The Self-Describing Web

TAG Finding 07 February 2009

This version:
Latest version:
Previous versions:
selfDescribingDocuments-2009-01-15.html, selfDescribingDocuments-2009-01-14.html, selfDescribingDocuments-2009-01-07.html, selfDescribingDocuments-2008-12-03.html, selfDescribingDocuments-2008-09-08.html, selfDescribingDocuments-2008-05-12.html, http://www.w3.org/2001/tag/doc/selfDescribingDocuments-2008-02-08, http://www.w3.org/2001/tag/doc/selfDescribingDocuments-2007-05-24, http://www.w3.org/2001/tag/doc/selfDescribingDocuments-2007-02-25
Noah Mendelsohn, IBM Corp. <Noah_Mendelsohn@us.ibm.com>

This document is also available in these non-normative formats: XML.


The Web is designed to support flexible exploration of information by human users and by automated agents. For such exploration to be productive, information published by many different sources and for a variety of purposes must be comprehensible to a wide range of Web client software, and to users of that software.

HTTP and other Web technologies can be used to deploy resource representations that are self-describing: information about the encodings used for each representation is provided explicitly within the representation. Starting with a URI, there is a standard algorithm that a user agent can apply to retrieve and interpret such representations. Furthermore, representations can be what we refer to as grounded in the Web, by ensuring that specifications required to interpret them are determined unambiguously based on the URI, and that explicit references connect the pertinent specifications to each other. Web-grounding ensures that the specifications needed to interpret information on the Web can be identified unambiguously. When such self-describing, Web-grounded resources are linked together, the Web as a whole can support reliable, ad hoc discovery of information.

This finding describes how document formats, markup conventions, attribute values, and other data formats can be designed to facilitate the deployment of self-describing, Web-grounded Web content.

Status of this Document

This document has been produced by the W3C Technical Architecture Group (TAG). The TAG approved this finding at its 22 January 2009 teleconference. Additional TAG findings, both accepted and in draft state, may also be available. The TAG may incorporate this and other findings into future versions of the [AWWW]. Please send comments on this finding to the publicly archived TAG mailing list www-tag@w3.org (archive).

Table of Contents

1 Introduction
2 The Web's Standard Retrieval Algorithm
3 Use of Widely Deployed Standards and Formats
4 Creating New Formats and Standards
    4.1 Use Existing URI Schemes, Protocols, and Media Types
    4.2 URI-based Extensibility
        4.2.1 Example: The Atom Syndication Format
        4.2.2 Example: Microformats
        4.2.3 Self-describing XML Documents
5 RDF and the Self-Describing Semantic Web
    5.1 Using RDFa To Produce Self-describing HTML
    5.2 Using GRDDL to Bridge From XML To RDF
6 Accountability and Grounding Information in the Web
    6.1 Grounding New Specifications in the Web
7 Conclusions
8 References


A Diagram of the Web's Retrieval Algorithm

1 Introduction

Supporting ad-hoc exploration is a goal of the Web. Users must therefore be able to get useful information from documents prepared by people whom they don't know, and with whom they have not coordinated in advance. The chapters below explain in more detail how the following techniques can be used to create, deploy and access self-describing Web resource representations that can be correctly interpreted using only widely available information (note that the terms resource and representation, as used here, are defined in [AWWW]):

Furthermore, when self-describing representations are linked together, the Web as a whole can support reliable, ad hoc discovery of information.


Self-describing resources promote ad hoc discovery of information.

Good Practice

Web resource representations should be self-describing.

The sections below discuss in more detail the techniques needed to create self-describing content for the Web, how to extend the Web with new formats that are themselves self-describing, how to publish self-describing Semantic Web data, why it's important that interpretation of Web representations be grounded unambiguously in the core specifications of the Web, and how a standard HTTP-based algorithm enables users to retrieve and interpret self-describing resource representations.

2 The Web's Standard Retrieval Algorithm

HTTP is the most widely deployed protocol on the Web, and it is designed to facilitate the deployment of self-describing Web resource representations. Indeed, there is a standard algorithm that a user agent can employ to attempt to obtain and interpret the representation of any Web resource that is accessible using the HTTP protocol. Consider the following example, which is representative of many simple Web interactions:

Bob is reading a Web page which includes a link to http://example.com/todaysnews. Bob has had no previous contact with the owner of the referenced resource, and his browser has not been specially configured for access to it. The steps taken by Bob's browser when he clicks the link illustrate a typical path through the standard retrieval algorithm of the Web (readers unfamiliar with the HTTP protocol may find it useful to consult either [HTTP], or one of the many HTTP introductions available on the Web). Bob's browser...

Neither Bob nor his browser has any advance knowledge of the nature of the resource or the fact that its representation is provided in HTML, yet the browser successfully retrieves the representation, determines its format, and renders it for him. The link could have been to an image/jpeg picture, an application/atom+xml feed, or to a document containing application/rdf+xml data. Bob's browser could in each case determine the format. Indeed, as Bob continues to browse the Web, his browser is able to determine the format of each representation that is retrieved, and can determine how to present it to him. This example shows how HTTP enables the deployment of self-describing Web resources.

Consider instead a different example, in which Bob clicks on a link to ftp://example.com/todaysnews. Although Bob's browser can easily open an FTP connection to retrieve a file, there is no way for the browser to reliably determine the nature of the information received. Even if the URI were ftp://example.com/todaysnews.html the browser would be guessing if it assumed that the file's contents were HTML, since no normative specification ensures that data from ftp URIs ending in .html is in any particular format.

The Web's retrieval algorithm works best when used with the core suite of protocols and formats that are most widely deployed, and that are capable of supporting retrieval of self-describing representations. These core technologies include: DNS, HTTP 1.1, HTML 4, XML, as well as widely deployed image formats such as image/jpeg and image/gif. As discussed in 5 RDF and the Self-Describing Semantic Web, RDF, OWL and GRDDL are among the additional core technologies that enable self-describing Semantic Web content. A flow diagram illustrating more details of the Web's standard retrieval algorithm is provided in A Diagram of the Web's Retrieval Algorithm.

3 Use of Widely Deployed Standards and Formats

Successful communication depends on the supplier and the consumer(s) of a document having a shared understanding of the information conveyed, and that in turn requires at least some shared assumptions about the form in which the information is represented. The simplest way to achieve this is if the media type, the document encoding, and any other conventions used for the representation are standards and are widely deployed.

Consider Susan, who buys a new digital camera. The software supplied with her camera uploads photos to the Web using the widely-deployed image/jpeg media type, and her Web server correctly labels served representations with that Content-Type. Millions of user agents deployed around the world are preconfigured to display Susan's photographs and to extract metadata such as camera settings from them. Search engines are likely to index them in helpful ways too.

Now consider instead Mary, who buys a different camera with software that does not use widely deployed Web formats. Indeed, the camera's manufacturer has invented a new "raw" file format that takes advantage of the camera's special features. The provided photo management software not only uses that format locally, it also uploads photos to Mary's Web server in that same form. Indeed, it even uploads a .htaccess file, configuring the server to label served representations with the proprietary Content-Type image/x-fancyrawphotoformat. In this example, there are no outright violations of Web architecture, but the decision to use an uncommon, proprietary, unregistered and apparently experimental media type is unfortunate. No existing Web user agents recognize the image/x-fancyrawphotoformat media type, search engine spiders are unlikely to extract useful information from pictures in that format, and so on. Unlike Susan's, which can be viewed by almost anyone, Mary's photos are at best useful to a few people who have the proprietary software needed to decode them.

Good Practice

Web resource representations should be published using widely deployed standards and formats.

4 Creating New Formats and Standards

The techniques described above apply in the many cases where widely deployed media types such as image/jpeg are sufficient, but the Web is used for a broad and continually growing range of information. No fixed set of formats and standards can fully meet the need to encode all such information for machine processing. Of course, ways can be found to convey almost any information using standard media types. An employment record, for example, can be transmitted as either text/plain or text/html. The resulting document may be quite suitable for browsing, but it might not facilitate automated discovery of the employee's name, his or her date of hire, and so on. To meet such needs, new standards must be created, e.g. for marking up the names and dates. Similarly, the need may arise to use new values for individual fields such as rel attributes on HTML link elements (see [TAGIssue51]).

So, although the Web requires self-describing documents that can be understood using only widely deployed standards, there is also a continual need for new formats and encoding conventions. How can new formats and encodings be deployed in a manner that is self-describing? The following sections explore ways of creating new formats and encoding conventions that maximize interoperability with existing Web infrastructure, and that can be used to create self-describing documents.

4.1 Use Existing URI Schemes, Protocols, and Media Types

Innovations can be introduced to the Web at many different architectural layers. For example:

  • New URI schemes can be introduced

  • New transfer protocols can be deployed

  • New media types can be introduced

  • New namespace-qualified markup can be defined for XML

  • New RDF properties and ontologies can be defined for the Semantic Web

Often, a given capability could in principle be deployed at any of several different layers. For example, new sorts of content, such as movies, could be made available using new URI schemes and/or with new protocols, but doing so would require updating hundreds of millions of user agents, servers, proxies, and so on to understand these changes to the core mechanisms of the Web. Usually it is preferable to leverage the existing core mechanisms of the Web, such as http-scheme URIs and the HTTP protocol, as these are widely deployed. Indeed, one should usually leverage as many existing layers of the Web's architecture as is practical when introducing new function.

Good Practice

When extending the Web with new formats and functions, use existing URI schemes, protocols, and media types wherever practical.

One way to do this is to use URI-based extensibility within existing media types, as described in the sections below.

4.2 URI-based Extensibility

Many documents, particularly those that convey machine-readable data or messages, encode information using specifications that are specialized to particular purposes. Such specifications may cover details of particular data formats such as lists of customers or inventory records, results of scientific experiments, listings for television shows, details of university course offerings, information about molecular structures or drug tests, etc. Because of the great variety and number of such formats and their specifications, it's not practical to assume that even most of them will be directly implemented by typical Web user agents. Instead, the Web provides means by which the necessary specifications can be discovered, and to a significant degree implemented, dynamically and automatically. This is done by:

  • ensuring that every specification, and in many cases each markup tag name or data value used, is identified with a URI

  • ensuring that such URIs are used in the instance either directly as data values or tag names, or else to identify the encodings used

  • including in Web representations URIs that identify the specifications needed to interpret those representations

In other words, it should be possible to discover from each Web representation the conventions used to encode it, and particularly in cases where those conventions are not widely deployed, to find within the representations links to specifications, ontologies and/or programs necessary for interpreting the representation. So, just as the Web may be used to dynamically discover a great wealth of resources, it can also be used to dynamically discover the specifications, ontologies, or programs needed to interpret the representations of those resources.

Good Practice

Web representations should link to the information needed to support automatic processing of those representations.

4.2.1 Example: The Atom Syndication Format

The Atom Syndication Format [ATOM] is an XML-based format for syndicating information about blogs and other Web resources. Atom syndications are served with Content-Type application/atom+xml, and thus can be recognized by user agents. ATOM entries can include <atom:link> elements such as the following:

  <title>An interesting picture</title>
  <link rel="enclosure" type="image/jpeg" length="12345"
    <content type="xhtml" xml:lang="en"
      <div xmlns="http://www.w3.org/1999/xhtml">
        <p><[Update: Here's an interesting picture.]</p>

The link elements identify external resources, in this case an image/jpeg photograph. Furthermore, each link can carry a rel attribute which specifies the relationship between the resource and the ATOM entry that links to it. In the example above, the relationship is specified as enclosure which, according to the ATOM specification, indicates that the photograph may have been too large for inline processing with the rest of the feed.

What's of interest for this finding is the fact that values of the rel attribute are URIs (actually [IRI]s, which are the internationalized form of URIs), or else the values can be mapped to URIs. This means that anyone, anywhere can invent a new sort of link relationship, can assign a URI to identify that relationship, and can use that value in the rel attribute. For example:

  <title>An interesting picture</title>
  <link rel="http://example.org/SomeNewATOMRelationship"
        type="image/jpeg" length="12345"
    <content type="xhtml" xml:lang="en"
      <div xmlns="http://www.w3.org/1999/xhtml">
        <p><[Update: Here's an interesting picture.]</p>

Furthermore, anyone doing this can (and indeed should) provide information about that new relationship via HTTP from the assigned URI. For convenience, the ATOM specification also provides that short form names such as enclosure in the first example can be registered with IANA, and ATOM provides a deterministic mapping to a URI for each of these. These URIs are formed by prepending the fixed base URI http://www.iana.org/assignments/relation/ to the short form. Thus, the first example above is in fact using the relationship http://www.iana.org/assignments/relation/enclosure.

This example shows how use of URIs for data values enables distributed assignment of new values. More importantly for this finding, the use of URIs for such values provides the opportunity for information about those values to be discovered dynamically on the Web.

4.2.2 Example: Microformats

[Microformats] provide a simple means of marking up data in HTML Web pages. The presence of a microformat is typically indicated by the appearance of an identifying value such as vcard in an HTML class attribute, and particular data items are usually marked with other class values. Indeed, somewhat confusingly, the [hCard] specification requires that hCards be identified with the root class name vcard, as in the example below. This hCard provides contact information for the North American office of the W3C:

    <div class="vcard">
      <a class="fn org url" href="http://www.w3.org/">World Wide Web Consortium</a>
      <div class="adr">
        <span class="type">Work</span>:
        <div class="street-address">32 Vassar Street</div>
        <span class="locality">Cambridge</span>,  
        <abbr class="region" title="Massachusetts">MA</abbr>  
        <span class="postal-code">02139</span>
        <div class="country-name">USA</div>
      <div class="tel">
       <span class="type">Work</span> +1-617-253-2613
      <div class="tel">
        <span class="type">Fax</span> +1-617-258-5999

In general, microformats such as hCard are not self-describing, because there is no requirement in the HTML media type specifications that class attribute values such as vcard or type be interpreted per the hCard specification. Indeed, lacking any specific indication that the resource owner has intended this interpretation, it is dangerous for clients to assume hCard semantics — there is a real risk that some HTML Web pages use values like type, value or even in principle vcard for other purposes.

Unlike some other microformats, hCard does provide an option for deploying in a way that is self-describing. The hCard profile specifies a value for the profile attribute of the HTML 4.01 [HTML401] <HEAD> element:

    <head profile='http://www.w3.org/2006/03/hcard'>

and presence of this profile value indicates that class attributes can be reliably interpreted per the hCard specification. (Note, however, that there is ongoing discussion as to whether the profile attribute will be included as part of HTML 5, and if not, whether some other mechanism will be provided for signaling the use of extensions such as microformats.)

So, microformats are self-describing only when profiles or other means licensed by a pertinent media type specification are used to enable them. Unfortunately, few microformats have such profiles, and even when profiles are available, evidence suggests that they are not universally applied. User agents that infer the presence of microformats without reliable indicators such as <HEAD> element profiles are at risk of extracting incorrect data from Web pages.

4.2.3 Self-describing XML Documents

XML Namespaces [XMLNamespaces] facilitate the creation of self-describing XML documents. When namespace-qualified attribute and element names are used in XML, recursive processing from the root element may be applied to determine not just the overall nature of the document, but also the meaning in context of all sub-elements. Doing this, however, requires understanding of the semantics of each named element. Although a few specific XML variants such as application/xhtml+xml may be directly supported by some user agents, no user agent can build in support for the ever growing set of XML languages used for Web representations. This section describes how namespace documents, discoverable from the XML tag names in the markup, can be used to make such languages self-describing, and to enable automated processing of them.

When XML namespaces are used, each XML element is named with a qualified name, consisting of a prefix and a local name. In the following example, the root element has the qualified name <inventory:inventoryItem>:


Qualified names map to expanded names such as {http://example.org/inventoryNamespace,inventoryItem}, comprised of a namespace name URI (http://example.org/inventoryNamespace) and a local name (inventoryItem). The namespace name URI serves at least two roles: the most obvious and the most widely understood is to distinguish expanded names in one namespace from those in another; the other role, and the one that is most important for purposes of this finding, is that it provides Web identification for the namespace itself. The namespace is a Web resource, and like any other resource, it can and should provide representations using HTTP. A user agent processing an XML document can retrieve descriptions of the namespaces used in that document, and can use that retrieved information to determine how to correctly process the XML markup. The TAG Finding "Associating Resources with Namespaces" [NamespaceDocuments], recommends the use of [RDDL] as a preferred means of documenting namespaces. RDDL is itself extensible, but it is commonly used to suggest XML Schemas (in any of several languages), XSLT Stylesheets, etc. that are usable with markup from the namespace being described.

Example: assume that user Bob is browsing the Web, and that he follows a link to a resource that returns the XML above as its representation. Specifically, Bob's browser uses 2 The Web's Standard Retrieval Algorithm to retrieve the representation, to determine its character encoding, and to discover that its Content-type is application/inventory+xml. Of course, it's very unlikely that Bob's browser has built in knowledge of the inventory XML language, but the Content-type makes clear [XMLMediaType] that the representation can be interpreted as XML with Namespaces. The root element tag is from namespace http://example.org/inventoryNamespace, which uses the http scheme, so Bob's browser does an HTTP GET from that URI. What comes back is a RDDL document containing the following <rddl:resource> element:

   xlink:title="Transform Inventory XML to HTML for Browsing">

This designates a stylesheet (http://example.org/InventoryToBrowsableHTML.xslt) that can be applied to format the inventory XML as HTML — the browser automatically retrieves and applies the stylesheet, producing HTML that is rendered on the screen. Without any manual intervention from Bob, his browser automatically displays the inventory record in a format that is convenient to read and print. Bob's browser may also be enabled for XML validation, in which case it can look in the RDDL for a link to a schema to validate the inventory markup.

Bob's browser has, by retrieving and processing the RDDL document, extended itself for processing of the inventory markup language. Unless the RDDL provides a link to one or more executable programs that process inventory records, it's unlikely that Bob's browser can automatically discover everything that one might reasonably want to know about processing inventory markup. Still, even the limited automatic function described above is very useful, and RDDL is an extensible framework that can be easily adapted to provide new kinds of information about namespaces. Note that because RDDL documents are themselves XML, GRDDL can be applied to derive RDF statements from them (see 5.2 Using GRDDL to Bridge From XML To RDF). In this way, self-describing XML documents can be integrated with the self-describing Semantic web. The "Associating Resources with Namespaces" Finding describes this technique in more detail.

Careful readers of RFC 3023, which governs the XML media types, will note that it only allows namespace URIs to be interpreted as specifying the semantics of qualified names, without requiring this. This means that processors that do not behave as the one described in the example above are not in violation of the RFC. But processors that do behave in that way are not only allowed by the RFC, but support the integration of XML into the self-describing web.

5 RDF and the Self-Describing Semantic Web

[RDF] provides an interoperable means of publishing and linking self-describing Web data resources, and for integrating representations rendered using other technologies such as XML. The result is a single, global self-describing Semantic Web that integrates not only resources that are themselves built or represented using RDF, but also the other Web resources to which that RDF links, as well as those that can be mapped to RDF using technologies such as [GRDDL] . Readers unfamiliar with RDF should consult the RDF primer [RDFPrimer] or the N3 Primer [N3Primer] as a prerequisite to understanding the discussion below.

Each RDF statement is a triple consisting of a subject, a predicate (typically the identifier for a property, or for a relationship between two Web resources), and an object (the value of the property or the referent of the relationship). The subject, the predicate, and often the object as well, are themselves identified by URIs, enabling the dynamic discovery introduced in 4.2 URI-based Extensibility above — if a user agent has no built in knowledge of some particular RDF subject, relationship, or object, it can often use the URI to retrieve the information necessary for processing. Indeed, RDF's Schema [RDFSchema] and OWL Ontology technologies [OWL] together offer a standard, machine-processable means of describing relationships between RDF statements, e.g. that one property is an rdfs:subPropertyOf of another.

As described in 2 The Web's Standard Retrieval Algorithm, the principal purpose of the Web's core retrieval algorithm is to obtain self-describing representations of Web resources. For the self-describing Semantic Web, the algorithm is extended to achieve a more particular goal: to directly obtain RDF triples that represent or indirectly obtain RDF triples that describe the referenced resource.

Consider Amy, who uses an RDF-enabled user agent to retrieve an RDF/XML document containing the following element:

<rdf:RDF xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#"
  <employeeData:employee rdf:about="http://example.org/Employees#BobSmith">
    <employeeData:name>Bob Smith</employeeData:name>
    <employeeData:email rdf:resource="mailto:BobSmith@example.org"/>

The user agent is general purpose, and although it has rules for certain commonly used ontologies, it has no built in code to handle the employeeData properties in the above example. To dynamically acquire the necessary function, the agent does an HTTP GET for http://example.org/EmployeeInformation. The GET returns an OWL ontology, from which the agent discovers that http://example.org/EmployeeInformation#email is rdfs:subPropertyOf the http://www.w3.org/2001/vcard-rdf/3.0#email property, one that the agent recognizes as designating a person's e-mail address. The agent offers Amy the option to send e-mail to Bob Smith. Amy's browser has, like Bob's in the example above, automatically extended itself for processing the employee data. Note that in this case, the user agent has built in capabilities corresponding to the URI, HTTP, XML, RDF, and RDFS specifications, and awareness of the vCard ontology. Additional capability was enabled dynamically based on the ontology information acquired at runtime by following links.

Good Practice

Representations provided directly in RDF, or those for which automated means can be used to discover corresponding RDF, contribute to the self-describing Semantic Web.

Because its model is uniform, because all of its self-description is provided in the same model as the data itself, and because all RDF information is linked into the Web as a whole, RDF provides uniquely powerful facilities for dynamic integration of a self-describing Web. Therefore, it's particularly important that information not originally supplied in an RDF-specific format be convertible into RDF. The sections below discuss two means of doing this: the first shows how RDFa can integrate HTML documents into the Semantic Web, and the second illustrates the use of GRDDL to extract RDF from XML documents.

5.1 Using RDFa To Produce Self-describing HTML

[RDFa] is a W3C Recommendation for embedding Semantic Web statements into XHTML Web pages (see also [RDFaSyntaxandProcessing]). The following example illustrates how RDFa can integrate HTML into the self-describing Semantic Web:

Mary is exploring the Web using a browser that has been enhanced with capabilities for interpreting RDFa. Her browser knows to look through each XHTML Web page that she browses, picking out information from the RDFa, and helping her to use it. For example, the page might contain the following HTML, which represents a FOAF-based contact listing. (This example is adapted from one in [RDFa]):

<?xml version="1.0" encoding="UTF-8"?>
<html xmlns="http://www.w3.org/1999/xhtml" 
    version="XHTML+RDFa 1.0"
  <title>FOAF/RDFa Demo</title>
    <div typeof="foaf:Person" xmlns:foaf="http://xmlns.com/foaf/0.1/"
    <p property="foaf:name">
    Alice Birpemswick
    Email: <a rel="foaf:mbox" href="mailto:alice@example.com">alice@example.com</a>
    Phone: <a rel="foaf:phone" href="tel:+1-617-555-7332">+1 617.555.7332</a>

Even though this document is of media type application/xhtml+xml [XHTMLMediaType], which is not a member of the RDF family of media types, an RDFa-enabled user agent can extract RDF from this document. This document conveys as RDF a set of semantic Web statements about the Web resource http://example.org/staff/Alice. The predicates are all named with the same base URI http://xmlns.com/foaf/0.1/, for which the shorthand prefix foaf is established in the HTML. Using this syntax, the RDFa carries triples for relationships such as the full name of the contact (http://xmlns.com/foaf/0.1/#name), which is Alice Birpemswick, the e-mail address (http://xmlns.com/foaf/0.1/#mbox) which is mailto:alice@example.com, and so on.

An RDFa-enabled user agent can extract these triples and use them to help Mary work with the data they contain, or to integrate with other Semantic Web information. Indeed RDF is designed for such use because, as discussed above in 5 RDF and the Self-Describing Semantic Web, Semantic Web triples are inherently self-describing. If a user agent needs more information about the processing of the email triple it can, like Amy's user agent, do an HTTP GET for URI http://xmlns.com/foaf/0.1/, and use the results to get more information. That information may allow the agent to automatically discover that, in the example, mailto:alice@example.org can indeed be used to send mail to the person named Alice Birpemswick. The browser can then offer Mary the option to send e-mail to Alice, or to add Alice to her address book.

Good Practice

To integrate HTML information into the self-describing Semantic Web, use RDFa.

RDFa documents such as the one shown above are grounded in the Web of data when served using the Content-type application/xhtml+xml. The specification for that media type [XHTMLMediaType] designates it as a member of the family of application/____+xml media types, and the specification for those types [XMLMediaType] in turn provides for markup to be interpreted based on the namespaces used in the document. Finally, the namespace document for XHTML [XHTMLNamespace] allows use of XHTML with RDFa. So, taken together, these specifications provide normatively for use of RDFa in documents served with Content-type application/xhtml+xml. Note that in this case, the user agent has built in capabilities corresponding to the URI, HTTP, XML, XHTML and RDFa and FOAF specifications. The server used these technologies to publish, and there was successful communication.

5.2 Using GRDDL to Bridge From XML To RDF

RDFa provides a standard means of encoding RDF information in XHTML documents, but many other XML variants lack that capability. Furthermore, RDFa requires explicit encoding of each triple in the XHTML instance, and that may in some cases be impractical. [GRDDL] provides a standard means of extracting triples from a broad range of XML document formats. Each GRDDL-enabled XML document links to a transformation that, when applied to the document, produces RDF triples. Typically, the same GRDDL transformation can be used on entire families of similar XML documents.

For example, assume that Albert uses a GRDDL-enabled user agent to retrieve an XML document containing the following fragment:

<employees xmlns="http://example.org/employeeNS">
  <employee name="Bob Smith">

Note that, unlike the earlier examples, this is neither in HTML nor in RDF; we can assume that http://example.org/employeeNS is a namespace created by some particular business for use in its own business documents. Albert's agent has no built in knowledge of this namespace, and so can not do much with it. Now assume that Albert instead retrieves a different document. Most of the markup and data in it is identical to the first, but this document is GRDDL enabled:

<employees xmlns="http://example.org/employeeNS"
  <employee name="Bob Smith">

Albert's user agent is GRDDL aware, so it transforms the <employees> information to RDF using the supplied GRDDL_For_employeeNS.xsl transformation, producing RDF triples from a vCard-based ontology — the agent has knowledge of this ontology, and it uses the triples to determine Bob Smith's email address. As in the example above, Albert's agent offers to send mail to Bob Smith. (The agent might also dynamically discover how to interpret the triples using the techniques described above in 5 RDF and the Self-Describing Semantic Web.)

Good Practice

To integrate information conveyed in XML into the self-describing Semantic Web, use GRDDL.

Note that in this case, the user agent was programmed to understand URI, HTTP, XML, XHTML, GRDDL and the vCard ontology. The information needed for translation from the XML vocabulary into RDF was loaded dynamically.

6 Accountability and Grounding Information in the Web

The Web is an important medium for publishing information, and so it is often important to get agreement about what has in fact been published, and by whom. Consider the following example:

Senator Smith alleges that he has been libeled by an article published on the Web, and so the senator files suit against the purported publishers. He produces in court the log of a response to an HTTP GET for URI http://publisher.example.com/oursenator.html:

HTTP/1.1 200 OK
Date: Tue, 22 Oct 2008 02:43:22 GMT
Server: Apache
Content-Type: text/html; charset=utf-8

<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Strict//EN"
<html xmlns="http://www.w3.org/1999/xhtml" xml:lang="en" lang="en">
<title>The problem with Senator Smith</title>
  <h1>Senator Smith is a Liar and a Thief!</h1>
  <p>Our senator, Mr. Smith, steals money from children, 
     and he lies on his income tax returns!

When confronted by the judge, the owners of the Web site attempt to claim that they have not in fact published information about the senator. "Yes", they say, "our server did accept a TCP/IP connection at port 80 and it responded with some bits, but those bits don't mean what you think they mean. They might look to you like HTML, but that's not what we intended. We haven't said anything derogatory about the senator."

In fact, the specifications for the Web can be useful in establishing that a document was indeed published containing the statement: "Senator Smith is a Liar and a Thief!". The pertinent Web specifications indicate that the bits returned are to be interpreted as the UTF-8 encoding of Unicode characters, that those characters are to be interpreted as HTML, that the text within the HTML is to be read in English, and thus that the entity body is indeed to be read as containing the sentence quoted above.

How, though, can we know whether these specifications apply at all? Perhaps the publishers could claim that some other specifications applied? Unfortunately for them, the Web is clear on that point as well. Starting with a URI such as http://localnewspaper.example.com/oursenator.html and the specification for URIs [URI], all the applicable specifications for interpreting an http-scheme Web representation can be unambiguously discovered by, recursively, following references to other specifications. Using this specific example to illustrate:

Of course, determining whether Senator Smith has been libeled is beyond the scope of Web Architecture. The architecture does, however, provide the means by which one can determine unambiguously the specifications that apply to interpretation of the data retrieved. In general, it is possible to determine which specifications apply to interactions with resources identified by any particular URI; as explained above, these specifications can be located by recursively following references from the specifications for URIs themselves. Whatever the other legal arguments about the case, the interpretation of the document is well defined.

Documents published using HTTP (or other schemes and protocols that are appropriately registered) are thus self-describing not just in the general sense of being interpretable using widely available information, but in the particular sense of having an interpretation that follows from the URI used for access and from the core specifications of the Web. We therefore refer to such documents as being grounded in the Web. From the URI referenced and the specification for URIs [URI], the other specifications for interacting with the resource and for interpreting its representations are determined unambiguously. Furthermore, this characteristic implies an important degree of accountability for those who serve information on the Web. As illustrated by the example above, Web architecture settles many important questions relating to what information has been published, and by whom.

6.1 Grounding New Specifications in the Web

As explained in 4 Creating New Formats and Standards, new formats and protocols are occasionally needed for use with the Web. For information published using those new technologies to be grounded in the Web, it's essential that the pertinent specifications be reachable by references, preferably in the form of Web links, from the specifications for existing Web technologies. Means by which this can be achieved include:

  • The new specification can be registered in a suitable registry. Examples of registries that are available today include the IANA registries for URI schemes [IANASchemeRegistry] or Internet media types [IANAMediaTypeRegistry]. Suitable registries are those that are themselves normatively linked or referenced from applicable specifications for the Web (e.g. the IANA scheme registry is discussed in [BCP35], which in turn is referenced from [URI]). Registries like this can be extremely valuable, but there is also a logistical and social cost to maintaining them, and a cost to developers who must arrange for the registration of new entries. In return for these costs, the community benefits from avoidance of collisions in the allocation of identifiers, and in many cases from careful review of new registrations.

  • Existing specifications can be edited and republished to link explicitly to specifications for the new technology.

  • In certain cases, the new specification may be discoverable automatically. For example, when a new namespace is used in XML, it may be sufficient to publish a namespace document (see [XMLNamespaceDocuments]), as [XMLMediaType] is itself properly linked, and it already provides for extension of XML using Namespaces.

Good Practice

Specifications for interactions with Web resources, and for interpretation of resource representations, should be linked (directly or indirectly) from the specification for URIs [URI].

7 Conclusions

Ad hoc exploration of the Web is possible only if resource representations are self-describing. Using the techniques described above and starting with an http- or https-scheme URI, a user agent can proceed step by step to retrieve a representation, reliably discover the conventions that have been used to encode it, and if necessary, dynamically find instructions for processing it. Furthermore, it is possible starting with a URI and the specification for URIs [URI], to follow successive references to specifications that apply to interactions with the identified resource. These specifications, recursively, define the interpretation of information published in the Web.

Those who invent new document formats, new markup tags, or new conventions for encoding particular data values should use the techniques described above to make those formats self-describing, and to ensure that the pertinent specifications can be discovered by following references from existing ones. When these techniques are used, and when self-describing representations are linked together, the Web as a whole can support reliable, ad hoc discovery of information, and can grow to include ever more powerful languages and systems.

8 References

M. Nottingham, R. Sayre (Eds.) RFC 4287: The Atom Syndication Format. December 2005 (See http://tools.ietf.org/html/rfc4287.)
R. Fielding, I. Jacobs, Authoritative Metadata. W3C Technical Architecture Group Finding, April, 2006. (See http://www.w3.org/2001/tag/doc/mime-respect.)
I.Jacobs, N. Walsh, Architecture of the World Wide Web. W3C. December, 2004. (See http://www.w3.org/TR/webarch/.)
R. Petke, and I. King,Registration Procedures for URL Scheme Names . November, 1999. (See http://tools.ietf.org/html/bcp35.)
P. MockapetrisRFC 1034:DOMAIN NAMES - CONCEPTS AND FACILITIES . November, 1987. (See http://tools.ietf.org/html/rfc1034.)
D. Connolly, Gleaning Resource Descriptions from Dialects of Languages (GRDDL), W3C Candidate Recommendation, May, 2007 (See http://www.w3.org/TR/grddl/.)
T. Çelik, B. Suda, hCard Specification, Microformats Wiki, December, 2008 (See http://microformats.org/wiki/hcard.)
D. Raggett, A. Le Hors, I. Jaobs (Eds.)HTML 4.01 Specification, W3C Recommendation, December, 1999 (See http://www.w3.org/TR/html401/.)
D. Connolly, L. Masinter,The 'text/html' Media Type. November, 1999. (See http://www.rfc-editor.org/rfc/rfc2854.txt.)
J. Gettys, J. Mogul, H. Frystyk, L. Masinter, P. Leach, T. Berners-Lee RFC 2616: Hypertext Transfer Protocol - HTTP/1.1. June 1999 (See http://www.ietf.org/rfc/rfc2616.txt.)
Internet Assigned Numbers AuthorityUniform Resource Identifer (URI) Schemes. (See http://www.iana.org/assignments/media-types/.)
Internet Assigned Numbers AuthorityUniform Resource Identifer (URI) Schemes. (See http://www.iana.org/assignments/uri-schemes.html.)
M. Duerst, M. Suignard RFC 3987: Internationalized Resource Identifiers (IRIs). January 2005 (See http://www.ietf.org/rfc/rfc3987.txt.)
microformats.org, About Microformats (See http://microformats.org/about/.)
Tim Berners-Lee Primer: Getting into RDF & Semantic Web using N3.August, 2005. (See http://www.w3.org/2000/10/swap/Primer.html.)
N. Walsh, Associating Resources with Namespaces. W3C Technical Architecture Group Finding, June, 2008. (See http://www.w3.org/2001/tag/doc/nsDocuments/.)
D. McGuinness, F. van Harmelen (Eds.) OWL Web Ontology Language Overview . W3C Recommendation, February 2004. (See http://www.w3.org/TR/owl-features/.)
J. Borden, T. Bray, Resource Directory Description Language (RDDL). W3C. February, 2002. (See http://www.rddl.org/.)
G. Klyne, J. Carroll (Eds.) Resource Description Framework (RDF): Concepts and Abstract Syntax. W3C Recommendation, February 2004. (See http://www.w3.org/TR/rdf-concepts/.)
B. Adida, M. Birbeck RDFa Primer. W3C. (working draft) October, 2008. (See http://www.w3.org/TR/xhtml-rdfa-primer/.)
B. Adida, M. Birbeck, S. McCarron, S. Pemberton RDFa in XHTML: Syntax and Processing W3C. Proposed Recommendation, October, 2008 (See http://www.w3.org/TR/2008/REC-rdfa-syntax-20081014/.)
F.Manola, E. Miller (Eds.) RDF Primer. W3C Recommendation, February 2004. (See http://www.w3.org/TR/rdf-primer/.)
D. Birckley, R.V. Guha (Eds.) RDF Vocabulary Description Language 1.0: RDF Schema. W3C Recommendation, February 2004. (See http://www.w3.org/TR/rdf-schema/.)
J. Postel,Media Type Registration Procedure . March, 1994. (See http://www.ietf.org/rfc/rfc1590.txt.)
W3C Technical Architecture Group Issue standardizedFieldValues-51: Squatting on link relationship names, x-tokens, registries, and URI-based extensibility (See http://www.w3.org/2001/tag/group/track/issues/51.)
T. Berners-Lee, R. Fielding, L. MasinterRFC 3986: Uniform Resource Identifier (URI): Generic Syntax. January 2005 (See http://www.ietf.org/rfc/rfc3986.txt.)
M. Murata, S. St. Laurent, D. Kohn RFC 3023: XML Media Types. January 2001 (See http://www.ietf.org/rfc/rfc3023.txt.)
M. Baker, P. StarkRFC 3236: The 'application/xhtml+xml' Media Type. January 2002 (See http://www.ietf.org/rfc/rfc3236.txt.)
Pemberton, S.XHTML namespace. October 2008 (See http://www.w3.org/1999/xhtml/.)
H. S. Thompson, N. Walsh Associating Resources with Namespaces. W3C TAG Finding, June, 2008. (See http://www.w3.org/2001/tag/doc/nsDocuments/.)
T. Bray, D. Hollander, A. Layman, R. Tobin, Namespaces in XML 1.1. W3C, August, 2006 (2nd Edition). (See http://www.w3.org/TR/xml-names11/.)

A Diagram of the Web's Retrieval Algorithm

Flow diagram of the Web's retrieval algorithm