RDF Primer — Turtle version

2.1 Basic Concepts

Imagine trying to state that someone named John Smith created a particular Web page. A straightforward way to state this in a natural language such as English would be in the form of a simple statement such as:

http://www.example.org/index.html has a creator whose value is John Smith

Parts of this statement are emphasized to illustrate that, in order to describe the properties of something, there need to be ways to name, or identify, a number of things:

• the thing the statement describes (the Web page, in this case)
• a specific property (creator, in this case) of the thing the statement describes
• the thing the statement says is the value of this property (who the creator is), for the thing the statement describes

In this statement, the Web page's URL (Uniform Resource Locator) is used to identify it. In addition, the word "creator" is used to identify the property, and the two words "John Smith" to identify the thing (a person) that is the value of this property.

Other properties of this Web page could be described by writing additional English statements of the same general form, using the URL to identify the page, and words (or other expressions) to identify the properties and their values. For example, the date the page was created, and the language in which the page is written, could be described using the additional statements:

http://www.example.org/index.html has a creation-date whose value is August 16, 1999
http://www.example.org/index.html has a language whose value is English

RDF is based on the idea that the things being described have properties which have values, and that resources can be described by making statements, similar to those above, that specify those properties and values. RDF uses a particular terminology for talking about the various parts of statements. Specifically, the part that identifies the thing the statement is about (the Web page in this example) is called the subject. The part that identifies the property or characteristic of the subject that the statement specifies (creator, creation-date, or language in these examples) is called the predicate, and the part that identifies the value of that property is called the object. So, taking the English statement

http://www.example.org/index.html has a creator whose value is John Smith

the RDF terms for the various parts of the statement are:

• the subject is the URL http://www.example.org/index.html
• the predicate is the word "creator"
• the object is the phrase "John Smith"

However, while English is good for communicating between (English-speaking) humans, RDF is about making machine-processable statements. To make these kinds of statements suitable for processing by machines, two things are needed:

• a system of machine-processable identifiers for identifying a subject, predicate, or object in a statement without any possibility of confusion with a similar-looking identifier that might be used by someone else on the Web.
• a machine-processable language for representing these statements and exchanging them between machines.

Fortunately, the existing Web architecture provides both these necessary facilities.

As illustrated earlier, the Web already provides one form of identifier, the Uniform Resource Locator (URL). A URL was used in the original example to identify the Web page that John Smith created. A URL is a character string that identifies a Web resource by representing its primary access mechanism (essentially, its network "location"). However, it is also important to be able to record information about many things that, unlike Web pages, do not have network locations or URLs.

The Web provides a more general form of identifier for these purposes, called the Uniform Resource Identifier (URI). URLs are a particular kind of URI. All URIs share the property that different persons or organizations can independently create them, and use them to identify things. However, URIs are not limited to identifying things that have network locations, or use other computer access mechanisms. In fact, a URI can be created to refer to anything that needs to be referred to in a statement, including

• network-accessible things, such as an electronic document, an image, a service (e.g., "today's weather report for Los Angeles"), or a group of other resources.
• things that are not network-accessible, such as human beings, corporations, and bound books in a library.
• abstract concepts that do not physically exist, such as the concept of a "creator".

Because of this generality, RDF uses URIs as the basis of its mechanism for identifying the subjects, predicates, and objects in statements. To be more precise, RDF uses URI references [URIS]. A URI reference (or URIref) is a URI, together with an optional fragment identifier at the end. For example, the URI reference http://www.example.org/index.html#section2 consists of the URI http://www.example.org/index.html and (separated by the "#" character) the fragment identifier Section2. RDF URIrefs can contain Unicode [UNICODE] characters (see [RDF-CONCEPTS]), allowing many languages to be reflected in URIrefs. RDF defines a resource as anything that is identifiable by a URI reference, so using URIrefs allows RDF to describe practically anything, and to state relationships between such things as well. URIrefs and fragment identifiers are discussed further in Appendix A, and in [RDF-CONCEPTS].

To represent RDF statements in a machine-processable way, RDF normatively uses the Extensible Markup Language [XML], but other syntaxes are also possible. The XML serialization syntax (referrred to as RDF/XML) is described in a separate document [RDF-SYNTAX]. This document uses the Turtle syntax [TURTLE]. An example of Turtle was given in Section 1. Turtle content can contain Unicode [UNICODE] characters, allowing information from many languages to be directly represented. The specific Turtle syntax is defined in [TURTLE]

2.2 The RDF Model

Section 2.1 has introduced RDF's basic statement concepts, the idea of using URI references to identify the things referred to in RDF statements, and Turtle as a machine-processable way to represent RDF statements. With that background, this section describes how RDF uses URIs to make statements about resources. The introduction said that RDF was based on the idea of expressing simple statements about resources, where each statement consists of a subject, a predicate, and an object. In RDF, the English statement:

http://www.example.org/index.html has a creator whose value is John Smith

could be represented by an RDF statement having:

• a subject http://www.example.org/index.html
• a predicate http://purl.org/dc/elements/1.1/creator
• and an object http://www.example.org/staffid/85740

Note how URIrefs are used to identify not only the subject of the original statement, but also the predicate and object, instead of using the words "creator" and "John Smith", respectively (some of the effects of using URIrefs in this way will be discussed later in this section).

RDF models statements as nodes and arcs in a graph. RDF's graph model is defined in [RDF-CONCEPTS]. In this notation, a statement is represented by:

• a node for the subject
• a node for the object
• an arc for the predicate, directed from the subject node to the object node.

So the RDF statement above would be represented by the graph shown in Figure 2:

Figure 2: A Simple RDF Statement

Groups of statements are represented by corresponding groups of nodes and arcs. So, to reflect the additional English statements

http://www.example.org/index.html has a creation-date whose value is August 16, 1999
http://www.example.org/index.html has a language whose value is English

in the RDF graph, the graph shown in Figure 3 could be used (using suitable URIrefs to name the properties "creation-date" and "language"):

Figure 3: Several Statements About the Same Resource

Figure 3 illustrates that objects in RDF statements may be either URIrefs, or constant values (called literals) represented by character strings, in order to represent certain kinds of property values. (In the case of the predicate http://purl.org/dc/elements/1.1/language the literal is an international standard two-letter code for English.) Literals may not be used as subjects or predicates in RDF statements. In drawing RDF graphs, nodes that are URIrefs are shown as ellipses, while nodes that are literals are shown as boxes. (The simple character string literals used in these examples are called plain literals, to distinguish them from the typed literals to be introduced in Section 2.4. The various kinds of literals that can be used in RDF statements are defined in [RDF-CONCEPTS]. Both plain and typed literals can contain Unicode [UNICODE] characters, allowing information from many languages to be directly represented.)

Sometimes it is not convenient to draw graphs when discussing them, so an alternative way of writing down the statements, called triples, is also used. In the triples notation, each statement in the graph is written as a simple triple of subject, predicate, and object, in that order. For example, the three statements shown in Figure 3 would be written in the triples notation as:

<http://www.example.org/index.html> <http://purl.org/dc/elements/1.1/creator> <http://www.example.org/staffid/85740> .
<http://www.example.org/index.html> <http://www.example.org/terms/creation-date> "August 16, 1999" .
<http://www.example.org/index.html> <http://purl.org/dc/elements/1.1/language> "en" .

Each triple corresponds to a single arc in the graph, complete with the arc's beginning and ending nodes (the subject and object of the statement). Unlike the drawn graph (but like the original statements), the triples notation requires that a node be separately identified for each statement it appears in. So, for example, http://www.example.org/index.html appears three times (once in each triple) in the triples representation of the graph, but only once in the drawn graph. However, the triples represent exactly the same information as the drawn graph, and this is a key point: what is fundamental to RDF is the graph model of the statements. The notation used to represent or depict the graph is secondary.

The full triples notation requires that URI references be written out completely, in angle brackets, which, as the example above illustrates, can result in very long lines on a page. For convenience, the Primer uses a shorthand way of writing triples (the same shorthand is also used in other RDF specifications). This shorthand substitutes a qualified name (or QName) without angle brackets as an abbreviation for a full URI reference . A QName contains a prefix that has been assigned to a namespace URI, followed by a colon, and then a local name. The full URIref is formed from the QName by appending the local name to the namespace URI assigned to the prefix. So, for example, if the QName prefix foo is assigned to the namespace URI http://example.org/somewhere/, then the QName foo:bar is shorthand for the URIref http://example.org/somewhere/bar. Primer examples will also use several "well-known" QName prefixes (without explicitly specifying them each time), defined as follows:

prefix rdf:, namespace URI: http://www.w3.org/1999/02/22-rdf-syntax-ns#
prefix rdfs:, namespace URI: http://www.w3.org/2000/01/rdf-schema#
prefix dc:, namespace URI: http://purl.org/dc/elements/1.1/
prefix owl:, namespace URI: http://www.w3.org/2002/07/owl#
prefix ex:, namespace URI: http://www.example.org/ (or http://www.example.com/)
prefix xsd:, namespace URI: http://www.w3.org/2001/XMLSchema#

Obvious variations on the "example" prefix ex: will also be used as needed in the examples, for instance,

prefix exterms:, namespace URI: http://www.example.org/terms/ (for terms used by an example organization),
prefix exstaff:, namespace URI: http://www.example.org/staffid/ (for the example organization's staff identifiers),
prefix ex2:, namespace URI: http://www.domain2.example.org/ (for a second example organization), and so on.

Using this new shorthand, the previous set of triples can be written as:

ex:index.html  dc:creator             exstaff:85740 .
ex:index.html  exterms:creation-date  "August 16, 1999" .
ex:index.html  dc:language            "en" .

ex:index.html  dc:creator  exstaff:85740 .
ex:index.html  dc:creator  exstaff:27354 .
ex:index.html  dc:creator  exstaff:00816 .

These examples of RDF statements begin to illustrate some of the advantages of using URIrefs as RDF's basic way of identifying things. For instance, in the first statement, instead of identifying the creator of the Web page by the character string "John Smith", he has been assigned a URIref, in this case (using a URIref based on his employee number) http://www.example.org/staffid/85740 . An advantage of using a URIref in this case is that the identification of the statement's subject can be more precise. That is, the creator of the page is not the character string "John Smith", or any one of the thousands of people named John Smith, but the particular John Smith associated with that URIref (whoever created the URIref defines the association). Moreover, since there is a URIref to refer to John Smith, he is a full-fledged resource, and additional information can be recorded about him, simply by adding additional RDF statements with John's URIref as the subject. For example, Figure 4 shows some additional statements giving John's name and age.

Figure 4: More Information About John Smith

These examples also illustrate that RDF uses URIrefs as predicates in RDF statements. That is, rather than using character strings (or words) such as "creator" or "name" to identify properties, RDF uses URIrefs. Using URIrefs to identify properties is important for a number of reasons. First, it distinguishes the properties one person may use from different properties someone else may use that would otherwise be identified by the same character string. For instance, in the example in Figure 4, example.org uses "name" to mean someone's full name written out as a character string literal (e.g., "John Smith"), but someone else may intend "name" to mean something different (e.g., the name of a variable in a piece of program text). A program encountering "name" as a property identifier on the Web (or merging data from multiple sources) would not necessarily be able to distinguish these uses. However, if example.org writes http://www.example.org/terms/name for its "name" property, and the other person writes http://www.domain2.example.org/genealogy/terms/name for hers, it is clear that there are distinct properties involved (even if a program cannot automatically determine the distinct meanings). Also, using URIrefs to identify properties enables the properties to be treated as resources themselves. Since properties are resources, additional information can be recorded about them (e.g., the English description of what example.org means by "name"), simply by adding additional RDF statements with the property's URIref as the subject.

Using URIrefs as subjects, predicates, and objects in RDF statements supports the development and use of shared vocabularies on the Web, since people can discover and begin using vocabularies already used by others to describe things, reflecting a shared understanding of those concepts. For example, in the triple

ex:index.html   dc:creator   exstaff:85740 .

the predicate dc:creator, when fully expanded as a URIref, is an unambiguous reference to the "creator" attribute in the Dublin Core metadata attribute set (discussed further in Section 6.1), a widely-used set of attributes (properties) for describing information of all kinds. The writer of this triple is effectively saying that the relationship between the Web page (identified by http://www.example.org/index.html ) and the creator of the page (a distinct person, identified by http://www.example.org/staffid/85740 ) is exactly the concept identified by http://purl.org/dc/elements/1.1/creator. Another person familiar with the Dublin Core vocabulary, or who finds out what dc:creator means (say by looking up its definition on the Web) will know what is meant by this relationship. In addition, based on this understanding, people can write programs to behave in accordance with that meaning when processing triples containing the predicate dc:creator.

Of course, this depends on increasing the general use of URIrefs to refer to things instead of using literals; e.g., using URIrefs like exstaff:85740 and dc:creator instead of character string literals like John Smith and creator. Even then, RDF's use of URIrefs does not solve all identification problems because, for example, people can still use different URIrefs to refer to the same thing. For this reason, it is a good idea to try to use terms from existing vocabularies (such as the Dublin Core) where possible, rather than making up new terms that might overlap with those of some other vocabulary. Appropriate vocabularies for use in specific application areas are being developed all the time, as illustrated by the applications described in Section 6. However, even when synonyms are created, the fact that these different URIrefs are used in the commonly-accessible "Web space" provides the opportunity both to identify equivalences among these different references, and to migrate toward the use of common references.

ex:index.html  dc:creator  exstaff:85740 .

fy:joefy.iunm  ed:dsfbups  fytubgg:85740 .

The result of all this is that RDF provides a way to make statements that applications can more easily process. An application cannot actually "understand" such statements, as noted already, any more than a database system "understands" terms like "employee" or "salary" in processing a query like SELECT NAME FROM EMPLOYEE WHERE SALARY > 35000. However, if an application is appropriately written, it can deal with RDF statements in a way that makes it seem like it does understand them, just as a database system and its applications can do useful work in processing employee and payroll information without understanding "employee" and "payroll". For example, a user could search the Web for all book reviews and create an average rating for each book. Then, the user could put that information back on the Web. Another Web site could take that list of book rating averages and create a "Top Ten Highest Rated Books" page. Here, the availability and use of a shared vocabulary about ratings, and a shared group of URIrefs identifying the books they apply to, allows individuals to build a mutually-understood and increasingly-powerful (as additional contributions are made) "information base" about books on the Web. The same principle applies to the vast amounts of information that people create about thousands of subjects every day on the Web.

RDF statements are similar to a number of other formats for recording information, such as:

• entries in a simple record or catalog listing describing the resource in a data processing system.
• rows in a simple relational database.
• simple assertions in formal logic

and information in these formats can be treated as RDF statements, allowing RDF to be used to integrate data from many sources.

2.3 Structured Property Values and Blank Nodes

Things would be very simple if the only types of information to be recorded about things were obviously in the form of the simple RDF statements illustrated so far. However, most real-world data involves structures that are more complicated than that, at least on the surface. For instance, in the original example, the date the Web page was created is recorded as a single exterms:creation-date property, with a plain literal as its value. However, suppose the value of the exterms:creation-date property needed to record the month, day, and year as separate pieces of information? Or, in the case of John Smith's personal information, suppose John's address was being described. The whole address could be written out as a plain literal, as in the triple

exstaff:85740   exterms:address   "1501 Grant Avenue, Bedford, Massachusetts 01730" .

However, suppose John's address needed to be recorded as a structure consisting of separate street, city, state, and postal code values? How would this be done in RDF?

Structured information like this is represented in RDF by considering the aggregate thing to be described (like John Smith's address) as a resource, and then making statements about that new resource. So, in the RDF graph, in order to break up John Smith's address into its component parts, a new node is created to represent the concept of John Smith's address, with a new URIref to identify it, say http://www.example.org/addressid/85740 (abbreviated as exaddressid:85740). RDF statements (additional arcs and nodes) can then be written with that node as the subject, to represent the additional information, producing the graph shown in Figure 5:

Figure 5: Breaking Up John's Address

or the triples:

exstaff:85740       exterms:address    exaddressid:85740 .
exaddressid:85740   exterms:street     "1501 Grant Avenue" .
exaddressid:85740   exterms:city       "Bedford" .
exaddressid:85740   exterms:state      "Massachusetts" .
exaddressid:85740   exterms:postalCode "01730" .

This way of representing structured information in RDF can involve generating numerous "intermediate" URIrefs such as exaddressid:85740 to represent aggregate concepts such as John's address. Such concepts may never need to be referred to directly from outside a particular graph, and hence may not require "universal" identifiers. In addition, in the drawing of the graph representing the group of statements shown in Figure 5, the URIref assigned to identify "John Smith's address" is not really needed, since the graph could just as easily have been drawn as in Figure 6:

Figure 6: Using a Blank Node

Figure 6, which is a perfectly good RDF graph, uses a node without a URIref to stand for the concept of "John Smith's address". This blank node serves its purpose in the drawing without needing a URIref, since the node itself provides the necessary connectivity between the various other parts of the graph. (Blank nodes were called anonymous resources in [RDF-MS].) However, some form of explicit identifier for that node is needed in order to represent this graph as triples. To see this, trying to write the triples corresponding to what is shown in Figure 6 would produce something like:

exstaff:85740   exterms:address      ??? .
???             exterms:street       "1501 Grant Avenue" .
???             exterms:city         "Bedford" .
???             exterms:state        "Massachusetts" .
???             exterms:postalCode   "01730" .

where ??? stands for something that indicates the presence of the blank node. Since a complex graph might contain more than one blank node, there also needs to be a way to differentiate between these different blank nodes in a triples representation of the graph. As a result, triples use blank node identifiers, having the form _:name, to indicate the presence of blank nodes. For instance, in this example a blank node identifier _:johnaddress might be used to refer to the blank node, in which case the resulting triples might be:

exstaff:85740   exterms:address     _:johnaddress .
_:johnaddress   exterms:street      "1501 Grant Avenue" .
_:johnaddress   exterms:city        "Bedford" .
_:johnaddress   exterms:state       "Massachusetts" .
_:johnaddress   exterms:postalCode  "01730" .

In a triples representation of a graph, each distinct blank node in the graph is given a different blank node identifier. Unlike URIrefs and literals, blank node identifiers are not considered to be actual parts of the RDF graph (this can be seen by looking at the drawn graph in Figure 6 and noting that the blank node has no blank node identifier). Blank node identifiers are just a way of representing the blank nodes in a graph (and distinguishing one blank node from another) when the graph is written in triple form. Blank node identifiers also have significance only within the triples representing a single graph (two different graphs with the same number of blank nodes might independently use the same blank node identifiers to distinguish them, and it would be incorrect to assume that blank nodes from different graphs having the same blank node identifiers are the same). If it is expected that a node in a graph will need to be referenced from outside the graph, a URIref should be assigned to identify it. Finally, because blank node identifiers represent (blank) nodes, rather than arcs, in the triple form of an RDF graph, blank node identifiers may only appear as subjects or objects in triples; blank node identifiers may not be used as predicates in triples.

The beginning of this section noted that aggregate structures, like John Smith's address, can be represented by considering the aggregate thing to be described as a separate resource, and then making statements about that new resource. This example illustrates an important aspect of RDF: RDF directly represents only binary relationships, e.g. the relationship between John Smith and the literal representing his address. Representing the relationship between John and the group of separate components of this address involves dealing with an n-ary (n-way) relationship (in this case, n=5) between John and the street, city, state, and postal code components. In order to represent such structures directly in RDF (e.g., considering the address as a group of street, city, state, and postal code components), this n-way relationship must be broken up into a group of separate binary relationships. Blank nodes provide one way to do this. For each n-ary relationship, one of the participants is chosen as the subject of the relationship (John in this case), and a blank node is created to represent the rest of the relationship (John's address in this case). The remaining participants in the relationship (such as the city in this example) are then represented as separate properties of the new resource represented by the blank node.

Blank nodes also provide a way to more accurately make statements about resources that may not have URIs, but that are described in terms of relationships with other resources that do have URIs. For example, when making statements about a person, say Jane Smith, it may seem natural to use a URI based on that person's email address as her URI, e.g., mailto:jane@example.org. However, this approach can cause problems. For example, it may be necessary to record information both about Jane's mailbox (e.g., the server it is on) as well as about Jane herself (e.g., her current physical address), and using a URIref for Jane based on her email address makes it difficult to know whether it is Jane or her mailbox that is being described. The same problem exists when a company's Web page URL, say http://www.example.com/, is used as the URI of the company itself. Once again, it may be necessary to record information about the Web page itself (e.g., who created it and when) as well as about the company, and using http://www.example.com/ as an identifier for both makes it difficult to know which of these is the actual subject.

The fundamental problem is that using Jane's mailbox as a stand-in for Jane is not really accurate: Jane and her mailbox are not the same thing, and hence they should be identified differently. When Jane herself does not have a URI, a blank node provides a more accurate way of modeling this situation. Jane can be represented by a blank node, and that blank node used as the subject of a statement with exterms:mailbox as the property and the URIref mailto:jane@example.org as its value. The blank node could also be described with an rdf:type property having a value of exterms:Person (types are discussed in more detail in the following sections), an exterms:name property having a value of "Jane Smith", and any other descriptive information that might be useful, as shown in the following triples:

_:jane   exterms:mailbox   <mailto:jane@example.org> .
_:jane   rdf:type          exterms:Person .
_:jane   exterms:name      "Jane Smith" .
_:jane   exterms:empID     "23748"  .
_:jane   exterms:age       "26" .

(Note that mailto:jane@example.org is written within angle brackets in the first triple. This is because mailto:jane@example.org is a full URIref in the mailto URI scheme, rather than a QName abbreviation, and full URIrefs must be enclosed in angle brackets in the triples notation.)

This says, accurately, that "there is a resource of type exterms:Person, whose electronic mailbox is identified by mailto:jane@example.org, whose name is Jane Smith, etc." That is, the blank node can be read as "there is a resource". Statements with that blank node as subject then provide information about the characteristics of that resource.

In practice, using blank nodes instead of URIrefs in these cases does not change the way this kind of information is handled very much. For example, if it is known that an email address uniquely identifies someone at example.org (particularly if the address is unlikely to be reused), that fact can still be used to associate information about that person from multiple sources, even though the email address is not the person's URI. In this case, if some RDF is found on the Web that describes a book, and gives the author's contact information as mailto:jane@example.org, it might be reasonable, combining this new information with the previous set of triples, to conclude that the author's name is Jane Smith. The point is that saying something like "the author of the book is mailto:jane@example.org" is typically a shorthand for "the author of the book is someone whose mailbox is mailto:jane@example.org". Using a blank node to represent this "someone" is just a more accurate way to represent the real world situation. (Incidentally, some RDF-based schema languages allow specifying that certain properties are unique identifiers of the resources they describe. This is discussed further in Section 5.5.)

ex2terms:book78354   rdf:type          ex2terms:Book .
ex2terms:book78354   ex2terms:author   "Jane Smith" .

ex2terms:book78354   rdf:type         ex2terms:Book .
ex2terms:book78354   ex2terms:author  _:author78354 .
_:author78354        rdf:type         ex2terms:Person .
_:author78354        ex2terms:name    "Jane Smith" .

2.4 Typed Literals

The last section described how to handle situations in which property values represented by plain literals had to be broken up into structured values to represent the individual parts of those literals. Using this approach, instead of, say, recording the date a Web page was created as a single exterms:creation-date property, with a single plain literal as its value, the value would be represented as a structure consisting of the month, day, and year as separate pieces of information, using separate plain literals to represent the corresponding values. However, so far, all constant values that serve as objects in RDF statements have been represented by these plain (untyped) literals, even when the intent is probably for the value of the property to be a number (e.g., the value of a year or age property) or some other kind of more specialized value.

For example, Figure 4 illustrated an RDF graph recording information about John Smith. That graph recorded the value of John Smith's exterms:age property as the plain literal "27", as shown in Figure 7:

Figure 7: Representing John Smith's Age

In this case, the hypothetical organization example.org probably intends for "27" to be interpreted as a number, rather than as the string consisting of the character "2" followed by the character "7" (since the literal represents the value of an "age" property). However, there is no information in Figure 7's graph that explicitly indicates that "27" should be interpreted as a number. Similarly, example.org also probably intends for "27" to be interpreted as a decimal number, i.e., the value twenty seven, rather than, say, as an octal number, i.e., the value twenty three. However, once again there is no information in Figure 7's graph that explicitly indicates this. Specific applications might be written with the understanding that they should interpret values of the exterms:age property as decimal numbers, but this would mean that proper interpretation of this RDF would depend on information not explicitly provided in the RDF graph, and hence on information that would not necessarily be available to other applications that might need to interpret this RDF.

The common practice in programming languages or database systems is to provide this additional information about how to interpret a literal by associating a datatype with the literal, in this case, a datatype like decimal or integer. An application that understands the datatype then knows, for example, whether the literal "10" is intended to represent the number ten, the number two, or the string consisting of the character "1" followed by the character "0", depending on whether the specified datatype is integer, binary, or string. (More specialized datatypes could also be used to include the units information mentioned at the end of Section 2.3, e.g., a datatype integerYears, although the Primer will not elaborate on this idea.) In RDF, typed literals are used to provide this kind of information.

An RDF typed literal is formed by pairing a string with a URIref that identifies a particular datatype. This results in a single literal node in the RDF graph with the pair as the literal. The value represented by the typed literal is the value that the specified datatype associates with the specified string. For example, using a typed literal, John Smith's age could be described as being the integer number 27 using the triple:

<http://www.example.org/staffid/85740>  <http://www.example.org/terms/age> "27"^^<http://www.w3.org/2001/XMLSchema#integer> .

or, using the QName simplification for writing long URIs:

exstaff:85740  exterms:age  "27"^^xsd:integer .

or as shown in Figure 8:

Figure 8: A Typed Literal for John Smith's Age

Similarly, in the graph shown in Figure 3 describing information about a Web page, the value of the page's exterms:creation-date property was written as the plain literal "August 16, 1999". However, using a typed literal, the creation date of the Web page could be explicitly described as being the date August 16, 1999, using the triple:

ex:index.html  exterms:creation-date  "1999-08-16"^^xsd:date .

or as shown in Figure 9:

Figure 9: A Typed Literal for a Web Page's Creation Date

Unlike typical programming languages and database systems, RDF has no built-in set of datatypes of its own, such as datatypes for integers, reals, strings, or dates. Instead, RDF typed literals simply provide a way to explicitly indicate, for a given literal, what datatype should be used to interpret it. The datatypes used in typed literals are defined externally to RDF, and identified by their datatype URIs. (There is one exception: RDF defines a built-in datatype with the URIref rdf:XMLLiteral to represent XML content as a literal value. This datatype is defined in [RDF-CONCEPTS], and its use is described in Section 4.5.) For instance, the examples in Figure 8 and Figure 9 use the datatypes integer and date from the XML Schema datatypes defined in XML Schema Part 2: Datatypes [XML-SCHEMA2]. An advantage of this approach is that it gives RDF the flexibility to directly represent information coming from different sources without the need to perform type conversions between these sources and a native set of RDF datatypes. (Type conversions would still be required when moving information between systems having different sets of datatypes, but RDF would impose no extra conversions into and out of a native set of RDF datatypes.)

Since the value that a given typed literal denotes is defined by the typed literal's datatype, and, with the exception of rdf:XMLLiteral, RDF does not define any datatypes, the actual interpretation of a typed literal appearing in an RDF graph (e.g., determining the value it denotes) must be performed by software that is written to correctly process not only RDF, but the typed literal's datatype as well. Effectively, this software must be written to process an extended language that includes not only RDF, but also the datatype, as part of its built-in vocabulary. This raises the issue of which datatypes will be generally available in RDF software. Generally, the XML Schema datatypes that are listed as suitable for use in RDF in [RDF-SEMANTICS] have a "first among equals" status in RDF. As noted already, the examples in Figure 8 and Figure 9 used some of these XML Schema datatypes, and the Primer will be using these datatypes in most of its other examples of typed literals as well (for one thing, XML Schema datatypes already have assigned URIrefs that can be used to refer to them, specified in [XML-SCHEMA2]). These XML Schema datatypes are treated no differently than any other datatype, but they are expected to be the most widely used, and therefore the most likely to be interoperable among different software. As a result, it is expected that much RDF software will also be written to process these datatypes. However, RDF software could be written to process other sets of datatypes as well, assuming they were determined to be suitable for use with RDF, as described already.

In general, RDF software may be called on to process RDF data that contains references to datatypes that the software has not been written to process, in which case there are some things the software will not be able to do. For one thing, with the exception of rdf:XMLLiteral, RDF itself does not define the URIrefs that identify datatypes. As a result, RDF software, unless it has been written to recognize specific URIrefs, will not be able to determine whether or not a URIref written in a typed literal actually identifies a datatype. Moreover, even when a URIref does identify a datatype, RDF itself does not define the validity of pairing that datatype with a particular literal. This validity can only be determined by software written to correctly process that particular datatype.

For example, the typed literal in the triple:

exstaff:85740  exterms:age  "pumpkin"^^xsd:integer .

or the graph shown in Figure 10:

Figure 10: An Invalid Typed Literal for John Smith's Age

is valid RDF, but obviously an error as far as the xsd:integer datatype is concerned, since "pumpkin" is not defined as being in the lexical space of xsd:integer. RDF software not written to process the xsd:integer datatype would not be able to recognize this error.

However, proper use of RDF typed literals provides more information about the intended interpretation of literal values, and hence makes RDF statements a better means of information exchange among applications.

2.5 Concepts Summary

Taken as a whole, RDF is basically simple: nodes-and-arcs diagrams interpreted as statements about things identified by URIrefs. This section has presented an introduction to these concepts. As noted earlier, the normative (i.e., definitive) RDF specification describing these concepts is RDF Concepts and Abstract Syntax [RDF-CONCEPTS], which should be consulted for further information. The formal semantics (meaning) of these concepts is defined in the (normative) RDF Semantics [RDF-SEMANTICS] document.

However, in addition to the basic techniques for describing things using RDF statements discussed so far, it should be clear that people or organizations also need a way to describe the vocabularies (terms) they intend to use in those statements, specifically, vocabularies for:

• describing types of things (like exterms:Person)
• describing properties (like exterms:age and exterms:creation-date), and
• describing the types of things that can serve as the subjects or objects of statements involving those properties (such as specifying that the value of an exterms:age property should always be an xsd:integer).

The basis for describing such vocabularies in RDF is the RDF Vocabulary Description Language 1.0: RDF Schema [RDF-VOCABULARY], which will be described in Section 5.

Additional background on the basic ideas underlying RDF, and its role in providing a general language for describing Web information, can be found in [WEBDATA]. RDF draws upon ideas from knowledge representation, artificial intelligence, and data management, including Conceptual Graphs, logic-based knowledge representation, frames, and relational databases. Some possible sources of background information on these subjects include [SOWA], [CG], [KIF], [HAYES], [LUGER], and [GRAY].

4.1 RDF Containers

There is often a need to describe groups of things: for example, to say that a book was created by several authors, or to list the students in a course, or the software modules in a package. RDF provides several predefined (built-in) types and properties that can be used to describe such groups.

First, RDF provides a container vocabulary consisting of three predefined types (together with some associated predefined properties). A container is a resource that contains things. The contained things are called members. The members of a container may be resources (including blank nodes) or literals. RDF defines three types of containers:

• rdf:Bag
• rdf:Seq
• rdf:Alt

A Bag (a resource having type rdf:Bag) represents a group of resources or literals, possibly including duplicate members, where there is no significance in the order of the members. For example, a Bag might be used to describe a group of part numbers in which the order of entry or processing of the part numbers does not matter.

A Sequence or Seq (a resource having type rdf:Seq) represents a group of resources or literals, possibly including duplicate members, where the order of the members is significant. For example, a Sequence might be used to describe a group that must be maintained in alphabetical order.

An Alternative or Alt (a resource having type rdf:Alt) represents a group of resources or literals that are alternatives (typically for a single value of a property). For example, an Alt might be used to describe alternative language translations for the title of a book, or to describe a list of alternative Internet sites at which a resource might be found. An application using a property whose value is an Alt container should be aware that it can choose any one of the members of the group as appropriate.

To describe a resource as being one of these types of containers, the resource is given an rdf:type property whose value is one of the predefined resources rdf:Bag, rdf:Seq, or rdf:Alt (whichever is appropriate). The container resource (which may either be a blank node or a resource with a URIref) denotes the group as a whole. The members of the container can be described by defining a container membership property for each member with the container resource as its subject and the member as its object. These container membership properties have names of the form rdf:_n , where n is a decimal integer greater than zero, with no leading zeros, e.g., rdf:_1, rdf:_2, rdf:_3, and so on, and are used specifically for describing the members of containers. Container resources may also have other properties that describe the container, in addition to the container membership properties and the rdf:type property.

It is important to understand that while these types of containers are described using predefined RDF types and properties, any special meanings associated with these containers, e.g., that the members of an Alt container are alternative values, are only intended meanings. These specific container types, and their definitions, are provided with the aim of establishing a shared convention among those who need to describe groups of things. All RDF does is provide the types and properties that can be used to construct the RDF graphs to describe each type of container. RDF has no more built-in understanding of what a resource of type rdf:Bag is than it has of what a resource of type ex:Tent (discussed in Section 3.2) is. In each case, applications must be written to behave according to the particular meaning involved for each type. This point will be expanded on in the following examples.

A typical use of a container is to indicate that the value of a property is a group of things. For example, to represent the sentence "Course 6.001 has the students Amy, Mohamed, Johann, Maria, and Phuong", the course could be described by giving it a s:students property (from an appropriate vocabulary) whose value is a container of type rdf:Bag (representing the group of students). Then, using the container membership properties, individual students could be identified as being members of that group, as in the RDF graph shown in Figure 14:

Figure 14: A Simple Bag Container Description

Since the value of the s:students property in this example is described as a Bag, there is no intended significance in the order given for the URIrefs of the students, even though the membership properties in the graph have integers in their names. It is up to applications creating and processing graphs that include rdf:Bag containers to ignore any (apparent) order in the names of the membership properties.

Example 12 describes the graph shown in Figure 14:

@prefix rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#>.
@prefix s:   <http://example.org/students/vocab#>.

<http://example.org/courses/6.001>
    s:students [
        a rdf:Bag;
        rdf:_1 <http://example.org/students/Amy>;
        rdf:_2 <http://example.org/students/Mohamed>;
        rdf:_3 <http://example.org/students/Johann>;
        rdf:_4 <http://example.org/students/Maria>;
        rdf:_5 <http://example.org/students/Phuong>.
    ].

The graph structure for an rdf:Seq container, and the corresponding Turtle, are similar to those for an rdf:Bag (the only difference is in the type, rdf:Seq). Once again, although an rdf:Seq container is intended to describe a sequence, it is up to applications creating and processing the graph to appropriately interpret the sequence of integer-valued property names.

Note that this example also shows a Turtle abbreviation for the property http://www.w3.org/1999/02/22-rdf-syntax-ns#type in the form of the special predicate name "a". Because using rdf typing is very frequent, this abbreviation is very useful in practice.

To illustrate an Alt container, the sentence "The source code for X11 may be found at ftp.example.org, ftp1.example.org, or ftp2.example.org" could be expressed in the RDF graph shown in Figure 15:

Figure 15: A Simple Alt Container Description

Example 13 shows how the graph in Figure 15 could be written in Turtle:

@prefix rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#>.
@prefix s:   <http://example.org/packages/vocab#>.

<http://example.org/packages/X11>
    s:DistributionSite [
        a rdf:Alt;
        rdf:_1 <ftp://ftp.example.org>;
        rdf:_2 <ftp://ftp1.example.org>;
        rdf:_3 <ftp://ftp2.example.org>.
    ].

An Alt container is intended to have at least one member, identified by the property rdf:_1. This member is intended to be considered as the default or preferred value. Other than the member identified as rdf:_1, the order of the remaining elements is not significant.

The RDF in Figure 15 as written states simply that the value of the s:DistributionSite site property is the Alt container resource itself. Any additional meaning that is to be read into this graph, e.g., that one of the members of the Alt container is to be considered as the value of the s:DistributionSite site property, or that ftp://ftp.example.org is the default or preferred value, must be built into an application's understanding of the intended meaning of an Alt container, and/or into the meaning defined for the particular property (s:DistributionSite in this case), which also must be understood by the application.

Alt containers are frequently used in conjunction with language tagging. (Turtle permits the use of a language tag to indicate that the element content is in a specified language. The use of this tag is described in [TURTLE], and illustrated later in Section 6.2.) For example, a work whose title has been translated into several languages might have its title property pointing to an Alt container holding literals representing the titles expressed in each of the language variants.

The distinction between the intended meanings of a Bag and an Alt can be further illustrated by considering the authorship of the book "Huckleberry Finn". The book has exactly one author, but the author has two names (Mark Twain and Samuel Clemens). Either name is sufficient to specify the author. Thus using an Alt container for the author's names more accurately represents the relationship than using a Bag (which might suggest there are two different authors).

Users are free to choose their own ways to describe groups of resources, rather than using the RDF container vocabulary. These RDF containers are merely provided as common definitions that, if generally used, could help make data involving groups of resources more interoperable.

Sometimes there are clear alternatives to using these RDF container types. For example, a relationship between a particular resource and a group of other resources could be indicated by making the first resource the subject of multiple statements using the same property. This is structurally different from the resource being the subject of a single statement whose object is a container containing multiple members. In some cases, these two structures may have equivalent meaning, but in other cases they may not. The choice of which to use in a given situation should be made with this in mind.

Consider as an example the relationship between a writer and her publications, as in the sentence:

Sue has written "Anthology of Time", "Zoological Reasoning", and "Gravitational Reflections".

In this case, there are three resources each of which was written independently by the same writer. This could be expressed using repeated properties as:

exstaff:Sue   exterms:publication   ex:AnthologyOfTime .
exstaff:Sue   exterms:publication   ex:ZoologicalReasoning .
exstaff:Sue   exterms:publication   ex:GravitationalReflections .

In this example there is no stated relationship between the publications other than that they were written by the same person. Each of the statements is an independent fact, and so using repeated properties would be a reasonable choice. However, this could just as reasonably be represented as a statement about the group of resources written by Sue:

exstaff:Sue   exterms:publication _:z
_:z           rdf:type            rdf:Bag .
_:z           rdf:_1              ex:AnthologyOfTime .
_:z           rdf:_2              ex:ZoologicalReasoning .
_:z           rdf:_3              ex:GravitationalReflections .

On the other hand, the sentence:

The resolution was approved by the Rules Committee, having members Fred, Wilma, and Dino.

says that the committee as a whole approved the resolution; it does not necessarily state that each committee member individually voted in favor of the resolution. In this case, it would be potentially misleading to model this sentence as three separate exterms:approvedBy statements, one for each committee member, as shown below:

ex:resolution   exterms:approvedBy   ex:Fred .
ex:resolution   exterms:approvedBy   ex:Wilma .
ex:resolution   exterms:approvedBy   ex:Dino .

since these statements say that each member individually approved the resolution.

In this case, it would be better to model the sentence as a single exterms:approvedBy statement whose subject is the resolution and whose object is the committee itself. The committee resource could then be described as a Bag whose members are the members of the committee, as in the following triples:

ex:resolution      exterms:approvedBy   ex:rulesCommittee .
ex:rulesCommittee  rdf:type             rdf:Bag .
ex:rulesCommittee  rdf:_1               ex:Fred .
ex:rulesCommittee  rdf:_2               ex:Wilma .
ex:rulesCommittee  rdf:_3               ex:Dino .

When using RDF containers, it is important to understand that the statements are not constructing containers, as in a programming language data structure. Instead, the statements are describing containers (groups of things) that presumably exist. For instance, in the Rules Committee example just given, the Rules Committee is an unordered group of people, whether it is described in RDF that way or not. Saying that the resource ex:rulesCommittee has type rdf:Bag is not saying that the Rules Committee is a data structure, or constructing a particular data structure to hold the members of the group (the Rules Committee could be described as a Bag without describing any members at all). Instead, it is describing the Rules Committee as having characteristics corresponding to those associated with a Bag container, namely that it has members, and their order of description is not significant. Similarly, using the container membership properties simply describes a container resource as having certain things as members. This does not necessarily say that the things described as members are the only members that exist. For example, the triples given above to describe the Rules Committee say only that Fred, Wilma, and Dino are members of the committee, not that they are the only members of the committee.

@prefix rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#>.
@prefix s:   <http://example.org/packages/vocab#>.

<http://example.org/packages/X11>
    s:DistributionSite [
        a rdf:Alt;
        a rdf:Bag;
        rdf:_2 <ftp://ftp.example.org>;
        rdf:_2 <ftp://ftp1.example.org>;
        rdf:_5 <ftp://ftp2.example.org>
    ].

4.2 RDF Collections

A limitation of the containers described in Section 4.1 is that there is no way to close them, i.e., to say "these are all the members of the container". As noted in Section 4.1, a container only says that certain identified resources are members; it does not say that other members do not exist. Also, while one graph may describe some of the members, there is no way to exclude the possibility that there is another graph somewhere that describes additional members. RDF provides support for describing groups containing only the specified members, in the form of RDF collections. An RDF collection is a group of things represented as a list structure in the RDF graph. This list structure is constructed using a predefined collection vocabulary consisting of the predefined type rdf:List, the predefined properties rdf:first and rdf:rest, and the predefined resource rdf:nil.

To illustrate this, the sentence "The students in course 6.001 are Amy, Mohamed, and Johann" could be represented using the graph shown in Figure 16:

Figure 16: An RDF Collection (list structure)

In this graph, each member of the collection, such as s:Amy, is the object of an rdf:first property whose subject is a resource (a blank node in this example) that represents a list. This list resource is linked to the rest of the list by an rdf:rest property. The end of the list is indicated by the rdf:rest property having as its object the resource rdf:nil (the resource rdf:nil represents the empty list, and is defined as being of type rdf:List). This structure will be familiar to those who know the Lisp programming language. As in Lisp, the rdf:first and rdf:rest properties allow applications to traverse the structure. Each of the blank nodes forming this list structure is implicitly of type rdf:List (that is, each of these nodes implicitly has an rdf:type property whose value is the predefined type rdf:List), although this is not explicitly shown in the graph. The RDF Schema language [RDF-VOCABULARY] defines the properties rdf:first and rdf:rest as having subjects of type rdf:List, so the information about these nodes being lists can generally be inferred, rather than the corresponding rdf:type triples being written out all the time.

Turtle provides a special notation to make it easy to describe collections using graphs of this form. To illustrate how this notation works, the Turtle from Example 15 would result in the RDF graph shown in Figure 16:

@prefix rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#>.
@prefix s:   <http://example.org/students/vocab#>.

<http://example.org/courses/6.001>
    s:students (
        <http://example.org/students/Amy>
        <http://example.org/students/Mohamed>
        <http://example.org/students/Johann>
    ).

@prefix rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#>.
@prefix s:   <http://example.org/students/vocab#>.

<http://example.org/courses/6.001> s:student _:sch1.

_:sch1 
    rdf:first <http://example.org/students/Amy>;
    rdf:rest  _:sch2.
        
_:sch2 
    rdf:first <http://example.org/students/Mohamed>;
    rdf:rest  _:sch3.

_:sch3 
    rdf:first <http://example.org/students/Johann>;
    rdf:rest  rdf:nil.

4.3 RDF Reification