Automatic Proficiency Assessment in L2 English Learners

TL;DR

The paper addresses the need for scalable, objective L2 English proficiency assessment by leveraging deep learning on both speech and transcription, covering structured and spontaneous dialogue. It compares diverse audio models (2D CNN, Frequency-axis CNN, ResNet, wav2vec 2.0) and a text model (BERT) under speaker-independent conditions, plus a dialogue-specific preprocessing workflow. Key findings show wav2vec 2.0 as the strongest performer for speech-based proficiency, with multi-task learning offering gains, while text-only BERT models lag behind audio; longer audio segments provide richer cues, and interviewer speech can enhance performance in audio but noisy content in text. The results underscore the potential of automated, multimodal L2 assessment and point to future work in multimodal fusion and resource-efficient models for real-world dialog evaluation.

Abstract

Second language proficiency (L2) in English is usually perceptually evaluated by English teachers or expert evaluators, with the inherent intra- and inter-rater variability. This paper explores deep learning techniques for comprehensive L2 proficiency assessment, addressing both the speech signal and its correspondent transcription. We analyze spoken proficiency classification prediction using diverse architectures, including 2D CNN, frequency-based CNN, ResNet, and a pretrained wav2vec 2.0 model. Additionally, we examine text-based proficiency assessment by fine-tuning a BERT language model within resource constraints. Finally, we tackle the complex task of spontaneous dialogue assessment, managing long-form audio and speaker interactions through separate applications of wav2vec 2.0 and BERT models. Results from experiments on EFCamDat and ANGLISH datasets and a private dataset highlight the potential of deep learning, especially the pretrained wav2vec 2.0 model, for robust automated L2 proficiency evaluation.