Hi Barry, thanks for your answer. Yes, I'm talking about "tree-to-tree" models. After reading a lot of different papers I've concluded that such a model is just as helpless as an old-fashioned transfer system facing nominalization: the need to translate a verb phrase to a noun phrase. I just wanted to check if I've overlooked some useful feature of Moses that might address the problem.
I've glanced through the paper you linked to, but I'm not sure if the described strategy would help to overcome this particular problem. Yours, Per Tunedal On Mon, Jun 17, 2013, at 10:08, Barry Haddow wrote: > Hi Per > > I guess you're talking specifically about "tree-to-tree" models? It's > generally recognised that extracting only those rules that respect the > tree structure on both sides is far too restrictive - but I'm not sure > that translation of verb phrase to a noun phrase is actually impossible. > > In any case, there have been many attempts to overcome the restrictions > of "pure" tree-to-tree models, for example Zollman's SAMT and David > Chiang's fuzzy tree-to-tree models > (http://www.isi.edu/~chiang/papers/acl2010-chiang.pdf) - to name but two, > > cheers - Barry > > On 17/06/13 08:44, Per Tunedal wrote: > > Hi again, > > I guess the answer is no. > > After rereading the Moses Syntax tutorial, I conclude that hierarchical > > models with syntax cannot handle nominalization: > > > > A rule translating a verb phrase to a noun phrase cannot be extracted. > > > > Yours, > > Per Tunedal > > > > On Sat, Jun 15, 2013, at 14:11, Per Tunedal wrote: > >> Hi, > >> working with French and Swedish I'm wondering if Moses and in particular > >> an Hierarchical model with syntax on both sides can handle > >> nominalization. > >> > >> Swedes often prefer to express certain kinds of meaning with a verb, > >> where Frenchmen prefer a noun: > >> > >> Jag är förvånad över att HAN LYCKADES. (he succeeded) > >> Je m'étonne de SA RÉUSSITE. (his success) > >> > >> Looking at a description of synchronous context-free grammar I conclude > >> that a rule must have one and only one symbol to the left. This would > >> imply that verb phrases could only be translated to verb phrases, noun > >> phrases only to noun phrases. Is this true for Moses or are the > >> tree-rules constructed in a way that masters this problem? > >> > >> I haven't yet managed to verify this in my tests. > >> > >> Yours, > >> Per Tunedal > >> > >> _______________________________________________ > >> 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 > > > -- > The University of Edinburgh is a charitable body, registered in > Scotland, with registration number SC005336. > _______________________________________________ Moses-support mailing list [email protected] http://mailman.mit.edu/mailman/listinfo/moses-support
