[libvoikko] Introducing VFSTs (Varissuo Finite State Transducers)
hatapitk at iki.fi
Thu Mar 1 16:17:42 EET 2012
On Wednesday 29 February 2012, Sam Hardwick wrote:
> Very interesting! I'd be interested in hearing more about VFST, sounds
> very useful.
I can write something about the format in the wiki soon.
> 1) Does the VFST lookup do result caching? That's known to give a big
> speed boost.
No. Spellers that are built on top of the analyzers do cache the spelling
results but analysis results are not cached. This was designed this way
because Malaga results were complex and hard/expensive to cache. Of course now
that the results are just strings it would be possible to move the cache from
spell checker down to analyzer.
> 2) 80.4 s for 80 000 words sounds slow for hfst-optimized-lookup, even
> with OmorFi. There's probably something a bit off here.
I suppose so. Maybe the tool was configured with debug settings or something
like that. I'll check later today.
More information about the Libvoikko