Challenges and Overcoming Methods for Word Sense Disambiguation

Disambiguation of Word Sense is a challenging and tough task in computational linguistics which still sustains as a tricky problem for a long period of research. It is important and difficult, since the words may not be directly related to each other. WSD corpora leads to an expensive annotation process because of its small size. Almost all the issues in linguistics have a development in its resolving methods for at least 90% but the only challenging task which has not improved more than 80% till now is WSD because of its language semantics i.e. time taken is much larger to get the semantic annotations for a considerably more count of sentences. In this paper, a survey is undertaken on various challenges with WSD and overcoming algorithms to disambiguate words. The algorithms are specified based on three categories knowledge-based, supervised, and unsupervised techniques. The old methods of resolving solved nearly 75% and the new tools managed to achieve 85% which is much more satisfiable. From this work, it is concluded that (i) accuracy differs based on methods, (ii) the algorithm depends on the size of the used data set (iii) most of these approaches can be implemented for multiple languages successfully (iv) Hybrid models shall be involved to get better results (v) Recent trending NLP tools namely BERT, ELMO, Roberta, GlossBERT, SenseEMBERT, etc resolves the issues of WSD more efficiently.