Near Human-Level Performance in Grammatical Error Correction with Hybrid Machine Translation
Roman Grundkiewicz, Marcin Junczys-Dowmunt
Abstract
We combine two of the most popular approaches to automated Grammatical Error Correction (GEC): GEC based on Statistical Machine Translation (SMT) and GEC based on Neural Machine Translation (NMT). The hybrid system achieves new state-of-the-art results on the CoNLL-2014 and JFLEG benchmarks. This GEC system preserves the accuracy of SMT output and, at the same time, generates more fluent sentences as it typical for NMT. Our analysis shows that the created systems are closer to reaching human-level performance than any other GEC system reported so far.
Results
| Task | Dataset | Metric | Value | Model |
|---|---|---|---|---|
| Grammatical Error Correction | CoNLL-2014 Shared Task | F0.5 | 56.25 | SMT + BiGRU |
| Grammatical Error Correction | JFLEG | GLEU | 61.5 | SMT + BiGRU |
| Grammatical Error Correction | CoNLL-2014 Shared Task (10 annotations) | F0.5 | 72.04 | SMT + BiGRU |
Related Papers
A Translation of Probabilistic Event Calculus into Markov Decision Processes2025-07-17Function-to-Style Guidance of LLMs for Code Translation2025-07-15Speak2Sign3D: A Multi-modal Pipeline for English Speech to American Sign Language Animation2025-07-09Pun Intended: Multi-Agent Translation of Wordplay with Contrastive Learning and Phonetic-Semantic Embeddings2025-07-09Unconditional Diffusion for Generative Sequential Recommendation2025-07-08GRAFT: A Graph-based Flow-aware Agentic Framework for Document-level Machine Translation2025-07-04TransLaw: Benchmarking Large Language Models in Multi-Agent Simulation of the Collaborative Translation2025-07-01CycleVAR: Repurposing Autoregressive Model for Unsupervised One-Step Image Translation2025-06-29