Identity, URIs and the Semantic Web

1.   Acknowledgements

• I'm fortunate to work with and learn from some of the proprietors of the Semantic Web
  • Most notably Tim Berners-Lee and Dan Connolly
• They're not in any way responsible for the opinions presented here
• My colleague Michael Sperberg-McQueen first enumerated for me RDF's particular qualities
  • and got me started on thinking about what significance this might have
• Harry Halpin has discussed these topics with me many times, and was the driving force behind the recent IRW workshop.
• My introduction to the problems of RDB->RDF conversion came in conversations with Kate Byrne.

2.   URIs unpacked

• 'URI' stands for

  Uniform

  There's a standard generic syntax (specified by IETF RFC 3986) and (partial) semantics

  Identifier

  URIs are identifiers, that is, names (not addresses)

  Resource

  What it is they identify -- anything at all

3.   More about syntax

• Organised hierarchically
  • Left-to-right dominance
  • scheme ":" hier-part [ "?" query ] [ "#" fragment ]
• Federated (each scheme defines its own syntactic details)
• Absolute or relative

4.   More about resources

• Who gets to say what the resource is that a URI identifies?
  • The owner of the URI
• Who is the owner (sometimes also called the 'minter')?
  • For URI schemes (such as http and ftp) whose hier-part starts with a domain name, the owner of that domain name.
  • For others, should (!) be specified by the scheme specification.

5.   More about identification

• In principle, all URIs are opaque
  • That is, no conclusions about the shape of the space of things identified should be drawn from the shape of the set of their identifiers.
• In practice, e.g. for http URIs, this is false
  • It's entirely up to minters to manage this, and document it (or not :-(
• As for (some) naming elements of ordinary language, URIs can be deiectic:
  • http://localhost/
  • file:///etc/hosts (and indeed, all file: URIs
  • http://www.guardian.co.uk/

6.   An aside about . . . terminology

• RFC 3986 says

  URIs have a global scope and are interpreted consistently regardless of context, though the result of that interpretation may be in relation to the end-user's context.

• This could use some help from the Philosophy of Language
  • From that perspective, we would say

    URIs have a single meaning but may have multiple interpretations

• In particular, the idea that meaning is a function from context to interpretation.

7.   What about web pages?

• Isn't that what URIs are really for?
• Short answer
  • No!
• It's perfectly OK to mint URIs with no intention to ever produce a web page at all
  • http://www.ltg.ed.ac.uk/~ht/Concepts/democracy
  • There, I just did it.
• Longer answer
  • 3 billion web page authors can't be wrong
  • But what they've authored are not the resources identified by the URIs
    • They're representations of those resources.
    • (from WebArch)

8.   Resources vs. representations

• Why make this distinction? That is, why isn't the web page itself the resource?
  • Some resources don't have web pages associated with them in any way
  • Many resources have more than one web page associated with them
  • Many URIs produce representations without there being any static web pages involved

9.   A digression about media types and encodings

• [Assume all URIs are http: URIs and all retrieval uses HTTP]
• A message, i.e. the response to an HTTP GET request, includes information about the character encoding and media type of the byte-string which follows.
• The character encoding tells you how to interpret the byte-string as a character string, e.g. iso-8859-1, shift-jis, utf-8.
• The media type tells you how to interpret the resulting character string, e.g. application/jpg, application/pdf, text/html

10.   Multiple representations

• So a server might have
  • utf-8 or iso-8859-1 encodings of
  • application/xml or text/html representations of
  • a document in English or Irish
  • representing today's weather report for Dublin
• 8 representations for one (time-varying) resource.
• WebArch uses the phrase 'information resource' for the subset of resources which can be pretty much completely captured in a message.
• So (today's) Dublin weather report is an information resource, but Dublin itself is not.

11.   Information resources

• Another way to look at this is that for information resources, the retrievable representations are (isomorphic to) the information resource itself -- consider
  • Beethoven's Symphony No. 6: http://downloads.bbc.co.uk/rmhttp/radio3/symphony6_beethoven_radio3.mp3
  • The Charge of the Light Brigade: http://eserver.org/poetry/light-brigade.html
  • The XML 1.0 Recommendation: http://www.w3.org/TR/REC-xml/
  • The W3C at 10 logo: http://www.w3.org/2004/09/logos/W3C10_logo_short_fc.png

12.   Open- vs. Closed-World Assumption

• What does all this have to do with the subject of this meeting?
  • Remember the network effect
• The history of XML is illuminating
  • XML was designed to solve the online-documentation problem
  • It succeeded for a completely different reason
    • The commercial value of data interchange
• The use of http: URIs as the foundation of the Semantic Web is intended to enable the same effect
  • But for machines this time, instead of people

13.   Architecture of the World Wide Web

• A grandiose concept
  • In practice an ongoing post-hoc analysis project with respect to a vast, messy, poorly delimited real-world phenomenon
• A document
  • The first publication from the Technical Architecture Group of the W3C (World Wide Web Consortium)
  • Architecture of the World Wide Web, Volume One
  • The goal of the document is to "preserve [the properties underlying the success] of the information space as the technologies evolve"
  • It contains a number of "Principles, Constraints and Good Practice Notes"

14.   Grandmother observations about the Web

• "Global naming leads to global network effects"
  • A prerequisite for citation
  • A challenge for many existing schemes
• "To benefit from and increase the value of the World Wide Web, agents should provide [http:] URIs as identifiers for resources"
  • I.e. not just any global name, but a particular kind of global name

15.   More from Grandma Webarch

• It's good for the ownership of a name (URI) to be manifest in the form of that URI
  • We need to know who is responsible for the name and its use
• "A URI owner should not associate arbitrarily different URIs with the same resource."
• "Agents do not incur obligations by retrieving a representation."
  • HTTP 'GET' is side-effect free (c.f. cookies)
• "A URI owner should provide representations of the resource it identifies"

16.   More on terminology and technology

• Retrieved representations are always accompanied by one crucial kind of metadata:
  • Its media type, e.g.
    • text/html
    • image/png
    • audio/mp3
    • See the IANA registry for a full list
  • That is, information sufficient to enable an agent to interpret the representation.
    • Some agents, e.g. web browsers, present the interpreted representations to human beings
    • Others, e.g. search engines, process the interpreted representations automatically for their own purposes
  • Retrieval may involve content negotiation
    • When multiple representations are available
      • JPG vs. PNG vs. SVG vs. GIF . . .
      • French vs. German vs. Chinese vs. . . .
      • DTD vs. XML Schema vs. . . .
• So information resources are not necessarily concrete
  • It's perfectly coherent to get real value from a URI which indentifies an abstraction
  • For example, http://www.guardian.co.uk/uklatest

17.   Connection with "persistent indentifiers"

• The TAG is concerned not only to protect the Web as it is
  • but also to extend its reach and value
• This often amounts to strenuous efforts to combat the NIH syndrome
  • It's so hard to accept that the solution to your problem is already available, off the shelf, but with someone else's name on it
• "No scheme or syntax guarantees persistence of any kind" (John Kunze)
  • Accordingly persistent identifier efforts can and should save huge amounts of fuss by focussing on the non-technology substrate issues involved in producing persistence
  • Just use http: URIs as the technology substrate

18.   Lookup plus hierarchy

• Virtually all public distributed naming system proposals we have seen have these properties
  • Some kind of global lookup to establish a context
  • Some kind of navigation through a structured space determined by that context
• These properties address the fundamental needs of such a naming system
  • It's distributed (no single central source of name bindings)
  • It's scalable (no intrinsic upper bound on what can be named)
• The http URI scheme is such a system
  • The domain name part (e.g. www.w3.org) can be resolved by appeal to a distributed lookup mechanism (DNS)
  • The path part (e.g. /2001/tag/webarch/) provides for scalable navigation to a particular name binding within that context
• There are huge benefits to using an existing, well-tried approach, that is, http URIs
  • So you need a very good reason to attempt to launch a new public distributed naming story

19.   (Parenthesis on distributed, secure and human-readable)

• You can have any two :-(
• See SFS (Self-certifying File System)
• The simplest kind of secure names are names which include a one-way unique hash of the thing that they name.

20.   Versioning and extensibility

• The scandal of computer science
• None-the-less, try not to forget that none of what you build will last forever
  • or even until next year

21.   Some folk wisdom

• The Web works because you can
  • View Source
  • Follow your nose
  • Write URIs on the side of a bus
  • Use generic tools
  • Redirect, cache and proxy
• The Web is hands-down the most successful distributed name-based system the world has yet seen
  • Postal addresses and the telephone system are not counterexamples
• Ergo anyone designing a persistent identifier system should start from the assumption that http: URIs are sufficient for their technology needs.
  • For an example of how to satisfy all the technical requirements of persistent identifiers using http: URIS see the ARK indentifier approach of John Kunze.

22.   So why is he shouting?

• What's this got to do with databases?
• Whether you translate once, or convert on demand, RDB->RDF requires new names
  • Since subjects, objects and relations all have to be URIs
• Working hypothesis: every attribute, and every key, will have to be converted to a URI
• What kind of URI?
• The observed fact of the matter is that communities confronting this question often feel they need to define their own URIs
  • A new URI scheme, e.g. xri:, info:
  • A new URN namespace, e.g. urn:lsid:..., urn:nzl:...
• There is no technical excuse for this
• The social dimension is, on the other hand, ineffable

23.   The Semantic Web

• A collection of technologies from the World Wide Web Consortium (W3C)
  • Inspired by Tim Berners-Lee
• The goal of the Semantic Web
  • Reproduce for machines what the original Web has done for people
• Alongside human beings finding information of the Web
• Computers finding information on the Semantic Web
• Distributed development of information resources which can be
  • Discovered
  • Merged
  • Exploited
• All without human participation.

24.   What's not special about the Semantic Web

• Two-place relations
  • understood as forming a graph
• Are nothing special
• An XML serialisation is nothing special
• Designing the language first
  • Then playing 'catch-up' with respect to inference
• Isn't new either, alas

25.   What is special about the Semantic Web

• It's KR for the Web
• That means one thing in particular
  • Coming from the fundamental architecture of the Web
• Everything has a universal name
• URIs (Uniform Resource Identifiers) are used for naming
  • Individuals
  • Classes
  • And relations.
• Usually http: URIs
  • http://www.w3.org/2002/07/owl#Class

26.   RDF example

RDF graph

relations

creator(homepage,ora)
name(ora,"Ora Lassila")
email(ora,"Lassila@w3.org")
ako(ora,Person)

RDF triples

http://www.w3.org/Home/Lassila
 http://purl.org/dc/elements/1.1/#creator
http://www.w3.org/staffId/85740
----------
http://www.w3.org/staffId/85740
 ???#Name
"Ora Lassila"
----------
http://www.w3.org/staffId/85740
 ???#Email
"lassila@w3.org"
----------
http://www.w3.org/staffId/85740
 http://www.w3.org/1999/02/22-rdf-syntax-ns#type
http://www.example.org/UpperOntology#Person

27.   The impact of URIs

• RDF graphs are composed of subject-relation-object triples
• All three parts are named with URIs.
• In principle, you can directly merge two independently-constructed RDF-encoded knowledge bases
  • Graph nodes named with the same URI can simply be merged
  • URI comparison is trivial and well-defined

28.   Merging ontologies

• The relations in a factual knowledge base
• Are the nodes (subjects or objects) in an ontology
• So in principle easy to imagine merging multiple ontologies

29.   Practical problems

• Independently developed knowledge bases
  • or ontologies
• Are unlikely to have chosen the same URIs for the same individuals
  • or concepts
• Even when the same URIs do turn up in two graphs
  • Identity of real-world denotation or ontological functionality depends on respect for the principle of URI ownership
  • As well as consistent interpretation of metadata by independent third parties
• This is sometimes called the SW 'Identity Problem'

30.   One clear benefit of URIs

• Using URIs for everything enables the 'follow your nose' strategy
• URIs identify resources
• URIs can be dereferenced
  • At least, http URIs can
• Dereferencing an SW URI should produce useful information about the thing/concept it is being used to identify.

31.   'Follow your nose' example

• Suppose I (or a program) find the following triple in a graph:

  http://www.example.org/Books/MobyDick
    http://purl.org/dc/elements/1.1/#creator
  http://www.example.org/Authors/HermanMelville

• Dereferencing the creator URI yields both human- and machine-readable information about the relation it identifies
  • Its official definition, in natural language/HTML
  • Its place in an RDF-encoded ontology (i.e. an OWL schema)

32.   The key prediction

• The ability to 'follow your nose' will solve the Identity problem.
• Creators of new knowledge resources will find existing related knowledge resources
  • (How?)
• Dereference the URIs of the subjects and relations therein
• Use the ontology they find instead of creating a new one
  • Or at least connect their ontology to the existing one(s)
• Then detecting and exploiting co-reference gets easier

33.   Conclusions, part 1

• The history of knowledge representation sends a clear message
  • Good data models are necessary, but not sufficient, for success
  • The ability to compute with the represented knowledge is the real test.
• In other words, ignore the inference engine problem at your peril.
• It seems likely the Semantic Web project will succeed in one way
  • As a knowledge representation system in its own right
• Much less clear if its larger goal can be achieved
  • The emergence of a virtual world-wide knowledge base.

34.   Conclusions, part 2

• One thing is clearly new and important
  • The Semantic Web project's focus on extending the boundaries of knowledge bases beyond the site at which they are developed
• Sends a clear message to the KR community
• Any knowledge resource creation effort
  • Particularly any one aimed at a large-scale and cooperatively developed knowledge resource
  should look very carefully at the ideology and the technology of the Semantic Web

35.   Conclusions, part 3

• No-one ever expects the network effect
• That is, committing to http: URIs for your (public) names will have unforseen benefits
  • And few if any real costs compared to any other naming scheme.
• Guidelines for automatic generation of URIs during RDB->RDF conversion are. . .
  • lacking.