RAVEL: Evaluating Interpretability Methods on Disentangling Language Model Representations

Jing Huang, Zhengxuan Wu, Christopher Potts, Mor Geva, Atticus Geiger

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

Individual neurons participate in the representation of multiple high-level concepts. To what extent can different interpretability methods successfully disentangle these roles? To help address this question, we introduce RAVEL (Resolving Attribute-Value Entanglements in Language Models), a dataset that enables tightly controlled, quantitative comparisons between a variety of existing interpretability methods. We use the resulting conceptual framework to define the new method of Multi-task Distributed Alignment Search (MDAS), which allows us to find distributed representations satisfying multiple causal criteria. With Llama2-7B as the target language model, MDAS achieves state-of-the-art results on RAVEL, demonstrating the importance of going beyond neuron-level analyses to identify features distributed across activations. We release our benchmark at https://github.com/explanare/ravel.

Original languageEnglish
Title of host publicationLong Papers
EditorsLun-Wei Ku, Andre F. T. Martins, Vivek Srikumar
PublisherAssociation for Computational Linguistics (ACL)
Pages8669-8687
Number of pages19
ISBN (Electronic)9798891760943
StatePublished - 2024
Event62nd Annual Meeting of the Association for Computational Linguistics, ACL 2024 - Bangkok, Thailand
Duration: 11 Aug 202416 Aug 2024

Publication series

NameProceedings of the Annual Meeting of the Association for Computational Linguistics
Volume1

Conference

Conference62nd Annual Meeting of the Association for Computational Linguistics, ACL 2024
Country/TerritoryThailand
CityBangkok
Period11/08/2416/08/24

All Science Journal Classification (ASJC) codes

  • Computer Science Applications
  • Linguistics and Language
  • Language and Linguistics

Cite this