Copy Link
Add to Bookmark
Report
NL-KR Digest Volume 01 No. 27
NL-KR Digest (12/05/86 10:54:07) Volume 1 Number 27
Today's Topics:
From CSLI Calendar, December 4, No. 9
Seminar - Syntactic Categories (TI)
Talk: Perrault on Applying Default Logic to Speech Acts
Talk: Winograd on NL Understanding
----------------------------------------------------------------------
Date: Wed 3 Dec 86 16:41:57-PST
From: Emma Pease <Emma@CSLI.STANFORD.EDU>
Subject: From CSLI Calendar, December 4, No. 9
Tel: (415) 723-3561
[Excerpted from CSLI Calendar]
THIS WEEK'S TINLUNCH
Reading: What to do with theta-Roles?
Discussion led by Annie Zaenen
December 4
When Extended Standard Theory won the linguistic wars (Newmeyer's
version of linguistic history), lexical semantics went out of fashion
in mainstream generative grammar but, as is often the case with
victories that are the results of power politics rather than reason,
the problems that were raised in the generative semantics research
remained unsolved and recent years have seen them resurface. At this
point several attempts to specify the role of lexical semantics in
syntax are under elaboration. Among the debated issues are (1) the way
semantic information has to be represented in the lexicon; (2) the
number and the properties of the levels of representation needed to
link semantics and syntax.
The paper tries to give a partial answer to these questions from a
Government Binding related view. I choose it because that point of
view will most likely not be widely represented among the live
participants at the TINLunch. The main purpose of the lunch should be
a discussion of the general issues raised in the paper rather than a
critique of the paper itself.
Other relevant recent writings on the topic include: Dowty (1986):
On the semantic content of thematic roles; Jackendoff (1986): The
status of Thematic Relations in Linguistic Theory; Foley and Van Valin
(1984): Functional Syntax and Universal Grammar; and Kiparsky's
manuscript on Morphosyntax.
--------------
THIS WEEK'S SEMINAR
Rational Behavior in Resource-bounded Agents
David Israel
December 4
Members of the Rational Agency Project at CSLI (RatAg) have been
involved in research to develop an architecture for the production of
rational behavior in resource-bounded agents. The overall aim of this
work is to combine techniques that have been constructed in artificial
intelligence for automating means-end reasoning with a computational
instantiation of techniques that have been developed in decision
theory for weighing alternative courses of action. The focus is on
ensuring that the resulting synthesis is a viable architecture for
agents who, like humans and robots, are resource-bounded, i.e., unable
to perform arbitrarily large computations in constant time.
Predicating the architecture on the fact that agents have resource
bounds will enable its use both as a device for producing rational
behavior in robots that are situated in dynamic, real-world
environments, and as a model of human rational behavior. In taking
seriously the problem of resource boundedness, we draw heavily on the
view of plans as ``filters'' on practical reasoning. We are concerned
with determining what regularities there are in the relationship
between an agent and her environment that can be exploited in the
design of the filtering process.
--------------
THIS WEEK'S COLLOQUIUM
Rational Speech Activity: The Case of Discourse Politeness
Asa Kasher
December 4
The paper will briefly outline the role to be played by rationality
considerations in governing understanding and production of speech
acts. It will be argued that a certain aspect of rationality
considerations, namely cost, has been neglected. Its importance will
be demonstrated in the case of discourse politeness as well as in some
apparent counter-examples to Grice's conversational maxims.
------------------------------
Date: WED, 10 oct 86 17:02:23 CDT
From: leff%smu@csnet-relay.arpa
Subject: Seminar - Syntactic Categories (TI)
Resent-Message-Id: <12260018636.13.LAWS@SRI-STRIPE.ARPA>
Subject: Syntactic Categories
Speaker: Dr. James D. McCawley
1:45pm, Friday, 5 DEC 1986, North Building C-1 Texas Instruments
In most modern approaches to syntax, expressions are assigned to
syntactic categories on the basis of criteria that are purely syntactic,
and the various category assignments are mutually exclusive (a particular
token of a word cannot be simultaneously a noun and an adjective). In
approaches that allow for multiple syntactic structures (i. e. the
various versions of transformation grammar) categories are not allowed
to differ from one structure to another (something is a verb phrase in surface
structure if and only if is one in deep structure).
I argue for the rejection of the above policies and develop an
alternative approach in which a syntactic category name is merely an
informal abbreviation for a complex of information not all of it syntactic in
nature that is relevant to the syntactic behavior of the item. This
approach enables one to do transformational grammar without needing to
accept many of its more dubious standard assumptions, and makes possible
a scenario for language aacquisition that allows for the acquisition of
highly specific syntactic structures before the child has more than a
fragmentary knowledge of the relevant syntactic rules.
------------------------------
From: dlm.allegra%btl.csnet@RELAY.CS.NET
Date: Sun 30 Nov 1986 14:10:57
Original-To: research!csnet!sri-stripe.arpa!ailist
Subject: Talk: Perrault on Applying Default Logic to Speech Acts
Title: An Application of Default Logic to Speech Act Theory
Speaker: C. Raymond Perrault, AI Center, SRI International
and CSLI, Stanford U.
Date: Thursday, November 13
Place: AT&T Bell Laboratories - Murray Hill
Abstract:
One of the central issues to be addressed in basing a theory of speech
acts on independently motivated accounts of propositional attitudes
(belief, knowledge, intentions, ... ) and action is the specification
of the effects of communicative acts. The very fact that speech acts
are largely conventional means that specifying, for example, the
effects of the utterance of a declarative sentence, or the performance
of an assertion, requires taking into consideration many possible
exceptions to the conventional use of the utterances (e.g., the
speaker may be lying, the hearer may not believe him, etc.). Previous
approaches to the problem have attempted to deal with these exceptions
using two kinds of techniques. The first is to make the statement of
the effects of the act conditional on certain conditions being true at
the time of utterance. The second is to stipulate the consequences of
the utterance as the strongest condition which is true in all possible
conditions of utterance. We will argue that there are problems with
both these approaches and present an approach to the problem within
the framework of an extension of Reiter's non-monotonic Default Logic.
Default rules are used to embody a simple theories of belief adoption,
of action observation, and of the relation between the form of a
sentence and the attitudes it is used to convey. This allows quite a
simple picture of the relation between certain illocutionary and
perlocutionary acts. The emphasis will be on uses of declarative
sentences.
Sponsor: Mitch Marcus
------------------------------
Date: Tue, 2 Dec 86 17:29 EST
From: dlm.allegra%btl.csnet@CSNET-RELAY.ARPA
Subject: Talk: Winograd on NL Understanding
Reply-To: miller@ACORN.CS.ROCHESTER.EDU
Organization: University of Rochester, Department of Computer Science
Postal-address: 617 Hylan Building, University of Rochester, Rochester NY 14627
Phone: 716-275-7747
Title: "Can We Expect Our Computers to Understand?"
Speaker: Terry A. Winograd
Associate Professor of Computer Science and Linguistics
Stanford University
Date: Tuesday, November 18, 1986
Place: AT&T Bell Laboratories - Indian Hill Auditorium
Simulcast Locations: Holmdel, Whippany, Murray Hill
AI Seminar Series
Abstract:
Many of the anticipated applications of computers in the coming years
will demand some kind of `understanding,' either of natural language
or in the more general sense of understanding situations and acting
appropriately in them. Proponents of `Fifth Generation Systems' see
them as revolutionizing the ways we use computers and do our work.
At the same time, the actual work on developing computer understanding
has been frustratingly slow. Computers cannot deal effectively with
natural language except in the most stringently restricted of contexts
and no expert systems are successful in areas that call for the ordinary
understanding of human common sense.
My own experience with these problems has led me to examine some
fundamental questions about the nature of intelligence, thought, and
language. The result (elaborated in a recent book with Fernando Flores,
entitled "Understanding Computers and Cognition") is a distinction
between different ways in which we understand. There is a critical
difference between the kinds of `rationalistic' reasoning that is so
well implemented on the computer, and the kind of situated understanding
that is rooted in human practices and ordinary language. Current
approaches to artificial intelligence will not bridge this gap, and we
need to reinterpret the role that computer reasoning will play in human
situations, in order to design more effective uses of machines.
Sponsor : William F. Opdyke ihesa!opdyke
------------------------------
End of NL-KR Digest
*******************