An analysis of word sense disambiguation in Bangla and English using supervised learning and a deep neural network classifier

Abstract

Word sense disambiguation is a significant task in natural language processing and addresses an old problem in the field of computational linguistics. Word sense disambiguation facilitates tasks like machine translation, information retrieval, text-to-speech and other application systems. Sense ambiguity is introduced because certain words for a given language can have multiple meanings. Word sense disambiguation involves identifying the correct sense of a word for a given context. Bangla language has a few cases of word sense ambiguity. Many machine learning algorithms have already been applied to disambiguate word sense including few implementations of neural networks. However, many existing word sense disambiguation systems using supervised learning and neural networks were focused only on English data. And an evaluation of their implementation on Bangla data is necessary. This thesis attempts to analyze and present results when using an updated deep neural network classifier and supervised learning algorithm for word sense disambiguation on a Bangla dataset and on an English dataset. A pre-processor for the dataset was constructed to appropriately extract features from the context sentences to build sets of refined feature vectors which are fed into the neural network.