The results of this study demonstrated that some components of artificial grammar learning were vulnerable to the pathology of schizophrenia. Patients' artificial grammar learning performances were significantly correlated with the severity of their clinical symptom dimension "Disorganization" that reflects poor abstract thinking, disorganization, and stereotyped thinking. Impaired abstract thinking in patients with schizophrenia may weaken their ability to extract the abstract rules that regulate lexical and color patterns. Patients' disorganized thoughts may reduce their sensitivity to the frequency information in the stimulus sequences. The observation favors the abstractionist account in terms of pathogenesis of the impaired artificial grammar learning. The results of our correlation study also suggested that these two underlying processes, the poor abstract thinking and disorganized thoughts, were not mutually exclusive [20
]. The correlation between lexical strings and color patterns learning performances deserves attention. As mentioned earlier that learning transfer between color patterns and lexical strings in healthy adults has been reported [28
]. The color patterns used in our experiments are identical to those used in that study. While a transfer study was not the goal of our study, we demonstrated that a possible common mechanism was present between these two types of stimuli. Further study is necessary to identify what this common mechanism is and whether it is responsible for learning transfer.
To what extent schizophrenia does affect perceptual priming component in artificial grammar learning? The less abnormal lexical strings learning and markedly impaired color patterns learning in patients with schizophrenia suggests perceptual priming contributed significantly to the overall performance. Lexical priming is thought to be intact in schizophrenia [29
]. This may explain Danion et al's finding of intact lexical strings implicit learning in schizophrenia even though they took great care to avoid learning from superficial similarities in the stimuli [32
]. On the other hand, we obtained but a borderline impairment of implicit memory function with lexical string learning if the performance of explicit verbal memory function was not taken into consideration. The difference would be thus pronounced after removing the confounding. This would imply that the impairment of implicit lexical string learning could be partially compensated by the explicit memory function.
The pronounced difference in the color pattern learning between patients and comparison subjects was in agreement with our previous studies [33
]. We have proposed that the impairment in color pattern learning resulted from information overloading. Support for this argument can be found in the event related potential (ERP) P50 sensory gating deficit in schizophrenic patients [34
]. We observed impaired color patterns artificial grammar learning was significantly correlated with impaired auditory P50 sensory gating while lexical strings was not. This finding suggested that faulty early pre-attention processing interfered with color pattern implicit learning. The perceptual category learning, i.e., the exemplar account, of the artificial grammar learning could thus be impaired.
Danion et al. reported intact implicit learning in patients with schizophrenia and they attributed their learning to the progressive abstraction of abstract rules [32
]. They accepted that depending on the specific constraints of the stimuli, different kinds of implicit knowledge might be acquired. They constructed the test material in a way that grammatical and non-grammatical strings were similar in chunk strengths, that is, the frequency with which the adjacent elements appear in the strings. This special arrangement allowed them to rule out learning by chunk-strengths. However, their claim of intact abstract grammatical rules learning in patients with schizophrenia was not supported by the evidence from our clinical symptoms correlation study. We observed moderate to severely reduced strengths in artificial grammar learning depending on the stimulus type and found strong correlations between learning performances and abstract thinking and/or disorganized thought symptoms. We suggest the artificial grammar learning effect observed by Danion et al. may be supported by lexical priming generally agreed to be intact in schizophrenia, however, as already mentioned, confounded by the explicit memory. To what extent abstract rules abstraction is preserved in patients with schizophrenia deserves further investigation. Given the demonstrated relationship between schizophrenia symptoms and the various sub-component of artificial grammar learning we suggest that in conjunction with other neurophysiological assays such as ERP, artificial grammar learning could be regarded as an index to the neural processing.
In summary, the impaired abstract thinking may be attributed to a frontal dysfunction while the perceptual learning deficits may be related to a pre-attention sensory gating deficit which is either a temporal or a fronto-temporal mechanism. The current data does not lead us to a conclusive source of dysfunction which also might imply either a multiple mechanism of the implicit learning or it takes the cooperation of several systems or locations of the brain to complete such function. This is the major limitation of our study that we do not have comprehensive data of the cognitive function test from the comparison subjects to contrast with the patients with schizophrenia. Nor do we have data of functional images to supplement the lesion localization.