Matxin

From Apertium
Jump to navigation Jump to search

Matxin is a free software machine translation engine related to Apertium. It allows for deeper transfer than can be found in Apertium. The linguistic data available under a free licence is a fraction of the data that is used in the papers and descriptions of the subject, so naturally the translations from the pair will be less good than you can find results in the papers.

Contact

Questions and comments about Matxin can be sent to their mailing list matxin-devel, or to the apertium-stuff list.

Prerequisites

  • BerkleyDB — sudo apt-get install libdb4.6++-dev
  • libpcre3 — sudo apt-get install libpcre3-dev

Install the following libraries in <prefix>,

If you're installing into a prefix, you'll need to set two environment variables: CPPFLAGS=-I<prefix>/include LDFLAGS=-L<prefix>/lib ./configure --prefix=<prefix>

Building

Checkout
$ svn co http://matxin.svn.sourceforge.net/svnroot/matxin

Then do the usual: 

$ ./configure --prefix=<prefix>
$ make

After you've got it built, do: 

$ su
# export LD_LIBRARY_PATH=/usr/local/lib
# export PKG_CONFIG_PATH=/usr/local/lib/pkgconfig
# make install

Executing

The default for MATXIN_DIR, if you have not specified a prefix is /usr/local/bin, if you have not specified a prefix, then you should cd /usr/local/bin to make the tests.

Bundled with Matxin there's a script called Matxin_translator which calls all the necessary modules and interconnects them using UNIX pipes. This is the recommended way of running Matxin for getting translations. 

$ ./Matxin_translator -f $MATXIN_DIR/share/config/es-eu.cfg

For advanced uses, you can run each part of the pipe separately and save the output to temporary files for feeding the next modules. 

$ export MATXIN_DIR=<prefix>  
$ echo "Esto es una prueba" |  \
./Analyzer -f $MATXIN_DIR/share/matxin/config/es-eu.cfg | \
./LT -f $MATXIN_DIR/share/matxin/config/es-eu.cfg | \
./ST_inter --inter 1 -f $MATXIN_DIR/share/matxin/config/es-eu.cfg | \
./ST_prep -f $MATXIN_DIR/share/matxin/config/es-eu.cfg | \
./ST_inter --inter 2 -f $MATXIN_DIR/share/matxin/config/es-eu.cfg | \
./ST_verb   -f $MATXIN_DIR/share/matxin/config/es-eu.cfg  | \
./ST_inter --inter 3 -f $MATXIN_DIR/share/matxin/config/es-eu.cfg | \
./SG_inter -f $MATXIN_DIR/share/matxin/config/es-eu.cfg | \
./SG_intra -f $MATXIN_DIR/share/matxin/config/es-eu.cfg | \
./MG -f $MATXIN_DIR/share/matxin/config/es-eu.cfg | \
./reFormat

Da proba bat hau

Speed

Between 25--30 words per second.

Troubleshooting

libfries

If you get the error to do with "strlen was not declared in this scope", add #include <string.h> to the file src/include/fries/language.h.

If you get the error to do with "set_union was not declared in this scope", add #include <algorithm> to the file src/libfries/RGF.cc.

Sometimes, people get the error: 

configure:2668: error: C++ compiler cannot create executables

Try installing libpcre3-dev and trying again.

libomlet

If you get the error to do with "exit was not declared in this scope", add #include <stdlib.h> to the file src/libomlet/adaboost.cc.

FreeLing

If you get the error to do with "exit was not declared in this scope", add #include <stdlib.h> to the files src/utilities/indexdict.cc, src/libmorfo/accents.cc, src/libmorfo/accents_modules.cc, src/libmorfo/dictionary.cc, src/libmorfo/tagger.cc, src/libmorfo/punts.cc, src/libmorfo/maco_options.cc, src/libmorfo/splitter.cc src/libmorfo/suffixes.cc src/libmorfo/senses.cc src/libmorfo/hmm_tagger.cc.

If you get the error to do with "strlen was not declared in this scope", add #include <string.h> to the files src/libmorfo/automat.cc, src/libmorfo/dates.cc, src/libmorfo/locutions.cc, src/libmorfo/maco.cc, src/libmorfo/np.cc, src/libmorfo/nec.cc, src/libmorfo/numbers.cc, src/libmorfo/numbers_modules.cc, src/libmorfo/quantities.cc, src/libmorfo/tokenizer.cc and src/libmorfo/dates_modules.cc.

If you get the error to do with "memcpy was not declared in this scope", add #include <string.h> to the files src/include/traces.h, src/libmorfo/dictionary.cc, src/libmorfo/traces.cc src/libmorfo/senses.cc, src/libmorfo/feature_extractor/fex.cc

If you get the error to do with "set_union was not declared in this scope", add #include <algorithm> to the file src/libmorfo/feature_extractor/RGF.cc.

To add the following strings 

#include <stdlib.h>
#include <string.h>
#include <algorithm>

in the top of every .cc file in the FreeLing-1.5 directory, you can use the following command: 

pasquale@dell:~/stuff/matxin/FreeLing-1.5$ ./configure
..
pasquale@dell:~/stuff/matxin/FreeLing-1.5$ find . -type f -name "*.cc" | awk '{ print "echo \"#include <stdlib.h>\n#include <string.h>\n\
#include <algorithm>\n\" > " $1 ".new && cat " $1 " >> " $1 ".new && mv " $1 ".new " $1 }' > k
pasquale@dell:~/stuff/matxin/FreeLing-1.5$ sh k
pasquale@dell:~/stuff/matxin/FreeLing-1.5$ make
..

If you get the error: 

In file included from analyzer.cc:72:
config.h:32:18: error: cfg+.h: No such file or directory

Run make like make CXXFLAGS=-I<prefix>/include

Matxin

If you get the error: 

g++ -DHAVE_CONFIG_H -I. -I..   -I/usr/local/include -I/usr/local/include/lttoolbox-2.0 -I/usr/include/libxml2  -g -O2 -ansi -march=i686 -O3 
-fno-pic              -fomit-frame-pointer -MT Analyzer.o -MD -MP -MF .deps/Analyzer.Tpo -c -o Analyzer.o Analyzer.C

--->Analyzer.C:10:22: error: freeling.h: Datei oder Verzeichnis nicht gefunden
 In file included from Analyzer.C:9:
 config.h: In constructor 'config::config(char**)':
 config.h:413: warning: deprecated conversion from string constant to 'char*'
 Analyzer.C: In function 'void PrintResults(std::list<sentence, std::allocator<sentence> >&, const config&, int&)':
 Analyzer.C:123: error: aggregate 'std::ofstream log_file' has incomplete type and cannot be defined
 Analyzer.C:126: error: incomplete type 'std::ofstream' used in nested name s...

Then change the header files in src/Analyzer.C to: 

//#include "freeling.h"

#include "util.h"
#include "tokenizer.h"
#include "splitter.h"
#include "maco.h"
#include "nec.h"
#include "senses.h"
#include "tagger.h"
#include "hmm_tagger.h"
#include "relax_tagger.h"
#include "chart_parser.h"
#include "maco_options.h"
#include "dependencies.h"

Upon finding yourself battling the following compile problem, 

Analyzer.C: In function ‘int main(int, char**)’:
Analyzer.C:226: error: no matching function for call to ‘hmm_tagger::hmm_tagger(std::string, char*&, int&, int&)’
/home/fran/local/include/hmm_tagger.h:108: note: candidates are: hmm_tagger::hmm_tagger(const std::string&, const std::string&, bool)
/home/fran/local/include/hmm_tagger.h:84: note:                 hmm_tagger::hmm_tagger(const hmm_tagger&)
Analyzer.C:230: error: no matching function for call to ‘relax_tagger::relax_tagger(char*&, int&, double&, double&, int&, int&)’
/home/fran/local/include/relax_tagger.h:74: note: candidates are: relax_tagger::relax_tagger(const std::string&, int, double, double, bool)
/home/fran/local/include/relax_tagger.h:51: note:                 relax_tagger::relax_tagger(const relax_tagger&)
Analyzer.C:236: error: no matching function for call to ‘senses::senses(char*&, int&)’
/home/fran/local/include/senses.h:52: note: candidates are: senses::senses(const std::string&)
/home/fran/local/include/senses.h:45: note:                 senses::senses(const senses&)

Make the following changes in the file src/Analyzer.C: 

   if (cfg.TAGGER_which == HMM)
-    tagger = new hmm_tagger(cfg.Lang, cfg.TAGGER_HMMFile, cfg.TAGGER_Retokenize, cfg.TAGGER_ForceSelect);
+    tagger = new hmm_tagger(string(cfg.Lang), string(cfg.TAGGER_HMMFile), false);
   else if (cfg.TAGGER_which == RELAX)
-    tagger = new relax_tagger(cfg.TAGGER_RelaxFile, cfg.TAGGER_RelaxMaxIter, 
+    tagger = new relax_tagger(string(cfg.TAGGER_RelaxFile), cfg.TAGGER_RelaxMaxIter,
 			      cfg.TAGGER_RelaxScaleFactor, cfg.TAGGER_RelaxEpsilon,
-			      cfg.TAGGER_Retokenize, cfg.TAGGER_ForceSelect); 
+			      false); 
 
   if (cfg.NEC_NEClassification)
     neclass = new nec("NP", cfg.NEC_FilePrefix);
 
   if (cfg.SENSE_SenseAnnotation!=NONE)
-    sens = new senses(cfg.SENSE_SenseFile, cfg.SENSE_DuplicateAnalysis);
+    sens = new senses(string(cfg.SENSE_SenseFile)); //, cfg.SENSE_DuplicateAnalysis);

Then probably there will be issues with actually running Matxin.

If you get the error: 

config.h:33:29: error: freeling/traces.h: No such file or directory

Then change the header files in src/config.h to: 

//#include "freeling/traces.h"
#include "traces.h"

If you get this error: 

$ echo "Esto es una prueba" | ./Analyzer -f /home/fran/local/share/matxin/config/es-eu.cfg 
Constraint Grammar '/home/fran/local//share/matxin/freeling/es/constr_gram.dat'. Line 2. Syntax error: Unexpected 'SETS' found.
Constraint Grammar '/home/fran/local//share/matxin/freeling/es/constr_gram.dat'. Line 7. Syntax error: Unexpected 'DetFem' found.
Constraint Grammar '/home/fran/local//share/matxin/freeling/es/constr_gram.dat'. Line 10. Syntax error: Unexpected 'VerbPron' found.

You can change the tagger from the RelaxCG to HMM, edit the file <prefix>/share/matxin/config/es-eu.cfg, and change: 

#### Tagger options
#Tagger=relax
Tagger=hmm

Then there might be a problem in the dependency grammar: 

$ echo "Esto es una prueba" | ./Analyzer -f /home/fran/local/share/matxin/config/es-eu.cfg 
DEPENDENCIES: Error reading dependencies from '/home/fran/local//share/matxin/freeling/es/dep/dependences.dat'. Unregistered function d:sn.tonto

The easiest thing to do here is to just remove references to the stuff it complains about: 

cat <prefix>/share/matxin/freeling/es/dep/dependences.dat | grep -v d:grup-sp.lemma > newdep
cat newdep | grep -v d\.class > newdep2
cat newdep2 | grep -v d:sn.tonto > <prefix>/share/matxin/freeling/es/dep/dependences.dat

Error in db

If you get:

  • SEMDB: Error 13 while opening database /usr/local/share/matxin/freeling/es/dep/../senses16.db

rebuild senses16.deb from source:

  • cat senses16.src | indexdict senses16.db
  • (remove senses16.db before rebuild)

Documentation

Retrieved from "https://wiki.apertium.org/w/index.php?title=Matxin&oldid=12573"