Skip to Content
Artificial intelligence

Baidu’s Chinese-to-English translator finishes your sentence for you

October 23, 2018

Baidu has developed a new technique for smoother real-time machine translation.

The challenge: Developing a simultaneous translation system has been a tough nut to crack because word order differs between different languages. Consider this sentence in English, then Chinese: “The US president meets with the English prime minister.” “美国总统与英国首相会晤.” In English, the word “meets” appears near the beginning; its Chinese counterpart, “会晤” (huiwu), appears at the end. Because of this, commercial “real-time” translation systems wait for the person to complete a sentence before translating it into the target language. The result is a clunky user experience with awkward delays.

How it works: Baidu’s new approach shortens the delay by starting before a sentence is finished. If the system were translating the sentence above from Chinese to English, for example, it would anticipate the English word “meets” after hearing the first part of the Chinese sentence, based on the likelihood that the US president would meet with someone. The idea was inspired by an anticipation technique commonly used by human translators to keep up with the speaker.

Work in progress: Like any predictive machine-learning model, the system only works with reasonable accuracy when trained on copious amounts of language data with similar sentence structures. As a result, there are still significant limitations to deploying this system at scale. Baidu researchers found that the system doesn’t perform as well as full-sentence translators in experiments on Chinese-to-English translation with a five-word delay. But the researchers are excited about the new path forward. “Simultaneous translating for human interpreters is extremely challenging and burdensome,” says Liang Huang, the principal scientist at Baidu Research, “so we’re hopeful machines can step in and really make this service more accessible for professionals and consumers.”

Deep Dive

Artificial intelligence

Large language models can do jaw-dropping things. But nobody knows exactly why.

And that's a problem. Figuring it out is one of the biggest scientific puzzles of our time and a crucial step towards controlling more powerful future models.

Google DeepMind’s new generative model makes Super Mario–like games from scratch

Genie learns how to control games by watching hours and hours of video. It could help train next-gen robots too.

What’s next for generative video

OpenAI's Sora has raised the bar for AI moviemaking. Here are four things to bear in mind as we wrap our heads around what's coming.

Stay connected

Illustration by Rose Wong

Get the latest updates from
MIT Technology Review

Discover special offers, top stories, upcoming events, and more.

Thank you for submitting your email!

Explore more newsletters

It looks like something went wrong.

We’re having trouble saving your preferences. Try refreshing this page and updating them one more time. If you continue to get this message, reach out to us at with a list of newsletters you’d like to receive.