Lesk

Lesk is a classic algorithm for word sense disambiguation (WSD) introduced by Michael Lesk in 1986. It uses a sense inventory, typically WordNet, and the dictionary definitions (glosses) of sense entries to determine which sense of a polysemous word best fits the surrounding context. The method is unsupervised and relies on textual overlap between the context and the glosses.

In its basic form, the algorithm takes a target word in a sentence and a set of candidate senses from the inventory. It builds a contextual window around the word (for example, surrounding words). For each candidate sense, it collects the gloss of that sense, and, in extended variants, glosses of related senses such as hypernyms or hyponyms. It then computes an overlap score by counting shared content words between the context and the gloss signature. The sense with the highest overlap is chosen as the disambiguated sense. A commonly discussed variant is the original or simple Lesk; a Simplified Lesk variant uses only the target sense gloss, while Extended Lesk broadens the signature with related definitions to improve coverage.

Strengths of Lesk include its interpretability and independence from large annotated corpora, making it a useful baseline in WSD research. Its effectiveness depends on the richness of the sense inventory and the presence of definitional vocabulary that matches the context. Limitations include reduced performance with short or noisy contexts, reliance on lexical resources, and potential miss of senses not well represented in the glosses. Lesk has paved the way for more sophisticated knowledge-based and hybrid WSD approaches and remains a reference point in the field.