Arguments Over Operator Arguments
A piece criticizing a discussion on MT
presented at the NY Academy of Science.
Published by Language International, 1990.
On February 5, Virginia Teller presented a talk with slides entitled Machine Translation: An Idea Whose Time Has Come (Again) at the New York Academy of Sciences as part of their ongoing program in Linguistics and Computer and Information Sciences. I was informed by a friend that Ms. Teller, who teaches at Hunter College and the CUNY Graduate Center, would be describing new advances she had made in this field, and so I came in a receptive mood, ready not only to believe what she might tell me but even to defend her against any possible detractors. This is because I have seen enough good for translators come from computers to believe that more may be on the way.
Her first half-hour contained nothing astounding: it was essentially a recap of MT history, from early growth to ALCAP Report gloom to present progress, followed by some rudimentary distinctions between types of computer-translator interfaces, pre-editing, post-editing, and the like. She also presented many of the time-honored examples of ambiguity in language and voiced the usual disclaimer that "Fully Automatic High Quality Translation" was not a likelihood in our time. But the full hall was already restive, and listeners began to pepper Ms. Teller with all the usual objections to MT, so much so that I was, as I had supposed, obliged to defend her to some extent.
Then something unexpected happened. After she carefully explained her working concepts of operator-argument grammar and the equivalence of sub-language patterns between two languages (in her project Japanese and English), she laid bare the statistical underpinning of her method. This took the form of a comparison of two texts, one in English and one in Japanese, whose results, she insisted, proved there was really not that much difference between the two languages, so that her approach was likely to work on far more extended translations between them. With seeming pride she pointed out that the two texts chosen for statistical comparison were a brief computer manual in English and a translation into Japanese of the same manual, made by humans at the computer company at a prior date.
At this point something strange happened to me, and I suddenly found myself joining the critical multitude rather than the dwindling minority of her defenders. I heard myself saying that I didn't think the two texts had been well chosen for her purposes, that they were unlikely to illustrate the real problems of translation between the two languages. I pointed out that when I had assigned my own class in translating French computer texts anything that had been translated from English, they spotted it immediately and even identified it as "garbage," since it presented few of the challenges offered by native French texts. I suggested to her that the results of her analysis were representative not so much of the differences between English and Japanese as of those between English and Nihonglish. Needless to say the speaker was not enthusiastic over this.
During refreshments afterward, I was forced to join the majority of those present who felt that nothing new had really been presented, no breakthroughs had been made for Machine Translation, and the only ones richer for this work were likely to be the speaker and her colleagues, if they received funding for the continuation of this project.
This article is Copyright © 1990
by Alexander Gross. It may be
reproduced for individuals and for
educational purposes only. It may
not be used for any commercial (i.e.,
money-making) purpose without
written permission from the author.
to linguistics page