John DeNero

John DeNero is co-founder and chief scientist of Lilt, an AI translation company that helps enterprises and government agencies translate their content with a mix of machine translation, AI agents and human linguists. He co-founded it with Spence Green, whom he met while both worked on Google Translate, and runs its research division. He is also a teaching professor of electrical engineering and computer sciences at the University of California, Berkeley.

Studying Machine Translation

DeNero earned a bachelor's degree in mathematical and computational science and symbolic systems and a master's degree in philosophy at Stanford, then a PhD in computer science at UC Berkeley, where his research was on machine translation, according to his LocWorld speaker biography. When he started in the field in 2005, he said on the Tomato podcast, automatic translation produced something useful but not very good, and "it was always easier to make fun of it than to use it."

Working on Google Translate

He joined Google Translate right after graduate school, even though he expected to become a professor one day, because the timing was so good: the service was by then used by hundreds of millions of people a month. "I wanted to have a positive impact on lots of human beings," he said. He studied what people typed into it and found that many used it to write love letters, while others pasted in short phrases without any context. That led to a project that showed users the common translations of a word, so they could tell apart, for example, the different Spanish translations of "post."

Building Lilt for Professional Translators

Professional translators were not finding Google Translate very useful, DeNero said, because publication work depends on context beyond a single sentence, such as what a document is for. Lilt, incorporated in March 2015, set out to build a system "that was great for professional translators" and that would not keep making the same mistakes. He points to the National Weather Service, which uses Lilt to translate weather alerts for communities that speak other languages. The goal, he said, is AI that does not "replace people or do a worse version faster," but makes skilled linguists faster and more accurate and takes the drudgery out of their work.

Teaching Data Science at UC Berkeley

At UC Berkeley he is faculty director of data science undergraduate studies and created Foundations of Data Science, the largest undergraduate course on campus; as a graduate student there he taught the only undergraduate course in artificial intelligence. "I think I've always been motivated by trying to build the courses and programs that I would have wanted to take when I was an undergraduate," he said. He argues that useful work with data needs computing, statistics and knowledge of the subject together: "In order to do anything useful with data, you have to understand what the data mean."