IRC log of erswad on 2002-06-24

Timestamps are in UTC.

10:34:49 [RRSAgent]
RRSAgent has joined #erswad
10:36:05 [maxf]
maxf has joined #erswad
10:37:13 [chaalsBRS]
== Specification organisation
10:37:26 [Giorgio]
hello. This is Giorgio.
10:37:34 [chaalsBRS]
people at meeting: Nadia Heninger, Ian Hickson, Liddy Nevile, Nick Kew,
10:38:00 [chaalsBRS]
Jim Ley, Wendy Chisholm, Max Froumentin, Nick Gibbins, Dan Brickley,
10:38:09 [chaalsBRS]
Libby Miller, Charles McCathieNevile
10:38:54 [chaalsBRS]
NH It is not always clear which parts of EARL are really important information and which are supplementary.
10:39:06 [chaalsBRS]
WC tried to shuffle this a bit in the new draft.
10:39:37 [chaalsBRS]
ACTION: NH try to re-order the spec to be clearer about the role of different pieces
10:39:49 [chaalsBRS]
==discussed this morning:
10:40:30 [chaalsBRS]
1. Building a corpus of EARL data to run tests against. We need to have different tools producing similar types of results (agreed to use accessibility testing) against similar and different material.
10:41:00 [chaalsBRS]
Plan to use at least Page Valet (automatic) and MUTAT (manual) and AccVerify to generate results
10:54:16 [maxf]
remote people, we're going to lunch and we'll be back in 1.5 hours.
12:17:04 [wendy]
wendy has joined #erswad
12:17:06 [maxf]
maxf has joined #erswad
12:17:22 [nadia]
nadia has joined #erswad
12:17:37 [libby]
libby has joined #erswad
12:19:11 [nmg]
nmg has joined #erswad
12:20:25 [chaalsXXX]
chaalsXXX has joined #erswad
12:21:37 [Giorgio]
can you read me?
12:28:25 [Giorgio]
Giorgio has joined #erswad
12:32:23 [Zakim]
Zakim has joined #erswad
12:41:38 [chaalsBRS]
giorgio, calling you for number...
12:42:29 [wendy]
agenda+ xpointer, location
12:42:55 [chaalsBRS]
+39 0432 57 53 82 /whois sbp
12:42:59 [wendy]
agenda+ cleaning up the schema (daml+oil needed? owl instead of daml+oil)
12:43:08 [wendy]
agenda+ extensions and interoperability
12:43:26 [wendy]
sbp are you here?
12:43:33 [sbp]
12:45:28 [sbp]
howdy ho, folks
12:45:32 [wendy]
do you want us to call you? near a phone?
12:45:33 [chaalsBRS]
sean, do you want to be on the phone?
12:46:03 [chaalsBRS]
Wendy said we should ahve a 90 minute lunch
12:46:18 [wendy]
12:46:19 [wendy]
12:46:20 [wendy]
12:46:21 [wendy]
12:46:36 [sbp]
what's the phone set up there? is anybody else participating by phone?
12:46:42 [chaalsBRS]
12:47:14 [Giorgio]
my phone is:
12:47:15 [sbp]
how is he finding it?
12:47:23 [sbp]
oh, hi there
12:47:30 [chaalsBRS]
we haven't connected to him yet.
12:47:33 [Giorgio]
+39 0432 575382
12:48:53 [Giorgio]
if you can't get a phone connection, we could do it over irc.
12:49:53 [libby]
we have a phone connection but wendy is trying to get a bridge so both you and sbp can call
12:50:42 [libby]
ok, calling giorgio: chaals is minuting to irc for sbp
12:50:53 [sbp]
12:51:36 [libby]
if we get a brifdge we can change that
12:52:12 [sbp]
I think that I prefer IRC anyway :-)
12:52:34 [chaalsBRS]
NK - conversation model.
12:52:59 [chaalsBRS]
3 players, producer (valet), database, consumer (another program)
12:53:15 [chaalsBRS]
consumer needs to know what producer is talking about
12:53:29 [chaalsBRS]
ways to identify what's being talked about:
12:53:35 [chaalsBRS]
Domain - base level
12:53:46 [chaalsBRS]
URI (has no focus)
12:54:01 [chaalsBRS]
Xpointer or HTTP metadata (but both of these have problems)
12:54:14 [chaalsBRS]
xpointer probs: tends to be overcomplicated.
12:54:41 [chaalsBRS]
doesn't apply to HTML - only works for XML
12:54:53 [chaalsBRS]
doesn't cope with content negotiation
12:55:07 [chaalsBRS]
not robust against changes
12:56:11 [chaalsBRS]
CMN if you pretty-format the source line-character references will die, and Xpointers will survive unless they rely on string ranges.
12:56:25 [chaalsBRS]
NK Right, we are looking for something more robust.
12:56:39 [chaalsBRS]
we see this with annotea deling with orphaned annotations.
12:57:58 [chaalsBRS]
DP If I parse an XML document and generate an ID on a node
12:58:11 [chaalsBRS]
...the requirements of Xpath state that it will be the same if it is the same document
12:58:34 [sbp]
who's DP?
12:58:49 [chaalsBRS]
... you could formulate xpath without generating IDs, but all implementations today will give the same value for same node if the document has not changed (based on node position within the tree)
12:58:53 [libby]
dave pawson
12:58:55 [chaalsBRS]
Dave pawson
12:59:43 [chaalsBRS]
JL If you insert an element above the generated ID will change - you can't tell that your pointer is pointing to something different
12:59:53 [chaalsBRS]
DP the pointer will idenitfy that it is new
13:00:04 [chaalsBRS]
NK right, but if the paragraph is still there we don't want to lose it.
13:00:15 [chaalsBRS]
JL there are lots of ways of finding out if the node has changed.
13:01:02 [chaalsBRS]
NK if you have two paragraphs and sdwitch them then the nodes point to something that seems right, but is a different paragraph. We want to be able to identify that.
13:01:17 [chaalsBRS]
JL because that's what people do - change the order of a couple of paragraphs, or bullet points...
13:02:48 [chaalsBRS]
NK So we figured that you could normalise HTML and then make an Xpointer, and we offered the idea to the HTML WG who didn't want it.
13:03:06 [chaalsBRS]
We can normalise, so why not? Which means the problem becomes how normalisation is defined
13:03:21 [chaalsBRS]
(traactor noise drowning us out briefly)
13:03:56 [chaalsBRS]
we could define a program as the normalisation, but we could make a Web service that does the normalisation, and use its URI as the reference.
13:03:58 [sbp]
you can have some kind of super-normalization property, and then define a subProperty of it for each specific normalization algorithm
13:04:29 [chaalsBRS]
NK Sean is coming from a formal point of view
13:04:36 [chaalsBRS]
13:04:53 [chaalsBRS]
NK Sean, are you thinking of coming up with a description in RDF?
13:05:26 [chaalsBRS]
DB Sean says that in RDF you can say that there are normalisations, and be more precise about which kind of normalisation you use.
13:05:30 [sbp]
yeah, I'm just thinking from the model POV. I don't think we can force people to use one set normalization algorithm, can we?
13:05:51 [sbp]
13:06:29 [chaalsBRS]
CMN how do we deal with someone who uses a better algorithm?
13:06:39 [chaalsBRS]
NK we say "you have to provide a service that does it"
13:06:59 [wendy]
cmn practial problem, 2 diff normalizers do it 2 diff ways.
13:07:31 [wendy]
cmn offer diff error mechanisms. could chain them, each link improving it, but changing the normalizations.
13:07:42 [wendy]
nk the html pointer+link to spec of what used to get it.
13:08:10 [chaalsBRS]
13:08:30 [chaalsBRS]
NG we haven't dealt with normalising document -we have been dealing with negotiated documents
13:08:43 [chaalsBRS] we have added statements about what the HTTP information sent includes
13:09:32 [wendy]
ack ChaalsBRS
13:09:42 [chaalsBRS]
People do conneg on all kinds of things server side, so you get different documents back sort of at random.
13:10:02 [chaalsBRS]
NK capturing those is one option
13:10:13 [chaalsBRS]
13:10:55 [ericP]
ericP has joined #erswad
13:11:09 [libby]
wendy says we have a bridge...
13:11:11 [danb_lap]
danb_lap has joined #erswad
13:11:29 [wendy]
bridge: zakim, passcode 92479
13:11:34 [chaalsBRS]
NG We have assumed that poepple will do the conneg based on what the client sends...
13:11:59 [chaalsBRS]
JL there are problems that can complicate life.
13:12:10 [ericP]
wendy, should i join the bridge?
13:12:28 [chaalsBRS]
well, so far we are using IRC.
13:12:42 [ericP]
i understand it's for only 3 people
13:12:43 [ericP]
13:12:58 [wendy]
eric, you are welcome to join. i don't think sbp will join.
13:13:21 [wendy]
hmmm. giorgio might not either.
13:13:21 [ericP]
ok, we can fight over it
13:13:22 [chaalsBRS]
giorgio is going to go just on IRC...
13:13:29 [wendy]
should we call you?
13:13:38 [wendy]
i guess it's hard to hear.
13:13:39 [libby]
bye giorgio
13:13:46 [Giorgio]
I'm here
13:13:57 [chaalsBRS]
13:14:04 [Giorgio]
13:15:05 [chaalsBRS]
NK: We have looked at HTML pointer, and at HTTP headers.
13:15:24 [chaalsBRS]
These are straightforwad things to do.
13:16:09 [chaalsBRS]
NK so if toolA annotates one version of a page, and toolB user has a different set of preferences they are not going to get the same information.
13:16:59 [danb_lap]
FWIW, I've posted my (personal/rough/etc) notes from this morning:
13:17:47 [chaalsBRS]
CMN so you could use a normalisation that was just whatever tool X does to the source, and then a line/character reference
13:17:57 [wendy]
giorgio? you're on the q - question?
13:18:03 [Giorgio]
why don't we frame
13:18:04 [chaalsBRS]
NK right. Not stable over changes to the content...
13:18:12 [chaalsBRS]
ack chaalsBRS
13:18:15 [danb_lap]
13:18:16 [chaalsBRS]
ack giorgio
13:18:21 [Giorgio]
the whole discussion of EARL within the context
13:18:40 [Giorgio]
of a service that allows end users to cpmare different evaluations?
13:18:47 [Giorgio]
I meant compare
13:18:58 [chaalsBRS]
13:19:07 [Giorgio]
so a user would be able to see how a page changes over time
13:19:21 [danb_lap]
zakim, list conferences
13:19:22 [Zakim]
I see WAI_ERTWG(SW f2f)9:00AM
13:19:23 [Giorgio]
or how a page is evaluated by toolA and tool B using the same guideline.
13:19:31 [danb_lap]
zakim, this is WAI_ERTWG
13:19:33 [Zakim]
ok, danb_lap
13:19:39 [danb_lap]
zakim, who is here?
13:19:39 [Giorgio]
13:19:40 [Zakim]
On the phone I see EricP, ??P1
13:19:41 [Zakim]
On IRC I see danb_lap, ericP, Zakim, Giorgio, chaalsBRS, nmg, libby, nadia, maxf, wendy, RRSAgent, sbp
13:19:54 [danb_lap]
zakim, ??P1 is BristolF2F
13:19:55 [Zakim]
+BristolF2F; got it
13:20:48 [danb_lap]
zakim, BristolF2F holds libby, davep, nick_k, chaals, ian, nadia, liddy, nick_g, max, jim, wendy, danbri
13:20:49 [Zakim]
+Libby, Davep, Nick_k, Chaals, Ian, Nadia, Liddy, Nick_g, Max, Jim, Wendy, Danbri; got it
13:21:48 [wendy]
yes, phone is hard to hear. that's why giorgio decided not to use it (is only on irc).
13:22:03 [chaalsBRS]
DP At day 0 you have a node with a given generateID. You record string_length...
13:22:31 [chaalsBRS]
DP at day3 if you don't get the same generateID and don't get the same string_length, you could find out if the string has been moved to somewhere else
13:22:43 [chaalsBRS]
NK that's something I am going to talk about in a minute
13:23:07 [wendy]
cmn we do consider earl in context of comparing results.
13:23:35 [wendy]
nk next part is markup measure.
13:23:41 [chaalsBRS]
NK Markup measures - change detection.
13:24:08 [wendy]
nk http metadata (last modified, content neg, e-tag, md5)
13:24:14 [wendy]
nk then tell if doc is the same.
13:24:14 [ericP]
ooo, this sound like something i want to hear
13:24:16 [wendy]
13:24:24 [wendy]
these are whole document measures.
13:24:44 [wendy]
..why not take the same measures of factored out subdocument.
13:25:04 [wendy]
.. if metadata says this doc has struct of h1->h2->...headings, and what between them (isn't relevant to the assertion)
13:25:12 [wendy]
.. md5 on doc, similar hash on just headers.
13:26:04 [wendy]
.. last autumn, we talked about it and i created a demo. it detected, through a variety of hashes, could tell if changes were significant to whole doc or just headings.
13:26:33 [wendy]
.. demoed w/cnn could see that some hashes changed each time content changed, but hashes on headers (i.e. structure) remained the sam.e
13:26:50 [wendy]
.. since then we've decided to make this useful to more devs and the dom might be better rep to work from.
13:26:56 [wendy]
.. we can take what want and ignore the rest.
13:27:06 [wendy]
.. e.g. pcdata of a node - take a hash - same idea.
13:27:45 [wendy]
.. can tell if banner has changed and thus that accessibilty not effected.
13:28:00 [wendy]
dp trust: if i make an assertion on that doc, and "that" = measures a,b,c,d...someone else says
13:28:06 [wendy]
.. it's not that doc it's that w/some changes.
13:28:17 [wendy]
.. need trust to say "i took out pieces." will need to be very clear.
13:28:30 [wendy]
nk we need to say exactly how the assertion was processed.
13:28:52 [wendy]
.. we measured these bits but not those. come up with an equivalence class. as far as the metadat concerns.
13:29:21 [wendy]
jl most eval wll stand up if text changes. although doesn't work for "cearl and simple writing" b/c if text changes does change that measure.
13:29:56 [wendy]
nk could just take text nodes, if that has changed then accessibility is likely to have changed...or just p and headings...not blockquote or address which are not likely to change.
13:30:11 [wendy]
jl if just add emphasis, clear and simple still stands up (content not likely to change)
13:30:19 [wendy]
dp missed
13:30:25 [wendy]
nk why do you need human trust there?
13:30:31 [wendy]
dp putting a confidence level on something.
13:30:41 [wendy]
.. if adverts changed, it's not the same doc.
13:30:55 [wendy]
jl if trusting doc can trust these elements.
13:31:29 [wendy]
nk the kinds of things we use to measure: when are changes signivicant, when docs are equiv, whether subsection is shared between more than one.
13:31:44 [wendy]
.. e.g. if an element is a toolbar we should be able to find that element in more than one page.
13:32:00 [wendy]
.. we can measure that perhaps by hashing content.
13:32:42 [wendy]
.. is subsection changed, or everything outside subsction changed.
13:33:05 [wendy]
.. when looking for robust pointers, take into account content neg and doc change.
13:33:19 [wendy]
.. doc summary of doc change:
13:33:29 [wendy]
1. decide what is significant and reduce doc to it
13:33:32 [wendy]
2. has what you have
13:33:56 [wendy]
3. save triple: xpointer+hash+spec for how got the hash (could be a web service)
13:34:21 [wendy]
.. once valey generates, fillyjonk can look at.
13:34:36 [wendy]
.. pointer, details of content neg, hash, reduction spec, and can reconstruct the request.
13:34:54 [wendy]
.. retrieve the doc. compute hash on pointer and tell whether pointer is still valid or not.
13:35:09 [wendy]
.. if not valid, look to other elements that the pointer points to to see if have same hash
13:35:48 [wendy]
.. last bit beyond what we have implemented. but can based on what we have done.
13:35:56 [chaalsBRS]
WAC have you imlemented that - how common a requirement is it in practise?
13:36:20 [chaalsBRS]
WAC you are sharing pointers now?
13:36:41 [chaalsBRS]
NK pointer, hash, spec, HTTP details
13:36:56 [chaalsBRS]
... haven't implemented it, did proof of concept for html pointer and hashing
13:37:05 [chaalsBRS]
... with hashing the annozilla guy has done it too.
13:37:17 [chaalsBRS]
WAC didn't know you are hashing elements
13:37:26 [chaalsBRS]
NK we are hashing at whatever point we are interested.
13:37:58 [chaalsBRS]
...a node and children, a node and some children, etc.
13:38:04 [chaalsBRS]
WAC how do you decide what to hash?
13:38:22 [chaalsBRS]
NK test decides what it needs and hashes that.
13:38:35 [chaalsBRS]
... wich is recorded in the data we are then sharing
13:39:05 [chaalsBRS]
WAC where does nortmalisation happen?
13:39:13 [wendy]
dp sounds reasonably complex, is it worth it?
13:39:22 [chaalsBRS]
NK that is at the first place to get a DOm to make pointers into
13:39:33 [wendy]
jl try to evaluate almost anything, you will fail b/c doc changes everytime you evaluate it.
13:39:41 [wendy]
nk worked out through a bit of trial and error.
13:40:18 [chaalsBRS]
NK what I am talking about here is where we are at in terms of reverse engineering our stuff into something formal
13:40:26 [chaalsBRS]
WAC intersted in feedback...
13:40:47 [chaalsBRS]
WAC They didn't want this in HTML?
13:41:03 [chaalsBRS]
NK We proposed a canonicall normalisation, but they didn't want to touch that.
13:41:29 [chaalsBRS]
... this is a new proposal that can allow the use of whatever normalisation someone wants, plus how to repeat that.
13:41:32 [chaalsBRS]
WAC 3 questions.
13:41:43 [chaalsBRS]
1. Reactions from people?
13:41:55 [chaalsBRS]
2. Seems like we need to be proposing this to someone, but who?
13:42:02 [chaalsBRS]
NK In principle anyone talking about markup
13:42:08 [chaalsBRS]
WAC yes, but it needs to be blessed...
13:42:12 [chaalsBRS]
13:42:45 [wendy]
cmn how much does this need to be blessed? not sure. don't think so. it's useful. it's a huge need for EARL.
13:43:04 [wendy]
.. needs to do html, not just xml. not situation for people developing xpath.
13:43:11 [wendy]
nk they might be itnerested in similarity measures.
13:43:16 [wendy]
jl i don't think it's sufficient as it is.
13:43:22 [wendy]
.. doesn't survive doc change.
13:43:30 [wendy]
cmn we can bless it if we find it useful.
13:43:41 [wendy]
. we're not reinventing someone else's wheel.
13:43:49 [wendy]
.. we're extending their wheels.
13:44:09 [chaalsBRS]
WAC where does it go in relation to the EARL spec?
13:44:17 [chaalsBRS]
... we would need some documentation of it.
13:44:19 [chaalsBRS]
13:44:27 [chaalsBRS]
NK I can hack that documentation together
13:44:54 [chaalsBRS]
ACTION: Nick Kew - write up HTML pointer documentation
13:45:24 [wendy]
cmn make sure it works alongside what people are doing. can we do it with 2 or 3 normalizations. can we map between them?
13:45:47 [wendy]
.. if your tool was integrated into a few comemrcial products, you would have to do more than one.
13:45:51 [wendy]
.. they don't normalize the same way.
13:46:25 [wendy]
.. at the moment they are using line/paragraph pointers. can you work that into the system to retrieve?
13:47:25 [wendy]
dp if you view the normalization as the first part of this interface, it should be viable.
13:47:38 [wendy]
cmn can we map between normalizations.
13:47:48 [wendy]
dp i can choose between diff parsers,
13:47:53 [wendy]
nk similar
13:48:19 [wendy]
dp you dno't care b/c input is normalized doc.
13:48:43 [wendy]
nk mozilla and x normalizes diff.
13:48:57 [wendy]
.. just what found when did it.
13:48:59 [Giorgio]
the comparisons service should be able to take as input EARL statements, the evaluated resource and store both of them and compute an internally used normalization.
13:48:59 [wendy]
jl not quite.
13:49:22 [wendy]
ack ericp
13:49:41 [Giorgio]
the service should get from EARL the problem location in the way in which it is computed by the tool.
13:49:58 [wendy]
ep i haven't heard much of what you said, but amaya has the ability to normalize one way
13:50:01 [wendy]
.. tidy does diff.
13:50:12 [wendy]
.. one question we run into is whether something like schema validation
13:50:20 [wendy]
.. was something you want to include in the xpath or not
13:50:40 [wendy]
nk tbody?
13:50:43 [wendy]
ep yes.
13:50:45 [wendy]
nk yes, have to consider.
13:51:03 [wendy]
.. leave that to the normalization, as long as specify can take it either way since can reconstruct.
13:51:26 [wendy]
ep require processing of xpath.
13:51:35 [wendy]
.. missed it (also hard to hear you)
13:52:00 [wendy]
wac acks cmn
13:52:06 [wendy]
wac acks chaalsbrs
13:52:06 [ericP]
a defined mapping to a well-formed XML document is all you need for XPath
13:52:10 [wendy]
wac acks chaalsBRS
13:52:22 [sbp]
13:52:25 [sbp]
I think
13:52:30 [ericP]
the question is then, why would you want to add a dependency on schema validation
13:52:43 [chaalsBRS]
13:52:49 [ericP]
one answer, some H
13:52:51 [chaalsBRS]
response to Giorgio:
13:53:15 [ericP]
XGML-MXML mappers do produce a PSV document
13:53:33 [wendy]
dp wasn't that the comment you (jl) was making? the simple URI is inadequate?
13:53:35 [wendy]
jl nods
13:53:49 [wendy]
dp yes, diff between time 0 and time 1.
13:54:24 [wendy]
nk reads giorgio's comments
13:54:48 [wendy]
cmn how's the tool going to compute what normalization was used
13:54:49 [wendy]
13:55:06 [Giorgio]
the service should use whatever data the tool provides
13:55:19 [chaalsBRS]
giorgio, do you mean that the normalisation used should be stored as part of the data?
13:55:26 [Giorgio]
13:55:29 [wendy]
nk earl:normalization
13:55:39 [Giorgio]
the tool should use ,say line numebers
13:56:03 [wendy]
dp store data "this is what i tested"
13:56:05 [Giorgio]
the service gets this and using some sort of normalization finds where the problem is located
13:56:11 [wendy]
nk it is excessive if doing much of it
13:56:13 [wendy]
dp good alternative
13:56:15 [chaalsBRS]
I mean should the earl include a note about what normalisation was used, or that tools should be able to automatically figure out which normalisation was used?
13:56:15 [wendy]
nk not for me
13:56:26 [wendy]
gl have to eval something. it will change. then do you store the new one?
13:56:35 [wendy]
13:56:44 [wendy]
jl storing it versus storing a measure of it.
13:56:54 [wendy]
dp if changes one iota retest
13:57:02 [wendy]
jl not usable for many of the use cases.
13:57:06 [wendy]
nk often not necessary
13:57:12 [wendy]
jl human eval expensive to redo.
13:57:22 [wendy]
.. don't want to invalidate just b/c one date changes.
13:57:37 [chaalsBRS]
I don't see that tools are going to be able to calcualte automatically the normalisation used by a different tool, unless it is specified and there is potenitally a service to show what that mapping is.
13:58:07 [sbp]
agreed. you have to note the normalization
13:58:24 [sbp]
unless you point to an archive of the normalized result
13:58:28 [wendy]
cmn still trying to understand what giorgio's saying.
13:59:09 [wendy]
.. to reuse that info (the earlier evaluation) and check later. you have to reproduce that normalization
13:59:13 [Giorgio]
My point is that it is far easier to make tool vendor to provide EARL statements with things like line number, offsets, DOM index
13:59:15 [wendy] order to find where that thing was.
13:59:21 [wendy]
.. and know what the original normalization.
13:59:29 [Giorgio]
.. rather than to agree to adopt a specific normalization.
13:59:34 [wendy]
.. then require in earl, carry problem, location of problem, part of location, what normalization is used.
13:59:51 [Giorgio]
.. That said, we need to work on the interface of a service that accepts this data and
14:00:18 [chaalsBRS]
I agree that we shouldn't require a particular normalisation - they can just specify whatever they do already.
14:00:24 [Giorgio]
.. somehow integrates it into the internally used normalized version of the document being tested.
14:00:53 [sbp]
<+q>the thing is, we can't require that people say what normalization was used: we can require that they note that *some* normalization was used, and recommend strongly that the ydocument it,a nd perhaps provide the hash and an archive link to the normalized output or whatever</+q>
14:00:55 [wendy]
nk you can look at line and column, what i'm saying is that we can do more useful things than that.
14:01:24 [chaalsBRS]
14:01:41 [wendy]
jl why can't we require them to provide which normalization was used?
14:01:46 [wendy]
.. even if private, they can say which was used.
14:01:55 [wendy]
nk they would have to share if they want it to be useful to a 3rd party.
14:01:55 [ericP]
i was going to ask that too
14:02:28 [wendy]
cmn rereads end of giorio's last comment
14:02:35 [wendy]
cmn take data from another tool...
14:02:42 [wendy]
nk each dev needs to specify what doing.
14:02:53 [sbp]
in reply to JL: well, what information are we going to say that we should require? a link to the source code of the normalization algorithm? an RDF representation of that algorithm? the name for the algorithm? the human-readable documentation for it?
14:03:06 [ericP]
just an identifier
14:03:23 [wendy]
people agree - just an identifier.
14:03:26 [wendy]
nk it could be any
14:03:26 [chaalsBRS]
agree with ericP
14:03:28 [wendy]
jl hopefully all
14:03:29 [ericP]
i invent an alghorythm and maybe i document it
14:03:34 [sbp]
right, hence that "we can require that they note that *some* normalization was used"
14:03:52 [ericP]
someone happens to re-implement that algorythm so they use the same identifier
14:03:58 [wendy]
jl they need to give a ref to the normalization, so can compare...if two people use same tool we can compare.
14:04:27 [Giorgio]
normalization = html parser?
14:04:41 [danb_lap]
Have you folks looked at the work XML signature etc are doing?:
14:04:45 [chaalsBRS]
a typical example.
14:04:46 [wendy]
nl typically (yes giorgio)
14:04:48 [sbp]
you can't require that nodes be identified by URIS in RDF, though. that's my point. peopel can use bNodes
14:04:57 [chaalsBRS]
Another normalisation is to run through tidy...
14:05:01 [ericP]
if people have to invent an identifier when inventing an algorythm, we get some opportunity for interop
14:05:19 [ericP]
we can get some false negatives, but no false positives
14:05:46 [chaalsBRS]
danbri: Look at ripping off the work of XML signature - they have been down this path. (See URI above)
14:05:50 [ericP]
also, once some become more popular, we have progressively better interop
14:05:53 [wendy]
action wendy take latest back to reagle for discussion
14:05:57 [sbp]
note to self: use this as an argument for deprecating bNodes on rdfig
14:05:59 [chaalsBRS]
ACTION: Wendy - investigate XML signature technique
14:06:49 [ericP]
sbp, what's your take on my argument?
14:07:03 [wendy]
14:07:33 [wendy]
ack ChaalsBRS
14:07:43 [wendy]
cmn sbp says we can't require that nodes have uris
14:07:45 [sbp]
my take is that you can't require people to use a URI as identifier in RDF, although perhaps danbri has some thoughts on that?
14:07:51 [chaalsBRS]
Sean, if we are specifying the RDF to be used for identifying the nodes why can't we require that they be a particular kind of thing (e.g. URI)
14:08:04 [wendy]
danbri - it's your data format, if you want uris there
14:08:09 [wendy]
nk seems an obvious mechanism
14:08:16 [sbp]
how do you say that in the schema?
14:08:18 [chaalsBRS]
14:08:20 [ericP]
sbp, not at a model level, but it can be a requirement at a higher level application level
14:08:25 [sbp]
the symbol is independent of the thing that it denotes
14:08:30 [sbp]
application level: oh. ugh
14:08:47 [chaalsBRS]
chair: Break time.
14:08:57 [danb_lap]
this isn't the sort of thing an rdf schema would express. But its the sort of thing the EARL spec might dictate.
14:09:10 [wendy]
14:09:10 [ericP]
if you don't have any app-level specification, you have no earl spec
14:09:14 [chaalsBRS]
WAC where do we go from here on this topic?
14:09:43 [RRSAgent]
I see 3 action items:
14:09:43 [RRSAgent]
ACTION: NH try to re-order the spec to be clearer about the role of different pieces [1]
14:09:43 [RRSAgent]
recorded in
14:09:43 [RRSAgent]
ACTION: Nick Kew - write up HTML pointer documentation [2]
14:09:43 [RRSAgent]
recorded in
14:09:43 [RRSAgent]
ACTION: Wendy - investigate XML signature technique [3]
14:09:43 [RRSAgent]
recorded in
14:10:49 [chaalsBRS]
20 minute break. Back at 3.30 local time to talk about cleaning up the schema.
14:11:19 [Zakim]
14:12:08 [Zakim]
14:12:09 [Zakim]
WAI_ERTWG(SW f2f)9:00AM has ended
14:21:52 [sbp]
[chuckle]: "(not yet a) W3C Working Draft"
14:23:08 [wendy]
:) you like that eh, sbp?
14:23:19 [wendy]
we'll pull that off once we go to tr.
14:23:34 [sbp]
very much so. TimBL and DanC were moaning about people publishing pseudo-WDs very recently on #rdfig
14:24:14 [sbp]
nadia: actually, TestCase is *not* the range of object; it's restricted to that when used on an earl:Assertion. the distinction between using range/domain and using DAML restrictions is something that shouldn't be missed
14:25:44 [sbp]
missed out in the specification, that is... dunno how best to represent it in that table
14:30:03 [nadia]
sean: have you looked at my reorganization that i just sent?
14:30:21 [nadia]
i'm not sure what you mean by what you said
14:30:28 [sbp]
yep, that's what I was commenting on
14:30:39 [sbp]
well... the range of rdf:subject is not earl:TestCase
14:30:51 [sbp]
otherwise reficiation would be pretty useless to 99% of the world :-)
14:30:57 [sbp]
I refer to the table in section... um...
14:31:06 [sbp]
"4. Properties of Classes"
14:31:07 [libby]
14:31:09 [nadia]
wendy did that
14:31:23 [libby]
do you want to dial in again?
14:31:36 [nadia]
look at sections 2 (missing the number) and 3
14:31:59 [nadia]
i wanted to replace the "properties" and "classes" sections with something that's a bit clearer on the actual structure
14:32:21 [wendy]
oops - yes, i see my error. what i was trying to say is that on an earl:assertion, the object is limited to earl:TestCase.
14:32:52 [wendy]
14:33:04 [nadia]
all i did was organize properties by their domain, and the minor classes by the property they were associated with
14:33:05 [sbp]
right, sorry... I thought that was one of nadia's amendments
14:33:26 [wendy]
in the earl world, the range of object is TestCase, no?
14:34:00 [nadia] for those who can't find the message
14:34:23 [wendy]
just playing around w/representing the data in various forms to make the schema easier to understand.
14:35:37 [wendy]
section: earl as rdf
14:36:09 [wendy]
title of doc: version 1.0 spec
14:37:55 [sbp]
the range of object is not TestCase, and it'd be confusing to say otherwise. I'm not sure how to put it clearly, but object has a DAML restriction on it such that the values that it can take when used on earl:Assertion are limited to instances of earl:TestCase
14:38:07 [wendy]
giorgio, sbp, eric? which of you wants to be on the phone for the next session?
14:38:20 [wendy]
we'll be talking about cleaning up the schema. danbri has posted a proposal.
14:38:47 [nadia]
um, wendy's tables are really useful as a quick reference
14:38:50 [danb_lap]
I have to admit that I _don't_ know the current EARL schema very well. So I resolved to keep my mouth shut rather than bluff.
14:38:57 [sbp]
yep, agreed
14:39:08 [nadia]
perhaps you could say "earl suggests TestCase" in the tables
14:39:22 [sbp]
(um, agreed that they're a very useful reference, that is)
14:40:06 [sbp]
if we went with DanBri's proposed line, this would no longer be a problem
14:40:16 [sbp]
since we wouldn't be using restrictions there at all
14:41:17 [sbp]
documentation for DAML restrictions:
14:41:43 [sbp]
14:41:44 [sbp]
14:41:45 [sbp]
Note: When restricting the range of a property, as we are doing here, there is an important difference between using a toClass restriction and using rdfs:range (as we did for the hasFather property). An rdfs:range element has a global scope: any use of the hasFather property for any class must always yield a male. A toClass restriction (as used in the Person class) on the other hand has a local scope: the parent of a person must be person, but the parent of any oth
14:41:46 [sbp]
14:42:14 [sbp]
hork, how much of that came through?
14:42:24 [danb_lap]
i saw it.
14:42:48 [wendy]
sbp - any reactions to danbri's suggestion?
14:42:52 [wendy]
14:43:10 [sbp]
14:43:11 [danb_lap]
(re daml etc -- Is this w.r.t. context-specific range/domain constraints? I think we can probably worry about that elsewhere.)
14:43:11 [wendy]
sbp - i think the object/testcase range issue is just typographical. error on my part. i don't need convincing.
14:44:03 [sbp]
well, the use of restrictions will be confusing for people that haven't come across them before
14:44:10 [sbp]
so we might want to link off to the above or something
14:44:28 [sbp]
it is minor
14:47:05 [danb_lap]
sean, what do you reckon to separating the DAML+OIL/WebOnt claims out into a separate doc? ie. have a simple vanilla RDF Schema to describe EARL, with richer description(s) separately.
14:48:05 [sbp]
agreed with nmg
14:48:06 [danb_lap]
I find the daml+oil stuff unreadable, but I like idea of trying to use D+O/WebOnt...
14:48:38 [sbp]
especially since the documentation parts are now being auto-extracted to the specification
14:48:44 [danb_lap]
one doc could describe basic classes, properties, and seeAlso to fancier stuff. nmg/sbp, do you have an example of something that wouldn't partition?
14:48:55 [nmg]
weeeell, owl should help on the readability front, when it arrives.
14:49:05 [sbp]
I mean, you don't have to have the D+O stuff in the specification, as long as the schema is normative
14:49:29 [sbp]
I don't expect OWL to be much different in the readability stakes from D+O
14:51:03 [danb_lap]
(discussion of role of DAML+OIL vs WeBont, using external specs in w3 docs etc)
14:51:19 [nmg]
nmg raises question about dependance of w3c specs on specs from external orgs.
14:51:34 [nmg]
like daml+oil - not actually backed by any standards org.
14:51:51 [nmg]
owl as alternative, but not available yet
14:52:13 [sbp]
define a standards org. I've seen recommendations pointing to RFCs; what makes the IETF necessarily more reputable than DARPA?
14:52:37 [sbp]
we could use the copy of the DAML+OIL schema that's in the W3C datespace
14:52:42 [danb_lap]
danbri: Concern about too much reliance on webont, however I'm keen to see this WG sanity check WebOnt/OWL work.
14:52:47 [nmg]
DARPA isn't the standards org - the Joint Committee is the closest to such
14:53:01 [danb_lap]
see also my review of webont requirements doc,
14:53:41 [sbp]
as I say, a W3C note was published detailing the D+O schema. however, using that would mean breaking implementations which recognize the URI-refs
14:54:14 [nmg]
maxf: owl redefines (is inspired by? - nmg) daml+oil - doesn't make normative references to daml+oil
14:54:16 [sbp]
moreover, the Semantic Web wouldn't get very far if people were afriad of using terms outside of the W3C
14:54:21 [danb_lap]
How many EARL implementations use the DAML+OIL aspects of the schema? In practical terms, what will we break?
14:54:49 [sbp]
well, a lot of my early schema munging used terms, but admittedly not allt hat much
14:55:20 [sbp]
but I'm still vehemently against it on grounds that it's rather anti-Semantic-Web-philosophy
14:55:33 [danb_lap]
danbri: I'd rather see this WG use a DTD and XML Schema to describe EARL 'file format', RDFS to describe the modelling constructs, and (separately) use WebOnt/OWL for fancier constraints.
14:56:18 [nmg]
nmg: difference between using terms defined outside the W3C (foaf, rss, etc) and languages defined outside (daml+oil)
14:56:26 [sbp]
you mean use a canonical subset of XML RDF?
14:56:38 [sbp]
how would people add their own extensions, which Nick et al. have already done? via. some m12n mechanism?
14:57:14 [danb_lap]
much as per rss 1.0
14:57:23 [sbp]
um... I don't agree with that at all
14:57:55 [sbp]
DAML+OIL is in every way as stable as those languages in my opinion, and perhaps more so than FOAF. just because it's more complex doesn't mean that it doesn't get maintained
14:58:30 [nmg]
my point is more that it stands to be superceded by OWL
14:58:50 [sbp]
so we have to wait for Godot before we can go to NOTE/Rec?
14:58:51 [nadia]
imho, i don't think the problem is that daml+oil might be unstable, i think just that it's unnecessary and confusing in the schema itself
14:59:14 [danb_lap]
ie you'd have an XML-based 'file format', with particular portions of the XML syntax allowing other namespaces.
14:59:27 [ericP]
nadia, i bet putting the rdfs stuff at the top would help
14:59:32 [sbp]
danbri: then you couldn't have a DTD/XML schema for it
14:59:36 [ericP]
it would give folks a lay of the land
14:59:50 [danb_lap]
sbp, I'm on the DAML+OIL committee. I don't see it being maintained. The DAML+OIL baton was passed to WebOnt WG.
14:59:59 [danb_lap]
sbp: yes you could.
14:59:59 [ericP]
sbp, when i last looked, the daml schema schema didn't use Disjoints for validation
15:00:48 [sbp]
danbri: would you be willing to produce the DTD/XML schema for EARL, then? I'm pretty sure that even in ANY, you have to decalre the elements being used
15:01:09 [danb_lap]
A schematron schema, perhaps...
15:01:10 [sbp]
as I found out when I tried to modularize RDF into HTML
15:01:23 [sbp]
but schematron is not endorsed by the W3C!
15:01:27 [ericP]
it would be feasible to write a DTD/XML schema/relaxNG schema to validate a particular form of an earl document.
15:01:40 [sbp]
15:01:54 [ericP]
it wouldn't be validate all the RDF isomorphisms of an instance of that document
15:02:05 [sbp]
nor extensions
15:02:08 [danb_lap]
that'd be ok. It'd be a syntactic profile of RDF>
15:02:09 [danb_lap]
15:02:18 [danb_lap]
You could do it in prose, even.
15:02:21 [ericP]
it also wouldn't permit extensibility except in ways that the schema designer expected and designed for
15:02:22 [sbp]
it's the extensions that I'm worried about
15:02:34 [libby]
group wonders whether to talk about other aspects of the schema
15:02:51 [ericP]
so then you get into the general problem of expressing constraints for an extensible object
15:02:59 [libby]
nmg brings up an issue about subclassing rdf:property
15:03:14 [ericP]
this is what description logic is good for
15:03:15 [sbp]
eric: well, XHTML m12n is the model to look at
15:03:21 [libby]
...which will confuse fancy inference engines
15:03:34 [sbp]
eric: i.e. the best precedent amongst W3C recommendations
15:04:03 [libby] separate out the ontology from the ontology langauge - not using rdf:propertyu breaks this divide
15:04:16 [ericP]
15:04:17 [sbp]
15:04:20 [libby]
wendy: what is the benfit of creating subclasss of property
15:04:20 [libby]
15:04:21 [sbp]
15:04:21 [ericP]
15:04:26 [ericP]
i'll bet it's that one
15:04:35 [ericP]
maybe multiplication
15:04:37 [libby]
sbp? what d'you reckon?
15:04:52 [libby]
wendy - maybe from a OO point of view
15:04:55 [sbp]
subClassOf Property where?
15:05:02 [nadia]
15:05:02 [sbp]
what are we referring to?
15:05:04 [libby]
nmg - subclassing is deemed ok...
15:05:05 [sbp]
15:05:54 [ericP]
sbp, do you have a specific proposal derivative of XHTML miscalculation or a hunch that that's the way to go?
15:05:56 [sbp]
well the range of rdf:predicate is rdf:Property, so I didn't have much of a choice :-)
15:06:15 [libby]
libby says her stuff can't deal with subproperties
15:06:51 [sbp]
eric: it's mainly a hunch that that would be the direction to take if we started producing XML schemata for EARL, but I'm not sure how much value that approach has anyway
15:07:04 [danb_lap]
danbri: I'm worried that EARL's use of RDF falls between two worlds. It does some stuff that Description Logic (DAML+OIL etc) reasoners will likely barf on. And things like having subclasses of property mean simple triple stores can get confused too.
15:07:17 [libby]
libby: so if we got rid of rdf reification the subclassing propeerrty would go too?
15:07:44 [sbp]
it'd be O.K. to infer out some triples, e.g. stating that earl:fails a rdf:Property
15:08:04 [libby]
nmg offers to go away and write a review of the schema - at the moment the rationale is ratther spare so outsiders find it difficult to underxstand who certain decisions were made
15:08:07 [sbp]
you can run over it or something (as I have done and posted in the past)
15:08:15 [ericP]
sbp, it seems like sort of a compromise between the overconstraining world of syntactic validation and the, um, very accepting world of DL.
15:08:29 [sbp]
eric: such is EARL :-)
15:08:35 [ericP]
s/it/XHTML multiplication/
15:08:37 [danb_lap]
DL world isn't very accepting!
15:08:52 [danb_lap]
BTW, a schematron schema for RSS, aka a syntactic profile of RDF:
15:08:57 [sbp]
it's about balancing constraints on the model with constraints on the syntax
15:09:01 [libby]
lookign at earl 1 schema .rdf
15:09:17 [ericP]
danbri, it's had to validate much with DL, or at least with DAML+OIL
15:09:26 [libby]
[missing some discussion]
15:09:49 [libby]
nmg giving example of possible confusiogn with no rationale
15:09:57 [maxf]
nmg mentioning that stuff appear to be not useful in the schema
15:10:01 [libby]
...useage examples...
15:10:11 [maxf]
like validity property
15:10:12 [libby]
wendy: asking those questions is very useful
15:10:18 [libby]
nmg agrees to do this
15:10:32 [wendy]
action nmg write up list of questions about schema
15:11:06 [libby]
wendy starts to list issues:
15:11:15 [libby]
simpler - getting rid of reification...
15:11:20 [danb_lap]
I'd like to see a plain RDF Schema version of the EARL schema, not least so we can compare it with an OWL/WebOnt-enriched version.
15:11:28 [libby]
...subclasses of rdf:property
15:11:57 [libby]
danbri would like to know what tools in practice expect of an earl document
15:12:05 [chaalsBRS]
DB: What do tools expect of an EARL document?
15:12:19 [libby]
...e.g. you expect some assertions in a pacrticular document
15:12:28 [libby]
...rdfs daml etc don;t do this per doc
15:12:43 [libby]
[this sound like xsd type thing to me...]
15:12:59 [sbp]
I've always thought of an EARL document as being an RDF document that has at least one instance of earl:Evaluation asserted inside it, and then some extra explanatory garbage
15:13:14 [chaalsBRS]
But which is useful for?
15:13:26 [chaalsBRS]
JL It depends on what kind of evaluation you are making and what you are doing?
15:14:11 [libby]
danbri: rdfs schema doesnt tell you what to do when you get a document
15:14:15 [sbp]
well, I think you're always going to need the evaluation part, otherwise it'd just be ARL. but as to how much of the context information etc. that you need, that is up to the application
15:14:37 [libby]
danbri/jim - steriotypical examoples would be/are good
15:14:42 [libby]
wendy: doing this for each usecase
15:14:54 [libby]
jim: an xmlsschema would be good
15:15:01 [sbp]
we need more usecase examples actually worked through
15:15:48 [wendy]
dp wrote an xml dtd. will send to list to see how well it matches w/rdf.
15:16:01 [chaalsBRS]
ACTION: Charles look over the use cases and see what features they need to use.
15:16:03 [wendy]
lm what if it looked more like [shows dp danbri's proposal sent during break]
15:17:00 [wendy]
ACTION: nmg write up list of questions about schema
15:17:51 [libby]
libby: rss excample - both a dtd and an rdf schjema
15:17:59 [libby]
danbri: rss spec has an exampel at the top
15:18:29 [libby]
dp: need something simple, like dp's DTD, then transform to rdfs
15:18:41 [libby]
dp: any decision made about N3/XML?
15:18:43 [sbp]
right, but a lot of problems have been caused by RSS's little syntactic constraints
15:18:57 [libby]
wendy:xml; sbp uses N3 and converts it
15:19:05 [libby]
in particular?
15:19:09 [libby]
15:19:12 [RRSAgent]
I see 5 action items:
15:19:12 [RRSAgent]
ACTION: NH try to re-order the spec to be clearer about the role of different pieces [1]
15:19:12 [RRSAgent]
recorded in
15:19:12 [RRSAgent]
ACTION: Nick Kew - write up HTML pointer documentation [2]
15:19:12 [RRSAgent]
recorded in
15:19:12 [RRSAgent]
ACTION: Wendy - investigate XML signature technique [3]
15:19:12 [RRSAgent]
recorded in
15:19:12 [RRSAgent]
ACTION: Charles look over the use cases and see what features they need to use. [4]
15:19:12 [RRSAgent]
recorded in
15:19:12 [RRSAgent]
ACTION: nmg write up list of questions about schema [5]
15:19:12 [RRSAgent]
recorded in
15:19:14 [libby]
urls at all?
15:19:32 [wendy]
15:21:23 [sbp]
check out the "RSS 1.0 conformance tests" thread
15:21:25 [libby]
grooup wonders whether to ajourn
15:21:45 [sbp]
15:21:46 [sbp]
I've seen the entire range of RSS compliance factors and it's a grim sight.
15:21:46 [sbp]
Very little beyond the core set is implemented. And even within a core set the
15:21:46 [sbp]
compliance has been weak.
15:21:48 [sbp]
]]] - Bill Kearney
15:22:01 [libby]
danbri talks about splitting the earl spec
15:22:17 [libby]
compliance factors?
15:22:30 [chaalsBRS]
rereading Giorgio's mail - <ClaimPackage>
15:22:30 [chaalsBRS]
<name>Bob Bobbington</name>
15:22:39 [libby]
danbri: modularised version of earl? sam ebucket currently
15:22:41 [libby]
nadia agrees
15:22:54 [chaalsXXX]
chaalsXXX has joined #erswad
15:22:54 [danb_lap]
danbri: some stuff seems core, architectural, structural; other bits are useful/handy/nifty properties useful for real world EARL apps
15:23:26 [sbp]
yeah. classifying the lot is surprisingly difficult, though
15:23:31 [danb_lap]
nadia: I'm in favour of cleaning up the schema. Folk aren't using the d+o stuff, and there are a bunch of properties being neglected; others that should be there but aren't.
15:23:45 [libby]
wendy: looks like we will reqwrite the schema
15:23:59 [libby]
sbp, wendy is asking if you agree split schema?
15:24:05 [wendy]
sbp are you in favor of splitting it into 2, ala nadia's proposal (from a while back)
15:24:08 [sbp]
s/but aren't/but have to be infered/
15:24:25 [danb_lap]
sbp, RSS has had troubles that weren't technical in origin
15:24:33 [chaalsBRS]
aah. Sorry about that before
15:24:51 [chaalsBRS]
rereading Giorgio's email -
15:25:05 [nadia]
15:25:32 [sbp]
heh, cool:
15:26:41 [sbp]
any reason why we can't have earl:asserts and earl:assertedBy and let people choose either?
15:26:57 [libby]
might make queries a touch trickier
15:26:57 [nadia]
it'll make parsing unknown earl reports more difficult
15:27:05 [libby]
wendy talks about things to discuss tomorrow
15:27:10 [nadia]
er what libby said
15:27:12 [libby]
- rdf database requirements
15:27:33 [libby]
- intereoperable extensions - hytime examples
15:27:36 [chaalsBRS]
Suggestion that the location of a problemidentify a normalisation that people can replicate. I guess that is the thing of whether tools will implement a normalisation just for the sake of EARL
15:27:42 [libby]
- tomorrow we can play!
15:27:54 [sbp]
are the cats out?
15:28:26 [libby]
15:28:46 [libby]
wendy would like to generate earl for the usercases that we don;t have earl for already
15:29:00 [sbp]
"when the cats are out, the mice shall play"
15:29:02 [sbp]
(or somethingorother)
15:29:05 [libby]
dp: is not sure could put the test requirements in earl, but certainly the test specification
15:29:10 [libby]
15:29:19 [libby]
the cats are off to the pub...
15:29:22 [danb_lap]
sbp, we're wrapping up. Any final thoughts...?
15:29:23 [sbp]
heh, heh
15:29:45 [sbp]
nah. I'll just rant via. email if I see fit
15:29:50 [danb_lap]
15:29:53 [danb_lap]
ok, take care
15:29:57 [sbp]
cheers, you too
15:30:00 [sbp]
have fun at the pub :-)
15:30:27 [danb_lap]
15:30:36 [libby]
thanks sbp :)
15:30:44 [danb_lap]
RRSAgent, pointer?
15:30:44 [RRSAgent]
15:43:16 [chaalsBRS]
bye folks