On 8/23/15 4:53 AM, Jan Voskuil wrote:
> Karin, thank you for your insights and the references!
>
> I agree that flexibility is desirable. Would this in your view also
> imply that the range declaration of dcterms:type as rdfs:Class should
> be dropped?
I have mixed feelings about the ranges in dcterms. On one hand, without
them one has no guidance on usage/expectations, and we know that it's
not easy to work with properties whose ranges can be either literals or
IRIs. On the other hand, not everyone has IRIs (yet) for their data.
(cf. the ISO language codes, ISBNs, etc.) The saving grace is that for
some common properties the 1.1 vocabulary exists, so if one needs to use
a literal for "type" one can use dce:type rather than dcterms:type.
>
> As far as the use of SKOS is concerned, the creation of a formal
> thesaurus sounds much more complicated than it is. Instead of
> publishing a list of literals, people should IMHO be stimulated
> wrapping the literals in concepts.
Well, in spite of your HO, you have no control over what others choose
to do. While SKOS is not difficult, we can't make others use it if they
decide not to. However, if we want Dublin Core to be useful throughout
the web of data, we need to keep it as unconstrained as possible.
kc
>
> I believe that from a practical perspective this method is easier
> than publishing formally defined Vocabulary Encoding Schemes (VES),
> which the DC-standard seems to prescribe wherever literals are used
> as property values (independently of whether this is actually done at
> a significant scale). Using a VES, one can write things like
> dcterms:subject "Napoleon"^^ex:myVES. Using simple tooling one could
> instead put all the literals in an Excel-sheet and import them in a
> thesaurus, so that you could then write dcterms:subject
> ex:NapoleonConcept. Subsequently, others could relate their
> terminology to the one in the thesaurus, using standard relations,
> which is not possible with a VES.
>
> Different communities could use different concept schemes (thesauri)
> in their APs to restrict the property values of the same properties.
> -j
>
>
> -----Original Message----- From: DCMI Architecture Forum
> [mailto:[log in to unmask]] On Behalf Of Karen Coyle
> Sent: dinsdag 18 augustus 2015 19:49 To:
> [log in to unmask] Subject: Re: dcterms:type and SKOS
>
> Jan,
>
> I believe I understand your use case, but it is only one possible use
> of Dublin Core terms, and I feel it is essential that DC terms be
> suitable for a wide variety of communities. What makes dcterms one of
> the most used vocabularies in the LOV dataset [1] is precisely its
> flexibility. Limiting dcterms:subject to skos:Concepts would likely
> discourage communities whose tradition does not include the creation
> of formal thesauri.
>
> It makes sense to have an agreement with your data sharing partners
> about the expected values for properties. This is the basis for the
> DCMI work on application profiles [2][3] which extend the basic
> vocabulary to meet specific needs. APs also relate to work in
> progress on RDF validation [4][5]. The "best of all possible worlds"
> would be a very general and flexible vocabulary that can be
> integrated into specific applications but that also allows
> interconnection between disparate communities. Machine-actionable
> application profiles could make that possible.
>
> As for the list of terms in the DC type vocabulary[9] - I see it as
> being rather naive. Library of Congress not only has its own list of
> genres [6], it has a list of lists of genres [7]. The total number
> must be in the high three digits. I also find interesting the
> FaBio/CiTO list, that is primarily based on academic articles[8].
> "Type" is definitely a concept within a context, so will be defined
> differently in different communities, as we see already.
>
> kc
>
>
> [1] http://lov.okfn.org [2]
> http://dublincore.org/documents/singapore-framework/ [3]
> http://dublincore.org/documents/profile-guidelines/ [4]
> http://wiki.dublincore.org/index.php/RDF_Application_Profiles [5]
> http://www.w3.org/2014/data-shapes/wiki/Main_Page [6]
> http://id.loc.gov/vocabulary/genreFormSchemes/marcgt.html [7]
> http://id.loc.gov/vocabulary/genreFormSchemes.html [8]
> http://sempublishing.sourceforge.net/ [9]
> http://dublincore.org/documents/dcmi-type-vocabulary/
>
> On 8/17/15 12:56 PM, Jan Voskuil wrote:
>> Hi Dan, and Karen, I've been thinking about the valuable point you
>> raise --- thanks for doing so.
>>
>> Before going on, let me first ask: would you agree on the basic
>> premise on which I started this discussion, namely, that the range
>> of dcterms:type should not be rdfs:Class?
>>
>> The next question is about declaring the range of dcterms:subject
>> and dcterms:type to be skos:Concept. I see your point about
>> usefulness. I think this has to do with two fundamentally different
>> approaches to the notion of "aboutness": the systematic approach
>> versus the encyclopedic approach.
>>
>> In the encyclopedic approach, an article is seen as a set of
>> statements about some RWO or FWO. The set of articles is flat,
>> without a structure (apart from alphabetic ordering by the name of
>> the RWO). This idea is driving Wikipedia and DBPedia and is part of
>> their success and effectivity.
>>
>> In the systematic approach, articles are thought of as being about
>> "subject headings". The subject headings together form a rich
>> associative structure, in which the subject headings are connected
>> to each other, which yields groupings that make sense from some
>> perspective or for some purpose. This implies a notion of
>> metamodeling. SKOS intends to capture this notion in a manner as
>> concise and simple as possible.
>>
>> Is the systematic approach useful? That is quite a question to
>> pose. It is certainly easy to make fun of (sometimes overly
>> ambitious) attempts made in the past. See Borges' famous and
>> hilarious taxonomy of animals in terms of those that belong to the
>> Emperor, those drawn with a very fine camel hair brush, and some
>> other such categories (see [1]). Another fundamental critique of
>> hierarchical schemes such as DDC and UDC is the " rhizome metaphor"
>> (see [2] and [3]). A beautiful, thoroughly non-philosophical
>> account of the development of ideas underlying the systematic
>> approach is [4]. That said, I think it is fair to say that the
>> systematic approach has its usefulness, warts and all. Probably
>> especially so in specific, delimited domains.
>>
>> Concluding, the statement that something is about something can
>> mean two fundamentally different things, depending on the spirit in
>> which of the two approaches the statement is made.
>>
>> There are two options for dcterms:subject (and dcterms:type).
>>
>> A. We leave the choice open to the user by not specifying a range
>> for dcterms:subject. The property can mean two different things
>> depending on the context in which it is used: a simple case of
>> polysemy or even homonymy (or punning if you like).
>>
>> B. We urge users to make the choice explicitly by declaring
>> dcterms:subject to have skos:Concept as range. In that case, one
>> would say that the Wikipedia article has foaf:focus Bill_Clinton
>> (to which it does not bear the dcterms:subject relation), while the
>> biography fits under a particular subject heading, so that it bears
>> the dcterms:subject relation to that heading, which in turn bears
>> the foaf:focus relation to Bill_Clinton.
>>
>> It seems to me that option B is to be preferred, independent of
>> your commitments towards either one of the two approaches towards
>> aboutness.
>>
>> The reason for this is data quality. In the old days, we were used
>> to, say, put the customer's date of birth in the field called
>> telephone number because that field was not used by applications
>> anyway. And everybody was happy. Now we put our data models on the
>> Web, hoping to achieve unprecedented levels of interoperability at
>> almost no cost at all. We do this in the realization that we cannot
>> expect every data source to always adhere to every minute detail of
>> the model. This is what Antoine pointed out previously: the
>> philosophy schema.org, which says: " In the spirit of "some data is
>> better than none", we will accept this markup [which does not
>> comply to expectations, JV] and do the best we can."
>>
>> This also means that refined data is better than some data. In
>> other words, the more careful data sources are about which
>> particular properties they use, the more value others can extract
>> from them. (Of course, up to a point, where the distinctions become
>> so subtle that they become difficult to understand.)
>>
>> Under option B, a data source that uses dcterms:subject to relate
>> articles to RWOs (and maybe even foaf:focus to relate articles to
>> subject headings) is not a problem. We can happily use the data
>> source, and even manipulate it using a range of methods to make it
>> as valuable as possible. At the same time, however, a "high
>> quality" data source that does make the distinctions as intended
>> yields the same value or more at lower cost. Under option A, there
>> is no real sense in which this data source is different in
>> quality.
>>
>> So, to drive home the point I am (albeit somewhat laboriously)
>> trying to make: in view of the existence and broad use of
>> foaf:focus, it makes sense to restrict dcterms:subject and make the
>> two disjoint in range. As opposed to leave it completely open to
>> use dcterms:subject either as a synonym of foaf:focus or as
>> something else.
>>
>> I hope I am not ranting. Does this make any sense? -Jan
>>
>>
>> [1]
>> https://en.wikipedia.org/wiki/Celestial_Emporium_of_Benevolent_Knowled
>>
>>
ge [2] https://en.wikipedia.org/wiki/Rhizome_(philosophy)
>> [3] http://rhizomik.net/html/rhizome/ [4]
>> http://www.catalogingtheworld.com/
>>
>>
>>
>>
>>
>> -----Original Message----- From: DCMI Architecture Forum
>> [mailto:[log in to unmask]] On Behalf Of Dan Matei
>> Sent: maandag 17 augustus 2015 11:46 To:
>> [log in to unmask] Subject: Re: dcterms:type and SKOS
>>
>> -----Original Message----- From: Jan Voskuil
>> <[log in to unmask]> Date: Mon, 17 Aug 2015 07:55:22 +0000
>>
>>>
>>> To express the relation between a metamodelling concept
>>> (":NapoleonConcept") and the RWO/FWO (":NapoleonBonaparte"),
>>> foaf:focus fits the bill quite nicely. (((---On a side note: I
>>> think that there should be an equivalent of this property within
>>> the SKOS-namespace.---)))
>>
>>
>> I have difficulties to understand the practical usefulness of the
>> distinction "NapoleonConcept" vs. RWO/FWO "NapoleonBonaparte" :-(
>>
>> How "NapoleonConcept" fits in the definitions:
>>
>> S: (n) concept, conception, construct (an abstract or general idea
>> inferred or derived from specific instances)
>> http://wordnetweb.princeton.edu/perl/webwn?s=concept&sub=Search+WordNe
>>
>>
t&o2=&o0=1&o8=1&o1=1&o7=&o5=&o9=&o6=&o3=&o4=&h=
>>
>> or
>>
>> A SKOS concept can be viewed as an idea or notion; a unit of
>> thought. However, what constitutes a unit of thought is subjective,
>> and this definition is meant to be suggestive, rather than
>> restrictive. http://www.w3.org/TR/skos-reference/#concepts
>>
>> ?
>>
>> Of course my idea of "Monica Bellucci" differs (somehow) of the
>> real Monica Bellucci :-) However...
>>
>> Yes, I can see the usefulness of the distinction between different
>> catalographic identities (as subjects), such as:
>>
>> Mark Twain vs. Samuel Langhorne Clemens
>>
>> Charles Lutwidge Dodgson vs. Lewis Carroll
>>
>> Enea Silvio Piccolomini vs. Pius II
>>
>> or even:
>>
>> Bill Clinton (as himself) Bill Clinton (as governor of Arkansas)
>> Bill Clinton (as president of USA)
>>
>> But to consider them concepts ? Useful ?
>>
>> Dan Matei
>>
>
> -- Karen Coyle [log in to unmask] http://kcoyle.net m: 1-510-435-8234
> skype: kcoylenet/+1-510-984-3600
>
--
Karen Coyle
[log in to unmask] http://kcoyle.net
m: 1-510-435-8234
skype: kcoylenet/+1-510-984-3600
