By Dr Zafar M. Iqbal
TCCI, Chicago, IL
Using the vast database, accumulated over the years, by international organizations like UN and EU which depends on professional translation of one language into another in the course of routine business, Google Translate has in the last four-five years developed a digital system for machine translation. It currently supports translations mostly in prose, to and from over 70 languages.
The ‘supported’ languages include Arabic, Persian, Turkish, Hindi and Chinese (in addition to French, German, Russian) but not Urdu, Bengali, Tamil or Telugu and other ‘alpha languages’ in which translations are less reliable now, though the improvements are reported to be furiously underway.
Translating services like Yahoo!, Babel Fish, among others, use an older system, Systran, founded in 1968 and used by US Department of Defense, mostly for translating Russian during the cold war. By the way, Urdu is among the languages Systran-7 supports.
Google Translate is a statistical machine translation system, which David Bellos of the Program in Translation and Intercultural Communication at Princeton, says “looks for similar sentences in already translated texts [by human translators] somewhere out there on the Web. Having found the most likely existing match through an incredibly clever and speedy statistical reckoning device, Google Translate coughs it up, raw or, if necessary, lightly cooked” but “it doesn’t try to unpick or understand anything.” It does not parse a sentence the way learned to do early in a school. Google Translate does not supply us the ‘correct translation’ but, according to Bellos, “only an expression consisting of the most probable equivalent phrases as computed by its analysis” of the database. This is unlike any other known language which has a lexicon and a grammar to generate infinite number of sentences.
This is also quite a leap from what Warren Weaver, the ‘creator’ of machine translation, had to depend on during the cold war for cracking the “code” in Russian language, using crude translating machines.
Bellos found the Google Translate’s rendering of a French phrase for ‘love’ from ‘Les Misérables’ (“On n’a pas d’autre perle à trouver dans les plis ténébreux de la vie”) to “There is no other pearl to be found in the dark folds of life,” was “creditable.” The reason it was “creditable” is probably because, as Bellos says, it also “happens to be identical to one of [its] many published translations.” However, the opening sentence of Proust’s “In Search of Lost Time,” was translated as “Long time I went to bed early,” which is not grammatical.
For Google Translate to work and work effectively and be “creditable,” first, there has to be a reservoir of human translations that Google Translate can tap into and then, find the most appropriate match. This obviously means: the more careful the human translations, the better Google Translate product. It is by analyzing patterns in hundreds of millions of documents already translated by human translator that Google Translate can come up with and select intelligent guesses to the appropriate translation. This “statistical machine translation" seeks patterns in available reservoir of the text database. Not all such translation can be expected to be perfect, and both the quality and accuracy in these translations often vary. Obviously, it all depends mostly on the extent of the human-translated documents available to Google Translate in a given language for finding a better quality and accuracy of translations.
Google Translate is proven very useful in many areas, but to translate and interpret truly original literary work into another language at similar quality level maybe too challenging for any machine translation at this time. This is not such an easy feat for most of the professional human translators, either.
Despite the progress being made, says Franz Josef Och, head of this translating service, "the trajectory we are on just doesn't seem likely to reach artificial intelligence." Other Silicon Valley experts believe, however, that “we are not far away from having a real-life version of HAL from the movie2001: A Space Odyssey,” and some day we may be able “to download our consciousness into” our computers.
When, however, the Google Translate gets “patchy,” as cited above, in translating French prose of Hugo and Proust to English, it may not be quite ready yet to render Urdu poetry and that too of Iqbal, with all his culturally-infused metaphors, linguistic charisma and depth, into English or any other Western language any time soon.
So, with no digital relief on the immediate horizon, we’d continue in the meantime to wonder how best to convey ‘Jigri doosth’, ‘Khoon-jiger’ or ‘Gray-baan chaak’ in a word or phrase in English, without losing any linguistic or cultural nuances.