aiqbench analysis
This research advances mechanistic interpretability by introducing DSRA, a methodology that maps internal transformer activations to abstract definitional structures rather than just factual retrieval tasks. By applying causal tracing to reverse dictionary tasks in both English and Spanish, the study reveals how MLP and MHA layers systematically encode compositional semantic roles like genus and differentia across model depths. These findings provide crucial insights into how large language models construct meaning from linguistic inputs, moving beyond token-level localization to understand conceptual interpretation.
Bridging Linguistic Structure and Mechanistic Interpretability for Conceptual Interpretation in Language Models
arXiv:2408.11827v2 Announce Type: replace Abstract: Understanding how language models compose meaning from linguistic input remains a central problem in interpretability research. Mechanistic studies have attributed functional roles to core transformer components; however, these findings derive largely from factual retrieval settings. Whether the same mechanisms support \textit{conceptual interpretation}, the compositional mapping from definitional expressions to abstract meaning, remains insufficiently characterised. We introduce \textit{DSRA} (Definitional Semantic Role Analysis), a methodology that applies causal tracing within the reverse dictionary task and augments restoration traces with definitional semantic roles (DSRs) grounded in Argument Structure Theory. This linguistic overlay identifies which compositional functions (e.g., genus, differentia quality) are associated with high-recovery states, extending activation patching beyond token-level localisation. Applied to GPT-J-6B (English) and BERTIN GPT-J-6B (Spanish), the results show that MLP layers associate content-bearing tokens with high-specificity DSR categories in early layers, MHA layers distribute integration across middle-to-upper layers with concentration at the final token, and hidden states aggregate information in upper layers. Alignment between restored states and DSR categories indicates systematic correspondence between internal activations and definitional structure, with consistent localisation patterns across both languages.
Abstract:Understanding how language models compose meaning from linguistic input remains a central problem in interpretability research. Mechanistic studies have attributed functional roles to core transformer components; however, these findings derive largely from factual retrieval settings. Whether the same mechanisms support \textit{conceptual interpretation}, the compositional mapping from definitional expressions to abstract meaning, remains insufficiently characterised. We introduce \textit{DSRA} (Definitional Semantic Role Analysis), a methodology that applies causal tracing within the reverse dictionary task and augments restoration traces with definitional semantic roles (DSRs) grounded in Argument Structure Theory. This linguistic overlay identifies which compositional functions (e.g., genus, differentia quality) are associated with high-recovery states, extending activation patching beyond token-level localisation. Applied to GPT-J-6B (English) and BERTIN GPT-J-6B (Spanish), the results show that MLP layers associate content-bearing tokens with high-specificity DSR categories in early layers, MHA layers distribute integration across middle-to-upper layers with concentration at the final token, and hidden states aggregate information in upper layers. Alignment between restored states and DSR categories indicates systematic correspondence between internal activations and definitional structure, with consistent localisation patterns across both languages.
Computation and Language (cs.CL)
arXiv:2408.11827 [cs.CL]
(or arXiv:2408.11827v2 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2408.11827 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 269 words
Related coverage
The Pragmatic Persona: Discovering LLM Persona through Bridging Inference
arXiv · Jun 12, 2026
Models Can't Remember Their Training. Neither Can You.
LessWrong · Jul 19, 2026
Semantic Abstraction for Natural Language Inference: a Methodological Framework for Discovering and Compensating Semantic Knowledge and Reasoning Gaps in Large Language Models
arXiv · Sep 23, 2026
The Ignition Index: Measuring Global Workspace Dynamics in Language Models
arXiv · Aug 7, 2026