Discourse-level Annotation for Investigating Information Structure

Discourse-Level Annotation for Investigating Information Structure Ivana Kruijff-Korbayov´a and Geert-Jan M. Kruijff Computational Linguistics, Saarland University, Saarbr¨ucken, Germany {korbay,gj}@coli.uni-sb.de Abstract We present discourse-level annotation of newspa- per texts in German and English, as part of an ongoing project aimed at investigating information structure from a cross-linguistic perspective. Rather than annotating some specific notion of information structure, we propose a theory-neutral annotation of basic features at the levels of syntax, prosody and discourse, using treebank data as a starting point. Our discourse-level annotation scheme cov- ers properties of discourse referents (e.g., semantic sort, delimitation, quantification, familiarity status) and anaphoric links (coreference and bridging). We illustrate what investigations this data serves and discuss some integration issues involved in combin- ing different levels of stand-off annotations, created by using different tools. 1 Introduction The goal of this paper is to present a discourse- level annotation scheme developed for the pur- pose of investigating information distribution in text from a cross-linguistic perspective, with a particular focus on the interplay of various fac- tors pertaining to the realization of information structure. Information Structure (IS) concerns utterance-internal structural and semantic proper- ties reflecting the speaker’s/writer’s communica- tive intentions and the relation of the utterance to the discourse context, in terms of the dis- course status of the content, the actual and at- tributed attentional states of the discourse partici- pants, and the participants’ prior and changing atti- tudes (knowledge, beliefs, intentions, expectations, etc.) (Kruijff-Korbayov´a and Steedman, 2003). In many (if not all) languages, differences in IS moti- vate variations in surface realization of utterances, such as syntactic structure, word order and intona- tion. But languages differ in the extent to which they employ various combinations of IS-realization means (Vallduv´ı and Engdahl, 1996; Kruijff, 2001). Modeling these phenomena and their interaction re- quires understanding IS and its role in discourse. IS is therefore an important aspect of meaning at the interface between utterance and discourse, which computational models of discourse processing should take into account. Unfortunately, there exists no theory that provides a comprehensive picture of IS, explaining its realization cross-linguistically, its rep- resentation at the level of linguistic meaning, and its interpretation in context. Employing corpora can help to deepen our intuitive understanding of IS, in order to construct explanatorily more adequate the- ories. While the phenomena involved in discourse and IS are themselves complex and not yet fully un- derstood, studying and modeling their interaction is made difficult by proliferating and often under- formalized terminologies, especially for IS (cf. the diverging dichotomies, e.g., Theme-Rheme, Topic- Comment, Topic-Focus, Background-Focus, Given- New, Contextually Bound-Nonbound). What is needed is further systematization of terminologies, formalization and computational modeling, and em- pirical and corpus-based studies. The goal of the MULI (MUltilingual Informa- tion structure) project is to contribute to this effort by empirically analyzing IS in German and English newspaper texts. For this, we designed annotation schemes for enriching existing linguistically inter- preted language resources with information at the levels of syntax, discourse semantics and prosody. The MULI corpus consists …

Content truncated to 500 words