This model has been trained on 80% of the COWS-L2H dataset and 80,984 SYNTHETICALLY-GENERATED errorful sentences for grammatical error correction of Spanish text. The corpus was sentencized, so the model has been fine-tuned for SENTENCE CORRECTION. This model will likely not perform well on an entire paragraph. To correct a paragraph, sentencize the text and run the model for each sentence.
The synthetic data was generated based on a rule-based algorithm from well-formed Spanish sentences. The code for synthetic generaton is available in the Github repo for this project: https://github.com/SkitCon/synth_gec_es
BLEU: 0.851 on COWS-L2H
Example usage:
from transformers import AutoTokenizer, BartForConditionalGeneration
tokenizer = AutoTokenizer.from_pretrained("SkitCon/gec-spanish-BARTO-SYNTHETIC")
model = BartForConditionalGeneration.from_pretrained("SkitCon/gec-spanish-BARTO-SYNTHETIC")
input_sentences = ["Yo va al tienda.", "Espero que tú ganas."]
tokenized_text = tokenizer(input_sentences, max_length=128, padding="max_length", truncation=True, return_tensors="pt")
input_ids = tokenized_text["input_ids"].squeeze()
attention_mask = tokenized_text["attention_mask"].squeeze()
outputs = model.generate(input_ids=input_ids, attention_mask=attention_mask)
for sentence in tokenizer.batch_decode(outputs, skip_special_tokens=True):
print(sentence)