Abstract
Previous research on artificial grammar has indicated that the human ability to classify sentences or letter strings according to grammaticality relies on two types of knowledge. One is a superficial, familiarity-based understanding of a grammar the other is the knowledge of rules and critical features underlying a grammar. The fundamentally different characteristics of these systems permit an analysis of receiver-operating characteristics (ROC), which measures the extent to which each type of knowledge is used in grammaticality judgments. Furthermore, violations of a grammar can be divided into hierarchical and local violations. The present study is the first to combine the use of ROC analyses, fMRI and a grammaticality dichotomy. Based on previous neuroimaging studies, it was hypothesized that judgments based on rule knowledge, as extracted from individual ROC analyses, involve the left inferior frontal gyrus (IFG), whereas similarity would involve right IFG, as well as left hippocampal regions. With regards to violation types, it was hypothesized that hierarchical violations would recruit the opercular part of the left IFG as well as the posterior operculum, whereas local violations would bilaterally activate the premotor cortex (PMC). Results indicated that for greater reliance on rule knowledge, a ventral part of the left PMC was activated for ungrammatical items, whereas other PMC areas show a differentiated response for grammaticality for individuals less reliant on similarity. The right IFG was related to ungrammatical items as a function of similarity. Results are discussed with regards to possible error detection systems and differentiated efficiencies for respective classification strategies.