[ont.events] U of Toronto Computer Science seminars, May 9-11

clarke@csri.toronto.edu (Jim Clarke) (04/19/89)

         (GB = Galbraith Building, 35 St. George Street)
    (SF = Sandford Fleming Building, 10 King's College Road)
       (Events usually begin at 10 minutes past the hour.)

SUMMARY:

COLLOQUIUM - Tuesday, May 9, 11 a.m. in SF 1105 -- Frances E. Allen
     "Program Optimization"

SYSTEMS SEMINAR - Tuesday, May 9, 2 p.m. in GB 244 -- Alexander Stoyenko
     "Concert: A High-Level Language Approach to
          Heterogeneous Distributed Systems"

AI SEMINAR - Thursday, May 11, 11 a.m. in SF 1105 -- Len Schubert
     "Generic Sentences and Story Understanding"

---------------------

COLLOQUIUM - Tuesday, May 9,  11 a.m.  in  Room SF 1105

                        Frances E. Allen
                 IBM T.J. Watson Research Center

                     "Program Optimization"

The theory and practice of program optimization is discussed in
this talk.  The focus is on "classical" optimizations for
sequential machines, e.g., code motion, strength reduction,
induction variable elimination, common subexpression elimination,
constant propagation, and register allocation.  The optimization
challenges of new architectures and new languages are also
considered.

SYSTEMS SEMINAR - Tuesday, May 9,  2 p.m.  in  Room GB 244

                       Alexander Stoyenko
                 IBM T.J. Watson Research Center

           "Concert: A High-Level Language Approach to
               Heterogeneous Distributed Systems"

Distributed systems involve increased programming complexity
along two dimensions: they require dealing with system-dependent
details of multitasking and communication, and they require
dealing with these details across a wide range of languages,
systems and networks.  This paper describes Concert, a high-
level-language-based approach to programming heterogeneous
distributed systems.  The Concert model introduces a small set of
language extensions into conventional procedural languages.
These language extensions support a cooperative peer process
model which addresses in the distributed environment the same
issues addressed by language semantics in the conventional
environment.  The Concert implementation provides layered support
for these language extensions, bridging a different source of
heterogeneity at each layer.  A prototype Concert system
currently includes C programs running on OS/2 on multiple PS/2s
communicating via calls with one another as well as with PL/I
programs running on VM/370.

AI SEMINAR - Thursday, May 11,  11 a.m.  in  Room  SF 1105

                          Len Schubert
                     University of Rochester

           "Generic Sentences and Story Understanding"

Our ability to comprehend narratives (and more generally, to cope
with the world around us) depends on our grasp of a multitude of
"general (or generic) facts", such as that a hungry person will
want food, a friend asked for a small favour is likely to oblige,
dogs are furry and can bite, TV talk shows are often late at
night, and so on.  While a good deal of effort in AI has been
devoted to the logic of simple generalizations with exceptions,
such as "Birds fly", little attention has been paid to the forms
and meanings of these generalizations and more complex ones in
English (or another language).  The premise of this talk is that
(as Montague maintained) language and logic are tightly linked,
so that a good strategy for gaining insight into the logic of
generics should be to study the logical form and formal semantics
of English generic sentences. Moreover, a clear understanding of
the mapping from English generic sentences to their underlying
logical form should enable us to impart general world knowledge
to computers in plain English. This would ease the problem of
formalizing the large amounts of knowledge needed for story
understanding and other intelligent activities.

     The talk will review past efforts to analyze the logical
form and formal semantics of generic sentences, the limitations
of these analyses, and will propose  possible ways of overcoming
some of these limitations. Roughly speaking, the idea is that
many generalizations, such as that a hungry person will want
food, can be interpreted as probabilistic statements about the
likelihood of certain conclusions being true, given a characteri-
zation of an object or situation (such as a situation in which
there is a hungry person). However, to account for the logical
form of significant classes of generic sentences in a systematic
way, one also has to allow for the "law-like" character of many
of these sentences, for complex anaphoric dependencies (like
those in so-called "donkey sentences"), and for the role of tense
and time adverbials.  Some of the ideas to be discussed have been
tried out in the context of a partially built story understanding
system.
-- 
Jim Clarke -- Dept. of Computer Science, Univ. of Toronto, Canada M5S 1A4
              (416) 978-4058
clarke@csri.toronto.edu     or    clarke@csri.utoronto.ca
   or ...!{uunet, pyramid, watmath, ubc-cs}!utai!utcsri!clarke