[RIFRAF] ACTION 173: Initial Ontology for Action Language Discriminators

Attached please find the initial pass at an ontology for action languages 
(Action 173). 


Some remarks on the ontology.

1.I built my ontology on top of the ontology that Allen Ginsberg had 
created for Action 173 (ontologizing semantic discriminators). This 
approach has the advantage of dealing with at least some integration 
issues from the start, instead of deferring them to a later date. 

2. As discussed during earlier telecons, I broadened the original mandate 
for this action, which was to create an ontology for discriminators for 
ECA (event-condition-action) rules, which are used mainly to describe 
updates to databases. I looked at the more general problem of 
discriminators for AI action languages.
These more general action languages would seem to be needed to represent 
the use cases in the UCR document. (For example, in the use case Ruleset 
Integration for Medical Decision Support, one reasons about various 
medical events, such as Bob’s Hb1AC levels increasing, the doctor 
prescribing various medications, Bob’s reactions to them, and Bob’s taking 
a medical test.) ECA rules, which are much narrower in scope, can be 
considered a subset of general action rules.

Examples of such general AI action languages include the situation 
calculus, the event calculus, the fluent calculus, temporal action logics, 
and  the action description languages \cal{A}, and \cal{C}

3. These languages share certain features, but differ on other features. A 
list of features of interest follows:
 
a. Division into sets of sentences: domain description, observation 
sentences, queries. (It is almost universally accepted to have at the 
first two classes of sentences in action languages.)
b. Intervals vs. time points. vs. both (E.g., the event calculus has both 
time points and intervals; sitcalc has situations/time points, \cal{A} has 
time points.)
c. Discrete time vs. continuous time (Situation calculus: discrete time; 
fluent calculus,  event calculus: continuous time)
d. Branching time vs. linear time; branching forward only vs. branching 
both forward and backward (Event calculus: linear time; situation 
calculus: forward branching time.)
e. Causation as an explicit relation vs. concept explicit in rule and/or 
material implication. (Explicit in \cal{C}; implicit in temporal action 
logics, EC, SC.)
f. Causal rules; state constraints
g. Concurrency: concurrency disallowed; concurrent processes allowed, but 
can’t have them starting at exactly the same time (asynchronicity).  
(Fluent calculus, event calculus: concurrency allowed; vanilla sitcalc; 
only one action at a time; extended (Reiter-style) situation calculus: 
asynchronicity.)
h. Explicit agent vs. implicit agent
i. Single agent vs. multiple agent
j. Determinism vs. non-determinism
k. Solving the frame problem: monotonic solutions (explanation closure 
axioms; Reiter) vs. nonmonotonic solutions (using circumscription, or 
answer-set semantics, e.g. together with an appropriate formulation of the 
commonsense law of inertia)
l. All actions have preconditions and effects. Can also have failure 
conditions and success conditions. (Success conditions different than 
preconditions.)

4. The different features are sometime superficial, but may reflect 
different deep-seated foundational assumptions. Different sets of 
assumptions underlying these languages could make translation difficult. 
Of importance is the growing set of results on methods of translations 
between various pairs of languages (e.g., between TAL and   sitcalc, 
fluent calc and various formalisms).

5. The exercise of constructing the ontology brought to light some 
interesting questions regarding the categorization of these features. Does 
the distinction between single agents and multiple agents belong to the 
model or the theory? What about the distinction between the concurrency 
and asynchronicity? I’ve done a first effort at addressing these issues, 
but they remain open for discussion.

Best regards,
Leora