Using text-to-speech in second language research and assessment: preliminary results from elicited imitation

Text-to-speech (TTS) is a technology widely used in second language learning and testing, for example in the case of learning material creation, pronunciation training, and speaking assessment. Such advances suggest that TTS may also hold great potential for automating performance-based measures of L2 knowledge and proficiency such as the Elicited Imitation (EI) test. However, questions remain about the impact of the technology on validity and fairness and consequently on how learners’ linguistic competence is interpreted. Both issues are crucial in the context of high-stakes language assessment but also for SLA research, where elicited performance is used to infer learners’ underlying interlanguage development. We show that EIs developed with TTS-synthesised stimuli demonstrated comparable construct validity as tests created with human-spoken stimuli and do not discriminate against test takers of varying proficiency or different first language backgrounds. EI scores were also not significantly different between human-spoken and TTS-synthesised items. These findings provide preliminary evidence for the use of TTS technology in language learning and language assessment contexts, with practical implications for researchers, educators, and test developers working in second language acquisition and bilingualism.