@inproceedings{zhang-etal-2026-honesty,
title = "An Honesty-Aware Multi-Agent Framework for Synthetic Psychiatric Intake Data Generation",
author = "Zhang, Xinyuan and
Dao, Chang and
Wang, Zijian and
Pang, Yechao and
Zhou, Juexiao",
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.2334/",
pages = "50172--50198",
ISBN = "979-8-89176-462-0",
abstract = "Psychiatric intake combines self-report scales with clinician-rated interviewing, but public full-trace data are scarce and most synthetic resources assume truthful reporting. We present an honesty-aware multi-agent framework for psychiatric intake simulation that explicitly models honest reporting, concealment, and malingering under asymmetric information. Starting from structured patient profiles, five role-specific agents generate self-report scale outputs, strategy-guided clinician-rated interviews, honesty inference from the interview transcript alone, and honesty-aware diagnostic fusion over four common psychiatric domains: major depressive disorder (MDD), anxiety disorder (AD), bipolar disorder (BD), and attention-deficit/hyperactivity disorder (ADHD). Using normalized clinical and healthy source profiles, we construct a dataset of 2,008 complete intake sessions with 58,918 dialogue turns, linking profiles, scales, strategy traces, inferred honesty, and final diagnostic reports. Experiments show that assigned reporting conditions are expressed in generated patient behavior and can be recovered from the interview transcript alone, with condition-level inference accuracy between 60.1{\%} and 94.8{\%}. Under hidden-information constraints, the best LLM achieves 85.4{\%} exact recovery of synthetic reference domains with 0.99 severity MAE and zero detected leakage. Honesty-aware fusion reduces deviation from benchmark reference labels relative to self-report alone, and strategy-guided interviewing attains higher accuracy with fewer evaluator questions than item-by-item questioning. These results support honesty-aware synthetic intake data as a controlled research benchmark, not a clinically validated pre-assessment system."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="zhang-etal-2026-honesty">
<titleInfo>
<title>An Honesty-Aware Multi-Agent Framework for Synthetic Psychiatric Intake Data Generation</title>
</titleInfo>
<name type="personal">
<namePart type="given">Xinyuan</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Chang</namePart>
<namePart type="family">Dao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zijian</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yechao</namePart>
<namePart type="family">Pang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Juexiao</namePart>
<namePart type="family">Zhou</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>Psychiatric intake combines self-report scales with clinician-rated interviewing, but public full-trace data are scarce and most synthetic resources assume truthful reporting. We present an honesty-aware multi-agent framework for psychiatric intake simulation that explicitly models honest reporting, concealment, and malingering under asymmetric information. Starting from structured patient profiles, five role-specific agents generate self-report scale outputs, strategy-guided clinician-rated interviews, honesty inference from the interview transcript alone, and honesty-aware diagnostic fusion over four common psychiatric domains: major depressive disorder (MDD), anxiety disorder (AD), bipolar disorder (BD), and attention-deficit/hyperactivity disorder (ADHD). Using normalized clinical and healthy source profiles, we construct a dataset of 2,008 complete intake sessions with 58,918 dialogue turns, linking profiles, scales, strategy traces, inferred honesty, and final diagnostic reports. Experiments show that assigned reporting conditions are expressed in generated patient behavior and can be recovered from the interview transcript alone, with condition-level inference accuracy between 60.1% and 94.8%. Under hidden-information constraints, the best LLM achieves 85.4% exact recovery of synthetic reference domains with 0.99 severity MAE and zero detected leakage. Honesty-aware fusion reduces deviation from benchmark reference labels relative to self-report alone, and strategy-guided interviewing attains higher accuracy with fewer evaluator questions than item-by-item questioning. These results support honesty-aware synthetic intake data as a controlled research benchmark, not a clinically validated pre-assessment system.</abstract>
<identifier type="citekey">zhang-etal-2026-honesty</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.2334/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>50172</start>
<end>50198</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T An Honesty-Aware Multi-Agent Framework for Synthetic Psychiatric Intake Data Generation
%A Zhang, Xinyuan
%A Dao, Chang
%A Wang, Zijian
%A Pang, Yechao
%A Zhou, Juexiao
%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 zhang-etal-2026-honesty
%X Psychiatric intake combines self-report scales with clinician-rated interviewing, but public full-trace data are scarce and most synthetic resources assume truthful reporting. We present an honesty-aware multi-agent framework for psychiatric intake simulation that explicitly models honest reporting, concealment, and malingering under asymmetric information. Starting from structured patient profiles, five role-specific agents generate self-report scale outputs, strategy-guided clinician-rated interviews, honesty inference from the interview transcript alone, and honesty-aware diagnostic fusion over four common psychiatric domains: major depressive disorder (MDD), anxiety disorder (AD), bipolar disorder (BD), and attention-deficit/hyperactivity disorder (ADHD). Using normalized clinical and healthy source profiles, we construct a dataset of 2,008 complete intake sessions with 58,918 dialogue turns, linking profiles, scales, strategy traces, inferred honesty, and final diagnostic reports. Experiments show that assigned reporting conditions are expressed in generated patient behavior and can be recovered from the interview transcript alone, with condition-level inference accuracy between 60.1% and 94.8%. Under hidden-information constraints, the best LLM achieves 85.4% exact recovery of synthetic reference domains with 0.99 severity MAE and zero detected leakage. Honesty-aware fusion reduces deviation from benchmark reference labels relative to self-report alone, and strategy-guided interviewing attains higher accuracy with fewer evaluator questions than item-by-item questioning. These results support honesty-aware synthetic intake data as a controlled research benchmark, not a clinically validated pre-assessment system.
%U https://aclanthology.org/2026.findings-emnlp.2334/
%P 50172-50198
Markdown (Informal)
[An Honesty-Aware Multi-Agent Framework for Synthetic Psychiatric Intake Data Generation](https://aclanthology.org/2026.findings-emnlp.2334/) (Zhang et al., Findings 2026)
ACL