Zero-Shot Cross-Lingual Transfer for Indonesian Natural Language Understanding Using Pre-Trained Transformers

Views: 0
Downloads: 0
DOI https://doi.org/10.55923/aksara.v3i1.1
Eko Prasetyo 1 , Arif Saputra 2 , Dian Gunawan 3 , Ahmad Lestari 4
1, 2, 4Department of Civil Engineering, Universitas Sumatera Utara , 3Department of Computer Science, Institut Teknologi Bandung
1 eko.prasetyo@usu.ac.id , 2 arif.saputra@usu.ac.id , 3 dian.gunawan@itb.ac.id , 4 ahmad.lestari@usu.ac.id

Downloads

Resource limitations in Indonesian linguistic training corpora complicate deep learning translation and intent extraction. This paper validates cross-lingual transfer methodologies fine-tuning multilingual mBART and IndoBERT models on downstream classification tasks, yielding substantial F1-score gains on regional dialect datasets.
Issue Vol. 3 No. 1 (2024)
Section Articles
Pages 129-142
Published 2024-06-30