Google Translate writes its own translation engine

I noticed on the Google Operating System blog yesterday that Google Translate had switched translation engines from Systrans widely used engine to an in-house statistical based translation engine.

Intrigued, I headed over to Antonio Ortiz’ blog (Antonio is based in Malaga and blogs in Spanish), copied some of the text from one of his posts and ran it through both engines.

This is the original Spanish:

Microsoft comprará 20 compañías al año durante los próximos cinco. Es lo que ha afirmado Steve Ballmer en la conferencia Web 2.0 de San Francisco, que también ha señalado que también incluirán fabricantes de software open source en la lista y que prefieren compañías pequeñas (pequeñas entre 50 millones y 1000 millones de dólares) antes de que se conviertan en inaccesibles (News.com).

Here is the Systrans translation:

Microsoft will buy 20 companies to the year during next the five. It is what it has affirmed to Steve Ballmer in the conference Web 2,0 of San Francisco, that also has indicated that also they will include software manufacturers open source in the list and who they prefer small companies (small between 50 million and 1000 million dollars) before they become inaccessible (News.com).

and now here is Google’s translation:

Microsoft will buy 20 companies a year for the next five. This is what has Steve Ballmer said at the Web 2.0 conference in San Francisco, which has also indicated that manufacturers will also include open source software on the list and who prefer smaller companies (small between 50 million and one billion dollars) before they become inaccessible (News.com).

While they are both far from perfect, I think you will agree the Google translation is at least as good as the Systrans one. Not bad for a new translation engine.

The Google Translation service has the option to “Suggest a better translation” – this is then fed back into the translation engine and it learns apparently!

What will they think of next!

6 thoughts on “Google Translate writes its own translation engine”

  1. Well, it seems to me the Google one is a little bit better than the systrans. Have you asked at home? 😉
    Take care!

  2. One of the drones told me about this product two or three years ago. It’s the result of throwing big statistical analysis at a corpus of documents that have been formally translated into a number of languages (French, English and Chinese is where they started). There was an academic paper about this resulting from some competition they entered. Apparently the Google system is particularly strong compared to the rest of the field as regards Chinese. I am not in a position to comment on whether this is true.

  3. The option to add human input is a big plus – maybe most of us translators will have to start looking for alternative sources of income in the near future…

  4. Am not sure how good is this translation as I don’t speak Spanish, but I must say that Google is done extremely well with arabic language.

  5. I also think the Google one is better 🙂

    I use it quite regularly. But I don’t think professional translators will be out of a job any time soon. Human language is so complex. But perhaps in the far, far future there will be more possibilities for them to translate faster with the help of programs like this.

Comments are closed.