Where are the Semantics in the Semantic Web?

1
Where are the Semantics in the Semantic Web?

• Michael Ushold
• The Boeing Company

2
Evolution of the Web

• Locating Resources evolving from keyword search
  to semantic search
• Users evolving from human only to human and
  machines
• Services evolving from a place to find things to
  a place to do things.
• Semantics evolving from little or no explicit
  semantics to rich semantic infrastructure

3
Semantic Web Vision

• No widespread agreement on exactly what the
  semantic web is
• Clear emphasis on
• Machine usable Web content
• Has more meaning
• Requires machines to
• know how to recognize content
• know what to do when they encounter it
• Machine access to the semantics of Web content
  is at the heart of confusion about the Semantic
  Web.

4
Semantics A Many-Splendored Thing

• Semantics meaning
• Kinds of semantics
• Real-world mapping of model into the real world
  for human interpretation
• Agent communication language performatives e.g.
  request or inform
• Axiomatic set of descriptions expressed in a
  logic language
• Model-theoretic describes conditions objects
  must satisfy to be assigned meaning
• Intended vs. actual meaning we usually intend to
  describe one model but actually describe several
• Things that have semantics
• Terms referring to real-world objects (e.g.
  semantic markup)
• Terms in agent communication languages (e.g.
  inform)
• Languages for representing these terms (e.g. OWL)

5
Semantic Continuum
reduce hardwiring and begin to infer something
hardwired
6
Machine Processible Semantics

• Possible for agents to automatically infer
  something
• How when never before encountered?
• Extremely difficult for humans, never mind
  machines
• Have to make assumptions
• Language heterogeniety a single language already
  known to the agent
• Incompatible conceptualizations (e.g. time
  intervals vs. time points) must be compatible
• Term heterogeneity impossible to guess intended
  meaning must correspond to a publicly declared
  concept (i.e. must be a concept in a shared
  ontology)

7
Formal Semantics for Machine Processing
Semantics of the ontology hardwired but the
agent can infer automatically.
8
Side Comments

• Usholds comments
• remains an unproven conjecture that such
  approaches will enhance search capabilities, or
  have significant impact on the Web.
• insufficient business drivers to motivate
  venture capitalists to heavily invest in Semantic
  Web companies.
• Other comments
• A computer doesnt truly understand anything,
  but computers can manipulate terms in ways that
  are useful and meaningful to the human user
  Berners-Lee
• Key Point the manipulation only has to be good
  enough. And thats our challenge and our
  opportunity!

9
Requirements for Machine Usable Content

• The machine needs to know what to do with the
  content it encounters.
• Since humans write programs, humans must know
  what to do.
• So, humans must know the meaning of the expected
  content.
• Then, humans can hardwire the agents
• Or, humans can hardwire the semantic
  specifications used by agents and can even share
  these specifications publicly.

10
Hardwiring

• Automatically determining meaning of Web content
  likely impossible
• Thus, humans will always be hardwiring
  semantics into Web applications
• Question What is hardwired and what is not?
• Hardwire term understanding into every agent?
• Hardwire the semantics using representation
  languages
• Only one language?
• Only one conceptualization? (or do we need
  mappings back and forth?)
• Shared, publicly declared ontologies? (or
  private?)

11
Agreeing

• The more agreement the better (?)
• Emerging standards (OWL)
• Without agreement
• effort required to make sure the concepts are
  right
• Guesswork may undermine the reliability of
  applications

12
Sharing

• Assumed we will share
• Standards emerging in a variety of sectors
• Dublin Core for elements like title, subject,
  date,
• NewsML and PRISM for news and magazine publishing

13
Web shopping agents work because

• Humans know how to proceed
• Everything hardwired
• Although no agreement, strong overlap in
  underlying concepts
• Semantics not specified, but generally understood
• Although no public standards, the general
  understanding makes them unnecessary
• The requirements are met
• Humans know the meaning
• Humans know what to do with the content
• Humans program the machine to know what to do

14
Side Questions

• Whats hardwired and whats not for our
  information extraction ontologies?
• How do we manage agreement?
• Do we (could we) share?

15
Vision of the Semantic Web

• Do Web shopping agents satisfy the vision?
• Probably not
• Degenerate case perhaps
• Genuine examples are there any?
• How to move toward the vision
• Move along the semantic continuum to more clearly
  specified (formal) semantics
• Note theres nothing inherently good about being
  further along the continuum
• What is good is what works.
• Reduce the amount of hardwiring
• Change which parts are hardwired
• Increase the amount of inference
• Increase the amount of public standards and
  agreements The more agreement there is, the
  less it is necessary to have machine processable
  semantics.
• Develop technologies for semantic mapping and
  translation

16
So, Where are the Semantics in the Semantic Web?

• Often just in the human
• In informal specification documents
• Hardwired in implemented code
• In formal specifications to help humans
  understand and/or write code
• Formally encoded for machine processing
• In axiomatic and model-theoretic semantics of
  representation languages