Anahit Martirosyan
Word Sense Disambiguation (WSD) is considered a hard problem in linguistic comprehension and Natural Language Processing. Most English words are ambiguous, i.e. they have different meanings in different contexts. WSD is a necessary step in Machine Translation, Information Retrieval, Text and Speech Processing etc. The main source of information which contributes to WSD is context, in which a given word occurs. Optimal number of context words and their distance from the word to be disambiguated play a fundamental role in statistical WSD. This presentation presents an experimental study of word sense ambiguity and context, applied to National Research Council WSD system. We present results of the experiments conducted on the system with different window of context words around the word to be disambiguated. Experiments used corpus-based and lexicon-based measures of semantic similarity. In both cases our results are in accordance with a few previous studies of the topic, proving that a successful window size of words in micro-context, immediately preceding and following the ambiguous word, is relatively small.