Dear Philipp, Thanks for the reply. I also have the same feeling about getting the CN output, so currently I perform simple processing on n-best list instead.
Regards, Thang On Fri, Dec 19, 2008 at 2:52 AM, Philipp Koehn <[email protected]> wrote: > Hi, > > Moses allows for confusion network and word lattice input, > but you seemed to be interested in confusion network output. > This, however, is in a typical scenario not a useful representation, > since reordering may happen during translation, which does not > fit nicely with the concept of a confusion network. > > -phi > > On Thu, Dec 18, 2008 at 10:10 AM, Thang Luong Minh > <[email protected]> wrote: > > Hi Josh & all, > > > > Tks for the suggestion about the output search graph! I'll use it once I > > totally understand. Currently, I'm using n-best-list with alignment info > to > > derive the different translation choices from a particular src > > words/phrases, which is for simplicity. > > > > After reading through the Moses webpage, I think what I'm interested in > is a > > confusion network (CN) representation instead of a word lattice one. > Thus, I > > have a few concerns, hopefully someone could help me answer them :) : > > 1) representation: the CN derived from the output search graph would mean > > edges between two consecutive nodes correspond to different translation > of > > the same src/word phrase? If so, am I right that we will obtain multiple > CNs > > from an output search graph since there's multiple ways of segmenting the > > source input. > > 2) CN for n-best list: if we want to extract CN network for an n-best > list, > > is that in the search graph, we need to traverse to those last hypotheses > in > > different search paths, and pick up those with top scores? It looks like > > there's some redundancy as the moses option to extract n-best list 's > > already done that? > > 3) ConfusionNet in Moses code: could I make use of this to obtain the > > ConfusionNet at the end of the decoding process? If so, where should I > > start, I mean the classes? > > > > Thanks in advance, > > > > Regards, > > Thang > > > > On Wed, Dec 17, 2008 at 4:44 PM, Josh Schroeder <[email protected]> > > wrote: > >> > >> Thang, > >> > >> I think you'll want to process the file generated by > -output-search-graph. > >> > >> More information: > >> http://www.statmt.org/moses/?n=Moses.AdvancedFeatures#ntoc10 > >> > >> Hope this helps, > >> Josh > >> > >> On 17 Dec 2008, at 09:17, Thang Luong Minh wrote: > >> > >>> Dear all, > >>> > >>> I would like to ask if there's an option in Moses to output n-best list > >>> in lattice form. I checked on the Moses website and the options, but it > >>> looks like there's no option yet. > >>> > >>> I would appreciate if someone could enlighten me on that, or any > specific > >>> files in Moses code I could look into. I'm interested in testing my > >>> morphological preference model to re-decide on the most probable word > forms > >>> among different options derived from the n-best list. > >>> > >>> Regards, > >>> Thang > >>> > >>> -- > >>> Luong Minh Thang > >>> WING group, School of Computing, National University of Singapore > >>> http://wing.comp.nus.edu.sg/~lmthang<http://wing.comp.nus.edu.sg/%7Elmthang> > >>> _______________________________________________ > >>> Moses-support mailing list > >>> [email protected] > >>> http://mailman.mit.edu/mailman/listinfo/moses-support > >> > >> > >> -- > >> The University of Edinburgh is a charitable body, registered in > >> Scotland, with registration number SC005336. > >> > > > > > > > > -- > > Luong Minh Thang > > WING group, School of Computing, National University of Singapore > > http://wing.comp.nus.edu.sg/~lmthang<http://wing.comp.nus.edu.sg/%7Elmthang> > > > > _______________________________________________ > > Moses-support mailing list > > [email protected] > > http://mailman.mit.edu/mailman/listinfo/moses-support > > > > > -- Luong Minh Thang WING group, School of Computing, National University of Singapore http://wing.comp.nus.edu.sg/~lmthang
_______________________________________________ Moses-support mailing list [email protected] http://mailman.mit.edu/mailman/listinfo/moses-support
