Research

The lab research program uses data from experiments to develop a detailed mathematical model of the dynamics of practice and forgetting in memory tasks. This modeling is applied to further the goal of maximizing learning in instructional settings. These applications tend to reveal limitations of the model and areas for further theoretical research. This in turn facilitates the formulation of new hypotheses to test in further experiments. This cycle of experimentation, model building and application keeps our research focused and integrated.

Areas of Interest The mechanisms involved in the change of a memory over time are fascinating. Many factors influence this change, including the well-researched effects of practice, forgetting, and the spacing of practice. Given this body of work, I initially expected that one might be able to use what is known to optimize the learning of something like simple facts. However, when I looked for such applications I found that what existed had serious theoretical problems and that further modeling was necessary to account for the important factors involved.     My research program at CMU has been directed to finding and filling those holes with the goal of furthering my understanding of some of the basic processes in human memory to the point where this understanding could be used in effective applications. In general, this means that my work touches on many of the factors controlling the retrieval of memories. For instance, much of my work so far has concentrated on the modulation of forgetting caused by the spacing of practice. Also known as the spacing effect, this is the advantage for widely spaced practice compared with more narrowly spaced practice. Understanding this effect is crucial to understanding the larger problem of how repetition improves memory strength because repetition is always spaced.     Using the independence afforded to me by my graduate advisor, John Anderson, I systematically addressed the theoretical questions outlined below using both mathematical modeling and experimentation. This work has focused on three areas:

Understanding the effect of repetition spacing on memory This line of research uses simple facts (Japanese – English vocabulary pairs) to explore the effect of different schedules of practice on longer-term retention. The results of some of this work, along with a model of the effects, are published in Cognitive Science (Pavlik & Anderson, 2005). This work is the first formalization of a new theory of the spacing effect that says that the advantage of spacing is due to a reduction in forgetting for more widely spaced practice. Understanding strategy, individual and item differences The first step in these investigations has been to describe the extent of performance variability in the data. I have found that significant variability exists in the memory ability of subjects, the difficulty of items, and the performance interaction between a subject and an item. Describing these sources of variability has enriched my work in the areas of theory, model building and application. Understanding these item differences has resulted in methods for fine-grained tracking of memory strengths in applied learning situations. This tracking of item strength allows more accurate prediction of the optimal way to schedule practice. Understanding the differences between study and test practice It has often been shown that a successful test provides more memory benefit than a study opportunity; however, this simple advantage has never been fully described or explained in the literature. To understand better this benefit of testing, an experiment in my dissertation asked what the benefit of a test is relative to a study presentation. This work looked at differences in encoding and forgetting between tests and studies of varying duration. Answering these questions has helped to build theories of the learning process and is directly relevant to modeling.

Modeling Modeling has been crucial to unifying my research program in a domain that might otherwise be intractably complex. The interplay between modeling and experimentation has helped me develop my theoretical perspective by allowing me to identify important factors and their interactions in the tasks I study. The organizational framework that modeling provides allows me to consider hypotheses and applications that I would have been unlikely to consider otherwise.     Most of my modeling has involved elaborating ACT-R's (Adaptive Character of Thought – Rational) declarative memory model to capture the effects I have investigated (Anderson and Lebiere, 1998). It provides a way of describing hypotheses about factors of the memory task and then seeing how well competing hypotheses can account for the data in the context of the overall model. Rather than describing how parts of the memory task might work in isolation, this model has the potential to provide a unified and constrained theoretical description of all the relevant interactions and effects in the data.

Application In my dissertation, I am using this model to explore ways to optimize a variety of factors in memory performance. The ACT-R declarative memory model I use accounts for all the effects I have mentioned and provides predictions of both latency and probability of performance. Because it measures the time cost of performances and has a dependent measure of learning, the model allows maximization of learning per second rather than simply learning per trial. This innovation is the first example I am aware of in which a model is used to maximize learning efficiency for individual trials (gain/cost) rather than just maximizing learning per trial. My dissertation work showed that paying attention to the efficiency of practice results in significant gains compared to a widely spaced practice control group. A later experiment has replicated this result for a larger training corpus over more training sessions.

Current Directions My current research has left me well positioned both to explore more complex aspects of memory and to further apply the work I have completed. This is reflected in my current projects

Understanding the temporal dynamics of complex memory structures Having a strong model of the paired-associate task, I am now seeking to broaden the paradigms I am investigating. My specific goal is to be able to apply the memory model I use in understanding the temporal dynamics of complex chunks of information. Assuming a model of a complex memory structures as a number of associations, and given a model of simple association dynamics, I believe it may be possible to look at the learning and forgetting functions of complex memory structures. This work aims at two things. First, this sort of modeling will provide new ways of modeling complex cognition such as analogy and problem solving because it will better represent the complex structures involved. Second, this sort of model will allow exact specification of dependent measures to optimize. This will be necessary to design applications to optimize the learning of complex information. Development of applications for the classroom The current system for tracking the memory strengths of individual simple memory items is near the point where further specification of mechanisms is unlikely to lead to further gains in learning. Some work may still be needed to refine the applications so they can track and optimize thousands of memory items sets over a period of months and many sessions. Following this work, I plan to enlist the cooperation of professors in departments teaching foreign languages so that the system can be validated as a tool to augment classroom learning. Undoubtedly, this classroom testing will present new challenges for theory and modeling. Theoretical and experimental investigations of memory dynamics Much of my current research has used experimentation and modeling to investigate the spacing effect and aspects of forgetting dynamics. While this work has a theoretical basis, I believe that this theoretical basis needs to be strengthened with further research. Specifically, I am interested in investigating in further detail how the spacing effect may be a neural adaptation to the episodic nature of memory needs in the environment. Similarly, I am interested in a stronger theoretical understanding of the differences in rates of forgetting as a function of the density of interference in the environment.