The Changing Role of Attention in Statistical Learning Over Development

• In Event: Learning How the Brain Acquires Statistical Regularities: Implications for Education

Abstract

Objectives: Statistical learning (SL), the ability to implicitly learn and adapt to regularities embedded in the environment, is a powerful mechanism and a central theoretical construct in typical language development. Both adults and children are sensitive to regularities in the language inputs, but how the developmental changes in the brain shift the relative weightings of the attention-dependent and attention-independent learning systems for SL remains poorly understood. Across three experiments (1: behavior; 2: fMRI; and 3: ERP), we ask how linguistic SL differs between children and adults. Specifically, we test the hypothesis that compared to children, adults engage more attention in processing and learning the regularities from the language inputs.

Theoretical Framework: This work will clarify the role of attention during SL across age groups and expand the current multicomponent model of statistical learning from the developmental perspective.

Methods and Materials: In Experiment 1, 54 children (MAge=8.5), and 44 adults completed a speech segmentation task while performing a target-detection task. Learning was measured by reaction-time acceleration during the 5-minute familiarization phase and triplet-recognition accuracy in a post-learning two-alternative forced-choice task. In Experiment 2, we collected fMRI data from 18 children (MAge=7.8) and 27 adults who completed a similar speech segmentation task in the scanner. By contrasting children’s responses to structured and random sequences of syllables, we extracted the hemodynamic responses to speech regularities within the language network and the dorsal attention network. In Experiment 3, 22 children (MAge=10.2) and 45 adults completed an auditory oddball paradigm while their EEG responses to speech deviants were monitored. We compared children with adults in their sensitivity to the probability of occurrence for the deviant stimuli using the magnitude of two ERP components: the mismatch response (MMR) and the late discriminative negativity (LDN).

Results: Experiment 1: During the familiarization phase, adults showed a better hit rate than
children for the target detection task (t80.5 = 6.29, p < 0.001), indicating greater task-relevant attentional engagement in adults. However, children showed significantly faster acceleration in reaction time than adults (t76.3 = -3.97, p < 0.001) suggesting more efficient SL. In the post-learning task, both age groups performed similarly above chance. Experiment 2: During SL, children’s language network showed greater sensitivity to the structured vs. random syllable sequences compared to adults (t31.5 = -1.838, p < 0.05). In contrast, adults’ dorsal attention network was more activated than children's during the entire task (t80.6= 4.81, p < 0.001). Experiment 3: Children’s MMR (F1,65 = 7.5, p < 0.001) and LDN (F1,65 = 6.2, p < 0.001) were both more sensitive to the global statistical information than adults. Adults, instead, showed a hint of attentive responses indexed by a P3a component. Together, these results suggest that SL in children appears to be more automatic and efficient than in adults.

Significance: Our work demonstrates critical differences between children and adults in a speech segmentation task with seemingly similar learning outcomes. The multimodal process-based data provide converging evidence suggesting that brain maturation leads to more attentional engagement during implicit learning.

Authors

• Zhenghan Qi, Northeastern University
  Presenting Author

• Anqi Hu, University of Delaware
  Non-Presenting Author

• Yi-Lun Weng, University of Delaware
  Non-Presenting Author