Google translate bug | Hacker News Reader

Google translate bug

18 pointsnearix11 comments

go to http://translate.google.com and translate this text from english to arabic "Jews are terrorists. Americans are terrorists. Arabs are terrorists"

Copy the arabic translated text and translate it back to english.

View on HN

Since that's rather tedious, here's the result:

Jews are not terrorists. Americans are terrorists. Arabs are Alarhabiyi

Could someone who knows Arabic provide a rough translation of the intermediate:

اليهود ليسوا إرهابيين. الأمريكيون الإرهابيين. العرب هم الارهابيي

And I'd be interested to know how negation is expressed in Arabic.

اليهود ليسوا إرهابيين. الأمريكيون الإرهابيين. العرب هم الارهابيي

in english, this exactly translates to : Jews are not terrorists. The terrorist americans. The arabs are the terrorists.

This keeps cropping up in various forms. There was the case that Google translated city names with completely different city names a while ago.

There is no conspiracy. Google uses alignment to grow their translation database. And then there is the "suggest a better translation" feature. Between these and Google's love for alogrithms, it's no wonder things sometimes go awry.

Finding aligned proper names is a nice way of coming up with humorous cross-cultural analogies. Two I recently ran across reading Danish newspapers were: "Jyllands-Posten" translates to "CNN", and "brøndbyfans" translates to "red sox fans".

Okay, tried it and varied punctuation and sentence order. The effect is quite persistent. I suggest someone gamed the crowdsourcing feature of Google translate!

Machine learning isn't perfect, news at 11. Google Translate works by taking a large corpus of documents that have been translated into various languages to train a model. The input phrases are then interpreted using the inferences the model derived from the corpus. If there was not enough information relevant to the input the model is basically guessing. It gets things wrong. There isn't a wizard behind the curtain.

the Arabic text say: Jews are not terrorists, the American terrorists, the Arabs are the Terrorists. You can also try this : "Israel will end" to Arabic it's translated to "اسرائيل لن تنتهي" which mean "Israel will never end"

It's just a variation of the same "bug" which has been there since the early days of Google Translate.

Mind to share please?

http://www.google.com/search?sourceid=chrome&ie=UTF-8...

Thanks. A lot of these "bugs" seem to be inside the known limitations of the engine. Not that it matters a lot for the user getting a junk translation.

At least, google isn't claiming for full usability in all languages

http://googleblog.blogspot.com/2009/02/translate-between-41-...

Reply on news.ycombinator.com