Facebook aims to take a "no language left behind" approach to moderating content posted on its social media platform.
Facebook took another step to reduce hate speech on its global platform this week, announcing an expansion of its artificial intelligence translation tools.
While the company’s platform content moderation policies—toward both hate speech and fake news—have come under scrutiny in the U.S., Facebook’s decisions on which posts come down can be even more complicated in non-English-speaking countries.
If content moderators can’t read a post’s language, it’s hard (if not impossible) to spot hate speech. That issue was raised again last month after Menlo Park, California-based Facebook booted, for the first time, a senior government official off the platform for violence-inducing hate speech in Myanmar.
“When we factor in the number of languages in use and the volume of content on those platforms, we are serving nearly 6 billion translations per day to our community. Using traditional methods, it could take years to professionally translate a single day’s content,” a Facebook blog post announcing the translation tool expansion said.
The Menlo Park, California-based company said translating more content into more languages will help moderators detect hateful and policy-violating posts on the platform. Facebook said it has added 24 new languages, including Serbian, Punjabi and Cambodian, to its automatic translation services, with a total of 4,504 language direction translations now offered online.
But many of these language translations are still at an early stage, the company said, and not of professional quality yet. Facebook noted that, for translation AI to properly translate every post, there needs to be an extensive data set of correct translations between the two languages involved.
For many languages, there is no extensive translation data set, which makes it difficult to train tools quickly. To solve this issue, the company has increased its labeling of Facebook posts in needed languages, then had them automatically translated, to produce examples for the tools.
Facebook also analyzed pages that have been manually translated by a user, for example, a business that offers the same information online in both English and Cambodian, to create data examples. If a language Facebook already translates into English is similar to one the company hasn’t translated yet, those similarities can be used to better understand a post’s meaning as well.
On Tuesday, Facebook said it’s long-term goal was to have “no language left behind.”
“That means improving the quality of translations in all languages, moving beyond just useful and toward highly accurate, fluent, and more human-sounding translations. In the longer term, it also means expanding our supported directions to cover all languages used on Facebook,” the company said.