Simulating the social dynamics of motherese in learning vowel postures
Most relevant references
Ilana Heintz, Mary E. Beckman,
Eric Fosler-Lussier, & Lucie Ménard (2009).
Evaluating parameters for mapping
adult vowels to imitative babbling.
Proceedings of InterSpeech2009.
Andrew R. Plummer, Mary E. Beckman, Mikhail Belkin,
Eric Fosler-Lussier, & Benjamin Munson (2010).
Learning speaker normalization using
semisupervised manifold alignment.
Proceedings of InterSpeech2010.
work supported by National Science Foundation grants
BCS-0729306, BCS-0729140, & BCS-0729277
Dynamics of Human Behavior/Collaborative Research:
Using machine learning to model the interplay of production dynamics and perception dynamics in phonological acquisition.
Mary E. Beckman
Linguistics, Ohio State University
Communication Disorders, University of Wisconsin
Computer Science and Engineering, Ohio State University
Speech-Language-Hearing-Sciences, University of Minnesota
Outcome: As part of a multi-site collaboration, linguist Mary Beckman and computer scientists Eric Fosler-Lussier and Mikhail Belkin at the Ohio State University are building computer models of how babies learn the consonants and vowels of their mother language, despite the fact that their vocal tracts are too small for them to acoustically match ambient speech sounds. One of the first models that we built demonstrated that a mother "mimicking back" sometimes when she hears babblings that she interprets as the vowels in "heat" and "hot" is enough to make the baby learn a mapping not just for those two sounds, but also for other intermediate sounds, such as the vowel in "hay".
Impact: In building different computer models, we vary the simulated input from the baby's own babbling (to reflect effects such as the silencing of tracheostomy) and also the simulated input from adult speakers (to reflect differences in the amount of speech that the baby hears). A better understanding of the role of input can help us develop more effective policies and programs to aid parents of babies at risk for language delay.
Background: It seems obvious that babies learn speech sounds by imitating the speech they hear. However, because babies' vocal tracts are tiny, the imitation cannot be a simple pattern match. For example, when a mother says the vowels in "heat" and "hot" the frequencies (in Hz) of the resonances that indicate the different mouth and tongue gestures for these vowels are completely non-overlapping with the frequencies that the baby's vocal tract can produce. And there is "misleading" overlap in the parts of the vowel resonance space that are shared. The babbling resonance pattern that the mother interprets as the vowel in "hoot" juts into the mother's space for her vowel in "hay", a sound halfway between "heat" and "hot" in terms of degree of mouth opening. So babies have to learn a complex "mapping" between their mother's speech and their own babbling.
We are trying to understand this complex learning process by building mathematical models called manifolds. A manifold describes what our brains might know about something that is very complex and multi-dimensional by building a much lower-dimensional "map" of it. For example, a map of the world is a two-dimensional manifold built to describe what we need to know to navigate the three-dimensional surface of our planet.
Our models combined real-world data from children's productions, real-world data on the speech spoken to children, and real-world data on adults' perception of the accuracy of children's productions. We have already shown that manifolds are a useful way of understanding some key aspects of early language acquisition, and we are working with manifolds to try to understand more key aspects.