nl-kr-request@CS.ROCHESTER.EDU (NL-KR Moderator Brad Miller) (02/24/88)
NL-KR Digest (2/23/88 14:10:46) Volume 4 Number 21
Today's Topics:
Re: What is a grammar (for)
BBN Lang. & Cognition Seminar
SUNY Buffalo Comp. Sci. Colloq: Hirst/Ambiguity
BBN AI Seminar -- Dietmar Roesner
Lang. and Cognition Seminar - McNeill
SUNY Buffalo Comp. Sci. Colloq: James Allen
Submissions: NL-KR@CS.ROCHESTER.EDU
Requests, policy: NL-KR-REQUEST@CS.ROCHESTER.EDU
----------------------------------------------------------------------
Date: Tue, 16 Feb 88 11:23 EST
From: Rick Wojcik <rwojcik@bcsaic.UUCP>
Subject: Re: What is a grammar (for)
In article <2628@dciem.UUCP> mmt@dciem.UUCP (Martin Taylor) writes:
>A question that is simple on the surface, but I suspect not so simple in
>implication: "What is a grammar, and what is a grammar for?"
These are very good questions. They lie at the foundations of
linguistic theory. The generally accepted (generative) position is that
a grammar is a set of rules which generate all and only the strings of a
language. The question of what a grammar is for is not so easily
answered. Generative grammars are supposed to have psychological
relevance in that they are directly linked with intuitions of
grammatical well-formedness. But I do not know of any generativist
works that make explicit how grammars are used to produce such
intuitions. It is easy to see that grammars generate all possible
interpretations for a given string of words (but perhaps not "pragmatic"
interpretations). The problem is that speakers don't perceive all
grammatical interpretations. For example, Bever's GP sentence 'The
horse raced past the barn fell' [read: 'the horse WHICH WAS raced past
the barn BY SOMEONE fell'] is normally interpreted as ungrammatical
outside of a context. Despite the fact that grammatical
intuitions are crucial to the validity of any data set examined by
generative grammarians, they have no principled method of connecting
intuitions to grammars.
A second problem has to do with how we produce and understand
utterances. Given that grammars somehow get used to produce
grammaticality judgments, what role do such judgments play in language
understanding? It is easy to imagine that grammars can be used to
render the interpretation of utterances more predictable. The problem
comes in when we try to understand how ungrammatical speech--a pervasive
phenomenon--is understood. Grammaticality intuitions may play a role in
in the interpretation of grammatical speech, but they are a distinct
liability in the interpretation of ungrammatical speech. So why does
the 'horse' sentence favor an ungrammatical reading outside of a
disambiguating context? Why don't we automatically perceive the
grammatical reading? Grammars are supposed to help us to perceive
grammatical readings, aren't they?
Generative grammarians have thought of these questions, but they have
yet to come up with useful answers. What are grammars and what are they
for? These are two very good questions.
--
Rick Wojcik csnet: rwojcik@boeing.com
uucp: {uw-june uw-beaver!ssc-vax}!bcsaic!rwojcik
address: P.O. Box 24346, MS 7L-64, Seattle, WA 98124-0346
phone: 206-865-3844
------------------------------
Date: Tue, 16 Feb 88 13:30 EST
From: Dori Wells <DWELLS@G.BBN.COM>
Subject: BBN Lang. & Cognition Seminar
BBN Science Development Program
Language and Cognition Seminar Series
THE ROLE OF EVENTS IN LEXICAL SEMANTICS
James Pustejovsky
Computer Science Department
Brandeis University
BBN Laboratories Inc.
10 Moulton Street
Large Conference Room, 2nd Floor
10:30 a.m., Thursday, February 18, 1988
Abstract: It is now generally accepted that a semantics based on events is
able to capture many significant generalizations missed by more
traditional approaches to meaning. Important as this change is, it
has had little impact on theories of lexical semantics and the nature
of the semantic functions (e.g. thematic relations) associated with
verbs. In this talk, we propose an event semantics, where it is
the topology of the event itself which defines the aspectual classification of
a verb or sentence. As a result of this finer-grained, subeventual
structure, thematic relations (or case roles) are a derivative notion
and play no primary role in determining the meaning of a verb (but may
play a role in language learnability). We define a calculus of aspect
where verbs are represented as a sequence of events and states. By
enriching the substructure of events, we overcome Parsons' and
Higginbotham's difficulties with the imperfective paradox and
adverbial modification. Finally, we explore the event-like properties
of relational nominals, and propose that their lexical semantics makes
reference to a ``hidden event'' variable. This has strong
implications for theories of polysemy and procedures in word-sense
disambiguation.
-------
------------------------------
Date: Thu, 18 Feb 88 11:17 EST
From: William J. Rapaport <rapaport@cs.Buffalo.EDU>
Subject: SUNY Buffalo Comp. Sci. Colloq: Hirst/Ambiguity
STATE UNIVERSITY OF NEW YORK AT BUFFALO
DEPARTMENT OF COMPUTER SCIENCE
COLLOQUIUM
THE REPRESENTATION AND RESOLUTION OF NATURAL LANGUAGE AMBIGUITY
Graeme Hirst
Department of Computer Science
University of Toronto
There are many different kinds of ambiguity in natural
language, and an NLU program needs to be able to deal with
all of them. Resolving ambiguity requires two actions:
determining what the possibilities are, and then choosing
among them. I will describe work on ambiguities of word
meaning, thematic structure, and description, with some
emphasis on the degree of psychological reality embodied in
each component.
Date: Thursday, 25th February, 1988
Time: 3:30 pm to 4:30 pm
Place: Bell 337, Amherst Campus
Danish and Coffee will be served at 4:30 pm at Bell 224.
For further information, call (716) 636-3199.
------------------------------
Date: Fri, 19 Feb 88 19:22 EST
From: Marc Vilain <MVILAIN@G.BBN.COM>
Subject: BBN AI Seminar -- Dietmar Roesner
BBN Science Development Program
AI Seminar Series Lecture
THE SEMSYN GENERATION SYSTEM:
INGREDIENTS, APPLICATIONS, PROSPECTS
Dietmar Roesner
Institute for Integrated Publication and Information Systems
Darmstadt, Germany
(unido!fistig!semsyn%uunet.uu.net@relay.cs.net)
BBN Labs
10 Moulton Street
2nd floor large conference room
10:30 am, Friday February 26
The SEMSYN generator for German was initially implemented for use within
a joint Japanese/German machine translation project. Since then it has
been applied to a variety of generation tasks in both machine
translation and text generation:
--generation from semantic structures produced by CMU's Universal
Parser,
--generation of news stories from data,
--generation of descriptive texts related to geometric constructions.
In recent experiments the SEMSYN generator has been extended to produce
(rudimentary) English as well.
We will review the various applications and discuss their importance
for the evolution of the system. Special emphasis will be placed on
questions related to future work towards multilingual generation.
*********************************************************
* *
* If machines are available, a demonstration of the *
* German/Japanese MT application will follow the talk. *
* *
*********************************************************
------------------------------
Date: Tue, 23 Feb 88 08:55 EST
From: Dori Wells <DWELLS@G.BBN.COM>
Subject: Lang. and Cognition Seminar - McNeill
BBN Science Development Program
Language & Cognition Seminar Series
TOWARDS A MODEL OF CONTEXTUALIZED SPEECH:
EVIDENCE FROM REAL TIME SPEECH/GESTURE SYNCHONIZATION
Professor David McNeill
Linguistics and Behavioral Sciences
University of Chicago
BBN Laboratories Inc.
10 Moulton Street
Large Conference Room, 2nd Floor
10:30 a.m., Thursday, March 3, 1988
Abstract: Gestures exhibit imagery that is closely
synchronized with the
semantically parallel parts of the stream of speech. But gestures
differ from speech in how they represent meaning and in their relative
lack of conventional specification. Thus gestures enable the analyst
to set a conventionalized system of linguistic code elements side by
side with a gesture performance that is not specifically
conventionalized, giving two coordinated but distinct views of what is
arguably one underlying process. Using examples from videotaped
materials, Dr. McNeill will illustrate some of the insights into the
mental operations carried out be speakers in real time that can be
inferred from consideration of coordinated gesture and speech data.
------------------------------
Date: Tue, 23 Feb 88 10:18 EST
From: William J. Rapaport <rapaport@cs.Buffalo.EDU>
Subject: SUNY Buffalo Comp. Sci. Colloq: James Allen
UNIVERSITY AT BUFFALO
STATE UNIVERSITY OF NEW YORK
DEPARTMENT OF COMPUTER SCIENCE
COLLOQUIUM
The Architecture of Discourse Systems
Dr. James Allen
Department of Computer Science
University of Rochester
A system that can understand and partake in an extended
dialog must be comprised of many diverse processing mechan-
isms: syntactic and semantic analysis, reference analysis,
speech act analysis, the recognition of the other speakers
plans and goals, the identification of topic structure, and
much more to do with generating appropriate responses. While
there has been alot of work in the last decade on each of
these problems, there has been very little work concerned
with how each process can be integrated into a complete sys-
tem. In this talk I will summarize some of work done in the
areas of reference, speech act analysis, plan recognition
and discourse structure and then suggest how this work might
be integrated into a complete system capable of participat-
ing in an extended two-person dialog. This framework is
currently being tested in an exploratory system under
development at Rochester.
Date: Tuesday, 1st March, 1988
Time: 3:30 pm to 4:30 pm
Place: Bell 337, Amherst Campus
Danish and Coffee will be served at 4:30 pm at Bell 224.
------------------------------
End of NL-KR Digest
*******************