Re: [PRD] review of the frozen draft of Nov 25

Christian, thanks for the corrections. They look good. There are some
responses below, embedded in the text.


 --michael  


On Thu, 04 Dec 2008 17:49:58 +0100
Christian de Sainte Marie <csma@ilog.fr> wrote:

> >   15. Sec 2.2.2, Operational semantics.
> > 
> >       Here the reader starts to run into notational problems. At least,
> >       I did, especially trying to get through the definitions that use
> >       symbols defined a page or two ago. For instance, L is used for
> >       actions, W is used for ground condition formulas, T for terminal
> >       states, etc. These sometimes occur with subscripts. At some point
> >       it becomes hard to follow. I suggest that better mnemonic symbols
> >       are used. For instance, gActions, FinalStates, gConds, etc.
> 
> Difficult to do at once. We will try to make the notation more palatable, but
> longer names for sets may make the notation heavier... 

Heavier?
I am sure that if I measure the weight of my laptop before and after you make
the change then it will stay the same :-)

> I am thinking about re-organising the whole section to make it an easier
> reasing, too.

Cool.

> Probably not much we can do for WD2, though: is that ok?

Yes, but the notation hurts. It makes the neurons in my brain spin faster...

> >   16. Definition of the PRD transition relation (Sec 2.2.2).
> >           * Here φ is a ground formula, right? Say this in the preamble
> >             of the definition.
> 
> I repeated that alpha is a ground atomic action: is that sufficient, or do
> you think we should specifically repeated that phi is a ground atomic formula?

I think it would be useful to throw phi in there as well.


> >           * In the assert and retract action bullets, why not simply say
> >             that w' = w+φ and w' = w - φ, respectively? This is not only
> >             simpler, but also correct. In contrast, the current
> >             definitions are wrong because of the problem with |= in the
> >             presence of negation, as mentioned previously.
> >           * In item 3, why not simply write w' = w
> >             -{f1,...,fn,m1,...,mn'}? (for the same reason)
> 
> It was indeed like that in an earlier version. I do not quite remember why we
> changed that, but I suspect that it might be because w' = w+phi
> looked to much like implementation for our comfort (at least
> for mine).

It does not look like an implementation to me. The thing is that production
rules are operational, and you are defining operational semantics in terms of
semantic structures. The two definitions are provably equivalent, so why not
use the simpler one?

> The point is that any condition that match phi must be satisfied after
> assert(phi), etc, and satisfaction and matching are defined in terms of
> entailment, so, defining the effect of the actions in terms of entailment
> seemed both safe and the right way to do.

Since they are equivalent, the simpler -- the better.
Also, I think your readership would be very surprised by the definition based
on entailment.

> But simpler is better, and if w' = w+phi defines the effect of asserting phi,
> in terms of what formula are satisfied, without making any assuption wrt how
> the action works, it is fine with me (unless I or somebody else remember why
> we did it the way it is).
> 
> I did not do the correction yet, waiting for your reply to the implicit
> question above...

If they are equivalent, as I believe they are, then the change should be made.
If they are not equivalent, then something strange must be going on, and you
would have to explain the differences.

> >   17. Sec 2.3.1, last sentence: repeats the same sentence in Sec 2.1.1
> >       for the third time.
> 
> Yes, indeed. Don't you like it? Oh, ok, I removed it :-)

I love it, but it is too much of a good thing.

> >   23. /The idea of describing PRS as a labeled ... which execution
> >       results ..."/
> > 
> >       Replace /which/ by /whose/.
> 
> Done. I thought that 'whose' was only for people.

No. See http://dictionary.reference.com/search?q=whose
Eg, a word whose meaning escapes me.

> >   26. In the rule instance definition: r^id .
> > 
> >       What is this thing that is uniquely identifying the rule? Is it
> >       the rule itself? You never talked about these identifiers before.
> >       Furthermore, why not define ground instances of the rules in the
> >       usual way, i.e., as the result of substitution of ground terms for
> >       variables?
> 
> I simplified that part and removed that unique identifier thing.
> 
> I have a more general question, however: supposing we could not have done
> without, who cares about what it is, anyway, if we only use its defining
> characteristic, which is to uniquely identify a specific rule?

I am not sure what you mean above (another example of a too long a sentence :-)
but my point was that most people prefer simpler definitions, and they should
not refer to undefined concepts.

> >   27. Three paragraphs later: FORMULAE. This symbol has never been
> >       defined. Why not just say "ground facts" here? The end of this
> >       sentence is garbled: /... are fired, in some sense to be further
> >       specified, in that state of facts/. Please rephrase.
> 
> FORMULAE, ATOMICS etc.. Remnants of versions past. Removed.
> 
> "firing" is commonly used for executing the action part of a rule instance. I
> rephrased the sentence: "all the instances of the rules in the
> considered ruleset ''RS'', whose action blocks are executed in that
> state of facts."

Firing was not the problem. The problem was that the sentence, which I quoted,
was unclear.  The new version is fine.

> >   29. Definition of INSTANTIATE, end of sentence: / in a sense that is
> >       determined by the specification of the function./
> > 
> >       Cannot understand what is meant here.
> 
> I rewrote it: "[...], where the detailled specification of ''INSTANTIATE''
> will define what it means, precisely, for a rule instance, ri, to
> satisfy a set of facts, w."

Looks better. But please include a pointer to where that detailed spec is,
to reassure the reader.

> I think that we should be done by tomorrow night.

Good. So, I'll read the new frozen draft when it is ready.


 cheers
   --michael

Received on Thursday, 4 December 2008 18:06:41 UTC