Combining Community-Engagement with AI: Tutorial for a New Model for Research Translation Equity (Preprint)

BACKGROUND Research often excludes the experiences of English-as-an-Additional-Language participants due to English-dominant practices. Among women living with HIV in Canada, this is particularly a concern given that an increased number of immigrants are testing positive for HIV. To increase representation of english-as-an-additional-language communities within the literature of women living with HIV, we developed and implemented a community-engaged, culturally sensitive translation process using artificial intelligence language platforms. OBJECTIVE The objective of this study was to increase research accessibility and equitability for English-as-an-Additional-Language immigrant women living with HIV by addressing language barriers. METHODS The British-Columbia CARMA-CHIWOS Collaboration is a community-based study of women living with HIV. We consulted immigrant women living with HIV to identify the most commonly spoken languages of their communities for translation of study materials. Participants identified seven languages for translation: French, Gujarati, Swahili, Shona, Hausa, Luganda and Zulu. We identified six top-performing artificial intelligence platforms through review of ratings and forum recommendations: Lingvanex, OpenL, MerlinAI, QuillBot, ChatGPT and DeepL. We translated samples of study materials using all six identified platforms for all seven community-identified key languages. We engaged immigrant women living with HIV, fluent in one of the target languages and English, in a second community consultation to provide feedback on all materials translated into the target language. We hired community members with lived experience and language expertise to refine translations, ensuring accuracy and cultural sensitivity. Completed knowledge mobilization materials were shared at a final community event. RESULTS Translations were completed in Gujarati, Hausa, Shona, Swahili, Zulu, Luganda, and French, with top-performing AI tools identified as OpenL for Luganda, DeepL for French, Lingvanex for Gujarati, Hausa, Shona, Swahili and Zulu. Costs of translation were reduced from $10,912 per language to $800.00. Community feedback highlighted linguistic nuances overlooked by AI. Participants appreciated this tailored approach, which improved accessibility, reduced costs, and provided paid opportunities for community. CONCLUSIONS This project highlights the feasibility and value of combining AI tools with community engagement to overcome language barriers to inclusion in health research. The British Columbia CARM-CHIWOS Collaboration plans to adopt this approach for future materials, offering a replicable model for enhancing inclusivity and representation in research.