Limited talker variability differentially impacts artificial grammar learning in preschoolers and adults

• In Event: 3-235 - Contending with talker variability across development

Integrative Statement

A well-known challenge for language learners is that the input is typically produced by a variety of talkers. While processing input from multiple talkers can impose processing costs (e.g. Jusczyk et al, 1992), increasing talker variability may also be helpful for language learning (Rost & McMurray, 2009). In the current study, we investigated how limited talker variability, receiving input from only two talkers, impacts learning. Previous infant research suggests that limited talker variability can interfere with speech segmentation (Estes & Lew-Williams, 2015) but both adults and children can use talker-specific information to learn talker preferences (Eisner & McQueen, 2005; Weatherhead & White, 2016). We extend this field by asking whether input presented by two talkers impacts learning an artificial grammar consisting of two categories of novel objects. We ask whether talker information interacts with the information to be learned, by testing learning of individual object labels and categories. We investigate these differences in preschoolers and adults, as reliance on talker-specific information is thought to depend on experience. We additionally test learning at multiple time points, as testing delays can highlight different learning outcomes (Vlach & Sandhofer, 2012).

We tested two groups of participants, 96 monolingual English speaking adults (=18.9y) and 48 preschoolers (=43m) at four separate time points: immediately after familiarization, after a short delay (10-15m), a long delay (1 week), and a brief reexposure. Within each group, participants were assigned to a condition with 1- or 2-speakers. Learners were exposed to sentences describing novel objects. The object names were divided into two categories that were defined by word-final vowels which differed across categories, and preceded by a category-specific article (Figure 1).

Mixed-effects models were conducted for each aspect of learning (noun, article) across the four time points for each group. There was no effect of condition for learning of the nouns for adults, t(91) = .8, p=.43, or preschoolers, t(46) = .29, p=.77. However, there was an effect of condition on the article test for adults, t(91) = 2.06, p=.04, and preschoolers, t(46) = 3.53, p=.001. Preschoolers in the 2-speaker condition trended towards above chance performance at the immediate test and subsequently exhibited above chance learning at the second time-point, whereas those in the 1-speaker condition never exhibited learning. For adults, the effect of condition was quantified by an interaction between test time and condition, t(276) = 2.27, p=.024, such that participants in the 2-speaker condition only exhibited learning of the article-noun pairs after a brief reexposure (Figure 2).

Results suggest that limited talker variability does not impact learning of novel object labels, but differentially impacts learning of article-noun pairings for preschoolers and adults at different time scales. Preschoolers are facilitated by the presence of limited variability, whereas the 2-talker condition inhibits learning for adults. A brief reexposure period allows adults to demonstrate learning even under conditions of limited variability. Taken together, the impact of limited talker variability may depend on the age of the learner, the amount of exposure to the variability, and the material to be learned.

Authors

• Federica Bulgarelli, Duke University
  Presenting Author

• Daniel J Weiss, Penn State University
  Non-Presenting Author