miCRoSCoPiC^eaRthLinG 0 Report post Posted March 28, 2007 Ventures into online machine based translations haven't met with much success so far - at least not to a degree that can be called phenomenal. While smaller words / phrases / sentences translate fine, the translator engines perform rather pathetically when it comes to large chunks of text. Translation engines have been around since the early 90's, when AltaVista came out with its famous Babelfish. In general, the online translators are pretty good and are able to convey the essence of the translated text. However, translator engines fair miserably while operating on non-Western languages specially languages which are based on pictograms / ideograms, like Chinese, Japanese etc. Google seems to have finally realised that, following an incessant barrage of whines and grumbles about it's own translation service over the past few years. Google envisions that in near future ... people will be able to translate documents instantly into the world's main languages, with machine logic, not expert linguists, leading the way. In order to make this happen, Google plans to bring some drastic reforms to the algorithms working behind scene. As is expected of a technology pioneer, they've devised a new approach to the whole issue - namely, "statistical machine translation" which differs from any of the past efforts in that it forgoes language experts who program grammatical rules and dictionaries into computers." The new process involves feeding pre-translated parallel text in various languages into computers and then relying on them to discern patterns for future translations. At the moment the quality offered by this mechanism isn't perfect either - but there's a distinct advantage to such a statistical analysis engine. With time, "the more data we feed into the system, the better it gets" said Franz Och, the head of Google's translation effort at its Mountain View, California headquarters. He further stated that "some people that are in machine translations for a long time and then see our Arabic-English output, then they say, that's amazing, that's a breakthrough." Read the full article Notice from vizskywalker: I know chaos-labratory is yours m^e, but search engines like Google don't know that, and even if they did, couldn't make the link. One of the reasons for quoting things (other than not gaining credits for work that isn't yours) used to be that search engines penalized for exact text found in multiple places. If that isn't true or isn't an issue anymore than I apologize for adding the quote tags. Otherwise, I feel they belong. Share this post Link to post Share on other sites
.:Brian:. 0 Report post Posted March 28, 2007 Well this is indeed a very interesting article....I can't wait to see this as it unfolds, hopefully they'll have something that people can see at work sometime in the future. I would assume that they have been working on it for some time now, or else they wouldn't have told people about it....I know that from my experience, I have found that the online translation services are really terrible....in general they are basically the same as going into a dictionary that has both the languages, and then translating each individual word...which in many cases doesn't work, due to differences in grammatical flow of sentences, as well as multiple word definitions...[also, just a side note: shouldn't this article have quote tags associated with it, seeing as most of the text is copied and pasted directly from the original article?] Share this post Link to post Share on other sites
vizskywalker 0 Report post Posted March 29, 2007 Sounds like something done by a neural network. At least from overhearing all the people who were in the neural networks class at my college talk about their projects. It seems like it could be a good approach, but it will still likely have issues with words that can mean many things, and more importantly, slang and connotative only meanings. It could take a lot of data to get this up to a reasonably useful level. What I'm wondering is what if they try mixing the two styles. Using grammar rules and dictionaries on a first run through, use the statistical translation the second time, and then try and merge the two into one sensible document. The third part should be fairly easy relative to the second and first parts, especially when considering how well modern content management systems are at merging.Also, Google's approach sounds (from the name they give to it) similar to the cryptograph solution method of finding the most common letters in the coded message and assuming they are the most common letter in the alphabet, working until it is easily guessed or a contradiction or unlikely situation occurs.It will be interesting to see how this compares to grammar and dictionary based translations (good ones anyway) for grammatically precise well conjugated languages such as Hebrew and Latin. From languages like English to those languages it probably compares well, but I'm interested more in going from those languages to English.~Viz Share this post Link to post Share on other sites
