Keith Johnson, "Four phonetic discoveries: How I spent my sabbatical 'vacation'"

Colloquium | September 24 | 3:10-5 p.m. | 370 Dwinelle Hall

 Keith Johnson, UC Berkeley

 Department of Linguistics

In a scientifically young field like phonetics it is possible to make some relatively basic discoveries. I’ll talk about four that I made last year while on sabbatical.

1) Perceptual voice stereotypes are based on selective memory of spoken exemplars. This acoustic and perceptual study (with Eric Tracy of UNC-Pembroke) of gay men’s speech shows that listeners’ judgement of what sounds ‘gay’ is based on a subset of exemplars - namely exemplars that have some acoustic feature that makes them stand out. A model built on such ‘accessible’ exemplars matches listener data better than a model that draws on a full set of exemplars produced by gay men. (One bonus observation here - we could build an acoustic/ phonetic ‘gaydar’ that was much more accurate than listeners.)

2) Speech perception may involve speaker normalization, after all. The ‘famous' conclusion in Johnson (1997) that speech perception doesn’t involve speaker normalization, may have been wrong. A new approach for finding the length of the vocal tract from acoustic vowel formants is much more accurate, and feasible in real time, than methods that I rejected in my earlier work. Perception of vocal tract length may be an evolutionarily old ability - useful for survival and sexual selection. The study concludes that a normalization method based on perceived vocal tract length may have a simple neural implementation.

3) Vocal tract length and palate doming have different effects on articulation in the same speakers. This is a study of speech articulation based on data from the Wisconsin x-ray microbeam database. Studies of individual differences in speech production have been somewhat "vocal tract provincial” - focussed on differences in pharynx length, or palate doming, but not both; and on pronunciation of particular vowels or consonants. I looked at all of the anatomical parameters explored in previous work and played them off against each other in the same group of speakers. Earlier observations about the relationship between vocal tract anatomy and speech production patterns were replicated. (These findings about individual differences are really important for the study of sound change.) In addition, vocal tract length effects were largely independent from palate shape effects.

4) Neural activity in Broca’s area is correlated with lexical word frequency. A single speaker (a neurosurgery patient) spoke 3000 words (reading from word lists) while a grid of electrodes on the surface of the brain recorded activity over a wide swath of the left hemisphere. Half a second before word onset, neural firing in Broca’s area (Inferior Frontal Cortex) was higher for high frequency words than for low frequency words. Shortly later, overlapping with word onset, neural firing in a portion of the motor cortex was LOWER for high frequency words. By the onset of speech, these differences were largely gone. I have no theory for this, so end the talk with an observation in need of an explanation.

 mikkelsen@berkeley.edu