aiqbench analysis
This paper introduces StructLens, a novel interpretability framework that analyzes the internal organization of language model representations using maximum spanning trees derived from residual streams. The study reveals how token relationships are structured across layers during pre-training, offering new insights into the mechanistic formation of semantic units within large models.
StructLens: A Structural Lens for Language Models via Maximum Spanning Trees
arXiv:2603.03328v3 Announce Type: replace-cross Abstract: Language exhibits inherent structures, a property that explains both language acquisition and language change. Given this characteristic, we expect language models to manifest their own internal structures as well. While interpretability research has investigated how models compute representations mechanistically through attention patterns and Sparse AutoEncoders, the organization of the resulting representations is overlooked. To address this gap, we introduce StructLens, a framework to analyze representations through a holistic structural view. StructLens constructs maximum spanning trees based on the semantic representations in residual streams, inspired by tree representation in dependency parsing, and provides summaries of token relationships in representation space. We analyze how contiguous tokens are also nearby in representation space and find that middle layers show the strongest local-span organization. Moreover, analysis of pre-training checkpoints reveals that smaller local units become detectable earlier in pre-training, and larger units later. Our findings demonstrate that StructLens provides insights into how models organize token representations across layers and training. Our code is available at https://github.com/naist-nlp/structlens.
Abstract:Language exhibits inherent structures, a property that explains both language acquisition and language change. Given this characteristic, we expect language models to manifest their own internal structures as well. While interpretability research has investigated how models compute representations mechanistically through attention patterns and Sparse AutoEncoders, the organization of the resulting representations is overlooked. To address this gap, we introduce StructLens, a framework to analyze representations through a holistic structural view. StructLens constructs maximum spanning trees based on the semantic representations in residual streams, inspired by tree representation in dependency parsing, and provides summaries of token relationships in representation space. We analyze how contiguous tokens are also nearby in representation space and find that middle layers show the strongest local-span organization. Moreover, analysis of pre-training checkpoints reveals that smaller local units become detectable earlier in pre-training, and larger units later. Our findings demonstrate that StructLens provides insights into how models organize token representations across layers and training. Our code is available at this https URL.
Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
arXiv:2603.03328 [cs.CL]
(or arXiv:2603.03328v3 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2603.03328 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 · 260 words