On Tue, Feb 04, 2020 at 12:55:55PM -0500, Daniel Swanson wrote: > > Do you have plans on doing tests > > on runtime efficiency, i.e. how fast it is to run the automata on texts? > > One thing that we found with flag diacritics on lexc is is that it's > > kindof possible to abuse them to optimise the compiled stuff and it'd > > probably be interesting to see here too, I see there's something with > > flags in the code already? > > It can compile with or without flag diacritics, though the flag mode was > mostly an afterthought and I haven't really tested it yet.
Yeah so if the flag mode stuff works it can be interesting for testing if flags optimise certain morphotactics or not. > For non-flag runtimes, the transducers should be the same as lexc + twoc, > apart from alignment differences (a:b c:0 vs a:0 c:b) and state numbers, so > I assumed it would have the same performance, but maybe I should double > check. Yeah, the --align option of hfst-lexc is there because alignment differences in worst cases are bad; especially as lexc is usually followed by further processing, I haven't checked the theory but from experience I'd estimate it can get exponentially worse, it certainly made some bigger langs uncompileable. In the end though this is all usually only noticeable with rather large old language models like Finnish or North Sámi. -- Regards, Flammie <https://flammie.github.io> (Please note, that I will often include my replies inline instead of top or bottom of the mail)
signature.asc
Description: PGP signature
_______________________________________________ Apertium-stuff mailing list Apertium-stuff@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/apertium-stuff
