In Depth
The Machine That Learns From You
What LILT actually does
Most translation tools work in one direction: machine generates, human checks, content ships. LILT's architecture flips the dynamic. As a human translator works, the engine watches - updating its predictions in real time, learning correction patterns, adapting to the specific domain, tone, and terminology of that translator's workflow. The result is a human-machine team that gets faster over time rather than plateauing.
Green describes this as the core insight: machine translation systems can generate but cannot verify. An enterprise shipping product documentation in 47 languages can't afford to find out there was an error after the fact. LILT's model keeps humans in the decision loop not because the AI isn't capable, but because the stakes require it.
The government bet that validated everything
In 2020, LILT was awarded an AFWERX SBIR Phase II contract - a U.S. Air Force program designed to identify and scale dual-use technology. The contract tasked LILT with tracking international aerospace and cyber developments, translating foreign-language intelligence to inform Air Force decision-makers. The company later partnered with In-Q-Tel, the CIA's venture arm, to serve additional government agencies.
This is where Green's Northrop Grumman background becomes legible. He knew how to build software that governments trust - software with strict quality requirements, defined accountability structures, and consequences for failure. Defense translation isn't forgiving. A mistranslated document in an intelligence context is worse than no document.
LILT's public sector business now runs in parallel with its enterprise commercial division, serving Fortune 500 companies localizing websites, marketing content, software products, and customer support.
The NAACL paper that caught the industry's attention
In 2021, LILT's research team - approximately 15 researchers split between SF and Berlin - won the Best Paper award from the North American Chapter of the Association for Computational Linguistics (NAACL). The paper described a neural automatic review and correction system designed to automate the reviewer step in translation workflows - the most expensive and time-consuming quality assurance bottleneck in the industry.
Academic translation research and commercial translation practice are historically disconnected. Green considers bridging that gap a core part of LILT's competitive advantage. The research team meets with the services team monthly - an intentional structure that prevents the company from drifting into academic irrelevance or commercial mediocrity.
The economics that drive the mission
Green has been blunt about the target: translation needs to cost roughly one dollar per piece before the problem is genuinely solved. Today's enterprise translation costs are prohibitive - which means the world's information is still siloed by language, still inaccessible to most of the people who need it.
This isn't altruism dressed up as a business plan. It's a market thesis. If translation costs drop far enough, the total addressable market expands from large enterprises with localization budgets to every organization that creates content. The potential is enormous. The technology path is difficult.
Green's view on AI hype is consistent with this pragmatism. He critiques claims about sentient machines and fully autonomous vehicles as distractions from the actually hard problems - like making it cheap and reliable for a hospital in Indonesia to publish accurate patient instructions in Bahasa. The stakes are high. The solutions need to be real.
The podcast and the long game
"The Last Mile" - Green's podcast hosted for LILT - takes its name from the hardest part of translation: not the machine pass, but the final human judgment that makes content actually work in context. The show explores AI adoption across industries, featuring conversations with executives and technologists navigating the shift.
It's also a signal about where Green places LILT in the market: not as a translation vendor, but as a voice in the broader conversation about how enterprises implement AI in workflows that matter. The company spent years building before the GenAI wave arrived. Green's bet - that human oversight plus machine efficiency, not automation alone, is the right architecture for high-stakes language work - now looks prescient.
