How to Build a Foundational Ontology

The Object-Centered High-level Reference Ontology (OCHRE)

Luc Schneider Institute for Formal Ontology and Medical Information Science Faculty of Medicine, University of Leipzig, Germany - [email protected] -

Introduction

Foundational Ontologies
An ontology is the formal specification of a conceptualisation shared by intelligent agents (Gruber).
Any odd theory, even a mere taxonomy can count as an ontology !

A top-level ontology is a representation of the general categories and relations in the agents environment. A foundational ontology is a top-level ontology which constitutes a (formalised and axiomatized) theory (Guarino). Foundational ontologies are repositories of highly general information modeling concepts that can be reused in the design of application ontologies for all kinds of domains. As toolboxes of standardised KR primitives, they enhance the semantic interoperability between communicating agents. They allow to describe the ontological commitment underlying the truth-conditions of agent messages in a uniform way.

Building a Foundational Ontology

The challenges of building a foundational ontology are unfamiliar to most knowledge engineers, insofar it is an eminently interdisciplinary task. Indeed, the design options for top-level ontologies are identical to the ontological choices discussed in the branch of philosophy called metaphysics as well as in the research on qualitative reasoning. As an illustration of current approaches to the design of foundational ontologies, I outline the formalisation of my Object-Centered High-level REference ontology OCHRE, elucidating and motivating its basic modeling decisions.

OCHRE in nuce

4D vs. 3D
The two basic criteria for evaluating ontological choices are formal economy and simplicity as well as descriptive or intuitive adequacy.
It is obvious that these criteria are orthogonal to each other and that one has to find a compromise, which will be always contentious.

Every-day intuition is notoriously unreliable, but it cannot be ignored totally, since the ontology has to account for the meaning of natural language sentences. Fourdimensionalism would be the most economical choice as an account of persistence, but it rejects the distinction between objects and events that seems to be central to common sense reasoning and natural language understanding. Threedimensionalism is intuitively more appropriate. However, conventional endurantist accounts require an arkward timestamping of assertions about objects.

Thin and Thick Objects

In order to give an endurantist account of change which does not require time-stamping every assertion about an object, we distinguish between thin and thick objects:
A thick object is a mereologically and topologically invariant threedimensional entity that roughly corresponds to the phase of a fourdimensional entity or perdurant. A thin object is the core of invariant characteristics that is shared by a series of succeeding thick objects.

Thick objects are stages of thin objects; thin objects are the substrates or haecceities of thick objects. Change is described as the succession of thick objects that share the same thin object as a common core of features. This account is akin to Chisholms theory of entia successiva.
Note that, in our account, only thick objects can be bearers of topological, i.e. spatial and temporal, relations.

Objects and Properties

Ontologically, one has to distinguish between repeatable and non-repeatable properties and relations.
Repeatables (Redness, Humanity) apply to more than one case. Non-repeatables (your weight and your colour) are particular instances of repeatables and are characteristics of single objects. Repeatables are called universals and non-repeatables tropes.

In order to treat qualitative change as mereological change, it is convenient to regard (thin and thick) objects as being constituted by particular instances of properties/relations, i.e. tropes (Williams, Campbell). Thick objects have fine parts, i.e. tropes (like your weight), and gross parts which are thick objects themselves (like your arm). A trope inheres in a thick object iff it is a fine part which does not overlap with any of its gross (proper) parts (e.g. the weight of your body vs. the weight of your arm).

Guises
Common-sense allows for distinct entities to be co-located, or coincident, e.g., a terracotta statue and the clay it is made of, or a person and her body. Many ontologists (e.g. Wiggins) assume the possibility for spatial objects, i.e. thick objects, to coincide. Instead, I adopt a particularist version of Castaedas guise theory. The main idea is that thick objects can have more than one thin object as an haecceity or substrate and hence can be subdivided in non-overlapping trope-bundles that constitute qualitative aspects of the same thick object. Guises are fine, not gross parts. A particular thick object that we identify as a terracotta statue made of clay contains two guises as fine parts, namely the statue and the amount of clay, each centered on a particular thin object: the functions of the artifact and the chemical characteristics of the material.

Events and Processes

Events as state-transitions are pairs of immediately succeeding thick objects which share the same substrate;
The change of a tomato's colour from green to red amounts to the succession of a red tomato-stage to a green tomatostage. The change of a memory cell from 0 to 1 is the succession of a charged cell-stage to an uncharged one.

Processes are arbitrary mereological sums of events; they can be recursively characterised with single events as a base case. Participation can be accounted for in terms of parthood: a thin object participates in a process iff it is the core of an event that is a part of the process.

OCHRE en detail

Mereology (1)
OCHRE contains a unique, atemporalised parthood relation, which is:
atomistic: everything has parts which have no proper parts. extensional: if x has the same atomic parts as y then x is part of y.

An atomistic picture of the world would seem to imply some kind of reductionism. However, this is a fallacy. E.g. to say that a wall is made out of bricks does not mean that the wall is a brick, nor that it is a bundle of bricks. Indeed, we assume that what makes the wall this wall is its structure, a set of (topological) relations between the bricks. OCHRE assumes a number of so-called formal, i.e. domainindependent, properties and relations which shape the ontologically relevant composite entities (like objects and events) and distinguish them from arbitrary bundles.

Mereology (2)

Mereology (3)

Theory of Similarity (1)

Tropes form families whose members can be can be weakly ordered or compared according to their intensity. Atomic non-repeatables are assumed to be ordered unidimensionally. Multidimensional variations as in the case of colours indicate that the attribute is composite. The relation of weak order being more or equally intense than is the qualitative basis of scales and metrics. Two tropes x and y are comparable iff either x is more or equally intense than y or vice versa. Two tropes are (exactly) similar iff they are equally intense. Bundles of tropes are comparable or (exactly) similar iff their component tropes can be mapped one-to-one in pairs of comparable or exactly similar tropes. The genus of a trope x is the property of being comparable to x; its species that of being similar to x.

Theory of Similarity (2)

Theory of Dependence (1)

An individual x is dependent on an individual y iff x can only be identified with respect to y (Strawson). An independent entity is ontologically autonomous, i.e. all its determations are immanent to it (Ingarden). An whole cannot exist without its parts, but identificationally it may be the parts which are dependent on the whole. We call thin objects the self-dependent particulars. We assume that no single trope is a thin object. Self-dependent entities form the ontological backbone of reality, as they constitute the basic framework for identification. Tropes are the building blocks of reality, but they cannot be conceived of separately from the (thin) objects they characterise.

Theory of Dependence (2)

Theory of Connection (1)

Thin objects are the ultimate nodes in the framework of identificational dependences. Similarly, thick objects are characterised by being the nodes in a pervading grid of spatiotemporal connections. Spatio-temporal connection is a reflexive, symmetric and intransitive relation. Its underlying intuition is that of spatiotemporal contact or immediate neighbourhood in space and time (cf. Casati and Varzi). A thick object is enclosed in another iff it is connected to everything the later is connected to. Mutual enclosure is assumed to entail identity. Thick objects compete for space and time, they cannot be co-located. Thick parthood is defined as parthood between thick objects. Thick parthood is assumed to entail enclosure.

Theory of Connection (2)

Theory of Inherence (1)

The formal relation of inherence which holds between (thin or thick) objects and their attributes can be accounted for in terms of parthood. Since a thick object may have other thick objects as parts, one has to specify whether a trope or a thin object is associated with that thick object or one of its thick parts. E.g., one would like to distinguish the weight of a body and the weight of its heart. Such distinctions can be made through the relation of direct parthood. An individual x is a direct part of an entity y iff x is a proper part of y, y is a thick object, and there is no thick proper part z of y such that x overlaps with z. Every trope is a direct part of some thick object; there are no homeless tropes. Furthermore, no two comparable tropes may be both direct parts of the same thick object. E.g. a thick object cannot have more than one mass or kinetic energy.

Theory of Inherence (2)

A thin object x that is a direct part of a thick object y is an essence of y. Each thick object has at least one essence. It seems counterintuitive that a thick object may have more than one essence, but this is the case for most every-day objects such as artifacts (or organisms) and the amount of material they are made of. Each thin object that is an essence of a thick object is the core of a bundle of attributes which constitutes a guise, or qualitative aspect, of that thick object. The so-called inherence principle links atomic direct parthood and dependence by stipulating that, for any trope x, thin object y, and thick object z, if x depends on y and is a direct part of z, then y is an essence of z.

Theory of Inherence (3)

Theory of Temporal Order (1)

Thick objects are also the only trope-bundles which exhibit temporal relations. We assume discrete time in the sense of an enumerable series of atomic intervals. There are two primitive temporal relations: the irreflexive, antisymmetric and intransitive relation of direct anteriority and simultaneity which is an equivalence relation. Thick objects are supposed to form synchronised discrete series of succeeding entities in time. Thus one has to postulate that the direct antecedents of a thick object have to be simultaneous; the same holds for its direct successors. The relation of anteriority is defined by recursion on the basis of direct anteriority. In order to exclude temporal gaps, one can postulate that any two thick objects are comparable in terms of temporal anteriority.

Theory of Temporal Order (2)

Two thick objects that are spatio-temporally connected, such that one is directly anterior to the other, and share the same thin object as a common core, are said to be in direct essential succession. Essential succession is defined recursively in terms of essential succession. Essentially succeeding thick objects act like temporal counterparts of their underlying thin object. They are proxytruthmakers of temporal statements about thin objects.

Theory of Temporal Order (3)

Theory of Temporal Order (4)

Events and Processes (1)

The basic perdurants are events as changes or state-transitions: Some x is an event in a thin object y iff x is the sum of two thick objects that are directly essentially succeeding with respect to y; we also say that y is the substrate of x. The definition implies that there are no instantaneous events, which is consistent with the doctrine that perdurants have at least two distinct temporal parts. The instantaneous left and right boundaries of perdurants are endurants, namely thick objects. Processes are arbitrary mereological sums of events; they can be recursively characterised with single events as a base case. OCHRE's particular account of perdurants in terms of endurants allows for participation to be defined as a special case of parthood. Indeed, a thin object x participates in a process y, iff x is the substrate of an event that is part of y.

Events and Processes (2)

Conclusions

Conclusions (1)
Part-whole reasoning in Atomistic General Extensional Mereology, the basis of OCHRE, is quite simple, since the parthood structure is considered to be a Boolean algebra (without a null element). The problem of the identity of objects through change motivates many ontologists to reject extensional mereology and to recur to an opaque temporalisation of properties and relations, amongst them parthood itself. Furthermore, the apparent coincidence of objects is often tackled by introducing a non-mereological relation of constitution between objects.

Conclusions (2)
OCHRE maintains the descriptively correct distinctions of common-sense while avoiding formal intricacies.
To account for change in objects, OCHRE emphasises the ambiguity of references to objects and distinguishes between thin objects and thick objects as their evanescent stages. Temporal statements about thin objects are translated into atemporal statements about their stages. Events are accounted for in terms of succeeding object-stages. Co-located distinct entities are reconstructed as qualitative aspects or guises of the same thick object.

Classical extensional mereology can be preserved throughout, assuming attributes as atoms out of which thick and thin objects are ultimately composed, leaving no space for unscrutable substrates.