Computational approaches to identification and interpretation of figurative text

Abstract

Figurative language is pervasive in the English language. From a computational linguistics point of view it is a serious problem in automated language processing. In this dissertation, we'll treat the coercion of logical metonymy by exploiting the selectional preferences of both the verbs and the nouns in a predicate argument structure. Induction of sectional preferences is done using topic models, which infer hidden thematic structures in collection of texts. The model will be able to treat for interpretations of metonymy and ranking in order of likelihood. In the end, the model's performance will be compared against human judgements to evaluate its performance on predicting the probability of a predicate and an argument co-occuring together.