---
language: 
- en

tags:
- gec

library_name: opennmt
license: mit
metrics:
- bleu

inference: false
---

### Introduction

This repository contains a description on how to use OpenNMT on the Grammar Error Correction (GEC) task. The idea is to approch GEC as a translation task

### Usage

Install the necessary dependencies:


```bash
pip3 install ctranslate2 pyonmttok
```


Simple tokenization & translation using Python:


```python
import ctranslate2
import pyonmttok
from huggingface_hub import snapshot_download
model_dir = snapshot_download(repo_id="jordimas/gec-opennmt-english", revision="main")

tokenizer=pyonmttok.Tokenizer(mode="none", sp_model_path = model_dir + "/sp_m.model")
tokenized=tokenizer.tokenize("The water are hot. My friends are going to be late. Today mine mother is in Barcelona.")

translator = ctranslate2.Translator(model_dir)
translated = translator.translate_batch([tokenized[0]])
print(tokenizer.detokenize(translated[0][0]['tokens']))
```

# Model

The model has been training using the [clang8](https://github.com/google-research-datasets/clang8) corpus for English language.

Details:
  * Model: TransformerBase
  * Tokenizer: SentencePiece
  * BLEU = 85.50

# Papers

Relevant papers:

* [Approaching Neural Grammatical Error Correction as a Low-Resource Machine Translation Task](https://aclanthology.org/N18-1055.pdf)
* [A Simple Recipe for Multilingual Grammatical Error Correction](https://arxiv.org/pdf/2106.03830.pdf)


# Contact

Email address: Jordi Mas: jmas@softcatala.org