A Proposal for WSD Using Semantic Similarity

The aim of this paper is to describe a new method for the automatic resolution of lexical ambiguity of verbs in English texts, based on the idea of semantic similarity between nouns using WordNet. 1 An outline of our approach. The method of WSD proposed in this paper is based on knowledge and consists basically of sense-disambiguating of the verb that appear in an English sentence. A simple sentence or question can usually be briefly described by an action and an object [1]. For example the main idea from the sentence "He eats bananas" can be described by the action-object pair "eat-banana". Our method determine which senses of these two words are more similar between themselves. For this task we use the concept of semantic similarity [2] between nouns based on WordNet [3] hierarchy. In WordNet, the gloss of a verb synset provides a nouncontext for that verb, i.e. the possible nouns occurring in the context of that particular verb [1]. The glosses are used here in the same way a corpus is used. Our method takes into consideration the verb-noun pair extracted from the sentence. This verb-noun pair is the input for the algorithm. The output will be the sense tagged verb-noun pair, so we assign the sense of the verb. The algorithm is described as follows: Step 1. Determine all the possible senses for the verb and the noun by using WordNet. Let us denote them by and Step 2. For each sense of verb vh and all senses of noun : 2.1. Extract all the glosses from the sub-hierarchy including vh. The sub-hierarchy including a verb vh is determined as follows: consider the hypernym hh of the verb vh and consider the hierarchy having hh as top [1]. 2.2. Determine the nouns from these glosses. These constitute the noun-context of the verb. Determine all the possible senses for all these nouns. Let us denote them by . 2.3. Then we obtain the similarity matrix (Sm) using the semantic similarity, where each element is defined as follows: Sm(i, j) = sim (xi, nj) For determining the semantic similarity (sim(xi, nj)) between each sense of the nouns extracted from the gloss of verb and each sense of the input noun, we use the formula followed: sim (xi, nj) = 1 – sd (xi, nj) 2 sd (xi, nj) = 2 1 . ( 1 1