CoChat
Sign Up
Cited by 44EFL/ESL Teaching and LearningArtificial Intelligence in EducationEnglish Language Learning and Teaching

Understanding English as a foreign language students’ idea generation strategies for creative writing with natural language generation tools

Open access
Woo, David James, Wang, Yanzhi, Susanto, Hengky, Guo, Kai·2023·Journal of Educational Computing Research
DOI 10.1177/07356331231175999

Published version of a preprint

Open this paper in CoChat — free account

Abstract

Natural language generation (NLG) is a process within artificial intelligence where computer systems produce human-comprehensible language texts from information. English as a foreign language (EFL) students’ use of NLG tools might facilitate their idea generation, which is fundamental to creative writing. However, little is known about how EFL students interact with NLG tools to generate ideas. This study explores strategies adopted by EFL students when searching for ideas using NLG tools, evaluating ideas generated by NLG tools and selecting NLG tools for idea generation. Four Hong Kong secondary school students attended workshops where they learned to write stories comprising their own words and words generated by NLG tools. After the workshops, they answered questions to reflect on their writing experience with NLG tools. In a thematic analysis of the written reflections, we found students may have existing ideas when searching for ideas and evaluating ideas with NLG tools. Students showed some aversion to ideas generated by NLG tools and selected NLG tools that generated a greater quantity of ideas. The findings inform our understanding of EFL students’ concerns when using NLG tools for idea generation and can inform educators’ instruction to implement NLG tools for classroom creative writing.

Verified 2026-10-09 · 2 references not checked yet

AI snapshot

Written from the paper’s full text.

TL;DR

This exploratory study analyzed written reflections from four Hong Kong EFL secondary students who used four prompt-based NLG tools during creative writing workshops. It identifies strategies for searching, evaluating, and selecting NLG tools, highlighting language-feature focus, idea aversion, and preference for tools producing more ideas.

Key findings

  • Thematic analysis revealed one main theme for searching for ideas, two for evaluating ideas, and one for selecting NLG tools. Students may have existing ideas when searching for and evaluating ideas, but their search descriptions focused on language features such as words or sentences rather than discourse features like story elements.
  • Evaluation involved “sensing a schemata,” where students tried to place AI-generated ideas into existing story ideas or elements, and “idea aversion,” where some students appeared more averse to novel or transformative AI ideas than others. For example, Student H preferred the sentence generator to stay “on the same page,” while Student W valued the next sentence generator because it “provides a wide variety of ideas.”
  • Students selected NLG tools that generated a greater quantity of ideas. No one preferred the Next Word Generator; Student S selected Next Paragraph Generators 1 and 2 to “pick out the best sentences,” and Student M used the sentence generator because “the words I will get in return will increase.”
  • The study collected 12 reflection instances from four participants (Student H: 3, Student M: 4, Student S: 2, Student W: 3).

Methods

  • Exploratory qualitative study at an all-girls secondary school in Hong Kong. Four students (ages 12–17) wrote English short stories of no more than 500 words using their own words and words generated by NLG tools, during seven voluntary hour-long workshops from March 1 to April 5, 2022; stories were submitted April 7, 2022.
  • Data included 12 written reflections from the four participants, plus workshop audio/video recordings and student screenshots. Students used four NLG tools developed on Hugging Face: Next Sentence Generator, Next Word Generator, Next Paragraph Generator 1, and Next Paragraph Generator 2.
  • Analysis used inductive thematic analysis (Braun & Clarke, 2006), coding for strategies of searching for ideas, evaluating ideas, and selecting NLG tools. The first author initial-coded, and the second and third authors reviewed the coding scheme.

Limitations

  • Stated: the study is exploratory with a small sample of four students, so findings are not statistical generalizations but theoretical contributions; themes are not exhaustive. The authors call for larger, more diverse samples, more rigorous instrumentation, and data such as screen recordings and audio narration rather than relying largely on introspective written reflections.
  • Stated: the NLG tools used open-source language models that were state-of-the-art but not the largest or most sophisticated available; the models were neither fine-tuned nor had their default parameters changed.
  • Evident: the sample was drawn from one all-girls, Cantonese-medium secondary school, recruited voluntarily through English-language advertisements, which may further limit diversity and generalizability beyond the four participants.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1177/07356331231175999

Cites

44 of 46 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

0 of 8 cited sources checked · 0 supported · 0 flagged

  • Findingown data
    § Tools Abstract

    EFL students showed some aversion to ideas generated by NLG tools.

    “Students showed some aversion to ideas generated by NLG tools and selected NLG tools that generated a greater quantity of ideas.”
    Quote in the paperSubstantial claim
  • Findingown data
    § Tools Abstract

    EFL students may have existing ideas when evaluating ideas with NLG tools.

    “In a thematic analysis of the written reflections, we found students may have existing ideas when searching for ideas and evaluating ideas with NLG tools.”
    Quote in the paperSubstantial claim
  • Findingown data
    § Tools Abstract

    EFL students may have existing ideas when searching for ideas with NLG tools.

    “In a thematic analysis of the written reflections, we found students may have existing ideas when searching for ideas and evaluating ideas with NLG tools.”
    Quote in the paperSubstantial claim
  • Findingown data
    § Tools Abstract

    EFL students selected NLG tools that generated a greater quantity of ideas.

    “Students showed some aversion to ideas generated by NLG tools and selected NLG tools that generated a greater quantity of ideas.”
    Quote in the paperSubstantial claim
  • Findingown data
    § Strategies For Evaluating Ideas Sensing a schemata.

    A common approach to evaluating ideas from NLG tools was trying to place them in existing ideas or elements of a story.

    “A common approach to evaluating ideas from NLG tools was trying to place them in existing ideas or elements of a story.”
    Quote in the paperNot a substantial claim
  • May combine two claims. 10
  • Classified as probably not a substantial claim. 15

  • May combine two claims.

    Findingown data
    § Findings

    Thematic analysis revealed one main theme for students' searching for ideas, two for evaluation of ideas, and one for selection of NLG tools.

    “Our thematic analysis revealed one main theme for students' searching for ideas, two for evaluation of ideas and one for selection of NLG tools.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data
    § Strategies For Searching for Ideas A mostly blank slate.

    Students described their search for ideas in terms of language features, but did not refer to discourse features such as elements of a short story.

    “This is because students described their search for ideas in terms of language features, but did not refer to discourse features, for instance, elements of a short story.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data
    § Strategies For Searching for Ideas A mostly blank slate.

    Students may have had ideas when prompting an NLG tool, but their search for story ideas appeared underdeveloped.

    “The first theme shows that although students may have had ideas when prompting an NLG tool, their search for story ideas appeared underdeveloped.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data
    § Strategies For Selecting NLG Tools Increasing the quantity of ideas.

    Students chose particular NLG tools because those tools produced a greater quantity of ideas than other tools, and selected NLG tools that generated a greater number of ideas which students could evaluate.

    “This theme shows students chose particular NLG tools because those tools produced a greater quantity of ideas than other tools.”
    • Also stated in: § Major Findings · hedged
      “Finally, we found students selected NLG tools that generated a greater number of ideas which students could evaluate.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning
    § Natural Language Generation Tools

    To effectively collaborate with prompt-based NLG tools for idea generation, students need to be taught strategies.

    “However, to effectively collaborate with prompt-based NLG tools for idea generation, it appears students will need to be taught strategies.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S4 (based).

    Interpretationnone given
    § Writers' Strategies For Using NLG Tools

    Prior studies on strategies of highly educated, experienced writers may not be applicable to EFL students as novice writers.

    “Although these studies can provide insight into ideas generation strategies with NLG tools, they may not be applicable to EFL students as novice writers.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: EFL students are novice writers.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductreasoning
    § Data Collection

    The six reflection questions were aligned with one or more of the research questions.

    “Since we sought students' strategies from their perspective, we formulated six reflection questions for students to answer, aligning each reflection question with one or more of the research questions (See Table 2).”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S4 (based). May combine two claims.

    Interpretationreasoning
    § Conceptualizing Student-AI Idea Generation

    To achieve effective collaboration with NLG tools and successful idea generation, students need to develop and use strategies for crafting prompts and evaluating ideas from NLG tools.

    “To achieve effective collaboration with NLG tools and successful idea generation, students need to develop and use strategies for crafting prompts and evaluating ideas from NLG tools.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductown data
    § EFL STUDENTS' IDEA GENERATION STRATEGIES

    The study collected audio and video recordings of workshops to triangulate students' written answers to reflection questions with what they shared verbally.

    “In addition, we collected the audio and video recordings of the workshops so as to triangulate students' written answers to reflection questions with what students shared verbally during workshop reflection time.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Premisecitation
    § Introduction

    The content of a prompt greatly impacts an NLG tool's ability to generate text.

    “Moreover, the content of a prompt greatly impacts this type of NLG tool's ability to generate text so that, for instance, changing the length of a prompt, changing key words in a prompt, or changing the order of words can impact the quality of text generated by the NLG tool.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites White et al., 2023

      Not checked in the free run.

  • Classified as probably not a substantial claim. May combine two claims.

    Interpretationnone given
    § Writers' Strategies For Using NLG Tools

    Writers require strategies to effectively search for ideas using NLG tools and to evaluate ideas generated from NLG tools.

    “Writers require strategies to effectively search for ideas using NLG tools and to evaluate ideas generated from NLG tools.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Interpretationreasoning
    § Creative Writing

    The integration of NLG tools presents a potential solution to the difficulty of finding a suitable collaborator for idea generation.

    “The integration of NLG tools presents a potential solution to this issue.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductown data
    § Data Analysis

    The first author initial-coded and the second and third authors reviewed the coding scheme, discussing to resolve disparities, propose new codes, and revise the coding scheme.

    “In this coding process, the first author initial-coded and the second and third authors reviewed the coding scheme, discussing with the first author to resolve coding disparities, to propose new codes and to revise the coding scheme.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation
    § Introduction

    A person needs to craft a prompt and then evaluate the quality of the text generated from the NLG tool before using the text in their own writing.

    “In this way, a person needs to craft a prompt and then evaluate the quality of the text generated from the NLG tool before using the text in their own writing (Liu et al., 2021).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Liu et al., 2021

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationreasoning
    § Limitations and Future Directions

    The themes found in this study are not exhaustive of all possible themes for idea generation.

    “Thus, the themes found in this study are not exhaustive of all possible themes for idea generation.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationreasoning
    § Limitations and Future Directions

    Another limitation of the study is the technical features of the NLG tools.

    “Another limitation is the technical features of the NLG tools.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: Large language models are superior to the open-source language models used in this study at present.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductcitation
    § Data Analysis

    The inductive coding approach enhanced the validity and relevance of the findings.

    “we designed a coding scheme using an inductive approach, which enabled us to remain open to the data and to identify themes that were truly reflective of the students' answers, thereby enhancing the validity and relevance of our findings (Saldaña, 2012).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Saldaña, 2012

      Not checked in the free run.

  • Classified as probably not a substantial claim. May combine two claims.

    Interpretationreasoning
    § Limitations and Future Directions

    The study is exploratory and its sample of four students is small, so its findings are not statistical generalizations but theoretical contributions to understanding EFL students' strategies in using NLG tools for creative writing.

    “The study is exploratory in nature and its sample of four students is small so that its findings are not statistical generalizations but theoretical contributions to our understanding of EFL students' strategies in using NLG tools for creative writing.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Premisecitation
    § Introduction

    Crafting an effective prompt that leads to a desired text generated by an NLG tool is not a straightforward process, but one of trial and error.

    “However, a challenge for nontechnical users is that crafting an effective prompt that leads to a desired text generated by an NLG tool is not a straightforward process, but one of trial and error (Dang et al., 2022).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Dang et al., 2022

      Not checked in the free run.

  • May combine two claims.

    Interpretationown data
    § Implications

    EFL students have language concerns when searching for ideas with NLG tools and evaluating ideas from NLG tools.

    “Our study shows that EFL students have language concerns when searching for ideas with NLG tools and evaluating ideas from NLG tools; and they are concerned with the quantity of ideas generated by NLG tools.”
    Quote in the paperSubstantial claim
20 of 20

  • The paper relies on this without support.

    Little is known about how EFL students interact with NLG tools to generate ideas.

    “However, little is known about how EFL students interact with NLG tools to generate ideas.”

  • The paper relies on this without support.

    EFL students are novice writers.

    “they may not be applicable to EFL students as novice writers.”

  • The paper relies on this without support.

    Schools have rarely deployed NLG tools for classroom writing activities.

    “However, to date, schools have rarely deployed NLG tools for classroom writing activities.”

  • The paper relies on this without support.

    Large language models are superior to the open-source language models used in this study at present.

    “Although the open-source language models used in this study's NLG tools are state-of-the art, large language models are superior at present and we foresee advances in performance for open-source language models.”
4 of 4

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

  • The open-ended and dialogic nature of the NLG tools encouraged creative risk taking, particularly among students who previously expressed anxiety about storytelling in English.

    “The open-ended and dialogic nature of these tools encouraged creative risk taking, particularly among students who previously expressed anxiety about storytelling in English.”

    From AI-Integrated Scaffolding to Enhance Agency and Creativity in K-12 English Language Learners: A Systematic Review (2025) · cites this paper as [36]

    The Conceptualizing Student-AI Idea Generation section describes students repeatedly prompting NLG tools and evaluating the ideas they produce. In Idea aversion, one student says unexpected words helped make her writing more creative, while another says trying potentially odd suggestions was “still worth a try.” These passages offer limited support for creative experimentation, but the paper does not establish that the tools encouraged risk taking, particularly among students who had previously expressed anxiety about storytelling in English. (LLM, conf=0.84)

    Experimental — accuracy not yet measured

  • Woo et al. found a high degree of experimentation, with students blending human-authored and AI-generated text to enhance plot coherence, character development, and stylistic depth.

    “The findings revealed a high degree of experimentation, with students blending humanauthored and AI-generated text to enhance plot coherence, character development, and stylistic depth.”

    From AI-Integrated Scaffolding to Enhance Agency and Creativity in K-12 English Language Learners: A Systematic Review (2025) · cites this paper as [36]

    The Methods section says students wrote stories using their own words and NLG-generated words, and Major Findings says they evaluated AI ideas against existing ideas or story elements. The paper does not report a high degree of experimentation or show that this blending enhanced plot coherence, character development, or stylistic depth. The indexed passages provide section labels but no page numbers. (LLM, conf=0.84)

    Experimental — accuracy not yet measured

  • By reducing cognitive load in areas such as grammar correction and basic organization, AI tools free up mental resources for higher-order creative tasks.

    “Studies by Song and Song [7] and Woo et al. [36] reveal that, by reducing cognitive load in areas such as grammar correction and basic organization, AI tools free up mental resources for higherorder creative tasks.”

    From AI-Integrated Scaffolding to Enhance Agency and Creativity in K-12 English Language Learners: A Systematic Review (2025) · cites this paper as [36]

    The Introduction (page unavailable) says NLG tools might reduce EFL students’ cognitive barriers when writing and might benefit their creative writing. It does not say they do so by handling grammar correction or basic organization, nor does it establish that this frees mental resources for higher-order creative tasks. (LLM, conf=0.82)

    Experimental — accuracy not yet measured

  • Woo et al. [36] and Kim et al. [10] emphasize the importance of designing writing tasks that require students to engage critically with AI-generated suggestions rather than accepting them passively.

    “Woo et al. [36] and Kim et al. [10] emphasize the importance of designing writing tasks that require students to engage critically with AI-generated suggestions rather than accepting them passively.”

    From AI-Integrated Scaffolding to Enhance Agency and Creativity in K-12 English Language Learners: A Systematic Review (2025) · cites this paper as Woo et al. [36]

  • Woo et al.'s investigation of AI-generated text revealed comparable benefits for both high- and low-performing students in terms of content development and organizational clarity.

    “This finding was corroborated by Woo et al. [36], whose investigation of AI-generated text revealed comparable benefits for both high-and low-performing students in terms of content development and organizational clarity.”

    From AI-Integrated Scaffolding to Enhance Agency and Creativity in K-12 English Language Learners: A Systematic Review (2025) · cites this paper as [36]

    The Major Findings section reports a thematic analysis of four students’ strategies for searching for and evaluating AI-generated ideas, not a comparison of writing outcomes. The Limitations and Future Directions section says the small study does not support statistical generalizations. Searches for high- versus low-performing students, content development, and organizational clarity found no passage reporting comparable benefits. (LLM, conf=0.76)

    Experimental — accuracy not yet measured

Cites

Flagged first, then by citations

Research works printed but not linked: 9

  1. Biderman S., 2022, Neural language models are effective plagiarists
  2. Graham S., 2007, A report to carnegie corporation of New York
  3. Hugging Face, What is text generation?—Hugging face
  4. Lin M. P.-C., 2020, Journal of Educational Technology and Society, 23, 78
  5. OpenAI, 2022, ChatGPT: Optimizing language models for dialogue
  6. Radford A., 2019, Language models are unsupervised multitask learners
  7. Rini T. A., Proceedings of the 2nd Early Childhood and Primary Childhood Education, 51
  8. UNESCO, 2022, K-12 AI curricula: A mapping of government-endorsed AI curricula—UNESCO digital library
  9. Yang D., Zhou Y., Zhang Z., Li T. J.J. (2022). AI as an active writer: Interaction strategies with generated text in human-AI collaborative fiction writing. Joint Proceedings of the ACM IUI Workshops, Helsinki, Finland, 2022, 10. CEUR-WS Team.

Cited by

Sorted by citations

Related

Authors
Woo, D. J., Wang, Y., Susanto, H., & Guo, K.
Affiliations
  • Precious Blood Secondary School, Hong Kong, China
  • New York University (US)
  • Education University of Hong Kong (HK)
  • University of Hong Kong (HK)
Corresponding author
Woo, David James
ORCID
Venue
Journal of Educational Computing Research
Publisher
SAGE Publications
Published
2023-05-16
Volume
61
Issue
7
Pages
1464-1482
ISSN
  • 0735-6331 (ISSN-L)
  • 1541-4140
DOI
10.1177/07356331231175999
OpenAlex
W4376636280
Semantic Scholar
3f35e2b369cd975224675016cf44db0b450c0231
Cited by
44
References
43
Citations by year
2026: 11 · 2025: 21 · 2024: 9 · 2023: 3
FWCI
49.26
Type
journal article
Language
en
Open access
Licence
Locations
  • Journal of Educational Computing Research (publishedVersion)
  • arXiv (Cornell University) (submittedVersion)
  • EdArXiv (OSF Preprints) (submittedVersion, cc-by-sa)
  • EdArXiv (OSF Preprints) (submittedVersion)
  • arXiv (Cornell University) (cc-by)
Topics
  • Social Sciences › Arts and Humanities › Language and Linguistics › EFL/ESL Teaching and Learning
  • Physical Sciences › Computer Science › Information Systems › Artificial Intelligence in Education
  • Physical Sciences › Computer Science › Information Systems › English Language Learning and Teaching
Keywords
natural language generation, idea generation, creative writing, English as a foreign language, EFL learners, student strategies, idea evaluation, tool selection, high school students, story writing, student reflection, AI-assisted writing

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-09.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-09.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review