Hi Adam, (01)
I think that your experiences with SUMO and Sigma are particularly
relevant to this discussion. Thanks for taking the time to enlighten! (02)
I agree 100%... a very simple API for passing semantically encoded
messages between OOR components. However, I suspect we'll need a
_small_ set of commands in addition to the ask/tell interface. This
is what I'm proposing by drawing attention to the FIPA message
structure spec... one very small spec within the much larger FIPA
specification set. (03)
This specific API has one method. How this method is processed by an
agent depends on the various parameters contained with the method
call. One of the parameters is the message payload. This would be
the set of sentences to be asked or 'telled' to the receiving agent.
The FIPA architecture supports a multitude of encodings,
representation languages and ontologies describing the message
payload. However, any given agent need only support some subset of
these. Other parameters on the method call describe the encoding,
representation language and ontology of the specific message payload.
>From this the agent can determine whether or not it can process the
message. (04)
The receiving agent considers the performative parameter on the method
call to determine what action it should perform. I don't propose that
we use the performatives defined in the FIPA spec as they are tailored
to agent interactions and probably not well suited to the OOR. As a
start we could define performatives for ask, tell and untell and add
others if/when we need to. IMHO whether or not we define these as
parameters on one method, or break them out as distinct methods of the
API is irrelevant. The so called "surface area" of the API is the
same either way. (05)
The FIPA message structure also defines parameters related to the
conversational aspects of agent interaction. I don't think that these
apply within the context of the OOR's Service Oriented Architecture,
so there would be no analogy within the protocol we define. (06)
This proposal is based on the assumption that we can't be so
prescriptive as to mandate one language as de facto within the OOR.
However, doing so does provide simplifications and this is the
approach that I'm taking with the Samian Platform which uses the CL
abstract syntax as an internal representation. If this assumption is
incorrect, we could do away with some of what I'm proposing. (07)
Thoughts? (08)
Cameron. (09)
Sent from my iPhone (010)
On Sep 18, 2010, at 10:01 AM, Adam Pease <adampease@xxxxxxxxxxxxx> wrote: (011)
> Hi Cameron,
> Please forgive the barrage of responses from me of late.
>
> I want to caution against protocols of this sort. As I see it, there
> are really just two (maybe three) operations for a knowledge base (and
> maybe some optional parameters) - "ask" and "tell" (maybe "untell" if
> you don't want to implement a direct contradiction as retraction). Just
> about anything else starts to move portions of the language into the
> interface.
> Another view might be is that if we really want to get an implemented
> OOR, start simple. Allow just CLIF and those three operations. Once
> that actually works, create additional features.
> Yet another caution - even if we choose a powerful language like CLIF
> as the core, if we want to allow translations to and from any other
> language, even one with closely equivalent expressiveness, it's
> non-trivial. SUO-KIF and TPTP are very close in expressiveness.
> Creating an initial version of a translator was not too hard, given that
> we'd already spent several years figuring out how to convert SUO-KIF to
> syntactically acceptable strict first order. But in practice we found a
> number of additional constraints that all theorem provers that
> implemented TPTP had assumed. It took two iterations of the yearly CASC
> competition to get it right (assuming there isn't something else yet
> undiscovered).
> Since, as far as I know, no prover natively and completely implements
> CLIF at the moment, that would also leave us with a moving target.
> There have been so many big (and even well-funded) efforts to create
> elaborate languages and protocols that never were completely
> implemented.
>
> Adam
>
> On Sat, 2010-09-18 at 09:09 -0400, Cameron Ross wrote:
>> Hi,
>>
>>
>> During yesterday's OOR meeting, Immanuel proposed that we design the
>> OOR as a collection of disparate web services. There was some
>> agreement with this proposal. We also discussed some of the
>> challenges with supporting multiple representation languages.
>>
>>
>> Perhaps we could adapt some of the ideas used within the FIPA
>> architecture to help achieve these objectives. Specifically, the FIPA
>> ACL Message Structure Specification may have some value
>> (http://www.fipa.org/specs/fipa00061/SC00061G.html). Basically, FIPA
>> ACL messages carry with them meta-data describing the language,
>> encoding and ontology associated with the payload. I'm not sure that
>> the performatives or the control of conversation aspects make sense
>> for OOR, but the meta-data describing the pay load might.
>>
>>
>> Just a thought.
>>
>>
>> Cameron.
>> _________________________________________________________________
>> Message Archives: http://ontolog.cim3.net/forum/oor-forum/
>> Subscribe: mailto:oor-forum-join@xxxxxxxxxxxxxxxx
>> Config/Unsubscribe: http://ontolog.cim3.net/mailman/listinfo/oor-forum/
>> Shared Files: http://ontolog.cim3.net/file/work/OOR/
>> Wiki: http://ontolog.cim3.net/cgi-bin/wiki.pl?OpenOntologyRepository
>
>
>
> _________________________________________________________________
> Message Archives: http://ontolog.cim3.net/forum/oor-forum/
> Subscribe: mailto:oor-forum-join@xxxxxxxxxxxxxxxx
> Config/Unsubscribe: http://ontolog.cim3.net/mailman/listinfo/oor-forum/
> Shared Files: http://ontolog.cim3.net/file/work/OOR/
> Wiki: http://ontolog.cim3.net/cgi-bin/wiki.pl?OpenOntologyRepository (012)
_________________________________________________________________
Message Archives: http://ontolog.cim3.net/forum/oor-forum/
Subscribe: mailto:oor-forum-join@xxxxxxxxxxxxxxxx
Config/Unsubscribe: http://ontolog.cim3.net/mailman/listinfo/oor-forum/
Shared Files: http://ontolog.cim3.net/file/work/OOR/
Wiki: http://ontolog.cim3.net/cgi-bin/wiki.pl?OpenOntologyRepository (013)
|