I just tried it for Dutch, and it got one out of the ten 5-word sentences I tried, correct. Some sentences only got one word right, and I sure was articulating like they do on national television ;)
They just got a much greater sample size for English, matter of time for the rest of the languages I hope.
"... and I sure was articulating like they do on national television" Not sure what they trained their algorithm on, but this articulation might have been the culprit. If they have a corpus of "normal people" speaking Dutch, and then you go and articulate, you will get poor results.