09/22/07Andrew Frank1 Data Quality Ontology: An Ontology for Imperfect Knowledge Andrew U. Frank Geoinformation TU Vienna frank@geoinfo.tuwien.ac.at .

09/22/07Andrew Frank 1

Data Quality Ontology:An Ontology for Imperfect Knowledge

Andrew U. FrankGeoinformationTU Viennafrank@geoinfo.tuwien.ac.atwww.geoinfo.tuwien.ac.at

09/22/07Andrew Frank 2

Ontologies describe (perfect) conceptualizations of a (perfect) world

Our knowledge of the world is usually imperfect, incomplete, vague, uncertain and with errors.

Can we build an ontology of imperfect knowledge?

09/22/07

Motivation:

I have heard (and probably said myself):

“All human knowledge is imperfect.”

09/22/07

Counterexample:

A bill of $50 is exactly 50 Australian dollars, not 49.99 or 50 with an error (normally distributed) of 2 cents.

The knowledge is accurate.

I will show in this talk that the types of imperfections are related to ontological tier.

09/22/07

Structure of this presentation

1. Tiers of ontology2. Ontological Commitments for an

ontology of imperfect knowledge3. What can an ontology of imperfection be

used for?

09/22/07

Tiers of ontology:

0. Physical reality1. Point property values2. Object properties3. Socially or subjectively constructed

objects

09/22/07

Ontologies are described through the ontological commitments they imply.

Ontological commitments state the obvious.

It is useful to write them down to investigate what they entail.

Considering the negation of a commitment demonstrates often its validity.

09/22/07

Commitments for a Geographic Ontology• a single world• the world exists in space and evolves in

time• actors can observe some states of the

world• actors can change states of the world

Example of an entailment: Commitments (3) and (4) allow communication between agents.

09/22/07

Commitments for a GIS

• information models reality• information causation is different from

physical causation

09/22/07

Partial Knowledge:

• only part of the world is known• not all states of the world are

observable

09/22/07

Limited Capacity of Biological Agents for Observation and Information Processing:

• concentration on discontinuity• object boundary location is uncertain• multiple ways to cut reality in objects• objects formed with respect to

interactions• preferred objects endure in time• object properties derived from

observable states

09/22/07

Mental Classification of Objects

Classification reduces the information processing load.

• membership of an object in a class is based on object properties

• reasoning uses default values

09/22/07

Social or Subjective Construction

X counts as Y in context Z.(John Searle)

Example:This piece of paper counts as 50 Australian dollars here (i.e., in the context of the Australian regulation for commerce)

09/22/07

Commitments for Constructions:

• all constructions are physically grounded in an object or an action (no “freestanding Y terms”).

• the context is a set of rules connecting to other constructions:- personal history- social conventions- legal, scientific rules

09/22/07

What use is an Ontology of Imperfection?1) process based definition of (top level)

ontology2) theory for imperfection in knowledge3) definition of data quality terms (vague,

precise, accurate, ..)

09/22/07

Imperfections in the data must be linked to the ontology

An ontology can be formally modeled through the processes which produce the knowledge.

The conceptualization of the imperfections in the data must be linked to information processes in the ontology.

09/22/07

Information processes between the ontological tiers:0. Physical reality

observations of phys. properties at points ->

1. Point property valuesobject formation (granulation) and determination of object properties ->

2. Object with propertiessocial construction ->

3. Socially or subjectively constructed objects

09/22/07

Basic observations of physical propertiesThis is the only source of knowledge about

the physical reality of the world.Imperfect realizations as physical process

(scale, random error).

09/22/07

Physical object formation (Granulation)Objects as regions with uniform properties

-> threshold -> boundary

Object property as integration of some observable value over object region.

Modeling of propagation ofimperfections from observations is possible.

Example:Where is the boundary of the mountain?

09/22/07

Classification of objects by use:

What use do the physical properties of the object permit?

Typically a fuzzy classification!

Example: Is region x a mountain?

09/22/07

Imperfections of constructed facts:Is this $50 note a fake?

• Who makes authoritative decision? by what process?

• What are the physical properties determining the constructed fact?

• What is the context? (space and time dependent).

Possible theories:- supervaluation- f-theories of Zadeh

09/22/07

Information processes cause the imperfections in the data The processes which produce our

knowledge about the world are responsible for the imperfections in our knowledge.

All knowledge based on observation of the physical world is imperfect (but not knowledge about socially constructed facts!)

09/22/07

Conclusion:

An ontology must explain how we construct knowledge from observing realityincluding the imperfections in the observationsand other information processes used.

Imperfections in the data are caused by imperfection in the information processes.

09/22/07

Metaconclusion:

Are the attempts to discuss ontology and data quality separately comparable to the attempts to capture geographic space and time separately in the 1990s?