Adversarial Grammatical Error Generation: Application to Persian Language

Grammatical error correction (GEC) greatly benefits from large quantities of high-quality training data. However, the preparation a amount labelled data is time-consuming and prone to human errors. These issues have become major obstacles in GEC systems. Recently, performance English systems has drastically been enhanced by application deep neural networks that generate synthetic limited samples. While extensively studied languages such as Chinese, no attempts made for improving Persian Given substantial grammatical semantic differences language, this paper, we propose new learning framework create enough sentences are grammatically incorrect A modified version sequence generative adversarial net with policy gradient developed, which size model scaled down hyperparameters tuned. The generator trained an on dataset 8000 Our proposed achieved bilingual evaluation understudy (BLEU) scores 64.5% BLEU-2, 44.2% BLEU-3, 21.4% BLEU-4, outperformed conventional supervised-trained long short-term memory using maximum likelihood estimation recently labeler machine translation augmentation. This shows promise toward generating