Using corpus-driven TLU domain analysis to increase the authenticity of listening passages

Many high-stakes language tests use scripted listening passages that lack many features of authentic speech (Wagner, 2016), potentially resulting in construct under-representation. One solution is to ‘authenticate’ or revise passage scripts to include more natural language features (Wagner et al., 2021, p. 206). Empirical analysis of real-world language use can help ensure effective authentication. The present study analyses the use of turn openers, a prevalent oral language feature that is crucial to turn-taking, in graduate academic U.S. English. A subset of the Michigan Corpus of Academic Spoken English (MICASE) (Simpson et al., 2002) was analysed to identify common turn openers (the first sound, word, or phrase of a new speaker turn; Tao, 2003) and their clusters, functions, and relationships to contextual variables (discipline, interactivity, speaker status, graduate student role). The nine most frequent forms (including yeah, okay, mhm) clustered and functioned differently and seemed unimpacted by contextual variables. Implications for listening script authentication are discussed. This article was published open access under a CC BY licence: https://creativecommons.org/licenses/by/4.0/ .