A Comprehensive Comparison of Knowledge-Based, Supervised, and Unsupervised Techniques for Word Sense Disambiguation

Shruti Garg

Abstract: Word Sense Disambiguation (WSD) is a cornerstone and persistent challenge in Natural Language Processing (NLP), critical for enabling machines to achieve a human-like understanding of language. The task involves computationally identifying the intended meaning of a polysemous word within a specific context. This paper presents a comprehensive survey and a rigorous comparative analysis of the three dominant paradigms in WSD: Knowledge-Based, Supervised, and Unsupervised methods. We provide an in-depth examination of the core methodologies, tracing their evolution from early heuristic and graph-based approaches to modern deep learning and sense embedding techniques. The comparison is structured across multiple dimensions, including performance, data dependency, computational efficiency, robustness, and interpretability. Our analysis confirms that while supervised deep learning models achieve state-of-the-art results on benchmark tasks, they are fundamentally constrained by the knowledge acquisition bottleneck?the scarcity of sense-annotated data. Knowledge-based methods offer greater domain independence and interpretability but often lag in accuracy. Unsupervised methods and, more recently, knowledge-informed neural models present a promising path forward by leveraging large, unlabeled corpora and structured lexical resources. The paper concludes that the optimal WSD technique is highly application-dependent, and future breakthroughs will likely stem from hybrid architectures that seamlessly integrate the robustness of knowledge bases with the representational power of contextualized language models.

Keywords: Word Sense Disambiguation, Natural Language Processing, Computational Linguistics, Senseval, SemEval, Lesk Algorithm, Supervised Learning, Unsupervised Learning, Neural Networks, Word Embeddings, BERT

How to Cite?: Shruti Garg, "A Comprehensive Comparison of Knowledge-Based, Supervised, and Unsupervised Techniques for Word Sense Disambiguation", Volume 14 Issue 11, November 2025, International Journal of Science and Research (IJSR), Pages: 847-850, https://www.ijsr.net/getabstract.php?paperid=SR251110145436, DOI: https://dx.doi.org/10.21275/SR251110145436