Knowledge Classes or Canonical Representation


PhD Thesis Notes, Pedro Ferraz de Abreu, 1996

My approach to deal with the multiple-domain / multiple source problem in building knowledge bases is to establish a non-ambiguous, mutually exclusive classification of different types of knowledge, in other words, a canonical representation. The rationale is that by encapsulating each and all knowledge units in one of these categories, we create a virtual level of knowledge representation where the dominant traits are not domain-dependent, since we can define them at a syntactic level, instead of a semantic level.

This canonical representation was achieved by reviewing a large set of multi-domain vocabulary (more than one thousand items) and several field taxonomies (from different school curricula, job market demand and supply on domain qualifications, etc.). As a result, I identified the following categories: Term; Concept; Definition; Model; Rule; Norm; Procedure; Methodology; Description.

Term:

• Short word or sentence ;

• Represents an element of technical, scientific or cultural vocabulary; or a variable in an algebraic expression;

• May be defined in a simpler and less technical language (Glossary);

• Does not require extensive explanations or complex theoretical foundation;

• Definition may contain other terms only .

Concept:

• Word or sentence ;

• Represents an idea or abstraction (technical, scientific or cultural), or a knowledge domain (class, sub-class, domain);

• May be explained in lay language, eventually requiring more or less complex theoretical foundation;

• Explanation may contain terms or other concepts, of similar or lesser complexity.

Definition:

• One or more sentences;

• Represents the exact, non ambiguous explanation of a term or concept; or establishes an axiom, which should, in this case, be considered a term or concept;

• There may be more than one definition per concept, and they may or not contradict themselves (if they do, it implies the co-existance of several truth/belief systems);

• Explanation may contain other terms and concepts, other than the object being defined, of similar or lesser complexity.

Model:

• One or more algebraic expressions (set of variables linked by algebraic or logical operators );

• May establish an axiom ( variables must also be considered terms).

Rule:

• Regular expression [IF precedent THEN consequent], in which precedent and consequent are a set of one or more conditions linked by the logical operator AND, where condition is a 3-tuple variable-operator algebraic-value;

• Represents a causal or dependency relationship between phenomena, identified through investigation and not arbitrarly set.

Norm:

• Regular expression [IF precedent THEN consequent], in which precedent and consequent are a set of one or more conditions linked by the logical operator AND, where condition is a 3-tuple variable-operator algebraic-value, and the consequent part may be a set of conditions or a set of procedures;

• Represents a causal relationship resultant from arbitrary determination.

Procedure:

• One or more phrases or images;

• Represents a sequence of one or more acts ( operations, interventions ) of one or more agents acting on one or more target-objects (people, things, entities, etc.);

• Is conditioned by rules or norms.

Methodology:

• Set of norms and procedures.

Description:

• One or more phrases, images or sounds;

• Factually represents things, people, entities, places, events, situations or states ;

• May contain models, terms, concepts and other descriptions.