Pure perceptual-based learning of second-, third-, and fourth-order sequential probabilities

There is a growing body of evidence that the visuospatial attention system and the oculomotor system are tightly linked. For example, a covert shift of attention to a location is accompanied by the preparation of an eye movement to that location (Belopolsky & Theeuwes, 2009; Van der Stigchel & Theeuwes, 2007). Also, covert shifts of attention activate brain regions that are nearly identical to those activated by saccadic eye movements (de Haan, Morgan, & Rorden, 2008; Ikkai & Curtis, 2008). And, a covert shift of attention cannot be directed to a location that cannot be accessed physically by an eye movement (Craighero, Nascimben, & Fadiga, 2004; Smith, Ball, Ellison, & Schenk, 2010). Thus sequence learning in the perceptual SRTT might involve the oculomotor system. However, until more is known about the relationship between the visuospatial attention system and the oculomotor system, it will be difficult to establish the latter system's role in sequence learning in the perceptual SRTT.

There may also be some learning of the sequence of visuospatial locations in the traditional SRTT (Clegg, 2005; Keele, Jennings, Jones, Caulton, & Cohen, 1995; Stadler, 1989). Such learning, however, appears to have minimal influence on performance measures of sequence learning because, as noted earlier, the presence of a distractor that captures visuospatial attention impairs learning of a sequence of visuospatial locations but has no effect on performance measures of sequence learning in the traditional SRTT.

One might argue that the awareness survey is not optimally sensitive to explicit knowledge of the second-order adjacent probabilities because it does not reinstate all of the cues that were present during the perceptual SRTT (e.g., the underline moving from one target location to the next). The empirical evidence, however, suggests the survey is sensitive. A recognition test where sequences of target locations are presented as sequences of digits is (a) as sensitive as a recognition test that requires participants to respond to the sequences of target locations as in the SRTT (Willingham, Greeley, & Bardone, 1993), (b) capable of detecting explicit sequence knowledge after very limited exposure to the sequence during the SRTT, suggesting that such a recognition test can pick up relatively low levels of awareness (Perruchet, Bigand, & Benoit-Gonin, 1997, Experiment 3), and (c) able to discriminate participants who are instructed to try to explicitly learn a sequence while performing the SRTT from participants who are not given such instructions (Curran, 1997). Also, the awareness survey is able to detect explicit knowledge of subtle first- and second-order adjacent dependencies of 0.40 versus 0.60 after extended training on the SRTT (Remillard & Clark, 2001, Experiment 4).

The central value of 10.9 ms for the 90% CI is the RT difference between L and H successors (i.e., L − H) averaged across the three epochs. The central value of 7.9 ms/epoch for the 95% CI is the slope of the regression line through the L successor RTs minus the slope of the regression line through the H successor RTs.

RTs to the last element of each of the four types of runs in Table 4 were also examined. There were 96 (6 × 2⁴) possible five-element sequences of which 24 were EE runs, 24 were UE runs, 24 were EU runs, and 24 were UU runs. For each participant and epoch, the median RTs to the last element of each of the 96 five-element sequences were computed, and then the 72 RTs for EE runs (24 RTs per epoch × 3 epochs) were averaged, as were the 72 RTs for UE runs, the 72 RTs for EU runs, and the 72 RTs for UU runs. Mean RTs on EE, UE, EU, and UU runs were, respectively, 496, 500, 497, and 490 ms. The RT difference between EE and UE runs was significant, p = 0.010, as was the difference between EU and UU runs, p = 0.001. The pattern was similar in Experiments 2 and 3.

An effect of successor in the absence of a Successor × Session interaction occurs frequently in the traditional SRTT when there are six target locations and each target location has two possible successors (Remillard, 2008a, 2010; Remillard & Clark, 2001). Thus, this pattern of results is not unusual. Indeed, Rowland and Shanks (2006a) note that "Learning tends to be quite rapid in the probabilistic SRT task… so the critical statistical evidence for learning is a main effect of target probability rather than a Probability × Block interaction" (p. 291).