Training corpus for Brill Tagger in other languages than English

Question

Does anyone know any tagged corpus or lexicon for using the Brill Part-of-Speech (POS) tagger in other languages than English?

Thanks!

Answer 1

If you're using NLTK (http://nltk.org/) and coding with python you can do as follows. You don't even need to code your own brill-tagger since it's already inside the library, http://nltk.org/_modules/nltk/tag/brill.html.

def train_brill_tagger(train_data):
    # Modules for creating the templates.
    from nltk.tag import UnigramTagger
    from nltk.tag.brill import SymmetricProximateTokensTemplate, ProximateTokensTemplate
    from nltk.tag.brill import ProximateTagsRule, ProximateWordsRule
    # The brill tagger module in NLTK.
    from nltk.tag.brill import FastBrillTaggerTrainer
    unigram_tagger = UnigramTagger(train_data)
    templates = [SymmetricProximateTokensTemplate(ProximateTagsRule, (1,1)),
                            SymmetricProximateTokensTemplate(ProximateTagsRule, (2,2)),
                            SymmetricProximateTokensTemplate(ProximateTagsRule, (1,2)),
                            SymmetricProximateTokensTemplate(ProximateTagsRule, (1,3)),
                            SymmetricProximateTokensTemplate(ProximateWordsRule, (1,1)),
                            SymmetricProximateTokensTemplate(ProximateWordsRule, (2,2)),
                            SymmetricProximateTokensTemplate(ProximateWordsRule, (1,2)),
                            SymmetricProximateTokensTemplate(ProximateWordsRule, (1,3)),
                            ProximateTokensTemplate(ProximateTagsRule, (-1, -1), (1,1)),
                            ProximateTokensTemplate(ProximateWordsRule, (-1, -1), (1,1))]

    trainer = FastBrillTaggerTrainer(initial_tagger=unigram_tagger,
                                   templates=templates, trace=3,
                                   deterministic=True)
    brill_tagger = trainer.train(train_data, max_rules=10)
    print
    return brill_tagger

# To train and test using Brown Corpus.
from nltk.corpus import brown
brown_train = list(brown.tagged_sents(categories='news')[:500])
brown_test = list(brown.tagged_sents(categories='news')[500:600])
brown501 = brown.tagged_sents(categories='news')[501]

bt = train_brill_tagger(brown_train)

# To tag one sentence.
print bt.tag(brown501)
print

# To evaluate tagger.
print 'Accuracy of Brill Tagger:', bt.evaluate(brown_test)

There's a list of corpus with corpus readers already pre-coded in NLTK: http://nltk.googlecode.com/svn/trunk/nltk_data/index.xml

Here's an example to apply the brill tagger to a Dutch corpus:

# To train and test using Alpino Corpus (Dutch).
from nltk.corpus import alpino
alpino_tagged_sents = alpino.tagged_sents()
# Split corpus into train/test.
datasize = len(alpino_tagged_sents)
trainsize = int(datasize*90/float(100))
alpino_train = list(alpino_tagged_sents[:trainsize])
alpino_test = list(alpino_tagged_sents[trainsize+1:])
alpinotest1 = [i for i,j in alpino_tagged_sents[trainsize+1]]

bt_nld = train_brill_tagger(alpino_train)
print 'Test sentence:', alpinotest
print bt_nld.tag(alpinotest1)
print
print bt_nld.evaluate(alpino_test)
print

In fact, if you're hardworking enough to read until this point, here's the trick to train a brill tagger in NLTK by just inputting the corpus =)

from nltk.corpus import LazyCorpusLoader

def train_brill_with_corpus(nltkcorpus, train_percent = 90/float(100)):
    if not isinstance(nltkcorpus, LazyCorpusLoader):
        raise NameError("Please use a pre-coded corpus from NLTK.")

    tagged_sents = nltkcorpus.tagged_sents()
    if not tagged_sents: # i.e. tagged_sents == []
        raise NameError("This corpus doesn't have POS tags.")
    trainsize = int(train_percent*len(tagged_sents))
    corpus_train = list(tagged_sents[:trainsize])
    corpus_test = list(tagged_sents[trainsize+1:])
    bt = train_brill_tagger(corpus_train)
    return bt, corpus_test  

Essentially with train_brill_tagger() and train_brill_with_corpus(), you can just do this:

# To train and test using CESS_ESP Corpus (Spanish).
from nltk.corpus import cess_esp
bt_spa, cess_test = train_brill_with_corpus(cess_esp)
cesstest1 = [i for i,j in cess_test[0]]
print 'Test sentence:', cesstest1
print bt_spa.tag(cesstest1); print
print bt_spa.evaluate(cess_test)