Scholarlink

<p>Grammatical Error Correction (GEC) constitutes a critical component for enhancing text quality across a diverse range of Natural Language Processing (NLP) applications. While pre-trained transformer models have achieved notable advancements in GEC, they frequently encounter challenges with complex grammatical structures and long-range dependencies within sentences. This paper presents AdaMHCG-AE, a novel autoencoder architecture incorporating Adaptive Multi-Head Attention with Contextual Gating (AMH-CG). This mechanism is designed to dynamically modulate attention weights based on contextual information, thereby facilitating the effective capture of both local and global dependencies within sentential contexts. Through rigorous experimentation on established benchmark datasets, including JFLEG, Lang-8, CoNLL-2014, FCE, and Write&amp;Improve, AdaMHCG-AE demonstrates superior performance compared to state-of-the-art models such as T5, BART, and PEGASUS. This performance advantage is consistently observed across key evaluation metrics, encompassing accuracy, precision, recall, <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\hbox {F}_{0.5}\)</EquationSource> </InlineEquation>-score, BLEU, and GLEU. These findings underscore the efficacy of the proposed AMH-CG mechanism in significantly enhancing the performance of autoencoder-based GEC systems.</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

A dynamic hybrid attention-based autoencoder model with adaptive contextual attention for grammatical error correction

  • 期刊论文

  • 发表日期：

    2026年1月17日

查看全文
  • Lazhar Farek,
  • Amira Benaidja

摘要

Grammatical Error Correction (GEC) constitutes a critical component for enhancing text quality across a diverse range of Natural Language Processing (NLP) applications. While pre-trained transformer models have achieved notable advancements in GEC, they frequently encounter challenges with complex grammatical structures and long-range dependencies within sentences. This paper presents AdaMHCG-AE, a novel autoencoder architecture incorporating Adaptive Multi-Head Attention with Contextual Gating (AMH-CG). This mechanism is designed to dynamically modulate attention weights based on contextual information, thereby facilitating the effective capture of both local and global dependencies within sentential contexts. Through rigorous experimentation on established benchmark datasets, including JFLEG, Lang-8, CoNLL-2014, FCE, and Write&Improve, AdaMHCG-AE demonstrates superior performance compared to state-of-the-art models such as T5, BART, and PEGASUS. This performance advantage is consistently observed across key evaluation metrics, encompassing accuracy, precision, recall, \(\hbox {F}_{0.5}\) -score, BLEU, and GLEU. These findings underscore the efficacy of the proposed AMH-CG mechanism in significantly enhancing the performance of autoencoder-based GEC systems.