Predicting children's word recognition accuracy from accent distance metrics

Abstract

Compared to adults, children show lower word identification accuracy under both environmental- and source-related listening challenges (e.g., speech in noise or unfamiliar accents). Early school-aged children have difficulty understanding both unfamiliar native and nonnative accents. Although adults typically have difficulty only when unfamiliar accents are mixed with noise, children show word identification decrements for some unfamiliar accents under quiet conditions. Furthermore, children's word recognition accuracy varies widely across accents in both quiet and noise-added conditions. The developmental differences appear to stem partially from children's less robust use of semantic/syntactic contextual cues; less well understood is the role of specific accent characteristics leading to reduced word identification. We are currently assessing the relation between a number of accent distance metrics and children's word recognition accuracy. These metrics, which quantify the distance from the home dialect, include segmental (i.e., Levenshtein distances), holistic signal (i.e., dynamic time warping), holistic perceptual (i.e., listener rankings from the native dialect standard), and suprasegmental (e.g., articulation rate, speech rhythmic metrics) measures. This work may provide insight into what aspects of the signal cause children more difficulty than adults when mapping unfamiliar pronunciations onto their lexical representations. Work supported by the National Science Foundation, Grant No. 1941691.]