From Recall to Discrimination: The Dynamic Neural Field Theory Generalizes Across Tasks and Development

From Recall to Discrimination: The Dynamic Neural Field Theory Generalizes Across Tasks and Development Vanessa R. Simmering (vanessa-simmering@uiowa.edu) John P. Spencer (john-spencer@uiowa.edu) Department of Psychology, University of Iowa Iowa City, IA 52242 USA Gregor Schoner (gregor.schoener@neuroinformatik.ruhr-uni-bochum.de) Institut fur Neuroinformatik, Ruhr-University 44780 Bochum, Germany behavior-in-the-moment. That is, these models are strong on generalization, but weak on the details so central to process- based models of performance in particular tasks. The research presented here seeks to achieve both specificity and generality. In particular, we report theoretical and empirical work that generalizes a Dynamic Neural Field Theory (DNFT) of spatial working memory (SWM) beyond the spatial recall task that was central to the development of this theory. We extend the DNFT to capture performance in a position discrimination task, and then use this theory to generate specific behavioral predictions which we test in an experiment with adult participants. Next, we probe whether a core developmental hypothesis implemented in the DNFT—the spatial precision hypothesis—that has captured developmental changes in spatial recall can predict developmental changes in position discrimination. Finally, we test these predictions in an experiment with 3- to 6-year- old children. Results from our theoretical and empirical work demonstrate that the DNFT effectively generalizes across tasks and development. Importantly, such generality does not come at the cost of behavioral specificity due to our consistent emphasis on the processes that create behavior in the moment. Abstract Many models developed to account for tasks in cognition and development are well-suited to a single task or phenomenon. In recent years, modelers have begun to generalize their approaches to incorporate other tasks and/or developmental change. Few models, however, have attempted to generalize across both tasks and development. This paper presents one such model, the Dynamic Neural Field Theory (DNFT). The DNFT was developed to account for biases in spatial recall across development. A second line of research, presented here, generalizes the DNFT to a new task, position discrimination, and predicts a complex pattern of performance with both children and adults. We conclude with implications for future theoretical and empirical work. Specificity vs. Generality in Models of Cognition and Development Human behavior is rich and complex. As such, a central goal in cognitive science is to develop models that can capture this complexity and predict behavior in a meaningful way. In many cases, models in cognitive science are exclusively linked to particular tasks (for a discussion of this concern in categorization, see, e.g., Murphy, 2002). This approach is sensible, as it is often very difficult to capture the details of performance in a task in a way that sheds light on the processes that underlie behavior. Although theoretical specificity is essential to capturing the details of complex human behavior-in-the-moment, it often comes at the cost of generalization. For instance, there are relatively few detailed, process-based models that generalize across multiple tasks (for one exception, see Love, Medin, & Gureckis, 2004). Similarly, there are few process-based models that generalize across development. And the set of models that lie at the intersection of these two types of generalization is very sparse indeed (for examples, see Huttenlocher, Hedges, & Vevea, 2000; Morton & Munakata, 2002). The difficulty of balancing specificity on one hand and generality on the other is not simply reflected in the relatively small number of models that generalize; it is also reflected in the nature of models that have effectively moved beyond specific tasks. For instance, the few models that have generalized across tasks and development have relatively weak ties between the model and the details of Dynamic Neural Field Theory: Overview The DNFT is a process-based theory of spatial working memory (SWM) instantiated in a neural network model that captures children’s and adults’ spatial recall performance (Schutte, Spencer, & Schoner, 2003; Spencer & Schoner, 2003, 2006). In spatial recall tasks, participants are asked to remember a target location within a large, homogeneous space. After a short delay (0 – 20 s), adults show systematic biases away from the midline symmetry axis of the task space (e.g., Spencer & Hund, 2002). In this same task, young children show the opposite pattern of bias, that is, they show systematic bias toward midline (e.g., Schutte & Spencer, 2002). In the following section, we outline the DNFT and how this theory accounts for changes in spatial recall biases over development. Figure 1 shows a simulation of the DNFT 1 during a single Note that for simplicity we show only the relevant portion of the full model here. For a complete description, see Spencer, Simmering, Troob, Schutte, & Schoner (2006).