arXiv:2511.11884v1 Announce Type: new 
Abstract: Mental health illness represents a substantial global socioeconomic burden, with COVID-19 further exacerbating accessibility challenges and driving increased demand for telehealth mental health support. While large language models (LLMs) offer promising solutions through 24/7 availability and non-judgmental interactions, pre-trained models often lack the contextual and emotional awareness necessary for appropriate therapeutic responses. This paper investigated the application of supervised fine-tuning (SFT) and reinforcement learning (RL) techniques to enhance GPT-2's capacity for therapeutic dialogue generation. The methodology restructured input formats to enable simultaneous processing of contextual information and emotional states alongside user input, employing a multi-component reward function that aligned model outputs with professional therapist responses and annotated emotions. Results demonstrated improvements through reinforcement learning over baseline GPT-2 across multiple evaluation metrics: BLEU (0.0111), ROUGE-1 (0.1397), ROUGE-2 (0.0213), ROUGE-L (0.1317), and METEOR (0.0581). LLM evaluation confirmed high contextual relevance and professionalism, while reinforcement learning achieved 99.34% emotion accuracy compared to 66.96% for baseline GPT-2. These findings demonstrate reinforcement learning's effectiveness in developing therapeutic dialogue systems that can serve as valuable assistive tools for therapists while maintaining essential human clinical oversight.

تمثل مشاكل الصحة النفسية عبئًا اجتماعيًا واقتصاديًا كبيرًا على مستوى العالم، وقد زادت جائحة كوفيد-19 من الطلب على خدمات الصحة النفسية عن بُعد. على الرغم من أن النماذج اللغوية الكبيرة مثل GPT-2 تقدم حلولًا واعدة من خلال توافرها على مدار الساعة وتفاعلاتها غير الحكمية، إلا أن النماذج المدربة مسبقًا غالبًا ما تفتقر إلى الوعي السياقي والعاطفي اللازم للاستجابات العلاجية المناسبة. تبحث هذه الورقة في تطبيق تقنيات الضبط الدقيق الخاضع للإشراف والتعلم المعزز لتعزيز قدرة GPT-2 على توليد حوارات علاجية.

Los problemas de salud mental representan una carga socioeconómica global significativa, agravada por la COVID-19, que ha aumentado la demanda de servicios de telemedicina. Aunque los grandes modelos de lenguaje (LLM) como GPT-2 ofrecen soluciones potenciales a través de su disponibilidad constante y sus interacciones sin juicios, a menudo carecen de la conciencia contextual y emocional necesaria para un diálogo terapéutico efectivo. Este estudio investiga el uso de técnicas de ajuste fino supervisado y aprendizaje por refuerzo para mejorar la capacidad de GPT-2 para generar diálogos terapéuti…

Les problèmes de santé mentale représentent un défi socio-économique mondial important, aggravé par la COVID-19, qui a augmenté la demande de services de télésanté. Bien que les grands modèles de langage (LLM) comme GPT-2 offrent des solutions potentielles grâce à une disponibilité constante et des interactions non-jugeantes, ils manquent souvent de la conscience contextuelle et émotionnelle nécessaire pour un dialogue thérapeutique efficace. Cette étude explore l'utilisation de l'affinage supervisé et de l'apprentissage par renforcement pour améliorer la capacité de GPT-2 à générer des conver…

Mental health issues pose a significant global socioeconomic challenge, worsened by COVID-19, which has increased the demand for telehealth services. While large language models (LLMs) like GPT-2 provide potential solutions through constant availability and non-judgmental interactions, they often lack the necessary contextual and emotional awareness for effective therapeutic dialogue. This study explores the use of supervised fine-tuning and reinforcement learning to improve GPT-2's ability to generate therapeutic conversations by processing contextual information and emotional states simultan…

Context-Emotion Aware Therapeutic Dialogue Generation: A Multi-component Reinforcement Learning Approach to Language Models for Mental Health Support

arXiv:2512.25026v2 Announce Type: replace 
Abstract: Transformer language models can generate strikingly natural text by modeling language as a sequence of tokens, but by relying primarily on surface-level co-occurrence statistics they fail to form globally consistent latent representations of entities and events, which contributes to poor relational generalization (the reversal curse), contextualization errors, and data inefficiency. Cognitive science, by contrast, shows that human comprehension converts linguistic input into compact, event-like representations that persist in memory while verbatim form is short-lived. Motivated by these findings, we introduce the Thought Gestalt (TG) model, a recurrent transformer that models language at two levels of abstraction: tokens and sentence-level "thought" states. TG generates one sentence at a time while cross-attending to a working memory of prior sentence representations. Token and sentence representations are generated using a shared stack of transformer blocks and trained with a single objective, next-token prediction loss. By retaining the computation graph of sentence representations written to working memory, gradients from future token losses flow backward through cross-attention to optimize the parameters that generate earlier sentence vectors. In scaling experiments, TG consistently improves data and parameter efficiency compared to matched GPT-2 runs and other baselines, with scaling fits indicating GPT-2 requires ~5-8% more data and ~33-42% more parameters to match TG's test loss. TG also reduces errors in relational-direction generalization on a father-son reversal curse probe.

أدت التطورات الأخيرة في نماذج اللغة القائمة على المحولات إلى تقديم نموذج Thought Gestalt (TG)، الذي يهدف إلى تحسين توليد النصوص الطبيعية من خلال نمذجة اللغة كسلسلة من الأفكار. يعمل هذا النموذج على مستويين من التجريد، حيث يقوم بإنشاء تمثيلات على مستوى الجمل مع الحفاظ على ذاكرة عمل للتمثيلات السابقة، مما يعالج مشكلات التعميم العلاقي وأخطاء السياق.

Los avances recientes en los modelos de lenguaje transformadores han llevado a la introducción del modelo Thought Gestalt (TG), que busca mejorar la generación de texto natural al modelar el lenguaje como una secuencia de pensamientos. Este modelo opera en dos niveles de abstracción, generando representaciones a nivel de oración mientras mantiene una memoria de trabajo de oraciones anteriores, abordando así problemas de generalización relacional y errores de contextualización.

Des avancées récentes dans les modèles de langage transformateurs ont conduit à l'introduction du modèle Thought Gestalt (TG), qui vise à améliorer la génération de texte naturel en modélisant le langage comme une séquence de pensées. Ce modèle fonctionne à deux niveaux d'abstraction, générant des représentations au niveau des phrases tout en maintenant une mémoire de travail des phrases précédentes, abordant ainsi les problèmes de généralisation relationnelle et d'erreurs de contextualisation.

Recent advancements in transformer language models have led to the introduction of the Thought Gestalt (TG) model, which aims to improve the generation of natural text by modeling language as a sequence of thoughts. This model operates on two levels of abstraction, generating sentence-level representations while maintaining a working memory of prior sentences, addressing issues of relational generalization and contextualization errors.

Modeling Language as a Sequence of Thoughts

One More Thing in AI – Your Shortcut to AI Mastery

Context-Emotion Aware Therapeutic Dialogue Generation: A Multi-component Reinforcement Learning Approach to Language Models for Mental Health Support

Was this article worth reading? Share it

One More Thing in AI

LucidQuery AI

DBTAI

PrettyPolly

Kansei

Graza.ai

Ready to build your own newsroom?