@article{fdi:010097586,
  title = {{P}rompting large language models for {CEFR}-{EBCL}-aligned chinese {L}2 learning : an empirical study of sinographic constraint compliance},
  author = {{L}in-{Z}ucker, {M}. and {B}ellassen, {J}. and {Z}ucker, {J}ean-{D}aniel},
  editor = {},
  abstract = {{L}arge {L}anguage {M}odels ({LLM}s) are increasingly used in {C}hinese as a {S}econd {L}anguage ({L}2) learning, yet their ability to comply with pedagogical constraints specific to the {C}hinese writing system remains underexplored. {T}his study examines whether system prompts aligned with the {CEFR}-{EBCL} framework enable {LLM}s to generate learner-facing {C}hinese texts that respect sinographic thresholds at the {A}1, {A}1+ and {A}2 levels. {W}e conducted controlled experiments using two models ({GPT}-4.1 and {GPT}-4.1-mini) across ten {EBCL}-related written tasks. {P}rompt conditions with and without explicit character lists were compared. {M}odel outputs were automatically analyzed to quantify instruction deviation, defined as the proportion of characters outside the target {EBCL} set. {R}esults indicate that including explicit character lists significantly reduces out-of-list character production at the {A}1 and {A}1+ levels, particularly with {GPT}-4.1. {A}t the {A}2 level, this effect becomes marginal. {T}hese findings provide empirical evidence on the pedagogical value and limits of prompt-based control of {C}hat{GPT} outputs for {CEFR}-{EBCL}-aligned {C}hinese {L}2 learning.},
  keywords = {{C}hinese as a {S}econd {L}anguage ; {L}arge {L}anguage {M}odels ; {P}rompt {E}ngineering ; {CEFR}-{EBCL} {A}lignment ; {S}inographic {C}onstraints ; {CHINE}},
  booktitle = {},
  journal = {{J}ournal of {T}echnology and {C}hinese {L}anguage {T}eaching},
  volume = {17},
  numero = {1},
  pages = {1--29},
  ISSN = {1949-260{X}},
  year = {2026},
  URL = {https://www.documentation.ird.fr/hor/fdi:010097586},
}