Difference between revisions of "Corpus based preposition selection - HOWTO"

From Apertium
Jump to navigation Jump to search
 
(23 intermediate revisions by 2 users not shown)
Line 1: Line 1:
  +
{{TOCD}}
 
The general algorithm for performing corpus based preposition selection is as follows:
 
The general algorithm for performing corpus based preposition selection is as follows:
 
* Download a parallel corpus
 
* Download a parallel corpus
 
* Extract patterns which contain prepositions from the source-language corpus
 
* Extract patterns which contain prepositions from the source-language corpus
 
* Align the patterns to their translations in the target-language corpus
 
* Align the patterns to their translations in the target-language corpus
* Extract the features and label (the correct preposition from the target-language corpus) for classification.
+
* Extract the features and corresponding labels (the correct preposition from the target-language corpus) for classification.
 
* Train a model
 
* Train a model
* Use the trained model in the pipeline
+
* Use the trained model in the translation pipeline
   
 
The general toolkit for performing these tasks can be found [https://apertium.svn.sourceforge.net/svnroot/apertium/branches/gsoc2012/fpetkovski/morph-parser/ here].
 
The general toolkit for performing these tasks can be found [https://apertium.svn.sourceforge.net/svnroot/apertium/branches/gsoc2012/fpetkovski/morph-parser/ here].
  +
=== Extracting training data for your classifier ===
 
  +
== The toolkit ==
For the purpose of extracting training data for your classifier, you can use the preposition-extraction tool.
 
  +
=== Preposition Extraction ===
  +
The tool for preposition extraction takes a stream in the format ^lemma<tags>$ ^lemma<tags>$ on standard input and outputs a list of extracted patterns which are later used in the alignment process.
  +
The program will match patterns in the form of: (n|vblex) (pr) (adj | det)* (n|vblex)
  +
  +
Example:
  +
<pre>
  +
echo "^Косово<np><top><nt><sg><nom>$ ^испитува<vblex><imperf><tv><pres><p3><sg>$ ^процес<n><m><sg><nom><def>$ ^на<pr>$ ^приватизација<n><f><sg><nom><ind><@P←>$^.<sent>$
  +
" | ./preposition-extraction.bin
  +
output:
  +
^процес<n><m><sg><nom><def>$.. ^на<pr>$.. ^приватизација<n><f><sg><nom><ind><@P←>$
  +
</pre>
  +
  +
These patterns will later be translated using apertium and matched in the target-language corpus.
  +
  +
=== Preposition alignment ===
  +
  +
The patterns that were extracted in the previous process need to be aligned to their translations in the target language so the correct preposition can be extracted as a label. This way a training set can be created.
  +
  +
Usage: ./preposition-aligner.bin -s source-file -t target-file -tr translations-file -n number-of-features [-asrc allow-source] [-atrg allow-target]
  +
Options:
  +
-sl, --source a file with the sorce language sentences
  +
-tl, --target a file with the target language sentences
  +
-tr, --translations a file with the translations of the files
  +
-n number of features
  +
-atrg, --allow-only-target path to a file containing the allowed source-language prepositions
  +
-asrc, --allow-only-source path to a file containing the allowed target-language prepositions
  +
  +
Example:
  +
<pre>
  +
preposition-aligner.bin -sl training-patterns-mk -tl training-patterns-en -tr extracted-patterns-nodef-train -n 2 | head -n 10
  +
  +
output:
  +
пара--од$од--продажба$from
  +
полза--од$од--приватизација$from
  +
ефикасност--како$како--концепт$as a
  +
префрли--во$во--банка$to
  +
резултат--на$на--приватизација$of
  +
земја--во$во--регион$in
  +
план--за$за--развој$for
  +
злоупотреба--на$на--положба$of
  +
стави--под$под--контрола$under
  +
сметка--со$со--содржина$with
  +
</pre>
  +
  +
== The complete training phase ==
  +
The training phase is done in two steps:
  +
* Extract patterns in the form of (n|vblex) (pr) (adj | det)* (n|vblex) from the source language corpus and translate the using apertium to the target language.
  +
* Go through the source language file again, matching those same patterns and trying to find their translations in the target language. If a translation is found, extract the features and correct preposition as a training-set example. You could theoretically choose any combination of features, however, the tools provided so far support only 3 different combinations:
  +
** 1-feature model -- extract an example in the following format: sl_nv1-sl_pr-sl_nv2<delimiter>tl_pr
  +
** 2-feature model -- extract an example in the following format: sl_nv1-sl_pr<delimiter>-sl_nv2<delimiter>tl_pr
  +
** 3-feature model -- extract an example in the following format: sl_nv1<delimiter>sl_pr<delimiter>sl_nv2<delimiter>tl_pr
  +
  +
sl_nv1, sl_nv1 and sl_pr stand for the first and second source language noun or verb, and for the source language preposition. tl_pr stands for the target language preposition, and that is the actual label used in classification
  +
  +
== Example ==
  +
This is an example script that uses these two tools to create a training set:
  +
<pre>
  +
cat setimes.mk | head -n 150000 | apertium -d ~/Apertium/apertium-mk-en mk-en-pretransfer > training-patterns-mk
  +
cat training-patterns-mk | ~/Apertium/fpetkovski/morph-parser/preposition-extraction \
  +
| lt-proc -g ~/Apertium/apertium-mk-en/en-mk.autogen.bin \
  +
| apertium -d ~/Apertium/apertium-mk-en/ mk-en-postchunk > extracted-patterns-train
  +
  +
# In Macedonian, the definiteness of the noun is encoded in the noun itself,
  +
# while in English it is denoted by the article before the noun.
  +
# As a result, the extracted patterns after translation can have up to 5 tokens instead of the desired three.
  +
# That's why we want to remove the articles from the translated patterns.
  +
  +
# remove articles
  +
cat extracted-patterns-train | sed 's/[ ]*\^[Tt]he<[^\$]*\$[ ]*//g' > extracted-patterns-nodef-train;
  +
  +
# tag the tl set
  +
cat setimes.en | head -n 150000 | apertium -d ~/Apertium/apertium-en-es en-es-tagger > training-patterns-en
  +
  +
# alignment
  +
preposition-aligner -sl training-patterns-mk -tl training-patterns-en -tr extracted-patterns-nodef-train -n 2 > training-set
  +
</pre>
  +
  +
And the output:
  +
  +
<pre>
  +
  +
head -n 10 training-set
  +
  +
пара--од$од--продажба$from
  +
полза--од$од--приватизација$from
  +
префрли--во$во--банка$to
  +
резултат--на$на--приватизација$of
  +
земја--во$во--регион$in
  +
план--за$за--развој$for
  +
злоупотреба--на$на--положба$of
  +
сметка--со$со--содржина$with
  +
вработи--во$во--медиум$in
  +
слобода--на$на--говор$of
  +
</pre>
  +
where the '$' character here serves as a delimiter. <br/>
  +
Now you have a training set which you can use to train a classifier.
  +
  +
It should be noted that you can specify a list of both source-language and target-language prepositions that you want to allow in your training set. If such a list is specified for source-language prepositions, then patterns that do not contain those prepositions will not be extracted for the training set.
  +
  +
If a list is specified for target-language prepositions, then for those prepositions which are not in the list a new class will be created (class 'other'). This means that it will be left up to apertium to decide how to translate the source-language preposition if the classifier labels some example as a member of the class 'other'.
  +
  +
It is recommended that you use such a 'white-list' for target-language prepositions, and put the most common prepositions there, since for the less common ones there won't be enough coverage for those classes to be learned.
  +
  +
== Applying the model ==
  +
  +
In order avoid depending on an external library, a naive bayes classifier was manually constructed, since that was the one used in the experiments. It can be found in the morph-parser directory and it can be used for training a model.
  +
  +
Once you have trained a model, you can insert it in the pipeline so it can be applied in the translation process. For the purpose of applying a naive bayes model, the preposition-selection tool was created which takes a biltrans output as an stream on standard input.
  +
  +
Usage: ./preposition-selection.bin [ -t | -l ] data_file -d delimiter
  +
Options:
  +
-t, --train use the data_file to train a model
  +
-l, --load load a trained model from the data_file
  +
-d, --delimiter sets the delimiter
  +
  +
=== Example ===
  +
<pre>
  +
cat ~/Desktop/setimes-en-mk-nikola/setimes.mk.fixed | tail -n 50000 | apertium -d ~/Apertium/apertium-mk-en mk-en-biltrans |\
  +
./preposition-selection --train "training-set" -d "$" | ./biltrans-to-end
  +
</pre>
  +
or
  +
<pre>
  +
cat ~/Desktop/setimes-en-mk-nikola/setimes.mk.fixed | tail -n 50000 | apertium -d ~/Apertium/apertium-mk-en mk-en-biltrans |\
  +
./preposition-selection --load "model" -d "$" | ./biltrans-to-end
  +
</pre>
  +
  +
The biltrans-to-end script should go through the rest of the pipeline, executing the transfer and generation processes.
  +
  +
For mk-en:
  +
<pre>
  +
/usr/local/bin/apertium-transfer -b /home/philip/Apertium/apertium-mk-en/apertium-mk-en.mk-en.t1x /home/philip/Apertium/apertium-mk-en/mk-en.t1x.bin \
  +
|/usr/local/bin/apertium-interchunk /home/philip/Apertium/apertium-mk-en/apertium-mk-en.mk-en.t2x /home/philip/Apertium/apertium-mk-en/mk-en.t2x.bin \
  +
|/usr/local/bin/apertium-postchunk /home/philip/Apertium/apertium-mk-en/apertium-mk-en.mk-en.t3x /home/philip/Apertium/apertium-mk-en/mk-en.t3x.bin \
  +
| sed 's/\[/\\[/g' | lt-proc -g ~/Apertium/apertium-mk-en/mk-en.autogen.bin \
  +
| lt-proc -p ~/Apertium/apertium-mk-en/mk-en.autopgen.bin | apertium-retxt
  +
</pre>

Latest revision as of 13:15, 23 August 2012

The general algorithm for performing corpus based preposition selection is as follows:

  • Download a parallel corpus
  • Extract patterns which contain prepositions from the source-language corpus
  • Align the patterns to their translations in the target-language corpus
  • Extract the features and corresponding labels (the correct preposition from the target-language corpus) for classification.
  • Train a model
  • Use the trained model in the translation pipeline

The general toolkit for performing these tasks can be found here.

The toolkit[edit]

Preposition Extraction[edit]

The tool for preposition extraction takes a stream in the format ^lemma<tags>$ ^lemma<tags>$ on standard input and outputs a list of extracted patterns which are later used in the alignment process. The program will match patterns in the form of: (n|vblex) (pr) (adj | det)* (n|vblex)

Example:

echo "^Косово<np><top><nt><sg><nom>$ ^испитува<vblex><imperf><tv><pres><p3><sg>$ ^процес<n><m><sg><nom><def>$ ^на<pr>$ ^приватизација<n><f><sg><nom><ind><@P←>$^.<sent>$
" | ./preposition-extraction.bin 
output:
^процес<n><m><sg><nom><def>$.. ^на<pr>$.. ^приватизација<n><f><sg><nom><ind><@P←>$

These patterns will later be translated using apertium and matched in the target-language corpus.

Preposition alignment[edit]

The patterns that were extracted in the previous process need to be aligned to their translations in the target language so the correct preposition can be extracted as a label. This way a training set can be created.

Usage: ./preposition-aligner.bin -s source-file -t target-file -tr translations-file -n number-of-features [-asrc allow-source] [-atrg allow-target] Options: -sl, --source a file with the sorce language sentences -tl, --target a file with the target language sentences -tr, --translations a file with the translations of the files -n number of features -atrg, --allow-only-target path to a file containing the allowed source-language prepositions -asrc, --allow-only-source path to a file containing the allowed target-language prepositions

Example:

preposition-aligner.bin -sl training-patterns-mk -tl training-patterns-en -tr extracted-patterns-nodef-train -n 2 | head -n 10

output:
пара--од$од--продажба$from
полза--од$од--приватизација$from
ефикасност--како$како--концепт$as a
префрли--во$во--банка$to
резултат--на$на--приватизација$of
земја--во$во--регион$in
план--за$за--развој$for
злоупотреба--на$на--положба$of
стави--под$под--контрола$under
сметка--со$со--содржина$with

The complete training phase[edit]

The training phase is done in two steps:

  • Extract patterns in the form of (n|vblex) (pr) (adj | det)* (n|vblex) from the source language corpus and translate the using apertium to the target language.
  • Go through the source language file again, matching those same patterns and trying to find their translations in the target language. If a translation is found, extract the features and correct preposition as a training-set example. You could theoretically choose any combination of features, however, the tools provided so far support only 3 different combinations:
    • 1-feature model -- extract an example in the following format: sl_nv1-sl_pr-sl_nv2<delimiter>tl_pr
    • 2-feature model -- extract an example in the following format: sl_nv1-sl_pr<delimiter>-sl_nv2<delimiter>tl_pr
    • 3-feature model -- extract an example in the following format: sl_nv1<delimiter>sl_pr<delimiter>sl_nv2<delimiter>tl_pr

sl_nv1, sl_nv1 and sl_pr stand for the first and second source language noun or verb, and for the source language preposition. tl_pr stands for the target language preposition, and that is the actual label used in classification

Example[edit]

This is an example script that uses these two tools to create a training set:

cat setimes.mk | head -n 150000 | apertium -d ~/Apertium/apertium-mk-en mk-en-pretransfer > training-patterns-mk
cat training-patterns-mk | ~/Apertium/fpetkovski/morph-parser/preposition-extraction \
| lt-proc -g ~/Apertium/apertium-mk-en/en-mk.autogen.bin \
| apertium -d ~/Apertium/apertium-mk-en/ mk-en-postchunk > extracted-patterns-train

# In Macedonian, the definiteness of the noun is encoded in the noun itself, 
# while in English it is denoted by the article before the noun. 
# As a result, the extracted patterns after translation can have up to 5 tokens instead of the desired three. 
# That's why we want to remove the articles from the translated patterns.

# remove articles
cat extracted-patterns-train | sed 's/[ ]*\^[Tt]he<[^\$]*\$[ ]*//g' > extracted-patterns-nodef-train;

# tag the tl set
cat setimes.en | head -n 150000 | apertium -d ~/Apertium/apertium-en-es en-es-tagger > training-patterns-en

# alignment
preposition-aligner -sl training-patterns-mk -tl training-patterns-en -tr extracted-patterns-nodef-train -n 2 > training-set

And the output:


head -n 10 training-set

пара--од$од--продажба$from
полза--од$од--приватизација$from
префрли--во$во--банка$to
резултат--на$на--приватизација$of
земја--во$во--регион$in
план--за$за--развој$for
злоупотреба--на$на--положба$of
сметка--со$со--содржина$with
вработи--во$во--медиум$in
слобода--на$на--говор$of

where the '$' character here serves as a delimiter.
Now you have a training set which you can use to train a classifier.

It should be noted that you can specify a list of both source-language and target-language prepositions that you want to allow in your training set. If such a list is specified for source-language prepositions, then patterns that do not contain those prepositions will not be extracted for the training set.

If a list is specified for target-language prepositions, then for those prepositions which are not in the list a new class will be created (class 'other'). This means that it will be left up to apertium to decide how to translate the source-language preposition if the classifier labels some example as a member of the class 'other'.

It is recommended that you use such a 'white-list' for target-language prepositions, and put the most common prepositions there, since for the less common ones there won't be enough coverage for those classes to be learned.

Applying the model[edit]

In order avoid depending on an external library, a naive bayes classifier was manually constructed, since that was the one used in the experiments. It can be found in the morph-parser directory and it can be used for training a model.

Once you have trained a model, you can insert it in the pipeline so it can be applied in the translation process. For the purpose of applying a naive bayes model, the preposition-selection tool was created which takes a biltrans output as an stream on standard input.

Usage: ./preposition-selection.bin [ -t | -l ] data_file -d delimiter Options: -t, --train use the data_file to train a model -l, --load load a trained model from the data_file -d, --delimiter sets the delimiter

Example[edit]

cat ~/Desktop/setimes-en-mk-nikola/setimes.mk.fixed | tail -n 50000 | apertium -d ~/Apertium/apertium-mk-en mk-en-biltrans |\
./preposition-selection --train "training-set" -d "$" | ./biltrans-to-end

or

cat ~/Desktop/setimes-en-mk-nikola/setimes.mk.fixed | tail -n 50000 | apertium -d ~/Apertium/apertium-mk-en mk-en-biltrans |\
./preposition-selection --load "model" -d "$" | ./biltrans-to-end

The biltrans-to-end script should go through the rest of the pipeline, executing the transfer and generation processes.

For mk-en:

/usr/local/bin/apertium-transfer -b /home/philip/Apertium/apertium-mk-en/apertium-mk-en.mk-en.t1x /home/philip/Apertium/apertium-mk-en/mk-en.t1x.bin \
|/usr/local/bin/apertium-interchunk /home/philip/Apertium/apertium-mk-en/apertium-mk-en.mk-en.t2x  /home/philip/Apertium/apertium-mk-en/mk-en.t2x.bin \
|/usr/local/bin/apertium-postchunk /home/philip/Apertium/apertium-mk-en/apertium-mk-en.mk-en.t3x /home/philip/Apertium/apertium-mk-en/mk-en.t3x.bin \
| sed 's/\[/\\[/g' | lt-proc -g ~/Apertium/apertium-mk-en/mk-en.autogen.bin \
| lt-proc -p ~/Apertium/apertium-mk-en/mk-en.autopgen.bin | apertium-retxt