@inproceedings{zhao-coulson-2026-dissecting,
title = "Dissecting Lie Expectancy in Language Models",
author = "Zhao, Xingyuan and
Coulson, Seana",
editor = "Arad, Dana and
Belinkov, Yonatan and
Chen, Hanjie and
Feldhus, Nils and
Mueller, Aaron and
Mohebbi, Hosein and
Sarti, Gabriele and
Tutek, Martin",
booktitle = "Proceedings of the 9th {B}lackbox{NLP} Workshop: Analyzing and Interpreting Neural Networks for {NLP}",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.blackboxnlp-1.42/",
pages = "726--738",
ISBN = "979-8-89176-476-7",
abstract = "When the speaker has a motive to lie {---} profit, face-saving, fear of punishment {---} an informed reader expects the lie. Forming this expectation requires three distinct abilities: representing what is actually true in the described situation, aligning linguistic expectations with the speaker{'}s incentives to be truthful, and adapting those expectations across situations. We study whether twelve open Qwen3 and OLMo2 models possess these abilities, asking how each is organized, whether they are separable, and when each is acquired. Within each model, fact information is linearly decodable at mid depth but becomes directly expressed only in the final third of the network; motive recognition emerges slightly later, and both grow with scale. Two-layer ablations dissociate them: removing a mid-depth band of layers weakens fact expression and motive recognition together. Ablating a deeper pair of layers, in contrast, leaves both essentially intact while eliminating human alignment: being sensitive to motives and being sensitive in the way humans are turn out to be separable abilities. Across 19 public checkpoints of OLMo2-7B, fact representation and motive recognition become detectable at around 0.3T tokens and grow steadily throughout pretraining, while human alignment stays non-significant until mid-training {---} a brief continued training on curated data. Altogether, representing what is true and recognizing whether a speaker is motivated to lie are closely coordinated; matching human judgments about when lying is likely is a later and partially separable achievement."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="zhao-coulson-2026-dissecting">
<titleInfo>
<title>Dissecting Lie Expectancy in Language Models</title>
</titleInfo>
<name type="personal">
<namePart type="given">Xingyuan</namePart>
<namePart type="family">Zhao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Seana</namePart>
<namePart type="family">Coulson</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 9th BlackboxNLP Workshop: Analyzing and Interpreting Neural Networks for NLP</title>
</titleInfo>
<name type="personal">
<namePart type="given">Dana</namePart>
<namePart type="family">Arad</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yonatan</namePart>
<namePart type="family">Belinkov</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hanjie</namePart>
<namePart type="family">Chen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Nils</namePart>
<namePart type="family">Feldhus</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Aaron</namePart>
<namePart type="family">Mueller</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hosein</namePart>
<namePart type="family">Mohebbi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Gabriele</namePart>
<namePart type="family">Sarti</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Martin</namePart>
<namePart type="family">Tutek</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">979-8-89176-476-7</identifier>
</relatedItem>
<abstract>When the speaker has a motive to lie — profit, face-saving, fear of punishment — an informed reader expects the lie. Forming this expectation requires three distinct abilities: representing what is actually true in the described situation, aligning linguistic expectations with the speaker’s incentives to be truthful, and adapting those expectations across situations. We study whether twelve open Qwen3 and OLMo2 models possess these abilities, asking how each is organized, whether they are separable, and when each is acquired. Within each model, fact information is linearly decodable at mid depth but becomes directly expressed only in the final third of the network; motive recognition emerges slightly later, and both grow with scale. Two-layer ablations dissociate them: removing a mid-depth band of layers weakens fact expression and motive recognition together. Ablating a deeper pair of layers, in contrast, leaves both essentially intact while eliminating human alignment: being sensitive to motives and being sensitive in the way humans are turn out to be separable abilities. Across 19 public checkpoints of OLMo2-7B, fact representation and motive recognition become detectable at around 0.3T tokens and grow steadily throughout pretraining, while human alignment stays non-significant until mid-training — a brief continued training on curated data. Altogether, representing what is true and recognizing whether a speaker is motivated to lie are closely coordinated; matching human judgments about when lying is likely is a later and partially separable achievement.</abstract>
<identifier type="citekey">zhao-coulson-2026-dissecting</identifier>
<location>
<url>https://aclanthology.org/2026.blackboxnlp-1.42/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>726</start>
<end>738</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Dissecting Lie Expectancy in Language Models
%A Zhao, Xingyuan
%A Coulson, Seana
%Y Arad, Dana
%Y Belinkov, Yonatan
%Y Chen, Hanjie
%Y Feldhus, Nils
%Y Mueller, Aaron
%Y Mohebbi, Hosein
%Y Sarti, Gabriele
%Y Tutek, Martin
%S Proceedings of the 9th BlackboxNLP Workshop: Analyzing and Interpreting Neural Networks for NLP
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-476-7
%F zhao-coulson-2026-dissecting
%X When the speaker has a motive to lie — profit, face-saving, fear of punishment — an informed reader expects the lie. Forming this expectation requires three distinct abilities: representing what is actually true in the described situation, aligning linguistic expectations with the speaker’s incentives to be truthful, and adapting those expectations across situations. We study whether twelve open Qwen3 and OLMo2 models possess these abilities, asking how each is organized, whether they are separable, and when each is acquired. Within each model, fact information is linearly decodable at mid depth but becomes directly expressed only in the final third of the network; motive recognition emerges slightly later, and both grow with scale. Two-layer ablations dissociate them: removing a mid-depth band of layers weakens fact expression and motive recognition together. Ablating a deeper pair of layers, in contrast, leaves both essentially intact while eliminating human alignment: being sensitive to motives and being sensitive in the way humans are turn out to be separable abilities. Across 19 public checkpoints of OLMo2-7B, fact representation and motive recognition become detectable at around 0.3T tokens and grow steadily throughout pretraining, while human alignment stays non-significant until mid-training — a brief continued training on curated data. Altogether, representing what is true and recognizing whether a speaker is motivated to lie are closely coordinated; matching human judgments about when lying is likely is a later and partially separable achievement.
%U https://aclanthology.org/2026.blackboxnlp-1.42/
%P 726-738
Markdown (Informal)
[Dissecting Lie Expectancy in Language Models](https://aclanthology.org/2026.blackboxnlp-1.42/) (Zhao & Coulson, BlackboxNLP 2026)
ACL
- Xingyuan Zhao and Seana Coulson. 2026. Dissecting Lie Expectancy in Language Models. In Proceedings of the 9th BlackboxNLP Workshop: Analyzing and Interpreting Neural Networks for NLP, pages 726–738, Budapest, Hungary. Association for Computational Linguistics.