Disambiguation with Feature Selection and Semi - Supervised Learning ”

1. Objective Word Sense Disambiguation (WSD) is the task of determining the right sense of a polysemous word in a given context. This study aims to enhance the performance of supervised-based word sense determination by focusing on feature selection and using bootstrapping techniques. Senses determination of a word is essentially based on the information extracted from the context in which this word appears. The information is represented by a set of features: unordered words, ordered words, collocations (sequences of words including the target word), and grammatical constituents. With an observation that: some of these features may be redundant and various ways of combining these features will cause different results of the WSD task, we focused our first work on feature selection. Another problem is that of lacking labeled data for supervised methods. While labeled data is scarce or expensive, an abundance of unlabeled data is available but it is not be utilized by conventional supervised methods. Therefore the second problem in my study is utilizing unlabeled data in WSD by using semi-supervised learning algorithms (or bootstrapping algorithms).