Closed fooSynaptic closed 5 years ago
I have not tried it with NMT datasets. In general, they will improve the readability and diversity of the answers. However, researchers of NMT are too focused on getting the highest BLEU scores for which the exact match is more important. Thus, they do not try paraphrasing models to augment their answers.
I think its really a good try for NMT engineer or researcher. Maybe paraphrase generation can fix the syntex error in the NMT output?
which still the highest value is applied to sequence generation like NMT and text summary?