Defining Antonymy: A Corpus-based Study of Opposites Found by Lexico-syntactic Patterns

Using small sets of adjectival seed antonym pairs, we automatically find patterns where these pairs co-occur in a large corpus of Dutch, and then use these patterns to extract new antonym pairs. Evaluation of extracted pairs by five human judges showed that automatic scores correlate with human evaluation and that pattern based methods can be used to extract new antonym pairs. The majority of extracted pairs were noun-noun pairs, contrary to expectations based on previous research. Additionally, the method identifies a subgroup of co-hyponyms that frequently function antonymously, and together with more traditional antonyms makes up a wider class of incompatibles, suggesting that antonymy is a diverse relation that includes pairs of different types and categories that are not captured by any single linguistic theory. Comparison with Dutch WordNet and an online Dutch dictionary shows that only a handful of extracted pairs are currently listed in these existing resources, emphasizing the usefulness of the project.