Generating lexical-selection rules from a parallel corpus
This page is deprecated. For further information see: Learning rules from parallel and non-parallel corpora
If you have a parallel corpus, one of the things you can do is generate some lexical selection rules from it, to improve translation of words with more than one possible translation.
You will need[edit]
Here is a list of software that you will need installed:
- Giza++ (or some other word aligner)
- Moses (for making Giza++ less human hostile)
- All the Moses scripts
- lttoolbox
- Apertium
- apertium-lex-tools
Furthermore you'll need:
- an Apertium language pair
- a parallel corpus (see Corpora)
Installing prerequisites[edit]
See Minimal installation from SVN for apertium/lttoolbox.
See Constraint-based lexical selection module for apertium-lex-tools.
For Giza++ and moses-decoder, etc. you can do
$ mkdir ~/smt $ cd ~/smt $ mkdir local # our "install prefix" $ wget https://giza-pp.googlecode.com/files/giza-pp-v1.0.7.tar.gz $ tar xzvf giza-pp-v1.0.7.tar.gz $ cd giza-pp $ make $ mkdir ../local/bin $ cp GIZA++-v2/snt2cooc.out ../local/bin/ $ cp GIZA++-v2/snt2plain.out ../local/bin/ $ cp GIZA++-v2/GIZA++ ../local/bin/ $ cp mkcls-v2/mkcls ../local/bin/ $ git clone https://github.com/moses-smt/mosesdecoder $ cd mosesdecoder/ $ ./bjam
Now e.g. the clean-corpus and train-model scripts referred to below will be in ~/smt/mosesdecoder/scripts/training/clean-corpus-n.perl See http://www.statmt.org/moses/?n=Development.GetStarted if you want to install the binaries to some other directory.
Getting started[edit]
Important: If you don't want through the whole process step by step, you can use the Makefile script provided in the last section of this page.
We're going to do the example with EuroParl and the English to Spanish pair in Apertium.
Given that you've got all the stuff installed, the work will be as follows:
Prepare corpus[edit]
To generate the rules, we need three files,
- The tagged and tokenised source corpus
- The tagged and tokenised target corpus
- The output of the lexical transfer module in the source→target direction, tokenised
These three files should be sentence aligned.
Make a folder called data-en-es. We are going to keep all the generated files there.
The first thing that we need to do is tag both sides of the corpus:
$ nohup cat europarl.en-es.en | apertium-destxt |\ apertium -f none -d /home/fran/source/apertium-en-es en-es-pretransfer > data-en-es/europarl-en-es.tagged.en & $ nohup cat europarl.clean.es | apertium-destxt |\ apertium -f none -d /home/fran/source/apertium-en-es es-en-pretransfer > data-en-es.europarl-en-es.tagged.es &
Then we need to remove the lines with no analyses on and replace blanks within lemmas with a new character (we will use `~`):
$ paste data-en-es/europarl-en-es.tagged.en data-en-es/europarl-en-es.tagged.es | grep '<' | cut -f2 | sed 's/ /~/g' | sed 's/\$[^\^]*/\$ /g' > data-en-es/europarl-en-es.tagged.new.es $ paste data-en-es/europarl-en-es.tagged.en data-en-es/europarl-en-es.tagged.es | grep '<' | cut -f3 | sed 's/ /~/g' | sed 's/\$[^\^]*/\$ /g' > data-en-es/europarl-en-es.tagged.new.en
Next, we need to clean the corpus and remove long sentences. (Make sure you are in the same directory as the one where you have your europarl corpus)
$ perl (path to your mosesdecoder)/scripts/training/clean-corpus-n.perl data-en-es/europarl-en-es.tagged.new es en data-en-es/europarl-en-es.tag-clean 1 40 clean-corpus.perl: processing europarl-v6.es-en.es & .en to data-en-es/europarl-en-es.clean, cutoff 1-40 ..........(100000)... Input sentences: 1786594 Output sentences: 1467708
We're going to cut off the bottom 67,658 for testing (also because Giza++ segfaults somewhere around there).
$ mkdir testing $ tail -67658 data-en-es/europarl-en-es.tag-clean.en > testing/europarl-en-es.tag-clean.67658.en $ tail -67658 data-en-es/europarl-en-es.tag-clean.es > testing/europarl-en-es.tag-clean.67658.es
$ head -1400000 data-en-es/europarl-en-es.tag-clean.en > data-en-es/europarl-en-es.tag-clean.en.new $ head -1400000 data-en-es/europarl-en-es.tag-clean.es > data-en-es/europarl-en-es.tag-clean.es.new $ mv data-en-es/europarl-en-es.tag-clean.en.new data-en-es/europarl-en-es.tag-clean.en $ mv data-en-es/europarl-en-es.tag-clean.es.new data-en-es/europarl-en-es.tag-clean.es
These files are:
data-en-es/europarl-en-es.tag-clean.en: The tagged source language side of the corpusdata-en-es/europarl-en-es.tag-clean.es: The tagged target language side of the corpus
Check that they have the same length:
$ wc -l europarl.* 1400000 data-en-es/europarl-en-es.tag-clean.en 1400000 data-en-es/europarl-en-es.tag-clean.es 2800000 total
Align corpus[edit]
Now we've got the corpus files ready, we can align the corpus using the Moses scripts:
nohup perl (path to your mosesdecoder)/scripts/training/train-model.perl -external-bin-dir \ ~/smt/local/bin -corpus data-en-es/europarl-en-es.tag-clean \ -f en -e es -alignment grow-diag-final-and -reordering msd-bidirectional-fe \ -lm 0:5:/home/fran/corpora/europarl/europarl.lm:0 >log 2>&1 &
Note: Remember to change all the paths in the above command!
You'll need an LM file, but you can copy it from a previous Moses installation. If you don't have one, make an empty file and put a few words in it. We won't be using the LM anyway.
This takes a while, from a few hours to a day. So leave it running and go and make a soufflé, or chop some wood or something.
Extract sentences[edit]
After the sentences are aligned, we need to trim unnecessary tags from the tokens, and generate a biltrans file.
zcat giza.en-es/en-es.A3.final.gz | ~/source/apertium-lex-tools/scripts/giza-to-moses.awk > data-en-es/europarl.phrasetable.en-es
cat data-en-es/europarl.phrasetable.en-es | sed 's/ ||| /\t/g' | cut -f 1 \
| sed 's/~/ /g' | ~/source/apertium-lex-tools/multitrans ~/source/apertium-en-es/en-es.autobil.bin -p -t > tmp1
cat data-en-es/europarl.phrasetable.en-es | sed 's/ ||| /\t/g' | cut -f 2 \
| sed 's/~/ /g' | ~/source/apertium-lex-tools/multitrans ~/source/apertium-en-es/en-es.autobil.bin -p -t > tmp2
cat data-en-es/europarl.phrasetable.en-es | sed 's/ ||| /\t/g' | cut -f 3 > tmp3
cat data-en-es/europarl.phrasetable.en-es | sed 's/ ||| /\t/g' | cut -f 2 \
| sed 's/~/ /g' | ~/source/apertium-lex-tools/multitrans ~/source/apertium-en-es/en-es.autobil.bin -b -t > data-en-es/europarl.clean-biltrans.en-es
paste tmp1 tmp2 tmp3 | sed 's/\t/ ||| /g' > data-en-es/europarl.phrasetable.en-es
rm tmp1 tmp2 tmp3
Then we want to make sure again that our file has the right number of lines:
$ wc -l data-en-es/europarl.phrasetable.en-es 1400000 data-en-es/europarl.phrasetable.en-es
Then we want to extract the sentences where the target language word aligned to a source language word is a possible translation in the bilingual dictionary:
$ ~/source/apertium-lex-tools/scripts/extract-sentences.py data-en-es/europarl-en-es.phrasetable.en-es data-en-es/europarl-en-es.biltrans-tok.en-es \ > data-en-es/europarl-en-es.candidates.en-es
These are basically sentences that we can hope that Apertium might be able to generate.
Extract bilingual dictionary candidates[edit]
Using the phrasetable and the bilingual file we can extract candidates for the bilingual dictionary.
python3 ~/source/apertium-lex-tools/scripts/extract-biltrans-candidates.py data-en-es/europarl-en-es.phrasetable.en-es data-en-es/europarl-en-es.biltrans-tok.en-es > data-en-es/europarl-en-es.biltrans-candidates.en-es 2> data-en-es/europarl-en-es.biltrans-pairs.en-es
where data-en-es/europarl-en-es.biltrans-candidates.en-es contains the generated candidates for the bilingual dictionary.
Extract frequency lexicon[edit]
The next step is to extract the frequency lexicon.
$ python ~/source/apertium-lex-tools/scripts/extract-freq-lexicon.py data-en-es/europarl-en-es.candidates.en-es > data-en-es/europarl-en-es.lex.en-es
This file should look like:
$ cat europarl.lex.en-es | head 31381 union<n> unión<n> @ 101 union<n> sindicato<n> 1 union<n> situación<n> 1 union<n> monetario<adj> 4 slope<n> pendiente<n> @ 1 slope<n> ladera<n>
Where the highest frequency translation is marked with an @.
Note: This frequency lexicon can be used as a substitute for "choosing the most general translation" in your bilingual dictionary.
Generate patterns[edit]
Now we generate the ngrams that we are going to generate the rules from.
$ crisphold=1.5 # ratio of how many times you see the alternative translation compared to the default $ python ~/source/apertium-lex-tools/scripts/ngram-count-patterns.py europarl.lex.en-es data-en-es/europarl-en-es.candidates.en-es $crisphold 2>/dev/null > data-en-es/europarl-en-es.ngrams.en-es
This script outputs lines in the following format:
-language<n> and<cnjcoo> language<n> ,<cm> lengua<n> 2 +language<n> plain<adj> language<n> ,<cm> lenguaje<n> 3 -language<n> language<n> knowledge<n> lengua<n> 4 -language<n> language<n> of<pr> communication<n> lengua<n> 3 -language<n> Community<adj> language<n> .<sent> lengua<n> 5 -language<n> language<n> in~addition~to<pr> their<det><pos> lengua<n> 2 -language<n> every<det><ind> language<n> lengua<n> 2 +language<n> and<cnjcoo> *understandable language<n> lenguaje<n> 2 -language<n> two<num> language<n> lengua<n> 8 -language<n> only<adj> official<adj> language<n> lengua<n> 2
The + and - indicate if this line chooses the most frequent transation (-) or a translation which is not the most frequent (+). The pattern selecting the translation is then shown, followed by the translation and then the frequency.
Filter rules[edit]
Now you can filter the rules, for example by removing rules with conjunctions, or removing rules with unknown words.
Generate rules[edit]
The final stage is to generate the rules,
python3 ~/source/apertium-lex-tools/scripts/ngrams-to-rules.py data-en-es/europarl-en-es.ngrams.en-es $crisphold > data-en-es/europarl-en-es.ngrams.en-es.lrx
Makefile[edit]
For the whole process you can run the following Makefile:
CORPUS=europarl PAIR=en-es SL=en TL=es DATA=/home/philip/Apertium/apertium-en-es LEX_TOOLS=/home/philip/Apertium/apertium-lex-tools SCRIPTS=$(LEX_TOOLS)/scripts MOSESDECODER=/home/philip/mosesdecoder/scripts/training TRAINING_LINES=200000 BIN_DIR=/home/philip/giza-pp/bin LM=/home/philip/Apertium/gsoc2013/giza/dummy.lm crisphold=1 all: data-$(SL)-$(TL)/$(CORPUS).ngrams.$(SL)-$(TL).lrx data-$(SL)-$(TL)/$(CORPUS).biltrans-entries.$(SL)-$(TL) # TAG CORPUS data-$(SL)-$(TL)/$(CORPUS).tagged.$(SL): $(CORPUS).$(PAIR).$(SL) if [ ! -d data-$(SL)-$(TL) ]; then mkdir data-$(SL)-$(TL); fi cat $(CORPUS).$(PAIR).$(SL) | head -n $(TRAINING_LINES) \ | apertium-destxt \ | apertium -f none -d $(DATA) $(SL)-$(TL)-tagger \ | apertium-pretransfer > $@; data-$(SL)-$(TL)/$(CORPUS).tagged.$(TL): $(CORPUS).$(PAIR).$(TL) if [ ! -d data-$(SL)-$(TL) ]; then mkdir data-$(SL)-$(TL); fi cat $(CORPUS).$(PAIR).$(TL) | head -n $(TRAINING_LINES) \ | apertium-destxt \ | apertium -f none -d $(DATA) $(TL)-$(SL)-tagger \ | apertium-pretransfer > $@; # REMOVE LINES WITH NO ANALYSES data-$(SL)-$(TL)/$(CORPUS).tagged.new.$(SL): data-$(SL)-$(TL)/$(CORPUS).tagged.$(SL) data-$(SL)-$(TL)/$(CORPUS).tagged.$(TL) paste data-$(SL)-$(TL)/$(CORPUS).tagged.$(SL) data-$(SL)-$(TL)/$(CORPUS).tagged.$(TL) \ | grep '<' \ | cut -f1 \ | sed 's/ /~/g' | sed 's/$$[^\^]*/$$ /g' > $@ data-$(SL)-$(TL)/$(CORPUS).tagged.new.$(TL): data-$(SL)-$(TL)/$(CORPUS).tagged.$(SL) data-$(SL)-$(TL)/$(CORPUS).tagged.$(TL) paste data-$(SL)-$(TL)/$(CORPUS).tagged.$(SL) data-$(SL)-$(TL)/$(CORPUS).tagged.$(TL) \ | grep '<' \ | cut -f2 \ | sed 's/ /~/g' | sed 's/$$[^\^]*/$$ /g' > $@ # CLEAN data-$(SL)-$(TL)/$(CORPUS).tag-clean.$(SL) data-$(SL)-$(TL)/$(CORPUS).tag-clean.$(TL): data-$(SL)-$(TL)/$(CORPUS).tagged.new.$(SL) data-$(SL)-$(TL)/$(CORPUS).tagged.new.$(TL) perl $(MOSESDECODER)/clean-corpus-n.perl data-$(SL)-$(TL)/$(CORPUS).tagged.new $(SL) $(TL) data-$(SL)-$(TL)/$(CORPUS).tag-clean 1 40; # ALIGN model: data-$(SL)-$(TL)/$(CORPUS).tag-clean.$(SL) data-$(SL)-$(TL)/$(CORPUS).tag-clean.$(TL) -perl $(MOSESDECODER)/train-model.perl -external-bin-dir $(BIN_DIR) -corpus data-$(SL)-$(TL)/$(CORPUS).tag-clean \ -f $(TL) -e $(SL) -alignment grow-diag-final-and -reordering msd-bidirectional-fe \ -lm 0:5:$(LM):0 2>&1 # EXTRACT AND TRIM data-$(SL)-$(TL)/$(CORPUS).phrasetable.$(SL)-$(TL): model zcat giza.$(SL)-$(TL)/$(SL)-$(TL).A3.final.gz | $(SCRIPTS)/giza-to-moses.awk > $@ cat data-$(SL)-$(TL)/$(CORPUS).phrasetable.$(SL)-$(TL) | sed 's/ ||| /\t/g' | cut -f 1 \ | sed 's/~/ /g' | $(LEX_TOOLS)/multitrans $(DATA)/$(TL)-$(SL).autobil.bin -p -t > tmp1 cat data-$(SL)-$(TL)/$(CORPUS).phrasetable.$(SL)-$(TL) | sed 's/ ||| /\t/g' | cut -f 2 \ | sed 's/~/ /g' | $(LEX_TOOLS)/multitrans $(DATA)/$(SL)-$(TL).autobil.bin -p -t > tmp2 cat data-$(SL)-$(TL)/$(CORPUS).phrasetable.$(SL)-$(TL) | sed 's/ ||| /\t/g' | cut -f 3 > tmp3 cat data-$(SL)-$(TL)/$(CORPUS).phrasetable.$(SL)-$(TL) | sed 's/ ||| /\t/g' | cut -f 2 \ | sed 's/~/ /g' | $(LEX_TOOLS)/multitrans $(DATA)/$(SL)-$(TL).autobil.bin -b -t > data-$(SL)-$(TL)/$(CORPUS).clean-biltrans.$(PAIR) paste tmp1 tmp2 tmp3 | sed 's/\t/ ||| /g' > data-$(SL)-$(TL)/$(CORPUS).phrasetable.$(SL)-$(TL) rm tmp1 tmp2 tmp3 # SENTENCES data-$(SL)-$(TL)/$(CORPUS).candidates.$(SL)-$(TL): data-$(SL)-$(TL)/$(CORPUS).phrasetable.$(SL)-$(TL) data-$(SL)-$(TL)/$(CORPUS).clean-biltrans.$(PAIR) python3 $(SCRIPTS)/extract-sentences.py data-$(SL)-$(TL)/$(CORPUS).phrasetable.$(SL)-$(TL) \ data-$(SL)-$(TL)/$(CORPUS).clean-biltrans.$(PAIR) > $@ 2>/dev/null # FREQUENCY LEXICON data-$(SL)-$(TL)/$(CORPUS).lex.$(SL)-$(TL): data-$(SL)-$(TL)/$(CORPUS).candidates.$(SL)-$(TL) python $(SCRIPTS)/extract-freq-lexicon.py data-$(SL)-$(TL)/$(CORPUS).candidates.$(SL)-$(TL) > $@ 2>/dev/null # BILTRANS CANDIDATES data-$(SL)-$(TL)/$(CORPUS).biltrans-entries.$(SL)-$(TL): data-$(SL)-$(TL)/$(CORPUS).phrasetable.$(SL)-$(TL) data-$(SL)-$(TL)/$(CORPUS).clean-biltrans.$(PAIR) python3 $(SCRIPTS)/extract-biltrans-candidates.py data-$(SL)-$(TL)/$(CORPUS).phrasetable.$(SL)-$(TL) data-$(SL)-$(TL)/$(CORPUS).clean-biltrans.$(PAIR) \ > $@ 2>/dev/null # NGRAM PATTERNS data-$(SL)-$(TL)/$(CORPUS).ngrams.$(SL)-$(TL): data-$(SL)-$(TL)/$(CORPUS).lex.$(SL)-$(TL) data-$(SL)-$(TL)/$(CORPUS).candidates.$(SL)-$(TL) python $(SCRIPTS)/ngram-count-patterns.py data-$(SL)-$(TL)/$(CORPUS).lex.$(SL)-$(TL) data-$(SL)-$(TL)/$(CORPUS).candidates.$(SL)-$(TL) $(crisphold) 2>/dev/null > $@ # NGRAMS TO RULES data-$(SL)-$(TL)/$(CORPUS).ngrams.$(SL)-$(TL).lrx: data-$(SL)-$(TL)/$(CORPUS).ngrams.$(SL)-$(TL) python3 $(SCRIPTS)/ngrams-to-rules.py data-$(SL)-$(TL)/$(CORPUS).ngrams.$(SL)-$(TL) $(crisphold) > $@ 2>/dev/null
