Published online by Cambridge University Press: 02 November 2017
Using speakers of either African American English or Southern White English, we asked whether a working memory measure was linguistically unbiased, that is, equally able to distinguish between children with and without specific language impairment (SLI) across dialects, with similar error profiles and similar correlations to standardized test scores. We also examined whether the measure was affected by a child's nonmainstream dialect density. Fifty-three kindergarteners with SLI and 53 typically developing controls (70 African American English, 36 Southern White English) were given a size judgment working memory task, which involved reordering items by physical size before recall, as well as tests of syntax, vocabulary, intelligence, and nonmainstream density. Across dialects, children with SLI earned significantly poorer span scores than controls, and made more nonlist errors. Span and standardized language test performance were correlated; however, they were also both correlated with nonmainstream density. After partialing out density, span continued to differentiate the groups and correlate with syntax measures in both dialects. Thus, working memory performance can distinguish between children with and without SLI and is equally related to syntactic abilities across dialects. However, the correlation between span and nonmainstream dialect density indicates that processing-based verbal working memory tasks may not be as free from linguistic bias as often thought. Additional studies are needed to further explore this relationship.