semantic mining unsupervised acquisition of multilingual
play

Semantic mining: Unsupervised acquisition of multilingual semantic - PowerPoint PPT Presentation

Journe des Doctorants du LIMSI 30/05/2017 Semantic mining: Unsupervised acquisition of multilingual semantic classes from texts Presenter: Zheng ZHANG Supervisors: Pierre ZWEIGENBAUM & Yue MA Research group: ILES Doctoral


  1. Journée des Doctorants du LIMSI 30/05/2017 Semantic mining: Unsupervised acquisition of multilingual semantic classes from texts Presenter: Zheng ZHANG Supervisors: Pierre ZWEIGENBAUM & Yue MA Research group: ILES Doctoral school: STIC Funding: CSC Date

  2. Distributional hypothesis Distributional hypothesis: Words that occur in the same • contexts tend to have similar meanings (Harris, 1954). Take a word and its contexts: • pomme rouge • pomme délicieuse • … • By looking at a word’s context, one can infer its meaning • 1

  3. Distributional similarity Distributional hypothesis: Words that occur in the same • contexts tend to have similar meanings (Harris, 1954). Take a word and its contexts: • pomme rouge • pomme délicieuse • … • By looking at a word’s context, one can infer its meaning • 1

  4. Multilingual semantic classes pomme Semantic class: A group of words • clustered by using distributional pêche fruits similarity measures. prunelle poire couleurs 2

  5. Multilingual semantic classes FR EN peach pear fruits pomme pêche apple fruits prunelle colors poire couleurs 2

  6. Multilingual semantic classes FR EN peach pear fruits pomme pêche apple fruits prunelle colors poire couleurs 2

  7. Applications: Unknown words “translation” FR EN peach pear fruits pomme pêche apple fruits prunelle colors poire roux couleurs 3

  8. Applications: Unknown words “translation” FR EN peach pear fruits pomme pêche apple fruits prunelle colors poire roux couleurs 3

  9. Applications: Universal classes extraction FR EN peach pear fruits pomme pêche apple fruits prunelle poire Universal class of “fruits” 4

  10. Merci

Recommend


More recommend