Lecture 17 Chunking

1
Lecture 17 Chunking
CSCE 771 Natural Language Processing

• Topics
• Chunking
• Overview of Meaning
• Readings
• Text 13.5
• NLTK book 7.2

July 7, 2009
2
Overview

• Last Time (Programming)
• Probabilistic parsing
• Features and Unification
• Complexity of Language
• Meaning representations
• Projects???
• Today
• Chunking
• Meaning representations
• Description Logics
• Projects???
• Readings
• Chapter 17.1
• 771 Website Resources
• Next Time Semantics

3
Harvesting the Web Homework

• You should write a python program to open a URL,
  read the page, and save it to a file.
• You should figure out how to find (recognize
  regexpr for) and build a list of links.
• Combine the previous two to have a program that
  will do a breadth first search to a certain level
  starting with a URL.
• You should find a tool or routine that will strip
  the HTML for the web page read leaving plain
  text.
• You should find a tool or routine to convert a
  pdf file into plain text.

4
Find a Topic

• Suggested projects include
• Reading a collection of research papers and
  identifying references and extracting a reference
  graph.
• Constructing a similarity measure between
  documents by constructing bigram vectors.
• Read a collection of documents and build a parse
  tree distribution.
• Knowledge extraction
• Plagiarism detection.
• Page Rank procedures
• Authorship identification

5
Partial Parsing / Chunking

• Information extraction typically does not require
  a complete parse.
• Consider ungrammatical fragments
• Chunking is the indentifying and classifying
  non-overlapping portions of a sentence that
  constitute the basic non-recursive phrases
• E.g. noun phrases, verb phrases, adj. phrases,
  prepositional phrases
• NP the morning flight PP from NP Denver
  VP has arrived

6
Focus on NP

• NP the morning flight from NP Denver has
  arrived

7
Finite State Rule Based Chunking

• Rules developed based on regular expressions for
  the particular application
• No recursion
• NP ? (DT) NN NN
• NP ? NNP
• VP ? VB
• VP ? Aux VB

8
Cascaded Finite State Machines

• .

9
IOB Tagging

• B beginning
• I internal parts of the chunk
• O outside the chunk
• Example
• the morning flight from Denver has
  arrived
• B_NP I_NP I_NP B_PP B_NP B_VP
• Focussing on NPs
• the morning flight from Denver has
  arrived
• B_NP I_NP I_NP O B_NP O O

10
Chunking Systems Evaluation

• Accuracy of chunking systems can be evaluated
  using techniques from information retrieval
• Precision
• Recall
• F- measure (Rijsbergen) combines these into one
  measure
• where ß gt 1 favors recall and where ß lt
  1 favors precision

11
Information Extraction Chap 7 NLTK Book

• OrgName LocationName
• Omnicom New York
• DDB Needham New York
• Kaplan Thaler New York
• BBDO South Atlanta
• Georgia-Pacific Atlanta
• gtgtgt print org for (e1, rel, e2) if rel'IN' and
  e2'Atlanta'
• 'BBDO South', 'Georgia-Pacific'

http//nltk.googlecode.com/svn/trunk/doc/book/ch07
.html
12
Information Extraction from Text

• The fourth Wells account moving to another agency
  is the packaged paper-products division of
  Georgia-Pacific Corp., which arrived at Wells
  only last fall. Like Hertz and the History
  Channel, it is also leaving for an Omnicom-owned
  agency, the BBDO South unit of BBDO Worldwide.
  BBDO South in Atlanta, which handles corporate
  advertising for Georgia-Pacific, will assume
  additional duties for brands like Angel Soft
  toilet tissue and Sparkle paper towels, said Ken
  Haldin, a spokesman for Georgia-Pacific in
  Atlanta.

http//nltk.googlecode.com/svn/trunk/doc/book/ch07
.html
13
Information Extraction Architecture

• .

http//nltk.googlecode.com/svn/trunk/doc/book/ch07
.html
14

• gtgtgt def ie_preprocess(document)
• ... sentences nltk.sent_tokenize(document)
• ... sentences nltk.word_tokenize(sent) for
  sent in sentences
• ... sentences nltk.pos_tag(sent) for sent in
  sentences

http//nltk.googlecode.com/svn/trunk/doc/book/ch07
.html
15
Noun Phrase Chunking using NLTK

• gtgtgt sentence ("the", "DT"), ("little", "JJ"),
  ("yellow", "JJ"),
• ... ("dog", "NN"), ("barked", "VBD"), ("at",
  "IN"), ("the", "DT"), ("cat", "NN")
• gtgtgt grammar "NP ltDTgt?ltJJgtltNNgt"
• gtgtgt cp nltk.RegexpParser(grammar)
• gtgtgt result cp.parse(sentence)
• gtgtgt print result (S (NP the/DT little/JJ
  yellow/JJ dog/NN) barked/VBD at/IN (NP the/DT
  cat/NN))
• gtgtgt result.draw()

http//nltk.googlecode.com/svn/trunk/doc/book/ch07
.html
16
Tag Patterns used in Chunk Grammars

• Tag pattern - a sequence of part-of-speech tags
  delimited using angle brackets, e.g.
  ltDTgt?ltJJgtltNNgt

http//nltk.googlecode.com/svn/trunk/doc/book/ch07
.html
17
nltk.app.chunkparser()
http//nltk.googlecode.com/svn/trunk/doc/book/ch07
.html
18
Chunking with Regular Expressions
http//nltk.googlecode.com/svn/trunk/doc/book/ch07
.html
19

• gtgtgt nouns ("money", "NN"), ("market", "NN"),
  ("fund", "NN")
• gtgtgt grammar "NP ltNNgtltNNgt Chunk two
  consecutive nouns"
• gtgtgt cp nltk.RegexpParser(grammar)
• gtgtgt print cp.parse(nouns) (S (NP money/NN
  market/NN) fund/NN)

http//nltk.googlecode.com/svn/trunk/doc/book/ch07
.html
20

• gtgtgt cp nltk.RegexpParser('CHUNK ltV.gt ltTOgt
  ltV.gt')
• gtgtgt brown nltk.corpus.brown
• gtgtgt for sent in brown.tagged_sents()
• ... tree cp.parse(sent)
• ... for subtree in tree.subtrees()
• ... if subtree.node 'CHUNK' print subtree
• ...
• (CHUNK combined/VBN to/TO achieve/VB)
• (CHUNK continue/VB to/TO place/VB)
• (CHUNK serve/VB to/TO protect/VB)
• (CHUNK wanted/VBD to/TO wait/VB)
• .

http//nltk.googlecode.com/svn/trunk/doc/book/ch07
.html
21
Chinker.py Figure 7.5

• Natural Language Toolkit code_chinker
• grammar r"""
• NP
• lt.gt Chunk everything
• ltVBDINgt Chink sequences of VBD and
  IN
• """
• sentence ("the", "DT"), ("little", "JJ"),
  ("yellow", "JJ"),
• ("dog", "NN"), ("barked", "VBD"), ("at",
  "IN"), ("the", "DT"), ("cat", "NN")
• cp nltk.RegexpParser(grammar)

http//nltk.googlecode.com/svn/trunk/doc/book/ch07
.html
22

• gtgtgt print cp.parse(sentence)
• (S
• (NP the/DT little/JJ yellow/JJ dog/NN)
• barked/VBD
• at/IN
• (NP the/DT cat/NN))

http//nltk.googlecode.com/svn/trunk/doc/book/ch07
.html
23
Developing and Evaluating Chunkers

• chunk.conllstr2tree() builds a tree
  representation from one of these multi-line
  strings

http//nltk.googlecode.com/svn/trunk/doc/book/ch07
.html
24
Reading IOB Format and the CoNLL 2000 Corpus
http//nltk.googlecode.com/svn/trunk/doc/book/ch07
.html
25

• gtgtgt text '''
• ... he PRP B-NP
• ... accepted VBD B-VP
• ... the DT B-NP
• ... position NN I-NP
• ... of IN B-PP
• ... vice NN B-NP ... chairman NN I-NP ... of IN
  B-PP ... Carlyle NNP B-NP ... Group NNP I-NP
• ... , , O
• . . O
• ... '''
• gtgtgt nltk.chunk.conllstr2tree(text,
  chunk_types'NP').draw()

http//nltk.googlecode.com/svn/trunk/doc/book/ch07
.html
26

• gtgtgt from nltk.corpus import conll2000
• gtgtgt print conll2000.chunked_sents('train.txt')99
  
• (S
• (PP Over/IN)
• (NP a/DT cup/NN)
• (PP of/IN)
• (NP coffee/NN)
• ,/,
• (NP Mr./NNP Stone/NNP)
• (VP told/VBD)
• (NP his/PRP story/NN)
• ./.)

http//nltk.googlecode.com/svn/trunk/doc/book/ch07
.html
27

• gtgtgt from nltk.corpus import conll2000
• gtgtgt cp nltk.RegexpParser("")
• gtgtgt test_sents conll2000.chunked_sents('test.txt
  ', chunk_types'NP')
• gtgtgt print cp.evaluate(test_sents)
• ChunkParse score
• IOB Accuracy 43.4
• Precision 0.0
• Recall 0.0
• F-Measure 0.0

http//nltk.googlecode.com/svn/trunk/doc/book/ch07
.html
28

• gtgtgt grammar r"NP ltCDJNP.gt"
• gtgtgt cp nltk.RegexpParser(grammar)
• gtgtgt print cp.evaluate(test_sents)
• ChunkParse score
• IOB Accuracy 87.7
• Precision 70.6
• Recall 67.8
• F-Measure 69.2

http//nltk.googlecode.com/svn/trunk/doc/book/ch07
.html