Language - Computational Explorations in Cognitive Neuroscience - page 331

No. Layer(s) lesioned Dyslexia Type

table 10.4. For each type of lesion, we remove units
in increments of 10 percent from the layer in question,
observing the effects on reading performance.
The first two lesion types damage the semantic path-
way hidden layers ( OS_Hid and SP_Hid ), to simulate
the effects of surface dyslexia. The next type damages
the direct pathway ( OP_Hid ), to simulate the effects of
phonological dyslexia, and at high levels, deep dyslexia.
The next two lesion types damage the semantic pathway
hidden layers again ( OS_Hid and SP_Hid ) but with
a simultaneous complete lesion of the direct pathway,
which corresponds to the model of deep dyslexia ex-
plored by Plaut and Shallice (1993). Finally, the last
lesion type damages the direct pathway hidden layer
again ( OP_Hid ) but with a simultaneous complete le-
sion of the semantic pathway, which should produce
something like an extreme form of surface dyslexia.
This last condition is included more for completeness
than for any particular neuropsychological motivation.

0 Surface

OS Hid

1 Surface

SP Hid

2 Phono

OP Hid

3

OS Hid + Comp Dir

Deep

4

SP Hid + Comp Dir

Deep

5

OP Hid + Comp Sem

Surface

Table 10.4: Types of lesions performed automatically by the
simulator. The lesion number (No.) corresponds to the batch
counter variable in the simulator. Comp Dir is a complete le-
sion of the direct pathway (in addition to partial lesions of in-
dicated layer), and Comp Sem is similarly an additional com-
plete semantic pathway lesion. The dyslexia type indicates
which type of dyslexia the lesion simulates.

The text log also has columns for blend responses
(when the distance to the closest valid output pattern is
greater than 1), and a catch-all other column when none
of these criteria apply (due to space constraints, neither
of these are shown in the figures, 10.8 and 10.9).
To summarize the results so far, we have seen that
a lesion to the semantic pathway results in purely vi-
sual errors, while a lesion to the direct pathway results
in a combination of visual and semantic errors. To a
first order of approximation, this pattern is observed in
surface and deep dyslexia, respectively. As simulated in
the PMSP model, people with surface dyslexia are actu-
ally more likely to make errors on low-frequency irreg-
ular words, but we cannot examine this aspect of perfor-
mance because frequency and regularity are not manip-
ulated in our simple corpus of words. Thus, the critical
difference for our model is that surface dyslexia does
not involve semantic errors, while the deep dyslexia
does. Visual errors are made in both cases.

To set up the simulation for the lesioning tests,
iconify the current text logs, set lesion_path back
to NO_LESION , and do View , LESION_LOGS from the
overall control panel.
You will see two graph logs appear. The graph
on the left ( RepiBatch_3_GraphLog ) displays the
automatically-scored error types as a function of level
of damage of a given type. The graph on the right
( LesAmtBatch_3_GraphLog ) displays the results
from the last lesion level (90%) for each different type
of lesion. The lesion types are numbered across the X
axis as in table 10.4.

Do LesionTest to run the lesion test.
The network display will flash and update several
times at the beginning of each lesion (even if the
Display is turned off), with only the final of these
flashes corresponding to the actual lesion (the others oc-
cur as the network is rebuilt, reconnected, and reloaded
between each lesion). Then, all the items will be “read”
by the network, and scored (you probably want to keep
Display off for this). You will be able to see the le-
sioned units in the network, but don't be confused by the
fact that several of the layers in the intact network (fig-
ure 10.6) have “missing” units compared to the square
layer box (this was done to better organize the units in
meaningful groups within the layers).

, !

Reading with Partial Pathway Lesions

We next explore the effects of more realistic types of le-
sions that involve partial, random damage to the units in
the various pathways, where we systematically vary the
percentage of units damaged. The simulation is config-
ured to automatically step through a series of six differ-
ent lesion types, corresponding to damaging different
layers in the semantic and direct pathways as shown in