Translation quality is evaluated using both automatic
metrics and human evaluation, also using the MQM error typology. While
as expected NMT outperforms PB-SMT in terms of committing fewer inflectional
morphology and word-order errors, the authors find that, contrary to previous
results, NMT is better across all language pairs on both short and long sentences
as regards fluency, while the results are less clear when it comes to adequacy
نکته :
MQM : multidimensional quality metrics
NMT : neural machin translation
PB-SMT : phrase-based statistical machin translation