Since 2021, the National Weather Service (NWS) has experimented with artificial intelligence (AI) to translate weather warnings into multiple languages, cutting translation times to about a minute – but recent contract lapses and budget constraints have forced the agency to scale back.
The Government Accountability Office (GAO) said in a Monday report that the AI pilot launched by NWS to translate weather alerts into other languages needs a clear long-term plan to ensure that communities with limited English proficiency can remain safe in severe weather-related events.
Currently, 36 local NWS offices and the National Hurricane Center offer AI-powered translations in Spanish, Simplified Chinese, Vietnamese, French, and Samoan for certain weather forecasts, watches, warnings, hazard messages, tropical weather outlooks, and other statements.
While those offerings can improve safety in communities with limited English proficiency, GAO said that NWS has no updated long-term plan and faces staffing, funding, and quality-control challenges.
NWS officials and stakeholders told GAO that clearer goals, stronger oversight, and human review of translations are needed to prevent missed milestones and ensure accuracy.
Last April, GAO said NWS paused its pilot and then later reduced the scope and funding for the project after the contract supporting the effort lapsed and took several weeks to be resigned and revised.
The new contract was delayed due to a recent Trump administration policy requiring a department-level review of all National Oceanic and Atmospheric Administration (NOAA) contracts over $100,000. The NWS is a component of NOAA.
The new agreement cut pilot funding by 40%, and while NWS said it will continue to offer translations in the five languages it previously supported, it cannot begin developing new language translation models before April 2026.
“NWS also has limited ability to update and improve the language translation models under the new contract, according to [NWS] officials,” GAO said.
GAO said that NWS officials also reported that expanding AI translations into more languages hinges on funding. Still, NWS sought no dedicated funds for the project in its fiscal year (FY) 2024-2026 congressional budget justifications.
Despite limited funding, NWS plans to expand its translations for operational use in the future, according to GAO, which added that the pilot has been both cost-effective and time-efficient.
“According to NWS officials, it takes about 3 months to initially train the AI model for a new language, and models are trained and improved over time,” GAO wrote. “To train the models, NWS bilingual staff and the contractor’s linguists review translations of weather products and provide corrections to the model.”
The translation process takes about one minute and is integrated into the system that forecasters use to generate weather products, GAO said.
However, GAO said AI-translated weather alerts remain difficult for the public to find because they are not available on the NWS main website or local forecast pages, requiring users to navigate a separate translation site and multiple drop-down menus.
According to the report, NWS officials said a redesigned weather.gov website – which is in development – and an outreach campaign could eventually make those translations easier to access.
GAO recommended that NWS develop an updated implementation plan for its AI language translation project that includes resources needed to achieve each goal. NWS agreed with GAO’s recommendation.