probability and linguistic theory - sci.lang

Viesti keskustelusta probability and linguistic theory

Hubey

Näytä profiili

Lisäasetukset 9 maalis 1992, 08:47

Uutisryhmät: sci.lang

Lähettäjä: hu...@pilot.njin.net (Hubey)

Päivämäärä: 8 Mar 92 21:42:23 GMT

Paikallinen: Su 8 maalis 1992 23:42

Aihe: Re: probability and linguistic theory

  Välitä | Tulosta | Näytä samanaiheinen viestijoukko
  | Näytä alkuperäinen
  | Raportoi tästä viestistä | Etsi tämän kirjoittajan viestit
  

In article <203...@unix.cis.pitt.edu> dt...@unix.cis.pitt.edu (David M Tate) writes:

> Hmmm. I'd think there's a (slippery) giant step between "frequency of
> lexical items" and "stochastic models of syntax". Frequencies are *not*
> probabilities, as anyone who pays car insurance can attest.

I'm not sure I understand this statement. The probability density
function is sometimes called the frequency function. Probability is
a mass phenomenon. It is considered to be physical property like
mass, conductivity etc. i.e. the probability of a number on a die
is either a priori (theoretically) determined to be 1/6 (for a fair
die) or can be experimentally determined from either tossing many many
dice at once or one representative die thrown many times. It is an
attribute a lot like its mass. Car insurance rates can similarly
be thought of as being experimentally determined i.e. the
probability of a young, unmarried male being involved in an accident
is accepted to be higher than other lower risk groups a lot like
an attribute of young males and undoubtedly experimentally determined.

similar statements can be made with regard to word frequencies in
languages.

> As with quantum mechanics, there is a "hidden variables" question here.
> Do observed frequencies arise out of an inherent indeterminacy of the
> underlying (psychological) productive processes, or do they reflect a
> deeper (but deterministic) structure, in the way that tossing a coin is
> subject to deterministic physical laws, but at a level of detail which
> exceeds our information-processing capacity, and therefore appears "random"?

This argument can also be used against correlation-regression
analysis. It's always possible that a given C-R analysis can
be incorrectly interpreted as a cause-effect between indirectly
related variables. You are correct again of course that measurement
difficulties are passed of as randomness. However, there are
stochastic methods that take into account the deterministic components
along with the randomness in physical phenomena.

Random differential-difference equations is a case in point. In this
case, the deterministic component (as manifested in the difference
or differential equation) along with the random component which can be
either additive or multiplicative more correctly describes many
phenomena. There are different methods of solving equations of
this type, such as mean square methods and Fokker-Planck-Kolmogorov
methods. There is no reason why these methods cannot be used
in linguistics.

> Or, to put it another way, are the people who work with stochastic models
> claiming that they can say interesting things about production by looking
> at aggregate frequencies with which people choose this or that syntax, or
> that the choice of syntax in individual utterances is nondeterministic?

If you are criticizing only looking at frequencies, of course it
is the simplest way of applying stochastic methods but that doesn't
mean that more can't be done. Only time will tell.
--

mark

hu...@pilot.njin.net hu...@apollo.montclair.edu
...!rutgers!pilot!hubey hu...@logos.montclair.edu

Välitä