The Influence of Bilingualism on Statistical Learning With Multiple Inputs

Restricted (Penn State Only)
Poepsel, Timothy John
Graduate Program:
Doctor of Philosophy
Document Type:
Date of Defense:
August 29, 2016
Committee Members:
  • Dan Weiss, Dissertation Advisor
  • Dan Weiss, Committee Chair
  • Judith Kroll, Committee Member
  • Matthew Carlson, Committee Member
  • Kristin Buss, Outside Member
  • Statistical learning
  • language acquisition
  • bilingualism
  • speech segmentation
  • word learning
Statistical learning is a fundamental component of language acquisition, yet to date, relatively few studies have examined whether these abilities differ in bilinguals, especially when the learning task presents multiple statistically independent distributions. Furthermore, extant work comparing monolinguals and bilinguals has been undertaken in a single statistical learning paradigm representing only one of many components of the process of language acquisition. In this dissertation, we addressed these gaps in the statistical learning literature by comparing the performance of monolinguals and late-learning bilinguals in two types of cross-situational statistical learning (CSSL) tasks that contained multiple statistical distributions. In Experiment 1, learners (English monolinguals, English-Spanish bilinguals, and Chinese-English bilinguals) were asked to form both one-to-one and two-to-one word-object mappings, and were tested at three points during training. All groups performed identically on one-to-one mappings, but bilinguals outperformed monolinguals on two-to-one mappings, acquiring these mappings both more quickly and proficiently. In Experiment 2, learners (English monolinguals and English-Spanish bilinguals) were asked to form one-to-one mappings and two-to-one mappings that varied with respect to the amount of evidence supporting either side, and were tested at a single point after training. The average performance of monolinguals and bilinguals on both one-to-one and two-to-one mappings did not differ; however, bilinguals more frequently acquired both sides of two-to-one mappings when evidence for each side was equal, and also showed increased sensitivity to two-to-one mappings at lower thresholds of evidence. In Experiment 3, learners (English monolinguals and English-Spanish bilinguals) completed a speech segmentation task presenting multiple statistically distinct streams that differed in the degree to which their components (i.e., syllable inventories) overlapped. Here, we did not find a difference in the ability of monolinguals to acquire multiple inputs or manage the degree of statistical interference between two inputs, but we did note a significant effect of degree of overlap on learning, so that greater overlap correlated with decreased learning. In accord with previous research, the results of these studies suggest that the fundamental ability to track the statistics of language input may not be affected by bilingualism. However, we found distinct advantages for bilinguals in the acquisition of multiple distributions, and that this effect is modulated by the type of statistical learning task presented to learners. Overall, our results suggest that bilingual experience may impact a learner’s response to variability in an input, and that the statistical learning mechanism may be comprised of multiple subcomponents that function and respond differently based on both a learner’s experience and task-specific demands.