Enhancing Wordnet with Morphological Knowledge

1
Enhancing Wordnet with Morphological Knowledge

• May 28th 2007
• Emilia Apostolova
• School of Computer Science, Telecommunication and
  Information Systems
• DePaul University
• emilia.aposto_at_gmail.com

2
Enhancing Wordnet with Morphological Knowledge

• Background
• Wordnet
• The Princeton Wordnet (PWN)
• Limitations of PWN
• Problem definition Motivation and Hypothesis
• Design Data and Data Analysis
• Deliverables and Schedule

3
Background - WordNet

• Wordnet
• A semantic lexicon A machine-readable lexical
  database organized by meanings
• The Princeton Wordnet (PWN)
• Started in 1985 at the Cognitive Science
  Laboratory of Princeton University.
• As of 2006, the database contains about 150,000
  words, a total of 207,000 word-sense pairs.
• Groups words into synsets, provides short,
  general definitions, and records the various
  semantic relations between these synonym sets.
  (antonyms, broader term, narrower term, used for,
  etc)

4
Background PWN
5
Background PWN

• Motivation
• Traditional dictionaries organized on
  historical principals.
• Advances in psycholinguistics and NLP.
• Purpose
• To produce a combination of dictionary and
  thesaurus that is more intuitively usable.
• To support automatic text analysis and
  artificial intelligence applications.

6
Background WordNets

• EuroWordNet - Dutch, Italian, Spanish, German,
  French, Czech and Estonian
• The Global WordNet Association Wordnets being
  developed for 52 languages.

7
Background Limitations of PWN

• No morphological knowledge.
• Morpheme
• The smallest meaningful unit of a language.
• joy-less
• re-create
• happi-ness
• comput-er-ize
• tree-s

8
Background Limitations of PWN Contd

• Originally no plans to include any morphological
  knowledge.
• Inflectional morphological parser interface
  built.
• Miller, Fellbaum, and Miller suggest that it
  became obvious that programs dealing with
  derivational morphology would greatly enhance the
  value of WordNet, but that more ambitious project
  hasnt been undertaken yet
• English has a limited inflectional system, but
  a very complex and productive derivational
  morphology.
• compute (root) - computer, computerize,
  computerization, recomputerize, noncomputerized,
  computerless, etc.
• Impossible to list exhaustively in a lexicon.

9
Problem Definition

• Give Wordnet the ability to recognize the
  derived forms of a word (including coined terms
  or inventive uses of language) that might occur
  in natural text by proposing an integrated
  morpheme dictionary and a derivational parser
  implementation. The expectation is that
  morphological parsing and morphemic semantic
  knowledge will increase significantly the
  semantic meaning derived from the texts using the
  PWN system.

10
Problem Definition Contd

• Integrated morpheme dictionary
• Bound and free morphemes computer-less,
  happi-ness, re-create
• Derivational parser interface
• A parser that can break down an English word
  into its morphemes
• re-computer-ize

11
Motivation

• English has a rich derivational morphology. PWN
  used in numerous research projects (more than 420
  listed at the official Wordnet bibliography
  site). PWN lacks morphological knowledge.
• An increasing number of research projects aim at
  adopting WordNet to specialized scientific
  domains to aid NLP tasks in the corresponding
  fields. The relative frequency of
  neologisms(coined words) in scientific texts
  appears to be higher than in general public
  texts.
• WordNet is the structural basis for wordnets in
  different languages, most of which have much
  richer morphology than the English language.

12
Hypothesis

• Morphological parsing and morphemic semantic
  knowledge will increase significantly the
  semantic meaning derived from natural texts using
  the PWD system.

13
Design

• Data
• Two different types of English language corpora
  will be used to evaluate the system - general
  public news texts and specialized computer
  science research papers.
• Wordnet will be used to derive the meaning of
  words for the sample texts. The number of misses
  (words for which WordNet returned no meaning) for
  each type of text with and without derivational
  logic will be recorded.
• Independent variable
• Presence of morphological semantic knowledge and
  parser.
• Dependent variable
• Number of misses (words without definition) in
  the sample texts.

14
Deliverables and Schedule

• Integrated Morpheme Dictionary (4 weeks)
• - compile a suitable format bound morpheme
  dictionary
• - extend the WordNet schema and integrate into
  the existing database
• Integrated derivational parser interface (4
  weeks)
• attempt to modify and integrate, if
  possible, existing English language derivational
  parser, for example PC-KIMMO
• Find and prepare English language corpora text
  samples (1 week)

15
Deliverables and Schedule Contd

• Prepare and run a script to extract word
  definitions for the words in the sample text.
  Provide a list of newly found words and their
  meanings. (1 week)
• Manually inspect the new meanings found through
  morphological knowledge and mark as hits only
  ones whose meaning is, at least partially, a sum
  of the meaning of their morphemes. (1 week)
• Analyze the results and prepare a final report.
  (2 weeks)

16
??