Attention was described by William James (1890, cited in Eysenck & Keane, 2000, p130) as “the taking possession of the mind, in clear and vivid form , of one out of what seem several simultaneously possible objects or trains of thought. Focalisation, concentration of consciousness are of its essence. ” This definition emphasises how attention is thought of as a selective process.
It seems clear from common sense that we cannot attend to all stimuli at once, so some kind of selection must take place as to what information we attend to and process further, and what is disregarded. Since the 1950’s, there has been a great deal of research into selective attention, both auditory and visual. Several different theories and models of selective attention have been proposed.
One central and on-going debate in attention research has been that between early and late selection theories, i. . at what stage of processing a stimulus does selection occur? This essay will compare and contrast early and late selection models of attention The main examples used to illustrate similarities and differences will be Broadbent’s (1958) filter theory model (as cited in Driver, 2001) which was the first cognitive model of auditory attention, and an extreme example of early selection,(and the rival late selection model proposed by Deutsch & Deutsch (1963) .
It will then go on to evaluate these along with other models as including Treisman’s (1960, as cited in Driver, 2001) attenuation model, as to how well they are able to explain the phenomenon of selective attention. Both early and late selection models of selective attention were originally derived from research into auditory attention, attempting to explain how the human auditory system is able to process mixed Various dichotic listening experiments were conducted (Driver, 2001; Naish, 2010), in which participants had different messages played into each ear and were asked to shadow, i. e. epeat the message from just one ear. (They would then be asked questions regarding what they remembered from the message which had been played into the other, non-shadowed ear. In most cases it was found that participants could remember almost nothing about the message in the non-shadowed, i. e. unattended ear) Driver (2001, p55) demonstrates how both early and late selection models can be represented as very simple two stage flow diagrams, illustrating how different early and late models of selective attention all appear to be based on Broadbent’s (1958, as cited in Driver, 2001) original filter theory.
Both early and late selection models can be thought of as having a selective filter or bottleneck (McLeod, 2007; Eysenck & Keane, 2000), which extracts the attended information for further processing while filtering out irrelevant (unattended) information. Both types of model assume that initial processing of all stimuli takes place in parallel, prior to the bottleneck filter, after which the selected information is thought to undergo deeper, serial processing. The main difference between early and late selection models is the position of the bottleneck.
Broadbent’s (1958, as cited in Driver, 2001) model assumes that the bottleneck occurs very early in processing, (near to the stimulus end if the model is represented as a flow diagram) it is assumed that only simple physical properties of a stimulus are extracted in the parallel pre-attentive stage prior to filtering, therefore the unattended stimulus does not undergo any processing for meaning, but only for simple physical characteristics, e. g. the location of the speaker or whether the voice was male or female.
These simple physical characteristics are all that can usually be remembered about the unattended message by participants in dichotic listening tasks. Broadbent (1954, as cited in Naish, 2010) also discovered that if both messages were very short, participants could remember the message from the unattended ear. This led to the assumption that there was a sensory buffer, a very short-lived memory store also known as echoic memory, which could hold on to unattended material for just a few seconds prior to selective filtering (Naish, 2010).
By contrast, late selection models, e. g. Deutsch and Deutsch (1963) place the bottleneck much nearer to the response end of processing. Their model assumes that all incoming stimuli are automatically processed and analysed for meaning, regardless of whether they are consciously attended to or not, with selective filtering occurring only after meaning has been extracted. Late selection models provide a possible explanation for results obtained in some dichotic listening experiments where processing of unattended stimuli did seem to take place.
For example, Corteen and Wood (1972, as cited in Naish, 2010), paired electric shocks with certain words, so that a conditioned galvanic skin response (GSR) took place. Later, when these words were again presented to the unattended ear , (without electric shocks), the GSR still occurred for these words as well as other words from the same category, indicating that processing for meaning had indeed taken place.
Late section theories could also be used to explain the cocktail party effect (Naish , 2010) i. e. if someone is attending to one conversation at a party and their name is mentioned in another conversation in the room, they are able to hear their name and switch their attention to the other conversation. The above examples lead us to consider some limitations of a strict early selection model such as Broadbent’s (1958, as cited in Driver, 2001) filter theory.
Whilst this model explains the results of early shadowing experiments (Driver, 2001), the inflexible nature of Broadbent’s model means thatit cannot fully account for the cocktail party effect, or for the findings of a number of experiments in which varying degrees of processing of unattended stimuli are observed to take place. One such study was carried out by Treisman (1960, as cited in Driver, 2001) in which she found that while shadowing a message played into one ear, participants would sometimes switch to shadowing the other, previously unattended ear when the message they had been shadowing was switched over, i. . they followed the message over to the other ear. Treisman went on to develop an alternative selective attention theory (Treisman, 1960, as cited in Naish, 2010 & Driver, 2001) which argued that unattended stimuli were not completely filtered out, but turned down or attenuated. Normally this would make them too weak to be available for semantic processing, but in certain circumstances, e. g. when words from the unattended message had special significance, such as one’s own name, or words relevant to the attended message, these words would have a lower threshold for identification and therefore would be processed.
Treisman’s model can be considered an early selection one, as it is basically a modified version of Broadbent’s (1958, as cited in Driver, 2001) filter model, where the fixed bottleneck is replaced by a more flexible “attenuator” (Eysenck and Keane, 2000). Treisman’s model can be used to explain the cocktail party effect as well as the other experimental findings mentioned above. In the study by Corteen & Wood (1972, as cited in Naish, 2010), words associated with electric shocks would likely be very significant to the participants, therefore the threshold would be low enough for identification to occur.
The selective ttention models discussed above have all had a great deal of influence in attention research, and have certainly been useful in aiding our understanding of how certain processes might occur. Broadbent’s filter theory (1958, as cited in Driver, 2001) in particular has been immensely influential, with many subsequent models of selective attention in auditory and visual research being based upon its simple, logical structure. However, it should be noted that in real life, selectively attending to information is a very complex process carried out by the brain, and cannot be fully explained by such a simple computational model.
This point has been made by Allport (1980, 1987, 1992, as cited in Driver, 2001) and will be returned to later. There are a number of methodological issues which can be used in criticism of selective attention theories, particularly of Broadbent’s model (Driver, 2001). One of these is that in early dichotic listening experiments, participants were unfamiliar with the shadowing task, so it would have placed heavy demands on their processing capacity just to be able to shadow the attended message (Eysenck and Keane, 2000).
Secondly, participants in these early experiments were questioned retrospectively about the message played to the unattended ear. Their reports of knowing very little about the message could therefore be due to them having forgotten it, rather than there being no processing of the message at all. This problem was addressed in later research by devising experiments in which indirect measures of processing were used, one example mentioned previously was Corteen and Wood (1972, as cited in Naish, 2010) where GSR was measured, giving results inconsistent with Broadbent’s theory.
Another important methodological issue, and according to McLeod (2007), a problem with all dichotic listening experiments is the possibility that participants could simply switch their attention from one channel to the other, leading researchers to falsely suppose that the unattended message was being processed. This point is also emphasised by Lachter, Forster and Ruthruff (2004), who attempted to control for this phenomenon in visual attention experiments, and whose findings support Broadbent’s (1958, as cited in Driver, 2001) filter model.
Driver’s (2001) review provides many examples of similarities between auditory and visual attention research in terms of the early vs late selection debate. Sperling (1960, as cited in Driver, 2001) found evidence for a very brief short-term memory buffer, which was analogous to Broadbent’s echoic memory, and termed ‘iconic memory’. Rock and Gutman’s (1981, as cited in Driver, 2001) findings were consistent with early selection model derived from shadowing experiments, and also subject to the same methodological concerns, as they used retrospective questioning to assess whether unattended information had been processed.
Treisman’s (1988, as cited in Driver, 2001) feature integration theory can be said to bear a very strong resemblance to Broadbent’s (1958) model, this is illustrated very well by Driver (2001, p55) where he simplifies it into a two stage flow diagram consisting of extraction of physical features, followed by integration of features for the selected object. Rival late selection theories include those of Tipper (1985, as cited in Driver, 2001) and various others, as reviewed in Driver (2001), and were based on studies involving negative priming effects, as well as other indirect measures indicating that unattended stimuli were fully processed.
A possible resolution to the long standing early vs late selection debate was proposed by Lavie (1995, 2000, as cited in Driver, 2001) in the form of a perceptual load theory, which, based on the assumption that the system had limited capacity, could incorporate findings in favour of both early, and late selection models. Lavie conducted an extensive review of the literature, as well as conducting her own experiments, and argued that results supporting late selection were normally obtained in situations of low perceptual load, e. g. n undemanding task involving one target and one distractor. The system would therefore have spare capacity for processing of non-target information. Conversely, in situations where perceptual load is higher, as in more difficult target identification tasks, then an early selection explanation tended to be more appropriate as little or no spare capacity would be available. Evidence from neuroscience should also be considered in this evaluation. Woldorff et al (1993, as cited in Naish, 2010) recorded data from event related potentials (ERPs) in the brain following auditory stimuli.
The results provided very strong evidence in support of both early selection and attenuation, attending away from a stimulus reduced the intensity of the signal in the brain. Driver (2001) who was once in favour of late selection, now argues that late selection has been conclusively falsified by evidence from neuroscience. Driver (2001) also reviews further evidence from neuroscience which reminds us that attention is a complex process involving different brain areas and top-down, as well as bottom-up processes, therefore cannot be adequately represented by simple box flow diagrams.
In conclusion, it can be argued, on the basis of the evidence presented in the above discussion, that an early selection model provides a better explanation of the way in which we attend to information than a late selection one. As there is evidence for attenuation, perhaps Treisman’s (1960, as cited in Driver, 2001) attenuation theory is more appropriate than Broadbent’s (1958, as cited in Driver, 2001) theory. Finally, it must be re-stated that attention is a very complex set of processes and cannot be fully explained by any of the simple models discussed here, however they have been very useful in aiding our understanding.