How Reliable is Adult Statistical Learning? A Critical Review of Literature Since Saffran, Newport, and Aslin (1996)

Saffran, Newport, and Aslin (1996) demonstrated that adults can recognize words of an artificial language at an above chance level after mere exposure to statistical regularities in the speech stream. This seminal work led to a resurgence of interest in how experience contributes to language acquisition. Although statistical learning in adults has been replicated in many studies, null effects have also been reported in the literature. In this paper, I critically review the statistical analyses performed by Saffran, Newport, and Aslin (1996) and discuss how some practices could have increased the false positive rate and caused difficulties in interpreting the results, including the lack of generalizability. As effect sizes are commonly inflated in earlier publications, I also conduct a publication-bias-adjusted meta-analysis focused on 20 studies that closely followed the original experimental design. The adjusted effect size is between the two original effect sizes from Saffran, Newport, and Aslin’s (1996) Experiments 1 and 2. Suggestions are given for future research on adult statistical learning and the broader field of psycholinguistics.