Cognitive Reframing of Negative Thoughts with Iterative In-context Clinical Grounding Feedback

ACL ARR 2026 May Submission13952 Authors

Keywords: Cognitive reframing, LLM agents, computational psychotherapy, computational psycholinguistics, in-context learning, human-AI interaction/cooperation

Abstract:

This paper introduces a new framework for cognitive reframing of negative thoughts. Unlike prior methods that consider reframing as a
shallow generation task, the framework guides LLMs to follow Cognitive Behavioral Therapy (CBT). To do that, we first design a multi-
agent backbone that integrates CBT for clinical grounding. We next introduce an objective function that assesses psychological, semantic, and stylistic aspects. In each iteration, the function’s scores are converted into prompt edits that steer the next iteration. Zero-shot evaluation on two datasets shows that the framework outperforms shallow- and single-agent baselines on automatic and human-rated metrics, including those for safety and hallucination.

Paper Type: Long

Research Area: Linguistic theories, Cognitive Modeling and Psycholinguistics

Research Area Keywords: LLM agents, computational psycholinguistics, cognitive modeling, human-AI interaction/cooperation, prompting, healthcare applications, clinical NLP

Contribution Types: NLP engineering experiment, Publicly available software and/or pre-trained models

Languages Studied: English

EMNLP 2026 AI Reviewing Experiment: yes

Submission Number: 13952