Carbajal, J. & Peperkamp, S.
Ecole Normale Supérieure
Research using preferential listening paradigms has found that 11-month-old infants learning different languages prefer listening to lists of familiar as opposed to unfamiliar words (e.g. Hallé & de Boysson-Bardies, 1994; Vihman & DePaolis, 1999; Vihman et al., 2004, Swingley, 2005). This has been taken as evidence for the existence of a recognition lexicon. In two experiments with French-learning infants, we failed to consistently find the same effect. A meta-analysis including 12 published and unpublished experiments revealed a medium-sized overall effect (Hedge?s g = 0.57, p < .0001). In order to observe this effect with a power of 80%, a sample size of at least 26 infants is needed; most studies, though, have smaller sample sizes. Additionally, a funnel plot analysis suggests a publication bias, such that there might be unpublished null results (correlation between SE and Hedge?s g: r = 0.72). We discuss problems related to the paradigm, such as the fact that a single recognized familiar word can drive the effect, as well as its sensitivity to trial order and the presence of outliers. We conclude that infants? developing recognition lexicon should best be studied using methods allowing for item-level analyses.