Student-AI Creative Writing: Pedagogical Strategies for Applying Natural Language Generation in Schools

AI natural language generation (NLG) is a process where computer systems generate human-comprehensible language texts from information. It can become an integral part of a human's creative writing process. Importantly, youths can learn to apply NLG in mainstream education and become better prepared for AI-enhanced writing jobs and other writing endeavors. To explore how students apply NLG to creative writing, we designed and implemented the 1st Human-AI Creative Writing Contest in a Hong Kong secondary school. In this contest, each student participant wrote a short story of up to 500-words using the student's own words and words generated by a computer and built on open-source language models. We designed four text generators for the contest as the computer's text entry. Additionally, using design-based research, we developed seven workshops where students learned to write with the four text generators and answered reflection questions. In analyzing four students' short stories and adjudicators' scores for the stories, we found different strategies in terms of the number and the type of text generator words that students used. Some strategies appeared more sophisticated than others. In analyzing students' reflections, we found students could describe text generator input and output as units of thought. Besides, students showed preferences for text generators; and they expressed a range of feelings when writing with text generators. The findings provide design implications not only for NLG applications in formal schooling but also suggest pedagogical strategies for AI curriculum.