Models of Attention

Models of Attention

Early selection theories

Broadbent’s model of selective attention or Filter theory

The theories described here belong to the group of filter and bottleneck theories. A filter blocks some of the information going through and thereby selects only a part of the total of information to pass through to the next stage. A bottleneck slows down information passing through. The models differ in two ways. First, do they have a distinct “filter” for incoming information? Second, if they do, where in the processing of information does the filter occur (early or late)?

Broadbent’s Model

According to one of the earliest theories of attention, we filter information right after we notice it at the sensory level. Multiple channels of sensory input reach an attentional filter. Those channels can be distinguished by their characteristics like loudness, pitch, or accent.

The filter permits only one channel of sensory information to proceed and reach the processes of perception. We thereby assign meaning to our sensations. Other stimuli will be filtered out at the sensory level and may never reach the level of perception.

Broadbent’s theory was supported by Colin Cherry’s findings that sensory information sometimes may be noticed by an unattended ear if it does not have to be processed elaborately (e.g., you may notice that the voice in your unattended ear switches to a tone). But information requiring higher perceptual processes is not noticed if not attended to (e.g., you would likely not notice that the language in your unattended ear switches from English to German).

Selective Filter Model

The evidence began to suggest that Broadbent’s model must be wrong. Moray found that even when participants ignore most other high-level (e.g., semantic) aspects of an unattended message, they frequently still recognize their names in an unattended ear. He suggested that the reason for this effect is that messages that are of high importance to a person may break through the filter of selective attention. But other messages may not. To modify Broadbent’s metaphor, according to Moray, the selective filter blocks out most information at the sensory level. But some personally important messages are so powerful that they burst through the filtering mechanism.

Attenuation Model

To explore why some unattended messages pass through the filter, Anne Treisman conducted some experiments. She had participants shadowing coherent messages, so they must have been somehow processing the content of the unattended message.

Moray’s modification of Broadbent’s filtering mechanism was clearly not sufficient to explain Treisman’s findings. Her findings suggested that at least some information about unattended signals is being analyzed. Treisman proposed a theory of selective attention that involves a later filtering mechanism (Figure 4.9). Instead of blocking stimuli out, the filter merely weakens (attenuates) the strength of stimuli other than the target stimulus. So when the stimuli reach us, we analyze them at a low level for target properties, we pass the signal on to the next stage; if they do not possess those target properties, we pass on a weakened version of the stimulus. In a next step, we perceptually analyze the meaning of the stimuli and their relevance to us, so that even a message from the unattended ear that is supposedly irrelevant can come into consciousness and influence our subsequent actions if it has some meaning for us.

Late-Filter Model

Deutsch and Deutsch developed a model in which the location of the filter is even later. They suggested that stimuli are filtered out only after they have been analyzed for both their physical properties and their meaning. This later filtering would allow people to recognize information entering the unattended ear.

Note that proponents of both the early and the late-filtering mechanisms propose that there is an attentional bottleneck through which only a single source of information can pass. The two models differ only in terms of where they hypothesize the bottleneck to be positioned.

Late selection theories

Neisser’s synthesis model

A Synthesis of Early-Filter and Late-Filter Models

Both early and late selection theories have data to support attention models. In 1967, Ulric Neisser synthesized the early-filter and the late-filter models and proposed that there are two processes governing attention:

• Preattentive processes: These automatic processes are rapid and occur in parallel.

They can be used to notice only physical sensory characteristics of the unattended message. But they do not discern meaning or relationships.

• Attentive, controlled processes: These processes occur later. They are executed serially and consume time and attentional resources, such as working memory.

They also can be used to observe relationships among features. They serve to synthesize fragments into a mental representation of an object.

A two-step model could account for Cherry’s, Moray’s, and Treisman’s data. The model also nicely incorporates aspects of Treisman’s signal-attenuation theory and of her subsequent feature-integration theory.

Feature‐Integration Theory

Treisman (1980) started this theory with simple foundations, however the current form of the theory is complex and complicated. Three different aspects

of the theory are considered.

a. Basic elements of the theory

b. Research on the theory

c. Current status of the theory

Basic Elements

1. Treisman suggests that we look at a scene, two kinds of attentional processes take place.

2. In the beginning, all parts of the scene are processed at the same time called distributed attention (divided attention). Such attention is

automatic, effortless, fast and is carried out as a parallel process.

3. The second kind is focused attention, which involves consciously driven, effortful, slow processing carried out as a serial process.

Research on Theory

1. Treisman suggests that we look at a scene, two kinds of attentional processes take place.

2. In the beginning, all parts of the scene are processed at the same time called distributed attention (divided attention). Such attention is

automatic, effortless, fast and is carried out as a parallel process.

3. The second kind is focused attention, which involves consciously driven, effortful, slow processing carried out as a serial process.

Illusory Conjunctions

1. Later research showed that we make in search when two features in two objects are confused. Presented with blue N and green T, results in blue T.

2. For our auditory sense we make similar errors.  Given “dax” and “kay” as nonsense syllables participants respond by the word “day”.

3. Such errors (illusory conjunctions) are cussed by object proximity or if there is distraction in attention to begin with.

4. Presence of illusory conjunctions suggests that features are processed separately and then put together later in our minds (binding problem).

5. Lack of focused attention leads to illusory conjunctions and thus form the basis of integrated perception of features or binding problem.

Current Status of the Theory

1. Over 25 years Feature‐integration theory has gone through major revisions and has become a complex theory.

2. Errors based on illusory conjunctions can be reduced with practice (Coren et al., 2004; Treisman, 1992)

3. For some tasks distributed and focused attentional processes loose distinction.

4. Theory awaits new neurophysiological insights. However serves as a strong framework to explain visual attention.

Models of Attention

Late Selection Theories

A second general type of structural theory of attention assumed that selection took place late in the processing sequence, after all inputs were processed for meaning.  An outline of a late-selection model of attention, known as the Pertinence Model is presented in Figure 6.12.

Deutsch and Deutsch - The Pertinence Model

The pertinence model was first proposed by Deutsch and Deutsch in 1963 and was revised several times by Norman (1969, 1976). This model introduces the idea of late selection and puts the bottleneck of processing information later than the attenuator model does. The pertinence model suggests that all information is initially processed and that selection only takes place after each input (or message) has been analysed in the memory system. In this model, selection for attention is closer to the responses (output) than the senses (input). Selection then takes place based on the pertinence of the information. Information that is the most pertinent (e.g. relevant and/or important) is most likely to be selected (see Figure).

The pertinence of information is influenced by factors such as expectations and context (these are called top-down factors).

There is some support for the pertinence model from research that shows that information in an 'unattended" message is processed and analysed. MacKay (1973) found that, when participants were asked to shadow an ambiguous sentence in one car, the perception of the meaning of the sentence was influenced by the message in the other ear.

A potentially serious problem of this model is the finding that the information from most unattended messages is completely lost (e.g. Cherry, 1953). Norman (1969) has suggested that this is because the information is stored in short-term memory and is forgotten very quickly. He found that participants could remember the last few words of an unattended message if asked to recall them immediately after the message ended. (In most experiments the participants are asked to recall the information after they have finished shadowing and, since this carries on after the messages have finished, the unattended message has been forgotten.)

However, there are a number of other problems with the pertinence model. As Solso (1979) points out, this model is uneconomical. It suggests that all information is analysed for meaning, yet most of it would not be required or used. In addition, if all incoming information is analysed then we would need a very large processing capacity.

Most of the experimental evidence discussed so far indicates that information in the unattended is processed to some degree. Both the Treisman and the Deutsch and Deutsch models can explain this. The essential difference between the models is where the selection for further attention takes place - early (in the Treisman model) or late (in the Deutsch and Deutsch model).

Limited capacity models of attention

Capacity Models

Psychologists have become more interested in capacity Demands of different tasks (Kahneman,1973). Like there is information coming in and then goes to filter and only relevant information is selected. Different tasks different demands attention diverted to one task to other task. Tasks require mental effort. People may have some control over where the bottleneck occurs (Johnston& Heinz, 1978)

Kahneman's Capacity Model

Attention and Effort" was a major work of kahneman (Kahneman, 1973). He shifted the focusfrom bottleneck to capacity. There is flexibility in attention, like we can change our attention fromone thing to other thing. There is a lot of evidence that our bottle neck is actually adjustable and itcan be move early to late. A general limit on a person's capacity to perform mental work. Aperson has considerable control over how this capacity is allocated.

Daniel Kahneman

He has been doing his work at Princeton University. He has been a pioneer of cognitive psychology not only in the area of attention but on other areas.

Kahneman's Capacity Model

In this model there are many miscellaneous determinants that impact sensory system. Something happens that trigger arousal. Arousal means some activity starts. Arousal has many manifestations. Available capacity of attention will be allocated depending on the state of arousals. So there is allocation policy.  Then there are some possible activities. Like the amount of attention is paid on a task. At the same time there is a feedback loop. We evaluate how much attention is needed for the task. Then we readjust more capacity to that task.

The factors that have impact on allocation policy are:

Arousal

A physiological state that influences the distribution of mental capacity in the various tasks.

Enduring Disposition

An automatic influence where people directtheir attention.

Monetary Intentions

A conscious decision to allocate attention to certain tasks or aspects of the environment.

Example: World Trade Centre

Boeing 707 flying in cloudy weather, at an altitude 200 feet below the top of the WTC was reported by the airport controller. Alarm buzzed in the airport control tower to signal the danger. The controller radioed the crew to turn around and climb up to 3000 feet. The controller was monitoring seven other planes at the time. His attention was diverted to other planes so he could not pay attention to that plane. So alarm diverted his attention toward that plane.

Neisser's Synthesis model

• Preattentive processes  

–Parallel

– Note physical characteristics

• Attentive processes

– Controlled processes occur serially

–Occur in working memory

In 1967, Ultric Neisser synthesized the early filter and the later filter models. Synthesis of early-filter and late-filter models where there are two processes governing attention: preattentive and attentive processes.

Neisser's Preattentive Processes: These automatic, rapid processes occur in parallel on physical sensory characteristics  They are used to notice only physical sensory characteristics of the unattended message. However, they do not discern meaning or relationships.

Neisser's Attentive Processes: These processes occur after preattentive processes. They are executed serially and consume time and attentional resources, such as working memory. They are also used to observe relationships among features and also synthesize fragments into a mental representation of an object.

More recent work in attention builds on Neisser's distinction between preattentive and attentive processes. It focuses only on the consciously controlled aspects of attention (Cowan, 1995).

Consider a different view of the two processes (McCann & Johnston, 1992). Ac­cording to these researchers, physical analysis of sensory data occurs continually, but semantic analysis of stimuli occurs only when cognitive capacity (in the form of working memory) is not already overtaxed; the capacity also must be sufficient to permit such analysis. Supportive evidence is that people show much faster reaction times when re­sponding to physically discriminable stimuli than to semantically discriminable stimuli.

A two-step model of some sort could account for Cherry's, Moray's, and Treisman's data. Evidence of fully automatic versus fully controlled processes also seems to support this model. Automatic processes may be governed only by the first step of attentional processing. Controlled processes additionally may be governed by the second of the two steps. The model also nicely incorporates aspects of Treisman's signal-attenuation the­ory and of her subsequent feature-integration theory. According to this latter theory, discrete processes for feature detection and for feature integration occur during searches.