As a background to this post, I want to reference a post by Bob called "SubclassOrNot". I discovered this page on an early foray into the TDWG website labyrinth and it has been very influential on my thinking since then. The idea Bob discusses is central to what I'm writing below so if you haven't read it you might want to do so first. You can probably skip the "OWL Inference" section and still get the point which is described in the first two sections of his post. The URL for the page is http://wiki.tdwg.org/twiki/bin/view/TAG/SubclassOrNot .
To preface what I'm going to say below, I want to put Darwin Core Occurrences in the context of what Bob wrote. In my mind, one of the hallmarks of the Darwin Core standard and one thing that makes it a great improvement over previous versions is that the decision was made to use what Bob called the "has a" approach rather than the "is a" approach. In particular, the Darwin Core standard has a single class called dwc:Occurrence rather than subclasses called "Specimen", "Observation", and other possible things. The way that we differentiate among different kinds of Occurrences is by using the DwC types which are the controlled values for the term dwc:basisOfRecord. Thus we say an Occurrence "has a" basisOfRecord=PreservedSpecimen rather than saying it "is a" PreservedSpecimen. We say an Occurrence "has a" basisOfRecord=HumanObservation rather than saying it "is a"HumanObservation". This approach has greatly reduced the number of different terms in the standard since we don't have to have separate "ObservedBy" and "CollectedBy" terms, but rather can just have a single "RecordedBy" term that applies to both specimens and observations. The same thing applies to many other things, like eventDate rather than DateCollected and DateObserved, locality rather than collectionLocality and observationLocality, etc. With the ratification of Darwin Core, this decision is now a fait acompli and not a subject of discussion or something optional for users of the standard. It also seems to be clear that as necessary new terms can be added to the DwC types which would then be valid controlled values for basisOfRecord.
Since the adoption of the DwC standard, the approach to Occurrences has been what I would describe as "I know an Occurrence when I see one". I consider this as a pretty sloppy practice and as I indicated in my post last night, I think there is enough consensus about what an Occurrence is that we can come up with a better definition than "an occurrence is the category of information pertaining to evidence of an occurrence...". Another part of what I would characterize as sloppiness is the lack of a clear definition of what exactly basisOfRecord means. When I wrote my attempt at summarizing consensus last night, I dodged the question about what I called the "token". This "thing" has been called various names. In the previous discussion on the list, it was sometimes called "the evidence" of the occurrence. In the past I have called it "a representation" - however, I now think the term "token" is better because "representation" has a different technical meaning in the context of content negotiation. When we type an Occurrence by saying it has a basisOfRecord=PreservedSpecimen, we are saying that this Occurrence has as supporting evidence, or as a "token" if you prefer, all or part of the dead remains of the organism (i.e. what I'm calling "the Individual") that was being documented by the Occurrence. When we type an Occurrence by saying it has a basisOfRecord=LivingSpecimen, we are saying that this Occurrence has as a "token" the entire organism that was being documented (or some vegetative part of the live organism that was propagated). When we type an Occurrence by saying it has a basisOfRecord=HumanObservation, we are saying that the Occurrence has no supporting evidence other than the reputation of the observer to accurately record the metadata about the Occurrence. In other words, we "tag" a instance of a core class (to use Bob's words), Occurrence, by telling a metadata consumer what kind of token we are using as evidence of the Occurrence.
A fundamental part of creating a clear definition of what an Occurrence is, is to define exactly what we are including in the concept of Occurrence. One possibility is to (1) say that the two boxes at the right side of the diagram at http://bioimages.vanderbilt.edu/pages/occurrence-diagram.gif are fused and that both the Occurrence metadata and its associated token are what we consider to be "the Occurrence". Another approach (2) would be to say that the actual Occurrence as an entity is only the metadata part and that the token is a separate thing. A third approach is to say (3) that everything with the blue dotted lines is considered a part of the Occurrence (i.e. the metadata, the token, the event, and the locality). I don't think in an absolute sense, any one of these approaches is "right". The problem is that these approaches are used inconsistently, sometimes even by the same person, depending on the basisOfRecord. Differences in ways of thinking about this issue is a part of why people aren't understanding the way other people are approaching the structuring of metadata. I have tried to consistently take the approach (1) that the two boxes on the right are fused, i.e. that the Occurrence metadata and the token should both be considered part of the entity that we call "an Occurrence". I think this is why Rich was confused in http://lists.tdwg.org/pipermail/tdwg-content/2010-October/001666.html when I said that it was "wrong" to assert that a scientific name is a property of an Occurrence - obviously it is silly to say that the token (photons on a film, sound patterns in a digital file) has a scientific name. Yet that is exactly what people do routinely when the token is a branch cut off a tree and glued to a piece of paper. They say that they are "identifying a specimen". What I am asking (actually demanding) is that the TDWG community get its act together and come to some consistency on this. If we are going to take the approach (2), then we need to take specimens off their pedestal and treat them like we do any other token that we are using as evidence that an Occurrence happened. If we are going to do what was suggested for the BioBlitz in http://lists.tdwg.org/pipermail/tdwg-content/2010-October/001603.html, i.e. to call Occurrences "observations" and then link the tokens to them by associatedMedia, ResourceRelationship, or some other means (approach 2) then do it consistently for every kind of token, including specimens, and don't single out media tokens for punishment.
I have in a sense "thrown down the gauntlet" on this issue by proposing that DigitalStillImage be added as a DwC type and as a controlled value for basisOfRecord (http://code.google.com/p/darwincore/issues/detail?id=68). I know what some people are going to say in response to this proposal. "Why do you need to have 'DigitalStillImage' as a value for basisOfRecord when you can just say that the resource's dcterms:type=StillImage?" The answer goes back to Bob's point. If we are going to go the "has a" path (which we already have in DwC for Occurrences) rather than subclassing everything, then we need to provide an appropriate value for the "tag" for any type of resource that a reasonable number of users will want to use as a token. I think it is clear from this and other Bioblitzes, my work in Bioimages, the whale tracking project, and many other examples, that there are plenty of people who are already using DigitalStillImages as tokens and we all need a controlled value to use for basisOfRecord.
The other thing that we accomplish when we type an Occurrence by its basisOfRecord is to tell a consumer what kind of metadata to expect to get about the token in addition to the generic metadata that is provided for all Occurrences. Thus for a LivingSpecimen we expect to be told what zoo, botanical garden, bacterial collection, etc. contains the specimen. For a PreservedSpecimen we expect to be told the preparation type, the location of the repository, etc. For a DigitalStillImage we expect to be told the file type, accessURL, etc. Simply providing a value for dcterms:type=StillImage doesn't indicate whether the image is a physical one (i.e. on film) or a digital one. It is also unreasonable to expect a client to have to be checking two different terms (basisOfRecord and dcterms:type) to find out what they could learn from one (basisOfRecord). Of course it would be advisable to provide a value for dcterms:type as well for clients outside the biodiversity community who may not "understand" what basisOfRecord means.
I hate to keep bringing my posts back to the RDF issue, but thinking about how one would write RDF forces clear thinking about how metadata should be structured. If we intend to separate tokens as entities from their associated Occurrence metadata, i.e. approach (2), then we open up a whole other can of worms. To associate the occurrence resources (i.e. the metadata) with the "different" resource (i.e. the token), we will have probably have to be able to create URIs for the tokens and separate RDF metadata blocks which will have to be rdfs:type'd. What are we going to use for that rdfs:type - create another Darwin Core class? I simply don't think that is a complicated road that we want to travel. It would be far easier to just say that every Occurrence has a one-to-one relationship with its token (which could be "the empty set" for observations). This would not work for people who want to hang multiple tokens on a single observation event, but I think that itself is a bad idea because it makes it even harder to have "flat" occurrence datasets. Just say that every time we collect a different token (or make an observation that has no token), it is a new Occurrence record. Realistically, a single collector can't actually take a picture of a plant at the same time he or she collects it for a specimen anyway. Those really should be considered two different events because they happen at different times.
OK, enough said. Consider this my defense of my proposal "issue 68" to add DigitalStillImage. I would urge the powers that be to respond to the issues that I've raised here before having any kind of "vote" (or whatever is ultimately going to happen when there is an up or down decision about the proposal).
Steve
Steve Baskauf wrote:
After the flurry of emails recently, I had an opportunity to carefully read all the way through the threads again, followed by enforced "think time" during my long commute. I was actually pretty cheerful after that because I think that in essence, most of the conversation about what constitutes an Occurrence really boils down to the same thing. So I have sat down and tried to summarize what seems to me to be a consensus about Occurrences. To follow my points, please refer to the diagram at: http://bioimages.vanderbilt.edu/pages/occurrence-diagram.gif
Consensus on relationships
- The fundamental definition of an Occurrence involves evidence that a
representative of a taxon occurred at a place and time. Note 1.A: For clarity, I have modified John's statement in his last email by replacing "taxon" with "representative of a taxon". I'm considering a taxon to be an abstract concept that is applied to individuals or groups of organisms. Note 1.B. This definition is far more useful than the official definition of the class Occurrence "The category of information pertaining to evidence of an occurrence..." which is essentially circular. Note 1.C: This statement is extremely broad because the evidence could be of many sorts, the representative could range from a single individual to all organisms on the earth, the taxon could be anyone's definition at any taxonomic level, the place could range from a GPS point with uncertainty of less than 10 meters to the entire planet earth, and the time could range from a shutter click of less than one second to 3.4 billion years. 2. The diagram is an attempt to summarize in pictorial form statements and relationships that have been described in the thread. The taxon representative is recorded as existing at a particular time and place (the arrow) and the result is an Occurrence record. That Occurrence record exists as metadata which may be associated with a token that can be used to voucher the fact that the taxon representative existed. That token may be the organism itself (or a living part of it as in a twig for grafting), all or part of the organism in preserved form, an electronic representation such as an image or sound recording, and other kinds of things like tissue or DNA samples. There may also be no token at all, in which case we call the Occurrence record an observation. Based on direct observation of the taxon representative, examination of one or more tokens, or both, some determiner asserts that a taxon concept applies to the taxon representative and as a result a scientific name can be used to "identify" the taxon representative. (There may be a lot of other complicated stuff above the Identification box, but that will have to be filled in by the taxonomists.) Note 2.A: I have mapped onto this diagram the letters that John used in his last email to refer to entities that are involved in an Occurrence (T, E, L, O, and G). I will beg the forgiveness of fossil people because I don't really know how the geological context fits in. I'm assuming that it is a way of asserting time and location on a much broader scale than we do for extant organisms. Note 2.B: I have put a dotted line around the part of the diagram that I think includes all the things that people might consider part of the Occurrence itself. I have left out "T" and the other parts related to identification because it seems to me that you can have an occurrence that you document which does not yet (and perhaps never will) have an identification. The Occurrence still asserts that a taxon representative existed at a time and place; we just don't yet know what the taxon is. 3. The red lines indicate the relationships that connect the various entities (I'm going to go ahead and call them resources). Consistent with popular opinion, the Occurrence record is the center of the universe and most things are connected to it. Note 3.A: I am sticking to my guns and refuse to connect the Identification directly to the Occurrence. It is the taxon representative that is being identified, not the occurrence. One can assert another sort of relationship between the identification and the occurrence if one wants to say that one consulted the occurrence metadata and token in order to decide about the identification, but it is not correct to say that the Identification identifies either the Occurrence metadata or the token (as Rich pointed out).
OK, so that's step one - defining what is related to what. If anyone disagrees with these relationships, please clarify or create your own diagram.
Complicating circumstances/caveats
- It is noted and recognized that some users will not care to include
all of these relationships in their models. In the interest of simplification or "flattening" the relationships, they may wish to collapse some parts of this diagram (e.g. incorporate time and location metadata within the Occurrence metadata rather than considering them separate resources, applying scientific names directly to the taxon representatives without defining a taxon concept or recording the determination metadata, connecting identifications directly to the occurrence, etc.). This doesn't mean that the relationships don't exist, it just means that some users don't care about them. 2. It is recognized that different users will be interested in or able to specify the various resources to differing degrees of precision. Examples: A photographer might record times to the nearest second, a collector may only be interested in noting the date on which a specimen was collected. A location may be specified to the precision of a GPS reading or be defined as some geographic or political subdivision. The taxon representative may be an individual organism, a flock or clump, or some larger aggregation of taxon representatives.
That's step two. If I've missed any complications, please point them out.
My opinions about the implications of this diagram
- The circle I've labeled as "taxon representative" is the resource
type that I'm proposing to be represented by the class Individual. You will note that in both the definition of dwc:individualID ("An identifier for an individual or named group of individual organisms...") and the proposed class definition ("The category of information pertaining to an individual or named group of individual organisms represented in an Occurrence"), groups of individual organisms are included. Thus John's example of a fossil having myriad individuals, or Richard's examples of thousands of plankton, a large school of fish, herd of wildebeest, flock of birds, could all be categorized as "Individual" under this definition if there is a reasonable expectation that all of the individuals in the group are members of the same taxon. Perhaps there is a better name for this resource, but since dwc:individualID was already extant, I chose Individual as the class name for consistency with the pattern established with other classes and their associated xxxxID terms. 2. Although in note 1.C. I have given the ranges of the various resources to their logical extreme (as was done previously in the thread), I think that as a practical matter we can adopt guidelines to set reasonable values for the "normal" ranges of the resources. One such guideline might be that we suggest a range that can accommodate about 95% of the user needs within the community (this came from Rich's comment about satisfying 95% of the user need with an establishmentMeans controlled vocuabulary). For example, it was suggested that the range for the location of an Occurrence could span the entire planet Earth. True enough, but virtually nobody would find such a span useful. 95% of users would probably find a range between a GPS reading with 10 meter precision and the extent of a county or province useful for recording the location of an Occurrence. I can suggest similar "useful" ranges: one second to one day for an event time (excluding fossils), one individual organism to the number of organisms that would fit within a 50 meter radius for an "individual", and taxon identified to family for plants and maybe mammals, genus for birds, and order for insects. So framing the definition of an Occurrence in these terms it would be something like: "An occurrence involves evidence (consisting of a physical token, electronic record, or personal observation) that a representative (ranging from a single individual to the number that would fit on a football field) of a taxon (hopefully identified to some lower taxonomic level) occurred at a place (determined to a precision between that of a GPS reading and the size of a county/province) and time (spanning one second to one day)." A few people might object to this level of restrictiveness, but I would guess that it would make 95% of us happy. 3. With the exception of the "missing" class Individual, every resource type on this diagram except for the "token" and Scientific name has a Darwin Core class. Every resource type on the diagram except for "token" has a dwc:xxxxID term that can be used to refer to a GUID for the resource. The implication of this is that any resource on this diagram except for the token and taxon representative (i.e. Individual) is ready to be represented in RDF by Darwin Core terms in the sense that the relationships (red lines) can be represented by the xxxxID terms and that the resources can be rdfs:type'd using Darwin Core classes. (Lacking a class for the scientific name doesn't seem like a big deal to me since the scientific name can be a string literal - but then I'm not a taxonomist.) 4. OK, I've avoided it as long as I can, so I'm going to confess now to the RDF-phobes. The red lines and shapes are something pretty close to an RDF graph. What that means is that if the community can agree that this diagram correctly represents the relationships among the kinds of biodiversity resources that we care about, then the matter of providing guidelines on how to represent Darwin Core in RDF suddenly gets a lot simpler. Just convert the "picture" of the RDF graph into XML format and we have a template. Alright, that's an oversimplification, but I think it is essentially true because the most difficult part of achieving a consensus on RDF representations is to decide how we connect the resource types, not on the literals that we hang onto resources as properties. 5. While I'm beating the RDF drum again, the importance of my opinion number 2 can be extended into the GUID adoption process. In my comments to Kevin about the Beginner's Guide to Persistent Identifiers, I think I commented on the question of how one decides whether a GUID needs to be assigned to something or not. I believe that the answer to that question boils down to this: we need a GUID for any resource that will be referenced by more than one other resource. Do we need to be able to assign a GUID to Taxon concepts? Yes, because it is likely that many identifications will want to reference a particular taxon concept. Do we need to be able to assign a GUID to an Event? Maybe or maybe not. If every occurrence has its own separate time recorded, then no GUID is needed because the time is just a part of every separate occurrence record. If the event is defined to be a time range that represents a collecting trip, then there may be many Occurrences that are associated with that trip and all of them could reference the GUID for that event rather than repeating the event information for every Occurrence. The point here is that every shape (class of resources) on this diagram at least has the POTENTIAL to be a node connecting multiple resources and therefore should have the capability of being assigned a GUID, having its own RDF record, and being appropriately typed (presumably by a DwC class). So this is a final technical argument for why we need to have the DwC class Individual. Whether or not people ultimately choose to assign GUIDs to particular resource types or not is their own choice, but they need to at least be ABLE to if they need that resource to serve as a node given the structure of their metadata.
We need to clarify how the "token" thing fits in, but I'm stopping there for now. I would very much appreciate responses indicating that:
A. you agree with the diagram and connections (and consider this definition and diagram a consensus) B. you disagree with the diagram (and articulate why) C. you provide an alternative diagram or explanation of the relationships among the classes related to Occurrences.
Thanks for you patience with another tome. Steve
-- Steven J. Baskauf, Ph.D., Senior Lecturer Vanderbilt University Dept. of Biological Sciences
postal mail address: VU Station B 351634 Nashville, TN 37235-1634, U.S.A.
delivery address: 2125 Stevenson Center 1161 21st Ave., S. Nashville, TN 37235
office: 2128 Stevenson Center phone: (615) 343-4582, fax: (615) 343-6707 http://bioimages.vanderbilt.edu
tdwg-content mailing list tdwg-content@lists.tdwg.org http://lists.tdwg.org/mailman/listinfo/tdwg-content .