Semantic Web Interest Group IRC Chat Logs for 2002-05-28

00:13:59 <AaronSw> MarkB, in the email archives

00:14:02 <AaronSw> for ws-arch

01:26:57 <MarkB> thanks Aaron, I found them. He was being facetious in that email.

01:30:41 <AaronSw> oh, ugh

01:36:43 * eikeon is back (gone 07:16:36)

06:42:18 <tav|sleep> tav|sleep is now known as tav

06:49:13 * eikeon is away: zzz

09:58:55 <danbri>http://www.jibbering.com/2002/4/shepherds-1000.svg

09:58:56 <dc_rdfig> A: http://www.jibbering.com/2002/4/shepherds-1000.svg from danbri

09:59:23 <danbri> A:|Nice demo Jim made, using a raster2vector tool (url???) and SVG to show how image region outlining can be machine assisted

09:59:23 <dc_rdfig> titled item A

10:26:44 <edd> hey danbri

10:26:52 <danbri> mornin'

10:27:18 * edd fiddling with redland til late last night, just woke up. still confused about timezones.

10:27:25 * danbri playing 3D Monster Maze on virtual ZX81

10:27:53 <danbri> my sympathy. I'm back on a pretty ideal schedule. Tired late evening, wake up 6am. Notmy usual schedule, but makes a change.

10:28:07 <danbri> 10 PRINT "SEMANTIC WEB VAPOURWARE CONSIDERED HARMFUL"

10:28:09 <danbri> 20 GOTO 10

10:28:20 <danbri> ah, the old days...

10:28:33 <edd> i remember well my 10-entry addressbook that i wrote.

10:28:37 * danbri wipes a retro-chic fake tear from eye

10:28:47 <edd> took the screen 30 seconds to render.

10:29:12 * edd waves a happy goodbye to the days of RAM pack wobble.

10:29:15 <danbri> I knew z80 machine code on zx81, though never wrote anything useful. I've got progressively less technical since then.

10:29:24 <danbri> Nothing bluetac couldn't fix :)

10:29:56 <danbri> redland: interesting. playing w/ foaf aggregator ideas, or other stuff?

10:30:29 <danbri> Did you see I got it build under cygwin w/ no problems (use the cvs version...). And running outside cygwin environment seems easy, .DLLs can be generated etc too.

10:30:59 <danbri> So C-based apps, maybe scripted from VB, should be possible. This was btw w/out the BerkleyDB stuff, though that should work too I'd guess.

10:31:09 <danbri> it=redland

10:45:21 * edd on phone. i basically got it compiled and installed then looked at the py interface, which needs some changing. i'll either persuade dajobe or do them myself.

10:53:16 <danbri> there's changing vs glueing it to the other Py APIs around... same re Java etc... -- interesting study in cross-language api issues

10:53:49 <edd> i'd settle for fixing it as a first port in the storm.

10:55:12 <danbri> ah ok, just a bug. I thought you meant aesthetics :)

10:56:00 * danbri wonders if redland has an operation for 'rename all uses of this bNode-identifier to be xyz'

10:56:23 <danbri> need something like that for foaf / smushing

10:56:44 <danbri> .google rdf smush

10:56:45 <xena> rdf smush: http://rdfig.xmlhack.com/2001/07/20/2001-07-20.xml

10:56:50 <danbri> boo

10:56:55 <danbri> .google rdf smush design

10:56:56 <edd> i daresay i'll get near that. the short term aim is FOAF

10:57:03 <xena> rdf smush design: http://rdfweb.org/2001/01/design/smush.html

10:57:22 <danbri> Really early on w/ foaf, I found a real world vs RDF problem with our old Perl RDF API.

10:58:01 <danbri> Soon as you start merging RDF from multiple places, you find a need to deal with node-renaming etc. The old Perl stuff didn't support this, hence the broken state of the perl / rdfweb /foaf demos.

10:58:20 <danbri> Libby's was SQL-backed, where much easier to do such things (UPDATE yadda WHERE etc...).

10:58:39 <danbri> re short term aim is FOAF: :) :)

10:58:51 <edd> i'll daresay i'll have to bug dajobe. it's fun being a Customer.

10:59:48 <danbri> even more fun being a customer with use cases. that smush.html above has a crap example case, but tries to talk thru the design issue that FOAF threw up re RDF data merging.

11:00:51 * edd reads that smush.html

11:01:42 * edd still on phone

11:03:04 <edd> hmm, reading smush.html... there are bnodes there. this is some future rdf/xml?

11:04:06 <danbri> bNodes are just the name we have for what we used to call 'anonymous nodes' (or worse, 'anonymous resources'). They were in the original RDF spec, but ill-explained.

11:04:23 <danbri> It's just where you mention something without giving a URI for it. Real RDF is full of that...

11:04:28 <edd> yes, i understand.

11:04:47 <edd> last night i was told something that conflicts. hold on, let me research...

11:04:49 <danbri> why ask re 'future'? ('dark triples' and similar voodoo are future, perhaps neverever, features..)

11:04:52 <danbri> ok

11:08:55 <edd> hmph, someone must have been tired last night when they told me stuff. bah.

11:09:15 <danbri> bNodes are definitely in.

11:09:34 <edd> yeah, i figured so. just the midnight hacking haze, i guess.

11:10:14 <edd> ok, i'm happy. on with hacking.

11:10:17 * danbri pleased that http://rdfweb.org/2001/01/design/smush.html begins with semi-plausible application scenario, and ends with links to actual implementations (though the bit in between needs work...)

11:10:19 <danbri> ok!

12:15:35 <JibberJim> A:Created with help from [CR2V|http://www.celinea.com].

12:15:35 <dc_rdfig> added comment A1

12:16:01 <JibberJim> A:Seemingly no longer in development, and no source.

12:16:01 <dc_rdfig> added comment A2

12:26:37 <JibberJim> A:[Other conversion tools which may help|http://www.w3.org/Graphics/SVG/SVG-Implementations#convert]

12:26:37 <dc_rdfig> added comment A3

12:55:17 <nadia> is there anybody awake who can help me with a problem I'm having with the Jena library?

12:56:29 <ambient> maybe

12:56:35 <ambient> what's the problem?

12:57:23 <nadia> it's throwing a MalformedURLException when i'm asking a model to read some rdf...

12:57:49 <nadia> i'm playing with rdfpic, and the rdf in question was produced by a previous version of rdfpic... it validates in the rdf validator

12:58:03 <ambient> strange

12:58:22 <ambient> the rdf validator and jena both use ARP, so both should either validate or not validate

12:58:56 <nadia> that's what i thought...

12:59:04 <ambient> why not post the RDF to the jena mailing list

13:00:23 <nadia> what's the address?

13:00:39 <ambient> jena-dev@yahoogroups.com

13:00:47 <nadia> thanks

13:31:21 <danbri> nadia, is the rdf file online anywhere?

13:32:07 <nadia> i suppose i could put it online

13:32:17 <nadia> i'm reading it from the jpeg file

13:33:52 <danbri> I'd take a look if you can get it online easily

13:34:01 <nadia> i posted it to the jena-dev group

13:34:47 <JibberJim> RDF inside the JPEG? - How do you get it out?

13:35:14 <nadia> it's written to a field of the jpeg

13:35:51 <JibberJim> Do you know of a writeup on how to extract/add to ?

13:36:29 <nadia> umm... i don't know of any writeups, but if you look at the code you can probably figure it out..

13:36:45 <nadia> each field has a marker... you scan for the markers and then read the contents of the field

13:37:41 <JibberJim> Actually no, I'm not going to be distracted, I'm going to use Annotea for image metadata.

13:37:53 <nadia>http://dev.w3.org/cvsweb/java/classes/org/w3c/tools/jpeg/

13:37:54 <dc_rdfig> B: http://dev.w3.org/cvsweb/java/classes/org/w3c/tools/jpeg/ from nadia

13:38:27 <nadia> you could use rdfpic to mark them up :)

13:38:52 <nadia> i'm about to come out with a new version that reads exif headers and translates them to rdf

13:39:33 <JibberJim> B:|JPEG comment writer source code

13:39:33 <dc_rdfig> titled item B

13:39:47 <JibberJim> B:Add RDF to jpegs.

13:39:47 <dc_rdfig> added comment B1

13:40:31 <JibberJim> The photo-rdf namespace of rdfpic doesn't say the sort of things I want to say...

13:40:46 <ambient> nadia: when you read the RDF into the ModelMem, do you pass a Base URI?

13:40:48 <JibberJim> I don't even know what my camera is for example...

13:41:45 <nadia> ambient: no

13:42:18 <ambient> nadia: I think you need one, because of the rdf:about='' line

13:42:20 <nadia> ambient: what i'm mainly confused about is that rdf that looks almost exactly the same works just fine...

13:42:49 <ambient> Jena needs a base URI to turn '' into an absolute URI

13:43:57 <nadia> ah ha... even if the base URI is just given as ""?

13:44:03 * nadia thinks she sees what the problem is..

13:44:13 <ambient> well, "" is not a valid URI

13:44:24 <ambient> hence the exception

13:44:53 <nadia> well... i just noticed that in the case that worked i was giving "" as the base uri... and in the other case i gave nothing

13:47:11 <nadia> ok, thanks a lot, it works now with "" as my base uri... should i expect that not to work in the future?

13:47:18 <ambient> yes :)

13:48:40 <nadia> meanies

13:48:51 <nadia> :)

13:49:43 <nadia> is "#" a valid uri?

13:50:45 <ambient> i think all uris need colons in them

13:52:39 <ambient> i know that jena doesn't check uri syntax in some places

13:53:26 <larsbot> all absolute uris do; the question is if "#" is relative

13:53:35 <larsbot> as an absolute URI it's invalid

13:53:56 <nadia> well, jena doesn't like it

13:53:58 <ambient> and i think Model.read() in jena needs an absolute URI

13:54:02 <nadia> i think i need to use ""

13:55:00 <larsbot> why? neither "" nor "#" makes much sense as an absolute URI

13:55:36 <JibberJim> but they do make sense for saying this metadata is about the document it's contained in.

13:55:55 <larsbot> "" does for that, yes, but ambient seems to imply that the URI is absolute

13:56:39 <AndyS> ARP does check URI syntax (it uses code from Xerces, I think). The rest of Jena is less check-ful.

13:57:41 <AndyS> The W3C validator isn't running the latest ARP as far as I know.

13:57:56 <nadia> "" also makes more sense for my purpose..

14:11:33 <danbri> <JibberJim> Do you know of a writeup on how to extract/add to ?

14:11:55 <danbri> I once extracted the Java classes from RDFPic that do this. Simple enough API, if I remember right

14:12:07 <nadia> yes

14:12:23 <nadia> see url B from above

14:12:30 <nadia> those are the ones rdfpic uses

14:12:38 <JibberJim> yeah but some of us are naturally anti Java.

14:12:42 <nadia> hehe

14:13:16 <danbri> Yes, the URIs in the RDF graph need to be absolute. RFC2396 tells you some ways of getting there. Passing in some URI that you own, or a mid or uuid URI, should do it.

14:14:01 <danbri> 'malformed' is perhaps not most helpful of error msgs, if this is really the prob.

14:15:32 <nadia> well ok... but i have a philosophical/practical issue with real absolute uris from rdfpic's point of view

14:16:11 <nadia> if i mark up an image while it's at one location, and then move it...

14:17:03 <danbri> Yes, that's an RDFPic problem: it should omit the about='' from the markup, imho.

14:17:16 <danbri> And use some other means of identifying the image

14:17:18 <danbri> eg.

14:18:25 <danbri> <image sha1:sum="34254829423984238942349" dc:title="Some photo" dc:description="This JPEG when fed to sha1sum algorithm has the specfied value" dc:format="image/jpeg"/>

14:18:34 <danbri> rather than

14:18:39 <danbri> <image rdf:about="" sha1:sum="34254829423984238942349" dc:title="Some photo" dc:description="This JPEG when fed to sha1sum algorithm has the specfied value" dc:format="image/jpeg"/>

14:19:06 <danbri> the former will create a blank-node in Jena (hopefully). The latter won't, and requires a base URI context

14:19:11 <JibberJim> but you'd have to extract the comments from the jpeg before sha'ing and then which comments would you remove etc?

14:19:13 <danbri> Some folk are happy with that...

14:19:26 <danbri> Hey, yeah, tricky point.

14:19:31 <danbri> All my photo metadata lives outside.

14:20:00 <danbri> it'd need a sha1:sum-when-metadatafieldsblankedout property

14:20:25 <danbri> Or sha1sum or 1st or last n bytes or something.

14:20:50 <danbri> Same issue w/ Adobe XMP btw

14:21:27 * JibberJim is going to store his photo metadata in annotea.

14:21:55 <danbri> .zip files w/ a manifest would be my preference, if everyone did it

14:21:58 * JibberJim is also shortly going to begin friendly IE interface to annotea with the nice folks at http://www.meadroid.com/zeepe/

14:22:11 <danbri> I just put rdf files in the web scattered around at random

14:22:28 <JibberJim> Or as an SVG file, with inlined JPEG, and metadata section?

14:22:33 <danbri> annotea-type-servers are indexers, not main stores

14:22:41 <danbri> inlined ascified jpeg?

14:22:51 <nadia> ew

14:22:55 <danbri> annotea: cool!

14:22:59 <danbri> yes, ew :)

14:23:17 <JibberJim> hey it's part of the SVG spec...

14:23:51 <danbri> a workaround for lack of packaging conventions for web content, i guess

14:26:42 <JibberJim> e.g. http://www.showcaster.com/conference/showcaster/intro/download.svgz which probably doesn't work properly yet.

15:11:41 * eikeon is back (gone 08:22:26)

15:43:59 <rreck> so much for rdfing the imdb

15:44:23 <rreck> 3. Specifically the files may NOT be used to construct any kind of on-line database

15:46:17 <rreck> i need to create a scheme where people give ME data, that then I hold the rights to

15:53:38 <Seth> what's the difference between a predicate and a property?

15:56:07 <sbp> see #swhack

15:56:07 <sbp> [[[

15:56:08 <sbp> <sbp> the predicate is the thing in the middle of a triple, a property is the class of things that join a subject and object

15:56:08 <sbp> <sbp> so in rdfs:comment rdfs:label "comment" . both rdfs:comment and rdfs:label are properties, but only rdfs:label is the predicate in this triple

15:56:08 <sbp> ]]]

15:57:08 <rreck> hi

15:58:45 * jhendler hi - filling out paperwork to continue your appt - assume you're okay with this :->

16:00:28 <edd> we could debate :)

16:00:54 <Acapnotic> AaronSw: did your internet leave? I can't get through to i.e. logicerror.com

16:01:30 <AaronSw> yeah, logicerror and blogspace.com (but not google.blogspace.com) are down

16:02:17 <AaronSw> is there something i can help you with?

16:02:55 <AaronSw> ah, so you go to oberlin? Hm, looks like they deleted your website...

16:02:57 <Acapnotic> not particularly, I was just hitting some pages I bookmarked in the last few days and got some timeouts.

16:03:22 <Acapnotic> AaronSw: went there. it's been a few years now, I guess.

16:03:47 <AaronSw> ah

16:04:15 <AaronSw> yeah, sorry. they should be back up soon

16:07:16 * Acapnotic . o O ( Where did you find a pointer to me@oberlin? )

16:07:26 <AaronSw> via advogato

16:08:15 <Acapnotic> ah, thought that might be it. I lost my password there a while back =(

16:09:18 <Acapnotic> as evidenced by the fact that my last blog entry was fall 2000 (!)

16:11:27 <Seth> sbp, im gonna make a case to respectfully disagree .. i think that distinction between property and predicate is vague and ambiguous ... is that distinction written up formally in any w3c spec?

16:11:46 <sbp> um... the RDF specification, I'd presume

16:12:05 <Seth> which one?

16:17:12 <AaronSw> M&S, probably

16:17:19 <AaronSw> sbp had a really good explanatiopn, i thought

16:17:19 <Seth> from M&S we have [[A property is a specific aspect, characteristic, attribute, or relation used to describe a resource]]

16:18:07 <Seth> which i would say matches your definition above for property

16:18:21 <Seth> now about 'predicate'

16:19:11 <Seth> also form M&S [[These three individual parts of a statement are called, respectively, the subject, the predicate, and the object.]]

16:19:16 <sbp> you can think of predicate as being a property that links a statement (ID) to its arc. the range of the property is rdf:Property

16:19:34 <sbp> rdf:predicate rdfs:range rdf:Property .

16:23:58 <Seth> In the statement {Seth isa person,human.} we would both agree that 'isa' identifies a property ... and that 'isa' identifies the predicate of that satement ... right?

16:24:57 <sbp> isa is the predicate of the triple yes, and (therefore) it belongs to the class rdf:Property

16:25:36 <Seth> please restate because {Seth isa person,human.} (not isa) triple.

16:26:23 <Seth> rather it is a sentence .. you could phrase it as a N3 sentence too.

16:26:37 <Seth> it actually expressess 2 triples.

16:27:35 <Seth> and i agree that 'isa' identifies the rdf:Property isa.

16:27:42 <Acapnotic> Instead of defining a predicate as a property, I would have defined a Property as something that can be a predicate. "predicate" is a grammatical term from RDF's logic underpinnings, whereas "Property" describes a slightly more application-specific concept.

16:28:59 <Seth> acapnotic, actually my point in in another direction ... i would agree that only a property can label a predicate.

16:29:30 <sbp> in the N3 "@prefix : <#> . :x :y :z ." :y is the predicate of the only triple that appears within it. since all predicates are members of the class rdf:Property, :y is therefore an rdf:Property.

16:29:45 <Seth> my point would be that in my example the predicate is "isa person,human".

16:30:41 <Seth> but i really dont know .. this is a learning experience here

16:30:48 <sbp> no, you have two objects in there too

16:31:36 <jonb> seth: to distinguish, find a way that you use one term in a way that is different than you use the other term

16:33:05 <Seth> jonb, the predicate of {seth isa person,human} is 'isa person,human' and the property of that predicate is isa.

16:33:47 <jonb> sorry, makes no sense to me, try Ntriples so that we both don't get confused

16:33:50 <Seth> but sean, and M&S and the grammar are saying things differently:

16:34:59 <Seth> they say the predicate of {seth isa person,human} is 'isa' and 'isa' identifies a property.

16:36:27 <Acapnotic> which is what you said. the label of the predicate, "isa", is the identifier for a Property.

16:37:04 <Seth> i guess i have to go with M&S, which apparently is askance to the usage of these words in traditional English language grammar.

16:39:21 <Seth> iow: in M&S grammar, a predicate is the name of the grammatic slot .. and a property identifier is what can exist in that slot.

16:40:18 <Seth> but in English grammar, the predicate slot is filed with both the verb and its objects.

16:41:46 <sbp> yes, but only if you're talking about "predicate" in the linguistic sense

16:41:55 <sbp> since RDF is not natural language, I'd have to guess that we're not :-)

16:42:54 <Seth> huh, i dont grok your distinction ... rdf is a language and we are talking about it in a linguistic sense .. what do you mean?

16:44:33 <Seth> for the record .. from English grammar: [[Each sentence contains (or implies) two parts: a subject and a predicate. The predicate is what is said about the subject.

16:44:36 <Seth> ]]

16:45:56 <sbp> well, RDF isn't English. what can I say? it's not a natural language. if you're saying that the authors of M&S abused the word "predicate", then I think I could take you up on that point too, since the word is much like "context" and "ontology" in that nobody really has a clue what it means in any given situation :-)

16:46:04 <Seth> so i got a choice: go with M&S and against English grammar, or go with English grammar usage ... me thinks ill choose the latter.

16:47:29 <Seth> well we could argue long and hard about whether rdf could become a natural language or not. we were all over that in swag. rdf has some properties of a formal language and some of the properties of a natural language .. so i guess it's really a hybred.

16:47:29 <sbp> whatever. just keep in mind that RDF M&S is (unusually!) clear on this topic, and that RDF != English or any other natural language

16:48:37 <Seth> sbp, i agree with you on that .. M&S is very clear on this .. and its quite clear that RDF != English.

16:49:19 <Seth> but Semenglish != RDF, so i will be making some choices that squew it closer to English, than to RDF.

16:50:56 <sbp> for the purpose of being clear about terminology, it's important to ground terms about RDF in terms of the RDF specification, otherwise you introduce confusion. All I can say is that I advise you to use predicate--when talking about RDF--to refer to the middle term in a triple. if not, you run the risk of being misinterpreted

16:51:14 <Seth> in particula, in the Pentals grammar, i think i'll invent a new grammatic identifier ... probably something like "predicateSlot" and use that instead of predicate.

16:52:10 <Seth> which does ground the terms in RDF, yet still allows me to talk about a predicate as something said about the subject.

16:54:35 <Seth> thing is in my python tree, the predicate is a branch that (imho) includes the object leaves as well as the property in the predicateSlot

16:55:46 <Seth> ... just testing out my term usage on you guys

16:56:07 <jonb> seth: a basic way to deal with terms in any language that you ever encounter is to understand that the meaning of any term is always dependent on the context in which it is used

16:56:34 <Seth> yep

17:07:30 <Seth> another question, doesn't the N-triples grammar definition need to be changed for the new definition of literals?

17:09:04 <Acapnotic> "new definition" where?

17:09:19 <Seth> and if so, where is the new EBNF for N-triples?

17:09:43 <dajobe> in the N-Triples document

17:09:55 <dajobe> which you has been published as a WD for a month approx

17:10:20 <Seth> im looking at http://www.w3.org/2001/sw/RDFCore/ntriples/

17:10:47 <dajobe> and what does it say in STATUS OF THIS DOCUMENT?

17:11:45 <Seth> ok thanks

17:11:58 <dajobe> :)

17:13:05 <AaronSw> I think it might be good to split n-triples back out, so that people can use it on its own

17:13:23 <dajobe> that might encourage people to think it wasn't for test cases :) :)

17:13:31 <AaronSw> exactly!

17:13:44 <AaronSw> well, it's not _just_ for test cases.

17:14:24 <Seth> i agree with aaron

17:14:53 * sandro_ quietly leaves the room.

17:15:19 <sandro_> sandro_ is now known as sandro

17:15:42 <AaronSw> ;-)

17:44:52 <Seth> isn't there better URL for URI References than http://www.w3.org/TR/rdf-testcases/#sec-uri-encoding ?

18:53:52 <sandro> In an OOP language, what do you call the class which represents things about which one has RDF information?

18:54:43 <Acapnotic> well, in RDF applications that's generally the "Resource"

18:59:20 <sandro> Even when it's a literal?

19:00:20 <bijan> I've seen them separated into two classes. Resource adn Literal :)

19:00:41 <sandro> I guess I'll stick with Object.

19:01:18 <bijan> Little nasty because of 1) many oo systems have an Object and 2) the "object" position of an rdf statement.

19:01:26 <bijan> But hardly enough to quibble over.

19:01:50 <sandro> Agreed. I guess DAML went with "Thing" for the second reason.

19:02:05 <bijan> Node is another one with tradeoffs.

19:02:09 <bijan> cwm might use thing.

19:02:39 <jhendler> we couldn't find a term everyone could agree to -- all the good ones have been taken by multiple research communities - a real mess.

19:02:49 <sandro> yeah, node seems quite wrong for predicates.

19:03:18 <bijan> lets see, redfoot uses resoruce and literal.

19:03:39 <sandro> And resouce and literal don't inherit from some common superclass?

19:04:16 <bijan> They do...str, it looks like.

19:04:32 <bijan> Nodes have two types - resource (used for subject, predicate and object) and literal (only used for object). Resources have URIs, Literals have string content plus optional XML Language, XML space statement and a flag if the content is XML which is used for supporting RDF parseType="Literal".

19:04:32 <bijan>

19:04:36 <bijan> Redland

19:04:52 <dajobe> (I nicked terminology from stanford java api)

19:04:59 <dajobe> (like jena, IIRC)

19:05:25 <dajobe> ugh, s/xml space//

19:06:06 <bijan> wilbur seems to use 'node'

19:06:42 <bijan> Which i can rationalize, but won't :)

19:07:21 <jhendler> kind of like node - a graph should have nodes and links...

19:07:44 <sandro> but the links in RDF are the same kind of thing as the nodes. :-)

19:07:53 <bijan> Yep.

19:07:59 <bijan> Since the object of a statement can be either a resource or a literal, the getObject() method returns an object typed as RDFNode, which is a common superclass of both Resource and Literal. The underlying object is of the appropriate type, so the code uses instanceof to determine which and processes it accordingly.

19:08:09 <bijan> Jena.

19:08:26 <bijan> Heya jim.

19:08:30 <sandro> So the arcs are RDFNodes are well, I guess. (looking...)

19:08:48 <bijan> I would think so.

19:09:18 <bijan> Jena has object classes to represent graphs, resources, properties and literals.

19:10:09 <sandro> I think Node or RDFNode is worse than Thing or Object. Ho hum.

19:10:39 <bijan> Eh.

19:10:43 <bijan> Prolly.

19:10:46 <bijan> Marginally.

19:11:31 <bijan> Aside from the (minor) quibbles iwth Object, both Object and Thing are *prime* ontology terms.

19:11:52 <bijan> Keeping those namespaces somewhat distinguishable is prolly a good idea.

19:12:09 <bijan> For many purposes.

19:12:44 <bijan> I kinda of like Resource and Literal myself.

19:13:08 <sandro> But what's their superclass?]

19:13:20 <bijan> Do they need one?

19:13:33 <bijan> AbstarctRDFResourceOrLiteral :)

19:13:40 <sandro> Yeah, sure -- otherwise what does getObject return, etc.

19:13:41 <AaronSw> Literals are Resources, no?

19:13:52 <bijan> Aaron, not in the MT, I believe

19:14:11 <sandro> Well, I certainly don't want to call them "resources" anyway. :-)

19:14:34 <AaronSw> I think this is a job for... Ted Nelson!

19:15:07 <bijan> Jena has object classes to represent graphs, resources, properties and literals.

19:15:09 <bijan> Oops.

19:15:14 <bijan> RDF uses two kinds of referring expression; urirefs and literals. We make very simple and basic assumptions about these. Urirefs are treated as logical constants, i.e. as names which denote something (the things are called 'resources', but no assumptions are made about the nature of resources.) Literals are treated as denoting themselves, ie each literal is essentially a quoted character string. (A future extension of RDF is planned which will allow da

19:15:34 <bijan> We will define LV to be the set of all possible literals, ie all character strings that can occur as a literal label in an RDF graph.

19:15:59 <bijan> IR is the set of resources.

19:16:16 <bijan> I believe that LV need not be a subset of IR, but that's not clear.

19:16:20 <bijan> (to me)

19:16:21 <bijan> (yet)

19:17:46 <sandro> How so, Aaron?

19:17:49 <bijan> Have I mentioned that I love the MT? I love the MT.

19:18:09 <bijan> It's *such* a good read.

19:18:17 <dajobe> bijan: LV; it need not but it is written so that it can be a subset of IR

19:18:35 <bijan> Right.

19:18:37 <AaronSw> Ted Nelson is sure to come up with a cool new word for this superclass. Like dag, humber, etc.

19:18:37 <dajobe> IIRC from earlier drafts

19:18:45 <bijan> That's what I thought.

19:19:04 <bijan> It seems a distinction worth marking in your api, but that's heavily dependant on the sort of app you're writing.

19:19:14 <dajobe> this is due to the RDF core issue that we punted on "literals are resources" (which we mostly agree with I think) but didn't decide to decide. ahem :)

19:19:43 * eikeon is just working on a doc to outline Redfoot's new generator based triple store interface: http://redfoot.sourceforge.net/1.5.2/doc/tripleStore.html

19:19:49 <bijan> Yay!

19:19:57 <bijan> Looking forward to that eikeon.

19:20:16 <bijan> Sorry not to have responded to your note. Just got back from apt hunting.

19:20:27 <eikeon> Also noticed that we could use on for the 'objects' and perhaps call them something better?

19:20:44 <dajobe> what's generator?

19:21:00 <AaronSw> generator is a python idiom for basic coroutines

19:21:05 <bijan> Ok, this...*this* is a *sentence*:

19:21:10 <bijan> """Implementors who wish to check any kind of entailment should use a process which is optimised for the combinatorics of the particular set of use cases that are most likely to arise in a given application area. """"

19:21:23 <bijan> Have I mentioned that I *love* the MT?

19:21:43 <bijan> 'combinatorics'...what a great word.

19:21:44 <dajobe> it's those lovely pics

19:21:54 <bijan> The pics are nice, I grant.

19:22:02 <bijan> But the sentence i cnped is my fav.

19:22:15 <bijan> That's where I went over from admiration to adoration :)

19:22:20 <AaronSw> You know, perhaps future editions of the MT should just be mailed to bijan and PFPS...

19:22:44 <bijan> Actually, I think this draft should be fairly accessible.

19:22:57 <bijan> I had a pretty tough time with some of the earlier drafts.

19:23:03 <bijan> But this was dreamy.

19:23:12 <Acapnotic> I read the MT the other day. I think I got most of the way through and only dropped out in a few spots.

19:23:49 <bijan> There are still a few less that clear spots.

19:24:07 <bijan> Some quite easily clarified.

19:24:35 * dajobe lol at "dreamy"

19:24:48 <bijan> :)

19:25:04 * jonb agrees that the MT is a terrific document in all respects

19:25:05 <bijan> Hey! We logic geeks gotta have fantasy's too! :)

19:25:38 <jonb> it actually has good prose descriptions of what it is doing

19:25:41 <bijan> Well, the pseudo n3 Qname presentation syntax has to go, but yes :)

19:25:42 <bijan> Yes.

19:25:50 <bijan> Excellent ones, in fact.

19:27:19 <sandro> I should read the MT in some newer draft. I fear for all its beauty, it may say the wrong thing in some places.

19:27:43 <jonb> wrong for whom?

19:27:54 <dajobe> well, since it says 0 on reification and containers ....

19:28:10 <jhendler>http://www.openhealth.org/WOWG/DTTF-20020527

19:28:10 <dc_rdfig> C: http://www.openhealth.org/WOWG/DTTF-20020527 from jhendler

19:28:19 <jhendler> C:| Latest on Dark Triples

19:28:19 <dc_rdfig> titled item C

19:28:36 <jhendler> C: JonB's posting to WebOnt of a summary of the DT issue

19:28:36 <dc_rdfig> added comment C1

19:28:38 <dajobe> oh, I'm wrong. http://www.w3.org/TR/rdf-mt/#Reif

19:28:40 <AaronSw> It was either going to be DataTypes or Dark Triples. We're doomed either way

19:28:52 <danbri> sandro, yes do please read the MT

19:28:55 <jhendler> C: nice summary of how unasserted triples matter

19:28:55 <dc_rdfig> added comment C2

19:29:05 <jhendler> Datatypes is next on the agenda :->

19:29:09 <danbri> review comments -> mailto:www-rdf-comments@w3.org v welcome

19:29:24 <danbri> and RDFS, Syntax, Primer too!

19:29:33 <jhendler> c2: nice summary of why unasserted triples matter

19:29:40 <sandro> danbri, I think I should do my formal semantics for PTL and/or CLR, to push back on the model theory, before I study it more.

19:30:01 <danbri> I don't know PTL, CLR

19:30:04 <jhendler> C: | Latest on Unasserted (nee dark) Triples

19:30:05 <dc_rdfig> added comment C3

19:30:15 <AaronSw> C:"The effort to both develop OWL as an extension to RDF and assert each and every triple, appears to have been akin to an attempt to stuff a balloon filled with water into a cube whose volume is less than that of the balloon: every attempt to stuff the balloon in results in some part of the balloon squeezing itself out somewhere."

19:30:15 <dc_rdfig> added comment C4

19:30:18 <danbri> not sure what you mean re 'push back'

19:30:21 <jhendler> C:| Latest on Unasserted (nee dark) Triples

19:30:21 <dc_rdfig> titled item C

19:30:40 <jonb> he he

19:30:43 <dajobe> C:can this be posted to www-archive for archiving?

19:30:43 <dc_rdfig> added comment C5

19:30:54 <sandro> "push back" == really be testing it, against something that will want it changed.

19:31:19 <jonb> google has probably already archived it, plus I posted all the text to webont wg in an email

19:31:25 * eikeon is away: back in a few hours

19:31:39 <danbri> testing is good; reviewing the docs on their own terms the main thing though, even if PTL/CLR would do it differently (even better).

19:31:54 <jonb> ptl/clr??

19:31:59 <jhendler> C3: discussion thread at [http://lists.w3.org/Archives/Public/www-webont-wg/2002May/0229.html|webont archive]

19:32:00 <dc_rdfig> replaced comment C3

19:32:12 <danbri> =sandro work in progress, i think

19:32:38 <dajobe> C:"Conclusion ... Two mechanisms are provided:" those are potential solutions, not provided

19:32:38 <dc_rdfig> added comment C6

19:32:51 * jhendler finally learned the right syntax for chumping ...

19:32:58 <sandro> Positive Triples Logic (like cwm-- something we beleive can be done in RDF), and Classical Logic for RDF (something I suspect can't be done, because it has negation, but want to try for myself).

19:33:13 <jonb> ah

19:33:20 <Acapnotic> hmm. may I not subscribe to webont, though the archives are public?

19:33:42 <jonb> dajobe: expand?

19:35:30 <dajobe> the "Implementing .. in RDF" section seems like these are all the options. There are more, and we've discussed them here.

19:36:05 <jonb> oh, sorry, what others would you like to see?

19:36:07 <dajobe> at what point should rdf core take this doc (say) and pick some solution(s), add more

19:36:19 <jhendler> acapnotic - strangely enough, that is the case - something to do with a combination of ACLs and who can post where.

19:37:07 <Acapnotic> That's unfortunate.

19:37:12 * jhendler one of those mysteries of life that only DanCon fully understands

19:38:15 <jonb> dajobe: i just discussed the options for unasserted triples that the webont group specifically discussed, i was summarizing the WG discussion, not attempting to give the topic a more comprehensive treatment (scary that after >300 emails there might be something more comprehensive)

19:38:17 <AaronSw> Heh.

19:38:54 <dajobe> jonb: oh, I thought this doc was going to rdf core with an "provide this please" header :)

19:39:07 <jonb> yes :-)

19:39:42 <jonb> well, really, provide this facility, and by the way these are several ways to do it, but you decide

19:40:24 <dajobe> and pick others?

19:40:40 <jonb> sandro: why do you say that cwm can be done in RDF? cwm makes extensive use of unasserted contexts and other non-RDF thingies

19:41:04 <jonb> dajobe: if that's what rdfcore decides

19:41:09 <AaronSw> I'm not following this problem. If OWL can express a paradox, why is that our fault?

19:41:27 <bijan> It's *always* your fault, Aaron.

19:41:32 <AaronSw> and what does the student example have to do with anything?

19:41:40 <bijan> That's web axiom #pi

19:41:52 <jonb> webont is chartered to layer on RDF and OWL needs this in order to do it

19:41:54 <AaronSw> I know, I know, but are there any _other_ reasons?

19:41:59 <AaronSw> (to bijan)

19:42:05 <bijan> Web Axiom #2pi is: even if it's not aaron's fault, blame him anyway.

19:42:21 <jonb> student example is a simple example that illustrates point

19:42:33 <bijan> Web Axiom #pi/2: And misspell "aaarron" as much as possible :0

19:42:44 <jonb> any use of "intersectionOf" will do, really

19:42:57 <AaronSw> I don't see why adding new entailments requires dark triples.

19:43:06 <bijan> The student example is an example of an entailment webont should support.

19:43:18 <jonb> it doesn't, but without unasserted triples you get paradoxes

19:43:28 <jhendler> Aaron - seriously, there's two choices - WebOnt goes its own way, but that bothers RDF purists (and I'm one), other is RDF creates some mechanism to say to an rdf interpreter "ignore this" (i.e. <rdf:unasserted> ) so that something else can interpret it - I don't like that a lot better.

19:43:51 <sandro> jonb, I'm not sure what you mean by "in RDF". I think much of what cwm does can be guided by RDF M&S 1.0 along with my PTL ontology, in a semantically sound way. I don't know how to do cwm's log:notIncludes in a semantically sound way, but I haven't tried really hard. (it's notInlcudes, not notImplies, so it may be possible.)

19:43:59 <bijan> It's not *just* new entailment, it's supporting certain entailments, plus everything asserted, plus a few other things.

19:44:01 * jonb thinks that the average database geek could care less whether a triple is asserted in any case

19:44:14 <jhendler> the choice I like best is not viable to most people -- that would be for RDF (and probably WebOnt too) to tear up their model theories for a few years until we've worked this out...

19:44:25 <jonb> i.e. a triple is a triple

19:44:41 <jonb> and webont uses some triples as syntax and others have meaning

19:44:45 <AaronSw> What I'm looking for is an RDF entailment that causes problems for WebOnt entailements, I guess.

19:44:48 <jonb> s/webont/OWL

19:44:52 <sandro> What matters is who claims that the triple represents some fact.

19:45:00 <jonb> right.

19:45:02 <sandro> ... (and for what time-range they claim it.)

19:45:12 <bijan> So, if you relax various combinations of these constraints, you avoid paradox. But you have to eliminate *something*.

19:45:16 <jhendler> However, what I can live with is that we can hide all these rdf:unasserted tags in the OWL namespace, and only those who care about this level of reasoning know they're there.

19:45:32 <jonb> and OWL could care less about the 'fact' that _:L1 owl:first <#student> .

19:46:05 <jonb> sandro: these owl:List thingies aren't intended to be facts

19:47:00 <jonb> nor is [this owl:include <http://example.org/myOnt.owl>]

19:47:14 <AaronSw> why isn't that a fact?

19:47:14 <sandro> and yet they are..... there IS something which has a first element which is some-student.

19:47:24 <bijan> Jonb, wouldn't it be better to say that webont uses some triples for their *rdf* meaning, and some not?

19:47:37 <jonb> that's the point, OWL doesn't care which is first or rest

19:47:46 <jonb> bijan: exactly

19:48:07 <jonb> asserted triples are given meaning by the RDF MT, and unasserted triples by the OWL MT

19:48:17 <sandro> The problem as far a I can tell comes from the RDF inference of subgraphing. It's a real pain to have parts of your logic sentences disappearing on you.

19:48:20 <dajobe> isn't that pat hayes' solution in (hold on) ...

19:48:40 <jonb> so in order for OWL to have something to layer on, it needs some unasserted triples to give meaning to

19:48:48 <bijan> sandro, what do you mean?

19:49:02 <dajobe> in ... http://lists.w3.org/Archives/Public/www-webont-wg/2002Apr/0290.html

19:49:18 <dajobe> [[3. OWL introduces a special class name owl:Dark (which can be an

19:49:18 <dajobe> rdfs:Class). Any assertion of the form

19:49:18 <dajobe> AAA rdf:type owl:Dark

19:49:20 <dajobe> ]]

19:49:33 <jonb> sandro: tarski creates the 'liar's paradox' from any 'completely specified' language

19:50:28 <bijan> (note that patH has claimed that RDF MT permits dark triples...and I believe him, fwiw...thus it's unclear what the problem with inferencing would be)

19:50:35 <bijan> (If you mean *rdf* inference)

19:50:45 <sandro> That's why I'm trying to do PTL -- to completely specifiy a language, and see where the paradox might be. (I've been assuming if I can find a model [using MACE, probably], people will give up trying.)

19:50:48 <jonb> right, bijan, there is no problem w.r.t RDF

19:51:06 <jonb> sandro: are you the modern day Hilbert?

19:51:21 <sandro> I rather doubt it.

19:51:35 <sandro> I don't see how one can construct a paradox with negation, mostly.

19:51:53 <jonb> can you prove that?

19:52:01 <bijan> Er.. you mean without?

19:52:06 <jonb> i don't just mean fail to find it.

19:52:12 <sandro> If I found a model, would that be proof?

19:52:19 <sandro> yes, I meant without. sorry.

19:52:39 <jonb> does owl:complementOf, owl:disjointWith etc. count?

19:52:45 <bijan> Tarksi's conditions are: FOPL and containing one's own (complete, unambiguous) truth predicate.

19:52:52 <bijan> (I think that's all you need)

19:53:33 <sandro> Yes, those do count as negation. Anything that lets you make a contradiction.

19:53:51 <jonb> well daml+oil/owl has these...

19:54:08 <bijan> If you drop certain things you can avoid the liar paradoxes (i.e., regain consistency) but at the price of expressivity.

19:54:14 <sandro> Right. I'm not saying DAML+OIL or OWL can work with RDF as is -- I am saying PTL can.

19:54:16 <jonb> right

19:54:38 <jonb> ah, ok. so we need to see how useful it is for solving problems.

19:54:49 <bijan> jonb, yes.

19:54:52 <sandro> Right -- that's something people have been trying with cwm for a while now.

19:55:03 <jonb> in the mean time, if unasserted triples allow me to solve problems, then i am all for it.

19:55:17 <bijan> Will folks want to trade off *that* expressivity to gain consistency, or some *other* bit of expressivity.

19:55:20 <jonb> sorry sandro, cwm has contexts

19:55:39 <sandro> No more than PTL does.

19:55:49 <sandro> They're just not well defined.

19:55:53 <jonb> err, then what are we talking about?

19:56:15 <timbl> Not well defined?

19:56:28 <timbl> cwm doesn't have not.

19:56:50 <jonb> cwm has formula, contexts, and hence unasserted triples

19:56:54 <sandro> We need to stipulate a mapping from n3 (as cwm uses it, with log:forSome and {} and everything) into RDF. PTL is that mapping.

19:57:09 <sandro> But pre PTL, that mapping is not well defined.

19:57:15 <bijan> (doesn't have not as a free floating sentential operator of general scope)

19:57:31 <timbl> cwm has unasserted triples. I don't know whetehr it has "dark triples" as I don't know what they are.

19:57:44 <bijan> dark triples == unasserted triples

19:58:02 <Seth> what is PTL ?

19:58:09 <timbl> sandro, so you want to basically make a --flat that works?

19:58:17 <sandro> Yes, Tim.

19:58:34 <sandro> One whose output can be read and used again.

19:59:09 <sandro> AND which never infers more when you drop a triple or URI (that is, when you perform RDF inference).

19:59:55 <sandro> Seth, PTL is http://www.w3.org/2002/05/positive-triples

19:59:55 <sandro> [ which is unfinished, alas! ]

20:00:39 <sandro> I want this, Tim, so that people can talk about cwm logic *without* talking about n3.

20:01:57 <AaronSw> formulas could be written out using a fixed sort of reification

20:02:08 <jonb> timbl: "dark triple" is just another name for unasserted triple

20:02:22 <jonb> reification is apparently a no go with the new MT

20:02:46 <jonb> unasserted triples would replace that use for reification

20:02:56 <bijan> Aaron: see the mcdermott paper.

20:04:08 <sandro> I don't see how the MT can stop me from doing my own kind of reification.

20:05:10 <bijan> jonb, er...what about: http://www.w3.org/TR/rdf-mt/#Reif?

20:05:36 <jonb> pat h. tells me this

20:06:00 <jonb> reification is a legacy concept

20:06:11 <bijan> PatH doesn't tell me anything, save through specs :)

20:06:46 <jonb> bijan: i don't pretend to understand this fully

20:07:02 <bijan> Admittedly, reification in that section isn't particularly *interesting* :)

20:07:13 <jonb> that's the point

20:07:18 <jhendler> JimH rereads PTL, but doesn't see where reification is mentioned

20:07:52 <timbl> I feel that reification should not be in RDF core.

20:08:09 * jhendler learns to tell difference between sandro and aaron - oops.

20:08:23 <sandro> I have longer hair, Jim.

20:08:25 <timbl> For one thing, it isn't used by many important things which build on core, inc. RDFS and damlplusoil.

20:08:26 <bijan> That is a widespread feeling tim :)

20:08:29 <AaronSw> I don't have a beard.

20:08:40 <timbl> Secondly, the reification you use is in fact a bit arbitrary.

20:08:54 <timbl> This also applies for reifying N3.

20:09:02 <jonb> agree about reification, which is why we would like a workable and simple way to have an unasserted triple

20:09:18 <sandro> What IS an unasserted triple?

20:09:26 <bijan> Reification can be used to (clumsily) do that for some things.

20:09:42 <jonb> a triple which doesn't itself assert a truth value

20:09:50 <jonb> i.e. it is syntax

20:09:50 <timbl> I can chose to reify a formula as a bnode which is the formula whose statementlist is the following first/rest list of statements.

20:09:55 <bijan> "Bijan does not assert the following triple: :a :b :c."

20:10:00 <bijan> The triple in there is unasserted :)

20:10:12 <timbl> Sandro, in a language, I take it as any triple whcih isn't in eth ortermost context.

20:10:27 <sandro> PTL is about relating the truth of some RDF triples to the truth of other RDF triples. Those RDF triples are identified by giving their subject/predicate/object -- that's reificaiton, I think.

20:10:29 <bijan> Not what you wanted, of course, sandro, but sometimes an example is more fun to supply :)

20:10:49 <jhendler> timbl - isn't in the outermost context or otherwise asserted by some other mechanism

20:10:57 <timbl> :sandro :wrote { :sky :color :blue}. has 1 asserted and one unasserted triple in N3.

20:11:00 <jonb> bijan, that is tarski's liar's paradox in sheep's clothing, eh?

20:11:21 <bijan> No, jonb.

20:11:33 <bijan> Actually, its just true :0

20:12:00 <jonb> what about: this a log:falsehood .

20:12:00 <bijan> My asserting that there are no assertions is a bit closer.

20:12:02 <jhendler> :jimhlogic :asserts { :sky :color :blue} should be allowed, no? (would assert the inside triple)

20:12:07 <bijan> That is, yes.

20:12:25 <timbl> jhendler, no not from the language POV. In other words you can infer triples as being true from the context and orteh stuff but the darknes I assume is a language thing - they are dark in a given expression.

20:12:31 <bijan> jhendler, actually, I would say that *doesn't* assert the inside triple, on standard readings.

20:12:33 <jonb> oh, i see the difference... sorry

20:13:21 <bijan> There are standard pragramtics which permit the implicature from my assertion of "I assert the following three things" to an assertion of those three things.

20:13:25 <timbl> jonb, though log:Falsehood isn't a term i'm actually using, if it was a part of the logic still your statement is nota problem unless yo assume everything has to be either true or false.

20:13:27 <sandro> And what's wrong with saying :jimhlogic :asserts _:x. _:x rdf:subject :sky; rdf:predicate :color; rdf:object :blue. Doesn't that give us the same thing as unasserted triples, with no special features?

20:13:42 <jhendler> yes, okay, I see the point -- agree the way Tim puts it - but the MT view could sometimes have those inside things assertable - but would rather not.

20:13:54 <jhendler> (err, I mean I would rather it not)

20:13:55 <timbl> sandro, yes, exactly.

20:14:04 <bijan> That doesn't actually give us an unasserted triple, sandro.

20:14:20 <bijan> It gives us a bunch of asserted triples.

20:14:25 <sandro> Oh, Bijan? THen what would you call _:x?

20:14:31 <Guha_> JimH --- if you allow that :jimhlogic :asserts { :sky :color :blue} asserts the inside triple, and further allow for a statement that all the statements a certain agent makes are true, then you have your paradox

20:14:32 <sandro> _:x denotes an unasserted triple.

20:14:56 <bijan> sure, but there is no unasseted triple in the formula/graph.

20:15:04 <sandro> So?

20:15:14 <timbl> Dark triples are language things. Sandro's formula has 4 asserted triples and the same semantics as the formula with the dark triple, because of the meaning he gives to erf:predicate etc.

20:15:25 <bijan> That's where they are needed :)

20:15:27 <Seth> graph1 formulae {:sandro :wrote { :sky :color :blue}.}. graph2 formulae {:wrote rdfs:domain :person; rdfs:range :document}.

20:15:42 <bijan> And the point is you *still* don't haven't given an unasserted triple.

20:15:51 <Seth> timbl, is the graph2 asserted triples in the graph1 ?

20:15:56 <sandro> Why are they needed in the syntax, if we have them in the semantics?

20:16:02 <bijan> You've *talked* about an unasserted triple...maybe.

20:16:29 <sandro> You haven't done any more than talk about it either.

20:16:40 <timbl> seth, don't understand the question.

20:17:13 <bijan> :bijan :sez {:sky :is :blue}.

20:17:25 <bijan> I've certainly done more than talked about it.

20:17:36 <bijan> I've uttered the unasseted triple (in some sense of utter)

20:18:12 <sandro> Utterance is irrelevant. Only assertion matters.

20:18:20 <bijan> Heh. Ok. Sure.

20:18:21 <Seth> timbl, if the definitions of the properties are in another graph (set of formula), how do they get asserted in graphs which use those properties?

20:19:01 <sandro> ... otherwise issues of beleive and provenance and change-over-time get exponentially more complex.

20:19:04 <timbl> Sandro, re "why do we need the syntax", it is that the reification is ugly, and arbitrary in how it is done, while the formula which it allows you to express is more fundamental.

20:20:10 <sandro> The syntax is just as arbitrary, and much more ugly for RDF processors....

20:20:13 <Guha_> tim, do you consider dark triples as reification done more cleanly?

20:20:42 <sandro> I'm fine with { ... } for human use, of course.

20:20:50 <timbl> (Seth, your graph2 doesn't define :wrote. It just says a little about it. I meant that a system doesn't understand it until it has a whole lot of new axioms which follow from Sandro's reification and Guha's truth model that you belive tehings sandro :wrote. )

20:21:32 <bijan> Guha: I should point out that DTs, themselves, don't give you reification.

20:21:43 <timbl> Guha, I don't know enough about dark triples as seen on openhealth.org etc to answer how things are cleaner... must do.

20:21:44 <bijan> They allow you to build what is, IMHO, a cleaner reification.

20:22:35 <Guha_> Ok, glad to hear that I am not the only one who doesn't understand DTs.

20:22:38 <bijan> (Or, perhaps better, to achieve what is generally claimed for M&S reification)

20:22:45 <Seth> timbl, ok fine, but how does the triple {:wrote rdfs:domain :person} get asserted in the graph {:sandro :wrote {...}} ?

20:22:54 <Guha_> BTW, a good paper to read on all this : http://www-formal.stanford.edu/jmc/concepts.html

20:23:17 <bijan> Indeed.

20:23:53 <Guha_> Summary of that paper : Using a reification like construct to do propositional-attitudes

20:24:06 <bijan> Guha_, DTs are just triples which are present in the Graph, but are not asserted. I.e., they have no RDF sanctioned meaning.

20:24:19 <bijan> As such, they are "just syntax".

20:24:28 <timbl> I heard that the layering problem is that "If all triples are asserted then all deductions licenced under RDF can be made under the RDF MT, and so adding more terms can't extend them", which seems to me to be flawed. Because RDF should be defined to include the sepc of any properties used as predicates, so that to be strict, every time you meet a new property, you may have to extend your model theory.

20:25:01 <danbri> re reification being 'in' rdfcore; we present the reification vocab as part of RDFS now, but pretty low key. I think that's OK

20:25:21 <danbri> even though I think it's useless for intended purpose without being augmented in fiddly ways...

20:25:21 <Guha_> Bijan, thats where I don't understand them --- there is no such thing as "asserted vs not-asserted" in model theory.

20:25:45 <bijan> timbl, in so far as I followed what you said you heard, I don't think that that is the layering problem.

20:25:52 <Guha_> In that sense, a DT in the graph is just the same as all random permutations and combinations of triples that a computer may generate

20:26:01 <sandro> When when you query (the deductive closure of) a graph with DTs in it, what do you get?

20:26:12 <bijan> I don't follow that, Guha_

20:26:22 <jonb> i don't agree that that is a summary of the layering problem, rather:

20:26:23 <jhendler> Guha, you lost me there.

20:26:25 <Guha_> That is the crucual point.

20:26:41 <jonb> the current daml+oil mt, combined with the current rdf mt leads to paradoxes.

20:26:44 <Guha_> Lets say you have a DT p(a, y). i.e., it shows up in the graph, but not in the models

20:26:54 * DanCon wants to join the discussion if it's about mime types for RDF/WebOnt, but not if it's about dark triples.

20:27:05 <bijan> IT's about DT :)

20:27:23 <bijan> Sorry, DanCon.

20:27:32 <Guha_> You want p(a, y) to have some special significance not accorded to random p(x, y) pairs generated by a program for the heck of it.

20:27:43 <Guha_> Right? In which case, you have to tell me what that difference is.

20:28:02 <sandro> ... and how is that difference going to show up in a user interface?

20:28:10 <DanCon> actually, jonb, the current DAML+OIL MT comined with current RDF MT doesn't lead to paradox; it just doesn't lead to very many interesting conclusions. it's the strengthened DAML MT that would lead to paradox.

20:28:13 <sandro> ... or in program behavior?

20:28:20 <Guha_> we are not talking about those yet ;-)

20:28:33 <bijan> Yes, that's what gets pushed to subsequent "layered" specs.

20:28:48 <Guha_> Sorry, don't understand that.

20:28:48 <jonb> yes dancon

20:28:49 <sandro> (I'm afraid if we don't Guha, the answer will be too fluffy.)

20:29:09 <Guha_> not fluffy ---- just mathematical.

20:29:11 * timbl proposes "+rdf" a la "+xml" for mime types.

20:29:14 <bijan> So, pure DTs, sure. There is no RDF meaning assigned. They are, so to speak, "Dark".

20:29:31 * DanCon hopes +rdf was a joke

20:29:38 <Guha_> Look, the MT folks want to be mathematical and have DTs, fine.. But they need to tell me what a DT is.

20:29:40 * sandro worries.

20:29:58 <bijan> The reason to add DT is not to have random dark triples, but so taht specific applications (e.g., WEbOnt) can have some triples for which they can specify the meaning.

20:30:02 <jonb> a DT is a language statement, a formula

20:30:07 <Guha_> In terms of math. As far as I can make out, all they are saying is that they want to drop one or two assertions made by RDF/RDFS.

20:30:27 <jonb> a DT has no meaning except that given by a layered MT

20:30:40 <bijan> A triple which is dark to RDF need not be dark to, e.g., WebOnt.

20:31:03 <jonb> right, webont may use DTs to create sets out of owl:Lists

20:31:16 <jhendler> Guha - approximate (but not perfect) analogy - DT's provide a semantic extension mechanism the way the old "parsetype" permitted a syntactic extension mechanism

20:31:23 <Guha_> Which is fine. Why add something to RDF/RDFS to remove something from it.

20:31:38 <bijan> Why not?

20:31:54 <sandro> Why not make all triples dark, until lit? That seems much cleaner.

20:31:54 <bijan> Adding *just the right* holes to things is a tricky business!

20:32:00 <jhendler> Guha - because the unextended meaning and the extended meaning may conflict

20:32:01 <jonb> because the default is to assert.

20:32:05 <bijan> Ask swiss cheese makers :)

20:32:11 <Guha_> Jimh: intuitively, I understand the motivation ... I am just failing to map it into tarski's language

20:32:24 <jonb> it would be fine to make the default unasserted

20:33:04 * Seth thinks that DT are irrelivant to an api that has multiple graphs ...{...} ... {...} ... goes back to coding

20:33:15 <bijan> Hmm. Maybe this analogy would be helpful, Guha_. Though I'm making up off the top of my head.

20:33:23 <Guha_> JimH --- then, in order to keep things simple and have a hope in hell of explaining things to developers, we should consider removing the offending statements from RDF/RDFS.

20:33:25 <jonb> guha: what about tarksi's "The inconsistency of semantically closed languages" ?

20:33:39 <bijan> Take a standard represenation of First Order Propositional calc.

20:33:48 <Guha_> And if we decide we can't do that, *resolve* the tension between the two.

20:34:01 <bijan> Where atomic sentences are represented by capital letters, e.g., A, B, C...

20:34:22 <Guha_> The two meaning OWL wanting classes to note denote and rdfs wanting them to denote

20:34:56 <jhendler> Guha - the idea as I see it is this -- have RDF have a mechanism it allows, but doesn't use in it's namespace called RDF:unasserted. The MT knows how to handle it

20:35:11 <bijan> Now imagine that instead of A B C we used rather more complext glyphs. Say, Aa, Bab, or even (x)Ax

20:35:52 <bijan> From the POV of our FOPropCalc, the extra structure of these "glyphs" is dark.

20:35:59 <jhendler> Then OWL is able to have in its namespace "owl:list a rdf:unasserted" this is an elegant solution for those who want to see minimal user impact, because the average person needs to know about neither of them. The implementor of a reasoning system, will need to go to the "expert" part of the documents and learn about this.

20:36:15 <bijan> But, we can layer a predicate calc on top of our propositional mt.

20:36:40 <jonb> or as pat suggests, owl could otherwise do: owl:list a owl:unasserted

20:36:47 <bijan> Which fills in the meaning for those bits of syntax.

20:36:55 * bijan hopes the logger enjoyed that ;)

20:37:11 <sandro> But Jim, RDF inference lets someone drop those triples, right? So the unassertedness has to go into RDF syntax.

20:37:16 <Guha_> what about solving the root of this problem?

20:37:26 <bijan> Switching rdf for kif?

20:37:27 <bijan> :)

20:37:51 * bijan totally joking. Please don't hurt me.

20:37:55 <jonb> allowing unasserted triples does solve the root of the problem

20:38:09 <Guha_> No, that is a solution for the tension ...

20:38:12 <jonb> s/the problem/a problem/ :-)

20:38:31 <Guha_> the root of the problem is that owl and rdfs differ on an important and fundamental aspect of ontology architecture.

20:38:33 <jhendler> sandro - don't understand. The unasserted property is legal RDF, not used in a visible way. The owl uses this rdf:unasserted syntax when it needs to

20:38:50 <Guha_> We can come up with clever hacks to get MT consistency, but the difference remains.

20:39:01 <bijan> Guha_, wait, before solivng the root of the problem, are you satisfied that DTs are intellible, model theoretcially speaking?

20:39:04 <sandro> So owl needs access to the RDF source text, not just the n-triples that come out of the parser....?

20:39:04 <jonb> not a hack

20:39:12 <Guha_> From an architecture perspective, it would be much cleaner to resolve this tension.

20:39:26 <jonb> sandro: no

20:39:33 <jhendler> sandro - no, in this scheme - that's what I like about it

20:39:37 <danbri> owl needs the xml form? ugh....

20:39:38 <jhendler> Guha - agreed, but how?

20:39:42 <Guha_> Bijan, I don't understand the model theory of it yet. I don't even know if it should be reflected in the model theory.

20:39:46 <bijan> I'm not clear what you, Guha_, think the "tension" is, exactly.

20:39:51 <jonb> danbri: no, that's the point

20:39:57 <sandro> But it needs ALL the triples -- if one of the triples gets dropped (a valid RDF inference) -- then OWL would get things wrong.

20:40:27 <Guha_> Lets get into a room and decide once and for all, whether on the semantic web, Classes should denote. Either they do or don't. We should not have different answers for rdfs and owl

20:40:31 <jhendler> danbri - no - the graph looks the same, but the owl:lists that are used now are linked to rdf:unasserted through the namespace - usable by the reasoner.

20:40:49 <bijan> Er...classes denote?

20:40:57 <Guha_> in rdfs they do.

20:40:58 <jonb> guha: in owl classes can denote

20:41:12 <bijan> I need that description connected up to the more or less standard discussion of the layering problem.

20:41:27 <bijan> yes, especially in light of my agreement with jonb :)

20:41:28 <jonb> that's not the tension, just possibly owl:list, owl:first, owl:rest, owl:nil -- the syntax stuff

20:41:36 <jhendler> Guha - that is the heart of this. But I suspect we need to wait for the SWAG (Sem Web Architecture Group) to come along and fix this someday :->

20:42:07 <bijan> (And classes don't denote in RDFS, do they? They have instances and subclasses...)

20:42:13 <jonb> sandro: what do you mean 'needs all the triples' ?

20:42:25 <Guha_> Ok, I am not being precise ... in rdfs, everythign denotes, type is just another property, etc. ... the accumulation of these leads to the problem with the daml+oil mt.

20:42:25 <danbri> a class is a thing in the domain of discourse, for rdfs, no?

20:42:30 <jhendler> jonb - unfortunately, that's not resolved PFPS argues virtually everything in OWL might have to be dark (i.e. he wants these not to denote) - but unclear he'll win this argument.

20:42:33 <Guha_> bijan --- yes, classes denote in rdfs.

20:42:38 <danbri> DL folk don't like that.

20:42:43 <bijan> (Classes aren't linguistic items in RDFS, are they? I thought URI(refs) and literals denoted)

20:42:45 <jonb> jhendler: yes but pat thinks differently

20:43:03 <bijan> Guha, please point me to the part of the spec (either the CR or the current WD) where this is said.

20:43:14 <bijan> And what do they denote?

20:43:25 <sandro> never mind, jonb. :)

20:43:38 <jonb> in any case in the RDF MT, there is a set of asserted triples, and a set of triples

20:43:46 <jhendler> jonb - yes. was just saying why I agree w/Guha - at heart this denote vs. non-dentoe is the heart of DL vs. non-DL semantics. I'm agnostic on the issue, Guha is in the anti-DL camp, pfps is in the pro-DL

20:44:16 <bijan> Wow. I'm off the map. I didn't think denotation entered the picture!

20:44:25 <jonb> right, but the unasserted triples idea is apropos regardless of DL vs not.

20:44:48 <jhendler> personally, I can live in either world because the semantics I care about is the simpler stuff like "points to the same URI" :->

20:44:51 <danbri> where in spec -- maybe not explicitly said. but the rdf approach, using the normal rdf machinery to talk about vocabularies, puts Classes and Properties and rdf:Statements and wn:Cats and xyz:Dogs on a par. They're all types of thing in the world.

20:45:12 <jonb> ok danbri, but owl:List?

20:45:21 <danbri> yeah, that's the problem.

20:45:33 <jonb> that's like requiring a triple ex:foo rdf:ID "myObject" .

20:45:55 <danbri> rdf:ID being a property?

20:46:04 <Guha_> bijan --- section 3.3 of the rdf mt

20:46:05 <danbri> Not such a bad idea...

20:46:14 <jonb> well, RDF has tokens that it uses for syntax, eh?

20:46:36 <Guha_> jimh : I agree. Either world will be ok.

20:47:12 <jonb> danbri likes an anonymous world, interesting idea

20:47:25 <bijan> Guha_, thanks! do you mean:

20:47:27 <bijan> """Although not strictly necessary, it is convenient to state the RDFS semantics in terms of a new semantic construct, a 'class', i.e. a resource which represents a set of things in the universe which have the same value of the rdf:type property. """

20:47:49 <Guha_> Jimh --- clarification. I've been thinking about it a lot and I've decided I am not anti-DL or pro-DL. I am just pro-kiss

20:48:04 <Guha_> bijan: yes. a class is a resource.

20:48:22 <jonb> yes, but the saying is "as simple as possible"

20:48:30 <bijan> I will have to reflect on that, but my intital thought is that this isn't denotation from the MT perspective.

20:48:45 <bijan> But I'm not clear yet.

20:48:47 <DanCon> nope; there's no problem with owl:Lists being like cats, dogs, and properties. At least: not a problem that I see.

20:49:06 <jonb> not a problem necessarily or always

20:49:38 <bijan> (The key is that it's often unwise to *both* have them like cats, dogs, etc. *and* have special status in the logic)

20:49:38 <jonb> but we want to say (a b c ... n) == (n a b c ... n-1)

20:50:27 <bijan> Eek! Gotta run. Guha_, I would like to take this up again sometime.

20:50:35 <jonb> so the intended _meaning_ of these lists for OWL is not that they are ordered. why assert foo owl:first bar?

20:50:39 <bijan> Ta all.

20:50:44 <jonb> adios

20:50:49 <Guha_> bye

20:51:07 <jhendler> Guha - pro kiss makes sense.

20:51:21 <Guha_> so, why don't we make it so?

20:51:38 <jhendler> I'd label myself as "pro-inclusive" that is, I want something all the communities can live with.

20:51:50 <jonb> this a rdf:closedIssue

20:51:52 <jhendler> Ok, wave my magic wand to "make it so" -- would love a process suggestion

20:52:10 <danbri> Guha, do you consider DLs simple?

20:52:57 * danbri can see it both ways; they inhabit a simple minded universe (but implementing them could be pretty complex)

20:53:02 <jonb> to be really simple why not just allow negation, its a very simple concept, even a two year old can understand

20:53:05 <danbri> esp in the web

20:53:09 <danbri> quite

20:54:00 <jonb> In order to simplify the inner workings of your MT, you are greatly complicating the life of simple minded users, why not just allow logic as learned in high school?

20:54:39 <jonb> e.g. (this and not(that) or if(that-other-thing) then whatever)

20:54:51 <jonb> what could be simpler that if-then?

20:55:09 <danbri> Or play them gramaphone recordings of Russell and Quine while they're in the womb?

20:55:48 * jonb wonders if his kids understand: "if (you hit your sister) then (timeout)" ?

20:55:48 <sandro> (2-year-old is more right than high school -- my 2 year old can interpret fairly complex boolean expressions.)

20:55:57 <Guha_> jimh: a similar issue came up in June '97 on the issue of whether RDF should have a model theory or not. Ralph got the main parties together (me, tim bray, jean paoli, eric miller, andrew layman, eve maler and ora lassila) into a room in cambridge and wouldn't let us go back until we all agreed.

20:56:10 <Guha_> Then we took that to the working group and all was well

20:56:33 <Guha_> correction --- I meant to write data model, not model theory

20:56:42 <sandro> "If you or your brother does this again, without specific permission, then ... " is understood just fine.

20:57:04 <jonb> understood and ignored in my case

20:57:28 <sandro> :-)

20:57:31 <jhendler> sandro - it's the interpretation that's the problem, not the understanding (but you never said I couldn't hit her when I WAS NOT in my room) :->

20:57:59 <sandro> :-)

20:58:31 <jhendler> Guha - as a chair I'd love to see it happen - problem is we have something that seems to fall in the cracks between WGs, and there is no obvious place to resolve it - the OWL only solution is the one that is least "layered" and some people seem not to like that.

20:59:45 <jhendler> a suggestion to create a joint task force was vetoed in the CG, so we seem to be at the mercy of either an infinite argument or just taking some solution now and revisiting in RDF 3.0/Owl 2.0 if we got it wrong.

21:00:13 <jhendler> personally I'd just like to find soem cats to herd - would be easier...

21:00:15 <danbri> Or not getting to OWL 1.0 if it doesn't layer on RDF...

21:00:21 <danbri> whatever layer on means...

21:00:27 <jonb> how would people feel about [owl:first a owl:Unasserted] ?

21:00:51 <jonb> similar to (this a log:falsehood :-))

21:00:58 <timbl> timbl is now known as tim-away

21:02:13 <jonb> we might not get to OWL 1.0 if the issue goes away, either because everyone will continue to argue, or everyone will quit and there will be no one left to write the docs...

21:02:38 <jonb> s/goes away/doesn't go away/

21:03:33 <DanCon> any other efax users out there? my efax account seems to have died, and all the accounting features that used to be part of the free accounts are now only available to paying customers.

21:03:45 <jonb> i genuinely think that when there is so much controversy, there must be some reality

21:04:14 <DanCon> timbl convinced me that the solipsistic approach is acceptable, btw.

21:04:44 <jonb> to whom? it looks too weak to do the stuff i want to do, btw.

21:04:57 <DanCon> to me

21:05:47 <jonb> i can't imagine using a solipsitic system to allow me to classify any real patient (individual) into any diagnosis(class) that is at all complicated

21:06:50 <DanCon> no? it doesn't matter how complicated the class is, as long as you're willing to state the existence of the class, rather than expecting the system to infer it.

21:07:46 <jonb> yes but i want to be able to infer "intersectionOf(A B) => intersectionOf(B A)" right?

21:08:32 <jonb> or am I missing something?

21:08:34 <DanCon> you might want lots of stuff. but you might be able to do your job without it

21:08:44 <jonb> can I?

21:08:49 <DanCon> dunno.

21:09:04 <jonb> that's an honest question.

21:09:45 <jhendler> DanCon - I thought the solipcistic stuff was still considered researchy - Jermey withdrew it from consideration because he became convinced it had holes. It may hold hope for a future solution, but I'd like something practical for the short term so the many D+O users can know what to expect...

21:09:53 <DanCon> yes, if you've said that intersection(A, B) and intersection(B, A) exist, and you've said x is in one of them, a solipsistic approach can conclude x is in the other.

21:10:21 <jonb> yes but suppose "intersectionOf(a1 ... a100)"

21:10:25 <DanCon> no, solipsistic wasn't researchy; it was dead simple: the system doesn't infer the existence of any classes. you have to mention all the relevant ones.

21:10:41 <Guha_> YES YES YES!

21:11:02 <jonb> do I need to explicitly mention intersectionOf(a1 a100 a2 ... a99) etc etc etc

21:11:03 <Guha_> Infering the existence of classes gets us into all kinds of problems.

21:11:34 <jonb> I want to say the class is the intersection of the following _set_ of classes

21:11:46 <DanCon> jonb, if you want any conclusions from a solipsistic system about those classes, yes, you have to mention them. It remains to be seen whether you can get your job done without that, though.

21:12:22 <jhendler> DanCon - problem was at f2f pfps derived a contradiction in Jeremy's formation of solipcistic and Jeremy was not convinced it could be solved easily and right away - I must admit that I couldn't follow the details, but following that discussion, Jeremy changed his tune.

21:12:38 <jonb> so is the equation, for each diagnosis that has n symptoms, i need to mention n-factorial classes?

21:12:57 <DanCon> no, pfps found problems with Jeremy's "comprehensive entailments", not with solipsistic.

21:13:11 <DanCon> the solipsistic answer to the "DAML+OIL semantics is to weak" issue is: tough.

21:13:23 <jonb> again, I want common sense to work here, and a good explanation of why I don't get common sense

21:13:40 <jhendler> jonb - possibly, but only "in theory" - for example many systems use numbers based on the conviction that they are well behaved without having the formal axioms of numbers needing to be entered - could be like that

21:14:10 <DanCon> I don't see any need to mention n-factorial classes. If you mention a diagnosis that has N symptoms, a solipsistic system can classify a patient that presents all of them.

21:14:15 <jonb> yes but that is basically saying, to heck with the MT, I will make the program work, argument

21:14:33 <jhendler> DanCon - at the f2f it was solipcism being discussed - but it may be the issues got blurred somehow - I just know the result - Jeremy said "never mind" specifically to his solipcism proposal - feeel free to contact him.

21:15:05 <DanCon> I don't have any need to contact him.

21:15:27 <jonb> dancon: can you assure me this? this is what i have asked, and all I get are very fidgety answers in the best of circumstances -- not something that gives me confidence

21:16:07 <Guha_> There are two reasons for MTs

21:16:20 <DanCon> assure... not offhand. I'd have to try it "in anger" for a while to see if it works out.

21:16:20 <Guha_> One is to mathematically constrain denotation

21:16:38 <Guha_> The other is to specify soundness/correctness of programs operating on the axioms.

21:16:44 <Guha_> We are not using it for the former.

21:16:56 <Guha_> For the latter, there are many many ways.

21:17:11 <jonb> dancon: really all i want is to be assured that i can write a system to make proper diagnoses given proper symptoms

21:17:22 <jonb> and a proper disease classification

21:17:48 <jonb> so "soundness/correctness" is very important

21:18:12 <Guha_> yes, but remember, soundness is defined *relative* to something.

21:18:30 <Guha_> MT is one way. FP semantics is another.

21:18:57 <Guha_> The MT approach will get into all kinds of problems when we start pushing the boundaries of expressiveness

21:19:00 <jonb> sure, but you agree that RDF and OWL should use the same mechanism, whatever that be

21:19:26 <DanCon> it seems to me that for medical diagnosis, it's *very* unlikely that a machine can determine a "correct" result; what you want is a useful result. In particuar, you want to be able to encode the knowledge of experts so that the machine can simulate what they'd say.... *not* so that the machine can necessarily diagnose patients perfectly.

21:19:32 <Guha_> No, that is not clear. What is clear that they should agree

21:19:48 <DanCon> in short: it seems perfectly reasonable for a diagnostic-machine to say "I don't know'

21:20:27 <jhendler> Dan - are you saying Sound but not coplete is okay? If so, I wouldn't disagree.

21:20:35 <jonb> dancon: certainly, in some cases, but that would be malpractice if the diagnosis is staring you right in the face

21:20:41 <DanCon> yes, sound but not complete seems ok

21:20:48 <jonb> agreed.

21:21:24 <Guha_> DanCon : That is *inevitable*

21:21:34 <jonb> but to be useful, the degree of completeness needs to mirror the complexity of the classification, i.e. it is pointless to create complex classifications if only trivial classifications can be made

21:22:09 <Guha_> but be very careful. Before you know it, non-complete turns into not-sound (in the precense of NM)

21:22:21 <jonb> and medical ontologies are >400,000 terms

21:22:42 <DanCon> yes, jon, but the question is: do you need the system to infer the existence of unmentioned classes in order to do non-trivial classification? or does it suffice if the machine only reasons about classes that you mentioned? I don't see much evidence one way or the other.

21:23:34 <jonb> dancon: well i'd -love- for the system to be able to pick up a new pattern of a disease, e.g. an outbreak of bioterrorism etc.

21:23:47 <Guha_> jonb .. I take it you are refering to snomed. Well, every one of those terms is mentioned.

21:24:39 <jonb> yes but my question is: intersectionOf(A B) is that the same as mentioning "intersectionOf(B A)" to common sense it is

21:25:34 <jonb> and no one, except DanCon here, has been willing to tell me that solispsism will allow common sense inferences to be made

21:26:14 <jonb> but if it does work for snomed/medical diagnosis, then i really really have no objection

21:27:02 <jonb> (gosh i thought a room full of the world's experts on DAML+OIL would have been able to give me some assurance)

21:30:17 <AaronSw> semantic web is searched for on google 400 times a day

21:31:24 <jhendler> But Aarin - I don't type fast enough to bring that to 500!!

21:31:43 <DanCon> got any other stats for comparison, AaronSw?

21:32:01 <jonb> dancon: seriously though, there isn't a formal "solipsistic" argument on the table.

21:32:14 <AaronSw> britney spears is searched fro 92,400 times a day

21:32:20 <jhendler> dancon - chairing-wise, this is exactly the problem.

21:32:30 <AaronSw> argh, this site breaks the back button

21:32:35 <jhendler> Aaron - how about W3C or something like that

21:32:51 <AaronSw> w3c: 2200/day

21:32:56 <jonb> aaronsw is learning how the world really thinks

21:33:15 <DanCon> er... the "too weak" issue isn't open, is it, jim?

21:33:26 <jonb> the entire function of the SW will be to find the best photo's of britany spears ...

21:33:37 <AaronSw> connolly: 2400/day; berners-lee: 200

21:33:51 <jhendler> way to go Dan!

21:33:54 * DanCon wonders which connolly they're looking for

21:34:04 <AaronSw> .google connolly

21:34:05 <xena> connolly: http://www.w3.org/People/Connolly/

21:34:16 <jonb> there are 2400 connollys in boston alone, each with one search/day

21:34:53 <AaronSw> dan connolly: 0; jennifer connolly: 600

21:35:04 <jhendler> lol

21:36:33 * jonb wonders what the unspoken resistance to unasserted triples is

21:37:35 <DanCon> what's unspoken about it?

21:37:47 <jhendler> unspoken???

21:38:24 <jonb> it seems so logical, but yet there is resistance that i can't see

21:38:42 <DanCon> RDF is real simple: from <Book><title>abc</title><author>Fred</author></Book>, I can conclude <Book><author>Fred</author></Book>. Proposals for dark triples undermine that most basic principle.

21:39:09 <jonb> yes but reification was (an attempt) to have unasserted triples from day one

21:39:21 <DanCon> yes, but reification doesn't break the principle of erasure.

21:39:29 <jonb> ?

21:39:46 <DanCon> erasure: From "A and B" conclude A"

21:39:52 <DanCon> erasure: From "A and B" conclude "A"

21:40:12 <jonb> how do unasserted triples break that?

21:40:50 <DanCon> from <List><first>Employee</first><rest>...</res></List>, I'm not supposed to conclude <List><first>Employee</first></List>.

21:42:11 <jonb> well the first <List> ... </List> and the second are both unasserted so why does this break the principle of erasure?

21:42:44 <jonb> (not to say that we couldn;t have a list construct that was intended to be asserted also)

21:42:51 <DanCon> ???. if it's unasserted, how can it be the basis of a conclusion?

21:43:09 <DanCon> "dark" means "doesn't lead to any conclusions", i.e. "is semantically invisible", no?

21:43:10 <jonb> right but "A and B" is a premise

21:43:56 <jonb> and the owl:List would be semantically visible to OWL (under its MT which would provide an interpretation)

21:44:46 <jonb> under your example if "From -blank- conclude -blank-"

21:44:47 <DanCon> I can't make any sense of what you're saying. I understood "dark" to mean: can't be seen by RDF inference gizmos.

21:44:57 <jonb> right.

21:45:27 <DanCon> then how can my RDF inference gizmo see it, in order to conclude <List><first>Employee</first></List>?

21:45:38 <jonb> from what?

21:45:52 <DanCon> from <List><first>Employee</first><rest>...</rest></List>

21:46:13 <jonb> yeah but both are blank i.e. From -blank- conclude -blank-

21:46:22 <jonb> what is the problem with that?

21:46:32 <DanCon> no, <List><first>Employee</first></List> isn't blank.

21:46:41 <jonb> yes it would be.

21:46:44 <DanCon> ;lewit uveairha]

21:46:53 <DanCon> I'm telling you how my RDF inference gizmo works.

21:47:13 <DanCon> and the way it works is: given triples A and B, it's licensed to conclude B.

21:47:32 <DanCon> B being, in this case, _:something ont:first "Employee".

21:47:33 <jonb> but <owl:List> ... </owl:List> would always be blank to RDF, either as premise or as conclusion

21:47:48 <DanCon> if it's blank, how does it get into a conclusion?

21:47:53 <AaronSw> it doesn't!

21:47:58 <DanCon> that's my point.

21:47:59 <jonb> that's the point

21:48:05 <jonb> ok, so what is the problem?

21:48:10 <DanCon> ARGH!

21:48:15 <jonb> seriously.

21:48:19 <DanCon> the problem is: I can no longer conclude B from "A and B"

21:48:21 <AaronSw> if the premise "A and B" is false than the implication (erasure) must hold

21:48:34 <DanCon> g3JF'3AHrew

21:48:34 <jonb> sorry, your "A and B" doesn't apply

21:48:35 <AaronSw> well, it's not disproved at least

21:49:05 <DanCon> AaronSw, entailment isn't about judging premises independently of conclusions.

21:49:20 <AaronSw> on the other hand, i think Dan's point is that there's no equivalent erasure principle for dark triples

21:49:25 <jonb> when you say "A and B" I assume you are asserting A and you are asserting B

21:49:55 <jonb> well the OWL MT has to be written to work correctly ... yes that is true if OWL is braindead

21:50:16 <DanCon> let's start again, shall we? RDF is nothing if not the principle of erasure: that from "A and B", I can conclude "B"; i.e. in RDF, A and B are independent claims. saying them next to each other is to claim their conclusion, from which follows, e.g. B. if B is "dark", I can't do this any more.

21:51:34 <DanCon> (erasure is actually one of a few critical features, the others being literals, URIs, and existential introduction, i.e. blank nodes. But erasure is right up there near the top, if not at the top)

21:51:42 <jonb> why would you try to assert the truth of a piece of syntax?

21:51:49 <DanCon> cuz it's not syntax.

21:52:04 <jonb> unasserted triples are syntax

21:52:10 <jonb> just syntax

21:52:23 <DanCon> but in RDF, all triples are claims; none of them is syntax.

21:52:32 <jhendler> Dan - I've been lurking and reading your stuff carefully - I see the problem. But here is the situation as I see it:

21:52:42 <jonb> no, there is a set of asserted triples

21:52:47 <jonb> and a set of triples

21:53:17 <jhendler> first, if you come in w/an RDF reasoner that doesn't know anything about OWL, it is going to be able to conclude the inner thing.

21:53:25 <DanCon> this language (which is not RDF) which has asserted triples and other triples doesn't have the erasure principle.

21:53:38 <DanCon> "inner thing"?

21:53:41 <jonb> by this language, the MT WD?

21:53:54 <DanCon> ?

21:53:57 <jonb> or a private definition?

21:53:58 <jhendler> If you come in with an OWL aware tool, it will check for things inheriting from owl:dark and realize that it cannot conclude that.

21:54:15 <jhendler> therefore it introduces a discontinuity between RDF and OWL. Is that correct?

21:54:41 <jhendler> (sorry - I mean trying to sum up your DanC's argument)

21:54:46 <DanCon> I don't follow, jim. I suggest being much, much more concrete.

21:55:12 <jhendler> Dan - I was just trying to see if I understood the ramification of what you are saying.

21:55:34 <DanCon> well, I don't follow.

21:55:52 <jhendler> i.e. Both the RDF tool and the OWL tool work fine, but there's a fact that the OWL tool would not conclude that the RDF tool would, and that causes grief

21:56:02 <DanCon> "inheriting from owl:darl" <- can't make heads or tail of that.

21:56:33 <jhendler> sorry - that was shorthand for the owl:list a owl:unasserted (or RDF:unasserted, may not matter)

21:57:08 <jhendler> (can also be implemented as owl:first subclass owl:darkProperty )

21:58:13 <jonb> yes but if rdf:unasserted were used, the presumably an RDF aware tool would know this also, so _no_ discrepency

21:58:25 <DanCon> yes, it would cause grief if from { _:L ont:first "x". ont:first a ont:unasserted } I was not licensed to conclude { _:L ont:first "X" }. Cuz there's piles and piles of code out there that already comes to that conclusion. We could try to recall it, but I'd want to know why, first.

21:58:55 * jonb goes to search for 'weasel wording'

21:59:28 <DanCon> weasel wording was never agreed by the WG; WG recently decided against weasel wording; I think the most recent WD reflects this

21:59:44 <jonb> ah removed from current WD, at your request

22:00:26 <jonb> yes i see. in any case prior WD provided for set of asserted triples (and by implication set of unasserted triples)

22:00:42 <DanCon> but there was never any syntax for unasserted triples

22:01:12 <jonb> correct, and the current proposal only requires a class in the RDF namespace, no changes to RDF syntax itself

22:01:13 <jhendler> no that's not what I think would happen. that would be fine - it's that from <List><first>Employee</first><rest>...</rest></List> you wouldn't be able to conclude <list><first>employee</first><list>

22:01:26 <DanCon> (er... actually, the '98 spec had a magic namespace. WG thought about it and removed it, cuz nobody implemented it and nobody saw any need for it)

22:02:19 <jhendler> dan - the proposal is now to have there be no special syntax for unasserted triples, but to ask implementors of reasoners to know how to derive what was owl:dark

22:02:46 <jhendler> might be a lot like the magic namespace, unclear if anyone would ever use it, but there's folks claiming there's aneed for it

22:02:57 <jonb> yes and if RDF reasoners understood rdf:unasserted then everyone agrees

22:04:27 <DanCon> jim, you raise a good point: that folks would be expected to be aware of unstated (or: externally stated) "property P is dark" bits. which brings us back to 1st principles for RDF: it's syntactically evident what formula you're looking at, and in particular, you can conclude any subset of the triples from the ones there in the document.

22:04:57 <DanCon> (which 1st principles are broken by dark triples proposals)

22:05:59 <DanCon> note that reification doesn't break the principle of erasure: from "there's a statement whose subject is Mary, whose predicate is hit, and wholse object is theBall" it's safe to conclude "there's a statement whose subject is Mary".

22:06:00 <jonb> well you have the nested <rdf:RDF> explicit syntax

22:06:27 <DanCon> nested rdf:RDF gives us a different language; one I'm willing to discuss.

22:06:38 <jhendler> dancon - I don't think this is broken - because the graph woudl explicitely include the info, but you would need the whole graph (including what is in the owl:namespace) but that is always true (because a document might assert some rdf:type to something in a different document ro the like)

22:07:16 <jhendler> jonb - let's not go there, I'll quit before I would accept the nested syntax - I'm not going to waste my time creating a language no one could (or would) use...

22:07:24 <DanCon> no, Jim, it's *not* always true that you need the whole graph, including bits from schemas. This is the big feature of RDF: the principle of erasure works WITHOUT ACCESS TO SCHEMAS.

22:07:38 <jhendler> i.e. the rules on when to use the <rdf:rdf> stuff would be a pain...

22:08:26 <jonb> hmmm perhaps the <rdf:RDF> element might contain an attribute for parsers e.g. <rdf:RDF rdf:unasserted="owl:List owl:first">

22:09:04 <jonb> or perhaps <rdf:RDF rdf:unassertedNS="owl"> ... something like that...

22:09:07 <jhendler> yes - Dan I understand - but I was probing this before - in the owl only solution, each tool (RDF tool; OWL tool) would have principle of erasure of it's own, but there'd be a problem that when an RDF tool read an OWL document it would make some unwaranted assumptions from an OWL viewpoint

22:09:14 <DanCon> note that [rdf:unasserted attr] would be ignored by existing RDF tools, jonb. i.e. you're talking about a whole new language, not layering on RDF.

22:09:26 <jhendler> and I'm trying to see if that is what you don't like.

22:09:49 <DanCon> yes, I consider it unacceptable for owl to license fewer conclusions than RDF does.

22:10:05 <jonb> dancon: the request is for the current RDFCore WD, note that I might consider banning the unqualified "about" or "ID" atts as a new language

22:10:07 <DanCon> from the same document.

22:10:50 <jonb> or perhaps OWL can layer on RDF M&S which means writing its own MT from the ground up...

22:11:28 <jhendler> right - that is what I thought - so you are looking for a solution where everything RDF would conclude, OWL would conclude as well, but OWL could have things RDF doesn't. What would you be willing to give up for that? ANything?

22:11:46 <DanCon> yes, well, the WG saw motivation for about/ID change: consistency with XML stuff. (XPath, XSLT, etc.). Perhaps there's an analagous argument for WebOnt, but I don't see it. I don't see how WebOnt is going to use unassertedness in a way that's going to help anything.

22:12:16 <jhendler> that is, if the DT folks are right and it is impossible to have cake and eat it too, is there a way out. (let's play the game for a minute of supposing it is impossible)

22:12:54 <DanCon> no, my technical position is that there' nothing I'd trade, Jim. (I'm duty-bound to trundle along even if I don't get my way, technically, but I won't like it)

22:14:02 <jhendler> ok, interesting point, my chair training is to look for consensus/compromises but there is the alternative of saying "I don't like it, but we'll fix it in the next version when you guys realize I was right"

22:14:55 <jhendler> assuming we'll make all the right decisions first time out of the box is something WOWG has trouble giving up... (probably true of all first working groups for a language)

22:16:18 <DanCon> the first "dark triples" I saw were things like log:forAll. i.e. I've been that way, as an implementor. There lies madness... well, technically: unsoundness.

22:16:42 <DanCon> we haven't fixed this bug in N3, but it's acknowledged as a bug.

22:16:55 * jhendler foo - just when I thought I was actually going to have a couple of days to catch up, I receive a 13M, 155 p. word document I need to edit by Friday - foo.

22:17:47 <jhendler> dancon - was wondering about how you dealt with the contexts in N3 - dont they cause these same sort of erasure problems when mapped to RDF, or does the python converter do lots and lots of reification to get around it?

22:19:03 <DanCon> N3 {} aren't expressible in RDF.

22:19:23 <DanCon> i.e. the converter spits out nonsense.

22:19:34 <DanCon> another known bug

22:20:04 <DanCon> i.e. another reason we say that the N3 language defintion is in the tests, not the code. There are no tests for converting {} to RDF.

22:21:25 <jhendler> so if you want to fix this, won't you need something akin to a DT solution? To put it another way - the only argument I see for rdf:unasserted, instead of owl:unasserted, is because we will someday need rules:unasserted; logic:unasserted; proofs:unasserted, etc. If RDF is to be the bottom of the SW, it will need some sort of extension mechanism

22:22:33 <DanCon> umm... yes, I need something that's like DT in that it's not expressible in current RDF; i.e. we'll need a new language. But it won't be like DT in that it won't pretend to be RDF triple syntax.

22:23:04 <jhendler> personally, I can handle having an OWL only solution until something else comes along, and if it causes a problem somewhere (i.e. either w/occasional times something is not concluded by OWL that is by RDF) then we can fix it once we know if it is serious...

22:23:06 <DanCon> plus, N3 {}s have clear analogs in existing languages: subformulas are common. I don't know what to compare DTs to.

22:23:29 <jhendler> hmm, but I'm tied by charter and Director's rule to build OWL on RDF, so this isn't an option at this point.

22:24:12 <DanCon> but OWL doesn't need subformulas! it's dirt-plain obvious that we can state that one class is disjoint from another using 2-place predicates.

22:25:16 <DanCon> jim, I thought you weren't tied, but that the WG had agreed to build on RDF in Amsterdam. Oh well.

22:25:22 <jhendler> yes, but not dirt clear to me we can say something like "one of the things in the following list is true" without some sort of extension - and we do have predicates that have that sort of effect.

22:25:56 <DanCon> jim, you're making me CRAZY! noone, NO ONE is proposing "one of the things in the following list is true"

22:26:41 <jhendler> sorry Dan, you're right - I was being informal again - didn't mean to make you crazier.

22:26:43 <DanCon> ont:unionOf is a relation between a class and a list of classes. simple 2-place predicate.

22:30:10 <jhendler> Dan - I'm working on an erasure example, but don't think I'll have time to get details right - if I have X oneof (disjoint (list A B C)) would erasure force me to consider X oneof A, X one of B, or does the "scoping" solve the problem (this is off the topic of DT, I'm trying to understand erasure)

22:38:51 <danbri> boy, you guys are still going!

22:39:00 <danbri> hope someone feels like summarising...

22:39:13 <danbri> AaronSw -- how do you get those google stats?

22:39:54 <danbri> are historicals available?

22:39:54 <AaronSw> not that i know of

22:39:54 <AaronSw> i got them from adwords.google.com

22:40:36 <danbri> ta

22:41:47 <Guha_> dancon ... you there?

22:42:04 * danbri treats google http server to all combinations of my lowish security userid/passwds, fails :(

22:44:21 * DanCon is only sorta here

22:46:31 <AaronSw> where's this password-protected server?

22:46:36 <Guha_> ok, will send you email

22:46:48 <danbri> I thought it wanted me to login to adwords...

22:47:39 * danbri not here any more

22:47:40 <AaronSw> oops, full URL was https://adwords.google.com/AdWords/main?cmd=Preview

22:47:50 <AaronSw> then click continue

22:48:08 <AaronSw> (misuse of POST...)

22:50:12 <sbp> wow, the "absoluteURI" production in N-Triples is really confusing. People are bound to think that imported from RFC 2396

22:50:26 <dajobe> eh?

22:50:33 <AaronSw> yeah, those were my thoughts

22:50:43 <AaronSw> also

22:50:59 * AaronSw translates some of Google into Pig Latin

22:51:01 <dajobe> it'll be pointing elsewhere when the graph desc is written

22:51:02 <sbp> in RFC 2396, absoluteURI is a non-relative URI without a fragment

22:52:00 <dajobe> so absoluteURI-reference is *not* really confusing?

22:52:47 <AaronSw> absoluteURIref would work too

22:52:58 <sbp> yeah, that'd be much better

22:53:08 <dajobe> I dunno about *much*, but noted :)

22:54:05 <sbp> just a minor point. had to think twice about it

22:55:02 * DanCon works on test case for http://www.w3.org/2001/sw/WebOnt/webont-issues.html#4.5-InverseOf

22:55:43 <AaronSw> hm, that's not really WD-track, is it?

22:56:28 <jhendler> aaronsw - issues list went from "informal" to "formal" and needs a lot of rewrite - group mainly understands the issues, but we need to clean up the document.

22:56:41 <DanCon> absoluteURIref is the thing-without-a-standardized-name

22:58:03 <DanCon> hmm... test isn't working...

22:58:11 <DanCon> family time in T-2min

22:58:20 * DanCon debugs at high speed...

22:59:10 * DanCon adds inverseOf axiom to ontAx.n3 ...

22:59:28 <DanCon> aha! works...

23:02:16 * DanCon checks in, chacls...

23:07:41 * jonb faimly time adios

23:09:16 * DanCon sends test case, wanders off

23:26:57 * edd having great fun with redland, foaf and smushing.

23:59:11 <danbri> really? do tell!

23:59:25 <danbri> smushing was the death of my perl::rdfweb module.

23:59:30 <danbri> it made my head hurt...