Understanding Asian American Help-Seeking through NLP: Towards Culturally Responsive Suicide Prevention

Background and Purpose: This conceptual paper addresses the importance of inclusive technology to ensure that no community is left behind in the digital mental health services. The current Natural Language Processing (NLP) models for suicide prevention are trained largely on Western language databases and expressions of distress. The help-seeking behaviors among Asian Americans facing emotional distress are often expressed indirectly, as treatments for somatic symptoms, and are influenced by cultural values. Asian American populations are often facing decisions to treat mental health problems are impeded by cultural values that prioritize family reputation over help-seeking. The objective of this paper is to develop a conceptual framework for assessing cultural appropriateness across diverse non-Western populations of NLP models designed to detect help-seeking. The hypothesis of this paper is that Asian American populations prefer more indirect or implicit language when expressing suicidal ideation and that Western-centric NLP models often fail to detect emotional distress among Asian Americans. By presenting a culturally grounded framework for detecting suicide risk for historically marginalized populations, this paper aims to democratize one technological approach and ensure more equitable outcomes.

Methods: This study is applied to a systematic integrative review methodology. Peer-reviewed articles were identified through comprehensive searches of multidisciplinary databases, including PubMed, PsycINFO, Social Work Abstract, and SAGE Journals. Selection criteria focused on empirical studies published within the last 10 years that examine (1) NLP-based suicide detection, (2) Asian American mental health help-seeking, and (3) cultural linguistic variations in distress expressions. A total of 50 empirical and methodological articles were selected for synthesis based on their relevance to the intersection of computational linguistic and cultural theory. The studies and findings were summarized to identify systemic gaps in suicide detection.

Results: The study results suggest that existing NLP suicide detection models demonstrate a “Western bias” and lower sensitivity for identifying indirect and culturally nuanced help-seeking cues among Asian American populations. To address these issues, a taxonomy of somatic symptoms and a conceptual framework are proposed that integrate culturally informed annotation, context aware modeling, and social work ethical principles. Applying this framework when using NLP models to identify suicide risk and help-seeking among Asian Americans will help improve equitable outcomes in detection and intervention.

Conclusions and Implications: This study proposes a culturally informed framework to promote risk identification and suicide prevention using NLP models with Asian American populations. This framework could inform social work interventions that are more culturally nuanced and go beyond applying “one size fits all” linguistic models to diverse populations. Such social work interventions hold promises for improving the accuracy of digital risk screening, identifying help-seeking behaviors, and improving culturally aware NLP models. The goal of this study is to provide insights for future research that integrates culture, theory, and technology in suicide prevention studies.