SFST
SFST (Stuttgart Finite State Toolkit) is a set of programs that can be used for writing morphological analysers. It is also one of the possible backends for HFST.
Installation
Prerequisites
On Ubuntu/Debian:
sudo apt-get install libreadline5-dev
Download, compile, install
wget ftp://ftp.ims.uni-stuttgart.de/pub/corpora/SFST/SFST-1.4.2.tar.gz tar -xzvf SFST-1.4.2.tar.gz cd SFST/src make sudo make install sudo make libinstall
Note: on Arch Linux, you may have to do make CFLAGS=-fpic instead of make.
Usage
To try SFST out, you can start by compiling the German transducer, SMOR, that comes with the package:
$ cd data/SMOR $ make
Wait some time, and you will have a file called smor.a, now you need to compact this so it can be read by fst-proc,
$ fst-compact smor.a smor.ac
Now you can use it,
$ cd ../../src $ echo "Ich habe ein Bier" | fst-proc ../data/SMOR/smor.ac ^Ich/<CAP>ich<+PPRO><pers><1><Sg><NoGend><Nom>$ ^habe/haben<+V><1><Sg><Pres><Konj>/haben<+V><3><Sg><Pres><Konj>/haben<+V><Imp><Sg>/haben<+V><1><Sg><Pres><Ind>$ ^ein/ein<+ART><Indef><Masc><Nom><Sg>/ein<+ART><Indef><Neut><Nom><Sg>/ein<+ART><Indef><Neut><Akk><Sg>$ ^Bier/*Bier$
It should also work with deformatters and reformatters,
$ echo "Ich habe <em>ein</em> bier" | apertium-deshtml | ./fst-proc ../data/SMOR/smor.ac ^Ich/<CAP>ich<+PPRO><pers><1><Sg><NoGend><Nom>$ ^habe/haben<+V><1><Sg><Pres><Konj>/haben<+V><3><Sg><Pres><Konj>/haben<+V><Imp><Sg>/haben<+V><1><Sg><Pres><Ind>$[ <em>]^ein/ein<+ART><Indef><Masc><Nom><Sg>/ein<+ART><Indef><Neut><Nom><Sg>/ein<+ART><Indef><Neut><Akk><Sg>$[<\/em> ]^Bier/*Bier$.[][
Note: fst-proc currently does a rather crude tokenisation based on spaces, so multiwords currently aren't possible.
Morphologies
SFST has the following morphologies available for download:
- Morph-IT! (Italian, 34,968 lemmas, LGPL)
- Omorfi–SFST implementation of word form morphology of Finnish (Finnish, 93,510 lemmas, LGPL)
- For further information see: Omorfi
- SMOR — comes in the SFST distribution (German, 1,096 lemmas, GPL)
- trmorph (Turkish, wide coverage)
- For further information see: Turkish
Performance
The analysers produced are fast. For a 1.3Mb analyser (SMOR), it processes ~1,100 words per second. Compare with lttoolbox which processes ~5,000 words per second.
