<![CDATA[The development of Large Language Models (LLMs) and Artificial Intelligence (AI)-based technologies has increased the demand for multilingual chatbots for AI ethics education. However, language differences between chatbot training data and user language remain a challenge that can affect the interaction quality.Although machine translation has been widely used to support multilingual chatbots, studies comparing the impact of different translation models on translation quality, particularly in the domain of AI ethics, remain scarce. This study aims to compare and select the best machine translation model in the field of artificial intelligence ethics. The dataset was obtained from UNESCO’s Recommendation on the Ethics of Artificial Intelligence document and generated using a Retrieval-Augmented Generation (RAG) approach based on LLMs. The dataset consisted of 1,000 English-language questions that were later translated into Indonesian using an LLM and manually validated. The Indonesian-language dataset was used as input for back-translation into English using several machine translation methods, namely Google Translate, MarianMT, and M2M-100. The evaluation was conducted using BLEU and METEOR metrics. The results indicate that Google Translate achieved the highest performance, with a BLEU score of 52.2% and a METEOR score of 81.7%, whereas the lowest performance was observed in MarianMT Multi-EN, with a BLEU score of 18.95% and a METEOR score of 56.19%. The findings also indicate that increasing the number of parameters in the M2M-100 model improved the translation quality. This study demonstrates that machine translation has significant potential for supporting multilingual chatbots, particularly in the field of AI ethics.]]>
