Identification and Disambiguation of Lexical Cues of Rhetorical Relations across Different Text Genres

Lexical cues are linguistic expressions that can signal the presence of a rhetorical relation. However, such cues can be ambiguous as they may signal more than one relation or may not always function as a relation indicator. In this study, we first conduct a corpus-based analysis to derive a set of n-grams as potential lexical cues. These cues are then utilized in graph-based probabilistic models to determine the syntactic context in which the cue is signaling the presence of a particular relation. Evaluation results are reported for various cues of the CIRCUMSTANCE relation, confirming the value of syntactic features for the task of cue disambiguation in the context of Rhetorical Structure Theory. Moreover, using a graph to encode syntactic information is shown to be a more generalizable and effective approach compared to the direct usage of syntactic features.