The translations are terrible because the program can't understand anything beyond the words. Context, register and tone are all lost. Each word is translated based on some fancy maths (algorithms or something) that works out how frequently words are used together and picks what it thinks is the best option.
If you've got a typo things become even worse. Humans are pretty good at working out when words shouldn't be there. That's because they understand the sentence. If we wrote though instead of thought it wouldn't be picked up by the program since both are correct spellings of words. There's no semantics involved with Google Translate. Only words that are "related" to other words in other languages.
For example, the word fan can refer to people who support or like something or the device used to circulate air. Many, if not most, languages have two distinct words for these two concepts. Google Translate will have to pick one. In Spanish there are a couple options... it could choose aficionado for the people and ventilador for the machine.
 |
| Cool... |
In the sentence "There are a lot of fans on the ceiling," the online translator still picks aficionado, although most humans would know from context that we're probably not talking about people, as the machines tend to be installed in the ceiling. Not Google... it sees the words and guesses.
 |
| Well done Google... |
We're not offering a simple solution to this modern age dilemma. However, you should be proud that your brain is powerful enough to deduce things like this. The amount of processing power required to calculate these things is immense. If you remember our post on speech recognition, we already know that people are definitely smarter than machines. Don't worry about a Matrix-esque uprising any time soon!
Google Translate is apparently better at some languages now because rather than translate, it looks for the same sentence that exists in its database (ie all the text of google books and other stuff on the internet) in both languages. So if it's a sentence that's already been written in both, like if a book exists in original French and in English translation, it just picks out the translation rather than do it from scratch. This is what I gather from reading reviews of Alex Bellos' 'Is that a fish in your ear' - I haven't read the whole book yet. Google Translate is still bad when your sentence hasn't been written in either or both languages though.
ReplyDeleteThat's interesting, Laura. We're also reading "Is That a Fish in Your Ear" and really enjoying it.
ReplyDeleteIt is terrible for Turkish to English. The verb in Turkish is at the end of the sentence with suffixes for tense and person. Turkish also has clauses within clauses before the verb pulls it all together. Google gets very confused...
ReplyDeleteOh, Catherine, I think that's a very mild way to put it :)
ReplyDeleteTurkish = agglutination magic
(as in Finnish, though Finnish has become quite good now)
"Yizın Paristeydik."
"We were in Paris during the summer".
Google Translate?
FORGET IT. The "machine" is just beaten by this plethora of combinations in endings.