@inproceedings{gao-etal-2026-hybridthink,
title = "{H}ybrid{T}hink: Learned Interleaving of Continuous and Discrete Reasoning in Large Language Models",
author = "Gao, Jing and
Jin, Weiqiang and
Zhao, Guoshuai and
Ye, Peigen and
Qian, Xueming and
Du, Hongyang",
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.91/",
pages = "2136--2164",
ISBN = "979-8-89176-459-0",
abstract = "Reasoning in Large Language Models (LLMs) is primarily conducted in a discrete token space. Although recent approaches employing continuous concept tokens facilitate abstract reasoning, they introduce significant stability challenges. The accumulation of Out-of-Distribution (OOD) states frequently causes manifold deviation, while the absence of clear boundaries leads to indeterminate termination, often referred to as the anchoring problem. To address these limitations, we propose HybridThink. We formulate mode selection in LLM reasoning as a sequential decision process. A lightweight gating policy is trained with RL to decide at each step whether to propagate a continuous concept token or generate a discrete text token to anchor the semantic trajectory.This learned interleaving limits manifold deviation and resolves the fragility of static rules. Furthermore, we ensure stable convergence through a training strategy that incorporates Error-Centric Experience Replay (ECER) and Inductive Exploration Shaping (IES), prioritizing learning from failure modes, specifically drifting and premature termination. Empirical evaluations on mathematics and coding benchmarks demonstrate that HybridThink outperforms standard CoT and Soft Thinking baselines, achieving superior accuracy with reduced token consumption."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="gao-etal-2026-hybridthink">
<titleInfo>
<title>HybridThink: Learned Interleaving of Continuous and Discrete Reasoning in Large Language Models</title>
</titleInfo>
<name type="personal">
<namePart type="given">Jing</namePart>
<namePart type="family">Gao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Weiqiang</namePart>
<namePart type="family">Jin</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Guoshuai</namePart>
<namePart type="family">Zhao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Peigen</namePart>
<namePart type="family">Ye</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xueming</namePart>
<namePart type="family">Qian</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hongyang</namePart>
<namePart type="family">Du</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>Reasoning in Large Language Models (LLMs) is primarily conducted in a discrete token space. Although recent approaches employing continuous concept tokens facilitate abstract reasoning, they introduce significant stability challenges. The accumulation of Out-of-Distribution (OOD) states frequently causes manifold deviation, while the absence of clear boundaries leads to indeterminate termination, often referred to as the anchoring problem. To address these limitations, we propose HybridThink. We formulate mode selection in LLM reasoning as a sequential decision process. A lightweight gating policy is trained with RL to decide at each step whether to propagate a continuous concept token or generate a discrete text token to anchor the semantic trajectory.This learned interleaving limits manifold deviation and resolves the fragility of static rules. Furthermore, we ensure stable convergence through a training strategy that incorporates Error-Centric Experience Replay (ECER) and Inductive Exploration Shaping (IES), prioritizing learning from failure modes, specifically drifting and premature termination. Empirical evaluations on mathematics and coding benchmarks demonstrate that HybridThink outperforms standard CoT and Soft Thinking baselines, achieving superior accuracy with reduced token consumption.</abstract>
<identifier type="citekey">gao-etal-2026-hybridthink</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.91/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>2136</start>
<end>2164</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T HybridThink: Learned Interleaving of Continuous and Discrete Reasoning in Large Language Models
%A Gao, Jing
%A Jin, Weiqiang
%A Zhao, Guoshuai
%A Ye, Peigen
%A Qian, Xueming
%A Du, Hongyang
%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 gao-etal-2026-hybridthink
%X Reasoning in Large Language Models (LLMs) is primarily conducted in a discrete token space. Although recent approaches employing continuous concept tokens facilitate abstract reasoning, they introduce significant stability challenges. The accumulation of Out-of-Distribution (OOD) states frequently causes manifold deviation, while the absence of clear boundaries leads to indeterminate termination, often referred to as the anchoring problem. To address these limitations, we propose HybridThink. We formulate mode selection in LLM reasoning as a sequential decision process. A lightweight gating policy is trained with RL to decide at each step whether to propagate a continuous concept token or generate a discrete text token to anchor the semantic trajectory.This learned interleaving limits manifold deviation and resolves the fragility of static rules. Furthermore, we ensure stable convergence through a training strategy that incorporates Error-Centric Experience Replay (ECER) and Inductive Exploration Shaping (IES), prioritizing learning from failure modes, specifically drifting and premature termination. Empirical evaluations on mathematics and coding benchmarks demonstrate that HybridThink outperforms standard CoT and Soft Thinking baselines, achieving superior accuracy with reduced token consumption.
%U https://aclanthology.org/2026.emnlp-main.91/
%P 2136-2164
Markdown (Informal)
[HybridThink: Learned Interleaving of Continuous and Discrete Reasoning in Large Language Models](https://aclanthology.org/2026.emnlp-main.91/) (Gao et al., EMNLP 2026)
ACL