International Conference on Computer and Knowledge Engineering

Home / 14th International Conference on Computer and Knowledge Engineering

Enhancing Persian Word Sense Disambiguation with Large Language Models: Techniques and Applications

Authors :

Fatemeh Zahra Arshia¹ Saeedeh Sadat Sadidpour²

1- Faculty of Electronic & Computer Engineering, Malek Ashtar University of Technology 2- Faculty of Electronic & Computer Engineering, Malek Ashtar University of Technology

Keywords :

Word Sense Disambiguation (WSD)،Large Language Models (LLMs)،Persian Disambiguation

Abstract :

WSD means the task of word sense disambiguation, which is a very important task in NLP. It assigns not only the meaningful word to the source text but also the proper meaning of the word according to the context. Hence, it is key to the proper accomplishment of NLP in Persian—a language rich in morphology and great polysemy. The recent improvements in LMs have greatly advanced the capabilities of NLP, opening further improvement avenues in WSD performance. This paper presents the integration of LLMs for improving WSD in Persian, considering the linguistic challenges related to this language. In this study, we consider four models of the Persian language: FaBERT, AriaBERT, GPT-2 Persian, and PersianMind-v1.0. We use the supervised fine-tuning method on the SBU-WSD-Corpus. Our methodology will consist of preprocessing the Persian WSD corpus, then fine-tuning the models with the mentioned corpus, and measuring their performance. Results indicate that methods using LLMs significantly improve WSD accuracy against traditional methods, with FaBERT achieving the best accuracy. We have further expounded on their real-life applications, such as sentiment analysis, to show the consequential effect of this advancement on general NLP tasks. The study is concluded with some insights into future research directions, underlining the potential role that LLMs can play in further transforming WSD and related fields.