23 Entity-Event Ontology Construction by Conceptualization …
345
broad range of learning methods like formal concept analysis, clustering, association rules, linguistic patterns, regular expressions, statistical methods, graph theory,
probabilistic graphical models, and so on [4].
Among widely acknowledged problems like axiom learning, there is also ongoing
research in the field of language-independent methods that are maintenance-friendly
and capable of learning structure on their own. One of the most promising methods
is based on neural networks and treating expressive ontology learning as a machine
translation task and mapping sentences to axioms [5].
23.3 Construction of Entity-Event Ontologies from Texts
Ontology learning is one of the methods for extracting conceptual knowledge from
texts. Those ontologies are used to partly represent the meaning of the text or at least
its structure. Often there are predefined classes in ontology, and the task is to extract
their instances by using pattern matching or other extraction methods. Another case
is when there is fixed base ontology and the task is to construct a domain ontology
by the means of extracting relations, classes, their instances, and so on. The term
“base ontology” is used here to denote that it can be both upper-level ontology with
a specific focus or just a sufficiently high-level domain ontology.
As mentioned earlier, most automatically constructed ontologies are focused on
universal linguistical concepts and usually employ abstract relations like IS-A, PARTOF, INSTANCE-OF, HYPONYM, HAS-VALUE, and others. Although suitable for
artificial intelligence research, it has low practical value in applied knowledge storage
and representation tasks. Applied knowledge bases and knowledge-based systems
typically utilize domain ontologies as they benefit from their focus on specifics and
predefined relevant formalism.
However, there is a space in between the following: some problems require
capturing of new ontological knowledge in a broad, but highly structured domain.
This kind of problems can be tackled by using two levels of ontologies—template
and instance ones—as outlined below:
• Template ontology acts as a description of what can happen in a world from
some practical viewpoint and sets a general structure for knowledge in supported
domains.
• Instance ontology is formed by extracting information from a set of documents
according to template ontology and de-facto constitutes a knowledge base.
Template or base ontology can be just an upper ontology like DOLCE, SUMO,
Cyc, and others, but that is not always the case, as it can be domain ontology as well.
It is quite common for applied knowledge bases and some cases of expert systems.
For example, one could define template ontology for representing knowledge
in court orders and then build a court order knowledge base in form of ontology.
Another case is a knowledge base of the company’s legal documents that can be
345
broad range of learning methods like formal concept analysis, clustering, association rules, linguistic patterns, regular expressions, statistical methods, graph theory,
probabilistic graphical models, and so on [4].
Among widely acknowledged problems like axiom learning, there is also ongoing
research in the field of language-independent methods that are maintenance-friendly
and capable of learning structure on their own. One of the most promising methods
is based on neural networks and treating expressive ontology learning as a machine
translation task and mapping sentences to axioms [5].
23.3 Construction of Entity-Event Ontologies from Texts
Ontology learning is one of the methods for extracting conceptual knowledge from
texts. Those ontologies are used to partly represent the meaning of the text or at least
its structure. Often there are predefined classes in ontology, and the task is to extract
their instances by using pattern matching or other extraction methods. Another case
is when there is fixed base ontology and the task is to construct a domain ontology
by the means of extracting relations, classes, their instances, and so on. The term
“base ontology” is used here to denote that it can be both upper-level ontology with
a specific focus or just a sufficiently high-level domain ontology.
As mentioned earlier, most automatically constructed ontologies are focused on
universal linguistical concepts and usually employ abstract relations like IS-A, PARTOF, INSTANCE-OF, HYPONYM, HAS-VALUE, and others. Although suitable for
artificial intelligence research, it has low practical value in applied knowledge storage
and representation tasks. Applied knowledge bases and knowledge-based systems
typically utilize domain ontologies as they benefit from their focus on specifics and
predefined relevant formalism.
However, there is a space in between the following: some problems require
capturing of new ontological knowledge in a broad, but highly structured domain.
This kind of problems can be tackled by using two levels of ontologies—template
and instance ones—as outlined below:
• Template ontology acts as a description of what can happen in a world from
some practical viewpoint and sets a general structure for knowledge in supported
domains.
• Instance ontology is formed by extracting information from a set of documents
according to template ontology and de-facto constitutes a knowledge base.
Template or base ontology can be just an upper ontology like DOLCE, SUMO,
Cyc, and others, but that is not always the case, as it can be domain ontology as well.
It is quite common for applied knowledge bases and some cases of expert systems.
For example, one could define template ontology for representing knowledge
in court orders and then build a court order knowledge base in form of ontology.
Another case is a knowledge base of the company’s legal documents that can be