Semantic Relations . Theory and Applications 18 May 2010 Editors :

Electronic sense inventories are needed for Word Sense Disambiguation (WSD) in Natural Language Processing (NLP) applications. However, existing sense inventories are criticized for being inadequate for this task. The fine granularity of the senses described in these resources is not always needed in NLP applications. Furthermore, the senses are listed without any information on their distinguishability and their relations. However, this type of knowledge can be automatically extracted from textual data using machine learning techniques. In this paper, we show how an unsupervised sense induction method permits to capture two different types of semantic relations: first, semantic similarity relations between the translation equivalents of ambiguous words; second, relations between their senses, which are automatically identified by means of semantic clustering. We analyze the results of this method and compare them with the semantic descriptions provided in a typical multilingual semantic resource.