Background

There are two major search problems addressed by semantic interoperability between data systems:

1. We cannot find all the data we are seeking.
2. We get too many results and they are difficult to classify.

Ontologies are mechanisms that can be used to help solve these problems. An ontology is a type of controlled vocabulary, which provides for categories, relationships, rules, and axioms among metadata values. Typically a hierarchy of terms, an ontology is a machine-readable way of relating metadata terminology.

Ontologies provide many capabilities. They can satisfy any requirement that a controlled vocabulary can satisfy (although some reformatting might be needed), including providing definitions, controlling possible answers to questions, and ensuring uniform spelling. Because ontologies support a rich set of relationships among different vocabulary terms, they enable a much fuller understanding of terminology and concepts than most controlled vocabularies. For example, the MMI Platforms Ontology (under development) currently includes categorizations of different kinds of oceanographic platforms, as well as complex properties, such as types of mobility, and other platform qualities.

How Ontologies Can Help

Ontologies can be used by automated tools to power advanced services such as more accurate web search, intelligent software agents, and knowledge management. By formalizing relations between concepts of one or more collections in a machine-readable language, ontologies can facilitate interoperability. These concept descriptions determine the format in which the information is kept, and establish the actual conceptual information, or semantic content, that is defined in the ontology. Agreements also should be reached about the community and technical processes used to modify the ontology. Finally, ontologies are designed to be computer-usable (also known as 'computable') - their format and rules are specified so that the information can be found, exchanged, and applied by computer systems, without additional human intervention.

Some examples of how ontologies can faciltate interoperability:

• Knowledge of a Domain

• This image of a natural catastrophe ontology demonstrates how an ontology can represent a domain of interest (from Robert Laurini INSA –Lyon). Ontologies can fully represent a domain of interest (using concept terms and relationships) and thus enhance interoperability.

• Mappings Between Controlled Vocabularies

• Controlled vocabularies are important, but there is rarely only one controlled vocabulary relevant to a domain of interest. Different funding sources, project purposes, program histories, etc. lead to different controlled vocabularies for a given domain. Mappings between controlled vocabularies, normalized in ontology representation languages such as the Resource Description Framework or the Web Ontology Language (OWL), can consist of identifying terms in each vocabulary as equivalent to, broader than, narrower than, or a subclass of terms in another vocabulary. Such ontology representations and mappings can enhance interoperability between data systems in that the use of specific search terms is no longer necessary. The mappings between terms in different controlled vocabularies used in different data systems can allow the user to find additional information. For example, at the MMI Advancing Domain Vocabularies Workshop in 2005, we demonstrated the enhanced ability to quickly find sea surface temperature data sources (regardless of whether "SST", "sea surface temperature", "Ocean Temperature" variations were used), using an MMI semantic mediation service called Semor. Semor is a semantic mediation service for earth science terminologies. Terminologies are expressed in ontologies following the RDF model. Users can query terminologies using RDF query languages or simple text matching queries. This service helps users discover what a term means and its relationships to other terms.

• Mappings Between Categories/Hierarchies of Concepts

• Taxonomies (or other hierarchies) used by different data systems, as well as within a data system, may vary. Ontologies, and mappings between ontologies, can facilitate interoperability between these higher level categorizations. For example, the Oregon Coastal Atlas and the Marine Irish Digital Atlas, which strive to interoperate as components of an International Coastal Atlas Network, use different classifications for grouping their mapping data sets to help users find data sets of interest. MMI is working with this group to create an interoperability prototype between the two atlases, using an upper ontology, as well as mappings between classifications and terms.

How Ontologies Work

Ontologies can represent concepts (as classes), individuals (members or instances of the classes), characteristics of each concept (as properties), and relationships between the concepts in a machine-usable language, based on the Resource Description Framework. RDF is a graph data model, where concepts are represented by nodes, and the relationships between them represented by the lines linking them. RDF, which employs subject-property-object triples and Uniform Resource Identifiers to define ontologies, makes it possible for computers to readily use the information directly represented in ontologies. In addition, depending on the precise type of relationships allowed in a given ontology, computers can 'reason' in various ways -- drawing real-time conclusions relating to terms and data -- using the knowledge embodied in the relationships of the ontology. The standard framework that ontologies provide for representing information and relationships enable many general-purpose software capabilities that would not be feasible with other technologies. For example, search engines can 'understand' that a person looking for 'coastline' is also interested in 'shoreline' within physical coordinates that correspond to an oceanographic geospatial location. Training and testing tools can be written that leverage the information in an ontology, without changing the tool, to present new information to a student. And as 'reasoning systems' become more advanced, the 'raw knowledge' in the ontology can be leveraged with other systems and ontologies, giving computer systems a much more general ability to deal with environmental concepts.

A controlled vocabulary in a simple format, such as ASCII, can be converted into RDF (by using the MMI tool Voc2RDF, for example). The terms in this new ontology can then be mapped to terms from another ontology that covers the same subject matter. Tools exist that facilitate mappings between vocabularies that are formatted as RDF or OWL files, and save the result in a similar format. (For example, one free tool is the MMI tool, VINE.) The resulting OWL-formatted mappings file can then be used in web services to search (or otherwise interoperate between) multiple data systems for the domain of interest. This advanced search/interoperability is founded on the computers' new "understanding" of the meanings of terms and the relationships between them across the different data systems.

Have a specific question about ontologies? Ask MMI!