Control of Response Initiation: Mechanisms of Adaptation to Recent Experience

Control of Response Initiation: Mechanisms of Adaptation to Recent Experience Michael C. Mozer Department of Computer Science & Institute of Cognitive Science University of Colorado Sachiko Kinoshita MACCS & Department of Psychology Macquarie University We describe shortcomings of existing theoretical frame- works that have tried to account for these data. We then present a framework that successfully explains key phe- nomena and makes further predictions which we have verified through additional behavioral studies. Abstract In most cognitive and motor tasks, speed-accuracy trade offs are observed: Individuals can respond slowly and accurately, or quickly yet be prone to errors. Control mechanisms governing the initiation of behavioral responses are sensitive not only to task instructions and the stimulus being processed, but also to the recent stimu- lus history: when stimuli can be characterized on an easy- hard dimension (e.g., word frequency in a naming task), an easy item is responded to more slowly when inter- mixed with hard items than when presented among other easy items; likewise, hard items are responded to more quickly when intermixed with easy items. We propose a mathematical theory with three components: a model of temporal dynamics of information processing, a decision criterion specifying when a response should be initiated, and a mechanism of adaptation to the stimulus history. Performance during the course of an experimental trial is cast in terms of a utility function that increases with accu- racy and decreases with response time. We assume a deci- sion criterion that initiates a response at the point in time that maximizes expected utility. We posit that the effect of the stimulus history arises because information con- cerning recent trial difficulty is incorporated into the util- ity estimate. We present further behavioral studies to validate predictions of the theory. The Blocking Effect To understand the control mechanism that initiates responses, consider the variables that affect its opera- tion. The mechanism is influenced by task instructions: individuals can choose to emphasize speed or accuracy. The mechanism is also influenced by recent perfor- mance: participants often slow down after producing an error (Rabbit & Vyas, 1970). Finally, even in the absence of errors, the mechanism is sensitive to the recent stimulus environment (Kiger & Glass, 1981): when items are presented in a sequence or block, reac- tion time (RT) and error rate to an item depends on the immediately preceding items. This blocking effect is generally studied by manipu- lating item difficulty. Some items are intrinsically easier than others, e.g., 10+3 is easier than 5+8, whether due to practice or the number of cognitive operations required to determine the sum. By definition, individuals have faster RTs and lower error rates to easy problems. How- ever, the RTs and error rates are modulated by the com- position of a block. Consider an experimental paradigm consisting of three trial blocks: just easy items (pure easy), just hard items (pure hard), and a mixture of both in random order (mixed). When presented in a mixed block, easy items slow down relative to a pure block and hard items speed up. Thus, the control mechanism that initiates responses uses information not only from the current stimulus, but also adapts to the stimulus environ- ment in which it is operating. Table shows a typical blocking result for a word reading task, where word fre- quency is used to manipulate difficulty. Based on our review of the blocking-effect literature (e.g., Lupker, Brown & Columbo, 1997; Lupker, Kinoshita, Coltheart, & Taylor, 2000; Taylor & Lupker, 2001), we summarize the central, robust phenomena as follows. Consider a simple task in which you are asked to name the sum of two numbers, such as 14+8. Given sufficient time, you presumably produce the correct result; how- ever, under speed pressure, mistakes can occur. In most all cognitive and motor tasks, such empirical speed- accuracy trade offs are observed: Individuals can respond slowly yet accurately, or quickly and be prone to errors. Speed-accuracy trade offs are due to the fact that evidence accumulates gradually in response sys- tems over time (Rabbitt & Vyas, 1970). Responses initi- ated earlier in time will be based on lower quality information, and hence more likely to be incorrect. This paper addresses a simple yet fundamental form of cog- nitive control—the mechanism that governs the initia- tion of a behavioral response, and therefore, where an individual operates on the speed-versus-accuracy con- tinuum. In the following section, we describe data that place constraints on the nature of control mechanisms. TABLE 1. RTs and Error Rates for Blocking study of Lupker, Brown, & Columbo (1997, Experiment 3) Easy Item Hard Item Colin J. Davis MACCS Macquarie University Sydney, NSW 2109 Pure Block 488 ms (3.6%) 583 ms (12.0%) Mixed Block 513 ms (1.8%) 559 ms (12.2%) Difference +25 ms (–1.8%) –24 ms (+0.2%)