Comparing Conceptual Metaphor Theory-Related Features Using Classification Algorithms in Searching for Expressions Used Figuratively Within Japanese Texts

Abstract

In this work we propose an approach towards identifying expressions used figuratively in Japanese literary texts by means of the classification algorithm. Our considerations are inspired mostly by the epoch-making Conceptual Metaphor Theory which once presented by Lakoff and Johnson [13] has instantly become a central problem to address not only from the perspective of cognitive linguistics but also other domains of scholarship. In NLP-related research, approaches of this sort - although already present - seem to be rarely applied to models dedicated to Japanese language. In our preliminary approach to build a trustworthy model able to create a large corpus of figurative expressions, we compare efficiency of several features selected as possibly relevant in the task of detecting non-literal language use. Our algorithm does not require complex system of rules which was the case in some of the computational models presented by other authors so far.