[Prev][Next][Index]

CSLI Calendar, 31 May, vol. 5:30




       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
-----------------------------------------------------------------------------
31 May 1990                     Stanford                       Vol. 5, No. 30
_____________________________________________________________________________

    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, 31 MAY 1990

12:00 noon		TINLunch
      Cordura 100	Spatial Mental Models
			Barbara Tversky
			Department of Psychology
			Stanford University
			(bt@psych.stanford.edu)
			Abstract in last week's Calendar

 2:15 p.m.		CSLI Seminar
      Cordura 100	Controversies in Natural-Language Research 7
			led by Stanley Peters 
			(peters@csli.stanford.edu)
                        Title: Propositional Attitudes and Russellian
			Propositions
			Speaker: Robert C. Moore
			Artificial Intelligence Center, SRI International
			(bmoore@ai.sri.com)
			Abstract in last week's Calendar
			     ____________

	    CSLI ACTIVITIES FOR NEXT THURSDAY, 7 June 1990

12:00 noon		TINLunch
      Cordura 100	Computers and Musical Structure
			Jonathan F. Hallstrom
			Department of Music
			Colby College
			(visiting CCRMA)
			(colby!jfhallst@uunet.uu.net)
			Abstract below

 2:15 p.m.		CSLI Seminar
      Cordura 100	Controversies in Natural-Language Research 8
			led by Stanley Peters 
			(peters@csli.stanford.edu)
			Title: Possible Worlds Semantics and Situation
			Semantics
			Stanley Peters
			Abstract below
			     ____________
				   
			 NEXT WEEK'S TINLUNCH
		   Computers and Musical Structure
			Jonathan F. Hallstrom

Music, like language, is much more than the sum of its parts.  Yet
musicians traditionally spend the bulk of their training learning only
those parts (scales, intervals, etc.), with little time devoted to
addressing in any firsthand fashion how the parts combine to make the
musical structures that we actually hear when we listen to music.  The
ability of computers to create and manipulate musical "macros" (e.g.,
motives, phrases, sections, etc.)  provides interesting opportunities
for a deepened understanding of how music really works.
			     ____________
				   
		       NEXT WEEK'S CSLI SEMINAR
	     Controversies in Natural-Language Research 8
	  Possible Worlds Semantics and Situation Semantics
			    Stanley Peters

Stanley Peters will present arguments for advantages of situation
semantics relative to possible worlds semantics in analyzing such
"generalized modal" notions as conditionals and attitudes.  The two
semantic approaches take quite different strategies to explaining
phenomena such as the limited transitivity of conditionals, and Peters
will show, for instance, that possible worlds semantics requires
certain apparently false conditionals to be true, while situation
semantics does not.

Mark Gawron will respond briefly in support of possible worlds
semantics, to initiate general discussion of the issues raised.
			     ____________
				   
			SYMBOLIC SYSTEMS FORUM
		     Thursday, 31 May, 4:15 p.m.
			Building 60, Room 61G

Seniors in the Symbolic Systems honors program present their work.
			     ____________
				   				   
		   PHILOSOPHY DEPARTMENT COLLOQUIUM
	 Natural Right, Absolutism, and Political Obligation
			  Pasquale Pasquino
	 Centre National de la Recherche Scientifique, Paris
		      Friday, 1 June, 3:15 p.m.
			Building 90, Room 91A

No abstract available.
			     ____________
				   
		  LINGUISTICS DEPARTMENT COLLOQUIUM
			   Formal Phonology
		       (dissertation proposal)
			    Andra's Kornai
		      (kornai@csli.stanford.edu)
		      Friday, 1 June, 3:30 p.m.
			     Cordura 100

More than ten years after the pioneering work of Leben (1973),
Williams (1971, 1976), Goldsmith (1976), Clements (1976), McCarthy
(1979), and many others, fundamental questions relating to the basic
assumptions, notions, and utility of autosegmental phonology have
still not received satisfactory answers.  For example, it is not yet
determined what a well-formed autosegmental representation should be;
what a well-formed rule or a properly stated constraint would look
like; or what the denotation of the autosegmental notation should be
taken to be.

Even more disturbing, whether it is important to answer these
questions is not at all certain, since even more important issues
relating to the inherent usefulness of autosegmental phonological
theory remain unresolved. To date, there is no theoretical proof that
autosegmental phonology is more (or less) powerful than linear (SPE)
phonology.  Furthermore, no practical demonstration in the form of a
speech recognition/synthesis application has shown autosegmental
phonology to be superior to linear phonology. On the contrary. Systems
based on more sophisticated linguistic theories -- including
autosegmental phonology -- have been outperformed at every turn by
Hidden Markov speech recognition devices.  Such models, based on the
completely discredited (Chomsky 1957) finite state model of language
production, have made tremendous gains, from 60% (Lowerre 1976) to 96%
(Lee 1989) recognition accuracy, and are now the undisputed industry
standard.

This thesis explores the possibility of turning autosegmental
phonology into a theory useful for speech recognition.  A
formalization of autosegmental phonology is undertaken in order to
establish that sophisticated nonlinear notation has the same
generative capacity as the mundane finite state mechanism. In order to
demonstrate the usefulness (if any) of autosegmental ideas, I
introduce a class of of Markov-like models that have their structure
determined by "feature geometry" constraints.

The central empirical issue of the thesis is timing. Autosegmental
theory assumes that phonetic events and subevents are concatenated in
a nonlinear manner (permitting a certain amount of desynchronization
among tiers), while Markov models assume simple (linear)
concatenation.  While the duration characteristics of individual
phones are predicted to follow the same (Gamma) distribution in both
models, autosegmental theory offers the possibility of making the
parameters of the model more perspicuous.

The main theoretical contribution of the thesis is the formalization
of autosegmental phonology that fulfills two functions. On the one
hand, it opens the way for a theoretical assessment of autosegmental
theory in terms of mathematical linguistic (generative capacity) and
information-theoretic (channel capacity) results. On the other hand,
the formalization links autosegmental theory to Markov modeling, and
thus makes it possible to evaluate competing versions of autosegmental
theory in a new, nontypological fashion.
			     ____________
				   
			  PHONOLOGY WORKSHOP
			     Polytonicity
			     Ove Lorentz
		     University of Tromso, Norway
		       (visiting UC Santa Cruz)
		       (lorentz@ucscc.uscs.edu)
		      Tuesday, 5 June, 7:30 p.m.
			      Ventura 17

Polytonic languages, according to Jakobson (1931) (TCLP 4), have
different tone sequences assigned to morae in stressed syllables.  It
is usually assumed that such a distinction is present in the lexicon
in languages with polytonic stress (tonal accent), such as
Scandinavian or Serbo-Croatian, but not in monotonic languages, such
as English.

In this talk, I will present an alternative analysis of Scandinavian
polytonicity, which factors out one tone as a default tone given to
syllables with primary stress by a rule of the postlexical phonology.
Such a default tone is also assigned in English and other monotonic
languages. The default tone is found in both accent 1 and accent 2
words in Scandinavian, but only accent 2 words bear lexical tone.

The assignment of a default primary stress tone to a word already
bearing a lexical tone has interesting effects.  One is the so-called
delayed tone-effect, which is thus accounted for. In the case of
similar tones we see OCP-related buffering effects.  One of the
several phenomena that illustrates this is the so-called double-peaked
accent of many dialects.  Docking of tones subject to expulsion from
primary stressed syllables, has repercussions for parametric theory,
which will be briefly discussed.

An attempt will also be made to extend the analysis to other languages
with lexical tone specifications, such as Chinese.
			     ____________
				   
			   POETICS WORKSHOP
	   Unstressed Syllables in the Rhythmic Patterns of
		      Contemporary Russian Verse
			   Vycheslav Ivanov
			  Moscow University
		      Thusday, 7 June, 4:00 p.m.
			      Ventura 17

This talk, which was scheduled for 29 May, had to be postponed to this
new date.
			     ____________