[Top] [All Lists]

Re: [ontolog-forum] open knowledge

To: ontolog-forum@xxxxxxxxxxxxxxxx
From: "John F. Sowa" <sowa@xxxxxxxxxxx>
Date: Tue, 30 Nov 2010 20:08:27 -0500
Message-id: <4CF5A00B.7030804@xxxxxxxxxxx>
Ed,    (01)

I'm happy that we agree about the first point, and I believe that I
can make a strong case for the second.    (02)

>> the term "much darker and deeper hole" [Wittgenstein's later philosophy]
>> is not a bad way to characterize the complexity of NL semantics.
>> ...
>> I believe that AI would be doing true NL understanding today
>> if Quine had devoted equal time to teaching Peirce and Frege.    (03)

> Now if software could really understand this email, just imagine the
> amount of background knowledge it would take, and the nature of the
> philosophical commitment it would take, for the software not to see
> a contradiction here.    (04)

There is no contradiction between those two statements, and any person
or computer system that tried to understand the connections could go
to online resources for the background knowledge.  We (VivoMind) have
implemented systems that routinely search for background knowledge
from other documents than the one they're reading.  See slides 33
to 40 of the following presentation:    (05)

    http://www.jfsowa.com/talks/pursue.pdf    (06)

Note that the system successfully uses background knowledge from
a textbook in order to find connections and draw inferences from
research reports.  I won't claim that this system today "truly"
understands English.  But I believe that this level of NLP is an
important step along the way, and it could have been implemented
in the 1990s if NL researchers had paid more attention to Peirce
and Wittgenstein than to Frege, Russell, Carnap, and Quine.    (07)

> I simply don't think the choice of academic deity of 1950 has had
> a significant impact on our current technical capabilities (others,
> for example, would see Chomsky as the villain of the piece).    (08)

I consider Chomsky one of the villains (even though he admired
Peirce), but much earlier events had warped Chomsky's teachers.    (09)

1950 was at least 20 or 30 years too late to catch up.  In fact, the
major damage was done in the 1890s and early 1900s when Peirce was
systematically shut out of any kind of academic position -- partly
because of a vendetta by Simon Newcomb and partly because of his
"loose morals".  (He had the audacity to marry his second wife
shortly after a divorce from his first, even though they had been
separated for over 7 years.)    (010)

In addition to Peirce's misfortunes, other historical events disrupted
the development of linguistics, logic, psychology, and philosophy
during the first half of the 20th century:    (011)

  1. Compare the two volume _Principles of Psychology_ written in 1890
     by William James (Peirce's best friend for many years) to the
     _Cognitive Psychology_ by Ulrich Neisser in 1967.  In the topics
     covered and experimental techniques, Neisser's book looks as if
     it had been written about 20 years after James's, not 77 years.    (012)

  2. Leonard Bloomfield wrote the first edition of his book on language
     in 1914.  In it, he included a considerable amount of semantics
     based on Wundt's psychology.  (Saussure's work had not yet been
     published, and Peirce's work was almost unknown in the US.)
     But during the 1920s, he was seduced by the behaviorists into
     deleting the semantics (because it was too "mentalistic").  His
     1933 introduction to linguistics became the primary textbook
     on the subject for the next 20 years.    (013)

  3. One of the great tragedies was the death of Frank Ramsey in 1930
     at the age of 26.  Ramsey was a brilliant logician who had read
     some of Peirce's writings and had recommended them to Wittgenstein.
     In the preface to his _Philosophical Investigations_, LW credited
     Ramsey with helping him to see the weaknesses in his first book.
     LW undoubtedly took Ramsey's advice, because he wrote a letter
     to his sister in which he recommended a collection of Peirce's
     writings published as the book _Chance, Love, and Logic_.  If
     Ramsey had survived, he was sufficiently strong as a logician
     to develop a serious alternative to the Vienna Circlers.    (014)

  4. Many European linguists and psychologists were untouched by the
     behaviorist plague and continued to do good work -- Otto Selz
     and the Gestaltists in Germany, Bartlett in England, Vygotsky
     in Russia, Tesniere in France, and logicians in Poland -- but
     their work was severely disrupted by WW II.    (015)

  5. In the US, C. I. Lewis studied with Peirce's buddies James and
     Royce at Harvard, went to teach at Berkeley, and was invited back
     to Harvard to teach in 1920.  Lewis spent two years reading the
     manuscripts, which Peirce's widow had donated to Harvard in 1914.
     That led him to do his most creative work, including inventing
     the modern systems of modal logic.    (016)

  6. Although Lewis was a good logician, Quine was better, but he had
     a very strong prejudice against the ways of thinking by Peirce,
     Lewis, and even his thesis adviser, Whitehead.  After earning a
     PhD under Whitehead, Quine spent a year in Europe (mostly Vienna)
     where he absorbed the logical positivist doctrine from Carnap.
     He also adopted the view that NL semantics is hopeless, and it
     must be replaced by a purified version of logic
     as the foundation for science.  Quine overshadowed Lewis at Harvard
     and indoctrinated many generations of logicians with a behaviorist
     attitude and a disdain for even trying to understand NL semantics.    (017)

  7. Chomsky learned linguistics from Bloomfield-inspired teachers
     and was also inspired by Carnap's _Logical Syntax of Language_.
     From the beginning (his 1955 thesis), Chomsky assumed that the
     only "scientific" study of language had to begin with syntax,
     At MIT, he took money from DoD projects on machine translation,
     but he never got his hands dirty in doing what he was paid to do.    (018)

We can't create an alternate universe that emphasizes Peirce's ideas
over Frege's, but the Cambridge Language Research Unit (CLRU) was
founded in the 1950s with a strong Wittgensteinian flavor.  Two of
the founders were Margaret Masterman, who had been a former student
of LW, and Michael Halliday, a linguist who had a strong influence
on NLP work in semantics.  For a brief summary of their work, see
the two book reviews I wrote about their writings:    (019)

    Review of _Language, Cohesion and Form_ by Margaret Masterman    (020)

    Review of _Construing Experience through Meaning: A Language-Based
    Approach to Cognition_ by Halliday & Mattheissen    (021)

Many of the leaders in NLP got their start at CLRU, but that was not
sufficient to produce a system that understands language today.  But
just imagine what could have been done if Ramsey and Wittgenstein had
continued their collaboration from 1930 to 1950.  The hardware wouldn't
have been adequate in the '50s, but by 1970 they could have developed
far more sophisticated prototypes than Terry Winograd's SHRDLU or
Bill Wood's Q/A system about moon rocks.    (022)

As another possibility, suppose that Quine and Lewis had collaborated
at Harvard instead of feuding.  (Given Q's personality, that would be
a stretch, but let's suppose a kinder and gentler Quine.)  By the 1950s,
they could have developed modal logic and Peirce's semiotics into a
much more realistic theory of language than Montague Grammar in 1970.
For an outline of what such a theory could have been, see the following
two papers, which summarize the theoretical assumptions at VivoMind:    (023)

    Answers to five questions on epistemic logic    (024)

    The role of logic and ontology in language    (025)

These are just two scenarios that could have led to much more realistic
theories of language that could have produced good prototypes in the
1970s and '80s, practical systems by the 1990s, and systems that could
really understand language today -- perhaps not as well as the best
trained humans, but well enough to read and learn from textbooks.    (026)

John    (027)

Message Archives: http://ontolog.cim3.net/forum/ontolog-forum/  
Config Subscr: http://ontolog.cim3.net/mailman/listinfo/ontolog-forum/  
Unsubscribe: mailto:ontolog-forum-leave@xxxxxxxxxxxxxxxx
Shared Files: http://ontolog.cim3.net/file/
Community Wiki: http://ontolog.cim3.net/wiki/ 
To join: http://ontolog.cim3.net/cgi-bin/wiki.pl?WikiHomePage#nid1J
To Post: mailto:ontolog-forum@xxxxxxxxxxxxxxxx    (028)

<Prev in Thread] Current Thread [Next in Thread>