Utilizing Semantic Composition in Distributional Semantic Models for Word Sense Discrimination and Word Sense Disambiguation

Semantic composition in distributional semantic models (DSMs) offers a powerful tool to represent word meaning in context. In this paper, we investigate methods to utilize compositional DSMs to improve word sense discrimination and word sense disambiguation. In this work, we rely on a previously proposed multiplicative model of composition. We explore methods to extend this model to exploit richer contexts. For word sense discrimination, we build context vectors, which are clustered, from the word representations based on the extended compositional model. For word sense disambiguation, we augment lexical features with their word representations based on the same extended compositional model. For both tasks, we achieve substantial improvement.