Facebook Algorithm can directly translate texts between 100 languages
Facebook has developed a machine-learning algorithm that can perform direct pair translations between 100 languages without relying on the English translation, as existing systems do. The accuracy of the translations was estimated by experts at about 90%.
Most services for the automatic translation of the text into another language first translate the texts into English, and then into the desired foreign language. Due to this, the accuracy of some elements of the text is lost in the process. According to a Facebook researcher, the Facebook new translate algorithm will be able to perform direct translations without the use of English.
Also Read: Facebook AR Glasses Aria Release Date
In training the algorithm, the Facebook team used 7.5 billion text examples from across the Web. The model was taught by focusing on languages that are usually translated into each other. She then grouped the languages into 14 separate collections based on geographical and cultural similarities.
For some language pairs, the new system has shown significant improvements in translation compared to existing results. For example, translating from Spanish to Portuguese using a new algorithm is particularly effective because Spanish is the second most widely spoken language worldwide and researchers have had access to a large amount of data.
According to BLEU – the algorithm for assessing the quality of translated text – the quality of translation is estimated at 10 points higher than using the English-oriented model. Although this system is not used on the platform, Facebook plans to launch it soon.