Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Sólo Escúchame: Spanish Emotional Accompaniment Chatbot

Bruno Gil Ramírez, Jessica López Espejel, María del Carmen Santiago Díaz, Gustavo Trinidad Rubín Linares

TL;DR

This work introduces Sólo Escúchame, the first open-source Spanish emotional-accompaniment chatbot built on LLaMA-2-7b-Chat to address gaps in Spanish mental-health support. It introduces HEAR, a two-stage Spanish-emotion dataset (HRECPW-based emotion recognition and GPT-3.5-Turbo-generated generic responses) used to fine-tune the model via LoRA and instruction-following prompting, with CPU-efficient quantization. The authors propose a semi-automatic evaluation framework using Active Listening and Socratic Method criteria, assigning $20\%$ weights to each of five criteria and validating with GPT-4-instruct and human raters; Sólo Escúchame outperforms baselines (GPT-3.5, LLaMA-2-7b-Chat, Mixtral 8x7b, GPT2-124M) across both metrics, e.g., $90.67$ vs $87.62$ in Active Listening and $77.12$ vs $67.84$ in Socratic Method. They also provide a conversation example to demonstrate practical, empathetic Spanish dialogue. The work offers a practical, open-source, CPU-friendly tool for Spanish emotional support and outlines future extensions, such as recordings and voice-input integration, to better capture non-verbal cues.

Abstract

According to the World Health Organization (WHO), suicide was the fourth leading cause of death in the world for individuals aged 15 to 29 in 2019. Given the rapid increase in mental health issues, providing psychological support is both crucial and urgent. In this paper: (1) we propose Sólo Escúchame, the first open-source Spanish emotional assistance chatbot, based on LLaMA-2-7b-Chat. (2) We introduced the HEAR (Hispanic Emotional Accompaniment Responses) dataset, compiled from multiple English sources translated into Spanish, as well as generic data generated using ChatGPT-3.5-Turbo. Finally, (3) we propose an evaluation metric based on two semi-automatic assessment methods. Our system outperforms a range of state-of-the-art models in providing psychological assistance in Spanish. Our models and datasets are publicly available to facilitate reproducibility.

Sólo Escúchame: Spanish Emotional Accompaniment Chatbot

TL;DR

This work introduces Sólo Escúchame, the first open-source Spanish emotional-accompaniment chatbot built on LLaMA-2-7b-Chat to address gaps in Spanish mental-health support. It introduces HEAR, a two-stage Spanish-emotion dataset (HRECPW-based emotion recognition and GPT-3.5-Turbo-generated generic responses) used to fine-tune the model via LoRA and instruction-following prompting, with CPU-efficient quantization. The authors propose a semi-automatic evaluation framework using Active Listening and Socratic Method criteria, assigning weights to each of five criteria and validating with GPT-4-instruct and human raters; Sólo Escúchame outperforms baselines (GPT-3.5, LLaMA-2-7b-Chat, Mixtral 8x7b, GPT2-124M) across both metrics, e.g., vs in Active Listening and vs in Socratic Method. They also provide a conversation example to demonstrate practical, empathetic Spanish dialogue. The work offers a practical, open-source, CPU-friendly tool for Spanish emotional support and outlines future extensions, such as recordings and voice-input integration, to better capture non-verbal cues.

Abstract

According to the World Health Organization (WHO), suicide was the fourth leading cause of death in the world for individuals aged 15 to 29 in 2019. Given the rapid increase in mental health issues, providing psychological support is both crucial and urgent. In this paper: (1) we propose Sólo Escúchame, the first open-source Spanish emotional assistance chatbot, based on LLaMA-2-7b-Chat. (2) We introduced the HEAR (Hispanic Emotional Accompaniment Responses) dataset, compiled from multiple English sources translated into Spanish, as well as generic data generated using ChatGPT-3.5-Turbo. Finally, (3) we propose an evaluation metric based on two semi-automatic assessment methods. Our system outperforms a range of state-of-the-art models in providing psychological assistance in Spanish. Our models and datasets are publicly available to facilitate reproducibility.
Paper Structure (26 sections, 2 equations, 1 figure, 7 tables)

This paper contains 26 sections, 2 equations, 1 figure, 7 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Psychological support chatbots.
  4. Large Language Models (LLMs).
  5. Sólo Escúchame Language Model
  6. Dataset
  7. Stage 1 : Create the Hispanic Emotion Recognition Based on Plutchik's Wheel (HRECPW) Dataset.
  8. Stage 2: Create the Hispanic Emotional Accompaniment Responses (HEAR) Dataset.
  9. Model
  10. Fine-Tuning.
  11. Prompting.
  12. Experiments
  13. Evaluation Protocol
  14. $\bullet$ Active Listening:
  15. $\bullet$ Socratic Method:
  16. ...and 11 more sections

Figures (1)

  • Figure 1: Overview of the procedure to generate generic data using GPT-3.5.