Developing a Multilingual Spontaneous L2 Speech Corpus for Automated Proficiency Assessment

Abstract

Currently, most accessible multilingual L2 spontaneous speech corpora primarily include L2 English from speakers of various L1 backgrounds, with few and often small-scale corpora available for non-English L2s. Annotated assessment data from expert raters is especially rare. This study addresses this gap by constructing a large-scale dataset of spontaneous L2 speech in seven languages (Chinese, Japanese, English, French, German, Spanish, and Russian) from Korean L1 speakers, accompanied by detailed assessments using a carefully designed rubric. The dataset includes extensive metadata analysis and validation processes to ensure the reliability of subjective assessments. To our knowledge, this is the first large-scale, comprehensive corpus of its kind, featuring diverse L2 spontaneous speech from Korean speakers with expert-annotated assessments.