pywsd

Why we included this project

Word sense disambiguation is the task of picking the right meaning for a word like 'bank' from context, and pywsd packages the classic algorithms for it in pure Python. You get the original Lesk method, an adapted version, a cosine-based overlap variant, and several WordNet similarity measures (path, Wu-Palmer, Resnik, Jiang-Conrath, Lin), plus simple baselines to compare against. The API is small: you can disambiguate one word in a sentence or run all-words disambiguation over a whole utterance, and you get back WordNet synsets with definitions and lemmas ready for downstream use. It runs on Open English WordNet 2024, which it downloads automatically on first use, and bundles the information-content files the similarity metrics need. That makes it a useful reference implementation for NLP researchers and students, and a practical starting point for engineers who want lexical-semantics features without jumping straight to a heavy supervised model.