Google Translation: Getting More Intelligent?

February 7, 2019

Translation has never been easier with AI and NLP tools. It is amazing for people who cannot speak foreign languages to communicate with the assistance of translation apps, like Google Translation. While there are many translation apps on the market, Google is by far the best free one. As with many of its products and services, Google spends countless hours perfecting its language algorithms. The Verge published “Google’s Head of Translation On Fighting Bias In Language And Why AI Loves Religious Texts.”

Macduff Hughes heads Google’s translation and in the interview discusses how Google has moved from translating word by word but entire sentences. The new and smarter translation method is called “neural machine translation,” it uses machine learning, and a lot of its data comes from religious texts. One problem Google Translation faces is gender biased language. In order for translation AI to learn, it needs to be fed a lot of accurate and diverse data. These data sources, however, reflect societal biases which the AI can learn and replicate, such as doctors are male and nurses are female. The goal is to overcome these limitations so people know there is more than one way to phrase something as well as explain the differences.

Google is addressing three big bias and nuance initiatives. The first is to expand full sentence gender translation to more languages, the second is improving document translation based on context, and the third is addressing gender neutral languages. On a funnier and conspiracy based note is in 2018, when people typed nonsense words into Translate it spat back religious information. The explanation is a logical way of teaching AI:

“Usually it’s because the language you’re translating to had a lot of religious text in the training data. For every language pair we have, we train using whatever we can find on the world wide web. So the typical behavior of these models is that if it gets gibberish in, it picks out something that’s common in the training data on the target side, and for many of these low-resource languages — where there’s not a lot of text translated on the web for us to draw on — what is produced often happens to be religious.”

Translation is becoming a tool to organize more of the world’s information, according to Hughes, because it allows more people to access stuff that was in a different language. The naysayers argue that Translation provides a very shallow translation and Hughes acknowledges that. However, Translation works for basic translation and someday AI might have the skills of a professional linguist. It is not perfect, but Google Translate gets you to the train station and the bathroom.

Whitney Grace, February 6, 2019


Comments are closed.

  • Archives

  • Recent Posts

  • Meta