Concatenating familiar movement sequences: the versatile cognitive processor

Abstract

Earlier studies demonstrated that practicing a series of key presses in a fixed order yields memory representations (i.e., motor chunks) that can be selected and used for sequence execution as if familiar key pressing sequences are single responses. In order to examine whether these motor chunks are robust in different situations and whether preparation for one sequence may overlap with execution of another one, two experiments were carried out in which participants executed two highly practiced keying sequences in rapid succession in response to two simultaneously presented stimuli. The results confirmed robustness of motor chunks, even when the sequences included only two elements, and showed that preparation (and in particular, selection) of a forthcoming sequence may occur during execution of the earlier sequence. Sequences including only two keys appeared to be slowed more by concurrent preparation than longer sequences. Together these results suggest that the execution of familiar keying sequences is predominantly carried out by a dedicated motor processor, and that the cognitive processor can be allocated to preparing a forthcoming sequence (e.g., during execution of an earlier sequence) or, some times, to selecting individual sequence elements in parallel to the motor processor.

Introduction

This article has been written on the occasion of the retirement of professor Andries F. Sanders. One of the main themes in his work was disentangling the processing stages used in choice RT tasks, the merits of which should “make sense beyond the realm of additive factor application” in order to “advance performance theory” (Gopher & Sanders, 1984, p. 250). Based on a long line of research, Sanders, 1980, Sanders, 1990 proposed a model in which it is assumed that information processing in choice RT tasks involves six successive stages. In the present article, I shall extend Sanders’s line of thinking by reporting indications that the response selection and motor processing stages he proposed, may be carried out by two independent processors that can be active simultaneously.

Let me start off by noticing that much of our daily life involves the virtually automatic production of series of actions. Intelligent, goal directed behavior would be impossible if we would continuously be engaged in controlling every minute aspect of our behavior. Several researchers have argued that we are capable of skilled action because, with extensive practice, behavioral sequences, such as pronouncing phonemes, writing letters, and typing short words are executed to an increasing degree as if they were single movements (e.g., Fowler, 1985, Van Galen, 1991, Inhoff, 1991). Elsewhere, I called these sequence-specific memory representations, motor chunks(Verwey, 1994). The availability of motor chunks would eliminate the need to select each element in a familiar keying sequence in a demanding and slow manner.

The present article investigates, first, the merits of the motor chunking concept in a situation that two familiar keying sequences are carried out in rapid succession. One possibility is that when familiar movement sequences are part of a longer behavioral sequence, the original motor chunks are still used for sequence control. In line with some older ideas, this would imply that familiar movement sequences constitute ‘building blocks’ of behavior (e.g., Eysenck and Frith, 1977, Gallistel, 1980, Paillard, 1960). Before the notion can be accepted that motor chunks underlie these building blocks of skilled behavior, there is a need to confirm that motor chunks are robust in the sense that they can also be used in different situations than the one they were practiced in. Alternatively, a new organization could develop and successive familiar sequences could be produced as a single, new sequence. This is especially likely when two successive sequences are short so that the elements of both sequences could be easily prepared together Sternberg et al., 1978, Verwey, 1996.

In an earlier study, I reported how participants in a practice phase cycled through a series of nine key presses (Verwey, 1996). This series included a fixed order of keys and participants were forced to wait for a relatively long time at two or three positions in the sequence. Subsequently, they were required to cycle through the sequence at maximum speed, that is, without waiting at any position. The results showed that even at maximum speed, the interkey intervals at the positions where the pauses used to be, were much longer than the remaining interkey intervals. I argued that during practice, the pauses had separated the sequences in parts, and that for each part a different motor chunk had developed. Executing the sequence at maximum rate forced participants to prepare the ensuing motor chunk at the positions in the sequence where they used to pause during practice.

In the present study, I focus on a second finding in Verwey (1996), also see Verwey and Dronkert (1996): it turned out that execution rate of each sequence part between pauses had slowed too. This was attributed to the fact that at least part of the preparation of an oncoming sequence, overlapped with the execution of the preceding one. This was not unexpected as it is in line with other findings that preparation of the forthcoming movement slows down the execution rate of the ongoing movements (e.g., Brown, McDonald, Brown, & Carr, 1988; Portier, van Galen & Meulenbroek, 1990). However, in another study, I found that when participants selected a key press during execution of a familiar keying sequence, execution rate was not affected by the compatibility of the imperative stimulus and the associated key press that immediately followed the familiar keying sequence (Verwey, 1995). How can we explain that the mere presence of concurrent preparation slows ongoing execution while the demands of concurrent preparation do not?

The common explanation for interference between concurrent processes is that preparatory and motor processes tap the same limited processing resource (e.g., Kahneman, 1973). This was actually the explanation I advanced (Verwey, 1996), but in Verwey (1995), I had explained the finding that manipulation of response selection demands did not affect sequence execution by the notion that execution and preparation (in terms of selection) tap different resources (e.g., Wickens, 1984). Obviously, it is unsatisfactory to assume a single-resource model for the comparison of execution with and without concurrent preparation, and a multiple-resource model for the situation in which the demands of concurrent preparation are manipulated.

Therefore, a second purpose of this study is developing an alternative for the resource interpretation of slowed sequence execution. In line with various other models, I would like to propose that there are two different processors, a cognitive and a motor processor (or system; e.g., Allport, 1980, MacKay, 1982, Pew, 1966, Schmidt, 1988). For example, Shaffer (1991) argued that skilled action “is arranged at two levels in the brain, (a) in a cognitive system which plans and represents (symbolically) a goal structure of action, and (b) in a motor system with organizes movements appropriate to the goal” (p. 372). An unfamiliar movement sequence would be controlled by a cognitive processor selecting individual sequence elements on the basis of a symbolic representation. A familiar movement sequence would be controlled by the cognitive processor in that it selects a single representation – a motor chunk – for the entire sequence which is subsequently read and executed by a dedicated motor processor. Compared to the production of unfamiliar sequences, this then reduces processing load on the cognitive processor.

If an additional assumption is made, then this dual-processor model can explain why familiar sequences are slowed by concurrent preparation and not by the demands of concurrent preparation. The assumption is that when participants are asked to generate a single, familiar keying sequence as fast as they can (which is usually what they are instructed to do), they will have the cognitive and the motor processors operate in parallel in selecting the individual sequence elements. That is, the two processors will both produce the same information needed for executing an element in the sequence. On a particular trial, the processor providing its output first determines reaction time. Obviously, the cognitive processor is able to execute sequence elements as it had done so in early practice too. This notion stresses the versatility of the cognitive processor for, in case of familiar movement sequences, it may support the motor processor in producing the sequence elements and it may be allocated to other tasks.

Of course, parallel operation of the cognitive and motor processors is likely only if it will actually speed up sequence execution. It appears that when the time taken by the cognitive and motor processors to produce their output is distributed and the distributions of both processors overlap, the average time to execute a single sequence element will reduce. This has been named statistical facilitation(Raab, 1962). In other words, execution rate will increase when two processors operate in parallel even when, on the average, one is slower than the other. The finding that execution rate was reduced by concurrent preparation, as compared to the condition without concurrent preparation (Verwey, 1996), can now be explained in terms of the cognitive processor being allocated to preparing the forthcoming sequence and, hence, being withdrawn from (assisting the motor processor with) sequence execution. The absence of an effect of the demands of concurrent preparation on execution rate (Verwey, 1995), is then explained by the notion that, once the cognitive processor had been allocated to another task, the actual demands placed upon the cognitive processor do not affect execution as execution was already controlled by the motor processor alone. In my view, the strength of this dual-processor model is not only that it provides an explanation for the effects of concurrent preparation on sequence execution, but also that it is in line with the parallel architecture of the brain (see Section 4).

As interpretation of the execution rate effects in terms of the dual-processor model is still speculative at present, the purpose of the present study is testing some of its assumptions. First, it examines whether motor chunks are used also when two familiar sequences are executed in rapid succession. Next, it examines if preparation for one keying sequence overlaps with execution of the preceding keying sequence. Finally, it tests whether preparing a forthcoming sequence during execution of an earlier sequence, slows execution of the earlier sequence while the demands of concurrent preparation do not affect execution of the earlier sequence.