Plain-language summary
Cross-lingual question-answering systems can struggle when training labels are limited or when the meaning of a question is uncertain. CenterDistill uses weak supervision and teacher–student learning to transfer useful signals while explicitly accounting for ambiguity.
Research contribution
- Weakly supervised training for cross-lingual QA with limited labeled data.
- Knowledge distillation that transfers information from a stronger teacher model to a compact student model.
- Ambiguity-aware reasoning designed to reduce overconfident answers when evidence is uncertain.
Publication status
Published in Engineering Applications of Neural Networks, edited by L. Iliadis et al., in the Communications in Computer and Information Science series, volume 3025. Springer Nature Switzerland, 2026, pages 1–15. DOI: 10.1007/978-3-032-31141-2_11.
Code and dataset
Code and dataset links are not yet public.
Recommended citation
Chakraborty, S., Naskar, S., Paul, S., Jana, A., Chakraborty, N. and Gayen, A. (2026) ‘CenterDistill: Weakly-supervised distillation for ambiguity-aware cross-lingual QA’, in Iliadis, L. et al. (eds.) Engineering Applications of Neural Networks. Communications in Computer and Information Science, vol. 3025. Springer Nature Switzerland, pp. 1–15. doi: 10.1007/978-3-032-31141-2_11.
BibTeX
@inproceedings{chakraborty2026centerdistill,
title={CenterDistill: Weakly-supervised distillation for ambiguity-aware cross-lingual QA},
author={Chakraborty, Somyajit and Naskar, S. and Paul, S. and Jana, A. and Chakraborty, N. and Gayen, A.},
editor={Iliadis, L. and others},
booktitle={Engineering Applications of Neural Networks},
series={Communications in Computer and Information Science},
volume={3025},
pages={1--15},
publisher={Springer Nature Switzerland},
year={2026},
doi={10.1007/978-3-032-31141-2_11}
}