🤖 Phrase, up until recently known as Memsource, announces the release of NextMT: the first neural machine translation service from the very beginning designed to be used in a CAT tool. And in its own CAT tool only: NextMT will be available solely within the Phrase translation platform.
NextMT will be actively using existing translations done by people (translation memories) and term glossaries, i.e., it is the so-called trained machine translation (trained MT) by default.
Other MT services can also train MT. But NextMT offers another, unique function that has not been introduced by anyone so far.
It is believed that while working in a CAT program it is better to translate fuzzy matches with a match value of less than 85% by editing machine translation than adapting the fuzzy match from the translation memory (naturally, this value depends on the language direction, text topic, etc.).
NextMT developers have attempted to combine the best of the two worlds of “human” and machine translation. Upon coming across a fuzzy match, the system will not translate the segment from scratch but use the existing “human” translation as much as possible, completing only the small difference between the fuzzy and 100% matches with the help of MT.
This approach is estimated to be able to increase translation quality by 50%. Whether it is true, we will see quite soon.
#memsource #phrase #machinetranslation
NextMT will be actively using existing translations done by people (translation memories) and term glossaries, i.e., it is the so-called trained machine translation (trained MT) by default.
Other MT services can also train MT. But NextMT offers another, unique function that has not been introduced by anyone so far.
It is believed that while working in a CAT program it is better to translate fuzzy matches with a match value of less than 85% by editing machine translation than adapting the fuzzy match from the translation memory (naturally, this value depends on the language direction, text topic, etc.).
NextMT developers have attempted to combine the best of the two worlds of “human” and machine translation. Upon coming across a fuzzy match, the system will not translate the segment from scratch but use the existing “human” translation as much as possible, completing only the small difference between the fuzzy and 100% matches with the help of MT.
This approach is estimated to be able to increase translation quality by 50%. Whether it is true, we will see quite soon.
#memsource #phrase #machinetranslation