How Cultural Diversity in the NWS Workforce Combines Forces with AI to Improve Alerting to Limited English Proficiency Communities During Weather Events

Meteorologists and hydrologists with the National Weather Service (NWS) struggle with their limited capability to adequately communicate weather hazards in languages other than English. In order to start addressing this gap in service, the NWS needed to tap into the cultural diversity of its workforce and find bilingual staff that were willing to serve by providing translations into non-English languages. For decades, the San Juan Weather Forecast Office staff has assisted the National Hurricane Center (NHC) in translating tropical weather products utilized across the Caribbean, Central, and South America Countries. Since 2017, the Multimedia Assistance in Spanish and the Spanish Outreach Teams have aided and expanded Spanish translation efforts throughout the contiguous US, especially during high impact events. These two teams manually translate short weather messages and safety tips on social media to alert the Hispanic community of hazards, impacts, and threats.

While the NWS has continued to make great strides in providing services in Spanish, there have been increased calls for the federal government to better serve non-English speaking communities in recent years. In addition, the recently published Hurricane Ida Service Assessment “represents the first time NWS has reviewed service delivery formally through the lenses of equity and underserved communities.” A finding from the Ida assessment revealed that the “NWS has no current capacity for the multitude of other languages spoken by residents or visitors to the U.S. and only limited capabilities to address communities with disabilities that limit communications.”

Recent advances in Machine Learning and Artificial Intelligence (AI) technology offer the possibility of translating NWS products into Spanish much faster. During the 2022 hurricane season, AI-based language models drastically reduced the time required for translation of the NHC Tropical Cyclone Public Advisories and Tropical Weather Outlooks using human-in-the-loop interactive training and leveraging “model memories” to recognize and use past human-verified translations automatically.

With the previous success with the AI models translating weather products into Spanish, the NWS is working with multilingual groups within the agency to use AI to assist with additional languages, such as Simplified Chinese, during an experimental period. This presentation will highlight examples of how culturally diverse groups within the NWS workforce leverage their linguistic expertise to reduce language barriers preventing preparedness action needed for weather events. In addition, this talk will also share the success and methodology behind how these multilingual employees approached experimenting with automated translations in an operational environment.