Topic 3: Part of Speech Tagging

Question 1

What is part of speech?

Answer 1

word classes/syntactic categories reveals alot of word and its neighbours

example:
noun is likely to be preceeded by determiners and adjective
ver is likely to be proceeded by noun

syntactic structure..nouns are part of noun phrases

Question 2

Application of POS

Answer 2

1. parsing
2. labelling named entities in IR
3. Co-reference resolution..example
4. speech recognition or synthesis

Question 3

What are the 2 categories of POS

Answer 3

closed class and open class

Question 4

Closed class

Answer 4

relatively fix membership, preposition or new preposition are rarely coined

generally function words
occur frequently, short and often have structuring uses in grammer
E.g: of, it, and or you

Question 5

Open Class

Answer 5

noun and verbs..can have new word..
new noun : iPhone
new verb: to fax
are continually being created or borrowed

In English: there are 4 (nouns, verbs, adjectives, adverbs)
many languages have the 4 but not all

Question 6

Open Class: Noun

Answer 6

Give example
Proper noun (Penang samsung, IBM, Intel)
Common noun (cat, pencil
    count noun ( cat, cats)  - can be enumerated grammatically
    mass noun (salt, snow)

Question 7

Open Class: Verb

Answer 7

have inflections (a change of the form of word)
non-third-person-sg (eat), third-person-sg (eats)
progressive (eating), past participle (eaten)

Question 8

Open Class: Adjective

Answer 8

terms for properties and qualities

e.g concepts of color (blue, yellow), age, value

Question 9

Open Class: Adverb

Answer 9

view as modifying something
directional/locative (here, downhill)
degree (very, extremely)
manner (slowly, delicately)
temporal (yesterday, monday)

Question 10

Closed class

Answer 10

list and examples
prepositions
particles
determiners
pronouncs
conjunctions
auxiliary verbs
numerals

Question 11

Closed class: Prepositions

Answer 11

occur before noun phrases. semantically often indicate spatial preposition or temporal relations

Question 12

Closed class: Particles

Answer 12

resembles a preposition or adverb but used in combination with verb for extended meanings
eg over : she turned the paper over

Question 13

Closed class: Phrasal verb

Answer 13

verb and particle that act as single syntactic and/or semantic unit
example:
turn down
find out
go on

Question 14

Closed class: Determiners

Answer 14

the closed class that occurs with nouns, marking the beginning of noun phrase
eg a, an, the, this, that

Question 15

Closed class: Conjunction

Answer 15

word that join 2 phrases, clauses or sentences

eg: and, or but, that

Question 16

Closed class: Pronoun

Answer 16

forms that often act as a kind of shorthand for referring to the same noun phrase or entity or event
eg: you, she, i, it, me, (personal pronoun)

my, your, his, her, its, one’s, our, their (possessive pronoun)

Question 17

Penn Treebank Tagset

Answer 17

45-tag Penn Treebank tagset used to label many corpora

in labelling, POS generally represented by placing the tag after each word delimited by a slash

Question 18

Choosing tagset

Answer 18

Coarse tagsets:
8 parts of speech which is noun, verb, pronoun, preposition, adverb, conjunction, participle, article

Finer grained tagset:
45 tag “ Penn Treebank Tagset”

More finer grained tagset
87-tag tagset from Brown tagset

Question 19

Tagged corpora

Answer 19

corpora labelled with POS tag. cruicial for training statistical tagging algorithm

created by running automated POS tagger on texts then human annotators hand-corrected the tags

words are generally tokenized before tagging

Three main tagged corpora commonly used
Brown corpus
WSJ corpus
Switchboard corpus

Question 20

Role of tokenization in tagging

Answer 20

Treebank tagset assumes tokenization of multipart words done at whitespace process..seperated by whitespaces

Question 21

recap tagging what is POS tagging and the model

Answer 21

process of assigning POS marker to each word in input text

input: sequence o tokenized word and tagset
function: tagging algorithm
output: sequence of tags, one per token

Question 22

Challenges in tagging and give example

Answer 22

• words are ambiguous
  tagging is a task to disambiguate

ambiguity - one of more ambiguous possible POS. goal is to find the correct tag

For example
a verb (book that flight)
a noun (hand me that book)

Question 23

High ambiguous word

Answer 23

that, back, down, put and set..

example
there are 6 different POS for "back"
JJ
NN
VBP
VB
RP
RB

Question 24

Method for POS tagging

Answer 24

rule based…tag based on hand-written disambiguation rules

probabilistic/stochastic tagger
resolve tagging ambiguities by training corpus.
compute probability of given word having given tag in a given context
HMM tagger

Question 25

Simple baseline algorithm

Answer 25

idea of more likely POS

for example “ a” can be a determiner or the letter “a”.
but it is more likely a determiner.

simplistic baseline algo for POS choose tag that most frequent in training corpus given the ambiguous word.

Question 26

POS Accuracy

Answer 26

standard performance measure
percentage of tags correctly labelled matching human labels test set.

always compare classifier against baseline at least as good as the most frequent class baseline. MFCB

Question 27

Rule based POS tagger

Answer 27

assign list of potential POS tag to each word based on dictionary

manual rules for out of vocab words

apply handwritten constrains until each word has only one possible POS

example

1. DT cannot immediately precede with a verb
2. no verb can immediately precede a tensed verb
3. eliminate VBN if VBD is an option

Question 28

A probabilistic method for POS

Answer 28

consider all possibile sequences of classes

chose tag sequences which is most probable given the observation sequence of n words

Question 29

Estimating probability

Answer 29

word likelihood prob * tag transition prob