Evaluating Human Correction Quality for Machine Translation from Crowdsourcing

28901

Knowledge Triples