Facebook has been trying to offer translations to users on its marquee social media platform for quite some time and now they’re harnessing the power of their artificial intelligence unit to improve the translations on the platform.
With over 2 billion users and support for 45 languages, Facebook’s translation services account for over 2,000 translation directions and 4.5 billion translations on a daily basis.
Previously, the company was using phrase-based statistical techniques for translation but stated that the methodology had its own limitations such as translating between languages which had different word orderings.
“One of the main drawbacks of phrase-based systems is that they break down sentences into individual words or phrases, and, thus, when producing translations, they can consider only several words at a time,” the researchers said.
Facebook has started working on a type of recurrent neural network, known as sequence-to-sequence LSTM (long short-term memory) to replace the phrase-based system.
“We recently switched from using phrase-based machine translation models to neural networks to power all of our back-end translation systems. These new models provide more accurate and fluent translations, improving people’s experience consuming Facebook content that is not written in their preferred language,” the company stated.
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. This will create more accurate translations.
“With the new system, we saw an average relative increase of 11 percent in BLEU ((bilingual evaluation under study) – a widely used metric for judging the accuracy of machine translation – across all languages compared with the phrase-based systems,” the team noted.