New Computational Tool Classifies Manner and Result Verbs at Scale

A new scalable computational approach has been created by researchers to pinpoint manner and result verbs within sentence contexts, filling a void in developmental language studies. By employing linguistically informed prompts, they produced sentence-level annotations using large language models on data from MASC and InterCorp, broadening the coverage from earlier annotated segments of VerbNet to 436 classes. A RoBERTa-based classifier was trained with these annotations and tested against three gold-standard datasets, which included previously annotated items and a new set annotated by experts. The model demonstrated encouraging performance with an average accuracy. This innovative tool facilitates extensive measurement of a distinction that has been challenging to examine due to a scarcity of annotated resources.