This article reports a multifaceted comparison between statistical and neuralmachine translation (MT) systems that were developed for translation of data frommassive open online courses (MOOCs). The study uses four language pairs: English toGerman, Greek, Portuguese, and Russian. Translation quality is evaluated using automaticmetrics and human evaluation, carried out by professional translators. Resultsshow that neuralMTis preferred in side-by-side ranking, and is found to contain feweroverall errors. Results are less clear-cut for some error categories, and for temporaland technical post-editing effort. In addition, results are reported based on sentencelength, showing advantages and disadvantages depending on the particular languagepair and MT paradigm.
Evaluating Machine Translation for Massive Open Online Courses: A Multifaceted Comparison between Phrase-Based Statistical Machine Translation and Neural Machine Translation Systems
Federico Gaspari;
2018-01-01
Abstract
This article reports a multifaceted comparison between statistical and neuralmachine translation (MT) systems that were developed for translation of data frommassive open online courses (MOOCs). The study uses four language pairs: English toGerman, Greek, Portuguese, and Russian. Translation quality is evaluated using automaticmetrics and human evaluation, carried out by professional translators. Resultsshow that neuralMTis preferred in side-by-side ranking, and is found to contain feweroverall errors. Results are less clear-cut for some error categories, and for temporaland technical post-editing effort. In addition, results are reported based on sentencelength, showing advantages and disadvantages depending on the particular languagepair and MT paradigm.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
