Numerous scientific reports have previously proposed that infants group sounds into native vowel- and consonant-like phonetic labels through a statistical clustering system called ‘distributional learning.’
This early phonetic learning concept that infants learn phonetic categories has been recently challenged by new research published in the Proceedings of the National Academy of Sciences. In the study, a multi-institutional team of scientists and computational linguists has come up with a quantitative modeling framework based on a simulation of the infants’ language learning process.
Early Language Acquisition
Using computational efficient machine learning methods, this new concept allows learning mechanisms to be consistently associated with testable predictions regarding infants’ adjustment to their native language.
Thomas Schatz, the lead of the study and a postdoctoral associate in the University of Maryland Institute for Advanced Computer Studies (UMIACS), explained: “Hypotheses about what is being learned by infants have traditionally driven researchers’ attempts to understand this surprising phenomenon. We propose to start from hypotheses about how infants might learn.”
For their research, the scientists simulated the learning process in infants by training a computationally efficient grouping algorithm on realistic speech input. The algorithm was catered with auditory features resembling spectrograms at regular timeframes taken from naturalistic speech recordings in a specific language.
This resulted in a candidate model for the early phonetic knowledge of a, for instance, Japanese infant, the scientists explained. Next, they asked two questions of the trained models: ‘Could they explain the noticed difference in how Japanese and English-learning infants make the difference between speech sounds?’ and ‘Did the models learn consonant- and vowel-like phonetic categories?’
The team found that the answer to the first question was positive, but the second question’s answer was negative. The outcome suggests a massive reinterpretation of the existing literature on early phonetic learning. Challenges in scaling up distributional learning of phonetic categories to realistic learning circumstances may be better considered as questioning the concept that what infants learn are phonetic categories, instead of the idea that how infants learn is via distributional learning.
The scientists believe their approach and ongoing efforts in the field to gather empirical information on a large scale opens the path toward a better understanding of early language acquisition.