Markov.py

Alex Martelli aleaxit at yahoo.com
Wed Jan 24 05:40:27 EST 2001


"Carel Fellinger" <cfelling at iae.nl> wrote in message
news:94l59e$4kc$1 at animus.fel.iae.nl...
    [snip]
> > I have a tender spot in my heart for HMM's ever since using
> > them in the early '80s for speech recognition and natural
> > language modeling... they give you all that 'neural networks'
> > are supposed to, with a very solid and mature mathematical basis
> > ["everything is an application of Bayes' Theorem"!-)] and, as
> > a result, lots of confidence in the results you're getting!
>
> Funny, in those days I was more into Montague Grammers and Natural
> Language Machine Translation (mainly on the implementation side of a
> very formal approach under the guidance of Jan Landsbergen).  We used
> to look with dedain to people using stochastic models in the field:)

Yep, I remember the debates!-)  The little fact that our
approach *WORKED* (allowing a 20,000-word vocabulary real
time recognizer to be first built in 1984), of course,
only made the barbs & disdains from the Theoreticians
more bitter and envenomed:-).  [That's also when I started
lining up theoretical support from NON-computational kinds
of linguists & scholars, such as Wittgenstein and Tullio
de Mauro, for the general thesis that "deterministic"
modeling of natural language is an oxymoron:-)].


Alex






More information about the Python-list mailing list