Hi, the best way to understand reordering, you have to keep in mind the alignment to the source.
If f1, f2, and f3, are the source phrases of the sentence in that order, and you translate f1->e1, f3->e2, f2->e3, meaning you first translate f1, then f3, and then f2, then f1->e1 is consider monotone, f3->e2 discontinuous (because it has connection to the previous phrase), and f2->e3 a swap (because it swaps order with the previous phrase). The IWSLT paper gives some illustration of it, I hope this helps. -phi > On 19/03/2011 15:05, Neda NoorMohammadi wrote: > > Hi, > > The documentation is particularly scarce on lexicalized reordering , > so I don’t understand the detail! Also, i studied Koehn’s IWSLT paper. > How is lexicalized reordering in moses? > > Whether the reordering in moses just is done on segments of source > sentence? if so, when a segment for translating is selected, one of > the hypothesis that is produced is: > with considering the probability of swap orientation this phrase > with previous phrase, the translation of this phrase with translation > of previous phrase in old hypothesis is swapped and is build a new > hypothesis. A simple sample is: > > e1 e2 e3 : old hypothesis > e4: the translation of new segment > (e3,e4) --> swap > e1 e2 e4 e3: new hypothesis > > is it correct? > > Also, what is subcategory of lexicalized reordering code in moses? > > Thanks. > > _______________________________________________ > 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 > > _______________________________________________ Moses-support mailing list [email protected] http://mailman.mit.edu/mailman/listinfo/moses-support
