Silicon Valley AI wants to solve translator shortages between the US and Ukraine

Training manuals and tutorials for weapons ranging from the United States to Ukraine are piling up, and few Americans are able to translate them into Ukrainian or Russian.

There are a limited number of U.S. citizens who speak Ukrainian who can obtain top secret clearance to process documents related to national security. US trainers have acknowledged some of the difficulties in teaching Ukrainian troops to use the millions of dollars worth of equipment provided by the US and other allies.

“There are the natural challenges associated with training in a second language, ranging from overcoming a language barrier to certain technical elements,” said Brig. General Joseph Hilbert, head of the 7th Army Training Command in Europe, who oversees the training of small groups of Ukrainians in Grafenwoehr, Germany.

But the federal government spends nearly 200 times more on procurement for translation and other language requirements that depend on people than it does on AI-powered services.

Artificial intelligence-based language translation company Lilt wants to solve the government’s language problem with its technology, CEO and founder Spence Green said.

A US soldier sits in front of the barrel of a tank at the US Army military training base in Grafenwoehr, southern Germany, on March 11, 2022. (Photo by Christof STACHE / AFP) (Photo by CHRISTOF STACHE/AFP via Getty Images)

The U.S. government has 83% of the required manpower for Slavic linguists, and there is a need for at least 100 more Slavic linguists for the U.S. intelligence community alone, according to Green’s conversations with U.S. government officials. .

“You have all this information that suddenly has to go into Ukrainian, and you have a fixed group of people in the United States in a classified environment,” Green said. “One solution is to train more of them. The other solution is to use machine learning to make that small group of people five times larger.

Translate on the battlefield

“The solution of sending people to a language school for 12 months does not solve President Zelenskiy’s problems,” Green added, referring to Ukrainian President Volodymyr Zelenskiy.

The Defense Language Institute’s Foreign Language Center in Monterey, Calif., and Washington, DC, conducts immersion courses for land-based military intelligence jobs and positions with the FBI and National Security Agency. The languages ​​taught depend on the needs of the services and the Ministry of Defense, according to Natela Cutter, spokesperson for the institute.

The center currently does not teach any Ukrainian language classes, with busy classes in Chinese, Korean, Russian, Arabic, Spanish and Persian Farsi, she said.

“I don’t think you can say there’s a ‘vacuum’ in the Slavic languages,” Cutter said. “It’s fair to say that after the fall of the Berlin Wall and the changes in Eastern Europe in the early 1990s, we stopped teaching a number of Eastern European languages.”

Rusty Shughart, a retired Air Force colonel who chairs the American Translators Association’s government linguist outreach task force, said managing language crises can be challenging in terms of scope and duration, and must involve not only the government and the military, but also volunteers. language specialists, U.S. allies, and nongovernmental organizations, among others.

But he was skeptical that AI could work in Ukraine.

“AI is making pretty significant progress, and especially in machine translation,” he said. “I’m very impressed with what AI can do, especially in other related areas like neural machine translations etc., but I don’t think we’re at the point where we can apply those things to this field. special battle.”

AI translation market

Since fiscal year 2018, the contract segment of the artificial intelligence and machine learning market defined by the Bloomberg government for translation has amounted to less than $25 millionincluding $2.3 million to Lilt. The total for all translation and lawful interception services during this period is $4.7 billion.

Competition in the US startup market for translation services is thin, with most of the space dominated by large tech companies such as Microsoft and Google. Green cited a Chinese company with $400 million in public funding as an example of non-US development in AI-based translation. Baidu Inc., a Chinese company that hosts one of China’s largest search engines and other AI projects, is developing a portable translator.

Baidu is leading a years-long race for China’s first driverless taxis

Other competitors include Unbabel Inc., Reverso, DeepL GmbH, NeuralSpace, Meta AI, Amazon Translate, Bing Microsoft translator, and Google Translate.

Lilt’s argument is that it keeps “humans up to date” with the AI ​​algorithm for translating written content, allowing the company to provide translation at a lower cost than hiring human translators. , and more specifically than using only AI.

Interview requests to AFWERX, a US Air Force technology branch that awarded Lilt a small business innovation research contract, were not immediately returned.

But Intel Corp. invested in Lilt in 2019 and they have been working with them ever since. Mark Rostick, vice president and senior managing director of Intel Capital, said he viewed Lilt as a “really exciting new company that was really trying to change a market that hadn’t changed in 30 or 40 years.”

Rostick described big tech companies’ big translation platforms as “in a good enough phase,” but added that Lilt’s ability to learn from a specific data set and level of accuracy make it a challenge in this space.

“There are times when you can’t afford to make a mistake,” he said of accurate translations.

Lobbying congress

Lilt’s human-in-the-loop technology can translate over 70 languages ​​and is currently under contract with the intelligence-sharing alliance known as “Five Eyes”, consisting of the United States, Canada, UK, Australia and New Zealand. It also associates with the United States Aviation, navy, and the FBI.

The language-agnostic and data-driven system learns from “parallel texts”, i.e. processing a sentence in English and the same sentence in Ukrainian, for example. The software is self-learning and does not require a network connection, which sets it apart from some of the other major players in the space.

Lilt is currently working with Invariant, a government affairs and public relations firm run by Heather Podesta, to secure further government support, including in the Ukraine emergency funding request. The $40 billion supplement for Ukraine that the House voted on Tuesday did not include specific funds for translation, but did mention broader reimbursement for Pentagon “defense services.”

The company is also focused on fiscal year 2023 funding for classified programs, as well as research and development accounts that may be part of annual authorization and defense appropriations bills that Congress works on.

Lilt has already spent $130,000 in lobbying fees at Invariant since late last year, according to the Senate Lobbying Disclosure Database. Lilt previously spent $320,000 with Banner Public Affairs LLC, according to the database.

Lilt and Invariant are meeting with lawmakers and government officials “to emphasize that with an overwhelming flood of foreign language content and data in the digital age, national security agencies simply cannot hire enough qualified American translators. and authorized to brutally force out of the problem,” Green said in an email.

To contact the reporters on this story: Patty Nieberg in washington at [email protected]; Roxana Tiron in washington at [email protected]

To contact the editors responsible for this story: Amanda H. Allen to [email protected]; Anna Yukhananov to [email protected]