This is an old revision of the document!
Table of Contents
Lab 5
Try to complete as many problems as you can. Hand in a python code file (fullName_lab5.py) with what you have finished in MySchool before midnight today (17 September).
If you can't manage to complete a particular problem please hand in your incomplete code – comment it out if it produces an error.
1. Down the Garden Path
Lets get started by trying out the POS-Tagger in NLTK. See if you can think of some ambiguous sentences that confuse the tagger – or google for some garden path sentences.
import nltk text = nltk.word_tokenize("The horse raced past the barn fell.") tagged_text = nltk.pos_tag(text) print(tagged_text)
Use nltk.help.upenn_tagset() to look up the meaning of the tags.
nltk.help.upenn_tagset('NN') nltk.help.upenn_tagset('NN.*') #fyi, there is also a lookup function for the tagset used for the Brown corpus: nltk.help.brown_tagset('NN')
See if making minor changes to the wording of the sentences is enough for the tagger to tag it correctly. For example adding that was after horse in the example above.
2. Training and Testing Data and Finding the Baseline
from nltk.corpus import brown
Get the tagged sentences from the Brown corpus and separate them into training (90%) and testing parts (10%). (Note: we will use them also in the two remaining problems). You can limit the corpus to the news category during testing but use the whole corpus for the handin.
Train the Unigram Tagger on the training sentences using Default Tagger as backoff. Use 'NN' as the default tag.
Evaluate the taggers performance on the testing sentences. How well does it do? Do you think this is a fair baseline the measure other taggers against?
3. A Cascade of Taggers
Extend the tagger combination in 5.4 Combining Taggers by defining a TrigramTagger called t3, which backs off to t2 as suggested.
t0 = nltk.DefaultTagger('NN') t1 = nltk.UnigramTagger(brown_train_sents, backoff=t0) t2 = nltk.BigramTagger(brown_train_sents, backoff=t1) print(t2.evaluate(brown_test_sents)) [...]
Evalute both t2 and t3 and report the difference.
4. Most frequent tag of Hapax legomenon
Instead of using the most common tag overall for the Default Tagger some say that hapax legomenon is a better model for unknown words. That is, that words that occur only once in the corpus are likely to be representative of the words that never occur, the unseen words.
See if you can write code to find the most common tag in the set of words that occur only once in the Brown corpus. You might prefer to use brown.tagged_words() here, and even brown.tagged_words(categories='news') during testing.
What is the most common tag of hapax legomenon?