Translation Memory (TM) vs Machine Translation (MT)

I twitted Jost's video on the way TM and MT are evolving into one -or at least as complimentary technologies for the translator. There will be times, sooner rather than later I expect, where recalls from a TM and MT output will be so close or at least so reliable that the difference between them will become somewhat fuzzy. Anti-machine translation activists or those worried by the death of the translator need not worry. Machine Translation (and particularly Statistical Machine Translation) has not been designed to corner translators. It is in fact a way of improving the way translators work. I like to remind professional linguists that the mission of the translation community is to help and facilitate communication, not making money as language intermediaries or as language agents. The key to the matter will be how different machine translation vendors link up or make the technology available via TM vendors, who obviously control the tool market. It will be more difficult for a machine translation company to enter the mass translation market, although customization may open new ways and translation memory leverage and repositories are likely to become just another asset in the process. Combined with machine translation, the playfield might become level. Many people who are afraid of machine translation forget that in the case of the nascent Statistical Machine Translation, it is precisely the work of human translators (translation memories) what feeds algorithms that learn language patterns. Through alignments and by having enough human examples, the algorithms learn to reconstruct sentences together, in a process that is not too dissimilar to combining group of words and smoothing the final process out. The reconvergence of TM and MT by Jost. Good video summary of the situation from the translator's perspective Translatorstraining.com http://bit.ly/6HWInU