some thoughts about machine translation

I have discovered two websites: Bad Translator and Translation Party. They are two interesting websites, which firmly tell us the disadvantage and disability of machine translation.

They both make use of APIs provided by Google to build up the website. And you may agree that, Google Translate is one of the best machine translating tools (and Microsoft Translator is another good choice), if is not the best, for us. However, it seems disappointed and not so useful as we expect.

I think, in human translation, it’s an easier job to translate a paragraph than to translate a few words, because we can “feel” the meaning through reading a passage easily. But in the case of Google Translate, on the contrary, the machine “understand” single word better than paragraphs. Absolutely the result is the representation of the intelligence: the translating programs only “understand” the dictionaries that have been already located in the hard disks, but it can NOT quite exactly know the relations between words and words, phrases and phrases, sentences and sentences. Computers are still unable to analyse the “core meaning” hidden in the appearances of grammars and vocabularies.

How to make the computers “less stupid”? Well, I am just expressing my thoughts.

I am sure that Google has some algorithms: if one user submit a suggestion of translation, the computer “remember” and “compare” the difference between results by machine and human translation. Since Google has an enviable huge database of content of WWW, I think it may especially index the content from professional translating websites and analyse them. For example, YeeYan is a famous website for translating in China, and it provides a platform for us to translate passages in English into Chinese by human. And Google may index the original and the translated edition of the passages and analyse them to improve the result of its machine translation. Of course I am just telling an example and Google should take the law problem seriously.

What’s more, I think, we can never translate any literature by machine translation, unless we fully understand how human brains understand and translate the meaning of passages.

Update: Google is trying poetic machine translation. In fact, the engineers and scientists are doing these by generating translations with fixed rhyme and meter schemes.

Update: In fact, Google has described how the translation service works.