Cognitive Reframing of Negative Thoughts through Human-Language Model Interaction

Ashish SharmaKevin RushtonInna E. LinDavid WaddenKhendra G. LucasAdam S. MinerTheresa NguyenTim Althoff

article2023ACL139 citationsOutstanding Paper Award

Demonstrates how language models can assist in cognitive therapy by generating and controlling linguistic attributes of reframed thoughts, backed by a randomized field study of over 2,000 participants showing that users favor empathic and specific reframes over overly positive ones.

Listen

Cognitive reframing is an established psychological technique that helps individuals overcome distressing negative thoughts by replacing them with constructive, alternative perspectives. While highly effective, widespread access to this intervention is severely limited by clinician shortages, high financial costs, and social stigma. The article addresses whether language models can assist individuals by automatically generating relatable, helpful, and memorable reframed thoughts in real time.

The main objective of the article is to define a measurable framework of reframing attributes, develop an artificial intelligence method to generate and control these reframes, and evaluate which linguistic characteristics make reframed thoughts most effective for people experiencing negative thoughts.

To achieve this, the authors collaborated with clinical psychologists to establish a framework of seven linguistic attributes: addressing thinking traps, rationality, positivity, empathy, actionability, specificity, and readability. They created an expert-annotated dataset of 600 situations, thoughts, and reframes sourced from mental health practitioners. Using this data, they designed a retrieval-enhanced language model that generates reframes and modulates their specific attributes. They subsequently validated the approach through automated metrics, clinical expert ratings, and a month-long, randomized field study involving 2,067 consented participants on Mental Health America, a major national mental health platform.

The field study and expert evaluations revealed several key findings regarding what users value in cognitive reframing. Highly empathic reframes were preferred 55.7% more often than low-empathy reframes, and highly specific reframes were preferred 43.1% more often than generic ones. In contrast, reframes with high positivity were preferred 22.7% less often than those with lower positivity, indicating that overly optimistic messaging can alienate individuals facing emotional distress. Additionally, reframes grounded in rationality were 10.8% more relatable, while reframes that directly addressed thinking traps, offered actionable guidance, or provided situational specificity scored significantly higher in helpfulness and long-term memorability. In technical benchmarks, the retrieval-enhanced approach outperformed standard baseline models in both linguistic overlap metrics and practitioner ratings for helpfulness and relatability.

These findings suggest that artificial intelligence can provide valuable, scalable scaffolding for in-the-moment cognitive support if properly guided. Crucially, the results show that simply maximizing positive sentiment is counterproductive; effective reframing relies instead on empathy, logical soundness, and actionable steps. System safety proved high, with only 0.56% of suggestions flagged by users, none of which involved harmful or unsafe content.

Organizations developing digital mental health tools should design generative systems that prioritize empathy, specificity, and cognitive distortion resolution over superficial optimism. Before deploying these systems widely or across clinical workflows, stakeholders must conduct further research across diverse demographic groups, evaluate non-English languages, and measure longitudinal psychological outcomes beyond single-session interactions. High confidence in these findings is supported by the randomized field trial with over 2,000 real-world users, though caution is warranted regarding long-term clinical efficacy and generalizability outside English-speaking web audiences.

Allen Institute for AI Mental Health America Stanford University University of Washington n-gram language models statistical language models cognitive reframing Human-LM interaction cognitive distortions

arXiv: 2305.02466 behavioral-data/Cognitive-Reframing

No sufficiently relevant recommendations were found.

Cover for Cognitive Reframing of Negative Thoughts through Human-Language Model Interaction

Abstract

A proven therapeutic technique to overcome negative thoughts is to replace them with a more hopeful “reframed thought.” Although therapy can help people practice and learn this Cognitive Reframing of Negative Thoughts, clinician shortages and mental health stigma commonly limit people’s access to therapy. In this paper, we conduct a human-centered study of how language models may assist people in reframing negative thoughts. Based on psychology literature, we define a framework of seven linguistic attributes that can be used to reframe a thought. We develop automated metrics to measure these attributes and validate them with expert judgements from mental health practitioners. We collect a dataset of 600 situations, thoughts and reframes from practitioners and use it to train a retrieval-enhanced in-context learning model that effectively generates reframed thoughts and controls their linguistic attributes. To investigate what constitutes a “high-quality” reframe, we conduct an IRB-approved randomized field study on a large mental health website with over 2,000 participants. Amongst other findings, we show that people prefer highly empathic or specific reframes, as opposed to reframes that are overly positive. Our findings provide key implications for the use of LMs to assist people in overcoming negative thoughts.

Topics

n-gram language models statistical language models

Keywords

cognitive reframing Human-LM interaction cognitive distortions mental health conversations

1 Introduction
2 Problem Definition and Goals
3 Framework of Linguistic Attributes of Reframed Thoughts
4 Data Collection
4.1 Curated Situations & Negative Thoughts
4.2 Annotation Task and Procedure
4.3 Ethics and Safety
5 Method
5.1 Measuring Reframing Attributes
5.2 Reframe Generation
5.3 Controlling Linguistic Attributes of Generated Reframes
6 Experiments and Results
6.1 Construct Validity of Linguistic Attributes
6.2 Reframe Generation Performance
7 Randomized Field Study on a Large Mental Health Platform
7.1 Model Deployment
7.2 What types of reframed thoughts do people prefer?
7.3 How do the linguistic attributes of reframed thoughts relate to the desired outcomes of cognitive reframing?
8 Related Work
9 Conclusion
10 Ethics Statement
11 Limitations
Acknowledgements
References
A Method
A.1 Linguistic Attributes of Reframed Thoughts
A.3 Hyperparameter Choices for our Proposed Retrieval-Enhanced In-Context Learning Method
B Reproducibility
C Flagged Reframes
D List of Thinking Traps
E Example Illustrating Our Rationality Measurement
F Randomized Field-Study Interface
G Data Collection Instructions
H Consent Form Used in the Randomized Field Study on MHA

Knowls

Knowl 1 — Seven linguistic attributes define the reframing framework
definition
Cognitive reframing takes a situation $S$ , a negative thought $T$ associated with that situation, and produces an alternative thought $R$ . The framework characterizes reframes along seven linguistic attributes: (1) thinking-trap addressing—whether the reframe challenges at least one distorted pattern in the negative thought, such as fortune telling or all-or-nothing thinking; (2) rationality—whether its reasoning is supported by sound evidence rather than unrealistic assumptions; (3) positivity—how strongly it emphasizes positive perspectives, with balanced reframes distinguished from exaggeratedly positive ones; (4) empathy—whether it acknowledges or validates feelings associated with the thought; (5) actionability—whether it suggests a concrete action or offers a perspective that is easy to act on; (6) specificity—how directly it addresses the particular situation and thought rather than applying generically; and (7) readability—how linguistically simple or complex it is to read. The framework treats these as distinct properties: for example, a reframe can be positive without being rational, or specific without suggesting an action.
Knowl 2 — Expert-annotated dataset of situations, thoughts, and reframes
data/table
The authors created a dataset of 600 expert-written reframes for 300 situation–thought pairs. The pairs comprise 180 manually curated examples from the Thought Records Dataset and 120 manually curated self-reports collected from Mental Health America (MHA) visitors. Fifteen mental-health practitioners and clinical-psychology graduate students with practical reframing experience each wrote two different reframes for a presented pair, labeled the thinking traps addressed by each, and compared the two reframes on rationality, positivity, actionability, empathy, specificity, and readability. The result is a resource pairing situations and negative thoughts with expert reframes and annotations, used to develop and evaluate the generation and attribute-measurement methods.
Knowl 3 — Retrieval-enhanced in-context generation of reframes
model/method
For a new situation $S$ and negative thought $T$ , the generation method encodes the situation and thought with RoBERTa embeddings, concatenates the representations of $S$ and $T$ , and retrieves the five most similar situation–thought pairs from the expert dataset by cosine similarity. It places the retrieved examples and their reframes in a prompt to GPT-3, specifically text-davinci-003, to generate a reframe for the new input. The authors selected five retrieved examples after qualitative assessment of 100 manually written inputs, comparing retrieval counts of 1, 5, 10, and 20. The method was designed to supply relevant examples for varied situations and thoughts, addressing failures observed when prompting with a fixed set of examples that did not cover the input.
Knowl 4 — Prompt-based control of reframing attributes
model/method
The controllable generation method starts from a reframe $R$ and generates alternatives that target a chosen attribute. For thinking-trap addressing, it prompts GPT-3 with separate examples whose reframes address at least one trap and examples whose reframes do not, producing one version in each category; if the original thought has multiple traps, addressing any one counts as addressing a trap. For each of the other six attributes—rationality, positivity, empathy, actionability, specificity, and readability—it generates a higher-scoring and a lower-scoring alternative using five expert-annotated contrastive examples. To make a higher-attribute version, the prompt examples map a lower-rated expert reframe to a higher-rated one; for a lower-attribute version, the mapping is reversed. The procedure is intended to vary the requested attribute while retaining the original situation and thought as context.
Knowl 5 — Automated operationalizations of the seven attributes
model/method
The paper operationalizes each reframing attribute as follows. Thinking-trap addressing is a multi-label classification task: GPT-3 is fine-tuned on expert labels to identify the traps addressed by a reframe, given its situation and negative thought. Rationality is estimated by prompting GPT-3 to generate explanations supporting and refuting a reframe, estimating the probabilities that the reframe is labeled sound or flawed from next-token probabilities, and recursively assessing the strength of the generated explanations; the reported implementation uses a recursion depth of three and generates three explanations for each side at each step. Positivity is scored with a RoBERTa sentiment classifier fine-tuned on TweetEval. Empathy is scored from 0 to 6 with a RoBERTa empathy model further fine-tuned on 300 manually labeled reframes. Actionability combines a few-shot GPT-3 binary prediction of whether the reframe contains a concrete action with the average pairwise cosine similarity among embeddings of five GPT-3-generated possible next actions; the latter is treated as action coherence, and the two quantities are added. Specificity is the RoBERTa sentence-embedding similarity between the reframe and the concatenation of its situation and negative thought. Readability is measured with the Coleman–Liau Index, $0.0588L - 0.296S - 15.8$ , where $L$ is the average number of letters per 100 words and $S$ is the average number of sentences per 100 words.
Knowl 6 — Attribute metrics correlate with practitioners’ judgments
empirical result
The authors assessed construct validity by correlating each automated attribute measure with mental-health experts’ judgments. Pearson correlations were 0.680 for thinking-trap addressing, 0.448 for rationality, 0.550 for positivity, 0.575 for empathy, 0.647 for actionability, 0.427 for specificity, and 0.331 for readability. The paper reports significance markers of * for $p<0.05$ , ** for $p<0.001$ , and *** for $p<10^{-5}$ : trap addressing, positivity, empathy, and actionability received ***, rationality and specificity received **, and readability received *. Thus, the measures aligned most strongly with expert judgments for trap addressing and actionability and least strongly for readability; the correlations provide validation evidence, not proof that the metrics perfectly capture the constructs.
Knowl 7 — Retrieval-enhanced GPT-3 leads on automatic and expert evaluations
empirical result
The authors split the 600 expert-annotated examples 70:30 into training and test sets and used top- $p$ sampling with $p=0.6$ . They compared retrieval-enhanced GPT-3 with retrieval-only output, a positive-reframing model, DialoGPT, T5, and GPT-3 prompted with five randomly selected training examples. Automatic metrics were BLEU, ROUGE-1, ROUGE-L, and BERTScore; three mental-health practitioners also rated test outputs for relatability and helpfulness on 1–5 scales. In the order BLEU / ROUGE-1 / ROUGE-L / BERTScore / relatability / helpfulness, scores were: Retrieval Only, 21.6 / 18.8 / 14.2 / 86.7 / 2.58 / 3.14; Positive Reframing, 24.4 / 23.6 / 17.6 / 87.6 / 2.67 / 2.40; DialoGPT, 22.5 / 17.4 / 13.5 / 86.3 / 2.49 / 3.21; T5, 24.9 / 23.4 / 17.8 / 87.2 / 2.51 / 3.30; GPT-3 Only, 25.0 / 23.9 / 18.0 / 88.3 / 2.97 / 3.98; and the retrieval-enhanced model, 27.8 / 26.0 / 19.9 / 88.6 / 3.10 / 4.11. The proposed model scored highest on all six reported measures in this comparison. The positive-reframing baseline had the lowest helpfulness score.
Knowl 8 — MHA field study tested attribute preferences and immediate ratings
experimental setup
The authors deployed the model for one month on Mental Health America and enrolled 2,067 adult visitors who consented to participate. The interface asked participants to enter a negative thought and a related situation, showed possible thinking traps for the participant to select, and then presented AI-generated reframes. In the attribute-preference task, participants saw multiple reframes whose values varied for a randomly selected attribute and selected the one they found most relatable, helpful, and memorable. In a separate rating task, participants saw one generated reframe, without a choice among alternatives, and rated its relatability, helpfulness, and memorability on 1–5 scales to avoid selection effects. The study had IRB approval and informed consent; responses matching a list of 50 self-harm or suicidal-ideation expressions were filtered, and crisis resources were provided. The study measured preferences and self-reported immediate judgments, not clinical outcomes or long-term effects.
Knowl 9 — Participants favored empathic and specific reframes over highly positive ones
empirical result
In the field study’s comparison of alternative reframes, highly empathic versions were selected more often than lower-empathy versions: 39.7% versus 25.5%, a reported 55.7% relative increase ( $p<10^{-5}$ ). Highly specific versions were selected more often than lower-specificity versions: 39.2% versus 27.4%, a reported 43.1% relative increase ( $p<10^{-5}$ ). The direction differed for positivity: high-positivity reframes were selected less often than lower-positivity reframes, 29.6% versus 38.3%, a reported 22.7% relative decrease ( $p<10^{-5}$ ). Participants also favored medium-readability reframes over the very simple or very complex options. These results describe selections among generated alternatives, not clinical effectiveness.
Knowl 10 — Different reframe attributes align with different desired outcomes
empirical result
For the single-reframe rating task, the authors compared the first and fourth quartiles of each automated attribute score and measured participant ratings on a 1–5 scale. Higher-rationality reframes had higher mean relatability ratings than lower-rationality reframes: 3.91 versus 3.53, a reported 10.8% increase ( $p<0.05$ ). For helpfulness, reframes that addressed thinking traps scored 3.39 versus 3.19 when they did not, a reported 6.3% increase ( $p<0.01$ ); higher-actionability reframes scored 3.41 versus 3.20, a 6.6% increase ( $p<0.05$ ); and higher-specificity reframes scored 3.42 versus 3.12, a 9.6% increase ( $p<0.01$ ). For memorability, higher-actionability reframes scored 3.67 versus 3.40, a 7.9% increase ( $p<0.01$ ), and higher-specificity reframes scored 3.70 versus 3.48, a 6.3% increase ( $p<0.05$ ). These are associations between measured attributes and self-reported ratings, not evidence that changing an attribute causes a better outcome.

Coverage note — The paper’s broader scope limitations are included in the field-study knowl; the appendix’s full thinking-trap inventory and the detailed breakdown of flagged suggestions are omitted as secondary reference and audit material rather than core method or outcome findings.

References

1.Ashley Batts Allen and Mark R Leary. 2010. Self-compassion, stress, and coping. Social and personality psychology compass.
2.Tim Althoff, Kevin Clark, and Jure Leskovec. 2016. Large-scale analysis of counseling conversations: An application of natural language processing to mental health. Transactions of the Association for Computational Linguistics.
3.Francesco Barbieri, Jose Camacho-Collados, Luis Espinosa Anke, and Leonardo Neves. 2020. TweetEval: Unified benchmark and comparative evaluation for tweet classification. In EMNLP Findings.
4.Aaron T Beck. 1976. Cognitive therapy and the emotional disorders. International Universities Press.
5.Judith S Beck. 2005. Cognitive therapy for challenging problems: What to do when the basics don’t work. Guilford Press.
6.Chandra Bhagavatula, Ronan Le Bras, Chaitanya Malaviya, Keisuke Sakaguchi, Ari Holtzman, Hannah Rashkin, Doug Downey, Wen-tau Yih, and Yejin Choi. 2020. Abductive commonsense reasoning. In ICLR.
7.Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared J Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, et al. 2020. Language models are few-shot learners. NeurIPS.
8.Franziska Burger, Mark A Neerincx, and Willem-Paul Brinkman. 2021. Natural language processing for cognitive therapy: Extracting schemas from thought records. PloS one.
9.Hannah A Burkhardt, George S Alexopoulos, Michael D Pullmann, Thomas D Hull, Patricia A Areán, and Trevor Cohen. 2021. Behavioral activation and depression symptomatology: longitudinal assessment of linguistic indicators in text-based therapy sessions. JMIR.
10.David D Burns. 1980. Feeling good: Thenew mood therapy. New York.
11.Meri Coleman and Ta Lin Liau. 1975. A computer readability formula designed for machine scoring. Journal of Applied Psychology.
12.Peter Damielson, Robert Audi, Cristina Bicchieri, et al. 2004. The Oxford handbook of rationality. Oxford University Press, USA.
13.Daniel David, Steven Jay Lynn, and Albert Ellis. 2009. Rational and irrational beliefs: Research, theory, and clinical practice. Oxford University Press.
14.William N Dember and Larry Penwell. 1980. Happiness, depression, and the pollyanna principle. Bulletin of the Psychonomic Society.
15.Sona Dimidjian, Manuel Barrera Jr, Christopher Martell, Ricardo F Muñoz, and Peter M Lewinsohn. 2011. The origins and current status of behavioral activation treatments for depression. Annual review of clinical psychology.
16.Xiruo Ding, Kevin Lybarger, Justin Tauscher, and Trevor Cohen. 2022. Improving classification of infrequent cognitive distortions: Domain-specific model vs. data augmentation. In Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies: Student Research Workshop.
17.Robert Elliott, Arthur C Bohart, Jeanne C Watson, and Leslie S Greenberg. 2011. Empathy. Psychotherapy.
18.Manas Gaur, Amanuel Alambo, Joy Prakash Sain, Ugur Kursuncu, Krishnaprasad Thirunarayan, Ramakanth Kavuluru, Amit Sheth, Randy Welton, and Jyotishman Pathak. 2019. Knowledge-aware assessment of severity of suicide risk for early intervention. In WWW.
19.Ari Holtzman, Jan Buys, Li Du, Maxwell Forbes, and Yejin Choi. 2020. The curious case of neural text degeneration. In ICLR.
20.Kokil Jaidka, Niyati Chhaya, Saran Mumick, Matthew Killingsworth, Alon Halevy, and Lyle Ungar. 2020. Beyond positive emotion: Deconstructing happy moments based on writing prompts. In ICWSM.
21.Jaehun Jung, Lianhui Qin, Sean Welleck, Faeze Brahman, Chandra Bhagavatula, Ronan Le Bras, and Yejin Choi. 2022. Maieutic prompting: Logically consistent reasoning with recursive explanations. In EMNLP.
22.Carole A Kaplan, Anne E Thompson, and Sheila M Searson. 1995. Cognitive behaviour therapy in children and adolescents. Archives of disease in childhood.
23.Allison Lahnala, Yuntian Zhao, Charles Welch, Jonathan K Kummerfeld, Lawrence C An, Kenneth Resnicow, Rada Mihalcea, and Verónica Pérez-Rosas. 2021. Exploring self-identified counseling expertise in online support forums. In ACL-IJCNLP Findings.
24.Fei-Tzin Lee, Derrick Hull, Jacob Levine, Bonnie Ray, and Kathleen McKeown. 2019. Identifying therapist conversational actions across diverse psychotherapeutic approaches. In Proceedings of the Sixth Workshop on Computational Linguistics and Clinical Psychology.
25.Rensis Likert. 1932. A technique for the measurement of attitudes. Archives of psychology.
26.Chin-Yew Lin. 2004. Rouge: A package for automatic evaluation of summaries. In Text summarization branches out.
27.Inna Lin, Lucille Njoo, Anjalie Field, Ashish Sharma, Katharina Reinecke, Tim Althoff, and Yulia Tsvetkov. 2022. Gendered mental health stigma in masked language models. In EMNLP.
28.Alisa Liu, Swabha Swayamdipta, Noah A Smith, and Yejin Choi. 2022a. Wanli: Worker and ai collaboration for natural language inference dataset creation. arXiv preprint arXiv:2201.05955.
29.Jiachang Liu, Dinghan Shen, Yizhe Zhang, William B Dolan, Lawrence Carin, and Weizhu Chen. 2022b. What makes good in-context examples for gpt-3? In Proceedings of Deep Learning Inside Out (DeeLIO 2022): The 3rd Workshop on Knowledge Extraction and Integration for Deep Learning Architectures.
30.Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, and Veselin Stoyanov. 2019. Roberta: A robustly optimized bert pretraining approach. arXiv preprint arXiv:1907.11692.
31.Aman Madaan, Amrith Setlur, Tanmay Parekh, Barnabás Poczós, Graham Neubig, Yiming Yang, Ruslan Salakhutdinov, Alan W Black, and Shrimai Prabhumoye. 2020. Politeness transfer: A tag and generate approach. In ACL.
32.Tara Matthews, Kathleen O’Leary, Anna Turner, Manya Sleeper, Jill Palzkill Woelfer, Martin Shelton, Cori Manthorne, Elizabeth F Churchill, and Sunny Consolvo. 2017. Stories from survivors: Privacy & security practices when coping with intimate partner abuse. In CHI.
33.Adam S Miner, Nigam Shah, Kim D Bullock, Bruce A Arnow, Jeremy Bailenson, and Jeff Hancock. 2019. Key considerations for incorporating conversational ai in psychotherapy. Frontiers in psychiatry.
34.Robert R Morris, Stephen M Schueller, and Rosalind W Picard. 2015. Efficacy of a web-based, crowdsourced peer-to-peer cognitive reappraisal platform for depression: randomized controlled trial. JMIR.
35.Usman Naseem, Adam G Dunn, Jinman Kim, and Matloob Khushi. 2022. Early identification of depression severity levels on reddit using ordinal classification. In WWW.
36.Mark Olfson. 2016. Building the mental health workforce capacity needed to treat adults with serious mental illnesses. Health Affairs.
37.Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu. 2002. Bleu: a method for automatic evaluation of machine translation. In ACL.
38.Charles Sanders Peirce. 1974. Collected papers of charles sanders peirce. Harvard University Press.
39.Sachin R Pendse, Kate Niederhoffer, and Amit Sharma. 2019. Cross-cultural differences in the use of online mental health support forums. CSCW.
40.Verónica Pérez-Rosas, Kenneth Resnicow, Rada Mihalcea, et al. 2022. Pair: Prompt-aware margin ranking for counselor reflection scoring in motivational interviewing. In EMNLP.
41.Verónica Pérez-Rosas, Xinyi Wu, Kenneth Resnicow, and Rada Mihalcea. 2019. What makes a good counselor? learning to distinguish between high-quality and low-quality counseling conversations. In ACL.
42.Yada Pruksachatkun, Sachin R Pendse, and Amit Sharma. 2019. Moments of change: Analyzing peer-based cognitive support in online mental health forums. In CHI.
43.Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, Peter J Liu, et al. 2020. Exploring the limits of transfer learning with a unified text-to-text transformer. JMLR.
44.Emily Reif, Daphne Ippolito, Ann Yuan, Andy Coenen, Chris Callison-Burch, and Jason Wei. 2022. A recipe for arbitrary text style transfer with large language models. In ACL.
45.Koustuv Saha and Amit Sharma. 2020. Causal factors of effective psychosocial outcomes in online mental health communities. In ICWSM.
46.Jessica L Schleider, Michael C Mullarkey, Kathryn R Fox, Mallory L Dobias, Akash Shroff, Erica A Hart, and Chantelle A Roulston. 2022. A randomized trial of online single-session interventions for adolescent depression during covid-19. Nature Human Behaviour.
47.Raj Sanjay Shah, Faye Holt, Shirley Anugrah Hayati, Aastha Agarwal, Yi-Chia Wang, Robert E Kraut, and Diyi Yang. 2022. Modeling motivational interviewing strategies on an online peer-to-peer counseling platform. CSCW.
48.Ashish Sharma, Monojit Choudhury, Tim Althoff, and Amit Sharma. 2020a. Engagement patterns of peer-to-peer interactions on mental health platforms. In ICWSM.
49.Ashish Sharma, Inna W Lin, Adam S Miner, David C Atkins, and Tim Althoff. 2021. Towards facilitating empathic conversations in online mental health support: A reinforcement learning approach. In WWW.
50.Ashish Sharma, Inna W. Lin, Adam S. Miner, David C. Atkins, and Tim Althoff. 2023. Human–AI collaboration enables more empathic conversations in text-based peer-to-peer mental health support. Nature Machine Intelligence.
51.Ashish Sharma, Adam S Miner, David C Atkins, and Tim Althoff. 2020b. A computational approach to understanding empathy expressed in text-based mental health support. In EMNLP.
52.Eva Sharma and Munmun De Choudhury. 2018. Mental health support and its relationship to linguistic accommodation in online communities. In CHI.
53.Siqi Shen, Verónica Pérez-Rosas, Charles Welch, Soujanya Poria, and Rada Mihalcea. 2022. Knowledge enhanced reflection generation for counseling dialogues. In ACL.
54.Siqi Shen, Charles Welch, Rada Mihalcea, and Verónica Pérez-Rosas. 2020. Counseling-style reflection generation using generative pretrained transformers with augmented context. In SIGDIAL.
55.Amy E Sickel, Jason D Seacat, and Nina A Nabors. 2014. Mental health stigma update: A review of consequences. Advances in Mental Health.
56.C Estelle Smith, William Lane, Hannah Miller Hillberg, Daniel Kluver, Loren Terveen, and Svetlana Yarosh. 2021. Effective strategies for crowd-powered cognitive reappraisal systems: A field deployment of the flip* doubt web application for mental health. CSCW.
57.Ian Stewart, Charles Welch, Lawrence An, Kenneth Resnicow, James Pennebaker, and Rada Mihalcea. 2023. Expressive interviewing agents to support health-related behavior change: A study of covid-19 behaviors. JMIR formative research.
58.Michael J Tanana, Christina S Soma, Vivek Srikumar, David C Atkins, and Zac E Imel. 2019. Development and evaluation of clientbot: Patient-like conversational agent to train basic counseling skills. JMIR.
59.David Wadden, Tal August, Qisheng Li, and Tim Althoff. 2021. The effect of moderation on online mental health conversations. In ICWSM.
60.Charles Welch, Allison Lahnala, Verónica Pérez-Rosas, Siqi Shen, Sarah Seraj, Larry An, Kenneth Resnicow, James Pennebaker, and Rada Mihalcea. 2020. Expressive interviewing: A conversational system for coping with covid-19. In Proceedings of the 1st Workshop on NLP for COVID-19 (Part 2) at EMNLP 2020.
61.Xinnuo Xu, Ondřej Dušek, Ioannis Konstas, and Verena Rieser. 2018. Better conversations by modeling, filtering, and optimizing for coherence and diversity. In ACL.
62.Diyi Yang, Zheng Yao, Joseph Seering, and Robert Kraut. 2019. The channel matters: Self-disclosure, reciprocity and social support in online cancer support groups. In CHI.
63.Justine Zhang and Cristian Danescu-Niculescu-Mizil. 2020. Balancing objectives in counseling conversations: Advancing forwards or looking backwards. In ACL.
64.Justine Zhang, Robert Filbin, Christine Morrison, Jaclyn Weiser, and Cristian Danescu-Niculescu-Mizil. 2019a. Finding your voice: The linguistic development of mental health counselors. In ACL.
65.Tianyi Zhang, Varsha Kishore, Felix Wu, Kilian Q Weinberger, and Yoav Artzi. 2019b. Bertscore: Evaluating text generation with bert. arXiv preprint arXiv:1904.09675.
66.Yizhe Zhang, Siqi Sun, Michel Galley, Yen-Chun Chen, Chris Brockett, Xiang Gao, Jianfeng Gao, Jingjing Liu, and Bill Dolan. 2020. Dialogpt: Large-scale generative pre-training for conversational response generation. In ACL, system demonstration.
67.Caleb Ziems, Minzhi Li, Anthony Zhang, and Diyi Yang. 2022. Inducing positive perspectives with text reframing. In ACL.

Citation

MLA

Sharma, A., et al. “Cognitive Reframing of Negative Thoughts Through Human-Language Model Interaction”. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), 2023, pp. 9977–10000, https://doi.org/10.18653/v1/2023.acl-long.555.

APA

Sharma, A., Rushton, K., Lin, I., Wadden, D., Lucas, K., Miner, A., Nguyen, T., & Althoff, T. (2023). Cognitive Reframing of Negative Thoughts through Human-Language Model Interaction. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), 9977–10000. https://doi.org/10.18653/v1/2023.acl-long.555

Chicago

Sharma, A., K. Rushton, I. Lin, et al. 2023. “Cognitive Reframing of Negative Thoughts Through Human-Language Model Interaction”. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), 9977–10000. https://doi.org/10.18653/v1/2023.acl-long.555.

Harvard

Sharma, A. et al. (2023) “Cognitive Reframing of Negative Thoughts through Human-Language Model Interaction”, Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). Association for Computational Linguistics, pp. 9977–10000. Available at: https://doi.org/10.18653/v1/2023.acl-long.555.

Vancouver

1. Sharma A, Rushton K, Lin I, Wadden D, Lucas K, Miner A, Nguyen T, Althoff T (2023) Cognitive Reframing of Negative Thoughts through Human-Language Model Interaction. In: Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). Association for Computational Linguistics, pp 9977–10000

BibTeX

@inproceedings{sharma-etal-2023-cognitive,
    title = "Cognitive Reframing of Negative Thoughts through Human-Language Model Interaction",
    author = "Sharma, Ashish  and
      Rushton, Kevin  and
      Lin, Inna  and
      Wadden, David  and
      Lucas, Khendra  and
      Miner, Adam  and
      Nguyen, Theresa  and
      Althoff, Tim",
    editor = "Rogers, Anna  and
      Boyd-Graber, Jordan  and
      Okazaki, Naoaki",
    booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)",
    month = jul,
    year = "2023",
    address = "Toronto, Canada",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2023.acl-long.555/",
    doi = "10.18653/v1/2023.acl-long.555",
    pages = "9977--10000"
}

Metadata:ACL Anthology

Source Code

This paper has an official code repository available. Click below to access the source code.

View Repository

Access the Paper

This paper is available from its original source. Click below to access the PDF.

Open PDF

License: https://creativecommons.org/licenses/by/4.0/

Cognitive Reframing of Negative Thoughts through Human-Language Model Interaction

Abstract

Topics

Keywords

Table of Contents

Knowls

Knowl 1 — Seven linguistic attributes define the reframing framework

Knowl 2 — Expert-annotated dataset of situations, thoughts, and reframes

Knowl 3 — Retrieval-enhanced in-context generation of reframes

Knowl 4 — Prompt-based control of reframing attributes

Knowl 5 — Automated operationalizations of the seven attributes

Knowl 6 — Attribute metrics correlate with practitioners’ judgments

Knowl 7 — Retrieval-enhanced GPT-3 leads on automatic and expert evaluations

Knowl 8 — MHA field study tested attribute preferences and immediate ratings

Knowl 9 — Participants favored empathic and specific reframes over highly positive ones

Knowl 10 — Different reframe attributes align with different desired outcomes

References

Citation

BibTeX

Source Code

Access the Paper