Exploration of implicit artificial grammar learning in Parkinson's disease.

Parkinson's disease (PD) patients and matched control subjects were compared in an artificial grammar learning task, one of the main paradigms of implicit learning. The evaluation material was constructed in such a way that grammaticality judgements (classification task) made on the test strings could not be based on some superficial features of the learning strings: the grammatical and nongrammatical test strings did not differ according to different measures of chunk strength (based on the frequency with which their bigram and trigram components appear in the learning strings). Unknown to participants, two successive presentations of the set of test strings were allowed during the classification task. Results show that PD patients and controls performed at the same level during the first presentation of the test strings series, which suggests that the striatum is not (crucially) implicated in the ability to learn implicitly the complex conditional associations between elements present in a set of examples generated by a finite-state grammar. However, and contrary to control subjects, the classification performance of PD patients was at chance during the second presentation of the test strings. We argue that this latter result could be the consequence of the attentional deficit of PD patients.