The NLP task of determining which meaning of a word is intended in a given context.
Word Sense Disambiguation (WSD) is the computational linguistics task of determining which sense (meaning) of a polysemous word is intended in a given context. The word 'bank' can mean a financial institution, a riverbank, or the act of tilting an aircraft — WSD determines which meaning is correct based on surrounding context. WSD is fundamental to accurate natural language understanding, as most common words have multiple senses.
WSD has become critical for enterprise AI accuracy. When AI agents process business documents, they must correctly interpret domain-specific terminology — 'pipeline' means something different in sales, software engineering, and oil & gas. Without WSD, AI systems misinterpret terms and produce incorrect outputs. Modern LLMs handle WSD implicitly through attention mechanisms, but explicit WSD systems are still used in high-stakes applications where precision is required.
WSD approaches include knowledge-based methods (using WordNet or domain ontologies to select the sense that best fits the context), supervised learning (training classifiers on sense-annotated corpora like SemEval), and unsupervised methods (clustering word usages to discover senses). Modern approaches use contextual embeddings from transformer models — the embedding of a word in context naturally encodes its sense, enabling sense disambiguation through nearest-neighbor classification.
A financial AI processes the sentence 'The bank approved the loan.' WSD correctly identifies 'bank' as a financial institution (not a riverbank) based on the context of 'loan.' In a different sentence, 'The pilot banked the aircraft to the left,' WSD correctly identifies 'banked' as the aviation maneuver sense. This precision is critical when extracting financial events from news articles.