mlsanityData to 7 Oct 2026

CEC-Zero: Zero-Supervision Character Error Correction with Self-Generated Rewards

arXiv 2512.23971 ↗First posted 30 Dec 2025cs.CLcited by 10

On the map: Synthetic data and self-improvement for LLMs → Self-rewarding alignment and self-training

Abstract (arXiv)

Large-scale Chinese spelling correction (CSC) remains critical for real-world text processing, yet existing LLMs and supervised methods lack robustness to novel errors and rely on costly annotations. We introduce CEC-Zero, a zero-supervision reinforcement learning framework that addresses this by enabling LLMs to correct their own mistakes. CEC-Zero synthesizes errorful inputs from clean text, computes cluster-consensus rewards via semantic similarity and candidate agreement, and optimizes the policy with PPO. It outperforms supervised baselines by 10--13 F1 points and strong LLM fine-tunes by 5--8 points across 9 benchmarks, with theoretical guarantees of unbiased rewards and convergence. CEC-Zero establishes a label-free paradigm for robust, scalable CSC, unlocking LLM potential in noisy text pipelines.

Key facts. Cited by 10 papers in the map (data through 7 Oct 2026). Placed in Self-rewarding alignment and self-training, one of 13 directions in Synthetic data and self-improvement for LLMs. Builds on 19 of its 50 references that are in the map. First posted on arXiv 30 Dec 2025.

Builds on

19 of 50 references.

Closest ancestors · 1

Same direction, newest first

  1. 2024Self-Evolved Reward Learning for LLMs1 Nov

Other references · 18

  1. 2017Proximal Policy Optimization Algorithmscited by 14,441
  2. 2025TTRL: Test-Time Reinforcement Learningcited by 253
  3. 2025Self-Evolving Curriculum for LLM Reasoningcited by 107
  4. 2024RL on Incorrect Synthetic Data Scales the Efficiency of LLM Math Reasoning by Eight-Foldcited by 99
  5. 2020Spelling Error Correction with Soft-Masked BERTcited by 90
  6. 2024RLHF Deciphered: A Critical Analysis of Reinforcement Learning from Human Feedback for LLMscited by 89
  7. 2023NDC-Scene: Boost Monocular 3D Semantic Scene Completion in Normalized Device Coordinates Spacecited by 88
  8. 2024On Designing Effective RL Reward at Training Time for LLM Reasoningcited by 87
  9. 2025Temporal Working Memory: Query-Guided Segment Refinement for Enhanced Multimodal Understandingcited by 28
  10. 2023FT2TF: First-Person Statement Text-To-Talking Face Generationcited by 26
  11. 2024Swift Sampler: Efficient Learning of Sampler by 10 Parameterscited by 25
  12. 2024Prompt-Based Length Controlled Generation with Multiple Control Typescited by 19
6 more
  1. 2025Learning Musical Representations for Music Performance Question Answeringcited by 17
  2. 2023Chinese Spelling Correction as Rephrasing Language Modelcited by 15
  3. 2025Reinforcement Learning vs. Distillation: Understanding Accuracy and Capability in LLM Reasoningcited by 14
  4. 2022CSCD-NS: a Chinese Spelling Check Dataset for Native Speakerscited by 13
  5. 2025Does Bigger Mean Better? Comparitive Analysis of CNNs and Biomedical Vision Language Modles in Medical Diagnosiscited by 9
  6. 2025Rainbow Noise: Stress-Testing Multimodal Harmful-Meme Detectors on LGBTQ Contentcited by 6

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 10.

10 papers.

  1. 2026EEO-TFV: Escape-Explore Optimizer for Web-Scale Time-Series Forecasting and Vision Analysiscited by 9
  2. 2026One Refiner to Unlock Them All: Inference-Time Reasoning Elicitation via Reinforcement Query Refinementcited by 6
  3. 2026Time-TK: A Multi-Offset Temporal Interaction Framework Combining Transformer and Kolmogorov-Arnold Networks for Time Series Forecastingcited by 2
  4. 2026CoT2-Meta: Budgeted Metacognitive Control for Test-Time Reasoningcited by 2
  5. 2026Sharpness-Aware Minimization for Generalized Embedding Learning in Federated Recommendationcited by 0
  6. 2026Taming the Long Tail: Efficient Item-wise Sharpness-Aware Minimization for LLM-based Recommender Systemscited by 0
  7. 2026From Solver Feedback to Faithful Plans: Multi-Role Reinforcement Learning for Symbolic Planningcited by 0
  8. 2026AsyncCouple-Flow: Asynchronous Cross-Modal Coupling and Flow Matching for Spatio-Temporal Forecastingcited by 0
  9. 2026CoRe: Coherence and Relational Alignment for Multivariate Time Series Forecastingcited by 0
  10. 2026VAA-CSEC: Vote-guided Advantage Allocation for Chinese Semantic Error Correctioncited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025CEC-Zero: Chinese Error Correction Solution Based on LLM
  2. 2026CSRP: Chain-of-Thought Reasoning for Chinese Text Correction via Reinforcement Learning with Efficiency-Aware Rewards
  3. 2023Rethinking Masked Language Modeling for Chinese Spelling Correction
  4. 2025Chinese Spelling Correction: A Comprehensive Survey of Progress, Challenges, and Opportunities
  5. 2025RAIR: Retrieval-Augmented Iterative Refinement for Chinese Spelling Correction
  6. 2023An Error-Guided Correction Model for Chinese Spelling Error Correction
  7. 2025Harnessing Rule-Based Reinforcement Learning for Enhanced Grammatical Error Correction
  8. 2023RSpell: Retrieval-augmented Framework for Domain Adaptive Chinese Spelling Check