Primitive-Based Word Sense Disambiguation for SENSEVAL-2

This paper describes a descriptive-semantic-primitive-based method for word sense disambiguation (WSD) with a machine-tractable dictionary and conceptual distance data among primitives. This approach is using unsupervised learning algorithm and focuses only on the immediately surrounding words and basis morphological form to disambiguate a word sense. This approach also agrees with past observations that human only requires a small window of a few words to perform WSD. (Choueka & Lusignan, 1985). In additional, this paper also describes our experience in doing the English all-word task in SENSEVAL-2. Then, we will discuss the results in the SENSEVAL-2 evaluation. Apart from the description of current system, possibilities for future work are explored