Re: THEORY: Parsing for meaning.

On Sun, 25 Jun 2006 23:11:08 -0400, Gary Shannon <fiziwig@...> wrote:

> For example, no amount of purely linguistic knowledge
> will help you to parse "Join me in a song." Is "song"
> an enterable object? How do I go "into" a song? The
> fact is, it's an idiomatic phrase that should NOT be
> "parsed" at all, but merely recognized by simply
> looking it up in a list of patterns.
That is exactly it, but the Big Idea is to let the computer determine what
those patterns are, at first randomly (or at least stochastically). Using
methods something like a neural network, with the feedback source being a
human operator, the computer then comes to its own conclusions about how a
given language works, without the operator telling it anything except a
sequence of "marks out of ten" for each machine-created response to a
series of queries. Additional entries in the corpus also provides data on
what is correct and acceptable, and what is not. In my head, operator
interference would be kept to a minimum, and the vast bulk of learning
would come from studying a huge corpus.

Of course, this is nothing but an entertaining diversion unless the
machine is taught two (or more) languages at once. Posit a room full of
machines, all of which have been taught a natlang of their own, and the
same shared (con|nat)lang, and things get a lot more interesting.




Paul