NLP | Combining NGram Taggers

NgramTagger has 3 subclasses

• UnigramTagger
• BigramTagger
• TrigramTagger

BigramTagger subclass uses previous tag as part of its context
TrigramTagger subclass uses the previous two tags as part of its context.

ngram – It is a subsequence of n items.
Idea of NgramTagger subclasses :

• By looking at the previous words and P-O-S tags, part-of-speech tag for the current word can be guessed.
• Each tagger maintains a context dictionary (ContextTagger parent class is used to implement it).
• This dictionary is used to guess that tag based on the context.
• The context is some number of previous tagged words in the case of NgramTagger subclasses.

Code #1 : Working of Bigram tagger

# Loading Libraries
from nltk.tag import DefaultTagger
from nltk.tag import BigramTagger

from nltk.corpus import treebank

# initializing training and testing set
train_data = treebank.tagged_sents()[:3000]
test_data = treebank.tagged_sents()[3000:]

# Tagging
tag1 = BigramTagger(train_data)

# Evaluation
tag1.evaluate(test_data)

Output :

0.11318799913662854

Code #2 : Working of Trigram tagger

# Loading Libraries
from nltk.tag import DefaultTagger
from nltk.tag import TrigramTagger

from nltk.corpus import treebank

# initializing training and testing set
train_data = treebank.tagged_sents()[:3000]
test_data = treebank.tagged_sents()[3000:]

# Tagging
tag1 = TrigramTagger(train_data)

# Evaluation
tag1.evaluate(test_data)

Output :

0.06876753723289446

Code #3 : Collectively using Unigram, Bigram and Trigram tagger.

# Loading Libraries

from nltk.tag import TrigramTagger
from tag_util import backoff_tagger
from nltk.corpus import treebank

# initializing training and testing set
train_data = treebank.tagged_sents()[:3000]
test_data = treebank.tagged_sents()[3000:]

backoff = DefaultTagger('NN')
tag = backoff_tagger(train_sents,
                     [UnigramTagger, BigramTagger, TrigramTagger],
                     backoff = backoff)

tag.evaluate(test_sents)

Output :

0.8806820634578028

How it works ?

• The backoff_tagger function creates an instance of each tagger class.
• It gives previous tagger and train_sents as a backoff.
• The order of tagger classes is important: In the code above the first class is UnigramTagger and hence, it will be trained first and given the initial backoff tagger (the DefaultTagger).
• This tagger then becomes the backoff tagger for the next tagger class.
• Final tagger returned will be an instance of the last tagger class – TrigramTagger.

Code #4 : Proof

print (tagger._taggers[-1] == backoff)

print ("\n", isinstance(tagger._taggers[0], TrigramTagger))

print ("\n", isinstance(tagger._taggers[1], BigramTagger))

Output :

True

True

True