[libvoikko] Setting up HFST morphology backend

[Harri Pitkänen]

On Monday 26 April 2010, Sjur Moshagen wrote:
> One of the things I'm going to do - irrespective of which backend is used -
>  is to add voikko support for our proofing tools test bench. For that I
>  need some help with running voikkospell and voikkohyphenate, that is, I
>  need an overview of the command line options for the two commands. I could
>  not find anything online or in the documentation in svn.

The tools have manual pages that should get automatically installed with the 
library. So

  MANPATH=$libvoikko_installation_prefix/share/man man voikkospell

should give you information about all command line options. Just replace 
$libvoikko_installation_prefix with the directory where libvoikko got 
installed.

You can also see slightly outdated version online at

  http://manpages.ubuntu.com/manpages/lucid/man1/voikkospell.1.html

> 
> What I want to do:
> - run a file containing one word pr line through the speller
> - get back not only a simple C or W evalutation, but also - if W - the list
>  of suggestions, possibly with some weighting info as well

"voikkospell -s" will give you the suggestions. Weighting info is calculated 
internally but it cannot be printed at the moment.

> The result is compared to the input strings as well as with the expected
>  behavior (taken from gold standard documents/hand-annotated corrected
>  texts), and precision & recall, spelling error statistics, suggestion
>  statistics etc. are automatically calculated.
> 
> We have gold standard documents for three sámi languages, and it is easy to
>  add for new languages. For two sámi languages we have working lexicons for
>  both hunspell and our MS Office speller, and it is thus straightforward to
>  present comparisons of the type you were looking for in your e-mail a
>  couple of days ago.
> 
> The things I want to do are pretty straightforward, but not easy without
>  documentation (a simple -h / --help option would have been enough) ;)

Good idea, I'll add these options.

Harri