Applying PSO to natural language processing tasks: Optimizing the identification of syntactic phrases

George Tambouratzis

Abstract: The present article discusses the use of Particle Swarm Optimisation (PSO) in a natural language processing task, namely the creation of a phrasing model which splits any sentence into linguistically-motivated phrases. This involves taking a limited-size training dataset, where sentences are split into syntactically motivated sentences, and learning how to best segment arbitrary sentences into their corresponding phrases. The extrapolation of phrases is needed in numerous applications involving the generation of texts in natural language. One such application is machine translation, namely the automatic translation of unconstrained text from a source language to a target language. The phrasing model to be learnt comprises a number of parameters which need to be optimized based on the training data. To that end, a machine learning approach has been developed, which is based on the concept of attractive and repulsive forces. Instead of previous efforts using manual tuning of the parameters, here PSO is used to achieve the optimization of the model parameters. Experimental results indicate that the proposed PSO approach is promising, giving the most accurate phrasing in this specific application, with statistically significant improvements over earlier results.