TasTe: Teaching Large Language Models to Translate through Self-Reflection

Large language models (LLMs) have exhibited remarkable performance in various natural language processing tasks. Techniques like instruction tuning have effectively enhanced the proficiency of LLMs in the downstream task of machine translation. However, the existing approaches fail to yield satisfactory translation outputs that match the quality of supervised neural machine translation (NMT) systems. One plausible explanation for this discrepancy is that the straightforward prompts employed in these methodologies are not able to fully leverage the acquired instruction-following capabilities. To this end, we propose the TASTE framework, which stands for translating through selfreflection. The self-reflection process includes two stages of inference. In the first stage, LLMs are instructed to generate preliminary translations and conduct self-assessments on these translations simultaneously. In the second stage, LLMs are tasked to refine these preliminary translations according to the assessment results. The evaluation results across four language directions on the WMT22 benchmark reveal the effectiveness of our approach when compared to the existing methods. Our work presents a promising approach to unleash the potential of LLMs and enhance their capabilities in machine translation. 044 Numerous studies are underway to unlock 045 the vast potential of machine translation within 046 LLMs. Prompt engineering aims to design effective 047 prompt templates to guide LLMs in accomplishing 048 specific language tasks. Some approaches attempt 049 to integrate supplementary information pertinent to 050 the translation task to enhance the performance of 051