LEOPARD: A Structured LLM Orchestration Framework for Precision-Centric Grammatical Error Correction
Open accessThe DOI matches its Crossref record; 14 of the 14 references it cites were found in a registry or an index.
Abstract
The application of Large Language Models (LLMs) to Grammatical Error Correction (GEC) is fundamental to Computer-Assisted Language Learning (CALL), where the primary goal is to provide non-native learners with grammatically faithful edits. However, LLMs often fail as pedagogical tools due to a "Fluency Paradox": rather than identifying and fixing specific errors, they prioritize stylistic regeneration over minimal intervention. This tendency leads models to perform a wholesale restructuring of student attempts—transforming valid but non-native phrasing into professional L1-grade English. Such over-correction is pedagogically counterproductive; it conflates stylistic preference with grammatical error, erases the learner’s original intent, and obscures the specific proficiency gaps required for targeted instruction. Despite the rapid advancement of LLMs, this tension between creative fluency and diagnostic precision remains largely unaddressed in current GEC literature. In this work, we introduce LEOPARD (LLM Orchestration for Precision-Aware Repair & Diagnosis), a multi-agent framework designed to enforce precision in GEC. LEOPARD employs a deterministic bottleneck architecture where a high-creativity "Boss Agent" proposes corrections, while a precision-focused "Purist Agent" acts as a grammatical guardrail, vetoing unnecessary stylistic edits to preserve the learner’s original voice. We demonstrate that this architecture improves the F0.5score by 14.6% over a zero-shot GPT-4 baseline on the CLC-FCE benchmark. Our findings suggest that decoupling generative capability from verification logic is essential for recovering the precision needed to deploy trustworthy and helpful AI tutors.
Overview
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
Citations
Cites
The Role of Consciousness in Second Language Learning1
SCHMIDT, R. W. (1990) · Applied Linguistics · doi:10.1093/applin/11.2.129 · Cited by 5,435
Classification assessment methods
Tharwat, A. (2021) · Applied Computing and Informatics · doi:10.1016/j.aci.2018.08.003 · Cited by 2,417
Self-Refine: Iterative Refinement with Self-Feedback
Madaan, A., Tandon, N., Gupta, P., Hallinan, S., Gao, L., Wiegreffe, S., Alon, U., Dziri, N., Prabhumoye, S., Yang, Y., et al. (2023) · arXiv · doi:10.48550/arxiv.2303.17651 · Cited by 2,314
Reflexion: Language Agents with Verbal Reinforcement Learning
Shinn, N., Cassano, F., Berman, E., Gopinath, A., Narasimhan, K., & Yao, S. (2023) · arXiv · doi:10.48550/arxiv.2303.11366 · Cited by 1,087 · cited 1×
“Similarly, Shinn et al. [11] proposed “Reflexion,” using verbal reinforcement learning to guide agents.”
ReAct: Synergizing Reasoning and Acting in Language Models
Yao, S., Zhao, J., Yu, D., Du, N., Shafran, I., Narasimhan, K., & Cao, Y. (2022) · arXiv · doi:10.48550/arxiv.2210.03629 · Cited by 608
The CoNLL-2014 Shared Task on Grammatical Error Correction
Ng, H. T., Wu, S. M., Briscoe, T., Hadiwinoto, C., Susanto, R. H., & Bryant, C. (2014) · Proceedings of the Eighteenth Conference on Computational Natural Language Learning: Shared Task · doi:10.3115/v1/w14-1701 · Cited by 577
A New Dataset and Method for Automatically Grading ESOL Texts
Yannakoudakis, H., Briscoe, T., & Medlock, B. (2011) · Cited by 482
Is ChatGPT a Highly Fluent Grammatical Error Correction System? A Comprehensive EvaluationPreprint
Fang, T., Yang, S., Lan, K., Wong, D. F., Hu, J., Chao, L. S., & Zhang, Y. (2023) · arXiv (Cornell University) · doi:10.48550/arxiv.2304.01746 · Cited by 250
Analyzing the Performance of GPT-3.5 and GPT-4 in Grammatical Error CorrectionPreprint
Coyne, S., Sakaguchi, K., Galván-Sosa, D., Zock, M., & Inui, K. (2023) · arXiv (Cornell University) · doi:10.48550/arxiv.2303.14342 · Cited by 131
Automatic Annotation and Evaluation of Error Types for Grammatical Error Correction
Bryant, C., Felice, M., & Briscoe, T. (2017) · Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · doi:10.18653/v1/p17-1074 · Cited by 100
Cited by
Still being checked.
Metadata
- Authors
- Ashrafi, N. S., Mediani, M., & Bouktif, S.
- Affiliations
- United Arab Emirates University (AE)
- Venue
- 2026 International Conference on Frontiers of Engineering and Emerging Technologies (FET)
- Publisher
- IEEE
- Published
- 2026-04-22
- Pages
- 1-6
- OpenAlex
- W7169574080
- Cited by
- 0
- References
- 14
- FWCI
- 0.00
- Type
- conference paper
- Language
- en
- Open access
- hybrid
- doi.org
- Locations
- International Conference on Frontiers of Engineering and Emerging Technologies (FET) (publishedVersion, cc-by)
- Funders
- United Arab Emirates University
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Natural Language Processing Techniques
- Social Sciences › Arts and Humanities › Language and Linguistics › EFL/ESL Teaching and Learning
- Physical Sciences › Computer Science › Artificial Intelligence › Speech and dialogue systems
- Keywords
- grammatical error correction, large language models, multi-agent systems, LLM orchestration, overcorrection, computer-assisted language learning, non-native English learners, AI tutors
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.
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
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: LEOPARD: A Structured LLM Orchestration Framework for Precision-Centric Grammatical Error Correction. https://app.cochat.ai/papers/10.1109/fet68771.2026.11601896
- The paper
- Ashrafi, N. S., Mediani, M., & Bouktif, S. (2026). LEOPARD: A Structured LLM Orchestration Framework for Precision-Centric Grammatical Error Correction. 2026 International Conference on Frontiers of Engineering and Emerging Technologies (FET). https://doi.org/10.1109/fet68771.2026.11601896
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.