23. IAM overview
● Three layers of interconnected neurons (units)
– Layers:
● Word: each neuron corresponds to a word
● Letter: each neuron corresponds to a letter
● Feature: each neuron corresponds to a “feature” (edges in this case)
● Interactivity:
– Activation of the word level alters the lower levels (“top-down
information”)
● Lateral-inhibition:
– Within each layer, the neurons inhibit one-another resulting in
maximal activity for only one neuron (“winner-takes-all”)
24. Spreading Activation
● Each neuron either inhibits or
activates its neighbours
● Same with the interconnected
layers
25. From data to theory
IAM was originially devised to help account for the
so-called word-superiority effect, which is the
observation that letters are recognized faster
within words in comparison to nonwords
28. Hypotheses
● At least two hypotheses can be derrived from
the model
– Different levels of abstraction interact
● interactivity assumption
– Neurons within each layer compete
● Lateral inhibition assumption
29. From theory to data
● Lexical decision task:
– Subjects are presented with either words or non-
words on the screen, and asked to decide if each
item was a word or a non-word
● They are instructed to respond as fast as possible
● Response-Time (RT) is the primary measure of
performance
30. From theory to data
● Segui & Grainger (1990) tested the interactivity
assumption, in several lexical decision
experiments
● They used two different versions of the model
to simulate the data obtained from human
participants
– IAM: standard IAM model
– NIAM: non-interactive activation model
33. What now?
● Interactivity might not be necessary, since
lateral inhibition alone seems to account for the
data
● Suggests that the model might be too complex
than it should be (not good)
– See also Mewhort & Johns (1988)
● More scrutiny is needed!
34. references
● McClelland, J. L., & Rumelhart, D. E. (1981). An interactive activation
model of context effects in letter perception: I. An account of basic
findings. Psychological review, 88(5), 375.
● Jacobs, A. M., & Grainger, J. (1992). Testing a semistochastic variant
of the interactive activation model in different word recognition
experiments. Journal of Experimental Psychology: Human Perception
and Performance, 18(4), 1174.
● Mewhort, D. J. K., & Johns, E. E. (1988). Some tests of the
interactive-activation model for word identification. Psychological
Research, 50(3), 135-147. Chicago
35. “...the logical 'and' function of a computer can be
realized with switches operated by relays, with
vacuum tubes, or with transistors. A computer-
logic designer does not have to know the physics
of transistors to design with components based on
transistors. So, something in addition to structural
analysis is needed.”
From Sparse Distributed Memory,
by Pentti Kanerva
great book!
Kevin Shabahang: k.shabahang@gmail.com
BSc Psychology (honours), Queen's Universitry
You can usually find me in the Human
Information Processing Lab at 315
Humphrey Hall...
Dr. D. J. K. Mewhort