A COMPARATIVE ANALYSIS OF WORD SENSE DISAMBIGUATION METHODS FOR IDENTIFYING POLYSEMY AND HOMONYMY IN ENGLISH

Abstract

Abstract. This study presents a comparative and empirical analysis of Word Sense Disambiguation (WSD) methods in identifying polysemy and homonymy in English. The research evaluates three major approaches-knowledge-based, statistical, and deep learning-under a unified experimental framework using a semantically annotated dataset derived from SemCor and WordNet. The performance of the models was assessed using standard evaluation metrics, including accuracy, precision, and recall. The results demonstrate that the knowledge-based method shows limited effectiveness in complex contexts due to its reliance on lexical overlap, while the statistical approach provides more stable performance by capturing contextual co-occurrence patterns. The highest results were achieved by the transformer-based BERT model, which consistently outperformed other methods by effectively modeling contextual semantics at multiple levels. The findings highlight that the ability to capture deep contextual information is a critical factor in resolving semantic ambiguity, particularly in cases of highly polysemous and homonymous words. In addition to quantitative evaluation, qualitative error analysis revealed that metaphorical and idiomatic usages remain challenging for all approaches, indicating the need for further improvements. The study contributes to the field by providing a systematic comparison of WSD methods under consistent conditions and offers practical insights for improving natural language processing applications such as machine translation and automated text analysis.

Annotatsiya. Mazkur tadqiqot ingliz tilidagi polisemiya va omonimiyani aniqlashda so‘z ma’nosini ajratish (Word Sense Disambiguation, WSD) usullarining qiyosiy va empirik tahliliga bag‘ishlangan. Tadqiqotda bilimga asoslangan, statistik va chuqur o‘rganishga asoslangan yondashuvlar yagona eksperimental muhitda baholandi hamda SemCor va WordNet asosida shakllantirilgan semantik belgilangan ma’lumotlar to‘plamidan foydalanildi. Modellar samaradorligi aniqlik, precision va recall kabi mezonlar orqali baholandi. Natijalar shuni ko‘rsatdiki, bilimga asoslangan yondashuv lug‘aviy moslikka tayanishi sababli murakkab kontekstlarda past samaradorlikni namoyon qiladi, statistik model esa kontekstual bog‘liqliklarni hisobga olgani tufayli nisbatan barqaror natijalarni beradi. Eng yuqori natijalar transformer arxitekturasi asosidagi BERT modeli tomonidan qayd etildi, bu esa uning kontekstni ko‘p qatlamli semantik darajada qayta ishlash qobiliyati bilan izohlanadi. Tadqiqot natijalari chuqur kontekstual tahlilning semantik noaniqlikni bartaraf etishda hal qiluvchi omil ekanligini tasdiqlaydi. Shuningdek, xatoliklar tahlili metaforik va idiomatik qo‘llanishlar barcha modellarda murakkablik tug‘dirishini ko‘rsatdi. Ushbu tadqiqot WSD usullarini bir xil sharoitda qiyosiy baholab, tabiiy tilni qayta ishlash tizimlarida semantik aniqlikni oshirish uchun amaliy tavsiyalar beradi.