Anaphoricity in connectives: From corpus analysis to lexical description and consequences for discourse parsing

点击翻译按钮获取中文摘要

In language, discourse connectives are used to signal the presence of coherence relations between two text segments, such as causal, temporal or contrastive relationships. From the syntactic viewpoint, connectives can be conjunctions or certain adverbials. While conjunctions have been studied intensively on various levels of description, many questions are open for the discourse adverbials, regarding the mapping to their arguments (the text segments being related) and their role for constructing the discourse structure, which in turn is important for the reader, e.g. when resolving anaphoric pronouns.A central role in our investigations of discourse adverbials is played by the notion of anaphoricity, which in many connectives is given by an overt anaphoric morpheme. Our project adopts a bilingual perspective and uses German, English, and parallel bilingual text corpora for, in the first step, studying the usage conditions of these connectives: How do readers handle ambiguities, when a connective can signal different coherence relations? Under what circumstances and by what means is an argument of a connective being "constructed" during reading, in case it is not explicitly given in the text? How do readers determine the boundaries of the text segments being related - when are they identified clearly, and when do they remain rather vague, as it can happen for instance with a sentence-initial "but"? How do readers cope with potentially complex interactions between multiple connectives present within the same sentence?Studying these questions for two languages on the one hand allows for systematic comparison of German and English connectives; on the other hand it broadens the perspective for the consequences that discourse adverbials have for the level of discourse structure. Earlier empirical work has been done largely on English data; yet discourse-structural factors should be studied from a multi-lingual perspective if possible. We plan to accomplish this through comparisons to the earlier results for English.In addition, the results will be transferred to work in automatic text analysis ("text mining"). To this end, the information gathered for individual connectives will be modeled in formal lexicon entries that can then be used by an automatic discourse parser; a prototype of such a system will be developed as part of the project.