It has been proposed that procedural learning is mediated by the striatum and, it has been reported that patients with Parkinson's disease (PD) are impaired on the weather prediction task (WPT) which involves probabilistic classification learning with corrective feedback (FB). However, PD patients were not impaired on probabilistic classification learning when it was performed without corrective feedback, in a paired associate (PA) manner; suggesting that the striatum is involved in learning with feedback rather than procedural learning per se. In Experiment 1 we studied FB- and PA-based learning in PD patients and controls and, as an improvement on previous methods, used a more powerful repeated measures design and more equivalent test phases during FB and PA conditions (including altering the FB condition to remove time limits on responding). All participants (16 PD patients, H&Y I–III and 14 matched-controls) completed the WPT under both FB and PA conditions. In contrast to previous results, in Experiment 1 we did not find a selective impairment in the PD group on the FB version of the WPT relative to controls. In Experiment 2 we used a between groups design and studied learning with corrective FB in 11 PD patients (H&Y I.5–IV) and 13 matched controls on a more standard version of the WPT similar to that used in previous studies. With such a between groups design for comparison of FB and PA learning on the WPT in PD, we observed impaired learning in PD patients relative to controls across both the FB and PA versions of the WPT. Most importantly, in Experiment 2 we also failed to find a selective impairment on the FB version of the WPT coupled with normal learning on the PA version in PD patients relative to controls. Our results do not support the proposal that the striatum plays a specific role in probabilistic classification learning with feedback.
It has been suggested that the striatum and its cortical projections support procedural (sometimes called implicit or unconscious) learning; whereas the cortico-limbic-diencephalic structures are considered the substrate for declarative (sometimes referred to as explicit or conscious) learning (Butters, Wolfe, Martone, Granholm, & Cermak, 1985; Cohen & Squire, 1980). The question of whether procedural learning can be genuinely unconscious in humans is controversial (e.g. Wilkinson & Shanks, 2004).
One task that has been employed to study procedural learning in humans is the ‘weather prediction task’ (WPT) initially devised by Knowlton, Squire, and Gluck (1994) and subsequently employed by several other investigators (e.g. Aron, Gluck, & Poldrack, 2006; Gluck, Shohamy, & Myers, 2002; Hopkins, Myers, Shohamy, Grossman, & Gluck, 2004; Knowlton, Mangels, & Squire, 1996; Knowlton, Squire, et al., 1996; Lagnado, Newell, Kahan, & Shanks, 2006; Poldrack, Prabhakaran, Seger, & Gabrieli, 1999; Poldrack et al., 2001 and Sage et al., 2003). The WPT is a non-motor probabilistic classification task involving incremental learning over many trials which is considered to occur without any explicit knowledge. On each trial participants are presented with a particular arrangement of 1, 2, or 3 of 4 possible tarot cards each of which have different patterns on them (e.g. squares, diamonds, circles or triangles). Participants are required to use the cards presented on each trial to predict a binary outcome, i.e. whether the weather will be rainy or fine. Each of the four cards is independently associated with the two possible outcomes with a fixed probability and overall, each outcome occurs equally often. For example, the squares, diamonds, circles and triangles respectively predict the outcome ‘fine’ with a fixed probability of 0.2, 0.4, 0.6, and 0.8. Typically participants perform around 200 trials of the WPT with corrective feedback on each trial. The feedback consists of a ‘thumbs up’ or ‘thumbs down’ message following correct and incorrect responses respectively.
By learning the independent cue-outcome associations across trials, participants can improve their predictive accuracy on the WPT. For instance in one study, healthy controls were able to achieve 74.5% predictive accuracy across all trials (well above chance) despite the fact that they performed poorly on subsequent tests of explicit knowledge (Gluck et al., 2002). In particular, in post-experiment questionnaires the participants gave inaccurate estimates of cue-outcome probabilities and there was little correspondence between how they reported that they learned the task and their actual task performance. Such findings have led to the conclusion that participants have little or no awareness of the cue-outcome contingencies that they acquire during the WPT (Gluck et al., 2002) and this apparent dissociation between learning and awareness has been taken as evidence for the existence of two separate procedural and declarative learning systems.
