"Using Caffe2, we significantly improved the efficiency and quality of machine translation systems at Facebook". "Creating seamless, highly accurate translation experiences for the 2 billion people who use Facebook is hard", explained the company in a blog post.
This is the reason Facebook adopted neural machine technology and abandoned the old-school phrase-based technology. 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.
The Caffe2 team today also announced that in part due to work done around translation, the framework is now able to work with recurrent neural networks. Compare these two examples from Facebook of a Turkish-to-English translation.
FAIR was recently in news for reportedly shutting down one of its AI systems as chatbots defied the human-generated algorithms and started communicating in their own language.
Charles Oakley accepts plea deal, banned one year from Madison Square Garden
The firm had revenue of $386 million during the quarter, compared to analysts' expectations of $365.57 million. The Madison Square Garden had a negative net margin of 3.81% and a negative return on equity of 1.57%.
The new system put in place will not just translate a phrase from the sentence but will take the entire sentence and translate from that context. "As a result, all machine translation models at Facebook have been transitioned from phrase-based systems to neural models for all languages".
That essentially makes for "more accurate and fluent translations", they wrote, noting an average relative increase of 11 percent in a commonly-used metric used for scoring the accuracy of machine translations.
This leads to difficulty translating between languages with markedly different word orderings. A translation of that word is searched for in a sort of in-house dictionary built from Facebook's training data, and the unknown word is replaced. Now, however, the site is using neural networks to power its translations, which can take into account full sentences as well as their context, generating much more accurate translations. That allows abbreviations like "tmrw" to be translated into their intended meaning - "tomorrow".