Article: Research on intelligent recognition of English machine translation errors using joint contrastive learning Journal: International Journal of Business Intelligence and Data Mining (IJBIDM) 2026 Vol.28 No.10 pp.120 - 146 Abstract: This study presents an intelligent method for detecting English machine translation errors, combining fuzzy neural networks with joint contrastive learning. Low-rank features are extracted from the corpus via feature decomposition and non-negative subspace analysis, with statistical associations quantified in a rule table. An adaptive DPKM algorithm determines the fuzzy neural network's membership function parameters, dynamically optimised by a pigeon swarm module. The core innovation is a joint contrastive learning mechanism that employs a multi-perspective loss function to refine feature representations, enhancing semantic error discrimination. A multi-label active learning strategy then selects informative samples using uncertainty and diversity criteria. Evaluated on the WMT2021 dataset, the method achieves a recognition accuracy of 92.3% and an F1-score of 0.94, outperforming existing techniques and validating its superiority for translation error detection. Inderscience Publishers - linking academia, business and industry through research

Title: Research on intelligent recognition of English machine translation errors using joint contrastive learning

Authors: Yungang Wang

Addresses: College of English and Literature, Xi'an Fanyi University, Xi'an, Shaanxi, 710105, China; College of Liberal Arts and Communication, Da La Salle University, Dasmariñas, Cavite, 4115, Philippines

Abstract: This study presents an intelligent method for detecting English machine translation errors, combining fuzzy neural networks with joint contrastive learning. Low-rank features are extracted from the corpus via feature decomposition and non-negative subspace analysis, with statistical associations quantified in a rule table. An adaptive DPKM algorithm determines the fuzzy neural network's membership function parameters, dynamically optimised by a pigeon swarm module. The core innovation is a joint contrastive learning mechanism that employs a multi-perspective loss function to refine feature representations, enhancing semantic error discrimination. A multi-label active learning strategy then selects informative samples using uncertainty and diversity criteria. Evaluated on the WMT2021 dataset, the method achieves a recognition accuracy of 92.3% and an F1-score of 0.94, outperforming existing techniques and validating its superiority for translation error detection.

Keywords: English machine translation; joint comparative learning; misidentification; fuzzy neural network; FNN.

DOI: 10.1504/IJBIDM.2026.156188

International Journal of Business Intelligence and Data Mining, 2026 Vol.28 No.10, pp.120 - 146

Received: 23 Mar 2026
Accepted: 02 Jun 2026
Published online: 07 Sep 2026 *

Title: Research on intelligent recognition of English machine translation errors using joint contrastive learning

Keep up-to-date