ABSTRACT
Skillful performance in a human language often involves a composite of elementary skills, such that language skills, cognitive skills, and social skills can be conflated in the judgment of a human listener. A recent, computerbased method (Versant for English) provides an estimate of spoken language skills that is relatively independent of the speaker’s other social and cognitive skills. The Versant system has been implemented using interactive voice response to administer a standardized spoken English test that measures speaking and listening during a 12-minute telephone call. The system calculates scores on five performance subscales from a set of more basic measures that are produced by automatic speech recognition of examinee responses. Item response theory is used to analyze and scale aspects of examinee performance. The scores are also related to performance rubrics used in criterion-based human scoring of similar responses. This chapter outlines the test construct and describes the scaling methods and validation process with reference to experimental procedures used in developing the test.
