Hi Marcin, Sounds like a good idea (clustering and returning representative translations for each cluster).
As far as I know, there has not been much work on generating diverse lists of translations, but one notable exception is "Trait-Based Hypothesis Selection for Machine Translation" by Jacob Devlin and Spyros Matsoukas ( http://www.aclweb.org/anthology-new/N/N12/N12-1059.pdf). Maybe others know of other relevant work. Kevin On Sat, Jul 27, 2013 at 4:53 AM, Marcin Junczys-Dowmunt <[email protected]>wrote: > Hi list, > is the --distinct parameter currently the only option to generate more > diverse n-best lists? > > I have the following scenario: > Human translators use Moses like a TM via a Trados Plugin, upon request > they may see a list of m alternatives, which is just a list of the first > m sentences from a bigger n-best list. Usually those alternative are not > very useful (confirmed by translators), as they are still very similar > to each other and the best sentence. Current idea: generate bigger > n-best list, cluster using nifty similarity function, display only > cluster representatives. Somehow I believe something like that should > have been done before, have you heard of anything like that? > Best, > Marcin > _______________________________________________ > Moses-support mailing list > [email protected] > http://mailman.mit.edu/mailman/listinfo/moses-support >
_______________________________________________ Moses-support mailing list [email protected] http://mailman.mit.edu/mailman/listinfo/moses-support
