Lexical Effects in Verb Sense Disambiguation

First, we propose a unified framework for evaluating verb sense in a selectional restriction-based dictionary architecture, including both generalised and fixed verb senses. The proposed methodology is primarily based around the restrictiveness of the selectional restrictions on each case slot, evaluation of which is made through calculation of the average depth of semantic classes governing the case slot in question. As a spin-off of this restrictiveness measure, we develop the ability to predict likelihood of sense in the case of underspecification, and go on to combine the measure with methods to variously handle word order and surface case variation. By way of experimentation, a test dictionary of 148 verb entries lexically deriving from the Japanese verb mi(-ru) "(to) see", is used to empirically evaluate the effectiveness of the overall system on an open test set of 291 clauses. The overall accuracy on this test set is 92.10%. 1 Introduction Word sense disambiguation (WSD) refe..