Article: Intelligent assessment method of Japanese Kana writing trajectory based on ConvLSTM and CRF Journal: International Journal of Information and Communication Technology (IJICT) 2026 Vol.27 No.51 pp.36 - 52 Abstract: Japanese kana writing is fundamental to learning the Japanese language, and its standardisation has a significant impact on language learning outcomes. To address the inefficiency and subjectivity of traditional manual evaluation, this study proposes an intelligent evaluation model that integrates a convolutional long short-term memory (ConvLSTM) network with a conditional random field (CRF). First, the model utilises the ConvLSTM to efficiently extract spatiotemporal features of handwriting traces. Second, the CRF layer optimises sequence annotation to achieve automatic quantitative evaluation of kana writing accuracy, fluency, and structural standardisation. Finally, a self-constructed dataset containing 2,000 handwriting trace samples from five common hiragana and five katakana categories was used for evaluation experiments. The results show that the model achieved a 98.2% accuracy rate in kana character recognition, a Pearson correlation coefficient of 0.91 between its writing style score and expert evaluations, and a 91.2% accuracy rate in kana stroke regularity assessment, significantly outperforming the single LSTM and CNN-CRF models. Inderscience Publishers - linking academia, business and industry through research

Title: Intelligent assessment method of Japanese Kana writing trajectory based on ConvLSTM and CRF

Authors: Zheng Cao

Addresses: School of Foreign Languages, Shanghai Zhongqiao Vocational and Technical University, Shanghai, 200000, China

Abstract: Japanese kana writing is fundamental to learning the Japanese language, and its standardisation has a significant impact on language learning outcomes. To address the inefficiency and subjectivity of traditional manual evaluation, this study proposes an intelligent evaluation model that integrates a convolutional long short-term memory (ConvLSTM) network with a conditional random field (CRF). First, the model utilises the ConvLSTM to efficiently extract spatiotemporal features of handwriting traces. Second, the CRF layer optimises sequence annotation to achieve automatic quantitative evaluation of kana writing accuracy, fluency, and structural standardisation. Finally, a self-constructed dataset containing 2,000 handwriting trace samples from five common hiragana and five katakana categories was used for evaluation experiments. The results show that the model achieved a 98.2% accuracy rate in kana character recognition, a Pearson correlation coefficient of 0.91 between its writing style score and expert evaluations, and a 91.2% accuracy rate in kana stroke regularity assessment, significantly outperforming the single LSTM and CNN-CRF models.

Keywords: writing trajectory evaluation; ConvLSTM; CRF; Japanese kana; intelligent evaluation; sequence labelling.

DOI: 10.1504/IJICT.2026.153703

International Journal of Information and Communication Technology, 2026 Vol.27 No.51, pp.36 - 52

Received: 27 Oct 2025
Accepted: 16 Dec 2025
Published online: 21 May 2026 *

Title: Intelligent assessment method of Japanese Kana writing trajectory based on ConvLSTM and CRF

Keep up-to-date