Using Word Space Models for Enriching Multilingual Lexical Resources and Detecting the Relation Between Morphological and Semantic Composition

Over the past 15 years, WordSpace models have been increasingly used as practical ways of capturing and representing distributional information about words. Indeed, various experiments reported in the literature demonstrate that WordSpace models can be used to create or enrich Ontologies. However, the results of such experiments have been difficult to formalize, or to demonstrate that WordSpace based methods are accurate enough to reliably replace or at least supplement the work of professional knowledge engineers. In this paper we review two specific proposals, in which we hope to demonstrate that WordSpace models can be used for clearly defined linguistic tasks such as learning translation pairs and producing meaning representations for morphological compositions. As motivation, we describe the family of WordSpace models, review experiments in the literature that give us initial confidence, and explain how these can be taken much further.