In today’s Daily Telegraph, James Titcomb writes about Artificial Intelligence and suggests that we are getting a bit too worked up about the possibility of killer robots, etc. In his article he includes the suggestion that Google Translate is now approaching superhuman accuracy, linking to a further article in the Washington Post which claims that this is the case.
To test this, I tried Google Translate with one of the best loved phrases in the English language, “the cat sat on the mat”, and asked it to do its superhuman stuff. Now my Spanish is of a reasonable level, but I am not fluent. However, I do know that the word “alfombra” in Spanish means “carpet” and the word “alfombrilla” means “mat” (literally, a little carpet). There are also other words for “mat” in ‘Spanish, “la estera” and “el felpudo”, the latter being a mat by a door.
Guess which word the superhuman Google Translate chose? Correct, it went with “alfombra”. The cat sat on the carpet.
Yes, machine translation is getting better and will continue to do so, but that doesn’t excuse lazy journalism such as this. More pertinently, if this small detail is wrong, why should I believe the rest of the argument that Mr Titcomb makes in his piece in the Telegraph? And even more pertinently, if you’re striving for real accuracy, don’t rely on “superhuman” computers just yet. Just a normal human being, fluent and trained in the use of another language, is still the best thing you can get.
Rosetta Stone, Global Connects