Creating Multilingual Mental Health Dialogue Datasets: Limits of Persona-Based Localization via Nationality and Language

2026-06-17 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Human-Computer Interaction · Depth: Expert, quick

Summary

Researchers investigated the efficacy of using persona-based methods to generate multilingual mental health dialogue datasets, specifically for Mandarin, Bengali, and Hindi. This study aimed to address the critical shortage of high-quality data for training AI and large language models (LLMs) in global mental health. By modifying nationality and language parameters in synthetic clinical personas, dialogues were generated and then evaluated by different LLM judge models for depression severity against an English baseline. Findings indicate that merely adding nationality and language parameters can introduce clinical inconsistency across languages. LLM judges frequently showed inaccuracies in assessing depression severity in non-English texts, with performance varying significantly among models. This highlights systemic limitations of applying English-centric personas to multilingual contexts and underscores the urgent need for culturally responsive data generation.

Key takeaway

For AI Scientists and NLP Engineers developing global mental health support systems, you must move beyond simple persona localization. Relying solely on nationality and language parameters for multilingual data generation risks introducing clinical inconsistencies and inaccurate LLM evaluations. Instead, prioritize investing in culturally responsive data generation methods to ensure your systems provide equitable and accurate support across diverse linguistic and cultural contexts.

Key insights

Persona-based localization via nationality and language is insufficient for generating consistent multilingual mental health datasets.

Principles

• English-centric personas introduce clinical inconsistency in multilingual contexts.
• LLM judge accuracy varies significantly across non-English texts.

Method

Modified nationality and language parameters in clinical personas to generate dialogues in Mandarin, Bengali, and Hindi, then evaluated LLM judges on depression severity assessment.

In practice

• Avoid direct translation of English-centric personas for multilingual data.
• Prioritize culturally responsive data generation methods.

Topics

• Multilingual Datasets
• Mental Health AI
• Large Language Models
• Persona-Based Generation
• Clinical Inconsistency
• Culturally Responsive AI

Best for: Research Scientist, AI Scientist, NLP Engineer, AI Ethicist

Related on AIssential

• Psychological Digital Twins : Conceptual Models, AI Foundations and Ethical Challenges — Robust Psychological Digital Twins require hybrid AI, combining LLM flexibility with symbolic guardrails for precision mental health.
• Can we trust AI to detect healthy multilingual English speakers among the cognitively impaired cohort in the UK? An investigation using real-world conversational speech — AI models for cognitive decline detection exhibit bias against multilingual English speakers, particularly those with specific accents.
• Benchmarking GPT Models for Conversational AI Systems: Can AI Read a Doctor’s Notes? — LLMs struggle with medical abbreviation expansion, especially context-dependent ones, despite larger models performing better.
• Towards AI-augmented decision making in psychiatry — AI-augmented decision-making can standardize care for heterogeneous psychiatric disorders.
• AI’s fluency in other languages hides a Western worldview that can mislead users − a scholar of Indonesian society explains — LLMs exhibit "epistemological persistence," retaining a Western worldview despite multilingual fluency due to English-centric training data.
• ChatHealthAI: Aligning Electronic Health Record Representations with Large Language Models for Grounded Clinical Reasoning — ChatHealthAI integrates EHR representations with LLMs for grounded, interpretable clinical reasoning and accurate patient prediction.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.