http://searchenginewatch.com/showPage.html?page=3627520
Machines In Translation: Do MT Engineers Dream of Selectric Sheep? - Part 3
By Kevin Heisler, Search Engine Watch, Nov 6, 2007
[ ... ]
Machine Translation: Desktop to Mobile Language
Why did I get addicted to Google Blogoscoped? In part, because it provides the language data that helps decipher patterns in MT and other types of computer language translation. The primary reason? To get help translating Google "big picture" news into actionable search marketing data. I needed to separate the wheat from the chaff.
For example, how will desktop translate into mobile search and local search? On April 7, 2005, Phil announced that Google’s "mighty" Q&A feature can tell searchers who wrote a novel. He gave a number of funny results. In some cases, Google got it wrong. He tested the query: Who wrote Lolita? Vladimir Nabokov. (Correct.)
Google Q&A (natural language query) was first noticed by searchers as an experiment on the Google search engine. Most people accessed the feature via a laptop or desktop computer. When Google Answers went silent, there was an outcry among industry pundits.
Now, Google has translated the Google Q&A concept from desktop to mobile "language." It's not an exact translation. It's a close approximation, though.
What's Google Q&A today? According to Google, Q&A enables Google SMS users to get quick answers to straightforward questions. Need to know the population of Japan, or the author of "Hamlet"? Send any fact-based question or query through Google SMS, and Google will scour their resources to find the answer and cite the source.
Sting can now SMS to find out the name of that book by Nabokov.
Nabokov Blogoscoped
In "Lectures on Russian Literature," Vladimir Nabokov pointed out -- with wry, dry humor -- the challenges humans face translating other humans. His "lecture" was an essay entitled "The Art of Translation."
The question explored: how do translators do evil?
I'll excerpt the opening in its entirety, so readers can see how language has evolved, but the problems inherent in translation remain. The other reason, of course, is fear of Nabokov, who castigates translators who skip words or passages.
"The Art of Translation" was reprinted in "Lectures on Russian Literature" with permission of "The New Republic," who first published the essay in 1941. The estate of Nabokov owns the copyright to the book, and renewed it in 1981. During that 40 year period, computers and machine learning have evolved almost as fast as the meaning of the word "queer."
"Three grades of evil can be discerned in the queer world of verbal transmigration," he wrote. "The first, and lesser one, comprises obvious errors due to ignorance or misguided knowledge. This is mere human frailty and thus excusable."
Nabokov, forgiving human weakness, goes on to describe the reader's descent into translator Hell.
"The next step to Hell is taken by the translator who intentionally skips words or passages that he does not bother to understand or that might seem obscure or obscene to vaguely imagined readers; he accepts the blank look that his dictionary gives him without any qualms; or subject scholarship to primness: he is as ready to know less than the author as he is to think he knows better."
Whether it's an author, or a business person hoping for a solid translation, the problem is simple. The solution is complex. Nabokov saves his greatest wrath for a mistake software would make if tasked with translating literature.
"The third, and worst, degree of turpitude is reached when a masterpiece is planished and patted into such a shape, vilely beautified in such fashion as to conform to the notions and prejudices of a given public. This is a crime to be punished by the stocks as plagiarists were in the shoebuckle days."
The word planished means "to smooth (metal) by rolling or hammering."
Google, in an attempt to uncover a searcher's intentions, poses the helpful question, "Did you mean: punished?"
Yes, in a way. Nabokov goes into great detail describing how works of Russian literature have been punished by translators. He does it in a humorous way because the subject would otherwise be unreadable. In any language.
Key takeaway: statistical machine translation was never intended to translate literature. Nor does MT have designs on replacing U.N. translators. As a business tool, machine translation is an incredibly valuable tool for businesses expanding internationally.
The device that provides access to machine translation -- whether Google, Yahoo, MSN, Ask or other language search engines -- doesn't really matter. What does matter: whether the source can be trusted.
Until android dreams become real, all eyes are on the wrong prize.
Next time: Russian Brin's all in, Ochs talks, PageRank trust, and more LI vox populi.