Linguistic Cues for Distinguishing Literal and Non-Literal Usages

摘要

We investigate the effectiveness of differ-ent linguistic cues for distinguishing lit-eral and non-literal usages of potentially idiomatic expressions. We focus specif-ically on features that generalize across different target expressions. While id-ioms on the whole are frequent, instances of each particular expression can be rela-tively infrequent and it will often not be feasible to extract and annotate a suffi-cient number of examples for each expres-sion one might want to disambiguate. We experimented with a number of different features and found that features encoding lexical cohesion as well as some syntac-tic features can generalize well across id-ioms.