Towards a Conceptual Representation of Lexical Meaning in WordNet

Knowledge acquisition is an essential and intractable task for almost every natural language processing study. To date, corpus approach is a primary means for acquiring lexical-level semantic knowledge, but it has the problem of knowledge insufficiency when training on various kinds of corpora. This paper is concerned with the issue of how to acquire and represent conceptual knowledge explicitly from lexical definitions and their semantic network within a machine-readable dictionary. Some information retrieval techniques are applied to link between lexical senses in WordNet and conceptual ones in Roget's categories. Our experimental results report an overall accuracy of 85.25% (87, 78, 89, and 87% for nouns, verbs, adjectives and adverbs, respectively) when evaluating on polysemous words discussed in previous literature.