Replies: 7 comments 21 replies
-
|
As cataloguing processes move towards entity management, the use and addition of persistent identifiers and URIs in records (or metadata works) is a necessity. Further, enabling linked data and explore task presupposes that third-party identifiers and URIs will be likely used. Library of Congress has published its data in id.loc.gov and has also provided guidelines for the use of the $0 $1 subfields, i.e., Report, FAQs and Wikidata page. LoC differentiates between a RWO and the information resource describing it. LoC suggests that the URIs for the RWOs can be used in $1, while the URIs for the information resource (bibliographic and authority records) can be used in $0. In the MARC proposals related to the use of $0 and $1 subfields, e.g., MARC Proposal No. 2017-08, the previously-mentioned distinction is also made by providing these two definitions:
In these 2 definitions, two modelling issues must be highlighted:
When checking the 1st issue, one can conclude that LoC does not implement this modelling. LoC does not keep separated the information about the RWO and the authority record referring to that RWO. Let's take Homer as an example (Check Homer RDF files.zip). The authority record is an instance of mads:Authority class (which is not considered RWO by LoC), and of skos:Concept class (which is a RWO). It includes both information about Homer (e.g., gender, occupation, associated language) and information about the use of the heading Homer in LoC (e.g., citation sources, LC classifications, editorial notes, use of heading Homer as subject in LoC bibliographic records). The RWO RDF file presents an instance of mads:RWO, bf:Person, and foaf:Person. The description of Homer as a RWO includes information about Homer (e.g., gender, occupation, associated language) BUT the properties used for this descriptions are not from Bibframe (bf does not have such properties), nor from foaf (foaf has appropriate properties). MADS properties are used instead. Checking the MADS ontology, the relevant properties are modelled having as domain mads:RWO. So, from a model's point of view LoC may be consistent, yet IMHO it is strange modelling. Another important issue is that information about the heading Homer in the LoC catalog is maintained in the RWO, e.g. subjectOf. To be honest, I did not expect a subjectOf relationship in the Homer RWO RDF file, but an author relationship relating Homer to his Works. When checking the 2nd issue, one can conclude that there is a semantic problem. SKOS was developed for KOS, concepts and their relationships between them. An authority record is a metadata WORK. So, I believe that it is wrong to model an authority record as an instance of skos:Concept. The LoC practice that skos:Concepts are not RWOs originates from this wrong modelling. In relation to our mapping exercise, I propose
PS. I have not touched the issue of what is a RWO. To me it is something that can be dereferenced. Since there is not currently a tool to check if and how id.loc.gov URIs are dereferenced, I have focused on the modelling part of this discussion. |
Beta Was this translation helpful? Give feedback.
-
|
Sofia, I agree with a lot of this, but have some comments and one big difference. I have to leave the meeting early on Wed. (if it happens), so here are my thoughts
Nothing prevented LOC from making a RWO IRI and associating it with the SKOS Concept/MADS Authority IRI in the data about that IRI in the way they did for names:
when creating data for subjects and other vocabularies. But, they didn't. |
Beta Was this translation helpful? Give feedback.
-
|
I already offered some thoughts in regards to skos:Concept and how I think LOC uses it correctly. SZ: " I did not expect a subjectOf relationship in the Homer RWO RDF file, but an author relationship relating Homer to his Works." I fear in responding to the more proposals, I may be somewhat out-of-my-depth, but here goes: You probably know by now that I don't agree with your "4. Concepts from VES." Concepts in KOS's are mostly not RWOs -- of course it depends how they're modeled by the data producer. But the SKOS reference is pretty clear that SKOS was intended to be used that way; for example, it says, among many other things, that skos concepts "...cannot be reliably interpreted as either formal axioms or facts about the world... they serve only to provide a convenient and intuitive map of some subject domain" (that's in infamous section 1.3). Although consider Laura's remark, "SKOS properties like "broader", "inScheme", "historyNote", "preflabel" don't really have meaning outside the context of a specific knowledge organization scheme." If you ask me, that's a pretty good clue that we're not talking about an RWO. |
Beta Was this translation helpful? Give feedback.
-
|
Thank you for beginning this discussion. I don't think we can say we've properly discussed RWOs if it doesn't involve some pain, as it's a beastly topic. Anyway... I disagree with the direction our group has taken that maintains that instances of skos:Concept are necessarily RWOs. The SKOS specification allows ambiguity here, so we do seem to have a choice of creating a vocabulary where URIs are assigned to RWO concepts typed as skos:Concept. However, anything typed as a skos:Concept is not necessarily a real world concept; in fact, I would argue that a skos "concept" is a "concept" in a knowledge organization system and, as such, more appropriately considered documentary in nature. When we retrieve the document -- and this is complicated by the recommendation (Architecture of the World Wide Web, 2004) that a web document is actually content that has, potentially, multiple representations (an html representation, an RDF/XML representation, a Turtle representation, etc.) -- that is exactly what we are retrieving, "content" describing an entry in a KOS. That KOS is a document presenting a model that helps us organize a domain of knowledge but is not a "knowledge model." If you read the SKOS specification's introduction -- especially section 1.3 -- I don't how anyone can maintain that anything typed skos:Concept should be understood as a RWO; in fact, I would expect the reader to emerge thinking a skos:Concept is not an RWO. I would be very pleased if someone could show me one authoritative document that informs us that a skos:Concept is a RWO. You can find affirmations that concepts are RWOs, but a skos:Concept is not the same thing as a concept in the world. |
Beta Was this translation helpful? Give feedback.
-
|
Reading the SKOS spec has me agreeing with this, Theo. I don't think we can use the class skos:Concept to indicate anything about the "realness" of the thing it classifies. The problems I have with things classed and treated as a sort of mixture between an authority and a RWO remain problematic, though. |
Beta Was this translation helpful? Give feedback.
-
|
What would be an example of a thing classed and treated as a sort of mixture between an authority and a RWO? |
Beta Was this translation helpful? Give feedback.
-
|
Any NACO record was what I was thinking of, but that also applies to BIBFRAME works/hubs/expressions. The way it's set up, there are names and variant names and authority-type data, admin metadata about the record itself, and data about the "thing", such as name of publisher/place of publication. Even when a distinct RWO URI is provided for things, as in NACO, which triples apply to which? If an authority IRI and RWO IRI are assigned to the same description set, the more important question may be, "does it matter"? |
Beta Was this translation helpful? Give feedback.
-
|
What is an RWO. I would say that I take the views of a day-to-day librarian who just wants to apply the rules. I think there are more ambitious people that want to change the architecture of the web. In that imaginary, I believe that saying a RWO is something that can be dereferenced is an attempt to rethink our architecture. In the current architecture, there's only one "thing" that can dereferenced and that is a web document. In the current architecture, there are only two things in the world: (1) a web document, (2) everything else. Only web documents can be exchanged on the web. We call everything else a "real world object" and we cannot exchange it over the web; instead, we offer up some description in a web document. We've been given recommendations on how to do this (Cool URIs, 2008): we either redirect to a web document when RWOs are requested over HTTP, or we create fragment identifiers (hash URIs) in a web document where we describe RWOs in that document (of course actual practice mostly does not follow these recommendations). So an RWO, in what I call the current architecture, is anything that cannot be dereferenced directly. So I don't understand the idea that an RWO is anything that dereferences, as dereferencing means accessing a referenced resource. We can't access an actual elephant over http, only a document describing the elephant. |
Beta Was this translation helpful? Give feedback.
-
|
Ah, Theo, this is almost a philosophical discussion. It's late and I never took philosophy, but isn't it ontologically correct, that the "realness" of a concept is in my mind, and yours and those of others, if we believe we share the concept? So it's still in a place where we can't dereference it. |
Beta Was this translation helpful? Give feedback.
-
|
Agreed! And what about the concept that's widely-used, like "terrorism", whose meaning changes across social boundaries? Still, I have to say, as interesting and sometimes illuminating as it can be, thinking like this is something I would like to avoid professionally. I just want to apply the "rules." Show me the rules! I keep talking about "web architecture" because that's documented. I really like having a pint of ale and spinning-my-wheels with friends, but when there's a job to do, I sure hope there's a document somewhere with a solution! |
Beta Was this translation helpful? Give feedback.
-
|
Laura's question is the same one on my mind. When we look at something like the LC versions of MARC value vocabularies for the 007/008, are these concepts solely in the sense of a concept in a controlled vocabulary? Or, are they "real" concepts we share in our collective mental space? This is where the RWO/authority boundary gets blurry in my mind and makes it very difficult to figure out what the rules are. |
Beta Was this translation helpful? Give feedback.
-
|
For the record, I think this is connected with the "problem of universals" (see Wikipedia). The Platonic view says there is a 'real' universal concept of, say, "glassness" in the context of "materialness", but I go with the Aristotlean view that "glass" is just a method of categorizing the concept of "material" that is just a value of the predicate "has material". I recall an interesting discussion of this in Pirsig's Zen and the art of motorcycle maintenance ;-) |
Beta Was this translation helpful? Give feedback.
-
|
I don't think we have to define what an RWO is for the purposes of the project, although it is an interesting side discussion and should be part of a basic understanding of RDF and linked open data. [Nonetheless, a couple of points have been made that I think are incorrect:
For the project, we need to know of what RDF type (class/entity) is the instance represented by an IRI. Note that every IRI represents an instance of something (some "thing"). Example: The instance IRIs http://id.loc.gov/rwo/agents/nb2001072552 and http://id.loc.gov/authorities/names/nb2001072552 that are both labelled "Dunsire, Gordon" in id.loc.gov. Both instance IRIs have multiple types. This means the instance is a member of each of the type classes and may be described using typed-domain properties of any of the classes (including their super-classes). The RDF type graphs for each instance are:All I don't think we have to define what an RWO is for the purposes of the project, although it is an interesting side discussion and should be part of a basic understanding of RDF and linked open data. [Nonetheless, a couple of points have been made that I think are incorrect:
For the project, we need to know of what RDF type (class/entity) is the instance represented by an IRI. Note that every IRI represents an instance of something (some "thing"). Example: The instance IRIs http://id.loc.gov/rwo/agents/nb2001072552 and http://id.loc.gov/authorities/names/nb2001072552 that are both labelled "Dunsire, Gordon" in id.loc.gov. Both instance IRIs have multiple types. This means the instance is a member of each of the type classes and may be described using typed-domain properties of any of the classes (including their super-classes). The RDF type graphs for each instance are:
The dotted line property is rdfs:type; the unbroken line property is rdfs:subClassOf. Type statements are entailed upwards; for example: http://id.loc.gov/rwo/agents/nb2001072552 rdfs:type bf:Person . These statements should be sufficient to determine the appropriate RDA class with which to type each instance IRI. This is the same as determining a map from the MARC 21 classes to RDA classes. To determine the mapping from each class we have to compare the semantics of the source and target classes. The semantics are indicated by definitions, etc., subclass declarations, and the properties (in that order and recursively) but are not indicated by class labels. The MARC 21 types involve the following classes and definitions:
The equivalent RDA classes are:
The MARC 21 class definitions are significantly less precise than the RDA definitions. The MADS/RDF definitions are circular, inconsistent, and reference the definitions of other properties in the namespace. In particular, the MADS/RDF RWO definition can be unpacked to: RWO: An abstract entity that identifies a Real World Object (RWO) identified by the label of an idea or notion with a controlled label or a former idea or notion with a controlled label. I think this means that a MADS RWO is any thing that can or does have a label, and any thing that cannot be labelled is not an RWO. I note that a blank node is not an instance of an RWO, because there is no label (IRI to stringify). If so, we can ignore the MADS RWO class because an RDA instance must have an appellation (i.e. label). It seems clear from the definitions that RDA Person has a narrower coverage than BF Person, and therefore: rdac:C10004 rdfs:subClassOf bf:Person . All RDA persons are BIBFRAME persons, but not all BIBFRAME persons are RDA persons, so a transformation from BF/MARC 21 to RDA Person is unsafe. I think this analysis is applicable to all entities with authorities in id.loc.gov. At some point in the chain of classes and definitions, fictitious instances of the entity are included in the scope, so the corresponding RDA entity must be a subclass and the instance IRI cannot be transformed safely. |
Beta Was this translation helpful? Give feedback.
-
|
GD: “An RWO is just the thing (instance) that is represented by an IRI; if IRI, then corresponding RWO.” Also, “every IRI represents an instance of something (some "thing").” GD: “the MADS/RDF RWO definition can be unpacked to: RWO: An abstract entity that identifies a Real World Object (RWO) identified by the label of an idea or notion with a controlled label or a former an idea or notion with a controlled label. I think this means that a MADS RWO is any thing that can or does have a label, and any thing that cannot be labelled is not an RWO.” |
Beta Was this translation helpful? Give feedback.
-
|
@gerontakos: My statement is based on the definition in RDA Terms: real-world object: An instance of an entity or a term from a vocabulary encoding scheme that is referenced by an Internationalized Resource Identifier for representation in Resource Description Framework. The RDA definition is consistent with the MARC 21 Bibliographic definition of $1 - Real World Object URI: "Subfield $1 contains a URI that identifies an entity, sometimes referred to as a Thing, a Real World Object or RWO, whether actual or conceptual. When dereferenced, the URI points to a description of that entity. A URI that identifies a name or label for an entity is contained in $0." Assuming that "Thing" refers to owl:Thing, the OWL Guide 3.1.1 says "Every individual in the OWL world is a member of the class owl:Thing." madsrdf:RWO is explicitly declared to be a subclass of owl:Thing, but this is implicit in the definition of owl:Thing and does not need to be explicitly declared. It may be the case that the intention is to indicate that madsrdf:RWO is a restricted class of RDF resources (RDF Primer: "documents, people, physical objects, and abstract concepts"), but it is not at all clear what is being excluded, other than resources with a controlled label. If "controlled label" indicates a necessary KOS, why not say so? As it is, the definitions are self-referential, and that leads to significant differences in interpretation, I think. I am not saying that madsrdf:RWO is not consistent with web architecture; I am saying that there is nothing useful about madsrdf:RWO outside of itself, so we can ignore this rabbit-hole :-) |
Beta Was this translation helpful? Give feedback.
-
|
The key to understanding the id.loc.gov modelling is the property foaf:focus. This is a super-property of madsrdf:identifiesRWO which relates an authority to its RWO. The domain of the property is madsrdf:Authority and the range is madsrdf:RWO. ex:IRI1 madsrdf:identifiesRWO ex:IRI2 . This is a subpropertyof foaf:focus. The domain of the FOAF property is skos:Concept, the range is owl:Thing. ex:IRI1 madsrdf:identifiesRWO ex:IRI2 . This is why an instance IRI is typed as a MADS Authority and as a SKOS Concept, and why MADS Authority is declared a subclass of SKOS Concept. Otherwise, id.loc.gov cannot use foaf:focus. The "focus" property was added to FOAF in 2010, well after the publication of the original ontology; see FOAF spec revised - addtion of foaf:focus, a skos extension linking topical and factual information. [There seems to be a problem with the FOAF Vocabulary Specification which displays an older version without the "focus" property.] It is a bit strange that this property was added to FOAF and not to SKOS: it has a SKOS class as a domain. I think the original modelling decision at id.loc.gov was to use SKOS irrespective of the distinction between a value vocabulary and a dataset that is discussed in Library Linked Data Incubator Group: Datasets, Value Vocabularies, and Metadata Element Sets. This report was published in 2011. SKOS is not really intended for datasets of instances of entities such persons, works, and places, but it is fine for thesauri and other kinds of value vocabulary of instances of concepts such as carrier types. There is an interesting discussion about much of this, also from 2011, in Things & their conceptualisations: SKOS, foaf:focus & modelling choices. I have always been puzzled as to why id.loc.gov did not consider using the SKOSXL ontology: SKOS Simple Knowledge Organization System eXtension for Labels (SKOS-XL) Namespace Document - HTML Variant. This was published in 2009, and is closer to the idea of LRM Nomen. Finally, foaf:focus has no declared super-property. As @szapoun points out, RDA takes a different approach to relating a description to its subject that avoids confusion between labels, concepts, and subject relationships ("aboutness") by using the property rdax:P00030 "RDA entity described with metadata by" with domain RDA Entity and range (metadata) Work. Definitions:
The RDA definition unpacks to: A work that is a metadata statement or a metadata description set for an RDA entity. |
Beta Was this translation helpful? Give feedback.
-
|
I’m not seeing how understanding the relation to foaf:focus completely clarifies the modeling. I know it’s the historical importance of foaf:focus that Gordon is talking about, but I think it’s an abandoned property in the id.loc.gov vocabularies? I only spot-checked. As for foaf:focus, semweb people wanted it to extend SKOS. SKOS lacked an instrument to link the documentary "concept" to its RWO. As Dan Brinkley wrote (in the document Gordon references), "...a new term, foaf:focus that links a skos:Concept to the thing that the concept stands for." It seems pretty clear just from that what the purpose of foaf:focus was. I don’t know if Gordon is implying that an individual cannot be both a madsrdf:Authority and a skos:Concept. Whatever the case, it doesn’t seem to be a problem, as both authority data and a concept-in-a-KOS are documentary. I find Johnston’s blog entry well-aligned with web standards and recommended practices. Johnston looks at the LOC instances of skos:Concept and distinguishes them from RWOs. They are entries in a KOS. If a property calls for a value that is a RWO then, strictly speaking, the property requires that we do not use the skos:Concept. The skos:Concept uses foaf:focus to link the concept-in-a-KOS (which I claim is documentary) to its RWO (just as we now use madsrdf:RWO). He does recommend an easing of that requirement, but the easing would be enacted by changing properties of properties (mostly by changing ranges), not just by descending into chaos and, say, entering concept-values where RWOs are expected (I believe we're seeing that in current BIBFRAME metadata creation). Finally, this confusion between labels, concepts, etc. may prove more navigable than the use of “RDA entity described with metadata by,” which involves “frbr-izing” web documents to create the metadata w-e-m-i. The example we concocted for mapping MARC 561 looked like an abomination. Personally, I don't think labels, concepts, etc. is all that complicated if understood in the context of the architecture of the (semantic) web, where there are just two types of things: RWOs which cannot be returned over http with a 2XX response code, and web documents, which are the only things that can be returned over http with a 2XX response code. It is an excellent starting point when addressing the problem, and is something that, I admit, I’ve fashioned out of the W3C Technical Architecture Group’s proposed solution to the httprange-14 problem. |
Beta Was this translation helpful? Give feedback.
-
Beta Was this translation helpful? Give feedback.
-
|
Hesitating to weigh in here because of the danger of just adding confusion rather than clarity, but want to say something about @GordonDunsire's statement
As I see it the RDA definition is not consistent with the MARC21 Bibliographic definition of $1. because of that "or" clause. MARC21's $1 definition makes no mention of vocabulary encoding schemes, just an entity. I tend to think the RDA definition is problematic for this reason (hence, these arguments) and should be changed. An entity and a label for it in a scheme are not the same "thing". The label, a (hopefully human-readable) definition, and other data pertaining to an entity IRI may serve to allow machines and humans to have an idea what entity is supposed to be identified by this otherwise meaningless IRI string. The fuzziness in the RDA definition reminds me of the difference between owl:sameAs and the Schema.org sameAs property, which defines sameAs as "URL of a reference Web page that unambiguously indicates the item's identity. E.g. the URL of the item's Wikipedia page, Wikidata entry, or official website" . Lotta "or"s there, and a lot of potential confusion to programmers. Finally, a W3C 2008 email discussion on this topic that I came across (thanks Google) might be interesting. I had never heard of MOAT (meaning of a tag) before... https://lists.w3.org/Archives/Public/public-esw-thes/2008May/0017.html |
Beta Was this translation helpful? Give feedback.
-
|
@lake44me: The RDA definition covers the two kinds of sources of values identified in the W3C LLD report: datasets (instances of entities) and value vocabularies (concepts from a KOS or encoding scheme, including authority files). I think this corresponds to "actual" and "conceptual" in the MARC 21 definition: what is a "conceptual entity" if it isn't a concept and an instance of skos:Concept (if it is part of a KOS). In RDA, a label is an instance of an entity (Nomen), not an instance of skos:Concept. RDA, and LRM, is quite clear on the relationship between a label and its referrant, as is SKOSXL; it is the MADS model of authority files that is fuzzy. A subject authority file is primarily a KOS (a set of concepts) and an agent authority file is primarily a set of instances (of Agent). A name authority file is a set of instances of Nomen. MADS conflates these, and the result is neither a KOS nor a set of instances, but a set (file!) of documents that focuses on part of itself. Of course, a KOS is a particular kind of entity authority file; the entity is Concept. The difference is one of granularity: an entity authority file focuses on the entity, while a KOS focuses on specific groupings of the entity (concept schemes). |
Beta Was this translation helpful? Give feedback.
-
|
RDA definition of a RWO: An instance of an entity or a term from a vocabulary encoding scheme that is referenced by an Internationalized Resource Identifier for representation in Resource Description Framework. |
Beta Was this translation helpful? Give feedback.
-
|
RDA definition of VES: A named structured list of representations of controlled values for elements. A vocabulary encoding scheme includes an RDA list of terms or their corresponding value vocabularies in the RDA Registry, an ISO code list, a standard terminology, an authority control system, etc. Simple keyword indexes are excluded. |
Beta Was this translation helpful? Give feedback.
-
|
I've been looking at 'my' examples in more depth. A de-reference of the id.loc.gov 'name' IRI includes the following statement: http://id.loc.gov/authorities/names/nb2001072552 http://www.loc.gov/mads/rdf/v1#hasCloseExternalAuthority http://www.wikidata.org/entity/Q60661673 . As already noted, the subject IRI is typed as a mads PersonalName, a mads Authority, and a skos Concept. But http://www.wikidata.org/entity/Q60661673 is typed as wd Human. The mads property is defined as "Records a relationship between an Authority and one that is closely related from a different MADS Scheme." and it is declared as an owl symmetric property; the domain and range are mads Authority. I assume that mads Authority (< skos Concept) and wd Human are kinda disjoint, in which case the mads statement is incoherent. Or are we all just figments of the Matrix? Does MADS know something we don't? |
Beta Was this translation helpful? Give feedback.
-
|
"Oh me oh my." They seem to have done this for many authority records. In addition to Gordon's observation, I wonder why they feel at ease re-typing the Wikidata Human as an Authority (and also as a skos:Concept in the HTML). Why would they squat thus? Note: this is an issue we have not resolved ourselves, IRI "squatting." See Issue #336. |
Beta Was this translation helpful? Give feedback.
-
|
I don't find Toolkit examples less confusing. http://id.loc.gov/authorities/names/no91000939 http://rdaregistry.info/Elements/e/P20065 http://vocab.getty.edu/page/aat/300256135 http://rdaregistry.info/Elements/e/P20065 has domain Expression. I made a document last year of all similar examples I could find in the Toolkit. |
Beta Was this translation helpful? Give feedback.
-
Beta Was this translation helpful? Give feedback.
-
|
Agreed. I have never met him but happy to send him an email to invite him to a meeting? |
Beta Was this translation helpful? Give feedback.
-
|
On the subject of asking LC NetDev to come speak with us, see #388 , Questions for LC Net Dev. |
Beta Was this translation helpful? Give feedback.
-
This discussion is related to this discussion $0's and $1's #328
Beta Was this translation helpful? Give feedback.
All reactions