How to Design Translation Prompts for ChatGPT: An Empirical Study

• DOI: 10.48550/arxiv.2304.02182
• Publication Date: 2023-01-01
• Authors: Yuan Gao; Ruili Wang; Feng Hou
• Keywords: FOS: Computer and information sciences; Computer Science - Computation and Language; Computation and Language (cs.CL)

The recently released ChatGPT has demonstrated surprising abilities in natural language understanding and generation. Machine translation relies heavily on the of Thus, this paper, we explore how to assist machine with ChatGPT. We adopt several prompts a wide range translations. Our experimental results show that designed can achieve comparable or better performance over commercial systems for high-resource further evaluate quality using multiple references, achieves superior compared systems. also conduct experiments domain-specific translations, final is able comprehend provided domain keyword adjust accordingly output proper At last, perform few-shot consistent improvement across different base prompts. work provides empirical evidence still great potential