Pizza & Wine: The Need for Educational Tools for Foundational Ontologies Stefan SCHULZa,b,1, Martin BOEKERb, José A. VERA RAMOSa,c and Ludger JANSENd a Institute for Medical Informatics, Statistics and Documentation, Medical University of Graz, Austria b Institute for Medical Biometry and Statistics, Faculty of Medicine and Medical Center - University of Freiburg, Germany c Faculty of Biology, University of Murcia, Spain d Institute of Philosophy, University of Rostock, Germany Abstract. The educational ontologies PIZZA and WINE&FOOD have been widely used to teach ontology methods and tools. However, the two ontologies have remained largely unchanged for many years, despite increasing awareness of foundational principles for good ontology design (GoodOD). Recognizing that the two ontologies are lagging behind such principles, we analysed and re-designed them by strictly adhering to established upper-level ontology constraints provided by the foundation ontology BioTopLite2 (BTL2). As a result, the redesign required clarifying the ontological commitment of the PIZZA and WINE&FOOD classes by assigning them to top-level classes. We expect the ongoing redesign effort to yield a harmonized PIZZA&WINE&FOOD ontology. The redesign reflects the change in requirements for educational ontologies, which have now to take into account the privileged role that should be granted to foundational ontologies in Applied Ontology education. Keywords. Foundational Ontologies, Educational Ontologies 1. Introduction Teaching ontology to a broad variety of users requires a representational domain that is both intuitively understandable and complex enough to demonstrate all important features of ontology languages and editors. This is the reason why the ontologies WINE&FOOD and PIZZA have found broad acceptance in educational settings which focus on Description Logics (DLs) and the Semantic Web. WINE&FOOD was originally developed for teaching CLASSIC [1], an early dialect of description logics (DL), and was later transformed into Protégé Frames, for which ontology-oriented tutorials existed [2]. In the last stage of its current development, it was transformed into OWL-DL, and examples taken from WINE&FOOD were used to explain syntax and semantics in the first OWL-DL guide [3]. PIZZA was developed for training courses at the University of Manchester with the goal to teach the Web Ontology Language (OWL) and the ontology editor Protégé [4], where examples from PIZZA are used in its printed documentation. The two ontologies are, hence, primarily tailored to demonstrate the power of DL representation and reasoning (set theory, restrictions and 1 Corresponding author: Stefan Schulz, Medical University of Graz, Auenbruggerplatz 2/V, 8036 Graz (Austria), E-mail: stefan.schulz@medunigraz.at. quantifications), and to acquaint the students with Protégé. However, both ontologies have not been updated for years, and they have not kept up with more recent tendencies in ontology development, such as the good practice in the new technical discipline of Applied Ontology [5] to employ principles of the philosophical discipline of Ontology when engineering ontology artefacts. Both educational ontologies that are under scrutiny here are agnostic regarding ontological upper-levels and naming conventions, as well as regarding the de facto adoption of the language profiles OWL-EL and OWL- DL as the ones supported by the most popular DL reasoners. The objective of this paper is to evaluate the two educational ontologies on the background of these new developments. We first analyse how PIZZA and WINE&FOOD, in their current state, are positioned regarding quality criteria for good ontology design. Second, we test whether they can be redesigned to meet these criteria by aligning them with a pre-existing foundational ontology. Finally, we investigate whether the redesigned ontologies support two learning objectives, viz. (i) mastering OWL and description logics and (ii) understanding foundational ontologies. This paper primarily describes the alignment of these two ontologies with a foundational ontology. But besides giving an insight into redesign aspects (which could be an educational goal per se), we will also discuss the goals of ontology education, possible methods, the role of foundational ontologies in this process and the usefulness of PIZZA and WINE&FOOD when seen in this light. 2. Materials We use the PIZZA ontology version 2.0.02 and the WINE&FOOD ontology, which is provided as an annex to the 2004 OWL Web Ontology guide.3 “P” is used as namespace prefix for PIZZA, “W” for WINE&FOOD. PIZZA contains 100 classes and five individuals plus five object properties with three corresponding inverses (Table 1). Most of its content is under Food, which branches into the three disjoint classes Pizza, PizzaBase, and PizzaTopping. Pizza has subclasses like: CheeseyPizza, MeatyPizza, InterestingPizza, VegetarianPizza, etc. These subclasses of Pizza classes are subsumed under the helper class NamedPizza. The class PizzaTopping has subclasses like CheeseTopping, FishTopping, MeatTopping, etc. All toppings are primitive classes apart from SpicyTopping and VegetarianTopping. Axioms describe the composition of pizzas according to their base, toppings and country of origin of the recipe. Only the more general classes of the Pizza ontology are fully defined, e.g.: P:VegetarianPizza equivalentTo P:Pizza and not (P:hasTopping some P:MeatTopping) and not (P:hasTopping some P:FishTopping) PIZZA uses role hierarchies, nominals, inverse roles, unqualified number restrictions and data types [6]. The ontology includes six individuals, five of which are used in the extensional definition of the class P:Country. Most classes have English labels as annotation properties, according to the W3C standard [3] and many also Brazilian Portuguese ones. The annotation property ‘comment’ is mostly left blank, in 2 https://protege.stanford.edu/ontologies/pizza/pizza.owl 3 https://www.w3.org/TR/owl-guide/wine.rdf some cases it repeats the formal definition of a class; in other cases it has educational content. Table 1. Object properties of PIZZA with algebraic properties, domain and range. Numbers indicate usage, as displayed by Protégé Object property / Inverse Algebraic properties Domain Range P:hasCountryOfOrigin (14) – owl:Thing owl:Thing P:hasIngredient (17) / transitive P:Food P:Food P:isIngredientOf(17) P:hasTopping (301) / (inv) functional P:Pizza P:PizzaTopping P:isToppingOf (15) P:hasBase (22) / P:isBaseOf (16) (inv) functional P:Pizza P:PizzaBase P:hasSpiciness (77) functional owl:Thing P:Spiciness Table 2. Object properties of WINE&FOOD with algebraic properties, domain and range. Numbers indicate usage, as displayed by Protégé. Dashes are used to represent absence of algebraic properties Object property / Inverse Algebraic Domain Range properties W:adjacentRegion(10) symmetric W:Region W:Region W:course (10) – W:Meal W:MealCourse W:hasDrink (132) – W:MealCourse W:PotableLiquid W:hasFood (56) – W:MealCourse W:EdibleThing W:hasMaker (116) / functional owl:Thing owl:Thing W:producesWine (6) W:hasVintageYear(12) functional W:Vintage W:VintageYear W:hasWineDescriptor(14) – W:Wine W:WineDescriptor W:hasBody (162) functional owl:Thing W:WineBody W:hasColor (94) functional W:Wine W:WineColor W:hasFlavor (170) functional owl:Thing W:WineFlavor W:hasSugar (168) functional owl:Thing W:WineSugar W:locatedIn (188) transitive owl:Thing W:Region W:madeFromFruit (10) – W:ConsumableThing W:Fruit W:madeFromGrape(92) / – W:Wine W:WineGrape W:madeIntoWine (6) For WINE&FOOD we use the namespace prefix “W”. The ontology contains 138 classes, 16 object properties, one data type property and 206 individuals (e.g. with wine colours such as W:Red, W:Rose and W:White as individuals). The main classes are: W:Wine, W:Region, W:Vintage, W:VintageYear, W:Winery, and W:WineDescriptor with the subclasses W:WineColor and W:WineTaste. Table 2 shows the object properties. There is one datatype property, W:yearValue. 3. Methods We evaluate the ontologies in the light of the quality criteria stated in the GoodOD guideline for good ontology design of OWL ontologies with Protégé [7]. On the assumption that usability, maintainability and sustainable evolution of ontologies are important desiderata, the GoodOD guideline and its recommendations aim at optimising user-friendliness and interoperability of DL-based ontologies. GoodOD takes into account established best practice rules in ontology design [8] and class labelling [9], and reflects good practice of many previously published OWL ontologies. For the present study, we use the following quality requirements in particular: ● DL semantics should not be compromised: without exception, class-level axioms must hold true for all members of a class. ● There should be a robust and interoperable top-level ontology whose classes cover the domain exhaustively. For better usability, the top level is designed in a way that all domain classes can be uniquely attributed to one top-level class. ● Naming and annotation conventions should be consistently followed. ● Classes should always correspond to ‘repeatable’ features of the domain. E.g., the class P:Pizza can be instantiated by many particular pizzas; and P:Spicy is repeated in every particular spiciness. ● The division between individuals and classes should reflect the ontological or set-theoretic interpretation of ontology classes and not be motivated by mere subjective criteria for particular use cases. ● The object properties provided by the upper level ontology should be largely sufficient for the domain ontology. For new predicates, preference should be given to express them by means of process classes rather than as new ad hoc object properties (reification). ● The interpretation of the meaning of classes and object properties should be clear and understandable. We checked the two ontologies against these quality criteria (Sect. 4.1), from which we then derived principles for redesign. The two ontologies have then been redesigned accordingly while preserving all entities (Sect. 4.2). The resulting artefacts are then assessed against the questions stated in section 1 (Sect. 5). 4. Results 4.1. Quality analysis A positive point of PIZZA is the availability of an up-to-date, extensive and didactically optimized tutorial. We found, however, the following shortcomings: ● No reference to any foundational ontology. ● The top-level bipartition into the classes labelled “Domain concepts” and “Value partitions” seems to be purely navigational, its ontological significance remains unclear. ● Use of non-rigid classes [10], e.g. P:Food. ● Extensional class definitions: P:Country defined as having exactly five members, which is objectively wrong. ● Idiosyncratic object properties, with their meaning restricted to the Pizza domain, partly unclear such as P:hasCountryOfOrigin. ● Naming lacks precision, e.g. P:TabascoPepperSauce is surprisingly a descendent of P:PizzaTopping, because it means topping with tabasco pepper. In the WINE&FOOD ontology, we found the following shortcomings: ● No reference to any foundational ontology. ● Very domain-specific top-level partition into consumable and non- consumables, regions, vintages, wineries and wine descriptors. ● Idiosyncratic object properties: e.g. highly domain-specific as in PIZZA. ● Numerous potentially ambiguous labels, e.g. W:Loire intended to mean wine from the Loire region and not the river Loire. ● Unprincipled instance/class division: W:Chianti is a class, whereas W:ChiantiClassico is an individual. ● Extensional definitions, e.g. W:WineSugar defined as {W:Dry, W:OffDry, W:Sweet}. ● The ontology is completely devoid of metadata. ● In the tutorial [2], numerous references still point to the old Protégé frames version, using frame terminology, like “concept”, “slot” etc. 4.2. Redesign Applying the recommendations of the GoodOD guideline, we refined the two educational ontologies into the re-modelled counterparts PIZZA+ and WINE&FOOD+, with the namespace identifiers “P+” and “W+”. The new ontologies import the upper- level ontology BioTopLite2 (BTL2, namespace “btl2”) [11]. Accordingly, in PIZZA+, P+:Food is a subclass of btl2:PolyMolecularCompositeEntity. Since an instance of this class is required to have clear unity and identity criteria [10], we changed its name to “FoodItem”. The class is subclass of the top-level class btl2:MaterialObject, defined as having one mass and one volume at a time. Also, the object properties of BTL2 are intended to be exhaustive. In accordance with the GoodOD criteria we strived to fully express the meaning of the PIZZA+ and WINE&FOOD+ axioms by using exclusively BTL2 object properties. 4.2.1. PIZZA+ The re-implementation of PIZZA under BTL2 required numerous changes. First, we replaced all ontology-specific object properties by BTL2 object properties, e.g. P:hasTopping by btl2:hasComponentPart, a non-transitive subproperty of btl2:hasPart, which relates components with a compound. Components strictly partition the compound, and the compound is the mereological sum of its components. A loss of some component affects the integrity of the compound and can change the type it instantiates; e.g., it can change from a complete to a defective organism. The relation btl2:isBearerOf (inverse: btl2:inheresIn) relates realizable entities (qualities, roles, functions, dispositions, or information objects) with the physical entity they depend on. In BTL2, general localization is expressed by the object property btl2:includes (inverse btl2:isIncludedIn). It is transitive and relates a place with an entity which occurs, inheres, or is part of it. E.g., the original PIZZA included the following closure axioms: P:AmericanPizza subClassOf P:NamedPizza and P:hasTopping only (P:MozzarellaTopping or P:PeperoniSausageTopping or P:TomatoTopping) P:AmericanPizza subClassOf P:NamedPizza and P:hasTopping some P:MozzarellaTopping and P:hasTopping some P:PeperoniSausageTopping and P:hasTopping some P:TomatoTopping These axioms were transformed into PIZZA+ axioms by using only BTL2 object properties: P+:AmericanPizza subClassOf P+:NamedPizza and (btl2:hasComponentPart only (P+:MozzarellaTopping or P+:PeperoniSausageTopping or P+:TomatoTopping or P+:PizzaBase)) and (btl2:hasComponentPart some P+:TomatoTopping) and (btl2:hasComponentPart some P+:MozzarellaTopping) and (btl2:hasComponentPart some P+:PeperoniSausageTopping) We had to add some classes that were not included in the original ontologies in order to replace old ad hoc object properties by more precise constructs. The object property P:hasCountryOfOrigin had remained undefined in PIZZA. Common sense suggests that P:hasCountryOfOrigin is not meant to relate an individual pizza with the country where it was produced. In contrast to the wine and fruit, pizzas are not shipped around between countries. Therefore, we assume that an P+:AmericanPizza is a P+:Pizza that has been produced according to a recipe which was created in America. This leads us to create a class P+:Recipe, which is a btl2:Plan and therefore a btl2:InformationObject. btl2:Plan represents a thing in which several targeted steps (processes) are to be performed to reach a planned goal. A btl2:Plan can only be realized by some btl2:Process. It is a piece of information that exists independently of a particular material carrier. P+:Recipe has the subclass P+:PizzaRecipe with following definition: P+:PizzaRecipe isEquivalentTo P+:Recipe and (btl2:hasRealization only (btl2:Action and (btl2:hasOutcome some P+:Pizza))) In BTL2 an action is a process that has an agent [12]. Preparing food or creating a recipe should therefore be placed under the class btl2:Action: btl2:Action isEquivalentTo btl2:Process and btl2:hasAgent some owl:Thing Accordingly: P+:PreparingFoodAction subClassOf btl2:Action and btl2:hasOutcome some P+:Food P+:PreparingPizza EquivalentTo P+:PreparingFoodAction and btl2:hasOutcome some P+:Pizza P+:AmericanPizzaRecipe EquivalentTo P+:PizzaRecipe and (btl2:isOutcomeOf some (P+:CreatingRecipe and (btl2:isIncludedIn value P+:NorthAmerica))) By these means we can redefine P+:AmericanPizza as follows: P+:AmericanPizza EquivalentTo P+:Pizza and btl2:hasComponentPart only (P+:MozzarellaTopping or P+:PeperoniSausageTopping or P+:PizzaBase or P+:TomatoTopping) and btl2:hasComponentPart some P+:MozzarellaTopping and btl2:hasComponentPart some P+:PeperoniSausageTopping and btl2:hasComponentPart some P+:TomatoTopping and btl2:isOutcomeOf some (P+:PreparingPizza and (btl2:isRealizationOf some P+:AmericanPizzaRecipe)) The class P+:Spiciness in PIZZA+ describes a value partition for the classes P+:Hot, P+:Medium or P+:Mild. It is a btl2:ObjectQuality with the superclass btl2:Quality. 4.2.2 WINE&FOOD+ To redesign WINE&FOOD under BTL2, we aligned its class structure to BTL2 and replaced the WINE&FOOD object properties by BTL2 ones. E.g., the relations that indicate the qualities of a wine, viz. W:hasBody, W:hasColor, W:hasFlavor and W:hasSugar (in the sense of having a sweet flavour) were replaced with btl2:isBearerOf. Furthermore, new labels were added to several W+:AnimalOrganismPart classes to avoid ambiguity in an out-of-context situation. The class W+:ConsumableThing with the subclasses W+:EdibleThing, W+:Meal, W+:MealCourse, W+:PotableLiquid were defined as subclasses of btl2:PolyMolecularCompositeEntity. In WINE&FOOD some classes were subsumed by W:EdibleThing, e.g., W:Fowl, W:Meat and W:Seafood. The re-modelling requires changes in view of the classification, e.g., the classes W+:Fowl, W+:Meat and W+:Seafood are now subclasses of both W+:AnimalOrganismPart and W+:EdibleThing, the latter being defined as follows: W+:EdibleThing subClassOf W+:ConsumableThing and (btl2:isBearerOf some (btl2:Disposition and (btl2:hasRealization only W+:BeingEatenByHumans))) In order to distinguish between countable and mass entities, W:Fruit was re-labelled W+:FruitUnit to denote a countable thing; also its classification was changed: W+:FruitUnit subClassOf btl2:PolyMolecularCompositeEntity Not all fruits are edible, therefore we added: W+:EdibleFruitUnit equivalentTo W+:EdibleThing and W+:FruitUnit W+:NonSweetFruitUnit subClassOf W+:EdibleFruitUnit W+:SweetFruitUnit subClassOf W+:EdibleFruitUnit W+:MixedFruitUnit subClassOf W+:EdibleFruitUnit W:WineDescriptor is re-modelled as subclass of btl2:Quality: W+:WineDescriptor with the subclasses W+:WineColor and W+:WineTaste : W+:WineDescriptor subclassOf btl2:inheresIn some W+:Wine W+:WineDescriptor subclassOf btl2:inheresIn only W+:Wine The object property W:madeFromGrape was dropped and reified by a new subclass of btl2:Action, viz. W+:WineMaking:, a class representing the process of which wine is an outcome: W+:WineMaking isEquivalentTo btl2:Action and (btl2:hasOutcome some W+:Wine) and btl2:hasPatient some W+:Grape) W :Wine subClassOf (btl2:isOutcomeOf some W+:WineMaking) + The meaning of W:madeFromFruit was represented similarly by W+:FruitJuiceMaking, defined as a subclass of btl2:Action: W+:FruitJuiceMaking isEquivalentTo btl2:Action and (btl2:hasOutcome some W+:FruitJuice) and (btl2:hasPatient some W+:Fruit) Hence: W+:FruitJuice isEquivalentTo W+:PotableLiquid and btl2:isOutcomeOf some W+:FruitJuiceMaking Individuals like W:CabernetFrancGrape, W:GamayGrape, W:MalbecGrape, etc. are now typed as W+:WineGrape, whereas geographical regions are represented as classes like W+:AlsaceRegion, W+:AustralianRegion. The underlying definitions are of the following type: W+:AlsaceRegion isEquivalentTo W+:GeographicRegion and btl2:isIncludedIn value W+:Alsace Wine attributes like W:Red, W:Rose, W:Full, W:Delicate, W:Dry, etc. are classes in W+ and descendants of btl2:Quality. 5. Discussion The first aim of this paper was to evaluate the two popular educational ontologies PIZZA and WINE&FOOD with respect to common quality criteria. We have found that the two ontologies were not designed for interoperability and did not make any effort to align their classes with any foundational ontology. Whereas the PIZZA creators took care in good documentation, which partly elucidates the design decisions taken, WINE&FOOD reflects a pure computer-science approach towards ontologies as knowledge-representation devices for certain purposes, reflecting the notion of ontologies as “conceptualisations” [13], which characterised the discourse in the 1990s. This is concretised by ontology specialists’ recommendation that the decision “class or instance” is just a matter of convenience [2], to be driven by what the application requires. This postmodern flavour was then heavily attacked by proponents of “reality representation” like Barry Smith [14], who also were fierce advocates of foundational ontologies. In the meanwhile, this controversy has cooled down, mainly because interoperability-driven standardisation gained more importance in Applied Ontology, which prohibits overly idiosyncratic approaches anyway. The creation of BTL2 and the development of GoodOD were driven by the pragmatic motivation to simplify ontology construction, predominantly in the realm of biology and medicine. This is also the reason for its underlying naïve realism, which set it in close proximity to the Basic Formal Ontology (BFO) and the OBO Relation Ontology (RO). Second, though PIZZA and WINE&FOOD fared badly in this evaluation, we found that they can be re-designed such that their content is aligned with a pre-existing foundational ontology. Our redesigned ontologies demonstrate a successful proof of concept for such a redesign. In particular, the inventory of object properties contained in BTL2 proved to be sufficient for the domains covered. Idiosyncratic object properties in the source ontologies could be either replaced by existing BTL2 ones, or reconstructed by introducing reifying process classes. A considerable benefit of the avoidance of introducing new primitive object properties is the need to re-think the meaning of ambiguous and possibly fuzzy predicates like hasCountryOfOrigin, an object property that has turned out to have different meanings when applied to pizzas and to wines, respectively, which requires a thorough elucidation exceeding the expressivity of a single binary OWL object property. Thirdly, we intended to compare the educational value of the original and redesigned ontologies. PIZZA and WINE&FOOD describe material things, their constitution, qualities and processes they are involved in. In the terms of Karl Popper [15], these are “first-world” entities for which there is a large consensus that a realistic approach is adequate. Due to everyone’s familiarity with food, this subject matter constitutes an easy and straightforward entry point to ontology engineering for a broad range of learners. Domains that require to model mental, social, or legal entities would spark more controversy about upper-level divisions, provide less clear criteria for distinguishing between individuals and repeatables, and would therefore not be well served by a foundational ontology like BTL2. Regarding the two main learning objectives of (i) mastering OWL and description logics, on the one hand, and (ii) understanding foundational ontologies, on the other hand, the answers are clearly split. In the original Manchester tutorial for PIZZA, the focus was set (besides developing skills in using Protégé) on understanding the whole breadth of description logics including all technical intricacies connected with automated reasoning. This scope is clearly narrowed down by our redesigned PIZZA and WINE&FOOD ontologies, because the very point of using a highly constraining foundational ontology like BTL2 is to limit modellers’ degrees of freedom. This means for instance, that the manipulation of the “R-Box”, i.e., the creation and customisation of object properties, with their algebraic features, hierarchies, domain and range restrictions and CGIs is no longer admissible (nor necessary, as long as the premise holds that all necessary ontological relations are already provided by the upper-level ontology). On the other hand, with PIZZA+ and WINE&FOOD+ the ontology engineers’ curriculum now includes an in-depth understanding of a foundational ontology. This prevents naïve modelling decisions such as introducing ad hoc primitives like hasCountryOfOrigin. An ontologically proper representation of such predications is, however, not trivial and requires at least as many in-depth skills and knowledge in Formal Ontology as logic knowledge is needed for mastering the syntax and semantics of the representational language. Taken singly as well in combination, these ontologies can be applied for educational purposes, accompanied by a new tutorial which follows a clear educational strategy: After an introduction to tools (editor, reasoner), language, upper-level ontology and naming conventions, the modelling tasks are guided by textual characterisations of the entities and their relationships. This could be performed in two parallel groups, one working on the pizza domain and one working on the wine domain. The curriculum can then be concluded by a fusion of these two ontologies, demonstrating the interoperability of the two ontologies and thus proving the value of principled ontology design for interoperability of separately created ontology artefacts. 6. Conclusion The educational ontologies PIZZA and WINE&FOOD have been widely used to teach ontology methods and tools. However, the two ontologies have remained largely unchanged for many years, despite increasing awareness of foundational principles (GoodOD) for good ontology design. Recognizing that the two popular educational ontologies PIZZA and WINE&FOOD were lagging behind design principles and do not commit to any foundational ontology, we analyzed and re-designed them by adhering to upper level ontology constraints provided by the foundation ontology BTL2. As a result, the redesign required clarifying the ontological commitment of the PIZZA and WINE&FOOD classes by assigning them to top-level ontology classes. The ongoing redesign effort4 is expected to yield a harmonized combined PIZZA&WINE&FOOD+ ontology. As much as we acknowledge the value of educational ontologies, especially of the well-documented PIZZA, we are convinced that the education of ontologists must not be limited to training in the use of tools and the understanding of their logical foundations. Foundational ontologies ought to play a central role in the future of Applied Ontology as an engineering discipline; and this needs to be addressed by appropriate educational ontologies. The redesigned PIZZA+ and WINE&FOOD+ might be a first step in this direction. A detailed investigation of the reasoning results of the original and redesigned ontologies, their assessment against competency questions as well as benchmarking with common DL reasoners will be described elsewhere. 4 Documented in the GitHub repository: http://purl.org/biotop References [1] R.J. Brachman, D.L. McGuinness, P.F. Patel-Schneider, L.A. Resnick, A. Borgida, Living with CLASSIC: When and how to use a KL-ONE-like language, Principles of semantic networks (1991), 401–456. [2] N.F. Noy, D.L. McGuinness, Ontology development 101: A guide to creating your first ontology, Stanford knowledge systems laboratory technical report KSL-01-05; Stanford medical informatics technical report SMI-2001-0880, 2001. http://www-ksl.stanford.edu/people/dlm/papers/ontology- tutorial-noy-mcguinness-abstract.html. [3] M.K. Smith, C. Welty, D.L. McGuinness, OWL Web ontology language guide, W3C Recommendation, 10 February 2004. http://www.w3.org/TR/owl-guide/. [4] A. Rector, N. Drummond, M. Horridge, J. Rogers, H. Knublauch, R. Stevens, H. Wang, C. Wroe, OWL Pizzas: Practical experience of teaching OWL-DL: common errors & common patterns, Lecture Notes in Computer Science 3257 (2004), 63–81. [5] B. Smith, Applied ontology: A new discipline is born, Philosophy Today 12. 29 (1998), 5–6. [6] F. Baader, D. Calvanese, D.L. McGuinness, D. Nardi, P.F. Patel-Schneider, The Description Logic Handbook: Theory, Implementation, and Applications, 2nd Edition, Cambridge University Press, 2007. [7] S. Schulz, D. Seddig-Raufie, N. Grewe, J. Röhl, D. Schober, M. Boeker, L. Jansen, Guideline on Developing Good Ontologies in the Biomedical Domain with Description Logics, University Medical Center, Freiburg, 2012. http://www.purl.org/goodod/guideline [8] L. Jansen, S. Schulz, The ten commandments of ontological engineering. H. Herre et al. Ontologies in Biomedicine and the Life Sciences (OBML 2011), 2011/01: IMISE-Reports, Universität Leipzig, 2011. [9] D. Schober, B. Smith, S.E. Lewis, W. Kusnierczyk, J. Lomax, C. Mungall, C.F. Taylor, P. Rocca-Serra, S.A. Sansone, Survey-based naming conventions for use in OBO Foundry ontology development, BMC Bioinformatics 10.1 (2011),125. [10] N. Guarino, C.A. Welty, An Overview of OntoClean. In S. Staab and R. Studer (eds) Handbook on Ontologies, 201–220. Springer, Berlin/Heidelberg, 2009. [11] S. Schulz, M. Boeker, C. Martinez-Costa, The BioTop family of upper level ontological resources for biomedicine, Studies of Health Technology and Informatics 235 (2017), 441–445. [12] D. Seddig-Raufie, J. Jansen, D. Schober, M. Boeker, N. Grewe, S. Schulz, Proposed actions are no actions: re-modeling an ontology design pattern with a realist top-level ontology, Journal of Biomedical Semantics 3 Suppl. 2 (2012) S2. [13] T.R. Gruber, Towards principles for the design of ontologies used for knowledge sharing, In N. Guarino and R. Poli, editors, Formal Ontology in Conceptual Analysis and Knowledge Representation. Kluwer Academic Publishers, Dordrecht, 1993. [14] B. Smith, Beyond concepts: ontology as reality representation, Proceedings of the third international conference on formal ontology in information systems (FOIS 2004), 73–84. IOS Press, Amsterdam, 2004. [15] K. Popper, Three worlds, The Tanner lecture on human values, Univ. Michigan, 1978. http://tannerlectures.utah.edu/_documents/a-to-z/p/popper80.pdf.