Autonomous translation

Siri, do my Spanish homework

First, let’s note that while automatic translation services like the original Google Translate or Facebook’s translation button have their own ethical issues, here we’re talking about AI-driven systems. Automatic translation isn’t even good enough to let you cheat on your Spanish homework most of the time, whereas AI-assisted translation is now being used in life or death situations.

Work on this technology is not new. DARPA and other parts of the U.S. military have been developing something akin to Star Trek’s universal translator or The Hitchhiker’s Guide To The Galaxy’s Babelfish for almost a decade. Chinese Internet powerhouse Baidu has also been developing their own real-time pocket translator since 2015. We’ve seen the release of Google Buds that claim to do real-time translation and the recent announcement of Microsoft that it created the closest thing to human-like translation, but most people agree that we’re still miles away from being able to accomplish this task.

Human language is incredibly complex. We don’t use words accurately, our slang changes over time and by location, we use idioms to express ourselves, and we’ve come to accept bad pronunciation and grammar as the norm. Good translation requires more than getting just the gist of a conversation, especially in business or military engagements. It’s nearly impossible to teach a computer all the intricacies of a language AND enable them to know when to use what usages.

New deep-learning AI programs are designed to solve the problem of immediate, on-the-ground translation for people who need to communicate and act on important information when a translator is not available.

Currently, our speech recognition programs have trouble with noisy, distorted, or degraded speech, and so we still need to allocate significant human resources to this task in important situations, such as diplomatic missions.

There’s no end in sight to the push to create smarter AI translations, but there are some things to consider as we continue to develop this technology:

  • Will a reliance on technology strip us of not only language but basic communication skills as we increasingly talk through machines?
  • To what extent can we ever rely on a machine in a life or death situation, such as on the battlefield or in the hospital?
  • Will the billions of dollars we spend developing translation software be a good investment in a more connected future?
  • Can a machine ever be as nuanced as a human?
  • Will humans have to dumb down their communications in certain situations in order to be understood by machines?
  • Who is responsible when AI translations go horribly awry, resulting in lost money or lives?


Microsoft Says Its AI Can Translate Chinese As Well As A Human (Engadget, 2018)

Translation, Ethics, And Social Responsibility (The Translator, 2017)

DARPA page on Project RATS  (and here)

Why Computers Still Can’t Translate Languages Automatically (Slate, 2012)