Poster #113 - Statistical Learning Performance in the Visual Domain Predicts Early Language Skill

• In Event: 3-053 - Poster Session 10
  In Poster Session: PS 10 Section - Language, Communication

Integrative Statement

Research demonstrates that infants can learn word segmentation, based solely on statistical information (Saffran et al., 1999; Kirkham et al., 2002). Based on these studies, the statistical learning (SL) theory argues that infants possess a domain general learning mechanism that allows individuals to detect and learn from statistical information in their environment. Further, this mechanism constitutes the foundation of structuring language information, allowing word comprehension (Lany & Saffran, 2013; Pelucchi et al., 2009). However, for the theory of SL to hold true, a longitudinal relationship between domain general SL and early language skill is a necessity. The performance in a non-verbal domain should be able to predict learning and/or performance in the language domain – which there is little empirical evidence of. Thus, this study aimed to investigate whether non-verbal SL is predicative of early language skill.

We used three eye-tracking tasks that were conceptualized to tap into non-verbal SL. Here, we will focus on the Visual Sequence task (see the other tasks in Figure 1). This task was measured at 10 and 18 months, and aimed to evaluate whether infants could learn the specific pattern of a sequence. To assess the infants’ language skill, the participants’ mothers answered the Communicative Development Inventories (CDI) questionnaire at 18 months. The CDI consisted of two subscales; CDI Know and CDI Know and say. The current study is a part of a longitudinal cohort. The initial sample consisted of 118 infants (50% females). Of this initial sample, 110 returned for the second lab visit at 10 months of age. At the forth lab visit, 104 participants returned, at 18 months of age.

A zero order correlation suggested that learning in the Visual Sequence task, at both 10 and 18 months, was associated with the CDI. The other two eyetracking tasks seemed to be associated with the Visual Sequence task. Based on this, we conducted a path analysis, where it was investigated if early outcome measurements could predict later outcome, and if any task was predicative of the CDI. The two subscales of the CDI were modelled separately. Several fit indices indicated that the model had a good fit; χ2 (4) = 3.742, p = .442; CFI = 1.0; TLI = 1.04; RMSEA < 0.01; SRMSR = .058. The model revealed that the Visual Sequence task at 10 months could predict the CDI Know subscale. Further, and of great importance, it could also predict the performance in Visual Sequence task at 18 months. The performance at 18 months could, in turn, predict both CDI subscales (see Figure 1).

In line with the SL theory, the model provides further evidence that infants possess a domain general mechanism that is important for learning language, demonstrating that learning in the visual domain predicts performance in the language domain. Further, the results indicate that sequential learning might be central for language learning.

Authors

• Anton Gerbrand, Uppsala University
  Presenting Author

• Gustaf Gredebäck, Uppsala University
  Non-Presenting Author

• Martina Hedenius, Institution of neuroscience, Logopedics, Uppsala University
  Non-Presenting Author

• Linda Forssman, Uppsala University
  Non-Presenting Author

• Marcus Lindskog, Uppsala University
  Non-Presenting Author