Nicola Guarino started his course on “Ontological Analysis and Conceptual Modeling” in the last slot on Monday. The course was divided in four parts, going on until Thursday.
Nicola started with a summary of the course: the ontological level — the need for ontologically grounded, non-neutral representation constructs; the basic “tools” of formal ontological analysis: unity, essence, identity, dependence, etc.; the impact of ontological analysis on the practice of conceptual modeling — OntoClean, OntoUML; the DOLCE ontology and its key distinctions — endurants, perdurants, qualities, etc.; from Entity-Relationship conceptual modeling to Episode-centric conceptual modeling.
He characterized applied ontology as an interdisciplinary field that builds on philosophy, cognitive science, linguistics and logic with the purpose of understanding, clarifying, making explicit and communicating people’s assumptions about the nature of structured world.
What distinguishes applied ontology from philosophical ontology is this orientation towards helping people understanding each other. Related to that, there’s the activity of ontological analysis, which is the study of content (of these assumptions about the world) as such (i.e., independently of the way they are represented).
He then reminded that Conceptual Modeling was also in the title of the talk. One classical definition is that of Mylopoulos: “Conceptual Modeling is the activity of formally describing some aspects of the physical and social world around us for the purposes of understanding and communication.” The similarity with ontological analysis is visible.
The main phenomenon studied by ontological analysis is truth-making. What makes them true? Where? When? Who? Why?
Nicola then provided his definition of what a computational ontology is, which allowed him to quickly mention ontology quality and then make the case of ontological precision: the lack of it may lead to false agreement which may lead to errors in systems that are based on the model.
Furthermore, precision alone is not enough. Non-intended models may be excluded, but the rules for the competent usage of a concept in different situations may not be captured and also lead to problems.
He ended this part of the course with the notion of kind, going from the logical level, to the epistemological level and finally to the ontological level. At the logical level, all predicates are the same, whereas at the ontological level not all properties are the same. Using his classical red apple example, you would say there is an apple with color red, but not a red with a shape apple. This is due to the fact that apple carries an identity condition, whereas red does not.
The discussion of kinds and non-kinds will continue in the next part. Slides for the first part were made available at the summer school’s website.
Nicola continued where he stopped, saying that not all properties are the same. Take a person — John. “height = 6 feet” is an intrinsic quality; “right-leg = broken” is a part relation; “mother = Jane” is a role; “kissed = Mary” is an external relation; “job = researcher” is a relational quality. We need different primitives to express these different things.
These tools come from Philosophy: theories of essence and identity, parts (Mereology), unity and plurality, dependence, composition and constitution, properties and quality. This forms the basis for a common ontology vocabulary.
Currently, this is not being used by computer scientists in general. Take the Semantic Web for instance: the initial architecture proposed by Tim Berners-Lee had an ontology layer but its more recent revision, which is a more accurate view of what it currently is, does not. Description Logics tends to collapse properties as of being of the same type.
Nicola then presented the tool of ontological analysis and formal ontology, exemplifying it with a discussion on Mereology. He followed with a discussion on unity and plurality.
Towards the end, Nicola mentioned that Software Engineers usually run out of arguments when discussing models and stating that they prefer one model or another. Ontological Analysis gives you the tools to justify why you prefer one model or another.
Slides for the second part were made available at the summer school’s website.
On Wednesday Nicola talked about OntoClean. He started discussing essential properties for concepts. He stressed that us, Software Engineers, are in a comfortable position of being able to decide what is essential or not depending on the domain we’re working on (e.g., for a university information system being a student or a professor may be declared essential). He then related the concepts of essence and unity.
He went on to present the concepts of rigidity, non-rigidity and anti-rigidity; identity criteria and moving from identity criteria to identity conditions. From identity conditions he defined Sortals (kinds that carry identity condition, usually nouns) and Non-sortals (kinds that do not, usually adjectives). He also presented the distinction between supplying identity (+O) and carrying identity (+I), which could come from a concept that is subsumed.
Identity Criteria (IC) impose an important constraint: properties with incompatible IC are by definition disjoint. Unity can be a special case of Identity Condition, in which case properties with incompatible unity conditions are also disjoint. So OntoClean also has meta-properties for unity: +U (all instances have a common unity criteria), ~U (no instances have unity criterion) and -U (some instances have unity criteria).
Nicola motivated the use of these meta-properties in order to improve communication between modelers and resolving ontological conflicts. It could even help on a harder thing, which is discovering that there is a disagreement.
Finally, he showed an example of application of OntoClean: assign the meta-properties, remove non-rigid properties, analyze (and fix) taxonomic links according to the meta-properties constraints, resulting on the backbone of the ontology. The process goes on to include the non-rigid properties and possible missing types.
Slides for the third part were made available at the summer school’s website.
In the fourth and last part of his course, Nicola focused on DOLCE. DOLCE is a foundation ontology that focuses on particulars (does not include universals). It has a descriptive attitude (they don’t try to convince people of how the world is), emphasis on cognitive invariants, and is a rigorous model (standing on the shoulders of philosophers).
Nicola stressed the importance of language on the definition of the concepts of the ontology. By analyzing sentences in natural language you can identify multiple co-located objects, multiple co-located events and individual (reified) qualities. These concepts are then included in the ontology in order to represent these characteristics.
He then presented the basic taxonomy of DOLCE, which is mainly divided in objects (endurants, 3D continuants), events (perdurants, 4D occurrences), qualities and abstract entities. He pointed out the importance of associating the technical terms endurant and perdurant to the common terms object and event, which are very overloaded in natural language and don’t represent exactly what these concepts mean. What followed was a long discussion about individual qualities.
Nicola went on to present the four-category ontology — Kinds (substantial universals), Objects (substantial particulars), Attributes (non-substantial universals) and Modes (non-substantial particulars) — and how the individual qualities would fit such an ontology. To accomodate it, he four-category ontology would be extended to have 7 categories: Kinds, Quality Kinds, Quality Spaces and Qualia on the universals side; Objects, Qualities and Quality Manifestations on the particulars side. Nicola also made the case that Quality Manifestations are, in fact, perdurants (events).
In the last five minutes, Nicola briefly presented the ongoing work with Giancarlo and others on episodes, which are maximal occurrents. He also sees episodes as truth-makers of relations.
Nicola closed with his favorite conclusion: “Ah, but this is hard. Of course it is hard! Why should it be simple? There’s no reason for it to be simple.”
Slides for the fourth part were made available at the summer school’s website.