Semantic-based linking

Laurent Martelli
05 Jan 1999 09:08:37 +0100

>>>>> "Tril" == Tril  <> writes:

    Tril> On 3 Jan 1999, Laurent Martelli wrote:

    >> >>>>> "Tril" == Tril <> writes: I fond this notion
    >> semantic-based linking very interesting. I've been thinking
    >> about the problem of equality of information that arises when
    >> two systems want to communicate. I think that the problem of
    >> equality of 'procedure' that you're talking about is just a
    >> special of the one of equality of information, because
    >> procedures and code are just data used by a machine that
    >> interprets them.

    Tril> I don't know what you mean "equality" and I'm not sure what
    Tril> I said that made you bring it up.

Well, you didn't mention communcation between instances of the system,
but I was thinking about, and I found a relation ship with what you
said. If on system A I create a function f1 which is linked with is
semantically linked with f2, and I want to give it to system B, I
don't know if it has f2. f2 is an object, and I must know if B
contains an object which is "equal" to the object that I call f2 in
A. Equal here, has the meaning of "semantically equal". 

    >> But let's examine the semantic-based linking a litlle bit
    >> closer. The specification of the function is itself information
    >> that has to be transmitted between two instances of the
    >> system. If we apply the same principle for the transmition of
    >> the specifications, we should transmit an abstract
    >> representation of the specification, which itself is some
    >> pieces of information. Obviously, we need to stop the recursion
    >> sometime before the sun dies.

    Tril> The principle in tunes is to have standards, but they are
    Tril> formed by agreement of all parties (communities of users),
    Tril> not by us (system designers).

I am not sure that we have a community of users right now ... 

    >> And I don't how to do this whithout having some guid
    >> somewhere. The protocole of communication between instances of
    >> the systems should specifiy some guids for some objects. And
    >> the objects that have a guid form the minimal system. And each
    >> instance of the system should have at least those objects so
    >> that they can communicate with each other.

    Tril> Global unique IDs. I had to figure that one out.  I don't
    Tril> know if they will be required or not. If they are, I hope
    Tril> they will remain hidden as much as possible.  And that they
    Tril> can be changed without too much trouble.

Yes, but I think that Guids are important anyway because users want
them. It can have any kind of appearance, but we need them I
think. Otherwise we cannot visualize information on a limited
"display". If I want to visualize an object which is linked to others,
I'll need to visualize these, which in turn are linked to other
objects, and we end up visualizing almost the whole system. So it
seems a good idea to refer to linked objects through an identifier
which can have a "small" representation.

There we find again the problem of visualizing information
independantly of its semantics. 

    >> Well, this is my point of view, but if you have a better
    >> solution, I study it with great care.
    >> Otherwise, we need to specify of communication protocol so that
    >> any instances of the system care share information and
    >> communicate.

    Tril> It doesn't seem too important right now, but when the time
    Tril> comes in my opinion protocols should be invented on the fly
    Tril> to contain just the information necessary.  Maybe a standard
    Tril> meta-protocol to decide on special purpose protocols would
    Tril> work.

Yes, but even a standard meta-protocol will need to exchange
information, so the problem remains.