I’ve recently decided that I am going to go to graduate school after attending the California Cognitive Science Conference (put on by the amazing Cognitive Science Student Association at Cal). There’s a lot of schools to look into, though, and I am not yet sure what I want to study. I’ll know it when I see it, though, so here goes the journey through various schools and labs in-
- Language Learning
- Computational Neuroscience
- Biological Neuroscience
- Cognitive Linguistics
- Cognitive Psychology
- Memory, Attention, and Retention
- and above all….
- Learning!
My goal at first is to gain an understanding of who is doing what work. Today I read the research statement and a few recent articles by Michael Ramscar and his crew at the Cognition, Language, and Learning Lab at Stanford. Here’s a relevant quote about what he calls Feature-Label Orientation (FLO):
[He is discussing how children have a problem learning color words, which he believes is because children see way too many colors at once to be able to tell which one is being discussed.]
“To get around this problem, children’s attention needs to be directed to particular aspects of
visual scenes as labels are heard, so that the process of discerning the informative cues to labels can
be simplified (a ball can be simply red or blue, even if the world never is). Language can provide
children with this direction, as long as they already know the names of other things, such as objects,
and as long as language is structured appropriately. To explore these ideas, we conducted a training
experiment with two-year-olds that manipulated the order in which color words were used in
English. What we found was that children who heard object names predict color labels (hearing “this
ball is blue” when shown a blue ball, FL) consistently sorted novel objects above chance levels,
whereas those who heard color labels predict object names (“this is a blue ball,” LF) did not [5].”
WHOA! So noting the object first leads to better learning? Or it comes down to directing the student to the object first (ball) and then noting the new characteristic (blue), vs the reverse where you note “blue” first (could be anywhere), then the “ball”. Makes sense I guess. How are you supposed to draw a conclusion from “blue”, which could be anywhere, being associated with ball. That would make you look for a “blue ball” (as 1 unit), I guess? It seems the second way (LF) leads to more ambiguity in the statement, meaning that it will only work if the student understand the meaning of “blue” already. It’s a test more than a teaching. Very deep implications for my work with students – I find that many have a predisposition to learn the structure of the math but not the meaning (for example, they know how to follow the “steps” of long division but don’t understand what division actually is, what a remainder is, why you are left with a decimal…), and it seems to me that they have learned in an LF style – they have learned that there is a relationship between the words (or between the steps and dividing, here), but not why that exists or why it works that way – no meaning. I have long held that that means that they are missing some crucial information, and thinking about this in terms of FL/LF orientation gives me a solid tool to demonstrate it to them.
“In recent studies I have applied this technique [LFO] for analyzing information structure to other
developmental questions, such as the way children learn numbers. A formal analysis of the problem
of number set identification showed that the limits on set size identification (our ability to “see” a set
of, say, three objects; usually called “subitization”) arise naturally in a discrimination-learning model.
This is because of two factors: 1) the increased confusability of larger sets (a set of two objects also
contains sets of one object, and a set of three objects also contains a set of two, and sets of one
object, etc), and 2) the way sets are distributed in the learning environment (set salience appears to
follow a power law distribution, with one being the most frequent and salient set). These factors
make it easy to learn to discriminate smaller sets, but not larger ones.
As well as naturally explaining how subitization “capacity limits” arise out of experience, the
model makes a novel prediction: Because our training deliberately changes the background rates of
cues that aren’t informative to number words, training children feature to label (FL) on the sets 2, 4
and 6 should also improve their ability to discriminate 3, 5 and 7! This striking prediction was
confirmed in tests [8]. Given that the age at which children master the subitization of small sets is an
important predictor of their later mathematical abilities, I am currently working on developing
educational applications of this successful training paradigm.”
Fascinating! And again, confirms my belief that what counts in learning is amount of experience (and as we saw above, quality of experience). Knowing what parts are actually analyzed and learned in any learning task is critical to knowing what learning outcomes will result – something that the current school reform movement is very interested in. I will look into the applications Michael discusses – I have the sense that they will be very similar to the Montessori approach, which uses the same principle Ramscar is discussing – limit the inputs, be specific in your reference and direct students’ attention, to direct a certain learning outcome – and judging by the results I’ve heard Montessori kids achieve – teaching 1st graders to read at a 5th grade level, mastering long div and fractions in 2nd grade – I am excited to see the results!
The Blog of Frank Kotsianas 