Arika Okrent reports on a fascinating study about how much grammar people can deduce without explicit instruction:
The answer is: quite a bit. People, even as babies, are good at pulling out grammatical structure from patterned data. But the artificial learning experiments are necessarily small and limited, so it’s unclear how much they can tell us about language learning in the real world.
As it turns out, there has been a large-scale natural test of statistical learning out there all along in the practice of Qur’anic memorization. There are Muslims all over the world who do not speak Arabic (in Indonesia, Pakistan, and Turkey, for example), but who as part of religious practice memorize the Qur’an for recitation, often starting as children and continuing memorization training for years. That training is often unaccompanied by any explicit Arabic instruction or direct translation of the memorized text. They get the statistics of the pattern without the meaning.
A recent paper in Cognition by Fathima Manaar Zuhurudeen and Yi Ting Huang takes advantage of this “natural experiment” to test whether simple exposure to the patterned properties of the Classical Arabic in the Qur’an results in implicit grammatical knowledge. They compared four groups: memorizers who also had classroom Arabic lessons, memorizers with no classroom exposure, non-memorizers with classroom exposure, and a group with no Arabic exposure of any kind.
The groups that had classroom experience had explicitly learned things like what the first person pronoun “I” looks like and how it attaches to verbs, or what the second person possessive pronoun “your” like looks like and how it attaches to nouns. The group without classroom experience but with memorization training had never had these things explained. Had they absorbed the rules of how they worked simply by hearing and repeating them in memorized text?
Yes. The memorizers without classroom Arabic did better than any of the other groups at demonstrating knowledge of the rules. This knowledge was not explicit; they could not explain how pronouns, verbs, and nouns worked, but they could judge whether a sentence they had not heard before was correct or not with accuracy.
Surprisingly, the memorizers with no classroom Arabic did better than those who had had lessons, suggesting that a “top-down approach” that explains the rules of language “may negatively impact learners’ sensitivity to the bottom-up statistics of a language.”
I welcome this as another nail in the coffin of the Chomskyite dogma that grammar must be innate because we couldn’t possibly learn it from simply hearing a language spoken.
Unsurprising reaction from the resident Chomskyite: we don’t have a coffin, and if we did, this wouldn’t qualify as a nail!
We already know that kids learn language without relevant instruction, just from being exposed to the data–plus, we think, the innate properties of their minds that allow them to construct a grammar from the data. The experiment demonstrates that Qur’anic memorizers can also do something like this. No reason for my fellow Chomskyites to be alarmed; we can say the same thing about the memorizers that we do about the kids.
Heh, I was kind of expecting you to drop by and provide the counterpoint! I’m actually glad to have the Chomskyite perspective; while I’ll never accept that position, I’m not quite as rabid as the character I play on the internet.
Surprisingly, the memorizers with no classroom Arabic did better than those who had had lessons, suggesting that a “top-down approach” that explains the rules of language “may negatively impact learners’ sensitivity to the bottom-up statistics of a language.”
It has been my own experience that this is true, in a sense, from a rather early point onwards in language learning. However, the contrast between “top-down approach” and “sensitivity to the bottom-up statistics” is redonkulous, however much it may please devotees of the formulaic phrase.
How about meaning, context and motivation ? Sure, in learning a language various kinds of counting are involved, and of keeping track of counts, contexts and meanings. But this is not a practice of statistics, no more than it is an application of innate rules. Statistics tries to subsume differences under generalizations. To learn a language, you must learn how to cope with the differences in the absence of generalizations.
Note that we don’t necessarily know anything about the quality or approach taken in the classroom instruction. Would it be all that surprising if the formal instruction was so full of bad information (what’s Classical Arabic for “prescriptivist poppycock”?) as to be an affirmative impediment to the task of quickly judging sentences as syntactically well-formed versus ill-formed?
It’s one thing to have people memorize the Qu’ran without learning Arabic, and another to claim that they have memorized it with no knowledge whatsoever of its contents. I very much doubt the latter is true.
It might depend on how vast a quantity is memorized. I learned a handful of songs in French as a kid on a purely phonetic basis — I was singing pure nonsense syllables as far as having any ability to parse them syntactically or gloss them semantically (couldn’t even tell you where the word boundaries fell) but singing the same ones each time. Even where I knew an English alternative version of the same song (e.g. Frere Jacques) I didn’t know how free or loose the translation was. I know just enough French as an adult (picked up haphazardly w/o ever taking a class) that for some lines I learned as pure phonetic nonsense I can now pick out which syllables are probably such-and-such meaningful morpheme. But that’s decades after having learned the text by heart.
Just a few words from avowed non-specialist. The difference between memorizers and non-memorizers was very modest 84% to 73%, which basically shows, that the test was calibrated so that participants could’ve inferred the grammar during the test itself (they had a preliminary warm-up part). Memorizers did better on a test maybe just because they are trained in memorizing and has developed the capacity of retaining large chunks of meaningless text in memory. They could’ve reused this knowledge at the test part.
Would it depend on the type of classroom instruction they received? Ideally, ‘grammar learning’ and ‘pattern learning’ should occur hand in hand. If classroom instruction was closely tied to memorised content, surely there would be some kind of synergistic effect. My suspicion is that too much grammar instruction is pure theory, divorced from actual practice, resulting in abstract knowledge in a vacuum.
But can a chimaera bombinating in a vacuum devour second intentions?
“The answer is: quite a bit” mumble grumble. The answer is specifically, that they’ll get one additional question right out of ten, where the questions are ones that someone with no Arabic can get seven of the ten. You wouldn’t notice the difference between the ‘memorizers’ and the ‘zero Arabic’ if you were just asking questions, not without a scorecard.
So is that “quite a bit”? Matter of taste I suppose. All looks to me like we enjoy telling science stories so we turn a blind eye to the effect size.