Unraveling Verb Root Semantics: A Scalable Approach to Manner and Result Categorization

ACL ARR 2025 February Submission4410 Authors

Abstract: In this paper, we address the computational identification and categorization of verbs into result, and manner verbs—a distinction that has been shown to influence child vocabulary acquisition and later difficulties with language learning such as Developmental Language Disorder (DLD). Within this framework, manner verbs encode the dynamic “how” of an action, and result verbs, denote a change in outcome. Prior work has been limited to a narrow subset of VerbNet, and relied mainly on human linguistic reasoning without scalable computational methods. In contrast, we leverage Large Language Models (LLMs) as an expert annotator to generate synthetic annotations on 436 out of 487 VerbNet classes over sentences drawn from MASC and InterCorp dataset. These annotations serve as training data for a RoBERTa-based classifier, which achieves an accuracy of 89.6% overall on gold annotated datasets. To the best of our knowledge, this work presents the first large-scale computational approach to result and manner verb classification.

Paper Type: Long

Research Area: Semantics: Lexical and Sentence-Level

Research Area Keywords: natural language inference, word embeddings, linguistic theories

Contribution Types: Model analysis & interpretability, NLP engineering experiment, Publicly available software and/or pre-trained models, Data analysis

Languages Studied: English

Submission Number: 4410