My main research focus concerns the role of parent-child interaction in early acquisition of speech and language. I study human infants' early language and communicative development with a particular interest in how perceptual history (or lack thereof) may give rise to learning in social environments. How might lack of perceptual experience change learning scenarios for infants who are surrounded by much more perceptually mature adults?
Graded learning: What is the mechanism underlying infant vocal learning? Previous work suggests the power of social feedback in guiding vocal learning in humans, songbirds and primates. In humans, patterns of parents’ responses to their infant’s vocalizations are linked with healthy development, however the mechanisms underlying infants’ moment-by-moment learning from their parents’ speech are not well understood. Work with human participants highlights two key components of a social mechanism for vocal learning, one temporal and one distributional. First, results suggest that infants learn the vocal content of their parents’ speech better when parents respond promptly to infant babbling (i.e., provide contingent feedback). Second, it has been demonstrated that infants are better able to generalize learnt information to new input if previous training data contained sufficient exemplar variability. In this project, I ask how infant vocal learners incorporate patterns — which are perceptually rare to them — into their vocal production as a function of timing and exemplar variability of parents' speech responses to their babbling in real-time unstructured play.
The simplification effect: What is the function of infant babbling? Our recent findings suggest that infants’ immature vocalizations may elicit simplified linguistic responses from their caregivers. Already by 5 months of age, infants expect that their own vocalizations will elicit responses from their caregivers. By 9 months of age, infants regularly incorporate the content of parents' speech which is temporally organized around infants' own vocal behavior. Our results suggest that infants elicit simplified language structure from their caregivers — language infants have come to expect and are learning from. In this project, we are exploring possibilities these results implicate: multiple pathways along which infants may actively change their learning environment to contain more simplified, learnable information.
Automated babbling analytics: Do the acoustics of infants' babbling exhibit reliable and quantifiable signatures over the course of vocal development? Infants' prelinguistic vocal development is an early prerequisite of spoken language and communication. The classification of this development and its acoustic and infraphonological correlates require a high degree of ear-training and time-intensive procedure in manually annotating infants' vocalizations. While large-scale recordings of infant's vocalizations during vocal development are becoming available, automated detection of acoustic advances in their vocal productions is still best left to human coders and human ears. Currently, I am exploring the time-course of Mel-frequency cepstral coefficients derived from high-quality recordings of infant vocalizations during unstructured play with their caregivers. Mel-frequency trajectories may provide a quantifiable source of detection of infants' advancement in vocal forms during early vocal learning.