Acoustic and Temporal Analysis for Assessing Speaking

Oral assessment in language learning has received increasing attention among second language acquisition (SLA) researchers. This growing interest is likely a product of the increased interpretability of test scores and potential validity of the scores when linked to real-world criteria (Bonk & Ockey, 2003). However, assessing speaking skill can be more challenging than assessing other skills because of the possible subjective nature of listener comprehension, the complexity of rater reliability, and the validity of the performance itself. Of these challenges, the potential variability in rater judgments has been of particular concern for language assessment as a source of measurement error (e.g., Bachman, Lynch, & Mason, 1995). A variety of human rater biases are attested to in the perceptions of speaking proficiency, and the speaking assessment may have a limited basis in the linguistic characteristics of the speaker’s oral production. Although sophisticated statistical techniques derived from Rasch scaling or generalizability theory (G-theory) can in principle equate practiced ratings which may display different degrees of rigor or leniency among raters (Lumley & McNamara, 1995), a technology-based measurement strategy that compensates for the variation in rater judgments of oral proficiency is much to be desired (Kang, Rubin, & Pickering, 2010). In fact, certain acoustical and temporal features of non-native speakers’ (NNSs’) pronunciation, measurable by means of instrumentation rather than by listener impressions, can now provide supplementary parameters for “degree of accentedness.” Thanks to advances in speech science, we can readily identify acoustic and temporal features of pronunciation that affect listeners’ comprehensibility. That is, computer-assisted instruments can conveniently examine some elements of the physical facts of human utterances. In this chapter, the primary focus lies in