@inproceedings{satouf-etal-2026-storm,
title = "{STORM}: Stepwise Token Optimization with Reward-Guided Beam Search",
author = "Satouf, Arthur and
D{'}Erasmo, Giulio and
Zong, Yuxuan and
Boubacar, Habiboulaye Amadou and
Piantanida, Pablo and
Piwowarski, Benjamin",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-main.2149/",
pages = "48856--48871",
ISBN = "979-8-89176-459-0",
abstract = "Modern retrieval increasingly relies on dense and learned-sparse neural models that are effective but require encoding the entire corpus into a specialized index, which must be rebuilt whenever the model changes. Lexical retrievers like BM25 stay efficient, transparent, and run on a standard inverted index that need not change as models evolve, but suffer from vocabulary mismatch. LLM query rewriting can help, yet prompted rewriters emit well-formed but retrieval-ineffective{---}or harmful{---}terms, and training against a retrieval reward gives only delayed, sequence-level supervision that obscures which terms helped. We introduce STORM (Stepwise Token Optimization with Reward-guided beaM search), a self-supervised framework with retrieval feedback for lexical query expansion. STORM trains the rewriter through generation guided by retrieval metrics: at each step, a beam of candidate expansions is scored against the BM25 index and low-reward continuations are pruned, turning the retrieval reward into a token-level signal that concentrates exploration on retrieval-effective vocabulary. Across TREC DL and BEIR, STORM lets 0.6B{--}8B backbones match or surpass competitive LLM rewriters and far larger proprietary models while retrieving as fast as plain BM25, and transfers zero-shot to 18 languages (MIRACL), beating dedicated multilingual dense retrievers. STORM is thus a competitive, infrastructure-light alternative to dense neural retrieval."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="satouf-etal-2026-storm">
<titleInfo>
<title>STORM: Stepwise Token Optimization with Reward-Guided Beam Search</title>
</titleInfo>
<name type="personal">
<namePart type="given">Arthur</namePart>
<namePart type="family">Satouf</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Giulio</namePart>
<namePart type="family">D’Erasmo</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yuxuan</namePart>
<namePart type="family">Zong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Habiboulaye</namePart>
<namePart type="given">Amadou</namePart>
<namePart type="family">Boubacar</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Pablo</namePart>
<namePart type="family">Piantanida</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Benjamin</namePart>
<namePart type="family">Piwowarski</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hung-Yi</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sunipa</namePart>
<namePart type="family">Dev</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jessy</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Desmond</namePart>
<namePart type="family">Elliott</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-459-0</identifier>
</relatedItem>
<abstract>Modern retrieval increasingly relies on dense and learned-sparse neural models that are effective but require encoding the entire corpus into a specialized index, which must be rebuilt whenever the model changes. Lexical retrievers like BM25 stay efficient, transparent, and run on a standard inverted index that need not change as models evolve, but suffer from vocabulary mismatch. LLM query rewriting can help, yet prompted rewriters emit well-formed but retrieval-ineffective—or harmful—terms, and training against a retrieval reward gives only delayed, sequence-level supervision that obscures which terms helped. We introduce STORM (Stepwise Token Optimization with Reward-guided beaM search), a self-supervised framework with retrieval feedback for lexical query expansion. STORM trains the rewriter through generation guided by retrieval metrics: at each step, a beam of candidate expansions is scored against the BM25 index and low-reward continuations are pruned, turning the retrieval reward into a token-level signal that concentrates exploration on retrieval-effective vocabulary. Across TREC DL and BEIR, STORM lets 0.6B–8B backbones match or surpass competitive LLM rewriters and far larger proprietary models while retrieving as fast as plain BM25, and transfers zero-shot to 18 languages (MIRACL), beating dedicated multilingual dense retrievers. STORM is thus a competitive, infrastructure-light alternative to dense neural retrieval.</abstract>
<identifier type="citekey">satouf-etal-2026-storm</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.2149/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>48856</start>
<end>48871</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T STORM: Stepwise Token Optimization with Reward-Guided Beam Search
%A Satouf, Arthur
%A D’Erasmo, Giulio
%A Zong, Yuxuan
%A Boubacar, Habiboulaye Amadou
%A Piantanida, Pablo
%A Piwowarski, Benjamin
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-459-0
%F satouf-etal-2026-storm
%X Modern retrieval increasingly relies on dense and learned-sparse neural models that are effective but require encoding the entire corpus into a specialized index, which must be rebuilt whenever the model changes. Lexical retrievers like BM25 stay efficient, transparent, and run on a standard inverted index that need not change as models evolve, but suffer from vocabulary mismatch. LLM query rewriting can help, yet prompted rewriters emit well-formed but retrieval-ineffective—or harmful—terms, and training against a retrieval reward gives only delayed, sequence-level supervision that obscures which terms helped. We introduce STORM (Stepwise Token Optimization with Reward-guided beaM search), a self-supervised framework with retrieval feedback for lexical query expansion. STORM trains the rewriter through generation guided by retrieval metrics: at each step, a beam of candidate expansions is scored against the BM25 index and low-reward continuations are pruned, turning the retrieval reward into a token-level signal that concentrates exploration on retrieval-effective vocabulary. Across TREC DL and BEIR, STORM lets 0.6B–8B backbones match or surpass competitive LLM rewriters and far larger proprietary models while retrieving as fast as plain BM25, and transfers zero-shot to 18 languages (MIRACL), beating dedicated multilingual dense retrievers. STORM is thus a competitive, infrastructure-light alternative to dense neural retrieval.
%U https://aclanthology.org/2026.emnlp-main.2149/
%P 48856-48871
Markdown (Informal)
[STORM: Stepwise Token Optimization with Reward-Guided Beam Search](https://aclanthology.org/2026.emnlp-main.2149/) (Satouf et al., EMNLP 2026)
ACL
- Arthur Satouf, Giulio D’Erasmo, Yuxuan Zong, Habiboulaye Amadou Boubacar, Pablo Piantanida, and Benjamin Piwowarski. 2026. STORM: Stepwise Token Optimization with Reward-Guided Beam Search. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 48856–48871, Budapest, Hungary. Association for Computational Linguistics.