The main novel aspect of this work is that it is the only AGL study that incorporates a knowledge factor, constructed toreflect people’s general knowledge and expectations and dissociated from purely structural aspects of the stimuli. More specifically,we investigated how (in)congruence of prior knowledge with structure affects the development of knowledge concerningboth the relevant general knowledge (the distance factor) and purely structural information (grammaticality andsimilarity), and we also examined the implicitness and/or explicitness of each of the three knowledge types. Congruenceof prior knowledge with structure led to an advantage only in terms of grammaticality accuracy and only under baselinetraining, thereby replicating Pothos’s (2005)finding. However, a surprising finding is that we identified training conditionsand knowledge types, such that inconsistency between expectations and stimulus structure led to a clear advantage. Finally,we have identified conditions where expectations-based knowledge may involve both explicit and implicit knowledge andmay co-exist with both explicit and implicit knowledge of purely structural aspects. The complex pattern of results we identifiedindicates that, plausibly, the same kind of knowledge can be extracted via multiple routes. Clarifying the relevantissues (e.g., the interplay between alternative learning routes; Ashby, Alfonso-Reese, Turken, & Waldron, 1998) in AGL isan under-researched, though important direction for future research (cf. Chang & Knowlton, 2004; Pothos& Wood, 2009).