SINAI/ALIA-es-clinical-psychology-dialogues

DISCLAIMER: This dataset is not clinically validated. It is a research proof-of-concept. It should not be used as clinical truth or as a replacement for qualified human professional consultation.

Dataset Introduction

The ALIA Spanish Clinical Psychology Dialogues Corpus is a curated conversational instruction-tuning resource in Spanish created under the ALIA project. It was designed to train and evaluate language models in empathetic therapeutic dialogue and psychological counseling.

This dataset consists of two distinct subsets (configurations):

1. clinic-roleplay (University Psychology Clinic): 66 multi-turn clinical dialogues developed through roleplay by mental health professionals at the University.
2. TerapIA (TerapIA App Conversations): 320 anonymized real-world multi-turn conversational interactions between users and the digital AI assistant Pía in the TerapIA application.

Summary of Subsets

Subset / Config	Source	Num Instances	Avg Turns / Session	Total Tokens (Approx.)	License	Authors
clinic-roleplay	University Psychology Clinic	66	~30	~150,000	CC BY-SA 4.0	SINAI Research Group
TerapIA	TerapIA App Conversations (Pía Assistant)	320	~44	~387,000	CC BY-NC-SA 4.0	Inteligencia Artificial Generativa y Procesamiento de Lenguaje Natural, S.L.

---

Dataset Details

Dataset Description

1. clinic-roleplay Configuration (University Psychology Clinic)

This subset contains multi-turn conversational interactions between a patient (user) and an empathetic therapist (assistant), developed through roleplaying by mental health professionals (profesionales de la salud mental). Each dialogue session is seeded with a system prompt that steers the assistant to act as an empathetic, active-listening therapist who validates patient emotions and explores their experiences.

The corpus includes dialogues where the therapist responses prioritize validation, empathy, active listening, and safe clinical guidance in Spanish, avoiding generic, flat, or dry AI-generated advice.

2. TerapIA Configuration (TerapIA App Conversations)

This subset contains multi-turn conversational interactions in Spanish between real users (user) and the digital assistant Pía (assistant). The dialogues cover typical psychological and emotional concerns (e.g., anxiety, stress, relationship issues, self-esteem, depression) and showcase Pía's empathetic, active-listening guidance, prioritizing validation and safe therapeutic strategies (such as cognitive restructuring and relaxation techniques).

---

Uses

This dataset is intended for:

• Supervised Fine-Tuning (SFT) of conversational LLMs for psychological counseling and active listening in Spanish.
• Research in empathetic communication, active listening, and clinical dialogue strategies.
• Evaluation of emotional alignment and role-playing capabilities in Spanish mental health contexts.

---

Dataset Structure

Data Instances

clinic-roleplay Configuration

Each instance follows this structure:

{
  "conversations": [
    {
      "from": "system",
      "value": "Eres un terapeuta psicológico empático y profesional. Escucha activamente al paciente y responde de forma apropiada, validando sus emociones y explorando su experiencia."
    },
    {
      "from": "user",
      "value": "Hola Pues muy nervioso, no paro de darle vueltas a las cosas"
    },
    {
      "from": "assistant",
      "value": "Entiendo. ¿Este es el motivo por el que has decidido solicitar ayuda psicológica?"
    }
  ]
}

TerapIA Configuration

Each instance follows this structure:

{
  "id": "0",
  "messages": [
    {
      "role": "user",
      "content": "Hola."
    },
    {
      "role": "assistant",
      "content": "¡Hola, Laura! Qué alegría saludarte de nuevo. Cuéntame, ¿cómo has estado desde la última vez que hablamos?"
    },
    {
      "role": "user",
      "content": "He estado un poco mejor en general, pero esta semana he tenido mucho estrés."
    },
    {
      "role": "assistant",
      "content": "Entiendo, el estrés puede ser muy desafiante, especialmente si ya has estado lidiando con ansiedad..."
    }
  ],
  "last_conversations_summaries": "## Resúmenes de las últimas sesiones de más recientes a más antiguas\nDurante esta sesión, Laura Fernández expresó preocupación por...",
  "summary": "Durante esta sesión, Laura Fernández compartió sentimientos de estrés tanto en el ámbito laboral como personal..."
}

---

Data Fields

clinic-roleplay Configuration

• conversations (list of dicts): The sequence of conversational turns in the dialogue session.
  • from (string): The role of the speaker (system, user, or assistant).
  • value (string): The text content of the turn.

TerapIA Configuration

• id (string): Unique identifier of the conversation.
• messages (list of dicts): The sequence of conversational turns in the dialogue session.
  • role (string): The role of the speaker (user or assistant).
  • content (string): The text content of the turn.
• last_conversations_summaries (string): Contextual summary of the patient's previous sessions (from newest to oldest).
• summary (string): The clinical summary of the current session.

---

Data Splits

Distribution by split:

Configuration	Split	Num Instances	Percentage
clinic-roleplay	train	66	100.00%
TerapIA	train	320	100.00%

---

Example Usage

from datasets import load_dataset

# 1. Load the clinic-roleplay (University Psychology Clinic) subset
dataset_uja = load_dataset(
    "json",
    data_files="ALIA-es-clinical-psychology-dialogues.jsonl",
    split="train"
)
print(f"UJA Subset - Number of examples: {len(dataset_uja)}")

# 2. Load the TerapIA subset
dataset_terapia = load_dataset(
    "json",
    data_files="ALIA-TerapIA-es-clinical-psychology-dialogues.jsonl",
    split="train"
)
print(f"TerapIA Subset - Number of examples: {len(dataset_terapia)}")

---

Dataset Creation

Curation Rationale

This corpus was created to provide high-quality, domain-specialized conversational supervision for Spanish psychological support tasks within the ALIA project. The objective is to train models that exhibit genuine empathy, active listening, and emotional validation rather than generic AI advice.

Source Data

• clinic (University Psychology Clinic): Multi-turn dialogues generated via roleplay by mental health professionals (profesionales de la salud mental). The dialogues cover typical psychological concerns (e.g., anxiety, grief, relationship issues, self-esteem, depression) in Spanish, simulating authentic clinical/counseling interactions.
• TerapIA (TerapIA App Conversations): Conversational interactions between real users and the digital assistant Pía from the TerapIA application. The dialogues cover typical psychological concerns in Spanish, representing authentic, real-world support interactions.

Data Collection and Processing

clinic-roleplay Configuration

1. Mental health professionals conducted clinical roleplay sessions to simulate patient-therapist interactions, structuring them into multi-turn dialogues.
2. A clinical-domain system prompt is provided to establish the therapeutic context.
3. Therapist responses (assistant) were developed and refined by these professionals to reinforce key therapeutic strategies such as cognitive restructuring, active listening, and emotional validation.

TerapIA Configuration

1. Conversational data was collected from real-user interactions with the assistant Pía in the TerapIA application between November 25, 2024 and May 28, 2025.
2. The dataset was processed to extract the conversation history, along with the clinical summaries of the sessions (summary and last_conversations_summaries).
3. Anonymization pipelines were applied to ensure all personal and sensitive information was completely removed.

The resulting .jsonl files are formatted for seamless instruction-tuning using frameworks like Axolotl.

---

Annotations

No manual annotations are included in the metadata fields. Roles and content fields are structured during consolidation.

Personal and Sensitive Information

• clinic-roleplay: The dataset contains simulated therapeutic scenarios based on roleplay. No real patients were involved.
• TerapIA: The dataset contains anonymized real-user conversations. Standard filtering and de-identification pipelines were applied to ensure that no personally identifiable information of the users is present.

---

Considerations for Using the Data

Discussion of Biases

Potential biases include:

• Culturally-specific psychological language and idioms of Spain/Spanish-speaking regions.
• Focus on specific types of clinical dialogue patterns and patient scenarios.
• Style homogenization typical of curated conversational templates and the assistant Pía's conversational design.

Citation

% Citation for the clinic-roleplay subset
@misc{ALIA-es-clinical-psychology-dialogues,
  title={ALIA Spanish Clinical Psychology Dialogues Corpus},
  author={SINAI Research Group},
  year={2026},
  publisher={HuggingFace},
  howpublished={\url{https://huggingface.co/datasets/SINAI/ALIA-es-clinical-psychology-dialogues}}
}

% Citation for the TerapIA subset
@misc{ALIA-TerapIA-es-clinical-psychology-dialogues,
  title={ALIA-TerapIA Spanish Clinical Psychology Dialogues Corpus},
  author={Inteligencia Artificial Generativa y Procesamiento de Lenguaje Natural, S.L.},
  year={2026},
  publisher={HuggingFace},
  howpublished={\url{https://huggingface.co/datasets/IAG-PLN/ALIA-TerapIA-es-clinical-psychology-dialogues}}
}

---

Funding

This work is funded by the Ministerio para la Transformación Digital y de la Función Pública - Funded by EU – NextGenerationEU within the framework of the project ALIA.

---

Contact: ALIA Project - SINAI Research Group

More Information: SINAI Research Group