SWD Boston F2F - Day 1

tbl: meeting of a group of people interested in SW in the Cambridge area
... 15 people interested
... no particular agenda
... mainly a social thing
... discussion, brainstorming
... in this room (Kiva)

Guus: short round of introductions

Introductions

Alistair: we have several implementations of SKOS now

Jon: picked up on SKOS at Dublin Core Madrid workshop

Bernard: U. Manchester is adding SKOS to COHSE

Guus: interoperability of vocabularies is core to our work at Vrije University

Diego: my research work is on semantic search

Fabien: my research group is interested in graph-based reasoning on SemWeb

Ivan: every day I take a tram that goes by Guus' office

tbl: i'm not here as W3C director.
... interested in the discussions on RDFa and recipes ("slash")
...I taught a 1-week course [2 weeks ago] and one of the biggest problems was how to configure apache
... if it came out-of-the-box with application/rdf+xml support things would be a *lot* easier

RalphS: the activity of this group is very important, great impact
... very busy, reduced dedication to this group
...but hope to increase my time in SWD again

[Tom calling from Berlin, Elisa calling from Los Altos]

Tom: project looking at model-based metadata; includes Dublin Core and eventually SKOS

Elisa: working with organizations who are keenly interested in metadata about their ontologies ... core business of SandPiper is ontology development

Guus: three objectives of this meeting
... 1) skos use cases: discuss them
... 2) as a result, obtain a list of requirements for SKOS
... 3) review issue list, priorize them, select critical ones

SKOS use cases and requirements

<Antoine> http://www.w3.org/2006/07/SWD/wiki/UCRMaterial?action=AttachFile&do=get&target=ucr-20070117.html

Antoine: 12 use cases in the document
... there are more than 20 contributions
... some are not edited (yet), but available at the wiki

<Antoine> http://www.w3.org/2006/07/SWD/wiki/RawUCs

Antoine: thinks the response of the community was good
... will summarize each one of the UC next

UC1

Guus: two different hierachies for the same thesaurus, is this a requirement?

aliman: multi-hierarchy is an important requirement

Guus: shows an example of the getty vocab

Guus: google for 'tgn getty', then enter 'boston'

Guus: two record types: administrative and geographical, non exclusive

... (shows Utrecht next)

Guus: looks up for an example of a record with two types

Antoine: complex lexical info in the context of this application

Guus: Boston has several alternative names

Guus: Getty shows English, Vernacular, and Historical names

... e.g. Tokyo has 'Edo' historical name

Aliman: multilingual labels are already solved
... but language is not enough in some cases (see the Boston and Tokyo examples)
... the issue is there are different scripts for some language, but only a language tag

Alistair: potential issues with cardinality constraints and preferredLabel properties if there are multiple scripts in which the label might be written

Guus: this is probably out of scope of this WG

ivan: this WG should not worry about this issue. Maybe forward the issue to the RDF core WG

RalphS: asks for clarification of the multi-hierarchy issue

Alistair: a conceptual node may have more than one parent

guus: back to the issue of making a statement about a label

aliman: we should provide a framework to allow that

guus illustrates his point with an example in the whiteboard

Guus: how would we say the label "Edo" is valid only between 1600 and 1800 AD ?

Alistair: annotation properties

aliman: when you use an annotation propierty, you are not limited to a literal value
... you can use a resource as a value of the annotation property. Model annotation as an n-ary relation.

ivan: is this a possible use of reification?

guus: seems to be 2 options: to reify, or to lose information

timbl: another option is to put the statement in another document

SKOS annotated label whiteboard discussion

<aliman> http://www.w3.org/2004/02/skos/core/proposals.html#thesaurusRepresentation-11

Alistair: the old issue notes a place-holder item for this

... "SKOS does not provide support for ... any type of annotation associated with a non-descriptor"

<timbl> http://dig.csail.mit.edu/2007/Talks/0108-time-tbl/ are a few slides about options for modeling things which vary with time

Guus: not sure this is the same thing

ACTION: Alan write up the preferredLabel modelling issue [recorded in http://www.w3.org/2007/01/22-swd-minutes.html#action01]

AlanR: just joined the WG representing Science Commons ... active on HCLS IG

UC 2: Iconclass

... has descriptor concepts and non-descriptor concepts

Guus: this case helps define SKOS scope
... Iconclass is a grammar
... permits adding things to parts of the vocabulary
... I'd like to make this feature out of scope for SKOS
... e.g. KEY; it's not pre-defined where in the vocabulary this is used

Antoine: finding modifiers while browsing a vocabulary -- "post coordination"

aliman: this mechanism allows to create new concepts by combination of existing concepts

Guus: shows an example of the vocabulary ("Animal")

aliman: this is related to the "qualifiers" of the MESH medical vocabulary

Alistair: terms in MESH have flags that indicate they can be used with an additional qualifiers vocabulary

<aliman> http://www.w3.org/2004/02/skos/core/proposals.html#coordination-8 example of coordination

Alistair: ... e.g. 'aspirin' combined with 'sideEffects' means 'sideEffectsOfAspirin'

Alistair: BLISS classification scheme has similar aspects

Antoine: we lost the possibility to attach qualifiers
... cannot represent hierarchies of qualifiers

aliman: ambiguity can arise from the use of qualifiers

Alistair: in my master's thesis I conclude that it is an application-specific decision whether order of coordination is significant

aliman: if you don't have a mechanism to attach the qualifier to particular individuals

Guus: Iconclass also has a notion of 'opposite', or counter-example, done by doubling the letter; e.g. 25FF

<aliman> Examples of using bliss classification http://www.sid.cam.ac.uk/bca/bcclass.htm

Guus: this feature is also used in Iconclass to do male-female distinction

Antoine: 13,000 concepts
... in this vocabulary
... qualifiers allow to reduce the number of concepts
... indexing can use multiple concepts

<scribe> ACTION: Antoine to provide more use cases of uses of qualifiers [recorded in http://www.w3.org/2007/01/22-swd-minutes.html#action02]

Alistair: library world talks about "synthetic" and "enumerative" classification schemes; "synthetic" scheme is meant to be used in combinations to synthesize categories

[15 minute coffee break]

<aliman> http://dlist.sir.arizona.edu/651/ Core requirements for automation of analytico-synthetic classifications

<aliman> (I just found this paper, looks highly relevant to preceding discussion.)

antoine: we have to decide at some point what goes into the document

guus: we should keep the overview and the details of examples

ralph: features in the use cases that are important for skos has to be brought out from the examples (for those who do not know the details)

alistair: we could move the examples from the vocabulary to the example, but what ralph said made me think again...

guus: it is good to have that on our list...

UC 3 - Medieval illuminated manuscripts (mandragore)

antoine: next use case an integrated view to mediaval manuscripts
... there are collections and bridges among these
... we always have info on which vocabularies are used
... an issue of alignment of vocabularies
... it uses the iconclass vocabulary
... and another one that comes from the French national library
... the latter is 15000 subject, simple labels (simple and alternate)
... it is probably a flat list, and they introduce a set of classes for browsing purposes
... you got between 15000 descriptors, and each is linked to a class that is more general

alistair: is it essentially a tree level hierarchy, but you can use the descriptors on the bottom only

antoine: yes, only the leaves of the tree can be used as descriptors

guus: this is a feature I have not distilled yet
... this problem of representing mandragore
... there are 2 issues coming out: (1) requirement for mapping, you need equivalence
... (2) you have the notion of abstract classes
... things that are not for indexing

Guus: abstract classes appear in AAT also

alistair: i think it is a use cases that has some basic requirements for vocabulary mapping among themselves
... there is also a requirement to map between combination of concepts

Alistair: "11U4 Mary and John the Baptist ..."

alistair: 11U4 in the description
... i think that will be a common requirement

antoine: the mapping points that there could be a link between the non descriptor items
... a descriptor on the one side and a qualifier on the other side, the latter is never be found as a descriptor

guus: is it fair to say we have a mapping requirement and two basic requirements

scribe: with respect to the conjunction type of thing, that is an issue (or a requirement)

alistair: it comes up often in my experience
... there is a british standard wg rewriting the thesaurus standard
... working on how to represent mapping between thesauri
... i would think that they will come up with something how to model it

bernard: is there a requirement to map the iconclass to mandragore to identify the ??
... it seems that mandragore is a different type of mapping

Alistair cited ISO 2788 parts 3 and 4 (under development) work on mapping

guus: rephrase the question: do we need more specific than broad and narrow, ie, owl or rdfs vocabulary

bernard: yes, this is what I am asking
... what is the broader term of XX

alistair: there is a browser for mandragore, can we see how this looks like?

antoine showing the mandragore browser example

antoine shows the iconclass vocabulary, one can see the vocabulary and the specialization of the concept

scribe: on the right are the images from the collection (from the BNF) which have not been indexed against iconclass
... you browse your vocabulary, then you have access to the images

alistair: can you browse against the mandragore images only?

Project STITCH : Semantic Interoperability to access Cultural Heritage

alistair: when you do a mapping to mandragore, do you use a second level only?

Antoine: there are 15,000 alignment relations in the mapping

guus: I try to summarize, three things
... (1) need for an equivalence mapping
... (2) a less or more specific mapping, should it be more specific than broad/narrow
... (3) links between compostionals
... we recently linked a nist vocabulary for video tracking
... we got into a similar situation
... we got both the conjunctive and disjunctive form
... may be it should be a requirement, or maybe we can handle outside skos

ralph: there a reference to optional rejective forms
... is that from iconclass?

antoine: this comes from the French vocabulary

ralph: guus showed the double letter example, is it similar

antoine: they are more similar concepts, synonyms

guus: it is quite similar to preferred and non-preferred label

"optional rejected form" means "synonym but deprecated"

alistair: when it comes to mapping requirement, we need to have in mind of the functionality it is used for and focus on that
... that might help us in passing by other representations

guus: in this particular domain mappings are the only thing that adds something to the existing functionalities to musea
... if you open up the collections to browse to other vocabularies that you get new things
... mapping is 100% crucial,
... the only added value, and a big one
... in medicine it may be different

UC 4 - bio-zen

antoine: 4th example bio-zen
... wait for AlanR to come back on that one

UC 5 - multilingual agricultural thesauri

... the 5th use case: semantic search accross multilingual thesauri (agricultural domain)
... these are mostly mulitlingual and to provide open access to these vocabularies
... it is an interesting use case is for multilingual vocabularies
... there are 12 languages, with other terms, related terms, etc
... illustrates some typical usage like skos notes
... use some more complex links
... you can also use more specialized versions, subclasses,
... you find again the links between terms
... representing the terms of, eg, translations
... there is also a representation for mapping links, at the end of the use case
... they are using equivalence links, links between a concept and a combination of concepts
... conjunctions and disjunctions

alistair: that may be like a union
... the last example in the use case they use the mapping vocabulary as it is right now in skos
... it also has 'and or not'
... the second example is exactly an 'or'

guus: ie, they also have the 'and or not' in their usage?

bernard: the more these vocabularies are mashed, they have similar like narrow and broader

alistair: these can be ambigous...

guus: we already have this on the list of the issues (whether we need to represent a specific semantics to broad and narrow)

<scribe> ACTION: guus to check that this issue of more specialization than broad/narrow is on the issues' list [recorded in http://www.w3.org/2007/01/22-swd-minutes.html#action03]

guus: you can say we build into the skos vocabularies that we define, eg, two subclasses
... or we can say that we leave that to the vocabulary, the authors has the guideline to present this as a subproperty to broad/narrow
... the issue is to resolve this

alistair: ie, if people want to do more specific, how would they do it?

guus: yes, and whether this is part of the skos vocabulary or not

Ivan: were problems with representing multilingual scripts found?

... is there enough in RDF to represent this?

Alistair: there were some interesting language problems in the Chinese mapping

Antoine: but I think they succeeded in representing everything they wanted to represent in RDF, though they needed more than SKOS

<aliman> http://www.fao.org/docrep/008/af241e/af241e04.htm a document about mapping between agrovoc and chinese agricultural thesaurus

guus: term-to-term relationship?

antoine: the problem of having several labels for the same concepts that comes up, they want o be able to line up the literal translations to one another

guus: why not use for each preferred and alternative lables

alistair: the the preferred label in chinese may be the third alternative example in english

timbl: cat translates to 'chat' in French, you have to label in french

alistair: you are making a link between translations and labels

antoine: a concept in one vocabulary has a latin name for the pref label, and an alternative label the common name
... the same in the french versions
... and you want to point ot the fact to translate from the two alternative labels

guus: then the latin is a lingua franca

alistair: another thing they wanted that the label in French has been derived from that and than alternative label in English

guus: we may have an issue of relationship of linguistic labels
... not clear to me what to do with this

alistair: we have to be careful with a use case like this is what they do to exactly with this information
... why do they use it

<scribe> ACTION: antoine to capture the issue on capturing relationships between labels [recorded in http://www.w3.org/2007/01/22-swd-minutes.html#action04]

Antoine: e.g. acronym link
... an example of a semantic relationship between labels

UC 6-7 - tactical situations, product life cycle

antoine: use cases 6 and 7 are similar on features
... representing quite simple vocabularies, one is on tactical situation objects
... a list of unstructured terms
... each term has some label and a note
... when it should be used
... the support life cycle is similar

ralph: in #6 it was difficult to see what it says about skos

alistair: me too
... this is not the sort of use case i am familiar with

antoine: i tried to interpret it, but apart form simple labelling i did not find anything

alistair: we could ask them what they want to do

guus: this is what they have...

ralph: maybe we want to ask submittors to point at the wg on areas they want additional things

alistair: use case 7 actually adds a question mark on skos (or owl)
... it is not clear why they want to skos

guus: i could think of reasons

antoine: they were search of standard ways

guus: the problem with use case #7 that it is out of scope
... or am i misunderstanding

ralph: it would be interesting question to ask them what they want skos for

bernard: may be a marketing issue

<scribe> ACTION: antoine to contact the submittors of #7 to see what they want to use skos for (as opposed to, say, owl) [recorded in http://www.w3.org/2007/01/22-swd-minutes.html#action05]

alistair: it seems that they have a requirement to capture lots of things, that may need to extend skos

antoine: no, they really need only flat things...
... they need a structure to represent a natural language representation without a reasoner

UC 8-9 - GTAA (audiovisual archives) and CHOICE@CATCH (radio/TV)

antonie: number #8 gtaa web browser, accessing thesaurus
... want to provide the user with a sophisticated vocabulary

guus: there is an archive for tv and radio programs
... they do annotation inside the content but also coming from broadcasting companies
... on the top level there are 8 different facets
... and several of the sub hierarchies have separate classifications
... and that is the whole thing
... they are specific for a facet

alistair: there is a thematic and a named hierarchy, and they are orthogonal

guus: we can get test cases out of it

ralph: 'only keyword and genres can also have broader/narrower relation', is that a restriction?

guus: this is a very flat structure, this is not really a restriction

antonie: use case #9, another use of the same vocabulary of use case #8,
... using a special algorithm that provides the user an indexer
... the idea is to explore the different links in the thesaurus to rank the concepts
... if you have to index a document with a set of candidate terms, if the thesauri includes these terms, than that hierarchy is also presented

guus: I would have personally merged #8 and #9

antoine: #9 provided in a functional view
... adding a representation to an applicaiton is nice

guus: people in computer science like automatic things
... but these people like to manually check

ralph: even though it does not add anything technically, it adds a new aspect, good for 'marketing' reasons

alistair: if you look a traditional model, you manually build a vocabulary and index
... in this case the vocabulary is done manually, but an automatic indexing is good
... a use case document should have a business model section to show how different scenarios are used

guus: summary: #9 does not add anything to the requirements, but is an interesting use case scenario to keep

alistair: applications might want the integrity of their data, and expressing the constraints is a requirement

guus: there is already and issue on the level of semantics that skos has

<scribe> ACTION: alistair to summarizes the aspects of semantics of the skos data model [recorded in http://www.w3.org/2007/01/22-swd-minutes.html#action06]

JonathanRees: I'm part of Science Commons

alistair: a question on #8, relationships on terms between facets were computed
... question is how were these computed?

guus: the general problem was that there were lack of relationships
... but I do not think there were much semantics

alistair: it also says the precomputed terms were not part of the iso standards

guus: good question, I do not know

GTAA use case as submitted

<scribe> ACTION: guus to check with veronique on the terms being outside the iso standard [recorded in http://www.w3.org/2007/01/22-swd-minutes.html#action07]

UC 4 - bio-zen

Biozen use case as submitted

antoine: use case #4, bio-zen ontology framework
... the main point to represent these medical vocabularies, keeping all the infos that are useful for application
... the application was not really detailed
... gene ontology and mash are the two examples for applications
... it has an example of representation of a term
... the main point is the fact that the representation they mix all kind of different of metadata vocabularies
... they created some sort of metamodel using owl, and uses pieces of other vocabularies
... they use all these meta models to represent the medical vocabularies
... they use, eg, dublin core plus skos terms together
... they created an owl specification to mix these metamodel features

guus: why there was within the definition there is a representation of the part of relationship
... does the mesh have its own hierarchy

alan: 'is a' is not 'part of', careful about that

guus: in skos we use the broader and narrower terms which are less defined

alan: obo originates in the gene ontology
... the latter has is a and part of relationships in it
... there has been a number of threads using this
... one thread is to translate obo to other formats, people used, eg, skos
... they have to decide where broader, etc, are used
... these actually threw away information but they are part of skos
... from my understanding at the time at least
... there is an effort to translate this into owl
... second thread of discussion is the 'quality' of the whole thing

<Elisa> There is a recently released related portal - Daniel Rubin and his group have created this and are working to develop it as a part of their NCOR work: http://www.bioontology.org/tools/portal/bioportal.html

alan: what can be related to what, what are the description of that, more philosophical stuff

guus: some people make subproperties from, say, skos broader
... then you do not throw away things

alan: i had the issue on putting it with owl-dl

guus: that is a separate issue on the agenda (relationship to owl-dl)

Alan: Matthias is asking that as we fiddle with SKOS, we try to keep it OWL-DL compatible

Alistair: it's already not OWL-DL

alistair: if you go into library sciences, you will find papers on classification
... people there define fundamental facets, time, space, etc
... there are discussions on what these fundamental facets are
... that might come to the skos spec
... but if you want to do that, this should be done as an extension of skos (in my view)

guus: b.t.w., the relationship to owl-dl should be part of our issues list, not requirement
... maybe if we define a set of constraints, that might lead to skos-dl...
... but this is a topic for discussion

alistair: it is tricky, extension by requirements is one of the major way of extending skos, and all of those are annotation properties, and that leads to problem

<scribe> ACTION: alistair to rephrase the old issue of skos/owl-dl coexistence and semantics [recorded in http://www.w3.org/2007/01/22-swd-minutes.html#action08]

bernard: it was good in the owl days to have implementations submitted, too

guus: for the moment it is good to collect the information, it is good to use them as test cases
... but this group is much smaller than the old owl group, and we have a resource problem

alistair: there are two wiki pages, and the shiny new skos web site: SKOS home page

<alanr> http://biopaxwiki.org/cgi-bin/moin.cgi/InohMolecularRoleInSkos

alistair: the idea that anyone who has implementation should be able to add it

UC 10 - BirnLEX

antoine: use case #10 birnlex, lexion for neurosciences
... aims at providing several vocabularies
... they are the same as the bio-zen use case
... there is a mixture of different metadata models, skos, dc, foaf, etc

alistair: all they want is the some of the properties like pref label, alt label, not in the structure of label
... if skos has good annotating support, people may just want to use that

guus: i interpreted this as having a lot of need to various type of relations
... there are many things in the examples term relations with other semantics

alan: the argument is that there is a desire of the part of type of relationships that we may need in general
... ie, people insert tags into the rdf labels,
... shows the importance of this issue

Alan: the BIRNlex use case may bring in issues for our vocabulary management work

alistair: this is a bit of annotating just about everything

UC 11 - OBI (Ontology for Biomedical Investigation)

<alanr> https://www.cbil.upenn.edu/obiwiki/index.php/OntologyMetadataAnnotation

alistair: they do not want skos broader and narrower
... it is more that they want all type of documentation/annotation support

guus: the issue here is that you have your concepts
... how to document/annotate various concepts
... and what skos give you on that

<scribe> ACTION: alan to write down the general documentation requirements, in particular to those that are related to literal values, and how to represent that in skos [recorded in http://www.w3.org/2007/01/22-swd-minutes.html#action09]

antoine: use case #11 quite similar
... I have not read it in much details
... it is once again to represent all these various vocabularies and linking/importing skos concepts to an 'own' ontology
... and extending skos relations

guus: my proposal: there are still use cases coming in
... we have to include facilities to evaluate use cases
... we should go through the list of the requirements and see if we can refine this
... and go through the issues' list

SKOS Requirements (sandbox)

-> http://www.w3.org/2006/07/SWD/wiki/UCRMaterial?action=AttachFile&do=get&target=ucr-20070117.html#SKOS SKOS Requirements sandbox

--R0. Information accessible in distributed setting

Guus: is this a requirement on SKOS?

Antoine: doesn't seem to change anything about SKOS or what it represents

Guus: seems to be a general Web requirement

Ralph: comes with RDF and the Semantic Web

RESOLUTION: drop R0. Information accessible in distributed setting as not SKOS-specific

--R1. Representing relationships between concepts

Bernard: "displaying or searching concepts" might give the impression of constraining our scope
... e.g. excluding annotation

Guus: how about "representing relationships between concepts"
... the ability to represent hierarchical and non-hierarchical relationships between concepts

-- R2. Representing basic lexical values (labels) associated to concepts

Antoine: "basic" as in "simple" as compared to more sophisticated scope notes

Guus: basic lexical _information_ or do you really mean to restrict to _labels_ ?
... "access to" not needed

-- R3. Representing links between labels associated to concepts

Guus: we have an issue related to this
... this requirement may change after resolving the issue

Alistair: we could suggest a point of view without making a hard requirement
... while reviewing all these requirements today

Guus: I suggest that any requirement with a related issue be marked as "soft"

-- R4. Representation of glosses and notes attached to vocabulary concepts

Antoine: "notes" means "scope notes"

Guus: so use the well-known term "scope notes"

Antoine: should we include administrative notes?

Jon: suggest "glossaries" instead of "glosses"

Guus: I thought there is a distinction between a glossary and a scope note

Alistair: what's the difference between 'gloss' and 'definition', then?
... SKOS hasn't used the term 'gloss' previously

Guss: "representation of textual descriptions ", with text mentioning definitions, scope notes, ...

R{6,5}. Multilinguality

Tom: suggest "Representation of lexical information in multiple languages"

Bernard: multiple _natural_ languages?

Guus: yes, good addition

-- R6. Descriptor concepts and non-descriptor ones

Guus: distinction between concepts intended to be used for indexing and other concepts?

Antoine: yes
... what I had in mind was the existing skos:subject
... some concepts cannot be used as subject relationships

Guus: qualifiers are still relevant to indexing
... e.g. AAT vocabulary
... Furnishings ... furniture ... <furniture by form or function> ... screens
... the terms in <...> are not meant for indexing

Alistair: many folk would not consider the <...> to be concepts; they call them "node labels"
... they are labels for a grouping of concepts, the groupings are called 'arrays'
... they say the node label does not represent a 'concept'
... in the British standard it is quite clear that the node labels are only used in a certain way
... but AAT adds things to the thesaurus beyond the British standard
... it's just a matter of us wording this requirement correctly
... consider Mandragore; you're not supposed to use things from levels 1 and 2
... but the British standard demonstrates a requirement to be able to label groupings

Guus: propose to rephrase as "the ability to distingish between concepts to be used for indexing and for non-indexing"

Bernard: is this really a requirement or just an issue?

Guus: is this in the ISO standard?

Alistair: no, in ISO thesaurus any concept can be used for indexing
... there's no a-priori reason why something not intended for indexing in one context would be inappropriate for use in another context

Guus: suggest R6 is a soft requirement
... and add a new requirement having to do with grouping
...
... "the ability to include grouping constructs in concept hierarchies" -- as a soft requirement

Alistair: hierarchies are not the only place where node labels can be used
... node labels are also used in related terms

<aliman> see z39.19

-- R7. Composition of concepts

Guus: is this like conjunction and disjunction?

Alistair: the terms 'conjunction' and 'disjunction' don't really make sense as we're not talking about sets of things
... the classical way of talking about this is to talk about 'coordination', and 'coordination of things'
... I'm afraid to use set-theoretic language, as this would be jumping the gun
... we're not talking about True and False or sets, rather we're talking about concepts
... 'compound concepts' is a term used in the thesaurus world
... 'post-coordination' usually means that things are coordinated at search time but it typically really just means queries with more than one thing
... I don't recommend referring to pre- or post-coordination

Guus: I recommend linking 'coordination' to an explanation

Alistair: I'd be happy using 'composition' rather than 'coordination'

Guus: let's categorize into 'candidate requirements' and 'accepted requirements' (rather than 'hard' and 'soft')

-- R8. Vocabulary interoperability

Guus: mapping at the level of equivalence, more specific, less specific
... further things under discussion
... suggest dropping this, as we need to be able to test

Ralph: is R8 the general case and R12 a specific case?

Jon: I have another use case; our system supports the expression of relationships between terms in vocabularies we own and terms in vocabularies we don't own
... the reciprocal relationship would need to be endorsed by the owner of the other vocabulary

Guus: I can make equivalence statements in my own ontology and others can choose to use mine or not use mine
... valid to have different statements about mapping and determine to which you commit

Jon: imagine two indexing systems but a single retrieval system

Alan: a search for A should include B but not vice-versa?

Jon: yes

Fabien: is this specific to equivalence or is it a filter on the source?

Guus: back when we did OWL, I had to spend a long time defending owl:imports
... this may be outside the SKOS language, at a different level of the SemWeb stack

Jon: this is not about trust but about representing the intent of the thesaurus writer

Guus: but it's at a reasoning level

Alistair: we refer to 'SKOS concepts' and 'SKOS concept schemes'; perhaps we can also talk about 'mapping schemes'

Guus: like provenance?

Bernard: why isn't a concept scheme the same as a mapping scheme

Alistair: they're handled differently by applications
... an application wouldn't display a mapping scheme as a hierarchy

Bernard: but if you dereference all the concepts in a mapping scheme wouldn't you end up with a concept scheme?

Alistair: there's current a loose recommendation that two concepts in a single concept scheme do not share a label
... this might be expressed as a logical constraint on a concept scheme
... but this constraint would be inappropriate for a mapping scheme
... if someone wants to capture in their RDF graph that there exists a set of mappings that he authored ....
... a concept scheme has a notion of 'containment'
... different integrity constraints if you're just collecting some mappings

Alan: if you use owl:sameAs, you're making a bi-directional assertion
... but the author of a vocabulary might want only a one-way assertion

Antoine: is related to the issue Sean raised about containment

Bernard: is there something about R8 that is not included in R12?
... "it shall be possible to record provenance information on mappings between concepts in different vocabularies"

RESOLUTION: R8 reworded to "it shall be possible to record provenance information on mappings between concepts in different vocabularies"

Tom: we use "vocabulary", 'concept scheme', and 'SKOS model'; let's stick to one term

Guus: I propose we drop 'concept scheme' and use only 'vocabulary'

Alistair: the ISO standard does not distinguish between term-oriented and concept-oriented; I made up this distinction
... you _can_ talk about whether the data model is term-oriented or concept-oriented, but not the vocabulary itself

Guus: consider a 'bank' vs. 'financial institution' example
... 'bank' implicitly defines a concept, implicitly it's a lexical label
... consequence for a thesaurus is that the term 'bank' cannot be used anywhere else
... in practice this distinction is useful, which is why I'd prefer to not use the term 'concept scheme'
... 'vocabulary' is more general and makes less commitments

Tom: R10 is really talking about the extension of the SKOS vocabulary, not the SKOS model
... could be confusing if we use the term 'vocabulary' generically

Alistair: the 'concept scheme' idea came from DCMI

Ralph: it's probably easier to refer to "the SKOS vocabulary" and "a SKOS concept scheme" to differentiate between the SKOS terms and a thesaurus written using the SKOS terms

Bernard: let's define terms near the start of the document

Antoine: I tried to consistently use 'vocabulary' for applications of SKOS and 'model' for SKOS itself

Alistair: there are implicit integrity constraints currently expressed only in prose

Tom: this is a big question that deserves more thought, let's not decide now
... SKOS model is like an application model (in DCMI world)
... R10 talks about extending both the SKOS model and the vocabulary of properties

Guus: for the time being, let's distinguish between the terms in the SKOS vocabulary and the application terms that use SKOS

Guus: for now, let's use "SKOS vocabulary" and "concept scheme", respectively, for these two

-- R9. Extension of vocabularies

now called: "R9. Extension of concept schemes"

Alistair: how do I express that I want to import another concept scheme into my own, or import only a part of another concept scheme?

Bernard: why import, just reference?

Alistair: how does a browser know the boundary of a new concept scheme?

Bernard: related to Protege issue of how to represent externally-defined items

Guus: do we include maintenance properties, revision information, etc.?
... I suggest we add a requirement related to versioning information

Bernard: R9 is about tools

Ralph: I suggest we keep the vocabulary management work, including versioning, as a separate task and not mix it into SKOS right now

Jon: example; replacing a single term with two terms
... not necessarily establishing a broader/narrower relationship but dropping the old term

Guus: we handle this in OWL by deprecating old terms

Alistair: my approach is to worry first about how to represent a static model

Guus: suggest deferring versioning questions to separate vocabulary management work and later evaluate whether any SKOS-specific properties are needed

Alistair: we have requests to be able to define concept schemes as 'we use everything in that scheme with the following additions'
... Alistair: I need to find a better use case to motivate this

-- R10. Extendability of SKOS model

now "R10. Extendability of SKOS vocabulary"

Guus: means "local specialization of SKOS vocabulary"
... propose to rename this to "local specialization of SKOS vocabulary"
... get this for free

-- R11. Attaching resources to concepts

Antoine: this is skos:subject; annotating resource

Fabien: inverse of dc:subject?

Alistair: skos:subject is dc:subject with a range constraint

Guus: propose to rename this to "Ability to represent the indexing relationship between a resource and a concept that indexes it"
... I suggest this is a candidate requirement

-- R12. Correspondence/Mapping links between concepts from different vocabularies

now "... different concept schemes"

Bernard: related to mapping between labels in different concept schemes; that can be a separate requirement

Guus: at a minimum, equivalent, less/more specific, and related

Alistair: also composition

Alan: is 'related' a superproperty of 'broader' or 'narrower'?

Alistair: no

Alan: please document this explicitly in the spec

Guus: propose a new candidate requirement: Correspondence mapping links between lexical labels of concepts in different concept schemes

-- R13. Compatibility between SKOS and other metadata models and ontologies

Antoine: may not bring any additional requirements on representational features

Guus: what other models do we want to be compatible with?

Alistair: Dublin Core
... note that changes have been made to Dublin Core specifically to align it with SKOS

<Elisa> Another metadata standard we should consider here is ISO11179

Elisa: ISO 11179 is another related standard, on which Daniel Rubin and I have spent time recently
... Daniel is interested in 11179 because many biomedical ontologies use it
... by mapping 11179 to SKOS we bring a lot of those into the RDF world

Alistair: 2788 is a thesaurus standard and is very different from 11179, which is a metadata model
... there is a particular part of 11179 that is intended to talk about classification schemes
... it's obvious how SKOS and that part of 11179 relate

Alan: what does "compatible with" mean?

Bernard: does "compatible with" mean "does not violate the [Dublin Core] abstract model"?

Ralph: what sort of test cases could we construct to decide "is compatible" or "is not compatible"?

Alistair: could we translate a data instance using 11179 to a data instance in SKOS? how much data loss? how much data loss in transforming back?

Alan: the scope of 11179 is much larger than that of SKOS

Guus: it would be good to identify specific other models
... e.g. 2788, 11179 [part 3]

Alistair: 5964 (multilingual)
... I'd put 2788 as a stronger requirement than 5964
... interpretation of 5964 is harder

Alan: is SKOS a "metadata model"?

Guus: propose omitting the general requirements R13 and R15 and adding specific requirements for 2788, 11179.3

Alan: is all of 11179.3 relevant to SKOS? there's a lot of stuff in there

Elisa: I am happy to help narrow the scope
... and the US contingent in the 11179 group are physically close to me

-- R.14 OWL-DL compatibility

Guus: we can talk about a SKOS representation that is OWL-DL compliant

Alan: make it formal that annotation [sub]properties are allowed?

Guus: we have to be sure that we can complete our deliverables without requiring another WG to be rechartered
... we can make comments to the OWL comment list about annotation properties

Alan: there's a partial workaround available to SKOS

Ivan: what does DL compatibility mean when you have a processing model that includes some rules into SKOS?
... regardless of annotation properties, SKOS is already out of DL

Alistair: the rules don't have to be used

-- R.16 Checking the consistency of a vocabulary

Guus: issue raised earlier about semantics

[I think Tom's suggestion is agreed implicitly]

Jon: I've been updating the sandbox wiki in realtime

<JonP> http://www.w3.org/2006/07/SWD/wiki/CandidateReqList#preview

Guest: Steve Williams, hyperforms technologies, participating in W3C 2-3 in binary XML WGs
...interest in semweb, relatied technologies, interested in AI.

Guest: opera software, interested in semweb since 98, bumped into danbri, then graduate student of astrophysics, then hired by opera, mostly programmin, chaals getting me into more on semantic web.
... responsible for my opera foaf stuff

Best Practice Recipes

Current issues list: http://www.w3.org/2006/07/SWD/wiki/BestPracticeRecipesIssues

guus: moving on to discussion of recipes for publishing RDF, have as input recipes document from SWBPD, incomplete document, Jon action to generate issues list, now on wiki

guus: suggest briefly review this list, pick out critical issues, spend time discussing critical issues, diego can play a role because has proposed resolution for one of these issues (we can discuss and decide on)

jon: first four issues left over from previous working group, diego's been working on first issue.

<ivan> http://lists.w3.org/Archives/Public/public-swd-wg/2007Jan/0033.html

Diego's proposed resolution to COOKBOOK-I1.1

diego: [issue 1.1] already on mailing list, there was a TODO tag, issue regards configuration of apache to serve vocabularies, apache uses configuration files with directives, one of these directives is the overrides ...

<JonP> Diego's verification email: http://lists.w3.org/Archives/Public/public-swd-wg/2007Jan/0005.html

diego: in original doc there was a TODO tag next to overrides, to verify this is correct, I checked this and discovered that the line was correct, no additional overrides required, both overrides are required.
... proposed to remove this TODO tag.

@@TODO from Recipes WD

jon: is that resolution acceptable? how do we handle? seems to be fine.
... vote as a group?

guus: can we write a test case?

diego: I have test case.

jon: diego sent around email, describing test cases and results.

<berrueta> test cases

guus: further discussion?

PROPOSED to resolve issue 1.1 as per email of http://lists.w3.org/Archives/Public/public-swd-wg/2007Jan/0005.html

ralph seconds

no objections

RESOLVED

<scribe> ACTION: jon to update issue list as per resolution of issue 1.1 [recorded in http://www.w3.org/2007/01/22-swd-minutes.html#action10]

jon: skip over second issue, because the TODO is that it references 6 which doesn't exist
... issue 1.2 and 1.4 are essentially the same.

guus: PROPOSED to drop issue 1.2

diego seconds

no objections

RESOLVED to drop issue 1.2

<scribe> ACTION: jon to update issue list as per dropping of 1.2 [recorded in http://www.w3.org/2007/01/22-swd-minutes.html#action11]

jon: issue 1.3 - why performing content negotiation on the basis of the "user agent" heading.
... is not considered good practice.

bernard: whole section should be in an appendix, why in main body of text?

aliman: karl suggested move whole content negotiation to appendix

timbl: if you add features to a user agent, because it stumps the deployment of new browsers, e.g. folks at opera, resulted in user agents lying about who they are
... e.g. some browsers ship with lying user agent fields, unsatisfactory, better to look at the mime types
... sometimes in practice necessary to look at user agent field to pick up bugs, where you know there are specific bugs, particular trap for particular browser.

jon: potential resolution is to explain the problem with using user agent, as per stunting development?

timbl: yes

diego: esiting doc to explain this?

ralph: we didn't want to break semantic web applications which don't include accept header, so set RDF as default response

timbl: two cases, one is your serving data, but if you are trying do the trick of doing either rdf html version, but only put if you are content negotiating ...TAG says something about identity of resourse

aliman: but use 303 so don't have to have same info content

timbl: yes

bernard: uncomfortable with hack

jon: real world, applies to IE7?

<timbl> The TAG I think says the same URI may conneg go to different representations ... but they should convey the same information

guus: someone take action to look at IE7

jon: regardless of IE7, should still leave hack in.

aliman: I agree

jon: issue 1.3 is actually to explain why the hack is slightly bad

ralph: new issue would be to look again at the hack

jon: two separate issues

ralph: if IE7 does the wrong thing, leave it, if IE7 does the right thing then drop the hack (except if you have to support a specific ocmmunity)

bernard: I will raise this issue

<scribe> ACTION: bernard to raise new issue re IE6 hack [recorded in http://www.w3.org/2007/01/22-swd-minutes.html#action12]

<scribe> ACTION: diego to look at IE7 accept headers [recorded in http://www.w3.org/2007/01/22-swd-minutes.html#action13]

ralph: test cases?

aliman: just whar's in the document already.

jon: 1.3 issue has been raised.

ralph: I move we conside 1.3 open - there is a TODO that needs to be done, timbl how likely is that TAG write something about use of user agent header?

timbl: may be something already, otherwise need to send email to TAG

ralph: I will own this issue

<scribe> ACTION: ralph propose resolutition to issue 1.3 [recorded in http://www.w3.org/2007/01/22-swd-minutes.html#action14]

jon: move on to issue issue 1.4 ... recipe 6 is not there

bernard: do we need a recipe 6

Alistair: one of the reasons people like slash namespaces is because the response to a GET is specific to the requested resource and you can incrementally learn more with additional GETs
... in recipe 5, if you request RDF you get redirected to a namespace document that describes everything
... recipe 6 was intended to permit serving just a relevant chunk of RDF data

<timbl> see: http://www.w3.org/DesignIssues/LinkedData

TimBL: d2rdf does this ... a SPARQL query can navigate a graph by recursively pulling in documents
...e.g. d2r server does that virtually, enthusiastic about this group pushing linked data, critical thing about linked data is that when you derefrence linked data you get all arcs in and out then human being can navigate the graph, then also SPARQL query can find all the graphs by pulling in all the relevant docs, not as efficient but good, important to make all the backlinks.
... it's good to remind people to include backlinks; dereferencing a student should give you a pointer back to the class
...minimum spanning graph, RDF molecule
... patrick stickler CBD only arcs out: Concise Bounded Description [Stickler, 2005] ... this is important recipse to include
... proposed workshop for web conference about linked data, didn't have space.

<timbl> The D2R Server generates linked data automatically.

jon: this should be opened, this recipe should be written.

guus: open means we work on it now.

timbl: two separate points to be made, first is the recipe redirects you to a file, but the file is vritual, most web servers are virtual, if you redirect to a SPARQL query, doesn't mean that the SPARQL query is evident in the URI, inside that you use a rewrite rule, give a nice URI to your _part_ then use a rewrite rule to create the SPARQL query, hide the SPARQL query.

jon: useful to say that as part of the recipe, best handled by a service that will handle on the fly.

timbl: important to separate, e.g. FOAF files do it by hand, in some cases there is a lot of hand written stuff, fact that something is genearated automatically may apply to other recipes also.

ralph: we are a deployment group, rewriting nice URIs to query URIs, good to show this to get more deployment.

guus: we are in a position to write a resolution to this section, who can do?

ralph: examples of sparql services we can use?

timbl: geonames? d2r server. one does 303 redirect to URI encoded SPARQL query.

ralph: sounds like some code existed.

jon: two parts to this, first part is data, second part is server configuration. We're looking for a document fragment example and server config.

ralph: wordnet is an obvious choice, but the W3C need to commit to D2R service.

diego: I will own the issue

guus: maybe diego can talk with Ralph about wordnet, nice use case, widely used.

ralph: need to get W3C web servers to support the service, but plausible.

jon: issue 2.1 (QA comments) ... karl raised wordsmithing and structural comments, lots, something for each section, I couldlnt' break out individual issues, I'd like to propose we simply open this, I'll take ownership, I'll implement most of his suggestions and propose as modification to the document.

guus: comments from QA people, we owe them a response. Need to go through each response, say what we did.

jon: Issue 2.1 ... raised by me (wiki lies) recipes are specific to apache server, may be applicable in non-apache environments, do we want to provide general template that describes recipes in general, or say that recipes can be implemented by a script.

bernard: general template is possible?

ralph: is there is one web master would recognise then can look at it. cookbook is very practical, make it simple for server admin to do it, if someone wants to submit recipes for other environments then good.

jon: common principles e.g. redirect based on content negotiation, I don't konw enough about other environments to say how.

ralph: like to encourage others to contribute recipes

jon: suggest we provide a place for people to submit new recipes for other environments

ralph: happy with mailing list for proposed translations.

ivan: wiki? ... esw wiki?

alistair: the diagrams were intended to provide a schematic overview of the behavior we were trying to implement
... hopefully this would give people enough information to implement in other environments

jon: leave this in a raised state? open it?

ivan: resolution to open wiki page.

ralph: willing to own issue, proposed resolution to create wiki page.

guus: publication schedule, can say we don't think it's a high priority if we think resources are limited.

ralph: should be ok, but may get not good configs

jon: issue 3.1 (raised by me) discussion about differentiating between versions, one reason we use redirects to supply most recent snapshot of a vocabulary
... actualy document.

Alistair: consider Dublin Core; it has a fixed URI that is redirected to the current version of the vocabulary
... an application may be able to deal with versioned URIs, and access older snapshots of the vocabulary

jon: why not use mod_rewrite instead of redirects, can use redirects to make version ???
... proposing a complete suggestion for a naming convention for handling this sort of thing in a recipe ...
... link is in extended requirements section of original doc

extended requirements

<timblwanted to say 1) the redirect does not convey that semantics and so the semantics need sto be onveyedelsewhere and (b) the redirect is an overhead
... and (c) metadata in URI is covered by the TAG in a new finding and is in general bad

timbl: problem with redirects is twofold, part is when you redirect, redirect doesn't say anything about what the relationship between source and target is, doesn't deliver the smeantics you want, also takes time - overhead - adding it thourgh overhead is to be avoided if possible. However to be able to track differences between versions of an ontology is very useful, on web different translations, different content types, design issues note about this /gen/ont
... TAG is aware of this but hasn't tackled from RDF point of view, I wrote an ontology: http://www.w3.org/2006/gen/ont.n3

timbl: gen ont if best practices could just dump the map, some metadata can get from apache, if running CVS can generate metadata from previous versions if you've got web CVS. If you've got content negotiation can find out what all the options are from apache config
... that would be ideal, ideal pattern which nobody does at the moment - don't know if I should ask this group or another to look into this.
... suggest this group push this out.

ralph: I'd like us to consider this as a candidate requirement for discussion, but not for this document, because this doc was one of what was expected to be a collection of docs coming out of SWBPD, several aspects of VM e.g. how to serve them, versioning, properties about a vocab e.g. provenance, best practices for all that. SWBPD imagined there would be other docs to go along with this, I image this would be part of our work, Jon's soltuoin is plausible but
... consider this as candidate requirement for other VM work, not to try and solve for this doc (recipes).

jon: ralph is suggestins part of resolution of this issue is to start another document, and that this part of recipes should point to that documeent. Currently recipes punt, lots of different ways to do it, nonbody says what is best way. Part of utility is to say here is a recipe, a way to do it, this is a generic enough recipe to work in enough cases.
... so we should reword this doc?

ralph: shows up on our deliverables page ... "principles for management..." we can point to this document.

guus: propose to leave this issue as raised, go on with recipes without resolving it, indicate that this issue is intended to be resolved by another doc.

Ralph: specifically, our deliverable 3. Principles for Managing an RDF Vocabulary

Alistair: in anticipation of needing to use RDF to describe the relationship between a dynamic thing and static snapshots of that thing, I've published net/d4: http://purl.org/net/d4. d4 may provide a basis for discussion
... also GRDDL seems to be in a similar space; making assertions about the relationship between various documents you might be able to access from a namespace

jon: issue 3.2 (testing) diego has written some unit tests, useful if we could provide a service for developers who wanted to utilise cookbook recipes, provide as a server validation service officially
... this would allow you to specify you wanted to test a particular server against a particular recipe

guus: not an issue with the current doc,

ralph: intermediate step to publish the test cases?

jon: thinking more like RDF validation service, you point service at URL and say which recipe.
... diego has already written the code, we just don't have the service.

ralph: you have test service?

timbl: great idea, presentation suggest that you may get more people who validation than go to the doc, so service could point people out to the doc, start from existing situation and lead people buy the hand to appropriate recipes.

<timbl> (Service could be implemented in Javascript within the document ;-) not.

guus: question of timing, has to be synchronised.. From pragmatic view, suggest take an action to look at possibilities and report back, time frames and synchronisation.

ralph: can you commit to hosting?

diego: I can write the code.. .

timbl: can it run in a browser?

diego: runs on server side.

guus: before resolving, do some suggestion on the list about how to realise this.

ralph: I could put this in category of vocabulary mangement validator, then falls into big validator project we have.

timbl: rethink about how to support validators, logically if you're going to validate, you can go so many ways ... top of an iceberg.

guus: open issue, diego is owner, first to propose a timescale.

jon: issue 3.3 raised by diego, mod_rewrite is required for all recipes, but we don't say so.

ralph: and apparently not there by default.

guus: jon is to be issue owner.

ralph: probably worth saying here's what apache config file to go to to cause it to be loaded.

guus: close this dicussion on the recipes, thanks to all, moved to state where can see progress.
... have to think carefully about status of RDFa, Note? relationship with HTML? also think carefully about time horizon for SKOS recommendation, test cases and implementations worry, can see six months document to go to last call, also need a test suite in place, similar to OWL so tool developers can test stuff. Document itself in good shape, but getting to candidate rec may take more time.

l... to keep to schedule, we need to have test suite stage by the summer.