Hacker News Reader

Comment by dwc | Hacker News Reader

I study Norwegian, and I've seen some odd things like that in Google Translate. For instance, one time I translated a sentence containing a prominent Norwegian university that got translated to Princeton in English. I'm inclined to think that this wasn't malicious or a joke, but simply a deficiency in the automation. I've read a little bit about how Google Translate's automated stuff works, but not enough to feel confident to suggest how or why these kinds of errors happen.

View on HN

My favorite example (since fixed) was the word "amistad" translated from Spanish to English. It translated to "friendship," of course. You could add exclamation marks to it, like "amistad!" and it would translate to "friendship!" You could add more, and it would add more. But if you translated "amistad!!!!!" with precisely five exclamation marks, not four nor six but exactly five, it would instead translate to "murder!"

This sort of thing was mentioned in an early talk given by someone at Google Translate. (Sorry, I have neither the link nor the name of the person handy.)

As it makes use of parallel corpora, there were examples of documents mentioning (say) a famous person in the United States, but when the document was manually translated into French, the human translator selected a different famous person more well-known in France. It made perfect sense in the context of the document, but it threw a wrench into the automated translation based on those documents.

Both this and the article remind me of a strange Google Translate problem where the translation of "Austria" varied, depending on how many exclamation marks it was followed by:

http://itre.cis.upenn.edu/~myl/languagelog/archives/005485.h...

I would throw a wild guess that it tries to replace phrases with ones of equivalent meaning, so that the gist of a sentence is translated.

Things like "raining cats and dogs" doesn't probably translate well literally, for instance. It probably failed to categorise these universities as things that should not be translated in this manner.

I've seen this with measurements and currency.

Danish "... 15 meters ... 100kr" becomes "... 15 feet ... $100" which isn't helpful. (45 feet would not be helpful either.)

Reply on news.ycombinator.com