Empirical Antonym Implementation in the UMLS SPECIALIST Lexicon

Antonyms are words that have opposite or contrasting meanings in a specific domain. For example, "increase" is the opposite of "decrease" in the domain of "quantity". Antonyms play an important role in NLP applications to improve performance. This paper describes a systematic approach to generate antonyms. We developed five source models using negation rules, derivational morphology, co-occurrences in a corpus and semantic networks. As a result, 13K of canonical antonyms along with their associated features, such as bounded types, canonical domains, and negations are generated. They are released with the SPECIALIST Lexicon 2025 to provide broad coverage of antonyms with comprehensive features needed for NLP applications. Finally, analyses on antonym sources, canonicity, Part-of-Speech, domains and usage of types and negation are reported.