New Developments in WordNet:

1
New Developments in WordNet An Electronic
Lexical Resource
Helen Langone Columbia University Princeton
Cognitive Science Laboratory
2
Classic WordNet

• A lexical semantic network relating word forms
  and lexicalized concepts (i.e., concepts that
  speakers have adopted word forms to express)
• Main relationshyponymy/troponymy
  (kind-of/way-to), meronymy (part-whole),
  synonymy, antonymy
• Predominantly hierarchical, few relations across
  grammatical class, glosses example sentences do
  not participate in network
• Nouns organized under 9 unique beginners
• Command-line interface C library
• Prehistoric (but greppable!) db format

3
(No Transcript)
4
(No Transcript)
5
(No Transcript)
6
Unique beginner synsets
7
(No Transcript)
8
What's ahead/new...

• Instance-of pointers
• Morphosemantic links
• Ontological reorg
• Glosses disambiguated, parsed and translated into
  logical forms
• SQL database access Perl library
• Alternative forms, inflected forms

9
Instance-of pointer

• Classic WordNet hyponymy relation made no
  distinction between subsumption (type or kind of
  x) and instantiation (is an instance of an x)
• e.g. George Washington is an Instance-of
  United_States_President, not a Kind-of
• e.g. a playwright is a Kind-of writer, author,
  while Shakespeare is an Instance-of the type
  dramatist, playwright

10
Morphosemantic links

• Links between word forms motivated by both form
  and meaning.
• verb-noun (write/writer), noun-adj
  (duplicity/duplicitous), verb-adj
  (talk/talkative), noun-noun (friend/friendship)
• Not the sense of morphosemantics used in
  morphology for form-meaning correspondences.
• For polysemous forms, relates only sense-specific
  derivationally-related forms e.g.
  digest/digestion. 2 distinct senses of the verb
  digest (physiological, psychological) correspond
  with 2 distinct senses of the noun digestion.
• How is this different from Porter the like?

11
Porter stemmer

• Pattern-matches on word endings
• Misses many regular form-form correspondences
• biography/biographer/biographical
• song/sing/songster
• deception/deceive
• Conflates unrelated forms
• amor--gtamorous, amorously, amorousness, amoral,
  amorally, amoralism, amorality...!

12
Derivationally-related forms of myth
mythological
mythologic
mythology
mythicise
mythologize
mythologization
myth
mythologise
mythicize
mythologisation
mythic
mythical
mythologist
13
Porter-stemmed forms of myth
mytholog-
mythological
mythologic
mythicis-
mythology
mythicise
mythologize
mythologization
myth
mythologis-
myth-
mythologise
mythicize
mythologisation
mythic
mythical
mythologist
mythic-
mythologist-
14
(No Transcript)
15
Why should we care about the semantic part?

• Catvar (Habash and Dorr, 2003) clusters over
  100,000 word forms into 63,000 clusters based on
  morphological relatedness.
• Viegas et al.'s (1996) lexical rules
  automatically derive related words from a shared
  stem.
• Neither consider polysemylike Porter, they lump
  together all forms having the same stem.
• Relating forms that derive from different senses
  will affect language understanding.

16
(No Transcript)
17
(No Transcript)
18
Why should we care about the semantic part?
19
Why should we care about the semantic part?
20
Reorg of the top levels

• All noun hierarchies are now subsumed under a
  single synset, entity .
• The reorganization brings WordNet more in
  alignment with ontologies that would map into it
  (e.g., SUMO)
• The noun file is now structured more like a
  general-purpose ontology, but be aware multiple
  inheritance exists.

21
(No Transcript)
22
Disambiguation parsing of the glosses

• Classic WordNet only synsets and entry word
  forms participate in the network of relations
• Sense-tagging is a process of disambiguationa
  word form is linked to its context-appropriate
  sense (e.g., run a company vs. run a race)
• Sense-tagging will do the equivalent of
  hyper-linking every open-class word in the
  glosses to every other semantically-related
  word/concept in WordNet
• Will add an additional 800,000 direct links to
  the 120,000 bidirectional links currently in
  place the number of words/synsets indirectly
  reachable will be far greater.
• Disambiguated glosses will be parsed and
  translated into first order predicate logic (by
  Jerry Hobbs at USC/ISI)

23
WNDEV

• Current database format poses some limitations
• Restricts the number of kinds of relations
  searches possible
• Separate files for each part of speech means that
  changes affecting more than one file requires
  coordination among lexicographers
• Unstructured format means that formatting errors
  may not be caught
• Byte offset access means maintaining separate
  editable and compiled versions of the data, and a
  lengthy grind-and-fix-parse-errors process
• Adding to the lexicon must be done manually using
  an arcane syntax no automatic means exists for
  loading entries

24
(No Transcript)
25
WNDEV

• WordNet development environment a suite of tools
  for developing, editing, and using WordNet
• SQL database Perl library
• Graphical interface for editing synsets, linking,
  creating new relations, etc.
• Tools for automatic loading of externally-created
  entry sets, database patches, and format
  conversions.