Homework 2

• Please study lec_05, lec_06, lec_07, and lec_08 for this homework

• There are three parts in the homework:

  Part 1. practice encode, decode, and unicode
  Part 2. practice using regular expression to extract information
  Part 3. practice Part-of-Speech tagging

• The deadline is: 2021.03.06 10:00pm

• Please run your code in Google Colab to avoid unnecessary confusion and the expected outputs are shown below each code cell.

  Write your code in the code cell commented as "write your code here / fill in the function"
  Please do not remove the comment "write your code here / fill in the function"
  Please do not change variable names

import nltk

nltk.download('book') # download the data used in nltk book

Part 1: practice encoding, decoding, and unicode

# read the polish text file

file = nltk.data.find('corpora/unicode_samples/polish-lat2.txt')

f = open(file, encoding='latin2') # tell the function that this file is encoded with latin2

f_lines = f.readlines()

for line in f_lines:

    line = line.strip() # remove the leading and the trailing characters (e.g., space, \n)

    print(line)

• check the unicode representation of each line

# write your code here

• Print letters that are not in the ASCII range
  • hint: ord(the letter) > 127
  • print the following information of the letters:
    • letter
    • the byte representation of the letter encoded with 'latin2'
    • the byte representation of the letter encoded with 'ISO-8859-2'
    • the byte representation of the letter encoded with 'utf8'
    • the Unicode representation
    • the name assigned to this letter

# write your code here

• Extract words consisting non-alphabet letters
  • non-alphabet letters such as: ń,ó
  • words consisting non-alphabet letters such as: państwowa

from nltk import word_tokenize

special_letters = [ ]  # define the list of special letters from the previous output 

# write your code here

# note: "śląsk" appears twice because it contains two special characters "ś" and "ą", it's fine if you want to remove duplicated words.

Part 2: access a webpage and extract required information using regular expressions

# get contents from a URL

from urllib import request

html = request.urlopen('http://nltk.org/').read().decode('utf8') 

print(html)

# use BeautifulSoup to pull text data from the HTML file

from bs4 import BeautifulSoup

raw_text =  # write your code here

print(raw_text)

• Write a regular expression to extract digits from the text

  • the order of elements in the output list doesn't matter

• hints:

  • re.findall()
  • think carefully about the patterns of these digits
    • one or more digits
    • digits separated by dot

import re

# write your code here

• Write a regular expression to extract upper-cased sequence of letters from the text
  • all letters in the sequence are in upper-case format
  • the sequence should have at least two upper-case letters
  • use set() function to remove repeated elements
  • the order of elements in the output doesn't matter

# write your code here

• Write a regular expression to substitute 'NLP' to 'Natural Language Processing' in the given sentence
  • re.sub()

sentence = 'NLTK provides wrappers for industrial-strength NLP libraries'

re.sub() # fill the function

• Write a regular expression to find words ending in ing
  • use set() function to remove repeated elements
  • the order of words in the output doesn't matter

# write your code here

• Write a regular expression to extract the url
  • re.search()

# write your code here

Part 3: practice POS tagging

text = "John's big idea isn't all that bad" 

• For the given sentence, check the POS tag with tagset=None

# write your code here

• Check the POS tag for the given sentence with tagset = 'universal'

# write your code here

# check the specific meaning of each POS tag in the Penn Treebank tagset

nltk.help.upenn_tagset()

• Specify the training data and test data
  • training data: brown corpus with news category
  • testing data: bron corpus with humor category

from nltk.corpus import brown

train_sents = # write your code here

test_sents = # write your code here

print("%d training sentences, %d testing sentences" % (len(train_sents), len(test_sents)))

• train a default tagger using the news category of brown corpus
  • hint:
    • find the most frequent tag of the news category
    • train a DefaultTagger with the most frequent tag

• Get the most frequent tag for the brown corpus training data (categories = 'news')

# write your code here

• Train a DefaultTagger with the most frequent tag

default_tagger =  # write your code here

• Apply the DefaultTagger to tag the given text

text = "John's big idea isn't all that bad" 

default_tagger.tag() # fill in the tag function

• Evaluate the DefaultTagger's performance on testing data

default_tagger.evaluate() # fill in the evaluate function

• Train a UnigramTagger with the training data (categories='news')

• Get the top-n most frequent words and the most frequent tag for each word (in the training data)

def topn_frequent_tags(topn):

    """

    - get topn frequent words

    - for each frequent word, get its most frequent tag

    - return: dict{word}=tag 

    """

    # write your code here

    return most_freq_wordtags

• The top-10 most frequent word and the corresponding tags in training data

topn_frequent_tags(topn=10)

• Build a UnigramTagger with the top-100 most frequent word tags

unigram_tagger_top100 = # write your code here

• Evaluate the performance of the unigram_tagger_top100 on test data

# write your code here

• Build a UnigramTagger with the top-1000 most frequent word tags

unigram_tagger_top1000 = # write your code here

• Evaluate the performance of the unigram_tagger_top1000 on test data

# write your code here

• Build a UnigramTagger with all the tagged training sentences

unigram_tagger = # write your code here

• Evaluate the performance of the unigram_tagger on test data

# write your code here

• Build a BigramTagger with all the tagged training sentences and evaluate its performance on test data

t2 = # write your code here

t2.evaluate(test_sents)

• Apply the t2 tagger on the given sentence

t2.tag() # fill in the tag function

• Add the UnigramTagger as a backoff for the BigramTagger and evaluate its performance on test data

t2_t1 = # write your code here

t2_t1.evaluate() # fill in the evaluate function

• Apply the t2_t1 tagger on the given sentence

t2_t1.tag() # fill in the tag function

• Add the DefaultTagger as a backoff for the BigramTagger and evaluate its performance on test data

t2_default = # write your code here

t2_default.evaluate() # fill in the evaluate function

• Apply the t2_default tagger on the given sentence

t2_default.tag() # fill in the tag function

• Summarize the performance of your taggers

print("The accuracy of DefaultTagger is %.3f" % default_tagger.evaluate(test_sents) )

print("The accuracy of UnigramTagger (trained with the top-100 most frequent word tags) is %.3f" % unigram_tagger_top100.evaluate(test_sents))

print("The accuracy of UnigramTagger (trained with the top-1000 most frequent word tags) is %.3f" % unigram_tagger_top1000.evaluate(test_sents))

print("The accuracy of UnigramTagger (trained with all tagged training data) is %.3f" % unigram_tagger.evaluate(test_sents))

print("The accuracy of BigramTagger is %.3f" % t2.evaluate(test_sents))

print("The accuracy of BigramTagger (with UnigramTagger as backoff) is %.3f" % t2_t1.evaluate(test_sents))

print("The accuracy of BigramTagger (with DefaultTagger as backoff) is %.3f" % t2_default.evaluate(test_sents))