pith. sign in

arxiv: 2606.08254 · v1 · pith:M4AZF5LOnew · submitted 2026-06-06 · 💻 cs.CL

SSR: Can Simulated Patients Learn to Stigmatize Themselves? Modeling Self-Stigma through Internal Monologue

Kunyao Lan , Bingrui Jin , Zichen Zhu , Mengyue Wu This is my paper

Pith reviewed 2026-06-27 19:39 UTC · model grok-4.3

classification 💻 cs.CL
keywords self-stigmaLLM simulationpatient agentsinternal monologuemental healthchain of thoughtdialogue dataset
0
0 comments X

The pith

By augmenting dialogues with internal monologues, LLMs can be trained to simulate patients who adjust self-stigma based on conversation.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper tries to fix a gap in LLM patient simulations for mental health training: they miss self-stigma, where patients internalize stereotypes and show resistant behaviors that depend on the situation. Instead of static or always-compliant responses, the goal is dynamic adjustment like avoidance or self-blame when triggers appear. To do this, the authors build an SSR dataset that adds internal monologues based on the 3A1H model to standard dialogues, then fine-tune LLMs with chain-of-thought to learn stigma-aware reasoning. Tests show these agents generate responses judged more authentic and fitting than those from other methods. This would matter for creating training tools that better prepare clinicians for real patient interactions involving stigma.

Core claim

Grounded in the 3A1H model, the Stigmatized Self-Reflection dataset augments mental health dialogues with internal monologues that capture stigma-aware reasoning; fine-tuning LLMs on this data via chain-of-thought enables patient agents to dynamically adjust their stigma level and expression according to conversational triggers, resulting in more authentic and situationally appropriate responses than specialized baselines.

What carries the argument

The SSR dataset and chain-of-thought fine-tuning that incorporates internal monologues from the 3A1H self-stigmatization model.

If this is right

  • Agents respond with context-sensitive resistance rather than uniform compliance.
  • Performance exceeds baselines in authenticity evaluations.
  • The simulation becomes suitable for clinical training scenarios involving self-stigma.
  • Empathetic dialogue systems can better handle patients who internalize negative views.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • This technique of adding internal reasoning could apply to modeling other internalized attitudes in AI agents.
  • Better patient simulations may lead to improved detection of stigma-related issues in real-world AI counseling tools.
  • The reliance on a specific psychological model raises questions about generalizing to diverse cultural expressions of self-stigma.

Load-bearing premise

The 3A1H model of self-stigmatization provides a valid basis for augmenting dialogues with internal monologues that enable LLMs to learn and express context-sensitive stigma behaviors, and that evaluations can reliably judge the outputs as more authentic.

What would settle it

A study where the SSR model and baselines generate responses to the same set of stigma-triggering prompts, and independent raters show no preference or difference in ratings for authenticity and appropriateness.

Figures

Figures reproduced from arXiv: 2606.08254 by Bingrui Jin, Kunyao Lan, Mengyue Wu, Zichen Zhu.

Figure 1
Figure 1. Figure 1: Proportion change in life-event mentions between stigma-triggering and non-stigmatizing dialogues across four mental health datasets. Bars show the relative increase or decrease in frequency when stigma is present, highlighting context-sensitive stigma triggers (e.g., Identity, Relationships). Baselines. We compare fine-tuned models with two prompt-based patient simulation baselines: • UPSD4 (Cao et al., 2… view at source ↗
Figure 2
Figure 2. Figure 2: Distribution of stigma expression types across [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: SSMIS and ISMI scale scores for SSR-fine-tuned and baseline models. SSR models pro￾duce higher stigma scores, reflecting more authentic self-stigmatization in simulated patient responses. The psychometric scale evaluations provide an￾other view of whether the simulated persona oc￾cupies a meaningful self-stigma range rather than merely producing more resistant dialogue acts. As illustrated in [PITH_FULL_I… view at source ↗
read the original abstract

Simulating patients with large language models (LLMs) is a promising tool for mental health training, but existing approaches fail to capture a key clinical reality: self-stigma. Patients experiencing self-stigma, the internalization of negative stereotypes, often exhibit context-sensitive resistance, such as avoidance, denial, or self-blame, which current models render as static or uniformly compliant behavior. To address this, we introduce a novel simulation framework grounded in the psychological 3A1H model of self-stigmatization. Our core innovation is the creation of a \textbf{Stigmatized Self-Reflection} (\textbf{SSR}) dataset, where we augment mental health dialogues with internal monologues that reflect stigma-aware reasoning. By fine-tuning LLMs with this data using a chain-of-thought approach, we train patient agents to dynamically adjust their level and expression of stigma based on conversational triggers. Evaluations demonstrate that our approach significantly outperforms specialized baselines, generating more authentic and situationally appropriate patient responses. This work provides a crucial step towards realistic stigma simulation for clinical training and empathetic dialogue systems.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 2 minor

Summary. The manuscript introduces the Stigmatized Self-Reflection (SSR) dataset and framework for simulating self-stigma in patient agents powered by LLMs. Grounded in the 3A1H model of self-stigmatization, the approach augments mental health dialogues with internal monologues to capture context-sensitive stigma behaviors such as avoidance and self-blame. LLMs are fine-tuned using a chain-of-thought method to dynamically adjust stigma levels based on conversational triggers. The authors report that this method significantly outperforms specialized baselines in producing authentic patient responses.

Significance. If the empirical results hold, this contribution could meaningfully improve the fidelity of LLM-based simulations for mental health training by addressing the clinically important phenomenon of self-stigma, which current models overlook. The use of an external psychological model and new dataset creation is a strength.

major comments (1)
  1. [Abstract] Abstract: the claim that the approach 'significantly outperforms specialized baselines' and generates 'more authentic and situationally appropriate patient responses' is presented without any reported evaluation metrics, baseline definitions, statistical tests, or details on how authenticity was measured or judged by humans or automated evaluators. This is load-bearing for the central empirical claim.
minor comments (2)
  1. The description of the 3A1H model application and the exact procedure for augmenting dialogues with internal monologues would benefit from additional concrete examples or pseudocode to support reproducibility.
  2. Clarify the criteria used to select 'specialized baselines' and ensure the evaluation protocol (including inter-rater reliability if human judgments are involved) is fully specified.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their thoughtful review and for highlighting the need for greater specificity in the abstract. We address the major comment below and commit to revisions that strengthen the presentation of our empirical claims without altering the underlying methodology or results.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the claim that the approach 'significantly outperforms specialized baselines' and generates 'more authentic and situationally appropriate patient responses' is presented without any reported evaluation metrics, baseline definitions, statistical tests, or details on how authenticity was measured or judged by humans or automated evaluators. This is load-bearing for the central empirical claim.

    Authors: We agree that the abstract, as currently worded, does not include the supporting evaluation details and that this weakens the central claim. The full manuscript already contains the requested information in Sections 4 (Experimental Setup) and 5 (Results), including quantitative metrics (e.g., stigma alignment scores, perplexity, and human-rated authenticity on a 5-point scale), baseline definitions (fine-tuned GPT-2, Llama-2 without SSR, and rule-based 3A1H simulators), statistical tests (paired t-tests with p<0.01), and evaluation protocol (both automated classifiers and blinded human raters from clinical psychology). In the revision we will condense these elements into the abstract (approximately 30–40 additional words) while preserving the word limit, ensuring the claim is properly grounded. revision: yes

Circularity Check

0 steps flagged

No significant circularity identified

full rationale

The paper grounds its framework in the external 3A1H psychological model, creates a new SSR dataset by augmenting dialogues with internal monologues, fine-tunes LLMs via chain-of-thought, and evaluates against baselines. No derivations, equations, or claims reduce by construction to fitted inputs, self-definitions, or self-citation chains. The central claim rests on new data creation and standard training/evaluation rather than circular reduction.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Only the abstract is available; the central approach rests on the validity of the 3A1H model as a domain assumption with no free parameters or invented entities specified.

axioms (1)
  • domain assumption The 3A1H psychological model accurately captures the process of self-stigmatization
    The framework is explicitly grounded in this model per the abstract.

pith-pipeline@v0.9.1-grok · 5730 in / 1136 out tokens · 23777 ms · 2026-06-27T19:39:44.270791+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

39 extracted references · 10 canonical work pages

  1. [1]

    Mental health atlas 2024 , year =

    World Health Organization , publisher =. Mental health atlas 2024 , year =

    2024

  2. [2]

    World mental health today: latest data , year =

    World Health Organization , note =. World mental health today: latest data , year =

  3. [3]

    Yao, Binwei and Shi, Chao and Zou, Likai and Dai, Lingfeng and Wu, Mengyue and Chen, Lu and Wang, Zhen and Yu, Kai , booktitle =. D4: a. 2022 , doi =

    2022

  4. [4]

    Zhu and Kunyao Lan and Zhiling Zhang and Lyuchun Cui , title =

    Siyuan Chen and Mengyue Wu and Kenny Q. Zhu and Kunyao Lan and Zhiling Zhang and Lyuchun Cui , title =. 2023 , url =

    2023

  5. [5]

    2024 , url =

    Kunyao Lan and Cong Ming and Binwei Yao and Lu Chen and Mengyue Wu , title =. 2024 , url =

    2024

  6. [6]

    Zhu and Mengyue Wu , title =

    Kunyao Lan and Bingrui Jin and Zichen Zhu and Siyuan Chen and Shu Zhang and Kenny Q. Zhu and Mengyue Wu , title =. 2024 , url =

    2024

  7. [7]

    Mrinank Sharma and Meg Tong and Tomasz Korbak and David Duvenaud and Amanda Askell and Samuel R. Bowman and Esin DURMUS and Zac Hatfield-Dodds and Scott R Johnston and Shauna M Kravec and Timothy Maxwell and Sam McCandlish and Kamal Ndousse and Oliver Rausch and Nicholas Schiefer and Da Yan and Miranda Zhang and Ethan Perez , booktitle =. Towards Understa...

  8. [8]

    Understanding the impact of stigma on people with mental illness , volume =

    Corrigan, Patrick W and Watson, Amy C , journal =. Understanding the impact of stigma on people with mental illness , volume =. 2002 , language =

    2002

  9. [9]

    The quest for machine translation quality at

    Stadler, Hans-Udo and Peter-Sp. The quest for machine translation quality at. Proceedings of Machine Translation Summit XI: Papers , editor =. 2007 , url =

    2007

  10. [10]

    and Kaye, Kimberly E

    Sadler, Melody S. and Kaye, Kimberly E. and Vaughn, Allison A. , journal =. Competence and warmth stereotypes prompt mental illness stigma through emotions , volume =. 2015 , abstract =. doi:https://doi.org/10.1111/jasp.12323 , url =

    work page doi:10.1111/jasp.12323 2015

  11. [11]

    Breaking the Stigma! Unobtrusively Probe Symptoms in Depression Disorder Diagnosis Dialogue , year =

    Cao, Jieming and Huang, Chen and Zhang, Yanan and Deng, Ruibo and Zhang, Jincheng and Lei, Wenqiang , booktitle =. Breaking the Stigma! Unobtrusively Probe Symptoms in Depression Disorder Diagnosis Dialogue , year =. doi:10.18653/v1/2025.findings-naacl.10 , isbn =

    work page doi:10.18653/v1/2025.findings-naacl.10 2025

  12. [12]

    What is Stigma Attributed to? A Theory-Grounded, Expert-Annotated Interview Corpus for Demystifying Mental-Health Stigma , year =

    Meng, Han and Chen, Yancan and Li, Yunan and Yang, Yitian and Lee, Jungup and Zhang, Renwen and Lee, Yi-Chieh , booktitle =. What is Stigma Attributed to? A Theory-Grounded, Expert-Annotated Interview Corpus for Demystifying Mental-Health Stigma , year =. doi:10.18653/v1/2025.acl-long.272 , isbn =

    work page doi:10.18653/v1/2025.acl-long.272 2025

  13. [13]

    , publisher =

    Goffman, E. , publisher =. Stigma: Notes on the Management of Spoiled Identity , year =

  14. [14]

    and Zhi, Jiayin and Eack, Shaun M

    Wang, Ruiyi and Milani, Stephanie and Chiu, Jamie C. and Zhi, Jiayin and Eack, Shaun M. and Labrum, Travis and Murphy, Samuel M and Jones, Nev and Hardy, Kate V and Shen, Hong and Fang, Fei and Chen, Zhiyu , booktitle =. 2024 , doi =

    2024

  15. [15]

    2025 , url =

    Mithat Can Ozgun and Jiahuan Pei and Koen Hindriks and Lucia Donatelli and Qingzhi Liu and Junxiao Wang , title =. 2025 , url =

    2025

  16. [16]

    , journal =

    Every‐Wurtz, Cheryl R. , journal =. Counseling and Psychotherapy Transcripts, Client Narratives, and Reference Works , volume =. 2009 , abstract =. doi:10.1108/09504120910945317 , issn =

    work page doi:10.1108/09504120910945317 2009

  17. [17]

    Understanding Client Reactions in Online Mental Health Counseling , year =

    Li, Anqi and Ma, Lizhi and Mei, Yaling and He, Hongliang and Zhang, Shuai and Qiu, Huachuan and Lan, Zhenzhong , booktitle =. Understanding Client Reactions in Online Mental Health Counseling , year =. doi:10.18653/v1/2023.acl-long.577 , url =

    work page doi:10.18653/v1/2023.acl-long.577 2023

  18. [18]

    Towards Emotional Support Dialog Systems , year =

    Liu, Siyang and Zheng, Chujie and Demasi, Orianna and Sabour, Sahand and Li, Yu and Yu, Zhou and Jiang, Yong and Huang, Minlie , booktitle =. Towards Emotional Support Dialog Systems , year =. doi:10.18653/v1/2021.acl-long.269 , url =

    work page doi:10.18653/v1/2021.acl-long.269 2021

  19. [19]

    Self-stigma among people with serious mental illnesses: The use of focus groups to inform the development of a brief video intervention , volume =

    Amsalem, Doron and Rogers, R Tyler and Stroup, T Scott and Dixon, Lisa and Pope, Leah G , journal =. Self-stigma among people with serious mental illnesses: The use of focus groups to inform the development of a brief video intervention , volume =. 2023 , abstract =

    2023

  20. [20]

    Public Stigma and Self-Stigma: Differential Association With Attitudes Toward Formal and Informal Help Seeking , volume =

    Elise Pattyn and Mieke Verhaeghe and Charlotte Sercu and Piet Bracke , journal =. Public Stigma and Self-Stigma: Differential Association With Attitudes Toward Formal and Informal Help Seeking , volume =. 2014 , doi =

    2014

  21. [21]

    and Ajdacic-Gross, V

    Oexle, N. and Ajdacic-Gross, V. and Kilian, R. and Müller, M. and Rodgers, S. and Xu, Z. and Rössler, W. and Rüsch, N. , journal =. Mental illness stigma, secrecy and suicidal ideation , volume =. 2017 , doi =

    2017

  22. [22]

    Stigma and its impact on disclosure and mental health secrecy in young people with clinical depression symptoms: A qualitative analysis , volume =

    Prizeman, Katie and McCabe, Ciara and Weinstein, Netta , journal =. Stigma and its impact on disclosure and mental health secrecy in young people with clinical depression symptoms: A qualitative analysis , volume =. 2024 , abstract =

    2024

  23. [23]

    Self-stigma in patients with schizophrenia: Impact and management , volume =

    Ociskova, Marie and Prasko, Jan and Holubova, Michaela and Latalova, Klara and Sollár, Tomáš and Zatkova, Marta and Slepecky, Milos and Bocek, Jonas , journal =. Self-stigma in patients with schizophrenia: Impact and management , volume =

  24. [24]

    and Wu, Mengyue , booktitle =

    Chen, Siyuan and Wang, Meilin and Lv, Minghao and Zhang, Zhiling and Ju, Qianqian and Dejiyangla and Peng, Yujia and Zhu, Kenny Q. and Wu, Mengyue , booktitle =. Mapping Long-term Causalities in Psychiatric Symptomatology and Life Events from Social Media , year =. doi:10.18653/v1/2024.naacl-long.306 , url =

    work page doi:10.18653/v1/2024.naacl-long.306 2024

  25. [25]

    The Holmes–Rahe Stress Inventory , volume =

    Noone, Peter A , journal =. The Holmes–Rahe Stress Inventory , volume =. 2017 , doi =

    2017

  26. [26]

    Ruengorn, Chidchanok and Awiphan, Ratanaporn and Wongpakaran, Nahathai and Wongpakaran, Tinakon and Nochaiwong, Surapon and. Association of job loss, income loss, and financial burden with adverse mental health outcomes during coronavirus disease 2019 pandemic in Thailand: A nationwide cross-sectional study , volume =. Depression and Anxiety , number =. 2...

    work page doi:10.1002/da.23155 2019

  27. [27]

    Symptom Identification for Interpretable Detection of Multiple Mental Disorders on Social Media , year =

    Zhang, Zhiling and Chen, Siyuan and Wu, Mengyue and Zhu, Kenny , booktitle =. Symptom Identification for Interpretable Detection of Multiple Mental Disorders on Social Media , year =. doi:10.18653/v1/2022.emnlp-main.677 , url =

    work page doi:10.18653/v1/2022.emnlp-main.677 2022

  28. [28]

    Aaron Grattafiori and Abhimanyu Dubey and Abhinav Jauhri and Abhinav Pandey and Abhishek Kadian and Ahmad Al-Dahle and Aiesha Letman and Akhil Mathur and Alan Schelten and Alex Vaughan and Amy Yang and Angela Fan and Anirudh Goyal and Anthony Hartshorn and Aobo Yang and Archi Mitra and Archie Sravankumar and Artem Korenev and Arthur Hinsvark and Arun Rao ...

    2024

  29. [29]

    2025 , url =

    An Yang and Anfeng Li and Baosong Yang and Beichen Zhang and Binyuan Hui and Bo Zheng and Bowen Yu and Chang Gao and Chengen Huang and Chenxu Lv and Chujie Zheng and Dayiheng Liu and Fan Zhou and Fei Huang and Feng Hu and Hao Ge and Haoran Wei and Huan Lin and Jialong Tang and Jian Yang and Jianhong Tu and Jianwei Zhang and Jianxin Yang and Jiaxi Yang and...

    2025

  30. [30]

    Weinberger and Yoav Artzi , booktitle =

    Tianyi Zhang and Varsha Kishore and Felix Wu and Kilian Q. Weinberger and Yoav Artzi , booktitle =. BERTScore: Evaluating Text Generation with BERT , year =

  31. [31]

    Sentence-

    Reimers, Nils and Gurevych, Iryna , booktitle =. Sentence-. 2019 , doi =

    2019

  32. [32]

    Pennebaker, James W and Boyd, Ryan L and Jordan, Kayla and Blackburn, Kate , title =

  33. [33]

    Chinese LIWC Lexicon Expansion via Hierarchical Classification of Word Embeddings with Sememe Attention , volume =

    Zeng, Xiangkai and Yang, Cheng and Tu, Cunchao and Liu, Zhiyuan and Sun, Maosong , journal =. Chinese LIWC Lexicon Expansion via Hierarchical Classification of Word Embeddings with Sememe Attention , volume =. 2018 , doi =

    2018

  34. [34]

    Self-stigma of mental illness scale--short form: reliability and validity , volume =

    Corrigan, Patrick W and Michaels, Patrick J and Vega, Eduardo and Gause, Michael and Watson, Amy C and R. Self-stigma of mental illness scale--short form: reliability and validity , volume =. Psychiatry Res. , month = aug, number =. 2012 , abstract =

    2012

  35. [35]

    and Watson, Amy C

    Corrigan, Patrick W. and Watson, Amy C. and Barr, Leah , journal =. The Self–Stigma of Mental Illness: Implications for Self–Esteem and Self–Efficacy , volume =. 2006 , abstract =. doi:10.1521/jscp.2006.25.8.875 , url =

    work page doi:10.1521/jscp.2006.25.8.875 2006

  36. [36]

    Internalized stigma of mental illness: psychometric properties of a new measure , volume =

    Ritsher, Jennifer Boyd and Otilingam, Poorni G and Grajales, Monica , journal =. Internalized stigma of mental illness: psychometric properties of a new measure , volume =. 2003 , abstract =

    2003

  37. [37]

    Internalized Stigma of Mental Illness (

    Boyd, Jennifer E and Adler, Emerald P and Otilingam, Poorni G and Peters, Townley , journal =. Internalized Stigma of Mental Illness (. 2014 , abstract =

    2014

  38. [38]

    Lysaker, Paul H and Roe, David and Yanos, Philip T. Toward understanding the insight paradox: internalized stigma moderates the association between insight and social functioning, hope, and self-esteem among people with schizophrenia spectrum disorders. Schizophr. Bull

  39. [39]

    Internalized stigma predicts erosion of morale among psychiatric outpatients

    Ritsher, Jennifer Boyd and Phelan, Jo C. Internalized stigma predicts erosion of morale among psychiatric outpatients. Psychiatry Res