Google translate, accuracy and languages that will never make it to google translate
2011/01/30 § Leave a comment
First of all I must say am a big fan of Google translate and I use it almost everyday, being in the foreign land where i can’t understand the local language, life would have been much harder if there had not been Google translate. And i suppose there the thousands of people with similar situation. Apart from that for the thousands of articles published in different languages, if your friends tweet or Facebook in diff languages or if you are in strange place or store you don’t understand any thing for all these problems there is single stop Google translate. Yes Google translate is really great service with support for 52 languages you can lower your pressure if you are visiting new pace you don’t understand local language or if want to get gist of the article published in another language.
What about the accuracy of the service, can it be used for the professional services, can it ever be in the state that we can use it for translation of text books or poems ? probably not. Before telling any thing about the service we should know how it works. Google uses really intelligent technology instead of manually giving all the the translation rule of the languages to the system Google feeds enormous amount of human translated text to the system so system analyses these huge amount of translated texts and try to get significant patterns for translation, once system find these meaningful patterns on these human translated texts later it uses same patterns to translate the text.
So basically, accuracy of the system depends on the amount of human translated text provided to the system. More the translation pattern system can recognise more accurate the system gets. What if the language doesn’t have enough translated text, in that case Google translation system can’t learn enough translation rules/patterns that is required for the accuracy of the translation as a result you will get very poor translations. And the worst if some languages have very few translated texts then you may never see that language in the Google translate.
What’s the solution if we want to make translation system for less used languages, Solution is to manually define the translation rules of the language to the system. We can build a system where translation rules of the languages can be define manually. I think the framework like Grammatical Framework are very useful tools for building rule based systems. Given enough research and time we may even get the system more accurate than Google translate but as of the current stare of Machine translation techniques it very hard to get the same accuracy of translation as humans do. This is because unless the machine translation can pick up on the cultural undertones and subtleties at play in language we can’t get the same level of richness in the translation and i think these things are still very far from the current state of Machine Translation technology.