No abstract

If strategy shifts speed up performance, learning curves should show discontinuities where such shifts occur. Relatively smooth curves appear consistently in the literature, however. To explore this incongruity, we examined learning when multiple strategies were used. We plotted power law learning curves for aggregated data from four mental arithmetic experiments and then plotted similar curves separately for each participant and strategy. We then evaluated the fits achieved by each group of curves. In all four experiments, plotting separately by strategy produced significantly better fits to individual participants' data than did plotting a single power function. We conclude that improvement of solution time is better explained by practice on a strategy than by practice on a task, and that careful assessment of trial-by-trial changes in strategy can improve understanding of the effects of practice on learning. 0The generality and precision simultaneously achieved by expressing empirical regularities as mathematical functions facilitates theoretical development, testing, and the application of scientific knowledge. Although mathematical laws are more prevalent in the physical sciences than in the social sciences, psychology's search for quantitative laws that describe human behavior is long-standing, dating back to the 1850s. A few notable successes have been achieved, including Fitts's law (1954) and the Hick-Hyman law (Hick, 1952;Hyman, 1953). Newell and Rosenbloom (1981) proposed another candidate for the status of quantitative psychological law. They argued that the power law of practice 1 offers a sufficiently accurate, general, and useful characterization of human skill acquisition. This article examines that proposal in the light of empirical evidence that strategy changes sometimes play an important role in cognitive skill acquisition. Such evidence raises questions about the adequacy of the regular power law as a complete descriptor of the temporal course of complex human learning. Our goal is to describe the tension arising between the general formulation of the regular power law and the data on strategy shifts and then to suggest a way to reconcile the two bodies of evidence.It is well established that practice on a task almost always improves performance, both by reducing the number of errors and by reducing the time required to perform the task. Many longitudinal studies using performance time (e.g., solution time for problems, reaction time to stimuli) to measure skill acquisition have shown a remarkable regular-0.1. This regularity was first noted by Lewis (1976) in an unpublished manuscript that Newell acknowledged reading.

No abstract

No abstract

ACT-R is a hybrid cognitive architecture. It is comprised of a set of programmable information processing mechanisms that can be used to predict and explain human behavior including cognition and interaction with the environment. We start by reviewing its history, which shapes its current form, contrasts and relates it to other architectures, and helps readers to anticipate where it is going. Based on this history, we then describe it as a theory of cognition that is realized as a computer program. After this, we briefly discuss tools for working with ACT-R, and also note several major accomplishments that have been gained by working with ACT-R in both basic and applied science, including summarizing some of the insights about human behavior. We conclude by discussing its future, which we believe will include adding emotions and physiology, increasing usability, and the use of nongenerative models.There are numerous reviews of ACT-R, many noted here. Therefore, in this study in addition to describing it as a theory, we start by briefly reviewing its history, which explains its current form and helps readers to anticipate where ACT-R is going. Similar to most reviews of cognitive architectures, we describe the theory and its structure (see Box 1 for key concepts), and how these are realized as a running computer program; we also briefly discuss tools for getting started and working with ACT-R and major accomplishments in both the scientific and applied science areas. This includes summarizing some of the insights about human behavior that have been gained by working with ACT-R. We conclude by discussing its future, which we believe will include emotions and physiology, usability, and the use of nongenerative models. We include explicit lessons for other architectures, but there are many implicit lessons as well. | HISTORYThe history of ACT-R is worth reviewing briefly for several reasons. First, the predecessors and previous iterations of the theory continue to shape ACT-R's current form. Second, it shows ACT-R's evolution from early theories of cognition and the influence of contemporaneous cognitive architectures (see Box 2 for types). Third, we hope the history will help readers and future researchers anticipate where ACT-R is going. Finally, the progression shows how cognitive architectures can evolve. The progression from HAM to ACT-R 7, as well as research that influenced ACT-R, is summarized in Figure 1.Every broad simulation tool has innate strengths attributable to its base simulation's original purpose (i.e., the phenomena that were modeled). For example, the HumMod physiology simulation was initially developed from a heart simulation before being developed into a unified model of human physiology "from birth to death" (Hester et al., 2011). HumMod's heart simulation is its most developed module. ACT-R fits this paradigm as well: ACT-R began as a model of human memory before being developed into a unified theory of cognition. As such, ACT-R is strongest when modeling memory.Work based on the ACT-R archit...