ChatGPT Self-Correction Outputs between Self-Provided and External Feedback in Translating Medical Texts

Abstract

This study aims to investigate both external and self-provided feedback strategies and their effects on ChatGPT’s self-correction for the purpose of improving its translation performance. The study evaluates the effectiveness of the proposed strategies and points out which strategy is more effective when translating English medical texts into Arabic. The researchers chose 15 English medical texts. These texts were translated into Arabic by ChatGPT using a default translation prompt. The translated texts are manually annotated and evaluated. The researchers, then, retranslate the texts where some of the texts were retranslated using the self-provided feedback and some were retranslated using external feedback. Both feedback strategies are used to exploit ChatGPT self-correction. Manual evaluation of the initial translation and the retranslation are performed to evaluate the effectiveness of the used feedback strategies. The effectiveness of each of these feedback strategies are evaluated based on the improvement rate of the retranslation. The study reaches a conclusion regarding the most efficient feedback strategy, which is error taxonomy feedback strategy. The study is very important for translators, post-editors, researchers, and developers of MT. It also provides recommendations for future work.