Patricia J. Brooks, “Individual Differences in Statistical Learning: Implications for Language Development”. Commentary by ARC Student Fellow Ian Phillips

In her recent talk at ARC titled Individual Differences in Statistical Learning: Implications for Language Development, Dr. Patricia J. Brooks presented ongoing research examining the relationship between individual differences in statistical learning ability and language development. In this talk, Brooks, Professor of Psychology at the College of Staten Island, City University of New York and ARC Distinguished Fellow, explored the relationship between statistical learning ability and both child first language acquisition and adult second language acquisition. The takeaway message is that statistical learning ability appears to influence the outcomes of both types of language acquisition, though this effect may be modulated by the quality and timing of feedback provided to the learner during language development.

Brooks started off by exploring how individual differences in statistical learning ability might underlie individual differences in linguistic skills. Statistical learning—aka procedural learning or implicit learning—is an inductive process in which the learner becomes sensitive to probabilistic patterns in the input. As Brooks pointed out, statistical learning is operative in infants and research suggests that it may facilitate acquisition of vocabulary, phonemic categories (language-specific sounds), and grammatical dependencies (e.g., subject-verb agreement). In further developing the relationship between statistical learning ability and language acquisition, Brooks presented her results from a meta-analysis of recent studies that show evidence that individuals with a language disorder known as Specific Language Impairment (SLI) also have a statistical learning deficit (Obeid, Brooks, Powers, Gillespie-Lynch, & Lum, 2015).

Brooks’s main argument is that immediate positive feedback is crucial to language development and this kind of feedback interacts with the statistical learning mechanisms that underlie language development. To illustrate how this interaction affects language development, Brooks weaved together evidence from three of her recent studies utilizing a variety of data collection and analysis techniques. Brooks first presented her research examining how environmental factors impact language outcomes in infants enrolled in the Early Head Start Research and Evaluation Project (EHSRE) to show that positive feedback from the caregiver is the biggest predictor of child language development (Poulakos, Brooks, & Jewkes, 2015). This study focused on investigating how the characteristics of mother, child, home, and social interaction when the infants were 14 months old affected language outcomes at 14 and 36 months, with the goal of determining whether the quality of social interaction with the caretaker impacts language development. In this research, Brooks analyzed data for 791 infants from low income families where English was the only language spoken at home and used a cumulative risk model to analyze the interactive effects that multiple factors—including maternal mental distress, negative interaction, maternal education, gestational age, child cognition, and child gender—had on language outcomes at 14 and 36 months. The analysis showed not only that a large percentage of the infants in the study were significantly delayed in language development at 36 months compared to all children in the EHSRE data but importantly at both 14 months and 36 months, joint attention—the factor indexing positive caretaker-child social interaction—was the biggest predictor of child language development.

After establishing the link between social interaction and language development, Brooks presented a second study in which she analyzes child-caretaker interactions in the CHILDES Clinical English Weismer SLI Corpus to determine which specific aspects of social interactions affect language development in late talking children (children using no words at 18 months, and fewer than 50 words and no word combinations at 24 months) (Che, Alarcon, Yannaco, & Brooks, 2015). This analysis shows two important findings: first, for late talkers at ages 30, 42, and 54 months, mothers and children show close to 20% overlap in each other’s speech, where overlap is defined as an imitation that may be either expanded or reduced; second, maternal overlap of child speech at 30 months is the single best predictor of a child’s mean length utterance (MLU; a standard measure of language proficiency) at 54 months of age—this factor is more predictive of language development at 54 months than child overlap, child MLU at 30 months, or the amount of mother speech. Brooks noted that this overlap is the critical just-in-time feedback needed for language development and cites research suggesting that this type of feedback provided by caretakers during social interaction facilitates language pattern extraction or rule learning.

After detailing how child language learning is affected by social interaction, Brooks shifted her focus to adult second language learning and presented a final experiment showing how individual differences in statistical learning ability predict language learning outcomes in adults (Brooks, Kwoka & Kempe, submitted). In this experiment, English-speaking college students completed three two-hour language learning sessions over 2-3 week period where they were exposed to spoken Russian phrases in a question-answer dialog using a computer program. Importantly, during the training phase of the language learning experiment, participants were provided with the type of immediate feedback that Brooks argues is crucial to language learning—they heard the correct phrases repeated immediately after answering each question. The results show that performance on both language comprehension and production tasks during each of the three language learning sessions is predicted by individual differences in statistical learning ability, as measured by two separate statistical learning tests.

There is much debate about the mechanisms that underlie both child first language acquisition and adult second language acquisition. In this talk, Brooks presented new evidence suggesting that general statistical learning ability plays an important role in language development in both children and adults. Brooks complemented research showing the importance of social interaction for child language learning with new evidence that suggests that what’s important about social interaction is the just-in-time positive feedback. While this work is important for language acquisition research, it also has implications for developing interventions to close reported gaps in child first language acquisition and applications for adult second language learning.