All research within the Cognitive Science Laboratories involves "computational modeling" of one sort or another: That is, we try to understand human cognition by developing computer models of how people process information. We are particularly interested in connectionist models and their application to various cognitive phenomena, in particular category learning and short-term memory. In pursuit of this goal, the models are first fit to existing results, and novel predictions of the models are then explored experimentally.
Category learning refers to people's ability to learn to classify objects into different categories; for example, as children we learn to differentiate between furry things that are dogs and other furry things that are cats. As adults, we learn to differentiate between meaningful email messages and spam. In all experiments within this line of research, people are taught arbitrary new categories involving artificial stimuli (e.g., line drawings). The aim of these experiments is to tell us something about the fundamental way in which people acquire, organize, and revise their knowledge.
Knowledge partitioning is the idea that knowledge may be held in separate, non-integrated parcels. These parcels may contain contradictory information and may be used without reference to other parcels, which can result in contradictory decision making. This has been shown to be the case in various function learning and categorization paradigms.
Knowledge restructuring refers to a shift in rules, strategies or representation that does not alter the learning environment. Despite the ease with which most of us think we can shift rules, knowledge restructuring has been notoriously hard to induce. This is true unless two preconditions are met: (1) the strategy that people switch from must be error-prone, and (2) people must be told that a better strategy exists.
Probabilistic category learning refers to category learning where the assignment of instances to a category is not perfect or deterministic. This is commonly investigated using the Multiple Cue Probability Learning paradigm (MCPL). We are currently investigating the existence knowledge partitioning and the ramifications of knowledge partitioning in MCPL. We are also investigating how people respond to shifts in reinforcement probabilities.
This new project combines the work conducted for the above funding periods under a common umbrella, by seeking ways in which individual differences in categorization can be explained. One common characteristic of all experiments involving knowledge partitioning, knowledge restructuring, and probabilistic categorization is that they give rise to striking individual differences: People approach the same task in very different ways. At present, we know little about what causes those individual differences, and this project seeks to explore differences in working memory capacity as a possible cause of those differences. Watch this space for news on this project.
The principal stream of research in the laboratory involves human memory, with particular emphasis on short-term memory (STM) and working memory (WM). This research has been funded continuously by various grants from the Australian Research Council since 1995.
Iterated learning refers to situations in which information or knowledge is transmitted across "generations" of learners. This paradigm permits us to study the evolution of knowledge in the laboratory, similar to the way in which language has evolved across generations of learners throughout history. In an iterated learning experiment, the first generation of participants is presented with stimuli (e.g., in a function learning experiment) that they have to learn. The next generation of participants then receives the responses of the preceding generation as stimuli, and so on. It turns out that this type of experiment gives rise to some surprising regularities that can be captured by a mathematical analysis.