Supervised, Semi-Supervised and Unsupervised WSD Approaches: An Overview

Word Sense Disambiguation (WSD) involves the identification of a correct sense of a word in a given sentence. WSD is considered to be an open and AI-complete problem of Natural Language Processing (NLP). WSD is found to be most important in many applications like Machine translation (MT), Information retrieval (IR), Information extraction (IE), text mining, and Lexicography. Supervised, Semi-supervised and Unsupervised Approaches to WSD are found to be important and very successful learning approaches. These methods are categorized based on the main source of knowledge used to differentiate senses or type and amount of annotated (labeled) corpora (data) required. Semi-supervised approach requires lesser quantity of annotated corpora as compared to supervised approaches which needs large amount of annotated corpora while unsupervised approach uses unannotated (unlabeled) corpora for training. In this paper, we will discuss all the three approaches and their respective methods in details