Saved in:
Bibliographic Details
Main Authors: Hernandez, Evan, Sharma, Arnab Sen, Haklay, Tal, Meng, Kevin, Wattenberg, Martin, Andreas, Jacob, Belinkov, Yonatan, Bau, David
Format: Preprint
Published: 2023
Subjects:
Online Access:https://arxiv.org/abs/2308.09124
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916127370641408
author Hernandez, Evan
Sharma, Arnab Sen
Haklay, Tal
Meng, Kevin
Wattenberg, Martin
Andreas, Jacob
Belinkov, Yonatan
Bau, David
author_facet Hernandez, Evan
Sharma, Arnab Sen
Haklay, Tal
Meng, Kevin
Wattenberg, Martin
Andreas, Jacob
Belinkov, Yonatan
Bau, David
contents Much of the knowledge encoded in transformer language models (LMs) may be expressed in terms of relations: relations between words and their synonyms, entities and their attributes, etc. We show that, for a subset of relations, this computation is well-approximated by a single linear transformation on the subject representation. Linear relation representations may be obtained by constructing a first-order approximation to the LM from a single prompt, and they exist for a variety of factual, commonsense, and linguistic relations. However, we also identify many cases in which LM predictions capture relational knowledge accurately, but this knowledge is not linearly encoded in their representations. Our results thus reveal a simple, interpretable, but heterogeneously deployed knowledge representation strategy in transformer LMs.
format Preprint
id arxiv_https___arxiv_org_abs_2308_09124
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Linearity of Relation Decoding in Transformer Language Models
Hernandez, Evan
Sharma, Arnab Sen
Haklay, Tal
Meng, Kevin
Wattenberg, Martin
Andreas, Jacob
Belinkov, Yonatan
Bau, David
Computation and Language
Much of the knowledge encoded in transformer language models (LMs) may be expressed in terms of relations: relations between words and their synonyms, entities and their attributes, etc. We show that, for a subset of relations, this computation is well-approximated by a single linear transformation on the subject representation. Linear relation representations may be obtained by constructing a first-order approximation to the LM from a single prompt, and they exist for a variety of factual, commonsense, and linguistic relations. However, we also identify many cases in which LM predictions capture relational knowledge accurately, but this knowledge is not linearly encoded in their representations. Our results thus reveal a simple, interpretable, but heterogeneously deployed knowledge representation strategy in transformer LMs.
title Linearity of Relation Decoding in Transformer Language Models
topic Computation and Language
url https://arxiv.org/abs/2308.09124