thomas leverett weblog

The vodka is good, but the meat is rotten (part 2)

When you translated, by machine, from English to Russian, "the spirit is willing, but the flesh is weak," then, turn around and translate the same expression back into English, you would get "the vodka is good, but the meat is rotten." This was used for many years to demonstrate that machine translation would never pick up the nuances of metaphorical meaning.

Metaphorical meaning is in fact a hard part of machine translation, but perhaps not as hard as you would imagine. It is not impossible, for example, to notice that most of the time, when the four words the+spirit+is+weak appear together, there is generally a metaphorical meaning. It is not impossible to identify that meaning. It is not impossible to get the computer to just deliver the metaphorical meaning upon command.

A whole supply of metaphoric expressions work this way; for example, "off the hook", "up the creek without a paddle," or "at loggerheads with each other". The computer looks foolish if it tries to translate one word at a time; you might lose that metaphoric meaning altogether. It becomes humorous to try to find metaphors in the jumbled productions that machine translate gives you. Songs, which are full of metaphor, are the worst; try crunching a few of them through a machine translator.

So this brings to mind several process that inevitably follow from this weakness. First, somebody has to go through the language, word for word, expression by expression, and make it so the computer can identify metaphoric ones with reasonable accuracy. Some, for example the one with the creek and the paddle, appear in pieces, or with different word combinations. Some, such as "off the hook" could have been written, originally, in absolute seriousness, with no metaphoric meaning whatsoever. The machine has a problem here, which is to make a statistical guess at the likelihood that you have a metaphoric meaning. But fortunately, machines are good at statistical guesses. A concordancer can give you the last 1000 times anyone used "at loggerheads with each other" and tell you with certainty how many of these actually had to do with sea turtles. Of course, the best metaphors are suggestive, and mean only what they say, but suggest other things, or don't, depending on the feeling of the writer. You could accuse me, for example, of making the title of this post metaphoric. And I could deny it. How is the computer supposed to know if I, the writer, don't know? Or lie?

So, there are times that the best the computer can do is point out that this is often used metaphorically or this could be metaphoric. Let the reader decide. Give the reader the whole picture. That's called good, careful reading; getting the whole picture, and deciding yourself. By the way there is another phenomenon that I should point out. Businessmen who use Google Translate, for example, for every trade, get out of the habit of using metaphors. In effect they say, "Well, I would use 'off the hook,' but the computer will butcher that (so to speak), so I'll just say, it's not your responsibility". And so their language and style actually change to fit the medium. Computer-translated language becomes less metaphoric naturally and by necessity.

But back to the main point, which is that what appeared to be a fairly inviolable line (computers will never be able to think metaphorically) turns out to be quite fluid, easy to beat, left behind (in the better programs) in a trail of dust. The better programs are flagging metaphoric expressions and identifying them as such. These programs have other problems, some of which impede meaning, but that isn't one of them.

It brings up the question of what happens when language learners stumble upon these machines, and use them in the process of learning. It's the teachers, by the way, who have trouble recognizing when a given body of jumbled up text has been created, essentially, by a computer dealing either with a student's original work or some other body of writing. Generally nobody has told the student that this is wrong, illegal, or whatever. Generally the teacher looks at the high-level vocabulary (translated directly from native words), the jumbled-up word order, metaphoric gaps, etc., and gets angry. How could a person of such limited awareness (of sentence structure) use such big words? Or, how can I possibly unravel this to figure out what you mean? Once I had a student who, asked to produce an English newspaper article, produced one that had run through the machine from Chinese. I looked at it and knew instantly that it was not standard English. But he genuinely didn't see the problem. One article full of English looked pretty much like the other to him: incomprehensible. He generally produced work similar to what is mentioned above, and he had flunked more than once up to that point, probably having made some teachers quite angry. It was his problem that he was at the cutting edge of the new technology.

I would call these learners false positives in the sense that they show great sophistication in areas where they really don't have the skill. The skill of turning high-level words into computer-generated glosses is a computer skill; often they have that skill, but not the one where you look at a simply misformed sentence and know it's wrong. And sometimes they'll slip each piece through several machines, hoping that more translation can only improve it. Start with google translate, then go through Ginger or Grammarly, and see where you end up. Teachers, unfortunately, are generally too busy to foresee what could have happened, or why they have a student with a paper that he or she can't even read. To the teacher it smacks of plagiarism, cheating, paper-buying. But it's not. The computer is like the purple pen, the fancy paper or the pretty font. It's part of the environment, and it's not labeled with a sign that says, "this tool will discourage your language learning" until we label it that way and actually prohibit it. But even then, even the phones are picking up skills. Why can't we just tell the phones what the problem is?

Labels: google translate, internet, writing

posted by tom @ 11:34 AM