Natural Language Processing and KOS as an Aid to Retrieval

1
Natural Language Processing and KOS as an Aid to
Retrieval

• Iolo Jones, Daniel Cunliffe and Doug Tudhope
• Hypermedia Research Unit, School of Computing,
  University of Glamorgan
• ijones6,
• djcunlif,
• dstudhope
• _at_glam.ac.uk
• ISKO 2004

2
Natural Language Processing and KOS as an Aid to
Retrieval

• KOS and Natural Language
• Review of relevant literature in the field of NLP
• Development of algorithms and experimental model
  use of Rogets Thesaurus and WordNet as
  linguistic tools
• Evaluation of results to date and directions for
  future work

3
KOS and Natural Language

• Past 150 years in LS and IS increased
  specificity in design of controlled/normalized
  vocabularies
• Why reintroduce NL into the equation?
• Need pointed out by Smeaton (1997), Hearst and
  Karadi (1997), Voorhees (1994).

4
Linguistic Information in KOS

• Especially in thesaurus design
• Homographs disambiguated by parenthetical
  qualifier, e.g. Mercury (Greek God), Mercury
  (Metal) and Mercury (Planet) Svenonius (2000)
• Relations of BT, NT, equivalence and related term
• Scope notes free text glosses, forming data for
  linguistic exploitation

5
General developments in NLP - I

• Development of general linguistic tools steadily
  maintained since late 1940s
• Kading, 1897 Boas, 1940 early use of corpora
• Masterman, 1947 machine translation using
  machine-implemented knowledge base (using Roget)
• Early 1960s AI techniques (e.g. Wilks,
  preference semantics)
• Early enthusiasm and optimism killed by 1966
  ALPAC report led to funding cuts

6
General developments in NLP - II

• Lexical approach (Amsler, 1989, Lesk 1986)
• Decade from 1990 statistical language
  processing (Charniak, 1997)
• Use of large corpora BNC, Penn Treebank
• Notable advances in POS tagging (Brill, 1995)
• and in WSD (Hearst, 1991)

7
Specific applications

• NLIDB (Natural Language Interfaces to Databases)
  (Copestake Sparck Jones, 1990 Androutsopoulos
  et al., 1995)
• E-commerce (Winiwarter and Ibrahim, 2000)
• Knowledge-based search (Clark et al., 2000)
• NL IR in digital libraries (Strzalkowski et al.,
  1996)
• Query paraphrase of a NL question (Radev et al.,
  2001)
• General conclusion
• NL techniques quite successful in specific,
  limited domains

8
Preparing Rogets Thesaurus for use as a general
linguistic resource

• Non-proprietary download from Project Gutenberg
  http//www.gutenberg.net
• 1911 edition 1000 new words
• Needed corrections (e.g sowing ones mild oats)
  and a word-to-head index (but see recent work by
  Old)
• Rationale 1) Abstract universal classification,
  2) provision of synonym sets (Wilks, 1998)

9
Word Sense Disambiguation

• The problem
• I thought it was rum (Hodges)
• Owing to the licensing regulations, children
  will not be served at the bar (Electricity Club,
  Cardiff)
• Man jailed in tomato case (South Wales Echo)
• Flying planes can be dangerous(Chomsky)

10
Using Rogets in implementing the Semantic
Overlap process

• Starting point parsing of NL string into POS and
  constituent phrases
• LINK Parser (Sleator and Temperley, 1993)
• Focus on Noun Phrases (NPs)
• Disambiguation of homographs within the parsed
  string
• especially those marked as homographs or bound
  terms within the KOS (using AAT )

11
The Semantic Overlap process - I

• Take a known ambiguous word in the text string,
  assign to it a list of candidate Rogets
  categories try to match these candidate
  categories with a list of candidate categories
  for any other noun phrases in text string.
• Apply same process to homographs inside the KOS.

12
(No Transcript)
13
The Semantic Overlap process - II

• Im looking for clubs or sticks used for
  fighting

14
Homograph Example - I

• pins (jewelry)

15
Homograph Example - II

• pins (fasteners)

16
(No Transcript)
17
Semantic Overlap process on KOS

• Assign homographs within the AAT to Rogets
  categories (here using equivalent terms)
• 45
• anchor
• fastener
• 367
• annual
• plant
• 593
• annual
• publication
• Bound Term apron pieces position in thesaurus
  3285
• 225
• apron
• skirt

18
WordNet - I

• In an attempt to compare/improve upon
  performance, we use WordNet (Fellbaum, ed. 1998)
• Using JWordNet jwn.sourceforge.com
• Lexical reference system whose design is inspired
  by current psycholinguistic theories of human
  lexical memory

19
WordNet - II

• English nouns, verbs, adjectives and adverbs are
  organized into synonym sets, each representing
  one underlying lexical concept
• Different relations link the synonym sets
• Hierarchically organized hypernymns, hypoyms
  (is-a), unique beginners (25 for noun source
  files)

20
WordNet - III

• 1. (1) pin -- (a piece of jewelry that is pinned
  onto the wearer's garment)
• 2. fall, pin -- (when a wrestler's shoulders are
  forced to the mat)
• 3. peg, pin -- (small markers inserted into a
  surface to mark scores or define locations etc.)
• 4. personal identification number, PIN, PIN
  number -- (a number you choose and use to gain
  access to various accounts)
• (11 senses in all)

21
WordNet - IV

• Measure of fit given by
• (depth of match of ambiguous word in hierarchy)
• (total depth of hierarchy)
• This value is minimised for best fit

22
measure of fit 7/8
23
measure of fit 2/8 1/4
24
WN Homograph Example - I

• pins (jewelry)

25
WN Homograph Example - II

• pins (fasteners)

26
Further Developments - I

• Automatic document indexing and classification
  use all the modules developed so far working in
  the other direction, from the target document to
  the indexing language
• Seek to formalise a facet grammar especially
  using citation order, rules (e.g. AAT rules for
  constructing strings)

27
Conclusions

• Problem of disambiguation often left in the hands
  of the user (e.g., Google etc.)
• Aim to provide some semantic processing of query
  to sort results by meaning, aiding user in
  disambiguation process
• Powerful tools at our disposal aim is to
  harness them successfully to the application at
  hand
• to exploit knowledge implicit in structure of
  KOS and in the structures of natural language

28
Bibliography - I

• AAT http//www.getty.edu/research/tools/vocabulary
  /aat/.
• Androutsopoulos, I., Ritchie, G. D. and Thanisch,
  P. (1995) Natural Language Interfaces to
  Databases -- an introduction., Journal of Natural
  Language Engineering, 1, 29-81.
• Brill, E. (1995) Transformation-Based
  Error-Driven Parsing, Computational Linguistics.
• Charniak, E. (1997) Statistical Techniques for
  Natural Language Parsing, A I Magazine, 18,
  33-44.
• Clark, P., Thompson, J., Holmback, H. and Duncan,
  L. (2000) Exploiting a Thesaurus-Based Semantic
  Net for Knowledge-Based Search, Proc. 12th Cong
  on Innovative Applications of AI (AIII/IAAI '00),
  988-995.
• Cunningham, H. (1999) A Definition and Short
  History of Language Engineering, Journal of
  Natural Language Engineering, 5, 1-16.
• Hearst, M. (1991) Noun Homograph Disambiguation
  Using Local Context in Large Text Corpora, Proc.
  7th Annual Conference of the University of
  Waterloo Centre for the New OED and Text
  Research, Oxford.
• Hodge, G. (2000) Systems of Knowledge
  Organization for Digital Libraries Beyond
  Traditional Authority Files, Council on Library
  and Information Resources (CLIR), Washington,
  D.C.
• Ide, N. and Véronis, J. (1998) Introduction to
  the Special Issue on Word Sense Disambiguation
  The State of the Art, Computational Linguistics,
  24, 140ff.

29
Bibliography - II

• Jarmasz, M. and Szpakowicz, S. (2001) Roget's
  Thesaurus as an Electronic Lexical Database, In
  NIE BEZ ZNACZENIA. Prace ofiarowane Profesorowi
  Zygmuntowi Saloniemu z okazji 40-lecia pracy
  naukowej(Eds, Gruszczynski, W. and Kopcinska, D.)
  (to appear), Bialystok.
• Mason, O. (2000) Programming for Corpus
  Linguistics How to do Text Analysis with Java,
  Edinburgh University Press, Edinburgh.
• McEnery, T. and Wilson, A. (1996) Corpus
  Linguistics, Edinburgh University Press,
  Edinburgh.
• Radev, D. R., Qi, H., Zheng, Z.,
  Blair-Goldensohn, Zhang, Z., Fan, W. and Prager,
  J. (2001) Mining the web for answers to natural
  language questions, Proceedings of ACM CIKM,
  143-150.
• Sleator, D. D. and Temperley, D. (1993) Parsing
  English with a Link Grammar, Third International
  Workshop on Parsing Technologies.
• Strzalkowski, T., Perez-Carballo, J. and
  Marinescu, M. (1996) Natural language information
  retrieval in digital libraries, Proceedings of
  the first ACM international conference on Digital
  libraries., 117-125.
• Svenonius, E. (2000) The Intellectual Foundation
  of Information Organization, MIT Press,
  Cambridge, MA.
• Wilks, Y. (1998) Language Processing and the
  Thesaurus, Proc. National Language Research
  Institute, Tokyo, Japan.
• Winiwarter, W. and Ibrahim, I. K. (2000) A
  Mulitilingual Natural Language Interface for
  E-Commerce Applications, Proc. of the 13th
  International Conference on Applications of
  Prolog, Tokyo, Japan.
• Yarowsky, D. (1992) Word-Sense Disambiguation
  using Statistical Models of Roget's Categories
  Trained on Large Corpora, Proceedings of
  COLING-92, 454-460.