Part-of-speech Tagging and Chunking

Question 1

Part of speech tagging

Answer 1

Map words to part of speech tags

Question 2

Part of speech tagging approaches

Answer 2

• Default tagger
• Morphology tagger
• Lookup tagger
• N-gram tagger

Question 3

Default tagger

Answer 3

Set everything to a noun

Good baseline

Question 4

Morphology tagger

Answer 4

Regular expression tagger

Question 5

Lookup tagger

Answer 5

Store a dictionary of the most hundred frequent words and their most frequent tags

Question 6

N-gram tagger

Answer 6

Look at sequences of words and tags

Deals with ambiguity and words that have more than one part-of-speech tag

Allows > 90% accuracy

Question 7

Hidden Markov Models, description

Answer 7

For the given sequence of words, what is the most likely sequence of tags?

Probabilistic approach

Want
t1,…,tn = argmax P( t1,…,tn | w1, …, wn)

Question 8

Hidden Markov Models, formula

Answer 8

For each tag sequence, maximize
the product of P (Ti | Ti-1) * P(Wi | Ti)

Probability of tag given previous tag * probability of word given tag

Question 9

Hidden Markov Models, two tricks

Answer 9

• Bayes’ Rule

P( t1,…,tn | w1, …, wn)
= P( w1,…,wn | t1, …, tn) * P(t1,…tn) / P(w1,…wn)

• Markov Assumption

Each word depends on its own tag
P( w1,…,wn | t1, …, tn)
= P(w1 | t1) * … * P(wn | tn)

Look back two tags
P(t1,…tn) = P(t1) * P(t2 | t1) * P (t3 | t2)

Ignore the probability of the words

Question 10

Dynamic Programming

Answer 10

Store intermediate results for shared subsequences rather than recomputing

Question 11

Chunking

Answer 11

Finding sequences of part-of-speech tags, such as noun phrases