@inproceedings{hagstrom-etal-2025-language,
title = "Language Model Re-rankers are Fooled by Lexical Similarities",
author = {Hagstr{\"o}m, Lovisa and
Nie, Ercong and
Halifa, Ruben and
Schmid, Helmut and
Johansson, Richard and
Junge, Alexander},
editor = "Akhtar, Mubashara and
Aly, Rami and
Christodoulopoulos, Christos and
Cocarascu, Oana and
Guo, Zhijiang and
Mittal, Arpit and
Schlichtkrull, Michael and
Thorne, James and
Vlachos, Andreas",
booktitle = "Proceedings of the Eighth Fact Extraction and VERification Workshop (FEVER)",
month = jul,
year = "2025",
address = "Vienna, Austria",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2025.fever-1.2/",
doi = "10.18653/v1/2025.fever-1.2",
pages = "18--33",
ISBN = "978-1-959429-53-1",
abstract = "Language model (LM) re-rankers are used to refine retrieval results for retrieval-augmented generation (RAG). They are more expensive than lexical matching methods like BM25 but assumed to better process semantic information and the relations between the query and the retrieved answers. To understand whether LM re-rankers always live up to this assumption, we evaluate 6 different LM re-rankers on the NQ, LitQA2 and DRUID datasets. Our results show that LM re-rankers struggle to outperform a simple BM25 baseline on DRUID. Leveraging a novel separation metric based on BM25 scores, we explain and identify re-ranker errors stemming from lexical dissimilarities. We also investigate different methods to improve LM re-ranker performance and find these methods mainly useful for NQ. Taken together, our work identifies and explains weaknesses of LM re-rankers and points to the need for more adversarial and realistic datasets for their evaluation."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="hagstrom-etal-2025-language">
<titleInfo>
<title>Language Model Re-rankers are Fooled by Lexical Similarities</title>
</titleInfo>
<name type="personal">
<namePart type="given">Lovisa</namePart>
<namePart type="family">Hagström</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ercong</namePart>
<namePart type="family">Nie</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ruben</namePart>
<namePart type="family">Halifa</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Helmut</namePart>
<namePart type="family">Schmid</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Richard</namePart>
<namePart type="family">Johansson</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Alexander</namePart>
<namePart type="family">Junge</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2025-07</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the Eighth Fact Extraction and VERification Workshop (FEVER)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Mubashara</namePart>
<namePart type="family">Akhtar</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Rami</namePart>
<namePart type="family">Aly</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Christos</namePart>
<namePart type="family">Christodoulopoulos</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Oana</namePart>
<namePart type="family">Cocarascu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhijiang</namePart>
<namePart type="family">Guo</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Arpit</namePart>
<namePart type="family">Mittal</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Michael</namePart>
<namePart type="family">Schlichtkrull</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">James</namePart>
<namePart type="family">Thorne</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Andreas</namePart>
<namePart type="family">Vlachos</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Vienna, Austria</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">978-1-959429-53-1</identifier>
</relatedItem>
<abstract>Language model (LM) re-rankers are used to refine retrieval results for retrieval-augmented generation (RAG). They are more expensive than lexical matching methods like BM25 but assumed to better process semantic information and the relations between the query and the retrieved answers. To understand whether LM re-rankers always live up to this assumption, we evaluate 6 different LM re-rankers on the NQ, LitQA2 and DRUID datasets. Our results show that LM re-rankers struggle to outperform a simple BM25 baseline on DRUID. Leveraging a novel separation metric based on BM25 scores, we explain and identify re-ranker errors stemming from lexical dissimilarities. We also investigate different methods to improve LM re-ranker performance and find these methods mainly useful for NQ. Taken together, our work identifies and explains weaknesses of LM re-rankers and points to the need for more adversarial and realistic datasets for their evaluation.</abstract>
<identifier type="citekey">hagstrom-etal-2025-language</identifier>
<identifier type="doi">10.18653/v1/2025.fever-1.2</identifier>
<location>
<url>https://aclanthology.org/2025.fever-1.2/</url>
</location>
<part>
<date>2025-07</date>
<extent unit="page">
<start>18</start>
<end>33</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Language Model Re-rankers are Fooled by Lexical Similarities
%A Hagström, Lovisa
%A Nie, Ercong
%A Halifa, Ruben
%A Schmid, Helmut
%A Johansson, Richard
%A Junge, Alexander
%Y Akhtar, Mubashara
%Y Aly, Rami
%Y Christodoulopoulos, Christos
%Y Cocarascu, Oana
%Y Guo, Zhijiang
%Y Mittal, Arpit
%Y Schlichtkrull, Michael
%Y Thorne, James
%Y Vlachos, Andreas
%S Proceedings of the Eighth Fact Extraction and VERification Workshop (FEVER)
%D 2025
%8 July
%I Association for Computational Linguistics
%C Vienna, Austria
%@ 978-1-959429-53-1
%F hagstrom-etal-2025-language
%X Language model (LM) re-rankers are used to refine retrieval results for retrieval-augmented generation (RAG). They are more expensive than lexical matching methods like BM25 but assumed to better process semantic information and the relations between the query and the retrieved answers. To understand whether LM re-rankers always live up to this assumption, we evaluate 6 different LM re-rankers on the NQ, LitQA2 and DRUID datasets. Our results show that LM re-rankers struggle to outperform a simple BM25 baseline on DRUID. Leveraging a novel separation metric based on BM25 scores, we explain and identify re-ranker errors stemming from lexical dissimilarities. We also investigate different methods to improve LM re-ranker performance and find these methods mainly useful for NQ. Taken together, our work identifies and explains weaknesses of LM re-rankers and points to the need for more adversarial and realistic datasets for their evaluation.
%R 10.18653/v1/2025.fever-1.2
%U https://aclanthology.org/2025.fever-1.2/
%U https://doi.org/10.18653/v1/2025.fever-1.2
%P 18-33
Markdown (Informal)
[Language Model Re-rankers are Fooled by Lexical Similarities](https://aclanthology.org/2025.fever-1.2/) (Hagström et al., FEVER 2025)
ACL
- Lovisa Hagström, Ercong Nie, Ruben Halifa, Helmut Schmid, Richard Johansson, and Alexander Junge. 2025. Language Model Re-rankers are Fooled by Lexical Similarities. In Proceedings of the Eighth Fact Extraction and VERification Workshop (FEVER), pages 18–33, Vienna, Austria. Association for Computational Linguistics.