A well-spent afternoon...
Jul. 7th, 2001 11:23 pm![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png){kind=small}
alexpgpI went home at noon, leaving Drew in charge of the store. After a quick lunch, I sat down and attacked the first translation.
The document was a letter addressed to a member of my client's staff. What was unusual about the file was that a translation was prepended to the Russian text. The recipient had sent the document out for translation because the English translation was, um, difficult to read.
And no wonder. It's a machine translation.
The translated letter opens:
In other words, the state of the art is progressing.
Many of my colleagues dismiss machine translation with derision, maintaining that a computer program will never be able to do as good a job as a human. Maybe so. Maybe that ought to be slightly modified, to read: "as an experienced human translator."
But will it be good enough?
By analogy, let me zip across the epistemological firmament and talk a bit about chess.
The late Claude Shannon described the overall design of a chess-playing computer back before 1950. Practical implementation of the idea was, of course, a ludicrous notion at the time.
"A machine would never be able to play chess," said the skeptics, "or at least, not a creditable game of chess. Certainly never a good game of chess." All sorts of reasons were given, the most convincing being one that borrowed a bit of chess lore about the number of possible ways the first dozen or so moves could be played. That number has a lot of zeros in it and can only begin to be comprehended by noting that if each atom in the known universe made one move every second, then all of them together, working in parallel, would require several billion years to exhaust all the possibilities... just for the first dozen moves or so.
Then, around 1970, a bunch of students at MIT (if memory serves) were allowed to enter a computer program they'd written - named MacHack IV - as a contestant in a USCF-sponsored chess tournament. Out of six games, it lost five and drew one; its first provisional rating was 1100 or so, which put it at the rank beginner level.
But it drew a game against a human who was not particularly interested in making academic points, but caissic ones.
The skeptics drew back. "An 1100 player is a patzer. Chess programs will never beat a master...certainly a program will never beat a grandmaster."
I remember sometime in the 80s, the English International Master David Levy bet something like 20,000 pounds that, five years hence, no program would exist that could beat him. Five years later, he won the bet, but declined to renew it.
A couple of years ago, a massively parallel computer - "Big Blue" - put away Garry Kasparov, the then-reigning World Champion.
Machine translation has come a long was since "out of sight, out of mind" was rendered as "blind lunatic" back in the early days. The text I saw today was unintelligible to the average user, but as I examined it, I saw the spark of something.
Is my job safe for the foreseeable future? I think so, and probably for the next 20 years as well. I just have to stay... one... step... ahead.
By the way, I got the job done in plenty of time to go with Drew and Shannon to Durango to stock up on staples. Among other goodies, I got the 7.1 version of Red Hat's "Deluxe Workstation" software, which is installing right now as I lay these lines down on phosphor. I decided to play it safe and have the installer reformat the disk and do a disk scan. Unlike the Mandrake product, the overall install time (excluding the format) has not risen above 34 minutes for the whole shebang, and there's only about 5 minutes left in the process, according to the screen.
Tomorrow will be devoted to clean-up and the other assignment I have to do. I've already asked Drew to be prepared to handle the store single-handedly on Monday and Tuesday, while I attend to translations. He responded by reserving Wednesday morning for a trip to Durango with Shannon and her mom.
The install is almost complete... gotta go.
Cheers...
The document was a letter addressed to a member of my client's staff. What was unusual about the file was that a translation was prepended to the Russian text. The recipient had sent the document out for translation because the English translation was, um, difficult to read.
And no wonder. It's a machine translation.
The translated letter opens:
I write you about one problem irrelevant with subjects of our current correspondence. Above this problem, I have though some time ago. My opinions I has impart to Mr. A and Mr. B. It would be desirable also to discuss this questions with you.After I read the original Russian, this is what I produced:
I am writing to you about an issue that is not related to the themes of our ongoing correspondence. I got to thinking about this issue some time ago, have shared my views with Mr. A and Mr. B, and would like to discuss them with you as well.What I find noteworthy about the machine translation is that, although it is pretty bad, it is not as bad as what one could expect of software from, say, five years ago.
In other words, the state of the art is progressing.
Many of my colleagues dismiss machine translation with derision, maintaining that a computer program will never be able to do as good a job as a human. Maybe so. Maybe that ought to be slightly modified, to read: "as an experienced human translator."
But will it be good enough?
By analogy, let me zip across the epistemological firmament and talk a bit about chess.
The late Claude Shannon described the overall design of a chess-playing computer back before 1950. Practical implementation of the idea was, of course, a ludicrous notion at the time.
"A machine would never be able to play chess," said the skeptics, "or at least, not a creditable game of chess. Certainly never a good game of chess." All sorts of reasons were given, the most convincing being one that borrowed a bit of chess lore about the number of possible ways the first dozen or so moves could be played. That number has a lot of zeros in it and can only begin to be comprehended by noting that if each atom in the known universe made one move every second, then all of them together, working in parallel, would require several billion years to exhaust all the possibilities... just for the first dozen moves or so.
Then, around 1970, a bunch of students at MIT (if memory serves) were allowed to enter a computer program they'd written - named MacHack IV - as a contestant in a USCF-sponsored chess tournament. Out of six games, it lost five and drew one; its first provisional rating was 1100 or so, which put it at the rank beginner level.
But it drew a game against a human who was not particularly interested in making academic points, but caissic ones.
The skeptics drew back. "An 1100 player is a patzer. Chess programs will never beat a master...certainly a program will never beat a grandmaster."
I remember sometime in the 80s, the English International Master David Levy bet something like 20,000 pounds that, five years hence, no program would exist that could beat him. Five years later, he won the bet, but declined to renew it.
A couple of years ago, a massively parallel computer - "Big Blue" - put away Garry Kasparov, the then-reigning World Champion.
Machine translation has come a long was since "out of sight, out of mind" was rendered as "blind lunatic" back in the early days. The text I saw today was unintelligible to the average user, but as I examined it, I saw the spark of something.
Is my job safe for the foreseeable future? I think so, and probably for the next 20 years as well. I just have to stay... one... step... ahead.
By the way, I got the job done in plenty of time to go with Drew and Shannon to Durango to stock up on staples. Among other goodies, I got the 7.1 version of Red Hat's "Deluxe Workstation" software, which is installing right now as I lay these lines down on phosphor. I decided to play it safe and have the installer reformat the disk and do a disk scan. Unlike the Mandrake product, the overall install time (excluding the format) has not risen above 34 minutes for the whole shebang, and there's only about 5 minutes left in the process, according to the screen.
Tomorrow will be devoted to clean-up and the other assignment I have to do. I've already asked Drew to be prepared to handle the store single-handedly on Monday and Tuesday, while I attend to translations. He responded by reserving Wednesday morning for a trip to Durango with Shannon and her mom.
The install is almost complete... gotta go.
Cheers...
