Multi-Task Transformer with Paradigm-Aware Embedding for Large-Scale Second German Inflectional Error Detection and Correction

Yingying  He

doi:10.6180/jase.202608_31.035

Multi-Task Transformer with Paradigm-Aware Embedding for Large-Scale Second German Inflectional Error Detection and Correction

Computer Science and Information Engineering

Yingying HeThis email address is being protected from spambots. You need JavaScript enabled to view it.

School of Foreign Languages and International Trade, Hubei University of Automotive Technology, 442002 China

Received: December 18, 2025
Accepted: February 7, 2026
Publication Date: February 26, 2026

Copyright The Author(s). This is an open access article distributed under the terms of the Creative Commons Attribution License (CC BY 4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are cited.

Download Citation: ||https://doi.org/10.6180/jase.202608_31.035

German poses significant challenges for second language (L2) learners in mastering inflectional rules, leading to frequent inflectional errors in written production. Existing approaches for second German inflectional error detection and correction (IEDC) suffer from three key limitations. (1) Separating modeling of detection and correction tasks leads to information loss and suboptimal performance. (2) Insufficient utilization of German inflectional paradigms contains inherent morphological regularity. (3) It has poor generalization on large-scale and diverse corpora due to limited representation learning. To address these issues, this paper proposes a multi-task Transformer with paradigm-aware embedding (MT-PAE), an unified framework for end-to-end second German IEDC. Specifically, we first design a paradigm-aware embedding (PAE) module that encodes inflectional paradigm knowledge into contextualized word representations. Second, we adopt a multi-task learning paradigm that jointly optimizes error detection (as a binary classification task) and error correction (as a sequence generation task) via a shared Transformer encoder and task-specific decoders. We construct a large scale second German inflectional error corpus (GL2IEC) containing 1.2 million sentences with 420,000 manually annotated inflectional errors, covering three major error types and eight sub-types. Extensive experiments on GL2IEC and two public benchmarks (GLC and DE-learner) show that MT-PAE outperforms state-of-the-art (SOTA) baselines by 3.2 −7.8% in F1-score for detection and 2.5−6.3% in BLEU score for correction. This work provides a new perspective for leveraging morphological paradigm knowledge in error correction tasks and offers a high-performance solution for large-scale second German IEDC.

Keywords: Second German; Inflectional Error Detection and Correction; Multi-Task Transformer; Paradigm-Aware Embedding; Morphological Knowledge

[1] P. Auer,(2025)“Dialect(non-)acquisition and use by young people of migrant background in Germany "Journal of Multilingual and Multicultural Development 46(4):954–968.DOI:10.1080/01434632.2020.1722680.
[2] C. Burisch, A. Bellary, F. Breuckmann, J. Ehlers, S. C. Thal, T. Sellmann, and D.Gödde,(2025)“ChatGPT 4 Performance on German Continuing Medical Education—Friendor Foe(TrickorTreat)? Protocol for a Randomized Controlled Trial "JMIR Research Protocols 14(1):e63887.DOI:10.2196/63887.
[3] J. König, R. Ligtvoet, S. Klemenz, and M. Rothl and, (2025)“Discontinued knowledge growth: on the development of teachers’ general pedagogical knowledge at the transition from higher education into teaching practice " Teachers and Teaching31(5):781–799.DOI:10.1080/13540602.2024.2308895.
[4] N. Dünkel, M. Knigge, and H. Brandt,(2025)“Language exposure within peer and family contexts and bilingual reading profiles of German–Russian and German Turkish adolescents in Germany "International Journal of Bilingualism 29(1):128–146. DOI:10.1177/13670069231221103.
[5] P. Auer and U. Røyneland,(2025)“Modelling acquisition and use of dialectal, standard and multiethnolectal features in migratory contexts across Europe "Journal of MultilingualandMulticulturalDevelopment46(4): 1035–1042.DOI:10.1080/01434632.2020.1730385.
[6] F.Woltran,S.Hassani,andS.Schwab,(2025)“German language support in Austria: feasibility, effectiveness and legitimacy from the perspective of public administration" Language and education 39(4): 1021–1038. DOI: 10.1080/09500782.2024.2371403.
[7] D. N. Jakobi, T. Kern, D. R. Reich, P. Haller, and L. A. Jäger, (2025) “PoTeC: A German naturalistic eye-tracking-while-reading corpus" Behavior Research Methods57(8): 1–37. DOI: 10.3758/s13428-024-02536-8.
[8] R.J. George, C. Eisenach, U. Ghai, D. Perrault-Joncas, A. Anandkumar, and D. Foster, (2025) “BRIDGE: Building Representations In Domain Guided Program Verification" arXiv preprint arXiv:2511.21104: DOI: 10.48550/arXiv.2511.21104.
[9] X. Yu, C. Wang, W. Wu,andS.Xiong, (2025) “A real time skeleton-based fall detection algorithm based on temporal convolutional networks and transformer encoder" Pervasive and Mobile Computing 107: 102016. DOI: 10.1016/j.pmcj.2025.102016.
[10] J. Yu, L. Zhao, S. Yin, and M. Ivanovi´c, (2024) “News recommendation model based on encoder graph neural net work and bat optimization in online social multimedia art education" Computer Science and Information Systems 21(3): 989–1012. DOI: 10.2298/CSIS231225025Y.
[11] S. N. Mehta, A. Roth, C. Munteanu, and S. Chandna. “AI-Based Pronunciation Assessment and Grammatical Error Correction with Feedback for the German Language”. In: International Conference on Human Computer Interaction. Springer. 2025, 388–407. DOI: 10.1007/978-3-031-93415-5_23.
[12] R. Sennrich, B. Haddow, and A. Birch. “Improving neural machine translation models with monolingual data”. In: Proceedings of the 54th annual meeting of the association for computational linguistics (volume 1: long papers). 2016, 86–96. DOI: 10.48550/arXiv.1511.06709.
[13] L. Qiu, C.-S. Wu, W. Liu, and C. Xiong, (2022) “Structure extraction in task-oriented dialogues with slot clustering" arXiv preprint arXiv:2203.00073: DOI: 10.48550/arXiv.2203.00073.
[14] M. Kowsher, A. A. Sami, N. J. Prottasha, M. S. Are f in, P. K. Dhar, and T. Koshiba, (2022) “Bangla-bert: transformer-based efficient model for transfer learning and language understanding" IEEE Access 10: 91855–91870. DOI: 10.1109/ACCESS.2022.3197662.
[15] P. Gupta and S. S. Jamwal, (2025) “Enhancing NLP for Low-Resource Language by Developing Deep Learning Powered Morphological Analysis of Dogri: An End-to End Pipeline from Corpus Construction and Linguistic Annotation to Model Training and Deployment" SN Computer Science 6(8): 928. DOI: 10.1007/s42979 025-04429-9.
[16] E. Coghill, (2025) “Borrowed adjectival inflection in the North-Eastern Neo-Aramaic dialects" Journal of Semitic Studies: fgae045. DOI: 10.1093/jss/fgae045.
[17] A. K. Ramarao, K. Tang, and D. Baer-Henney. “Frequency matters: Modeling irregular morphological patterns in Spanish with Transformers”. In: Findings of the Association for Computational Linguistics: ACL2025. 2025, 4474–4489. DOI: 10.18653/v1/2025.findings-acl.230.
[18] S. Yin, L. Wang, T. Chen, H. Huang, J. Gao, J. Zhang, M. Liu, P. Li, and C. Xu, (2025) “LKAFormer: A lightweight kolmogorov-arnold transformer model for im age semantic segmentation" ACM Transactions on In telligent Systems and Technology: DOI: 10.1145/ 3759254.