Does F3-F2 distance predict transcriptions of preschoolers’ /r/ productions?

Abstract

Children learning to speak American English are highly variable in early productions of /r/, often producing [w] as an error. One acoustic indicator of /r/ goodness is the difference between the second (F2) and third formant (F3), with smaller F3-F2 differences being associated with a ‘better’ /r/. This study analyzed the effectiveness of this acoustic measure in characterizing children's productions of the /r/-/w/ contrast on acoustic information extracted automatically from a large data set, where hand measurements are infeasible. LPC formant measures were extracted using Praat from productions of /r/- and /w/-initial words (n = 19) by 3.5- to 4.5-year old monolingual preschoolers (n = 136). These were rated by naive adult listeners (n = 142) on a visual-analog scale of phoneme goodness, and were narrowly transcribed by trained researchers into four categories: [r], [w], and two intermediate categories. Data visualization showed a weak relationship between F3-F2 measures, transcriptions, and ratings. Generalized linear mixed effects regressions found that the F3-F2 difference only modestly predicted transcription categories (R2 = 0.39) and listener ratings (R2 = 0.34). Hence, a description of preschoolers’ mastery of the /r/-/w/ contrast based on automatic analyses should include additional measures beyond the F3-F2 measures.