Hunmorph

From Apertium
Revision as of 14:34, 10 June 2012 by H2m (talk | contribs) (Undo revision 34592 by H2m (Talk): Reason: Changed wrong link (sorry))
Jump to navigation Jump to search

hunmorph is an set of programs for making morphological analysers and generators written largely in Ocaml. Analysers and generators made with these tools could be integrated into an Apertium-based machine translation system, although they would need to be hacked to change the output / input format. Currently it only seems to support Hungarian.

Requirements

On Debian you will need:

  • ocaml
  • ocaml-libs
  • ocaml-tools
  • ocaml-compiler-libs
  • ocaml-nox

Get them all by issuing:

sudo apt-get install ocaml ocaml-libs ocaml-tools ocaml-compiler-libs ocaml-nox

Compiling

Check out the code from CVS, compile the Ocaml code, C bindings and the C wrapper around the morphological analyzer:

cvs -d :pserver:anonymous:anonymous@cvs.mokk.bme.hu:/local/cvs co ocamorph
cd ocamorph
./build.sh build
cd src/lib
make
cd ../bindings/c
make
cd ../../wrappers/ocamorph
make

If you get the error, /usr/bin/ld: cannot find -lunix, then check the Makefile and the include -I and library -L paths, probably they don't point to the right place. On Debian, I had to change the /usr/lib/ocaml/3.09.1 for /usr/lib/ocaml/3.10.1 and change /usr/local/lib to /usr/lib. After you've compiled this you should have an ocamorph binary in wrappers/ocamorph. Now go back to the root of your CVS tree.

You can test ocamorph with the binary distribution available [ ftp://ftp.mokk.bme.hu/Tool/Hunmorph/Resources/Morphdb.hu/morphdb-hu-20060525.tgz here]. If you untar the file in ~/source/ you should see:

$ ls ~/source/morphdb.hu/
AUTHORS  CVS  doc  LICENCE  morphdb_hu.aff  morphdb_hu.dic  README

You can then test it with:

$ echo "programot" | ocamorph --aff ~/source/morphdb.hu/morphdb_hu.aff --dic ~/source/morphdb.hu/morphdb_hu.dic
> programot
program/NOUN<CAS<ACC>>
Compiling the lexicon

First check out the morphdb.hu and hunlex.

cvs -d :pserver:anonymous:anonymous@cvs.mokk.bme.hu:/local/cvs co lexicons/morphdb.hu 
cvs -d :pserver:anonymous:anonymous@cvs.mokk.bme.hu:/local/cvs co hunlex

Enter the hunlex/ directory, and run make. If there are errors, ignore them, providing that an executable hunlex is created under the src/ sub-directory.

Next enter the lexicons/morphdb.hu directory, edit the Makefile and change the paths of HUNLEXMAKEFILE and HUNLEX to point to the directory where you just installed in. Uncomment them if necessary, something like this will probably be the result.

HUNLEXMAKEFILE=../../hunlex/adm/HunlexMakefile
HUNLEX=../../hunlex/src/hunlex

Now issue the make command. Note, compiling the lexicon could take a long time, and will possibly take up a lot of CPU cycles, consider running it on a different machine than your desktop. The resultant .aff and .dic files will be found in the out/ directory.

Performance

For a 10,000 line test file, with a analyser with support for 4,000,000 word forms.

$ time cat /tmp/test | ocamorph --aff ~/source/morphdb.hu/morphdb_hu.aff --dic ~/source/morphdb.hu/morphdb_hu.dic > /dev/null
real    0m47.224s
user    0m41.859s
sys     0m0.620s

Compile the lexicon using:

$ echo "programot" | ocamorph --aff ~/source/morphdb.hu/morphdb_hu.aff --dic ~/source/morphdb.hu/morphdb_hu.dic --bin hu.morph.bin

You seem to be required to attempt to analyse something in order to compile. Then re-test:

$ time cat /tmp/test | ocamorph  --bin hu.morph.bin > /dev/null
real    0m15.023s
user    0m14.625s
sys     0m0.344s

Final size of the compiled binary is 22Mb.

Further reading

  • Trón, V., Németh, L., Halácsy, P., Kornai, A., Gyepesi, G., and Varga, D. (2005) "Hunmorph: open source word analysis". Proceedings of the ACL 2005 Workshop on Software. pp. 77--85

External links