Interesting Misreadings
In John & Belle Have A Blog: The Advantages and Disadvantages of Adaptation For Life I read a passage which I thought went
Show a cobbler a cow. Note his trade union obtuseness in relation to all kine! He simply cannot see how fine they are! ‘Ah yes,’ he will say, ‘there’s many a fine pair of shoes in that animal.’ Show this or that parrot an equestrian statue and he will say ‘Hah! Pretty big job that….
Which I thought was pretty funny, in an almost Pratchett-esque way. When it continued “That’d take the 24-foot ladder and a double-handled gauge-4 saw.” I realized that I must have made an error, and backtracked, to find that it was “this or that patriot” not “parrot.”
Misreading is interesting, not only in that it often produces funny sentences, but the way it points up facts about how we read. I mentioned that I’ve been reading Steven Pinker’s The Language Instinct (and becoming more dissatisfied as I go, but that’s a different post), and I think that misreading presents a problem for the way that he describes our internal mental grammars as operating. Now, I’ve got no training in formal linguistics—I just picked it up the way all the kids did, in the street—so maybe I’m talking through my hat. In fact, I probably am, but I might as well toss this out there.
The whole surface-structure is consumed and then transformed to deep-structure where it can be mentally aprehended thing seems like it needs to be a linear process, where the entire sentence needs to be consumed before its transformed, unless we’re wildly trying to transform each substring so far into every possible d-structure that could have yielded it. One problem with the consume and then transform model is that we seem to be able to notice misreadings that make no semantic sense before we’ve read enough to tell whether the sentence as a whole makes grammatic sense, but if I understand Pinker it’s the d-structure not the s-structure that determines whether something is grammatical and what its semantic sense is.
So since we’re apprehending the semantics before we’ve finished with the s-structure, it would seem either that our semantic apprehension is operating directly on the s-structure, or it is operating on partial, multiple, d-structures without our noticing the multiples that we’re subconsciously considering as we read. A problem with the latter is that we do notice it if a clause has two possible d-structures in other cases, as in Pinker’s example “discuss sex with Dick Cavett.” We’ve read “discuss sex” and as soon as we read “with” our partial d-structure calves, becoming (at least) two partial d-structures both of which are under consideration from that point out, and both of which are the products of transformation rules applied to tree structures that are still missing nodes. Maybe Pinker would regard it as perfectly obvious that’s what happens, but if that’s the case, it’s hard for me to see probabilistic super-rules operating on partial input being much more elegant or powerful an explanation than the “word-chain” explanations that Pinker derides.
And yet, if that’s not the case, if we can figure out the semantics on the s-structure as we go, what do we need the d-structure to explain? Am I completely missing the boat on what role Pinker thinks the d-structure plays?
Another aspect of misreading is how it demonstrates that reading is an active process, not a passive reception of input. (Not news in a lot of ways, but it seems curiously missing from Pinker’s account of Chomsky’s theory, where the transformation from s to d seems to take place in a completely abstract mathematical space) Consider the old phrase
Paris in the
the spring
Many people, particularly when first encountering it, don’t notice the extra “the” which makes it ill-formed. I think it’s generally accepted that it gets skipped because our minds are actively anticipating, based on what has come so far, what parts of speech are likely to come next. It seems to me, though, that’s a curious fact from the point of view of d-structure. Maybe it’s blindingly obvious to the language module that given the d-structure(s) so far from “Paris in the”, there is no transformation rule whatsoever that could make a determiner legal as the result of transformation on the next token. Seems like a lot of processing power to deduce that, but ok. So why doesn’t it get noticed and flagged as odd the way
Paris in the
run spring
does? From the point of view of s-structure, no problem (I think–I could be wrong), it’s just a repeated token. We know we don’t read every word, and that we have to make a special effort to read things as they really are, rather than as we anticipate; moreover it seems obvious that anticipation is, must be, of s-structure. From the point of view of d-structure, with the capacity of transformation rules to rearrange and insert things according to universal grammar, I would think it puzzling that the next thing is a “whatever the token is that results in a determiner of the reference-to-a-unique function type” is more invisible to the reader than any other illegal term, just because the last thing assigned to some part of the tree was the same kind. Again, maybe everybody who knows anything about linguistics knows that this is a non-issue, and there has to be substantial processing of syntax at the s-structure level to explain this sort of thing, but nevertheless there is something real and important going on that can only be explained by d-structure analysis. So far, though, I’m more puzzled than illuminated by this theory.
June 11th, 2004 at 11:59 am
The whole surface-structure is consumed and then transformed to deep-structure where it can be mentally aprehended thing seems like it needs to be a linear process,
Classical generative grammar is explicitly not a theory of parsing - it is merely a description of the grammatical sentences in the language (what was called “comptence”), not any of the mental processes that actually process sentences (”performance”).
You may think this is a crock (I do) but it is not the crock you’re looking for.
June 11th, 2004 at 1:49 pm
Really? You’d never know it from The Language Instinct. Pinker definitely seems to be presenting it as “this is what the mind does”, not “here’s one possible abstract description of permissible sentences.”
June 11th, 2004 at 8:11 pm
I would be interested in why you didn’t enjoy The Language Instinct. I found it very interesting and persuasive. I have also ordered Words and Rules: The Ingredients of Language and am waiting for it to be delivered either today or on Monday.
June 14th, 2004 at 7:05 am
Pinker himself is a psychologiste, and his favourite exercise is to make up Just So stories to account for the evolution of psychological modules alleged to underlie Universal Grammar. Chmosky would certainly be more into the psychology if it hadn’t been for the persistent tendency of experimental evidence not to support his hypotheses.
June 21st, 2004 at 10:15 am
It’s been a long time since I read Pinker and I probably had no nuanced understanding of his model of d-structures and s-structures anyway, but how relevant is “Paris in the the spring” to “language”? I thought that reading and writing were supposed to be a meta- or para-linguistic process, with lots of funky cognitive quirks of their own. To get a clear counterexample to Pinker you’d need to come up with a purely heard/spoken case.
Which isn’t to say that reading and writing can’t contribute to some useful hypotheses. Another impression I had is that there’s very little in cognition that is linear. So I would expect all sorts of mutual interference among syntax, semantics, and the look and sound of words before we settle on an interpretation, if indeed we do settle on one. In actual conversation, my subjective sense of it is that we can keep multiple possible interpretations going in our heads at once, and backtrack to adjust their relative probabilities based on new information.
I must say that I’m impressed by how you and some of the other self-taught linguabloggers can intelligently debate this stuff. When I read Pinker, say, I know just enough to point out the occasional crock but not to debate the details.
June 22nd, 2004 at 4:03 pm
It’s probably not that relevant, but it was the first example that came to mind of a clear case of cognitive interference where what we know about the language gets in the way of perceiving what’s clearly there. A better example is probably mondegreens. I think Pinker is right that the fact that mondegreens are often a less likely interpretation than the actual one, and sometimes even use nonce words, spells trouble for a naive theory of probabilistic matching. On the other hand, it cuts both ways. Since everybody presumably has pretty much the same dictionary when it comes to the words “the”, “this”, “sky”, and “guy”, and Pinker argues that the rules are pretty much built-in except for some parameters the fact that most people hear it the right way and the occassional person consistently hears the mondegreen seems to indicate that we’re bringing enough idiosyncratic psychological state to the table to guide the low-level identification and parsing of words, at least some of the time.