Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Language-Agnostic Suicidal Risk Detection Using Large Language Models

June-Woo Kim, Wonkyo Oh, Haram Yoon, Sung-Hoon Yoon, Dae-Jin Kim, Dong-Ho Lee, Sang-Yeol Lee, Chan-Mo Yang

TL;DR

This work tackles the scalability bottleneck of language-specific suicidal risk detection by proposing a language-agnostic framework that converts speech to Chinese text via ASR, uses large language models to extract bilingual suicidal-risk features, and fine-tunes language models in Chinese and English. It integrates text and speech modalities through multimodal fusion, enabling cross-lingual transfer where features extracted in one language can inform models in another. Evaluated on the SW1 dataset, the approach achieves performance comparable to or better than monolingual baselines and direct ASR fine-tuning, while highlighting strengths in cross-lingual feature extraction and multisensory integration. The findings suggest robust, scalable potential for multilingual adolescent mental-health assessment and suicidal-risk detection, with future work focusing on prompt optimization, generalization, and multilingual adaptability.

Abstract

Suicidal risk detection in adolescents is a critical challenge, yet existing methods rely on language-specific models, limiting scalability and generalization. This study introduces a novel language-agnostic framework for suicidal risk assessment with large language models (LLMs). We generate Chinese transcripts from speech using an ASR model and then employ LLMs with prompt-based queries to extract suicidal risk-related features from these transcripts. The extracted features are retained in both Chinese and English to enable cross-linguistic analysis and then used to fine-tune corresponding pretrained language models independently. Experimental results show that our method achieves performance comparable to direct fine-tuning with ASR results or to models trained solely on Chinese suicidal risk-related features, demonstrating its potential to overcome language constraints and improve the robustness of suicidal risk assessment.

Language-Agnostic Suicidal Risk Detection Using Large Language Models

TL;DR

This work tackles the scalability bottleneck of language-specific suicidal risk detection by proposing a language-agnostic framework that converts speech to Chinese text via ASR, uses large language models to extract bilingual suicidal-risk features, and fine-tunes language models in Chinese and English. It integrates text and speech modalities through multimodal fusion, enabling cross-lingual transfer where features extracted in one language can inform models in another. Evaluated on the SW1 dataset, the approach achieves performance comparable to or better than monolingual baselines and direct ASR fine-tuning, while highlighting strengths in cross-lingual feature extraction and multisensory integration. The findings suggest robust, scalable potential for multilingual adolescent mental-health assessment and suicidal-risk detection, with future work focusing on prompt optimization, generalization, and multilingual adaptability.

Abstract

Suicidal risk detection in adolescents is a critical challenge, yet existing methods rely on language-specific models, limiting scalability and generalization. This study introduces a novel language-agnostic framework for suicidal risk assessment with large language models (LLMs). We generate Chinese transcripts from speech using an ASR model and then employ LLMs with prompt-based queries to extract suicidal risk-related features from these transcripts. The extracted features are retained in both Chinese and English to enable cross-linguistic analysis and then used to fine-tune corresponding pretrained language models independently. Experimental results show that our method achieves performance comparable to direct fine-tuning with ASR results or to models trained solely on Chinese suicidal risk-related features, demonstrating its potential to overcome language constraints and improve the robustness of suicidal risk assessment.

Paper Structure

This paper contains 21 sections, 1 equation, 1 figure, 6 tables.

Table of Contents

  1. Introduction
  2. Preliminaries
  3. Dataset Description
  4. Training Details
  5. Evaluation Metrics
  6. Method
  7. Text Modality
  8. Text Data Acquisition using ASR
  9. Bilingual Suicidal Risk-Related Feature Extraction with LLMs
  10. Language Models Fine-tuning
  11. Speech Modality
  12. Multi Modality and Final Prediction
  13. Results
  14. Text Modality
  15. Speech Modality
  16. ...and 6 more sections

Figures (1)

  • Figure 1: Illustration of the proposed language-agnostic suicidal risk detection framework using LLMs. Speech data is processed through a pretrained speech model, while an ASR system converts speech to text. LLMs extract suicidal risk-related features in Chinese and English, which are used to fine-tune respective language models. Finally, multimodal fusion combines speech and text representations for suicidal risk prediction.