[Prev][Next][Index]

XTAG meeting on Feb 3rd, 10:30am


William Schuler is going to talk this week about implementation issues for
compositional semantics with STAGS. The relevant background papers are
given below (we've already read (1) and (3) - if anyone doesn't have a
copy for (3), let me know):

(1) Stuart Shieber and Yves Schabes (1990). "Synchronous Tree-Adjoining
        Grammars". In Proceedings of the 13th International Conference on
        Computational Linguistics, volume 3, pages 1-6, 1990.
        http://www.eecs.harvard.edu/~shieber/papers.html
        
(2) Stuart Shieber (1994). "Restricting the Weak-Generative Capacity of
        Synchronous Tree-Adjoining Grammars". Computational Intelligence,
        volume 10, number 4, pages 371-385, November 1994.
        http://www.eecs.harvard.edu/~shieber/papers.html  
        
(3) Laura Kallmeyer and Aravind Joshi (1999), "Factoring Predicate
        Argument and Scope Semantics: Underspecified Semantics with LTAG"
        
------------------ 
Williams' Summary:
---------------------------------------------------------------------------
I am going to talk about implementability issues with Synchronous TAGs and
the compositional semantics of Kallmeyer & Joshi (1999).  Basically, I
will present the following analysis:

  1) that the synchronous TAG formalism of 1990, while
     pedagogically useful, and potentially able to handle
     long distance quantifier raising, is probably
     impossible to implement on the shared forest produced
     by a parser (which would allow pruning the chart
     based on semantic constraints).
     
  2) that the synchronous TAG formalism of 1994 is
     relatively easy to implement on the shared forest,
     but is only capable of handling *local* quantifier
     movement (i.e. not across `NP islands').
        
  3) that the Kallmeyer/Joshi compositional semantics
     can handle long-distance quantifier movement, and
     is easier to implement than 1990 synch-TAG, but
     still has some complications over the 1994 synch-TAG 
     -- particularly, that it leaves the enumeration
     of quantifier scopes for a separate post-process.
        
--------------------------------------------------------------

- Rashmi