Perception and Attention - Computational Explorations in Cognitive Neuroscience - page 260

However, when we apply this normalizing procedure
to the patient's data, the results do not fit well. Thus, we
again divide 350 ms by the the 640 ms valid RT's, giv-
ing a ratio of .55. Then, we do 760 ￿ :55 = 418 ,which
is substantially slower than the 390 ms invalid times for
the adult normals (table 8.3). This makes it clear that
the patients are specifically slower in the invalid trials
even when their overall slowing has been taken into ac-
count by the normalizing procedure. This differential
slowing is what led to the hypothesis that these patients
have difficulty disengaging attention.
Now, we lesion the model, and see if it simulates the
patient's data. However, because the model will not suf-
fer the generalized effects of brain damage (which are
probably caused by swelling and other such factors),
and because it will not “age,” we expect it to behave just
like a adult subject that has only the specific effect of the
lesion. Thus, we compare the model's performance to
the normalized patient values. Although we can add in
the 310 ms constant to the models' settling time to get
a comparable RT measure, it is somewhat easier to just
compare the difference between the invalid and valid
cases, which subtracts away any constant factors (see
the Diff column in table 8.3).

for the opposite configuration of the cuing task, where
the cue is presented in the lesioned side of space, and
the invalid target is thus presented in the intact side. In-
terestingly, data from Posner et al. (1984) clearly show
that there is a very reduced invalid-valid reaction time
difference for this condition in the patients. Thus, it
appears that it is easier for the patients to switch atten-
tion to the intact side of space, and therefore less of an
invalid cost, relative to the normal control data. Further-
more, there appears to be less of a valid cuing effect for
the patients when the cue and target are presented on
the damaged side as compared to the intact side. Let's
see what the model has to say about this.

Set env_type to REVERSE_POSNER (and Apply ),
and do a Batch run for this case.
You should see that the network shows a reduced dif-
ference between the valid and invalid trials compared to
the intact network (an average of roughly 55 cycles for
valid, 61 for invalid, for a difference of only around 6
compared to around 41 for the intact network). Thus,
the cue has less of an effect — less facilitation on valid
trials and less interference on invalid ones. This is ex-
actly the pattern seen in the Posner et al. (1984) data. In
the model, it occurs simply because the stronger intact
side of space where the target is presented has less diffi-
culty competing with the damaged side of space where
the cue was presented. In contrast, the disengage theory
would predict that the lesioned network on the reverse
Posner task should perform like the intact network on
the standard Posner task. Under these conditions, any
required disengaging abilities should be intact (either
because the network has not been lesioned, or because
the cue is presented on the side of space that the le-
sioned network should be able to disengage from).
As mentioned previously, additional lesion data
comes from B alint's syndrome patients, who suffered
from bilateral parietal lesions. The most striking fea-
ture of these patients is that they have simultanagnosia
— the inability to recognize multiple objects presented
simultaneously (see Farah, 1990 for a review). The
more complex spatial attention model described in the
next section will provide a more useful demonstration
of this aspect of the syndrome. Interestingly, when such
subjects were tested on the Posner task (e.g., Coslett &
Saffran, 1991), they exhibited a decreased level of at-

To lesion the model, press Lesion in the
attn_ctrl control panel, and select SPAT1_2 to le-
sion both levels of spatial representations, and HALF
and HALF to lesion one half of the locations, and 1 out
of the 2 spatial units in each location. Select r.wt and
confirm that these units (the back 2 units on the right
for Spat1 , and the back right unit for Spat2 ) have their
weights zeroed out. Batch run the lesioned network.

, !

Question 8.10 (a) Report the resulting averages. (b)
Compute the invalid-valid difference, and compare it
with the patient's data, and with the intact network.
(c) Turn the network display back on, select act , and
Step through the events. Explain why the lesioned
model is slower on the invalid trials in terms of the ac-
tivation dynamics of the network.

You should have found that you can simulate the ap-
parent disengage deficit without having a specific “dis-
engager” mechanism. One additional source of support
for this model comes from the pattern of patient data