Re: Inconsistent models, mapping, interoperability, and the SUO
On Sunday 27 March 2005 00:58, Rich Cooper wrote:
>
> Machine learning of English statements requires both true and false
> examples of each concept.
You are stating misinformation as fact, Rich.
Supervised learning is one approach, unsupervised is another.
Without knowing what kind of information you want I can't make any concrete
suggestions, but for instance a classification of drugs into antisyphilitics
might be learned by clustering all the words which precede "cures syphilis",
and "is a drug", and... well, whatever antisyphilitics do. You get a
distribution and draw the line where you have enough specificity.
There must be other techniques. That is one I'm familiar with.
Machine learning has quite a rich literature. As I say I think it has been
limited historically only by the misconception the structure found should be
coherent and consistent. People have assumed their techniques were flawed,
when in fact what was flawed was their concept of a solution (single, not
multiple.)
By associating meaning with patterns in data I assert machine learning will
always work.
> But using WordNet class structures at least gives structure
> to generalizations and specializations of sentences based
> on replacing one member of a class with some other member
> of the same class. Knowing the class names and the class
> members is very valuable in reducing human effort required for
> this.
Perhaps there hasn't been any point to the discussions over the last few
weeks, after all.
We are going in circles.
WordNet forces one perspective on the data. I'm presenting this as the
historical error which has prevented effective knowledge representation.
With WordNet this misconception is built into the design of the system. So
long as you work with WordNet, I believe, you will not do any better than any
of the other thousands who have tried before. The error is already there.
For all its theoretical flaws, if WordNet at least contained the information
you seek I might understand. Then at least you could build a "quick and
dirty", "low hanging fruit" application using it.
But WordNet doesn't even have the information you seek.
True, WordNet may get you funding. For those who prefer familiar problems to
unfamiliar solutions, WordNet is the leading brand.
But it always pays to remember, when you are dealing with unsolved problems
too much familiarity is dangerous. When it comes to unsolved problems, by
definition, the historical solutions are the only ones we know for sure do
not work.
> > Can you give me some examples of exactly the kind of information
> > you need, in a structure you could use.
>
> I can't get too specific due to proprietary nature of the
> application. I want to be able to separate the tools from
> the application, and getting too specific works against
> that goal.
If telling me about the tools would reveal too much about your application, it
would seem you have already failed in your goal of separation!
-Rob