Friday, April 18, 2008

Lost in Translation — Um . . . Disfluencies and Translation

Let us take a moment to revisit the idea of scanning brain activity to guide the translation process. I suggested that the UT may be able to detect thought patterns that correspond to certain categories of words - such as nouns, verbs, or specific kinds of nouns (people, animals, food, numbers). I recently came across an interesting book that indicates that the human brain actually does work that way.

The books is Um . . . Slips, Stumbles, and Verbal Blunders, and What They Mean, by Michael Erard. The book compiles historical, literary, anecdotal, and linguistic evidence that show the inner workings of the mistakes we make when speaking. In the process of studying these speech disfluencies the underlying structure of language and thought patterns begins to emerge. Erard explains:
For instance, when you accidentally swap one word for another, you always choose a word of the same part of speech, which is why I heard someone say, "That's the cake on the icing" (where two nouns swap) and not, "That's the on icing the cake" (where a noun swaps with a preposition). Linguists use this regularity as evidence that before a person utters something, the brain is constructing a sort of frame with empty slots for words, and words can fit only into the slots designed for them. If they don't fit, they're kicked out, or the speaker notices the impending error and stops the sentence to revise it. (pp. 64)
These patterns hold true across all languages that have been studied. The mistakes people make in speaking - saying the wrong word, the wrong sound, the wrong name, regressing and starting over, the placement of ums and uhs - always follow the rules of grammar of whatever language the person is speaking. The book even shows how studying these mistakes in children has allowed linguists to create models of how our ability to speak develops.

When you read the book you quickly start to see these patterns all around you. For example, my father comes from a family of seven children. My grandfather was famous for running through all of their names, the names of the neighbor kids, and the names of the family pets before coming up with the name he was actually trying to say. But while he might say Mark, Martha, John, David, or even Sparky instead of Mike, he would never say Gravity or Spatula by mistake. His brain was working with a slot for proper names, so it would only allow proper names.

This suggests that a device such as the UT would have some chance of working. Detecting the patterns manifest in these pre-spoken "slots" could be a very useful guide to translation.

Erard also discusses how verbal slips rarely occur more than a few words apart. This is because we do not speak word by word, nor do we speak sentence by sentence. We speak in chunks of words, and certain speech disfluencies consistently occur in the breaks between these chunks. Understanding this would help alleviate (although not complete solve) the difficulty of real-time translation (see the section “Don’t Wait for the Translation, Answer Me Now!”). Rather than having to wait for entire sentences, there might be detectable mental chunks that could be processed and translated before the rest of the sentence was even spoken.

So category based thought patterns might actually be . . . um . . . might actually be a real possibility.

You can learn more about Um at the official website:

NEXT: Lost in Translation — There Be Whales Here