Copy Link
Add to Bookmark
Report
NL-KR Digest Volume 03 No. 28
NL-KR Digest (10/01/87 18:29:22) Volume 3 Number 28
Today's Topics:
right term for CD and similar systems
Long words are needed
metaphor blindness
Re: Indescribably Delicious
Re: Natural kinds
Re: Neural computing / Speech processing
----------------------------------------------------------------------
Date: Sat, 19 Sep 87 19:48 EDT
From: LEWIS%cs.umass.edu@RELAY.CS.NET
Subject: right term for CD and similar systems
A fairly simple question (perhaps with no answer): what is the right term to
use in referring to a system such as Schank's Conceptual Dependency or
Cullingford's ERKS? Some of the terms I've seen (in references of various
vintages) are: "meaning representation", "representational system",
"representational formalism", "representation langauge",
"representation scheme",etc. I'm wondering if there's an agreed-upon term
by now, and in particular if there are terms that clearly distinguish things
like CD which are essentially abstract qualitative models from things
like KL-ONE or KRYPTON which are essentially programming languages for
building knowledge representations (correct me if I have my examples wrong!).
Thanks, David D. Lewis
University of Massachusetts, Amherst
LEWIS@cs.umass.edu
------------------------------
Date: Tue, 22 Sep 87 09:58 EDT
From: "Martin Minow THUNDR::MINOW ML3-5/U26 223-9922" <minow%thundr.DEC@decwrl.dec.com>
Subject: Long words are needed
Recently, it was pointed out that the Scandinavian languages have no
boundaries on word length due to a very productive compounding mechanism.
Unless I misunderstood the posting, at least one respondant questioned
whether these were "real" compounds, or just a writing convention.
I would suggest that the compounding is real, and required in order
to tell the reader how to pronounce the word -- the compound is stressed
as one word. (Compare the similar phenomena in English words such as
"shoe maker" and "shoemaker".) By the way, the longest word I have
seen in Swedish -- at least, it was the longest word in all active
Swedish laws up to 1972 -- is
"arbetsmarknadsstyrelsenskonflictresolutionskommitte", meaning
"labor relations board's conflict resolution committee"
In my non-native pronunciation, it gets primary stress on the last
syllable (the French borrowing committee) and a secondary stress
on the first syllable of "konflict" (probably an English borrowing).
In modern (1980's) Sweden, practically the entire population (and certainly
all literate adults born after World War II) are fluent in English.
This has led to all manner of unassimilated English constructions entering
the language, such as "baby" for young infant (displacing "spaedbarn"),
and the basterdized Swedish/english plural "babisar." These new words
are also productive in compounding contexts. For example, Swedish
televison was described as following the "publicservicemodellen."
Somehow, the Swedish speaker recognizes the English words, even
when embedded in a compound. Convincing a computer to recognize
this is another matter, of cours.
Martin Minow
minow%thundr.dec@decwrl.dec.com
------------------------------
Date: Tue, 22 Sep 87 10:06 EDT
From: sas@bfly-vax.bbn.com
Subject: metaphor blindness
A friend of mine who works in the AI business once told me that a lot
of AI researchers have trouble understanding metaphors. We could only
come up with one example at the time, but I've kept my eyes open. Did
anyone else notice a recent NL-KR contributor who was confused by the
translation of:
The first pancake comes out as a blob.
as:
Practice makes perfect.
to an issue of word meaning or work order ambiguity? You usually
don't get such a clear cut case of metaphor blindness. Is this
problem more common among computer types than others?
Amused?
Seth
------------------------------
Date: Mon, 14 Sep 87 11:57 EDT
From: Stephen Sommerville <steves@cs.qmc.ac.uk>
Subject: Re: Indescribably Delicious
Since Frege and Russell are two of my favourite philosophers, I have to
broadcast, with a reluctance born of a lifelong net silence, as brief a note
as possible to warn of some of the misrepresentations and misinterpretations
in the article posted by Berke ( sci.lang: berke@CS.UCLA.EDU (Peter Berke)
"Indescribably Delicious".
I have no solutions to offer to the purported conundrum of how to
interpret phrases like "indescribably A", since I suspect such phrases only
appear puzzling to one with an overly simple notion of 'meaning' ( -- one
which is not unlike what Ryle used to call the "'Fido' - Fido" theory, i.e.
just as I call my dog Fido by uttering "Fido", so words/expressions ordinarily
have meaning by standing for things. This raises the wholly manufactured
problem of how to explain what things expressions like "but", "three" and
"ineffable mystery of life, the universe and everything" stand for). I do take
exception, however, when confusion is propagated in the name of either of my
two favourites! The following is intended to point up some of the more glaring
misinterpretations in Berke's posting:
>Church has shown the necessity for abstract objects if one adopts
>a Fregean theory of meaning. The abstract objects, whether called
>'concepts' or 'intensionalities', are required by the assumption
>that names name (denote) things.
Fregean 'concepts' are certainly not what Church (in "Logic of Sense and
Denotation" or in "The Need for Abstract Entities in Semantics") meant by
"sense". Frege also has the notion of "Sense (Sinn)" in "Ueber Sinn und
Bedeuting", as contrasted with the term "Begriff (concept)". Fregean concepts
can be referred to (they are the references of 'concept expressions'
(predicates)), whereas an expression's 'sense' can only be referred to by the
idiosyncratic description "The sense of ...." or indirectly within an oratio
obliqua context (and an indirect reference is not a form of naming). To say
that abstract semantic entities (like 'senses') are
> required by the assumption that names name things <
is precisely to repeat the mistake against which Frege was arguing in "Ueber
Sinn und Bedeutung" - one which thereby manufactures the puzzle with which the
article opens over "The Morning Star = The Evening Star".
I'm afraid that equating Fregean concepts with Carnap's "intensions" (or
"intensionalities") won't do either. This is a very well known Carnapian
blunder in interpreting Frege - believing that, when a referring expression
occurs in an indirect or oblique context (e.g. quoted or after a modal
operator) it loses its direct reference and refers instead to an intension
(which Carnap supposed was also the 'meaning' of a concept-expression). Frege
did maintain that any expression in an indirect context had indirect
reference, which was usually the sense of the expression (though not if the
context is doubly indirect as in "John believes that it is possible that
<expression>. . ."). Carnap's blunder was to invent the notion of "intension"
to cover both Fregean "references of expressions in indirect contexts" and
meanings of predicates. The blunder has long been recognised, see M. Dummett,
"Frege's Philosophy of Language" and articles by Furth, Copi, Dummett on
Frege.
>When we use a word, we usually (purport to) denote an object and
>express a concept. There are problems with this. The main one
>is called the 'paradox of the name relation by Church', the
>'antinomy of the name relation' by Carnap. It was discovered
>by Frege when he asked "How can A=B, if true, differ in content
>from A=A?"
Sorry, this is just plain wrong. The antinomy of the name relation as cited by
Carnap in "Meaning and Necessity" (p.133+) concerns the failure of
intersubstitutivity of terms in modal (or any other kind of 'intensional'
(sic!) context). For example, though "Necessarily (9 > 7)" and "The number of
planets = 9" are both true, "Necessarily (the number of planets > 7)" is
false. The example from Frege concerning "A=B" is not an antinomy - but an
illustration of the need to distinguish sense from reference ("Ueber Sinn und
Bedeutung", p. 1). It is confusing to conflate problems of semantically
interpreting identity statements with that of failures of extensionality and
intersubstitutivity in modal contexts! (As an aside, I resist the imputation
of:
>When we use a word, we usually (purport to) denote an object
and express a concept . . . <
This simply restates a version of the "'Fido'-Fido" theory - the temptations
of which can be resisted by repeating two central semantic principles: (Frege:
only in the context of a proposition does a name have meaning (bedeutung), and
Austin: (paraphrased) to infer from the statement that every word has meaning
that there is something that every word means is logically every bit as bad as
inferring from everyone loves someone that there is someone everbody loves!))
>It is commonly thought that Russell's theory of
>descriptions solves this paradox, but it does not. Russell's
>theory of meaning requires intensionalities as does Frege's.
This is misleading. The notion of "intensionality" derives from Carnap's work,
dating at the earliest from the 1930's. Russell wrote "On Denoting" (which
sets out the theory of descriptions) in 1905, whilst Frege recognised the need
for "concepts" as the referents of predicates by about 1890. Neither would
have accepted the imputation they 'require' intensions (as meanings of
descriptions/ predicates). In particular, Russell's theory of descriptions is
precisely an analysis designed to show that descriptive phrases of the form
"the so-and-so . . ." (i) only have meaning in the context of a complete
proposition (not on their own), and (ii) do not have meaning by denoting
anything at all (certainly not intensionalities. . .). Descriptions are
disguised existential assertions and are semantically interpreted in the same
way as existentially quantified propositions. About the only vaguely correct
claim, here, is that Frege was openly committed to the need for abstract
entities in a proper semantics of language. Russell disagreed, but could not
articulate a non-conceptualist semantics until he came to accept
Wittgenstein's Tractarian doctrines of Logical Atomism (by about 1917). (This
is not an endorsement of logical atomism, nor an endorsement of Russell's
version of Wittgenstein's Tractatus - both of which are problematic. It
mislocates Russell's theory of incomplete symbols (of which the theory of
descriptions is a part) to suppose it concerns the "paradox of naming".
Russell's concern in 1905 was to overcome the paradoxes of set theory
(including his own). As is noted in his Autobiography, the theory of
descriptions was his first clue to how to overcome them - not a revamping of
Frege's sense/reference distinction!)
>Russell too
>insisted on the formal expression of logic, and wholly adopted Frege's
>language. Russell linearized it - Frege's was two-dimensional. Russell added
>ramified type theory to avoid certain paradoxes.
This is the worst misinterpretation of all. To claim that the logic of
"Principia Mathematica" is just a linearised version of Frege's
"Grundgesetze", with ramified type theory tacked on as an afterthought,
entirely misrepresents both works.
The logic of "Principia Mathematica"(First Edition, 1910 - 1913) IS ramified
type theory, with a full-blown omega-hierarchy of types and orders for every
proposition and propositional function. The "Grungesetze" has a hierarchical
structure. It is at most finite, and (so Frege suggests) rarely requires
expressions of level greater than level-3. Frege's theory of levels (of
quantified statement) does not resolve the set-theoretic paradoxes (as Frege
admits in his hastily prepared Appendix to the "Grundgesetze", added after the
announcement of Russell's paradox). Ramified Type Theory does resolve the
paradoxes, but at an unacceptable cost to Russell's programme to establish the
identity of logic and mathematics. What makes this such a misinterpretation
is, of course, the author's ingenuous lack of familiarity with the paradoxes
("certain paradoxes"). It was Russell's discovery of the paradox which bears
his name IN the logic of Frege's "Grundgesetze" which prompted his ten-year
effort to solve this (and the other related vicious-circle paradoxes). That
ten-year project is what motivated "Principia Mathematica" (> a mere
linearised version of Frege < ???).
>Frege thought up the idea of a variable as a place-holder, and invented
>quantification to explain our way of speaking, e.g., "Let x be a
>number greater than 5," or "consider all the blue things in your
>office."
Nope! The idea of a variable as a "gap" or "place-holder" is at least as old
as Leibniz - though there is suggestive evidence in the work of the 10th
century Arabian logician al-Farabi (Abu Nasr). What Frege's analysis of
quantification and use of variables adds to the logician's corpus is the
ability to handle inferences involving multiply-quantified and relational
statements, together with the fullest recognition of the semantic principle of
compositionality (that sentences can be decomposed syntactically into function
and argument (predicate and variable) in a way which matches their separable
semantically significant units).
>It enabled Goedel to do his work, and Turing to invent computers. I call
>that useful.
If this is referring to Goedel's use of primitive recursive functions in the
proofs of the Incompleteness Theorems, Frege would certainly have rejected the
overt "Formalism" of Godel's approach (particularly the "arithmetisation of
syntax", since Frege's "Concept-script" was not intended to be 'merely' a
syntax).In any case, Goedel was expressly directing his attention upon the
Consistency Problem as formulated by Hilbert. Frege would not have accepted
that axioms for his logic needed consistency proofs - they were "selbstandig
(self-evident)". If anything, Frege's insistence upon ONE correct logic to
underpin mathematics inhibited the emergence of the kind of many-faceted and
algebraisized logics out of which so much of computer science theory has
grown.
There is considerably more of a very confused/ing nature in the rest of
Berke's article - but my patience is exhausted. Beware the enthusiastic
Hegelian who sees all things as related to everything else (Frege to
Wittgenstein, Russell to Husserl) - they are often simply molding history to
their own purposes!
Steve Sommerville
------------------------------
Date: Sat, 19 Sep 87 16:03 EDT
From: Albert Boulanger <ABOULANGER@G.BBN.COM>
Subject: Generalization & Natural Kinds
[Excerpted from AIList]
To add some beef to much of this natural kinds discussion, I suggest
that those interested in the issue of natural kinds and generalization take
a look at a recent paper by Roger Shepard:
"Toward a Universal Law of Generalization for Psychological Science"
Science, 11 September 1987, 1317-1323
From the abstract:
A psychological space is established for any set of stimuli
by determining metric distances between the stimuli such
that the probability that a response learned to any stimulus
will generalize to any other is an invariant monotonic function
of the distance between them. To a good approximation, this
probability of generalization (i) decays exponentially with this
distance, and (ii) does so in accordance with one of two
metrics, depending on the relation between the dimensions along
which the stimuli vary. These empirical regularities are
mathematically derivable from universal principles of natural
kinds and probabilistic geometry that may, through evolutionary
internalization, tend to govern the behaviors of all sentient
organisms.
Albert Boulanger
BBN Labs
------------------------------
Date: Mon, 21 Sep 87 12:31 EDT
From: cugini@icst-ecf.arpa
Subject: Natural kinds
[Excerpted from AIList]
Gilbert Cockton writes:
> I'd like to continue the sociological perspective on this debate.
> Rule number 1 in sociology is forget about "naturalness" - only
> sociobiologists are really into "nature" now, and look at the foul
> images of man that they've tried to pass off as science (e.g. Dworkin).
This seems a somewhat abrupt dismissal of natural kinds, which has
lately attracted some support by people such as Saul Kripke, who is
neither a computer scientist, dumb, nor politically unreliable
(although he IS a philosopher, and is thereby suspect, no doubt).
The (philosophically) serious question is to what extent our shared
concepts ("dog", "star", "electron", "chair", "penguin", "integer",
"prime number") are merely arbitrary social conventions, and to what
extent they reflect objective reality (the old nominalist-realist
debate). A sharper re-phrasing of the question might be:
To what extent would *any* recognizably rational being share our
conceptual framework, given exposure to the same physical environment?
(Eg, would Martians have a concept of "star"?).
I believe there have been anthropological studies, for instance,
showing that Indian classifications of animals and plants line
up reasonably well with the conventional Western taxonomy.
If there are natural kinds, their relevance to some AI work seems
obvious.
John Cugini <Cugini@icst-ecf.arpa>
------------------------------
Date: Tue, 22 Sep 87 15:34 EDT
From: Gary Cottrell <gary@sdcsvax.ucsd.edu>
Subject: Re: Neural computing / Speech processing
In article <8709220532.AA12712@castor.cs.rochester.edu> you write:
>Date: Thu, 17 Sep 87 13:21 EDT
>From: Andrew Jeavons <andrewj@crta.UUCP>
>Subject: Neural computing / Speech processing
>
>Does anyone have any information on work using systems derived
>from neuronal networks theory being used for speech recognition ?
>Any references to literature or names of companies would be
>appreciated .
>
See ICS Tech Report "The hidden Structure of Speech" by Elman & Zipser.
Available from
The Institute for Cognitive Science C-015
UCSD,
La Jolla, CA. 92093
Also, Lokendra Shastri and a student have a system -
he's at
Computer Science Department
U. of Penn,
Phila, Pa (sorry no zip)
Also, Bob Port in Indiana is working on such a system;
Robert Port
Linguistics Dept.
Indiana University
Bloomington, Ind.
>Thanks
>---
>Andrew Jeavons
You're welcome,
gary cottrell
Computer Science and Engineering C-014
UCSD,
La Jolla, Ca. 92093
gary@sdcsvax.ucsd.edu (ARPA)
{ucbvax,decvax,akgua,dcdwest}!sdcsvax!sdcsvax!gary (USENET)
------------------------------
End of NL-KR Digest
*******************
