Cristia, A. 1 , Seidl, A. 2 & French, B. F. 3
1 Laboratoire de Sciences Cognitives et Psycholinguistique, CNRS, IEC-ENS, EHESS
2 Purdue University
3 Washington State University
An emergent line of work suggests vocabulary size can be predicted from infant speech perception measures. Here, we critique and augment this literature. First, we meta-analyzed 18 studies that linked speech perception before 12 months and vocabulary size. The median effect size was significant for all linguistic levels (sounds r = .35 [confidence interval .22;.47], words r = .28 [.14;.4], and prosody r = .42 [.18;.61]). These effect sizes overlap with those for well-established non-linguistic predictors arising from habituation (r = .45 [.2;.65]), dishabituation (r = .42 [.29;.54]), and rapid auditory processing (r = .54 [.25;.74]). This overlap suggests that infant speech perception tasks are capturing individual variation that is stable enough to result in significant correlations with vocabulary outcomes. However, this may also be interpreted as evidence that all measures are assessing a similar construct, including something as non-specific as `task performance'. We addressed this possibility through a multivariate approach. We gathered multiple measures from 45 infants tested at 5-6.5 months on a linguistic (preference for trochees) and a cognitive task (Visual Recognition Memory), and at 6.5 to 8 months on another linguistic (vowel discrimination) and another cognitive task (A-not-B). If all tasks measure a single performance construct, correlations across all measures should be comparable. This prediction was not met, since only the linguistic tasks showed a significant association (r = .3) and correlations between the linguistic tasks and the cognitive tasks were markedly lower (-.2 to .1). Moreover, these results are most compatible with a view of language development in which advancement in processing of vowels is more independent from cognitive processing than from prosodic knowledge. While we have not yet collected outcome vocabulary measures, these data already suggest that a multivariate approach may greatly inform our understanding of how infants begin to build language.