Kahneman and Treisman (1984, p.55) have succinctly described the main disagreement between early selection and late selection theories of attention: 'The classic question of attention theory has always been whether attention controls the build-up of perceptual information, or merely selects among the responses associated with currently active percepts.' Early selection theories hold that attention serves to select which one of a number of stimuli will be further semantically processed and stored in long term memory. On the other hand, more recent late selection theories maintain that selective attention operates after all stimuli are semantically processed. This essay briefly examines and discusses the main findings and criticisms that have caused the transition from early selection models to late selection models of attention. Apart from these two extreme positions, theories that combine aspects of both theories are also discussed.

The essential question which, if answered, would provide evidence on whether stimuli are semantically processed before or after selection is what people know about the unattended information. If the subject does not know the meaning of unattended information, this would mean that it has been discarded before it has been processed for meaning. One of the first approaches to this question was attempted by Cherry (1953, in Parkin, 1999) who carried out an experiment in which subjects were required hear two different messages simultaneously, one in each ear, but only pay attention to one of them. To make sure that subjects were not attending to the other message, they were asked to shadow the attended message, that is to recite aloud everything they heard. Subjects could report the physical characteristics of the voice in the unattended message, but not its meaning. Cherry concluded that selective attention completely overshadows the meaning of unattended stimuli. Broadbent's (1958, in Eysenck & Keane, 1995) 'filter' model of attention was based on a similar experiment. Subjects were presented dichotically with pairs of digits and were instructed to shadow whichever message they preferred. A strong preference was shown for attending ear-by-ear. Broadbent theorised that stimuli are selected according to their physical attributes and unattended messages are not semantically processed and discarded. According to Broadbent, switching attention from one stimulus to another is also quite difficult.

These two previous experiments had numerous weaknesses. In Cherry's experiment subjects were asked about information in non-shadowed messages after they had completed the shadowing phase. This information may have been perceived and processed, but forgotten later. In Broadbent's experiment there was no particular meaning to be extracted from the pairs of digits presented, so there was no sufficient evidence to draw conclusions about semantic processing. Moreover, it was later made clear that it is possible to easily switch attention from one ear to another, when the meaning of the unattended message was relevant to the shadowed message.

Experiments by Grey and Wedderburn (1960, in Parkin, 1999) and Treisman (1964, in Parkin, 1999) demonstrated that subjects select which message to attend to by following the meaning of a sentence. This suggests that humans can select a stimulus according to its meaning not only its physical cues. This could directly provide evidence for a late-selection model. However, Treisman (1960, in Kahneman and Treisman, 1984) accounted for this finding with the 'attenuator' model, which is still an early selection theory. She proposed that the filter described by Broadbent is actually more flexible and the unattended message can be partly semantically processed. Treisman's attenuation theory was a good alternative to the filter model, as it could explain the new findings. However, it has been criticised for lack of clarity and for being a half-measure. Since subjects' attention appears to be directed according to the meaning of a sentence, one could also claim that they can actually fully process stimuli for meaning before selection.

In the area of visual attention, Johnston and Dark (1986, in Eysenck & Keane, 1995) presented evidence that unattended visual stimuli undergo little or no semantic processing. However, counter evidence comes from recent studies with subjects suffering from simultagnosia. Patients are unable to attend to more than one object at any given time, even when objects are close together. However, when an unattended word is semantically related to the attended word, patients can report both words (Caslett & Suffran, 1991, in Allport, 1993).

Experiments on which early-selection models have been based appear problematic. Many authors (e.g. Deutsch & Deutsch, 1963, in Eysenck, 1995; Parkin, 1999) have argued that the shadowing task is too demanding for the subjects to be able to process the unattended stimuli in a way equivalent to real world conditions. It should also be noted that theories for attention were developed on the basis of either auditory or visual experiments solely, although it was not evident that the same mechanisms are employed in all types of attention.

Following the findings by Gray and Wedderburn (1960, in Parkin, 1999), Deutsch and Deutsch (1963, in Eysenck & Keane, 1995) formed the first account of a late selection theory, proposing that all stimuli are fully analysed and attention merely selects the most important stimulus to respond to. However, it was difficult to provide evidence for this position, as an experiment by Treisman and Riley (1989, in Eysenck & Keane, 1995) showed that subjects detected much less target words in non-shadowed messages. More evidence for late selection came from studies on subliminal perception. Some studies have shown that an unnoticed aspect of a stimulus, can influence subjects' behaviour (Parkin, 1999). This indicates that unattended features of stimuli can be semantically processed. However, there is still no evidence that unattended messages are fully analysed.

In the face of lack of evidence for the Deutsch and Deutsch theory, Johnston and Heinz (1978, in Parkin, 1999; 1978, in Eysenck & Keane, 1995) proposed a less rigid model in which selection is possible both before and after semantic processing of information. They assumed that there is a limited central capacity, which is carefully allocated to various tasks including reasoning, auditory attention, visual attention and other functions. Early selection was considered to consume fewer resources than late selection. Thus, when a task involves physical cues early selection should be preferred. However, there is evidence that dissimilar tasks, for example visual attention and auditory attention, interfere less with each other when performed simultaneously than more similar tasks do (e.g. Allport, 1972; Brooks, 1967, both in Parkin, 1999). Relevant evidence from neuropsychological studies (review by Allport, 1993) seems to suggest that rather separate mechanisms control different kinds of attentional tasks. The concept of a central capacity, therefore, appears ill defined. It is still possible, however, that we can process information semantically both before and after selection according to the task demands on the capacity of the processing system, which, according to Welford's (1952,in Eysenck & Keane, 1995) well supported 'bottleneck' idea, is unable to make two decisions about the appropriate responses to two different stimuli at the same time.

The evidence and theories mentioned in this essay lead to the conclusion that humans can detect meaning in unattended stimuli. There is no evidence, however, that this semantic processing is as complete as the processing of attended stimuli. Nevertheless, it could also be said that this conclusion is a rough generalisation, as there is no clear indication that there is a unique attentional mechanism responsible for attending all kinds of stimuli. Visual attention, for example may be controlled by different mechanisms than auditory attention. There is still no theory that provides a clear and testable explanation for all the known evidence and localises attentional selection. The main reason for this difficulty is that it is almost impossible to devise an experiment that will localise selection without interfering with the subjects' natural functioning.

REFERENCES

Allport, A. (1993). 'Attention and Control: Have We Been Asking the Wrong Questions? A Critical Review of Twenty-Five Years'. In D.E. Meyer & S. Kornblum (Eds.), Attention and Performance (Vol. XIV). London: MIT Press.

Ellis, H.C. & Hunt, R.R. (1993). Fundamentals of cognitive psychology (5th ed). Brown & Benchmark. Madison

Eysenck, M.W. & Keane, M.T.(1995), Cognitive psychology : a student's handbook (3rd ed.), Erlbaum, Hove

Kahneman, D. & Treisman, A.(1984). 'Changing Views of Attention and Automaticity. In Parasuraman, R. & Davies, D.R. (Eds.). Varieties of attention. Academic P. Orlando

Kinchla, R.A.(1992).'Attention'. Annual Review of Psychology. 43: 711-42

Parkin, A.J.(1999).Essential Cognitive Psychology. Draft 23-3-99. University of Sussex.