Difference between revisions of "Modes"

From Apertium
Jump to navigation Jump to search
(15 intermediate revisions by 4 users not shown)
Line 1: Line 1:
  +
[[Les fichiers modes|En français]]
There are a few ways you can use [[pipeline]]s in Apertium. One of them is '''Modes''' files. Modes files (typically called <code>modes.xml</code>) are XML files (see [http://apertium.svn.sourceforge.net/viewvc/apertium/trunk/apertium/apertium/modes.dtd modes.dtd]) which specify which programs should be run and in what order. Normally each linguistic package has one of these files which specifies various ways in which you can use the data to perform translations.
 
   
 
There are a few ways you can use [[pipeline]]s in Apertium. One of them is '''Modes''' files. Modes files (typically called <code>modes.xml</code>) are XML files (see [https://github.com/apertium/apertium/blob/master/apertium/modes.dtd modes.dtd]) which specify which programs should be run and in what order. Normally each linguistic package has one of these files which specifies various ways in which you can use the data to perform translations.
See the [http://apertium.svn.sourceforge.net/viewvc/apertium/trunk/apertium-es-ca/modes.xml modes file from es-ca] for an example. The modes which do not say <code>install="yes"</code> are only usable with the -d switch to apertium, these are typically used during development (eg. ca-es-anmor which only performs morphological analysis on Catalan and nothing else).
 
   
 
See the [https://github.com/apertium/apertium-spa-cat/blob/master/modes.xml modes file from Spanish-Catalan pair] for an example. The modes which do not say <code>install="yes"</code> are only usable with the -d switch to apertium, these are typically used during development (eg. ca-es-anmor which only performs morphological analysis on Catalan and nothing else).
   
  +
See [[Writing_Makefiles#Modes]] on how to ensure modes that say install="yes" are installed.
== Statistics mode ==
 
  +
  +
== Naming conventions ==
  +
  +
The main translation mode is always named "from-to", e.g. "sme-nob". The debug modes each have a suffix, e.g. "sme-nob-morph".
  +
  +
Common debug mode names:
  +
  +
* -anmor or -morph run the morphological analysers
  +
** these are used equivalently
  +
* -disam runs up until morphological (CG) disambiguation
  +
* -syntax runs up until syntactical (CG) disambiguation
  +
* -tagger runs up until probabilistic (apertium-tagger) disambiguation (or, if no .prob, up until the last disambiguation step)
  +
* -biltrans runs up until the bidix
  +
* -lex runs up until lexical selection
  +
* -transfer runs up until (1-stage) transfer
  +
* -chunker runs up until the first stage of 3-or-more-stage transfer
  +
* -interchunk runs up until the second stage of 3-stage transfer
  +
** -interchunk1 and -interchunk2 are used when the pair has 4-stage transfer
  +
* -postchunk runs up until the last stage of transfer
  +
* -dgen run up until generation (using lt-proc -d to include debug symbols)
  +
  +
== Autogenerating debug modes ==
  +
It's a drag to create all the regular install=no debug modes all the time (e.g. foo-bar-tagger, foo-bar-chunker, etc.).
  +
  +
If you put '''<code>gendebug="yes"</code>''' on a &lt;mode&gt; element, debug modes will be created automatically for you! If a mode comes out with the wrong suffix, you can override the guess by adding the <code>debug-suff</code> attribute.
  +
  +
For example, the following entry
  +
  +
<pre>
  +
<mode name="sme-nob" install="yes" gendebug="yes">
  +
<pipeline>
  +
<program name="hfst-proc --weight-classes 1 -w -p">
  +
<file name="sme-nob.automorf.hfst"/>
  +
</program>
  +
<program name="cg-proc" debug-suff="disam">
  +
<file name="sme-nob.mor.rlx.bin"/>
  +
</program>
  +
<program name="cg-proc -1 -n -w" debug-suff="syntax">
  +
<file name="sme-nob.syn.rlx.bin"/>
  +
</program>
  +
<program name="apertium-pretransfer"/>
  +
<program name="lt-proc -b">
  +
<file name="sme-nob.autobil.bin"/>
  +
</program>
  +
<program name="cg-proc" debug-suff="lex">
  +
<file name="sme-nob.lex.bin"/>
  +
</program>
  +
<program name="apertium-transfer -b">
  +
<file name="apertium-sme-nob.sme-nob.t1x"/>
  +
<file name="sme-nob.t1x.bin"/>
  +
</program>
  +
<program name="apertium-interchunk" debug-suff="interchunk1">
  +
<file name="apertium-sme-nob.sme-nob.t2x"/>
  +
<file name="sme-nob.t2x.bin"/>
  +
</program>
  +
<program name="apertium-interchunk" debug-suff="interchunk2">
  +
<file name="apertium-sme-nob.sme-nob.t3x"/>
  +
<file name="sme-nob.t3x.bin"/>
  +
</program>
  +
<program name="lt-proc $1">
  +
<file name="sme-nob.autogen.bin"/>
  +
</program>
  +
</pipeline>
  +
</mode>
  +
</pre>
  +
would make all these modes automatically:
  +
* sme-nob.mode
  +
* sme-nob-morph.mode
  +
* sme-nob-disam.mode
  +
* sme-nob-syntax.mode
  +
* sme-nob-pretransfer.mode
  +
* sme-nob-biltrans.mode
  +
* sme-nob-lex.mode
  +
* sme-nob-chunker.mode
  +
* sme-nob-interchunk1.mode
  +
* sme-nob-interchunk2.mode
  +
* sme-nob-postchunk.mode
  +
* sme-nob-dgen.mode
  +
(but only install the sme-nob.mode), where -disam, -syntax, -interchunk1, -interchunk2 are manually specified names, and the rest are default names based on program names.
  +
  +
== Modes hacks ==
 
=== Statistics mode ===
   
 
In order to get some statistical information about translations made using Apertium, we've hacked the main translation mode, pausing the pipeline just after disambiguation and saving the output into a temp file. After that, pipeline is resumed with temp file as stdin.
 
In order to get some statistical information about translations made using Apertium, we've hacked the main translation mode, pausing the pipeline just after disambiguation and saving the output into a temp file. After that, pipeline is resumed with temp file as stdin.
Line 16: Line 99:
 
/usr/local/share/apertium/apertium-es-ca/trules-ca-es.bin /usr/local/share/apertium/apertium-es-ca/ca-es.autobil.bin \
 
/usr/local/share/apertium/apertium-es-ca/trules-ca-es.bin /usr/local/share/apertium/apertium-es-ca/ca-es.autobil.bin \
 
|/usr/local/bin/lt-proc $1 /usr/local/share/apertium/apertium-es-ca/ca-es.autogen.bin \
 
|/usr/local/bin/lt-proc $1 /usr/local/share/apertium/apertium-es-ca/ca-es.autogen.bin \
|/usr/local/bin/lt-proc -p /usr/local/share/apertium/apertium-es-ca/ca-es.autopgen.bin
+
|/usr/local/bin/lt-proc -p /usr/local/share/apertium/apertium-es-ca/ca-es.autogen.bin
 
</pre>
 
</pre>
   
Line 30: Line 113:
 
When translation is done, we can process the log created in order to get statistics.
 
When translation is done, we can process the log created in order to get statistics.
   
==See also==
+
=== Mixed modes ===
   
* [[Mixed modes]]
+
See [[Mixed modes]]
  +
  +
==See also==
  +
* https://victorio.uit.no/langtech/trunk/tools/CorpusTools/corpustools/modes.py is a Python module that parses modes files into a <code>Pipeline</code> class
   
 
[[Category:Documentation]]
 
[[Category:Documentation]]

Revision as of 22:32, 23 March 2019

En français

There are a few ways you can use pipelines in Apertium. One of them is Modes files. Modes files (typically called modes.xml) are XML files (see modes.dtd) which specify which programs should be run and in what order. Normally each linguistic package has one of these files which specifies various ways in which you can use the data to perform translations.

See the modes file from Spanish-Catalan pair for an example. The modes which do not say install="yes" are only usable with the -d switch to apertium, these are typically used during development (eg. ca-es-anmor which only performs morphological analysis on Catalan and nothing else).

See Writing_Makefiles#Modes on how to ensure modes that say install="yes" are installed.

Naming conventions

The main translation mode is always named "from-to", e.g. "sme-nob". The debug modes each have a suffix, e.g. "sme-nob-morph".

Common debug mode names:

  • -anmor or -morph run the morphological analysers
    • these are used equivalently
  • -disam runs up until morphological (CG) disambiguation
  • -syntax runs up until syntactical (CG) disambiguation
  • -tagger runs up until probabilistic (apertium-tagger) disambiguation (or, if no .prob, up until the last disambiguation step)
  • -biltrans runs up until the bidix
  • -lex runs up until lexical selection
  • -transfer runs up until (1-stage) transfer
  • -chunker runs up until the first stage of 3-or-more-stage transfer
  • -interchunk runs up until the second stage of 3-stage transfer
    • -interchunk1 and -interchunk2 are used when the pair has 4-stage transfer
  • -postchunk runs up until the last stage of transfer
  • -dgen run up until generation (using lt-proc -d to include debug symbols)

Autogenerating debug modes

It's a drag to create all the regular install=no debug modes all the time (e.g. foo-bar-tagger, foo-bar-chunker, etc.).

If you put gendebug="yes" on a <mode> element, debug modes will be created automatically for you! If a mode comes out with the wrong suffix, you can override the guess by adding the debug-suff attribute.

For example, the following entry

  <mode name="sme-nob" install="yes" gendebug="yes">
    <pipeline>
      <program name="hfst-proc --weight-classes 1 -w -p">
        <file name="sme-nob.automorf.hfst"/>
      </program>
      <program name="cg-proc" debug-suff="disam">
        <file name="sme-nob.mor.rlx.bin"/>
      </program>
      <program name="cg-proc -1 -n -w" debug-suff="syntax">
        <file name="sme-nob.syn.rlx.bin"/>
      </program>
      <program name="apertium-pretransfer"/>
      <program name="lt-proc -b">
        <file name="sme-nob.autobil.bin"/>
      </program>
      <program name="cg-proc" debug-suff="lex">
        <file name="sme-nob.lex.bin"/>
      </program>
      <program name="apertium-transfer -b">
        <file name="apertium-sme-nob.sme-nob.t1x"/>
        <file name="sme-nob.t1x.bin"/>
      </program>
      <program name="apertium-interchunk" debug-suff="interchunk1">
        <file name="apertium-sme-nob.sme-nob.t2x"/>
        <file name="sme-nob.t2x.bin"/>
      </program>
      <program name="apertium-interchunk" debug-suff="interchunk2">
        <file name="apertium-sme-nob.sme-nob.t3x"/>
        <file name="sme-nob.t3x.bin"/>
      </program>
      <program name="lt-proc $1">
        <file name="sme-nob.autogen.bin"/>
      </program>
    </pipeline>
  </mode>

would make all these modes automatically:

  • sme-nob.mode
  • sme-nob-morph.mode
  • sme-nob-disam.mode
  • sme-nob-syntax.mode
  • sme-nob-pretransfer.mode
  • sme-nob-biltrans.mode
  • sme-nob-lex.mode
  • sme-nob-chunker.mode
  • sme-nob-interchunk1.mode
  • sme-nob-interchunk2.mode
  • sme-nob-postchunk.mode
  • sme-nob-dgen.mode

(but only install the sme-nob.mode), where -disam, -syntax, -interchunk1, -interchunk2 are manually specified names, and the rest are default names based on program names.

Modes hacks

Statistics mode

In order to get some statistical information about translations made using Apertium, we've hacked the main translation mode, pausing the pipeline just after disambiguation and saving the output into a temp file. After that, pipeline is resumed with temp file as stdin.

As an example, you can see the /broken/ pipeline for ca-es, installed as ca-es-estadistiques.mode

/usr/local/bin/lt-proc /usr/local/share/apertium/apertium-es-ca/ca-es.automorf.bin > $LOGSDIR$SEC.tmp;
/usr/local/bin/apertium-tagger -g /usr/local/share/apertium/apertium-es-ca/ca-es.prob < $LOGSDIR$SEC.tmp \
|/usr/local/bin/apertium-pretransfer|/usr/local/bin/apertium-transfer /usr/local/share/apertium/apertium-es-ca/apertium-es-ca.trules-ca-es.xml \
/usr/local/share/apertium/apertium-es-ca/trules-ca-es.bin  /usr/local/share/apertium/apertium-es-ca/ca-es.autobil.bin \
|/usr/local/bin/lt-proc $1 /usr/local/share/apertium/apertium-es-ca/ca-es.autogen.bin \
|/usr/local/bin/lt-proc -p /usr/local/share/apertium/apertium-es-ca/ca-es.autogen.bin

And an example of calling apertium with this mode would be the following

LOGSDIR=~/logs/apertium/; SEC=`date +%s`;
echo "Ara Apertium permet extraure estadístiques" | apertium ca-es-estadistiques

In that example, $LOGSDIR is a folder where the logs will be saved, and $SEC is an unique ID for that log.

When translation is done, we can process the log created in order to get statistics.

Mixed modes

See Mixed modes

See also