The way they tested this was a little strange. They had him looking at a computer screen, and tracked his eye movements. Then they put on the screen some strange abstract shapes, that I guess were supposed to represent the "categories" of objects he was supposed to learn (but I was holding him and watching the screen, and I have no idea what was similar or different about the objects). The idea, I think, was that as he got used to the "categories", he would spend less time looking at things in that category, but when a category he hadn't seen before comes up, he'd spend (relatively) more time looking at it.
At least, I think that was it. Some grad-student type person ran the experiment, and didn't have access to our specific data (it was all just fed into some program). Which I guess is a good research method, but frustrating for me as I try to figure out what was going on.
There also seem to be a bunch of innate confounding variables in the experiment. For example, the kinds of families that have the time and resources to spend an afternoon taking their child into the campus for the experiment on a Wednesday afternoon for no money kind of self-selects your pool a bit. Or the fact that babies are pretty easily distracted and may look someplace randomly.
I think these are pretty typical problems for social science research, because they didn't seem to worried about them. It's a very different world from what I'm used to, where I can make a computer program do pretty much exactly what I want it to (and if I can't, it's a flaw that needs to be fixed)