A Comparison of Machine Translation Paradigms for Use in Black-Box Fuzzy-Match Repair

摘要

Fuzzy-match repair (FMR). which combines a human-generated translation memory (TM) with the flexibility of machine translation (MT), i.s one way of using MT to augment resources available to translators. We evaluate rule-based, phrase-based, and neural MT systems as black-box sources of bilingual information for FMR. We show that FMR success varies based on both the quality of the MT system and the type of MT system being used.