Learnable artificial grammar rules are only learned explicitly

In most rule-learning experiments subjects (Ss) are trained with both positive and negative instances of the rule. However, in most traditional artificial grammar learning (AGL) experiments Ss are trained with positive instances only and using very complex rules. In a typical training phase Ss are unaware of the underlying rules governing the stimuli, and are instead instructed to do an irrelevant task. After training they are told about the rules, and then have to differentiate between positive and negative stimuli. Ss’ typical performance is significantly better than chance, although Ss are unable to verbalise the rules and think they are guessing. This dissociation of performance and verbalisation led Reber (e.g. 1967) to conclude that Ss care acting on “implicit” (i.e. unconscious), abstract knowledge. However, it has also been argued that Ss are not learning the abstract rules but are basing their classification on memorised fragments of the stimuli. In two experiments in this thesis, it was shown that Ss seem to be memorising fragments of the stimuli, rather than learning the underlying rules of the stimuli. It was further shown that presenting Ss with positive and negative evidence in the training phase was detrimental to subsequent test performance. If Ss were simply memorising fragments of stimuli, negative evidence would simply add to the memory load, and performance would thus decrease. In the main experimental series in this thesis, the critical antecedent step of testing the learnability of rules was taken, and an easy, medium and hard set of rules were constructed. It was seen that only very simple rules could be mastered to 100% during an experimental session. In several web-based experiments, groups of Ss were trained (1) with and without corrective feedback, (2) with an active response or mere passive exposure to the stimuli, (3) with forewarning about the existence of rules, forearmed with the actual rules, or with no prior knowledge of the existence of rules, and (4) with positive and negative stimuli, positive stimuli only, or negative stimuli only. It was found that Ss with high performance rates could also verbalise the rules, i.e. learning was explicit, and that passively exposed to stimuli resulted in better performance than actively responding to the stimuli. Established AGL effects may merely be artefacts of the fact that traditional artificial grammars were too complex and unlearnable within the scope of an experimental session, leaving the memorisation of fragments as the only basis for any improvement.