Scholarlink

<p>In recent years, robot-assisted learning systems have emerged as promising solutions for enhancing interactive language education. This research introduces a robot-assisted Japanese Conversation Training and Intelligent Feedback System designed to support pronunciation, fluency evaluation, grammatical correction, and contextual appropriateness in Japanese conversational practice. Unlike traditional computer-assisted approaches, the proposed system employs an Efficient Coyote Optimized Bidirectional Encoder mutated Long Short-Term Memory (ECO-BiE-LSTM) model to automate speech recognition and fluency analysis. Additionally, BERT supports language understanding, grammar correction, and contextual responses, enabling natural and adaptive conversation training. The data is evaluated using a publicly available robot-assisted Japanese conversation dataset comprising 5000 annotated conversational samples created for robot-assisted language learning applications. The dataset includes paired speech-text data with labels for pronunciation quality, fluency scores, grammatical correctness, conversational context, and sentiment. Pre-processing techniques such as Japanese tokenization and min-max normalization are applied to ensure data consistency and effective feature learning. The system dynamically analysis simulated Japanese dialogues inputs by evaluating pronunciation, grammar, fluency, and pragmatic suitability, and generates multimodal feedback delivered verbally by a robot and visually through on-screen prompts. Sentiment-aware feedback mechanism enables adaptive interaction strategies based on emotional cues detected from speech. Experimental evaluations conducted in Python demonstrate that the ECO-BiE-LSTM framework achieves high computational efficiency with GFLOPs (10.28), and an average inference time (5.32 ms), ensuring responsiveness suitable for robotic interaction. The results indicate that speech recognition, conversation language modeling, and adaptive feedback support robot-assisted Japanese conversation training systems without relying on human-subject experimentation.</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

Robot-assisted Japanese conversation training and intelligent feedback system

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年3月4日

查看全文
  • Tingting Zhang

摘要

In recent years, robot-assisted learning systems have emerged as promising solutions for enhancing interactive language education. This research introduces a robot-assisted Japanese Conversation Training and Intelligent Feedback System designed to support pronunciation, fluency evaluation, grammatical correction, and contextual appropriateness in Japanese conversational practice. Unlike traditional computer-assisted approaches, the proposed system employs an Efficient Coyote Optimized Bidirectional Encoder mutated Long Short-Term Memory (ECO-BiE-LSTM) model to automate speech recognition and fluency analysis. Additionally, BERT supports language understanding, grammar correction, and contextual responses, enabling natural and adaptive conversation training. The data is evaluated using a publicly available robot-assisted Japanese conversation dataset comprising 5000 annotated conversational samples created for robot-assisted language learning applications. The dataset includes paired speech-text data with labels for pronunciation quality, fluency scores, grammatical correctness, conversational context, and sentiment. Pre-processing techniques such as Japanese tokenization and min-max normalization are applied to ensure data consistency and effective feature learning. The system dynamically analysis simulated Japanese dialogues inputs by evaluating pronunciation, grammar, fluency, and pragmatic suitability, and generates multimodal feedback delivered verbally by a robot and visually through on-screen prompts. Sentiment-aware feedback mechanism enables adaptive interaction strategies based on emotional cues detected from speech. Experimental evaluations conducted in Python demonstrate that the ECO-BiE-LSTM framework achieves high computational efficiency with GFLOPs (10.28), and an average inference time (5.32 ms), ensuring responsiveness suitable for robotic interaction. The results indicate that speech recognition, conversation language modeling, and adaptive feedback support robot-assisted Japanese conversation training systems without relying on human-subject experimentation.