Word Sense Disambiguation

A word may have more than one meaning or sense (polysemous). For instance, the word “light” may have a meaning of “a source of lighting or illumination” as in “Turn off the light”. It can also have a meaning of “not heavy” as in “My new phone is small and light”. The meaning can usually be differentiated or disambiguated by observing the context where the word is used. In a written text or sentence, the context can be described in the simplest case as the information of the surrounding words. In a more complex case, as in the use of word bank in “I’ll meet you at the bank”, the disambiguation requires more than just the information from the surrounding words. Since, the bank may have a meaning of “a financial institution (a building)” but can also have a meaning of “river bank”. Both meanings are possible and the correct meaning will depend on the shared knowledge between the speakers. Word sense disambiguation is the task to decide, given a text or sentence, the correct meaning of a certain word inside the text. In this project, we experimented in disambiguating the word senses of three different words: hard, line, and serve. We use the data developed by (Leacock). The data contains sentences with the target word (word to be disambiguated) tagged and disambiguated using the senses from WordNet (Fellbaum). There will be one word that we need to disambiguate in each sentence and hence, the meaning of the word will be disambiguated using this one sentence context. We will use three different machine learning methods to do the classification of the word into their respective sense. The performance of each method will be measured by its accuracy, i.e. the number of sentences with correct prediction divided by the total number of sentences predicted. We will describe the data and the features we use in Section 2. Section 3 will provide theoretical background on the selected machine learning methods. In Section 4, we will describe our experiments and our analysis of the results. The conclusions of the work will be given in Section 5.