Discourse Coherence Through the Lens of an Annotated Text Corpus: A Case Study

Discourse Coherence Through the Lens of an Annotated Text Corpus: A Case Study Eva Hajičová, Jiří Mírovský Charles University Faculty of Mathematics and Physics Institute of Formal and Applied Linguistics Malostranské nám. 25, 118 00 Praha 1, Czech Republic {hajicova, mirovsky}@ufal.mff.cuni.cz Abstract A corpus-based study of local coherence as established by anaphoric links between the elements in the thematic (Topic) and the rhematic (Focus) parts of sentences in different genres of discourse. The study uses the Czech data present in the Prague Dependency Treebank and annotated for surface and underlying syntactic relations, the contextual boundness of tree nodes (from which the bi- partition of the sentence into Topic and Focus can be derived) and the coreference and bridging relations. Among the four possible types of the relations between anaphoric links and the Topic–Focus bipartition of the sentence, the most frequently occurring type is a link between the Topic of the sentence to the Focus of the immediately preceding sentence. In case there is an anaphoric link leading from the Focus of one sentence to the Topic or Focus of the immediately preceding sentence, this link frequently leads from a contextually bound element of the Focus, which supports the assumption that it is convenient to distinguish between “overall” Topic and Focus and the local Topic and Focus and/or the anaphoric relation is of the type of bridging and the relationship is often interpreted as a contrast. As for the relationship between the relations of the Topic-to-Topic type, due to the word order typological difference for Czech and English, these relations in Czech are not at all related to the syntactic function of subject. Keywords: discourse, coherence, coreference 1. Introduction One way how to look at discourse is to view it as a sequence of utterances linked by coherence relations (Halliday and Hasan, 1976). There are several possible ways how to account for these relations: e.g. they may be described on the basis of coreferential links between the elements present in the utterances, or on the basis of some discoursive links between segments of the adjacent utterances. In the present contribution, we first characterize some of the hitherto described approaches to this issue passing over to an examination of a possibility to look at the text coherence taking into account both the information structure of the utterances and the anaphoric relations. The proposed approach is based on a sample of Czech text corpus (the Prague Dependency Treebank 3.0) annotated for deep syntactic relations, information structure, and coreference and discourse relations. 2. Related Theories One of the most deeply elaborated and best known theories of discourse (local) coherence is the so-called centering theory (Grosz, Joshi and Weinstein, 1983) based on the model of the local attentional states of speakers and hearers as proposed by Grosz and Sidner (1986). Each utterance in discourse is considered to contain a backward looking center, which links it with the preceding utterance, and a set of entities called forward looking centers; these entities are ranked according to language- …

Content truncated to 500 words