UUnniivveerrssiittyy ooff PPeennnnssyyllvvaanniiaa SScchhoollaarrllyyCCoommmmoonnss IRCS Technical Reports Series Institute for Research in Cognitive Science January 1995 CCeenntteerriinngg:: AA FFrraammeewwoorrkk ffoorr MMooddeelllliinngg tthhee LLooccaall CCoohheerreennccee ooff DDiissccoouurrssee Barbara J. Grosz University of Pennsylvania Aravind K. Joshi University of Pennsylvania, [email protected] Scott Weinstein University of Pennsylvania, [email protected] Follow this and additional works at: https://repository.upenn.edu/ircs_reports Grosz, Barbara J.; Joshi, Aravind K.; and Weinstein, Scott, "Centering: A Framework for Modelling the Local Coherence of Discourse" (1995). IRCS Technical Reports Series. 116. https://repository.upenn.edu/ircs_reports/116 University of Pennsylvania Institute for Research in Cognitive Science Technical Report No. IRCS-95-01. This paper is posted at ScholarlyCommons. https://repository.upenn.edu/ircs_reports/116 For more information, please contact [email protected]. CCeenntteerriinngg:: AA FFrraammeewwoorrkk ffoorr MMooddeelllliinngg tthhee LLooccaall CCoohheerreennccee ooff DDiissccoouurrssee AAbbssttrraacctt This paper concerns relationships among focus of attention, choice of referring expression, and perceived coherence of utterances within a discourse segment. It presents a framework and initial theory of centering which are intended to model the local component of attentional state. The paper examines interactions between local coherence and choice of referring expressions; it argues that differences in coherence correspond in part to the inference demands made by different types of referring expressions given a particular attentional state. It demonstrates that the attentional state properties modelled by centering can account for these differences. CCoommmmeennttss University of Pennsylvania Institute for Research in Cognitive Science Technical Report No. IRCS-95-01. This technical report is available at ScholarlyCommons: https://repository.upenn.edu/ircs_reports/116 The Institute For Research In Cognitive Science Centering: A Framework for Modelling the Local Coherence of Discourse by P Barbara J. Grosz Harvard University Aravind K. Joshi Scott Weinstein IRCS E University of Pennsylvania 3401 Walnut Street, Suite 400C Philadelphia, PA 19104-6228 N January 1995 Site of the NSF Science and Technology Center for N Research in Cognitive Science University of Pennsylvania IRCS Report 95-01 Founded by Benjamin Franklin in 1740 Centering(cid:0) A Framework for Modelling the Local (cid:0) Coherence of Discourse(cid:1) Barbara J(cid:0) Grosz Aravind K(cid:0) Joshi Scott Weinstein Division of Applied Sciences Computer and Information Science Philosophy Harvard University University of Pennsylvania University of Pennsylvania Cambridge(cid:1) MA Philadelphia(cid:1) PA Philadelphia(cid:1) PA grosz(cid:2)das(cid:0)harvard(cid:0)edu joshi(cid:2)linc(cid:0)cis(cid:0)upenn(cid:0)edu weinstei(cid:2)linc(cid:0)cis(cid:0)upenn(cid:0)edu (cid:0) WewanttothankBreck Baldwin(cid:0)FeliciaHurewitz(cid:0) AndyKehler(cid:0)KarenLochbaum(cid:0)ChristineNakatani(cid:0) Ellen Prince(cid:0) and Lyn Walker for their valuable comments which helped us improve both the content and the presentation of our paper(cid:1) We are also grateful to CarolynElken in helping us keep track of the various drafts of this paper and for providing valuable editorialhelp(cid:1) Partial support for the (cid:2)rst author was provided by grants NSF IRI(cid:3)(cid:4)(cid:5)(cid:3)(cid:5)(cid:4)(cid:5)(cid:6)(cid:7)and IRI(cid:3)(cid:4)(cid:8)(cid:3)(cid:5)(cid:7)(cid:6)(cid:9)(cid:8)(cid:10)the second author was partiallysupported by the ARO Grant DAAL(cid:5)(cid:8)(cid:3)(cid:7)(cid:4)(cid:3)(cid:5)(cid:5)(cid:8)(cid:6)and ARPA Grant N(cid:5)(cid:5)(cid:5)(cid:6)(cid:11)(cid:3)(cid:4)(cid:5)(cid:3)J(cid:3)(cid:6)(cid:7)(cid:12)(cid:8)(cid:1) (cid:3) Contents (cid:0) Preface (cid:1) (cid:2) Introduction (cid:1) (cid:3) Phenomena to be Explained (cid:4) (cid:1) Basic Center De(cid:5)nitions (cid:6) (cid:7) Claims of Centering Theory (cid:0)(cid:0) (cid:4) Factors Governing Centering (cid:0)(cid:2) (cid:8) Constraints on Center Movement and Realization (cid:0)(cid:4) (cid:9) Applications of the Rules (cid:0)(cid:8) (cid:6) Requisite Properties of Underlying Semantic Theory (cid:2)(cid:10) (cid:0)(cid:10) Related Work (cid:2)(cid:1) (cid:4) Abstract This paper concerns relationships among focus of attention(cid:0) choice of referring ex(cid:1) pression(cid:0) and perceived coherence ofutteranceswithin adiscourse segment(cid:2) Itpresents a frameworkand initial theory ofcentering which are intended to model the local com(cid:1) ponent of attentional state(cid:2) The paper examines interactions between local coherence and choice of referring expressions(cid:3) it argues that di(cid:4)erences in coherence correspond in part to the inference demands made by di(cid:4)erent types of referring expressions given a particular attentional state(cid:2) It demonstrates that the attentional state properties modelled by centering can account for these di(cid:4)erences(cid:2) (cid:5) (cid:0) Preface Our original paper (cid:6)Grosz(cid:1) Joshi(cid:1) and Weinstein(cid:1) (cid:3)(cid:7)(cid:8)(cid:5)(cid:9) on centering claimed that certain entities mentioned in an utterance were more central than others and that this property imposed constraints on a speaker(cid:10)s use of di(cid:11)erent types of referring expressions(cid:0) Centering was proposed as a modelthat accounted for this phenomenon(cid:0) We argued that thecoherence of discourse was a(cid:11)ected by the compatibility between centering properties of an utterance andchoiceofreferringexpression(cid:0) Subsequently(cid:1)werevisedandexpandedtheideaspresented therein(cid:0) We de(cid:12)ned various centering constructs and proposed two centering rules in terms of these constructs(cid:0) A draft manuscript describing this elaborated centering framework and presenting some initial theoretical claimshas been in wide circulation since (cid:3)(cid:7)(cid:8)(cid:13)(cid:0) This draft (cid:6)Grosz(cid:1) Joshi(cid:1) and Weinstein (cid:3)(cid:7)(cid:8)(cid:13)(cid:1) hereafter(cid:1) gjw(cid:0)(cid:1)(cid:9) has led to a number of papers by (cid:0) others on this topic and has been extensively cited(cid:1) but has never been published(cid:0) We have been urged to publish the more detailed description of the centering framework and theory proposed in gjw(cid:0)(cid:1) so that an o(cid:14)cial version would be archivally available(cid:0) The task of completing and revising this draft became more daunting as time passed and more and more papers appeared on centering(cid:0) Many of these papers proposed extensions to or revisions of the theory and attempted to answer questions posed in gjw(cid:0)(cid:1)(cid:0) It has become ever more clear that it would be useful to have a (cid:15)de(cid:12)nitive(cid:16) statement of the original motivations for centering(cid:1) the basic de(cid:12)nitions underlying the centering framework(cid:1) and the original theoretical claims(cid:0) This paper attempts to meetthat need(cid:0) To accomplish this goal(cid:1) we have chosen to remove descriptions of many open research questions posed in gjw(cid:0)(cid:1) as well as solutions that were only partially developed(cid:0) We have also greatly shortened the discussion of criteria for and constraints on a possible semantic theory as a foundation for this work(cid:0) (cid:1) Introduction This paper presents an initial attempt to develop a theory that relates focus of attention(cid:1) choice of referring expression(cid:1) and perceived coherence of utterances within a discourse seg(cid:17) ment(cid:0) The research described here is a further development of several strands of previous research(cid:0) It (cid:12)ts within a larger e(cid:11)ort to provide an overall theory of discourse structure and meaning(cid:0) In this section we describe the larger research contextof this work and then brie(cid:18)y discuss the previous work that led to it(cid:0) Centering(cid:12)ts withinthe theoryof discourse structuredevelopedbyGrosz and Sidner(cid:6)(cid:3)(cid:7)(cid:8)(cid:13)(cid:9)(cid:1) henceforth(cid:1) G(cid:19)S(cid:0) G(cid:19)S distinguish among three components of discourse structure(cid:20) a lin(cid:17) guistic structure(cid:1) an intentional structure(cid:1) and an attentional state(cid:0) At the levelof linguistic structure(cid:1) discourses divide into constituent discourse segments(cid:21) an embedding relationship may hold between two segments(cid:0) The intentional structure comprises intentions and rela(cid:17) tions among them(cid:0) The intentions provide the basic rationale for the discourse(cid:1) and the (cid:0)Early drafts of gjw(cid:0)(cid:1) were in circulation from (cid:6)(cid:4)(cid:7)(cid:8)(cid:1) Some citations to other work have dates between (cid:6)(cid:4)(cid:7)(cid:8) and (cid:6)(cid:4)(cid:7)(cid:12)(cid:1) This work utilized these earlier drafts(cid:1) (cid:22) relations represent the connections among these intentions(cid:0) Attentional state models the discourse participants(cid:10) focus of attention at any given point in the discourse(cid:0) Changes in attentional state depend on the intentional structure and on properties of the utterances in the linguistic structure(cid:0) Each discourse segment exhibits both local coherence (cid:23) i(cid:0)e(cid:0) coherence among the utterances in that segment(cid:23)andglobal coherence (cid:23) i(cid:0)e(cid:0)coherencewith other segmentsin the discourse(cid:0) Corresponding to these two levels of coherence are two components of attentional state(cid:21) the local level models changes in attentional state within a discourse segment(cid:1) and the global level models attentional state properties at the intersegmental level(cid:0) G(cid:19)S argue that global coherence depends on the intentional structure(cid:0) They propose that each discourse has an overall communicativepurpose(cid:1) the discourse purpose (cid:6)DP(cid:9)(cid:21) and each discourse segmenthas an associated intention(cid:1)its discourse segmentpurpose (cid:6)DSP(cid:9)(cid:0) The DP and DSPS are speaker intentions(cid:21) they are correlates at the discourse level of the intentions Grice argued underlay utterance meaning (cid:6)Grice(cid:1) (cid:3)(cid:7)(cid:13)(cid:7)(cid:9)(cid:0) If a discourse is multi(cid:17)party (cid:6)e(cid:0)g(cid:0) a dialogue(cid:9)(cid:1) then the DSP for a given segmentis an intention of the conversational participant who initiatesthatsegment(cid:0) Lochbaum(cid:6)Lochbaum(cid:1)(cid:3)(cid:7)(cid:7)(cid:22)(cid:9) employscollaborativeplans (cid:6)Grosz and Kraus(cid:1) (cid:3)(cid:7)(cid:7)(cid:5)(cid:9) to model intentional structure and is thus able to integrate intentions of di(cid:11)erent participants(cid:0) Satisfaction of the DSPs contributes to the satisfaction of the DP(cid:0) Relationships between DSPs provide the basic structural relationships for the discourse(cid:21) embeddings in the linguistic structure are derived from these relationships(cid:0) The global coherence of a discourse depends on relationships among its DP and DSPs(cid:0) G(cid:19)S model the global(cid:17)levelcomponent of the attentional state with a stack(cid:21) pushes and pops of focus spaces on the stack depend on intentional relationships(cid:0) This paper is concerned with local coherence and its relationship to attentional state at the locallevel(cid:0) Centeringisproposed as amodelofthelocal(cid:17)levelcomponentof attentionalstate(cid:0) Weexaminetheinteractionsbetweenlocalcoherenceandchoicesofreferringexpressions(cid:1)and argue that di(cid:11)erences in coherence correspond in part to the di(cid:11)erent demands for inference made by di(cid:11)erent types of referring expressions(cid:1) given a particular attentional state(cid:0) We describe how the attentional state properties modelled by centering can account for these di(cid:11)erences(cid:0) Three piecesof previous researchprovidethe background for thiswork(cid:0) Grosz (cid:6)(cid:3)(cid:7)(cid:24)(cid:24)(cid:9) de(cid:12)ned two levels of focusing in discourse(cid:1) global and immediate(cid:0) Participants were said to be globally focused on a set of entities relevant to the overall discourse(cid:0) These entities may either have been explicitly introduced into the discourse or be su(cid:14)ciently closely related to such entities to be considered implicitly in focus (cid:6)Grosz(cid:1) (cid:3)(cid:7)(cid:8)(cid:3)(cid:9)(cid:0) In contrast(cid:1) immediate focusing referred to a more local focusing process(cid:1) one that relates to identifying the entity that an individual utterance most centrally concerns(cid:0) Sidner (cid:6)(cid:3)(cid:7)(cid:24)(cid:7)(cid:9) provided a detailed analysis of immediate focusing(cid:1) including a distinction between the current discourse focus and potential foci(cid:0) She gave algorithms for tracking immediate focus and rules that stated how the immediate focus could be used to identify the referents of pronouns and demonstrative noun phrases (cid:6)e(cid:0)g(cid:0) (cid:15)this party(cid:1)(cid:16) (cid:15)that party(cid:16)(cid:9)(cid:0) Joshi and Kuhn (cid:6)(cid:3)(cid:7)(cid:24)(cid:7)(cid:9) and Joshi and Weinstein (cid:6)(cid:3)(cid:7)(cid:8)(cid:3)(cid:9) provided initial results on the con(cid:17) (cid:25) nection between changes in immediate focus and the complexity of inferences required to integrate a representation of the meaning of an individual utterance into a representation of the meaning of the discourse of which it was a part(cid:0) To avoid confusion with previous uses of the term (cid:15)focus(cid:16) in linguistics(cid:1) they introduced the centering terminology(cid:0) Their notions of (cid:15)forward(cid:17)looking(cid:16) and (cid:15)backward(cid:17)looking(cid:16) centers correspond approximately to Sidner(cid:10)s potential foci and discourse focus(cid:0) In all of this work(cid:1) focusing(cid:1) whether global or immediate(cid:1) was seen to function to limit the inferences required for understanding utterances in a discourse(cid:0) Grosz and Sidner were concerned with the inferences needed to interpret anaphoric expressions of various sorts (cid:6)e(cid:0)g(cid:0) pronouns(cid:1) de(cid:12)nite descriptions(cid:1) ellipsis(cid:9)(cid:0) They used focusing to order candidates(cid:21) as a result the need for search was greatly reduced and the use of inference could be restricted to determining whether a particular candidate was appropriate given the embedding utterance interpretation(cid:0) Joshi(cid:1) Kuhn(cid:1) and Weinstein were concerned with reducing the inferences required to integrate utterance meaning into discourse meaning(cid:0) They used centering to determine an almost monadic predicate representation of an utterance in discourse(cid:21) they then use this representation to reduce the complexity of inference(cid:0) In this paper(cid:1) we generalize and clarify certain of Sidner(cid:10)s results(cid:1) but adopt the (cid:15)centering(cid:16) terminology(cid:0) We also abstract from Sidner(cid:10)s focusing algorithm to specify constraints on the centering process(cid:0) We consider the relationship between coherence and inference load and examine how both interact with attentional state and choices in linguistic expression(cid:0) The remainder of this paper is organized as follows(cid:20) In Section (cid:5)(cid:1) we brie(cid:18)y describe the phenomena motivating the development of centering that this paper aims to explain(cid:0) Sec(cid:17) tion (cid:22) provides the basic de(cid:12)nitions of centers and related de(cid:12)nitions needed to present the theoretical claims of the paper(cid:0) In Section (cid:25)(cid:1) we state the main properties of the centering framework and the major claims of centering theory(cid:0) In Section (cid:13)(cid:1) we discuss several factors that a(cid:11)ect centering constraints and that govern the centering rules given in Section (cid:24)(cid:0) In Section (cid:8)(cid:1) we discuss applications of the rules and their ability to explain several discourse coherence phenomena(cid:0) In Section (cid:7)(cid:1) we brie(cid:18)y outline the properties of an underlying se(cid:17) mantic framework that are required by centering(cid:0) Finally(cid:1) in Section (cid:3)(cid:26) we conclude with a brief comparison of centering with the research that preceded it and a summary of research that expands on gjw(cid:0)(cid:1)(cid:0) In particular(cid:1) Section (cid:3)(cid:26) provides references to subsequent inves(cid:17) tigations of additional factors that control centering and examinations of its cross(cid:17)linguistic applicability and empirical validity(cid:0) (cid:2) Phenomena to be Explained Discourses are more than meresequences of utterances(cid:0) For a sequence of utterances to be a discourse(cid:1) it must exhibit coherence(cid:0) In this paper(cid:1) we investigate linguistic and attentional state factors that contribute to coherence among utterances within a discourse segment(cid:0) These factors contribute to the di(cid:11)erence in coherence between the following two discourse (cid:13) (cid:1) segments(cid:20) (cid:6)(cid:3)(cid:9) a(cid:0) John went to his favorite music store to buy a piano(cid:0) b(cid:0) He had frequented the store for many years(cid:0) c(cid:0) He was excited that he could (cid:12)nally buy a piano(cid:0) d(cid:0) He arrived just as the store was closing for the day(cid:0) (cid:6)(cid:4)(cid:9) a(cid:0) John went to his favorite music store to buy a piano(cid:0) b(cid:0) It was a store John had frequented for many years(cid:0) c(cid:0) He was excited that he could (cid:12)nally buy a piano(cid:0) d(cid:0) It was closing just as John arrived(cid:0) Discourse (cid:6)(cid:3)(cid:9) is intuitively more coherent than Discourse (cid:6)(cid:4)(cid:9)(cid:0) This di(cid:11)erence may be seen to arise from di(cid:11)erent degrees of continuity in what the discourse is about(cid:0) Discourse (cid:6)(cid:3)(cid:9) centers around a single individual(cid:1) describing various actions he took and his reactions to them(cid:0) Incontrast(cid:1) Discourse (cid:6)(cid:4)(cid:9) seemsto (cid:18)ip backand forth among severaldi(cid:11)erententities(cid:0) More speci(cid:12)cally(cid:1) the initial utterance (cid:6)a(cid:9) in each segment could begin a segment about an individual named (cid:27)John(cid:10) or one about John(cid:10)s favorite music store or one about the fact that John wants to buy a piano(cid:0) Whereas Discourse (cid:6)(cid:3)(cid:9) is clearly about John(cid:1) Discourse (cid:6)(cid:4)(cid:9) has no single clear center of attention(cid:0) Utterance(cid:6)(cid:4)b(cid:9)seemstobeabout thestore(cid:0) Ifareaderinferredthatutterance(cid:6)(cid:4)a(cid:9) was about John(cid:1) then that reader would perceive a change in the entity which the discourse seems to be about in going from (cid:6)(cid:4)a(cid:9) to (cid:6)(cid:4)b(cid:9)(cid:21) on the other hand(cid:1) if the reader took (cid:6)(cid:4)a(cid:9) to be about the store then in going to (cid:6)(cid:4)b(cid:9)(cid:1) there is no change(cid:0) In either case(cid:1) in utterance (cid:6)(cid:4)c(cid:9) John seems to be central(cid:1) requiring a shift from utterance (cid:6)(cid:4)b(cid:9)(cid:1) while the store becomes central again in utterance (cid:6)(cid:4)d(cid:9) requiring yet another shift(cid:0) This changing of (cid:27)aboutness(cid:10) (cid:6)in fact(cid:1) (cid:18)ipping it back and forth(cid:9) makes discourse (cid:6)(cid:4)(cid:9) less coherent than discourse (cid:6)(cid:3)(cid:9)(cid:0) Discourses (cid:6)(cid:3)(cid:9) and (cid:6)(cid:4)(cid:9) conveythe sameinformation(cid:1) but in di(cid:11)erent ways(cid:0) They di(cid:11)er not in content or what is said(cid:1) but in expression or how it is said(cid:0) The variation in (cid:27)aboutness(cid:10) they exhibit arises from di(cid:11)erent choices of the way in which they express the same propositional content(cid:0) The di(cid:11)erences can only be explained(cid:1) however(cid:1) by looking beyond the surface form of the utterances in the discourse(cid:21) di(cid:11)erent types of referring expressions and di(cid:11)erent syntactic forms make di(cid:11)erent inference demands on a hearer or reader(cid:0) These di(cid:11)erences in inference load underlie certain di(cid:11)erences in coherence(cid:0) The model of local attentional state described in this paper provides a basis for explaining these di(cid:11)erences(cid:0) (cid:1) This example(cid:0) and the others in this paper(cid:0) are single speaker texts(cid:1) However(cid:0) centering also applies to dialogue and multi(cid:3)party conversations(cid:1) Issues of the interaction between turn(cid:3)taking and changes in centering status remainto be investigated(cid:1) (cid:24)