Schneider and Shiffrin (1977)
Schneider and Shiffrin define selective attention as "control of information processing so that a sensory input is perceived or remembered better in one situation than another" (p. 4). The concept of selective attention is predicated on the assumption that attention resources are limited. That is, "It is because processing capacity is overloaded in numerous situations that a subset of information arriving must be given special attention" (p. 4).
Two types of selective attention deficits are distinguished. First, divided-attention deficits occur when the subject must allocate attention resources to additional inputs. Attempting to follow two conversations at once illustrates this deficit. Second, focused-attention deficits occur when the subject is distracted by irrelevant inputs, even though attention is directed to a particular input. Attempting to listen to a conversation without distraction from other conversations illustrates this deficit.
Early attention theorists such as Broadbent argued that processing is parallel up to a point, after which additional processing is limited by a filtering mechanism. In these models, the filter was presumed to act in an all-or-none fashion such that information not selected for processing was simply lost.
The notion of an all-or-none filter was challenged by findings showing that information on non-attended channels is sometimes processed. One possible explanation was that attention might be switched to the non-attended channel briefly. However, Treisman and Riley (1969) found that whether information presented on a non-attended channel is processed depends on information type. This finding could not be reconciled with an all-or-none attention filter, and Treisman proposed that the attention filter attenuates messages on non-attended channels, but that information arriving on these channels is processed at least to some extent.
Later, theorists proposed that the attention bottleneck occurs in response selection or memory search, and that all sensory inputs results in temporary memory activation. As Schneider and Shiffrin explain, "Thus, the limitation on the subjectís capacity lay in the limited rate at which the subject could search and decide about the active features before they were lost from short-term memory" (p. 6). Late selection models can be represented graphically as follows:
The goal of the current paper was to provide a theory for understanding more precisely the conditions under which attention limitations occur. Two visual search tasks were used. In the first task, subjects were shown stimuli in a rapid succession of displays. The goal was to judge whether a target stimulus (or stimuli) had been presented, and accuracy was the dependent measure. Frame display duration and size, memory set size, and consistency of target-distractor mappings were manipulated. In consistent trials, the targets and distractors were distinguished by category (e.g., letters or numbers). In the varied mapping trials, targets and distractors were from the same category.
M = 2 (F, R)
F = 4
In the second task, stimuli were presented in single frames. Memory set size, frame size, and target-distractor mapping consistency were manipulated. Reaction time was the dependent measure.
M = 1 (G)
F = 4
The major results can be summarized as follows. First, for both paradigms, performance in the VM condition was highly dependent on load and frame size. The relations were positive and linear, with steeper slopes for negative trials. Second, and by contrast, performance in the CM condition was largely independent of load and frame size. That is, the functions relating these variables to performance were close to zero.
Schneider and Shiffrin proposed two quantitatively and qualitatively distinct processes to account for their results: controlled search and automatic detection. They proposed that controlled search is a serial process in which a matching decision occurs after comparison of each item in the display to memory set items. The following illustrates:
By contrast, they Schneider and Shiffrin proposed that automatic detection operates in parallel and independent of attention. The mechanism underlying automatic detection is described as follows: "Automatic [processes] do not require attention, though they may attract it if training is inappropriate, and they do not use up short-term memory capacity" (p. 38). Furthermore, automatic processes are difficult to stop once initiated and are not easily modifiable.
Implications for Skill Acquisition
Schneider and Shiffrin proposed that three variables affect skill acquisition: consistent of target-distractor mapping, amount of training, and cognitive load. Presumably, with enough training, automatic responding will develop under consistent training, irrespective of load. Cognitive load will affect the rate of acquisition. One implication for training is that only consistent task components can be trained. Thus, a possible strategy for training is to identify consistent task components, through task analysis, and then to design training for these components.
Schneider and Shiffrin do not seem to make a distinction between automatic detection of featural information and automatic detection that requires more training. This is significant because others (e.g., Treisman) have argued that while simple featural search and trained automatic responding appear similar, they have different origins of independence from attention. Automatic gains its independence from attention through practice, and reflects automatic attention attraction. By contrast, automatic feature detection is an innate process supported by populations of feature detectors. Logan makes a further distinction between these processes by arguing that automatic detection is a post-attentive process that is dependent on attention.
Summary of Two Types of Processing
The major characteristics of automatic detection can be summarized as follows:
The major characteristics of controlled search can be summarized as follows: