WordNet: An Overview

1
WordNet An Overview

• Anubhav Madan
• anubhavm_at_comp.nus.edu.sg

2
Todays Discussion

• WordNet A Lexical Database
• WordNetSimilarity
• Some More Applications
• Limitations
• Tutorial

3
WordNet A Lexical Database

• Started in 1985
• Basic Unit Synset
• Hierarchical arrangement w/r/t definition
• Contains compounds phrasal verbs, collocations,
  and idiomatic phrases
• Bad Person _at_? offender, libertine
• Establishes a rich, dense network and
  establishes text coherence

4
WordNet The Facts

• A word or phrase is the basic unit
• Words are organized into synsets, which are a
  group of units that have the same sense.
• A gloss is a textual definition of the synset
• Words organized into hierarchies
• hypernym/hyponym concept IS-A concept
• meronym/holonym concept HAS-PART concept
• Types Nouns, Verbs, Adjectives
• 80,000 Nouns organized into 60,000 concepts

5
WordNet Architecture
6
WordNet Architecture

• Word/synset pairs stored in the WordNet DB.
• Word/List of Word Forms, Pointer to Lexical
  File, frames (for verbs), list of elements,
  (optional gloss), adjective cluster
• apple, edible_fruit,_at_ (fruit with red or yellow
  or green skin and crisp whitish flesh)
• Indexes Senses are Ordered
• Index of Familarity How well known is the word.
• Index and Data Files
• Sense Index
• The Grinder as a Converter takes Lexical Source
  Files written by Lexiographers and converts them
  into a format that is understandable and
  updatable for WN.

7
Todays Discussion

• WordNet A Lexical Database
• WordNetSimilarity
• Some More Applications
• Limitations
• Tutorial

8
WordNetSimilarity

• An application measuring closeness of concepts
  in terms of their definitions
• Main categories of measures
• Path based
• Depth based
• Information Content Based
• Gloss Based

9
WordNet Similarity Measures

• Path Finder
• Depth Finder
• Wup (Wu and Palmer) Shortest path by scaling sum
  of values b/w node, root
• Lch (Leacock and Chodrow) Shortest path by
  scaling the max path
• Path Inverse of the Shortest Path measures
• Information Content Finder
• Resnik Max Distance b/w concepts of both words
• Jcn (Jiang and Conrath) Inverses the difference
  between Sum and LCS
• Lin Scales LCS IC with the description
• Gloss Finder
• Lesk (Banerjee and Pederson) Finds and scores
  overlaps between glosses
• Vector (Padwardhan) Creates a co-occurrence
  matrix with glosses in vectors
• Hso (Hirst and St-Onge) Specifies Direction
  between Words

Demo
10
LCH
Root
2
Medium of Exchange
D5
1
1
Money
Credit
1
1
Cash
Credit Card
1
Coin
Lch Related (Money-Credit) -log (2/10)
0.70
11
WUP
Root
2
Medium of Exchange
D5
1
1
Money
Credit
1
1
Cash
Credit Card
1
Coin
Wup ConSim (Money-Credit) 4/6 0.67
12
Path
Root
2
Medium of Exchange
D5

• Inverse of the ShortestPath Measures

1
1
Money
Credit
1
1
Cash
Credit Card
1
Path (Money-Credit) 1/ min0.70, 0.67
1/0.67 1.5
Coin
13
Resnik
Medium of Exchange
Money
Credit
Cash
Credit Card
Coin
Resnik Sim (Money-Credit) -log (3/6) 0.30
14
Lin
Medium of Exchange
Money
Credit
Cash
Credit Card
Coin
Lin Sim (Money-Credit) log (6/6 3/6) 0.30
15
JCN
Medium of Exchange
Money
Credit
Cash
Credit Card
Jcn Dist (Money-Coin) log (3/6) log (2/6)
2log(6/6) 0.301 0.477 0.878
Coin
16
Lesk
17
Vector
18
HSO

• Classfies the relations in WordNet as having
  directions. 
• The Is-a relations are upwards.  The has-part are
  horizontal. 
• Establishes a relationship b/w words through a
  path that is neither too long nor changes
  direction very often.

19
Demo
20
Todays Discussion

• WordNet A Lexical Database
• WordNetSimilarity
• Some More Applications
• Limitations
• Tutorial

21
Applications

• Building Semantic Concordances
• Performance and Confidence in a Semantic
  Annotation Resnik Similarity Measure in Class
  Based Probabilities
• Lch WordNet Similarity Measure in Word Sense
  Identification
• Text Retrieval using Wordnet

22
Applications

• Lexical Chains as Representations of Context for
  the Detection of Correction of Malapropisms
• Temporal Indexing through Lexical Chaining
• COLOR-X
• Knowledge Processing on an Extended WordNet

23
Further Speculation

• Sense Disambiguation
• Information Retrieval
• Semantic Relations and Textual Coherence
• Knowledge engineering

24
The Limitations

• Relation IS-NOT or NOT-A-KIND-OF is inexpressible
• Relation IS-USED-AS-A-KIND-OF is also
  inexpressible
• No Explicit Distinction between Proper and Common
  Nouns It was too difficult to include this
  information
• Does not attempt to identify basic-level or
  generic categories. For the concepts in the
  middle of the lexical hierarchy, there can be
  many listed features that can identify the
  differences between words. WordNet doesnt
  support this.
• Not enough semantic relations in Wordnet.

25
Tutorial

• What is WordNet?
• Why is WordNet unique?
• What is the difference between WordNet and
  WordNetSimilarity
• What are some of the limiting features?
• Give an example of a human scenario, where
  WordNet would be instrumental

26
Tutorial

• What Similarity measure would you use if you had
  only the following information
• Path linkages between words in an ontology
• Information Content of the Words
• Gloss of the Words
• An ontology with direction

27
References

• Overview Pedersen, Ted and Patwardhan,
  Siddharth, and Michelizzi, Jason
  "WordNetSimilarity - Measuring the Relatedness
  of Concepts" In Proceedings of Fifth Annual
  Meeting of the North American Chapter of the
  Association for Computational Linguistics
  (NAACL-04), pp. 38-41, Boston, May 2004.
• Lch Leacock, C., and Chodorow, M. 1998.
  Combining local context and WordNet similarity
  for word sense identification. In Fellbaum, C.,
  ed., WordNet An electronic lexical database. MIT
  Press. 265283.
• Wup Wu, Z., and Palmer, M. 1994. Verb semantics
  and lexical selection. In 32nd Annual Meeting of
  the Association for Computational Linguistics,
  133138.
• Res Resnik, P. 1995. Using information content
  to evaluate semantic similarity in a taxonomy. In
  Proceedings of the 14th International Joint
  Conference on Artificial Intelligence, 448453.
• Lin Lin, D. 1998. An information-theoretic
  definition of similarity. In Proceedings of the
  International Conference on Machine Learning.
• Jcn Jiang, J., and Conrath, D. 1997. Semantic
  similarity based on corpus statistics and lexical
  taxonomy. In Proceedings on International
  Conference on Research in Computational
  Linguistics, 1933.
• Hso Hirst, G., and St-Onge, D. 1998. Lexical
  chains as representations of context for the
  detection and correction of malapropisms. In
  Fellbaum, C., ed., WordNet An electronic lexical
  database. MIT Press. 305332.
• Lesk Banerjee, S., and Pedersen, T. 2003.
  Extended gloss overlaps as a measure of semantic
  relatedness. In Proceedings of the Eighteenth
  International Joint Conference on Artificial
  Intelligence, 805810.
• Vector Patwardhan, S. 2003. Incorporating
  dictionary and corpus information into a context
  vector measure of semantic relatedness. Masters
  thesis, Univ. of Minnesota, Duluth.
• Links availiable at http//www.comp.nus.edu.sg/a
  nubhavm/reading.htm

28
Thank You

• Anubhav Madan
• anubhavm_at_comp.nus.edu.sg