Models of central capacity and concurrency

According to single channel theory, the ability of humans to perform concurrent mental operations is limited by the capacity of a central mechanism. The theory was developed by analogy with early computers which had a single central processing unit and required sequential processing. These limitations are not likely to be properties of the mind. But now computers have begun to employ extensive concurrent processing, because of the decreasing cost of the necessary hardware. In this review we will try to bring the computer analogy up to date. Theoretical issues important for concurrent systems may be of interest to psychologists and have applications to such problems as the speed-accuracy trade-off. Several hypotheses about the way signals gain access to the central mechanism are reviewed. Recent variations of the single channel theory are discussed, including the hypotheses that more than one process can use the central mechanism at a time, and that some processes do not use the central mechanism and can be executed concurrently with those that do. In addition, relevant concepts from scheduling theory and operating systems theory are introduced and difficulties encountered by concurrent systems, namely complexity, deadlocks, and thrashing, are discussed.     

Silent reading rate and memory span

This paper reports two experiments concerned with the form of the linear function relating memory span to pronunciation rate. Experiment 1 asked whether the intercept of the equation is attributable to an interaction between modality, word-length, and serial position. The results lead us to reject this possibility. Experiment 2 examined the shape of the span function for visual presentation of lists read aloud or silently. The results show that the intercept of the span function disappears for silent presentation but not for aloud presentation. The results are discussed in terms of a balance between two opposing effects of word-length.     

Response time distributions: some simple effects of factors selectively influencing mental processes

When hypotheses about mental processing are tested with response times, inferences are often based on means, and occasionally on variance or skewness. Calculations on entire distributions of response times are more informative and can be conveniently carried out. Recently investigators have been updating procedures primarily based on means (such as additive factors tests) to procedures employing entire distribution functions. In one such advance, Nozawa and Townsend upgraded earlier tests of whether factors selectively influence serial or parallel processes, and whether parallel processes enter AND gates or OR gates. We discuss generalizations of the tests to complex arrangements of processes in networks. Results for a particularly difficult network, the Wheatstone bridge, are presented here. We use simulations to demonstrate the feasibility of the tests, and the possibility of mimicking.     

Redintegration and the Useful Lifetime of the Verbal Memory Representation

Factors affecting immediate serial recall of individual items can be classified according to whether they influence the degradation of the representation, or influence the ability to redintegrate a degraded representation. For recall of entire lists, factors can be classified according to whether or not they influence the rate at which items are recalled. We review experiments in order to classify the factors of serial position, word length, word frequency, and lexicality. We propose that the two classification systems coincide, at least for those factors whose classification is known for both schemes. We end by considering whether probability of recall of an entire list might be predicted from probability of recall of individual items.     

Scheduling Processes in Working Memory: Instructions Control the Order of Memory Search and Mental Arithmetic

Humans must often use working memory to execute processes one at a time because of its limited capacity. Two experiments tested where limits in access to working memory occur. Subjects searched a short-term memory set for one stimulus digit and performed mental arithmetic with another stimulus digit. In one experiment, they were told to carry out the mental arithmetic before the memory search and to make the arithmetic response first. In the other, they were instructed to perform the tasks in the opposite order. The overt responses were executed in the prescribed order. Moreover, the covert working memory processes were executed in the prescribed order, as revealed by a critical path network analysis of reaction times. Results are explained in terms of a double-bottleneck model in which central processes and responses are constrained to be carried out for one task at a time.     

Separable effects of factors on activation functions in discrete and continuous models: d' and evoked potentials

This article derives tests for partial-output models of information processing. A process in a partial-output model sends output, while executing, to its successors. In a stage model a process sends output to its successors only when it stops. Both models can be analyzed by manipulating factors, each affecting a different process. In discrete serial models, such factors have additive effects on reaction time (Sternberg, 1969) and sometimes on log percent correct (Schweickert, 1985). In partial-output models, such factors produce a simple pattern, the rectangle condition, in performance curves. Conditions for a partial-output representation are based on decomposable structures (Krantz et al., 1971). Tests of discrete and partial-output models are simultaneously satisfied only if performance is a linear function of time. Evoked potentials (Coles et al., 1985) and d' (Dosher, 1981) illustrate results.     

The bias of an estimate of coupled slack in stochastic PERT networks

An earlier paper presented a method for using response times to construct a PERT network representing the mental activities in a task. A key role in this method is played by parameters called coupled slacks. The magnitudes of the coupled slacks are used to test the validity of a PERT model and to deduce aspects of the structure of the network. In a stochastic PERT network, the usual estimate of coupled slack based on the mean completion times can be biased. Distribution-free upper and lower bounds are derived for the bias, and it is shown that the bias approaches zero as the probability of relatively long prolongations increases. It is also shown that the expected value of the estimate of coupled slack between two activities is negative only if the two activities are in a Wheatstone bridge. Finally, an estimate of the increase in completion time due to prolonging an activity, given that the prolongation did in fact increase the completion time, is derived.     

The Cognitive Social Network in Dreams: Transitivity,Assortativity, and Giant Component Proportion Are Monotonic

For five individuals, a social network was constructed from a series of his or her dreams. Three important network measures were calculated for each network: transitivity, assortativity, and giant component proportion. These were monotonically related; over the five networks as transitivity increased, assortativity increased and giant component proportion decreased. The relations indicate that characters appear in dreams systematically. Systematicity likely arises from the dreamer's memory of people and their relations, which is from the dreamer's cognitive social network. But the dream social network is not a copy of the cognitive social network. Waking life social networks tend to have positive assortativity; that is, people tend to be connected to others with similar connectivity. Instead, in our sample of dream social networks assortativity is more often negative or near 0, as in online social networks. We show that if characters appear via a random walk, negative assortativity can result, particularly if the random walk is biased as suggested by remote associations.     

Short-term memory capacity: magic number or magic spell?

Previous experiments have found that memory span is greater for items that can be pronounced more quickly. For a variety of materials the span equals the number of items that can be pronounced in about 1.5 s, presumably the duration of the verbal trace. This suggests a model for immediate recall: The probability of correctly recalling a list equals the probability that the time to recite the list is less than the variable duration of the trace. Recall probability for lists of various lengths was determined for six materials. Later, subjects read the lists aloud. The standard normal deviates corresponding to probability of correct recall were linear in pronunciation time. Evidently, over subjects, a normal distribution is a reasonable approximation of the distribution of the trace duration. The mean and variance of the trace duration were estimated. The mean (1.88 s) agrees well with previous estimates, and the model accounts for 95% of the variance in immediate recall.