12 March 2001 ERT WG minutes



/* LK reads WL's e-mail Semantic EARL from 12 March */

WL What is meant rather than what it looks like. If someone clarifies their use of class in a style sheet, the kinds of things we argue about in GL deal with providing semantics. We need to give the user enough info so they can format information as they wish. We're looking for adding the generalizations. Moving towards a semantic web requires that what we seek are inherent in it. Saying what you mean rather than what you want done with it are key aspects of both accessibility and semantic web. They both deal with repurposing. I'm looking for an index of repurposeability.

LK Thoughts on what EARL predicates are relative to this? What statements? What would you be pointing to and what would you be saying about it?

SP Might be a good idea to look for presentation elements as well. Look for semantics in style sheets.

DD WAI guidelines are already organized in testable assertions (checkpoints). We're ahead of other W3C specs that use plain English (must/should/may). Do build a test suite you have to extract the testable assertions. We have that.

SP There should be no shoulds in recommendations, since can't test. Why should it be there?

WL If you use "should" then you should have the why or excuses for not with it.

DD That's a different point. I agree with that. I don't see what it has to do with testing?

SP There's nothing to test against.

DD It's in the spec.

SP A "should" means some places where not appropriate to use. Need to clearly document.

WL An authoring tool thing?

DD In other words, don't need rationale for "must" since have to do, but for "should" need since optional. In the EARL output, there should be a field for comment?

LK Therefore, looking for a specific type of comment field. Specify why did not implement something.

SP Right.

DD Looking for specifics like we geneated at the meeting.

WC Look at the mail archives?

SP Mid-February came up with several examples and vocabulary. Schemaat the bottom of the summary. Modularized.

DD This is difficult to understand. What is the subject? object?

SP It's basically a core vocabulary. This is what is needed. People can define own objects and statements. e.g. for ATAG. It's a bit out of date.

DD Why need person?

SP Person is an object. X is testing by Daniel. Daniel is a person. This needs to be human readable as well as machine readable, otherwise will get fragmentation.

LK Can we look at tstatements from the F2F, Daniel and see how they fit into this?

DD Result file be a triplet. I'm looking a the minutes from the f2f. There was subject of test, result, then more info on test case. I'd like a flat approach rather than object orienteded.

LK I asked Raph and Libby if we say "daniel asserts checkpoint 5 is passed" the usual way to do that in RDF or N3 is to make that a statement about a statement. The simpler way to do it (as in Annotea) you have a statement like "checkpoint x is not passed" then person asserting is another property hanign off that statement. What they would recommend is that statements about statements doens't buy you anything. Tehrefore, that tilts toward flatter model.

SP using anonymous nodes.

LK A property of a node rather than itself being a statement.

SP In 5.2 of the infomesh document we have ATAG level A, where the property is ATAG level A. It is an anonymous node expressing that something conforms to level A.

DD Are we going to try to evolve the notation to map the vocabulary or work on the vocabulary in free-style English then map onto notation?

WC From what I understood from Ralph, we need to figure out what we want to express before notation. Not sure we documented all that we think we want to express.

LK I need concrete triples. Otherwise, not sure we're saying the same thing.

WC Do it iteravly, give an example with notation. If can't say want you want using the notation then the language is missing something.

WL Wants an example of an example, LK?

LK Alt-text. This node is type "accessiblity statemtn" the statement has a bunch of properties, e.g. "does it exist?" and "quality" etc. :x person, :x checkpoint - those specific terms hanging off of the statement. That's the conceptual diagram. When write it in N3, keep saying ":x property value" then we don't have to keep repeating things. Then, get rid of :x by saying "a EARL statement" SP, you're defining an object system on top of N3?

SP So people can make their own anonymous nodes.

LK We have a class system on top of N3, N3 is a short hand for RDF. As much as I like the notation, I worry that it is reinventing the wheel from knowledge engineering. e.g. 15 years ago there were engines to do this stuff. Are these the best tools?

DD If you look at the rDF model, it explains the foundation of the work. It's not new, it's just a simpler subset. That's why we're using it.

WL Somebody who understood SGML made a subset that transformed the planet. This in part due to the simplification of the notation. I can wrap my head around it.

SP RDF is decentralized, since it uses URIs. It's the way the Web is. It's web-izing knowledge. There are myths that need exploding.

LK I'm not arguing to do this in XML. I want to use knowledge representation, there are just zillions of them.

WL Most of them fell on the altar of owning the roots of the ontology tree problem (my opinion). Never be universal if proprietary.

LK Do we want to try expressing the general predicates from the F2F in EARL.

WL Need 5.2, 5.3, etc. to see what needs to change to 5.1.

DD I antitipcated seeing a detailed vocabulary in plain english. to write down the properties and names. then agree on that (as a WG). then we can develop the notation. then decide n3 or rdf, classes? etc. First is to agree on the vocabulary.

LK Would we come up with a vocabularly expressed in triples independent of how we notate the triples?

WL Mostly a list of predicates?

DD Predicates and properties.

Action DD: write more English examples.

/* dd and wendy leave, LK scribes */

LRK: sean example statements?

SP: metadata folks was asking about metadata... lets find them... this one uses reification "this image has no alt text" this image does not conform to this particular wcag checkpoint Comment would be human suggested output. Also quality of alt text. Vocabulary has what we need.

LRK. checkpoint 1.1 is missing alt text

SP: {uri earl:conformsTo uri-of-wcag-checkpoint}a log:falsehood or {uri earl:doesNotConformTo uri-of-wcag-checkpoint}

LK: how do tools now handle "a log:falsehood"?

SP: CWM knows what how process it. Also could handle other way if we generate log:falsehood form doesNotConformTo.

WL: this is lost to most of world.

LK: This statement has just uri's. So how does tool know what to do?

SP: tool would know what the checkpoints mean and know to popup dialog box. And could have mapping of wcag 2 to wcag 1 checkpoint so both popup same results.

WL if you have a transform. E.g. when wendy makes mapping from wcag1 to 2

LK: always one to one to one mapping?

SP: lots of other equivalents. would be good if in one schema it was in another schema.

$Date: 2001/03/12 19:46:57 $ Wendy Chisholm