Investigation on Transformer-Based Translation From Mandarin to Hakka
This study explores Mandarin to Hakka translation using Transformer models, specifically addressing the challenges of dialectal variations and limited data. We developed specialized models tailored to Hakka's linguistic characteristics and expanded the Hakka corpus. Using the mT5-small model, we fine-tuned it with layer normalization adjustments and gradient clipping to better handle Hakka's unique linguistic features.
Authors
- Shih‐Cheng Horng (ORCID: https://orcid.org/0000-0001-7707-2481)
- Qiangfu Zhao (ORCID: https://orcid.org/0000-0003-3101-749X)
- Chenxi Li (ORCID: https://orcid.org/0009-0001-8465-9915)
Institutions
- Chaoyang University of Technology (TW)
- University of Aizu (JP)
Publication Details
- Published
- 2024-09-20
- DOI
- https://doi.org/10.1109/icmlc63072.2024.10934999
- Primary Topic
- Natural Language Processing Techniques
- Type
- article
- Field-Weighted Citation Impact
- 0.00