I Translate, Therefore I Be
Would machine translators free people up for other work? Well, that still remains to be seen as to which degree you would like to depend on it. While the need for crude machine translation may be useful in emergencies, it may fall short of the hopes to replace language learning for face-to-face communication.
- The targeted sentences supplied by Google Translate must never be mistaken for “correct translation,” which never exists for there are too many nonverbal and unexpected factors in human speech.
- Google Translate, instead of translating, actually pairing most equivalent sentences from an astronomically large corpus utilized by Google search engine. That means it can be cool when it comes to writing, because it has access to the documentation online like Google Books. However, I would not expect it to work miracle when I have to express myself at a meeting, or when I need to craft a prose of beautiful translation. Of course, it should be of no use for simultaneous interpretation.
- I think over the last decades there has been a resurgence of optimism in the research community around AI, machine learning and voice recognition, which has come really close to solving some core problems with AI and interpretation/translation. Machines learnings improves with experience, which means even robots can learn as they are being preprogrammed. With enough data and examples to look at, machines come up, hopefully, with functionaries on their own. Therefore, you feed the systems lots of speech to leave them a conclusion arrived at by induction. And, this gives Google a competitive edge because it has a huge database built and expanded by Google’s search engine.
That is an incredible accomplishment as to understanding human speech given the problems caused by intonation. I think they have a group of software engineers sitting there, trying to write down the rules of syllable, phoneme, semantics and sound. But, they will ultimately get to a point where unpredictable behavior emerges.
Given the progress that has been made so far, machines are moving toward self-education with a huge database fed by the Net and speech. But the final mile would be how to build the correlations between the examples, but causal inference is much more complicated for them to acquire because languages do not always follow the rules. After all, the world is not just a matter of only statistic intelligence and correlations.
Take raining and opened umbrellas as an example. We can train a system to sift through millions of images of people going outdoors with their umbrellas opened. Nonetheless it may come up with an incorrect inferrence that whenever people open their umbrellas, it is going to rain. Umbrellas and rain are linked with each other, but are not necessarily in an cause-and-effect relation.
Understanding and learning this cause and connections in languages is what makes chatting beyond preprogrammed rules, and, of course, algorithm’s capabilities. If not, then we are predictable machines, too.
1.Google Translate App Gets an Upgrade http //bits.blogs.nytimes.com/2015/01/14/google-translate-app-gets-an-upgrade/?smid=fb-share&_r=0
3.Why Google Translate Can’t Compare to a Human Translation Service http://verbalink.com/articles/why-google-translate-cant-compare-to-a-human-translation-service