for denite

anaphora in as a straightforward and computationally tractable manner as p ossible Secondlywe

h more complex pro cess than Foxwould have b elieve that discourse segmentation involves a muc

us b elieve as argued in Grosz and Sidner Nevertheless this study likeFoxs recognizes

the relevance of hierarchical discourse structure to pronominalization In this vein we shall see

that pronominalization can shed some lighttowards on identifying discourse segment b oundaries

although it is not a sucient indicator of discourse b oundaries Discourse segmentation is

signaled byvarious other factors as well such as tenseasp ect shift cue words and intonation as

well as pronominalization cf Grosz and Sidner Polanyi Ehrlich Webb er

among others

Kuno

Kuno p oints out that two opp osing suggestions have b een made in the previous literature on

what can serveasantecedents for zero pronouns in Japanese discourse Yoshimoto predicts

that only topicmarked NPs can serveasantecedents for zero pronouns whereas Kameyama

suggests that NPs that are not topicmarked can qualify as antecedents for zero pronouns

According to Kuno these suggestions are b oth correct but the two conicting views stem from

the fact that these two studies analyze two dierenttyp es of zero pronominals pseudozero

pronominals and realzero pronominals Pseudozero pronominals are those that have undergone

Kunos Pecking Order of Deletion Principle Delete less imp ortant information rst and more

imp ortant information last Kuno Kuno argues that the Pecking

Order of Deletion Principle is a crosslanguage discourse constraint that accounts for numerous

deletion facts in many languages including English Chinese Japanese Russian and Turkish

Therefore the twotyp es of pronouns he prop oses for Japanese can b e extended to other sub ject

and ob jectdrop languages According to Kuno pseudo zeropronouns are those that arise from

discourse deletion strategies in questionanswer pairs and in parallel structures with not criss

crossing allowed If the null pronouns are not in answers to questions or in parallel constructions

then they are realzero pronouns Kuno presents the following Japanese examples of a

realzero pronoun b and a pseudozero pronoun b

a Yamada wa syuran da

i

Yamada TOPIC violentwhen drunkis

Yamada gets violent when drunk

b Dakara dare mo paatii ni yobani

i

Therefore anybody partytoinvitenotPresent

Therefore nob o dy will invite him to parties

a Taro o ga Hanako ni Ziro o a gakkoo de syookaisiata no desu ka

i k m

NOM ACC to scho ol at intro duced QUEST

Did Taro o intro duce Hanako to Ziro o at scho ol

b Hai gakko o de syookaisiata no desu

i k m

intro duced Yes scho ol at

Yes he intro duced him to her at scho ol

In b the zeropronoun in the ob ject p osition is coreferential with the sub ject of the previous

utterance Only realzero pronouns can crisscross in this way On the other hand no criss

crossing can apply in b and the pronouns are deleted as required by the Pecking Order of

Deletion Principle and only the imp ortant information is retained along the lines of following

principle

Fo cusVerb Discourse Deletion Strategy

Retain the fo cus information and delete everything else If the fo cus is not the verb

of the sentence retain the verb also to yield a sentential answer

The immediately preverbal p osition is reserved for fo cus ie the most imp ortant information

Although Kuno in a fo otnote states that his notion of less imp ortant should not b e equated

with given information his analysis assumes that less imp ortant information roughly corresp onds to twotyp es of

given information prop osed by Prince Prince distinguish es twotyp es of shared knowledge which includes what

is represented by old and predictable information Prince states the twotyp es of shared knowledge are

represented by the most saliententity and what is merely given Kuno implicates that the most salient information

is the least imp ortant and sanctioned to b e null Nevertheless some given information can b e imp ortantin

Kunos terms provided that it receives phonological stress eg some strong pronouns Any element with phonetic

prominence cannot b e deleted regardless of its information status Therefore Kunos notion of less imp ortant

information is represented by an element that do es not have phonetic or informational fo cus Crucially the term

deletion should not b e literally taken to mean that the constituents are actually deleted but rather that they are

not phonological l y sp elt out and are recoverable from the discourse

in Japanese and Turkish unless the verb is fo cus itself For Turkish see Erguvanl Erku

Homan

In Japanese as wellasinTurkish and many other languages the word order of elements in a

sentence is conditioned by the following from oldtonewinformation principle

FlowofInformation Principle

Elements in a sentence are ordinarily arranged in the order of less imp ortant informa

tion rst and more imp ortant information last

According to these rules and constraints everything except the phonologically fo cused con

stituent and verb can b e deleted in yesno questions and parallel constructions the deletion of

the fo cused element violates the Pecking Order of Deletion Principle

In the Japanese example in Kuno Kuno suggests that Sp eaker B do es not delete

merely the ob ject French but the adverb hard as well The answer is not interpreted as Hanako

is studying French but rather as Hanako is studying French hard Kuno claims that if a zero

ob ject is stipulated a zero adverb also needs to b e stipulated He argues that this is not the case

and that it is more plausible that French hard is deleted in one swo op via the application of

acrosstheb oard discourse deletion

A Dare ga issyookenmei huransugo o b enkyo osite imasu ka

who hard French study is QUEST

Who is studying French hard

B Hanakogabenkyo osite imasu

studying is

Hanako is studying it hard

Since only answers to yesno questions and parallel structures can have pseudozero pronouns

Kuno suggests pseudozero pronouns and realzero pronouns are distinguished by the following

criteria

Identication of Pseudo Zero Pronouns

Pseudozero pronouns must follow the same order and same syntactic function as their

ecking Order of Deletion Principle source NPs They are p enalized if they violate the P

Identication of Real Zero Pronouns

The antecedent of a zero pronoun must b e an unpronominalized NP that is overtly marked

as a topic or an NP whose referentisintro duced into the discourse scene discourse mo del

Penalty for RealZero Pronouns with Nontopic Antecedents

Realzero pronouns with nontopic antecedents receive p enalties inversely prop ortionate to

their p osition in the syntactic hierarchy

Syntactic Hierarchy Sub ject Ob ject

Penalties Lower Higher

Nonfo cus Constraint on Realzero pronouns

A realzero pronoun must b e a nonfo cus element in the sentence that it app ears in

Crisscrossing Constraint on RealZero Pronouns

A sentence that involves a crisscrossing realzero pronoun is unacceptable less the verb

signals the crisscrossing relationship

Kuno then computes discourse coherence numerically by assigning p enalties using the condi

tions ab ove The p enalty scores determine whether pronouns are instances of real or pseudozero

pronominals For example the preferred interpretation in ambiguous c is predicted by deter

mining whichinterpretation has the lowest p enalty score

a Dare ga Rosai o matte iru no desu ka

who waiting is QUEST

Whoiswaiting for Rosa

b Mary ga matte iru no desu

i

is waiting for

Mary is waiting for her

c yuusyoku ni syo otaisita no desu

i k

dinner to invited

Strong preference Rosa invited Mary to dinner

Weak preference Mary invited Rosa to dinner

A Assumption Rosa wa Mary o

i k

Sentence interpretation Mary invited Rosa to dinner

and are pseudozero pronouns in c

i j

Pecking Order of Deletion Principle is observed the imp ortant information is retained

Total Penalty Score

B Assumption Rosa wa Mary o

i k

and are realzero pronouns in c

i k

Sentence interpretation Rosa invited Mary to dinner

The antecedentof is topicmarked no p enalty

i

The antecedentof is not topicmarked p enalty score

k

Total Penalty score

C Assumption Mary wa k Rosa o

i

and are realzero pronouns in c

i k

Sentence interpretation Mary invited Rosa to dinner

The antecedentof is topicmarked no p enalty

i

The antecedentof is a nontopic nonsub ject NP p enalty

k

Crisscrossing constraint on RealZero pronouns violated p enalty

Parallel Interpretation Preference p enalty

Total p enalty score

The p enalty scores in A C predict that the pronouns in c are pseudozero pronouns

b ecause the p enalty score for this assumption is If they are considered to b e realzero pronouns

they are p enalized as in B byapenalty score of In addition it is predicted that the

interpretation assumed in C yields an incoherent discourse in comparison to A and B

since it has the highest p enalty score

If Kunos distinction b etween pseudozero pronouns and realzero pronouns is correct Turk

ish also has pseudozero pronouns b ecause Pecking Order of Deletion also applies in Turkish

Kuno in his survey of the Pecking Order of Deletion Principle includes Turkish He

p oints out that the question particle m and its phonological variants that have undergone Vowel

Harmony in Turkish follow the informationally fo cused constituent as in Kuno

a Sen mi bunu Paristen aldn

you QUEST thisACC ParisABL buyPASTSG

Did you buy this from Paris

b Sen bunu mu Paristen aldn

You thisACC QUEST ParisABL buyPASTSG

Was it this that you b oughtfromParis

c Sen bunu Paristen mi aldn

You thisACC ParisABL QUEST buyPASTSG

Was it in Paris that you b ought this

d Sen bunu Paristen aldn m

You thisACC ParisABL buyP ASTSG QUEST

Did you buy or did you steal this in Paris

In an answer to a yesno question as in the fo cused constituent cannot b e null but

everything else can b e deleted Kuno writes since the interrogative m is attached

to the rightofParis from Parisin Paris in A is in informational fo cus of the question

As exp ected all the answers shown in B which delete this fo cus but retain the verb are

unacceptable as answers

A Sen bunu Paristen mi aldn

You thisACC ParisABL QUEST buyPASTSG

Was it in Paris that you b ought this

B a Evet b en bunu Paristen aldm

Yes I thisACC ParisABL buyPASTSG

YesIboughtitinParis

b Evet b en Paristen aldm

Yes I ParisABL buyPASTSG

Yes I b oughtitinParis

c Evet bunu Paristen aldm

Yes thisACC ParisABL buyPASTSG

Yes I b ought this in Paris

d Evet Paristen aldm

Yes ParisABL buyPASTSG

Yes I b oughtitinParis

e Evet b en bunu aldm

buyPASTSG Yes I thisACC

Yes I b ought this

As Kuno states the most natural answer to this question is Bd This answer retains Paris

the most imp ortant information in Kunos terms and it also retains the verb in order to giveits

sentential status According to Kuno b oth the sub ject and the ob ject are thus pseudozero

pronouns in this example since they have undergone the Pecking Order of Deletion Principle

Ha jicova

Ha jicova develops a pronoun resolution algorithm that is related to but distinct in crucial

ways from Fo cusing Theory One ma jor distinction is that Ha jicova attempts to combine the

linguistic and natural language pro cessing notions of fo cus Wehaveseenabove that the notion

of fo cus in Sidners system roughly corresp onds to the opp osite notion ie the notion of topic

in linguistics Ha jicovas pronoun resolution algorithm is based on the Prague scho ol notions of

topic and fo cus Fo cus is roughly dened as a constituent ie verb or one of its arguments that

conveys new information ie contextually nonb ound Ha jicova Those constituents

that do not provide new information b elong to the topic part of a sentence Ha jicova

states it as follows where fo cusL is linguistic fo cus new information and fo cusAI is the fo cus

in Fo cusing Theory AI for Articial Intelligence

a It is rather obvious that the items referred to by the parts of the fo cusL of the

immediately preceding sentence are the most activated ie salient ones at every time

p oint of the discourse

b If an item is referred to in the topic part of the sentence then at least two issues

are to b e taken into consideration

i A pronominal reference seems to strengthen the activation of the item referred

to a lesser degree than a reference with a full denite NP

ii The activation of the items referred to in the topic part of the sentence seems to

fade away less quickly than that of the items referred to in the fo cusL part of the

sentence

c If the degree of activation of an item x is b eing changed lowered or raised then

also the degree of activation of the items asso ciated with the ob ject referred to byx

is b eing changed in the resp ective direction It should b e taken into consideration

however that frequently a mentioning of a particular ob ject brings into the foreground

only a fraction of a set of ob jects that has b een activated earlier Also other scales and

hierarchies should b e considered thus there is a hierarchy of more or less asso ciative

relationships or that of prominence with regard to individual sentences and their

p ositions in the text

d If an item of the sto ckofsharedknowledge is neither referred to in the given

utterance nor included among the asso ciated ob jects then its activation lowers down

Ha jicova compares these suggestions with those of Fo cusing Theory In Fo cusing Theory

sp eakers have options of maintaining the fo cus changing it to another immediate fo cus or return

to another fo cus that is not in the immediate list of fo ci but that is in the stack These are

summarized b elow

a the immediate fo cus of a sentence is the CurrentFo cus CF

b there is partially ordered stack of fo ci Partial Fo ci List PFL

c there is fo cus stack where immediate fo ci are listed

Ha jicova p oints out the similarities of these transitions with her suggestions According to

tities are those that are the linguistic fo cus of the immediately preceding her the most salienten

utterance The next most saliententities are those that have topic status Finally inferrables

that are in a logical relationship with the topic in the immediately preceding utterance or other

entities that are in other preceding utterances are the next most salient

In Ha jicovas algorithm each utterance in a discourse is assigned numb ers as in where the

utterances with the same numb ers are those that are ab out the same entities The capital letters

denote the phonological fo cus of each clause

The scho ol garden was full of CHILDREN

a They talked NOISILY

b but the teachers didnt REPROVE them

c b ecause they weresoEXCITED

Outside PARENTS were waiting

a One of them a father sto o d in front of a MICROPHONE

b as if he were prepare to TALK

In the children is the entity that is predicted to b e talked ab out in the next utterance

b ecause it is the most saliententity with the intonation center and as predicted this entityis

talked ab out in the following utterance In b the new entity the teachers is intro duced but

it is not a brandnew entity b ecause it has a logical relationship with the scho ol and thus is

inferrable in the sense of Prince a Due to this familiarity the NP is denite In b the

entity the children are still available and is also talked ab out anaphorically with the pronoun they

In another inferrable entity ie parents is intro duced in the fo cus of the sentence Since

the other entities the teachers and the children are not mentioned in this utterance they b ecome

less salient for subsequent reference In Ha jicovas algorithm eachentity is assigned a number

that indicates the salience of that entity starting from zero which represents the highest ranked

entity ie the one in fo cus Anaphora tracking is represented in the discourse schematically by

w fading of discourse entities and their salience following the intro duction and slo

In sum Ha jicovas algorithm is similar to Fo cusing Theory and Centering in that all three

approaches show the interaction b etween discourse coherence and the degrees of salience of the

discourse entities The ma jor dierence in Ha jicovas algorithm is that the informationally fo cused

entity is the highest ranked The Centerin and Fo cusing Theory algorithms determine saliency

using factors that are often asso ciated with topicho o d such as pronominalization sub jectho o d

and agentho o d The fo cusing and Ha jicovas approaches stack the previous entities that are not

in the immediately preceding utterance while Centering do es not The latter describ es transitions

between adjacent utterances in order to capture local coherence in discourse

Conclusion

This chapter has intro duced the constraints and rules of Centering Theory and its precursor

Fo cusing Theory In addition some other accounts of discourse anaphora eg Hobbs Fox

Givon Ariel Kuno Ha jicova have b een reviewed and compared to Centering

Chapter

RANKING FORWARDLOOKING

CENTERS IN TURKISH

Forwardlo oking centers in an utterance are ranked according to the degree of salience assigned

to them Grosz Joshi and Weinstein prop ose that the Cflist is partially

determined by the grammatical conguration hierarchy the sub ject ranking higher than ob jects

This view has b een widely accepted in the subsequent studies done on Japanese English Italian

Yiddish and Turkish Kameyama DZmura Di Eugenio Prince

Homan among others Kameyama prop oses and Walker Iida and Cote

agree that in addition to the role of grammatical hierarchytwo sp ecial devices in Japanese

grammar contribute to the salience of an entity grammaticized topics and empathy see b elow

for the notion of Empathy Due to this observation Walker Iida and Cote and Cote

claim that ranking forwardlo oking centers is determined by languagesp ecic factors Thus

Rambow suggests that ranking in German might follow the surface order p osition as will b e

discussed b elow Likewise a series of exp eriments conducted by Gordon Grosz and Gilliom

suggest that sentenceinitial p osition seems to contribute to salience even of nonsub jects Sidner

has suggested that thematic relations are used for determining the ranking The purp ose of

this chapter is to examine these suggestions and to investigate the factors that determine ranking

Cflists in Turkish The chapter is organized as follows In Section the background assumptions

are presented Section addresses the role of sentence p osition as prop osed byRambow

and Grosz Gordon and Gilliom The implications of these studies for Turkish discourse will

b e explored In Section the role of grammatical relations will b e discussed and evidence for

and exceptions to this analysis will b e given The exceptions involve the b ehavior of Exp eriencer

ob jects of psychverbs In relation to these ob jects the role of Empathy and thematic relations

will b e discussed in some detail Then ranking in certain p ossessive constructions in Turkish

discourse will b e addressed Section presents a discussion on the relationship b etween Cf

ranking p oint of view and discourse segmentation In Section some problems in the analysis

of naturallyo ccurring data will b e presented and some solutions will b e prop osed Finally Section

summarizes conclusions drawn and suggests areas for further research

Background Assumptions

In this dissertation the following assumptions are held based on Centering Theory Cf Prince

b Kameyama Kerslake Walker Iida Cote Brennan

among others

There is precisely one backwardlo oking center Cb p er utterance

a The CbUn is the highest ranked entity of the CbUn

b If anything is a null pronoun the Cb is a null pronoun

c Otherwise an overt pronoun or a full NP can b e the Cb

The choice of referential expression is not arbitrary but dep ends on the degree of salience

which the antecedent in the Cflist is assigned by the sp eaker Thus

a The least explicit expression ie anull pronoun otherwise an overt pronoun is

reserved for the most saliententity

b The most explicit expression ie full NP is reserved for a less saliententityInother

words the explicitness of an expression is inversely prop ortional to the salience assigned

to its antecedent

The Role of Surface Position

Below the arguments made in Rambow and Gordon Grosz Gilliom will b e presented

and their implications for Turkish discourse discussed

Rambow

Rambow suggests that fronted entities b etween the nite and nonnite verbs in German

aect their discourse salience German is an underlyingly verbnal language in addition to having

aVerbsecond V constraint in ro ot clauses That is in ro ot clauses the nite auxiliary or

main verb moves to the second p osition This divides the clause into two parts the p osition

in fron t of the nite verb theVorfeld VF and the p osition b etween the nite and nonnite

verbs the Mittelfeld MF The VF must contain precisely one constituent which can b e any

argument adjunct or nonnite verb The MF contains the remaining arguments and adjuncts

of the sentence Rambow argues that the movement of a single constituent to the VF is an

instance of Topicalization and the movementofanynumb er of constituents within the MF is

scrambling Rambow states that it is generally assumed for German that the ordering in the

MF is determined by the theme rheme ie topiccomment status of the arguments An NP

may scramble over less rhematic NPs Rambow argues that the theme corresp onds to the Cb

which links an utterance to the previous discourse and is what the sentence is ab out However

the ordered Cflist is not represented by the theme rheme hierarchy but helps to determine

the theme rheme hierarchy of the next utterance Thus centers of Un that are not realized in

Un are more rhematic than any of those that are Under this condition Rambow prop oses the

following

NP A in Un that expresses a memb er of CfUn may scramble b eyond an NP B if

one of the following two conditions holds

B do es not express a memb er of CfUn

B expresses a memb er of CfUn that is ranked lower than A

Rambow prop oses that the Cf of an utterance is the list of constituents of the MF ranked in

their order of o ccurrence He indicates that this ordering holds in the following examples where

the pronoun sie she in B and B is ambiguous German is a language with grammaticized gender

and b oth eine solche Massnahme such a measure and russischen Wirtshaft Russian economy

should b e realized with the thirdp erson feminine pronoun sie

A Glaub en Sie dass eine solche Massnahme der russischen Wirtshaft helfen kann

i

Think you that a such measure the Russian economy help can

Do you think that such a measure can help the Russian economy

B Nein sie ist viel zu primitiv

i

No she is much to o primitive

No it she is much to o primitive

A Glaub en Sie dass der russischen Wirtshaft eine solc he Massnahme helfen kann

i

B Nein sie ist viel zu primitiv

i

In B and B the ambiguous pronoun sie she refers to the rst entity in the MF eine solche

Massnahme such a measure and russischen Wirtshaft Russian economy resp ectively in A and

A If the arguments of a sentence are ordered according to their salience hierarchy from left to

right as suggested by Rambow then the prep osed nonsub ject entity in A ranks higher

than the grammatically most saliententity ie sub ject in the Cflist and this predicts that the

Cb sie she in B is the highest ranked entity of the previous utterance Thus the readings in

B and B can b e predicted if the Cfs in MF are ranked from left to right in German

This suggestion made for German mayhave implications for ranking the Cfs in Turkish

Turkish whichisanSOV language allows scrambling of each argument of the verb to various

p ositions dep ending on appropriate discourse conditions The rich case system of Turkish iden

ties the predicateargument structure in a sentence The sentenceinitial p osition in Turkish

like in German is the theme p osition where discourseold information is represented whereas

the preverbal slot is reserved for new unpresupp osed and or phonetically prominent information

cfErguvanl Erku Homan In prep Turkish diers from German in two relevant

ways Turkish do es not haveaverbsecond constraint and it is a prodrop language

There have b een some preliminary studies of the eects of word order on Centering in Turkish

In these studies Homan and Turan and Homan show that Cbs tend to app ear in

sentenceinitial p ositions In their data in SOV sentences the sub ject tends to b e the Cb

of the time while in OSV sentences the ob ject is the Cb of the time Homan and Turan

Since sentenceinitial p osition in Turkish is reserved for an argument or adjunct to link

it to the previous discourse it is not surprising that Cbs tend to b e realized in sentence initial

p osition However Homan observes that a prep osed nonsub ject entity do es not necessarily

rank higher Although prep osed ob jects tend to b e the Cb of an utterance sp eakers usually do

not continue talking ab out them in the next utterance They do not tend to function as Cps In

her data prep osed or p ostp osed nonsub jects do not signicantly aect the Cfranking

In the following some examples from Turkish narratives will b e discussed to shed lighton

whether prep osed nonsub jects in Turkish rank higher in the Cf list Consider

a Heyetimizi Ali karslad

i k

DelegationPOSSPLU Ali meetPAST

Ali met our delegation

bve heyeti Basbakana goturd u

k i

and delegation Prime MinisterDATtakePAST

k

He to ok them to the Prime Minister

c onlar Ankarada misar etti

k i

theyACC AnkaraLOC guest domakePAST

k

He hosted them in Ankara

Zubuk Aziz Nesin

In a the direct ob ject heyetimizi our delegation is in sentenceinitial p osition According

to Rambows surface p osition hyp othesis it should b e the most saliententity in the Cfhierarchy

The Cb in b is realized with a null sub ject and is anaphorically related to the noninitial sub ject

of the previous utterance On the other hand the full NP in ob ject p osition in b realizes the

sentenceinitial ob ject In accordance with the assumptions given ab ove the noninitial sub ject

in a must b e ranked higher than the initial ob ject

In the prep osed direct ob ject in a is realized with a full NP when used in sub ject p osition

in b b efore the sp eaker continues to talk ab out that entity in the subsequent discourse cd

This strategy as will b e discussed later is common in Turkish discourse The sp eaker has to

makeanentity salientby bringing it to the center of attention b efore she can refer to it bya

null sub ject cf Brennan who discusses a similar strategy in English This indicates that

the sentenceinitial p osition do es not contribute to the salience of a nonsub ject

a Umiti bile unutmuyor Birsen

i k

UmitACC even forgetNEGPROG Birsen

Birsen do esnt even forget Umit

bUmit Amerikaya gitmis

i

Umit AmerikaDAT goPAST

Umit went to America

c milyoner olmus

i

millionaire b ecomePAST

i

She b ecame a millionaire

d Birsenle dugunum uze milyoner olarak geliyor

i

BirsenWITH weddingPOSSPLUDAT millionaire b ecoming comePROG

i

She comes to our wedding as a millionaire

Selim Ileri Cehennem Kralicesi

Now consider and designed to b e parallel to Rambows examples The sub ject whether

in noninitial p osition or not ranks higher than the ob ject in Turkish In b oth B and B the

null sub ject is interpreted as the sub ject of the previous utterance

aSon karar ekonomiyi d uzeltir mi

i

last decision economyACC improvePRES QUEST

Will the last decision improve the economy

bHayr cok zayf

i

No very weak

i

No its very weak

aEkonomiyi son karar duzeltir mi

i

economyACC last decision improvePRES QUEST

Will the last decision improve the economy

bHayr cok zayf

i

No very weak

i

No its very weak

Thus the set of examples discussed in this section suggests that sentenceinitial p osition do es

not increase salience in Turkish Wenow turn to Gordon Grosz and Gilliom

Gordon Grosz and Gilliom

Gordon Grosz and Gilliom also explore the function of sentenceinitial p osition as a p oten

tial factor in contributing to the salience of an entity in the Cflist In a set of psycholinguistic

exp eriments they test discourses as follows

a Susan gaveFred a hamster

b In his opinion she shouldnt have done that

These utterances precede the following two alternative sets of discourses

c SheSusan just assumed that anyone would love a hamster

d Giving a p et as a gift can b e somewhat of an imp osition

c HeFred do esnt haveanywhere to put a hamster cage

d Giving a p et as a gift can b e somewhat of an imp osition

In these discourses the third utterances are critical Gordon Grosz and Gilliom test whether

the rep eated name in the third utterance renders the discourse less coherent The discourse

is less coherent in their view if the utterance with a full NP takes longer to pro cess than the

utterance with a pronoun ie the Rep eated Name Penalty The decreased coherence stems from

the violation of the following centering rules

The Cb Un is the most saliententity in Un

The Cb must b e realized with a pronoun rather than a full NP

Rule A is the standard denition of Cb in Centering Theory as has b een previously discussed

Rule B is a result obtained by Gordon Grosz and Gilliom in a set of psycholinguistic

exp eriments Thus their hyp othesis is as follows if the entity realized with a full NP requires a

longer pro cessing time it must not b e very salient and therefore must not b e the Cb The Cb is

the most saliententity in the previous utterance

Their results show that if the Cb is realized with a full NP the critical sentences takelonger

to pro cess Their results show a Rep eated Name Penalty if the nonsub ject in sentenceinitial

p osition in Un is realized with a full NP in Un A similar eect is observed for the entity realized

as the noninitial sub ject in Un Gordon Grosz and Gilliom conclude that the sentenceinitial

p osition contributes to salience

In the data given by Gordon Grosz and Gilliom are translated into Turkish and relevant

cultural and linguistic changes are made to make the discourses coherent

aSuzan Ferdiye bir kedi verdi

k i

Susan FredDAT one cat givePAST

Susan gaveacattoFred

bFerdiye gore Suzan hata yapt

i k

FerdiDAT according Susan mistakemakePAST

According to Fred Susan made a mistake

c kedileri hic sevmez

i

catPLUACC never likeNEGAOR

i

He do esnt like cats

d Birisine hediye secmek cok zor

oneDAT gift selectINF very hard

It is very hard to select a gift for someone

c herkesin kedileri sevecegini sand

k

everyGENSG catPLUACC loveFUTPOSS thinkPAST

k

She thoughteveryb o dy would like cats

d Birisine hediye secmek cok zor

oneDAT gift selectINF very hard

It is very hard to select a gift for someone

Note that an initial nonsub ject and a noninitial sub ject can b e realized with null pronouns

in the subsequent utterances in c and c This indicates that b oth the sub ject and the sentence

initial entity are salient Compare b with b where Fredisinsentencenal p osition

aSuzan Ferdiye bir kedi verdi

k i

Susan FredDAT one cat givePAST

Susan gaveacattoFred

bHata yapt Suzan Ferdiye gore

k i

mistake makePAST Susan FredDAT according

Susan made a mistake according to Fred

c kedileri hic sevmez

i

catPLUACC never likeNEGAOR

i

He do esnt like cats

In c the null sub ject still realizes the entityinsentencenal p osition in b This suggests

that it is not the sentence p osition that contributes to the salience in Turkish The equal salience

of the sub ject and the nonsub ject entities b oth in English and in Turkish stems from the fact

that the b sentences are from Freds p oint of view and b oth Susan and Fred are equally salient

from Freds p ersp ective cf section b elow

In conclusion wehave seen that the discussion of Turkish data in this section has not provided

evidence for the sentenceinitial p osition contributing to the salience of entities

The Role of Grammatical Relations

Although it is accepted that many factors gure in determining the Cfranking Grosz Joshi

and Weinstein Kameyama and Brennan Friedman and Pollard argue

that sub jectho o d is a strong indicator of salience They claim that the ranking in English is

determined by grammatical roles

Sub ject Ob jects others

Likewise Kameyama and Walker Iida and Cote and Iida ar

gue that grammatical role in Japanese discourse is signicant in determining salience Japanese

has additional languagesp ecic factors contributing to salience such as the grammaticized topic

marked bywa and Empathy asso ciated with verbs for giving and receiving Each predicate in

Japanese selects one of its arguments as the lo cus of Empathytheargument p osition whose

referents view p oint the sp eaker identies with in varying degrees Kameyama The

unmarked and marked empathy lo ci are the sub ject and the indirect ob ject resp ectively

As in English and Japanese the ma jority of the Turkish data can b e accounted for if sub jects

are ranked higher than others An example is given in The null and overt pronouns in

b resp ectively realize the sub ject and the ob ject of the previous utterance The ob ject in

Un cannot b e realized with a null sub ject in Un if the sub ject is realized with a more explicit

expression as indicated in b and b On the other hand the sequence of utterances ab

is felicitous when the ob ject in Un is realized with a full NP sub ject in Un These coherence

eects stem from the fact that sub jects are more salient than ob jects

aAli Murat davet etti

i k

Ali MuratACC invitation domakePAST

Ali invited Murat

b ona icki ikram etti

i k

HeDAT drink serving domakePAST

i

He served him a drink

b ona icki ikram etti

k i

b Aliyeicki ikram etti

k

Intended reading He served Ali a drink

b Murat ona icki ikram etti

k i

Murat served him a drink

Likewise the ob ject in b is rep eated as a full NP in c while the sub ject can b e a null

pronoun The null sub ject signals the most saliententity in the previous utterance

a Elindeki raketi mindere att

i

handPOSS racketACC cushionDAT throwPAST

i

He threw the racket that was in his hand on the cushion

b Murata dogru ilerledi

i k

MuratDATtoward advancePAST

i

He advanced towards Murat

c delikanly kucaklad

i k

youngmanACC embracePAST

i

He embraced the young man

d Sevincinden neredeyse aglayacakt

i

joyPOSSABL whereCOND cryFUTUREPAST

i

He almost cried from happiness

Kemal Bilbasar Baska Olur Agalarn Du gunu

The Center Promotion Rule for Turkish

In the following discussion more evidence for the salience of sub jects is presented and a Centering

rule is intro duced whic h is used extensively in Turkish discourse illustrated in

As has b een previously stated if an entityisevoked in ob ject p osition it cannot b e referred

to with a null sub ject unless it is brought to the center of attention Notice that hamuru paste

in b Murata to Murat in a and kitabi theb o ok in b crucially ha vetobementioned

as a full NP in their resp ective utterances in the sub ject p osition Only then can they b e realized

with a null sub ject

aFatma Bac guldu

i

Fatma Baci smilePAST

Fatma Baci smiled

b Boynuna svanan hamuru silkeledi

i k

neckPOSS coverREL pasteACC shakePAST

i

She sho ok of the paste that was on her neck

cHamur gitti

k

paste goPAST

The paste went and

d Hatcann gozune yapst

k

HGen eyePOSSDATstickPAST

k

It stuckonHatcas eye

Kemal Bilbasar Baska Olur Agalarn Du gunu

a Murata bakracla ayran getirdi

i k

MuratDAT bucketwith yogurtdrink bringPAST

i

She broughta yogurt drink in a bucket for Murat

bMurat bakrac ald

k

Murat bucketACC takePAST

Murat to ok the bucket

c Kzn gozlerine bakarak agr agr icti

k i

girlGEN eyePLUPOSSDAT lo okADV slow slow drinkPAST

k

He drank it slowly lo oking at the girls eyes

Kemal Bilbasar Baska Olur Agalarn Du gunu

a Her bir parcasn yrtabilsin diye

i

each one partPOSSACC tearABILSUBJUS sothat

i

So that he could tear every single piece of it

a bolmeye calst kitab u c parcay

k

b o okACC three partDAT divideINFDAT tryPAST

he tried to divide the b o ok into three pieces

bAma kitap ciltliydi

k

but book b oundCOPPAST

But the b o ok was b ound

cKaln bir kapag vard

k

thickonecoverPOSS havePAST

It had a thickcover

d ve arkas iplikle dikilmisti

k

and backPOSS threadWITH sewPASSPASTPAST

and its backwas sewn with thread

Cicekoglu Ucurtmay Vurmasnlar

The same rule do es not apply if an entityisevoked and realized in the ob ject p osition in b oth

utterances for example the rep ort in

i

a Ceketinin cebinde sakladg rap oru ckard

i

jacketPOSS p o cketGEN keepREL rep ortACC takePAST

ept in the p o cket of his jacket He to ok out the rep ort he k

b Bulente uzatt

i

BulentDAT givePAST

He gaveittoBulent

Basargan Gurbet Sofras

Compare b with the following variant where we cannot refer to the rep ort with a null

sub ject b ecause the rep ort was intro duced in the ob ject p osition in a and is not the center of

attention

b Rap or cok uzundu

i i

rep ort very longCOPPAST

i

The rep ort was very long

The data indicate that sub jects and ob jects are not assigned equal salience in Turkish dis

course Based on the observations ab ove the following Center Promotion Rule is prop osed

Center Promotion Rule for Turkish CPR rst approximation

If an entitye isevoked in ob ject p osition in Un e cannot b e realized with a null

sub ject in Un

So far wehave discussed the b ehavior of ob jects which are not Cbs In the ob ject is also

the Cb and thus it can b e continued byanull pronoun in the sub ject p osition in c

aValiyle goru smesi ogleden sonrayakald

mayorWITH meetingPOSS afterno on after remainPAST

His meeting with the mayor was postponed until the afterno on

bVali onu saat u ce dogru cagrd

i

mayor heACC clo ck threeDATtowards callPAST

The mayor called him ab out three oclo c k

c Ko caman b oyuyla Valinin onunde bir kackez egildi

i

big heightPOSSWITH mayorGEN frontPOSSLOC one few times b endPAST

i

He b owed in front of the mayor a couple of times

Yasar Kemal DemircilerCar ss Cinayeti

Thus instantiated Cbs do not undergo the Center Promotion Rule As a result the rule needs

to b e revised as follows

Center Promotion Rule for Turkish CPR

If an entity e is evoked in ob ject p osition and is not the Cb in Un e cannot b e

realized with a null sub ject in Un

Psychological Verbs

Although entities in sub ject p osition in Turkish seem to b e more salient in most cases we shall see

b elow that the ob jects of some psychological verbs rank higher than the sub jects Furthermore

in certain typ es of p ossessive constructions the p ossessors are more salient than the entityevoked

by the p ossessed NPThus the discussion b elow will showthat

a Not all sub jects are equally salient

b Some ob jects are more salient than sub jects

After the facts in Turkish is presented related researchinFo cusing and Centering Theories will

b e discussed After these discussions we shall also see that in p ossessive constructions indicating

the cognitiveorpsychological state of the p ossessor or inalienable p ossession the p ossessor noun

is more salient than the entityevoked bytheNPFor instance in an NP like Johns mind

i k

John ranks higher than Johns mind

In the Exp eriencer ob jects in a utterances rank higher than the sub jects These

ob jects do not undergo the Center Promotion Rule and they are immediately realized with null

pronouns The sub jects on the other hand are rep eated as full NPs in the successive utterances

if they are promoted to sub ject p osition

aAlana inen ko caman ucak heyecanlandryor karsn

k i

p ortDAT landREL huge plane exciteCAUSPROG wifePOSS

The huge plane that lands causes his wife to b ecome excited

b Hemen yanndakilere soruyor

i

immediately nearLOCDAT askPROG

i

She immediately asks the p eople next to her

bu ucagn bunun ne ucag oldugunu

k k

this planePOSS thisGEN what plane b eNOMACC

which plane this is

cYanndaki bir ad soyluy or

nearPOSS one name sayPROG

The p erson near her utters a name

gum from Erol Toy Kordu

aTurk u Karaca Aliyi bile yumusatmst

k i

song Karaca AliACC even softCAUSPASTPAST

The song had softened even Karaca Ali

bKaraca Ali Yanndakilere uzun uzun bakt

i i

Karaca Ali NearPOSSLOCPLUDAT long long lo okPAST

i i

He lo oked at the p eople with him for a long time

Yasar Kemal UcAnadolu Efsanesi

In the ob ject do es not undergo the Center Promotion Rule CPR unlike the other ob jects

wehave seen in the previous section In b we cannot use the full NP to realize the ob ject

Karaca Ali intro duced in a is infelicitous In utterance b on the other hand the sub ject has

to b e realized with a full NP Therefore in this case the sub ject rather than the ob ject undergo es

the CPR

aTurk u Karaca Aliyi bile yumusatmst

k i

song Karaca AliACC even softCAUSPASTPAST

The song had softened even Karaca Ali

bBuk bu turk u Ahmetin b estesiydi

k k

thisthis song AhmetGEN comp ositionCOP PAST

Thisthis song was Ahmets comp osition

In examples psychological verbs in the a utterances assign Theme and Exp eriencer

theta roles to their sub jects and ob jects resp ectively As a result Exp eriencer ob jects rank higher

than Theme sub jects There are two alternative theories available in the literature to account

for these facts the notion of Empathy or Thematic Relations These will b e discussed in the

following sections

Empathy

Kuno and Kaburaki intro duce a theory of Empathy which is then elab orated in Kuno

Empathy is the sp eakers identication whichmayvary in degree with a p ersonthing

that participates in the event or state that he describ es in a sentence through the eye

of a camera

Kuno

Since Empathy is a signicant factor in Cfranking in Japanese Kameyama

Walker Iida and Cote Iida it is conceivable that it is a discourse univer

sal in determining Cfranking The examples discussed in mayinvolve an Empathy eect

which assigns the Empathy lo cus to the ob jects The sp eaker identies with the p erson who

is going through a psychological exp erience As a result the exp eriencing p erson in the ob ject

p osition b ecomes more salient

Though this prop osal seems to b e plausible in Turkish the salience of ob jects is limited

to the ob jects of psychological verbs and the notion of Empathy cannot distinguish b etween

psychological and nonpsychological verbs eg verbs of giving and receiving in Japanese involve

EmpathyThus for Turkish discourse a theory that distinguishes psychological verbs from others

is necessaryFurthermore Exp eriencer ob jects of psychological verbs still rank higher than Theme

sub jects in ob jective ie nonEmpathycontexts as will b e shown b elow

er conicts with the camera angle of Kuno states that the ob jectivity of the observ

the exp eriencer He prop oses that an adverb like understandably signals the sp eakers empathy

with the Exp eriencer while an adverb like probably makes the sp eakers empathy imp ossible

with the Exp eriencers psychological state Consider a variant of which is not from the

sp eakers empathy lo cus shown by the adverb b elki probably

aBelki turku Karaca Aliyi yumusatmst

k i

probably song Karaca AliACC softCAUSPASTPAST

Probably the song had softened Karaca Ali

bZaten bu turkuy u cok severdi

i k

In fact this songACC very likeAORPAST

i

In fact he liked this song very much

As noted whether the sp eaker empathizes with the Exp eriencer or not the Exp eriencer still

is the Cb in b namely the highest ranked entity in the preceding utterance

Another test to eliminate sp eakers Empathy mightbetoemb ed the sentence under a clause

like as herkes ackca gorduki everybody clearly saw that where the universal quantier

herkes everybody blo cks the Empathy lo cus However the null sub ject in b realizing the

ob ject in a is still coherent

aHerkes ackca gordu ki Turku Karaca Aliyi yumusatmst

i

everyb o dy clearly seePAST that song Karaca AliACC softCAUSPASTPAST

Everyb o dy clearly saw that the song had softened Karaca Ali

bZaten bu turkuy u cok severdi

i k

In fact this songACC very likeAORPAST

i

In fact he liked this song very much

In conclusion then the b ehavior of psychological verbs in Turkish cannot b e explained by

Kunos theory of Empathy

Thematic Hierarc hy

Thematic roles unlike Empathy can distinguish psychological verbs from others Psychological

verbs assign an Exp eriencer thetarole to one of their arguments either the sub ject or the ob ject If

the ob ject has an Exp eriencer thetarole the sub ject can receive either an Agent or a Theme role

Exp eriencer ob jects rank higher in the Cf list only when the sub ject has a Theme role otherwise

when the sub ject is an Agent it ranks higher For example the sub jects in a sentences in

are assigned Theme roles These sub jects do not have the p otential to act intentionally in order

to achieve the psychological changes on the ob jects Sentences in and show that neither

an adverb likekasten deliberately nor a purp ose clause can co ccur with these sub jects b oth of

which require volitional actors ie Agents

Bu yur uy u sler kasten kendini toparlamasna yardmc oluyordu

This walkPLU delib erately himselfACC pullINFDAT helpful b ePROGPAST

These walks delib erately help ed him to pull himself together

Turk u Karaca Aliyi bile sakinlesmesi icin yumusatmst

Song Karaca AliACC even calmINFACC forto softCAUSPASTPAST

The song had softened even Karaca Ali in order to calm himit down

The same psychological verbs discussed ab ovecanhave Agent instead of Theme arguments in

sub ject p osition In examples and the thematic roles of Ahmet and turk u song will b e

dierent since the former as a human b eing has the ability to b ehavevolitionally The adverb

kasten deliberately is acceptable when Ahmet is the sub ject but not when turk u is When

the predicate assigns an Agent role Ahmet can b e realized with a null sub ject due to thematic

salience as seen in b This indicates that an entitywithanAgent role is more salient than an

Exp eriencer Note that the null sub ject in b necessarily realizes the Agent sub ject of a not

the Exp eriencer ob ject

aAhmet Aliyi kasten yumusatmst

i k

Ahmet AliACC delib erately softCAUSPASTPAST

Ahmet calmed down Karaca Ali delib erately

b onu saatlerce ikna etmeye ugrast

i k

heACC hourPLUADV p ersuasion doINFDATtryPAST

i

He tried to p ersuade him for hours

On the other hand a null sub ject is infelicitous when the sub ject has a Theme role as in

either a demonstrative or a full NP is required

aTurk u Karaca Aliyi bile yumusatmst

i

song Karaca AliACC even softCAUSPASTPAST

The song softened even Karaca Ali

b Turk u Ahmet tarafndan b estelenmisti

i i

song Ahmet by comp osePAST

i

The song was comp osed by Ahmet

Note that Kunos notion of Empathy can account for the facts in equally well in that

the sp eaker is identifying with a human entity ie Ahmet in the sub ject p osition in a but

wever in order for Empathyto nds it hard to identify with the inanimate sub ject in a Ho

b e the right analysis it should also account for In the sub ject is an animate entitywith

a Theme role with whom the sp eaker empathizes Kuno prop oses that given an

expression x egJohn and another expression fx dep endent on x egJohns brother the

sp eakers empathy with x is greater than with fx

aAhmet karsn farkna varmadan yumusatmst

i

Ahmet wifePOSS awareness arriveNEGABL softCAUSPASTPAST

Ahmet calmed down his wife without b eing aware

b Artk eskisi gibi cok cabuk sinirlenmiyordu

i

anymore oldone likevery fast nervousNEGPROGPAST

i

She didnt b ecome nervous as quickly as she used to

In a the sp eaker is exp ected to empathize with the sub ject Ahmet rather than with an

expression that is dep endent on it ie his wife this predicts that the former not the latter will

b e salient However as seen ab ove the dep endent expression in ob ject p osition must rank higher

in the Cf list b ecause it is the Cb realized with a null sub ject in the subsequent utterance The

thematic role hierarchy provides an explanation for this The reason is that the ob ject in a is

Exp eriencer while the sub ject p erforming an action unintentionally is the Theme

The discussion so far suggests that entities in Turkish must b e ranked based on thematic roles

as in the following order

AGENT EXPERIENCER THEME

Related ResearchinFocusing

Sidner formulates the exp ected fo cus hierarchy using thematic relations She ranks the

argument with a Theme role as the most saliententityinanutterance

As intro duced in Chapter the twotyp es of fo ci in Sidners Fo cusing Theory Actor Focus

AF and Discourse Fo cus DF can b oth b e the entities in the center of attention Sidner

suggests that DF is the entity that the sp eaker is mainly concerned ab out and that the AF

ie the argument with an Agent role is less salient Sidners prop osal contradicts with the

suggestions of Jackendo Grimshaw and Pesetsky among others who claim

that for agentiveverbs the Agentalways ranks highest in a salience scale Kameyama

argues against Sidners suggestions and shows that a grammatical hierarchy can equally explain

b oth Sidners data and the facts in Japanese

Suri extends Sidners analysis In her analysis the CurrentFo cus CF is the ob ject or

the Theme in the utterance There is a separate Actor Fo cus AF identied with the entityin

an Agent role in the utterance Roughly akin to our Cflist Suri prop oses keeping two separate

Fo ci Lists for the CF and the AF The CF and the AF are listed in the Potential Fo cus List PFL

and in the Potential Actor Fo cus List PAFL resp ectively If an utterance has no Agent then

the stacked AF is retained PAFL contains all NPs that sp ecify an animate entity in the database

in an Agent role A pronoun in Agentroleisinterpreted as the AF in the previous utterance

Suri like Sidner suggests that the ob ject or the Theme should rank higher than the sub ject or

the Agent in the hierarchy

Wehave seen evidence that sub jects rank higher than ob jects in Turkish in the discussion

so far Furthermore given that there is precisely one Cb in Centering Theory the SidnerSuri

analysis cannot b e adopted here

Related ResearchinCentering

Brennan

Brennan carries out a set of psycholinguistic exp eriments to nd out when English sp eakers

use pronominal forms as opp osed to full NPs to refer to previously evoked entities In her data from

recorded conversations one of the sp eakers describ es the basketball game previously recorded on

the videotap e to a hearer who cannot see the screen Brennan cho oses a basketball game b ecause

it is dynamic and fast moving and it exerts a reasonably high degree of control over the center

of attention the basketball or the player with the basketball Furthermore it provides a domain

with many opp ortunities to refer to entities of the same numb er and gender establishing a test

for centering predictions

Brennan nds that prominententities are intro duced in sub ject p osition as opp osed to ob ject

p osition This is a signal to the hearer that the center of attention is the particular entityin

sub ject p osition This supp orts the centering hyp othesis that the highest ranked entity in the Cf

list is the sub ject

Brennans other ndings also supp ort centering predictions She argues that when entities

are intro duced as ob jects sp eakers are likely to rep eat them as full NPs in sub ject p osition b efore

pronominalizing them According to centering denitions the Cp the highest ranked entityisin

sub ject p osition and is the exp ected Cb of the subsequent utterance A more explicit referential

expression ie a full NP is used for a less saliententity whichwas not the sub ject The following

example from Brennan shows this strategy The basketball player fortyone is intro duced as

the ob ject in the rst utterance and is rep eated as a full NP in sub ject p osition in the subsequent

utterance b efore b eing pronominalized

a Numb er thirty passes it of tofortyone

b Fortyone go es up for the shot

c and he misses

Brennan notes that this strategy is employed even though describing a basketball game is

biased against centering b ecause a basketball game is often quite fast and sp eakers haveto

keep up the pace of their descriptions As a pronoun is shorter than a full name we might

except that sp eakers would avoid unnecessary rep etition However as the ob ject p osition is less

salient the sp eaker takes the time to bring the entityinto sub ject p osition with a full NP b efore

pronominalizing it This supp orts Centering Theory in that sp eakers do not use pronouns for all

entities that are previously mentioned but for those that are in the center of attention

Cote

Cote discusses the inadequacy of grammatical roles in determining the Cfranking in English

She p oints out that in Japanese as shown byKameyama and Walker Iida and Cote

not only grammatical roles but the topic marker wa and the Empathy markers on verbs are

relev ant as briey discussed ab ove Cote also cites Di Eugenio who shows that some verbal

forms carrying tense and agreement morphologymayoverride centering predictions in Italian

Cote states that this constitutes evidence that the Cfranking cannot b e comp osed entirely of

syntactic features

Cote presents additional evidence from null ob jects in English whichmotivate her use of

Jackendo s Lexical Conceptual Structures LCS She argues that null ob jects in English are not

phonologically null pronouns or the result of ellipsis but that they are lexically aected arguments

that interact with Centering Rules

Consider examples in from Cote Certain verbs like eat in English mayhavenull

ob jects but in fact these null ob jects aect the subsequent salience of the entity they realize

Notice that in the following example even when the verb eat has a null ob ject it still evokes a

discourse entity realized by it in the subsequent utterance

In a small numb er of cases where sp eakers did not center an entity b efore pronominali zin g it they lengthened

the pronoun slightly This may b e a strategy to signal to the hearer that the sp eaker is talking ab out another entity

and not the Cp

D We ate at Jorges

M Was it go o d

D Um it was all right

M You know I dont think thats as go o d as a lot of p eople think it is

Unlikeeatverbs like clean and iron cannot havenull ob jects that have salientantecedents

A E Whadiyah doin

N What am I doin

E Cleanin

N Im ironing wouldja b elieve that

E Oh bless its heart

B

E Whadiyah doin with your shirts

N What am I doin with them

E Cleanin them

N Im ironing them wouldja b elievethat

E Oh bless its heart

On the other hand some verbs can havenull ob jects only when they have a textually evoked

salient ob ject as in

She stopp ed at the Surclis after dusk and had a Scotchandso da She stayed to o late

and when she left itwas dark and time to go home and co ok supp er for her husband

A What did you do on vacation

B I left

All these verbs imp ose dierent idiosyncratic constraints on their ob jects According to Cote

this idiosyncrasy can b est b e represented by the LCS whichcombine the syntactic sub categoriza

tion frames and selectional restrictions of verbs

For example the LCS of a verb likeeatcanbegiven as follows when the ob ject is lexically

aected indicates that the ob ject can b e null

CAUSE Thing a Event GOThing Path TO Place IN Thing MOUTHOF a

Cote prop oses that LCS can b e used as Cftemplates an abstract construct used in ranking

the Cfhierarchyasfollows

Conceptual Structure Cftemplate for English

Begin by ranking arguments of the phrasal conceptual structure from lefttoright

Mark all arguments as pronominal for Un unless no other interpretation is p os

sible

The prop osal made for the Cfranking in this study is along the lines of Cotes prop osal since

thematic roles are a part of LCS As Jackendo cited in Cote writes thematic roles

are nothing but particular congurations in conceptual structures the names for them are just

convenient mnemonics for particularly prominent congurations

A similar ranking of the Cflist is prop osed by HudsonDZmura using Lexical Relational

Structures LRS HudsonDZmuras work will b e discussed in Section For the time b eing we

should note that LRS is analogous to LCS and b oth make use of thematic relations underlyingly

Consequently using LCS is akin to using thematic relations to determine the Cfhierarchy

The dierence b etween LCSLR and thematic relations is that the former also include event

structure as well as thematic structure They refer overtlytoevents and states Cote argues

that this is an imp ortantcontribution b ecause b oth pronouns and null ob jects mayhaveevent

antecedents and that LCS provide a to ol for the representation of these anaphoric relations

To conclude this section there is evidence for using thematic relations for the Cfranking in

b oth Turkish and in English motivated for dierent reasons The evidence in the two unrelated

languages suggests that factors determining Cfranking may in fact b e universal rather than

languagesp ecic Future research will rene and elab orate these observations and determine

whether thematic relations or more p owerful devices as LCS and LRS should b e used for Cf

ranking

Possessive Constructions

Ranking the Cflist using the thematic hierarchy predicts that an Exp eriencer in a genitive case

marked NP ranks higher than the Theme This is indeed the case as indicated in

aBu olay Alinini kalbini krmst

k

this event AliGEN heartPOSS breakPASTPAST

This event had broken Alis heart

hic unutamyordu b Bunu

k i

thisACC never forgetNEGABILPROGPAST

i

He couldnt ever forget this

In a Ali is genitive casemarked Note that the null sub ject in the subsequent utterance

b is interpreted as Ali rather than as Alis heart The demonstrative pronoun on the other

hand realizes the sub ject in a Theme role this event Ali also ranks higher than the entity

evoked by the whole p ossessive NP Alinin kalbi Alis heart Note that in a the breaking of

Alis heart is idiomatic Ali is the Exp eriencer and since kalbi krlmak break heart is p ossibly

listed in the lexicon as a complex verb it can b e argued that kalp heart do es not evokea

discourse entity As a result it cannot b e the Cb in the subsequent utterance

A similar explanation holds for niyeti olmak haveintention in a Intention do es not

evoke a discourse entity and thus Nesrin the Exp eriencer is the entity predicted to b e the Cb

of the next utterance

aNesrinin karaya ckmaya niyeti yoktu

i

NesrinGEN landDAT gooutINFDATintentionPOSS existNOTPAST

Nesrin had no intention to land

c balk tutacakt mutlaka

i

sh catchFUTPAST denitely

i

She would denitely catch some sh

Gulderen Bilgili Bir Gece Yolculugu

However p ossessive NPs mayevokemultiple discourse entities For example in d and e

the sub jects are inalienable p ossessive constructions Sulonun yuz u Sulos face el leri his hands

bacaklar his legs etc are his b o dy parts In these NPs the entityevoked by the genitivemarked

Sulo or his in his legs face etc can b e referred to anaphorically in the subsequent discourse In

f the null sub ject is interpreted as Sulo rather than his b o dy parts which is the Cb in e

aSulo kendini kayb etti

i

Sulo selfACC losePAST

Sulo passed out

Ayldg zaman kendini bir adamn kollarnda buldu b

i i

wakeup when selfACC one manGEN armsPOSSDAT ndPAST

i

When he woke up he found himself in the arms of a man

cAdam onu bir yere gotur uy ordu

i

man heACC one lo cationDATtakePROGPAST

The man was taking him somewhere

d Yuz unde ellerinde bacaklarnda bir sur u yara vard

i

facePOSSLOC handsPOSSLOC legsPOSSLOC one many bruise existPAST

There were many bruises on his face hands and legs

eBir iki yeri hzla kanyordu

i

one two placePOSS fast bleedPROGPAST

Some parts of his face legs etc were bleeding

f Kannn aktgn gorunce aglamaya baslad

i

blo o dGEN owNOMACC seeWHEN cryINFDATstartPAST

i

Lit He started to cry when he saw his blo o d was owing

Fakir Baykurt BarsCoregi

One might argue against this suggestion stating that the salience of Sulo in is due to

its status as an instantiated Cb rather than any other prop erty of p ossessive NPs Continue

transitions are preferred and so sp eakers may tend to talk ab out Cbs rather than newly intro duced

entities In order to eliminate any eects that may stem from the Cbstatus wemust explore

examples where b oth nouns in a p ossessiveNPevoke brandnew entities

Likewise in a in the p ossessiveobjectNP the null genitive is the most saliententityeven

though it is not in the sub ject p osition This is b ecause the Exp eriencer ob ject ranks higher than

the Theme sub ject Lumumba Universitesi Lumumba University Note that the null sub ject in

b is interpreted as the individual who go es through the state of mind and not as Lumumba

University or as his mind

a Aklna Lumumba Universitesi geldi

i

mindPOSSDAT Lumumba University comePAST

i

Lit The Lumuba University came to his mind

b Formul uarad ceketin cebinde

i

formula lo okforPAST jacketGEN p o cketPOSSDAT

i

He lo oked for the formula in the jackets pocket

Ozdemir Basargan Gurbet Sofras

To sum up then an Exp eriencer in a p ossessiveNP or a p ossessor of inalienable parts ranks

higher than the entityevoked by the genitive construction Note that none of the Exp eriencers or

p ossessors undergo the Center Promotion Rule in discourses but rather they are realized

with null sub jects in the immediately subsequent utterances This shows that they are indeed the

Cps

e construction is not more salient than On the other hand the genitive noun in a p ossessiv

the entityevoked by the whole NP if the latter has an Agent role For example in b the null

sub ject cannot b e interpreted as Canan but is unambiguously used to refer toCananin annesi

Canans mother

aCananin annesik kek pisirmis

i

CananGEN motherPOSS cake co okPAST

Canans mother baked a cake

b onu yemis

ki

itACC eatPAST

ki

She ate it

Prince and Walker tentatively prop ose the following assumption for p ossessiveNPs

The Complex NP Assumption

In English when an NP evokes multiple discourse entities suchasanNPwitha

p ossessive pronoun the Cf ordering is from left to right within the NP

Prince and Walker prop ose that in an utterance like Her mother knows Queen Elizab eth

i k

her ranks higher than her mother ie i k The assumption given in Prince and Walker is

mo died for Turkish as follows

Possessive NP Hyp othesis

In Turkish when a p ossessiveNPevokes multiple discourse entities the Cfranking

should b e from left to right if the genitive noun is

a the Exp eriencer eg in his heart was brok en his his heart

b the entityevoked is involved in a cognitive state eg in constructions like Mary

came to Johns mind John Mary Johns mind

c the p ossessor noun in an inalienable p ossession

To sum up the discussion so far the following ranking is assumed

AGENT EXPERIENCER Inalienable Possessor THEME

Discourse Point of View and Segmentation

This section discusses other typ es of elements that rank higher than sub jects and constitute

exceptions to the Center Promotion Rule Before the discussion on Cfranking a brief intro duction

to a concept called Discourse Point of View will b e presented Discourse Point of View is one

factor which aects discourse segmentation Other factors that aect discourse segmentation will

b e discussed in Section

Sub jective and Ob jectivePoints of View

An author mayintend for the reader to evaluate the utterances from an ob jective or a sub jective

p oint of view This intention aects the realization of the entities at the attentional state These

two comp onents of discourse structure trigger clues for discourse segmentation Ob jectiveand

sub jective p oints of view may o ccur in dierent discourse segments or they maybeemb edded

within the same segment in a hierarchically ordered discourse

Dep ending on hisher intention then an author may narrate a story from an ob jective p oint

of view or from the sub jective p oint ofviewofacharacter Ob jective and sub jectivepoints of

view are represented b y ob jective and sub jectivesentences resp ectively Ob jective sentences rep

resentevents ob jectively and indep endently of anycharacters consciousness Sub jective sentences

present the consciousness of an exp eriencing character within the storyacharacters evaluations

judgments uncertainties hisher represented thought emotion or p erception Baneld

Wieb e and Rappap ort Ehrlich Wieb e A private state is part of

the meaning of a sub jectivesentence Ob jectivesentences in a world created by the narrator are

immune to truth or falsity judgments They create a ctional world that is taken to b e ctionally

true Sub jective sentences on the other hand reect the b eliefs thoughts and emotions of the

characters whichmay b e true or false The reader do es not assume that those b eliefs thoughts

and emotions presented in sub jectivesentences are necessarily true cf Wieb e and the

references therein

A sub jectivesentence ab out a sub jectivecharacter is not directed to the hearer A sub jective

sentence might b e thought of as the reader overhearing the characters thoughts or the reader

p erceiving the world through a characters senses Wieb e These private states are not

op en to ob jective observation and verication Baneld Wieb e

Sub jectivesentences are marked by sub jective elements Some sub jective elements are ex

clamations which express emotion epithets such as idiot which express the evaluation of the

referent kinship terms such as daddy which indicate the relationship of the referent and the

character According to Baneld see also Wieb e Fleischman the sub

jectivity expressed is attributed to the sub ject of consciousness or the self In a conversation

where the rst p erson singular I would normally b e used to refer to the sp eaker the self is the

sp eaker so the sub jectivity that app ears is attributed to the character In a sub jective sentence

the sub jectivecharacter is the self even when she is referred to in the third p erson In an

utterance like The idiot was standing next to her the epithet the idiot indicates the sub jective

nature of the sentence and who ever the idiot is is standing next to the referent of her This is

exemplied by the Turkish discourse b elow where a character Gokmen is realized with a third

p erson singular pronoun in c

aDayanlmaz aclar yasyordu Gokmen

i

unb earable painPLU livePROGPAST Gokmen

Gokmen was going through unb earable pain

bYasamcok uzundu iste

life very longPAST EXCLAMATION

Life was to o long

cYoo baska hi c kimse ilgilendi rmiyordu onu

i

no other none nob o dy interestNEGPROGPAST heACC

No nob o dy else interested him

d Yeryuz une Belks yasamak icin gelmisti

i

worldDAT BelkisACC liveINF for comePASTPAST

i

He had b een b orn to live for Belkis

Selim Ileri Cehennem Kralicesi

In the example ab ove a intro duces a sub jective p oint of view describing Gokmens suering

and his judgments and emotions are presented in bd Note that the sub jectivepoint of view

is marked by sub jective elements eg the exclamation iste which expresses a value judgment

in b and yo o no informal in c These sub jective elements cannot o ccur in ob jective

sentences Furthermore as has b een previously stated though these utterances representthe

characters inner thoughts ie his sp eech to himself the pronoun in c is in third p erson

singular rather than rst p erson

A represented thought explicitly mentions only the ob ject of the private state the attitude

of the character is implicit Wieb e Changes in tense and asp ect are contextsensitive

markers of sub jectivity Ehrlich The b ehavior of pronominal anaphora is another

ers are linguistic clues of marker to b e discussed in the next section Note that all these mark

discourse segment b oundaries

Point of View Shifts and Centering Predictions

When there is a switch from an ob jective to a sub jectivepoint of view a nonsub ject rather than

the sub ject can b e preferred as the next Cb

Sub jectivesentences can b e intro duced with verbs denoting private states These verbs can

be intellectual like b elieve wonder rememb er emotive like hate like or p erceptual like see hear

etc the consciousness through which the events are pro jected For example b indicates a shift

from ob jective to sub jective p ointofviewbyanintellectual verb rememb er We the readers

nowgointo the mind of Dr Wiley and see the world through his eyes Note that the adverb

apparently indicates a judgment of the p erceiver which signals the sub jective view The third

p erson pronoun he in c is used unambiguously to refer to Charles not to the Cp Dr Wiley

This is b ecause Charles is now pro jected through DrWileys consciousness and the entityis

ranked in his cognition as Dr Wileys mind Cflist Charles

a As you wish said Dr Wiley

i

b He rememb ered Charles as a rash but thorough student

i k

c Apparently he hadnt changed

k

d Although Dr Wiley knew that

i

Robin Co ok Fever

Likewise In a the p erception verb lo ok at marks a switch to a sub jectivepoint of view

the way in which Charles p erceives the man in front of him Note that once more the Cb in b

cosp ecies with the entity in the ob ject p osition in a due to the reasons discussed in example

a Charles lo oked at the man in front of him

i k i

b He was a full head taller than Charles

k i

c His p erspiring face was so pudgy that his eyes were mere slits

k k

d He was dressed like the other men Charles had seen

k i

Robin Co ok Fever

The following examples indicate that in Turkish as in English the nonsub ject rather than

the sub ject is preferred and realized bynull sub jects when there is a switch from an ob jectivetoa

sub jective p oint of view These null sub jects should b e ambiguous for they can equally cosp ecify

b oth of the entities in the previous utterance but they are unambiguously interpreted Consider

aDumen basndaki motorcu Korkutu suzuy ordu

i

rudder frontLOC motorist KorkutACC examinecloselyPROGPAST

The motorist in front of the b oats rudder was lo oking at examining Korkut carefully

bb esb elli gunesve deniz artk onu ilgilendi rmiyordu

i

apparently sun and sea anymoreheACC interestNEGPROGPAST

Apparently the sun and the sea did not interest him any more

c usanmst

i

b etiredPASTPAST

i

he was tired

d Tekne hantald

motorb oat coarseCOPPAST

The motorb oat was coarsely built

Selim Ileri Cehennem Kralicesi

The utterance in a presents Korkuts state from the motorists view indicated by the p er

ception verb suzmek lo ok at examine closely Hence Korkut is seen from the motorists

p ersp ective enabling it to b e the Cp As can b e seen from the indices the null sub ject in b

realizes the ob ject of a

Likewise is an example of switch from an ob jective to sub jective p ointofview

aGozlerini kapad Cem

i

eyePLUPOSSACC closePAST Cem

es Cem closed his ey

b Keremi dusundu

i k

KeremACC thinkPAST

i

He thought ab out Kerem

c Az konusurdu

k

little sp eakPRESPAST

i

He would hardly sp eak

Selim Ileri Her GeceBodrum

In ab Cem is the sub ject However the p erception verb du sunmek think signals a

shift in p oint of view In this case once again the reader lo oks at the world from a characters

p ersp ective Thus Kerem is the entity salient to Cem and it can therefore b e realized with a

null sub ject in c

Likewise the p erception verb in b b elow bakmak lo ok puts the ob ject into p ersp ective

which makes it a saliententity

a Jeeplere bindiler

i

JeepPLUDAT getonPASTPLU

i

They got on the Jeeps

bGokmen gunesek bakt

m

Gokmen sunDAT lo okPAST

Gokmen lo oked at the sun

c Yine goz kamastrcyd

k

again eye dazzlingCOPPAST

k

It w as still dazzling

d ama iyice batya kaymst

k

but well westDAT slidePASTPAST

k

But it had prettymuch slid to the west

Selim Ileri Cehennem Kralicesi

Wehave seen that b oth in and English and in Turkish verbs of p erception trigger a sub jective

p oint of view which represents entities from an individuals consciousness As a result entities

are ranked in the order in which they are salient to the sub ject of consciousness

Discourse Point of View Rule

If any of the events or actions are presented from the p erception or p ointofviewof

tities are ranked through that individuals p ointof a sub ject of consciousness the en

view

What this rule states is that the salience of an entity dep ends on the way in whichitis

p erceived by the sub ject of consciousness Note that in the sub jectivepoint of view an ob ject

or an individual in the physical environment or in the mind of a sub ject of consciousness can b e

salient to himher as well as the author Thus returning to examples and cited in Section

as Gordon Grosz and Gilliom suggest twoentities have equal salience However contrary

to their suggestion the sentenceinitial p osition do es not contribute to the salience of the entity

making it as salient as the sub ject but rather in the sub jectivepoint of view b oth the self and

the individual are equally available to the sub ject of consciousness

Sub jective p oint of view can create a subsegment within an ob jective discourse segment In

these instances the discontinuous utterances in the ob jective segments are considered to b e adja

cent Consider Utterances from from d to q form a sub jectivepoint of view subsegment

intro duced by the p erceptual verb bakmak lo ok in c This is also evident from the sub jective

elements present ie the epithets mubarek the sacred thing in h veletler brats and hnzrlar

bastards in l and q resp ectively adjectives like hasarat naughty in l as well as the

rhetorical question in m and n Note that discourse segmentation is also indicated in the

change in asp ect in d The character Recep in the sub ject p osition in ac can b e realized with

epoint of view in r anull sub ject with the shift to ob jectiv

aRecep elma agaclarna bakyor

i

Recep apple treePLUDAT lo okPROG

Recep is lo oking at the apple trees

b bir kac adm yur uy or

i

one few step walkPROG

i

he takes a few steps

c durup tekrar bakyor

i

stopCONJ again lo okPROG

i

he stops and lo oks at them again

d Alt dallarda elmalar iyice seyreklesmis

lower branchPLUDAT applePLU very b ecomerarePAST

The apples in the lower branches havebecomevery rare

e Yesil yapraklarn aralarnda tek tuk gorun uy or

green leavePLUGEN b etweenPLPOSSLOC one or two seePASSPROG

A few apples could b e seen in b etween the green leaves

f Krmz gune s gibi yanyor her biri

e shinePROG eachone red sun lik

Each of them shines like a red sun

g Agaclarn diplerinde curuk binlerce elma

treePLUGEN b eneathPOSSLOC rotten thousands apple

Thousands of rotten apples b eneath the trees

h Mubarek ne de cok dokul uy or

sacredthing what to o much spillPASSPROG

The sacred thing Howmanyhavefallendown

i On elmadan sekizi curu ge ckyor

ten appleABL eightACC rotten goPROG

Eight of ten apples go bad

jColuk co cuk da gelip yerden alp yemiyor

childrenmildren to o comeCONJ groundABL takeCONJ eatNEGPROG

And the kids dont come pick them up from the ground and eat them

k gidip daldan koparyor

goCONJ branchABL pickPROG

i

They go pick them up from the trees

l Hepsi hasarat veletlerin

all naughty bratPLUGEN

All of the brats are naughty

m Ne olur canm elmalari toplayp yeseniz

what happ enAOR lifeS applePLUACC pickCONJ eatCONDPLU

e why cant you pick up and eat the apples For Go ds sak

n saglamn braksanz

freshACC leaveCONDPLU

and leave the fresh ones

o O da elma bu da elma fark ne

this to o apple that to o apple dierence what

Apples are apples What is the dierence

p Yok illa zarar verecekler

No surely harm giveFUTPLU

But no they have to harm things

q Hnzrlar

bastards

r Egilip yerden bir tane ald

i

b endCONJ groundABL one piece takePAST

i

He b ent and picked up one from the ground

Kemal Bilbasar Baska Olur Agalarn Du gunu

One question concerning Centering is whether the typ e of discourse in should b e considered

as a lo cal discourse segment c and r can b e said to b e two consecutive utterances with

resp ect to their discourse segment Or should the null sub ject in r b e resolved at the global

level Following Grosz and Sidner adjacent utterances can b e within dierent discourse

segments and two nonadjacent utterances can b e in the same segment Further researchis

needed to resolve what a lo cal discourse segmentisforCentering Theory

In this section discourse segmentation in relation to Cfranking has b een discussed It has

b een suggested that the sub jective p ointofviewintro duced with verbs of p erception requires the

entities to b e ranked from the p oint of view of the character In the following related work in

Centering will b e briey discussed

Related Work In Centering

Kameyama prop oses that the sub jectivepoint of view can p otentially b e explained as

Empathy lo cus of p erception verbs in English She states that p erception verbs like seehear

lo oksound etc anchor the sp eakers p ersp ective as do Japanese empathyverbs However in

sub jective p oint of view contexts the saliententity is not the empathized entity but the one that

is not empathized

HudsonDZmura also deals with the sub jective p oint of view intro duced by p erceptual

verbs She suggests that the b ehavior of p erception verbs can b e explained by lexical event

structures as discussed in Huang and their interaction with Centering rules

Huang argues that a part of the highly organized and constrained structure of thematic

relations is predicted from their event structure Huang represents this by the Lexical Relational

ve LRS are akin to Structure LRS due to Hale and Keyser As has b een p ointed out ab o

Jackendo s Lexical Conceptual Structures which Cote suggests should b e used to determine Cf

templates LRSs are elements ordered by the thematic hierarchy and event the structure of verbs

LRS also marks whether verbs have pure or complex event structures Verbs can b e decomp osed

into one or more pure or complex verb meanings which they share with a large number of verbs

plus idiosyncratic features which distinguish a verb from others LRSs are comp osites of twoor

more basic eventualities CAUSE DO UNDERGO BE HudsonDZmura suggests that in

the following discourse Dan is more likely to b e understo o d as the p erson going inside the store

following a than b

a Dan saw Ben approaching the store

b Dan lo oked at Ben approaching the store

c He went inside

According to HudsonDZmura the interpretation of the ambiguous pronoun in c dep ends

on the Lexical Relational Structures LRS of the verbs see and lo ok That is see is an example

of an UNDERGO and lo ok is an example of a DO verb The latter involves intentionality which

the former do es not That means that one can intentionally lo ok at something but cannot

intentionally see it See and lo ok are basic pure predicates whichcombine with another verb as

an op erator to mo dify the meanings as illustrated in

a lo ok x DO x lo ok

b see x UNDERGO x see

c sightxDOxCAUSE UNDERGOxsight

In the psychological exp eriment HudsonDZmura designs she asks whether the dierences

between intentionalityoftheevent of the vision verbs is related to the choice She places twelve

vision verbs in identical sentence frames eg Dan verb ed Ben approaching the store These

sentences are followed by He went inside The sub jects are instructed to read the material and

then cho ose the antecedent of the pronoun he The results suggest that when the vision verb is

intentional the antecedent of the sub ject pronoun is the nonCp whereas if the vision verb is not

intentional the Cp is preferred

Future research is needed to determine whether discourse p oint of view can b e explained in

more depth with a notion of intentionality asproposedby HudsonDZmura Nevertheless it is

interesting to note that Cote also discussed in this chapter uses Jackendo s Lexical Con

ceptual Structures whicharevery similar to the Relational Lexical Structure HudsonDZmura

uses to explain the discourse p oint of view expressed by p erception verbs

Problems in the Analysis of NaturallyOccurring Data

Intervening Utterances and Segmentation

In this section some problems encountered in analyzing naturallyo ccurring data will b e discussed

in relation to Cfranking and Centering analyses Although certain solutions will b e suggested

this section raises more questions than it answers

Centering Theory predicts that the transitions ie center continuation retention or shift are

wo adjacent utterances in a discourse segment This rule requires that only p ossible b etween t

every single entity listed in the Cflist of an utterance Un should b e realized in Un Grosz Joshi

and Weinstein write The Cb is strictly lo cal CbUn cannot b e from CfUn or other

prior sets of forwardlo oking centers In their fo otnote they state that It may on o ccasion

app ear to b e from CfUn or prior sets of forwardlo oking centers but then it is only b ecause

it is in CfUn also

Thus if there is an intervening utterance whose Cflist do es not contain the Cb it should b e

reintro duced by a full NP on its next mention Consider

aRecep Day bahce kapsndan girerken

i

Recep Uncle garden gatePOSSABL enterWhen

Uncle Recep while entering through the garden gate

az kald Murata carpacakt

k

little left MuratDAT hitFUTPAST

almost hit Murat

b Kenara cekildi yol verdi doktora

i k

sideDAT recedePAST way givePAST do ctorDAT

i

He drew to the side and gaveway to the do ctor

cMurat merhaba bile demeden gecti

k

Murat hi even sayNEG passPAST

Murat passed by without even saying hi

d Ihtiyar onun delikanln n bu haline sast

i k k

old heGENladGEN this stateDAT b esurprisePAST

The old man was sho cked by his b ehavior the b ehavior of the young man

Kemal Bilba sar Baska Olur Agalarn Du gunu

In a the most saliententity is Recep Day in sub ject p osition and is realized with a null

pronoun in b In c it is not clear whether the verb demek say sub categorizes for a dative

argument It may b e the case that there is no suchentity in the sub categorization frame of the

verb If so ihtiyar the old man will not b e listed in the Cflist of c Then the entityhasto

b e reintro duced with a full NP ihtiyar the old man in c

This requirement is violated in and where b and b do not contain the relevant

entity the Cb of the next utterance in their Cflists Nevertheless the intervention of these

utterances do es not blo ck the realization of that entity with a null sub ject in c and c

aBelks delikanlya Mehmeti anlatacakt

i k

Belkis youngmanDAT MehmetACC tellFUTPAST

Belkis would tell the young man ab out Mehmet

bMehmet yeryuz un un en karslksz askyd

k

Mehmet worldGEN most unrequited loveCOPPAST

The Mehmet aair was the most unrequited love in the world

c onu sevmis

i k

heACC lovePAST

i

She loved him

d onunla dans ederken mutlu olmus

i k

heWITH dance doWHEN happybePAST

i

She b ecame happy when she danced him

e ama Mehmete asla ulasamamst

i k

but MehmetDAT never reachNEGABILPASTPAST

i

but she could never reach him

Selim Ileri Cehennem Kralicesi

aC esmeden su almaya giden delikanll ar arasnda

fountainABL water takeINFDAT goRel ladPLU among

Among the young men who go to the fountain to get water

bir tek Nalbandn oglu Ibramdan hoslanyor

k i

one only blacksmithGEN sonPOSS IbrahimABL likePROG

i

she likes only the blacksmiths son Ibrahim

bO da paraszn biri

k

he to o moneyless one

And he is totally broke

c Onunla evlenemez

i k

heWITH marryABILNEGPRES

i

She cant marry him

u Kemal Bilbasar Baska Olur Agalarn Du gun

However note that b and b are sub jectivesentences as discussed in the previous section

These utterances reect the sub jectivepoint of view of the individuals their value judgments In

b her unrequited love for Mehmet is presented from Belkis consciousness In b likewise

the reader go es through the consciousness of the individual who makes a value judgment ab out

Ibrahims nancial status In that case the sp eaker can return talking ab out the previous Cb in

the ob jective segment as in b or continue reecting the sub jective judgments of the character

as in c Then the utterances where the null pronouns are interpreted as referring to the same

entity are considered to b e consecutive with resp ect to the same discourse segment

Now consider where the Cb Riza Bey is realized byanull pronoun throughout the

discourse The underlined utterance e do es not have Riza Bey in its Cflist This utterance

intervenes b etween the two utterances that realize this entity ie Riza Bey with null pronouns

The single entity in the Cf list of e bir mektup a letter do es not prevent reference to Rza Bey

in f with a null pronoun On the other hand mektup letter is realized with a full NP in the

ob ject p osition in f

aRza Bey dairesinin onune gelince

Riza Bey apartmentPOSSGEN frontPOSS comeWHEN

When Riza Bey came in the front of his apartment

b anahtarn ckard

i

keyPOSSACC takePAST

i

he to ok out his key

c kapy act

i

do orACC op enPAST

i

op ened the do or

d elektrigi yakt

i

electricity burnPAST

i

he turned on the light

e Kapnn ic yannda bir mektup duruyordu

k

Do orGEN inside sidePOSS one letter standPROGPAST

A letter was lying inside by the do or

f Egildi ald mektubu

i k

b endPAST takePAST letterACC

i

He b ent and picked up the letter

g ve dondu kald

i

and freezePAST stayPAST

i

And he froze in place with astonishment

Cetin Altan Riza Beyin Polisiye Oykuleri

This intervention in this case cannot b e explained by sub jectivepoint of view Consider

avariant of If b were presented from the p oint ofviewofthecharacter one could claim

that the intervening utterance constitutes a subsegment or else the Cflist of e contains the

sub ject of consciousness This would allow the reference with a null sub ject in d However

c shows that the character did not see the letter therefore the event is not presented from his

p ersp ective

a iceri girdi

i

inside enterPAST

i

he went inside

b Kapnn ic yannda bir mektup duruyordu

k

Do orGEN inside sidePOSS one letter standPROGPAST

A letter was lying inside by the do or

c mektubu gormedi

i k

letterACC seeNEGPAST

i

He didnt see the letter

d Televizyonu act

i

televisionACC op enPAST

i

He turned on the television

In the intervening utterances c and e do not prevent the realization of the Cb bya

null sub ject in two nonadjacent utterances

aTemel Trabzona donmek icin tren bileti alr

i

Temel TrabzonDAT returnINF for train ticketPOSS buyAOR

Temel buys a train ticket to go backtoTrabzon

bve b eklemeye ba slar

i

and waitINFDAT startAOR

i

and he starts waiting

cBir anons yaplr

one announcement makePASSPRES

An announcement is made

d gider bir kompartmana oturur

i

goAOR one compartmentDAT sitAOR

i

He go es sits down in one of the cars

e Kompartmanda da bir adam oturuyor

carDAT to o one man sitPROG

A man is sitting in the train car to o

f Baslar konusmaya

i

startAOR talkINFDAT

i

He starts talking

from an email message received on December

Initisobvious that Temel heard the announcement and it is conceivable that this awareness

triggers the sub jectivepoint of view licensing a null sub ject However consider c where the

reference with a null sub ject b etween b and d is felicitous

aTemel Trabzona donmek icin tren bileti alr

i

Temel TrabzonDAT returnINF for train ticketPOSS buyAOR

Temel buys a train ticket to go backtoTrabzon

bve b eklemeye baslar

i

and waitINFDAT startAOR

i

and he starts waiting

cBir anons yaplr

one announcement makePASSPRES

An announcement is made

d anonsu duymaz

i

announcementACC hearNEGPAST

i

He do esnt hear the announcement

e Gider caybahcesinde oturur

i

goPRES tea houseLOC sitAOR

i

He go es and sits in the tea house

Below the relationship b etween pronominalization and discourse segmentation will b e dis

cussed

Discourse Segmentation and Pronominalization

Since Centering mo dels attentional state within a lo cal discourse segment it is very imp ortant

to determine what exactly a discourse segment is Wehave seen that one typ e of segmentation is

signaled by shifts in the Discourse Point of View but this is not limited to the distinction b etween

ob jective and sub jectivepoints of view

Suri and McCoy suggest that intervening utterances like are problematic in Cen

tering Theory but not in Fo cusing Theory The latter unlikeCentering stacks old information

and hence the sp eaker has the option of returning to the Potential Fo ci List to use a pronoun

after an intervening utterance Suri and McCoy state

an explanation of pronoun resolution in such cases must rely on either a stacking

of fo cicenters or a claim that the intervening utterance constitutes a discourse

segmentunto itself whichinterrupts the discourse segmentcontaining the previous

and subsequent sentences and thus the pronoun resolution must rely on global fo cus

and the recognition of a discourse segment b oundaryHowever it is not clear what

linguistic information would mark the discourse segmentation needed by a second

approach The recognition of discourse segment b oundaries and their inuence on

pronoun resolution is still not welldened Without a theory of discourse segmentation

that accounts for such cases centering is at a loss for explaining pronoun use

Walker also addresses the problem of discourse segmentation for Centering Theory She

surveys p ositions taken with resp ect to discourse segmentation in the Centering literature and

p oints out that the cues for discourse segmentation ie tensemo dalitysyntactic structure cue

phrases linear sequence orthography and proso dy and their interaction with segment b oundaries

have not b een systematically investigated Moreover none of these cues are completely reliable

Walker lists some approaches so far taken in the Centering literature with resp ect to

segmentation

a Whenever Centering rules and constraints are violated there must b e a segment b oundary

b One must nd another way to ob jectively segment a discourse to test hyp otheses related to

Centering

c Any phenomenon o ccurring across discourse segments is not part of Centering and need not

b e accounted for

d Anaphora distribution can b e used to argue for one segmental analysis over another

Walker states that these claims arise from four p ossible relationships b etween Centering and

discourse segmentation implicit in the following contradicting hyp otheses

a Discourse segmentation o ccurs prior to Centering

b Centering is prior to discourse segmentation and constitutes one of the cues for segmentation

c Centering and discourse segmentation are interdep endent pro cesses

d Centering and discourse segmentation are cotemp oral and indep endent pro cesses

Walker suggests that a cannot b e right b ecause pronoun resolution is an online pro cess

and the intention recognition may not b e accomplished until the end of a discourse segment

As Grosz and Linde haveshown the use of pronouns is constrained by the discourse

segmentation Thus it would b e implausible to assume that d is correct The hyp othesis in

alker app ears to capture the relationship b etween Centering and discourse c according to W

segmentation Since her aim is to test Centering Theory and since hyp othesis b works out

for the segmentation and the pronoun resolution at the same time she explores hyp othesis b

limiting her study to discourse initial utterances signaled by the cue phrase now

Walker nds that Centering transitions correlate well with other cues to the discourse

structure Continue transitions are rare at discourse segment b oundaries Smo othShift and

RoughShift b eing the most common transitions Walker suggests that RoughShifts indicate

a transition to a new segment with a distinct set of entities

The underlined utterances discussed in the set of examples in then might indicate a

new subsegment within their resp ective segments However since they are emb edded they do

not blo ck the reference b etween the two utterances that are not linearly adjacent but consecutive

with resp ect to the same discourse segmentcontaining them

The underlined utterances in do not move the narrativeforward unlike the other ut

terances in these discourses In a narrative the default case is to present actions to movethe

narrative ahead Lab ov shows that in a narrative clauses are temp orally sequenced in the

order in which the events o ccur Lab ov calls these temp orally ordered clauses narrative

clauses The order of narrative clauses cannot b e changed b ecause there is a temp oral juncture

between these clauses

Consider and from Lab ov

a This b oypunched me

b and I punched him

c and the teacher came in

d and stopp ed the ght

Or else the plup erfect can b e used to reverse the order

a The teacher stopp ed the ght

b She had just come in

c I had punched this b oy

d He had punched me

Narrative can also b e moved backwards byashbacks cf Webb er and yet another option

is to interrupt talking ab out the action by giving descriptions of certain characters or places in

which stativeverbs are chosen These strategies interrupt the movement of the action in a

narrative Note that in all the utterances except the underlined ones are narrative clauses

while the intervening utterances are not Thus intervening utterances may form a subsegment

prop erly included in narrative segments not blo c king the coreference b etween the preceding and

following narrative clauses These preliminary suggestions should b e tested in future research

To sum up this Section addressed some problems encountered in Centering analysis of naturally

o ccurring data as intervening utterances and determining discourse segmentation The data dis

cussed suggest that pronominalization and Centering rules can shed some light on understanding

discourse segment b oundaries

In the following Chapters in Centering analyses a discourse segment is taken to b e a sequence

of utterances with one common Discourse segment Purp ose

Conclusions

This chapter has investigated ranking Cflists in Turkish discourse wehave seen that the gen

eral rule is sub jects rank higher than anyotherentity The two exceptions to this general rule

are Exp eriencer ob jects of psychological verbs that rank higher than Theme sub jects and the

sub jectivepoint of view that triggers the salience of ob jects as well as the character exp eriencing

the sub jectiveevent Wehave also seen that an entityevoked by a p ossessor noun ranks higher

than the one evoked by a p ossessive construction provided that the entity is an Exp eriencer or

the NP is an inalienable p ossessive construction

Some problems in analyzing naturallyo ccurring data have also b een addressed ie intervening

utterances and discourse segmentation Though p ossible solutions have b een prop osed further

research is needed to understand the interaction b etween segmentation and pronominalization in

narratives and other genres of discourse

Chapter

THE FUNCTIONS OF SUBJECTS

IN TURKISH DISCOURSE

The purp ose of this chapter is to investigate the diering discourse functions of null vs overt pro

nouns vs full NPs in the sub ject p osition in Turkish We discuss the correlation of these dierent

referential expressions with the sp eakers attentional state as expressed by Centering transitions

It will b e shown that the choice of an expression is not arbitrary but rather structurally con

strained when Centering transitions are taken into account Moreover the data suggest that

the choice of an expression sp ecies constraints on the discourse structure and the discourse

structure in turn constrains the sp eakers center of attention as discussed by Grosz and Sidner

The chapter is organized as follows Section presents a description of the data included and

excluded in the analysis Section discusses the general tendencies observed in the choice of third

p erson pronouns or full NPs in the sub ject p osition Section addresses the functions of sub jects

in Continue transitions null sub jects overt pronouns and full NPs in Continue transitions are

discussed in Sections and resp ectively Section is devoted to the functions of full

NPs in Retain transitions and in Section the use of imp ersonal pronouns in these transitions

are given Section is concerned with shifting the center of attention Section is concerned

with the functions of full NPs in RoughShift transitions Section argues that there must b e

twotyp es of Smo othShift based on the facts revealed by the analyzed data Sections

and present the functions of sub jects in these twotyp es of Smo othShifts Finally Section

presents the conclusions

The Data

The data in this study include selected discourse samples from published narratives Two nar

ratives from email messages are also included A total of tokens are analyzed of which

involve imp ersonal pronouns tokens are excluded b ecause they are expressed from the sub jec

tive p oint of view The following considerations are taken into account in the analysis and the

selection of the data

If an NP in the sub ject p osition did not contribute to the Cflist the next available entity

ranked in the Cflist is treated as the Cp For example in a sub ject yazk pityshame is

incorp orated into the verb to form an idiomatic expression it loses its indep endent status and

cannot b e talked ab out anaphorically Selim the ob ject thus is the Cp as the single entityin

the Cflist in a and the transition to b is thus a Retain Note that b contains an imp ersonal

pro sub ject

a Yazk oldu Selime

i

pity happ enPAST SelimDAT

It was a shame ab out Selim

b Ucuncuk ez ameliyat etmek zorunda kalyorlar

i

third time op eration make obliged stayProgPLU

They had to op erate on him for the third time

c Kurtulamyor

i

savePASSNEGPROG

He couldnt b e saved

Imp ersonal pro as in b and b is listed in the Cflist as the Cp The reason is that the

sp eaker has an option of continuing the imp ersonal pro as in c Even though the imp ersonal pro

as a sub ject and Agent is listed as the Cp as a working hyp othesis in the analysis the results show

y under discussion in the next utterance The discourse that it do es not app ear to b e the entit

b ecomes infelicitous if the sp eaker intends the imp ersonal pro to b e the entity under discussion

This is exemplied in d where the infelicity indicates that the do ctors who op erated on Selim

cannot b e at the center of attention The dierence b etween c and d is that in the former but

not the latter the patient Selim is realized byanull genitive sub ject in the NP beyninden from

his brain rather than the do ctors and this is the entity the utterance in c is ab out

a Yazk oldu Selime

i

pity happ enPAST SelimDAT

It was a shame ab out Selim

b Ucuncuk ez ameliyat etmek zorunda kalyorlar

i

third time op eration make obliged stayProgPLU

They had to op erate on him for the third time

cpro Beyninden ur alyorlar

i

pro brainPOSSABL tumor takePROGPLU

They to ok out a tumor from his brain

dproC ok iyi doktorlar

pro very go o d do ctorPLU

They are very qualied do ctors

Further discussion on imp ersonal pro and imp ersonal insan human corresp onding to one in

English is presented in Section

As stated in Chapter not all typ es of null sub jects can alternate with overt pronouns It was

shown that an overt pronoun can only b e used when b oth the pronoun and its corresp onding entity

have the same referent otherwise a null pronoun should b e used For example in a ogretmenler

odas teachers ro om is within the scop e of the quantied NP her okul every scho ol and do es

not denote a particular teachers ro om as a referent Thus the anaphoric link b etween the null

sub ject in b and its antecedent is not that of a coreference they do not realize the same referent

The sub ject in b has to b e n ull it cannot alternate with an overt pronoun although the denite

pronoun it in English is allowed in the same p osition as can b e seen in the translation In these

cases the transition from a to b is not taken into account in the analysis On the other hand

the transition from b to c is included b ecause they realize the identical entity and the null

sub ject can alternate with an overt pronoun

a Her okulda ogretmenler o das ilk katta olur

every scho olLOC teacherPLU ro om rst o or b eAOR

In every scho ol the teachers ro om is lo cated in the rst o or

b Ama o burada ikinci katta

i

but it hereLOC second o or

But here it is in the second o or

cO yuzden o cok scak oluyor

i

That reason it very hot bePROG

Thats whyitissohot

Likewise the anaphoric link b etween the null sub ject in b and kahve coee a nonreferential

expression in a is not considered and the transition not analyzed This is b ecause the indenite

null sub ject cannot alternate with an overt pronoun in b see Chapter for details

a Markette kahvekalmams

MarketLOC coee remainNEGPAST

There was no coee left in the market

b Yarn gelecekmis

tomorrow comeFUTEVID

They said it will come tomorrow

In the selection of data sequences of utterances that contain inferrable entities inferentially

related to another entity are not counted and not analyzed For example in the sp eaker

intro duces a woman Nuriye Bac and then her family memb ers her husband eg eniste brother

inlaw in d which is not marked by p ossessive sux thus there is no overt Cblink unlikeinthe

case of other entities ie buy uk oglu her older son ku cuk oglu her younger son in e and

f resp ectively which are marked with p ossessive morphemes The p ossessive morpheme or the

null genitive sub ject in these cases can b e considered to b e the Cb Note that these entities are

interpreted as Nuriyes sons an entity that is not realized in the immediately preceding utterance

The sp eaker then talks ab out all the memb ers of the family in g The data where reference is

made to such setp oset relationships are excluded in the analysis b ecause the order in which these

inferrable entities are ranked in the Cflist is not obvious For example in c the CbUn realized

by a thirdp erson plural is interpreted as functionally dep endent to the Cb Un Nuriye and

her family Th us one can claim that these twoentities are ab out the same referent and that

the center of attention is still on Nuriye thus the transition from b to c can b e analyzed

as a Continue transition A comp eting suggestion is that even though the entity in c includes

Nuriye it is not ab out her any more and that the center of attention has b een shifted Alternative

explanations along these lines can b e presented to account for the transitions in dg As a result

the analysis of reference b etween inferrable entities is left for a future study and discourses as in

are not used in this analysis

a Koylum bu NuriyeBac

i

Fellowvillager this Nuriye Baci

Nuriye Baci is a fellowvillager

b Biraz uzaktan hsm da olur

i

somewhat distant relativetoobeAOR

She is also a distant relative

cSimdi kasabada oturuyorlar

now townLOC livePROGPLU

Now they liveinthetown

deniste hastanede calsyor

brotherinlaw hospitalLOC workPROG

The brotherinlawworks at the hospital

e buy uk oglu sofor

older sonPOSS chauer

her older son is a chauer

f ku cuk oglu da tornaclk ogreniyor

little sonPOSS to o latheprofession learnPROG

her younger son learns how to use lathes

g Iyiymisler

nehearsayPLU

they are all doing ne

Ote Yakadaki Cennet Talip Apaydn

Some problems encountered in the analysis of naturallyo ccurring data were discussed in Chap

ter These will b e briey summarized here It was p ointed out that there are no ob jective criteria

to determine the lo cal discourse segment whichistaken to b e the domain where Centering rules

applyFollowing Grosz and Sidner intentional structure is assumed to determine segments

and thus a segmentistaken to b e a sequence of utterances that intuitively represent a single

Discourse Segment Purp ose DSP The transitions b etween strictly adjacent utterances are an

alyzed and sequences that consist anyintervening utterances where the reference is b etween the

entities of Un and Un are not analyzed see Chapter for examples

A Preliminary Overview of the Functions of Sub jects

In this section some general observations on the b ehavior of thirdp erson sub jects in centering

are discussed

Table presents the distribution of thirdp erson sub jects with regard to the Centering Transi

tions The distribution of thirdp erson singular sub jects is presented in Tables and thirdp erson

plural sub jects in Table

Table The Distribution of the Thirdp erson Sub jects

SUBJECT CONTINUE RETAIN SMOOTH SHIFT ROUGHSHIFT

Null pronoun

Overt pronoun

Full NP

TOTAL

Table shows that the distribution of referential expressions pattern neatly when Centering

transitions are analyzed The least explicit expression null sub ject is used to enco de Continue

transitions of the time On the other hand no null sub ject is used to enco de RoughShift

a transition where the most explicit expression the full NP is favored of the time Thus

it app ears that Continue and RoughShift app ear to b ehavelike a mirror image as predicted by

Centering Theory the former is the easiest to pro cess and the latter is the most complex It is

conceivable that the least and the most explicit expressions signal varying degrees of complexity

in pro cessing referential expressions However a Retain is predicted to followaContinue in

terms of the ease of pro cessing and yet in b oth a Retain and a RoughShift a full NP enco des

the transitions and of the time resp ectively Smo othShifts on the other hand are

enco ded byanull sub ject of the time but do es not app ear to favor any form of expression as

strongly as other transitions However a closer lo ok at Smo othShift transitions suggest that the

choice of expressions also pattern dep ending on whether the shift is to the Cp or to the nonCp

Each of these observations will b e discussed later

Tables and show that thirdp erson singular and plural sub jects b ehave alike when analyzed

on the basis of Centering transitions and that the discussion ab oveisvalid for b oth thirdp erson

singular and plural sub jects

Table The Distribution of Thirdp erson Singular Sub jects

SUBJECT CONTINUE

RETAIN

SMOOTH SHIFT

ROUGHSHIFT

null pronoun

overt pronoun

full NP

TOTAL

Table The Distribution of Thirdp erson Plural Sub jects

SUBJECT CONTINUE

RETAIN

SMOOTH SHIFT

ROUGHSHIFT

Null pronoun

Overt pronoun

Full NP

TOTAL

Note that in general a null sub ject enco des a Continue transition b oth as singular and plural

thirdp erson pronouns and resp ectively On the other hand a thirdp erson singular

full NP enco des a RoughShift and a thirdp erson plural NP of the time Likewise

along the lines of the discussion presented ab ove Retains are also enco ded by full NP singular

and plural sub jects and of the time resp ectively A Smo othShift is enco ded bynull

sub jects more than of the time

The similaritybetween a Retain and a RoughShift also shows up in the use of imp ersonal

pronouns There are a total numb er of imp ersonal pronouns in the data of which are used

to enco de Retains and to enco de RoughShifts

Sub jects in a Continue Transition

Continue is considered to b e the least complex transition to pro cess In a Continue transition the

center in the two successive utterances are the same and the Cb in the current utterance is also

the Cp By keeping the same Cb as the highest ranked entity the sp eaker signals to the hearer

that the center of attention is on that particular entity b eing talked ab out

The Functions of Null Sub jects in a Continue Transition

The data analyzed in this study show that null sub jects function as the default referential expres

sions for all p ersonal pronouns to enco de a Continue transition as in b elow In the sp eaker

cannot replace any one of the null sub jects with a more explicit expression without disturbing the

coherence of the discourse

Note that otobus bus and dolmus minibus do not contribute to the Cflists in bc resp ectively

a Taksim alanndan yldrm gibi gecti Naile

i

Taksim squareABL thunderstorm like passPAST Naile

Naile passed through Taksim square likethunderstorm

Cb

Cf Naile Taksim Square

b Otobus b ekleyemezdi

i

bus waitABILNEGPAST

She couldnt wait for a bus

CbNaile

Cf Naile

c Dolmus bile bekleyemezdi

i

minibus even waitABILNEGPAST

She couldnt even wait for a minibus

CbNaile

Cf Naile

Cb Cp Continue

d Bir taksi tutmalyd Elli Lira

i

one taxi rentMUSTPAST Fifty Liras

She should catch a taxi Fifty Liras

CbNaile

Cf Naile

CbCp Continue

Given the distribution of the data and exemplied as ab ove as a rst approximation the

function of null sub jects can b e summarized as follows

Null sub jects enco de Continue transitions in the default case

This can b e stated in Centering terms as follows

If CbUn CbUn and if CbUn CpUn a null sub ject is used

This app ears to b e a universal rule as will b e discussed in Chapter However although null

sub jects enco de Continue transitions in the default case certain factors concerning stress fo cus

y require a more explicit expression in Turkish and discourse segment b oundaries ma

Overt Pronouns in a Continue Transition

The tokens of overt pronouns in Continue transitions in the data all represent phonetically promi



nententities ie stressed pronouns Stressed pronouns receivefocus which is dened as a

As p ointed out in Chapter linguisti c fo cus is a distinct notion from the one in Fo cusing Theory

linguistic device that either imp oses phonetic prominence on an entity or up dates the information

structure of discourse ie marks the new information see the references in Chapter In some

of these cases overt pronouns are used in spite of the fact that there is no other comp eting entity

This is b ecause the fo cused information cannot b e null otherwise a violation of a set of discourse

wellformedness rules render the discourse incoherent Kuno As discussed in Chap

ter according to Kuno any constituentcanbenull in languages as in Japanese

and Turkish in yesno questions and parallel constructions so long as the fo cused ie stressed

information is retained

Fo cus is also dened as a variable in an op en prop osition in which the background knowledge

is presupp osed Prince b or entailed Wilson and Sp erb er The fo cus of an

utterance is an instantiation of the variable and constitutes the new information of the op en

prop osition Prince b According to Prague scho ol linguists cf Ha jicova

and Vallduv fo cus represents new information According to Chafe on the other

hand the fo cused entitydoesnothave to represent the new information and has little relevance

to the distinction b etween given and new information

Moser distinguishes twotyp es of fo cus stressfo cus and informationfo cus Stress

fo cus marks a constituentbyrelative phonetic prominence Informationfo cus on the other hand

involves the constituent whichconveys new information or makes the p oint of utterance

Overt pronouns in Continue transitions do not represent the new information but they func

tion as Cbs which represent salient Discourseold entities and they map onto existing le cards

What is new in the discourse though is the relationship of the Cb with resp ect to the rest of the

prop osition Kuno a Prince

Stressed pronouns receive stressfo cus and have a listing reading as dened by Kuno a

ys Moser They derive a set of alternatives memb ers of a set in contextually relevantwa

Namely among Z X and only X is Y When Z the range from whichXischosen is not sp ecied

in the context of such a sentence Z is taken to b e all memb ers of the set of which X is one

Kuno a

This set can b e derived explicitly or implicitly Consider example

a Nevin nisanlsn merak ediyor

Nevin ancePOSSACC curious domakePROG

Nevin is curious ab out her ance

b goru se gelemezmis

i

visitDAT comeABILNEGEVID

she says he couldnt come for a visit

co da kafesteymis bizim gibi

i

he to o cageLOCEVID weGEN like

He to o is in prison likeweare

Cicekoglu Ucurtmay Vurmasnlar

In there is a set of memb ers who are in the prison Nevins ance and the referents of

the rstp erson pronoun in c The overt pronoun which co o ccurs with the fo cus particle de

to oalso elicits an explicit listing reading cf Homan Homan writes the

predicate of the sentence is true for all memb ers of the set that the demarked entity b elongs to

In c b eing in the prison is true for the memb ers Nevins ance and a set of p eople including

the sp eaker Now consider

Munire comes to visit Cavide So on however she starts living in Cavides house p erma

nently She interferes with every household aair

a kilerin anahtarn b eline takms

i

cellarGEN keyPOSSACC waistPOSSDAT attachPAST

She had attached the cellar key to her b elt

b evin idaresini eline almst

i

houseGEN managementPOSSACC handDAT takePASTPAST

She had taken the control of housekeeping

cHizmetcileri o degistiriyor

i

servantPLUACC she changePROG

She is the one who changes the servants

eHer seyin yerini o biliyor

i

CC she knowPROG every thingGEN placePOSSA

She is the one who knows where everything is lo cated

f bakkaln cakkaln hesabn o biliyor

i

gro cerGENmro cer billingPOSS she knowPROG

She is the one who knows ab out the gro cersmro cers bills

Memduh Esendal Hava Paras

Utterances cf derive implicit memb ers of a set there must b e one or more candidates to

fulll housekeeping chores Among these candidates of the list Munire is in an elementset

relation she is not only one of the memb ers of this set p otentially resp onsible for the chores but

also the one who is the least qualied and contrarytoexp ectation one b ecause as a guest she is

not supp osed to b e involved in all these household chores Although null sub jects can felicitously

o ccur in cf this contrarytoexp ectation reading will not b e available unless overt pronouns

are used

In utterances cf the overt pronouns are in the preverbal p osition which marks fo cus in

Turkish Erguvanl Erku Homan In prep In the data analyzed in this studyall

of the overt pronouns in Continue transitions o ccur in the preverbal fo cus p osition and some

co o ccur with fo cus particles as de and only

Based on the discussion wecannow revise the functions of sub jects in Continue transitions

as follows

The Function of Sub jects in CONTINUEs

Null sub jects are used to enco de a Continue transition when the Cb is not a stressed

pronoun ie phonetically prominent

Overt pronoun sub jects can b e used in a Continue transition if the Cb receives phonetic

prominence

The Functions of Full NPs in a Continue Transition

Table shows that of the Continue transitions are enco ded by full NPs In the data analyzed

in this study full NPs app ear to havetwo functions when they enco de a Continue transition

One is to provide new additional information ab out an entityatthecenter of attention and the

second is to mark a discourse segment b oundary

In a the discourse entity Necla is a Discoursenew entityintro duced with a full NP and then

instantiated as a Cb in b by a stressed overt pronoun the asso ciate of the fo cus particle de to o

The transitions to b and c are Continues where the Cb and Cf in the previous and current

utterances are maintained Note that the full complex NP in c conveys additional information

ab out its referent that she was more chic in nature than the other women in comp etition with

her to marry the do ctor Note that c also seems to b e a discourse segment b oundary signaled

by the change in asp ect system and sgift from Neclas action to her attitude A null sub ject may

render a slightly less coherent discourse in the same p osition

The do ctor is an eligible bachelor who is very p opular among young women who try to

marry him

a Necla ortaya ckt

Necla surroundings app earPAST

Necla came into the picture

b Birkac haftadr o da kendisi denedi

A few weekFOR she to o selfACC tryPAST

hances for a few weeks She also tried her c

cYaradls daha sk olan bu kz

naturePOSS more chic b eSP this girl

This girl who was more chic by nature

sanyordu ki kadnlar isterlerse

thinkPROGPAST COMP womanPLU wantAORPLUCOND

thought that if women wanted to

erkeklerin bastan ckmayan olmaz

manPLUGEN headABL leaveNEGSPACC b eNEGAOR

there were no men who couldnt b e seduced

Memduh Esendal Hava Parasi

Another full NP o ccurs in e and it signals a discourse segment b oundary which is also

marked by the shift in tenseasp ect system In ad events are narrated in progressive and in

ef Pigmalyons p ersonality is describ ed in simple present tense The intention of the sp eaker

shifts from narrating events in ad to describing Pigmalyons p ersonality in ef Thus change

in intention Discourse Segment Purp ose divides the discourse into segments Also note that the

comparison of the p ersonality traits of Pigmalyon and Freda triggers a list reading and the sub ject

receives phonetic prominence A null sub ject in e renders the discourse incoherent

a Bir kac hafta gecince Pigmalyon da Murat Aliye alst

i k

one few week passWHEN Pigmalyon to o MAliDAT getusedtoPAST

A few weeks later Pigmalyon b ecame friendly with Murat Ali as well

b Geceleri onun o dasna geliyor

i k

nightPLU his ro omPOSSDAT comePROG

She would come to his ro om at nights

c saatlerce oturup konusuyor

i

hoursfor sitCONJ talkPROG

she would sit and talk for hours

d dertlesiyordu

i

problemRECIPPROGPAST

she would talk ab out her problems

e Pigmalyon Freda gibi degil

i i

Pigmalyon Freda like not

i

Pigmalyon is not likeFreda

f Guluy or kzaryor utanyor

i

smilePROG blushPROG embarrassPROG

She smiles blushes and b ecomes shyembarrassed

Memduh Esendal HavaParas

Discourse anaphora is known to b e sensitive to discourse segment b oundaries Grosz

shows that in English taskoriented dialogs whose structure closely corresp onds to the task b eing

p erformed the center of attention is inuenced by the task structure and that the pronominal

anaphora do not cross subtask b oundaries Walker prop oses that the use of a full NP in a

Continue transition in English in her corpus may also b e due to the sensitivity to the discourse

segment b oundary Wehave seen that this also app ears to b e the case in Turkish discourse

However segmentation includes many complex factors none of which is reliable on its own For

example consider e and e as variants of e again a discourse segment b oundary Note

that although the transition from d to e is a Continue b oth null and overt pronouns in e

are infelicitous However this infelicity can b e eliminated when a conjunction as ama but is

is used as in e Thus other complex factors of discourse coherence app ear to interact with

pronominalization in discourse segment b oundaries

d dertlesiyordu

i

problemRECIPPROGPAST

i

she would talk ab out her problems

e o Freda gibi degil

i i

she Freda likenot

i

she is not likeFreda

e Ama o Freda gibi degil

i

but she Freda like not

But she is not lik eFreda

Likewise in c which is arguably a discourse segment b oundarybothovert and null pronouns

are infelicitous unless an adverb as bugun to day or a conjunction as ama but or b oth are used

This is an indication that pronouns may fail to link utterances in discourse segment b oundaries

and additional links might b e required

a Ali Adanaya gitti

i

Ali AdanaDATgoPAST

Ali has gone to Adana

b Burada degil

i

hereLOC not

He isnt here

cAli o gelecek Bugun gelecek

i i i i

He comeFUT Today comeFUT

He will comeHe will come to da y

As a result then the form of referential expressions can b e an indicator of discourse segment

b oundaries and the distribution of discourse anaphora can b e argued for an analysis of one seg

mentation to another and yet the interaction of other coherence factors such as tense asp ect

mo dality cue words conjunctions etc must also b e considered

With this mind wenow turn to functions of full NPs in Continue transitions Possible Func

tions of full NPs in Continue transitions are as in the following

Full NPs can b e used in a Continue transition to

i represent new information ab out an entity

ii to mark discourse segment b oundaries

In the rst case the full NP is a signal to the hearer to add new information to the corresp ond

ing le card at the center of attention In the second case it signals the hearer that a p ossible

shift o ccurs in intentional structure of discourse

The use of full NPs to enco de Continue transitions app ears to b e more widespread in jour

nalistic genre than narratives This strategy may stem from the motivation to presentasmuch

information as p ossible in limited space A research pro ject comparing the use of full NPs in Con

tinues in journalistic writings and narratives may provide valuable insight on our understanding

of the functions of discourse anaphora presentation of information and discourse segmentation

Retaining the Center

In the Retain transition as in Continue the Cb is the same as in the previous utterance However

the current Cb is not the highest ranked entity since another entityisintro duced in the highest

ranked sub ject p osition CbUn CbUn Cb Cp

It has b een suggested that Retain is a smo oth way of preparing for a Shift to a new entityby

demoting the Cb to a less salient p osition Brennan Friedman and Pollard Even though it

is conceivable that a Retain prepares the hearer for a Smo othShift this is not its single function

Linson rep orts that in his English data a Retain is followed byaContinue of



the time and by a Shift only of the time In a Retain the sp eaker thus mayintend to

continue the Cb but if the Cb is not the p erformer of an action but rather its undergo er

recipient or b enefaciary etc it necessarily has to app ear in ob ject p osition when another entity

is intro duced in sub ject p osition as Agent Except in passive constructions Agents must pro ject

as sub jects in accordance with the Uniformity of Theta Hyp othesis UTAH prop osed byBaker

which states that the hierarchy of thematic relations is uniformly pro jected in the hierarchy

of grammatical relations see also Jackendo Grimshaw and Pesetsky

The Functions of Full NP Sub jects in Retain Transitions

Table shows that a Retain o ccurs with a full NP sub ject of the time and a RoughShift

of the time In a Retain since the Cb is not the Cp it is not the sub ject Since every sentence

has a sub ject some other constituentmust ll that p osition

Turan b rep orts that a Retain transition o ccurs with an overt ob ject of the time



In the Turkish data a Retain is followed by a Continue of the time This prop ortion would b e higher if

the p ossessor were not ranked higher than the p ossessiveNP as previously discussed For example in example

rep eated b elow the NP kafas his head his ranked higher than his head As a result the null sub ject of

the p ossessive NP is considered as the Cp thus the transition from a to b is a Continue If the null sub ject of

the p ossessiveNPwere not considered as the Cp in b then the transition from a to b would b e a Retain Hence

dep ending on the ranking the transition from b to c can b e considered as a Retain followed byContinue or a

Continue followed byaContinue

a Iceri girdi

i

inside enterPAST

he went in

b Kafasi karmakarskt

headPOSS confusedCOPPAST

His mind was confused

c yaz o dasna dogru yurud u

i

writing ro omDATtowards walkPAST

he walked through the study

The choice in ranking in this study decreased the prop ortion of Continues followed by a Retain

and with a null ob ject only of the time in a corpus of naturally o ccurring narratives This

includes instances when the sub ject is either imp ersonal or nonreferential as kimse nob o dy

Since the Cb in a Retain is realized byanovert pronoun most of the time it cannot comp ete

with a null sub ject in the same utterance for Cbstatus unless the latter is an imp ersonal pro

The sub ject in a retained utterance cannot b e an overt pronoun of the same numb er and p erson

b ecause utterances with multiple overt pronouns with the same Phifeatures are very dicult to



pro cess in Turkish

Multiple overt pronouns in the same utterance are not considered to b e go o d style in written

and sp oken language utterances as in b may confuse the hearer who may b e forced to ask for

a clarication question Who didnt see whom The reason is that in b the Cb p osition is

already lled with an overt ob ject the highest ranked entity of the previous utterance and this

excludes the p ossibility of using another pronoun for a less saliententity of the previous utterance

a Saitin yanndan gecti

i k

SaitGEN nextPOSS passPAST

She passed next to Sait

b Ama o Sait onu tanmad bile

k i

but heSait sheACC recognizeNEGPAST even

But Sait didnt even recognize her

To obtain the reading intended in b the sub ject should b e a full NP in accordance with the

rst part of the Center Promotion Rule discussed in Chapter rep eated b elow

Center Promotion Rule for Turkish CPR to b e revised

i Given an entityeevoked in a nonsub ject p osition in Un where e is not the

instantiated CbUn if e is realized in Un it must b e realized by a full NP sub ject

These observations in Turkish discourse supp ort Centering Rule that there must b e at most

one Cb p er utterance In addition it accounts for why full NPs rather than less explicit expressions

o ccur as sub jects in Retain transitions

tro duced in a Retain There are a few p ossibilities for how a sub ject maybein



However note that null sub jects and ob jects co o ccur in answers to yesno questions and also when the sub ject

continues to b e the Cb The following sequence of utterances is felicitous with multiple null expressions However

note that the sub ject is the Cb throughout These null ob jects may turn out to b e pseudozeropronoun s as discussed

by Kuno see Chapter

a Almanyadaki arkadasna bir mektup yazd

i

GermanyLOC friendPOSSDAT one letter writePAST

He wrote a letter to his friend in Germany

b Zarfa koydu

i k

envelop eDAT putPAST

He put it in an envelop e

c kapatt pullad

i k

closePAST stampPAST

He closed and stamp ed it

Ozdemir Basargan Gurbet Sofras

It can b e an entity among the CfUn or inferrable from one of them

Itcanbeanentity that was previously discussed but not in the attentional state b ecause

of the newly intro duced entities in the intervening utterances

It can b e a Hearerold Discoursenew entity

It can b e a Brandnew entity

It can b e a dummylike sub ject imp ersonal pro

All of the options in ad will b e discussed b elow and imp ersonal pronouns will b e discussed

in the next section

Let us assume that a is the case ie the new sub ject is one of the entities in the CfUn

Then it must not b e the CbUn since the singleton CbUn is retained in Un The new

sub ject must also not b e the CpUn since otherwise a Smo othShift would obtain as in c

a parkta yuruy ordu

i

parkLOC walkPROGPAST

She was walking in the park

b Sait onun yanndan gecti

k i

SaitGEN sheGEN nextPOSS passPAST

Sait passed by her

CbSevim CbUn CbUn

Cf SaitSevim Cb Cp Retain

c Ama onu tanmad bile

k i

but Sait sheACC recognizeNEGPAST even

But he didnt even recognize her

CbSait CbUn CbUn

Cf Sait Sevim Cb Cp Smo oth Shift

Under these circumstances any nonCp nonCb of Cf Un must undergo the Center Pro

motion Rule

Bearing this in mind we can now consider where b is a Retain The entityvali the

ma yor is in the Cflist of a As a nonCp nonCb if it is realized in b as a sub ject it has to

b e realized by a full NP in accordance with the CPR

a Valiyle goru smesi ogleden sonraya kald

mayorWITH meetingPOSSS no onABL afterDAT remainPAST

His meeting with the mayor was p ostp oned until the afterno on

Cb his

Cf his meeting mayor

b Vali onu saatu ce dogruca grd

Mayor heACC hour threeDATtowards callPAST

The mayor called him close to three oclo ck

Yasar Kemal DemircilerC arss Cinayeti

Cb his CbUn CbUn

Cf mayor him Cb Cp Retain

Similarly the nonCp nonCb ob ject of the p ostp osition in b has to b e realized by a full

NP if realized in the sub ject p osition in accordance with the CPR while the Cb is retained in

c

a Sevim bir acele giyinip sokaga frlad

i

Sevim one hurry dressedCONJ street runPAST

Sevim quickly dressed and ran out

b Saitin yanndan gecti

i k

SaitGEN nextPOSS passPAST

She passed by Sait

CbsheSevim

Cf sheSait Cb Cp Continue

c Ama Sait onu tanmad bile

i

but Sait sheACC recognizeNEGPAST even

But Sait didnt even recognize her

CbherSevim CbUn CbUn

Cf Sait Sevim Cb Cp Retain

cmiyordu dcunku gozleri se

b ecause eyePLUPOSSSG distinguishNEGPROGPAST

Lit b ecause his eyes couldnt distinguish things

b ecause he couldnt see very well Aziz Nesin Gol Kral

Note that the requirement of a full NP sub ject in a Retain transition is a consequence of the

interaction of the rules and constraints of Centering Theory presented in

If anyentity of CfUn is realized by a pronoun then the CbUn is

There is at most one Cb p er utterance

The Center Promotion Rule CPR

Given that there is a single Cb p er utterance and given that the Cp is the exp ected Cb of

the next utterance any instantiated Cb CpUn cannot b e realized by a full NP in ob ject

p osition For exampleSevim the null sub ject Cb in a cannot b e in a full NP in ob ject p osition

in b it must b e represented with a less explicit expression Since null ob jects are constrained

in Turkish discourse cf Turan b it has to b e realized byanovert ob ject pronoun As

predicted full NP ob ject in b is incoherent while the overt pronoun ob ject is felicitous

a Saitin yanndan gecti

i k

SaitGEN nextPOSS passPAST

She passed by Sait

b Ama Sait Sevimionu tanmad bile

i

but Sait sheACC recognizeNEGPAST even

But Sait didnt even recognize her

Likewise Grosz Joshi and Weinstein show that in English the instantiated CbUn

cannot b e realized with a more explicit expression in Un Example is cited in Grosz Joshi

and Weinstein

a Susan gaveBetsy a p et hamster

i k

b She reminded her that such hamsters were quite shy

i k

c She told Susan that she really liked the gift

k i

Utterance c is infelicitous b ecause Susan an instantiated Cb in b is realized by a full NP

This example suggests that there is at most one Cb p er utterance and that sp eakers cannot use

more explicit expressions once the Cb is instantiated unless the full NP has sp ecial functions as

discussed in Section

The discussion of the Turkish data has implications for the p ossible universal relevance of

Centering TheoryWe suggest that the CPR applies crosslinguisticall y dep ending on the avail

able distinguishing expressions in the sense of Dale Supp ort for this claim is provided in

Brennans English data as discussed in Chapter

However in a language like English whichmakes use of gender distinctions in pronouns it is

predicted that the CPR will apply in a more limited set of contexts than in Turkish ie only where

the two expressions are of the same gender p erson and numb er On the other hand Turkish a

language without gender distinctions has to use distinguishing expressions to resolveambiguity

among the memb ers of the Cflist of the same p erson and number Thus the most saliententity

will b e represented with a null sub ject or a nullovert ob ject while less saliententities will b e

realized by NPs with more distinguishing p ower Full NPs typically have more descriptiveand

distinguishing p ower than pronouns and thus the CPR follows Various expressions maybe

distinguishing relative to the other set of expressions in the same Cflist For example the NP the

young woman after the sentence Mary and Nancy met can b e distinguishing if one of the women is

younger but the NP the woman fails to distinguish the entities assuming b oth are adult females

Pronouns can also have distinguishing p ower for example is a coherent discourse b ecause

Mary and John can b e distinguished by the pronouns she and he resp ectivelyInTurkish on

the other hand pronouns cannot mark such a distinction and the utterance corresp onding to c

with twoovert pronouns is very dicult to pro cess As a result Turkish has to make use of the

distinguishing p ower of referential expressions

a Mary quickly got dressed and ran outside

i

b She passed byJohn

i k

c But he didnt even recognize her

k i

d Because his eyesight is so p o or

k

Compare with where c which is as infelicitous as its Turkish counterpart Note that

c is a violation of the CPR

a Mary quickly got dressed and ran outside

i

b She passed by Sue

i k

c But she didnt even recognize her

k i

d Because her eyesightissopoor

k

The coherentversion where Sue is realized by a full NP ob eys the CPR c is like the situation

in Turkish in general b ecause it contains entities that are not distinguished by the English gender

system

a Mary quickly got dressed and ran outside

i

b She passed by Sue

i k

c But Sue didnt even recognize her

k i

d Because her eyesightissopoor

k

Thus it is predicted that the CPR will b e a less common strategy in English than in Turkish

and that the prop ortion of sub ject pronouns in Retain transitions will b e higher in English

Although Turkish do es not make use of gender other Phifeatures as numb er and p erson are

marked and can therefore make pronouns distinguishing expressions Thus as exp ected a null

sub ject can indeed enco de a Retain transition as in b

a Sen Filizi tanmazsn

i

you FilizACC knowNEGA ORSG

You dont know Filiz

b Sizin kogusa yeni geldi

i

youGENPL wardDAT new comePAST

Shes new in your ward

Cicekoglu Ucurtmay Vurmasnlar

Nowwe return to the other functions of full NP sub jects As was stated in b an entity that

has left attentional state may o ccur as the sub ject of a Retain Since sp eakers have a limited center

of attention Chafe Grosz and Sidner Ha jicova previously centered entities

disapp ear from attentional state with time as new entities are intro duced and the discourse no

longer involved the previous entitites For example of the twoentities intro duced in a only

one is realized in b and since Sevim is absent from the Cflist of b the entityhastobe

reintro duced as a full NP in c

a Sait saygyla Sevimi goturd u

i k

Sait resp ectWITH SevimACC bringPAST

Sait to ok Sevim to her seat resp ectfully

b Simdi ne yapmas gerektigini bilmiyordu

i

now what doINFSG needPOSSACC knowNEGPROGPAST

He didnt know what he should do next

cSevim onu bos bir masayagoturd u

i

Sevim heACC empty one tableDAT takePAST

Sevim to ok him to an empty table

Aziz Nesin Gol Kral

Furthermore again as suggested in cd sp eakers mayintro duce Discoursenew and even

Hearernew entities into the discourse mo del while retaining a Discourseold entity In the default

case a full NP is used to intro duce such new entities into the discourse This is b ecause it has the

most descriptivepower compared to null and overt sub jects Thirdp erson pronouns have little

descriptivepower they make animacy gender p erson and numb er distinctions in English and

p erson and numb er distinctions in Turkish Void of any further con tent a new entity cannot b e

intro duced with a pronoun In the case of a Hearerold entity the hearer knows ab out the entity

but it is not in hisher attentional state and so it must b e broughtinto the discourse mo del bythe

use of a full NP On the other hand the hearer do es yet not haveany le card for a Hearernew

entity whichthus has to b e evoked by a full NP Inferrables are also intro duced with full NPs in

the default case

Due to these reasons in where the son of the laundry woman is at the center of attention

the evoked entity komsular the neighb ors has to b e represented by a full NP while the Cb of a

is retained byanovert pronoun

a Bir kimsesiz camasrc kadnn oksuz u idi

i

one anyonewithout launderer womanGEN orphanPOSS COPPAST

He was the orphan son of a laundry woman with no family

b Komsular ona acyp barndrmslard

i

neighb orPLU heDAT feelpityCONJ takecareofPLUPAST

The neighb ors felt sorry for him and to ok care of him

Memduh Esendal Hava Paras

The functions of Full NP Sub jects in Retain Transitions in Turkish can b e summarized as

follows

Full NP sub jects are used in Retain transitions in the default case

This is due to the following facts

i The sub ject of the current utterance is crucially not the Cb in Retain transitions otherwise

a Continue or a Smo othShift would obtain Since null and overt sub jects are reserved for the

Cbs the nonCb sub ject will tend not to b e represented by these Cbenco ding forms

ii The sub ject in the retained utterance

a may p ossibly b e realized in the previous utterance but not in sub ject p osition in which case it

must undergo the CPR while the CbUn must b e realized by a pronoun in Un if it is realized



in that utterance

He met John

John greeted him RETAIN

b Otherwise a Discourseold entitymay p ossibly not b e realized in the Cflist of the previous

utterance and this entity has to b e intro duced in to the discourse with a full NP

He met John

He also met Sandy

John told him RETAIN

Alternativelyitmay p ossibly b e intro duced into the discourse for the rst time as a Discourse

new entity in which case it must b e represented by a full NP

He wasatamovie

Aguysaw him RETAIN

However full NPs are not the only typ e of sub jects in Retain transitions Table ab ove shows

that Retains in the corpus havenull sub jects of the time One example is presentedine

aCo cuk one dogru bir adm att

i

child frontDATtowards one step takePAST

The child to ok a step forward

b Bu sk kyafetli amcadan korkmus olmalyd

i k

T scaredEVID b eMUSTPAST this chic outtWITH uncleDA

He must havebeenintimidated by this man with an elegant outt

cCevdet Bey onu en son alt yl once gormu stu

k i

Cevdet Bey heACC most last six year ago seePASTPAST

Cevdet b ey had last seen him six years ago

Cb the child CbUn CbUn

Cf CBeychild CbUn CpUn Retain

d O zamanu c dort yasnda gozuk uy ordu

i

that time three four yearLOC seemPROGPAST

He seemed to b e three or four years old at that time



Consider the following example where John undergo es the CPR in utterance b Since utterance a is assumed

to b e in discourse initial p osition it do es not haveany Cb the transition from a to b is not analyzed In utterance

b the Cb is instantiated

a Sam met John

i k

b John greeted him

k i

Cb the child CbUn CbUn

Cf child CbUn CpUn Continue

e Yanagn oksad

k

cheekPOSSACC caressPAST

He caressed his cheek

Orhan Pamuk Cevdet Bey ve Ogul lar

Cb the child CbUn CbUn

CfCBeychild CbUn CpUn Retain

The reason why e is analyzed as a Retain is b ecause even though the null sub ject representing

Cevdet Bey is not represented in the Cflist of d with a null or overt pronoun d is an utterance

representing Cevdet Beys inner thoughtaboutthechild Thus Cevdet Bey is the sub ject of

consciousness and b oth his SELF and the child are equally available to him from his sub jective

p ointofviewFor this reason Cevdet Bey can b e realized with a null sub ject in e whichisa

Retain

The Functions of Imp ersonal Sub jects in Retain Transitions

As previously mentioned a total number of tokens of imp ersonal sub jects are attested in the

data of which o ccur in Retain transitions and in RoughShifts In other words of Retain

transitions have an imp ersonal pronoun These imp ersonal sub jects and their functions in

Retain will now b e discussed

Turkish has an imp ersonal null pronoun which is a pure pronominal pronominal anaphor

with only a human reading with features thirdp erson plural verbal agreement in a tensed

clause It is always represented byanull sub ject the overt sub ject with the corresp onding Phi

features ie thirdp erson plural never having an imp ersonal interpretation In a nonimp ersonal

olve more than one reading with a null thirdp erson plural the sub ject is understo o d to inv

individual but this prop erty is absent in the imp ersonal reading where one or more individuals

are involved

In addition Turkish has another imp ersonal sub ject insan human corresp onding to one in

English Some dierences b etween imp ersonal pro and insan one are presented b elow

Imp ersonal pronouns in Turkish as in other prodrop languages like Italian and Spanish do

not app ear as sub jects of a passive construction cf Belletti Suner

Dun pro tedavi edildil er

Yesterday therapymakedoPASSPASTPLU

They were treated yesterday

Likewise they cannot app ear as sub jects of ergative constructions

pro Burada iyi calsrlar

hereLOC well workAORPLU

They work well here

pro Ksn tatile gitmezler

winter holiday goNEGAORPLU

They dont go for a vacation in winter

The use of imp ersonal pro then is conned to active transitive tensed clauses In contrast

insan human with an imp ersonal interpretation can o ccur in these constructions with third

p erson singular agreement morphology asinand

Insan burada iyi calsr

human here well workAOR

One works well here

Insan ksn tatile gitmez

human winter holidayDAT goNEGAOR

One do esnt go for a vacation in winter

Imp ersonal means denoting any one or more memb ers of a set whose referential identityis

irrelevant since either the sp eaker do es not know who the referent is or else she simply is not

really talking ab out the referent but rather ab out another entity In fact no Continue transition

o ccurs in the data with imp ersonal sub jects Thus it seems that the use of an imp ersonal

pronoun do es not prepare the hearer to shift to the newly intro duced Cp the imp ersonal sub ject

the identity of the referent b eing irrelevant

Imp ersonal insan like imp ersonal pro do es not representanyentity for the sp eaker or the

hearer One distinction b etween imp ersonal pro and insan concerns the interpretation of the

er sub ject Like in Spanish Suner Jaeggli Turkish imp ersonal pro excludes the sp eak

from the action denoted by the verb while this is not the case for insan human where the

sp eaker may b e included in the interpretation

We can now examine instances of imp ersonal pro in the data The imp ersonal pro in utterance

c do es not intro duce a discourse entity and c is still ab out murekk epci the ink seller who

undergo es the action of b eing buried The center of attention is not shifted to the imp ersonal pro

in the sub ject p osition

a Gun un birinde murekkepci on b esgun hasta yatt

i

DayGEN onePOSSLOC inkseller ten vedaysick layPAST

The ink seller was sick for fteen days

b ve oldu

i

and diePAST

and one day he died

cpro Onu mahalle camiinin bir kosesine gomduler

i

heACC district mosqueGEN one sidePOSSDAT buryPASTPLU

They buried him on one side of the district mosque

Memduh Esendal Hava Paras

Utterance in b also has an imp ersonal pro sub ject where the Cb is the overt ob ject Note

that the transition from b to c continues the previous Cb and the null sub ject in c is not

interpreted as imp ersonal pro but as the Cb onu him of the previous utterance

a Evin kucuguydu

i

houseGEN youngPOSS

He was the youngest in the house

b pro Sayrl Babayla basbasa brakvermislerdi onu

i

pro Sayril fatherWITH headtohead leavePASTPLUPAST heACC

They left him alone with Father Sayril

c Babayla ilgilenmek zorundayd Fuat Altnsoy Hendekler

i

fatherWITH interestINF obligedCOPPAST

He had to take care of the father

These facts show that imp ersonal pro is not salient and it neither interacts nor comp etes with

the overt ob ject for Cbstatus As we see in e insan like imp ersonal pro app ears to havea

very low ranking in the Cfhierarchy no doubt for the same reasons Indeed in f the sp eaker

go es on talking ab out Korkut rather than insan Furthermore f shows that the null sub ject is

understo o d as Korkut and not as insan This suggests that an entitylowered to ob ject p osition

by the intro duction of an imp ersonal sub ject has still not left the center of attention

a Ap ollana b enzemiyordu

i

Ap ollonDAT lo oklik eNEGPROG

He didnt lo ok like Ap ollon

b Hermese b enzemiyordu

i

HermesDAT lo oklikeNEGPROG

He didnt lo ok like Hermes

c Korkuttu oyleydi

i i

KorkutCOPPAST like thatCOPPAST

He was Korkut he was like that

d ve hep oyle kalacakt

i

and always likethat stayFUTPAST

and he would always b e like that

e Gelgelelim yardim edemiyordu insan ona

i

however help domakeNEGPROGPAST human heDAT

However one couldnt help him

f Sevgisi bir yara gibiydi

i

lovePOSS one wound likeCOPPAST

His lovewas like a pain

f S urekliuz ul uy ordu

i

always feelbadPROGPAST

He was always sad

Selim Ileri Cehennem Kralicesi

Thus wehave seen that neither imp ersonal pro nor insan prepare the hearer for a p ossible

Shift since these do not b ecome Cps Rather they indicate that the Cb although no longer the

sub ject has not necessarily left the center of attention If a sp eaker intends to shift the Cb we

do not nd these imp ersonal pronouns but rather a new discourse entity

The Functions of imp ersonal pro and insan in Turkish are as follows

Imp ersonal pro and insan may b e used as sub jects in a Retain when the Cb is inuenced

by an action p erformed by human Agent whose identity is irrelevant

Consequently then imp ersonal sub jects must b e ranked very low in the Cflist or they can

b e considered analogous to nonreferential expressions which are not listed in the Cflist

Shifting the Center

Shifting from one Cb to another is a signal to the hearer that the attentional state has changed from

the previously instantiated Cb to a new discourse entity In b oth Smo othShifts and RoughShifts

the Cbs of the two adjacent utterances are not the same The dierence b etween a Smo othShift

and a RoughShift is that the Cb is necessarily the sub ject in the former and not the sub ject

in the latter A Smo othShift is an indication that the sp eaker wishes to talk ab out the newly

intro duced entityby assigning it the highest salience in the current utterance In a RoughShift

on the other hand the sp eaker has no such desire to continue to talk ab out the entityWalker

Iida and Cote

In a Smo othShift the Cb and the Cp of the current utterance is necessarily the same On

the other hand in a RoughShift the Cb is not the Cp in the current utterance The following

example from BrennanFriedman and Pollard shows how a Smo oth and a RoughShift

dier in the last utterance with multiple ambiguous pronouns

a Brennan drives an Alfa Romeo

Cb Cf Brennan Alfa Romeo

b She drives to o fast

Cb she Brennan Cf Brennan

c Friedman races her on weekends

Cb her Brennan Cf F riedman Brennan Retain

d She often b eats her

Cb She Friedman Cf Friedman Brennan SShift

or Cb her Friedman Cf Friedman Brennan R Shift

Utterance d in this example has to b e a Shift since the Cb of the current utterance is

the highest ranked element of the previous utterance that is realized in the current utterance

In an earlier formulation of Centering Theory Grosz Joshi and Weinstein there is no

distinction b etween the twotyp es of Shifts a Shift would o ccur whenever the Cbs of the two

utterances are not the same and the denition of Shift do es not include whether the CbUn is

also the CpUn Brennan Friedman and Pollard argue that the Cp status in the current

utterance is imp ortant in the case of Shift as well as in Continue and Retain

The Functions of Full NP Sub jects in RoughShift Transitions

Twotyp es of RoughShifts are considered in this study One typ e involves those utterances with

a set of new entities and therefore with no Cb at all For example the transition from b to c

is a RoughShift since none of the entities intro duced in ab are realized in c

a Ertesi sabah evin kaps calnd

i

Next morning houseGEN do orPOSS kno ckPASSPAST

The next morning the do or of the house was kno cked

b Actm

i

op enPASTSG

I op ened the do or

cSinemadan bir adam geldi

cinemaABL one man comePAST

A man from the cinema came

UlkuT amer cited in Ozdemir

The other RoughShift involves transitions with a Cb as in b The Cb in a the null

sub ject is not the Cb of b In b kapy the do or is the Cb an entity from the Cflist of

a but is not the Cp thus a RoughShift obtains

a Kapy vurdu b ekledi

do orACC kno ckPAST waitPAST

He kno cked at the do or and waited

b Kapy Zeynep Hanm act

do orACC Zeynep Hanim op enPAST

Zeynep Hanim op ened the do or

Orhan Pam uk Cevdet Bey ve Ogul lar

The reason why full NP sub jects enco de RoughShift transitions is exactly the same as why

they enco de Retain transitions In b oth a Retain and a RoughShift the Cbs cannot b e the

sub ject The dierence is that in a Retain the CbUn is the same of the CbUn while in

RoughShift the Cb of the previous utterance should not b e the Cb of the current utterance

Since the Cb if there is one is not the sub ject in a RoughShift either a new entityisintro duced

into the discourse mo del or else one that has left the attentional state is reintro duced Again as

in Retains imp ersonal pronouns can o ccur in RoughShift transitions for the same reasons

TwoTyp es of Smo othShifts

It will b e shown b elow that sp eakers cho ose distinct forms of sub jects dep ending on whether they

shift to the previous sub ject or to the ob ject in Smo othShift transitions The data suggest that

null sub jects enco de Shifttothesub ject Shiftsub j while overt pronouns and full NPs enco de

Shifttotheob ject of the previous utterance Shiftob j The results are shown in Table b elow

Null sub jects are used in Shiftsub j transitions of the time whereas overt pronouns are used

only of the time On the other hand shiftob j disfavors null sub jects as will b e discussed

b elow

Utterance in c is a Shiftsub j while the one in c is a Shiftob j Note that in b oth c and

c the Cbs Zeliha Hanm and Temel resp ectively are not the same Cbs in b and b and the

Cbs of c and c are the Cps As a result b oth transitions are Smo othShifts The dierence

however is that c shifts to the sub ject of b while c shifts to the ob ject of b

a yine merdivenleri her zamanki alskanlgyla gurult uyle indi

i

again stairsACC all timeCONJ habitPOSSWITH noiseWITH godownPAST

He wentdown the stairs noisily which has b een his usual habit

b Zeliha Hanmk onu gulumsey erek kaslad

i

Zeliha Hanim heACC smilingly meetPAST

Zeliha Hanim met him smilingly

c Kahvaltnn hazr oldugunu soyledi

k

breakfastGEN ready b eNOMACC sayPAST

She said the breakfast was ready

Orhan Pam uk Cevdet Bey ve Ogul lar

Temel is traveling from one city to another on a bus However he sits in the wrong seat

and another guy tries to convince him to change his seat

a gider

i

goPRES

he go es

b Temelin kulagna bir seyler soyler

i k

TemelGEN earPOSSDAT one thingPLU sayPRES

he whisp ers something into Temels ears

cTemel hemen kalkar

k

Temel immediately standupPRES

Temel immediately stands up

d ve gider biletteki numaraya oturur

k

and go es ticketLOC numberDAT sitPRES

and he go es sits in the seat numb ered on his ticket

from an email message received on December



Shiftob j exemplied in c can b e further divided into twotyp es Shiftob j disfavors null

sub jects of the time and favors a more explicit expression overt pronouns or full NPs

Shiftob j represents instances where the entities in the Cflist cannot b e distinguished

from one another by the use of a null sub ject shiftob j is for the remaining cases

Consider In a the twoentities in the Cflist the null sub ject and the ob ject have b oth

thirdp erson singular features The Shift transition shiftob j in b must b e enco ded either

by a full NP or an overt pronoun the null sub ject is infelicitous b ecause it will b e interpreted as

the Cp of a A null sub ject in Turkish is reserved to realize the Cp of the previous utterance

or to enco de a Continue transition On the other hand more explicit sub jects the overt pronoun

and the full NP realize less saliententities nonsub jects and nonCbs

a Aliyi arad

i k

AliACC callPAST

He called Ali

b Ama Ali o evde degildi

k k k

but he homeLOC notCOPPAST

But he was not at home

Nevertheless as Table shows the data suggest that null sub jects can b e used to enco de

a Shiftob j dep ending on the distinguishing p ower of the NPs in the same Cflist Phifeatures

suchasnumb er and p erson have distinguishing p ower in Turkish hence when the entities in the

Cflist are of dierentnumb er and p erson Shiftob j is enco ded byanull sub ject of the time

as can b e observed in the table

Table rdp erson Singular and Plural Sub jects in Smo othShift Transitions

Sub jects Shiftsub j Shiftob j Shiftob j

rd Person Singular

null

overt

Plural NP

TOTAL

In Tables and the distribution of the thirdp erson singular and plural sub jects are pre

sented resp ectively Note that thirdp erson plural sub jects are rare Nevertheless b oth third

p erson singular and plural sub jects show the tendencies discussed ab ove Null sub jects are used

nearly of the time to enco de a Shiftsub j while more explicit expressions are preferred for

Shiftob j In these cases null sub jects are used only of the time in thirdp erson singular and

no null sub ject is attested in the thirdp erson plural sub jects The p ercentagesofnull sub jects

increases when the entities in the Cflist are marked for a dierentnumb er or p erson In that

case null thirdp erson sub jects enco de Shiftob j of the time

Table Thirdp erson Singular Sub jects in Smo othShift Transitions



A total numb er of tokens of null sub jects were excluded b ecause even though they app ear to b e shifttothe

ob ject at rst sight they are ranked from the p oint of view of a sub ject of consciousness as discussed in Chapter

Sub jects Shifttosub ject Shifttoob ject Shifttoob ject

null

overt

full NP

TOTAL

Table Thirdp erson Plural Sub jects in Smo othShift Transitions

Sub jects Shifttosub ject Shifttoob ject Shifttoob ject

null

overt

full NP

TOTAL

Null Sub jects in Shiftsub ject

Shiftsub j is enco ded byanull sub ject in the bulk of the data Thus we can conclude that this

is a general tendency suggesting that the null sub ject is used to realize the Cp of the previous

utterance

Smo othShiftsub j is analogous to Continue in that b oth favor null sub jects Why should this

b e It is true that in b oth Continue and Shiftsub j the Cps and the Cbs are the same in the

current utterance ie CbUnCpUn But this is also the case in Shiftob j Thus this is not a

distinctive feature of Continue and Shiftsub j transitions What is distinctive of these transitions

is the following the CbUn has to realize the CbUn the most saliententity which is exempt

from the Center Promotion Rule in b oth continues and shiftsub js The generalization is

Anull sub ject is used when its corresp onding entity in the Cflist of the Un is

exempt from the CPR

We can now add this observation to the functions of null sub jects previously discussed

The function of null sub jects in Turkish

Null sub jects are used to enco de Continue and Smo othShiftsub j transitions ie they

pickuptheentities that are exempt from the CPR ie null sub jects pick up either

Cps or Cbs

Full NPs and Overt Pronouns in Shiftob j

The second typ e of Smo othShift involves Shiftob j as exemplied in b The Centering analysis

of the transition from a to b is a Smo othShift b ecause the Cb the null sub ject in a is not

the same Cb in b kap the do or The Cb of b is the Cp at the same time The shift is to

the ob ject of a

a Melihann kapsn cekiyor

k i

MelihaGEN do orPOSSACC pullPROG

He pulls Melihas do or

b Kap kapanyor

i

do or closePROG

The do or closes

cMeliha yatt

Meliha liePAST

Meliha laydown

Selim Ileri Destan Gonul ler

The reason why full NPs enco de these transitions is clear when the CPR is considered Note

that the prop ortion of the full NPs that enco de such transitions are high of the time

Shiftob j are also enco ded byanovert pronoun of the time as seen in Table These

are exemplied in b and b In neither utterance can the sub jects b e represented byanull

pronoun b ecause a null sub ject would unambiguously pick up the Cp of the previous utterances

obtaining a Continue instead of a shift to a dierent Cb

a Bagslanmay dilemeliydi Ekremden

i k

pardonPASSINFACC askMUSTPAST EkremABL

He should ask Ekrem for ap ologies lit to b e pardoned

b O aldrssz gorunerek

k

he careless seemADV

Seeming careless he Ekrem

satrancoynamaya devam ediyordu

chess playINFDAT con tinue doPROGPAST

was continuing to playchess

Selim Ileri Bir Aksam Alacas

a Co cuklarn citlerin dibine

i k

childPLUPOSSACC fencePLUGEN b ottomPOSSDAT

They laid their children down by the fence

battaniyelerin ust une yatrmslar

blanketPLUGEN onPOSSDATlayPASTPLU

on the blankets they laid

b Agzlarna yalanc emzikleri vermisler

i k

mouthPLUPOSSDAT pacierPLACC givePASTPLU

They put paciers in their mouths

c Onlar da du surmu sler emzigi

k k

they to o dropPASTPLU pacierACC

They the children dropp ed the paciers

d aglyorlar

k

cryPROGPLU

They are crying

Fakir Baykurt BarsCoregi

tity in b is not realized with a full NP in accordance with One mightwonder why the en

the CPR There are three reasons for this First of all although the CPR is a very common

strategy in Turkish discourse as has b een discussed so far it is merely a tendency rather than an

absolute rule The second reason is that there is no other overt pronoun in the same utterance

that is comp eting for Cbstatus The third reason is that Ekremden from Ekrem is in sentence

nal p osition a place where only Discourseold entities are allowed Erguvanl Erku

Homan This p ostp osing indicates that the entity is not evoked into the discourse mo del

for the rst time but has b een in the discourse although not as the most saliententity The CPR

might apply to the entities evoked into the discourse for the rst time

To sum up wehave seen that Shiftob j should b e treated separately from Shiftsub ject This is

motivated by the correlation observed b etween the choice of expression and the typ e of transition

Wehave seen that Shiftob j is enco ded by a full NP in accordance with the CPR or byanovert

pronoun sub ject when there is no pronoun comp eting for Cbstatus in the same utterance For

ShiftSub js the iii part of the CPR can b e extended to include overt sub ject pronouns as follows

Center Promotion Rule for Turkish CPR revised

i If an entitye isevoked in a nonsub ject p osition and is not an instantiated CbUn

emust b e realized by a full NP sub ject in Un

ii Entity e should b e expressed by a full NP in Un b efore it can b e talked anaphori

cally in Un

iii Entity e can b e realized byanovert pronoun only if there is no other overt pronoun

in the utterance that is comp eting for Cbstatus

The discourse functions of full NP and overt pronoun sub jects in Shiftob j transitions is

summarized b elow

The Discourse Functions of full NPs in Shifttotheob ject Transitions

A full NP sub ject o ccurs in a Shiftob j if it is evoked as a nonCpUn

An overt pronoun sub ject o ccurs in a Shiftob j if there is no other pronoun in Un

comp eting for the Cbstatus and if it is Discourseold in Un

Null Sub jects in ShiftOb j



Smo othshiftob j is enco ded bya null sub ject of the time as shown in Table In these

cases the p erson and numb er Phifeatures distinguish the entities in the Cflist of the utterance

as in In a the Cflist contains a rstp erson singular and a thirdp erson singular entity

Since the entity in ob ject p osition has a distinguishing p ower relative to the sub ject it can b e

realized with a null sub ject in b

a Co cukluk asklar unutulmaz asl diyorum Melihaya

i

childho o d lovePLU forgetPASSNEGAOR in fact sayPROGSG MelihaDAT

I tell Meliha that childho o d loves cant in fact b e forgotten

b Inanmyor

i

b elieveNEGPROG

She do esnt b elieve itme

c Pek cok duygularma inanmyor

i

many feelingsPOSSSGDAT b elieveNEGPROG

She do esnt b elieve manyofmy feelings

Selim Ileri Destan Gonul ler

First and secondp erson pronouns are indeed dierent from thirdp erson pronouns in that

they are situationally evoked and their values are assigned dep ending on the sp eaker and the

addressee To b e more precise rst and secondp erson pronouns do not need antecedents within

the discourse mo del to b e interpreted the way thirdp erson pronouns do

An utterance is not pro duced in a vacuum but in a particular lo cation at a particular time

According to Kaplan an utterance is uttered at a particular time t in a particular context

c The context includes a sp ecication of the sp eaker of the utterance the audience the time

and the place of the utterance Ufs a t pg The interpretation of the rstp erson I and weand

the secondp erson you dep ends on the context of utterance Thus it makes sense to treat the

context of an utterance separately and to suggest that these indexical elements do not interact

with the ranking of textually evoked thirdp erson pronouns or full NPs Furthermore it seems



In the data tokens involved entities that are ranked from the sub jective p oint of view of a sub ject of

consciousness as discussed in Chapter They are not represented in any of the Tables These entities do not

undergo the CPR b ecause the entity in ob ject p osition is the Cp from the p oint of view of the individ ual in the

sub ject of consciousness and Cps are exempt from the CPR

that due to these separate levels of ranking rst and secondp erson pronouns in sub ject p osition

do not blo ck the next highest ranked entity from b eing realized as a null sub ject in the subsequent

utterance Perhaps they are elements of global discourse and not b e listed in the Cflist at all as

suggested byWalker

However rst and secondpronouns interact with thirdp erson pronouns when b oth are stressed

as in Note that the rstp erson singular pronoun b en I in c cannot b e null b ecause the

sp eaker compares himself to individuals who have a prop erty of b elieving in fake virtue

a Bilirim bu sozlere sinirlenecek

KnowPRESSG this wordPLUDAT b eangryRELFUT

saysz insan ckacaktr

numb erless human app earFUT

I know many p eople will b e upset by these words

bC unku onlar yalanl dolanl

Because they lieWith deceitWith

sahte namuslara inanrlar

fake virtuePLUDAT b elievePRESPLU

Because they b elieveinfake virtues full of lies and deceit

cBen iktisad yonuyle

I economy sideWith

duygusal yonuyle samiyetteki namusa inanrm

emotional sideWith sincerityLOC virtueDAT b elievePRESSG

I b elieve in the sincerity of virtue with its economic and emotional sides

If there were no interaction b etween the situationally evoked rst and secondp ersonpronouns

and textually evoked thirdp erson pronouns the pronoun in c could b e null Further research

is needed to determine how rst and secondp erson pronouns shoule b e treated in the global

discourse

Nevertheless the data suggest that the distinguishing p ower of Phifeatures has an imp ortant

eect on the choice of expression in the case of Shiftob j Not only are rst and secondp erson

pronouns distinct from the thirdp erson sub jects in numb er and p erson but also thirdp erson

pronouns can often b e distinguished from one another bynumb er As a result the thirdp erson

plural in b is distinct from the thirdp erson singular and therefore do es not have to undergo

the CPR In such cases a Shiftob j can b e enco ded byanull or an overt sub ject as in c

a Ali butun gun yemek yapt

k

Ali all dayfood co okPAST

Ali co oked all day

b Filizle Ayhan yemege cagrd

k i

FilizWITH Ayhan dinnerDAT invitePAST

He invited Filiz and Ayhan to dinner

c Onlar Cok mesgulduler

i i

they very busyCOPPLU

They were very busy

d Oste calstlar yemege gidemediler

i

oceLOC workPASTPLU dinnerDAT goNEGABILPASTPLU

They worked in the oce they couldnt go to the dinner

In b the overt pronoun is more coherent than a null sub ject however a null sub ject is also

p ossible

a Tarkla Murat oldukca pahal

i k

Tarikwith MuratACC rather exp ensive

vited Tarik and Murat to a rather exp ensive he in

ama b egenisiz bir lokantaya cagrdi

but tasteless one restaurantDAT invitePAST

but tacky restaurant he invited

b Onlar gitmek istemediler

k k

They goINF wantNEGPASTPLU

They didnt wanttogo

c Haklydlar

k

rightCOPPASTPLU

They were right

Selim Ileri Her GeceBodrum

As was stated in Chapter the thirdp erson plural mopheme is optionally marked in Turkish

If it is not present as in b a variantofbthenthenull sub ject is not allowed an overt

pronoun must b e used This is b ecause in b the agreement features cannot distinguish the

entities in a

a Tarkla Murat oldukca pahal

i k

Tarikwith MuratACC rather exp ensive

he invited Tarik and Murat to a rather exp ensive

ama b egenisiz bir lokantaya cagrdi

but tasteless one restaurantDAT invitePASTPAST

but tacky restaurant he invited

b Onlar gitmek istemedi

k k

they goINF wantNEGPAST

They didnt wanttogo

By the same token a null sub ject in b renders the discourse incoherent b ecause the third

p erson singular entities cannot b e distinguished by the thirdp erson singular verb morphology

This is so even when world knowledge can resolve the ambiguity in this case where we assume

that the p erson who did not wanttogomust b e the one who was invited rather than the one who

did the inviting

a Tarkla Murat oldukca pahal

i k

Tarikwith MuratACC rather exp ensive

he invited Tarik and Murat to a rather exp ensive

ama b egenisiz bir lokantaya cagrdi

but tasteless one restaurantDAT invitePASTPAST

but tacky restaurant he invited

b gitmek istememisti

k

they goINF wantNEGPASTPAST

He didnt want to go

Dale suggests that pronominalization is driven bytwo Griceantyp e conversation maxims

a Principle of Adequacy which requires that a referential expression identify the corresp onding

entity and provide sucient information for the reference and a Principle of Eciency which

pulls in the opp osite direction and requires that the referential expression should not contain

more information than required Note that these principles presume a notion of distinguishing

power such as Phifeatures

In order to distinguish a set of entities relative to one another the Phifeatures must b e repre

sented in the Cflist These features can b e gender numb er and p erson as well as grammaticized

topics honorics etc dep ending on the language In Turkish only numb er and p erson features

distinguish entities The facts observed here suggest that Phifeatures may b e listed in the Cflist

as follows

Cf xNUMB PERSSG yNUMB PERSPLU

The CPR given ab ove is revised to include the Phifeatures numb er and p erson as distinguish

ing an entity from others As has b een alluded to various times in this chapter the CPR will

apply to those entities that are not distinguished by Phifeatures The rst part of the CPR is

revised as follows

Center Promotion Rule for Turkish CPR revised

i If an entity e is evoked in a nonsub ject p osition and is not an instantiated CbUn

where e is not distinguishable from the Cp in the Cflist e must b e realized bya

full NP sub ject in Un

Anull sub ject cannot b e used if it cannot b e distinguished from other entities on the Cf list

particularly the Cp which is the preferred center for the next utterance Thus a full NP sub ject

must b e used when the conditions listed ab ove hold Moreover if a less saliententity can b e

distinguished from the Cp it can b e realized byanull sub ject

More Evidence for twotyp es of Smo othShift

It has b een argued that there should b e twotyp es of Smo othShifts based on the distribution of

sub jects Baldwin In prep prop oses that Centering transitions should b e reordered as follows

Continue Smo othShift Retain The original ranking prop osed by Grosz Joshi and Weinstein

is Continue Retain Smo othShift Baldwin argues that the order of frequency in

Linsons results should b e taken as correlating with the order of preference of the transitions

Continue

Smo othShift

RoughShift No Cb

Retain

In what follows it will b e argued that Baldwins observation makes b oth the right and the

wrong predictions and that wemust instead recognize that there are twotyp es of Smo othShift

which b ehave in dierentways For example consider The transition from a to b is a

Smo othShift In a Sevim is an instantiated Cb realized byanull sub ject The infelicityofb

can b e accounted for by the following predictions of Centering theory following Grosz Joshi and

tiated Cb and the Cp in a the Weinstein and the discussions ab ove The instan

null sub ject is realized by a full NP in b while the nonCb is realized with a full NP This is a

violation of Rule which states that if anyentity is pronominalized the Cb must Furthermore

the nonsub ject of a fails to ob ey the CPR As a result b which is a Smo othShiftob j is

infelicitous

a Saitin yanndan gecti

i

SaitGEN nextPOSS passPAST

She passed by Sait

b Ama o Sevimi tanmad bile

i

but he SevimACC recognize even

but he didnt even recognize Sevim

The coherent discourse must b e as in where the transition from a to b is a Retain In

b the CPR is ob eyed and the Cb is realized with a pronoun

a Saitin yanndan gecti

i

SaitGEN nextPOSS passPAST

She passed by Sait

b Ama Sait onu tanmad bile

i

but Sait sheACC recognize even

but Sait didnt even recognize her

The ranking prop osed in the original formulation of Centering Theory therefore correctly

predicts that retaining is preferred to shifting when b oth options are available This state of

aairs suggest that Baldwins prop osal do es not hold at least for Turkish

On the other hand seems to supp ort Baldwins prop osal The transition from a to b

is a Retain and the one from b to c is a Smo othShiftsub j Note that b oth of the entities in

b are realized in c Omer is represented with a null sub ject while Nazl the Cb of b is

represented with a full NP If Retain were preferred in c Omer would b e realized with a more

explicit expression while Nazl would still b e the Cb Therefore c suggests that Shiftsub j is

preferred to Retain

a haf kzarmst

k

slightly blushPASTPAST

She blushed slightly

b Omer ona bakmaktan cekindi

i k

Omer sheDAt lo okatABL hesitatePAST

Omer hesitated to lo ok at her

c Nazlya cekinmeden baktg icin

k

NazliDAT hesitateNEGABL lo okNOMACC for

For lo oking at Nazl without hesitation

teyzesine kzd

i

auntPOSSD AT getangryPAST

He got angry at her aunt

Orhan Pamuk Cevdet Bey ve Ogul lar

As a result the comparison of and suggest that they must indeed b e two dierenttyp es

of Smo othShift Shiftsub j and Shiftob j resp ectively

Conclusion

This chapter has examined the discourse functions of sub jects in Turkish and has prop osed a set

of discourselevel wellformedness conditions on their use

Null sub jects enco de a Continue or Shiftsub j transitions ie they realize the previous Cp

or Cb The Cb in a Continue transition is always realized byanull pronoun in Turkish unless

it is a stressedfo cused pronoun or the sp eaker wants to communicate some other information

such as the start of a new discourse segment An overt pronoun sub ject o ccurs in a Continue if

and only if the entity is phonetically prominent the fo cus of the utterance In most cases overt

pronoun sub jects o ccur in Shiftob j transitions A small numb er of full NP sub jects are used in

the analyzed data to enco de a Continue transition their function is to add new information ab out

the entity represented or to signal a new discourse segment

Full NP sub jects are used to enco de a Retain or RoughShift transition b ecause either a Brand

new entityisintro duced into the discourse or some other entity in CfUn a nonsub ject is

promoted into the center of attention in sub ject p osition in the current utterance This latter

case is describ ed by the following Center Promotion Rule for Turkish

Center Promotion Rule for Turkish CPR

i Given an entityeevoked in a nonsub ject p osition in Un where e is not an

instantiated CbUn and is not distinguishable by Phifeatures from the Cp in the

Cflist If e is realized in Un it must b e realized by a full NP sub ject

ii Entity e should b e expressed by a full NP in Un b efore it can b e talked anaphori

cally in Un

iii Entity e can b e realized byanovert pronoun only if there is no other overt pronoun

in the utterance that is comp eting for Cbstatus

In the next chapter it will b e argued that the generalizations discussed for Turkish sub jects

can b e extended for the facts in other languages

Chapter

A CROSSLINGUISTIC STUDY OF

CENTERING RULES

Intro duction

Centering Theory provides a formal means to crosslinguistically investigate the use of pronouns

in English Grosz Joshi and Weinstein Brennan Friedman and Pollard Cote

Gordon Grosz and Gilliom Gundel Prince and Walker in Japanese

Kameyama Walker Iida and Cote Nakatani Italian Di

Eugenio German Rambow Yiddish Prince and Turkish Homan

and this study The purp ose of this chapter is to review the studies that haveinvestigated

Centering in other languages and to explore the universal prop erties of Centering suggested in

these studies and elab orated in the present study

It will b e argued that Centering rules are universal and that parameters are set dep ending on

languagesp ecic prop erties In what follows a set of universal Centering rules will b e prop osed

and they will b e compared to the data provided in previous research These are as follows

The Center Promotion Rule prop osed in this dissertation is a universal rule which is parame

terized according to the distinguishing p ower made available through Phifeatures number

gender and p erson topic marker Empathy lo cus andor honoric features dep ending on

the language

The Discourse Point of View Rule prop osed in this dissertation is a universal rule

Continue transitions are enco ded by the least explicit pronouns null sub jects in prodrop

languages and unstressed pronouns in others

The Center Promotion Rule across Languages

It is suggested that the Center Promotion Rule prop osed in this study is a universal rule As was

discussed in Chapters and this is an extensive strategy used in Turkish discourse The rule

states that if an entityisevoked in a nonCb ob ject p osition then it cannot b e referred to with the

most salient anaphor sub ject eg a null sub ject in Turkish in the next utterance Wehave also

seen that Brennans ndings in English provide supp ort for this rule Brennan has found as

mentioned in Chapter that the entity in nonsub ject p osition is rep eated as a full NP verbatim

in sub ject p osition in the next utterance b efore b eing referred to anaphoricallyasshown in the

following example from Brennan

Rice with the ball in the middle over to Joub ert

Joubert with a nice move baskets o shots o

Furthermore Grosz Joshi and Weinstein show that sub jects in English are so salient

that failure to cho ose a sub ject as the antecedent of a pronoun leads to infelicityFor example

the choice of the sub ject pronoun in e to realize the ob ject in d renders the discourse incoherent

although one would exp ect that the world knowledge would disambiguate it That is a sp eaker

can deduce via world knowledge that a p erson who calls someone should b e awake and also that

the p erson that is called can b e sleeping at that particular time Thus world knowledge facts

might force a reading where Tonyistheantecedent for the pronoun rather than TerryHowever

the discourse is infelicitous A more natural and coherent sequence instead is obtained by a full

NPTony The infelicity of e can b e accounted for by the Center Promotion Rule as follows the

ob ject in d is realized with a pronoun that is not the Cb the sub ject pronoun is the Cb the

less salien tentity the ob ject undergo es the Center Promotion Rule

a Terry really go ofs sometimes

i

b Yesterdaywas a b eautiful day and he was excited ab out trying his new sailb oat

i i

c He wanted Tony to join him on a sailing exp edition

i k

d He called him at am

i k

e He Tony was sick and furious at b eing woken up so early

k k

Example demonstrates that world knowledge is not sucient for pronoun resolution and

that salience plays an imp ortant role in constraining the choice of antecedent for a pronoun

Therefore there are linguistic wellformedness conditions on pronouns that cannot b e overridden

by inferencing

However a pronoun in exactly the same discourse in e is felicitous when gender distinguishes

the twoentities

a Terry really go ofs sometimes

i

b Yesterdaywas a b eautiful day and he was excited ab out trying his new sailb oat

i i

c He wanted Lee to join him on a sailing exp edition

i k

d He called her at am

i k

e She was sick and furious at b eing woken up so early

k

This study prop oses that the factors that determine Cfranking are universal and that the

sub ject is the Cp However languages dier in that they have dierent linguistic devices for

distinguishing b etween entities English is a language that makes gender distinctions on pronouns

eg him and her which distinguish Terry MASC from Lee FEM in ce A set of entities

distinguished as such can b e realized by less explicit referential expressions These facts will b e

captured if the Phifeatures are represented in the Cflist

Let us now consider the Italian pronominal system in search of further supp ort for the sug

gestions made so far Di Eugenio presents an account of the discourse functions of null and

overt sub jects in Italian within the framework of Centering Theory Italian has a set of weak and

strong pronouns The weak pronouns are cliticized to the verb in ob ject p osition and are null

in sub ject p osition Since there is no neuter gender in Italian nouns that are used to refer to

inanimate ob jects are either masculine or feminine A strong pronoun for an inanimate entity has

to b e represented with a paraphrase or deictic

Di Eugenio hyp othesizes that null and overt sub jects in Italian have the following func

tions when Centering transitions are taken into account

uation is typically enco ded with a null sub ject in Italian Center contin

Center retention and shift are enco ded with overt pronouns

These hyp otheses are tested in Di Eugenio by studying naturallyo ccurring discourses

Di Eugenio nds that Continues are enco ded bynull sub jects in naturallyo ccurring texts She

also rep orts that Retains rarely are enco ded with null and overt pronouns note that this is

also the case in Turkish as discussed in Chapter However overt pronouns are not always

used to enco de Shifts Di Eugenio states that the signicant dierence b etween null and

overt pronouns in the case of Continues following a Continue and Shift disapp ears if the Continue

follows a Retain She questions whether the null sub ject tends to realize the Cp of the previous

utterance rather than its Cb but she leaves this for future research We shall return to this later

in the chapter

The second hyp othesis suggested in Di Eugenio is as follows

Anull pronoun can enco de a Retain or Shift if the agreement and tense morphology

and syntactic features are sucient to disambiguate it

Consider the following Italian example from Di Eugenio

a Maria voleva andare al mare

i

Maria wanted to go to the seaside

b Telefono a Giovanni

i k

She called Giovanni up

c Si arrabbio p erche non lo trovo a casa

i i

She got angry b ecause she not him found at home

Cf Maria Giovanni Cb Maria Continue

c Luik si arabbio p erche stava dormendo

k

He got angry b ecause he was sleeping

Cf Giovanni Cb Giovanni Smo othShift

c si arabbiato p erche stava dormendo

k k

He got angry b ecause he was sleeping

Cf Giovanni Cb Giovanni Smo othShift

In b and c Continue transitions are enco ded bynull sub jects In c a Smo othShift is

enco ded byanovert pronoun coreferential with the ob ject of the previous utterance Giovanni

The use of strong pronouns to enco de Smo oth Shiftob j has also b een observed in the Turkish

discourse Contrast b with c The null sub ject in c should realize the sub ject in b However

the null sub ject realizes Giovanni This is due to the fact that in c the verbal morphology forces

the null sub ject to b e Giovanni and not Maria since it is masculine

The following are other Italian examples from Di Eugenio

a Ieri Carlo ha incontrato Mario

i k

Yesterday Carlo has met Mario

b Non glik ha nemmeno detto ciao

ik

He not tohim has even said hi

a Ieri Carlo ha incontrato Maria

i k

Yesterday Carlo has met Maria

b Non glii ha nemmeno detto ciao

ik

He not tohim has even said hi

These discourses are exactly the same except that the referents of the ob jects Mario in a

and Maria in a are of dierent sexes male and female resp ectively The null sub ject in b

cannot realize the previous ob ject while in b it can The reason is that during the pro cessing

of b the p ossibility of the null sub ject realizing Carlo is ruled out when the masculine clitic

gli is encountered Gli has to b e used to refer to Carlo given that it is not reexive it cannot

realize the sub ject therefore the null sub ject is interpreted as Maria This disambiguation is not

p ossible in b since the referents of the p otential antecedents are b oth male and a strong pronoun

has to b e used to obtain a Shiftob j

The Italian facts as presented by Di Eugenio supp ort the prop osal made in this study that

Phifeatures should b e represented in the Cflist As discussed in Chapter Phifeatures also

play an imp ortant role in the use of null sub jects in a Shiftob ject in Turkish For example in

the entities are distinguished bythenumb er feature the sub ject in a is thirdp erson plural

After this distinguishing feature is checked at the Cflist the ob ject can b e represented with a

null sub ject in b b eing redeemed from the Center Promotion Rule

aCo cukluk asklar unutulmaz asl diyorlar Melihaya

i

childho o d loveaairPLUACC forgetPASSNEG in fact sayPROGPLU MelihaDAT

They tell Meliha that the childho o d love aairs are in fact unforgettable

b Inanmyor

i

b elieveNEGPROG

She do esnt b elieve

On the contrary in the Cflist of a Phifeatures cannot distinguish the twoentities therefore

the ob ject has to undergo the Center Promotion Rule ie has to b e realized with a full NP in

the subsequent utterance in sub ject p osition

aCo cukluk asklar unutulmaz asl diyor Melihaya

i

childho o d loveaairPLUACC forgetPASSNEG in fact sayPROG MelihaDAT

She tells Meliha that the childho o d love aairs are in fact unforgettable

b Meliha Inanmyor

i i

believeNEGPROG

She do esnt b elieve

These facts in English Italian and Turkish supp ort our suggestion that the Center Promo

tion Rule applies crosslinguistically and that Phifeatures with distinguishing features playan

imp ortant role in pronominalization and must b e represented and checked in the Cflist

The Discourse PointofViewasaUniversal rule

Another rule that may apply crosslinguistically is the Point of View Rule This aects the ranking

of the Cflist by determining whether the ranking is ob jective or from the p oint of view of a sub ject

of consciousness

Consider where the thirdp erson pronoun is supp osed to b e ambiguous

a Charles lo oked at the mank in front of him

i i

b He was a full head taller than Charles

k i

c His p erspiring face was so pudgy that his eyes were mere slits

k k

d He was dressed like the other men Charles had seen

k i

Robin Co ok Fever

Utterance c is from Charless p ersp ective The pronoun in c is not ambiguous contrary to

one might exp ect but realizes the ob ject in the previous utterance rather than the Cb sub ject

This has also b een discussed by HudsonDZmura as reviewed in Chapter

In Turkish as well as in English wehave observed that p erception verbs increase the salience

of the nonsub ject

a Gozlerini kapad Cem

i

eyePLUPOSSACC closePAST Cem

Cem closed his eyes

b Keremi dusundu

i k

KeremACC thinkPAST

He thought ab out Kerem

c Az konusurdu

k

little sp eakPRESPAST

He would hardly sp eak

Selim Ileri Her Gece Bo drum

In b Cem is the Cb and the Cp However the verb dusunmek think signals Cems inner

thoughts his sub jectivepoint of view The ob ject entityisthus ranked from his consciousness

and can b e realized byanull sub ject in c without the exp ected ambiguity

The Functions of Sub jects across Languages

This section argues that null and unstressed sub jects are used to enco de Continue and Shiftsub j

transitions crosslinguistically

HudsonDZmura designs a set of exp eriments to discover the discourse constraints on

anaphora in English in particular on the dierence b etween pronouns and nouns She argues

that pronouns indicate how sentences are related to another in a discourse segment Hence they

aect lo cal coherence Nouns on the other hand change the global fo cus of atten tion and aect

the global coherence of the discourse Note that in the Turkish data discussed in Chapter it

was shown that full NP sub jects in Continue transitions signal a discourse segment shift and it

was also noted that full NPs are observed in RoughShift transitions whichmay b e considered as

a shift in discourse segmentalso

HudsonDZmura rep orts that the Continued Cb is pro cessed in a shorter time and more easily

when it is a pronoun than when it is a noun and that garden paths o ccur when the Cb of the

current utterance is not the Cp of the previous utterance This nding supp orts the claim that a

pronoun is favored when the entity is in sub ject p osition in the previous sentence

Null sub jects in prodrop languages corresp ond to unstressed pronouns in English cf Kameyama

Kerslake Walker Iida and Cote Di Eugenio among oth

ers Kameyama Walker Iida and Cote show that Continue transitions are

enco ded bynull sub jects in Japanese In the following example from Japanese cited in Kameyama

c has a reading of Max invited Rosa to dinner Kameyama states an overt pro

noun or a full NP in c to realize the previous Cb would b e incoherent

a Max wa dare o matteiru no

i

Max TOP who OBJ iswaiting Quest

WhoisMaxwaiting for

b Rosa o matteiru nda yo

i

OBJ iswaiting

He is waiting for Rosa

c Rosakanozyo o yuusyoku ni sasotta nda

i

Rosaher OBJ supp er to invited

He invited Rosa to dinner

Likewise Di Eugenio as has b een previously mentioned shows that null sub jects are

used to enco de Continues in Italian She rep orts that the numb er of Retains is very low for b oth

null and overt pronouns in her corpus This is the case in Turkish as well as discussed in Chapter

This is b ecause a full NP is required if a Brandnew entityisevoked into the discourse or an

entity that has left the centerofattention is reintro duced

Di Eugenios other ndings suggest that null sub jects in Italian are analogous to those in

Turkish ie they may enco de Shiftsub j Di Eugenio rep orts that overt pronouns are not always

used to enco de Shift transitions contrary to her predictions This may b e b ecause the null sub ject

realizes the Cp of the previous utterance

In Yiddish sp eakers also tend to use a null sub ject when the Cb is continued Prince

shows that sub ject prodrop in Yiddish is syntactically more restricted than in many other lan

guages Unlike Italian Japanese and Turkish Yiddish disallows null sub jects following a lled

wed only in complementizer and in nonsentence initial p ositions Thus null sub jects are allo

matrix clauseinitial preverbal p osition Prince explores the discourse functions of null versus

overt sub ject pronouns in those environments where there is an option noting that even when

the syntactic constraints are met a great numb er of sub ject pronouns are not dropp ed This

contrasts with Japanese Italian and Turkish in whichnull sub jects are very frequent

Prince rep orts that in all instances of null sub ject clauses in her corpus where the null pronoun

would b e realized by a rst or thirdp erson pronoun or by a secondp erson plural the null pronoun

is the Cb and the transition is a Continue However when the null pronoun would b e the second

p erson singular pronoun this is not the case the null sub ject is not necessarily the Cb and even

when it is the transition typ e is not necessarily a Continue Thus Princes results show that if

null sub jects are sensitive to salience then they are used to enco de Continues in Yiddish

Prop ertysharing Constraint and Zero Topic Assignmentin

Japanese

Kameyama prop oses a prop ertysharing constraintonCentering which requires that

two pronominal entities realizing the same Cb in adjacent utterances should share a certain

grammatical prop erty eg sub ject or nonsub ject This constraint states that two unstressed

pronouns that represent the same Cb in adjacent utterances should share one of the sub ject

or nonsub ject prop erties Kameyama claims that this constraint can account for the cases of

ambiguous multipronouns as in

a Max is waiting for Fred

b He invited him to dinner

According to Kameyama the discourse is more coherent when the sub ject and ob ject in the

subsequent utterance are the same as the sub ject of and ob ject of the previous utterance For the

same reason the prop ertysharing constraint will also explain why the nonsub ject him in b

realizes the nonsub ject Tom in a in the cases of explicit parallelism as in

a Carl is talking to Tom in the Lab

b Terry wants to talk to him to o

Kameyama p oints out that this constraint can b e overridden by Empathy considerations in

Japanese

Walker Iida and Cote suggest that Kameyamas prop ertysharing constraintis

inadequate They argue that using Centering to the fullest extent will suce to account for the

facts of Japanese and that additional mechanisms as prop ertysharing constraint are redundant

Japanese is a language that diers from Turkish Italian Yiddish and English in that the

Japanese verbs do not have agreement morphology that lo cally identies null sub jects In addi

tion Japanese unlike the others has an additional grammaticized topic marked with wa and a

prevailing Empathylo cus

Consider the Japanese example from Walker Iida and Cote where c is ambiguous

with multiple zero pronouns

a Taro o wa ko o en o sanp ositeita

i

TOPIC park walk around

Taro o was walking around the park

oga yatto mituketa b Hanak

i

SUBJ nally found

Hanako nally found him

c yotei o setumeisita

ik ik

SUBJ OBJ schedule explained

He explained the schedule to her

She explained the schedule to him

To account for the ambiguity of pronouns in such cases Walker Iida and Cote

stipulate an optional assignment of topic to null pronouns by means of a Zero Topic Assignment

Rule They conjecture that the optional grammaticized topic marker wa which can b e attached

only to a Discourseold entity is represented with a null pronoun in ob ject p osition While topics

are usually sub jects sub jects and grammaticized topics need not coincide Walker Iida and Cote

Since Kameyama and Walker Iida and Cote have argued that topics in Japanese rank

higher than sub jects in the Cflist a Smo othShift can haveanull sub ject realizing this null topic

marked Cp by Zero Topic Assignment as follows

Zero Topic Assignment ZTA

When no Continuation transition is available and a zero pronoun in Un represents an

entity that was the Cb Un and if no other entityinUnisovertly marked as topic

that zero maybeinterpreted as the topic of Un

According to Walker Iida and Cote the ambiguity in c with twonull pronouns in the

sub ject and ob ject p ositions can b e resolved byZTA Rule The ambiguity is resolved dep ending

on whether the null ob ject in b is assigned a zero topic or not If the null ob ject is assigned a

zero topic it will b e the most saliententity allowing a Continue with the Taro o explained reading

If it is not then the null sub ject in c will realize the sub ject in b resulting in a Smo othShift

with the Hanako explained reading Hence b has two Cflists according to Walker Iida and

Cote as represented in the example

The grammaticized topic wa can function in Japanese in a sp ecial way enabling sp eakers

to entertain two Cflists as in ab ove It is conjectured in this study that topic marker has

a distinguishing p ower like p erson numb er and gender features in other languages Further

distinguishing descriptions in Japanese discourse may b e honorics This suggestion has to b e

tested in further research

Summary

Universal generalizations can b e made regarding the use of referential expressions based on Cen

tering analyses The Center Promotion Rule app ears to b e a universal rule Sub jects tend to rank

higher than nonsub jects across languages Furthermore languages tend to make use of various

devices of distinguishing expressions such as gender p erson and numb er or the grammaticized

topic in Japanese as well as honorics Future research is needed to elab orate more on these

preliminary observations

Chapter

SUMMARY AND CONCLUSIONS

This study has concentrated on three asp ects of discourse anaphora in sub ject p osition in Turkish

First the question of which NPs contribute to the Cflist is investigated in Chapter It

is shown that while only referential expressions are available for subsequent denite reference

nonreferential expressions can also serveasantecedents for nonsp ecic indenite null pronouns

In these cases overt pronouns can never b e used interchangeably with null pronouns It is shown

that overt pronouns in Turkish must b e strictly coreferential with their corresp onding entities As

a result overt pronouns cannot function as b ound variables or pronouns of laziness In contrast

null pronouns are not constrained in this way One of their uses is to function as nonsp ecic

indenite anaphora with nonreferential antecedents However if the utterance in whichsuch

anaphora is used asserts the existence of an entity these indenite null pronouns can evokea

discourse entity for subsequent reference by denite pronouns It is further p ointed out that in

referentially opaque contexts denite null sub jects can alternate with overt pronouns

The second question investigated in this study concerns the factors that determine the most

saliententity ie the plausible antecedent for denite anaphoric reference in the next utterance

Chapter concentrates on the factors that determine ranking the Cflist in Turkish It is argued

that sub jectho o d is an imp ortant factor for determining the Cp the entity that is predicted to

b e the center of attention in the next utterance On the other hand Exp eriencer ob jects of

psychological verbs rank higher than their Theme sub jects This fact can b e captured if thematic

roles are used for ranking or if Exp eriencer ob jects are treated as underlying DStructure

sub jects and Themes as DStructure ob jects It is also observed that the intentional structure of

discourse determines the sub jective and ob jectivepoints of view It is argued that salience may

b e ranked from the sub jective p oint of view of an individual a sub jectivepoint of view can b e

triggered by the use of p erception verbs and verbs of cognition or there may b e subtler cues

Finally the third question in this study is to determine the discourse functions of null vs

overt pronouns vs full NPs in sub ject p osition in Turkish Chapter discusses the discourse

functions of null vs overt pronouns vs full NPs and the way in which they pattern in Centering

transitions It is shown that there are discourselevel wellformedness rules eg the Center

Promotion Rule which is conjectured to b e a universal rule It is argued that the distinguishing

power of Phifeatures is signicant and that they should b e taken into account in Centering

analyses In addition the Turkish pronominalization system provides evidence for the Centering

claim that there is at most one Cb p er utterance Finally it is argued that an overt ob ject cannot

comp ete with an overt sub ject pronoun for Cbstatus and thus the sub ject must either b e null

indicating that it is the Cb or else it must b e realized with a full NP signaling that it is not

the Cb As a result multiple overt pronouns with the same p erson and numb er features are

infelicitous in Turkish discourse Chapter surveys some of the cross linguistic researchdone

within the Centering framework and discusses the universal features of pronoun systems and

their interaction with Centering transitions

An interesting pro ject for future research is to analyze intrasentential pronominalization rules

in Centering terms and to investigate to what extentemb edded clauses can b e considered as sepa

rate utterances Clearly a unied accountofsentencelevel and discourselevel pronominalization

would b e desirable if p ossible In addition future research will concentrate on the Centering anal

ysis of null vs overt objects in comparison to this study on Turkish sub jects in order to complete

our understanding of Turkish anaphora in discourse

BIBLIOGRAPHY

in

Allerton DJ Deletion and Proform Reduction in Journal of Linguistics

App elt Douglas E Some Pragmatic Issues in the Planning of Denite and Indenite

Noun Phrases in Asso ciation for Computational Linguistics

Ariel Mira Referring and Accessibility in Journal of Linguistics

Baker Mark Incorp oration A Theory of Grammatical Function Changing Chicago

Chicago University Press

Baldwin Breck Cogniac A Discourse Pro cessing Engine UniversityofPennsylvania

PhD Dissertation

Baneld Ann Unsp eakable Sentences London Routledge and Paul

Barwise Jon and John Perry Situations and Attitudes Cambridge Massachusetts

Bradford Bo oks

Belletti Adriana and Luigi Rizzi PsychVerbs and Theta Theory In Natural Lan

guage and Linguistic Theory

Belletti Adriana Morphological Passive and prodrop the Imp ersonal Construction

in Italian Journal of Linguistic Research

Bosch Peter Agreement and Anaphora A Study of the Role of Pronouns in Syntax

and Discourse New York Academic press

Brennan Susan E Centering Attention in Discourse to app ear in Language and

Cognitive Pro cesses

Brennan Susan Marilyn Walker Friedman and Carl Pollard A Centering Approach

to Pronouns in Pro ceedings of the th Annual Meeting Of ACL Stanford

Camp os Hector Indenite Ob ject Drop Linguistic Inquiry

Chafe Wallace Givenness Contrastiveness Deniteness Sub jects Topics and Point

of View in Charles N Li ed Sub ject and Topic New York Academic Press

Chomsky Noam Deep Structure Surface Structure and Semantic Interpretation In

D Steinb erg and Jo cob ovits eds Semantics Cambridge Cambridge University Press

Chomsky Noam Lectures on Government and Binding DordrechtForis Publications

Chomsky Noam Some Concepts and Consequences of the Theory of Governmen tand

Binding Cambridge Massachusetts The MIT Press

Cornish Francis Anaphoric Relations in English and French London Cro om Helm

Cote Sharon Discourse Functions of TwoTyp es of Null Ob jects in English pap er

presented at the th Annual Meeting of the Linguistic So ciety of America January

Cote Sharon Ranking Forwardlo oking Centers pap er presented at the Workshop on

Centering Theory in Naturallyo ccurring Discourse Institute for Research in Cognitive Science

UniversityofPennsylvania

Cote Sharon The Representation of Unexpressed Ob jects in English Unpublished

manuscript

Dale Rob ert Co oking up Referring Expressions in Asso ciation of Computational

Linguistics

Dede Mu serref Deniteness and ReferentialityinTurkish Verbal Sentences in D

Slobin and K Zimmer eds Studies in Turkish Linguistics Amsterdam John Benjamins

Di Eugenio Barbara Centering Theory and the Italian Pronominal System In COL

ING Pro c th International Conference on Computational Linguistics Helsinki Finland

Di Eugenio Barbara Centering in Italian to app ear in EPrince A Joshi and M

Walker eds Centering in Discourse Oxford University Press

Donnellan Keith S Reference and Denite Descriptions in Philosophical Review

Ehrlich Susan Asp ect Foregrounding and p oint of View Text

Ehrlich Susan Point of View A Linguistic Analysis of Literary Style

Enc Murv et Topic Switching and Pronominal Sub jects in Turkish in D Slobin and

K Zimmer eds Studies in Turkish Linguistics Amsterdam John Benjamins

Enc Marvel The Semantics of Sp ecicity Linguistic Inquiry

Erguvanl Eser The Function of Word Order in Turkish Los Angeles Universityof

California Press

ErguvanlTaylan Eser Pronominal versus Zero Representation of Anaphora in Turk

ish in D Slobin and K Zimmer eds Studies in Turkish Linguistics Amsterdam John Ben

jamins

Erku Feride Discourse pragmatics and word order in Turkish University of Minnesota

PhD Dissertation

Evans Gareth Pronouns in Linguistic Inquiry

Farrell Patrick Null Ob jects in Brazilian Portuguese Natural Language and Lin

guistic Theory

Fleischman Suzanne Verb Tense and Point of View in Narrative in S Fleischman

and L Waugh eds Discourse Pragmatics and the Verb New York Routledge

Fox Barbara Discourse Structure and Anaphora Cambridge Cambridge University

Press

Geach Paul Reference and generality Ithaca New York Cornell University Press

George Leland M and Jaklin Kornlt Finiteness and Boundedness in Turkish in

Binding and Filtering London Cro om Helm Ltd

Givon Talmy a Topic ContinuityinSpoken English in T Givn ed Topic Continuity

in Discourse A Quantitative CrossLanguage Study Philadelphi a John Benjamins Publishing

Company

Givon bTopic Continuity in Discourse The Functional Domain of Switch Reference in

John Haiman and Pamela Munro eds SwitchReference and Universal Grammar Philadelphia

John Benjamins Publishing Company

Gordon Peter C Barbara Grosz and Laura Gilliom Pronouns Names and the Centering

of Attention in Discourse Cognitive Science

Grimshaw Jane Argument Structure Cambridge MA The MIT Press

Grosz Barbara and Candace Sidner Attentions Intentions and the Structure of Dis

course Computational Linguistics

Grosz Barbara The Representation and Use of Fo cus in Dialogue Understanding Tech

nical Rep ort SRI International Ravenswood Ave Menlo Park CA USA

Grosz Barbara Aravind Joshi and Scott Weinstein Providing a Unied Accountof

Denite Noun Phrases in Discourse In Pro c st Annual Meeting of ACL Asso ciation of

Computational Linguistics

Grosz Barbara Aravind Joshi and Scott Weinstein Towards a Computational Theory

of Discourse Interpretation Unpublished manuscript

Grosz Barbara Aravind Joshi and Scott Weinstein Centering A Framework for

Mo deling the Lo cal Coherence of Discourse Computational Linguistics

Ha jicova Eva Issues of Sentence Structure and Discourse Patterns Prague Institute

of Theoretical and Computational Linguistics Charles University

Heim Irene The Semantics of Denite and iIndenite Noun Phrases Universityof

Massachusetts PhD Dissertation

Heim Irene File Change Semantics and the Familiarity Theory of Deniteness in R

Buerle C Schwarze and A von Stechow eds Meaning Use and Interpretation of Language

Berlin Walter de Gruyter

Hobbs Jerry Resolving Pronoun References in B Grosz K Sparck Jones and B

ebb er eds Readings in Natural Language Pro cessing California Morgan Kaufman Publish W

ers Inc

Ho ckey Beth Ann In prep The Interpretation and Realization of Fo cus Universityof

Pennsylvania PhD Dissertation

Homan Beryl Word Order Information Structure and Centering in Turkish ms

To app ear in EPrince A Joshi and MWalker eds Centering in Discourse Oxford University

Press

Homan Beryl and mit D Turan Word Order and Centering in Turkish pap er pre

sented in Workshop on Centering Theoryin Naturallyo ccuring Discourse Institute for Research

in Cognitive Science UniversityofPennsylvania Philadelphi a PA May

Homan Beryl Deniteness in Turkish Unpublished manuscript

Homan Beryl Marking SetMemb ership in Turkish The Penn Review of Linguistics

Homan Beryl In prep The Computational Analysis of the Syntax and Discourse Use of

Free Word Order in Turkish UniversityofPennsylvania PhD dissertation

Huang CT James b On the Distribution and Reference of Empty Pronouns Lin

guistic Inquiry

Huang James On Lexical structure and Syntactic Pro jection Unpublished manuscript

HudsonDZmura Susan The Structure of Discourse and Anaphor Resolution The

Discourse Center and the roles of Nouns and Pronouns UniversityofRochester PhD Disserta

tion

HudsonDZmura and Michael Tannenhaus Discourse Constraints on Anaphoric Pro

cessing pap er presented at the Workshop on Centering Theory in Naturallyo ccurring Discourse

Institute for Research in Cognitive Science UniversityofPennsylvania

HudsonDZmura Susan Control and Event Structure through the Mind The View

from Center To app ear in EPrince A Joshi and MWalker edsCentering in Discourse Oxford

University Press

Iatridou Sabine and David Embick Aprop os pro To app ear in Linguistic Inquiry

Jackendo Ray Semantic structures Cambridge Massachusetts The MIT press

Jaeggli Osvaldo Topics in Romance Syntax Dordrecht Foris Publications

Jaeggli Osvaldo Arbitrary Plural Pronominals In Natural Language and Linguistic

Theory

Joshi Aravind and Steve Kuhn Centered Logic The Role of Entity Centered Sentence

Representation in Natural Language Inferencing Pro ceedings of the Sixth International Joint

Conference on Articial Intelligence Tokyo

Joshi Aravind and Scott Weinstein Control of Inference Role of some Asp ects

enth International Joint Conference on of Discourse StructureCentering Pro ceedings of the Sev

Articial Intelligence British Columbia

Kameyama Megumi Zero anaphora the case of Japanese Stanford University PhD

dissertation

Kameyama Megumi A Prop ertysharing ConstraintinCentering Pro ceedings of the

th Annual Meeting of ACL

Kamp Hans A Theory of Truth and Semantic Representation in Truth Interpretation

and Information in Gro enendijk etal eds Formal metho ds in the study of language Amsterdam

Kaplan David Demonstratives in JAlmog JPerry and H Wettstein eds

Themes from Kaplan New York Oxford University Press

Karttunen Lauri Discourse referents in J McCawley ed Syntax and Semantics Vol

Karttunen Lauri Pronouns and Variables CLS Pap ers from the Fifth Regional Meet

ing of the Chicago Linguistic So ciety

Kerslake Celia Noun Phrase Deletion and Pronominalization in Turkish In HE

Bo eschoten and LT Verho even eds Studies on Mo dern Turkish Tilburg University Press

Knecht Laura Sub ject and ob ject in Turkish Massachusetts Institute of Technology

PhDDissertation

Kornlt Jaklin Case marking agreement and empty categories in Turkish Harvard

University PhD dissertation

Kornlt Jaklin Issues in Syntax in H Bo eschoten and LVerho even eds Turkish

Linguistics Today New York EJ Brill

Kuno Susumu and Etsuko Kaburaki Empathy and syntax in Linguistic Inquiry

Kuno Susumu Some prop erties of nonreferential noun phrases in R Jacobson and

S Kawamoto eds Studies in General and Oriental Linguistics

Kuno Susumu a Functional Sentence Persp ective A Case Study from Japanese and

English in Linguistic Inquiry

Kuno Susumu b Pronominalization Reexivization and Direct Discourse in Lin

guistic Inquiry

Kuno Susumu Principles of Discourse Deletion Case Studies From English Russian

and Japanese in Journal of Semantics

Kuno Susumu Identication of Zero Pronominal Reference in Japanese Unpublished

ms

Kuno Susumu Functional Syntax Chicago Chicago University Press

Kural Murat YantumcelerdeC ekim Ekleri Inectional Suxes in Sub ordinate

Clauses in K Imer A Ko caman and S Ozsoy eds Dilbilim Arastrmalar Linguistic Re

search Ankara Hacettep e Universitesi Yaynlar

Kuribayashi Yuu a Sozdizimsel Sozcuk birlesmesi Syntactic Word Comp ounding in

u Bildiril eri Pro ceedings of the IVth Linguistics Symp osium IV Dilbilim Semp ozyum

Kuribayashi Yuu b Syntactic Comp ounding in Turkish in GDurmusoglu K Imer

AKo caman and SOzsoy eds Dilbilim Arastrmalar Linguistic Research

Lab ov William Language in the Inner City Studies in the Black English Vernacular

Philadelphia UniversityofPennsylvania Press

Liejiong Xu Free Empty Category Linguistic Inquiry

Linde Charlotte Fo cus of Attention and the Choice of Pronouns in Discourse in T

Givon ed Discourse and Syntax New York Academic Press

Linson Brian A Distributional Analysis of Discourse Transitions pap er presented at

the Workshop on Centering Theory in Naturally o ccurring Discourse Institute for Researchin

Cognitive Science UniversityofPennsylvania

McKeown R Kathleen Fo cusing in Discourse in Text Generation London Cam

bridge University Press

Montalb etti Mario After binding on the interpretation of pronouns Massachusetts

Institute of Technology PhD Dissertation

Moser Megan G Fo cus Particles Their Denition and Relational Structure CLS

Chicago LInguistic So ciety

Nakatani Christine Global and Lo cal Functions of Topicmarking in Japanese pap er

presented pap er presented in Workshop on centering theory in naturallyo ccurring discourse

Institute for Research in Cognitive Science UniversityofPennsylvania Philadelphi a PA May

Nilsson Birgit Ob ject incorp oration in Turkish In Pro ceedings of the Turkish Lin

guistics Conference August

Nilsson Birgit Turkish semantics revisited in H Bo eschoten and L Verhoven eds

Turkish linguistics to day Leiden EJ Brill

Oakhill Jane Alan Garham and WietskeVonk The online construction of discourse

mo dels in Language and cognitive pro cesses

PesetskyDavid Zero syntax Exp eriencers and cascades Cambridge Massachusetts

The MIT Press

Polanyi Livia A formal mo del of the structure of discourse in Journal of Pragmatics

Portereld Leslie and Veneeta Srivastav Indeniteness in the absence of articles

Evidence from Hindi and Indonesian in Pro ceedings of the th West Coast conference on formal

linguistics

Prince Ellen F On the Function of Existential Presupp osition in Discourse In

DFarkas WJacobsen and KTo drys eds CLS Chicago Linguistic So ciety

Prince Ellen F a Toward a TaxonomyofGiven New Information in PCole ed

Radical Pragmatics New York Academic Press

Prince Ellen F b Topicalization Focus Movement and Yiddish Movement A Prag

matic Dierentiation in DAlford etal eds BLS Berkeley Linguistic So ciety

Prince Ellen F On the Syntactic Marking of Op en Prop osition In CLS Chicago

Linguistic So ciety

Prince Ellen F Discourse Analysis A Part of the Study of Linguistic Comp etence

in F J Newmeyer ed Linguistics The Cambridge Survey Vol

Prince Ellen F Sub jectPro drop in Yiddish pap er presented at the Fo cus Conference

Schloss Wolfsbrunnen Germany June

Rambow Owen Pragmatic asp ects of scrambling and Topicalization in German pap er

presented at the Workshop on Centering Theory in Naturallyo ccurring Discourse Institute for

Research in Cognitive Science UniversityofPennsylvania

Reinhart Tanya Pragmatics and Linguistics An Analysis of Sentence Topics in

Philosophica

Rigau Gemma Some Remarks on the Nature of Strong Pronouns in Nullsub ject

Languages in I Bordelois H Contreras and K Zagona eds Generative studies in Spanish

syntax Dordrecht Foris Publications

Rizzi Luigi Null Ob jects in Italian and the theory of pro Linguistic Inquiry

Rob erts Craige Mo dal sub ordination and pronominal anaphora in discourse in Lin

guistics and Philosophy

Rob erts Craige Centering and Anaphora Resolution in DRT pap er presented in

Workshop on Centering Theoryin naturallyo ccurring discourse Institute for Research in Cogni

tive Science UniversityofPennsylvania Philadelphia

RuhiS ukriye Zero NP Anaphora versus Full NP Anaphora in Turkish pap er presented

at the th International Turkish Linguistics Conference Anadolu University Eskisehir

Sidner Candace Towards a Computational Theory of Denite Anaphora Comprehen

sion in English Discourse Technical Rep ort Articial Intelligence Lab oratory Cambridge

Massachusetts Institute of Technology

Sidner Candace Fo cusing for Interpretation of Pronouns Computational Linguistics

Suner Margarita Proarb Linguistic Inquiry

Suri Linda and Kathleen McCoy Comparing Lo cal Fo cusing and Centering pa

p er presented in Workshop on Centering Theory in naturallyo ccurring discourse Institute for

Research in Cognitive Science UniversityofPennsylvania Philadelphia

Suri Linda Using Lo cal Fo cus to Correct Illegal NP Omissions University of Delaware

Technical Rep ort

Tn Erkan and Varol Akman Anaphora in Turkish A Computational Framework

pap er presented at the th International Turkish Linguistics Conference Anadolu University

Eskisehir

Tura Sabahat S Deniteness and ReferentialityinTurkish Nonverbal Sentences in

D Slobin and K Zimmer eds Studies in Turkish Linguistics Amsterdam John Benjamins

Turan Umit D a Zero Ob ject Arguments and ReferentialityinTurkish pap er presented

at the Seventh International Conference on Turkish Linguistics The Institute of Oriental Studies

Turkology Johannes Gutenb erg University Mainz

Turan Umit D b The Functions of Ob jects in Turkish Discourse Unpublished

manuscript

Vallduv Enric The Informational Comp onent New York Garland

Vallduv Enric Detachment in Catalan and Information Packaging in Journal of

Pragmatics

Walker M Evaluating Discourse Pro cessing Algorithms In Pro ceedings of th

Annual Meeting of the Asso ciation of Computational Linguistics

Walker Marilyn Masayo Iida and Sharon Cote Centering in Japanese Discourse

InCOLING Pro c of the th International Conference on Computational Linguistics Helsinki

Walker Marilyn Masayo Iida and Sharon Cote Japanese Discourse and the Pro cess

of Centering in Computational Linguistics

Walker Marilyn Initial Contexts and Shifting Centers pap er presented in Workshop

on Centering Theory in naturallyo ccuring discourse Institute for Research in Cognitive Science

UniversityofPennsylvania Philadelphia PA May

Webb er Bonnie L A Formal approach to discourse anaphora New York Garland

Webb er Bonnie L So what can we talk ab out now in M Brady and RC Berwick

eds Computational Mo dels of Discourse Cambridge Mass MIT Press

Webb er Bonnie L Tense as Discourse Anaphor in Computational Linguistics

Webb er Bonnie L Structure and Ostension in the Interpretation of Discourse Deixis

in Language and Cognitive Pro cesses

Wieb e J and William Rappap ort A Computational Theory of Persp ectiveand

Reference in Narrative in ACL

Wieb e Janyce M Identifying Sub jective Characters in Narrative in COLING Helsinki

Wieb e Janyce M References in NarrativeText in Nos

Wieb e Janyce M Tracking PointofViewinNarrative in ACL Asso ciation for Com

putational Linguistics

Wilson Deirdre and Dan Sp erb er Ordered Entailments An Alternative to Presupp osi

tional Theories in CK Oh and D Dineen eds Syntax and Semantics I I New York Academic

Press

ZribiHertz Anne Anaphor binding and narrativepoint of view English reexivepro

nouns in sentence and discourse in Language