Chapter summary
Most of the time we seem to be ‘bombarded' by numerous stimuli coming in through all of our senses at the same time. Despite this, we usually have no difficulty in focusing attention on one stimulus while ignoring others, and this suggests that we have an exceptional ability to filter out unwanted stimuli. Further, because this act seems effortless and because we are unaware of how we do it we have the illusion that it is a simple and straightforward process. In this chapter we discover how difficult the issue of selectivity in attention is to explain, and we review several theories and methodologies designed to understand attention.
Focused auditory attention
In this section you will be introduced to the issue of capacity limitations in attention, explored through the study of auditory attention. We illustrate the most important research methods used in this topic and present the main theories of selective attention.
The cocktail party effect
One of the first investigators of auditory attention was Colin Cherry in the 1950s. His interest, as an engineer, was whether one could build a machine that could selectively filter out unwanted sounds, in order to make sense of a particular signal (such as speech in a noisy environment). Humans can do this quite easily and without a great deal of effort, 36 most of the time. However, the ease with which we can do this betrays the fact that it is a complicated process.
Cherry (1953) presents us with the ‘cocktail party' effect, which is an example of the incredible achievement of auditory selective attention. Suppose you are having a conversation with someone at a cocktail party (as you do). Unless the surrounding noise is unusually loud, it is quite easy to hear and follow what they are saying (provided they are speaking a language you understand and are not mumbling, of course). You might not notice, for example, the particular piece of music being played in the background because you are focusing your attention on what the other person is saying. Also, around you lots of other conversations are going on and similarly you don't really hear what other people are saying because of your focus on the current one.
However, supposing the couple chatting a few metres away from you were to mention your name. It is quite likely, since we are all concerned to some extent over what other people say about us, that you will hear your name being mentioned. The question which arises is, how were you able to detect a word (your name) in a conversation that you were ignoring? How can you ignore something and yet at the same time hear it?
In order to try to understand this phenomenon Cherry (1953) developed the dichotic listening task that involves being presented with two different messages, one to each ear via a set of headphones. The task is to attend to one of the messages closely and involves ignoring the second message to an extent in order to do this successfully. To be sure that the participant is attending to the desired message, he or she is required to repeat the message out loud as accurately as possible, a process known as shadowing.
Early findings indicated that the task outlined in the Crucial concept above was more difficult if the physical features of the attended and unattended message were similar. For example, if both messages were spoken by the same person then performance was worse than if the messages were delivered by different speakers or if one was a male voice and the other a female voice.
A key question concerns how much of the unattended message participants could detect. In terms of physical characteristics. Early studies suggested that only gross physical features were noticeable. For example, if the sex of the voice of the unattended message changed participants would notice this, but would not notice if the message changed from English to German. Cherry reported that his participants were unable to recall any specific word in the unattended message and Moray (1959) claimed that participants were unable to detect a word even when it was spoken 35 times within a trial in the unattended message. From these studies it was concluded that only basic physical features could be detected in the unattended message and that this was because attention was selecting out the unattended message in order to focus on the task of shadowing one message.
Theories of focused auditory attention
Bottleneck theories
You might think that from these sorts of studies it would be relatively straightforward to produce a theory of selective auditory attention and the above findings imply some sort of filtering process whereby the attended message is filtered in and the unattended message is filtered out. This is the basis of Broadbent's (1958) filter theory. Since the amount of information coming in from both channels (the attended and unattended messages) is more than can be coped with reasonably well then one message needs to be inhibited or ignored.
The main features of the model are (see Figure 3.1) as follows:
• Sensory store
Incoming messages are held in a sensory store very briefly.
• Sensory filter
One message is filtered in and the remaining are filtered out (except for their basic physical features). Messages filtered in receive further processing, while filtered out messages are eventually lost.
Figure 3.1 Broadbent's filter model
These elements of attention are said to process messages in the following way:
• The filter operates on the basic physical characteristics of the messages (e.g. sex of the speaker, type of sound).
• Filtering is a ‘winner-takes-all' process, which means that only one message is selected for further processing, the rest are lost.
• Filtering is done consciously, in that people decide what they want to listen to.
Furthermore, people can switch their attention from one message to another.
• The first stage of further processing is to identify the main components of the message (e.g. the words).
Evaluation
Broadbent's model accounts for the apparent finding that very little information, except gross physical features, of the unattended message get noticed. Filter theory also accounts for the fact that performance on the shadowing task is reduced when the similarity between the two messages is increased. An important aspect of Broadbent's approach is that he adopted an information processing perspective and presented one of the first ‘box and arrow' diagrams of a mental process. Information processing theorists make the assumption that people's mental activity mainly involves processing information. They adopt the computer metaphor – that people process information in a similar fashion to the digital computer. We have input (sensory stimulation), we process the input (selectively filter them or abstract meaning), and we emit a response (talk back, laugh, and so on). One criticism of information processing theory is that it ignores meaning (see the discussion on Searle's Chinese Room example in Chapter 9), and that might be where Broadbent's theory begins to break down.
As well as the Gray and Wedderburn (1960) study, other studies reveal that the meaning of the unattended message can be processed. For example, Treisman (1964) has shown that participants will switch between the two channels when the messages themselves are switched. If the unattended channel was not processed for meaning then participants would not know that the messages have been switched, they would just assume that the A B stimuli sensory store further processing response.
If selective attention works on a winner-takes-all principle and is based on physical features rather than the meaning of the message, then if a participant was instructed to attend to the message in the left ear and ignore the right, this should not be difficult. However, suppose that the messages were presented like this:
Left ear Right ear Dear, 7, Jane 9, Aunt, 6 If the participant is asked to repeat all the items presented then Broadbent's theory predicts that recall would consist of either ‘Dear, 7, Jane' or ‘9, Aunt, 6' , since they would be attending to one channel only. However, not only can participants recall all of the items, they recall them as two units ‘Dear Aunt Jane' and ‘976'. In other words, participants do not group the items based on the channel they were presented in but on their meaning.
This finding is important because it suggests that more than just the physical characteristics are processed in the unattended message. Broadbent certainly did not consider that the meaning of the unattended message would receive processing. topic of the material has changed. Findings such as these led Treisman to develop her ‘attenuation' model of attention.
When thinking about the issues in this section, examine how you selectively attend to things in your environment. For example, when you are at home try writing an essay or reading a book with the radio on. Once you are deeply into your work make a note if anything, word or sound, on the radio catches your attention and distracts you from your work. Is it an unusual sound that distracts you or the voice of someone you like or strongly dislike, or is it a particularly favourite tune? Ask yourself how the theories discussed in this chapter fit in with your observation. Activities like this can often inform us about the plausibility of a theory.
Treisman's attenuation model
The key modification in Treisman's model is the assumption that, at the early stage at least, information is processed in parallel and selection is made at a later stage (see Figure 3.2).
Figure 3.2 Treisman's attenuation model
The key features of the model are that early processing is not an ‘all or none' or ‘winnertakes- all' affair but rather that the main message gets through with other information being attenuated (made weaker). Thus the weakened message is still processed to some extent (hence it is also referred to as a ‘leaky' filter model).
• Selection can be based on physical cues in the same way as Broadbent's model.
• If the attenuated information is consistent with the meaning of the message in the attended ear it can intrude and affect performance.
Evaluation
Treisman's model can account for the Gray and Wedderburn (1960) experiment, since the meaning of the message in the unattended ear is being attenuated (and still able to influence performance) rather than being completely ignored. Furthermore, unlike Broadbent's model, this model can better account for Cherry's cocktail party effect, since a significant message in the unattended channel (e.g. one's name) is being processed and not ignored, and hence attention can switch to that channel. However, the attenuation model denies the possibility that stimuli may be selected at an even later stage of processing. There is some evidence that even more information can be processed in the unshadowed message than is suggested by Treisman's model.
Von Wright et al. (1975) used a classical conditioning procedure to pair neutral words with mild electric shocks. As expected when presented with the neutral words, participants showed a small increase in the galvanic skin response (normally taken to mean that they were responding like this because they were expecting the shock). However, when these words were presented in the unattended channel in a dichotic listening procedure, the galvanic skin response was still present despite the fact that participants were not attending to the words. This is made even more interesting by the fact that participants often showed the inflated GSR to the words even when they reported not hearing them. The implication is that unattended information can be processed for meaning, even nonconsciously. Corteen and Wood (1972) have reported similar findings. A B stimuli sensory store further processing response attenuation.
Late selection models (Deutsch and Deutsch, 1963; Norman, 1968)
These results imply that all information is processed in parallel but that selection and filtering occurs much later on. Filtering is then based on whether the information is pertinent or not (i.e. the level of significance of the information to the individual).
Pertinence (the word salience is also used for the same concept) refers to the significance of a stimulus. Pertinent stimuli can either be those that are personally important to us, such as the sound of our own name or the sight of those we care about, or they are those that have a direct bearing on the current task (such as a set of traffic lights when we are driving). The concept is related to the ‘goal-directed' nature of human behaviour and mental processes – we don't passively sit around waiting for stimuli to hit us but rather we are always engaged in an activity and trying to achieve something. This concept may help explain the cocktail party effect and is something that Broadbent clearly did not give enough consideration to.
The pertinence model appears to be more parsimonious (it can explain things more simply and elegantly) as an explanation of selective attention than Treisman's model. Consider, for example, the findings of MacKay (1973). In this study sentences that contained words with more than one meaning were presented, e.g. ‘They were standing near the bank'. This could be taken to mean either ‘They were standing near the river bank' or ‘They were standing near the money bank'. Participants interpreted the sentence depending upon whether the word ‘river' or ‘bank' was presented in the unattended channel. The word provided a context in which to interpret the sentence (hence bank becomes pertinent to the word river or money ). Clearly, the meaning of the information in the unattended channel is not filtered.
In this section we will be addressing the issue of limitations in attention but from a different angle. Rather than get participants to focus on one task and then see how much they were aware of other information around them, an alternative is to find out whether attention can be divided among several tasks. We describe a number of research methods used to study this topic and present the main theories of divided attention.
Divided attention concerns our ability to ‘multitask', i.e. whether we can attend to more than one task at a time. While the dichotic listening task involves trying to attend to only one message, in studies of divided attention the task is to attend to more than one source of information. The questions concern how well we can do more than one task at a time, what kinds of dual tasks are achievable and which are not, and how can we explain this ability.
Early studies have shown two important factors that determine our ability to multitask:
• The similarity of the tasks. Allport et al. (1972) asked participants to learn a set of words while shadowing a spoken message. They found that the words could be learned when they were presented visually but not when they were presented as spoken words. However, if messages were sufficiently different then both could be attended to.
• How well practised we are at the task. Spelke et al. (1976) found that, with practice, students could learn to read a story while writing down a list of words read out to them.
Why do we have these limitations on performance?
Norman and Bobrow (1975) argued that the major factor that determines our ability to multitask is that our attentional resources are limited. Furthermore, this ability can also be affected by the quality of the information we receive (e.g. mumbled speech). Kahneman (1973) also suggested that the amount of resources we give to a task is flexible. For example, we are more alert at times or more motivated to complete a task than at other times. There may also be an upper limit to the amount of resources that are available. Kahneman argued that this upper limit and the amount and type of information we choose to attend to is determined by an ‘allocation policy', which is itself determined by several factors:
• Our physiological state. If we are feeling full of energy then we have more resources to allocate than when we feel tired.
• Our enduring dispositions. The amount of attention we give to tasks is dependent upon what type of people we are: our personality, our habits and our long-term goals.
• Our momentary intentions. Some stimuli may receive more attention than others because they are deemed more relevant to the task, the context we are in can affect what we attend to, and our current mood or personal concerns can affect what we attend to and what we choose to ignore.
This type of theory is known as a capacity theory since it is based on the notion that there is a general-purpose limited-capacity central processor:
• The central processor is a hypothetical mental structure that deals with ongoing tasks.
• It is said to be of limited capacity in that it has an upper limit in the amount of information it can deal with in any one moment.
• It is general-purpose in that it can deal with all kinds of information and tasks. For example, it is not restricted to dealing with information from a single modality.
Evaluation
The assumption of limited capacity has been challenged by several authors, since attention can be successfully divided under certain conditions and can be improved with practice. Furthermore, the assumption of a single general-purpose processor may be incorrect. It may be that several processors exist, each with a specific purpose. For example, consider the desktop PC. Although it has a main processor for doing such things as running programs, opening and saving files or making calculations in a spreadsheet, it also has a separate processor for playing CDs. In Kahneman's model the central processor divides its resources between the ongoing tasks, and while this is mostly true for the PC, the brain may have more specialised processors.
This idea has led to multiple-resource models of attention (e.g. Navon and Gopher, 1979). The idea is that it is not just the amount of resources a task requires that is important but the type of information required. Some tasks can interfere with each other not because they both require a large amount of resources but because they require the same type of resources. For example, reading a book requires processing language and counting to 48 in fours requires processing numbers. With a little practice, these tasks can be done at the same time without interfering with each other and causing errors. However, shadowing spoken words and reading a book both require language processing, so may easily interfere with each other even with practice. Such interference is referred to as ‘crosstalk' and a good example is the Stroop Effect (see Critical study below).
THE STROOP EFFECT (STROOP, 1935)
Imagine a list of words written in different ink colours. Suppose that you were asked to name the colour of the ink. The task is straightforward. However, if the word to be named was itself a colour word and it was printed in a different colour (e.g. the word BLACK printed in blue ink) then the task is not so easy. The Stroop effect is found when participants are slower to name the colour of printed words when the colour word and its printed colour mismatch than when they match. The effect has been replicated many times and is highly reliable. It has been used to explore the influence of anxiety and depression in selective visual attention (see Chapter 8).
Controlled and automatic processes
To some extent, the ability to perform more than one task depends on how accomplished we are at each task. Being able to drive and listen to a talk show on the radio (without crashing) depends on how experienced the driver is and their proficiency with the language and the topic of the discussion. For learner drivers, the radio discussion is a complete distraction and can interfere with driving (otherwise they take in little information from the discussion).
Schneider and Shiffrin (1977) have provided an extensive research programme to address the issue of how behaviour may become ‘automated'. They argue that when learning a new skill, a lot of attention or controlled processing is required, but when we are skilled we can do the task without paying much attention to it at all. During skill acquisition, the behaviours required to carry out the task well become automated, which means that they are performed so well that they do not require many attentional resources and in that sense can be performed automatically.
Generally, automatic processes are said to be:
• rapid – a skilled driver can change gear quicker than a novice;
• effortless and make few demands on attention – the skilled driver can change gear while concentrating on a conversation and not thinking about driving;
• unavailable to consciousness – a novice driver will probably recall most of his or her attempts to change gear during a lesson whereas if you were to ask a skilled driver to demonstrate how they change gear they will probably perform the task less well than if they just did it in a normal driving situation;
• unavoidable – when we repeatedly do something it becomes a habit and we cannot avoid doing it, like looking in the rear-view mirror before overtaking (well, for most people).
Controlled processes, on the other hand, are said to be:
• slow – since controlled processing is applied to novel situations, we aren't aware of a reliable way to respond quickly;
• effortful and make heavy demands on attention – the learner's concentration on the task is intense;
• completely conscious – a lot of conscious effort is allocated to a new task.
The results of a series of experiments reported by Schneider and Shiffrin (1977) may be explained by the distinction between automatic and controlled processes. The experimental design is quite complicated but to give you an idea think about the following everyday example. Suppose that you are driving from Oxford to York and you are not familiar with the route. You know that you have to head towards Birmingham , so you are looking out for signposts to Birmingham . However, when you are near to Birmingham you have to look out for signposts to Leicester . In doing so you will still see signposts to Birmingham but you will have to ignore them. This type of route we might call a varied mapping and this is so because the target city changes and as it does so we have to ignore targets that we were previously searching for (once we are on route for Leicester, we have to ignore signposts to Birmingham). Contrast this with going from Oxford to York needing only to follow signposts to York . We might call this type of route a consistent mapping because the target city is the same and we never have to ignore signposts to York at some later time.
The Schneider and Shiffrin (1977) experiment involves a search task, looking out for a target letter. The two conditions are illustrated below:
Trial 1: Search for T among Varied mapping: V C H T G R Y Consistent mapping: V C H T X R Y
Trial 2: Search for G among Varied mapping: P C H T G J Y Consistent mapping: X D R Y G P B
Trial 3: Search for J among Varied mapping: V T G H Y J M Consistent mapping: V F L J S M U
Note that for the varied mapping condition the search targets appear in other conditions, whereas in the consistent mapping condition the target only appears in the list of letters when it is a target and not in the other trials.
The number of distracters (the number of items appearing in the display during a trial) was varied, as was the presentation time of the list. Sometimes the target was present and sometimes it was not. The results showed that reaction time in the consistent mapping condition was the same regardless of the number of items in the display or the presentation time. Participants said that the target appeared to ‘pop out' and it did not require much effort. However, for varied mapping reaction time was longer the more items there were in the set. In an earlier study, Kristofferson (1972) trained participants on either a varied mapping or a consistent mapping scheme for 30 days. The results indicate that performance at the task with a consistent mapping improved with practice but did not do so with the varied mapping condition. Schneider and Shiffrin (1977) interpret these findings by arguing that automaticity of a task requires a consistent mapping. In our example, it implies that learning the route from Oxford to York is much easier when we only have to look out for the signposts to York than when we have to look out for signposts to a city and only later have to ignore signposts to that city.
Evaluation
The notion of automaticity has been challenged. Hirst et al. (1980) report individuals who were able to learn to read while taking dictation. Since both tasks are not in the same form as a ‘consistent mapping' there should not be, according to Schneider and Shiffrin's account, any opportunity for automaticity. Hirst et al. (1980) argued that automaticity had not taken place, but rather participants had learned to restructure the tasks so that they could do them simultaneously (for example by learning to share attentional resources between the two tasks).
ection 3
Typical Examination Questions
1. Which model of focused auditory attention can best explain the cocktail party effect?
2. What governs our ability or inability to perform two tasks at the same time?
Further reading
Coren, S., Ward, L. M. and Enns, J. T. (1994) Sensation & Perception. Harcourt Brace. Chapter 15.
Cowan, N. (1998) ‘Evolving conceptions of memory storage, selective attention and their mutual constraints within the human information processing system', Psychological Bulletin, 104, 163–91.
Johnston , W. A. and Dark, V. J. (1982) In defence of intraperceptual theories of attention. Journal of Experimental Psychology: Human Perception and Performance, 8, 407–21.
Internet sites
http://psych.hanover.edu/classes/hfnotes3/index.html (someone's lecture notes!)
http://ling.ucsc.edu/~chalmers/mind.html#attention (consciousness and attention essays)
Write up your lab report with this unique application. www.labwriteup.com
This book was first published in 2003 by Crucial, a division of Learning Matters Ltd [ISBN 1 903337 13 5] © 2003 Eamon Fulcher; © 2009 GEFT Consultance Services (geft.co.uk).
All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, photocopying, recording, or otherwise, without prior permission in writing from Geft Consultancy Services, who may be contacted via www.geft.co.uk.