Missatge de Jonathan Washington <[email protected]> del dia dt., 29 de gen. 2019 a les 6:37:
> > пн, 28 янв. 2019 г. в 15:32, Hèctor Alòs i Font <[email protected]>: > >> Missatge de Francis Tyers <[email protected]> del dia dl., 28 de gen. >> 2019 a les 21:13: >> >>> >>> There is a lot of functionality that is not used widely that could be >>> really >>> used to improve performance of language pairs. >>> >>> * apertium-separable >>> * weights in lttoolbox >>> * weighted transfer >>> >> >> I'd like to try them. I have already used apertium-separable en fra-cat >> and I really helps, >> > > Saluton Hèctor, > > I'm curious how you fared with apertium-separable. I have a specific > question about it: > > What is your pipeline now? My imagined pipeline is something like this: > [xxx tagger] → [xxx-yyy lsx] → [xxx-yyy dix] → [lrx] → [t*x] → > [yyy-xxx.reverse lsx] → [yyy generator] > > With this pipeline, you have to reverse-compile (in theory supported) the > apertium-seperable transducer for the opposite direction and include that > in your pipeline before morphological generation. > > However, dolphingarlic and I were having trouble getting that working in > eng-deu. Our work is in this branch: > https://github.com/apertium/apertium-eng-deu/tree/separable-words > The call to the apertium-separable module stands in two different places, depending on the side of the translation. For fra2cat it is placed just after pretransfer, before lexical selection and transfer. But in cat2fra it is like you say: between transfer and generation. ( https://github.com/apertium/apertium-fra-cat/blob/master/modes.xml ). In the fra-cat pair apertium-separable is used for multiword verbal expressions (like e.g. "avoir lieu") defined in the dictionaries. The problem is in the negative forms, e.g. "n'a pas lieu". In fra2cat it is changed to "n'a lieu pas" before it is searched in the dictionary. In cat2fra the usual pipeline gives something like "ne<morph stuff> avoir<morph stuff> lieu<morph stuff> pas<morph stuff>", so after transfer "pas<morph stuff>" is reordered and just after "n'a pas lieu" is generated. You can take a look on https://github.com/apertium/apertium-fra-cat/blob/master/apertium-fra-cat.fra-cat.l1x and https://github.com/apertium/apertium-fra-cat/blob/master/apertium-fra-cat.cat-fra.l2x Hèctor
_______________________________________________ Apertium-stuff mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/apertium-stuff
