We view ISO/IEC 11179 registries as a means of linking together fragments or references of "information artifact" schemas (e.g., data element specifications) to conceptual models (e.g., ontologies, etc.). The rationale for such linkages is to provide more precise, complete semantic specifications for information artifacts. Here we discuss proposed revisions to ISO/IEC 11179 to better capture such ontologies, especially formally encoded ontologies (e.g., in description logic).
Our motivation for adding "ontologies" to ISO/IEC 11179 registries is detailed further below. Here we note some major motivations:
Here we discuss a series of proposed changes to ISO/IEC 11179 Part 3 Edition 3 concerned with registering ontologies and formal statement sets (formerly axiom sets) in ISO/IEC 11179 metadata registries. . This is part of an effort to expand the scope of ISO/IEC 11179 from data elements and data element concepts to encompass terminologies, thesauri, database schemas, and ontologies. Each of the individual changes will be discussed in separate ISO/IEC 11179 issue documents. This document is intended to provide an overview of the proposed changes. Specifically we are concerned with supporting:
[Note that the currently proposed revisions to the metamodel may differ slightly from this discussion due to the currently rapid evolution of proposed revisions to the metamodel.]
Among the central problems confronting us in the design of a registry metamodel to support ontologies are:
From Tom Gruber:
Ontologies as a specification mechanism:
A body of formally represented knowledge is based on a conceptualization: the objects, concepts, and other entities that are assumed to exist in some area of interest and the relationships that hold among them (Genesereth and Nilsson, 1987) . A conceptualization is an abstract, simplified view of the world that we wish to represent for some purpose. Every knowledge base, knowledge-based system, or knowledge-level agent is committed to some conceptualization, explicitly or implicitly.
An ontology is an explicit specification of a conceptualization. The term is borrowed from philosophy, where an Ontology is a systematic account of Existence. For artificial intelligence (AI) systems, what "exists" is that which can be represented. When the knowledge of a domain is represented in a declarative formalism, the set of objects that can be represented is called the universe of discourse. This set of objects, and the describable relationships among them, are reflected in the representational vocabulary with which a knowledge-based program represents knowledge. Thus, in the context of AI, we can describe the ontology of a program by defining a set of representational terms. In such an ontology, definitions associate the names of entities in the universe of discourse (e.g., classes, relations, functions, or other objects) with human-readable text describing what the names mean, and formal axioms that constrain the interpretation and well-formed use of these terms. Formally, an ontology is the statement of a logical theory.[1]
An ontology consists of things (and their descriptions), and a set of formal statements for describing relationships among these things and allowable operations which can performed on these things. A formal statement is a sentence in logic.
Ontologies provide a formal approach to conceptual modeling of (usually) specific domains. Shared ontologies can be used as a means to anchor application specific schemas, i.e., as a mechanism to link together various database/data set schemas. The formality of ontologies, e.g., encoding in description logic or other logics (e.g., first order predicate logic) permits the possibility of certain types of automated reasoning, querying, consistency and constraint checking which is not feasible with more informal approaches to conceptual modeling - such as natural language definitions.
Ontologies are of interest to ISO/IEC 11179 registries for these reasons:
The third rationale, registering ontologies, arises due to the multiplicity of efforts to create various domain specific ontologies. Just as with database schemas, message schemas, and XML document schemas, the multiplicity of ontologies generates a need to record, register, and link the variety of ontologies to support applications which need to interact with more than one ontology.
The rationale concerning multiple types of metadata registries reflects our view that there will (in any case) be multiple ontologies and probably multiple ontology registries (for reasons indicated in previous paragraph). If ISO/IEC 11179 does not embrace registration of ontologies, some alternative standard for this will be adopted and metadata registrars will be confronted with the necessity of operating two types of metadata registries and coordinating content mappings among them.
The are many different formalisms for encoding ontologies and axiom sets: KIF, DAML+OIL, OWL DL, Common Logic, ... The question of how we will support the various encodings is one of the central issues affecting the registration of ontologies. This question is intertwined with the question of how detailed we want to make the registration of ontologies and/or axiom sets, i.e., do we want to register entire ontologies or axiom sets, subsets, or possibly individual fragment (entities, statements, ...)
Possibilities include:
Encoding the ontology as a CLOB, with an annotation of the type formalism is the simplest approach, but this approach sacrifices (or at least defers) the possibility of reasoning using the ontology.
Adding a symbol table to the CLOB permits searching on symbols and linking these symbols to data element concepts in the ISO/IEC 11179 registry. For example, the concept "motor vehicle" might be used to define various concepts including trucks, automobiles, motorcycles, tanks, etc.
However, CLOB + symbol table still lacks the ability to exploit the subsumption relations embedded in many ontologies. Thus it would be desirable to also capture these subsumption relations in the registry encoding of the ontology. Similarly it would also be desirable to capture instance-of relationships. Note that the proposed MMF (Meta Model Facility) standard supports the symbol table + subsumption relationships approach to ontology registration. In the ISO/IEC 11179 setting we might register both concepts and individuals, along with some predicates and operators. This is similar to what is done in the DIG interface specification across description logics.
Encoding the ontology as a parse tree or abstract syntax tree offers the prospect of querying and/or reasoning using the ontology (axioms). However, this approach requires that the registry understand the full syntax of the various ontology encodings.
Translation of ontologies to a common representations (such as Common Logic) would then permit querying and reasoning using the common representation. However, a suitably expressive common representation may be difficult to reason over.
See the work of the OMG's Ontology Definition Metamodel working group for discussions of proposals to map ontologies to a common formalism.
Formal statements (e.g., axioms, sentences in various logics) can be used to specify a wide variety of constraints concerning objects and relationships encompassed by an ontology. Such constraints include:
The use of formal statements to encode constraint specifications offers two possible advantages: a more precise specification of the constraints (cf. to natural language); and support for machine processing (e.g., consistency checking, compilation into run-time integrity checking code, etc.).
In order to support a variety of data integration and data exchange tasks it will be necessary to describe the mappings among concepts, data elements, etc. in one ontology or database schema to those in another ontology or database schema. Formal statements (axioms, sentences in logic, etc.) can be used to encode mappings among "objects/concepts" in various ontologies, terminologies, taxonomies, and/or database/message schemas. Again, the two advantages are a possible increase in precision of describing the mappings, and the possibility of machine processing (reasoning, query translation, etc.) of the formalized mappings. We anticipate the need to explicitly specify such mappings, since it will often be technically or politically impractical to directly merge two distinct ontologies.
Ontologies, databases, etc. differ by what assumptions they make concerning whether the system is an open world or closed world (as described below). Given this diversity we need to specify for each ontology, etc. which assumption is operative. All ontologies and databases make some such assumptions.
Closed worlds assume that the ontology/database encompasses the entire universe of discourse. Thus if a person is not in the personnel database then we may infer that he (or she) does not work for the organization. Closed World Assumptions (CWA) are nearly universal (unspoken assumption) in database systems.
In contrast, most knowledge representation systems, e.g, ontologies, etc., make an Open World Assumption (OWA), that the knowledgebase is presumed to be incomplete. Thus the absence of a particular fact from the knowledgebase does not permit the inference that the negation is true. This is very common in many scientific settings, in which the absence of a positive fact may simply indicate that no one has done a particular experiment.
Similarly, ontologies differ as to whether they make the unique name assumption (UNA) (described below. Hence, it is necessary to specify whether the unique name assumption holds for a particular ontology.
The unique names assumption (UNA) asserts that different names for objects implies that they are different individuals, i.e., individuals do not have aliases. Such an assumption is used in equality testing. Unique names are called "keys" in database parlance.
The absence of the unique names assumption implies that individuals may have many names and that no inference of distinctness may be drawn by the observation of differing names. On the World Wide Web (WWW) unique names assumptions are not valid, because hosts may have more than one name, and files may have multiple links to them. In many file systems which support multiple links to a file the UNA does not hold. Finally, in email systems a user may have multiple email addresses (identities).
In order to register a formal definition we will require:
We suggest creating a generalization hierarchy of definitions, with subclasses of formal definition and natural language definitions. Alternatively, one may wish to consider "formal definitions" as a subclass of "formal statement sets". There have also been suggestions to make "formal definitions" into "designated items" to facilitate naming them (in multiple languages).
See discussion above concerning registration of formal definitions.
Recall that administered items in ISO/IEC 11179 carry a variety of metadata indicating provenance, authorship, dates of last revision. Making axiom sets into administered items would facilitate attaching such metadata to axiom sets. There is some debate as to whether formal statements or sets of formal statements should be administered items.
More generally it is clear that there is a desire that individual formal statement sets (and definitions) should be individually addressable units.However, there is some reluctance to make individual formal statements into administered items for two reasons: excessive burden of recording administrative metadata, and a desire to treat blocks of statements (or perhaps entire ontologies) as the integral administrative units. This is akin to the version management and propagation problem in Computer Aided Design (CAD) databases, wherein changes to the lowest level components need to be propagated as version changes to higher level subsystems.
If we are to allow composite administered units of metadata, rather than performing all administration at the individual item level, then we will a two new classes, "administered item" and "administered unit of metadata", and a new metamodel relationship connecting them, "contained in administered unit of metadata". Administered items will not directly have administrative metadata, but will inherit it from the containing administered unit of metadata. [Note: This is not the most recent proposal by K. Keck for the metamodel.]
This discussion also bears on all of the new proposed administered data items.
There is a perceived need to break large ontologies into smaller modules, and to be able to construct larger ontologies by importing ontology modules (sometimes referred to as axiom sets). Arguably, it may be desirable to name such ontology modules.
In the most recent proposed revisions to the ISO/IEC 11179 metamodel we have conflated ontologies, ontology modules, and formal statement sets (formerly axiom sets) into "ontologies". There has been considerable debate as to whether this is the most appropriate name. Ontologies are currently comprised of concepts, relationships, predicates, definitions, and sub-ontologies. We propose to have a (recursive) importation relationship on "ontologies" (a.k.a. "concept systems", "ontology modules", ...) to support modularization.
This issue was proposed by Mala Mehrotra, from Pragati Systems. Pragati Systems is interested in partitioning large formal statement sets into manageable pieces. To this end they develop templates for specifying "rule templates" these templates are used to cluster together similar formal statements which are related (reference one another).
One desiderata is the ability to represent complete (or possibly fragments of) ontologies in the metadata registry. See discussion of aspect E, of differing formalisms for encoding the ontologies.
On many occasions it may be necessary to include alternative formal representations of an ontology. The alternatives may or may not use differing formalisms (e.g., types of description logic), or the same formalism with a different (equivalent) set of axioms. This suggests the need to be able to specify equivalences among ontologies (ontology modules) and perhaps mappings among ontology symbols.
Concepts may have (slightly) differing definitions in different contexts. We therefore need a means of specifying a "context" and annotating concept definitions, etc. Since natural language definitions already have contexts, if formal definitions are a subclasses of the (now generalized) definitions they will inherit the context attribute (relation).
In the most recent proposed revisions to the ISO/IEC 11179 metamodel we have proposed that concepts, formal definitions, relationships, predicates, etc. all are part-of exactly one ontology (module). Thus the "ontology" provides an effective context for all of these entities / relationships. If we allow concepts, relationships, et al. to belong to more that one "ontology" then we may need to explicitly specify an enclosing "context" for every such entity/relationship in an ontology.
This page is maintained by Frank Olken. It was last updated on 2005-09-19 5:05 PM PDT