Scholarlink

<p>Oral corrective feedback in English as a foreign language (EFL) teaching is often constrained by delayed feedback, inconsistent error judgement, and limited adaptation to learners’ interactional and sociocultural backgrounds. To avoid treating oral errors only as acoustic or grammatical deviations, this study develops an EFL oral error correction model that integrates selected dimensions of sociocultural theory with deep learning. The integration is specifically operationalized through three dimensions: peer–teacher scaffolding in corrective interaction, first-language-based cultural script transfer, and learner affective response during feedback. These dimensions are encoded as interaction weights, transfer-interference factors, and feedback-adjustment parameters within an “input–processing–output” collaborative system. The model combines multimodal feature representation, a two-stream Transformer for speech–text fusion, a Flow-based generative correction module, and a dynamic fluency intervention strategy based on complexity, accuracy, and fluency indicators. Experiments were organized around three research objectives: detecting oral errors, examining sociocultural transfer-related error patterns, and evaluating adaptive fluency intervention. Results on the L2-Arctic-based speech error analysis showed that the misreading detection F1 score reached 67.4%, representing an improvement over the baseline model. The Flow-based module reduced mel-spectrogram distortion to 4.8 dB, suggesting improved speech reconstruction quality while preserving speaker-related acoustic features. In task-based fluency intervention, reinforcement-learning-driven pause prediction was associated with a 29.8% increase in WPM in the shopping scenario. The findings suggest that incorporating explicitly defined sociocultural parameters may improve the interpretability and adaptability of EFL oral error correction systems. However, the observed age-, gender-, and L1-related patterns are treated as descriptive tendencies rather than causal conclusions, and further validation with larger and more balanced learner samples is required.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

An EFL oral error correction model integrating sociocultural theory into English teaching: a study on learners’ interlanguage development tracking and fluency intervention based on deep learning

  • 期刊论文

  • 发表日期：

    2026年7月7日

查看全文
  • Qingdi Si

摘要

Oral corrective feedback in English as a foreign language (EFL) teaching is often constrained by delayed feedback, inconsistent error judgement, and limited adaptation to learners’ interactional and sociocultural backgrounds. To avoid treating oral errors only as acoustic or grammatical deviations, this study develops an EFL oral error correction model that integrates selected dimensions of sociocultural theory with deep learning. The integration is specifically operationalized through three dimensions: peer–teacher scaffolding in corrective interaction, first-language-based cultural script transfer, and learner affective response during feedback. These dimensions are encoded as interaction weights, transfer-interference factors, and feedback-adjustment parameters within an “input–processing–output” collaborative system. The model combines multimodal feature representation, a two-stream Transformer for speech–text fusion, a Flow-based generative correction module, and a dynamic fluency intervention strategy based on complexity, accuracy, and fluency indicators. Experiments were organized around three research objectives: detecting oral errors, examining sociocultural transfer-related error patterns, and evaluating adaptive fluency intervention. Results on the L2-Arctic-based speech error analysis showed that the misreading detection F1 score reached 67.4%, representing an improvement over the baseline model. The Flow-based module reduced mel-spectrogram distortion to 4.8 dB, suggesting improved speech reconstruction quality while preserving speaker-related acoustic features. In task-based fluency intervention, reinforcement-learning-driven pause prediction was associated with a 29.8% increase in WPM in the shopping scenario. The findings suggest that incorporating explicitly defined sociocultural parameters may improve the interpretability and adaptability of EFL oral error correction systems. However, the observed age-, gender-, and L1-related patterns are treated as descriptive tendencies rather than causal conclusions, and further validation with larger and more balanced learner samples is required.