The current Web is primarily made up of an enormous number of documents that have been created using HTML. These documents contain significant amounts of structured data, which is largely unavailable to tools and applications. When publishers can express this data more completely, and when tools can read it, a new world of user functionality becomes available, letting users transfer structured data between applications and web sites, and allowing browsing applications to improve the user experience: an event on a web page can be directly imported into a user's desktop calendar; a license on a document can be detected so that users can be informed of their rights automatically; a photo's creator, camera setting information, resolution, location and topic can be published as easily as the original photo itself, enabling structured search and sharing.

RDFa Core is a specification for attributes to express structured data in any markup language. The embedded data already available in the markup language (e.g., HTML) can often be reused by the RDFa markup, so that publishers don't need to repeat significant data in the document content. The underlying abstract representation is RDF [[RDF11-PRIMER]], which lets publishers build their own vocabulary, extend others, and evolve their vocabulary with maximal interoperability over time. The expressed structure is closely tied to the data, so that rendered data can be copied and pasted along with its relevant structure.

The rules for interpreting the data are generic, so that there is no need for different rules for different formats; this allows authors and publishers of data to define their own formats without having to update software, register formats via a central authority, or worry that two formats may interfere with each other.

RDFa shares some of the same goals with microformats [[MICROFORMATS]]. Whereas microformats specify both a syntax for embedding structured data into HTML documents and a vocabulary of specific terms for each microformat, RDFa specifies only a syntax and relies on independent specification of terms (often called vocabularies or taxonomies) by others. RDFa allows terms from multiple independently-developed vocabularies to be freely intermixed and is designed such that the language can be parsed without knowledge of the specific vocabulary being used.

This document is a detailed syntax specification for RDFa, aimed at:

For those looking for an introduction to the use of RDFa and some real-world examples, please consult the [[RDFA-PRIMER]].

How to Read this Document

First, if you are not familiar with either RDFa or RDF, and simply want to add RDFa to your documents, then you may find the RDFa Primer [[RDFA-PRIMER]] to be a better introduction.

If you are already familiar with RDFa, and you want to examine the processing rules — perhaps to create an RDFa Processor — then you'll find the Processing Model section of most interest. It contains an overview of each of the processing steps, followed by more detailed sections, one for each rule.

If you are not familiar with RDFa, but you are familiar with RDF, then you might find reading the Syntax Overview useful, before looking at the Processing Model since it gives a range of examples of markup that use RDFa. Seeing some examples first should make reading the processing rules easier.

If you are not familiar with RDF, then you might want to take a look at the section on RDF Terminology before trying to do too much with RDFa. Although RDFa is designed to be easy to author — and authors don't need to understand RDF to use it — anyone writing applications that consume RDFa will need to understand RDF. There is a lot of material about RDF on the web, and a growing range of tools that support RDFa. This document only contains enough background on RDF to make the goals of RDFa more clear.

RDFa is a way of expressing RDF-style relationships using simple attributes in existing markup languages such as HTML. RDF is fully internationalized, and permits the use of Internationalized Resource Identifiers, or IRIs. You will see the term 'IRI' used throughout this specification. Even if you are not familiar with the term IRI, you probably have seen the term 'URI' or 'URL'. IRIs are an extension of URIs that permits the use of characters outside those of plain ASCII. RDF allows the use of these characters, and so does RDFa. This specification has been careful to use the correct term, IRI, to make it clear that this is the case.

Even though this specification exclusively references IRIs, it is possible that a Host Language will restrict the syntax for its attributes to a subset of IRIs (e.g., href in HTML5). Regardless of validation constraints in Host Languages, an RDFa Processor is capable of processing IRIs.

This is an Editorial Revision of the Recommendation published on the 22nd of August, 2013. That document was a revision of RDFa Syntax 1.0 [[RDFA-SYNTAX]]. There are a number of substantive differences between this version and Version 1.0, including:

  1. The removal of the specific rules for XHTML - these are now defined in XHTML+RDFa [[!XHTML-RDFA]].
  2. An expansion of the datatypes of some RDFa attributes so that they can contain Terms, CURIES, or Absolute IRIs.
  3. Host languages are permitted to define collections of default terms, default prefix mappings, and a default vocabulary.
  4. The ability to define a default vocabulary to use for Terms that are undefined.
  5. Terms are required to be compared in a case-insensitive manner.
  6. A richer behavior of the @property attribute, that can replace, in many cases the @rel attribute.
  7. A slightly different handling of @typeof, making it better adapted to practical usage.

There is a more thorough list of changes in Changes.

A sample test harness is available. This set of tests is not intended to be exhaustive. Users may find the tests to be useful examples of RDFa usage.

Motivation

RDF/XML [[RDF-SYNTAX-GRAMMAR]] provides sufficient flexibility to represent all of the abstract concepts in RDF. However, it presents a number of challenges; first it is difficult or impossible to validate documents that contain RDF/XML using XML Schemas or DTDs, which therefore makes it difficult to import RDF/XML into other markup languages. Whilst newer schema languages such as RELAX NG [[RELAXNG-SCHEMA]] do provide a way to validate documents that contain arbitrary RDF/XML, it will be a while before they gain wide support.

Second, even if one could add RDF/XML directly into an XML dialect like XHTML, there would be significant data duplication between the rendered data and the RDF/XML structured data. It would be far better to add RDF to a document without repeating the document's existing data. For example, an XHTML document that explicitly renders its author's name in the text — perhaps as a byline on a news site — should not need to repeat this name for the RDF expression of the same concept: it should be possible to supplement the existing markup in such a way that it can also be interpreted as RDF.

Another reason for aligning the rendered data with the structured data is that it is highly beneficial to express the web data's structure 'in context'; as users often want to transfer structured data from one application to another, sometimes to or from a non-web-based application, the user experience can be enhanced. For example, information about specific rendered data could be presented to the user via 'right-clicks' on an item of interest. Moreover, organizations that generate a lot of content (e.g., news outlets) find it easier to embed the semantic data inline than to maintain it separately.

In the past, many attributes were 'hard-wired' directly into the markup language to represent specific concepts. For example, in XHTML 1.1 [[XHTML11]] and HTML [[HTML401]] there is @cite; the attribute allows an author to add information to a document which is used to indicate the origin of a quote.

However, these 'hard-wired' attributes make it difficult to define a generic process for extracting metadata from any document since an RDFa Processor would need to know about each of the special attributes. One motivation for RDFa has been to devise a means by which documents can be augmented with metadata in a general, rather than hard-wired, manner. This has been achieved by creating a fixed set of attributes and parsing rules, but allowing those attributes to contain properties from any of a number of the growing range of available RDF vocabularies. In most cases the values of those properties are the information that is already in an author's document.

RDFa alleviates the pressure on markup language designers to anticipate all the structural requirements users of their language might have, by outlining a new syntax for RDF that relies only on attributes. By adhering to the concepts and rules in this specification, language designers can import RDFa into their environment with a minimum of hassle and be confident that semantic data will be extractable from their documents by conforming processors.

Syntax Overview

The following examples are intended to help readers who are not familiar with RDFa to quickly get a sense of how it works. For a more thorough introduction, please read the RDFa Primer [[RDFA-PRIMER]].

In RDF, it is common for people to shorten vocabulary terms via abbreviated IRIs that use a 'prefix' and a 'reference'. This mechanism is explained in detail in the section titled Compact URI Expressions. The examples throughout this document assume that the following vocabulary prefixes have been defined:

bibo: http://purl.org/ontology/bibo/
cc: http://creativecommons.org/ns#
dbp: http://dbpedia.org/property/
dbp-owl: http://dbpedia.org/ontology/
dbr: http://dbpedia.org/resource/
dc: http://purl.org/dc/terms/
ex: http://example.org/
foaf: http://xmlns.com/foaf/0.1/
owl: http://www.w3.org/2002/07/owl#
rdf: http://www.w3.org/1999/02/22-rdf-syntax-ns#
rdfa: http://www.w3.org/ns/rdfa#
rdfs: http://www.w3.org/2000/01/rdf-schema#
xhv: http://www.w3.org/1999/xhtml/vocab#
xsd: http://www.w3.org/2001/XMLSchema#

In some of the examples below we have used IRIs with fragment identifiers that are local to the document containing the RDFa fragment identifiers shown (e.g., 'about="#me"'). This idiom, which is also used in RDF/XML [[!RDF-SYNTAX-GRAMMAR]] and other RDF serializations, gives a simple way to 'mint' new IRIs for entities described by RDFa and therefore contributes considerably to the expressive power of RDFa. The precise meaning of IRIs which include fragment identifiers when they appear in RDF graphs is given in Section 7 of [[RDF-SYNTAX-GRAMMAR]]. To ensure that such fragment identifiers can be interpreted correctly, media type registrations for markup languages that incorporate RDFa should directly or indirectly reference this specification.

The RDFa Attributes

RDFa makes use of a number of commonly found attributes, as well as providing a few new ones. Attributes that already exist in widely deployed languages (e.g., HTML) have the same meaning they always did, although their syntax has been slightly modified in some cases. For example, in (X)HTML there is no clear way to add new rel values; RDFa sets out to explicitly solve this problem, and does so by allowing IRIs as values. It also introduces the concepts of terms and 'compact URI expressions' — referred to as CURIEs in this document — which allow a full IRI value to be expressed succinctly. For a complete list of RDFa attribute names and syntax, see Attributes and Syntax.

Examples

In (X)HTML, authors can include metadata and relationships concerning the current document by using the meta and link elements (in these examples, XHTML+RDFa [[!XHTML-RDFA]] is used). For example, the author of the page along with the pages preceding and following the current page can be expressed using the link and meta elements:

RDFa makes use of this concept, enhancing it with the ability to make use of other vocabularies by using full IRIs:

Because using full IRIs like those above can be cumbersome, RDFa also permits the use of compact URI expressions so an author can use a shorthand to reference terms in multiple vocabularies:

RDFa supports the use of rel and rev on any element. This is even more useful with the addition of support for different vocabularies:

Not only can IRIs in the document be re-used to provide metadata, but so can inline text when used with property:

If some displayed text is different from the actual 'value' it represents, a more precise value can be added using content. A value can also optionally be typed using datatype:

RDFa allows the document to contain metadata information about other documents and resources:

In many cases a block of markup will contain a number of properties that relate to the same item. It's possible with RDFa to indicate the type of that item using typeof:

When dealing with small amounts of markup, it is sometimes easier to use full IRIs, rather than CURIEs. The previous example can also be written as follows:

A simple way of defining a portion of a document using terms from a specific vocabulary is to use vocab to define a default vocabulary IRI. For example, to use FOAF terms:

The example above will produce the following triples, expressed here in Turtle syntax:

In simple cases the property property can also be used in place of rel. Indeed, in case when the element does not contain rel, datatype, or content, but there is, for example, a href, the effect of property is analogous to the role of rel. For example, the previous example could have been written:

RDF Terminology

The previous section gave examples of typical markup in order to illustrate the structure of RDFa markup. RDFa is short for "RDF in Attributes". In order to author RDFa you do not need to understand RDF, although it would certainly help. However, if you are building a system that consumes the RDF output of a language that supports RDFa you will almost certainly need to understand RDF. This section introduces the basic concepts and terminology of RDF. For a more thorough explanation of RDF, please refer to the RDF Concepts document [[RDF-SYNTAX-GRAMMAR]] and the RDF Syntax Document [[RDF-SYNTAX-GRAMMAR]].

Statements

The structured data that RDFa provides access to is a collection of statements. A statement is a basic unit of information that has been constructed in a specific format to make it easier to process. In turn, by breaking large sets of information down into a collection of statements, even very complex metadata can be processed using simple rules.

To illustrate, suppose we have the following set of facts:

This would be quite difficult for a machine to interpret, and it is certainly not in a format that could be passed from one data application to another. However, if we convert the information to a set of statements it begins to be more manageable. The same information could therefore be represented by the following shorter 'statements':

Triples

To make this information machine-processable, RDF defines a structure for these statements. A statement is formally called a triple, meaning that it is made up of three components. The first is the subject of the triple, and is what we are making our statement about. In all of these examples the subject is 'Albert'.

The second part of a triple is the property of the subject that we want to define. In the examples here, the properties would be 'was born on', 'was born in', and 'has a picture at'. These properties are typically called predicates in RDF.

The final part of a triple is called the object. In the examples here the three objects have the values 'March 14, 1879', 'the German Empire', and 'http://en.wikipedia.org/wiki/Image:Albert_Einstein_Head.jpg'.

RDFa supports internationalized characters in the subject, 'predicate', and the object.

IRI References

Breaking complex information into manageable units helps us be specific about our data, but there is still some ambiguity. For example, which 'Albert' are we talking about? If another system has more facts about 'Albert', how could we know whether they are about the same person, and so add them to the list of things we know about that person? If we wanted to find people born in the German Empire, how could we know that the predicate 'was born in' has the same purpose as the predicate 'birthplace' that might exist in some other system? RDF solves this problem by replacing our vague terms with IRI references.

IRIs are most commonly used to identify web pages, but RDF makes use of them as a way to provide unique identifiers for concepts. For example, we could identify the subject of all of our statements (the first part of each triple) by using the DBPedia [http://dbpedia.org] IRI for Albert Einstein, instead of the ambiguous string 'Albert':

IRI references are also used to uniquely identify the objects in metadata statements (the third part of each triple). The picture of Einstein is already an IRI, but we could also use an IRI to uniquely identify the country 'German Empire'. At the same time we'll indicate that the name and date of birth really are literals (and not IRIs), by putting quotes around them:

IRI references are also used to ensure that predicates are unambiguous; now we can be sure that 'birthplace', 'place of birth', 'Lieu de naissance' and so on, all mean the same thing:

Plain Literals

Although IRI resources are always used for subjects and predicates, the object part of a triple can be either an IRI or a literal. In the example triples, Einstein's name is represented by a plain literal, specifically a basic string with no type or language information:

A plain literal can also be given a language tag, to capture plain text in a natural language. For example, Einstein's birthplace has different names in English and German:

Typed Literals

Some literals, such as dates and numbers, have very specific meanings, so RDF provides a mechanism for indicating the type of a literal. A typed literal is indicated by attaching an IRI to the end of a plain literal, and this IRI indicates the literal's datatype. This IRI is usually based on datatypes defined in the XML Schema Datatypes specification [[!XMLSCHEMA11-2]]. The following syntax would be used to unambiguously express Einstein's date of birth as a literal of type http://www.w3.org/2001/XMLSchema#date:

Turtle

RDF itself does not have one set way to express triples, since the key ideas of RDF are the triple and the use of IRIs, and not any particular syntax. However, there are a number of mechanisms for expressing triples, such as RDF/XML [[!RDF-SYNTAX-GRAMMAR]], Turtle [[TURTLE]], and of course RDFa. Many discussions of RDF make use of the Turtle syntax to explain their ideas, since it is quite compact. The examples we have just seen are already using this syntax, and we'll continue to use it throughout this document when we need to talk about the RDF that could be generated from some RDFa. Turtle allows long IRIs to be abbreviated by using an IRI mapping, which can be used to express a compact IRI expression as follows:

Here 'dbp:' has been mapped to the IRI for DBPedia and 'foaf:' has been mapped to the IRI for the 'Friend of a Friend' vocabulary.

Any IRI in Turtle could be abbreviated in this way. This means that we could also have used the same technique to abbreviate the identifier for Einstein, as well as the datatype indicator:

When writing examples, you will often see the following IRI in the Turtle representation:

This indicates the 'current document', i.e., the document being processed. In the end there will always be a full IRI based on the document's location, but this abbreviation serves to make examples more compact. Note in particular that the whole technique of abbreviation is merely a way to make examples more compact, and the actual triples generated would always use the full IRIs.

Graphs

A collection of triples is called a graph. All of the triples that are defined by this specification are contained in the output graph by an RDFa Processor. For more information on graphs and other RDF concepts, see [[RDF-SYNTAX-GRAMMAR]].

Compact URI Expressions

In order to allow for the compact expression of RDF statements, RDFa allows the contraction of most IRI references into a form called a 'compact URI expression', or CURIE. A detailed discussion of this mechanism is in the section CURIE and IRI Processing.

Note that CURIEs are only used in the markup and Turtle examples, and will never appear in the generated triples, which are defined by RDF to use IRI references.

Markup Fragments and RDFa

A growing use of embedded metadata is to take fragments of markup and move them from one document to another. This may happen through the use of tools, such as drag-and-drop in a browser, or through snippets of code provided to authors for inclusion in their documents. A good example of the latter is the licensing fragment provided by Creative Commons.

However, those involved in creating fragments (either by building tools, or authoring snippets), should be aware that this specification does not say how fragments are processed. Specifically, the processing of a fragment 'outside' of a complete document is undefined because RDFa processing is largely about context. Future versions of this or related specifications may do more to define this behavior.

Developers of tools that process fragments, or authors of fragments for manual inclusion, should also bear in mind what will happen to their fragment once it is included in a complete document. They should carefully consider the amount of 'context' information that will be needed in order to ensure a correct interpretation of their fragment.

A Description of RDFa in RDF Terms

The following is a brief description of RDFa in terms of the RDF terminology introduced here. It may be useful to readers with an RDF background:

An RDF graph comprises nodes linked by relationships. The aim of RDFa is to allow a single RDF graph to be carried in various types of document markup. The basic unit of an RDF graph is a triple, in which a subject node is linked to an object node via a predicate. The subject node is always either a IRI reference or a blank node (or bnode), the predicate is always a IRI reference, and the object of a statement can be a IRI reference, a literal, or a bnode.

In RDFa, a subject IRI reference is generally indicated using about and predicates are represented using one of property, rel, or rev. Objects which are IRI references are represented using resource, src, or href, whilst objects that are literals are represented either with content or the content of the element in question (with an optional datatype expressed using datatype, and an optional language expressed using a Host Language-defined mechanism such as @xml:lang).

RDFa Processor Conformance

This specification uses the term output graph to mean all of the triples asserted by a document according to the Processing Model section. A conforming RDFa Processor MUST make available to a consuming application a single RDF graph containing all possible triples generated by using the rules in the Processing Model section. The term processor graph is used to denote the collection of all informational, warning, and error triples that MAY be generated by the RDFa Processor to report its status. The output graph and the processor graph are separate graphs and MUST NOT be stored in the same graph by the RDFa Processor. However, processors may permit the two graphs to be retrieved together; see Section 7.6.1 for details.

A conforming RDFa Processor MAY make available additional triples that have been generated using rules not described here, but these triples MUST NOT be made available in the output graph. (Whether these additional triples are made available in one or more additional RDF graphs is implementation-specific, and therefore not defined here.)

A conforming RDFa Processor MUST preserve white space in both plain literals and XML literals. However, it may be the case that the architecture in which a processor operates has made changes to the white space in a document before that document ever reaches the RDFa Processor (e.g., [[XMLSCHEMA11-1]] processors are permitted to 'normalize' white space in attribute values - see section 3.1.4). To ensure maximum consistency between processing environments, authors SHOULD remove any unnecessary white space in their plain and XML Literal content.

A conforming RDFa Processor MUST examine the media type of a document it is processing to determine the document's Host Language. If the RDFa Processor is unable to determine the media type, or does not support the media type, the RDFa Processor MUST process the document as if it were media type application/xml. See XML+RDFa Document Conformance. A Host Language MAY specify additional announcement mechanisms.

A conforming RDFa Processor MAY use additional mechanisms (e.g., the DOCTYPE, a file extension, the root element, an overriding user-defined parameter) to attempt to determine the Host Language if the media type is unavailable. These mechanisms are unspecified.

RDFa Host Language Conformance

Host Languages that incorporate RDFa must adhere to the following:

XML+RDFa Document Conformance

This specification does not define a stand-alone document type. The attributes herein are intended to be integrated into other host languages (e.g., HTML+RDFa or XHTML+RDFa). However, this specification does define processing rules for generic XML documents - that is, those documents delivered as media types text/xml or application/xml. Such documents must meet all of the following criteria:

  1. The document MUST be well-formed as defined in [[!XML10-4e]].
  2. The document SHOULD use the attributes defined in this specification in 'no namespace' (e.g., when the attributes are used on elements they are used with no qualifying prefix: ).

    It is possible that an XML grammar will have native attributes that conflict with attributes in this specification. This could result in an RDFa processor generating unexpected triples.

When an RDFa Processor processes an XML+RDFa document, it does so via the following initial context:

  1. There are default terms (e.g., describedby, license, and role), defined in http://www.w3.org/2011/rdfa-context/rdfa-1.1.
  2. There are default prefix mappings (e.g., dc), defined in http://www.w3.org/2011/rdfa-context/rdfa-1.1.
  3. There is no default vocabulary IRI.
  4. The base can be set using the @xml:base attribute as defined in [[!XML10-4e]].
  5. The current language can be set using @xml:lang attribute.

Attributes and Syntax

This specification defines a number of attributes and the way in which the values of those attributes are to be interpreted when generating RDF triples. This section defines the attributes and the syntax of their values.

about
a SafeCURIEorCURIEorIRI, used for stating what the data is about (a 'subject' in RDF terminology);
content
a CDATA string, for supplying machine-readable content for a literal (a 'literal object', in RDF terminology);
datatype
a TERMorCURIEorAbsIRI representing a datatype, to express the datatype of a literal;
href (optional)
a traditionally navigable IRI for expressing the partner resource of a relationship (a 'resource object', in RDF terminology);
inlist
An attribute used to indicate that the object associated with a rel or property attribute on the same element is to be added to the list for that predicate. The value of this attribute MUST be ignored. Presence of this attribute causes a list to be created if it does not already exist.
prefix
a white space separated list of prefix-name IRI pairs of the form
NCName ':' ' '+ xsd:anyURI
property
a white space separated list of TERMorCURIEorAbsIRIs, used for expressing relationships between a subject and either a resource object if given or some literal text (also a 'predicate');
rel
a white space separated list of TERMorCURIEorAbsIRIs, used for expressing relationships between two resources ('predicates' in RDF terminology);
resource
a SafeCURIEorCURIEorIRI for expressing the partner resource of a relationship that is not intended to be navigable (e.g., a 'clickable' link) (also an 'object');
rev
a white space separated list of TERMorCURIEorAbsIRIs, used for expressing reverse relationships between two resources (also 'predicates');
src (optional)
an IRI for expressing the partner resource of a relationship when the resource is embedded (also a 'resource object');
typeof
a white space separated list of TERMorCURIEorAbsIRIs that indicate the RDF type(s) to associate with a subject;
vocab
an IRI that defines the mapping to use when a TERM is referenced in an attribute value. See General Use of Terms in Attributes and the section on Vocabulary Expansion.

In all cases it is possible for these attributes to be used with no value (e.g., datatype="") or with a value that evaluates to no value after evaluation using the rules for CURIE and IRI Processing (e.g., datatype="[noprefix:foobar]").

Roles of attributes

The RDFa attributes play different roles in a semantically rich document. Briefly, those roles are:

White space within attribute values

Many attributes accept a white space separated list of tokens. This specification defines white space as:

whitespace      ::=  (#x20 | #x9 | #xD | #xA)+ 

When attributes accept a white space separated list of tokens, an RDFa Processor MUST ignore any leading or trailing white space.

This definition is consistent with the definition found in [[XML10]].

CURIE Syntax Definition

The working group is currently examining the productions for CURIE below in light of recent comments received from the RDF Working Group and members of the RDFa Working Group. It is possible that there will be minor changes to the production rules below in the near future, and that these changes will be backward incompatible. However, any such incompatibility will be limited to edge cases.

The key component of RDF is the IRI, but these are usually long and unwieldy. RDFa therefore supports a mechanism by which IRIs can be abbreviated, called 'compact URI expressions' or simply, CURIEs.

When expanded, the resulting IRI MUST be a syntactically valid IRI [[!RFC3987]]. For a more detailed explanation see CURIE and IRI Processing. The lexical space of a CURIE is as defined in curie below. The value space is the set of IRIs.

A CURIE is comprised of two components, a prefix and a reference. The prefix is separated from the reference by a colon (:). In general use it is possible to omit the prefix, and so create a CURIE that makes use of the 'default prefix' mapping; in RDFa the 'default prefix' mapping is http://www.w3.org/1999/xhtml/vocab#. It's also possible to omit both the prefix and the colon, and so create a CURIE that contains just a reference which makes use of the 'no prefix' mapping. This specification does not define a 'no prefix' mapping. RDFa Host Languages MUST NOT define a 'no prefix' mapping.

The RDFa 'default prefix' should not be confused with the 'default namespace' as defined in [[!XML-NAMES]]. An RDFa Processor MUST NOT treat an XML-NAMES 'default namespace' declaration as if it were setting the 'default prefix'.

The general syntax of a CURIE can be summarized as follows:

prefix      ::=   NCName

reference   ::=   ( ipath-absolute / ipath-rootless / ipath-empty ) 
                     [ "?" iquery ] [ "#" ifragment ] (as defined in [[!RFC3987]])

curie       ::=   [ [ prefix ] ':' ] reference

safe_curie  ::=   '[' [ [ prefix ] ':' ] reference ']'

The production safe_curie is not required, even in situations where an attribute value is permitted to be a CURIE or an IRI: An IRI that uses a scheme that is not an in-scope mapping cannot be confused with a CURIE. The concept of a safe_curie is retained for backward compatibility.

It is possible to define a CURIE prefix mapping in such a way that it would overshadow a defined IRI scheme. For example, a document could map the prefix 'mailto' to 'http://www.example.com/addresses/'. Then a resource that contained 'mailto:user@example.com' might create a triple with the object 'http://www.example.com/addresses/user@example.com'. Moreover, it is possible though unlikely, that schemes will be introduced in the future that will conflict with prefix mappings defined in a document (e.g., the newly proposed 'widget' scheme [[WIDGETS-URI]]). In neither case would this RDFa overshadowing of the underlying scheme alter the way other consumers of the IRI treat that IRI. It could, however, mean that the document author's intended use of the CURIE is mis-interpreted by another consumer as an IRI. The working group considers this risk to be minimal.

In normal evaluation of CURIEs the following context information would need to be provided:

In RDFa these values are defined as follows:

A CURIE is a representation of a full IRI. The rules for determining that IRI are:

See General Use of Terms in Attributes for the way items with no colon can be interpreted in some datatypes by RDFa Processors.

Why CURIEs and not QNames?

In many cases, language designers have attempted to use QNames for an extension mechanism [[XMLSCHEMA11-2]]. QNames do permit independent management of the name collection, and can map the names to a resource. Unfortunately, QNames are unsuitable in most cases because 1) the use of QName as identifiers in attribute values and element content is problematic as discussed in [[QNAMES]] and 2) the syntax of QNames is overly restrictive and does not allow all possible IRIs to be expressed.

A specific example of the problem this causes comes from attempting to define the name collection for books. In a QName, the part after the colon must be a valid element name, making an example such as the following invalid: isbn:0321154991

This is not a valid QName simply because "0321154991" is not a valid element name. Yet, in the example given, we don't really want to define a valid element name anyway. The whole reason for using a QName was to reference an item in a private scope - that of ISBNs. Moreover, in this example, we want the names within that scope to map to an IRI that will reveal the meaning of that ISBN. As you can see, the definition of QNames and this (relatively common) use case are in conflict with one another.

This specification addresses the problem by defining CURIEs. Syntactically, CURIEs are a superset of QNames.

Note that this specification is targeted at language designers, not document authors. Any language designer considering the use of QNames as a way to represent IRIs or unique tokens should consider instead using CURIEs:

Processing Model

This section looks at a generic set of processing rules for creating a set of triples that represent the structured data present in an RDFa document. Processing need not follow the DOM traversal technique outlined here, although the effect of following some other manner of processing must be the same as if the processing outlined here were followed. The processing model is explained using the idea of DOM traversal which makes it easier to describe (particularly in relation to the evaluation context).

Note that in this section, explanations about the processing model or guidance to implementors are enclosed in sections like this.

Overview

Evaluating a document for RDFa triples is carried out by starting at the document object, and then visiting each of its child elements in turn, in document order, applying processing rules. Processing is recursive in that for each child element the processor also visits each of its child elements, and applies the same processing rules.

In some environments there will be little difference between starting at the root element of the document, and starting at the document object itself. It is defined this way because in some environments important information is present at the document object level which is not present on the root element.

As processing continues, rules are applied which may generate triples, and may also change the evaluation context information that will then be used when processing descendant elements.

This specification does not say anything about what should happen to the triples generated, or whether more triples might be generated during processing than are outlined here. However, to be conforming, an RDFa Processor MUST act as if at a minimum the rules in this section are applied, and a single RDF graph produced. As described in the RDFa Processor Conformance section, any additional triples generated MUST NOT appear in the output graph. They MAY be included in the processor graph.

Evaluation Context

During processing, each rule is applied using information provided by an evaluation context. An initial context is created when processing begins. That context has the following members:

During the course of processing, new evaluation contexts are created which are passed to each child element. The initial rules described below will determine the values of the items in the context. Then the core rules will cause new triples to be created by combining information provided by an element with information from the evaluation context.

During the course of processing a number of locally scoped values are needed, as follows:

Chaining

Statement chaining is an RDFa feature that allows the author to link RDF statements together while avoiding unnecessary repetitive markup. For example, if an author were to add statements as children of an object that was a resource, these statements should be interpreted as being about that resource:

In this example we can see that an object resource ('German_Empire'), has become the subject for nested statements. This markup also illustrates the basic chaining pattern of 'A has a B has a C' (i.e., Einstein has a birth place of the German Empire, which has a long name of 'the German Empire').

It's also possible for the subject of nested statements to provide the object for containing statements — essentially the reverse of the example we have just seen. To illustrate, we'll take an example of the type of chaining just described, and show how it could be marked up more efficiently. To start, we mark up the fact that Albert Einstein had, at some point in his life, a residence both in the German Empire and in Switzerland:

Now, we show the same information, but this time we create an incomplete triple from the residence part, and then use any number of further subjects to 'complete' that triple, as follows:

In this example, the incomplete triple actually gets completed twice, once for the German Empire and once for Switzerland, giving exactly the same information as we had in the earlier example:

Chaining can sometimes involve elements containing relatively minimal markup, for example showing only one resource, or only one predicate. Here the img element is used to carry a picture of Einstein:

When such minimal markup is used, any of the resource-related attributes could act as a subject or an object in the chaining:

Note that, as noted above, in many situations the property and rel are interchangeable. This is not true for chaining. Taking the first example, if that example was used as follows:

The subject for 'the German Empire' would remain Albert Einstein (and that would, of course, be an error). This is the main difference between property and rel: the latter induces chaining, whereas the former, usually, does not.

CURIE and IRI Processing

Since RDFa is ultimately a means for transporting RDF, a key concept is the resource and its manifestation as an IRI. RDF deals with complete IRIs (not relative paths); when converting RDFa to triples, any relative IRIs MUST be resolved relative to the base IRI, using the algorithm defined in section 6.5 of RFC 3987 [[!RFC3987]], Reference Resolution. The values of RDFa attributes that refer to IRIs use three different datatypes: IRI, SafeCURIEorCURIEorIRI, or TERMorCURIEorAbsIRI. All these attributes are mapped, after processing, to IRIs. The handling of these attributes is as follows:

IRI
The content is an IRI, and is used as such.
SafeCURIEorCURIEorIRI
  • When the value is surrounded by square brackets, then the content within the brackets is evaluated as a CURIE according to the CURIE Syntax Definition. If it is not a valid CURIE, the value MUST be ignored.
  • Otherwise, the value is evaluated as a CURIE. If it is a valid CURIE, the resulting IRI is used; otherwise, the value is processed as an IRI.

A consequence of this is that when the value of an attribute of this datatype is the empty string (e.g., about=""), that value resolves to an IRI. An IRI of "" is a relative IRI that is interpreted as being the same as the base. In other words, a value of "" will usually resolve to the IRI of the current document.

A related consequence of this is that when the value of an attribute of this datatype is an empty SafeCURIE (e.g., about="[]"), that value does not result in an IRI and therefore the value is ignored.

TERMorCURIEorAbsIRI
  • If the value is a term then it is evaluated as a term according to General Use of Terms in Attributes. Note that this step may mean that the value is to be ignored.
  • If the value is a valid CURIE, then the resulting IRI is used.
  • If the value is an absolute IRI, that value is used.
  • Otherwise, the value is ignored.

Note that it is possible for all values in an attribute to be ignored. When that happens, the attribute MUST be treated as if it were empty.

For example, the full IRI for Albert Einstein on DBPedia is:

This can be shortened by authors to make the information easier to manage, using a CURIE. The first step is for the author to create a prefix mapping that links a prefix to some leading segment of the IRI. In RDFa these mappings are expressed using prefix:

Once the prefix has been established, an author can then use it to shorten an IRI as follows:

The author is free to split the IRI at any point. However, since a common use of CURIEs is to make available libraries of terms and values, the prefix will usually be mapped to some common segment that provides the most re-use, often provided by those who manage the library of terms. For example, since DBPedia contains an enormous list of resources, it is more efficient to create a prefix mapping that uses the base location of the resources:

Note that it is generally considered a bad idea to use relative paths in prefix declarations. Since it is possible that an author may ignore this guidance, it is further possible that the IRI obtained from a CURIE is relative. However, since all IRIs must be resolved relative to base before being used to create triples, the use of relative paths should not have any effect on processing.

Scoping of Prefix Mappings

CURIE prefix mappings are defined on the current element and its descendants. The inner-most mapping for a given prefix takes precedence. For example, the IRIs expressed by the following two CURIEs are different, despite the common prefix, because the prefix mappings are locally scoped:

In general it is a bad practice to redefine prefix mappings within a document. In particular, while it is permitted, mapping a prefix to different values at different places within a document could lead to confusion. The working group recommends that document authors use the same prefix to map to the same vocabulary throughout a document. Many vocabularies have recommended prefix names. The working group recommends that these names are used whenever possible.

General Use of CURIEs in Attributes

There are a number of ways that attributes make use of CURIEs, and they need to be dealt with differently. These are:

  1. An attribute may allow one or more values that are a mixture of TERMs, CURIEs, and absolute IRIs.
  2. An attribute may allow one or more values that are a mixture of CURIEs and IRIs. In this case any value that is not a CURIE, as outlined in section CURIE Syntax Definition, will be processed as an IRI.
  3. If the value is surrounded by square brackets, then the content within the brackets is always evaluated according to the rules in CURIE Syntax Definition - and if that content is not a CURIE, then the content MUST be ignored.

An empty attribute value (e.g., typeof='') is still a CURIE, and is processed as such. The rules for this processing are defined in Sequence. Specifically, however, an empty attribute value is never treated as a relative IRI by this specification.

An example of an attribute that can contain a CURIEorIRI is about. To express an IRI directly, an author might do this:

whilst to express the IRI above as a CURIE an author would do this:

The author could also use a safe CURIE, as follows:

Since non-CURIE values MUST be ignored, the following value in about would not set a new subject, since about does not permit the use of TERMs, and the CURIE has no prefix separator.

However, this markup would set a subject, since it is not a CURIE, but a valid relative IRI:

Note that several RDFa attributes are able to also take TERMS as their value. This is discussed in the next section.

General Use of Terms in Attributes

Some RDFa attributes have a datatype that permits a term to be referenced. RDFa defines the syntax of a term as:

term     ::=  NCNameStartChar termChar*
termChar ::=  ( NameChar - ':' ) | '/'
    

For the avoidance of doubt, this production means a 'term' in RDFa is an XML NCName that also permits slash as a non-leading character.

When an RDFa attribute permits the use of a term, and the value being evaluated matches the production for term above, it is transformed to an IRI using the following logic:

  • If there is a local default vocabulary the IRI is obtained by concatenating that value and the term.
  • Otherwise, check if the term matches an item in the list of local term mappings. First compare against the list case-sensitively, and if there is no match then compare case-insensitively. If there is a match, use the associated IRI.
  • Otherwise, the term has no associated IRI and MUST be ignored.

A local default vocabulary can be defined by the Host Language as part of the initial context, and can be overridden on the current element and its children using vocab.

Use of CURIEs in Specific Attributes

The general rules discussed in the previous sections apply to the RDFa attributes in the following ways:

  • about and resource support the datatype SafeCURIEorCURIEorIRI - allowing a SafeCURIE, a CURIE, or an IRI.
  • href and src are as defined in the Host Language (e.g., XHTML), and support only an IRI.
  • vocab supports an IRI.
  • datatype supports the datatype TERMorCURIEorAbsIRI - allowing a single Term, CURIE, or Absolute IRI.
  • property, typeof, rel, and rev support the datatype TERMorCURIEorAbsIRIs - allowing one or more Terms, CURIEs, or Absolute IRIs.

Any value that matches a defined term MUST be expanded into a reference to the corresponding IRI. For example in the following examples:

would each generate the following triple:

Referencing Blank Nodes

In RDFa, it is possible to establish relationships using various types of resource references, including bnodes. If a subject or object is defined using a CURIE, and that CURIE explicitly names a bnode, then a Conforming Processor MUST create the bnode when it is encountered during parsing. The RDFa Processor MUST also ensure that no bnode created automatically (e.g., as a result of chaining) has a name that collides with a bnode that is defined by explicit reference in a CURIE.

Consider the following example:

In the above fragment, two bnodes are explicitly created as the subject of triples. Those bnodes are then referenced to demonstrate the relationship between the parties. After processing, the following triples will be generated:

RDFa Processors use, internally, implementation-dependent identifiers for bnodes. When triples are retrieved, new bnode indentifiers are used, which usually bear no relation to the original identifiers. However, implementations do ensure that these generated bnode identifiers are consistent: each bnode will have its own identifier, all references to a particular bnode will use the same identifier, and different bnodes will have different identifiers.

As a special case, _: is also a valid reference for one specific bnode.

Sequence

Processing would normally begin after the document to be parsed has been completely loaded. However, there is no requirement for this to be the case, and it is certainly possible to use a stream-based approach, such as SAX [[SAX]] to extract the RDFa information. However, if some approach other than the DOM traversal technique defined here is used, it is important to ensure that Host Language-specific processing rules are applied (e.g., XHTML+RDFa [[!XHTML-RDFA]] indicates the base element can be used, and base will affect the interpretation of IRIs in meta or link elements even if those elements are before the base element in the stream).

In this section the term 'resource' is used to mean 'IRI or bnode'. It is possible that this term will be replaced with some other, more formal term after consulting with other groups. Changing this term will in no way change this processing sequence.

At the beginning of processing, an initial evaluation context is created, as follows:

Processing begins by applying the processing rules below to the document object, in the context of this initial evaluation context. All elements in the tree are also processed according to the rules described below, depth-first, although the evaluation context used for each set of rules will be based on previous rules that may have been applied.

This specification defines processing rules for optional attributes that may not be present in all Host Languages (e.g., href). If these attributes are not supported in the Host Language, then the corresponding processing rules are not relevant for that language.

The processing rules are:

  1. First, the local values are initialized, as follows:
    • the skip element flag is set to 'false';
    • new subject is set to null;
    • current object resource is set to null;
    • typed resource is set to null;
    • the local list of IRI mappings is set to the list of IRI mappings from the evaluation context;
    • the local list of incomplete triples is set to null;
    • the list mapping is set to (a reference of) the list mapping from the evaluation context;
    • the current language value is set to the language value from the evaluation context.
    • the local term mappings is set to the term mappings from the evaluation context.
    • the local default vocabulary is set to the default vocabulary from the evaluation context.
    Note that some of the local variables are temporary containers for values that will be passed to descendant elements via an evaluation context. In some cases the containers will have the same name, so to make it clear which is being acted upon in the following steps, the local version of an item will generally be referred to as such.
    Note that the local term mappings is always reset to a global value, provided by the initial context. Future versions of this specification may introduce a mechanism whereby the local term mappings can be set dynamically, in which case the local term mappings would inherit from the parent's values.
  2. Next the current element is examined for any change to the default vocabulary via vocab. If vocab is present and contains a value, the local default vocabulary is updated according to the section on CURIE and IRI Processing. If the value is empty, then the local default vocabulary MUST be reset to the Host Language defined default (if any).
    The value of vocab is used to generate a triple as follows:
    subject
    base
    predicate
    http://www.w3.org/ns/rdfa#usesVocabulary
    object
    value from vocab
    A Host Language is not required to define a default vocabulary. In such a case, setting vocab to the empty value has the effect of setting the local default vocabulary to null.
  3. Next, the current element is examined for IRI mappings and these are added to the local list of IRI mappings. Note that an IRI mapping will simply overwrite any current mapping in the list that has the same name;
    Mappings are defined via prefix. Values in this attribute are evaluated from beginning to end (e.g., left to right in typical documents). For backward compatibility, RDFa Processors SHOULD also permit the definition of mappings via @xmlns. In this case, the value to be mapped is set by the XML namespace prefix, and the value to map is the value of the attribute — an IRI. (Note that prefix mapping via @xmlns is deprecated, and may be removed in a future version of this specification.) When xmlns is supported, such mappings MUST be processed before processing any mappings from prefix on the same element. Regardless of how the mapping is declared, the value to be mapped MUST be converted to lower case, and the IRI is not processed in any way; in particular if it is a relative path it MUST NOT be resolved against the current base. Authors SHOULD NOT use relative paths as the IRI.
  4. The current element is also parsed for any language information, and if present, current language is set accordingly;
    Host Languages that incorporate RDFa MAY provide a mechanism for specifying the natural language of an element and its contents (e.g., XML provides the general-purpose XML attribute @xml:lang).
  5. If the current element contains no rel or rev attribute, then the next step is to establish a value for new subject. This step has two possible alternatives.
    1. If the current element contains the property attribute, but does not contain either the content or datatype attributes, then
      new subject is set to the resource obtained from the first match from the following rule:
      • by using the resource from about, if present, obtained according to the section on CURIE and IRI Processing;
      • otherwise, if the element is the root element of the document, then act as if there is an empty about present, and process it according to the rule for about, above;
      • otherwise, if parent object is present, new subject is set to the value of parent object.

      If typeof is present then typed resource is set to the resource obtained from the first match from the following rules:

      • by using the resource from about, if present, obtained according to the section on CURIE and IRI Processing;
      • otherwise, if the element is the root element of the document, then act as if there is an empty about present and process it according to the previous rule;
      • otherwise,
        • by using the resource from resource, if present, obtained according to the section on CURIE and IRI Processing;
        • otherwise, by using the IRI from href, if present, obtained according to the section on CURIE and IRI Processing;
        • otherwise, by using the IRI from src, if present, obtained according to the section on CURIE and IRI Processing;
        • otherwise, the value of typed resource is set to a newly created bnode.
        • The value of the current object resource is then set to the value of typed resource.
    2. otherwise:
      • If the element contains an about, href, src, or resource attribute, new subject is set to the resource obtained as follows:
      • otherwise, if no resource is provided by a resource attribute, then the first match from the following rules will apply:
        • if the element is the root element of the document, then act as if there is an empty about present, and process it according to the rule for about, above;
        • otherwise, if typeof is present, then new subject is set to be a newly created bnode;
        • otherwise, if parent object is present, new subject is set to the value of parent object. Additionally, if property is not present then the skip element flag is set to 'true'.
      • Finally, if typeof is present, set the typed resource to the value of new subject.

  6. If the current element does contain a rel or rev attribute, then the next step is to establish both a value for new subject and a value for current object resource:
    new subject is set to the resource obtained from the first match from the following rules:

    if the typeof attribute is present, set typed resource to new subject.

    If no resource is provided then the first match from the following rules will apply:

    • if the element is the root element of the document then act as if there is an empty about present, and process it according to the rule for about, above;
    • otherwise, if parent object is present, new subject is set to that.

    Then the current object resource is set to the resource obtained from the first match from the following rules:

    • by using the resource from resource, if present, obtained according to the section on CURIE and IRI Processing;
    • otherwise, by using the IRI from href, if present, obtained according to the section on CURIE and IRI Processing;
    • otherwise, by using the IRI from src, if present, obtained according to the section on CURIE and IRI Processing;
    • otherwise, if typeof is present and about is not, use a newly created bnode.

    If typeof is present and about is not, set typed resource to current object resource.

    Note that final value of the current object resource will either be null (from initialization) or a full IRI or bnode.

  7. If in any of the previous steps a typed resource was set to a non-null value, it is now used to provide a subject for type values;
    One or more 'types' for the typed resource can be set by using typeof. If present, the attribute may contain one or more IRIs, obtained according to the section on CURIE and IRI Processing, each of which is used to generate a triple as follows:
    subject
    typed resource
    predicate
    http://www.w3.org/1999/02/22-rdf-syntax-ns#type
    object
    current full IRI of 'type' from typed resource
  8. If in any of the previous steps a new subject was set to a non-null value different from the parent object;
    The list mapping taken from the evaluation context is set to a new, empty mapping.
  9. If in any of the previous steps a current object resource was set to a non-null value, it is now used to generate triples and add entries to the local list mapping:
    If the element contains both the inlist and the rel attributes the rel may contain one or more resources, obtained according to the section on CURIE and IRI Processing each of which is used to add an entry to the list mapping as follows:
    • if the local list mapping does not contain a list associated with the IRI, instantiate a new list and add to local list mappings
    • add the current object resource to the list associated with the resource in the local list mapping
    Predicates for the current object resource can be set by using one or both of the rel and the rev attributes but, in case of the rel attribute, only if the inlist is not present:
    • If present, rel may contain one or more resources, obtained according to the section on CURIE and IRI Processing each of which is used to generate a triple as follows:
      subject
      new subject
      predicate
      full IRI
      object
      current object resource
    • If present, rev may contain one or more resources, obtained according to the section on CURIE and IRI Processing each of which is used to generate a triple as follows:
      subject
      current object resource
      predicate
      full IRI
      object
      new subject
  10. If however current object resource was set to null, but there are predicates present, then they must be stored as incomplete triples, pending the discovery of a subject that can be used as the object. Also, current object resource should be set to a newly created bnode (so that the incomplete triples have a subject to connect to if they are ultimately turned into triples);
    Predicates for incomplete triples can be set by using one or both of the rel and rev attributes:
    • If present, rel must contain one or more resources, obtained according to the section on CURIE and IRI Processing each of which is added to the local list of incomplete triples as follows:
      • If the element contains the inlist attribute, then
        • if the local list mapping does not contain a list associated with the IRI, instantiate a new list and add to local list mappings.
        • Add:
          list
          list from local list mapping for this IRI
          direction
          none
      • Otherwise add:
        • predicate
          full IRI
          direction
          forward
    • If present, rev must contain one or more resources, obtained according to the section on CURIE and IRI Processing, each of which is added to the local list of incomplete triples as follows:
      predicate
      full IRI
      direction
      reverse
  11. The next step of the iteration is to establish any current property value;
    Predicates for the current property value can be set by using property. If present, one or more resources are obtained according to the section on CURIE and IRI Processing, and then the actual literal value is obtained as follows:
    • as a typed literal if datatype is present, does not have an empty value according to the section on CURIE and IRI Processing, and is not set to XMLLiteral in the vocabulary http://www.w3.org/1999/02/22-rdf-syntax-ns#.

      The actual literal is either the value of content (if present) or a string created by concatenating the value of all descendant text nodes, of the current element in turn. The final string includes the datatype IRI, as described in [[RDF-SYNTAX-GRAMMAR]], which will have been obtained according to the section on CURIE and IRI Processing.

    • otherwise, as a plain literal if datatype is present but has an empty value according to the section on CURIE and IRI Processing.

      The actual literal is either the value of content (if present) or a string created by concatenating the value of all descendant text nodes, of the current element in turn.

    • otherwise, as an XML literal if datatype is present and is set to XMLLiteral in the vocabulary http://www.w3.org/1999/02/22-rdf-syntax-ns#.

      The value of the XML literal is a string created by serializing to text, all nodes that are descendants of the current element, i.e., not including the element itself, and giving it a datatype of XMLLiteral in the vocabulary http://www.w3.org/1999/02/22-rdf-syntax-ns#. The format of the resulting serialized content is as defined in Exclusive XML Canonicalization Version 1.0 [[XML-EXC-C14N]].

      In order to maintain maximum portability of this literal, any children of the current node that are elements MUST have the current XML namespace declarations (if any) declared on the serialized element. Since the child element node could also declare new XML namespaces, the RDFa Processor MUST be careful to merge these together when generating the serialized element definition. For avoidance of doubt, any re-declarations on the child node MUST take precedence over declarations that were active on the current node.

    • otherwise, as a plain literal using the value of content if content is present.
    • otherwise, if the rel, rev, and content attributes are not present, as a resource obtained from one of the following:
    • otherwise, if typeof is present and about is not, the value of typed resource.
    • otherwise as a plain literal.

    Additionally, if there is a value for current language then the value of the plain literal should include this language information, as described in [[RDF-SYNTAX-GRAMMAR]]. The actual literal is either the value of content (if present) or a string created by concatenating the text content of each of the descendant elements of the current element in document order.

    The current property value is then used with each predicate as follows:

    • If the element also includes the inlist attribute, the current property value is added to the local list mapping as follows:
      • if the local list mapping does not contain a list associated with the predicate IRI, instantiate a new list and add to local list mappings
      • add the current property value to the list associated with the predicate IRI in the local list mapping
    • Otherwise the current property value is used to generate a triple as follows:
      subject
      new subject
      predicate
      full IRI
      object
      current property value
  12. If the skip element flag is 'false', and new subject was set to a non-null value, then any incomplete triples within the current context should be completed:
    The list of incomplete triples from the current evaluation context (not the local list of incomplete triples) will contain zero or more predicate IRIs. This list is iterated over and each of the predicates is used with parent subject and new subject to generate a triple or add a new element to the local list mapping. Note that at each level there are two lists of incomplete triples; one for the current processing level (which is passed to each child element in the previous step), and one that was received as part of the evaluation context. It is the latter that is used in processing during this step.
    Note that each incomplete triple has a direction value that is used to determine what will become the subject, and what will become the object, of each generated triple:
    • If direction is 'none', the new subject is added to the list from the iterated incomplete triple.
    • If direction is 'forward' then the following triple is generated:
      subject
      parent subject
      predicate
      the predicate from the iterated incomplete triple
      object
      new subject
    • If direction is 'reverse' then this is the triple generated:
      subject
      new subject
      predicate
      the predicate from the iterated incomplete triple
      object
      parent subject
  13. Next, all elements that are children of the current element are processed using the rules described here, using a new evaluation context, initialized as follows:
    • If the skip element flag is 'true' then the new evaluation context is a copy of the current context that was passed in to this level of processing, with the language and list of IRI mappings values replaced with the local values;
    • Otherwise, the values are:
      • the base is set to the base value of the current evaluation context;
      • the parent subject is set to the value of new subject, if non-null, or the value of the parent subject of the current evaluation context;
      • the parent object is set to value of current object resource, if non-null, or the value of new subject, if non-null, or the value of the parent subject of the current evaluation context;
      • the list of IRI mappings is set to the local list of IRI mappings;
      • the list of incomplete triples is set to the local list of incomplete triples;
      • the list mapping is set to the local list mapping;
      • language is set to the value of current language.
      • the default vocabulary is set to the value of the local default vocabulary.
  14. Finally, if there is one or more mapping in the local list mapping, list triples are generated as follows:
    For each IRI in the local list mapping, if the equivalent list does not exist in the evaluation context, indicating that the list was originally instantiated on the current element, use the list as follows:
    • If there are zero items in the list associated with the IRI, generate the following triple:
      subject
      current subject
      predicate
      full IRI of the local list mapping associated with this list
      object
      http://www.w3.org/1999/02/22-rdf-syntax-ns#nil
    • Otherwise,
      • Create a new ‘bnode’ array containing newly created bnodes, one for each item in the list
      • For each bnode-(IRI or literal) pair from the list the following triple is generated:
        subject
        bnode
        predicate
        http://www.w3.org/1999/02/22-rdf-syntax-ns#first
        object
        full IRI or literal
      • For each item in the ‘bnode’ array the following triple is generated:
        subject
        bnode
        predicate
        http://www.w3.org/1999/02/22-rdf-syntax-ns#rest
        object
        next item in the ‘bnode’ array or, if that does not exist, http://www.w3.org/1999/02/22-rdf-syntax-ns#nil
      • A single additional triple is generated:
        subject
        current subject
        predicate
        full IRI of the local list mapping associated with this list
        object
        first item of the ‘bnode’ array

Processor Status

The processing rules covered in the previous section are designed to extract as many triples as possible from a document. The RDFa Processor is designed to continue processing, even in the event of errors. For example, failing to resolve a prefix mapping or term would result in the RDFa Processor skipping the generation of a triple and continuing with document processing. There are cases where knowing each RDFa Processor warning or error would be beneficial to authors. The processor graph is designed as the mechanism to capture all informational, warning, and error messages as triples from the RDFa Processor. These status triples may be retrieved and used to aid RDFa authoring or automated error detection.

If an RDFa Processor supports the generation of a processor graph, then it MUST generate a set of triples when the following processing issues occur:

Other implementation-specific rdfa:Info, rdfa:Warning, or rdfa:Error triples MAY be generated by the RDFa Processor.

Accessing the Processor Graph

Accessing the processor graph may be accomplished in a variety of ways and is dependent on the type of RDFa Processor and access method that the developer is utilizing.

SAX-based processors or processors that utilize function or method callbacks to report the generation of triples are classified as event-based RDFa Processors. For Event-based RDFa Processors, the software MUST allow the developer to register a function or callback that is called when a triple is generated for the processor graph. The callback MAY be the same as the one that is used for the output graph as long as it can be determined if a generated triple belongs in the processor graph or the output graph.

A whole-graph RDFa Processor is defined as any RDFa Processor that processes the entire document and only provides the developer access to the triples after processing has completed. RDFa Processors that typically fall into this category express their output via a single call using RDF/XML, N3, TURTLE, or N-Triples notation. For whole-graph RDFa Processors, the software MUST allow the developer to specify if they would like to retrieve the output graph, the processor graph, or both graphs as a single, combined graph from the RDFa Processor. If the graph preference is not specified, the output graph MUST be returned.

A web service RDFa Processor is defined as any RDFa Processor that is capable of processing a document by performing an HTTP GET, POST or similar action on an RDFa Processor IRI. For this class of RDFa Processor, the software MUST allow the caller to specify if they would like to retrieve the output graph, the processor graph, or both graphs as a single, combined graph from the web service. The rdfagraph query parameter MUST be used to specify the value. The allowable values are output, processor or both values, in any order, separated by a comma character. If the graph preference is not specified, the output graph MUST be returned.

Processor Graph Terms

To ensure interoperability, a core hierarchy of classes is defined for the content of the processor graph. Separate errors or warnings are resources (typically blank nodes) of a specific type, with additional properties giving more details on the error condition or the warning. This specification defines only the top level classes and the ones referring to the error and warning conditions defined explicitly by this document. Other, implementation-specific subclasses may be defined by the RDFa Processor.

The top level classes are rdfa:Error, rdfa:Warning, and rdfa:Info, defined as part of the RDFa Vocabulary. Furthermore, a single property is defined on those classes, namely rdfa:context, that provides an extra context for the error, e.g., http response, an XPath information, or simply the IRI to the RDFa resource. Usage of this property is optional, and more than one triple can be used with this predicate on the same subject. Finally, error and warning instances SHOULD use the dc:description and dc:date properties. dc:description should provide a short, human readable but implementation dependent description of the error. dc:date should give the time when the error was found and it is advised to be as precise as possible to allow the detection of, for example, possible network errors.

The example below shows the triples that should be minimally present in the processor graph as a result of an error (the content of the literal for the dc:description predicate is implementation dependent):

A slightly more elaborate example makes use of the rdfa:context property to provide further information, using external vocabularies to represent HTTP headers or XPointer information (note that a processor may not have these information in all cases, i.e., these rdfa:context information are not required):

Vocabulary Expansion

Processors MAY perform vocabulary expansion by utilizing limited RDFS and OWL entailment rules, as described in RDFa Vocabulary Expansion.

RDFa Processing in detail

This section provides an in-depth examination of the processing steps described in the previous section. It also includes examples which may help clarify some of the steps involved.

The key to processing is that a triple is generated whenever a predicate/object combination is detected. The actual triple generated will include a subject that may have been set previously, so this is tracked in the current evaluation context and is called the parent subject. Since the subject will default to the current document if it hasn't been set explicitly, then a predicate/object combination is always enough to generate one or more triples.

The attributes for setting a predicate are rel, rev and property, whilst the attributes for setting an object are resource, href, content, and src. typeof is unique in that it sets both a predicate and an object at the same time (and also a subject when it appears in the absence of other attributes that would set a subject). Inline content might also set an object, if content is not present, but property is present.

There are many examples in this section. The examples are all written using XHTML+RDFa. However, the explanations are relevant regardless of the Host Language.

Changing the Evaluation Context

Setting the current subject

When triples are created they will always be in relation to a subject resource which is provided either by new subject (if there are rules on the current element that have set a subject) or parent subject, as passed in via the evaluation context. This section looks at the specific ways in which these values are set. Note that it doesn't matter how the subject is set, so in this section we use the idea of the current subject which may be either new subject or parent subject.

The current document

When parsing begins, the current subject will be the IRI of the document being parsed, or a value as set by a Host Language-provided mechanism (e.g., the base element in (X)HTML). This means that by default any metadata found in the document will concern the document itself:

This would generate the following triples:

It is possible for the data to appear elsewhere in the document:

which would still generate the triple:

In (X)HTML the value of base may change the initial value of current subject:

An RDFa Processor should now generate the following triples, regardless of the IRI from which the document is served:

Using about

As processing progresses, any about attributes will change the current subject. The value of about is an IRI or a CURIE. If it is a relative IRI then it needs to be resolved against the current base value. To illustrate how this affects the statements, note in this markup how the properties inside the (X)HTML body element become part of a new calendar event object, rather than referring to the document as they do in the head of the document:

With this markup an RDFa Processor will generate the following triples:

Other kinds of resources can be used to set the current subject, not just references to web-pages. Although not advised, email addresses might be used to represent a person:

This should generate the following triples:

Similarly, authors may make statements about images:

which should generate the following triple:

Typing resources with typeof

typeof defines typing triples. typeof works differently to other ways of setting a predicate since the predicate is always rdf:type, which means that the processor only requires the value of the type. The question is: which resource gets these typing information?

If the element has an about, which creates a new context for statements, the typing relationships are defined on that resource. For example, the following:

also creates the triple:

The about attribute is the main source for typing; if it is present on an element, it determines the effect of typeof with the highest priority. If about is not present, but the element is used only to define possible subject resources via, e.g., resource (i.e., there is no rel, rev, or property present), then that resource is used for the typed resource, just like about.

If an rel is present (and still no about) then the explicit object of the triples defined by rel is typed. For example, in the case of:

the generated triples also include:

Finally, typeof also has the additional feature of creating a new context for statements, in case no other attributes define any. This involves generating a new bnode (see below for more about bnodes). For example, an author may wish to create markup for a person using the FOAF vocabulary, but without having a clear identifier for the item:

This markup would cause a bnode to be created which has a 'type' of foaf:Person, as well as name and given name properties:

This usage of “isolated” typeof may be viewed as a shorthand for:

Similarly,

generates:

A bnode is simply a unique identifier that is only available to the processor, not to any external software. By generating values internally, the processor is able to keep track of properties for _:a as being distinct from _:b. But by not exposing these values to any external software, it is possible to have complete control over the identifier, as well as preventing further statements being made about the item.
Chaining with property and typeof

As emphasized in the section on chaining, one of the main differences between property and rel (or rev) is that the former does not induce chaining. The only exception to this rule is when typeof is also present on the element. In that case the effect of property is identical to rel. For example, the previous example could have been written as:

generating the same triples as before. Here again, a typeof without an about or a resource can be regarded as a shorthand for an additional resource attribute referring to the identifier of a fresh bnode.

Determining the subject with neither about nor typeof

As described in the previous two sections, about will always take precedence and mark a new subject, but if no about value is available then typeof will do the same job, although using an implied identifier, i.e., a bnode.

But if neither about or typeof are present, there are a number of ways that the subject could be arrived at. One of these is to 'inherit' the subject from the containing statement, with the value to be inherited set either explicitly, or implicitly.

Inheriting subject from resource

The most usual way that an inherited subject might get set would be when the parent statement has an object that is a resource. Returning to the earlier example, in which the long name for the German_Empire was added, the following markup was used:

In an earlier illustration the subject and object for the German Empire were connected by removing the resource, relying on the about to set the object:

but it is also possible for authors to achieve the same effect by removing the about and leaving the resource:

In this situation, all statements that are 'contained' by the object resource representing the German Empire (the value in resource) will have the same subject, making it easy for authors to add additional statements:

Looking at the triples that an RDFa Processor would generate, we can see that we actually have two groups of statements; the first group is set to refer to the about that contains them:

while the second group refers to the resource that contains them:

Note also that the same principle described here applies to src and href.

Inheriting an anonymous subject

There will be occasions when the author wants to connect the subject and object as shown above, but is not concerned to name the resource that is common to the two statements (i.e., the object of the first statement, which is the subject of the second). For example, to indicate that Einstein was influenced by Spinoza the following markup could well be used:

An RDFa Processor will generate the following triples:

However, an author could just as easily say that Spinoza influenced something by the name of Albert Einstein, that was born on March 14th, 1879:

In RDF terms, the item that 'represents' Einstein is anonymous, since it has no IRI to identify it. However, the item is given an automatically generated bnode, and it is onto this identifier that all child statements are attached:

An RDFa Processor will generate the following triples:

Note that the div is superfluous, and an RDFa Processor will create the intermediate object even if the element is removed:

An alternative pattern is to keep the div and move the rel onto it:

From the point of view of the markup, this latter layout is to be preferred, since it draws attention to the 'hanging rel'. But from the point of view of an RDFa Processor, all of these permutations need to be supported.

Completing incomplete triples

When a new subject is calculated, it is also used to complete any incomplete triples that are pending. This situation arises when the author wants to 'chain' a number of statements together. For example, an author could have a statement that Albert Einstein was born in the German Empire:

and then a further statement that the 'long name' for this country is the German Empire:

RDFa allows authors to insert this statement as a self-contained unit into other contexts:

But it also allows authors to avoid unnecessary repetition and to 'normalize' out duplicate identifiers, in this case the one for the German Empire:

When this happens the rel for 'birth place' is regarded as a 'hanging rel' because it has not yet generated any triples, but these 'incomplete triples' are completed by the about that appears on the next line. The first step is therefore to store the two parts of the triple that the RDFa Processor does have, but without an object:

Then as processing continues, the RDFa Processor encounters the subject of the statement about the long name for the German Empire, and this is used in two ways. First it is used to complete the 'incomplete triple':

and second it is used to generate its own triple:

Note that each occurrence of about will complete any incomplete triples. For example, to mark up the fact that Albert Einstein had a residence both in the German Empire and Switzerland, an author need only specify one rel value that is then used with multiple about values:

In this example there is one incomplete triple:

When the processor meets each of the about values, this triple is completed, giving:

These examples show how about completes triples, but there are other situations that can have the same effect. For example, when typeof creates a new bnode (as described above), that will be used to complete any 'incomplete triples'. To indicate that Spinoza influenced both Einstein and Schopenhauer, the following markup could be used:

First the following incomplete triple is stored:

Then when the RDFa Processor processes the two occurrences of typeof, each generates a bnode, which is used to both complete the 'incomplete triple', and to set the subject for further statements:

Triples are also 'completed' if any one of property, rel or rev are present. However, unlike the situation when about or typeof are present, all predicates are attached to one bnode:

This example has two 'hanging rels', and so two situations when 'incomplete triples' will be created. Processing would proceed as follows; first an incomplete triple is stored:

Next, the RDFa Processor processes the predicate values for foaf:name, dbp:dateOfBirth and dbp-owl:residence, but note that only the first needs to 'complete' the 'hanging rel'. So processing foaf:name generates two triples:

but processing dbp:dateOfBirth generates only one:

Processing dbp-owl:residence also uses the same bnode, but note that it also generates its own 'incomplete triple':

As before, the two occurrences of about complete the 'incomplete triple', once each:

The entire set of triples that an RDFa Processor should generate is as follows:

Object resolution

Although objects have been discussed in the previous sections, as part of the explanation of subject resolution, chaining, evaluation contexts, and so on, this section will look at objects in more detail.

There are two types of object, IRI resources and literals.

A literal object can be set by content or the inline text of element if property to express a predicate. Note that the use of content prohibits the inclusion of rich markup in your literal. If the inline content of an element accurately represents the object, then documents should rely upon that rather than duplicating that data using the content.

An IRI resource object can be set using one of rel or rev to express a predicate, and then either using one of href, resource or src to provide an object resource explicitly, or using the chaining techniques described above to obtain an object from a nested subject, or from a bnode. Alternatively, the property can also be used to define an IRI resource; this requires the presence of a resource, href, or src and the absence of rel, rev, datatype, or content.

Object resolution for the property attribute

An object literal will be generated when property is present and no resource attribute is present. property provides the predicate, and the following sections describe how the actual literal to be generated is determined.

Plain Literals

content can be used to indicate a plain literal, as follows:

The plain literal can also be specified by using the content of the element:

Both of these examples give the following triple:

The value of content is given precedence over any element content, so the following would give exactly the same triple as shown above:

Language Tags

RDF allows plain literals to have a language tag, as illustrated by the following example from [[RDF11-TESTCASES]]:

In RDFa the Host Language may provide a mechanism for setting the language tag. In XHTML+RDFa [[!XHTML-RDFA]], for example, the XML language attribute @xml:lang or the attribute @lang is used to add this information, whether the plain literal is designated by content, or by the inline text of the element:

Note that the language value can be inherited as defined in [[!XML10-4e]], so the following syntax will give the same triple as above:

Typed Literals

Literals can be given a data type using datatype.

This can be represented in RDFa as follows:

The triple that this markup generates includes the datatype after the literal:

XML Literals

XML documents cannot contain XML markup in their attributes, which means it is not possible to represent XML within content (the following would cause an XML parser to generate an error):

RDFa therefore supports the use of arbitrary markup to express XML literals by using datatype:

This would generate the following triple, with the XML preserved in the literal:

This requires that an IRI mapping for the prefix rdf has been defined.

In the examples given here the sup element is actually part of the meaning of the literal, but there will be situations where the extra markup means nothing, and can therefore be ignored. In this situation omitting the datatype attribute or specifying an empty datatype value can be used to create a plain literal:

Rendering of this page has highlighted the term the user searched for. Setting datatype to nothing ensures that the data is interpreted as a plain literal, giving the following triple:

The value of this XML Literal is the exclusive canonicalization [[XML-EXC-C14N]] of the RDFa element's value.

IRI object resolution

Most of the rules governing the processing of objects that are resources are to be found in the processing descriptions given above, since they are important for establishing the subject. This section aims to highlight general concepts, and anything that might have been missed.

One or more IRI objects are needed when rel or rev is present. Each attribute will cause triples to be generated when used with href, resource or src, or with the subject value of any nested statement if none of these attributes are present.

If rel or rev is not present, and neither is datatype or content, a property attribute will cause triples to be generated when used with href, resource or src. (See also the section on @property and @typeof for an additional special case involving property.)

rel and rev are essentially the inverse of each other; whilst rel establishes a relationship between the current subject as subject, and the current object resource as the object, rev does the exact opposite, and uses the current object resource as the subject, and the current subject as the object.

Using resource to set the object

RDFa provides the resource attribute as a way to set the object of statements. This is particularly useful when referring to resources that are not themselves navigable links:

The blockquote element generates the following triple:

Note that, in the example above, property could have been used instead of rel, yielding the same triple.

Using href or src to set the object

If no resource is present, then href or src are next in priority order for setting the object.

When a predicate has been expressed using rel, the href or src on the RDFa statement's element is used to identify the object with a IRI reference. Their types are an IRI:

It's also possible to use both rel and rev at the same time on an element. This is particularly useful when two things stand in two different relationships with each other, for example when a picture is taken by Mark, but that picture also depicts him:

which then yields two triples:

Incomplete triples

When a triple predicate has been expressed using rel or rev, but no href, src, or resource exists on the same element, there is a 'hanging rel'. This causes the current subject and all possible predicates (with an indicator of whether they are 'forwards, i.e., rel values, or not, i.e., rev values), to be stored as 'incomplete triples' pending discovery of a subject that could be used to 'complete' those triples.

This process is described in more detail in Completing 'Incomplete Triples'.

List Generation

An RDF graph is a collection of triples. This also means that if the graph contains two triples sharing the same subject and predicate:

There is no way for an application to rely on the relative order of the two triples when, for example, querying a database containing these triples. For most of the applications and data sets this is not a problem, but, in some cases, the order is important. A typical case is publications: when a book or an article has several co-authors, the order of the authors may be important.

RDF has a set of predefined predicates that have an agreed-upon semantic of order. For example, the publication: "Semantic Annotation and Retrieval, by Ben Adida, Mark Birbeck, and Ivan Herman" could be described in RDF triples using these terms as follows:

which conveys the notion of 'order' for the three authors. Admittedly, this is not very readable. However, Turtle has a syntactic shorthand for these structures:

It would of course be possible to reproduce the same structure in RDFa, using the RDF predicates rdf:first, rdf:rest, as well as the special resource rdf:nil. However, to make this easier, RDFa provides the inlist. What this attribute signals is that the object generated on that element should be put on a list; the list is used with the common predicate and subject. Here is how the previous structure could look in RDFa:

Note that the order in the list is determined by the document order. (The value of the inlist is not relevant, only its presence is.)

Lists may also include IRIs and not only literals. For example, two of the three co-authors could decide to publicise their FOAF address in the authors’ list:

yielding:

In the example above, rel could have been used leading exactly to the same triples:

Incomplete Triples can also be used in conjunction with lists when all list elements are resources and not literals. For example, the previous example, this time with all three authors referring to their FOAF profile, could have been written as:

Resulting in:

Note that it is also possible to express an empty list, without inlist, using:

RDFa Initial Contexts

RDFa permits Host Languages to define an initial context. Such a context is a collection of terms, prefix mappings, and/or a default vocabulary declaration. An initial context is either intrinsically known to the parser, or it is loaded as external documents and processed. These documents MUST be defined in an approved RDFa Host Language (currently XML+RDFa, XHTML+RDFa [[!XHTML-RDFA]], and HTML+RDFa [[!HTML-RDFA]]). They MAY also be defined in other formats (e.g., RDF/XML [[!RDF-SYNTAX-GRAMMAR]], or Turtle [[TURTLE]]). When an initial context document is processed, it is evaluated as follows:

  1. Parse the content (according to the processing rules for that document type) and extract the triples into a collection associated with that IRI. Note: These triples MUST NOT be co-mingled with the triples being extracted from any other IRI.
  2. For every subject with a pair of predicates that have the values rdfa:prefix and rdfa:uri, create a key-value mapping from the rdfa:prefix object literal (the key) to the rdfa:uri object literal (the value). Add this mapping to the list of IRI mappings of the initial evaluation context, after transforming the 'prefix' component to lower-case.
  3. For every subject with a pair of predicates that have the values rdfa:term and rdfa:uri, create a key-value mapping from the rdfa:term object literal (the key) to the rdfa:uri object literal (the value). Add this mapping to the term mappings of the initial evaluation context.
  4. For an extracted triple that has a predicate of rdfa:vocabulary, define the default vocabulary of the initial evaluation context to be the object literal of the rdfa:vocabulary predicate.

When an RDFa Initial Context is defined using an RDF serialization, it MUST use the vocabulary terms above to declare the components of the context.

Caching of the relevant triples retrieved via this mechanism is RECOMMENDED. Embedding definitions for well known, stable RDFa Initial Contexts in the implementation is RECOMMENDED.

RDFa Vocabulary Expansion

Since RDFa is based on RDF, the semantics of RDF vocabularies can be used to gain more knowledge about data. Vocabularies, properties and classes are identified by IRIs, which enables them to be discoverable. RDF data published at the location of these IRIs can be retrieved, and descriptions of the properties and classes using specified semantics can be applied.

RDFa Vocabulary Expansion is an optional processing step which may be added once the normal processing steps described in Processing Model are complete. Vocabulary expansion relies on a very small sub-set of OWL entailment [[!OWL2-OVERVIEW]] to add triples to the output graph based on rules and property/class relationships described in referenced vocabularies. Vocabulary expansion MAY be performed as part of a larger RDF toolset including, for example, an OWL 2 RL reasoner. Alternatively, using vocabulary data added to the output graph in processing step 2 of Sequence, expansion MAY also be done using a separate and dedicated (e.g., rule based) reasoner after the output graph has been generated, or as the last processing step by an RDFa processor.

It can be very useful to make generalized data available for subsequent usage of RDFa-embedded data by expanding inferred statements entailed by these semantics. This provides for existing vocabularies that extend well-known vocabularies to have those properties added to the output graph automatically. For example, the namespace document of the Creative Commons vocabulary, i.e., http://creativecommons.org/ns, defines cc:license to be a sub-property of dc:license. By using the vocab attribute, one can describe a licensing information as follows:

which results in the following output graph:

After vocabulary expansion, the output graph contains:

Other vocabularies, specifically intended to provide relations to multiple vocabularies, could also be defined by publishers, allowing use of terms in a single namespace which result in properties and/or classes from other primary vocabularies being imported. This benefits publishers as data is now more widely searchable and encourages the practice of referencing well-known vocabularies.

Details of the RDFa Vocabulary Expansion

Once the output graph is generated following the processing steps defined in Sequence, processors MAY perform the following processing steps on the output graph. It must do so only if the user of the processor explicitly asks for it, as prescribed in Vocabulary Expansion Control of RDFa Processors.

A vocabulary graph is created as follows: Each object IRI in the output graph that has a subject the current document (base) IRI and a predicate of rdfa:usesVocabulary is dereferenced. If the dereferencing yields the serialization of an RDF graph, that serialization is parsed and the resulting graph is merged with the vocabulary graph. (An RDFa processor capable of vocabulary expansion MUST accept an RDF graph serialized in RDFa, and SHOULD accept other standard serialization formats of RDF such as RDF/XML [[!RDF-SYNTAX-GRAMMAR]] and Turtle [[TURTLE]].)

Note that if, in the second step, a particular vocabulary is serialized in RDFa, that particular graph is not expected to undergo any vocabulary expansion on its own.

Vocabulary expansion is then performed as follows:

  1. The processor operates on the merge of the default and vocabulary graphs using RDFa Vocabulary Entailment.
  2. Add the new triples inferred from the output graph using this entailment to the (expanded) output graph. The processor SHOULD NOT add the triples appearing in the vocabulary graph only.
The goal of the second step is to avoid adding the "axioms", e.g., the sub-property definitions to the output graph. Applications usually do not require any of this additional information.

RDFa Vocabulary Entailment

For the purpose of vocabulary processing, RDFa used a very restricted subset of the OWL vocabulary and is based on the RDF-Based Semantics of OWL [[!OWL2-RDF-BASED-SEMANTICS]]. The RDFa Vocabulary Entailment uses the following terms:

  • rdf:type
  • rdfs:subClassOf
  • rdfs:subPropertyOf
  • owl:equivalentClass
  • owl:equivalentProperty

RDFa Vocabulary Entailment considers only the entailment on individuals (i.e., not on the relationships that can be deduced on the properties or the classes themselves.)

While the formal definition of the RDFa Entailment refers to the general OWL 2 Semantics, practical implementations may rely on a subset of the OWL 2 RL Profile’s entailment expressed in rules (section 4.3 of [[!OWL2-PROFILES]]). The relevant rules are, using the rule identifications in section 4.3 of [[!OWL2-PROFILES]]): prp-spo1, prp-eqp1, prp-eqp2, cax-sco, cax-eqc1, and cax-eqc2.

The entailment described in this section is the minimum useful level for RDFa. Processors may, of course, choose to follow more powerful entailment regimes, e.g., include full RDFS [[!RDF11-MT]] or OWL [[!OWL2-OVERVIEW]] entailments. Using those entailments applications may perform datatype validation by checking rdfs:range of a property, or use the advanced facilities offered by, e.g., OWL’s property chains to interlink vocabularies further.

Vocabulary Expansion Control of RDFa Processors

Conforming RDFa processors are not required to provide vocabulary expansion.

If an RDFa processor provides vocabulary expansion, it MUST NOT be performed by default. Instead, the processor MUST provide an option, vocab_expansion, which, when used, instructs the RDFa processor to perform a vocabulary expansion before returning the output graph.

Although vocabulary expansion is described in terms of a vocabulary graph and OWL 2 entailment rules, processors are free to use any process which obtains equivalent results.

Notes to RDFa Vocabulary Implementations and Publishing

For RDFa Processors caching the relevant graphs retrieved via this mechanism is RECOMMENDED. Caching is usually based on HTTP response headers like expiration time, cache control, etc.

For publishers of vocabularies, the IRI for the vocabularies SHOULD be dereferenceable, and should return an RDF graph with the vocabulary description. This vocabulary description SHOULD be available encoded in RDFa, and MAY also be available in other RDF serialization syntaxes (using content negotiation to choose among the different formats). If possible, vocabulary descriptions SHOULD include subproperty and subclass statements linking the vocabulary terms to other, well-known vocabularies. Finally, HTTP responses SHOULD include fields usable for cache control, e.g., expiration date.

CURIE Datatypes

In order to facilitate the use of CURIEs in markup languages, this specification defines some additional datatypes in the XHTML datatype space (http://www.w3.org/1999/xhtml/datatypes/). Markup languages that want to import these definitions can find them in the "datatypes" file for their schema grammar:

Specifically, the following datatypes are defined:

CURIE
A single curie
CURIEs
A white space separated list of CURIEs
CURIEorIRI
A CURIE or an IRI
CURIEorIRIs
A white space separated list of CURIEorIRIs
SafeCURIE
A single safe_curie
SafeCURIEorCURIEorIRI
A single SafeCURIE or CURIEorIRI
SafeCURIEorCURIEorIRIs
A white space separated list of SafeCURIEorCURIEorIRIs.
TERM
A single term
TERMorCURIEorAbsIRI
A TERM or a CURIEorIRI
TERMorCURIEorAbsIRIs
A white space separated list of TERMorCURIEorAbsIRIs

XML Schema Definition

The following informative XML Schema definition for these datatypes is included as an example:

XML DTD Definition

The following informative XML DTD definition for these datatypes is included as an example:

The RDFa Vocabulary

The RDFa Vocabulary has three roles: it contains the predicates to define the terms and prefixes in initial context documents, it contains the classes and predicates for the messages that a processor graph may contain and, finally, it contains the predicate necessary for vocabulary processing. The IRI of the vocabulary is http://www.w3.org/ns/rdfa#; the usual prefix used in this document is rdfa.

This vocabulary specification is available in XHTML+RDFa 1.1, Turtle, and in RDF/XML formats.

Term and Prefix Assignments

The RDFa Vocabulary includes the following triples (shown here in Turtle [[TURTLE]] format):

These predicates can be used to define the initial context for a given Host Language.

These predicates are used to 'pair' IRI strings and their usage in the form of a prefix and/or a term as part of, for example, a blank node. An example can be as follows:

which defines a prefix for the FOAF IRI.

Processor Graph Reporting

The Vocabulary includes the following term definitions (shown here in Turtle [[TURTLE]] format):

Term for vocabulary expansion

The Vocabulary includes the following term definitions (shown here in Turtle [[TURTLE]] format):

Changes

Major differences since the Last Published Recommentation

Major differences with RDFa Syntax 1.0

This specification introduces a number of new features, and extends the behavior of some features from the previous version. The following summary may be helpful to RDFa Processor developers, but is not meant to be comprehensive.

While this specification strives to be as backward compatible as possible with [[RDFA-SYNTAX]], the changes above mean that there are some circumstances where it is possible for different RDF triples to be output for the same document when processed by an RDFa 1.0 processor vs. an RDFa 1.1 processor. In order to minimize these differences, a document author can do the following:

When producing XHTML+RDFa 1.1 documents, it is possible to reduce the incompatibilities with RDFa 1.0 conforming processors by doing the following:

Acknowledgments

At the time of publication, the active members of the RDFa Working Group were: