Facebook translations haven't exactly been spot-on in the past, but that might change from here on out. Today marks the end of Facebook's transition period between their former phrase-based translation system and their new AI-driven neural machine translation (NMT) system.

While Facebook is a bit behind the curve with this move (tech companies like Google and Microsoft have been using NMT since last year) this technology should result in a significant boost to translation accuracy for the social media giant.

The issue with Facebook's previous phrase-based translation system (and phrase-based systems on a whole) is that it wasn't really capable of looking at entire sentences. Rather, the system would break down sentences into individual phrases and keywords, and then use probability calculations and search algorithms to attempt to create an accurate translation.

This might sound good on paper, but that sort of technology quickly falls apart when faced with the challenge of translating sentences between languages with vastly different sentence structures, such as Mandarin and English.

On the other hand, neural machine translation attempts to look at full sentences and generate translations based on context, rather than individual words or phrases. If you compare Google Translate several years ago to Google Translate today, you can clearly see the difference this sort of technology can make.

However, AI translation technology is still far from perfect, and Facebook faces a unique set of problems when it comes to translating content on their site. For example, Facebook engineering manager Necip Fazil Ayan pointed out in an interview with TechCrunch that Facebook is a platform that sees a lot of slang, acronyms and other informal language, which would likely make it more difficult for an NMT system to automatically translate text across the entire platform.

Probably unrelated to this story but timing coincides with recent news that Facebook researchers decided to shut down AI they invented after it started speaking its own made up language.