Page 1
Modeling vs Encoding for the Semantic Web
Werner Kuhn University of Münster
Institute for Geoinformatics Münster Semantic Interoperability Lab (MUSIL)
Kuhn, W. (2010). Modeling vs encoding for the Semantic Web. Semantic Web - Interoperability, Usability, Applicability, 1(1), 11–15
Thursday, September 6, 12
Page 2
Thursday, September 6, 12
Page 3
Thursday, September 6, 12
Page 4
Integrating spatial information across vocabularies
Thursday, September 6, 12
Page 5
Claims in the semantic web
Thursday, September 6, 12
Page 6
Stated
“... ontologies are expected to be used to provide structured vocabularies that explicate the relationships between different terms, allowing intelligent agents (and humans) to interpret their meaning flexibly yet unambiguously...” [Horrocks et al., JWS, 2003]
”Technically, Linked Data refers to data published on the Web in such a way that it is machine-readable, its meaning is explicitly defined, ...” [Bizer et al., IJSWIS, 2009]
Thursday, September 6, 12
Page 7
Implicit
• description logic statements are necessary and sufficient to capture what people mean when they use vocabularies
• ontology engineers can say something useful about the semantics of vocabularies by expressing themselves in an encoding language for machine reasoning
• decidability matters in designing semantic models.
Thursday, September 6, 12
Page 8
How we are “living up” to these claims
Thursday, September 6, 12
Page 9
the
sem
antic
web
bot
tlene
ckHow we are “living up” to these claims
Thursday, September 6, 12
Page 10
the
sem
antic
web
bot
tlene
ckHow we are “living up” to these claims
meaningful communication
concepts
words
predicates
sortals
sets
⊓⊔⊑
Thursday, September 6, 12
Page 11
the
sem
antic
web
bot
tlene
ck
thro
win
g se
man
tics
out
of t
he w
indo
w
How we are “living up” to these claims
meaningful communication
concepts
words
predicates
sortals
sets
⊓⊔⊑
Thursday, September 6, 12
Page 12
the
sem
antic
web
bot
tlene
ck
thro
win
g se
man
tics
out
of t
he w
indo
w
How we are “living up” to these claims
context
pragmatics
structure
processes
prototypes
similarity
meaningful communication
concepts
words
predicates
sortals
sets
⊓⊔⊑
Thursday, September 6, 12
Page 13
Acknowledgments to Ross Purves
schema.org
Thursday, September 6, 12
Page 14
What I mean by “meaning”
Thursday, September 6, 12
Page 15
Thursday, September 6, 12
Page 16
Thursday, September 6, 12
Page 17
conceptualize
Thursday, September 6, 12
Page 18
conceptualize
express
Thursday, September 6, 12
Page 19
refer to
conceptualize
express
Thursday, September 6, 12
Page 20
The Ogden and Richards (1923) semiotic triangle
Thursday, September 6, 12
Page 21
Meaning as Process
• meaning is a process, not an object [Putnam 1975]
• “words don’t mean, people do”
• information results from referring to things through symbols
• information users interpret such references
• information providers can constrain such interpretationsmeadow := extensively used grassland [CORINE land cover class 231]
Thursday, September 6, 12
Page 22
How can the semantic web
constrain interpretations ?
Thursday, September 6, 12
Page 23
The Standard ApproachIt is useful to ...
• ... equate terms with classes (which are sets)class meadow
• ... assign properties (sets, again) to classesmeadow hasUse extensive
• ... declare sub-class (sub-set) relationships meadow is-a grassland
The arguments supporting this approach are based on formalization (rather than modeling) requirements.
Thursday, September 6, 12
Page 24
But ...It may be harmful to ...
• ... equate terms and classes, because terms are used in some contexts (but not in others)”is this a meadow or a field?”
• ... assign properties to classes, because “some do, some don’t”myMeadow instanceOf meadowsmyMeadow hasUse intensive
• ... declare sub-class (sub-set) relationships, because compositionality rarely holdsgrassland hasUse extensive
≠ extensive ⊓ grassland
Resulting discussions about “correct terminology” are a waste of time.
Thursday, September 6, 12
Page 25
Alternative ApproachIt is useful to ...
• ... document actual uses of vocabularies as triplestriples showing who calls what a “meadow”
• ... treat these as inconsequential type declarationssame piece of land could be typed “field”
• ... define type classes for shared behavior as ontology patternsclasses ARABLE, SELLABLE, BUILDABLE, ...
• ... inherit behavior to types playing rolestype meadow instantiates ARABLE, SELLABLE
• ... reason with type classesfield sameAs meadow, if same behavior
Thursday, September 6, 12
Page 26
A Modeling Language: Haskell
class (LINK link from to, SUPPORT from for, SUPPORT to for, CONTAINMENT medium link) => PATH for link from to medium where
move :: for -> link -> from -> to -> medium -> for
instance PATH Car Link Node Node Air
The standard modern functional language• clean, higher order type system• executable algebraic specifications• multi-parameter type classes
Thursday, September 6, 12
Page 27
What kinds ofontology patterns
are useful?
Thursday, September 6, 12
Page 28
Example: PATH
One source: Image Schemata
Thursday, September 6, 12
Page 29
Example: PATH
One source: Image Schemata
Thursday, September 6, 12
Page 30
Example: PATH
One source: Image Schemata
Thursday, September 6, 12
Page 31
Example: PATH
One source: Image Schemata
Thursday, September 6, 12
Page 32
• structured, invariant, compositional, physically grounded ...
• (re)capture some context, pragmatics, processes, prototypes
• more examples: link, containment, support, center-periphery, part-whole, ...
Example: PATH
One source: Image Schemata
Thursday, September 6, 12
Page 33
OWL:Classpath
OWL:Class:medium
OWL:Class:surface
OWL:Class:name
has part
part of
has
OWL:Class:endObjectOWL:Class:
startObject
OWL:Class:description
inon
has
has
has
Cardinality of all relations [0..m]
http://vocamp.org/wiki/GeoVoCampSB2012
Thursday, September 6, 12
Page 34
OWL:Class motion
OWL:Classpath
OWL:ClassstartEvent
OWL:Class movingObject
OWL:ClassmovedObject
OWL:ClassendEvent
OWL:Classname
OWL:Classdescription
OWL:ClassreferenceFrame
has part
part of
has
has
hashas
hashas
has
has
Cardinality of all relations [0..m]
http://vocamp.org/wiki/GeoVoCampSB2012
Thursday, September 6, 12
Page 35
Conclusions
Thursday, September 6, 12
Page 36
Conclusions
1. The semantic web emphasizes encoding for automated reasoning over modeling meaning.
Thursday, September 6, 12
Page 37
Conclusions
1. The semantic web emphasizes encoding for automated reasoning over modeling meaning.
2. As a consequence, there is an over-emphasis on set-based types, at the expense of higher level structure.
Thursday, September 6, 12
Page 38
Conclusions
1. The semantic web emphasizes encoding for automated reasoning over modeling meaning.
2. As a consequence, there is an over-emphasis on set-based types, at the expense of higher level structure.
3. I propose to constrain interpretations through second order type classes (as ontology patterns).
Thursday, September 6, 12
Page 39
Conclusions
1. The semantic web emphasizes encoding for automated reasoning over modeling meaning.
2. As a consequence, there is an over-emphasis on set-based types, at the expense of higher level structure.
3. I propose to constrain interpretations through second order type classes (as ontology patterns).
4. These provide small theories, easily combinable, for big data.
Thursday, September 6, 12
Page 40
Conclusions
1. The semantic web emphasizes encoding for automated reasoning over modeling meaning.
2. As a consequence, there is an over-emphasis on set-based types, at the expense of higher level structure.
3. I propose to constrain interpretations through second order type classes (as ontology patterns).
4. These provide small theories, easily combinable, for big data.
5. Specifications have started at the GeoVoCamp series (Santa Barbara, Dayton, ...): http://vocamp.org/wiki/GeoVoCampDayton2012.
Thursday, September 6, 12
Page 41
But, will this solve the fax-number-of-mountains
problem?
Thursday, September 6, 12
Page 42
But, will this solve the fax-number-of-mountains
problem?
Yes! mountains areendObjects of PATHs for hiking (not faxing)
Thursday, September 6, 12
Page 43
But, will this solve the fax-number-of-mountains
problem?
Yes! mountains areendObjects of PATHs for hiking (not faxing)
Thank You!Thursday, September 6, 12