Comparing Frame Membership to WN-based Similarity and Distributional Similarity

Abstract

The presentation compares FrameNet membership to WordNet-based measures of similarity and suggests a distribution-based method for the automatic induction of lexical units in FrameNet. FrameNet membership correlated with some WN-based similarity measures, despite the gradable versus binary values and the inapplicability of some measures on co-LUs. Patwardhan’s (2003) vector, Resnik’s (1995) IC, Hirst’s and St-Onge’s (1998) lexical chain and Lin’s (1998) IC measures improved the precision of the distributional-based induction of lexical units.