Hybrid PoS-tagging: A cooperation of evolutionary and statistical approaches

The assigning of syntactic categories to words in a sentence, which is referred to as part-of-speech (PoS) tagging problem, plays an essential role in many natural language processing and information retrieval applications. Despite the vast scope of methods, PoS-tagging brings an array of challenges that require novel solutions. To address these challenges in a principled way, one solution would be to formulate the tagging problem as an optimization problem with well-specified objectives and then apply the evolutionary methods to solve the optimization problem. This paper discusses the relative advantages of different evolutionary approaches to handle Part-of-Speech tagging problem and aims at presenting novel language-independent evolutionary algorithms to solve the PoS tagging problem. We show that by exploiting statistical measures to evaluate the solutions in tagging process, the proposed algorithms are able to generate more accurate solution in a reasonable amount of time. The experiments we have conducted on few well known corpus reveal that the proposed algorithms achieve better average accuracy in comparison to other evolutionary-based and classical Part-of-Speech tagging methods.