In this paper, we describe an integrated model for neural control of both routine and nonroutine action selection. Functionally, the model is based upon an architecture originally proposed by Norman & Shallice (Norman and Shallice, 1986; Shallice 1988). This model includes a Contention Scheduler (CS) and a Supervisory Attentional System (SAS). The CS mechanism is based upon the CS architecture described by Presott, Redgrave and Gurney (1999). Little is known of the neural architecture of the prefrontal cortex (PFC) that might realise the SAS. Accordingly, we have developed a partial implementation in which the architecture is guided by an analysis of the functionality required by Shallice's SAS. The resulting model is used to control the behaviour of a simulated robot. This paper extends earlier work on normal executive control (Garforth, McHale, Meehan, 2003), describing lesion studies to explore the correspondence between the behaviour of the robot and behaviours seen in humans suffering lesions of the PFC. In doing so, we are able to assess the legitimacy of the SAS model as hypothesised and as implemented.
展开▼
