In New Jersey, Supplemental Nutrition Assistance Program (SNAP) services are vital for supporting food security among low-income households. However, significant accessibility barriers prevent many users from benefiting fully from the available resources. The current NJ SNAP website is designed solely in English and directs non-English-speaking users to Google’s translation service. This approach falls short in addressing the complexities of multilingual needs. While both English and non-English speakers encounter challenges navigating the static website, the inconsistent, confusing, and jargon-heavy translations exacerbate these obstacles for non-English-speaking households, leaving them at an even greater disadvantage.
Non-English-speaking users struggle with navigating a website designed with English speakers in mind, facing unfamiliar terminology and inconsistent translations that fail to adequately convey vital information. These shortcomings leave users unclear about SNAP eligibility requirements or application procedures. Additionally, without interactive support, users are forced to sift through static content and translated pages, which may not adequately answer their specific questions. For households with limited digital literacy, these issues are further magnified, making the website even harder to use.
To tackle these challenges, we have developed a custom chatbot prototype that aims to provide equitable access to SNAP services. Built on OpenAI’s API and leveraging the Retrieval-Augmented Generation (RAG) model, the chatbot bridges the gap between users and SNAP resources by delivering answers in English and Spanish (among other languages). The chatbot has preprocessed content from the NJ SNAP website and converted it into smaller chunks of information. This chunking enables the chatbot to retrieve the most relevant details for any given user query. Once the user submits a question, the system enhances it by appending the most pertinent chunk of verbiage and sends the enriched query to OpenAI’s API, which generates a tailored response in the same language as the original query.
Post-processing using the instruction tuning approach further refines the chatbot’s output to maximize usability. Responses are crafted for visual readability, ensuring they are easy to understand at a glance. They are written using 8th-grade-level vocabulary, making SNAP information accessible to a broader audience. Additionally, all chatbot responses are traceable to specific language from the original website, ensuring transparency and accuracy in the information provided.
While these enhancements benefit all users, they are particularly impactful for non-English-speaking households, who face additional challenges including navigation complexities, translation inconsistencies, and digital accessibility hurdles. The chatbot’s tailored multilingual support helps overcome these compounded barriers and fosters greater inclusivity.
Currently, the chatbot is undergoing evaluation to ensure its effectiveness in addressing user needs. Feedback from users and performance data will shape further refinements to the system, ensuring it continues to serve the community effectively.
By addressing language barriers, improving accessibility, and providing interactive support, this chatbot represents a significant step toward equitable SNAP services. It empowers users with varying levels of language proficiency or digital literacy to confidently access the resources they need. As technology evolves, tools like this chatbot pave the way for inclusive solutions that ensure public services reach everyone in the community.