[Prev][Next][Index]

CSLI Calendar, 12 April, vol. 5:23




       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
-----------------------------------------------------------------------------
12 April 1990                     Stanford                     Vol. 5, No. 23
_____________________________________________________________________________

    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, 12 APRIL 1990

12:00 noon		TINLunch
      Cordura 100	A Representationalist Theory of Intention
			Kurt Konolige and Martha E. Pollack
			(konolige@ai.sri.com, pollack@ai.sri.com)
			Abstract in last week's Calendar

 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 
			(lauri@csli.stanford.edu)
			Abstract in last week's Calendar

 4:15 p.m.		CSLI Colloquium
      Cordura 100	Precompiled Syntactic Constraints for
			Continuous Speech Recognition
			Pierre Dupont
			Philips Research Laboratory, Brussels
			(dupont@prlb.philips.be)
			Abstract in last week's Calendar
			(Note change of location)
			     ____________

	   CSLI ACTIVITIES FOR NEXT THURSDAY, 19 APRIL 1990

12:00 noon		TINLunch
      Cordura 100	Reading: Harmonic Grammar -- A Formal Multi-Level
			Connectionist Theory of Linguistic Well-Formedness:
			An Application
			by Geraldine Legendre, Yoshiro Miyata, and
			Paul Smolensky	
			Discussion led by Stanley Peters
			(peters@csli.stanford.edu)
			Abstract below

 2:15 p.m.		CSLI Seminar
      Cordura 100	Controversies in Natural-Language Research
			led by Stanley Peters 
			(peters@csli.stanford.edu)
			Title: Developments in Phonological and 
			Morphological Theory
			Speakers: Paul Kiparsky and Bill Poser
			(kiparsky@csli.stanford.edu, poser@csli.stanford.edu)
			Abstract below
			     ____________
				   
			 NEXT WEEK'S TINLUNCH
	  Reading: Harmonic Grammar -- A Formal Multi-Level
  Connectionist Theory of Linguistic Well-Formedness: An Application
      by Geraldine Legendre, Yoshiro Miyata, and Paul Smolensky
		   Discussion led by Stanley Peters

This paper might have been called "Unaccusativity Meets
Connectionism."  The authors observe that a semantic account of
unaccusativity in French, along the lines of Zaenen (1989), does not
cope adequately with the unaccusativity mismatches that plague a
syntactic account.  Unaccusatives are a subclass of intransitive verbs
(e.g., the boldface [marked here with _ before and after the word]
ones in (1a) and (2a)) that occur in certain constructions where other
intransitive verbs (unergatives) cannot (cf. (1b) and (2b)).

(1) (a) La glace est facile a` faire _fondre_.
    (b) *Les e'tudiants sont faciles a` faire _travailler_.

(2) (a) La neige _fondue_ a forme' de la boue.
    (b) *Les e'tudiants _travaille's_ e'taient fatigue's. 

Mismatches exist, as the intransitive verbs that can occur in (1a) are
not precisely the same as those that can occur in (2a).  In fact, a
different set of verbs occurs in each of the ten diagnostic
environments for unaccusativity that exist in French.

The authors argue that a connectionist principle -- neural nets
maximize harmony (a global measure of activation levels and connection
strengths) -- yields an account of unaccusativity in French that is
empirically superior to any categorical semantic or syntactic account.
They studied 190 different combinations comprising an intransitive
verb and its argument noun phrase, testing each for unaccusative
behavior in four major diagnostic contexts.  They trained a neural
network on raw acceptability judgments of the 760 resulting sentences;
it learned to use two semantic features of verbs and two semantic
features of their argument to assign graded acceptability judgments
with the correct polarity in 758 of the cases.

The authors contend that this local neural network can be seen as
approximating another network in which individual units have no
interpretation as linguistic entities (words, phrases, rules, or the
like) but instead "isoharmonically" distribute the computation of the
local network among themselves.  Thus, they contend, the connectionist
framework is able both to fit the facts of human linguistic
performance and to explain how the familiar sort of categorical
linguistic description can approximate it.
			     ____________
				   
		       NEXT WEEK'S CSLI SEMINAR
	Developments in Phonological and Morphological Theory
		     Paul Kiparsky and Bill Poser

_The Sound Pattern of English_, the work underlying virtually all
present-day work in phonology and morphology, provided an impoverished
phonological representation combined with a relatively unconstrained
theory of rules.  Although work in computational phonology and
morphology has in general not advanced beyond this primitive stage,
there have been many developments in morphological and phonological
theory, such as:

- the lexical/postlexical division
- underspecification
- templatic morphology
- tighter constraints on rule form
- the discovery of the role of prosodic constituency

We will review some of the more important of these developments.
			     ____________
				   
		   SEMINAR ON THOUGHT AND LANGUAGE
			   Julius Moravcsik
		      (julius@csli.stanford.edu)
		      Wednesdays, 6:00-8:00 p.m.
			     Cordura 100

There has been a change of location for this seminar, which will now
be held in Cordura 100.
			     ____________
				   
			SYMBOLIC SYSTEMS FORUM
		   The General Theory of Induction
			   Peter Cheeseman
		    Thursday, 12 April, 4:15 p.m.
			Building 60, Room 61G
       
The basic theory of induction of models (theories) from data has
evolved over the last three centuries, and thanks to cheap available
computing, this theory now gives practical solutions to inductive
problems.  Starting from very basic first principles, the Bayesian
calculus is shown to be the theory of belief under uncertainty.  This
theory automatically provides the necessary trade-off between the fit
to the data and the complexity of the model.

Many philosophers in the past have recognized that this trade-off is
necessary (e.g., Ockham's Razor), but the Bayesian theory provides a
quantitative solution.  This solution requires a prior probability
over possible models, and it is this requirement that causes many to
reject the Bayesian solution, because it appears to introduce a
subjective element into inductive inference.

This talk will show the necessity of prior knowledge, and show how
very weak prior knowledge can be translated into a bland
"uninformative" prior.  In addition, the Bayesian approach will be
shown to give answers that do not depend on the language used to frame
the inductive problem.  These philosophical points will be illustrated
by examples of Bayesian induction that far exceed human abilities.

Next talk, 19 April: Overview of Research at the Center for the Study
of Language and Information, Stanley Peters.
			     ____________
				   
		   PHILOSOPHY DEPARTMENT COLLOQUIUM
	  Has the Description Theory of Names Been Refuted?
			     Jerrold Katz
		Graduate School and University Center
		     City University of New York
		     Friday, 13 April, 3:15 p.m.
			Building 90, Room 92Q

No abstract available.
			     ____________
				   
		  LINGUISTICS DEPARTMENT COLLOQUIUM
		Variation in Negative Concord Systems
			     Bill Ladusaw
		 University of California, Santa Cruz
		     Friday, 13 April, 3:30 p.m.
			     Cordura 100

Negative concord is the multiple morphological marking of a single
clausal negation (as exemplified in the "nonstandard" English "I
didn't give nothing to nobody," synonymous with standard "I didn't
give anything to anyone").  The talk presents a semantic account of
negative concord and examines the syntactic variation among dialects
of negative concord in English and the Romance languages.

The gist of the semantic account is the treatment of concordant terms
as Heimian indefinites, which are necessarily bound in the scope of a
clause-level negation operator.  The principal variation in negative
concord systems is described in terms of well-formedness conditions on
the incorporation of categorial chains projected between the
concordant terms and the clause domain over which the negation takes
scope.  These well-formedness conditions state restrictions on the
identity and structural position of the topmost element in the
concordant chain.

Under this view, dialects with no negative concord are ones in which
every morphologically negative constituent must be the topmost element
in a chain.  Among systems that allow negative concord, the principal
variation consists in whether the negation associated with the
inflected head of the clause must be topmost or not.  This is what
distinguishes between the two principal concording dialects of
English.  In one, where the inflection-associated negation must be
topmost, (i) and (ii) are not synonymous.  In the other dialect, they
are.

(i)  Nobody left.
(ii) Nobody didn't leave.

This contrast also divides the Romance languages, putting standard
Spanish and standard Italian in the former group, with Catalan and
some nonstandard dialects of Spanish in the latter group.  (Given the
word-order properties of these languages, it is not possible to
clearly decide whether linear order or hierarchical structure is the
basis for the well-formedness condition, so information about negative
concord in non-SVO languages would be greatly appreciated.)

In concluding, I will explore the relationship between negative
concord and the licensing of the negative polarity determiner "any,"
presenting an argument that the two systems are closely related but
not subject to the same variation in well-formedness conditions.  I
will also examine the subject-object asymmetry in long-distance
negative concord, which is claimed by Kayne (1984) to support the 
Empty Category Principle as a well-formedness condition on logical
forms, and argue that the raising of the concordant term crucial to
this explanation makes incorrect semantic predictions.
			     ____________
				   
			   SYNTAX WORKSHOP
		  Presentative Structure in Chinese
				Fu Tan
		       (tan@csli.stanford.edu)
		     Tuesday, 17 April, 7:30 p.m.
			     Cordura 100

In Chinese, certain verbs may have their logical subject alternate
between a preverbal and postverbal position. Most of these verbs also
have a locative argument.  The locative argument has to be a PP, if
the logical subject is preverbal, but takes the form of a preverbal PP
or NP, if the logical subject is inverted. In the literature,
sentences whose logical subject is inverted are called presentative
sentences.

(a) Xuesheng  zai tai-shang zuo zhe. 
    students  PREP stage-top sit ASP
    "The students are sitting on the stage."

(b) (zai) tai-shang zuo zhe xuesheng.
    (PREP)stage-top sit ASP students
    "On the stage are sitting students."

Li and Thompson (1975, 1978) consider both the preverbal and the
postverbal students invariably as the grammatical subject and take
their varied word order in sentences as evidence for the nonsubject
prominence of Chinese.  Their argumentation is that a nonsubject
prominent language does not have any means to encode its subject as
subject-prominent languages do by word order, subject-verb agreement,
or case-marking, or redundantly by two or three of them at the same
time. The subject in Chinese, as they perceive it to be, is
inconsistently encoded, as far as the word order is concerned.

The subject/topic and subject/object verifying tests show that the
inverted LOC NP is the subject, not the topic, and the inverted
subject is the object, not the subject.  Chinese is a SVO language in
which there is no such thing as a postverbal subject.

With subjects playing such a central role in syntax, as shown in the
grammatical processes and consistently encoded by word order, Chinese
is just as subject-prominent as, say, English.  The only difference is
how languages display their subject-prominence: in English, a subject
not only responds to grammatical processes the same way a subject does
crosslinguistically, but also displays the prominence by case-marking
and agreement with the verb, whereas in Chinese, the subject displays
its prominence only by its stable preverbal position and its responses
to those processes.

The next workshop will be Tuesday, 1 May.  Speakers will be Joan
Bresnan and Sam Mchombo. 
			     ____________
				   
		   PSYCHOLOGY DEPARTMENT COLLOQUIUM
	 Psychological Influences on Coronary Artery Disease
			     Steve Manuck
		       University of Pittsburgh
		    Wednesday, 18 April, 3:45 p.m.
			Building 420, Room 050

No abstract available.
			     ____________
				   
			   NEW CSLI VISITOR
			    Yoshihiro Ueda
		       IAP Visiting Researcher
       ATR Interpreting Telephony Research Laboratories, Japan
		 Dates of visit: 1 April-31 July 1990

Yoshihiro has been working on unification-based bidirectional grammar
generation for dialogue translation.  While at CSLI, he would like to
incorporate the center's developing theories into his system of
generating sentences that appropriately reflect the speaker's
intentions.  His research interests also include the proper syntactic
and semantic representation of sentences, the relationship between
surface illocutionary forces and deep illocutionary forces, and new
mechanisms of unification-based generation.
			     ____________