Serial processing in human movement production

Abstract

Parallel processing is often considered to be synonymous with biological computation, but a great deal of evidence points to serial computation being used by animals to solve specific types of problems. In particular, the observation of movement intermittency (fluctuations in limb kinematic variables that cannot be explained by low-level dynamics of the system) seems to imply a serial temporal segmentation strategy in the planning of arm movements. This paper discusses prior observations of movement intermittency in different task contexts, possible theoretical and physiological origins of the phenomenon, and implications for human movement strategies.

Introduction

Extensive parallel processing is considered the hallmark of biological computation. Dissection reveals neural elements that are massively interconnected, and it is well known that animals are capable of solving formidable computational problems (in image processing, for example) that are ill-suited to serial computation methods. The computational systems of animals exhibit in many cases `graceful degradation', that is, the removal of a small set of elements has only a minimal impact on computation capability—further evidence of parallel computation.

Biological systems are often simulated using parallel distributed processing (PDP) models. These models feature large numbers of elements that are interconnected in circuits; the effect of one element's activity upon another is a function both of the element's output level as well as the strength of the connection. PDP models (and in particular layered network models) have been successful not because they are likely to be replicas of actual processing in the nervous system, but because they illustrate the types of operations that can be performed by networks of interconnected units. These networks also process data in parallel, have been demonstrated to solve problems of high complexity (Arbib, 1995), and like biological systems also exhibit graceful degradation.

But despite all the evidence for parallel computation in animals, certain behavioural phenomena hint strongly at serial computation being used to solve certain types of problems. Human communication, for example, involves sequentially stringing together groups of phonemes. It is extremely difficult to listen to more than one speaker at a time, even when they all can be heard quite clearly. A similar limitation occurs with visual stimuli. Although visual processing no doubt involves much parallel computation, the mechanism of selective attention forces us to concentrate on only one visual stimulus in the foreground; we are only dimly aware of the background, and so we can only focus on objects in a serial fashion.¹ Yet another example is mental rotation. Shepard discovered (Shepard and Cooper, 1982) that the time taken for mental rotation of three-dimensional shapes to perform a matching task was proportional to the amount of rotation. This implies that the mental rotations were probably performed sequentially; several mental rotation angles could not be attempted simultaneously.

This paper will concentrate on one example of serial computation: the apparent `temporal segmentation' of motor behavior, the idea that humans generate long or complex movements by piecing together shorter or simpler movements. Humans appear to have difficulty moving smoothly to an accurately specified target posture; instead they seem to break the task up by sequentially moving to several postures, each successive one being closer to the target posture. In addition, humans appear significantly limited in their ability to move slowly and smoothly, even when no target is specified. This may imply that movement itself is nothing more than sequential transitions between postures. In both of these cases, the behavior demonstrates a time-series of events, implying an event `decision-maker' upstream that computes its results serially.