An Unsupervised Approach To Semantic Tagging

• The objective: minimize the requirements about a human annotation by exploiting lexical resources (Wordnet) and robust parsing
  
  
• The idea: minimize the requirements about a human annotation by exploiting lexical resources (Wordnet) and r obust parsing
  

  Syntatic similarity suggests Semantic similarity Local syntatic features (e.g. VSubj and VObj pairs) Deriving class preferences for each feature Syntatic features as parameters for a probabilistic approach to tagging

• The method:
  

  Observe semantic similarity in Wordnet - p(Γ|w,r) Use similarity measures as estimation of the Γ model p(Γ,w,r)

• My role: design tools for the Wordnet hierarchy traversing and the mapping to Longman tags