Generating Benchmarks for Factuality Evaluation of Language Models

Dor Muhlgay*, Ori Ram, Inbal Magar, Yoav Levine, Nir Ratner, Yonatan Belinkov, Omri Abend, Kevin Leyton-Brown, Amnon Shashua, Yoav Shoham

*Corresponding author for this work

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

Abstract

Before deploying a language model (LM) within a given domain, it is important to measure its tendency to generate factually incorrect information in that domain. Existing methods for factuality evaluation of LLM generation focus on facts sampled from the LM itself, and thus do not control the set of evaluated facts and might under-represent domain specific or rare facts. We propose FACTOR: Factual Assessment via Corpus TransfORmation, a scalable approach for evaluating LM factuality. FACTOR automatically transforms a factual corpus of interest into a benchmark evaluating an LM's propensity to generate true facts from the corpus vs. similar but incorrect statements. We use our framework to create three benchmarks: Wiki-FACTOR, News-FACTOR and Expert-FACTOR. We show that: (i) our benchmark scores increase with model size and improve when the LM is augmented with retrieval; (ii) benchmark score and perplexity do not always agree on model ranking; (iii) when perplexity and benchmark score disagree, the latter better reflects factuality in open-ended generation, as measured by human annotators. We make our data and code publicly available.

Original languageAmerican English
Title of host publicationEACL 2024 - 18th Conference of the European Chapter of the Association for Computational Linguistics, Proceedings of the Conference
EditorsYvette Graham, Matthew Purver, Matthew Purver
PublisherAssociation for Computational Linguistics (ACL)
Pages49-66
Number of pages18
ISBN (Electronic)9798891760882
StatePublished - 2024
Externally publishedYes
Event18th Conference of the European Chapter of the Association for Computational Linguistics, EACL 2024 - St. Julian�s, Malta
Duration: 17 Mar 202422 Mar 2024

Publication series

NameEACL 2024 - 18th Conference of the European Chapter of the Association for Computational Linguistics, Proceedings of the Conference
Volume1

Conference

Conference18th Conference of the European Chapter of the Association for Computational Linguistics, EACL 2024
Country/TerritoryMalta
CitySt. Julian�s
Period17/03/2422/03/24

Bibliographical note

Publisher Copyright:
© 2024 Association for Computational Linguistics.

Fingerprint

Dive into the research topics of 'Generating Benchmarks for Factuality Evaluation of Language Models'. Together they form a unique fingerprint.

Cite this