Meta4meaning: Automatic Metaphor Interpretation Using Corpus-Derived Word Associations

We propose a novel metaphor interpretation method, Meta4meaning. It provides interpretations for nominal metaphors by generating a list of properties that the metaphor expresses. Meta4meaning uses word associations extracted from a corpus to retrieve an approximation to properties of concepts. Interpretations are then obtained as an aggregation or difference of the saliences of the properties to the tenor and the vehicle. We evaluate Meta4meaning using a set of human-annotated interpretations of 84 metaphors and compare with two existing methods for metaphor interpretation. Meta4meaning significantly outperforms the previous methods on this task.