Unlocking Parameter-Efficient Fine-Tuning for Low-Resource Language Translation
This work addresses the challenge of improving translation accuracy for low-resource languages with minimal computational resources, representing an incremental advancement in applying PEFT methods to specific domains.
The paper tackled the problem of varying effectiveness of parameter-efficient fine-tuning (PEFT) methods in low-resource language neural machine translation, showing that 6 out of 8 PEFT architectures outperform baselines in in-domain and out-domain tests, with the Houlsby+Inversion adapter achieving the best performance.
Parameter-efficient fine-tuning (PEFT) methods are increasingly vital in adapting large-scale pre-trained language models for diverse tasks, offering a balance between adaptability and computational efficiency. They are important in Low-Resource Language (LRL) Neural Machine Translation (NMT) to enhance translation accuracy with minimal resources. However, their practical effectiveness varies significantly across different languages. We conducted comprehensive empirical experiments with varying LRL domains and sizes to evaluate the performance of 8 PEFT methods with in total of 15 architectures using the SacreBLEU score. We showed that 6 PEFT architectures outperform the baseline for both in-domain and out-domain tests and the Houlsby+Inversion adapter has the best performance overall, proving the effectiveness of PEFT methods.