2 Lab 2: Attention: A Limited Resource

CogLab Exercise 8

Historical Background

The concept of attention has been a focus of study since the time of William James. James realized that there were only so many things that could occupy our consciousness at one time. Current investigations of attention, though, grew out of developments in England towards the end of World War II. Researchers in a field called communication theory realized that at some point, information overloaded those trying to process it. They then asked the question, how much can people attend to at one time?

To study this question, Donald Broadbent, from England, devised some very clever techniques. Broadbent took advantage of what was at the time a recent technical breakthrough: stereo reproduction. Broadbent devised a way to use stereo reproduction to see what kinds of information “get through,” even when we are not attending to them.

Actually, the work on attention that Broadbent developed was derived from work by Cherry (Cherry, 1953), using a technique called shadowing. In a shadowing experiment, subjects are seated in the lab, and headphones are placed on their heads. Two separate messages are played through the headphones to the subjects, a different one in each ear. Subjects are asked to verbally repeat (or “shadow”) one of the messages. Though this is not an easy task, with a little bit of practice it can be done.7

If we ask you questions after the presentation of the messages, we find that you most likely retained a good bit of the content of the message you shadowed. We’re more interested, though, in how much information you were able to understand from the ear that was not shadowed–what Cherry and Broadbent called the “unattended channel.” If subjects are asked what the unattended message was about, they are unable to report anything about it. Content, then, is “filtered out.” If subjects are asked to respond when the speaker in the unattended ear switches from English to Spanish, they also cannot do that. If subjects are asked to respond when the message switches from something meaningful (say, a speech by Winston Churchill) to a single word repeated over and over, they also cannot notice that. However, they can usually notice when the speech switches from that of a male to that of a female, or when speech is replaced by music. Subjects can also notice when the location of the unattended message “moves” from one ear to the top of the head.

Models of Attention

What characteristics do the kinds of changes that can be detected share in common? Broadbent noticed that they involved a change in the physical characteristic of the sound, not in the content. Subjects notice a change in pitch (male to female), timbre (speech to music), or location (side to the top of the head). As a result of these findings, Broadbent proposed a model of attention which has come to be known as the “early selection filter model,” shown in Figure 2.1.

Early Selection Filter Models

Notice that the model of attention postulated here has three stages of processing. The first is a detection stage. Information must be detected if it is to be attended. Detection clearly takes place in a shadowing paradigm. If we ask subjects to raise their hands when the message shuts off, or is turned back on, they have little trouble with this. The second stage, the recognition stage, involves analyzing the content of the message, to the point of recognizing what is said. Not only is the message detected, but the content of the message is understood. This takes more processing than simple detection. The last stage of the analysis involves some kind of additional processing. This is probably done by short-term memory, which we’ll discuss in the next section of experiments.

The early selection filter model (sometimes called a “switch” model, because attention can be switched from one channel to another) has some nice qualities. It provides a good explanation of Cherry’s shadowing experiments, and also seems to corroborate our intuitive understanding of attention. However, subsequent findings called this early selection model into question.

Figure 2.1. Broadbent’s Early Selection Filter Model of Attention.

One phenomenon that we are all familiar with has been labeled the “cocktail party effect.” We have all been at parties (or other social gatherings) where we are carrying on a perfectly pleasant conversation with someone. We may be discussing how our summer vacations went, or what we think of the Rangers’ chances next year. Then, all of a sudden, someone having another conversation away from us mentions our name, and we hear our name as clearly as if they had been speaking directly to us. Our attention is “grabbed,” involuntarily, by the conversation in which our names were spoken. A number of things can grab our attention this way. New parents can sleep through an earthquake, but awaken at the slightest cry of their newborn. Or, in a room full of crying children, parents ignore all the other cries and respond only when they hear their own child’s cry. Things like hearing someone mention your hometown or names of old friends seem to jump out as well.

The cocktail party effect presents problems for early selection models. These models contend that the attentional filters edit unattended material out before any recognition takes place. If this were so, however, we would not be able to recognize the distant mentioning of our names, or the cries of our children–these messages would have been filtered out already. A number of other experiments have been conducted which cast doubt on the early selection models. These are detailed in your textbook.

Late Selection Filter Models

Perhaps the most obvious alternative to an early selection filter model would be a late selection filter model. Such a model was proposed, jointly, by Deutsch and Deutsch (1963) and Norman (1968), and has since been called the Deutsch/Norman model. According to the Deutsch/Norman model, the filtering doesn’t take place until after the recognition stage. The Deutsch/Norman (also known as the “late selection filter”) model is shown in Figure 2.2

Though the late selection filter model certainly can account for the cocktail party effect, it has its own problems. Most importantly, it cannot account for the original findings of Cherry. If we attend to all messages to the level of recognition, we should then be able to tell when the content or the language of the spoken language changes in the unattended channel. Some modifications to the model have made it more flexible–Norman suggests that “recognition” is not always conscious recognition, and has some data to support this.8 Still, the idea that we analyze all messages to the point of recognition seems unlikely.

Treisman’s Attenuation Model

Anne Treisman (Treisman, 1960, 1964)has proposed that the idea of a “filter” in both these model is misguided. Instead, she proposes that attention is more like an attenuator, or “volume control.” In unattended channels, rather than filtering the information our entirely, we instead “turn the volume down.” Usually, that means that the message is reduced to a level where recognition will no longer take place. Some messages, though, come in with a very high initial level to begin with, and even when the “volume” of these messages is attenuated, they are still “loud” enough that they will be recognized. What kinds of things have high initial values? The kinds of things that result in cocktail party-like effects: our names, our hometowns, and so on.

Figure 2.2. The Deutsch/Norman Late Selection Filter Model of Attention.

Capacity Models of Attention: Attention as a Resource

Treisman’s model has proven to be quite influential. Deep down, though, the model has many things in common with filter models. In fact, her attenuator can be seen as a kind of a “leaky filter,” filtering some things out, but not others. In response to these criticisms, recent theories of attention, most prominently one by Kahneman (1973) have abandoned all notions of filters or attenuators. Instead, these models, called capacity models, view attention as a precious resource that can be used up. In other words, our attention store has a limited capacity. Any task that consumes attentional resources will interfere with additional tasks. Have you ever been taking notes in a class when all of the sudden you notice your phone screen light up with a text message? This momentary distraction probably caused you to lose your place in your notes, even if you didn’t pick up your phone to read the message! Shifting your attention to your phone, even for a brief moment, limits your ability to attend to the lecture that you are sitting in. This is an example of attentional blink. Attentional blink is a period of time after attending to a stimulus in which one can not fully process other stimuli. In our experiment for this week, you will examine the limits of attention.

Central to the notion of the capacity models is the distinction between automatic processing and controlled processing. Automatic processing does not consume valuable attention resources. Controlled processing does. Posner and Snyder (1975) contend that automatic processing is characterized by three things:

1) Automatic tasks occur without any intention on the part of the person. Word recognition, for skilled readers, does not require that the reader “intend” to recognize the word. Simply seeing the letters is enough to initiate the process.

2) Automatic tasks do not consume attentional resources. One of the classic questions in this area has been: How many things can a person do simultaneously? The answer, according to capacity theory, is “many,” as long as they require only automatic processing. You may be able to walk, chew gum, talk to a friend, fiddle with your backpack, and also avoid running into things, all at the same time. This is only true because all of those tasks are largely automatic.

3) Automatic processing is not something of which a person is normally aware. They just “happen.” Sometimes, we can become aware of these processes, but often, such awareness interferes with the performance of the automatic task. If you drive a car with a stick-shift transmission, try to count how many times you shift when you drive home. You might well find that becoming aware of this usually-automatic task impairs your performance.

Schneider and Shiffrin (Schneider & Shiffrin, 1977; Shiffrin & Schneider, 1977, 1984) had subjects practice a task many hundreds of times, and watched the conditions under which these skills became automatic. Two things are necessary for a task to become automatic. First, the tasks must be consistent from one trial to the next. If you have a word processor you routinely use, your use of the commands now are probably automatic. You know that <SHIFT->-B means “bold,” for example. You execute that command without even thinking about it. But if you use multiple word processing packages, where the commands vary between the different processors, you might find it difficult to automatically perform them. When you want to “bold” something, you have to stop and think about how to do it, because the command is not consistent from one package to the next.9

The second thing that is required for tasks to become automatic is practice, practice, practice. Some skills, such as reading, require many hundreds, maybe thousands of hours of practice to become fluent (Weaver, 1994). Motor skills, such as playing an instrument or learning a new sport, can also require hundreds of hours. Automaticity demands practice.

This lab will examine the limited resources of attention with rapidly changing stimuli. You will be shown many letters in rapid succession, which each letter overwriting the previous letter. You will have to determine whether or not a certain target letters were present in the sequence. Each sequence will differ in the spatial location of the targets in relation to each other. The COGLAB website has specific details concerning this experiment.

At this time, complete the experiment Attentional Blink in CogLab. Instructions can be found in Lab 8 of your COGLAB Lab Page.

Questions for Lab 2

1. What are the independent and dependent variables in this experiment?

2. What factors were controlled in this experiment? Why?

3. Graph both your data and the class data. Does it appear that temporal location of the target affects your identification? Explain.

4. What does attentional blink tell us about attention? What are some occupations in which a workers’ performance could be adversely affected by attentional blink? What types of problems/mistakes might occur?

5. Give three examples (not given in class or in the written material) of tasks that are now automatic for you.

6. What kinds of things preclude tasks from becoming automatic? Give examples.

7. Recently, Bahrick (1991)has found that long-term, automatic performance of algebra skills was best predicted by whether or not a person had completed a calculus course. Does that surprise you? Why or why not? How would could you use automaticity to explain this finding?

Data Sheet for Lab 2

NAME:

Report Average Percent Recorded:

Graphs for Lab 2

Attentional Blink

Individual Data

NAME:

Attentional Blink

Group Data

NAME: