Google PlusFacebookTwitter

Lost In Translation

By on Jan 15, 2009 in Tech Takes | 5 comments

Share On GoogleShare On FacebookShare On Twitter

A few days back I was talking to someone about natural language processing and the challenges it holds for the future. NLP has the potential to revolutionize the way we interact with computers, and all this may finally become possible with the increased processing power becoming available at cheaper price points. Cloud computing could be a way forward too; a few voice-based search services for cellphones record user input on a cell, transfer it via high-speed 3G data networks, and process it at the service provider’s grid.

Technology, currently, is far from accurate. So until some breakthrough takes place NLP will probably provide us a barrel of laughs. Lost In Translation utilizes BabelFish‘s SYSTRAN-based machine translation to convert input in English to some other language, and then use that output to convert it back to English. The outcome can be quite funny. Sample this (the one I tried out). I entered

Cups of filthy liquid almost, but not quite, entirely unlike tea.

…which became

It is different from brown, almost, but considerably the glass of the impure liquid, completely.

This is when it was converted back and forth between English and Japanese. Japanese is totally notorious because even intonation and slight differences in a word can make a huge differnce in meaning.

Some people might say that this is all a problem with the machine translation model which uses rigid rules of grammar. But statistics-based systems (for instance, the one used by Google Translate) are no better. The same test applied to the same phrase gave me

Brown, but most have been different, quite impure liquid glass is full.

What makes Google better, some say, is the fact that it allows you to suggest a better translation. I disagree and think this is risky. While it may theoretically be better in a perfect world, we know that we don’t live in Utopia. Google’s system once it gets critical mass could be open to gaming by people out to subvert the system or just for laughs. Like googlebombing, thousands of users across the globe may start ‘suggesting a better translation’ for the Japanese word for ‘miserable failure’ as ‘George Bush’.

Until NLP becomes more accurate stuff like Lost In Translation would remain as jolly good jokes. Try it out, and if you get something particularly funny then post it here in comments section for all to see. ;)



5 Comments

      • Naman Bagga

        January 16, 2009

        This is quite a common error.I’ve seen lots of these mainly on Eeshole’s blog.Still can’t figure out the reason for this.

  1. Anuj

    January 15, 2009

    Post a Reply

    Oh that’s quite right you know. NLP is the future.

    Just one question who were you talking to by the way?

Submit a Comment

Your email address will not be published. Required fields are marked *