Edited by Helmut Gruber and Gisela Redeker
[Pragmatics & Beyond New Series 254] 2014
► pp. 121–141
Automatic discourse parsing refers to the identification of coherence relations and deriving a structural description for a text. Such parsers can derive much information from the presence of surface cues, especially connectives. These lexical signals, however, are ambiguous: Many have additional, non-connective readings; also, many connectives can signal more than one coherence relation. In this paper, we discuss the first problem, focusing on English and German: How many connectives are ambiguous, and how frequent are these in the two languages? Then we examine computational approaches for resolving such ambiguities. For English, we provide an overview of relevant work by other researchers, while for German we largely present our own studies on the utility of part-of-speech tagging for connective disambiguation.