Distincin semntica de compuestos lxicos en Recuperacin de Informacin

1
Distinción semántica de compuestos léxicos en
Recuperación de Información
SEPLN 2002 Valladolid

• Anselmo Peñas, Julio Gonzalo y Felisa Verdejo
• Dpto. Lenguajes y Sistemas Informáticos, UNED

2
Content

• Phrase indexing in IR
• Types of lexical compounds
• Automatic classification through WordNet
• Retrieval experiments
• Conclusions
• Future work

3
Phrase indexing in IR

• Literature
• Phrase indexing doesnt improve retrieval
• Give partial credit to components
• Proximity better than adjacency
• Reason
• Compositional meaning of phrases
• No phrase distinction
• Our proposal
• Semantic distinction of phrases

4
Types of lexical compounds (in English)

• Endocentric compound
• toothed whale
• One component gives nuclear meaning whale
• toothed whale is_a_type_of whale

5
Types of lexical compounds (in English)

• Appositional compound
• folk song
• All components add nuclear meaning
• folk song is_a_type_of song
• folk song is_a_type_of folk

6
Types of lexical compounds (in English)

• Exocentric compound
• mentally retarded
• No components give nuclear meaning
• mentally retarded is_a_type_of people
• However they retain syntactic properties

7
Types of lexical compounds (in English)

• Copulative compound
• Miguel de Cervantes
• No components give nuclear meaning
• Referred to entities
• Loss of syntactic properties

8
Automatic classification through WordNet
song
folk
whale
is_a
is_a
is_a
folk song
mentally retarded
Miguel de Cervantes
toothed whale
Appositional
Exocentric
Copulative
Endocentric

• Endocentric one component is hyperonym
• Appositional all components are hyperonyms
• Exocentric no components are hyperonyms
• Copulative not in WordNet (Entity Recognition)

9
Effect in retrieval

• Compare
• Plain text
• All compounds
• Only exocentric compounds
• Experiment conditions
• INQUERY Search Engine
• OHSUMED collection
• 380 Mb
• 101 queries
• medical domain

10
(No Transcript)
11
Conclusions

• Automatic classification of lexical compounds in
  WordNet with semantic criteria
• Exocentric and endocentric compounds behavior are
  different
• Detection of exocentric compounds in queries
  seems to increase precision slightly
• Not significative results
• Very few exocentric compounds in queries (7)

12
Future Work

• Try a test collection with longer queries
  (narrative in TREC topics)
• Detect exocentric compounds in a pseudo-relevance
  feedback framework

13
Distinción semántica de compuestos léxicos en
Recuperación de Información
SEPLN 2002 Valladolid

• Anselmo Peñas, Julio Gonzalo y Felisa Verdejo
• Dpto. Lenguajes y Sistemas Informáticos, UNED

14
Phrase indexing
Texts ...a guide for the fisher
who... ...information on cat care... ...arboreal
carnivorous called fisher cat...
15
(No Transcript)