[Prev][Next][Index][Thread]

Re: SERIOUS concerns about implementation



At 06:50 AM 02/21/97 -0600, Robert Streich wrote:
>This seems a strange comment coming from you, David. Any time you reduce a
>grammar, you reduce expressibility. Any time you reduce expressibility, you
>lose information.

We should probably define "expressibility" before trying to evaluate these
axioms.

Reducing SGML's grammar to

   document ::= char*

would certainly add to expressibility in one sense, but drastically subtract
from it in another. so when we say 'expressibility' or 'expressiveness' or
'expressive power', we need to know whether we mean:

   * maximizing the number of ways i can say the same thing.

   * minimizing the inconsistencies with which the same thing can be said

   * making the maximum number of strings valid

   * making the minimum number of strings valid

   * making the language easiest to use so as to minimize errors

   * having the most 'custom' settings so everyone can make non-similar-
     appearing syntaxes that all still conform.

   * maximizing the range of structurally distinct data that can be represented.


or something else. the last one is the one i'd prefer; I don't consider
being able to do the same thing (namely, start a P element) in all these
ways  "expressiveness", except in the way clothing designers use the term
(all are valid SGML given the right context, and I haven't even used variant
delimiters):

   <P>
   <P/
   <P<
   <P</
   <>

   Hello, world
   &my.starttag.entity;
   &#RE;&#RS;

Oh, I suppose we should also define "information" before we evaluate whether
it's gained or lost. In Information Theory (see anything by Claude Shannon),
it would be hard to show that any of the above differ in information
content, though they might differ in the entropy with which a certain
quantitiy of information is being transmitted.
S


Follow-Ups: