@inproceedings{cho-etal-2026-adaptive,
title = "Adaptive Latent Block Insertion via Online Reward Optimization",
author = "Cho, Yewon and
Hwang, Hyeongsoon and
Park, Yein and
Hwang, Hyeon and
Kang, Jaewoo",
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.1833/",
pages = "41715--41736",
ISBN = "979-8-89176-459-0",
abstract = "Chain-of-thought prompting improves reasoning by generating intermediate steps as discrete text, but each step is forced to commit to a single token even when the model is uncertain. Continuous latent reasoning offers an alternative: a latent state can preserve multiple candidate interpretations without immediately selecting one token. Existing training-free approaches either place latent states only before decoding or switch between explicit and latent modes without optimizing the inserted state itself. We propose MidThink, a training-free framework that inserts and optimizes latent blocks both at the start of reasoning and adaptively during generation. When a confidence trigger detects sustained mid-stream uncertainty, MidThink inserts a new latent block and refines it with a rollout-based reward computed from short future continuations. We further show that initialization strongly affects whether a mid-stream block preserves a superposition over multiple tokens or collapses toward a single token. Across five mathematical reasoning benchmarks, MidThink improves over training-free latent-reasoning baselines, with the largest gains on the AIME benchmarks. Mechanistic analysis links these gains to three measurable properties of the latent block: distributional entropy across layers, attention centroid position, and stability under reinforcement learning updates."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="cho-etal-2026-adaptive">
<titleInfo>
<title>Adaptive Latent Block Insertion via Online Reward Optimization</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yewon</namePart>
<namePart type="family">Cho</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hyeongsoon</namePart>
<namePart type="family">Hwang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yein</namePart>
<namePart type="family">Park</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hyeon</namePart>
<namePart type="family">Hwang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jaewoo</namePart>
<namePart type="family">Kang</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>Chain-of-thought prompting improves reasoning by generating intermediate steps as discrete text, but each step is forced to commit to a single token even when the model is uncertain. Continuous latent reasoning offers an alternative: a latent state can preserve multiple candidate interpretations without immediately selecting one token. Existing training-free approaches either place latent states only before decoding or switch between explicit and latent modes without optimizing the inserted state itself. We propose MidThink, a training-free framework that inserts and optimizes latent blocks both at the start of reasoning and adaptively during generation. When a confidence trigger detects sustained mid-stream uncertainty, MidThink inserts a new latent block and refines it with a rollout-based reward computed from short future continuations. We further show that initialization strongly affects whether a mid-stream block preserves a superposition over multiple tokens or collapses toward a single token. Across five mathematical reasoning benchmarks, MidThink improves over training-free latent-reasoning baselines, with the largest gains on the AIME benchmarks. Mechanistic analysis links these gains to three measurable properties of the latent block: distributional entropy across layers, attention centroid position, and stability under reinforcement learning updates.</abstract>
<identifier type="citekey">cho-etal-2026-adaptive</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.1833/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>41715</start>
<end>41736</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Adaptive Latent Block Insertion via Online Reward Optimization
%A Cho, Yewon
%A Hwang, Hyeongsoon
%A Park, Yein
%A Hwang, Hyeon
%A Kang, Jaewoo
%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 cho-etal-2026-adaptive
%X Chain-of-thought prompting improves reasoning by generating intermediate steps as discrete text, but each step is forced to commit to a single token even when the model is uncertain. Continuous latent reasoning offers an alternative: a latent state can preserve multiple candidate interpretations without immediately selecting one token. Existing training-free approaches either place latent states only before decoding or switch between explicit and latent modes without optimizing the inserted state itself. We propose MidThink, a training-free framework that inserts and optimizes latent blocks both at the start of reasoning and adaptively during generation. When a confidence trigger detects sustained mid-stream uncertainty, MidThink inserts a new latent block and refines it with a rollout-based reward computed from short future continuations. We further show that initialization strongly affects whether a mid-stream block preserves a superposition over multiple tokens or collapses toward a single token. Across five mathematical reasoning benchmarks, MidThink improves over training-free latent-reasoning baselines, with the largest gains on the AIME benchmarks. Mechanistic analysis links these gains to three measurable properties of the latent block: distributional entropy across layers, attention centroid position, and stability under reinforcement learning updates.
%U https://aclanthology.org/2026.emnlp-main.1833/
%P 41715-41736
Markdown (Informal)
[Adaptive Latent Block Insertion via Online Reward Optimization](https://aclanthology.org/2026.emnlp-main.1833/) (Cho et al., EMNLP 2026)
ACL
- Yewon Cho, Hyeongsoon Hwang, Yein Park, Hyeon Hwang, and Jaewoo Kang. 2026. Adaptive Latent Block Insertion via Online Reward Optimization. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 41715–41736, Budapest, Hungary. Association for Computational Linguistics.