Low ambiguity first algorithm: A new approach to knowledge-based word sense disambiguation

Abstract

The Word Sense Disambiguation (WSD) problem has been considered as one of the most important challenging task in Natural Language Processing (NLP) research area. Even though, many of scientists applied the robust machine learning, statistical techniques, and structural pattern matching approach, the performance of WSD is still not able to bit human results due to the complexity of human language. In order to overcome this limitation, currently, the knowledge base such as WordNet has gained high popularity among researchers due to the fact that this knowledge base can extensively provide not only the definitions of nouns and verbs, but also the semantic networks between senses which were defined by linguists. However, knowledge bases are not fully dealing with entire words of human languages because maintaining and expanding the knowledge base is huge task which requires many efforts and time. Expanding knowledge base is not a big issue to concern however, a new approach is the major goal of this paper to solve WSD problem only based on limited knowledge resources. In this paper, we propose a method, named low ambiguity first (LAF) algorithm, which disambiguates a polysemous word with a low ambiguity degree first with given disambiguated words, based on the structural semantic interconnections (SSI) approach. The LAF algorithm is based on the two hypothesises that first, adjacent words are semantically relevant than other words far way. Second, word ambiguity can be measured by frequency differences between synsets of the given word in WordNet. We have proved these hypothesises in the experiment results, the LAF algorithm can improve the performance of traditional WSD results.