[tdwg-content] [tdwg-tag] Idea for Discussion, Differentiating between "type's" of identifiers
morris.bob at gmail.com
Tue Oct 19 15:21:32 CEST 2010
It is very important to keep the words "Linked Data" and "Linked Open
Data" clearly in the conversation and properly used. Neither of them
is equivalent to the less well-defined "Semantic Web".
Whether temporarily or not, neither LD and LOD address specific some
use cases that are important for other semantic applications. As an
example, some of the recommended practices for LOD do not currently
support tractable reasoning either fundamentally or with current
reasoners. Intractable reasoning means, among other things, that it is
possible to launch queries that will never complete, and for which it
is not possible to know in advance whether that is the case or not.
The current conversation is treading on rather deep issues, some of
which are on the bleeding edge of Knowledge Representation research.
Premature decisions or KR-naive decisions will likely revisit the
history of Darwin Core itself. That is, a tremendously useful solution
will go a long way, provoke misuse along the way, and then come up
against a stone wall requiring a major multi-year re-architecture
effort, perhaps with huge expense to retrofit to the previous uses.
For some insight into what the problems are, one could do worse than
read the thread that begins(?) with
thread addresses the current wobbly state of the FOAF ontology, which
to my knowledge still remains without an agreed upon form that
guarantees tractable reasoning.
Also, see especially the bullet points on "Most Notably" in the
Objectives of 1st Workshop on Knowledge Injection into and Extraction
from Linked Data
On Fri, Oct 8, 2010 at 8:07 AM, Thomas Bandholtz
<thomas.bandholtz at innoq.com> wrote:
> Hi all,
> having just joined this list, I find it a great idea to have such an RDF
> Task Group.
> I am going to publish a little species catalog for the Federal Environment
> Agency in Germany as Linked Data, and I am looking for the best way to
> express it in RDF.
> In the Linked Data cloud  I find several related contributions, such as
> Geospecies, TaxonConcept, EUNIS, and more.
> Comparing these approaches I prefer the idea of reusing SKOS  labels and
> hierarchical relations, as in the Geospecies example .
> It might be a good idea to apply the SKOS XL extension as well to go deeper
> into the taxon name properties.
> Finally, I would add the taxon ranks as a distinct concept scheme and link
> them to the taxon concepts with a mapping relation.
> Certainly this is not the only way to go, but it is rather simple and will
> be easily understood as SKOS is quite common in the Linked Data community.
> This might be called "Simple Darwin Core" and give room for more complex
> ontology approaches beyond that.
> Looking forward to discussion,
>  http://lod-cloud.net
>  http://www.w3.org/TR/2009/NOTE-skos-primer-20090818/
>  http://lod.geospecies.org/ses/v6n7p.rdf
> Peter J. DeVries
> Am 07.10.2010 19:29, schrieb Blum, Stan:
> Hi Steve,
> Sorry, I missed your message below (as well as your response to Roger)
> before I sent my reply about the utility of an RDF guide for DwC.
> Obviously, I think it’s a great idea. To do this within the “normal” TDWG
> process, this should be done as a Task Group. I could help you draft a
> charter for that, which would then need to be reviewed by the TAG and Exec.
> Once approved, we would put the charter up on the web site, and do our best
> to provide any other resources that would help speed the task. I don’t mean
> to slow you down. The Charter doesn’t have to be elaborate. It’s function
> is to let others in TDWG and beyond know that this task is proceeding, who
> to contact, how to get involved, etc. It also gives you the backing of the
> TDWG community.
> Let me know if you’d like to pursue this.
> On 10/7/10 7:41 AM, "Steve Baskauf" <steve.baskauf at vanderbilt.edu> wrote:
> I agree that it is best to avoid a proliferation of terms and I agree that
> it is best to keep Darwin Core technology independent to the maximum extent
> possible. However, I think that the case of facilitating HTTP URIs is a
> special one because of the requirements of GUIDs/Persistent Identifiers.
> Both the TDWG and GBIF guidelines such as they currently stand say that
> GUIDs must be resolvable, that in their resolution they must return RDF, and
> that the RDF has to be in an XML format. Like it or not, that is what we
> have. Given the amount of time that it seems to have taken to settle on
> that much, I think it is best for us to decide to live with it, warts and
> all, rather than re-opening the discussion and delaying the implementation
> of GUIDs for another five years.
> Given that assumption, there needs to be within Darwin Core some way to
> support this particular "technology" (Linked Data, RDF/XML) even if we don't
> do "special" things to support other technologies such as LSID, DOI, etc.
> The point is well taken that most of those other technologies have
> mechanisms for turning their identifiers into URIs and the aforementioned
> guidelines lay out how owl:sameAs can be used within the RDF to associate
> the non-HTTP-resolvable forms with the URIs. Based on my admittedly limited
> experience with trying to write RDF using Darwin Core terms, I think that in
> most cases there already exists appropriate terms for getting the job done.
> What may be lacking is concrete examples and community consensus on what
> terms to use for what. I also think that there are probably some "ID" terms
> where it isn't really very important (from an RDF point of view) that there
> exist both a URI form and a text string form. I'm thinking of something
> like dwc:identificationID, which is mostly likely to be needed to allow a
> machine to make a connection between some resource and its identification.
> The machine isn't going to care if there is a human-readable version. In
> contrast, something like dwc:collectionID is likely to need both a URI
> version (e.g. proxied version of the BCI LSID) for the machines and a string
> version (the name of the collection as it would be displayed) for humans. I
> think that trying to make example/template RDF for various types of
> resources will help make it clear in which cases one version (URI), the
> other (string), or both are actually necessary.
> I "volunteered" a couple weeks ago to have a go at writing an RDF guide for
> Darwin Core. I am still willing to do this, although I'm still getting
> caught up at work from being at the TDWG meeting. However, next week we
> have fall break and I will make it a priority to come up with a draft which
> can be the subject of discussion. As a part of this process, I think it
> would be good to create one or more "boilerplate" RDF files for the various
> kinds of resources that are likely to be identified with GUIDs (e.g.
> Occurrences, Taxa, etc.). This can also be a subject of discussion and I
> think it will help to clarify what will meet the actual needs that we have
> discussed in this thread. I have a pretty clear picture of what I think
> Occurrence RDF should look like. I'm going to have to depend on Pete and
> others to deal with the taxonomy part.
> Markus Döring wrote:
> Steve, Pete,
> Id like to draw your attention on a basic DarwinCore design pattern. Dwc has
> the goal of being technology independent by simply providing a list of
> abstract terms one can use in various arenas such as xml, rdf, xhtml, csv
> etc. And even within those there might be various ways of using them (e.g.
> we have a normalised and a simple flat xml schema), thats why we should have
> a guideline for each of them on how to use them. We are missing such a
> guideline for rdf currently, hence this debate.
> Whether scientificName is a literal string or some complex object shouldnt
> matter - its defined to be a scientific name. Such a dwc rdf property could
> either hold a literal string or a url to some name rdf:resource (potentially
> with a rdfs:label).
> With the introduction if many ID terms we have diluted that idea a little
> already in my mind. We could have as well used scientificName in xml to hold
> some identifier for that name. All URNs tell you what they are by their urn
> prefix (not necessarily how to resolve them), so you can easily detect a
> UUID, LSID, http(s) url, ftp, doi and apply the conventional resolution
> mechanism. The hardest problem are the local ids and other plain identifers.
> For those mainly we created the ID terms (at least in my mind). I am feeling
> rather uncomfortable discussing the introduction of specific dwc terms for
> each type of id. Maybe we should remove all id terms in dwc and use the
> specific guidelines to specify these? At least if you really think having
> all those id terms for rdf is a good thing I would feel much more
> comfortable going down this route instead of diluting dwc by adding more and
> more rather redundant terms. The abstract concept is key to a dwc term, not
> the actual data type fo
> rced by the technology you are using it with. Would you want several date
> terms for various date formats? In fact we do that already to some degree
> (eventDate, eventTime, year, month, day, verbatimEventDate) and I always
> felt this is not a good idea. There are also a number of verbatimXXX terms
> in dwc which also contradict this pattern.
> Talking about new dwc terms - in the examples given properties like
> "hasScientificName" is not strictly the correct dwc term, which is simply
> scientificName. I think it would be fine to have the convention in the rdf
> guidlines to use hasDwcTerm instead of dwcTerm, this is exactly what an rdf
> guideline is for. On the flip side I am sure this only applies to some
> terms, recordBy for example is likely to remain as it is. Its unclear to me
> what is best to do really. Always stick to the original dwc terms? Refine
> them through some rdfs or owl schema and define the relation to the original
> term? Should we still use the same namespace in this case?
> As an rdf beginner even after a few years exposed I wonder if we cant simply
> stick to the non ID terms and use them either as literals or with a uri
> pointer. As in the rdf world a resolvable http is really required for
> resource relations to work, why not simply mandate this in the guidelines?
> If you only happen to have non resolvable uris like lsid or dois the
> guidelines should be asking you to use proxied versions, knowing it will
> break rdf frameworks and lod conventions otherwise. On the resolving side
> one could always include such urns with owl:sameAs (or sth alike) I believe.
> But how many non resolvable ids with no matching http counterpart are really
> out there yet?
> - Markus
> On Oct 6, 2010, at 9:02, Peter DeVries wrote:
> Hi Steve,
> You are probably right that it might be best to use rdfs:Label, but I am
> thinking we might be able to get the same
> result my defining the string variants as subproperties of rdfs:Label.
> This would make them an rdfs:Label but a special kind of rdfs:Label.
> This is one of those things that I would test with Sindice and URIburner to
> see if they interpret these correctly.
> This would require a live vocabulary that Sindice could look at to determine
> that hasScientificName is to be
> treated as a rdfs:Label.
> - Pete
> On Mon, Oct 4, 2010 at 10:41 AM, Steve Baskauf
> <steve.baskauf at vanderbilt.edu> <mailto:steve.baskauf at vanderbilt.edu> wrote:
> Although this specific example deals with taxonomic name identifiers, it is
> related to a previous discussion on this list about how we should use the
> dwc:xxxxxID terms and other terms (such as recordedBy and identifiedBy) that
> could have either a string (literal) or URI form. Although I don't really
> want to see an unnecessary proliferation of Darwin Core terms, I think that
> in the interest of clarity (particularly where RDF is involved) there either
> should be multiple terms that make it clear what form of identifier is
> expected, or else there should be an understanding that in RDF the default
> for such a term is a URI which would then have an rdfs:Label property which
> was the string form. I think the former would be preferable to the latter.
> I came to this opinion when trying to write RDF describing an herbarium
> specimen. The collector should be the dwc:recordedBy property of the
> specimen. Optimally, there would be a database in which known collectors
> were assigned URIs so that "Glen N. Montz", "Glen Montz", "G. N. Montz",
> etc. would all be different labels for the same resource. However,
> realistically, I'm not going to drop what I'm doing to set up such a
> database (even if I were capable of doing it, which I'm not). So I ended up
> just writing it as <dwc:recordedBy>Glen N. Montz</dwc:recordedBy> even
> though I knew it wasn't probably the best thing. In a large Occurrence
> database that was compiled from the RDF created by a lot of people, there
> might end up being a mixture of strings and URIs for dwc:recordedBy
> properties of the specimens. It seems to me like it would be better to have
> properties like dwc:recordedBy for strings and dwc:recordedByURI for a
> corresponding URI (and I suppose dwc:reco
> rdedByLSID if anyone wants to use it). Of course, this would require a
> number of term additions to DwC and clarification in the DwC documentation
> that the generic version was intended for strings.
> With respect to the example
> I think you are right that (with the possible exception of rdfs:seeAlso)
> there is an expectation that an rdf:resource attribute will be a resolvable
> URI that produces RDF. So
> is probably better.
> Peter DeVries wrote:
> I have been thinking about the following pattern. In part after looking at
> the GBIF vocabulary.
> I am not sure if it is even a good idea but might be worth some discussion.
> For those fields that have both a string and "ID" form maybe the following
> pattern might be useful
> hasScientificName = string form
> hasScientificNameURI = Resolvable LOD compliant identifier
> hasScientificNameLSID = LSID identifier which could be resolvable once you
> add the "http:proxy" <http:proxy> etc.
> This allows all three forms to be included if desired, it also provides a
> hint as to how the field should be interpreted or resolved.
> One group could also provide a mapping service so that each record does not
> need to include all three forms, but would allow systems
> to find the matching LSID for a given URI or vs. versa.
> My concern was that it would be difficult to infer how a scientificNameID
> should be interpreted by other systems.
> Is this an LSD, is it a URI, is it a UUID etc. ?
> This impacts the structure of the RDF.
> * Note that the actual identifiers might not be correct, the example below
> is more about the form of the RDF
> * For instance, I don't think it is probably correct to see the COL LSID as
> just a namestring
> * Also in this example the GNI name does not exactly match the string name
> <dwc:hasScientificName>Puma concolor (Linnaeus 1771)</dwc:hasScientificName>
> Some system may choke on the LSID form assuming that it uses a standard
> resolution mechanism
> So it might be best to use this form
> - Pete
> Pete DeVries
> Department of Entomology
> University of Wisconsin - Madison
> 445 Russell Laboratories
> 1630 Linden Drive
> Madison, WI 53706
> TaxonConcept Knowledge Base / GeoSpecies Knowledge Base
> About the GeoSpecies Knowledge Base
> This body part will be downloaded on demand.
> Thomas Bandholtz, thomas.bandholtz at innoq.com, http://www.innoq.com
> innoQ Deutschland GmbH, Halskestr. 17, D-40880 Ratingen, Germany
> Phone: +49 228 9288490 Mobile: +49 178 4049387 Fax: +49 228 9288491
> tdwg-content mailing list
> tdwg-content at lists.tdwg.org
Robert A. Morris
Emeritus Professor of Computer Science
100 Morrissey Blvd
Boston, MA 02125-3390
Associate, Harvard University Herbaria
email: morris.bob at gmail.com
phone (+1) 857 222 7992 (mobile)
More information about the tdwg-content