Constructing a longitudinal learner corpus to track L2 spoken English

Abstract

The main purposes of this article are to provide an overview of a research project on a longitudinal learner spoken corpus and to share procedures related to the transcription of learners’ utterances from audio files using automated speech recognition (ASR) technology (IBM Watson Speech-to-text). The data of the corpus were collected twice or thrice a year for three consecutive years from 2016, creating eight data collection points altogether. They were gathered from 120 secondary school students who had been learning English in an English as a Foreign Language context for three years. The students were asked to take a monologue speaking test, the Telephone Standard Speaking Test, consisting of various tasks. The overall discussion of the article focuses on the details of this project and highlights how a methodological approach of combining electronic learner language data and ASR technology is useful in constructing learner spoken corpora.