[Corpora-List] mmorph tool

Francis Tyers ftyers at prompsit.com
Thu Apr 14 18:29:53 CEST 2011


El dj 14 de 04 de 2011 a les 12:29 -0400, en/na maxwell va escriure:
> On Thu, 14 Apr 2011 14:43:28 +0000, Francis Tyers <ftyers at prompsit.com>
> wrote:
> >> > There is HFST,[1]
> >>
> >> That footnote didn't come through here.
> >
> > Opa!
> >
> > http://www.ling.helsinki.fi/kieliteknologia/tutkimus/hfst/
>
> Now I'm confused. I thought when you mentioned HFST as a
> re-implementation of the Xerox lexc/xfst, that you were referring to Mans
> Hulden's foma implementation
> (http://dingo.sbs.arizona.edu/~mhulden/hulden_foma_2009.pdf,
> http://wiki.apertium.org/wiki/Foma). Yes, I know, "foma" doesn't look
> anything like "HFST"...that's what I get for not checking my references
> first!
>
> My current confusion is whether HFST supports the xfst notion of "replace
> rules," i.e. more or less normal phonological rules that get composed with
> an FST (usually on the surface side) and modify the network, e.g.
> b --> p / __ #
> At a quick glance, I don't see those in the pages at the above link to the
> HFST documentation. Does HFST implement them? They're pretty central to
> the way most people use the Xerox xfst.

In HFST, the replace rules are implemented as a wrapper around foma -- the program that compiles them (runs the XFST scripts) is called 'hfst-xfst'. I put Tommi Pirinen in CC, he is one of the main developers.

Fran



More information about the Corpora mailing list