This is my report of the 3rd day of the Annual Meeting of the Cognitive Science Society 2007 in Nashville, TN

Day 3: Saturday, August 4th

This was the third and final full day we attended the conference. There were no plenary talks, just normal sessions, and we were able to make the final poster session.


Session 8-04-1A: Language as a Dynamical System: In Honor of Jeff Elman 10:30am-12:30am

This session followed up on some of the ideas from Elman’s plenary talk the evening before, and all the talks were given by researchers who had been influenced at one point or another by Elman’s work and ideas.

The first speaker was Gerry Altman, who talked about the key roles of prediction and context in language. He described experiments in which subjects were presented with a visual scene, e.g., with a woman near an empty table and a wine glass on the floor. I can’t say I followed it all perfectly, but I think the gist was this: If subjects were presented with a statements indicating that the woman will move the glass to the table in order to fill it with wine, then they will tend to make anticipatory eye movements to the place where the glass would be on the table if it were moved. I forget the sequence of presentation, but at one point the experimenters replace the image with a white screen, and measure saccades in reaction to statements about future states of the glass. Altman drew the conclusion that visual and language processing share a common substrate, and that they are intimately coordinated with one another.

Mary Hare was up next. I have to admit that I didn’t follow her talk very well at all. It seemed overstuffed with unfamiliar jargon, and I ended up thinking more about the previous talk than focusing on the current one.

Ping Li was up next, discussing lexical competition and organization. He again referenced the DevLex model, and the work presented previously at the conference with DevLex II involving second-language acquisition. He also discussed neuroimaging studies that corresponded to the modeling data. I believe the reference was Chan 2007, showing patterns consistent with early L2 learning as more unified in brain structure, and late L2 learning displaying a more fragmented pattern.

Next was Ken McCrae, who presented a connectionist approach to hierarchical categorization. He presented an attractor network in which there was no explicit hierarchy in the topology, and demonstrated that hierarchy can emerge through training. I have to admit finding his approach a little strange, in the sense that we have very strong evidence that the cortex is arranged hierarchically, so what necessarily is the point of demonstrating that a network without hierarchical structure can represent a hierarchy? Unless he thinks the cortex is a disorganized, undifferentiated mass and that organization arises through interaction with stimuli. I don’t know.

Kim Plunkett spoke next about categorization in infants, and how labeling facilitates categorization, i.e. it’s easier for infants to categorize the world if they have labels for those categories. It looked like interesting stuff, but I had to duck out early, so I didn’t catch it all.


There were a number of publishers there with sample journals and new book releases. Many of the journals were complementary. I actually found an article in a sample of Neural Networks that was related to large-scale modeling of the neocortex, and since none of the first afternoon sessions really grabbed my interest, I spent that time reading the paper.


Session 8-04-3E: Development and Objects 3:30pm-5:00pm

This was one of the best sessions of the conference, so it was a nice one to end on.

The first speaker was Heidi Kloos who spoke on the transition from using perceptual knowledge and conceptual knowledge to make predictions. The idea in the experiment was to use shadows as cues to motion and determine whether children (aged 3-4) were relying simply on visual information to predict the paths of objects, or the conceptual knowledge that shadows are linked to the objects that cast them. The setup involved a video game in which subjects were presented with a display of an egg which could travel one of two places: diagonally up toward a chicken and horizontally toward a duck. In one condition the egg’s shadow was directly beneath it, in the other, a mouse that was roughly the same size and shape of the shadow was beneath the egg. I believe the way it worked was that the subjects needed to try to predict the movement of the egg based on the movement of the shadow or mouse. They found that the two groups of children, a 3 yr. old and 4 yr. old group exhibited no difference in performance between the shadow and mouse conditions, while adults were much better in the shadow condition. The researchers conclude that there is a transition at some point after four years old where humans are able to integrate conceptual knowledge in order to make better judgments about the world. In the Q&A someone asked what the nature of that change might be. Kloos really couldn’t answer that, but it’s a good question.

Next was Maria Chen, who tested infant ability to track multiple objects and distinguish between numbers of things and/or amounts. She referenced a paper by Wynn (1992), which I believe is the paper where infants see a single doll, then a screen goes up, and hand moves behind the screen carrying a second doll, and the hand pulls back not holding the doll. The implication is that there are now two dolls hidden behind the screen. The screen is lowered. In one condition, the two dolls are there. In the other, one doll has been removed. The looking time of the infants is measured to assess their degree of surprise. Infants tended to look longer at the unusual display (a single doll where there should have been two) than the normal condition. I think these children were 4-5 months old, and Wynn drew the conclusion that infants have the capacity to count. However, another possible explanation is that they are just cueing in to the amount of stuff, and not actually counting. So Chen’s experiment involved a display in which three red circles move down from the top of the display, then one moves to the right behind a screen and the other left behind a screen. Then the third circle moves either to the right or left behind the screen to join one of the other circles. The screen that should have two circles behind it is lowered, and the infant’s looking times are measured. There were, I think, four conditions: the circles are smaller, but there are still two of them, there is one big circle of the same areas as the two original circles, there are three new circles of the same area as the two previous circles, and one where the circles are unchanged. The researchers were attempting to control for both mass and numerosity. What they found was that these infants (two groups of 9 and 12 month olds) were sensitive to changes in both number and area, though they tended to be more sensitive to number than area. It was a nice experiment and a good talk.

The third speaker was Elizabeth Baraff Bonawitz, talking about how children’s theories about the world bias their play activity. First she noted that there is a developmental transition in which children go from being “center theorists”, believing that an object can always be balanced by placing the fulcrum at the geometric center of the object, to “mass theorists”, believe that an object can best be balanced by placing the fulcrum at the center of mass. This transition tends to take place between 6 and 7 years old. The experiment involved first splitting one group of children into two by testing out their balancing theories. Once they had them in two groups, they habituated all subjects to a piece of wood with one end bulkier than the other. Then the researchers balanced the piece of wood in such a way that defied the theories of the children. In the case of the center theorists, they simply balanced the stick at the center of mass. With the mass theorists, they manipulated the stimuli (with a magnet, I think) to balance at the geometric center, defying the children’s expectations. Then they introduced the children to a new toy, in addition to the stick, and measured the amount of time they played with either the new toy or the stick. They found that children spent much more time playing with the stick they had already been habituated to, after the demonstration that it had defied their theory about the world, rather than the new object. They went on to suggest that play is not a free-form, unstructured activity, but is guided by children’s theories. In this particular case, the children played more with an item that defied their theories, invoking curiosity and directing their play. Again, I thought this was interesting.

We skipped the last evening session to rest a bit and grab dinner, but we made the poster session from 7:00pm to 9:00pm.

There was some interesting stuff. I saw interesting posters on people’s justifications for believing or not believe in evolutionary theory, one on saccade patterns in detecting symmetry, and another on how reading direction in Hebrew and English speaking subjects affects their reaction times in distinguishing sequences presented from either left-to-right or right-to-left.

Lots of interesting stuff, but my head was cram-packed by the end of the three full days of sessions and talks. And to give you an idea, what I’ve talked about was probably less than 20% of the content at the conference. If I’d been militant about trying to see as much as I could, I would have been able to squeeze a bit more out of the conference, but still would have only seen a fraction of it.

It did get me pretty fired up and thinking about lots of different things. I heard that next year it will be held in Washington D.C. I’d like to attend, and hopefully present.

And that's it for CogSci 2007. I hope somebody found some of this interesting.