Nanzheng Xie, Ryuichi Ikeda and Suk Min Hwang, University of California, USA

Traditional measures of participation in group discussions rely on surface-level indicators such as speaking frequency, obscuring the qualitative value of individual utterances. This paper proposes an NLP-based framework for measuring meaningful contribution that emphasizes how utterances advance collective problem solving rather than how often participants speak. We introduce the Meaningful Contribution Score (MCS), an utterance-level, five-dimensional measure spanning semantic and interactional constructs (relevance, novelty, enablement, affect, and decision proximity). Focusing on construct validity, we validate MCS on the GAP corpus using a new human-annotated subset and a three-way comparison between human ratings, heuristic MCS components, and LLM-based judgments. Results show strong alignment for semantic relevance, partial alignment for novelty and enablement, and persistent unreliability for affect, highlighting where lightweight heuristics suffice and where context-aware models are needed for interpretable assessment of contribution.

Meaningful Contribution, Group Discussion,Construct Validity,Discourse Modeling,Sentence Embeddings