Frontal lesion effects on verbal fluency in a network model.

    Previous work on neural network modeling of frontallesion effects in neuropsychological tests is extended to the verbal fluency test.  In this test, the subject is asked to say all the words he or she can think of beginning with a specified letter in a minute, and the test runs over three minutes with three different initial letters.  Damage to the left frontal cortex interferes with this task in three ways: reducing the attention the subject pays to the task as a whole; increasing the likelihood that the subject will violate the rules, for example by repeating words or using proper names; and increasing the number of perseverative errors, which consist in this case of the subject saying words that started with a previously used but currently incorrect letter.  A simple neural network, whereby frontal damage is mimicked by lowered gain of reinforcement signals, is constructed that reproduces all these effects.