The Best of Creative Computing Volume 1 (published 1976)
Computer-Based Experiments in Cognitive Psychology (predictions, hierarchical levels, parallel processing, using the program)
The dependent variable is the average time it
takes to scan one line. This is calculated by
dividing reaction time by the line number at which
the target appears.
Predictions
Hierarchical Levels
For a task in which subjects search for a single
letter, Neisser (1964) suggests a Pandemonium
something like that shown in Figure 1. Only one
cognitive demon is present, the cognitive demons
for all letters but the target letter being effectively
turned off. This implies that the processing of a
letter needs to go only as far as the feature
extraction level until the target letter is found. lf,
for example, the search is for a Q, processing of
each letter terminates as soon as features not
associated with a Q are extracted. The recognition
process is completed (i.e., the Q cognitive demon
and the decision demon are activated) only when
features associated with a Q are extracted. When
searching for the absence of a letter (e.g., not-Q),
the structure of the Pandemonium is similar to that
used in searching for the presence of a letter in that
the cognitive demons for all letters but the critical
letter (e.g., Q) are turned off. As before, this
means that the recognition process is completed
only for the critical letter. However, since the
critical letter is present in all lines but one, the
recognition process must be completed at least
once in each line. Since completion of the
recognition process (activation of the cognitive and
decision demons) requires more time than
processing to only the feature extraction level, it should
take longer to search for the absence than for the
presence of a letter. One prediction derived from
the idea of hierarchical levels of processing, then, is
that scanning time per line should be greater when
the search is for not-Q than for Q, and for not-Z
than for Z.
Referring again to Figure 1, it is seen that
there are three levels of feature extraction. Level 1
extracts the crudest features, the simple presence
or absence of lines, angles and curves. At level 2,
the feature extraction is more detailed, producing
information regarding the presence or absence of
categories of lines (vertical, horizontal and
oblique), angles (right and acute), and curves (continuous
and discontinuous). Level 3 produces the
most detailed information, going beyond the mere
presence or absence of the feature categories of
level 2 to indicate the number of features in each
category. lt is assumed that these three levels of
feature extraction form a hierarchy, level 1
extraction occurring before level 2, and level 2 extraction
occurring before level 3. lt is further assumed that
the cognitive demon evaluates features in the order
of their extraction, level 1 features being evaluated
first, then level 2 features, and finally level 3
features. The cognitive demon evaluates the
features extracted by level 2 and level 3 only if the
features extracted by the preceding level, levels 1
and 2, respectively, indicate that the letter being
processed could be the target letter. Since more
processing time will be required when more feature
extraction levels must be evaluated, and since more
feature extraction levels must be evaluated with
greater similarity of target-letter and context-letter
features, a second prediction derived from the idea
of hierarchical levels of processing is that scanning
time per line should be greater for a Q-search than
for a Z-search in the round context and greater for
a Z-search in the angular context.
Parallel Processing
Searching for two target letters, e.g., searching
for either a Q or a Z, requires the addition of a
cognitive demon to the Pandemonium shown in
Figure 1. In general, there will be one cognitive
demon for each target letter. Since demons at a
particular level of the Pandemonium operate in
parallel, however, the addition of a cognitive
demon will have no effect on processing time.
Accordingly, there should be no difference in
scanning time per line in searching for a Q, a Z, or
either a Q or a Z. To be more specific, it should
take no longer to search for a Q when a Z is also a
possible target than to Search for a Q as a single
target, and vice versa.
USING THE PROGRAM
The program has five running options: (1) to
run a subject and the simulation or run the
simulation alone; (2) a choice of four targets (Q, Z,
Q or Z, or any letter of the alphabet); (3) a goal of
detecting the presence of the target in a line or a
goal of detecting its absence; (4) a choice of three
contexts (round -C, D, G, O, R and U;
angular - X, I, M, V, W and X; or any six letters of the
alphabet); and (5) to have the Pandemonium
search procedure printed or omitted. lt is suggested
that a student should run himself as a subject in
conjunction with the simulation before running his
own experiments on other students and the simulation.
The choice of target, goal (presence or
absence of the target), and context will be
determined by the experimental condition under which
the program is to be run. For the first run, this
should probably be determined for each student by
the instructor in order to ensure an approximately
equal number of subjects in each experimental
condition.
The printout of the Pandemonium search
procedure lists the feature extraction levels used in
processing each target and context letter. Examining
the procedure for each of the experimental
conditions should help the student to understand
the structure of the Pandemonium and how the
experimental effects predicted above are produced
by this structure. The search procedure should be
printed whenever the student runs a new
experimental condition.
