Attention: Pattern Recognition
Explanations of pattern recognition
Pattern recognition involves making sense of and identifying the objects we see. This topic is closely related to perception, which explains how the sensory inputs we receive are made meaningful.
Two explanations for how we perceive objects are the template matching hypothesis and the feature detection model.
A template is a pattern used to produce items of the same proportions. The template-matching hypothesis suggests that incoming stimuli are compared with templates in the long term memory. If there is a match, the stimulus is identified.
For example the letter A may appear in many forms:
Either all possible forms have their own template or, with a little 'tweaking', all of the patterns can match one template for the letter A. However, sometimes patterns are ambiguous and fit a template for another class of patterns:
Feature detection models, such as the Pandemonium system for classifying letters (Selfridge, 1959), suggest that the stimuli are broken down into their component parts for identification. For example, an E has three horizontal lines and one vertical line.
This explanation is less rigid than the template model and allows for a degree of variation in the stimuli. However, this greater flexibility means that the stimuli must be limited to letters. If this were not the case, we would identify any pattern with three horizontal and one vertical lines as an E.
Biological mechanisms and pattern recognition
Evidence from Hubel & Wiesel provides biological support for the feature detection model. Cells in the part of the cerebral cortex devoted to visual information are sensitive to a range of lines with differing orientations. These cells have been called feature detectors.
Feature detectors were found by using microelectrodes to measure the activity of neurons in the brains of anaesthetised monkeys and cats whilst bar shaped visual stimuli were being presented.
Hubel & Wiesel identified three different types of cells. Can you name them?
Feature detectors provide good evidence for pattern recognition having a biological basis. These experimental techniques cannot be used in humans so the results are generalised. However, the fact that humans are more likely to confuse letters that have similar features (for example, 'b' and 'd' rather than 'b' and 'z') suggests that feature detectors may be at work.
Context and pattern recognition
Biological explanations of pattern recognition do not account for the effect of context on perception. In other words, our own expectations and knowledge can influence our perception of patterns. This has been called top-down processing (whereas perception based on features of the stimuli is a bottom-up process).
Evidence that context influences perception comes from Palmer (1975). He presented participants with pictures of familiar scenes and then briefly flashed a picture of an object. If the object fitted with the context of the scene, identification was very good. However, participants performed poorly if the object was not related to the scene.
Scene:Object:Identification?BedroomAlarm clockVery goodBathroomSpadeVery bad
This suggests that people's expectations influence their ability to perceive objects.
The evidence suggests that the process of pattern recognition involves elements of both bottom-up and top-down processes. The two processes combine to help us to perceive stimuli quickly and to select what is most likely if patterns are ambiguous.
Research into face recognition
Face recognition is a special form of pattern recognition, which allows us to perform the highly adaptive task of identifying individuals and picking up information from their expressions.
In their experiment, Bradshaw & Wallace used Identikit faces in pairs. Participants were asked to decide whether the faces were the same or different. They found that participants made quicker decisions if more features of the face were different. Bradshaw & Wallace (1971) concluded that faces were recognised by a bottom-up process. In other words, by processing features such as eyes and nose separately we can build up a representation of the whole face.
This explanation is rather simple. Research from Sergent (1984) suggests that we process faces as a whole and not by serial processing of individual features: Often the spatial relationship between features is as important as the features themselves.
This notion is supported by research into Identikit faces made from slices of different celebrities' faces. It is harder to identify the celebrities if the sections of their faces are closely aligned because we see the two parts as one whole. If the sections are misaligned, identification is much easier because we do not view them as a whole (Young et al., 1987).
It seems that theconfiguration of the face is particularly important when we process it. This could include three aspects (Bruce, 1995):
spatial relationship between features (e.g. distance between nose and mouth)
interaction between features (e.g. how the shape of the nose affects perception of mouth shape)
holistic processing of faces (i.e. faces are not processed serially).
Theories explaining face recognition
Face recognition is a special form of pattern recognition, which allows us to perform the highly adaptive task of identifying individuals and picking up information from their expressions.
Identifying a face involves a number of stages, including recognising the face as familiar, working out where we have seen it before and putting a name to it. Failure at one stage causes particular problems in the process of identification.
Bruce & Young (1986) proposed a model to account for the different stages involved in face identification:
This model of face recognition has much support from research studies, particularly those involving the use of brain damaged patients (For example; Young et al., 1983) and PET scanning to view areas of activity in the brain whilst different tasks are performed (Sergent & Signoret, 1992).
One problem of the model is that the components are not all explained fully. This is particularly true of the 'cognitive system' which was included in the model to explain processing not accounted for by the other components.
Kahneman's capacity model
Sometimes we are able to attend to more than one input at a time. This notion of divided attention led Kahneman (1973) to suggest that a limited amount of attention is allocated to tasks by a central processor. Many factors determine how much attentional capacity can be allocated and how much is needed for each task.
Kahneman provided a more flexible explanation of attention than the focused attention theorists did - we canattend to more than one thing at a time, particularly if we are skilled at a task. However, the capacity model fails to explain exactly how the allocation decisions are made.
Allport's module resource theory
Allport (1980) proposed that a number of limited-capacity processing modules exist. This notion can explain how we can easily divide our attention between dissimilar tasks (using different modules) but not between similar tasks (competing for resources from the same module).
This theory does not state how many modules there are, so any number could be made up to explain good performance in tasks requiring divided attention.
Research into controlled and automatic processing
The more we practise, the less attention we have to allocate to a task. It may become automatic if we can carry out the task without having to think about it.
Schneider & Shiffrin (1977) make a distinction between controlled and automatic processing:
Controlled processingAutomatic processinglimited capacityno limited capacityserial processing - requires focused attentionparallel processing - requires no attentionenables flexibile processing of novel situationslearned through practice and difficult to alter
This distinction is supported by evidence that simple visual features, such as shape or colour, can be detected automatically but combined features, such as shape and colour, need directed attention. With practise, these controlled processes can become automatic and difficult to change.
It seems that automatic processes can interfere with tasks requiring controlled processing. An example is the Stroop effect. Try naming the colour these words are written in as fast as you can:
You probably found it hard to ignore the word's meaning - if you see blue ink but the word says "red", it is very difficult to choose between the colour and the meaning of the word. This is because the meaning is normally more important than the colour so we process that information automatically.
One criticism is that there is no explanation of how a controlled process becomes automatic. Does practice speed up the processing or change the way it is processed?
Slips associated with automatic processing
People often make mistakes when carrying out automatic processes. These action slips have been investigated byReason (1992), who asked participants to keep a diary of the mistakes they made over two weeks.
The errors could be placed into five categories:
Description:Example:Category:Mixing up the objects used for different tasksMistaking a pencil for a penDiscrimination failureForgetting what you have done and doing it againWriting the same thing twiceStorage failureLeaving out or muddling up the stages in a taskSealing the envelope before putting the letter inSubroutine failureForgetting what you intended to do and doing something elseGoing to the post box and ending up at the shopTest failureMixing up the stages of different tasksPosting your pen and putting the letter in your pocketProgramme assembly failure
Reason concluded that actions are controlled by:
closed loop: slow processing involving conscious and deliberate attention for learning and switching tasks
open loop: fast processing involving no attentional control for well practised tasks
Automatic tasks are not subject to conscious monitoring, therefore action slips occur in these tasks under open loop control.
However, the results collected from diary studies are unreliable because people may not be aware of all the slips they make and not write them down. Trying to create situations to result in action slips in the lab leads to problems of ecological validity.
Other researchers have explained action slips in terms of existing schemata (memory sets for processes, things and events) being used automatically when conscious control should have taken over (Norman, 1981).
Describe one explanation of focused attention.
Describe one explanation of pattern recognition.
Discuss research into slips associated with automatic processing.
Studies of focused attention
One of the earliest studies of focused attention involved the cocktail-party phenomenon(focusing on one conversation whilst ignoring others). Cherry (1953) investigated it using the dichotic listening task:
Method: Participants listened to a different message in each ear at the same time whilst shadowing (repeating) one of the messages.
The diagram below shows Cherry's method:
Results: Little or nothing of the unattended message could be recalled.
Conclusion: Attention can be focused on one stimulus and most other information is blocked.
Support for Cherry's work comes from Moray (1959). Seven words were repeated 35 times in the unattended ear, but subsequent recall was no better than chance.
However, Cherry found that some features of information presented to the unattended ear seem to get through the block, such as a change from a male to female or loud to quiet voice (physical features). The sound of your name (meaningful or semantic information) may also 'grab' your attention.
Broadbent (1954) used the split-span procedure to investigate focused attention:
Method: Three digits presented to one ear, whilst three different digits were presented to the other ear. Participants had to write down as many digits as they could remember. This was done in one of two ways:
The diagrams below show Broadbent's split-span procedure:
Results: Accuracy was better for ear-by-ear recall than pair-by-pair recall. Given the choice, participants preferred ear-by-ear recall.
Conclusion: Attention can only be focused on one 'channel' of information at a time and switching 'channels' is difficult.
Broadbent (1954) thought that we select one 'channel' of information for attention, for example the left or right ear, based on physical characteristics of the information. This means that information is not processed semantically (for meaning) until after it has been selected for attention. This is an early selection model.
Broadbent's model encouraged others to investigate the area of focused attention and is supported by evidence from the dichotic listening task and split-span procedure. However, it has a number of significant weaknesses:
Information can be processed for meaning before selection (e.g. sound of your name grabs your attention).
Attention can be switched easily from one channel to another (Gray & Wedderburn, 1960). The following pictures show the possible responses to Gray & Wedderburn's task, which do you think was most easy for participants?
A 'category-by-category' response was easiest for the participants, showing that it is possible to switch between different channels.
Treisman's attenuation model
Treisman's (1964) model found solutions to Broadbent's problems. It is also an early selection model based on physical characteristics of the information. However, Treisman's filter attenuates (weakens), rather than eliminates, the unattended information.
A second filter processes the information for meaning, which may result in an attenuated channel being selected if it is important, for example: your name or an alarm call like 'help' or, to use Treisman's term, reaches the threshold level of intensity.
Treisman's model is supported by the work of Gray & Wedderburn (1960), which showed that channels could be switched easily to produce more meaningful information. Her own work, using speech shadowing, showed that an unattended message can be incorporated into the attended message if it makes more sense (Treisman, 1960):
Critics of Treisman's model think that it is too complicated and does not explain how the process of attenuation occurs.
Deutsch & Deutsch late selection model
The Deutsch & Deutsch (1963) model is called a late selection model because they claim that all information (attended and unattended) is analysed for meaning in order to select an input for full awareness. Whether or not information is selected is dependent on how relevant it is at the time.
The same evidence supporting Treisman's model supports this theory. However, the Deutsch & Deutsch model explains the process of focused attention more simply. More support comes from Moray (1969), who paired electric shocks with a word to condition a galvanic skin response (GSR) when the word was spoken. A GSR was produced even when the word was presented to the unattended ear and the participants were unaware of it.
Further evidence for the late selection model is that unattended messages can influence participant's understanding of the meaning of ambiguous sentences (MacKay, 1973):
It seems unlikely that all information should be processed semantically before we are made aware of it. This suggestion is backed up by evidence that we are better at spotting key words in attended messages than unattended messages - according to Deutsch & Deutsch we should be equally as good at each.
In this Learn-it, you have reviewed the research studies and models into focused attention. Here's a little test to see how much you remembered:
Share| Email| PrintGlossary Action slipsPerforming unintended actions, particularly during automatic processes cell contentcell content Automatic processingProcessing that does not require any focused attention Bottom-up processingPattern recognition based on features of the stimulus Closed loopSlow processing involving conscious and deliberate attention for learning and switching tasks ConfigurationArrangement of features with respect to each other Controlled processingProcessing that requires focused attention Dichotic listening taskMethod for investigating focused attention involving listening to two messages simultaneously Divided attentionAttending to more than one stimulus simultaneously Feature detection modelPattern recognition based breaking down the stimulus into its component features Feature detectorsCells in the visual cortex that respond to lines or edges Focused attentionAbility to selectively attend to some information stimuli whilst ignoring the rest Galvanic skin responseChanges in skin resistance in response to changes in arousal IdentikitDrawings of facial features combined to construct a face Open loopFast processing involving no attentional control for well practised tasks Parallel processingAbility to process more than one stimulus simultaneously Pattern recognitionMaking sense of and identifying perceived objects PerceptionProcess of making sensory inputs meaningful SchemataMemory sets for whole processes, things and events Serial processingFocusing attention on one stimulus at a time ShadowingRepetition of a message during a dichotic listening task Split-span procedureMethod for investigating focused attention involving listening to different information in each ear and writing it down Stroop effectDemonstration of how automatic processes can interfere with tasks requiring controlled processing Template matching hypothesisIncoming stimuli are compared with templates in the long term memory Threshold level of intensityAccording to Triesman, this is reached if an attenuated channel is selected for attention Top-down processingExpectations and knowledge of a particular context influence the perception of patterns
(source: S-cool, the revision website)
