/l/ variation in american english

Jiahong Yuan,Mark Liberman

doi:10.20396/joss.v1i2.15025

Abstract

We investigated the variation of /l/ in a large speech corpus through forced alignment. The results demonstrated that there is a categorical distinction between dark and light /l/ in American English. /l/ in syllable onset is light, and /l/ in syllable coda is dark. Intervocalic /l/ can be either light or dark, depending on the stress of the vowels. There is a correlation between duration (the rime duration and the duration of /l/) and /l/-darkness for dark /l/, but not for light /l/. Intervocalic dark /l/ is less dark than canonical syllable-coda /l/, but it is always dark, even in very short rimes. Intervocalic light /l/ is less light than canonical syllable-onset /l/, but it is always light. We argue that there are two levels of contrast in /l/ variation. The first level is determined by its affiliation to a single position in the syllable structure, and the second level is determined by its phonetic contexto.

Highlights

It has long been observed that the lateral liquid /l/ in English has two main varieties: a light /l/ and a dark /l/
/l/ in V0LV1 is less light than in the canonical onset position, #LV1. This suggests that instances of word-medial intervocalic /l/ preceding a stressed vowel are phonetically “ambisyllabic”, it is generally agreed in phonology that such /l/ is syllable onset
The analysis above demonstrates that there is a categorical distinction between dark and light /l/ in American English. /l/ in syllable onset is light and /l/ in syllable coda is dark

Summary

Introduction

It has long been observed that the lateral liquid /l/ in English has two main varieties: a light /l/ and a dark /l/. In a corpus study of the variation of /l/ in American English, we used a forced alignment technique to measure the degree of darkness of /l/ in a large speech corpus. /l/ in V1LV0 was always dark, even for very short rime durations This result contradicts the proposal of Sproat and Fujimura (1993) that there is no categorical distinction between dark and light /l/. In Yuan and Liberman (2009), we developed an innovative technique to measure the “darkness” of /l/ based on probability scores from forced alignment In this method, we trained two models of /l/: one for light /l/ and another for dark /l/. (1) D log p(l | L2) log p(l | L1); (L1 and L2 are the models for light and dark /l/, respectively)

Data and procedure

Analysis

Findings

Discussion