ACT-R?

1
Introduction

• ACT-R?
• General system for modeling a wide range of
  higher level cognitive processes.
• Developed by Anderson(1993)
• We describe theory of visual attention and how it
  relates to the ACT-R theory of higher level
  cognition.
• Then we describe its extension to a
  menu-selection task and its ability to make some
  novel predictions about the task.
• Assumptions
• Two types of knowledge
• Declarative Procedural

2
A theory of the visual interface

• Theories of higher level cognition typically
  ignore lower level processes.
• Two stresses for the plausibility of the
  resulting models
• Basic overview of the ACT-R system

3
Visual attention

• ACT-R can select the scale of the features for
  which it searches and the size of the objects it
  is recognizing.
• The best way to understand how this theory works
  is to see it applied to various tasks involving
  visual attention.

Figure 2 ACT-R can see either the H or the Xs
comprising the H, Depending on how ACT-R sets its
feature scale.
4
Visual attention-Sperling Task(1/2)

• Whole-report condition
• Sperling presented subjects with brief
  presentations (50msec) of visual arrays of
  letters.
• Partial-report condition
• Subjects could report 3.3 letters in the selected
  row.
• Delaying the auditory cue lower the subjects
  recall
• See figure 4 in the next slide

He found that subjects could report back 4.4
letters.
5
Visual attention-Sperling Task(2/2)

• Two dimensions of significance
• Limitations of visual sensory memory
• How fast visual attention can move over an array.

6
Visual attention-Subitizing Task

• Another way to measure switching time for
  attention is to see how long it takes o attend to
  several objects on a screen. (e.g)How many
  objects are there?)
• There is an increase in latency with number of
  digits to be identified.

7
Visual attention-Visual search Task

• Another way to investigate the time to shift
  attention is to display an array and ask subjects
  to search among objects for a specific objects

Results of Visual search task
8
Visual attention-Conclusions

• We have shown that the ACT-R model is consistent
  with some of the classic results from visual
  attention

9
Application to menu-selection data

• Apply ACT-R to the Nilsen data
• These data are concerned with time to scan a menu
  as a function of the target position of the item
  in the menu.
• The menu consists of a set of digits(1 to 9)
  randomly ordered vertically
• A linear function is obtained with a slope of
  103msec per position.

10
Conclusions

• We properly model the basic processes of visual
  attention and that they matter in a traditional
  HCI task such as menu scanning.
• ACT-R provides an architecture in which to work
  out these complex interactions with visual
  attention for both simple and complex tasks.