Robust Data Augmentation for Neural Machine Translation through EVALNET

Since building Neural Machine Translation (NMT) systems requires a large parallel corpus, various data augmentation techniques have been adopted, especially for low-resource languages. In order to achieve the best performance through augmentation, NMT should be able evaluate quality of augmented data. Several studies addressed weighting assess quality. The basic idea adopted in previous is loss value that system calculates when learning from training weight derived data, simple heuristic rules or neural models, can adjust used next step process. this study, we propose EvalNet, evaluation network, NMT. EvalNet exploits value, cross-attention map, and semantic similarity between as its features. map an encoded representation layers Transformer, which base architecture system. cosine distance two embeddings source sentence target sentence. Owing parallelism combination proved effective features evaluation, besides value. first evaluator network introduces Through experiments, conclude yet beneficial robust outperforms evaluator.