Differentiating synonyms and adjective subclasses by syntactic profiling

Abstract

Twenty adjectives belonging to five intuitively recognizable semantic families, age (old, young), size (large, small, big), color (black, white, red, yellow, blue), modality (possible, impossible, necessary, likely, sure) and emotion (happy, sad, glad, sorry) were inventoried in a 5-million word corpus tagged for part of speech (POS) and lemma. Regular expressions were used to target specific contextual parameters including syntactic function (adnominal or predicative), determiners, grammatical subject and intensifiers, while manual estimates were obtained from random samples of 500 occurrences per adjective. Benchmark data, for purposes of comparison, were first collected for the category as a whole using POS tags, from which it was observed that four major determiners and six copula verbs were present in 80-90% of all occurrences. Adjective families were found to share common, distinctive syntactic profiles, with similar predispositions to adnominal or predicative function and similar affinities with specific determiners or intensifiers. Differences were observed as well between certain quasi-synonyms such as glad/happy with respect to one or more of these criteria.