@inproceedings{gupta-etal-2026-anchored,
title = "Anchored Turns: A Length-Driven Gradient in Chain-of-Thought Causal Necessity in Sharded Arithmetic Conversations",
author = "Gupta, Dhruv and
Majhi, Vasudev and
Challa, Jagat Sesh and
Mandal, Murari and
Kumar, Dhruv",
editor = "Zhang, Yue and
Xu, Ruifeng and
Purwarianti, Ayu and
Rong, Tong and
Suzuki, Jun",
booktitle = "Proceedings of the 15th International Joint Conference on Natural Language Processing and the 5th Conference of the {A}sia-Pacific Chapter of the {A}ssociation for {C}omputational {L}inguistics (Volume 1: Long Papers)",
month = nov,
year = "2026",
address = "Hengqin, China",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.aacl-long.117/",
pages = "2536--2558",
ISBN = "979-8-89176-439-2",
abstract = "Chain-of-thought (CoT) faithfulness is measured almost exclusively in single-turn settings, leaving open how the reasoning-trace-to-answer relationship evolves as conversations lengthen. We introduce turn-level CoT counterfactual deletion: at every assistant turn we regenerate the answer at five truncation levels of the current-turn thinking block under greedy decoding, and label a turn anchored{---}a descriptive label for causal redundancy of the trace, not a mechanistic claim{---}when all five regenerations agree on the same answer. Across 67 multi-turn sharded-GSM8K conversations from DeepSeek-R1-Distill-Qwen-7B, the anchored fraction grows 3.9{\texttimes} from short ({\ensuremath{<}} 10 turns, 7.2{\%}) to long ({\ensuremath{>}} 20 turns, 28.4{\%}). Strict early {\ensuremath{<}} mid {\ensuremath{<}} late monotone ordering holds in five of seven reasoning models from three base families and four post-training pipelines (binomial sign test, p=0.002). A supplementary pooled GEE across all seven models converges on this direction ({\ensuremath{\beta}}=1.12, p=5.1{\texttimes}10{\ensuremath{^-}}{\ensuremath{^8}}, n=2,838 turn-level rows). A surface-feature predictor flags anchored turns at AUC= 0.71 (+0.07 over a length-only baseline) without additional generation, providing a cheap online signal for CoT monitors calibrated on short conversations. Single-turn CoT benchmarks, by construction, cannot detect this gradient."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="gupta-etal-2026-anchored">
<titleInfo>
<title>Anchored Turns: A Length-Driven Gradient in Chain-of-Thought Causal Necessity in Sharded Arithmetic Conversations</title>
</titleInfo>
<name type="personal">
<namePart type="given">Dhruv</namePart>
<namePart type="family">Gupta</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Vasudev</namePart>
<namePart type="family">Majhi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jagat</namePart>
<namePart type="given">Sesh</namePart>
<namePart type="family">Challa</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Murari</namePart>
<namePart type="family">Mandal</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Dhruv</namePart>
<namePart type="family">Kumar</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-11</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 15th International Joint Conference on Natural Language Processing and the 5th Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics (Volume 1: Long Papers)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yue</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ruifeng</namePart>
<namePart type="family">Xu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ayu</namePart>
<namePart type="family">Purwarianti</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tong</namePart>
<namePart type="family">Rong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jun</namePart>
<namePart type="family">Suzuki</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Hengqin, China</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-439-2</identifier>
</relatedItem>
<abstract>Chain-of-thought (CoT) faithfulness is measured almost exclusively in single-turn settings, leaving open how the reasoning-trace-to-answer relationship evolves as conversations lengthen. We introduce turn-level CoT counterfactual deletion: at every assistant turn we regenerate the answer at five truncation levels of the current-turn thinking block under greedy decoding, and label a turn anchored—a descriptive label for causal redundancy of the trace, not a mechanistic claim—when all five regenerations agree on the same answer. Across 67 multi-turn sharded-GSM8K conversations from DeepSeek-R1-Distill-Qwen-7B, the anchored fraction grows 3.9× from short (\ensuremath< 10 turns, 7.2%) to long (\ensuremath> 20 turns, 28.4%). Strict early \ensuremath< mid \ensuremath< late monotone ordering holds in five of seven reasoning models from three base families and four post-training pipelines (binomial sign test, p=0.002). A supplementary pooled GEE across all seven models converges on this direction (\ensuremathβ=1.12, p=5.1×10\ensuremath⁻\ensuremath⁸, n=2,838 turn-level rows). A surface-feature predictor flags anchored turns at AUC= 0.71 (+0.07 over a length-only baseline) without additional generation, providing a cheap online signal for CoT monitors calibrated on short conversations. Single-turn CoT benchmarks, by construction, cannot detect this gradient.</abstract>
<identifier type="citekey">gupta-etal-2026-anchored</identifier>
<location>
<url>https://aclanthology.org/2026.aacl-long.117/</url>
</location>
<part>
<date>2026-11</date>
<extent unit="page">
<start>2536</start>
<end>2558</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Anchored Turns: A Length-Driven Gradient in Chain-of-Thought Causal Necessity in Sharded Arithmetic Conversations
%A Gupta, Dhruv
%A Majhi, Vasudev
%A Challa, Jagat Sesh
%A Mandal, Murari
%A Kumar, Dhruv
%Y Zhang, Yue
%Y Xu, Ruifeng
%Y Purwarianti, Ayu
%Y Rong, Tong
%Y Suzuki, Jun
%S Proceedings of the 15th International Joint Conference on Natural Language Processing and the 5th Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics (Volume 1: Long Papers)
%D 2026
%8 November
%I Association for Computational Linguistics
%C Hengqin, China
%@ 979-8-89176-439-2
%F gupta-etal-2026-anchored
%X Chain-of-thought (CoT) faithfulness is measured almost exclusively in single-turn settings, leaving open how the reasoning-trace-to-answer relationship evolves as conversations lengthen. We introduce turn-level CoT counterfactual deletion: at every assistant turn we regenerate the answer at five truncation levels of the current-turn thinking block under greedy decoding, and label a turn anchored—a descriptive label for causal redundancy of the trace, not a mechanistic claim—when all five regenerations agree on the same answer. Across 67 multi-turn sharded-GSM8K conversations from DeepSeek-R1-Distill-Qwen-7B, the anchored fraction grows 3.9× from short (\ensuremath< 10 turns, 7.2%) to long (\ensuremath> 20 turns, 28.4%). Strict early \ensuremath< mid \ensuremath< late monotone ordering holds in five of seven reasoning models from three base families and four post-training pipelines (binomial sign test, p=0.002). A supplementary pooled GEE across all seven models converges on this direction (\ensuremathβ=1.12, p=5.1×10\ensuremath⁻\ensuremath⁸, n=2,838 turn-level rows). A surface-feature predictor flags anchored turns at AUC= 0.71 (+0.07 over a length-only baseline) without additional generation, providing a cheap online signal for CoT monitors calibrated on short conversations. Single-turn CoT benchmarks, by construction, cannot detect this gradient.
%U https://aclanthology.org/2026.aacl-long.117/
%P 2536-2558
Markdown (Informal)
[Anchored Turns: A Length-Driven Gradient in Chain-of-Thought Causal Necessity in Sharded Arithmetic Conversations](https://aclanthology.org/2026.aacl-long.117/) (Gupta et al., AACL-IJCNLP 2026)
ACL