In the fast lane toward structure in implicit learning: Nonanalytic processing and fluency in artificial grammar learning

Abstract

This paper reports three experiments that investigated the relation between perceptual fluency and nonanalytic processing in artificial grammar classification. Nonanalytic processing was manipulated by invoking response time restrictions. In Experiment 1 there were no response time restrictions for classification decisions (response time: M=5950 ms), Experiment 2 invoked a response deadline of 2 s (response time: M=1310 ms), and Experiment 3 used a within-subjects response-signal procedure comparing classification after a 500 ms delay (response time: M=838 ms) with a 2000 ms delay (response time: M=2280 ms) The results showed that fluency created by masked priming affects classification in artificial grammar learning under conditions of relatively nonanalytic processing (Experiment 2 and the short condition of Experiment 3). The results are discussed in relation to different heuristics that participants can use in artificial grammar classification, the fluency heuristic being one of them. The experiments also provided some evidence that metaknowledge of grammaticality in artificial grammar learning may under some circumstances be enhanced by nonanalytic processing during testing.

Acknowledgements

The research described here was supported by grants from the Crafoord foundation and the Siegvald foundation. The author thanks Zoltan Dienes, Carl Martin Allwood, Mats Dahl, David Shanks, and two anonymous reviewers for valuable comments on previous versions of the manuscript.

Notes

¹In all experiments the participants were also given a questionnaire after completing the test phase, the results of which are not reported here. The questionnaire was included for exploratory purposes and contained questions about the purpose of the experiment, what kind of strategies the participants used, and about the masked priming procedure. The questionnaire was not designed to measure awareness of the rules or the primes explicitly, and would most likely be incomplete for that purpose. However, it is important to bear in mind that the masked priming procedure does not presuppose absolutely subliminal priming. It is perfectly possible that participants had some vague conscious experience of the prime or parts of it occasionally, just as it is perfectly possible that participants had some degree of conscious experience somewhere along the line in the test phase in the perceptual clarification procedure used by Kinder et al. (2003). However, if the fluency manipulation is all too obvious, then participants may discount it (Jacoby & Whitehouse, 1989).

²For example, the grammatical test items had higher bigram chunk strength than ungrammatical items, t(62) = 4.51, p<.01, and also higher anchor chunk strength, t(62) = 2.53, p<.05, but test items with high item-specific similarity did not differ from items with low item-specific similarity (both ps>.6). Both of the chunk strength indices are measures of the extent to which bigrams, i.e., two-letter combinations, in a test string occurred in the set of training strings (see, e.g., Tunney, 2005). Bigram chunk strength is based on all bigrams of a test string and anchor chunk strength is based on the first and last bigram of a test string. Of course, it cannot be ruled out that item-specific similarity is not also correlated with some nonspecific notion of similarity. Nevertheless, item-specific similarity is certainly less correlated than grammaticality with nonspecific notions of similarity.

³To the extent that the increase of the grammaticality effect under a response deadline is based on an increased effect of nonspecific similarity, one would expect the difference in nonspecific similarity between endorsed and rejected items to be higher under a response deadline as compared with free response time. Accordingly, the difference in average bigram chunk strength between endorsed and rejected items in Experiment 1 (free response time) was 0.41 (SE=0.05), i.e., higher bigram chunk strength for endorsed than rejected items. In Experiment 2 (deadline) the difference was 0.56 (SE=0.04). An independent-samples t-test showed these two difference scores to be significantly different from each other, t(62) = 2.41, p<.05, meaning that bigram chunk strength had a higher impact on classification in Experiment 2 than in Experiment 1. This shows that, even though grammaticality may be the important underlying variable, nonspecific similarity can at least not be excluded.

⁴It should be mentioned that even under the assumption that eventual effects of masked priming and of grammaticality are both based on fluency, the grammaticality effect will be the larger of the two effects most of the time, since the grammaticality effect is grounded in a training phase. This is especially so in light of the fact that perceptual fluency manipulations, like masked priming, are generally quite weak (Whittlesea, 1993). Also, the grammaticality effect might very well be based on other sources as well, apart from fluency.

⁵A previous version of this manuscript contained a different version of Experiment 3, which showed partly different results than the present one. However, an anonymous reviewer pointed out that in the old Experiment 3 there was an item confound for the priming and delay variables, which might explain the old results. In the present Experiment 3, the assignment of items to the priming and delay variables is counterbalanced across participants. I thank the anonymous reviewer for pointing out the original problem.