I highly doubt it. There are basically two models of human speak. Originally human speak was just single utterances suitable for every occasion. Then they started to fine tune those words modifying them a little. And eventually those modifications became their own words. But this happened: certain languages branched at the second stage and standardized those modification rules. Facebook and Google refuse to study and manually code in those rules, and translations are and will be shit. Neural network will not learn those rules, because the underlying doctrine is based on not modifying words, but building dictionaries and sentence models.<p>BTW. There was already perfect Finnish-English
translator 40 years ago based on taking words apart by the hard-coded rule-book, and constructing same sentence of separate words, and then translating those words.