@inproceedings{zhang-etal-2026-adaptive-reasoning,
title = "Adaptive Reasoning for Efficient Generative Retrieval",
author = "Zhang, Yingchen and
Zhang, Ruqing and
Guo, Jiafeng and
Peng, Wenjun and
Li, Sen and
Lv, Fuyu and
Cheng, Xueqi",
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.2224/",
pages = "50634--50647",
ISBN = "979-8-89176-459-0",
abstract = "Generative retrieval (GR) retrieves documents by directly generating document identifiers (DocIDs) from queries using a sequence-to-sequence model. Recent reasoning-enhanced GR methods improve retrieval effectiveness by generating intermediate reasoning before DocID decoding. However, they typically apply reasoning uniformly to all queries, even though many queries can already be handled effectively through direct DocID decoding. This introduces unnecessary token generation and retrieval latency when reasoning provides little additional benefit. In this work, we propose Generative Retrieval with Adaptive Reasoning (GEAR), a retrieval framework that dynamically allocates reasoning computation based on query-specific needs. Instead of treating reasoning as a fixed pre-retrieval procedure, GEAR learns when to reason, what retrieval-oriented context to generate, and when to transition from free-form generation to constrained DocID decoding. This adaptive behavior is implemented through a learnable retrieval transition policy optimized jointly for retrieval quality and reasoning cost. Experiments on NQ, MS MARCO, and TREC DL 2022 show that GEAR maintains retrieval effectiveness comparable to strong reasoning-enhanced GR methods while substantially reducing reasoning overhead, triggering non-trivial reasoning for only 37.1{\%} of queries and reducing both generated tokens and retrieval latency."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="zhang-etal-2026-adaptive-reasoning">
<titleInfo>
<title>Adaptive Reasoning for Efficient Generative Retrieval</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yingchen</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ruqing</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jiafeng</namePart>
<namePart type="family">Guo</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Wenjun</namePart>
<namePart type="family">Peng</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sen</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Fuyu</namePart>
<namePart type="family">Lv</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xueqi</namePart>
<namePart type="family">Cheng</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>Generative retrieval (GR) retrieves documents by directly generating document identifiers (DocIDs) from queries using a sequence-to-sequence model. Recent reasoning-enhanced GR methods improve retrieval effectiveness by generating intermediate reasoning before DocID decoding. However, they typically apply reasoning uniformly to all queries, even though many queries can already be handled effectively through direct DocID decoding. This introduces unnecessary token generation and retrieval latency when reasoning provides little additional benefit. In this work, we propose Generative Retrieval with Adaptive Reasoning (GEAR), a retrieval framework that dynamically allocates reasoning computation based on query-specific needs. Instead of treating reasoning as a fixed pre-retrieval procedure, GEAR learns when to reason, what retrieval-oriented context to generate, and when to transition from free-form generation to constrained DocID decoding. This adaptive behavior is implemented through a learnable retrieval transition policy optimized jointly for retrieval quality and reasoning cost. Experiments on NQ, MS MARCO, and TREC DL 2022 show that GEAR maintains retrieval effectiveness comparable to strong reasoning-enhanced GR methods while substantially reducing reasoning overhead, triggering non-trivial reasoning for only 37.1% of queries and reducing both generated tokens and retrieval latency.</abstract>
<identifier type="citekey">zhang-etal-2026-adaptive-reasoning</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.2224/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>50634</start>
<end>50647</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Adaptive Reasoning for Efficient Generative Retrieval
%A Zhang, Yingchen
%A Zhang, Ruqing
%A Guo, Jiafeng
%A Peng, Wenjun
%A Li, Sen
%A Lv, Fuyu
%A Cheng, Xueqi
%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 zhang-etal-2026-adaptive-reasoning
%X Generative retrieval (GR) retrieves documents by directly generating document identifiers (DocIDs) from queries using a sequence-to-sequence model. Recent reasoning-enhanced GR methods improve retrieval effectiveness by generating intermediate reasoning before DocID decoding. However, they typically apply reasoning uniformly to all queries, even though many queries can already be handled effectively through direct DocID decoding. This introduces unnecessary token generation and retrieval latency when reasoning provides little additional benefit. In this work, we propose Generative Retrieval with Adaptive Reasoning (GEAR), a retrieval framework that dynamically allocates reasoning computation based on query-specific needs. Instead of treating reasoning as a fixed pre-retrieval procedure, GEAR learns when to reason, what retrieval-oriented context to generate, and when to transition from free-form generation to constrained DocID decoding. This adaptive behavior is implemented through a learnable retrieval transition policy optimized jointly for retrieval quality and reasoning cost. Experiments on NQ, MS MARCO, and TREC DL 2022 show that GEAR maintains retrieval effectiveness comparable to strong reasoning-enhanced GR methods while substantially reducing reasoning overhead, triggering non-trivial reasoning for only 37.1% of queries and reducing both generated tokens and retrieval latency.
%U https://aclanthology.org/2026.emnlp-main.2224/
%P 50634-50647
Markdown (Informal)
[Adaptive Reasoning for Efficient Generative Retrieval](https://aclanthology.org/2026.emnlp-main.2224/) (Zhang et al., EMNLP 2026)
ACL
- Yingchen Zhang, Ruqing Zhang, Jiafeng Guo, Wenjun Peng, Sen Li, Fuyu Lv, and Xueqi Cheng. 2026. Adaptive Reasoning for Efficient Generative Retrieval. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 50634–50647, Budapest, Hungary. Association for Computational Linguistics.