Skip to content
Permalink
Branch: master
Find file Copy path
Find file Copy path
Fetching contributors…
Cannot retrieve contributors at this time
507 lines (342 sloc) 43.2 KB

Beginner's guide to RDF: 4. Vocabularies, DCAM, and RDFS

go to part 3: RDF basics

go to part 5: Discovery, Transmission, and Storage of RDF data

Introduction to the Guide

Contents

4.1. What is a vocabulary?

A vocabulary is a list of terms and their definitions. (1) In this context, "vocabulary" refers specifically to a controlled vocabulary in the sense that terms are included in the list by the declaration of some sponsoring body. Within a vocabulary, the best practice is to have only one term for a single thing and to have only a single definition for each term. Terms are often identified by a URI created from a text string which is generally understandable by humans to be related to the concept that is embodied in the term.

Although different terms may be named by URIs that use the same string after a prefix, e.g. ac:derivedFrom and dsw:derivedFrom, they do not necessarily mean the same thing. The term text strings are qualified by prefixing them in such a way that they form a URI which unambiguously relates them to their defining vocabulary. What each term means is defined in the vocabulary from which they came, and the data in the RDF cannot be interpreted without reference to that definition.

For example, the string "derivedFrom" can represent "a reference to an original resource from which the current one is derived" (the sense in which it is used in the draft Audubon Core standard (2) ). However, "derivedFrom" can also be a term representing the organism or Token from which another Token is derived (the sense in which it is used in darwin-sw (3) ). To differentiate between different definitions of terms, a namespace abbreviation is placed in front of the term to make it unique. In this example, ac: can abbreviate "http://rs.tdwg.org/ac/terms/" and dsw: can abbreviate "http://purl.org/dsw/". So when one writes ac:derivedFrom it is an abbreviation for the URI http://rs.tdwg.org/ac/terms/derivedFrom while dsw:derivedFrom is an abbreviation for the URI http://purl.org/dsw/derivedFrom . In this convention, the namespace provides a way to identify the defining vocabulary while the entire URI provides a globally unique identifier for the term. Often a variety of the namespace itself serves as an identifier for the entire vocabulary, known as the vocabulary URI.

Commonly vocabulary namespaces are either a "hash namespace" or a "slash namespace". URIs for terms in vocabularies that use hash namespaces are constructed by appending a "#" character to the end of the vocabulary URI, then appending the term string. For example, the SKOS core vocabulary has the URI http://www.w3.org/2004/02/skos/core. The URI for the term skos:closeMatch is formed by concatenating the vocabulary URI, "#", and the term name "closeMatch": http://www.w3.org/2004/02/skos/core#closeMatch. URIs for terms in vocabularies that use hash namespaces are constructed by appending the term name to the end of the vocabulary URI. For example, the Dublin Core "terms" vocabulary has the vocabulary URI http://purl.org/dc/terms/. The URI for the term dcterms:subject is formed by concatenating the vocabulary URI and the term name "subject": http://purl.org/dc/terms/subject. The choice of whether to use a hash or slash namespace depends on several factors, including the size of the vocabulary and the method by which the RDF describing the vocabulary will be served. This is described in detail in Appendix B of Best Practice Recipes for Publishing RDF Vocabularies.

Even though a vocabulary may use a URI to identify its terms, it should not be presumed that attempting to dereference the URI using a web browser or semantic client will necessarily return any useful information. If a URI does dereference, the information returned can range from a minimal human-readable text description of the term to a complex RDF description of the properties of that term and how it is related to other terms.

4.2. What is an ontology?

Although there is no strict definition of an ontology, it is generally recognized that an ontology describes the formal constraints of the terms in a vocabulary and expresses relationships among terms in the vocabulary using some kind of ontology representation language. An ontology can describe hierarchical relationships among terms (as in a taxonomy), describe associative relationships among terms (as in a thesaurus), and relate terms within the vocabulary to other terms described outside of the vocabulary. Basic RDF makes few distinctions in the ways that URIs can be used to represent predicates and objects. But other systems have been developed to extend basic RDF and provide ways to indicate more clearly the nature of resources and the relationships between them. RDFS (section 4.4.) and OWL (section 7.4.) provide increasingly complex means of describing such relationships.

There are tradeoffs involved in choosing to describe terms using an ontology rather than a simpler controlled vocabulary. Terms that are described as part of an ontology carry much more meaning and can be used to do more sophisticated machine reasoning (i.e. to "generate" new triples from existing ones). On the other hand, terms that are described by an ontology are far more encumbered and may have a much more restricted utility than terms that are described in a simple controlled vocabulary.

For example, Audubon Core is very vague about the types of resources that can be the subjects and objects of ac:derivedFrom. Because Audubon Core is concerned with media items, one might assume that it is intended to describe how one media item (e.g. an image) is related to another. For instance a particular StillImage might be ac:derivedFrom some particular MovingImage . However, there is nothing in the formal description of ac:derivedFrom that would prohibit someone from making the statement that an offspring is ac:derivedFrom its mother. Thus a simple description of a term allows it to be adapted to uses that were not conceived by the original creators of the term.

In contrast, the darwin-sw ontology defines dsw:derivedFrom to be a transitive property (i.e. if A dsw:derivedFrom B and B dsw:derivedFrom C, then A dsw:derivedFrom C) and is also defined to have the inverse property dsw:hasDerivative (i.e. if A dsw:derivedFrom B, then B dsw:hasDerivative A). So persons who are not careful in using dsw:derivedFrom may end up saying things that they do not intend. The problem of making unintended statements is particularly acute when an ontology declares terms to have range and domain properties (see sections 4.3.2.5. and 4.3.2.6.).

Ontologies and a language to describe them (OWL) are described in more detail in section 7. of this guide.

4.3. The DCMI Abstract Model (DCAM)

4.3.1. What is the DCMI Abstract Model?

In 2007, the Dublin Core Metadata Initiative (DCMI) (4) produced a Recommendation describing an abstract model for Dublin Core (DC) metadata. (5) Because DC is a widely recognized and used vocabulary, the DCMI Abstract Model (DCAM) has been used as a model for other vocabularies outside DCMI, with Darwin Core as a notable example within our community. The proposed Audubon Core TDWG standard is also modeled after the DCMI Abstract Model. As an abstract model, the DCMI Abstract Model is independent of any particular coding scheme. Thus it can be used as a model for vocabularies such as Audubon Core, whose normative definition is in human-readable terms and assumes no particular serialization. Nevertheless, the DCAM was built upon concepts defined in the World Wide Web Consortium's (W3C's) Resource Description Framework (RDF). So even vocabularies which are based on the DCAM but which are not expressed in RDF are structured in a way analogous to the structure of RDF.

4.3.2. Main features of the DCMI Abstract Model

4.3.2.1. Use of URIs and literals

Resources that are described using the DCAM are identified by URIs or are expressed as string literals. The nature of literals can be clarified by applying an ISO language tag or by indicating the structure of the string by associating it with a syntax encoding scheme (e.g. it is an integer or formatted date). See this for details of DCAM literal value surrogates.

4.3.2.2. Resource description set

A subject resource is described by a set of statements. Each statement is made up of a property-value pair.

4.3.2.3. Property-value pair

In a statement about a subject resource which consists of a property-value pair, the property (also known as an element) is a term from the vocabulary (e.g. the DC vocabulary) which relates the subject resource to the value. The value of the statement can be represented by either a URI or a literal. For example a book (the subject) could be described by the property dcterms:title and the literal value "A Tale of Two Cities". Optionally, the literal could have the tag xml:lang="en" or xml:lang="es" for property-value pair dcterms:title "Historia De Dos Ciudades".

The value surrogate (which represents the value resource) could also be a URI as in the property-value pair:

dcterms:type   http://purl.org/dc/dcmitype/Text

where http://purl.org/dc/dcmitype/Text is a URI which identifies the type of things that consist primarily of words for reading.

4.3.2.4. Class

The DCAM defines a special type of vocabulary term called a class. A class is a mechanism defining a group of resources of a particular type. A resource which is a member of a particular class of resources is called an instance of that class.

Note: a common convention is to capitalize the names of classes and to write the names of properties in camelCase. However, this convention is one of convenience and it is not safe to make assumptions about the nature of a term based on its capitalization.

4.3.2.5. has domain

Terms in a vocabulary which are properties that can describe all instances of a particular class are said to have a domain relationship with that class. For example, the property term dcterms:medium has a domain which is the class dcterms:PhysicalResource. So the property dcterms:medium will apply to all dcterms:PhysicalResource instances.

Many DC terms which are properties do not have domain assignments. Thus, although a resource which is an instance of the class dcterms:BibliographicResource may have the property dcterms:publisher, there is nothing which says that a thing which has the property dcterms:publisher must be an instance of the dcterms:BibliographicResource class. However, any thing which has the property dcterms:medium can be inferred to be an instance of the class dcterms:PhysicalResource because of the domain assignment in the definition of dcterms:medium.

4.3.2.6. has range

In some cases, it may be assumed that all values associated with a particular property would be instances of a particular class. This is called a range relationship between the property and the class. For example, if a subject resource has a creator, it can be inferred that the creator is some kind of agent which is capable of creating. Therefore, the definition of the property dcterms:creator states that it has the range dcterms:Agent. So if one makes a statement about the painting Starry Night that

dcterms:creator  people:Vincent_van_Gogh

one can infer that people:Vincent_van_Gogh is a dcterms:Agent

As was the case with domain, there is no requirement that any property term have a range assignment.

4.3.2.7. Sub-classes and sub-properties

The DCMI Abstract Model allows a class to be related to another class by a sub-class relationship. One class is a sub-class of another class if all instances of the first class are also instances of the second class. For example dcterms:FileFormat is defined to be a sub-class of dcterms:MediaType. Therefore, anything which is a dcterms:FileFormat is also a dcterms:MediaType. It is NOT true that anything which is a dcterms:MediaType will also be a dcterms:FileFormat.

The DCAM also allows a property to be related to another property by a sub-property relationship. A property is a sub-property of another property if all instances of the first property also have the same relationship as the second property. For example dcterms:creator is a sub-property of dcterms:contributor. So if the painting Starry Night is described by the property value pair:

dcterms:creator  people:Vincent_van_Gogh

then it can also be known that the property value pair:

dcterms:contributor  people:Vincent_van_Gogh

also applies to Starry Night even if it is not stated explicitly.

4.3.3. Implications of using the DCMI Abstract Model as the basis for a vocabulary

If the DCAM is used as the basis for a vocabulary, one can assume:

  • that the purpose of the vocabulary is to describe resources using sets of property-value pairs.
  • that a described (subject) resource may be identified by a URI, although this is not required.
  • that resources which are values may be described using URI references or literals.
  • that the properties will be identified by qualified URIs.

Although the DCAM assumes the existence of classes, it does NOT require that a vocabulary following its model define any classes. If the vocabulary defines classes the DCAM allows but does NOT require that those classes be associated with any property terms through domain or range assignments.

Although the DCAM was designed to be compatible with RDF, it is a generic information model and does not assume any particular encoding syntax.

4.4. RDF Schema and the relationship of the DCMI Abstract Model to RDF

4.4.1. RDFS

RDF was developed as a general purpose model for representing information on the Web. After the introduction of RDF, the W3C published a Recommendation known as the RDF Schema (RDFS) for specifying how RDF could be used to describe vocabularies. (6) RDFS extends the basic RDF syntax by defining additional terms which facilitate the categorization of resources into classes, and which describe how properties are related to those classes. Because the DCMI Abstract Model was developed based on the concepts defined in RDF and RDFS, there is a direct relationship between the concepts discussed in section 4.3. and those which are described formally in RDFS. In fact, although the DCAM does not assume any particular encoding syntax for vocabularies which follow it, the model itself is formally defined based on RDF and RDFS.

4.4.2. Correspondence between the structure of the DCAM and RDF/RDFS

A described resource of the DCAM corresponds to the subject of an RDF triple. The property-value pair of the DCAM corresponds to the predicate and object of an RDF triple. The DCAM states that the description of a resource is made up of one or more instances of property-value pairs which describe that single resource. RDF does not demand that resources be described in sets of triples related to a single resource. Rather, RDF triples stand as independent statements of fact which do not have to be grouped in any particular way. However, as a practical matter when RDF is serialized using XML or N3, the triples which describe a single resource are grouped together. For example the following DCAM description:

Described resource: http://bioimages.vanderbilt.edu/kirchoff/ac1490
Property-value pairs:
dcterms:title "Quercus michauxii (Fagaceae) - leaf"
dcterms:created "2010-09-01T03:41:31"
dcterms:dateCopyrighted "2011"

can be expressed in RDF/XML as:

<rdf:Description rdf:about="http://bioimages.vanderbilt.edu/kirchoff/ac1490">
  <dcterms:title>Quercus michauxii (Fagaceae) - leaf</dcterms:title> 
  <dcterms:created>2010-09-01T03:41:31</dcterms:created>
  <dcterms:dateCopyrighted>2011</dcterms:dateCopyrighted>
</rdf:Description>

or in N3 as

<http://bioimages.vanderbilt.edu/kirchoff/ac1490> 
        dcterms:title "Quercus michauxii (Fagaceae) - leaf"
        dcterms:created "2010-09-01T03:41:31";
        dcterms:dateCopyrighted "2011".

In both of these RDF serializations, the predicates and objects which describe the subject are presented in a list of pairs which are analogous to property-value pairs as described in the DCAM.

4.4.3. Correspondence between the terminology of the DCAM and RDF/RDFS

Because of the basis of the DCMI Abstract Model on the ideas embodied in RDF and RDFS, there are equivalencies between the abstract terminology used in the DCMI Abstract Model and the formally defined terms of RDF and RDFS. The following table summarizes these equivalencies (7):

DCAM RDF/RDFS
resource rdfs:Resource
property or element rdf:Property
class rdfs:Class
syntax encoding scheme rdfs:Datatype
has domain rdfs:domain
has range rdfs:range
sub-property of rdfs:subPropertyOf
sub-class of rdfs:subClassOf

4.4.4. Using rdf:type and ways to determine instances of classes

4.4.4.1. Declaring an instance of a class using rdf:type

The DCAM model presents the concept that a vocabulary term can be a class. However, it does not provide a direct means of declaring that a resource is an instance of a class. This can be done explicitly in RDF using the predicate rdf:type. If a triple states that the subject resource has rdf:type C, then C is inferred to be an rdfs:Class and the triple states that the subject resource is an instance of that class. (8)

4.4.4.2. Inferring an instance of a class based on rdfs:range and rdfs:domain properties of a RDF predicate

Because the RDFS terms rdfs:range and rdfs:domain correspond directly to the has range and has domain relationships described in the DCAM, the explanation provided in sections 4.3.2.5. and 4.3.2.6. apply to RDFS as well. (9) In the following example,

foaf: = "http://xmlns.com/foaf/0.1/" and 
kimage: = "http://bioimages.vanderbilt.edu/kirchoff/

The definition of the term foaf:depicts in RDF (10) contains the following triple:

foaf:depicts   rdfs:domain  foaf:Image

which says that a subject of the property foaf:depicts is an foaf:Image. If we make the statement:

kimage:ac1490  foaf:depicts   http://bioimages.vanderbilt.edu/uncg/14

then the following triple can be inferred even if it is not explicitly stated in any RDF describing kimage:ac1490 since kimage:ac1490 is the subject of a triple having foaf:depicts as its predicate.

kimage:ac1490  rdf:type   foaf:Image

Thus applying the property foaf:depicts to a resource implicitly asserts that the resource is an instance of the class foaf:Image.

Similarly, if a property term T is defined to have range C, then if resource O is the object of a triple containing the T as the predicate, it is implied that O rdf:type C and therefore O is an instance of class C.

The following example illustrates the danger of using a property term having an assigned domain as the predicate of a statement about a resource of a type not intended to be described by that term. If I would like to describe the nature of novels using terms I have created in a vocabulary defined under the namespace myNmspc: I might make the statement

http://dbpedia.org/resource/The_Lord_of_the_Rings   foaf:depicts   myNmspc:goodVsEvil

However, making this statement asserts that my descriptive term for novels myNmspc:goodVsEvil is an instance of the class foaf:Image which doesn't make very much sense.

4.4.4.3. Implying an instance of a class using the class name as the name of a container element in XML serialization of RDF

When the name of a class is used as the name of the XML container element for the description of a resource (instead of the generic rdf:Description element), a rdf:type statement is implied which indicates that the resource is an instance of that class. See section 3.3.3. for an example.

4.4.4.4. Inferring an instance of a class based on a rdfs:subClassOf declaration

The definition of the class foaf:Image in RDF (10) makes the statement

foaf:Image  rdfs:subClassOf   foaf:Document

Thus if one declares that

kimage:ac1481  rdf:type  foaf:Image

it can also be inferred that

kimage:ac1481  rdf:type  foaf:Document

Note that nothing prohibits a particular resource from being an instance of several classes. Also note that in section 4.4.4.2. it was shown that the triple

kimage:ac1490  foaf:depicts   http://bioimages.vanderbilt.edu/uncg/14

implies that kimage:ac1490 is an instance of the foaf:Image class. Given the rdfs:subClassOf statement in the definition of foaf:Image, the triple also implies that kimage:ac1490 is also an instance of the foaf:Document class.

4.4.4.5. The hazards of rdfs:subClassOf, rdfs:range, and rdfs:domain

The examples given above show that careless application of the rdfs:subClassOf, rdfs:range, and rdfs:domain properties in the definitions of terms or the careless use of terms having such properties can result in unintended consequences if users are not aware of the implications of their actions. For this reason, the three properties listed above (as well as rdfs:subPropertyOf) are generally used sparingly in term definitions of general purpose vocabularies.

4.5. The Darwin Core vocabulary as an implementation of the DCMI Abstract Model

The Darwin Core vocabulary (11) is a TDWG technical specification standard. (12) It follows the DCMI Abstract Model (13). In accordance with the DCAM, Darwin Core has terms that are classes or properties. Those properties could have domains which are Darwin Core class terms (although they don't have any that are explicitly declared in the term definitions). The properties are associated with values to form property-value pairs that describe the subject.

However, this particular version of the abstract model is described in the context of the guidelines for implementing application schemas as XML, so some of the statements made in those guidelines are more restrictive than the general DCAM. In particular, they restrict a property to describe at most one class. This is to promote consistency in implementation of the containment system that is described in the guide. The XML guide also restricts values to be literal strings contained within property elements as opposed to allowing values to be specified as XML attributes. So although the model described is considered an abstract model, it is focused on describing "flat" records ("Simple Darwin Core"). The guidelines specifically state that they apply in situations where RDF/XML are not appropriate and that they are NOT guidelines for the implementation of RDF/XML. (14) So the restrictive modifications of the DCAM described in the Darwin Core abstract model should not be considered to apply to any RDF implementation of Darwin Core.

4.6. The Darwin Core vocabulary normative definition as RDF

4.6.1. Informational properties

Although there is no existing implementation guide or statement of best practices indicating how Darwin Core should be implemented as RDF, the standard itself is described using RDF in the normative document for Darwin Core (15). Much of the information about the Darwin Core terms which are defined in this document is in the form of literal strings intended for human consumption, namely rdfs:label, rdfs:comment, and dcterms:description. These properties can be viewed at the human-readable Quick Reference Guide. (16) Each term has the property rdfs:isDefinedBy which has the value of the URI of the vocabulary ( http://rs.tdwg.org/dwc/terms/). Several additional well-known informational properties are given for each term: dcterms:issued and dcterms:modified which are untyped strings but are in ISO 8601 format, rdf:type which is either rdfs:Property or rdfs:Class, dcterms:hasVersion which references a permanent record in the term history (17), and dcterms:replaces which references dated terms in the term namespace which are not explicitly described there as subject resources. The properties also include three terms that are described in the Darwin Core attributes vocabulary (18): dwcattributes:abcdEquivalence, dwcattributes:status (with values of recommended or deprecated), and dwcattributes:organizedInClass. The first of these three terms maps Darwin Core terms to similar terms in the Access to Biological Collections Data (ABCD) schema.(19) The second suggests to humans whether the term is still recommended for use. The last term suggests to humans the class which a property term might describe; Darwin core terms have no formal rdfs:domain properties (and no rdfs:range properties as well).

4.6.2. rdfs:subProperty relationships to other properties

Of the possible DCAM/RDFS relationships which allow a machine to infer additional facts from a property statement (i.e. range, domain, subclass, and subproperty), only rdfs:subProperty is used in the Darwin Core term definitions. There are several types of rdfs:subProperty declarations:

  • ID properties which describe identifiers (e.g. dwc:acceptedNameUsageID and dwc:datasetID) which are declared to be rdfs:subProperty of dcterms:identifier.
  • properties involving dates (e.g. dwc:eventDate and dwc:georeferencedDate) which are declared to be rdfs:subProperty of dcterms:date.
  • dwc:namePublishedIn which is declared an rdfs:subProperty of dcterms:bibliographicCitation
  • "By" terms (dwc:georeferencedBy,dwc:identifiedBy, dwc:locationAccordingTo, dwc:measurementDeterminedBy, dwc:nameAccordingTo, dwc:recordedBy, dwc:relationshipAccordingTo) which are declared to be rdfs:subProperty of dwc:accordingTo, a Darwin Core-defined property which is an "Abstract term to attribute information to a source."
  • dwc:collectionCode which is declared an rdfs:subProperty of http://rs.tdwg.org/ncd/terms/Collection#Code
  • dwc:collectionID which is declared an rdfs:subProperty of http://rs.tdwg.org/ncd/terms/Collection#Id
  • dwc:institutionCode which is declared an rdfs:subProperty of http://rs.tdwg.org/ncd/terms/Institution#Code
  • dwc:institutionID which is declared an rdfs:subProperty of http://rs.tdwg.org/ncd/terms/Institution#Id
  • dwc:ownerInstitutionCode which is declared an rdfs:subProperty of http://rs.tdwg.org/ncd/terms/Institution#Code
  • dwc:ResourceRelationship (a Darwin Core class) which is declared an rdfs:subProperty of dcterms:relation (20)

Note: the URIs having namespace http://rs.tdwg.org/ncd/terms/Institution# seem to be incorrect as the namespace for the Natural Collection Description vocabulary is defined as http://rs.tdwg.org/ontology/voc/Collection# in the draft standard. (21)

4.6.3. Darwin Core terms not defined by Darwin Core

There are a number of terms included in the Darwin Core standard which are not defined by Darwin Core itself. Most of these terms are record-level property terms that are defined by Dublin Core (22), e.g. dcterms:type, dcterms:modified, etc. The exception is dcterms:Location which serves as one of the classes under which Darwin Core terms are organized.

A significant aspect of these terms is that unlike the terms defined in the Darwin Core namespace, the Dublin core terms may have domains and ranges specified in their definitions. This is significant in the case of dcterms:type which will be discussed in section 4.7.3.1. See also a discussion of problems involving Darwin Core guidelines for using Dublin Core terms.

4.7. Hic sunt dracones: Using classes and types

4.7.1. How does one indicate that a resource is a class?

There are two ways to assert that a resource is a class

4.7.1.1. Define a term to be a class directly by declaring its rdf:type to be rdfs:Class

As described in section 4.4.4.1., one can express that a term represents a class if it is declared explicitly to be a class in its description. For example, in the definition of dcterms:Agent the following RDF triple is used to declare it to be a class:

http://purl.org/dc/terms/Agent   rdf:type   http://www.w3.org/2000/01/rdf-schema#Class

If a term is declared to be an owl:Class, then it is also implicitly declared to be an rdfs:Class because owl:Class rdfs:subClassOf rdfs:Class. See section 7.6.2. of this guide for more on owl:Class.

4.7.1.2. Assert that a term is a class by using it as the object of an rdf:type triple

The RDFS schema defines the relationship between rdf:type and rdfs:Class.(23) The schema says that rdf:type is used to state that a resource is an instance of an rdfs:Class. Significantly, the range of rdf:type is defined to be rdfs:Class. Therefore, when one uses rdf:type to declare that a resource is an instance of a particular class, one is also asserting that the object of the triple IS an rdfs:Class even if the term which is the object isn't already explicitly defined to be one. (So the in the previous example, the triple not only asserts that dcterms:Agent is an instance of rdfs:Class, it is also asserting indirectly that rdfs:Class is an instance of rdfs:Class !)

For example, if one asserts that

http://bioimages.vanderbilt.edu/kirchoff/ac1490  rdf:type   http://purl.org/dc/dcmitype/StillImage

or in abbreviated form

kimage:ac1490  rdf:type  dctype:StillImage

not only does it assert that kimage:ac1490 is an instance of dctype:StillImage, but it also declares dctype:StillImage to be an rdfs:Class. As it turns out, dctype:StillImage is already declared to be an rdfs:Class in its term definition (24), but if it hadn't already been a class, one would be asserting that it was one by using it as the object of the type declaration.

4.7.2. Where can we find classes to use in rdf:type statements?

The RDFS schema is silent about what constitutes appropriate classes or types in a particular situation. Users are free to use any pre-existing types/classes or to define their own.

However, since the purpose of having classes seems to be to allow different users to know that they are both talking about the same kind of resource, it is a best practice to use well-known types rather than making up ad hoc types.(25) The wiki page ClassInventory lists classes defined in some well-known vocabularies and ontologies.

4.7.3. Type vocabularies

Both Dublin Core (26) and Darwin Core (27) define type vocabularies. In the RDF which defines both of these vocabularies, each of the listed types is declared to be an rdfs:Class. So as a practical matter, from the standpoint of RDF these terms are no different than the classes defined within the regular Dublin Core and Darwin Core term vocabularies. The classes in the type vocabularies are also listed on the page ClassInventory.

4.7.3.1. Why the DCMI and Darwin Core type vocabularies are defined separately from other terms

These type vocabulary terms are defined separately from the classes defined in the general terms vocabularies because these terms are intended to be used as controlled values for other property terms in the respective vocabularies. The DCMI Type Vocabulary provides a list of approved terms that may be used as controlled values for the dcterms:type property, while the Darwin Core Type Vocabulary "extends the DCMI Type Vocabulary and provides a list of approved values that may be used for the dwc:basisOfRecord term to identify the specific type of a resource." (27) In the case of a Simple Darwin Core (28) serialization, values are expressed as text literals. dcterms:type and dwc:basisOfRecord can be used as field names in a text file with the literals from the type vocabularies given as values for that field in a record. In Darwin Core XML (29), the type values are expressed as string literal content of dcterms:type and dwc:basisOfRecord elements.

In RDF, a type vocabulary term could be used as an object of dcterms:type and dwc:basisOfRecord properties. For example:

http://bioimages.vanderbilt.edu/kirchoff/ac1490  dcterms:type  dctype:StillImage
http://bioimages.vanderbilt.edu/specimen/ncu592805 dwc:basisOfRecord   dwctype:LivingSpecimen

However, as an rdfs:Class each of these type terms could also be used as the object of an rdf:type property as well.

http://bioimages.vanderbilt.edu/kirchoff/ac1490  rdf:type  dctype:StillImage
http://bioimages.vanderbilt.edu/specimen/ncu592805 rdf:type  dwctype:LivingSpecimen

The guidelines for expressing Dublin Core metadata as RDF (30) state that "It is recommended that RDF applications implementing this specification primarily use and understand rdf:type in place of dcterms:type when expressing Dublin Core metadata in RDF, as most RDF processors come with built-in knowledge of rdf:type." It also notes that dcterms:type has semantics that are similar to rdf:type, but that the precise relationship between those properties is undecided. There is no similar guideline for use of dwc:basisOfRecord vs. rdf:type, but using the same rationale, it would seem preferable to use rdf:type when Darwin Core is expressed in RDF.

Interestingly, Darwin Core includes two classes borrowed from Dublin Core: dcterms:Location as a class defined in the Darwin Core Terms vocabulary and dctype:Event in the Darwin Core Type vocabulary. However, Darwin Core also defines classes dwctype:Location and dwc:Event. Nothing in the RDF ties these similarly-named classes together, so a machine would not consider a resource with rdf:type dctype:Event to be an instance of the same class as another resource with rdf:type dwc:Event; likewise resources typed as dcterms:Location and dwctype:Location would be considered instances of different classes. In a serialization where the types were expressed as text literals rather than URIs, they would simple be "Event" and "Location" and not differentiable.

4.7.4. Why does this matter???

The creators of RDFS and the DCMI Abstract Model considered identification of the class of which a resource was an instance to be such a significant aspect of describing a resource that special structures and vocabulary were developed to declare or infer the type of a resource. In order for effective querying of an RDF database, or to conduct reasoning based on RDF data, it is important to be able to know whether resources described by different people or organizations are actually the same kind of thing. Thus it should be an important task of the community to reach a consensus about the precise URIs which should be used for typing, or if several URIs are acceptable, to define equivalency among terms from different vocabularies.

References

1 Definition of a vocabulary

For some descriptions of what a vocabulary is and how it is related to terms such as taxonomy, thesaurus, and ontology, see

http://infogrid.org/wiki/Reference/PidcockArticle

http://semwebtec.wordpress.com/2010/11/23/contolled-vocabulary-vs-ontology/

2 Term list for Audubon Core draft TDWG standard

http://species-id.net/wiki/Audubon_Core_Term_List#derivedFrom

3 Description of the Token class of Darwin-sw

http://code.google.com/p/darwin-sw/wiki/ClassToken

4 Dublin Core Metadata Initiative

http://dublincore.org/

There are important distinctions between the legacy terms in the dc: = http://purl.org/dc/elements/1.1/ namespace and the terms in the dcterms: = http://purl.org/dc/terms/ namespace. See the sections "Addition of domains and ranges for existing properties and creation of new properties" and "Subproperty relations between terms in 'the Dublin Core' " at http://dublincore.org/usage/decisions/2008/dcterms-changes/#sect-2 as well as http://dublincore.org/documents/2008/01/14/domain-range/ for details.

5 DCMI Abstract Model

http://dublincore.org/documents/abstract-model/

6 RDF Vocabulary Description Language 1.0: RDF Schema

http://www.w3.org/TR/rdf-schema/

7 Comparison of DCMI Abstract Model with RDF/RDFS

The information in this table is taken from

http://dublincore.org/documents/abstract-model/#sect-5

using the following namespace abbreviations to simplify the table:

rdf: = http://www.w3.org/1999/02/22-rdf-syntax-ns#

rdfs: = http://www.w3.org/2000/01/rdf-schema#

8 Description of rdf:type

http://www.w3.org/TR/rdf-schema/#ch_type

9 Definitions of rdfs:range and rdfs:domain

http://www.w3.org/TR/rdf-schema/#ch_range

10 Definition of the FOAF vocabulary in RDF

http://xmlns.com/foaf/spec/index.rdf

11 Darwin Core

http://rs.tdwg.org/dwc/

12 Categories of TDWG standards

http://www.tdwg.org/standards/status-and-categories/

13 Darwin Core abstract model

http://rs.tdwg.org/dwc/terms/guides/xml/index.htm#abstractmodel

14 Introduction to the Darwin Core XML Guide

http://rs.tdwg.org/dwc/terms/guides/xml/index.htm#introduction

15 Darwin Core terms described in RDF/XML

can be downloaded from

http://rs.tdwg.org/dwc/rdf/dwcterms.rdf

using most web browsers. However, it is not possible to view the file using most web browsers.

The file at http://code.google.com/p/darwincore/source/browse/trunk/rdf/dwcterms.rdf (viewable in a browser) is similar but may be an outdated version.

The official normative document (i.e. Type 1 document sensu the draft TDWG documentation standard) for Darwin Core is http://rs.tdwg.org/dwc/rdf/dwctermshistory.rdf viewable at http://code.google.com/p/darwincore/source/browse/trunk/rdf/dwctermshistory.rdf but as a practical matter, dereferencing term URIs leads to the dwcterms.rdf file since the terms in the dwctermshistory.rdf file have dates appended for versioning purposes.

16 Darwin Core Quick Reference Guide

http://rs.tdwg.org/dwc/terms/index.htm

17 Darwin Core term history

can be downloaded from

http://rs.tdwg.org/dwc/rdf/dwctermshistory.rdf

using most web browsers and which can be viewed in human-readable form at

http://darwincore.googlecode.com/svn/trunk/terms/history/index.htm

but which is not necessarily up to date.

18 Darwin Core attributes vocabulary

http://rs.tdwg.org/dwc/terms/attributes/

which is described by the document

http://rs.tdwg.org/dwc/terms/attributes/dwcattributes.rdf

19 The Access to Biological Collections Data (ABCD) schema

The ABCD schema is a TDWG Standard (http://www.tdwg.org/standards/115/) which can be used to describe collections, specimens, and observations in a highly structured fashion. It is in the form of an XML schema and its terms are not identified by URIs nor defined in RDF, so they cannot be used as URI object references in RDF.

20 dwc:ResourceRelationship is both a class and a property

see http://code.google.com/p/darwincore/issues/detail?id=135

21 Natural Collections Descriptions (NCD) draft TDWG standard

Human readable http://www.tdwg.org/standards/312/

RDF at

http://rs.tdwg.org/ontology/voc/Collection

http://rs.tdwg.org/ontology/voc/Institution

http://rs.tdwg.org/ontology/voc/InstitutionType

http://rs.tdwg.org/ontology/voc/ContactDetails

Use view page source to view RDF; however, a note at that page says management of the vocabulary is now at http://code.google.com/p/tdwg-ontology/source/browse/trunk/ontology/voc/

22 DCMI Metadata Terms

http://dublincore.org/documents/dcmi-terms/

23 Definition of rdfs:type

http://www.w3.org/TR/rdf-schema/#ch_type

24 Definition of dcterms:StillImage

http://dublincore.org/2010/10/11/dctype.rdf in RDF

http://dublincore.org/documents/dcmi-type-vocabulary/ in human-readable form

25 Typing using well-known vocabularies

The TDWG GUID Applicability Standard http://www.tdwg.org/standards/150/ states in Recommendation 11 that "Objects in the biodiversity domain that are identified by a GUID should be typed using the TDWG ontology or other well-known vocabularies in accordance with the TDWG common architecture." Thus in our community, it is considered to be a best practice to use a well-known vocabulary.

26 DCMI Type Vocabulary

Human readable at http://dublincore.org/documents/dcmi-type-vocabulary/#H7

RDF at http://dublincore.org/2010/10/11/dctype.rdf

See the section "The Vocabulary Encoding Scheme 'DCMI Type Vocabulary' and its member terms" at http://dublincore.org/usage/decisions/2008/dcterms-changes/#sect-2 for a history of the evolution of the DCMI Type vocabulary

27 Darwin Core Type Vocabulary

Human readable at http://rs.tdwg.org/dwc/terms/type-vocabulary/index.htm

RDF at http://code.google.com/p/darwincore/source/browse/trunk/rdf/dwctype.rdf

28 Simple Darwin Core

http://rs.tdwg.org/dwc/terms/simple/index.htm

29 Darwin Core XML Guide

http://rs.tdwg.org/dwc/terms/guides/xml/index.htm

30 Notes on semantics of Dublin Core as RDF

http://dublincore.org/documents/dc-rdf/#sect-5


Thanks to Paul Murray for helpful comments and suggestions on this page, including part of the definition of term.

Questions? Comments? Contact Steve Baskauf .

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

You can’t perform that action at this time.