RIF Telecon 8 May 2007

Admin

F2F6

Chris: anyone who hasn't filled out the survey for the F2F, please do so

<PaulVincent> Unless dinner is ++ better than the meeting...

Chris: 22 people who will attend the meeting, and 28 who will attend the dinner...

Sandro: make your hotel reservations soon if you haven't already

Liason

<PaulVincent> NO update from PRR

Chris: POWDER is another SemWeb WG working on content rating...

<ChrisW> POWDER working group

<sandro> http://www.w3.org/2007/powder/blog

<AxelPolleres> interesting, that fits with one of our use cases.

<sandro> http://www.w3.org/2007/powder/

Chris: they've identified some rules use cases and want some feedback from us
... I'll forward to our list; feel free to reply

Axel: I sent comments on SPARQL that could perhaps been endorsed by RIF. For now I just sent them as myself.

Chris: Alas, we didn't have time to make that formally from RIF.
... I think it's fine.

Core: Issue 31

<ChrisW> http://www.w3.org/2005/rules/wg/wiki/Issue-31

Chris: main discussion in the group is between ONDS and OS choices

Sandro:: translation to DS is hard, from DS is easy.

Chris: Straw Poll. Who likes OS?

<AxelPolleres> Can MK comment on that?

Chris: translating from a higher number choice back to lower number (esp 3 to 1) is hard

Hassan: why is it difficult to translate from higher number to lower?

<AxelPolleres> That was a question.

<DaveReynolds> Actually #2 (ONDS) to #3 (OS) is not quite trivial if you have equality

<AxelPolleres> what hassan mentions is 2 -> 1, that is easy.

<DaveReynolds> agreed 1 <-> 2 is trivial, it is 3 (OS) which makes it hard, both ways

Chris: there are some entailments that make it extremely difficult from 3 down to 1 (or even 2)

Hassan: if you interpret it then it has a unique meaning; you can spew it out unambiguously

MichaelK: hassan is right, translations in both directions are possible

<sandro> p(a), q(b), p=q, in OS entails p(b) and q(a), right.

<sandro> ChrisW, I'd be interested to hear the results of the straw poll, but I'd like to know if folks can vote for more than one option.

<Hassan> OSR is even better IMHO

Chris: michael, from OS into DS is easy?

MichaelK: it was proven in the HiLog paper ...
... depends on what you mean by easy; can use a predicate etc.

Chris: a predicate for each arity ... a combinatoric explosion of rules.

Hassan: where's the combinatorial explosion come from?

MichaelK: grown in the size of each rule, but the same number of rules, same number of symbols. I think the grown is actually linear.

Sandro: if no one is going to argue other side, no need for argument

Chris: yes, vote for more than one if you're happy with that
... or we'll just ask who is against
... who's opposed to choice 3 (OS)?

<sandro> (say "-1 OS" if you are opposed.)

<sandro> (no one opposed)

Chris: anyone opposed to 2 (ONDS)?

<sandro> -1 ONDS

<MichaelKifer> -1 onds

<DaveReynolds> preference against ONDS but not rule it out

Hassan: I'm opposed to it

<sandro> -1 DS

Chris: who's opposed to DS?

<AxelPolleres> -1 DS

<PaulaP> -1

<Hassan> hak -1

<MichaelKifer> -1 ds

Allen -1 DS

Chris: so no one's opposed to 3; sounds like consensus

Sandro: hassan seems to prefer 4, but we don't need to choose now: can just add reflection later

Hassan: better to have reflection from the start, especially if it's not too costly to add

Chris: is there an encoding of a reflective rule (rule that takes the syntax of the language and puts it into the language)?

MichaelK: I don't know; haven't discussed it

Sandro: by picking 3, we're not ruling out 4

MichaelK: we need to pick up one and define semantics accordingly

Sandro: have you thought about semantics of 3 vs 4?

MichaelK: I know how to do semantics for 3 etc. but there are options for 4 (even syntactic)... we cannot just leave it open

Chris: I agree with that; we should agree on 3 or 4

Hassan: Reflection is the possibility to describe your abstract syntax in your semantics.

MichaelK: How much introspection do you want to allow?

<Harold> Pat Hayes recently put forward the very

<Harold> general IKL system

<Harold> which may be of interest here

<Harold> regarding syntactic self-description:

<Harold> http://lists.w3.org/Archives/Public/public-rif-wg/2006Oct/0109.html

Sandro: let's resolve on 3 or 4 for now and discuss it more next week

Chris: sure

MichaelK: introspection for encoding rules as data isn't a problem, but if we get close to Liar Paradox, etc, then it's another matter.
... I'm concerned about what happens when we go into a first order dialect

<Harold> How much would we go in the direction of KIF and CG if we introduce quotation mechanism?

MichaelK: a first order extension of this could lead to paradoxes, e.g. whatever I'm saying is a lie

Hassan: a very simple reflection mechanism can't harm you

MichaelK: I don't know what the consequences are going to be if we allow reflection in the core; in a dialect only, it's not a problem

Hassan: if you're designing a language, having the possibility to describe your own syntax is an advantage
... I don't see any dire consequences

MichaelK: in FOL, if you allow self-reference, you have to be very careful

<Harold> Could we have a 1st-order-capable Core and a (Reflective) Extension Layer above it, of possible use for some Dialects.

Chris: everyone seems to be talking about a different aspect of reflection
... hassan, could you write up a proposal about reflection you have in mind?

Hassan: sure, but I'm just using general programming language notion
... something like Prolog's "univ" ( written "=.." ).

Chris: My concern is whether there is a way to support that in things like FOL.

Hassan: a language that does not support reflection doesn't have to worry about it

Chris: We're trying to make the the Core be something that can be translated to every other language.

Sandro: concrete example: we have a ruleset and we're translating it to FOL ... if the rule is trying to infer new data, does it affect rules themselves?

Chris: anyone else in favor of adding reflection to Core?

(no one)

Chris: hassan, please send a message to the list and we'll discuss next week
... if no other support, we'll have to move to a dialect

Sandro: let's resolve that we agreed on 3 or 4, not 1 or 2

Hassan: I won't object to 3; I won't be able to attend next week

<PaulVincent> I'd be interested in reading Hassan's arguments for 4...

MichaelK: I won't be on the telecon next week either

<sandro> PROPOSED: To go with 3 or 4, and next week we'll decide between 3 and 4, if Hassan's e-mail convinced anyone.

<sandro> PROPOSED: To go with 3 (OS) or 4 (OSR), and next week we'll decide between 3 and 4, if Hassan's e-mail convinced anyone.

Chris: any objections?

<sandro> RESOLVED: To go with 3 (OS) or 4 (OSR), and next week we'll decide between 3 and 4, if Hassan's e-mail convinced anyone. (cf http://www.w3.org/2005/rules/wg/wiki/Issue-31)

Core: Metadata

Chris: I'm not really prepared on this issue; christian not here

Sandro: I feel somewhat prepared
... DaveR may be more than me if he'd rather

DaveR: How do we represent Metadata? What core vocab should we have? And How extensible should it be?
... I think it should be, in the style of RDF, completely open ended.

Sandro: what if you use RDF in RIF Core format?

<Hassan> BTW: having meta-data is having some relection

Chris: we have to be careful about putting this in core because of problems in translating rules into languages that don't support it
... what if you have a rule like "All the rules authored by Chris are false" ?

Sandro: How about having a file have a Ruleset and a Metadata set, and you can merge them at your own risk.

Chris: one obvious way to express RDF as metadata would be to use triples ...
... what metadata do we need?

<sandro> 4.1.5. Embedded metadata

<sandro> RIF must support metadata such as author and rule name.

<Harold> A 'rulename' can be used to *attach* metadata to rules such as 'author'.

<PaulVincent> Rule metadata - see Dublin Core; properties like priority etc

Chris: just wondering about kinds of metadata that might impact the core.....

DaveR: date, provenance,

Chris: is Dublin Core enough?

DaveR: for round-tripping, we might need "the original form of name"

Paul: usually in rule-matching systems, data defined for customer

<DaveReynolds> +1 to Chris, translator hint annotations seem valuable

Paul: obvious one is original language in which rules are defined, also source rules as comments

Chris: we have this Dublin Core metadata, round-trip "preservation metadata", any other kinds?

Allen: dialect identification metadata?

Sandro: I don't think that's metadata

Harold: instead of giving dialect as metadata, could have different URLs pointing to different XML Schemas of RIF

<DaveReynolds> Harold - surely a URL refernce is just a form of metadata?

<PaulVincent> Reference: http://dublincore.org/

Sandro: I agree, could have different namespace for each dialect

Harold: XSD is quite weak; may involve other things like Schematron or even semantic validators (for results of static analysis)

<Harold> Dave, I meant the URL that points to the definition of a dialect such as to its xml schema.

Chris: does each dialect define set of metadata fields?

<Harold> Chris, about the direction of pointing between a RIF file and its metadata, I think essential metadata (about results of static analysis) cannot point from outside into their RIF file -- they must be attached to the ruleset iteself.

UCR

Chris: where do we stand?

Allen: need to add text for figures, otherwise nothing else

Chris: should we have a use cases section at F2F?

Allen: phase 2 requirements...

Chris: I think Paula started a list of phase 2 requirements

RIFRAF

Allen: I did the ontology thing using imports and sent it to the list a while ago... just wanted to make sure it's okay

<AxelPolleres> Suggestion: Can we postpone this discussion until in two weeks (will be)

<AxelPolleres> travelling next telecon!

Chris: sure

Chris: so schedule F2F time to talk about RIFRAF

<AxelPolleres> yes!

<AxelPolleres> sorry, cannot unmute, my skype is somewhat stuck

<Hassan> +1

<AxelPolleres> please schedule rifraf on day 1 (june 2)

<AxelPolleres> thanks!

- DRAFT -

RIF Telecon

8 May 2007

Attendees

Contents