Speaker:
Nemecio Chavez (Chito), Ph.D. Student in Computer Science

Time and Location:
noon,  2/13/04 (Friday) in Science Hall 124

Title: Commonsense Aspects of Story Understanding

Abstract:

While it is simple for us to read a story and then demonstrate our 
understanding of it (i.e., we can outline it, summarize it, answer 
questions about it, etc.), this has proven to be a difficult problem 
within the field of artificial intelligence. Perhaps one of the most 
promising and studied avenues of research is that of conceptual 
dependency [Schank1975]. The idea is simple in that commonly
encountered situations are broken up into scripts. For example, each 
of us may have a restaurant script in memory that provides us with 
all the information we know about restaurants.  Thus, the meaning of 
a story is simply some chain of scripts.
  
This method of story understanding has had some success, but has been 
criticized for the following points: data is typically tailor-made 
for the story at hand, finding the right set of primitives to build 
the scripts upon may be hard or impossible, scripts tend to force 
a literal interpretation of the story, and lastly this form of 
understanding is highly dependent on the amount of background (world) 
knowledge available [Mallery1988].
  
However, since that time several advances have appeared: 1) commonsense 
knowledge databases (world knowledge of the form, "If I throw a rock up, 
it will come down.") have evolved to the point where they can be of some 
use; 2) object-oriented techniques have had success in many problem 
domains, but in particular for modeling large, complex tasks.
  
This talk suggests that using commonsense databases [Cycorp, 
ThoughtTreasure, OpenMind] along with representing the scripts with 
respect to an object-oriented structure will resolve the criticisms 
of script-based story understanding and lend itself more naturally 
to future extensions. While this is the central focus, improvements in 
understanding such as learning as the text is read and updating 
understanding when being questioned are also suggested.