One-Topic-Doesn't-Fit-All: Transcreating Reading Comprehension Test for Personalized Learning

Jieun Han, Daniel Lee, Haneul Yoo, Jinsung Yoon, Junyeong Park, Suin Kim, So-Yeon Ahn, Alice OharXiv 2025 · DOI · arXiv

Introduction

In English as a Foreign Language (EFL) education, student engagement is a critical factor for success. Traditional instructional materials often rely on generic or Western-centric topics that fail to resonate with diverse student populations, leading to diminished motivation and lower reading comprehension scores. While personalized learning has long been considered a "holy grail" in education, the manual creation of high-quality, tailored reading materials is prohibitively expensive and difficult to scale.

This research addresses the challenge of transcreation: the process of adapting content across different contexts (e.g., from a generic topic to one aligned with a student's interest) while strictly preserving the original meaning, linguistic complexity, and pedagogical intent of the assessment.

Methodology: The Transcreation Pipeline

The researchers developed a structured five-step pipeline leveraging OpenAI’s gpt-4o to transform source reading materials into personalized versions. This process ensures that while the topic changes, the educational value remains constant.

The LLM-driven transcreation pipeline: from interest collection to personalized test generation.

The pipeline consists of the following stages:

  1. Topic Extraction: Identifying the core subject of a source passage based on standardized curricula.
  2. Question Analysis: Classifying existing questions using Bloom’s Taxonomy (e.g., Remember, Understand, Analyze) to maintain cognitive demand.
  3. Linguistic Feature Extraction: Identifying 41 specific grammatical structures and linguistic markers defined by educational standards.
  4. Reading Passage Transcreation: Rewriting the passage into the student’s area of interest (e.g., shifting from "senior citizen discounts" to "online education") while embedding the extracted linguistic features.
  5. Question Transcreation: Adapting multiple-choice questions and distractors to fit the new topic while maintaining the original question type and difficulty level.

Key Insights

The study highlights several critical concepts in personalized AI education:

  • Transcreation vs. Generation: By using a source test as a structural anchor rather than generating from scratch, the pipeline ensures the resulting materials function as valid assessment tools.
  • The "Carry-Over" Effect: Exposure to personalized content was found to have a lasting impact. Motivation built during personalized sessions helped students stay engaged even when they transitioned back to non-personalized, "uninteresting" topics.
  • Affective Engagement: Personalization focuses on the affective (emotional/interest) domain. By making the content interesting, students are more willing to exert the mental stamina required for high-level analytical reasoning.

Experimental Results

The researchers conducted a controlled experiment with 20 college-level EFL learners in South Korea, comparing a control group (Group A, random topics) with an experimental group (Group B, personalized topics).

Comparison of test performance between Group A (control) and Group B (personalized) across two test phases.

The results demonstrated significant benefits for the personalized group:

  • Performance Gains: Group B’s scores increased from an average of 73.33 to 86.67 when using personalized materials, whereas Group A showed no significant change.
  • Analytical Improvement: Group B showed a marked improvement in "Analyze"-level questions, suggesting that interest facilitates deeper cognitive processing.
  • Increased Efficiency: Students in the personalized group completed tests faster (reducing time from ~31 to ~28 minutes), while the control group's completion time actually increased.
  • Sustained Motivation: Using the Instructional Materials Motivation Survey (IMMS), the authors confirmed that Group B maintained higher confidence and engagement levels throughout the study.

Limitations and Future Directions

While promising, the study identified areas for improvement. GPT-4o occasionally struggled with complex temporal reasoning, requiring human validation. Additionally, broad interest categories (e.g., "Sports") can sometimes lead to misalignment if the specific sub-topic (e.g., history vs. gameplay) does not match the student's preference. Future work will focus on refining interest capture and expanding the study to larger, more diverse student cohorts.

Summary updated

Abstract

Personalized learning has gained attention in English as a Foreign Language (EFL) education, where engagement and motivation play crucial roles in reading comprehension. We propose a novel approach to generating personalized English reading comprehension tests tailored to students' interests. We develop a structured content transcreation pipeline using OpenAI's gpt-4o, where we start with the RACE-C dataset, and generate new passages and multiple-choice reading comprehension questions that are linguistically similar to the original passages but semantically aligned with individual learners' interests. Our methodology integrates topic extraction, question classification based on Bloom's taxonomy, linguistic feature analysis, and content transcreation to enhance student engagement. We conduct a controlled experiment with EFL learners in South Korea to examine the impact of interest-aligned reading materials on comprehension and motivation. Our results show students learning with personalized reading passages demonstrate improved comprehension and motivation retention compared to those learning with non-personalized materials.
BibTeX
@misc{han2025one,
  author = {Jieun Han and Daniel Lee and Haneul Yoo and Jinsung Yoon and Junyeong Park and Suin Kim and So-Yeon Ahn and Alice Oh},
  title = {{One-Topic-Doesn't-Fit-All: Transcreating Reading Comprehension Test for Personalized Learning}},
  year = {2025},
  eprint = {2511.09135},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2511.09135},
}
Figure