Salvato in:
Dettagli Bibliografici
Autori principali: Xia, Eric, Kalita, Jugal
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:https://arxiv.org/abs/2507.14640
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866916852132741120
author Xia, Eric
Kalita, Jugal
author_facet Xia, Eric
Kalita, Jugal
contents A two-part affine approximation has been found to be a good approximation for transformer computations over certain subject object relations. Adapting the Bigger Analogy Test Set, we show that the linear transformation Ws, where s is a middle layer representation of a subject token and W is derived from model derivatives, is also able to accurately reproduce final object states for many relations. This linear technique is able to achieve 90% faithfulness on morphological relations, and we show similar findings multi-lingually and across models. Our findings indicate that some conceptual relationships in language models, such as morphology, are readily interpretable from latent space, and are sparsely encoded by cross-layer linear transformations.
format Preprint
id arxiv_https___arxiv_org_abs_2507_14640
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Linear Relational Decoding of Morphology in Language Models
Xia, Eric
Kalita, Jugal
Computation and Language
A two-part affine approximation has been found to be a good approximation for transformer computations over certain subject object relations. Adapting the Bigger Analogy Test Set, we show that the linear transformation Ws, where s is a middle layer representation of a subject token and W is derived from model derivatives, is also able to accurately reproduce final object states for many relations. This linear technique is able to achieve 90% faithfulness on morphological relations, and we show similar findings multi-lingually and across models. Our findings indicate that some conceptual relationships in language models, such as morphology, are readily interpretable from latent space, and are sparsely encoded by cross-layer linear transformations.
title Linear Relational Decoding of Morphology in Language Models
topic Computation and Language
url https://arxiv.org/abs/2507.14640