Steering Large Language Models for Machine Translation with Finetuning and In-Context Learning
This addresses the problem of making LLM-based machine translation more robust and efficient for practitioners, though it is incremental as it builds on existing finetuning and in-context learning techniques.
The paper tackled the brittleness of LLM-based machine translation systems by showing that adapter-based finetuning with LoRA matches traditional finetuning performance with 50x fewer parameters, outperforms few-shot prompting, and eliminates post-processing needs, while also proposing a method to recover few-shot capabilities during finetuning.
Large language models (LLMs) are a promising avenue for machine translation (MT). However, current LLM-based MT systems are brittle: their effectiveness highly depends on the choice of few-shot examples and they often require extra post-processing due to overgeneration. Alternatives such as finetuning on translation instructions are computationally expensive and may weaken in-context learning capabilities, due to overspecialization. In this paper, we provide a closer look at this problem. We start by showing that adapter-based finetuning with LoRA matches the performance of traditional finetuning while reducing the number of training parameters by a factor of 50. This method also outperforms few-shot prompting and eliminates the need for post-processing or in-context examples. However, we show that finetuning generally degrades few-shot performance, hindering adaptation capabilities. Finally, to obtain the best of both worlds, we propose a simple approach that incorporates few-shot examples during finetuning. Experiments on 10 language pairs show that our proposed approach recovers the original few-shot capabilities while keeping the added benefits of finetuning.