Latent semantic similarity based interpretation of Chinese metaphors

Abstract

As a ubiquitous usage in both text and speech, metaphor now attracts more and more attention. Automatic metaphor processing can be divided into two subtasks: metaphor detection and metaphor interpretation. This paper describes an algorithm to interpret the Chinese nominal and verbal metaphors based on latent semantic similarity which we define in this paper. Our method extends the perceptual features of the source and target concepts using the synonyms in WordNet to discover the latent semantic similarity between them and thereby generates the interpretation of nominal metaphors. It is considered that if two words are latent semantic similar, not only is there an extension path in WordNet from one to the other, but also their sentiments should be consistent. So the sentiment of the word is used to constrain the extension. Without a context, we think that the results of interpretation may be multiple because there are several features of the source that can be used to describe the target. Thus, we use Google Distance to rank the interpretation results. This model achieves 85% accuracy in nominal metaphors and 86% accuracy in verbal metaphors.