Copy Link
Add to Bookmark
Report
NL-KR Digest Volume 02 No. 41
NL-KR Digest (5/15/87 09:10:36) Volume 2 Number 41
Today's Topics:
S as an Object of Grammatical Study
Natural/artificial languages
Re: leaks and technical sublanguages
Seminar - Causal Reasoning as Nonmonotonic Temporal Reasoning
From CSLI Calendar, May 14, No.28
ONTIC: Knowledge Representation for Mathematics (MCC)
Unframing the Frame Problem (UTexas)
----------------------------------------------------------------------
Date: Sat, 9 May 87 21:28 EDT
From: David Pesetsky <PESETSKY%cs.umass.edu@RELAY.CS.NET>
Subject: S as an Object of Grammatical Study
I have recently had a look at NL-KR and noted the long discussion
initiated by Mark Edwards of S as an object of study, the
relationship of GB to GPSG syntax, etc. As a syntactician who works
in large part on sentences, whose work falls roughly into what is
called "GB", I find a good deal of the discussion deplorable but,
alas, not surprising. With regard to the broad question of whether
it is right to work on "sentence syntax", the answer can only be
sought in the cogency of the results obtained and the coherence of
the data considered.
I cannot tell for sure, but I strongly suspect that many of the
discussants are not terribly familiar with the work they are
criticizing. Consider, for example, Luigi Rizzi's paper on the
so-called pro-drop parameter (in his book "Issues in Italian Syntax",
Foris 1984). He observed that languages like Italian or Spanish (we
might add Bani-Hassan Arabic, but not Lebanese == cf. Kenstowicz in
NELS 1984) differ systematically from languages like English or
French in 3 distinct areas:
1. the ability to have phonologically unexpressed subjects
E' arrivato in ritardo
2. the ability to place the subject after the verb
E' arrivato Gianni
3. the absence of subject/object asymmetries in WH-questions and
relatives:
cf. someone who I don't know whether Mary talked to yet...
*someone who I don't know whether talked to Mary yet...
both are OK in Italian
Rizzi showed that properties 1 & 2 derive from special properties of
the inflected verb in Italian, but most stunningly, really proved
that 3 in Italian is due to 2: in other words, subjects can be
extracted "long distance" in Italian because they can be placed after
the verb first, and (in effect) get treated on a par with other
post-verbal material, like objects.
He showed this by noting (A) that elements that are demonstrably
preverbal in Italian act just like preverbal elements in French or
English (certain quantifier scope facts, and (B) that other
properties of Italian point to 3 deriving from 2. Conclusion (B) was
soon supported strongly by independent research into clitic pronouns
in the dialects of North Italy by P. Cordin and by the work on Arabic
cited above. Thus, if a language has property 2 it will have
property 3, and if a language has the right kind of inflectional
elements, it will have properties 1 and 2 -- with implications for
what a child has to learn to grow up speaking
Italian/Spanish/Bani-Hassan Arabic vs. French/English/Lebanese
Arabic.
Now who can deny that an Italian speaker's choice of preverbal or
postverbal subject (or his choice to phonologically realize or leave
unpronounced a subject pronoun) has something to do with the
discourse in which a sentence is embedded. Nonetheless, I fail to
see how Rizzi's work -- work on the questions that Rizzi, and I, and
syntacticians find important -- suffers from lack of attention to the
discourse issue. I also fail to see how the world inside the
sentence -- the type of research many or most syntacticians do (this
is accurate) -- is intrinsically less interesting than discourse. I
cannot help but think that if some of the scoffers and doubters could
examine the best of the work being done by those they criticize they
would see that there is real, useful and important work being done on
real and substantive topics. It may not be the work that they want
to do -- but it needs doing, since there is syntax, and
Italian/Spanish/Bani-Hassan Arabic does differ in systematic and
interesting ways from French/English/Lebanese Arabic, and wherever
there is systematicity, one would like to know why.
Sure, we all make mistakes. Often a sentence is judged ungrammatical
when later investigation reveals that some non-syntactic factor is at
work. See, for example, my own paper in an MIT Press volume on
Indefiniteness, edited by E. Reuland and Alice TerMeulen (due out in
June), for an example of the interaction between discourse factors
and pure syntax. But this is to be expected, just as factors
irrelevant to someone's lab experiment (but possibly of intrinsic
interest on their own) may inadvertantly skew the results. And it is
equally as possible that discourse-oriented research may falter over
a sentence-syntactic property of some sort. I fail to see the need
or desirability of polemic over these issues, and yet there has been
so much of it over the years...
David Pesetsky
Dept. of Linguistics
South College, University of Massachusetts
Amherst, MA 01003
------------------------------
Date: Fri, 8 May 87 09:35:57 edt
From: jlynch@nswc-wo.ARPA
Subject: Natural/artificial languages
With regard to the latest ramblings on the topic, may I add the
following. A language is a representation, an abstract if you will,
of the environment which spawns it. Language cannot describe what
it doesn't know or cannot imagine. All languages are "artificial"
in that regard. To say that one language is more or less "natural"
than another is really describing the system that produces or uses
the language, not the language itself.
(I am a minor student of the subject, by the way, not a full-time
expert).
Jim Lynch
jlynch@nswc-wo.arpa
------------------------------
Date: Mon, 11 May 87 10:39:26 EDT
From: Bruce Nevin <bnevin@cch.bbn.com>
Subject: Re: leaks and technical sublanguages
I erroneously attributed a comment about `technical' and natural
languages to Mary Holstege (Holstege@STANFORD.Sushi.EDU). The author
was Mitchell Sundt (mitch@mitre.ARPA). The passage in my message of 1
May (v 2 # 36) should have read as follows:
There is also a confusion of formal languages with technical
sublanguages in the same message:
MS> Perhaps the distinction between technical and natural languages
MS> should be that the intent of the users of technical languages is
MS> more focused, thereby reducing the rate of leakage (except to
MS> further the intent of the speakers), whereas there is no such
MS> focused intent among the users of natural langauges, thereby
MS> hastening the leaking of the grammar.
The main point of my rejoinder here was that formal languages are
distinct from `technical languages' or sublanguages, which are instances
of natural language.
Apologies to Mary, and thanks for pointing out the error. My mailer
does not de-digest, and apparently my eyes do not always do so either!
I hope to get time to respond to her very thought-provoking observations
on written language, punctuation, etc.
In a rather different way, I missed the `real' ID near (but not at) the
end of the message from Obnoxious Math Grad Student (obnoxio@brahms).
Just before the edifying quotations at the end appears the line:
ucbvax!brahms!weemba Matthew P Wiener/Brahms Gang/Berkeley CA 94720
The immediate impetus for inviting readers to reconsider his
contribution came from the way it parallelled my challenge to translate
K&R into C. I also agree about the greater freedom in mathematics for
things like *-transforms, which helps to distinguish it from natural
language because the latter is constrained by both by convention and by
user-perceived correlation with (user-perceived) reality. I would add
that constraint by convention can interfere with constraint by (shall we
say) empirical evidence even in sublanguages of the sciences, and
definitely has the upper hand in less carefully controlled usages; and
that not all mathematicians and philosophers agree with the quote from
Einstein about math (if it's certain it doesn't refer to reality &
converse).
Bruce
------------------------------
Date: Wed, 13 May 87 17:11:10 PDT
From: Amy Lansky <lansky@venice.ai.sri.com>
Subject: Seminar - Causal Reasoning as Nonmonotonic Temporal
Reasoning (SU)
CAUSAL REASONING AS NONMONOTONIC TEMPORAL REASONING
Yoav Shoham (SHOHAM@SCORE.STANFORD.EDU)
Stanford University
11:00 AM, MONDAY, May 18
SRI International, Building E, Room EJ228
This talk will address the following topics:
* A definition of the problems of Qualification and Extended Prediction,
and their relation to the Frame Problem.
* An outline of a semantical approach to nonmonotonic logics.
* A definition of a specific nonmonotonic epistemic logic, the logic
of Chronological Ignorance, and a demonstration of its utility in
solving the two problems mentioned above.
I will argue that the above analysis explains the meaning of causation,
and its central role in commonsense reasoning.
VISITORS: Please arrive 5 minutes early so that you can be escorted up
from the E-building receptionist's desk. Thanks!
------------------------------
Date: Thu 14 May 87 10:00:48-PDT
From: Emma Pease <Emma@CSLI.Stanford.EDU>
Subject: From CSLI Calendar, May 14, No.28
Tel: (415) 723-3561
NEXT WEEK'S TINLUNCH
Reading: "Active Zones"
by Ronald W. Langacker
Discussion led by Chris J. Pinon
May 21
Although many linguists exclusively associate R. Langacker with the
discovery of the now immortal notion `command' back in the late '60s,
he has by no means gone into hiding ever since his disillusionment
with the transformational regime began in the mid '70s. Au contraire,
he has undertaken a fundamental rethinking of linguistic theory in the
meantime, most of which has appeared over the years in a battery of
publications, culminating last month with an article in `Language'
("Nouns and Verbs") and with his Stanford Univ. Press book
`Foundations of Cognitive Grammar' (v. 1). He is indisputedly
recognized as one of the founders and major proponents of `cognitive
grammar' (formerly known as `space grammar').
The article I have chosen is one of the shorter things Langacker
has written in recent times and it serves as an intriguing, useful,
and very readable piece on how he operates in linguistics
(specifically semantics). In this article Langacker introduces the
notion `active zone' ("facets of an entity which participate directly
in a given relation") and demonstrates its usefulness in accounting
for the semantics and syntax of a variety of sentence types previously
considered to be distinct (pay special attention to his account of
raising). Although itself a profitable topic for discussion, the
paper also serves as a springboard for how the notions `subject' and
`object' are reconstructed in Langacker's theory. Drawing from other
work of his, I will show how he views these grammatical relations to
be an instantiation of his `trajector/landmark' schema (which is
briefly mentioned in this article). This all raises an important
question: What ARE `subjects' and `objects'? We will consider
Langacker's answer, as well as how other theories have dealt with this
question.
--------------
NEXT WEEK'S CSLI SEMINAR
A Discussion of the Institute for Research on Learning
James G. Greeno
May 21
IRL is founded on two premises: (1) American education is failing, (2)
The most important thing to do about that is to conduct fundamental
research about the nature of learning.
This makes sense if you believe, as we do, that contemporary
educational practice rests on faulty epistemological premises. Of
course, educators did not invent this epistemology. They express in
their practice the views about structures and acquisition of knowledge
that permeate the sciences of artificial intelligence, cognitive
psychology, and linguistics. Our prime goal is to lead in a
fundamental reformulation of the epistemological grounding of
educational research and practice, and to back that with solid
scientific analyses and research.
Our current research activity consists of interdisciplinary
discussions to create significant common ground among individuals with
apparently incompatible perspectives and insights. One set of ideas
that is emerging attacks the currently received premise that knowledge
is basically context-free and disembodied. Alternative ideas are that
knowledge is capability for situated activities, including
communication with others, that these activities should be understood
as practices, and that to learn these practices, fundamentally
different forms of learning are needed, including intellectual
apprenticeship and collaborative learning.
------------------------------
Date: Fri 3 Apr 87 08:39:00-CST
From: Ellie Huck <AI.ELLIE@MCC.COM>
Subject: Seminar - ONTIC: Knowledge Representation for Mathematics
(MCC)
Please join the AI Group for the following talk:
David McAllester
MIT AI Lab
April 7 - 10:30am
MCC Auditorium
"Ontic: A Knowledge Representation
Language for Mathematics"
Ontic is an interactive system for developing and verifying
mathematics. The system appears to be able to verify "proofs" that
are only one to three times longer than corresponding previously
published English arguments. Furthermore, the structure of the
machine readable proofs closely matches the structure of the English
arguments. Ontic's ability to read concise proofs is based on a
mechanism for automatically finding and applying information from a
lemma library containing hundreds of mathematical facts. Starting
with only the axioms of Zermello Fraenkel set theory, the Ontic system
has been used to build a data base of definitions and lemmas
culminating in a proof of the Stone representation theorem for Boolean
lattices. This proof involves an ultrafilter construction and is
similar in complexity to the Tychonoff theorem that an arbitrary
product of compact spaces is compact. This talk will discuss the
structure of Ontic's machine readable proofs, the automatic theorem
proving mechanisms used, and the empirically observed differences
between Ontic's proofs and English arguments.
April 7 - 10:30am
MCC Auditorium
------------------------------
Date: Tue 14 Apr 87 13:17:08-CDT
From: AI.CHRISSIE@R20.UTEXAS.EDU
Subject: Seminar - Unframing the Frame Problem (UTexas)
UNFRAMING THE FRAME PROBLEM
Dennis de Champeaux
Hewlett Packard Labs
Palo Alto, California
Date: Thursday, April 16, 1987
Time: 3:00 pm - 4:00 pm
Where: Taylor Hall 3.128
The predicate calculus in uncommitted to any ontology. Consequently it has a
nearly boundless domain of application. The price of this generality is that
some pervasive properties of the domain are represented at great implementation
cost. The Frame Problem is an example par excellence. To cope with this
problem, we propose to employ a fragment of intensional logic. Consequently
frame axioms or their equivalent have only to be injected for those entities
that are affected by an event, i.e., those for which a new extension must be
introduced. The situation calculus allows the description of a state of
affairs to coexist with the history. This property is preserved in our
proposal. The formalism has been implemented in the context of program
verification.
------------------------------
End of NL-KR Digest
*******************
