AUSTIN, Texas — When it comes to severe weather, communication is key. That is why the National Weather Service (NWS) is turning to artificial intelligence in order to make sure everyone in the U.S. and its territories has access to the most up-to-date and accurate forecast products.
Monica Bozeman is the project manager of AI translation at National Oceanic and Atmospheric Association (NOAA). She said translation in the NWS used to be different.
“The San Juan Forecast office has been translating into Spanish for about 30 years or so,” she said.
In Puerto Rico, there is an abundance of talented meteorologists who are bilingual, but that is not always true for stateside operations like local forecast offices and even the National Hurricane Center, based in Miami.
"You know, [the San Juan forecast office] kept getting more requests for translations, and so they were unable to meet the needs sufficiently with the talent that they had, so they were reaching out for help, asking if there was another office that could assist in translations," Bozeman explained.
To end the large reliance on the San Juan office for translations, NOAA found that translations by way of artificial intelligence (AI) could keep non-English speaking homes up-to-date on weather events and save a massive amount of time.
"They were able to drop that time to translate down to about less than 10 minutes," Bozeman said.
That's opposed to about an hour if done manually.
While amazing, the program isn't perfect. The bilingual forecasters have been training the AI program in Spanish for three years now, so it’s pretty reliable. But now, the NWS is working on auto translations for Chinese, Vietnamese and even Samoan.
That being said, mistakes happen.
"I have had some bloopers," Bozeman said. "When we first started with Chinese, I can't read Chinese myself, but the symbols for tornado apparently were very similar to a torpedo. So we quickly trained that out of there. We didn't want to be sending out torpedo warnings."
So far, there is a team of 20 to 30 employees doing quality checks on the AI for a few hours a day. But since this is about communicating weather to the public, the National Weather Service wants public feedback too. It's created an experimental website: weather.gov/translate.
"We're taking comments from the public and partners for a year. We wanted to have a very long extended response and comment period to see how we're doing," Bozeman said.
As the NWS prepares to totally revamp its website in the near future, Bozeman said it hopes to integrate these automatic translations into it, so that local populations with large amounts of non-English speakers can benefit from the most accurate and up-to-date weather information.