Showing posts with label central bank statements. Show all posts
Showing posts with label central bank statements. Show all posts

Monday, February 13, 2012

Machine Translation, "The Shining," and My Crush on a Faceless Canadian Bureaucrat



Homer: So what do you think, Marge?  All I need is a title. I was
 thinking along the lines of "No TV and No Beer
Make Homer" something something. Marge: [timid] "Go Crazy"?
Homer: Don't mind if I do!
--The Simpsons, "Treehouse of Horror V"



Let me begin with an impossibly obscure instance of centralbankese. A policy statement is the kind of text that “Low Quality Translation” theorists say is ripe for computerization because it is repetitive. Check out this little quatrain, taken from a recent statement put out by the Bank of Canada:



To the extent that the expansion continues and the current material excess supply in the economy is gradually absorbed, some of the considerable monetary policy stimulus currently in place will be eventually withdrawn, consistent with achieving the 2 per cent inflation target.

After puzzling over this monstrosity for several minutes, I realized that most of the cognitive dissonance for me came from “current material excess supply” (it just rolls off the tongue, doesn’t it?). Let us run it through Google Translate:

"En la medida en que la expansión continúa y el suministro de material excedente actual en la economía se absorbe poco a poco, algunas de las medidas de estímulo monetario considerable actualmente en vigor con el tiempo se retiró, en consonancia con el logro de la meta de inflación del 2 por ciento."

GoogleT is correct in providing “en la medida en que” (I would have used “en la medida que,” but you have to give some points for Google Translate); the verb tense in "continúa" needs changing to the subjunctive "cointinúe"). All in all, the first snippet of the MT version is OK. But the GoogleT sentence gets more and more garbled after “current excess material supply.” What the phrase means, of course, is “current and significant excess supply.” “Material” here is an adjective (GoogleT guesses incorrectly that here it is a noun). But even if GoogleT had guessed right, it would still have a mountain to climb. The question here is: “excess supply” of what? Is it monetary supply or unused capacity utilization or unsold inventory? Or all of them? My hunch is that it refers to the supply of money, but that is a decision I would have to consult further.

The “Crap Quality” ideologue believes that this messy human element (the undecidability of this hyeroglyph) can be elided in the case of “non-literary” texts. And the solution goes something like what I am about to describe. Imagine, for example, that we take all of the central bank policy statements written since the dawn of time (yes, back even to the times of the Roman Imperial Central Bank) and take all of the translations of those statements and upload them into the Universal Central Bank Policy Statement Translation Engine (the UCBPSTE©). Chop it all up and run it through the algorithm.

Is the translation still deficient? “No problemo,” as Americans say. Just take every single text that has ever been translated since the time of Cain with its original and cram both into the engine. What? Still garbage? Nothing doing. Let’s get large crowds of hamsters to translate all manner of texts to enrich the database. Eventually, given enough data (pant, pant), we will have an answer for our troublesome “current material excess supply.” 

I am here to tell you that this is unlikely. That forgets two problems. The first (and lesser one) is that central bank policy statements are willfully obscure. The Canadian bureaucrat who wrote the unfortunate “current material excess supply” is not just a bad writer. He practices to be a bad writer. He will be cranking out impenetrable hieroglyphs until the universe stops expanding and collapses back upon itself and we are all crushed as matter expands further and further to create an infinitely cold, infinitely empty void.

The second (and more important) obstacle is the lack of repetitiveness in something as complex as language and history (two areas not bound by strict rules). The “Crap Quality” people talk about the Content Tsunami, right? That means there should be a large enough (and ever increasing) corpus to solve most translation conundrums. To gauge the scale of the challenge, let us ask ourselves how many times in the history of humanity have the words “current material excess supply” been strung together by a hominid with opposable thumbs. Let’s do a Boolean search on the Internet using quotation marks to get exact hits for the phrase.

Aha! We get 21,000 hits. This bodes pretty well for computer-capturable repetitiveness, since it increases the likelihood that this tough nut of a phrase was translated.

Alas! Our “Crap Quality” theorist will feel a distinct pang of disappointment. And perhaps an extreme terror, not unlike Jack Nicholson’s wife in The Shining as she inspects her hubby’s literary handiwork. Yes, it is as if the Internet has been maniacally scribbling “All work and no play makes Jack a dull boy” over and over again. (Oh, faceless Canadian bureaucrat, I love you!) The 21,000 instances of “current material excess supply” come from the same sentence quoted 21,000 times. The point is (if it needs explanation) that “current material excess supply” is made up of non-problematic components (taken singly); the problem is that when you string them together, they create a unit that is not susceptible to statistical deciphering. The further point (and the objection to “Lower Quality Translation”) is that there are potentially infinite strings like “current material excess supply” out there, unspoken. And they are far more frequent than engineers who have never translated a word in their life think. So a rose is a rose is not a rose.