Modeling Control Strategies in the N-Back Task

Two studies aiming to investigate the use of cognitive control strategies in the N-Back task are presented. The first study identified a behavioral effect that seemed indicative of participants’ proneness toward using highversus low-control strategies. Two ACT-R models of N-Back implementing the two hypothesized strategies were developed. Model simulations were used to identify the proneness toward using highversus low-control strategies by the individual participants in the second study. An independent measure of control – Stroop interference – was used to validate the predictions of the two models. Introduction and Background Cognitive control processes are often postulated to account for behavioral effects that cannot be explained based solely on relatively better-understood cognitive processes such as perception, language or memory. Some tasks are believed to require more cognitive control than others (Garavan, Ross, Li, & Stein, 2000). This paper aims to demonstrate that there are also differences among individuals with regard to whether or not certain cognitive control processes are employed in a particular task. The N-Back task requires judging whether a new item is identical to the nth-item back in a sequentially presented list of items (McElree, 2001). For example, in the sequence < M3 A2 R1 A0 > of the 2-back task, the current item (A0) is identical to the 2nd-item back (A2). The task requires keeping available the most recent n items to be compared with the incoming item. Although at each particular step only the nth-item can be a target, items with indexes between 0 and n must be remembered because they may be targets in the following steps (Awh et al., 1996). For concision, the set of the most recent n items will be referred to as the rehearsal window. The involvement of executive control processes in NBack is justified by the necessity to interleave different subtasks: processing incoming information, maintaining activation of recently processed and potentially relevant information (rehearsal), and discarding recently processed but irrelevant (potentially interfering) information. Evidence that these subtasks are concurrently executed comes from fMRI studies showing activation in Broca’s area (BA44) indicating articulatory rehearsal, Posterior parietal area (BA40) indicating short-term storage, and Dorso-Lateral Prefrontal Cortex (BA9/46) indicating excitatory or inhibitory modulation of activation in other areas (Cohen et al., 1997; Miller & Cohen, 2001; Owen, McMillan, Laird, & Bullmore, 2005). One of the functions of the cognitive system is to keep active the information that is relevant to the task at hand. Usually, the most frequently and recently processed information is most likely to be relevant to the current processing (Anderson, 1989). In this case, the relevant information is kept active simply because it has residual activation from recent processing. The residual activation that an item bears for a while after its use is beneficial when the item is reused, and the probability that an item will be reused gradually decreases with time. However, in the NBack task (as well as in other tasks of this type), the switch from relevant to irrelevant is instantaneous, instead of being gradual. When an item reaches the index n+1 it becomes totally irrelevant. In this case, its residual activation is not only useless but it may cause interference. An active control mechanism is needed to temporarily increase or decrease activation of a particular item depending on whether or not this item is relevant for the current state of the task.