Durational and spectral factors in judgements of American Raising

Abstract

Canadian Raising and its relatives in the USA can respond to underlying voicing of flapped /t/ (Raised “writer” vs. un-Raised “rider") from the earliest stages (Fruehwald 2016). How so, if Raising is phonologized from a phonetic precursor sensitive only to phonetic features? Proposal (Bermúdez-Otero 2019): Raising responds to duration, not voicing: diphthongs are shortened before underlyingly voiceless sounds by a pre-existing lexical phonological rule of Pre-Voiceless Clipping, after which postlexical Raising transparently affects the shortened diphthongs. Experiment: Participants (N = 141, US dialects) read wordlists with /ai/ and /ei/ in voiceless and voiced contexts, then sorted the words into groups judged to share a vowel (DiPaolo & Faber, 1990). Clipping and Raising were measured using duration, F1, and F2. Predictions: (1) across speakers, /ai/-Clipping and / ai/- Raising should be positively correlated. (2) /ai/-sorting should be better predicted by /ai/-Clipping than by /ai/-Raising, because lexical rules change phonemes, while postlexical ones are subphonemic. (3) /ei/-sorting should be positively correlated with /ai/-sorting, since Clipping affects all vocoids. Results: (1) /ai/-Clipping did *not* predict /ai/-Raising (r = −0.082); (2) / ai/-Clipping predicted word sorting marginally *worse* than /ai/-Raising (95% CI for rclipping-rraising = (−0.50, 0.03)). (3) /ei/ and /ai/ judgements *were* positively correlated (r = 0.37, 95% CI = (0.21, 0.51)).