[libvoikko] Sámi/HFST , anything new?

[Sjur Moshagen]

Den 2. jun. 2010 kl. 18.10 skrev Flammie Pirinen:

>> Even if the situation could not be solved in a way that would allow
>> including HFST 3 in Fedora or Debian it is still possible to have the
>> HFST backend enabled in libvoikko. Users could then install the
>> morphologies manually or distributions could include HFST 2 and those
>> morphologies that can be built with it. But naturally this would not
>> be an ideal solution.
> 
> Yes and as said, target is indeed to have voikko link libhfstol only,
> which also works around this problem.

Sorry if I misunderstand something, but AFAIU, it is generally a requirement in many FLOSS context, especially GPL, that the source code is included and buildable using tools with compatible licenses. That is, if we want to distribute our Sámi transducers (as spellers, hyphenators, whatever), we should ideally only distribute our source code with proper build instructions and dependencies, and then the binaries would be build upstreams by the packages and distributions using them.

I know we in the Divvun/Giellatekno teams are a long way from this goal (our build system is rather - unorthodox), but this is our long-term goal.

If the above is correct, then we would still depend on the whole HFST tool set and back-end libraries to be able to build - HFST2 is not enough, and weighted transducers would best be built using the OpenFST libraries. Of course, if fsm2 (as Krister mentioned) is a good-enough substitution for OpenFST (ie all functionality is retained, only the compilation is e.g. a bit slower and/or memory-hungry), then all is fine. If not, we still have a problem.

Please correct me if I'm wrong.

Sjur