[Prev][Next][Index]

CSLI Calendar, 5 April, vol. 5:22


       C S L I   C A L E N D A R   O F   P U B L I C   E V E N T S
-----------------------------------------------------------------------------
5 April 1990                     Stanford                      Vol. 5, No. 22
_____________________________________________________________________________

    A weekly publication of the Center for the Study of Language and
Information (CSLI), Ventura Hall, Stanford University, Stanford, CA 94305-4115
			     ____________

	   CSLI ACTIVITIES FOR THIS THURSDAY, 5 APRIL 1990

12:00 noon		TINLunch
      Cordura 100	Variation Across a Population of American
			English Speakers
			Meg Withgott and Francine Chen
			Xerox PARC
			(withgott.pa@xerox.com, fchen.pa@xerox.com)
			Abstract in last week's Calendar

	   CSLI ACTIVITIES FOR NEXT THURSDAY, 12 APRIL 1990

12:00 noon		TINLunch
      Cordura 100	A Representationalist Theory of Intention
			Kurt Konolige and Martha E. Pollack
			SRI International
			(konolige@ai.sri.com, pollack@ai.sri.com)
			Abstract below

 2:15 p.m.		CSLI Seminar
      Cordura 100	Controversies in Natural-Language Research
			led by Stanley Peters
			(peters@csli.stanford.edu)
			Title: Computational Morphology
			Speaker: Lauri Karttunen
			Xerox PARC
			(lauri@csli.stanford.edu)
			Abstract below

 4:15 p.m.		CSLI Colloquium
      Ventura 17	Precompiled Syntactic Constraints for
			Continuous Speech Recognition
			Pierre Dupont
			Philips Research Laboratory, Brussels
			(dupont@prlb.philips.be)
			Abstract below
			     ____________
				   
		      CSLI SPRING SEMINAR SERIES
	      Controversies in Natural-Language Research
			led by Stanley Peters
			 Thursdays, 2:15 p.m.
			     Cordura 100

The CSLI Seminar for spring quarter will start on 12 April, when Lauri
Karttunen will discuss research by Ron Kaplan, Martin Kay, and Korvo
Koskeniemmi, among others, on Finite State Morphology.  Then Paul
Kiparsky and Bill Poser will discuss current linguistic theories of
morphology on 19 April.  The following Thursday, 26 April, the five
CSLI members among these six will debate, to bring out
complementarities and conflicts of the approaches.

A similar sequence of three seminars on syntax, and a like sequence
on semantics, will take place in May and June.
			     ____________
				   
			 NEXT WEEK'S TINLUNCH
	      A Representationalist Theory of Intention
		 Kurt Konolige and Martha E. Pollack

We present a theory of intention that begins with an account of the
ways in which agents form, execute, monitor, and revise their plans in
response to multiple goals in dynamic environments.  Our account of
intention is developed by considering the internal planning structure
of agents, and relating it to the agent's behavior in the world.  The
resulting theory is a propositional attitude one in that it depends on
the actions that the agent performs; however, it goes further in
making a commitment to agents having internal plan structures that
affect their behavior.  Because we consider the plan structures to be
sentences in some internal language for describing plans, our approach
is _representationalist_.  We show how increasingly complex notions of
intention arise naturally in the theory, and argue that the
representationalist stance allows us to provide a better
characterization of intention than the most fully developed
alternative proposal, that of Cohen and Levesque.
			     ____________
				   
		       NEXT WEEK'S CSLI SEMINAR
		       Computational Morphology
			   Lauri Karttunen

Morphology, the least glamorous of the maids in the castle of
linguistic theory, is stuck in a bygone area.  In morphology it is
generally assumed that words arise from minimally specified underlying
representations through elaborate derivations.  These derivations
generally involve transformational rules, rule ordering, derivational
cycles, transderivational constraints, and other arcane machinery.
>From the looks of it, morphology is a very difficult subject.  Given
the Byzantine complexity of derivations, it is clear that complete
rule systems are unattainable at the present stage of knowledge.
Articles in morphology typically involve partial descriptions of a
handful of word forms.  These partial accounts only address the
problem of how surface forms are generated from underlying abstract
morphemes.  Generation being as hard as it is, with the tools at hand
recognition is clearly impossible.

Outside the castle of linguistic theory, work has progressed on the
construction of intelligent systems for text processing, machine
translation, voice recognition, and other such mundane practical
tasks.  For these applications it is essential to recognize and
identify words correctly.  A system of rules that can be used for
generation but not for recognition is useless.  Gross incompleteness
is unacceptable.  Recognition and generation must be accomplished in
real time.

Although the state of morphological theory, as practiced in the
castle, virtually precludes them, there now exist systems with
comprehensive lexicons and substantive rule components that generate
and recognize the vast majority of word forms in languages as diverse
as Finnish, English, and Arabic.  These applications are based on two
simple theoretical insights: (1) Morphological alternations can be
characterized mathematically as regular relations -- even in cases
where the traditional way of describing them suggests greater
complexity.  Consequently, rules that describe such alternations can
be implemented as finite-state transducers. (2) By and large, the set
of possible combinations of stems and affixes can be encoded as a
finite-state network.

Although this new technology in itself does not embody any new
linguistic or psychological insights, it raises fundamental issues
that classical morphology has mused over, but not very seriously.
What exactly are morphological derivations about?  What is the status
of rules?  Are they applied to something or true of something?  What
is the structure of the lexicon as a whole?  I will try to explain the
relevance of these issues in the context of computational morphology.
			     ____________
				   
		     NEXT WEEK'S CSLI COLLOQUIUM
	   Precompiled Syntactic Constraints for Continuous
			  Speech Recognition
			    Pierre Dupont

Language models are needed to force speech-recognition systems to
produce syntactically correct sentences.  However, integration of
linguistic knowledge in a recognition system raises two types of
problems.  First, the computational overhead required to handle the
linguistic constraints should be reduced to a minimum.  Second, the
syntactic rules interact with the acoustic decoding and this implies
in particular that the linguistic representation should be able to
work in predictive mode.  More precisely, given a partial sentence
hypothesis, the language model should predict the list of legal word
successors. Since, in fact the acoustic decoding is performed at the
phoneme level, this prediction may be formulated in terms of phonemes
that can follow an end of word hypothesis.

The present approach is based on a language model expressed in the
context-free formalism.  This formalism offers two advantages.  First,
it is more powerful than the classical language models used in
speech-recognition systems like regular grammars or word-pair models.
Second, the context-free formalism is the backbone of several
higher-level linguistic formalism, PATR-II amongst them.

An efficient interface between the acoustic decoding module and the
language model has been designed.  This interface consists in a
representation of the language model by means of recursive transition
networks, which can be dynamically and locally expanded in a
finite-state network during recognition.  Each network associated with
a nonterminal symbol is precompiled offline.  Consequently, the actual
syntactic work during recognition reduces to the insertion of
subnetworks copies. 
			     ____________
				   
		   SEMINAR ON THOUGHT AND LANGUAGE
			   Julius Moravcsik
		      (julius@csli.stanford.edu)
		      Wednesdays, 6:00-8:00 p.m.
			     Cordura 104

I will give a seminar based on material covered in my book _Thought
and Language (Routledge, 1990).  Graduate students from philosophy and
related disciplines can earn credit by signing up for Philosophy 286
(a final paper will be required).

I will spend the first five meetings on the ontology that we need for
an adequate theory of language; then two sessions on the relation of
this material to psycholinguistics or developmental cognitive
psychology; and then three or four sessions on lexical semantics.
Depending on the availability of a couple of visitors, we might switch
the semantics and psychology material around so that the latter will
be discussed at the last sessions.

For the first five sessions, our program is as follows:

-- general introduction, sketch of lexical semantics and the ontology
   it requires, and recent attacks on such ontologies (4/4);

-- the two circles (extensional and intensional) -- why should one
   have priority over the other? (4/11);

-- can there be a noncircular account of universals? (4/18);

-- identity, individuation, and persistence (4/25);

-- the reality of events and their equal status with material
   objects (5/2).

We will work out other topics as we go along.  For those with less
preparation, the first half of my book gives background; in the second
half, the topics mentioned above are discussed and specific proposals
presented.
			     ____________
				   
		   PHILOSOPHY DEPARTMENT COLLOQUIUM
		     A Disproof of Bell's Theorem
			      Yalcin Koc
		Bogazici University, Istanbul, Turkey
			     visiting at
	   Center for the Philosophy and History of Science
			  Boston University
		     Friday, 6 April, 12:00 noon
			Building 200, Room 305

No abstract available.
			     ____________
				   
		   PHILOSOPHY DEPARTMENT COLLOQUIUM
		    Bruno De Finetti's Bayesianism
			Maria Carla Galavotti
		     University of Bologna, Italy
		      Friday, 6 April, 3:15 p.m.
			Building 90, Room 92Q

No abstract available.
			     ____________
				   
		   PSYCHOLOGY DEPARTMENT COLLOQUIUM
	       Spatial Hearing: The Importance of Edges
			     Ervin Hafter
		  University of California, Berkeley
		    Wednesday, 11 April, 3:45 p.m.
			Building 420, Room 050

No abstract available.
			     ____________

			   NEW CSLI VISITOR
				   
			  Yasuyoshi Inagaki
			      Professor
		Department of Information Engineering
		    and Department of Electronics
		       Nagoya University, Japan
		 Dates of visit: 1 April-30 June 1990

Yasuyoshi is on sabbatical from Nagoya University for seven months.
He is interested in formal semantics of languages and theoretical
aspects of natural-language understanding as a foundation for
developing a highly friendly interface between humans and computers.
While at CSLI, he would like to learn about situation theory
and situation semantics and to work on papers on some related topics.
His research interests include reasoning under incomplete information
or commonsense reasoning as well as algebraic approaches to software
development.