Difference between revisions of "Foma"

From Apertium
Jump to navigation Jump to search
Line 81: Line 81:
This final step takes some time, up to 2—3 minutes. It also takes a lot of processing power and RAM. The final result will be:
This final step takes some time, up to 2—3 minutes. It also takes a lot of processing power and RAM. The final result will be:


<pre>
<pre> 76.4 MB. 160041 states, 5002206 arcs, Cyclic.
76.4 MB. 160041 states, 5002206 arcs, Cyclic.
foma[1]:
foma[1]:
</pre>
</pre>

Revision as of 08:59, 30 September 2009

foma is a finite-state toolkit that implements Xerox lexc and xfst. It can be used for building morphologies of natural languages.

Installation

Note: foma requires libreadline to be installed, on Debian or Ubuntu use apt-get install libreadline5-dev

  • Download the .tar.gz source from the website.
  • Untar
  • Run make

If you get an error Makefile:12: *** missing separator. Stop., edit the Makefile and add \ to the end of the lines 11--13.

  • This will create a binary foma, which should be copied into your PATH.

Example usage

First check out the Greenlandic (kal) morphology from Giellatekno SVN:

$ svn co https://victorio.uit.no/langtech/trunk/st/kal

Move to the src/ directory and combine all the lexc source files:

$ cat kal-lex.txt \
abbr-kal-lex.txt acro-kal-lex.txt \
noun-kal-lex.txt verb-kal-lex.txt \
ateq-kal-lex.txt ateq-kal-morph.txt \
punct-kal-lex.txt prt-kal-lex.txt num-kal-lex.txt > kal-lex-all.lexc

Next, remove the comments from the xfst rewrite rule file:

$ cat xfst-kal.txt | sed 's/\s\!.*$/ /g' | grep -v '^!' | sed 's/$/ /g' | grep -v 'echo' > xfst-kal.tmp

Compile the xfst code as follows, run foma and load the rewrite rules:

foma[0]: source xfst-kal.tmp 
Opening file 'xfst-kal.tmp'.
defined Vow: 348 bytes. 2 states, 6 arcs, 6 paths.
defined Cns: 741 bytes. 2 states, 19 arcs, 19 paths.
...
6.1 MB. 12474 states, 402541 arcs, Cyclic.
foma[1]: 

Note the [1], if you don't get this something has gone wrong.

Next, save the compiled transducer and quit:

foma[1]: save stack xfst-kal.bin
Writing to file xfst-kal.bin.
foma[1]: quit

Now we compile the lexc file and save the resulting transducer and quit:

$ foma
foma[0]: read lexc kal-lex-all.lexc
Root...8, Z1Zmorf...59, Z1SZmorf...56, Z1PZmorf...59, Z1+ssZmorf...59, ...
Building lexicon...Determinizing...Minimizing...Done!
85.5 MB. 154826 states, 5599566 arcs, Cyclic.
foma[1]: save stack kal-lex.save
Writing to file kal-lex.save.
foma[1]: quit
<pre>

The final step is to compose the two transducers (the lexicon and the rewrite rules),

<pre>
$ foma
foma[0]: regex [[@"kal-lex.save"] .o. [[@"kal-lex.save"].l .o. [@"xfst-kal.bin"]] ] ;

This final step takes some time, up to 2—3 minutes. It also takes a lot of processing power and RAM. The final result will be:

76.4 MB. 160041 states, 5002206 arcs, Cyclic.
foma[1]:

Then save the final transducer, and quit:

foma[1]: save stack kal.morph.bin
Writing to file kal.morph.bin.
foma[1]: quit

You can now use the transducer for analysis and generation, for example,

$ foma
foma[0]: load kal.morph.bin
76.4 MB. 160041 states, 5002206 arcs, Cyclic.
foma[1]: apply up nittartagaq
nittar+TAR+vv+TAQ+N+Abs+Sg
foma[1]: apply up kalaallisut
kalaaleq+N+Aeq+Pl
kalaaleq+N+Aeq+Sg
foma[1]: apply down kalaaleq+N+Aeq+Sg
kalaallitut
kalaallisut

External links