🚀 go-pugleaf

From: wahlster@cs.uni-sb.de (Wolfgang Wahlster)
To: mod-ki@unido.informatik.uni-dortmund.de
Date: Thu, 14 Mar 91 21:50:12 +0100
Subject: <mod-ki 0671> Prague Summer School 1991 (see also mod-ki 669)
Keywords: Schools

mod-ki message 0671
************************************************************************
ABSTRACTS OF THE COURSES

The Types of Lexical Information  for  a  Dictionary  in  an  Integrated
Linguistics Description
Jurij D. Apresjan, Academy of Sciences, Moscow, USSR

An  integrated  linguistic  description  is  a  unified  description  of
dictionary and grammar bringing those two components in  perfect  accord
with each other  in  the  types  of  information  included.  Under  this
principle every lexeme  (word  meaning)  in  the  dictionary  should  be
assigned many new  types  of  information  (phrasal  stress,  intonation
patterns, communicative structure, and word pragmatics included),  while
such  traditionally  recorded  types  of  information   as   morphology,
semantics, patterns of government, and selectional  restrictions  should
be considerably expanded. Failing this  dictionary  will  be  unable  to
interact with grammar effectively.

Lexicography for Computational Linguistics
B.T.Sue Atkins, Oxford University Press, UK

These sessions are designed for  computational  linguists  as  users  of
machine-readable  dictionaries  and  compilers   of   dictionaries   for
computers, and will consist of an overview of dictionary compiling,  and
a case study of a  specific  dictionary  entry.  The  objective  of  the
overview to offer an insight into the dictionary and how it is made, and
topics will include: dictionary design; the task of  the  lexicographer;
and the components of the dictionary entry. The case study  starts  from
an analysis of approximately 2,000 corpus citations for one  word  (held
in a relational database), and works step by step through the process of
compiling an entry for different types of dictionary  on  the  basis  of
facts from the corpus.

Meaning and Understanding in MT
Christian Boitet, G.E.T.A., Grenoble, France

   1.  Text, context, meaning and understanding
   2.  Human understanding in translation
       2.1. Knowledge types
       2.2. Understanding not too little and not too much
       2.3. Aspects of translation beyond understanding
   3. An introduction to MT
       3.1. 4 situations for M(A)T
       3.2. System architectures (External)
       3.3. System architectures (Internal)
   4. Meaning and understanding in MT
       4.1. Meaning and representation of meaning
       4.2. Implicit, explicit and indirect understanding
       4.3. Personal MT, indirect understanding and word senses

Computational Semantics
Jens Erik Fenstad, University of Oslo, Norway

Today we have several systems which search in, rearrange and check texts
for a number of purposes. These  systems  are  necessary  tools  in  any
processing of information. But they  have  limitations.  The  underlying
algorithms typically operate on the given text as a set  of  symbols  of
signs - as a  stream  of  machine  codes.  But  signs  are  carriers  of
structure and meaning. This is a fact that we have to take seriously  if
we are to develop "true" information processing  systems.  And  it  also
sets a goal for computational semantics, a field of study which lies  at
the intersection of three disciplines: linguistics, logic  and  computer
science. A text processing system should provide  a  coherent  framework
for relating linguistic form with semantic content. And the relationship
must be algorithmic.

Semantic Interpretation and Construction Grammar
Charles J. Fillmore, University of California, Berkeley, USA

The course will describe the effort to construct a formal grammar  of  a
language whose central notion is that of "grammatical  construction",  a
notion generalized to include lexical items,  since  lexical  items  are
described in terms of the grammatical constructions which can or must be
built around them. The  semantics  will  be  of  the  "frame  semantics"
variety, the general interpretive mechanism will involve unification and
various inference principles. Among the topics covered will be (1)  some
cross-linguistic observations on the semantics of conditional sentences,
(2) the semantics of scales and negation (drawing heavily on the work of
Paul  Kay),  (3)  the  grammar  and  semantics  of  "chance"  (based  on
collaborative work with B.T. Sue Atkins), and  (4)  new  thoughts  about
semantic roles.

A  Functional  Approach  to  the  Meaning  of  the   Sentence   and   to
Intersentential Links
Eva Hajicova and Petr Sgall, Charles University, Prague,  Czechoslovakia

Theta roles and other valency  relations  (kinds  of  complementations),
understood as the core of underlying sentence structure, are established
and classified by means of operational criteria. Their systemic ordering
allows for an economical description of topic, focus and underlying word
order, which are  semantically  relevant  for  scopes  of  negation,  of
quantifiers and of  focalizers.  Underlying  representations  reflecting
these aspects of sentence structure are  suitable  starting  points  for
semantic interpretation, inferencing and cognitive  analysis.  Reference
assingment and anaphoric relations are  handled  on  the  basis  of  the
salience in the stock of shared knowledge.

Cognitive Semantics
George Lakoff, University of California, Berkeley, USA

The course discusses recent research on conceptual systems and  the  way
that they function in grammar. Topics will include:
1. The  metaphor  system,  a  hierarchical   system   of   hundreds   of
   cross-domain mappings. Metaphorical  mappings  characterize  abstract
   concepts in terms of more concrete concepts and characterize abstract
   inference patterns in terms of spatial inference patterns.
2. Frame semantics: Frames are conventional conceptual  structurings  of
   situations and scenarios.  Lexical  items  are  defined  relative  to
   frames. Each semantic field is the set of lexical  items  defined  in
   terms of a frame.
3. Categories: Categories are not, in general, defined by necessary  and
   sufficient  conditions.  We  will  survey  common   category   types,
   especially basic-level categories and a variety  of  prototype  based
   categories, especially metonymic and radial categories.
4. Construction grammar: Constructions are direct pairings of conceptual
   and  communicative  conditions  with  surface  forms.  Grammars   are
   hierarchically structured networks of such constructions.

Contextual Influences on Meaning
Martha E. Pollack, SRI International, USA

Utterances do not occur  in  isolation.  The  meaning  of  an  utterance
depends heavily on ites context of use, including the  larger  discourse
of which it is a part. Moreover, the meaning of  a  whole  discourse  is
typically more than the sum of the meanings of the individual  sentences
it comprises. In this course, we will consider contextual influences  of
meaning, focusing on computational techniques that  have  been  proposed
for determining the meaning of utterances in context. In  particular  we
will discuss the  determination  of  discourse  structure,  the  use  of
context in resolving  phrase-level  phenomena  such  as  referential  or
syntactic ambiguity, aspects of utterance meaning  affected  by  context
(e.g., conversational implicature), and the question of how the  meaning
of an entire discourse or discourse segment can be derived.

Computational Lexical Semantics
James Pustejovsky, Brandeis University, USA

In  this  course,  we  will  examine  recent  computational   approaches
to lexical  semantics  and  how  they  relate  to  theories  of  lexicon
organization and knowledge representation.  In  particular,  we  explore
how computational approaches  to  inheritance  and  representation  have
reaffirmed the importance of lexical structure  within  linguistics  and
NLP. Topics will include: the structure of  lexical  knowledge;  methods
of lexical decomposition; compositionality in the lexicon; the  creative
use of words and the problem of polysemy; type coercion, type  shifting,
and semantic selection;  default  inheritance  and  global  organization
of the lexicon;  and  lexical  inference  procedures.  We  will  explore
object-oriented approaches to  semantics  based  on  richer  notions  of
compositionality.

Formal Semantics
Mats Rooth, AT&T Bell Labs, USA

Model theoretic approaches to natural language semantics  (for  instance
Montague grammar and discourse representation theory) aim to incorporate
a theory of meaning into generative grammar, in the form of a system  of
rules which associate semantic values (which  in  practice  are  objects
constructed with a mathematical toolkit) with the phrases of a syntactic
analysis. The course introduces the technique in the context of specific
problems (predication, anaphora, quantification, temporal phenomena) and
grammar fragments. There are no mathematical prerequisities.

New Developments in Grammar Formalisms
Hans Uszkoreit, University of Saarbruecken, Germany

A family of  new  grammar  formalisms  has  become  very  successful  in
theoretical linguistics as well as in natural language  processing.  The
members  of  this  family  share  the  utilization  of  complex  feature
descriptions and operations that construct, combine and test  them.  The
course is an introduction to the underlying notions  of  the  family  of
unification grammars. It will also give an overview of some  wide-spread
unification  grammar  models  such  as   Lexical   Functional   Grammar,
Generalized  Phrase  Structure  Grammar,  Head-driven  Phrase  Structure
Grammar, Functional  Unification  Grammar,  and  Categorial  Unification
Grammar. Special emphasis will be placed on the integration of different
linguistic  levels  in  some  of  these  models,  especially  on   their
syntax-semantics interface.

Discourse and User Models
Wolfgang Wahlster, University of Saarbruecken, Germany

It has become evident that flexible  user-oriented  dialog  behavior  of
natural language systems can be achieved only if the system  has  access
to a model of the  user  containing  assumptions  about  his  background
knowledge as well as his goals  and  plans  in  consulting  the  system.
Research in the field of user modeling investigates how such assumptions
can be automatically created, represented and exploited by  the  system.
This course surveys the methods for  constructing  and  exploiting  user
models and illustrates their most important applications by examples  of
realized systems such as GUMS, IMP, KNOME, ROMPER, SC-UM and TAILOR.  In
contrast to a user model, a discourse model contains descriptions of the
syntax and semantics of discourse segments, the attentional structure of
the dialog, and the individual utterances on the speech  act  level.  We
show how user and discourse models can be used to tailor the interaction
to the individual user, to increase the system's cooperativeness, and to
correct or even prevent misconceptions  by  the  user.  We  examine  the
impact  of  the  user  and  the  discourse  model  on  lexical   choice,
over-answering,  and  the  generation  of  definite   descriptions   and
anaphora. Finally, some hot topics in  current  research  and  important
open problems of user and discourse modeling are discussed.
************************************************************************