Knowledge Representation for NL understanding

  

 i. Understanding task 

  1) The task of understanding. Understanding (translation into a formal 

language) and recognition (relation to a class) as the 

boundaries of the spectrum of AI problems.

  Understanding&recognition are adequately posed relatively to PD.

  2)Speech - recognition (specific organization of paterns, 

NL - understanding

  Reduction of an utterance understanding to an addressee's reaction   



  3) 2 main requirements to pose the undertsanding problem:

    1) expressive language 

      in particular, reasoning about time/space, knowledge, belief 

      and action

    2) representation of PD     

      effectiveness: ability to query arbitrary combination of concepts.

      effective search)

  

 ii.  Engineering of speech and NL processing

     1) Pure speech recognition. visual speech and areas transformation.  

     2) Filtering of speech recognition results (pretenders list).

          learning: occurence matrix; mutual definition -> occurence matrix update.   

          real-time : calculation of maximal likelihood of local PD

     3) Return to primary processing. Examples of scene images,  speech-NL     

     4) combination of PD and PD manager. 

     5) generality manager



 iii.  NL understanding in expandable environment

     1) understanding of inquiries and definitions (examples)



 iv. Metalanguage - supported language.

      1) modal metapredicates

      2) formal scenarios vs traditional axiomatic methods.

      3) metaaxioms, which allow multiple meanings

      4) metalanguage inference rules

  v. Examples of formal scenarios