A Scenario-Generic Neural Machine Translation Data Augmentation Method

Amid the rapid advancement of neural machine translation, challenge data sparsity has been a major obstacle. To address this issue, study proposes general augmentation technique for various scenarios. It examines predicament parallel corpora diversity and high quality in both rich- low-resource settings, integrates low-frequency word substitution method reverse translation approach complementary benefits. Additionally, improves pseudo-parallel corpus generated by substituting words includes grammar error correction module to reduce grammatical errors The experimental are partitioned into scenarios at 10:1 ratio. verifies necessity pseudo-corpus Models methods chosen from backbone network related literature comparative experiments. findings demonstrate that proposed is suitable effective enhancing training improve performance tasks.