In the wake of the COVID-19 pandemic, people turned to search engines like Google for information. A recent study, “Language disparities in pandemic information: Autocomplete analysis of COVID-19 searches in New York” by Singh et al., highlights how these autocompletes can shape public perception and amplify health disparities. This is crucial for informing the IMPACT-NJ project, which focuses on policy-guided equitable chatbot technology development for New Jersey. The technology behind Google autocomplete is also fundamental to chatbot conversations, which rely on next-word prediction to simulate dialogue.
Key Findings of the Study
The study audited and compared Google’s autocomplete results for COVID-19-related searches in Spanish and English over a period of 100+ days in 2020. The findings revealed significant disparities between the two languages:
- Fewer Autocomplete Options in Spanish: Spanish queries yielded fewer autocomplete options compared to English queries. This limitation in choice could restrict the breadth of information accessible to Spanish-speaking users.
- Negative Sentiment in Spanish Results: The content of the Spanish autocompletes tended to be more negative. For example, while English autocompletes included neutral or informative suggestions, Spanish autocompletes often included more negative connotations.
- Differing Topical Coverage: Spanish autocompletes included themes related to religion and spirituality, which were notably absent in the English results. This divergence in topical coverage could influence how different communities perceive the pandemic and their responses to it.
The differences highlighted by the study underscore the role of search engines as gatekeepers of information. By providing varied results based on language, search engines may contribute to differing perceptions of the same public health crisis. In the context of a pandemic, this could lead to unequal access to accurate information and potentially exacerbate health disparities among different language-speaking communities.
Design Implications for Health Chatbot Technologies for New Jersey
The IMPACT-NJ project, which stands for “Innovative Policy-guided Approach to Chatbot Technology for New Jersey,” aims to create equitable chatbot systems guided by thoughtful policymaking. The findings from the above paper underscore the need for a critical lens and intentional design when creating equitable health technology for a state like New Jersey, which has a significant Spanish-speaking population.
- Multilingual Support: Ensuring that chatbots cater to New Jersey’s diverse population by offering robust support in multiple languages, including Spanish.
- Policy-Guided Development: Establishing policies that ensure chatbot algorithms are transparent and equitable, mitigating inequities that could affect minority communities.
- Continuous Auditing: Implementing ongoing audits of chatbot responses to ensure they remain accurate and unbiased over time.
The findings from this study are directly guiding our development of an equitable chat prototype, ensuring it serves all communities effectively. The conceptual findings might also be of interest to others interested in shaping New Jersey’s policy around health equity in health information systems. By addressing language disparities, we can foster a more inclusive and fair digital health landscape, setting a standard for equitable information dissemination.
Yonaira Rivera is an assistant professor of communication at the Rutgers School of Communication and Information and Vivek Singh is an associate professor of library and information science at the Rutgers School of Communication and Information.