Finetuning Large Language Models for Automated Depression Screening in Nigerian Pidgin English: GENSCORE Pilot Study

Language barriers to mental health care can be addressed through AI: The study demonstrates that fine-tuned language models can effectively screen for depression in Nigerian Pidgin English, a language spoken by millions but underrepresented in medical AI tools. This approach could help overcome linguistic obstacles that currently prevent people from accessing depression screening in their native language.

Existing mental health tools may not work for all populations: Traditional screening tools like the PHQ-9 were validated primarily in high-income countries and may not be linguistically or culturally appropriate for low- and middle-income countries. The researchers found that understanding local slang, idioms, and cultural context was essential for accurate depression screening, highlighting how one-size-fits-all approaches can fail vulnerable populations.

Cultural appropriateness matters as much as technical accuracy: While GPT-4.1 achieved 94.5% accuracy in predicting depression severity scores, the study emphasizes that qualitative measures—including cultural appropriateness, clarity, and relevance—were equally important evaluation criteria. This suggests responsible AI development requires more than just high accuracy numbers; it must account for how tools are perceived and trusted by the communities they serve.

Dataset quality and human expertise are critical for fair AI in healthcare: The study involved rigorous human annotation including semantic labeling and cultural interpretation of language. This labor-intensive process underscores that deploying AI for mental health screening in underserved communities requires investment in local expertise and careful data preparation, not just algorithm optimization.