********************************************************************
ELECTRONIC NEWSLETTER ON REASONING ABOUT ACTIONS AND CHANGE
Issue 97026 Editor: Erik Sandewall 24.11.1997
Back issues available at http://www.ida.liu.se/ext/etai/actions/njl/
********************************************************************
********* TODAY *********
Do you have a concrete picture of what the ETAI will look like? It's
called Electronic Transactions, but will it exist on paper? Will it
be just a web page with links to a number of articles? What will be
its look and feel? Will it be possible to feel it at all?
To get an answer to those questions, please take a look at the main
ETAI home page,
http://www.ida.liu.se/ext/etai/
and click for "accepted articles". This structure was empty before;
it now contains the first accepted article (the one by Paolo
Liberatore) and also an editorial note explaining the publication,
review and refereeing process by ways of that first example. This
is the beginning of Volume 1, Issue 1-3, and more articles will be
added on successive pages when some more are accepted.
The articles that are posted there are in the definite ETAI style:
your printout of them on your own printer will look like the ETAI
journal appearance, except for adjusting the paper size. There is
even a page for the cover, in color, ready to be printed on your
color printer. (A few minor corrections may be forthcoming in the
text, however). The point is: the look-and-feel is just like for any
other journal. And why should there be any difference?
Next, if you wonder about the reviewing process, take a look at that
editorial note. And if you should still have questions after that -
send me a note.
********* ETAI PUBLICATIONS *********
--- DISCUSSION ABOUT RECEIVED ARTICLES ---
The following debate contributions (questions, answers, or comments)
have been received for articles that have been received by the ETAI and
which are presently subject of discussion. To see the full context,
for example, to see the question that a given answer refers to, or to
see the article itself or its summary, please use the web-page version
of this Newsletter.
========================================================
| AUTHOR: Paolo Liberatore
| TITLE: The Complexity of the Language A
========================================================
--------------------------------------------------------
| FROM: The Author
| TO: Anonymous Referee
--------------------------------------------------------
> Section 6 in the article addresses domain descriptions in which some
> states are not reachable from the initial state. I think the modification
> of the semantics of cal-A as proposed in that section does not allow to
> solve the problem of unreachable states. Consider ...
You are right. In this case, the semantics does not capture the intuitive
meaning of the domain description.
> Generally, my feeling is that in order to solve the problem that your
> example highlights (and I think it is a problem) you cannot do in the
> language of cal_A. What you need is a notion such as ``action A is
> executable if ...''.
The intended semantics for the actions in A is: all actions are executable
in any state. When Psi_D(I,\sigma) is undefined, the choice of the original
semantics of A is that the whole domain description is inconsistent. What I
meant is instead that A is not executable in \sigma. In this sense, in A
*it is possible* to express propositions like "I is executable if", but only
in a non-intuitive way: in the example of the counter I is not executable
in {H,L}. We need an appropriate semantics for such propositions.
(If there is a proposition "I is executable if ..." there is no need to make
\Psi defined on I in states in which I is not executable. As a result, there is
nothing that prevent us from using the indefiniteness of Psi to infer
the non-executability of actions. The drawback is that this definition is less
intuitive.)
Analyzing the problem of unreachable states, my first idea for solving it was:
an interpreted structure (\sigma,\Phi) is a model of D if and only if
- each action is executable in any state reachable from \sigma
- each value proposition is satisfied (as in the old semantics of A)
However, I discarded this idea, and instead I defined the one that is presented
in the paper. Note that this semantics, although correct wrt your example, suffers
from some drawbacks. For example, if for any state there is an action not executable
there, then the domain description is inconsistent. Consider for example the case
I is executable if F
J is executable if \neg F
This domain is inconsistent, because in the initial state in which F is true
the action J is not executable, while in the initial state in which F is false
the action I is not executable. In such cases, it seems possible to infer that,
if the initial state is {F}, then the first action executed is I. For instance,
if there is also a proposition
initially F
the first executed action is I. The semantics of A does not infer any
statement about the execution of actions (in cal-E and similar languages
there are propositions like "I at 0" or "I happens at 0".) A statement like
"F after I" may be interpreted in two ways:
- if F *were* executed the result would be a state that implies F
- F is executable in the initial state, and the resulting state implies F
(The choice of the semantics of the paper is the first one, while the discarded
one uses the second one.) Statements like "I at 0" or "I happens at 0" means
that I is executed in the time point 0 (and thus the action must be
executable in that state).
I discarded the first semantics (preferring the one of the paper) because
the impossibility of executing an action in the initial state (or any other
state) may influence the set of the possible initial states. The semantics
of the paper first determines the set of possible initial states, and only
then determines which states are reachable from them. However, in the example
of the counter, the initial state {H,L} must be rejected *because* the action
I is not executable from there. This is why the semantics of the paper fails.
Perhaps the discarded solution is better then the chosen one. The problem is
left open. A minimal requirement for a semantics that takes into account
the reachability of states is that if D does not have "executable" propositions
(or, if Psi_D is total) then the new semantics must coincide with G&L's semantics.
The semantics of the paper (and the one of the previous paragraph) has this
property (clearly, this is not enough.)
> Also, on page 5 you write:
>
> > ``to prove that D entails F after A_1, ... ; A_m suffices to prove that D
> > \cup {\neg F after A_1, ... ; A_m} is inconsistent.''
>
>
> This reduction seems not to go through for your modified semantics of
> section 6.
Indeed, this property holds only for the classical semantics of A (this
property is used for proving the complexity of the entailment in the classical
semantics of A).
--------------------------------------------------------
| FROM: Thomas Drakengren
--------------------------------------------------------
Paolo,
Here are some questions and suggestions about details in your article:
page 16, line -14: What about the case "A causes G, A causes -G"?
That would be inconsistent. What is the intended meaning in this case?
page 17, line 8: This case is similar to the IJCAI '97 paper by myself and
Marcus Bja"reland, where we require one negative precondition and one
negative postcondition (which is of course equivalent, replacing true
and false). You can probably add some nondeterminism here, retaining
tractability, if you do not allow a precondition, the same way as we're
doing it (you can then have a Horn postcondition).
page 19, line 3: An inconsistent domain description could never entail
the same propositions as a consistent one. What is the intended meaning
in this case?
********************************************************************
This Newsletter is issued whenever there is new news, and is sent
by automatic E-mail and without charge to a list of subscribers.
To obtain or change a subscription, please send mail to the editor,
erisa@ida.liu.se. Contributions are welcomed to the same address.
Instructions for contributors and other additional information is
found at: http://www.ida.liu.se/ext/etai/actions/njl/
********************************************************************