Scholarlink

ChatGPT’s release in late 2022 threw generative AI (GenAI) chatbots into the spotlight, significantly disrupting the status quo of numerous fields of study. Translation studies quickly adopted GenAI as researchers compared it with current neural machine translation (NMT) systems and leveraged its multiple potentials. However, despite growing interest, few studies have evaluated and documented GenAI chatbots’ translation errors. To bridge this gap, a comparison of translations of a specialized text using ChatGPT and three traditional neural machine translation (NMT) systems is performed, and errors are documented. Furthermore, by recording the screens of posteditors, an analysis of the translation process is performed to: (1) Determine whether posteditors need more or less effort working on ChatGPT-translated texts and (2) Assess posteditors’ practices during the editing process. The multiple findings have significant practical and pedagogical implications vis-à-vis human–computer interactions during postediting.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Comparing Generative AI and Traditional Machine Translation Systems: An Assessment of Computer Errors and the Human Post-Editing Process

  • 章节

  • 发表日期：

    2025年

查看全文
  • Kizito Tekwa

摘要

ChatGPT’s release in late 2022 threw generative AI (GenAI) chatbots into the spotlight, significantly disrupting the status quo of numerous fields of study. Translation studies quickly adopted GenAI as researchers compared it with current neural machine translation (NMT) systems and leveraged its multiple potentials. However, despite growing interest, few studies have evaluated and documented GenAI chatbots’ translation errors. To bridge this gap, a comparison of translations of a specialized text using ChatGPT and three traditional neural machine translation (NMT) systems is performed, and errors are documented. Furthermore, by recording the screens of posteditors, an analysis of the translation process is performed to: (1) Determine whether posteditors need more or less effort working on ChatGPT-translated texts and (2) Assess posteditors’ practices during the editing process. The multiple findings have significant practical and pedagogical implications vis-à-vis human–computer interactions during postediting.