[SOUND] One way to consider how language is processed, how cognition occurs is to think of it as a non-linear dynamical process. Now, what would we describe as a non-linear dynamical process? Well, we can take very famous examples. For example, the idea that a butterfly might flap its wings over Africa and create a storm in the Gulf of Mexico. The idea is that very small changes can add up over time. And as they add up and they multiply, what happens is you get very big differences occurring, but they started out as very small differences. People have taken this non-linear dynamical view of cognition. The most famous example of this is by Mclelland and Rumelhart, who proposed what's called a parallel distributed processing approach, this idea that different streams of processing are occurring in parallel, and they help to form cognition. We can think of the brain as a big parallel distributed processor. So, an example I give often to my students is if you were to look across a street and see a red car moving in front of you, the redness, the car-ness, and the fact that it's moving are represented in slightly different brain circuits. And they're combined. And there are people who have damage to one of these circuits or another, and actually show deficits. So they may be able to locate where that is, but not know what it is. Again, the idea is that our brain is like a big parallel processor and the notion that Mclelland and Rumelhart brought forward was to start to think about cognitive psychology as a product of this parallel distributed process. The fact that we might have these small interactive features that are occurring, for example in recognizing a word. People may be recognizing the forms of the letters, putting the letters together into words, but that there may actually be some information bleeding from the recognition of letters into the recognition of words. But also information bleeding back from the recognition of words, back into the recognition of letters. And one classic example is the fact that people can recognize letters better in a word than they can in a pseudo-word like m-a-v-e. This word superiority effect suggests that people are recognizing letters and that the word information is speeding up their ability to recognize letters. And this would be considered an interactive process. They suggested that the acquisition of language involved competitive processing, what they termed the competition model, that different types of forms and functions fought it out, battled against each other, to form language. And we can think about this when we discussed the trisyllabic type of processing done by Spanish speakers, and the bisyllabic done by English speakers. But the idea that trisyllabic sound bytes, right, la casa, would allow them, Spanish speakers, to develop gender, and determiners that were la and el as part of the word. So again, you see this as an interactive process. And that was very similar to the idea proposed by Bates and MacWhinney. That, in fact, you had an interactive process, that you had aspects of grammar working in tandem with form to create a much richer knowledge base. One example that she used a lot was to consider a very simple sentence. Let's consider the sentence the dog is chasing the cow. I could ask you who did it. Most people would answer the dog did it, English speakers certainly would. Then I can ask you, the dogs are chasing the cow and again you would say the dogs. But then there was another condition she used a lot, along with Brian MacWhinney. The dogs is chasing the cow. Now I can ask you who did it. English speakers will say well, the dogs did it, for the most part. Spanish speakers, Italian speakers will say the cow. The cow is chasing. Not the dogs is chasing. Why is that? Because each language relies on different types of grammatical information to signal who is doing what to whom. And this was a classic example of how each language is tuned to provide communication about who is doing what to whom. But they use different markers to indicate that. The nature of this competitive process has also made its way into thinking about development, and specifically we can think about the model developed by my colleague, Ping Lee, Farkas, and MacWhinney. In their DevLex model, they considered how the lexicon, how the word store develops over time. So what they did is they fed this computer simulated network sets of words heard by two-year-olds, heard by three-year-olds, heard, heard by four-year-olds at different points in time. And then they asked, what does a network look like, as these different types of words are fed into it. And what they saw was, very quickly, an emergent structure, this idea that structure began to emerge in the network as it was given different types of words. And it began to develop these classes of words. So nouns would appear in one place, different types of nouns, so fruits, vegetables, furniture, began to segregate into their different neighborhoods. And so you got this very rich organization in the in, in the network. But all that was happening was that the network was fed words one at a time. Zhao and Lee, in a subsequent study, looked at a bilingual network. The idea was to introduce the two languages at different points in time. And what they found was that when both languages were introduced simultaneously the network organized itself in a certain way with very clear boundaries between languages. But when the languages were presented successively, then what happened was that one language was more connected to the other. So it's as if there was this huge swamp of the first language, with little puddles of the second language attached to the first. When they were learned simultaneously, you had two separate lagoons, one for each language. The images we get from these networks help to describe in some ways what we've discussed in this course. That is the idea that these sensory-motor aspects of language occur earlier in life and help to form some connection between languages in a way that's different when a language is learned later in life.