Article: Knowledge graphs meet deep learning for intelligent diagnosis of oral English proficiency Journal: International Journal of Information and Communication Technology (IJICT) 2026 Vol.27 No.3 pp.70 - 89 Abstract: To address the critical challenges of insufficient diagnostic granularity and limited interpretability in spoken English assessment, this research proposes an intelligent framework that synergistically integrating knowledge graph and deep learning technologies. We construct a structured oral knowledge graph using multidimensional error annotations from the Speechocean762 corpus and phoneme-level pronunciation data from L2-Arctic, and design a knowledge graph-enhanced multi-task learning model to achieve cross-dimensional joint optimisation. Experimental results show 12.3% reduction in pronunciation error rate and 14.7% improvement in grammatical diagnostic F1-score compared to mainstream baselines, with overall diagnostic accuracy reaching 86.2%. Ablation studies confirm the knowledge graph's pivotal role in error-path reasoning, while the meta-relation learner significantly enhances few-shot adaptation capability (31.2% F1-score gain). This framework provides interpretable diagnostic support for adaptive language learning systems, reducing error-correction cycles by 40.5% in real-world educational applications. Inderscience Publishers - linking academia, business and industry through research

Title: Knowledge graphs meet deep learning for intelligent diagnosis of oral English proficiency

Authors: Mengyan Li; Wenjing Yang

Addresses: Office of Educational Administration, Chengde College of Applied Technology, Chengde, 067000, China ' Organization of Personnel Division, Chengde College of Applied Technology, Chengde, 067000, China

Abstract: To address the critical challenges of insufficient diagnostic granularity and limited interpretability in spoken English assessment, this research proposes an intelligent framework that synergistically integrating knowledge graph and deep learning technologies. We construct a structured oral knowledge graph using multidimensional error annotations from the Speechocean762 corpus and phoneme-level pronunciation data from L2-Arctic, and design a knowledge graph-enhanced multi-task learning model to achieve cross-dimensional joint optimisation. Experimental results show 12.3% reduction in pronunciation error rate and 14.7% improvement in grammatical diagnostic F1-score compared to mainstream baselines, with overall diagnostic accuracy reaching 86.2%. Ablation studies confirm the knowledge graph's pivotal role in error-path reasoning, while the meta-relation learner significantly enhances few-shot adaptation capability (31.2% F1-score gain). This framework provides interpretable diagnostic support for adaptive language learning systems, reducing error-correction cycles by 40.5% in real-world educational applications.

Keywords: knowledge graph fusion; spoken English diagnosis; multi-task learning; fine-grained error analysis.

DOI: 10.1504/IJICT.2026.151491

International Journal of Information and Communication Technology, 2026 Vol.27 No.3, pp.70 - 89

Received: 25 Oct 2025
Accepted: 25 Nov 2025
Published online: 02 Feb 2026 *

Title: Knowledge graphs meet deep learning for intelligent diagnosis of oral English proficiency

Keep up-to-date