| |
Abstract:
Abstract: Similarity judgment is assumed to play a central
role in a variety of cognitive processes (e.g., object recognition,
categorization, analogy) yet mechanisms underlying similarity
judgment are poorly understood. We have employed a
neurally-inspired computational architecture previously used in
explaining other cognitive tasks (serial recall, the Stroop task,
spatial delayed response) to simulate similarity judgment. This
model assumes massive, bidirectional connectivity,
continuous-valued units, and Hebbian learning. Together, these
assumptions give rise to networks that settle into stored patterns
of distributed activity (attractors). The model explains observed
directional asymmetries in similarity judgments (e.g., people
judging North Korea as more similar to China than vice versa) in
terms of differences in attractor strengths (it is easier to move
from weak attractors (e.g., North Korea) to strong attractors
(e.g., China) than vice versa) and makes a number of novel
predictions. We present an experiment designed to test the first
and most straightforward of these predictions, namely, that
manipulating the frequency with which stimuli are presented would
lead to changes in similarity asymmetries by influencing the
strength of the underlying attractors. Using hues of blue and green
as stimuli, we obtained baseline measures of similarity asymmetry.
Presentation frequency of hues was manipulated during a training
phase. Post-training measures of similarity asymmetry between high
and low frequency hues increased compared to pre-training
baselines, consistent with the predictions of the model.
|
