BerlinMT

Discussion -- Machine Translation in DELPH-IN

It might be good to have mechanisms for translation rankings, so as to be able to choose optimal outputs for the translation task, from either the deep or the shallow components.

It might be worth having a hand-crafted parallel corpus. This could be used for a demo with a graphical interface. Such a corpus might come from a Wizard-of-Oz experiment, or some already existing source corpus, like Verbmobil, or data from MT competitions, which can be obtained by (having students, mainly) participating to such competitions.