Estimating the accuracy of word-initial consonant clusters in child speech data

Abstract

Children have difficulty in producing consonant clusters in adult-like manner. A widely used measure for the proximity of consonant clusters is the proportion of accurately produced clusters to intended clusters (PClC). Recently, Babatsouli and Sotiropoulos proposed the measure for cluster proximity (MCP) which differentiates inaccurate clusters in developmental stages. The purpose here is two-fold: to highlight the differences between the two measures for individual children in typical or atypical speech development, and to examine the correlation between the two measures for speech data from several children. The data comprise 33 speech samples each with 18 word-initial clusters from 17 children: 1 monolingual male at age 2 years and six months; 1 bilingual female at ages 2 years and seven months, three years, and three years and five months; and 15 other children in atypical development whose speech data was collected before and after therapy. The results show that the two measures are strongly and significantly correlated across children, cumulatively for all word-initial clusters and classes of clusters, cumulatively for all children. Therefore, a universal norm for MCP can be predicted linearly from a universal norm for PClC, with a high coefficient of determination.