You are on page 1of 1

158

Ontology

reality: for Aristotle and many others, primarily objects and processes. In the empirical 19th century, metaphysics was perceived as experimentally unverifiable and, therefore, irresponsible, and parts of it may have been so. There has been no getting away, however, from the necessity of understanding the basic categories underlying both human thinking and the spectacular successes of empirical sciences. In the 20th century, the concern about foundational issues of physics and other hard sciences has become common and acute. The term ontologywas used for a while as a politesubstitute for metaphysics, but towards the end of the century, the later term regained its respectability. Accordingly, ontology started being used more correctly as the actual result of metaphysical research, a specific system of basic concepts. The major issue of metaphysics is the status of properties. The realists recognize the existence of both individuals, such as houses or cars, which are spatio-temporal entities, and of their shared properties, such as costliness or redness, which are not spatio-temporal but rather abstract. Their opponents, the naturalists, recognize only individuals. Both camps, existing in a large number of variations, have a difficult philosophical issue to deal with: for the realists, the different status of properties; for the nominalists, the existence of those properties. For a specific ontology, the matter is only important in one sense, namely, whether the ontology is to contain properties or not, and the nature of any reasonable application usually imposes the realist solution. 2.2.2 Formal ontology Formal ontology (see Guarino 1995; 1998) was developed late in the 20th century as a cross of mathematical disciplines of mereology, which studies the relations between parts and wholes, theory of dependence, and topology. But it was originally Edmund Husserl, earlier in the century, who saw a need to establish formal ontology as the theory of things parallel to logic, the theory of truths. The agenda of formal ontology typically includes explorations into notions of time and space, modality, especially the deontic (imperative, must-do) phenomena, taxonomy of artifacts, and types of inheritance. A specific ontology, in its implementation, has to take a position on each of those and other issues of formal ontology, so the awareness of these issues is essential for the developers. 2.2.3 Ontologies as engineering artifacts The practical use of an ontology involves constructed reality (see Gruber 1995). The MikroKosmos ontology, containing over 7,000 actual and many more virtual concepts (i.e. concepts derivable from the actual concepts with the help of well-defined rules), was created by a team of computer scientists and linguists over almost a decade at considerable cost (see Nirenburg and Raskin 2004). After the original time- and effort-consuming basic research, which established the top-level concepts and the format of each type of entry, a major effort was devoted to the development of

semi-automatic tools of acquisition enabling the participation in the work, at its simplest level, of rank-and-file enterers with a minimal training. This was achieved by an inventive use of easy-to-understand templates for several basic types of concepts. The introduction of the tools made the acquisition of numerous concepts by minimally trained workers very speedy and inexpensive. The resulting ontology meets and exceeds the industrys standards of clarity, coherence, and extendibility, and it stands to reason to conclude that any ontology of the future, comparable to this particular one in size and depth, will have to be developed more or less along the same lines: a combination of the tiered approach to acquisition, the use of semiautomatic tools, and the emphasis on cheap labor whenever and wherever practical.

2.3

ONTOLOGY AND NATURAL LANGUAGE

It is in the development of MT and other natural language processing (NLP) systems that the ontological approach proved to be particularly successful (see Nirenburg and Raskin 2004). Early MT was of the transfer type: a system was designed on the basis of a minimally required rules of transfer from words of the source language into words of the target language. While the simplest transfer, word for word, does not work in translation, inventive rules were created to overcome this difficulty in many cases. The resulting translation was, however, pretty rough and not very accurate, making it hard to use without expensive human post-editing. The alternative to transfer MT was the interlingua approach. Interlingua was a formal representation of text. The input text was analyzed and represented as an interlingua text and the latter was used to generate an equivalent text in the target language. An ontology was introduced as a rich and sophisticated form of interlingua in the late 1980s and won a dominant position in the interlingua approach in the 1990s. The MikroKosmos ontology is the foundation of the whole knowledge- and meaning-based approach to language processing. As a universal language-independent resource, the ontology is the metalanguage for the lexicon: each lexical entry in a natural language is defined in ontological terms. In the simplest case, the meaning of a word corresponds exactly to an ontological concept; in this case, the entry contains a simple pointer. More often, the match is not exact and an additional constraint, formulated in terms of ontological properties, must be added to the entry. Yet in other cases, the entry must contain a reference not to a concept but rather to a certain slot and/or filler in its frame. Even more importantly, the ontology provides a conceptual foundation and world grounding for whole texts. Working with the lexical entries making up a sentence and using the pertinent syntactic information for the natural language being processed, the analyzer produces ontology-based formula, the text meaning representation, for the sentence.

You might also like