[libvoikko] Libvoikko compiled for Android

["Harri Pitkänen"]

ke 22.8.2012 17:26 Harri Pitkänen kirjoitti:
> On Tuesday 21 August 2012, Flammie Pirinen wrote:
>> Ok, I tried to test this with my minilaptop for now, and notice that
>> there is another codepath in ZHfstOSpeller for legacy speller, the
>> automatic test only tested the older Speller object. Should be finally
>> fixed in r2626
>
> Thanks, I'll test this and if it works I'll update the binary test package
> for
> Android to have this support enabled.

It seems that legacy HFST format works correctly with libvoikko now.
Thanks for that! The Android package will come soon.

But then I ran into another problem when comparing the results from HFST
and VFST spellers. It looks like hfst-fst2txt is somehow broken at least
when optimized lookup transducers are used as input. For Omorfi speller

$ hfst-fst2txt -D spl.hfstol | head
0       1       @C.AGREE@       @C.AGREE@
0       2       -       -
0       2       ‐       ‐
1       196922  @D.NEED@        @D.NEED@
1       196923  A       A
1       196924  B       B
1       196925  C       C
1       196926  D       D
1       196927  E       E
1       196928  F       F

Lines 2 and 3 are the same and then there is no transition from initial
state with label @U.NEED.NOMINAL@ which would be needed for some words
(such as "aamu"). Thus "aamu" is not recognized with VFST speller. Similar
issues break the Sami speller as well.

I don't have similar problems with the non-optimized hfst format which I
use for Suomi-malaga -> VFST porting:

$ hfst-fst2txt -D ~/tmp/voikkotest/build/fi-x-vfstd/vvfst.mor.hfst | head
0       1       ž       [Les]
0       35      å       [Les]
0       141     q       [Lep]
0       166     q       [Les]
0       184     q       [Ln]
0       216     z       [Lep]
0       266     z       [Les]
0       291     z       [Ln]
0       299     š       [Ll]
0       339     ö       [Ll]


Harri