Scholarlink

Writing historical fiction presents unique challenges in research-inspired plot design, historically-relevant language use, and contextually consistent character descriptions. Current generative AI tools provide ways to facilitate the historical fiction writing process, but we know little about how these tools affect different task workflows of professional writers in the writing process. Our study engaged historical fiction writers in three independent tasks supported by ChatGPT: character design, novel outlines, and checking for consistency. We found that ChatGPT often gives suggestions that skews the writers’ intent, but works well for suggesting plot outlines and filling in descriptive details for determined narratives. Writers overcome inaccuracy challenges by using active step-by-step prompting to set boundaries for ChatGPT output, or passive regeneration approaches. We show a need for context-based, genre-language-specific active adjustment of outputs tailored for different writing tasks. This work shows the multiple-task intricacies of the creative writing process and how they can be differentially supported by generative AI.

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

Once More with (the Right) Feeling: How Historical Fiction Writing Processes of Character Design, Plot Outline, and Context Checking Are Affected by Co-Writing with ChatGPT

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Yun Chen,
  • Yiwei Wang,
  • Antoni B. Chan,
  • Jixing Li,
  • Ray LC

摘要

Writing historical fiction presents unique challenges in research-inspired plot design, historically-relevant language use, and contextually consistent character descriptions. Current generative AI tools provide ways to facilitate the historical fiction writing process, but we know little about how these tools affect different task workflows of professional writers in the writing process. Our study engaged historical fiction writers in three independent tasks supported by ChatGPT: character design, novel outlines, and checking for consistency. We found that ChatGPT often gives suggestions that skews the writers’ intent, but works well for suggesting plot outlines and filling in descriptive details for determined narratives. Writers overcome inaccuracy challenges by using active step-by-step prompting to set boundaries for ChatGPT output, or passive regeneration approaches. We show a need for context-based, genre-language-specific active adjustment of outputs tailored for different writing tasks. This work shows the multiple-task intricacies of the creative writing process and how they can be differentially supported by generative AI.