Metaphor Detection Using Fuzzy Rough Sets

摘要

Recent works in metaphor detection highlight the importance of psychological features such as imageability and concreteness to identify metaphors in text. However, the aspect of imprecision that is intrinsic to cognitive concepts is yet to be explored fully. Furthermore, psychological features give us an approximate indication of whether a particular textual usage is metaphorical or not. In this paper, we reflect upon the problem of the inherent vagueness in psychological features and approximation in classification through the notion of fuzzy rough sets. We develop a fuzzy-rough rule-based classifier to detect metaphors in text and evaluate the performance of the proposed model on a dataset of nominal metaphors. The results indicate the suitability of incorporating fuzzy-rough sets over SVM and the traditional rough set model.