Evaluating Adjective-Noun Compositionality in LLMs: Functional vs Representational Perspectives
Computation and Language
2026-03-17 v2 Artificial Intelligence
Abstract
Compositionality is considered central to language abilities. As performant language systems, how do large language models (LLMs) do on compositional tasks? We evaluate adjective-noun compositionality in LLMs using two complementary setups: prompt-based functional assessment and a representational analysis of internal model states. Our results reveal a striking divergence between task performance and internal states. While LLMs reliably develop compositional representations, they fail to translate consistently into functional task success across model variants. Consequently, we highlight the importance of contrastive evaluation for obtaining a more complete understanding of model capabilities.
Cite
@article{arxiv.2603.09994,
title = {Evaluating Adjective-Noun Compositionality in LLMs: Functional vs Representational Perspectives},
author = {Ruchira Dhar and Qiwei Peng and Anders Søgaard},
journal= {arXiv preprint arXiv:2603.09994},
year = {2026}
}
Comments
Under Review