SemCor

SemCor (Semantic Concordance) is a large, manually sense-tagged corpus of English designed for word sense disambiguation research. It aligns tokens with sense identifiers from WordNet, providing a gold-standard resource for evaluating and training WSD methods. The corpus is annotated at the word sense level; each content word in the text is tagged with the intended WordNet sense, and the annotation includes part-of-speech information.

The corpus comprises texts drawn from various genres to ensure lexical and stylistic diversity. The tagging was carried out by linguists and annotators, with adjudication to resolve disagreements. The sense alignment follows WordNet's synsets, enabling comparisons across WSD systems and integration with WordNet-based NLP tasks. The corpus covers multiple parts of speech and includes disambiguation for verbs, nouns, adjectives, and adverbs.

SemCor has become a standard benchmark in the word sense disambiguation community and is frequently used for evaluating WSD algorithms, feature engineering, and for training supervised models. It has influenced the development of corpus-based approaches to lexical semantics and contributed to resources linking corpus data to WordNet.

Availability and history: SemCor was produced in the late 1990s to early 2000s by researchers affiliated with WordNet and related NLP initiatives. It has undergone several releases and, in addition to the original edition, later versions added more text and revised annotations. It is distributed for research use by linguistic data repositories and is widely cited in NLP literature.