Modelling object omission availability in English transitive verbs with corpus and psycholinguistic data

Kurzfassung/Abstract

This data-driven study investigates the licensing factors of object omission in English transitive verbs (e.g., I’m eating Ø vs. * I’m devouring Ø). Using corpus data from the British component of the International Corpus of English and psycholinguistic measurements from the South Carolina Psycholinguistic Metabase (SCOPE), generalised additive models were trained on distributional and semantic features of 972 verbs to examine linear and non-linear effects on omission availability. In a complementary analysis, linear discriminant analysis was applied to fastText word embeddings to assess the predictive value of distributional co-occurrence information. Discriminant scores emerged as the strongest predictor, followed by frequency and dispersion, while purely semantic predictors contributed comparatively little once distributional measures were controlled for. Object-omitting verbs cluster towards the stative and communicative end of the verb space. The results challenge earlier monofactorial accounts and support usage-based explanations of argument realisation, though the overlap between distributional and semantic measures suggests that future research should address potential confounds and consider semantic frames as a key missing predictor.