Phonetic variation in intervocalic onset /l/'s in English

Abstract The English /l/ is often described as having a finite number of contextually determined allophones: a backer, “velarize d”, allophone in syllable rimes, a fronter allophone in syllable onsets, and a syllabic allophone, a special variant of the back /l/, in stressless syllables. In contrast, Sproat & Fujimura (1993) argued that variation in /l/ quality is a more continuous phenomenon, governed by principles of phonetic implementation determining the timing and extent of an apical gesture and a dorsal gesture for /l/. They claimed that the /l/'s dorsum-backing gesture is attracted to the syllable nucleus. Therefore in rimes, dorsum backing begins well before the apical gesture, and the longer the rime, the more extensive the backing. In onsets, on the other hand, the dorsum gesture is either simultaneous with, or follows, the apical gesture, which means that /l/ is short, and there is little time for backing. The study reported here investigates the relation between timing and quality in intervocalic onset /l/'s following schwa, a segmental context expected to produce considerable temporal variation across tokens, but which, according to Sproat and Fujimura's model, should not show much variation in degree of backness. Acoustic data from eight female speakers reading in a fluid narrative style shows that onset /l/'s do vary in backness, and suggests that backness is related to gestural timing. Backer onset /l/'s are usually longer, which means that Sproat and Fujimura’s notions of gestural affinity and gestural separation require modification. The tongue dorsum gesture for /l/ appears to avoid temporal overlap with a stressed vowel, whether or not that vowel is in the same syllable. The data also show, however, that longer /l/'s are not always backer: the relation of duration and backness can be complicated by differences in coarticulatory effects of neighboring vowels, or by speaker-specific constraints on absolute degree of backness. All of these contributing factors can be accommodated within a model of phonetic implementation sensitive to prosodic structure.