Numerous experimental studies indicate that selective attention is impaired in schizophrenia. The literature stresses an abnormal vulnerability to distractors and difficulties in filtering out non-relevant information. Interference tasks test the ability to process a target selectively when distractors are present in the display. Impairment in filtering out non-relevant information should therefore be reflected in increased interference. Evidence for a larger magnitude of interference in patients with schizophrenia has been reported in a number of studies involving the Stroop color/word interference task.

Wapner and Krus (1960) tested 24 patients with schizophrenia and 24 controls in a Stroop task. Stimuli were displayed on cards containing 100 items. Performance was compared under three conditions: (1) a color naming task of colored rectangles, (2) a reading task of color names written in black ink and (3) a color naming task of color names written in a conflicting color (e.g., RED written in blue ink). Performance was measured in terms of the overall naming or reading time of the 100 items. Patients with schizophrenia were slower than controls under the three conditions but a disproportionate slowing was found under the conflicting condition. Wysocki and Sweet (1984) used the same stimulus material but measured performance in terms of the number of items read (or named) in 45 s. The number of items processed was lower for patients, particularly in the conflict condition. Abramczyk et al. (1983) examined performance in two conflict conditions. They tested the classical interference of words on color naming and also the effect of color on word reading (the reversed Stroop effect). They found disproportionate slowing for patients in both conflict conditions compared to control conditions (colored rectangles and black color names).

This convergence in the demonstration of increased interference in schizophrenia led Cohen and Servan-Schreiber (1992) to propose a connectionist model to simulate performance of patients with schizophrenia in the Stroop interference task. The basic idea was that the deficit in selective attention frequently reported in schizophrenia results from an impairment in the “ internal representation of context ”. This account includes different types of representations used to mediate behavioral responses such as task instructions and the processing of prior stimuli. The authors developed a connectionist model to simulate the two empirical effects reported in the literature, that is (1) the overall slowing in response time and (2) the disproportionate slowing in the conflict condition for patients with schizophrenia. The model consists of two independent pathways, one for color naming and one for word reading. Each pathway is composed of a set of input units, a set of intermediate units and a set of output units. Processing within the net is feed forward. Activation propagates gradually from the input to the output units. A response occurs when sufficient information from one of the output units exceeds a response threshold. Attentional modulation is simulated by altering the responsiveness of the processing units in one of the pathways. In the network attention is an additional input that provides contextual support for the processing of information within a selected pathway. The deficit in maintaining an internal representation of context was simulated by reducing the gain of units in the task demand module. This manipulation resulted in both an overall increase in response time and a disproportionate slowing in the conflict condition. In contrast, a manipulation of the “cascade rate”, that is a decrease in the rate at which information accumulates for units in the network, resulted in an overall increase in response time without disproportionate slowing in the conflict condition suggesting that the slower performance of patients with schizophrenia cannot account for the increase in the magnitude of interference in the conflict condition. In the model the reduction of the gain of units is related to the lower level of dopamine in the prefrontal cortex of patients with schizophrenia.

However, the results of more recent studies question the previous empirical and simulation data. Carter et al. (1993) used a computer trial-by-trial version of the Stroop task and they failed to replicate the disproportionate slowing found in the conflict condition for patients with schizophrenia in the card version of the Stroop task. Carter et al. (1993) reported an increased facilitation in the congruent condition (e.g., RED written in red ink) for undifferentiated patients with schizophrenia and increased interference in the conflict condition (e.g., RED written in green ink) for a paranoid subtype of schizophrenia suggesting that facilitation and interference reflect different cognitive processes. Carter et al.’s (1993) results were not replicated by Phillips et al. (1996) who found no effect of illness stage and subtype on the extent of facilitation. Like Carter et al., 1993 and Phillips et al., 1996 found an overall increase in RTs for schizophrenic subjects but the magnitude of interference in the conflict condition was in fact slightly larger for controls than for schizophrenic subjects in their study (82 vs 64 ms averaged over left and right visual fields). The lack of disproportionate slowing in the conflict condition was also reported by Carter et al. (1995) and Elvevaag et al. (1995).