Semantic Signatures for Example-Based Linguistic Metaphor Detection

Metaphor is a pervasive feature of human language that enables us to conceptualize and communicate abstract concepts using more concrete terminology. Unfortunately, it is also a feature that serves to confound a computer’s ability to comprehend natural human language. We present a method to detect linguistic metaphors by inducing a domainaware semantic signature for a given text and compare this signature against a large index of known metaphors. By training a suite of binary classifiers using the results of several semantic signature-based rankings of the index, we are able to detect linguistic metaphors in unstructured text at a significantly higher precision as compared to several baseline approaches.