Connective-based Local Coherence Analysis: A Lexicon for Recognizing Causal Relationships

Local coherence analysis is the task of deriving the (most likely) coherence relation holding between two elementary discourse units or, recursively, larger spans of text. The primary source of information for this step is the connectives provided by a language for, more or less explicitly, signaling the relations. Focusing here on causal coherence relations, we propose a lexical resource that holds both lexicographic and corpusstatistic information on German connectives. It can serve as the central repository of information needed for identifying and disambiguating connectives in text, including determining the coherence relations being signaled. We sketch a procedure performing this task, and describe a manually-annotated corpus of causal relations (also in German), which serves as reference data.