Swarm optimization for Arabic word sense disambiguation based on English pre-trained word embeddings

Abstract

In this article, we present a new approach to word sense disambiguation for Arabic language based on the notion of local and global algorithms. We are going to use LESK defined on a distributional semantic space to compute the gloss-context overlap for disambiguation of words in the local context and the Cuckoo Optimization Algorithm to propagate local measures at the upper level. This task needs lexical resources and since Arabic lacks them, we are using English pre-trained word embeddings. Experimental results show that the proposed WSD approach significantly improves the base-line word sense disambiguation method. Furthermore, it will be easier to compare our results to other methods. In addition, we compared different pre-existing word embeddings model in our approach.