Re: [X-logic-strategy] Strategy update note

[Neil N. <n_n...@pa...>]

Roger,

> > The
> > _meaning_ of an expression may be seen as the conse-
> > quence set in the context/theory of that expression.  This
> > then suggests that there may be some fundamental logic
> > application that could be applied to expressions for the
> > purpose of determining meaning.
>
> You can't really establish the meaning of an expression
> until the semantics is defined of the language to which the
> expression belongs.
> The semantics is essentially a prescription of how to
> determine the meaning of any expression.
> In principle one doesn't need a logic thus far, but
> in practice a logic is very handy in defining semantics,
> and in deriving specific conclusions from such a definition.

The idea that meaning of an expression could be defined as
the expression consequences of the initial expression is an
idea by one of the philosophers of language (whom, I do not
remember).  The idea is simply that we may ask someone what
they mean by an expression and their expression response
would be the meaning of the initial expression.  This does not
address what expressions ultimately mean beyond expressions,
but if one was able to uniquely identify the expression
consequences, then for the purpose of expressions--a
bounding condition not necessarily more restrictive than that
usually considered in this area of logic, language, or
mathematics--`meaning' becomes defined in a way that allows
fairly direct analysis.  We might think of this as _formalist_
meaning. Meaning as the relations strings (expressions) have
in a particular generating/computable sequence.

Logic for, say, _formalist_ meaning would be similar to the
Intuitionist's logic for Constructive mathematics.  An implication
is a construction/computation and arguments of implication
states or arguments of the computable function.  There is a
sense in which the Intuitionist's formalism is an abstraction of
arbitrary computation.

> There are indeed interesting connection with your audio work
>
> I previously mentioned that I felt that natural languages
> are best dealt with, so far as logic is concerned, by being
> first translated into a more formal language (even if only behind
> the scenes).
> The audio think is yet one more layer removed, so one
> might then have three layers as follows:
>
> AUDIO LAYER
> == speech recognition ===
> SYMBOLIC NATURAL LANGUAGE
> == formalisation =======
> FORMAL LANGUAGE
>
> A problem well known to AI workers is that these
> translations require a lot of knowledge, so you have to
> go top to bottom and more to disambiguate all sounds.

Yes, the intended audio utility should meet any audio
layer requirement for speech recognition.  At the moment
I estimate the time ratio on my still fairly fast machine will
be in excess of 100 to 1--100 seconds to process 1
second of monophonic sound.  But on the positive side,
the result will identify a worst-case complete solution.
I.e., it will only get better with more analysis and faster
equipment.  But even a high ratio initially can be used
in many applications in audio recording where lead times
for sound in, say, popular music and film may be from
weeks to a year or more.  A Beowulf cluster of about a
dozen of next year's PCs (P4), which would not be out-
of-hand for professional recording studio equipment and
considering that just about all other audio processing
equipment could be replaced, would quickly minimize the
time ratio problem.

I expect that the audio layer for current speech recognition
software is optimized for particular audio functions and
such that current machine capability restrictions trade off
the precision of the audio layer for precision of output.
When the complete solution is obtained it could be relaxed/
optimized for less precision toward a sufficient commercial
speed for speech recognition.

But going from natural language to symbolic natural language
and then to a formal language will have a strong demand
without speech recognition and is an interesting area.

> I am likely to progress more slowly on X-Logic in
> the near future since I have a new contract starting
> up which will take priority.

This is very good to hear.

I have beat the frequency crosstalk problem and am
now working on combining these finely divided vertical
frequency spectrums (spectrum slices) into individual,
horizontal frequency and amplitude functions.  Each
frequency's duration definition in a song (roughly a note)
might be seen as a different colored stand of spaghetti,
and at the moment all those different colored strands
have been finely diced and I need to collect those small
pieces of each strand back together into a whole strand.

Regards,

Neil Nelson