Newsgroups: comp.ai
Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!zaphod.mps.ohio-state.edu!wuarchive!m.cs.uiuc.edu!ibma0.cs.uiuc.edu!sunc4.cs.uiuc.edu!epstein
From: epstein@sunc4.cs.uiuc.edu (Milt Epstein)
Subject: Re: scripts revisted...
Sender: news@ibma0.cs.uiuc.edu
Message-ID: <27F652F7.3ADE@ibma0.cs.uiuc.edu>
Date: Sun, 31 Mar 1991 21:21:58 GMT
References: <9103290133.AA27890@enuxha.eas.asu.edu>
Organization: University of Illinois at Urbana-Champaign
Lines: 50

In <9103290133.AA27890@enuxha.eas.asu.edu> gomez@ENUXHA.EAS.ASU.EDU (Jose L. Gomez-Rubio) writes:

>Read Shank's readings on scripts.  Have a question though on the restaurant
>script:
>
>What if one develops many scripts for all the restaurants for a particular
>city but the system suggests all the restaurants irregardless of the 
>inputs provided?

You might want to look at some of Schank's later work on MOPs (memory
organization packets) -- Janet Kolodner also did a lot of work on
them.  The main idea was that episodes were organized hierarchically
from general to specific (so that, for example, general MOPs/scripts
for restaurants were above those for specific restaurants), and common
features were stored at the appropriate level, and distinguishing
features were used to index distinct MOPs.  If there's enough
indexing, the system will be able to pick an appropriate MOP/script
according to the features of the current situation.  Kolodner's system
was called CYRUS and she has a book on it (whose name I cannot
remember -- perhaps something like "The Structure and Organization of
Episodic Memory").  Very interesting stuff, and the basis for much of
the work today on case-based reasoning systems.


>I've read some criticisms of the script paradigm.  One by Hubert Dreyfus 
>(sorry if I got the name wrong), a noted critic of AI, he suggests the
>script approach is to ad hoc in its representation.  Can someone elaborate
>on this?

I'm not real sure about this.  Perhaps he meant that you could fill in
the script with anything, so it didn't have formal or consistent
underpinnings.  But Schank et al would probably argue that this is a
"feature" and corresponds to how people do it -- since everyone's
impressions of restaurants (or whatever) are different and are formed
by experience.


>I think that the script representation can't handle unexpected scenarios and
>is only designed for stereotypical situations.

I believe Schank clearly indicates that scripts are supposed to
represent stereotypical situations, and that he introduced plans and
goals and themes to account for more general scenarios.


-- 
Milt Epstein
Department of Computer Science
University of Illinois
epstein@cs.uiuc.edu
