[Prev][Next][Index]

Abstracts not contained in last week's CSLI Calendar

	    COMMONSENSE AND NONMONOTONIC REASONING SEMINAR

		  Incremental, Approximate Planning
			 (Preliminary Report)

			    Charles Elkan
		    Department of Computer Science
			University of Toronto

		      Monday, 27 November, 3:15
		       Margaret Jacks Hall 252

I shall present evidence for three claims.

(1) Locally stratified definite clauses with the perfect model
    semantics are adequate for specifying the Yale shooting problem
    and other traditional planning domains, using the following
    frame axioms:

	causes(A,S,P) => holds(P,do(A,S))
	holds(P,S) & ~ cancels(A,S,P) => holds(P,do(A,S))

(2) With the right control strategy, Green's method is a practical
    way of inventing plans given a locally stratified domain theory.

(3) Allowing default conditions to be assumed is an abductive
    planning strategy that produces plausible plans quickly, and those
    plans can be incrementally refined towards guaranteed plans.
			     ____________

			   SYNTAX WORKSHOP

      On Mandarin Chinese Predicates with Subcategorized TOPICs

			    Chu-Ren Huang
		       Academia Sinica, Taipei

		   Tuesday, 28 November, 7:30 P.M.
			     Cordura 100

A small set of two-place predicates in Chinese requires the surface
representation of NP-NP-Pred but disallows the canonical transitive
pattern NP-Pred-NP.  This is illustrated in (1).

(1a) 	zeijian shi, ni zuozhu
	this matter, you take-charge
(1b)*	ni zuozhu zeijian shi

I argue that these sentence initial NPs are subcategorized arguments:
they are semantically required by the predicates; the predicate
imposes selectional restrictions on them; Mandarin predicates are
divided in two classes depending on whether they take an argument in
this position or not.  Furthermore I show that the sentence initial
NPs are TOPICs: they have to be definite and the existence of a
referent is presupposed.  I give a straightforward LFG account in
which the predicates subcategorize for a subject and a topic. I then
discuss the implications of such assignments for Lexical Mapping
Theory.  Among the alternatives explored are an independently motivated
feature to distinguish TOPIC from other grammatical functions in the
LMT and Zaenen's proposal of matching a hierarchy of required thematic
roles with that of a hierarchy of grammatical functions instead of
default assignment of features.

The final workshop of this year will be Peter Austin, on 12 December.
			     ____________

		SEMINAR ON COMPUTERS, WORK, AND DESIGN

	   Conversational Support for the Software Process.

			   Simon M. Kaplan
		    Department of Computer Science
	      University of Illinois at Urbana-Champaign

		    Wednesday, 29 November, 12:15
			   CSLI, Ventura 17

We are investigating the application of the language/action
perspective to support tools for the software process.  We view the
process as:

o	Consensual, negotiative.  Most parts of the process are carried out
  	by groups of some kind.  The negotiation and consensus deriving
	from that negotiation drives the process.

o 	Not amenable to formalization.  The "process programming"
	perspective assumes we can circumscribe the "conceptual space"
	of software development and then write "programs" which automate
	the process.  We believe that the process is often highly creative
	and not amenable to this kind of circumscription as a result.

o	Largely historical.  The perspectives brought to a project 
	by the members of the groups at various times in its life
	shape the way problems are approached and the particular
	solutions chosen.  Recognizing this is of vital importance 
	when trying to support maintenance tasks.

o	Automatable in parts.  Some parts of the process (building 
	system releases, maintaining versions, certain kinds of analysis
	and automated code generation under some circumstances) can be
	performed by tools.  

Traditional approaches to supporting the software development process
have assumed the process can be modeled in some direct way, and that
model used in a tool.  (The process programming approach is a fine
example of this.)  Experience shows that this kind of approach soon
fails because the users of the tool do not follow the model the tool
uses in anything but the loosest ways, or because the model is in some
sense incomplete.  The hermeneutic perspective explains why this is
the case; the language/action perspective gives us a way to do
something about it.

The talk will overview our view of the software process and what
hermeneutics and the language/action perspective allow us to do about
supporting it.  I will then discuss the Conversation Builder, the tool
(or rather, the open system for tool integration through conversation)
that we are building to support the process from the language/action
perspective and give some examples of the kinds of demonstration
systems we are constructing using the Conversation Builder to show the
utility of our approach.