Arabic word sense disambiguation using sense inventories

Disambiguation of words that have more than one meaning in a context is one of the challenging tasks in natural language processing. The representation of words and context via word embeddings is used in unsupervised approaches to construct sense inventories. In this research, we propose a disambiguation approach that utilizes an unsupervised approach for building a sense inventory from pre-trained embeddings. Part of Speech tagging is applied to the retrieved senses and compared with the tag of the ambiguous word to improve the selection of an appropriate sense. Experiments are conducted on the sentence similarity by using the selected sense vector compared with that of utilizing an ambiguous word vector to evaluate the selected senses. The sense vectors significantly improve the sentence similarity in terms of Pearson correlation. The use of Aravec embeddings provides an enhanced correlation of 0.423. © 2023, The Author(s), under exclusive licence to Bharati Vidyapeeth's Institute of Computer Applications and Management.