ABSTRACT
In a variety of domains, adults who are given input that is only partially consistent do not discard the inconsistent portion (regularize) but rather maintain the probability of consistent and inconsistent portions in their behavior (probability match). This research investigates the possibility that adults probability match, at least in part, because of two pragmatic assumptions they bring to the learning problem: (a) that the variation they see is predictable rather than random and (b) that their goal is to correctly learn that variation. Evidence from two experiments demonstrates that when either assumption is eliminated, people probability match less and therefore regularize more. These results are discussed with respect to age and domain differences in regularization.
Funding
This work was supported by ARC DECRA Fellowship DE120102378.
Notes
1. Stems were DUT, SIL, ZEG, MAB, YOK, PIM, REN, JAF, WUX, and COV. Items used were babies, balls, beds, birds, books, cars, cats, cups, dogs, and shoes.
2. Normality assumptions for the accuracy scores were violated in all conditions in which the parametric Kruskal-Wallis test and post-hoc Wilcoxon tests were applied.
3. Complete descriptive statistics for all measures can be found in Appendix C.
4. The analyses were conducted without this exclusion. In order to identify any learning effects, all analyses were also repeated for just the second half of trials. In all cases the results were qualitatively identical.
5. Two pairwise tests (rather than a full 2 × 2 ANOVA on the four central conditions) were performed for two reasons. First, those two tests correspond to the specific hypotheses this paper focuses on: whether people’s regularization was influenced by their perception of whether variation was predictable (i.e., experimenter vs participant) or whether it depended on the nature of the affixes (similar vs typos). Second, the non-normality of the data meant that an ANOVA was inappropriate, and there is no equivalent non-parametric test for 2 × 2 data. That said, in keeping with the pairwise tests, that ANOVA revealed a significant effect of how the data originated (i.e., experimenter or participant) but no effect of the nature of the affix (i.e., similar vs typos) and no interaction (data origin: F(2) = 5.79, p = .004; nature of the affix: F(1) = 0.473, p = .493; interaction: F(1) = 0.58, p = .448).
6. As in Experiment A, all analyses were qualitatively identical when the exclusion criteria were dropped or were restricted to the second half of test trials.