Revisiting Round-Trip Translation with LLMs and Agentic Translation

Round-trip translation (RTT) is a tool for rectifying grammatical errors in text and facilitating automated program repair (APR). When combined with large language models (LLMs), RTT has demonstrated significant efficacy in enhancing APR. We aim to evaluate the performance of LLMs and agentic translation on RTT. We conducted experiments with GPT-4o-mini, finding that LLMs behave well in both contexts and the agentic translation, powerfully on APR through RTT, but lack highlight on natural language RTT. Our findings reveal the exceptional adaptability of LLMs in RTT and the capacity of LLMs and agentic translation to facilitate APR with RTT.