Logical Metonymy: Discovering Classes of Meanings

We address the problem of interpretation of logical metonymy using a statistical method. Previous approaches to logical metonymy produce interpretations in the form of verb senses, whereas our definition of the interpretation is a cluster of verb senses. Such a class-based computational model of logical metonymy is novel and more informative than the previous ones. It also complies with the linguistic theories, which we empirically validate. We propose feature sets not previously used for verb clustering. In addition to this we conduct an experiment to prove that our representation is intuitive to human subjects. The system clusters the senses with the F-measure of 0.64 as compared to the gold standard, given that the human agreement on the task is 0.76.