- From: Sandro Hawke <sandro@w3.org>
- Date: Fri, 26 Sep 2003 15:58:50 -0400
- To: Charles McCathieNevile <charles@w3.org>, "Shadi Abou-Zahra" <shadi@w3.org>
- Cc: Dan Connolly <connolly@w3.org>, www-qa@w3.org, WAI ER group <w3c-wai-er-ig@w3.org>, Eric Miller <em@w3.org>, eric@w3.org
Charles McCathieNevile said: > Well, you've reinvented the axle it goes on... Ooops. Sorry, I forgot all about EARL while doing this work. Well, mostly it never sunk into my brain that you were positioning EARL for such broad use; I thought it was just about reporting issues with content. Of course my ontology is much simpler and probably somewhat easier to use. I wonder when, if ever, it would pay off to for the OWL folk to use EARL extended as necessary.... I'm very, very happy with the adoption so far; I wish I knew if a more complex vocabulary would have hindered adoption. I guess I can ask the current sources how they feel about such a change. Do you have an equivalent of my report generator which would make nice pages about test results if these folks had used EARL instead? Or which would tell us which PROPOSED tests were passed by two systems, which APPROVED ones were passed by none, etc? > Your RDF models a lot less than EARL [1], and not much more (it does include > a specific property for the test duration, whereas EARL includes a general > comment property...) I don't see documentation for "cannotTell", so maybe it covers this, but one thing we need is to report when the system being tested failed to give any answer. This is different from giving the right answer (Passing) or giving the wrong answer (Failing). I called this "Incomplete" originally, but I'm now changing it to "Undecided". (That term is weak in that it suggests that the TESTER couldn't decide if the tested system passed or failed (which is probably what "cannotTell" means), but it matches many decades of decidability research.) I'm not sure how the term could apply to content or even user agents. Why not just use rdfs:Comment instead of earl:message? > Other simple differences are not as many types of result, Yeah, I figured "notTested" would just be no data, and if the test was notApplicable, then why would it be tested? That is, I don't think those actually are test results. > the fact that there > is no RDF description of things being tested (earl defines webContent and I > think userAgent), That would be in an ontology of Tested-Things (which I think of as "Systems", but you need something broader to cover content), which I don't need yet. Actually, I've worked on it a lot for OWL, but it's totally orthogonal. If you include it in the test ontology, people may well think that's the only kind of thing the test ontology is for (like I did with EARL). > or the way that they were tested (automatically, manually, > or heurisitically - i.e. deriving a conformance result from other conformance > results). I'm not sure what those terms would mean for OWL. BTW, I think you're using the word "heuristic" when you mean (and say in the explanation) "derived", or perhaps "implied", "inferred", or "deduced". For a test to be completed heuristically would, I think, mean that you guessed what the results would be and then verified that that's what they actually were. That's probably not what you meant. > The important difference is that you have no notion of provenance in your > model. Provenance is built into the EARL model as a protection against > conflicting claims (which would cause your stuff to just have a > contradiction) or to enable choosing how to deal with conflicting claims > (trust management). Indeed. Provenance is also orthogonal and does not belong in a test-results ontology, although I was tempted, too. For my application, I don't need provencance; the submitters are all essentially trusted. Meanwhile, if I had an implemented system for handling provenance, I'd want to use it for a lot more than just test results. Shadi Abou-Zahra said: > furthermore, in EARL the assertor that conducts the tests is included so > that a single report of test results could be a collection of tests > conducted by different sources. My approach doesn't prevent merging results -- in fact I dump all the results into one triple store before doing anything else -- it just doesn't yet keep track of what came from where. Yes, if we want to track the source of each fact and allow reporting of results from multiple sources in one file, then we'll need some kind of provenance vocabulary. But, again, that vocabulary should be outside of EARL, since it's needed in so many other cases; it's been the focus of an enormous amount of work and is very hard to get right. (I'm not complaining about EARL going their own way on this, if they needed something simpler and sooner, but I'm very reluctant to use it myself when I'm working so hard on the more general approach.) -- sandro
Received on Friday, 26 September 2003 15:58:35 UTC