On Jan 23, 2010, at 9:46 PM, Rob Freeman wrote: (01)
> On Sun, Jan 24, 2010 at 11:01 AM, Christopher Menzel
> <cmenzel@xxxxxxxx> wrote:
>> On Jan 22, 2010, at 7:43 PM, Rob Freeman wrote:
>>>
>>> So what is important is this idea that there are "predictive
>>> limitations of theories."
>>
>> A fact as well known (in its own manifestations) in physics as it
>> is in computer science. A fact
>> obvious to anyone who has taken a basic physics lab, let alone
>> studied a bit of nonlinear
>> dynamics or quantum mechanics. I'm not sure why you seem to be
>> suggesting you're on
>> to something particularly deep here. In computer science there are
>> inherent limits to
>> computation. In physics there are inherent limitations to
>> observation, calculation, and
>> measurement. These rather mundane facts immediately entail that
>> our theories have
>> predictive limitations.
>
> Your affection for the mundane is too strong, Chris. Quantum mechanics
> is not explained by "inherent limitations to observation, calculation,
> and measurement", the explanatory force is entirely in the other
> direction. (02)
I don't read Chris' prose as saying anything about the 'direction',
only that the inherent limitations do exist. Which of course they do.
So your insulting ad-hominem tone is doubly inappropriate. (03)
>
> The observations are there, I agree. And go way beyond not having a
> big enough microscope. Over time more and more people, in a diverse
> range of fields, physics, maths, computer science, weather
> forecasting..., and indeed linguistics, have faced up to the fact that
> their observations appear to be random to some extent. (04)
Well, now, that depends upon exactly what you mean. Any single
observation is of course not random, but it is approximate. The
approximation may be described in terms of probability distributions,
but not always or necessarily (especially in engineering, where limit
ranges are more usual.) A suite of observations may be interpreted as
having random error , but errors need not be purely random; they can
be biassed in various ways. (05)
>
> But I don't think this has yet caused a general crisis in the
> theoretical orthodoxies (other than physics.) (06)
Perhaps not a crisis, but it is a commonplace of virtually all
empirical investigations in science, engineering and commerce. What
kind of crisis would you expect to be seeing? (07)
>
> In particular in computational linguistics, still, nobody questions
> the role of grammar. They certainly don't imagine productions in
> natural language might be examples of "computationally irreducible"
> processes, perfectly tractable, simply admitting of no (grammatical)
> "shortcuts". (08)
Um... this seems to be a complete change of topic, but here goes
anyway. Some people DO question the role of grammar. Many people have
very different notions of what 'grammar' means than others do: there
are many theories of grammar in linguistics. Virtually every linguist
is aware that the empirical boundaries of what is grammatical are and
indeterminate and often hard to pin down, and in any case are
constantly changing. I am not sure what you mean by a "shortcut". (09)
>
> John Sowa's advocacy of Wittgenstein's language games hints at this
> idea of process which does not admit of abstraction, but he doesn't
> see a theoretical link.
>
> Faced with randomness theoreticians commonly conclude that their
> theory is right, but the problem is elsewhere. (010)
Well, no, they typically expect the experimentalists to come back with
confidence measures on their observations, and work with those. In
some case, as in QT, the probabilities are the stuff of the very
theory itself, as Im sure you know. (011)
> So in computational
> linguistics you get people saying things like "syntax is solved" the
> problem is with semantics. (012)
Do you, indeed? I have never seen or heard a computational linguist
(or indeed any other linguist) make that statement. (Unless they were
speaking in a loose, pragmatic way about the practical success of
their parser on some large corpus, and meaning that it worked well
enough for their purposes.) (013)
> Our theories of syntax don't solve the
> problem, but that's because all the hard decisions are based on
> something else called semantics. A case of "redefine and conquer". It
> was this kind of idea I was reminded of when Ali said:
>
> Ali, Jan 22: "While it's (usually) not hard for people to understand
> the semantics in natural language, computers often lack the necessary
> "background" (
>http://en.wikipedia.org/wiki/John_Searle#Intentionality_and_the_Background
> ) to do so."
>
>>> My comment to Ali was that the "predictive limitations of
>>> theories" might explain why our "theories" of
>>> language (grammars) have failed to usefully disambiguate natural
>>> language, and that we might be able
>>> to do better by treating syntactic predictions purely as a
>>> process, distinct from theories about that process.
>>
>> I have no idea what that means.
>
> I know.
>
> It is actually very interesting for me to see the issue through your
> eyes, when you stop simply hrmphing and disputing words. For instance
> you characterized undecidability as "...about the limits of
> computation." This reminds me of a quasi-religious interpretation
> along the lines of Goedel himself, or recently, famously, Roger
> Penrose and his "Emperor's New Mind",
> http://ase.tufts.edu/cogstud/papers/penrose.htm";
>
> Why anyone would characterize undecidability as a "limit of
> computation", I don't know. (014)
That word has been used with that meaning for over half a century. I
am not entirely sure, but I believe the word was actually coined on
order to convey exactly that meaning. (015)
> I guess it means your faith in an
> absolute, decidable, world is so great, that when computation fails to
> deliver, it is computation which has failed, not the perfect,
> decidable world. (016)
No, it means that the word, "undecideable" is typically used to refer
to computational/provability problems, just as Chris was using it. If
you intend it in some other sense, then you really need to explain
that sense very carefully (and hopefully without insulting your reader
while doing so.) (017)
> The world must be decidable, so if computation is
> not, then computation must not be adequate to model the world...
>
> Similarly John has clearly pigeon-holed Wolfram's computational
> irreducibility as computationally "intractable". That is how he
> ignores it. How do you get from "irreducible" to "intractable"? It is
> as if conformity with a theory is the goal, and if a computation
> cannot be reduced to a theory then it is "intractable", no good,
> useless. (018)
"Intractable" means, of irreducibly high complexity, where
"complexity" means that the computational cost (measured in space or
time requirements) of solving the problem increases rapidly as the
size of the problem increases. "Rapidly" typically means, faster than
any polynomial function, eg exponentially. None of this has any
emotional connotation, though one might become emotional while waiting
for a computer to solve an intractable problem. (019)
>
> These are perspectives which trap us.
>
> When you are stuck in such a perspective, of course, it is almost
> impossible to jump out. (020)
One might say the same for illusions, hallucinations and obsessions,
of course. (021)
Pat Hayes (022)
>
> -Rob
>
> _________________________________________________________________
> 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
>
> (023)
------------------------------------------------------------
IHMC (850)434 8903 or (650)494 3973
40 South Alcaniz St. (850)202 4416 office
Pensacola (850)202 4440 fax
FL 32502 (850)291 0667 mobile
phayesAT-SIGNihmc.us http://www.ihmc.us/users/phayes (024)
_________________________________________________________________
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 (025)
|