@inproceedings{mehrafarin-etal-2026-chain,
title = "When Chain-of-Thought Fails, the Solution Hides in the Hidden States",
author = "Mehrafarin, Houman and
Parekh, Amit and
Konstas, Ioannis",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {EMNLP} 2026",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-emnlp.2018/",
pages = "42939--42954",
ISBN = "979-8-89176-462-0",
abstract = "Whether intermediate reasoning is computationally useful or merely explanatory depends on whether chain-of-thought (CoT) tokens contain task-solving information. We present a mechanistic causal analysis of CoT on GSM8K using {\_}activation patching{\_}: transferring token-level hidden states from an arithmetic CoT generation to a direct-answer run for the same question, then measuring the effect on final-answer accuracy. Across models, patching can elicit generations that lead to correct answers, revealing that individual CoT hidden states can encode sufficient task-solving information even when the original CoT trace is incorrect. This information is more prevalent in correct than incorrect CoT runs and is unevenly distributed across tokens, concentrating in mid-to-late layers and appearing earlier in the reasoning trace. Moreover, patching from language tokens such as verbs and entities more frequently elicits correct reasoning traces, whereas patches from mathematical tokens more often produce answer-proximal generations that rarely succeed. Further analysis shows that earlier CoT tokens concentrate recoverable information, while later ones require broader search to find the correct answer. Together, these findings show that CoT encodes recoverable, token-level task-solving information, revealing how arithmetic reasoning is represented, accessed, and disrupted."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="mehrafarin-etal-2026-chain">
<titleInfo>
<title>When Chain-of-Thought Fails, the Solution Hides in the Hidden States</title>
</titleInfo>
<name type="personal">
<namePart type="given">Houman</namePart>
<namePart type="family">Mehrafarin</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Amit</namePart>
<namePart type="family">Parekh</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ioannis</namePart>
<namePart type="family">Konstas</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>Findings of the Association for Computational Linguistics: EMNLP 2026</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">979-8-89176-462-0</identifier>
</relatedItem>
<abstract>Whether intermediate reasoning is computationally useful or merely explanatory depends on whether chain-of-thought (CoT) tokens contain task-solving information. We present a mechanistic causal analysis of CoT on GSM8K using _activation patching_: transferring token-level hidden states from an arithmetic CoT generation to a direct-answer run for the same question, then measuring the effect on final-answer accuracy. Across models, patching can elicit generations that lead to correct answers, revealing that individual CoT hidden states can encode sufficient task-solving information even when the original CoT trace is incorrect. This information is more prevalent in correct than incorrect CoT runs and is unevenly distributed across tokens, concentrating in mid-to-late layers and appearing earlier in the reasoning trace. Moreover, patching from language tokens such as verbs and entities more frequently elicits correct reasoning traces, whereas patches from mathematical tokens more often produce answer-proximal generations that rarely succeed. Further analysis shows that earlier CoT tokens concentrate recoverable information, while later ones require broader search to find the correct answer. Together, these findings show that CoT encodes recoverable, token-level task-solving information, revealing how arithmetic reasoning is represented, accessed, and disrupted.</abstract>
<identifier type="citekey">mehrafarin-etal-2026-chain</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.2018/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>42939</start>
<end>42954</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T When Chain-of-Thought Fails, the Solution Hides in the Hidden States
%A Mehrafarin, Houman
%A Parekh, Amit
%A Konstas, Ioannis
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Findings of the Association for Computational Linguistics: EMNLP 2026
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-462-0
%F mehrafarin-etal-2026-chain
%X Whether intermediate reasoning is computationally useful or merely explanatory depends on whether chain-of-thought (CoT) tokens contain task-solving information. We present a mechanistic causal analysis of CoT on GSM8K using _activation patching_: transferring token-level hidden states from an arithmetic CoT generation to a direct-answer run for the same question, then measuring the effect on final-answer accuracy. Across models, patching can elicit generations that lead to correct answers, revealing that individual CoT hidden states can encode sufficient task-solving information even when the original CoT trace is incorrect. This information is more prevalent in correct than incorrect CoT runs and is unevenly distributed across tokens, concentrating in mid-to-late layers and appearing earlier in the reasoning trace. Moreover, patching from language tokens such as verbs and entities more frequently elicits correct reasoning traces, whereas patches from mathematical tokens more often produce answer-proximal generations that rarely succeed. Further analysis shows that earlier CoT tokens concentrate recoverable information, while later ones require broader search to find the correct answer. Together, these findings show that CoT encodes recoverable, token-level task-solving information, revealing how arithmetic reasoning is represented, accessed, and disrupted.
%U https://aclanthology.org/2026.findings-emnlp.2018/
%P 42939-42954
Markdown (Informal)
[When Chain-of-Thought Fails, the Solution Hides in the Hidden States](https://aclanthology.org/2026.findings-emnlp.2018/) (Mehrafarin et al., Findings 2026)
ACL