Folks - During the 12/4/03 teleconf we can give a status report on the exploration that Michael Gruninger and I have been doing around PSL and automata-based behavioral signatures. (Michael will have to miss the meeting, so I will present this.) [Minor issue: can someone scribe while I'm going through this material?] Some specific, high-level questions we are exploring are: 1) can we "embed" or "encode" a composition of e-services, described with behavioral signatures, into the PSL framework? What do the axioms look like, which fragement of PSL is appropriate, etc. Answer: yes - see attached short pdf file, which I will step through briefly 2) Can we use PSL as a framework to understand better the behavioral signatures that composite OWL-S services will have? Short examples Example 1: suppose we have 2 atomic services A - inputs: i1, outputs: o2, o3 B - inputs: i7, i8 outputs: o9 consider the composition C = sequence(A, B) with the additional restriction that sameValue(A.output.02, B.input.i7) If we view C as a "simple" service, then what is the i/o signature possible answer: inputs are i1, i8, and outputs are o3 and o9 should o2 also be considered as an output? note that two possible message patterns for C could occur ?i1 ?i8 !o3 !o9 ?i1 !o3 ?i8 !o9 the second alternative is interesting, if the service invoking C wants to use o3 to determine what to provide as i8. Example 2: consider now the composition C' = sequence(A, loop(B)) how do we specify the inputs for the different iterations of B? If C' is considered as a simple service, then does it have an unbounded number of inputs? We are exploring these issues from 2 angles a) We believe that SWSL needs to incorporate some constructs for specifying how data is passed between services in a composition. How to approach this -- Imperative? Functional? Declarative? A particular issue here concerns 2 different kinds of output, e.g., -- providing the price of something -- producing a purchase order for that thing One is just informational/epistemic, the other has the side-effect of producing a physical artifact (which also ahs an imlication that some actions are to be taken.) b) We plan to create a framework that allows us to i) extend OWL-S compositions with something about the messaging behavior ii) embed the larger thing into PSL (combining Michael's embedding of DAML-S services into PSL and the embedding of behavioral signature perspective into PSL) iii) study properties of such service descriptions, e.g., -- can we characterize "well-formed" compositions -- can we find descriptions of composite services that lie somewhere between the "full description" and the "description as a simple service" extremes -- if we "project" out the DAML-S fluents, can all possible behavioral signatures be acheived? If not, can we characterize the subclass of behavioral signatures that can be acheived? For your reading pleasure, here are some related questions that we have been thinking about and/or hope to think about in this context (in no particular order). Please tell us other suggested questions if you have any. a) Can we define a notion of "composition" of services, that is in terms of PSL trees and axioms, rather than the constructors provided by OWL-S. Do OWL-S compositions correspond to a proper subset of these "abstract" compositions, and if so, can we find a declarative characterization of the OWL-S compositions? b) Are there useful generalizations of the notion of behavioral signature. E.g., to focus on actions rather than messages. E.g., to have a more abstract/under-specified view of behavior b1) Can we characterize the class of PSL activity trees that correspond to the services described using a given kind of behavioral signature. (E.g., the message-based one) b2) It is natural to verify temporal properties (e.g., in LTL) of structures expressed using automata. Assuming that we find some kind of natural correspondence between the behavioral signatures and the OWL-S compositions, perhaps we can verify temporal properties of OWL-S compositions? b3) Logic-based languages permit specifications that are "under constrained". Does this give us some advantage when working with the behavioral signature model (which is typically viewed as completely defined/determined) as encoded into PSL? Thanks Rick H and Michael G