Grammatical Error Correction on NLPCC-18 Chinese GEC (test)
67.33PrecisionTraditional Voting
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Traditional VotingThreshold T=42023.05 | 67.33 | 14.96 | 39.61 | — | |
| Traditional VotingThreshold T=32023.05 | 60.81 | 21 | 44.09 | — | |
| TemplateGECBackbone=BART2024.02 | 54.5 | 27.4 | 45.5 | — | |
| AlirectorModel=Yi-6B2024.02 | 54.23 | 24.89 | 43.88 | — | |
| AlirectorBackbone=Baichuan2-7B2024.02 | 52.27 | 27.14 | 45.01 | — | |
| AlirectorBackbone=BART2024.02 | 51.76 | 33.49 | 46.67 | — | |
| AlirectorModel=ChatGLM3-6B2024.02 | 51.75 | 25.11 | 42.69 | — | |
| Vanilla FTBackbone=Baichuan2-7B2024.02 | 51.69 | 27.92 | 44.17 | — | |
| CopyBackbone=Baichuan2-7B2024.02 | 51.56 | 28.53 | 44.39 | — | |
| CopyBackbone=BART2024.02 | 51.25 | 32.55 | 45.97 | — | |
| Vanilla FTBackbone=BART2024.02 | 50.63 | 31.83 | 45.28 | — | |
| Vanilla FTModel=Yi-6B2024.02 | 50.61 | 25.22 | 42.13 | — | |
| CopyModel=Yi-6B2024.02 | 50.11 | 25.01 | 41.73 | — | |
| SynGECPLM=true, Syntax=true2022.10 | 49.96 | 33.04 | 45.32 | — | |
| SynGECBackbone=BART2024.02 | 49.96 | 33.04 | 45.32 | — | |
| CopyModel=ChatGLM3-6B2024.02 | 49.64 | 25.23 | 41.59 | — | |
| BaselinePLM=true, Syntax=false2022.10 | 49.07 | 32.8 | 44.64 | — | |
| Vanilla FTModel=ChatGLM3-6B2024.02 | 48.61 | 26.46 | 41.64 | — | |
| AlirectorBackbone=Chinese-LLaMA2-7B2024.02 | 47.43 | 26.96 | 41.18 | — | |
| CopyBackbone=Chinese-LLaMA2-7B2024.02 | 46.53 | 27.93 | 41.06 | — | |
| AlirectorBackbone=Transformer2024.02 | 45.98 | 22.87 | 38.25 | — | |
| Vanilla FTBackbone=Chinese-LLaMA2-7B2024.02 | 45.85 | 27.44 | 40.43 | — | |
| seq2seq-1Strategy=Single Model2023.05 | 43.93 | 28.21 | 39.52 | — | |
| CopyBackbone=Transformer2024.02 | 43.16 | 23.58 | 37.01 | — | |
| Zhang et al. (2022)PLM=true, Syntax=false2022.10 | 42.88 | 30.19 | 39.55 | — | |
| Traditional VotingThreshold T=22023.05 | 42.71 | 32.62 | 40.22 | — | |
| Vanilla FTBackbone=Transformer2024.02 | 42.37 | 23.49 | 36.5 | — | |
| TemplateGECBackbone=Transformer2024.02 | 42 | 22.2 | 35.6 | — | |
| SynGECPLM=false, Syntax=true2022.10 | 41.44 | 28.28 | 37.91 | — | |
| SynGECBackbone=Transformer2024.02 | 41.44 | 28.28 | 37.91 | — | |
| seq2seq-2Strategy=Single Model2023.05 | 40.79 | 29.59 | 37.92 | — | |
| BaselinePLM=false, Syntax=false2022.10 | 40.1 | 26.34 | 36.31 | — | |
| Average of 4Strategy=Single Model Ensemble2023.05 | 39.83 | 28.19 | 36.76 | — | |
| seq2edit-1Strategy=Single Model2023.05 | 38.42 | 26.79 | 35.35 | — | |
| Sentence-level EnsemblePLM Backbone=GPT2-Chinese2023.05 | 37.75 | 33.2 | 36.74 | — | |
| Sentence-level EnsemblePLM Backbone=BERT-base-Chinese2023.05 | 37.71 | 22.8 | 33.35 | — | |
| Sentence-level EnsemblePLM Backbone=MacBERT-base-Chinese2023.05 | 37.62 | 31.3 | 36.16 | — | |
| Span Correction2021.06 | 37.3 | 14.5 | 28.4 | 2.7 | |
| Edit-combination EnsemblePLM Backbone=MacBERT-base-Chinese2023.05 | 36.35 | 30.69 | 35.05 | — | |
| Edit-combination EnsemblePLM Backbone=GPT2-Chinese2023.05 | 36.23 | 33.29 | 35.6 | — | |
| seq2edit-2Strategy=Single Model2023.05 | 36.19 | 28.15 | 34.24 | — | |
| Transformer-bigbeam size=52021.06 | 36 | 17.2 | 29.6 | 1 | |
| Edit-level EnsemblePLM Backbone=MacBERT-base-Chinese2023.05 | 35.38 | 28.42 | 33.73 | — | |
| Edit-level EnsemblePLM Backbone=GPT2-Chinese2023.05 | 35.04 | 31.6 | 34.29 | — | |
| Edit-combination EnsemblePLM Backbone=BERT-base-Chinese2023.05 | 34.25 | 21.56 | 30.64 | — | |
| Edit-level EnsemblePLM Backbone=BERT-base-Chinese2023.05 | 33.19 | 20.59 | 29.57 | — | |
| Shallow Aggressive Decodingarchitecture configuration=9 encoder + 3 decoder layers2021.06 | 33 | 20.5 | 29.4 | 12 | |
| LaserTagger2021.06 | 25.6 | 10.5 | 19.9 | 38 | |
| Levenshtein Transformer2021.06 | 24.9 | 15 | 22 | 3.1 |