Front/non-front asymmetry in word-final English /l/-velarization: Evidence from preceding-vowel frontness*

Abstract

This study reanalyzes the acoustic data from the author’s doctoral dissertation, examining whether the acoustic endpoint of English word-final /l/-velarization varies with the frontness of the preceding vowel. The original study observed that /l/ following front vowels (/i:, e, eɪ/) converges toward a schwa-like region while /l/ following non-front vowels (/ɔ:, oʊ, u:/) converges toward a separate, peripheral region. Using linear mixed-effects models with speaker and vowel as random effects (n=7 American English speakers), the study found that this asymmetry is significant for F1 (p=.013) and marginal for F2 (p=.053), collapsing across the steady state and offset; the frontness-by-timepoint interaction is not significant for either formant. Token-level analyses corroborated the overall asymmetry, while showing that /ɔ:/ and /eɪ/ are classified near chance level and that /i:/ dissociates from /e/ and /eɪ/ in offset height despite patterning with them in frontness. Critically, a continuous coarticulatory model, based on the /l/'s own onset formants and vowel duration, fits the data as well as, and more parsimoniously than, the categorical grouping. These results indicate a front/non-front bias in the acoustic endpoint of /l/-velarization that is not clearly distinguishable from continuous coarticulation, and should be read as evidence about endpoint position rather than as an articulatory target.