Article: Research on modified Biaffine method for Chinese semantic role labelling Journal: International Journal of Reasoning-based Intelligent Systems (IJRIS) 2026 Vol.18 No.8 pp.33 - 43 Abstract: Semantic role labelling (SRL) is a core technology for semantic analysis. However, SRL methods based on pre-trained language models still face semantic ambiguity and training complexity. This paper proposes a Chinese SRL approach that integrates pre-trained language models with Biaffine technology to better capture semantic information in long sentences while alleviating training difficulty. By further incorporating pooling techniques and part-of-speech (POS) features, the model more accurately identifies semantic role boundaries. Experiments show that the RoBERTa-MPBF-CRF* model based on maximum pooling achieves an F1 score of 90.89% on the Chinese PropBank (CPB) dataset, outperforming CRF-only baselines. The introduction of POS features yields an average F1 improvement of about 1.5%, and the additional computational overhead remains acceptable relative to the performance gains. Inderscience Publishers - linking academia, business and industry through research

Title: Research on modified Biaffine method for Chinese semantic role labelling

Authors: Ning Ma; Jiahao Wang; Youqi Wang

Addresses: Key Laboratory of Linguistic and Cultural Computing, Ministry of Education, Northwest Minzu University, Lanzhou, 730030, China; Key Laboratory of China's Ethnic Languages and Intelligent Processing of Gansu Province, Northwest Minzu University, Lanzhou, 730030, China ' Key Laboratory of Linguistic and Cultural Computing, Ministry of Education, Northwest Minzu University, Lanzhou, 730030, China; Key Laboratory of China's Ethnic Languages and Intelligent Processing of Gansu Province, Northwest Minzu University, Lanzhou, 730030, China ' Key Laboratory of Linguistic and Cultural Computing, Ministry of Education, Northwest Minzu University, Lanzhou, 730030, China; Key Laboratory of China's Ethnic Languages and Intelligent Processing of Gansu Province, Northwest Minzu University, Lanzhou, 730030, China

Abstract: Semantic role labelling (SRL) is a core technology for semantic analysis. However, SRL methods based on pre-trained language models still face semantic ambiguity and training complexity. This paper proposes a Chinese SRL approach that integrates pre-trained language models with Biaffine technology to better capture semantic information in long sentences while alleviating training difficulty. By further incorporating pooling techniques and part-of-speech (POS) features, the model more accurately identifies semantic role boundaries. Experiments show that the RoBERTa-MPBF-CRF* model based on maximum pooling achieves an F1 score of 90.89% on the Chinese PropBank (CPB) dataset, outperforming CRF-only baselines. The introduction of POS features yields an average F1 improvement of about 1.5%, and the additional computational overhead remains acceptable relative to the performance gains.

Keywords: semantic role labelling; SRL; Chinese PropBank; pre-trained language models; Biaffine; pooling techniques; part-of-speech features; POS.

DOI: 10.1504/IJRIS.2026.151727

International Journal of Reasoning-based Intelligent Systems, 2026 Vol.18 No.8, pp.33 - 43

Received: 03 Nov 2025
Accepted: 18 Dec 2025
Published online: 17 Feb 2026 *

Title: Research on modified Biaffine method for Chinese semantic role labelling

Keep up-to-date