@inproceedings{ding-zhang-2026-causal,
title = "Causal Tongue-Tie: {LLM}s Can Encode Causal Direction, But Their Yes/No Outputs Fail to Express",
author = "Ding, Ziyi and
Zhang, Xiao-Ping",
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.217/",
pages = "4976--4998",
ISBN = "979-8-89176-459-0",
abstract = "We find a mismatch between the causal direction recoverable from a large language model{'}s hidden state and its final Yes/No answer. On anti-commonsense CLadder (Jin et al., 2024) items, a frozen linear probe recovers the evidence-supported direction with accuracy 0.97{--}1.00, while ordinary Yes/No accuracy remains 0.35{--}0.53. This effect disappears in randomly initialized models, so it depends on the trained representation rather than probe capacity alone. Erasing the recovered direction reduces the model{'}s internal comparison of paired prompts by 96{\%}, and replacing the full hidden state flips 57{\%} of previously wrong answers. A 2 {\texttimes} 2 interface control further shows that changing only the answer tokens does not fix the error; explicitly contrasting the two candidate directions raises accuracy to 0.95{--}1.00 across Qwen, Mistral, and Llama. We call this mismatch Causal Tongue-Tie. It does not establish human-like causal understanding. Our main finding is that a wrong Yes/No answer can coexist with an evidence-aligned direction that the tested probe recovers from the same forward pass. Output-only accuracy therefore cannot distinguish an absent internal signal from a present but unexpressed one."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="ding-zhang-2026-causal">
<titleInfo>
<title>Causal Tongue-Tie: LLMs Can Encode Causal Direction, But Their Yes/No Outputs Fail to Express</title>
</titleInfo>
<name type="personal">
<namePart type="given">Ziyi</namePart>
<namePart type="family">Ding</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xiao-Ping</namePart>
<namePart type="family">Zhang</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>We find a mismatch between the causal direction recoverable from a large language model’s hidden state and its final Yes/No answer. On anti-commonsense CLadder (Jin et al., 2024) items, a frozen linear probe recovers the evidence-supported direction with accuracy 0.97–1.00, while ordinary Yes/No accuracy remains 0.35–0.53. This effect disappears in randomly initialized models, so it depends on the trained representation rather than probe capacity alone. Erasing the recovered direction reduces the model’s internal comparison of paired prompts by 96%, and replacing the full hidden state flips 57% of previously wrong answers. A 2 × 2 interface control further shows that changing only the answer tokens does not fix the error; explicitly contrasting the two candidate directions raises accuracy to 0.95–1.00 across Qwen, Mistral, and Llama. We call this mismatch Causal Tongue-Tie. It does not establish human-like causal understanding. Our main finding is that a wrong Yes/No answer can coexist with an evidence-aligned direction that the tested probe recovers from the same forward pass. Output-only accuracy therefore cannot distinguish an absent internal signal from a present but unexpressed one.</abstract>
<identifier type="citekey">ding-zhang-2026-causal</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.217/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>4976</start>
<end>4998</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Causal Tongue-Tie: LLMs Can Encode Causal Direction, But Their Yes/No Outputs Fail to Express
%A Ding, Ziyi
%A Zhang, Xiao-Ping
%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 ding-zhang-2026-causal
%X We find a mismatch between the causal direction recoverable from a large language model’s hidden state and its final Yes/No answer. On anti-commonsense CLadder (Jin et al., 2024) items, a frozen linear probe recovers the evidence-supported direction with accuracy 0.97–1.00, while ordinary Yes/No accuracy remains 0.35–0.53. This effect disappears in randomly initialized models, so it depends on the trained representation rather than probe capacity alone. Erasing the recovered direction reduces the model’s internal comparison of paired prompts by 96%, and replacing the full hidden state flips 57% of previously wrong answers. A 2 × 2 interface control further shows that changing only the answer tokens does not fix the error; explicitly contrasting the two candidate directions raises accuracy to 0.95–1.00 across Qwen, Mistral, and Llama. We call this mismatch Causal Tongue-Tie. It does not establish human-like causal understanding. Our main finding is that a wrong Yes/No answer can coexist with an evidence-aligned direction that the tested probe recovers from the same forward pass. Output-only accuracy therefore cannot distinguish an absent internal signal from a present but unexpressed one.
%U https://aclanthology.org/2026.emnlp-main.217/
%P 4976-4998
Markdown (Informal)
[Causal Tongue-Tie: LLMs Can Encode Causal Direction, But Their Yes/No Outputs Fail to Express](https://aclanthology.org/2026.emnlp-main.217/) (Ding & Zhang, EMNLP 2026)
ACL