Context similarity based hybrid approach for extracting hindi multiword expressions

Abstract

In natural language processing, multiword expressions (MWEs) play a significant role in understanding the context and meaning of a sentence. A MWE comprises two or more words that are handled as if they were one. MWE has the property that the constituent words are consistent and are often used in related contexts. Hindi is used as a case study. We employed three properties: linguistic or syntactical pattern, a relationship between constituent words, and context similarity and proposed a three-phase hybrid approach to extract MWE from unstructured Hindi text. Experimental analysis and comparison of results on the TDIL dataset show the superiority of the proposed hybrid method over the context-based method and association-based methods.