A Knowledge-Driven Approach to Word Sense Disambiguation through Lexical Semantics and Ontologies Integration for Enhanced Precision in Natural Language Processing

Abstract

Statistical and corpus-based approaches to Word Sense Disambiguation (WSD) face significant limitations in resource-limited environments and in domains requiring specific adaptability, due to their dependence on large annotated datasets and their limited interpretability. To address these challenges, this research proposes a knowledge-driven WSD framework that leverages structured semantic information from WordNet, FrameNet, BabelNet, and domain-specific ontologies to enhance adaptability and precision. The framework integrates semantic role labelling, graph-based knowledge models, and knowledge-informed similarity calculation methods to  identify word senses across diverse linguistic contexts. Experimental evaluation on SemCor and OntoNotes benchmarks demonstrates that the KDF-WSD-LSO (Knowledge-Driven Framework for Word Sense Disambiguation with Lexical-Semantic Ontologies)  achieves an F1-score of 87.3%, a 12.1% improvement over corpus-only baselines—and maintains 91.2% precision in low-resource medical and legal domains. These results confirm that combining lexical semantics with ontological reasoning yields scalable, interpretable, and accurate WSD solutions for key NLP applications, especially where annotated data is scarce.