General Meeting - Leslie Li: How much rhythm in language is in short-time acoustics?
General Meeting - Leslie Li: How much rhythm in language is in short-time acoustics?
Friday October 15, we will have our first General Meeting, to share work across the various constituencies of our department. Leslie Li kicks it off with a presentation of her research in computational models of phonology, asking: "How much rhythm in language is in short-time acoustics?"
Theories proposed to characterize differences in the rhythm of languages did not predict the representation of such rhythm in the human mind. Furthermore, these theories are not generalizable to the form of speech (low-pass filtered speech) which infants were tested on. In this project, I focused on one recent hypothesis that rhythmic classes can be discriminated based on short-time acoustics independent of rhythm (Carbajal et al., 2016). I implemented an acoustic-phonetic model and simulated experiments of infant language discrimination on both natural speech and low-pass filtered speech. Results suggest that while the model successfully captures infants' behavior in natural speech, it fails to capture any realistic information under low-pass filtered speech. This suggests while acoustic-phonetic information is a great part of language discrimination across rhythmic classes, additional features in speech are used by infants and not captured by the current model.