The Use of Formal Ontology to Increase the Adaptability, Capacity and Efficiency of Natural Language Processing Software

The central hypothesis of the collaboration of L&C; and IFOMIS is that the methodology of formal ontology will benefit application ontologies such as L&C;’s Linkbase. In this paper we discuss one of the general procedures to be implemented, with examples of several areas in which it has already brought greater clarification and perspicuity (clarification of ambiguity, allowing better future algorithm design, i.e. less human operator reliance, as well as a framework for a future translation hub) to the Linkbase ontology. The general procedure has been the implementation of a meta-ontological definition space, in which definitions of all concepts and relations of Linkbase are standardized in a framework of first-order logic. We then describe how this standardization effort has led to improvement of Linkbase’s treatment of parthood relations, relation between processes and objects, treatment of absence, and of functions. Our description also points to ways in which application ontologies in general are forced to grapple with genuinely philosophical issues. General Procedures – Standardization Linkbase is a medical domain ontology designed to integrate different medical terminologies and ontologies for use in Natural Language Processing applications. This task turns out to be staggeringly complex, since the different terminologies/ontologies to be integrated are often ambiguous and internally inconsistent, and mutually inconsistent to an even greater degree. Linkbase provides a central “hub” ontology, with fixed structured definitions into which external medical terminologies/ontologies may be embedded. BFO (Basic Formal Ontology) is a philosophically inspired top-level ontology. For millennia, when we have encountered problems in understanding reality, we have turned to philosophers for solutions. Now, when we encounter problems in understanding how to represent reality, we must do the same. The cause of the aforementioned ambiguities and inconsistencies has been precisely the lack of a unified framework for understanding many of the basic formal relationships that structure reality (of object to process, of universal to particular, parthood, dependence, and so forth). BFO provides a coherent, unified understanding of these relationships. The implementation of BFO, therefore, as a top-level or “backbone” ontology for Linkbase, will not only provide a framework for the clarification of existing ambiguities and discrepancies in and between ontologies, but will also provide a template for future revision and augmentation of those ontologies. Thus, the implementation of a philosophically sound top-level ontology will provide the necessary link to successful integration, as well as be a useful guide for future algorithm development. The BFO ontology will provide Linkbase with standardized, formal (first-order) definitions of Linkbase elements (concepts and binary relations). This will disambiguate Linkbase itself, and isolate regularities which will facilitate axiomatic reasoning based on these formal definitions, and more generally the development of future algorithms. The standardization is an implementation of philosophical rigour in two dimensions. First, the first-order language used will be the language in which BFO is defined and axiomatized. Thus, the rigour of the BFO classification system is imported into Linkbase. This is a “metasystematic” importation of rigour, in that few changes are made to the elements themselves, but rather their place in a BFO-founded domain ontology is “tagged”. The second dimension of rigour will be of the conceptual analysis variety. Linkbase itself may be viewed as an “object language” or a surface structure. It consists of a number of concepts and binary relations between them. Its axioms at this stage are therefore merely a list of instantiated binary relations. Yet these relations and concepts are given only in natural language, and their grammatical form leads to various ambiguities. Thus, the project of defining a unique “deep structure” common to every such concept, relation, and axiom requires sound conceptual analysis. The BFO standardization provides for this. The analysis is to run as follows: 1) 1) For every Linkbase concept C, the definition is a mapping to a pair: 2) For every Linkbase relation R(X,Y), the definition is a mapping to a Π2 formula (where X and Y are variables ranging over Linkbase concepts): For all x: x is the universal named by X or x is in the extension of that universal, There is a y: y is the universal named by Y or y is an element in the extension of that universal, such that R*(x,y) (where R*(x,y) is a relation in the formal language of BFO) Certain relations will be tagged to specify that only the concept universal, or only its class of instances, serve as relata. 3) Axioms run according to the relation paradigm, with the variables replaced by specific Linkbase concepts. The Π2 structure was chosen since this turned out to be compatible with the intended reading of 99% of existing Linkbase elements. The remainder of this article is devoted to describing a small selection of cases where this philosophical scrutiny has improved Linkbase.