Analyzing Effects of Goal Competition and Task Difficulty in Multiple-Task Performance: Volitional Action Control within ACT-R

In this paper we present a cognitive-modeling analysis of processes of volitional action control in multiple-task performance. We simulated experimentally obtained effects of goal competition and task difficulty on processing strategies and performance in a high-level learning and problem-solving task within the ACT-R architecture (version 4.0). Beyond the model's capacity for explaining the empirical pattern of results, the fundamental processing mechanisms used for modeling, i.e., activation mechanisms and executive control productions are in line with current explanations of elementary executive control processes. Thus, we consider our modeling approach to be a solution to fill the gap between volitional control demands in real-world multiple-task performance and experimental findings on elementary executive functions. Elementary Executive Control Processes In recent years, research on action control in multipletask situations in the fields of experimental cognitive psychology and cognitive modeling has yielded promising insights (see Monsell & Driver, 2000, for an overview). With regard to modeling, mainly two approaches have been proposed for analyzing the management of component tasks in multiple-task performance. First, activation mechanisms have been proven useful to represent aspects of task coordination and task interference. Second, executive control productions have successfully modeled processes that act on task-related processes by enabling, preventing, and directing them. There are two ways to handle component tasks in a multiple-task situation, namely either simultaneously or successively. Simultaneous Task Procedures In simultaneous processing, people are required to perform two different tasks at the same time as in the psychological refractory period (PRP) procedure where two choice reaction time tasks have to be performed with a temporal overlap. If this overlap is short enough performance impairments on the second but not on the first task will result. However, the total time required to complete both tasks is often less than the sum of times for performing both tasks separately. Meyer and Kieras (1997) simulated these findings in an EPIC model by introducing executive production rules that schedule and control task-specific rules by monitoring task progress and by inserting and deleting task goals as well as strategy notes in working memory. Successive Task Procedures Successive processing requires the ability to alternate between different tasks and is often studied in the taskswitching paradigm where subjects either repeatedly have to perform the same task or to alternate between different tasks. Typically, alternating task sequences result in temporal switch costs in terms of a prolonged response time compared to task repetition. In their ACT-R model, Altmann and Gray (2000) simulated switch costs mainly as a result of proactive interference from previous tasks that may impede performance on a current task. The authors proposed active inhibition and automatic decay of memory elements as mechanisms that in combination may counteract proactive interference. According to this view, a memory element can be inhibited by an encoding process that increases the activation of a competing element. Additionally, the activation of an unused memory element decreases over time because of automatic decay. Rubinstein, Meyer, and Evans (2001) proposed an EPIC model of task switching with two complementary sets of production rules. Task processes are used for performing the component tasks involved in multipletask performance. Executive control processes coordinate the execution of various task and subtask procedures. Because activation mechanisms as well as executive control productions have been shown to influence task switching, Sohn and Anderson (2001) combined both