Article: Optimisation of intelligent English grammar error correction based on multi-strategy Pinyin detection and hierarchical enhancement Journal: International Journal of Continuing Engineering Education and Life-Long Learning (IJCEELL) 2026 Vol.36 No.8 pp.21 - 48 Abstract: This study proposes an intelligent grammar correction method integrating multi-strategy Pinyin detection and hierarchical data augmentation to address common errors in Chinese English learners' writing. A dual-strategy Pinyin detection algorithm combines syllable tree matching and linguistic rules to accurately identify and preserve Pinyin segments. A hierarchical data augmentation approach employs rule-based and model-based back-translation to build diverse parallel corpora targeting typical learner errors. Based on the Transformer architecture, the grammar correction model treats error correction as a sequence-to-sequence task. Results show the Pinyin detector achieves 99.95% accuracy, processing 5,386 words/second with 13.02 MB memory usage. The correction model attains a 40.58 F_{0.5} score and 49.56% accuracy on CoNLL-2014. On CLEC subsets, it achieves 87.5%, 90.2%, and 85.9% accuracy for article, subject-verb agreement, and verb tense errors, respectively. Pinyin false corrections dropped from 65% to 1.8%, demonstrating significant improvement in handling Chinese learners' English writing. Inderscience Publishers - linking academia, business and industry through research

Title: Optimisation of intelligent English grammar error correction based on multi-strategy Pinyin detection and hierarchical enhancement

Authors: Lingling Song

Addresses: Open Education Academy, Yantai Vocational College, Yantai, 264003, China

Abstract: This study proposes an intelligent grammar correction method integrating multi-strategy Pinyin detection and hierarchical data augmentation to address common errors in Chinese English learners' writing. A dual-strategy Pinyin detection algorithm combines syllable tree matching and linguistic rules to accurately identify and preserve Pinyin segments. A hierarchical data augmentation approach employs rule-based and model-based back-translation to build diverse parallel corpora targeting typical learner errors. Based on the Transformer architecture, the grammar correction model treats error correction as a sequence-to-sequence task. Results show the Pinyin detector achieves 99.95% accuracy, processing 5,386 words/second with 13.02 MB memory usage. The correction model attains a 40.58 F_{0.5} score and 49.56% accuracy on CoNLL-2014. On CLEC subsets, it achieves 87.5%, 90.2%, and 85.9% accuracy for article, subject-verb agreement, and verb tense errors, respectively. Pinyin false corrections dropped from 65% to 1.8%, demonstrating significant improvement in handling Chinese learners' English writing.

Keywords: grammar error correction; GEC; Pinyin detection; data augmentation; Transformer; English.

DOI: 10.1504/IJCEELL.2026.153060

International Journal of Continuing Engineering Education and Life-Long Learning, 2026 Vol.36 No.8, pp.21 - 48

Received: 13 Nov 2025
Accepted: 29 Jan 2026
Published online: 20 Apr 2026 *

Title: Optimisation of intelligent English grammar error correction based on multi-strategy Pinyin detection and hierarchical enhancement

Keep up-to-date