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

The paper discusses the creation of multilingual mental health dialogue datasets by modifying clinical personas through nationality and language parameters to generate dialogues in Mandarin, Bengali, and Hindi. It evaluates the performance of various LLMs in assessing depression severity in these languages compared to English, revealing that merely adding nationality and language may lead to clinical inconsistencies and inaccuracies in LLM assessments. This underscores the necessity for culturally responsive data generation to enhance the effectiveness of digital mental health support systems across diverse linguistic contexts.