Over the years, there has been an increasing adoption of ontologydriven conceptual models to represent the conceptual structure of critical domains in reality. Given the complexity of this task, there has been a growing demand for the development of proper engineering tools for supporting the design of these models. Despite a number of advances in this area, there is still a shortage of tools directed at novice and non-technical users that can, at the same time, address two competing requirements, namely: maintain modeling expressivity by being able to represent true ontological distinctions, while remaining intuitive and easy to learn by this class of users. In this article, we sketch a proposal in this direction by introducing the idea of an Ontology Model Canvas.
In recent years, there has been an increasing interest in the application of ontologies in
conceptual modeling, including the use of foundational ontological theories to improve
the theory and practice of this discipline [
Given the increasing complexity of Ontology-driven Conceptual Modeling, there is
an urging need for developing theoretically well-founded methodological and
computational tools for this discipline [
The results of these efforts have been noticed in academic, industrial and
governmental settings. For example, OntoUML has been successfully employed in
several industrial projects in different domains, such as petroleum and gas, digital
journalism, complex digital media management, off-shore software engineering,
telecommunications, retail product recommendation, and government [
Despite the theoretical and practical contributions of these efforts, these tools have
not been particularly developed with novice and non-technical users in mind.
Especially for these classes of users, we need tools that are able to, on one hand, deal
with the inherent complexity of the modeling problems at hand, and on the other hand,
shield these users as much as possible from this inherent complexity. In this paper, we
advance a proposal in this direction by conceiving an ontology-based modeling
representation mechanism that, despite having the expressivity of a large set of
ontological distinctions put forth by UFO, could in theory be easily learned and
employed by novice and non-technical users. It is important to highlight that we are
here specifically trying to avoid the route taken by modeling approaches that are
oblivious to true ontological distinctions and, hence, that are in our opinion too
simplistic for supporting the development of Reference Ontologies [
In this proposal, we took inspiration on a significant success case in the area of
Business Modeling, namely, the Business Model Canvas (BMC). The BMC, proposed
in [
The remainder of this paper is organized as follows. In the next section, we present a whirlwind introduction to those categories of UFO that are germane to the purposes of this article. In section 3, which is the core of this article, we present our preliminary proposal of an Ontology Modeling Canvas. Section 4 presents some additional 2 http://www.menthor.net/model-repository.html examples that serve to illustrate the modeling expressivity that can be achieved with such an approach. Finally, section 5, presents some final considerations of the paper.
In the sequel, we briefly explain a selected subset of the ontological distinctions put
forth by the Unified Foundational Ontology (UFO). For an in depth discussion,
philosophical justifications, formal characterization and empirical support for these
categories one should refer to [
Take a domain in reality restricted to endurants [
Object kinds and subkinds represent essential properties of objects (they are also
termed rigid or static types [
Kinds, Subkinds, Phases and Roles are categories of Object Sortals. In the
philosophical literature, a sortal is a type that provides a uniform principle of identity,
persistence and individuation for its instances [
A Relationship Kind represents clusters of relational properties that “hang together”
by a nexus (provided by that relationship kind). In other words, relationships (e.g.,
marriages, enrollments, employments, presidential mandates, citizenships) are
fullfledged endurants, i.e., entities that endure in time bearing their own essential and
accidental properties and, hence, first-class entities that can change in a qualitative
manner while maintaining their identity [
Objects participate in relationships playing certain “roles”. For instance, people play the role of spouse in a marriage relationship; a person plays the role of president in a presidential mandate. Spouse and President (but also typically student, teacher, pet) are examples of what we technically term a role in UFO, i.e., a relational contingent sortal (since these roles can only be played by entities of a unique given kind). There are, however, relational and contingent role-like types that can be played by entities of multiple kinds. An example is the “role” Customer (which can be played by both people and organizations). We call these role-like types that classify entities of multiple kinds Rolemixins. In general, types that represent properties shared by entities of multiple kinds are termed Non-Sortals. In UFO, besides rolemixins, we have two other types of non-sortals, namely Categories and Mixins. Categories represent necessary properties that are shared by entities of multiple kinds (e.g., the category Physical Object represent properties of all kinds of entities that have masses, spatial extensions, etc.). In contrast, mixins represent shared properties that are necessary to some of its instances but accidental to others (e.g., the mixin Red Object can be thought as representing properties that are necessary to entities of certain kinds – for instance, rubies, while being accidental to entities of other kinds – for instance, apples). Categories and mixins are, in contrast to rolemixins, considered as Relationally Independent Non-Sortals.
Figure 1 illustrates the different regions comprising our tentative proposal for an Ontology Model Canvas (OMC). These regions are elaborated in subsections from 2.1 to 2.5 below. Loosely speaking, the central part of the canvas is divided in four regions. The two regions to the left (i.e., regions corresponding to categories, and kinds/subkinds) should be read in as “the ways things necessarily are” (according to a given conceptualization of the domain); the two regions on the right (i.e., regions corresponding phases and roles) should be read as the “the ways things can possibly be”. The region corresponding to Mixins “crosscut” these two macro-regions, given that mixins capture semi-rigid types, i.e., types are necessarily instantiated by some instances and contingently (possibly) instantiated by other instances. In the bottom part of the canvas, we represent relationship kinds and subkinds.
The regions partitioning this canvas reflect the UFO categories of types that are the
most recurrent in models according to the OntoUML repository [
In this region of the canvas, we have all the rigid object sortals of the domain at hand.
As previously mentioned, kinds tessellate the universe of objects in the domain, i.e.,
kinds are exhaustive (all objects belong to a kind) and all kinds are mutually disjoint
(no object belongs to more than one kind). In this region, different kinds are
represented in a different color. The color variable is used here to trace the line of
inheritance of the principle of identity provided by a kind [
As figure 2 shows, we use indentation (tabs) to represent a subtyping relation. For
example, Man and Woman are subtypes of Person. In fact, these are subkinds of
Person, i.e., rigid subtypes of a kind that share the principle of identity provided by that
kind (hence the representation of subkinds using the same color of their respective
kinds). We assume that unless explicitly specified, all subtypes of a type defined in an
indentation scope (all those subkinds in subsequent lines that share the same
indentation) form a partition of that given type (i.e., they form a disjoint and complete
generalization set [
In order to visually differentiate the multiple regions of the canvas, we make use of
signs (indexes, actually, in the semiotic sense [
After considering the kinds of entities that exist in a given domain, one can move
on to consider what kinds of relationships can hold between the entities in that domain.
Here, once more, the different relationship kinds tessellate the space of possible
relationships and, again, we use different colors to trace the identity of different
relationship kinds and make use of indentation (tabs) and consistency of color to
represent subkinds of a relationship kind. In figure 2, we have two relationship kinds,
namely, Employment and Car Rental. One should remember that in the theory of
relationships in UFO, relationships are not n-uples but full-fledged endurants [
We use a “double ring/chain” sign to represent the bond or tie between the entities a relationship connects. Here as well, this choice is meant to address the requirement of perceptual immediacy. One should notice that relationship kinds are also rigid types (e.g., the particular employment of John in the United Nations is necessarily an employment relationship) that provide a uniform principle of identity for their instances. However, the use of the “chain” sign here is meant to highlight the primary ontological status of relationships as being existentially dependent on the entities they bind (mediate, relate).
Both roles and rolemixins are generically relationally dependent types [
In this region, we make use of the color scheme of types to indicate which kinds of
entities play a particular role. In other words, as formally proved in [
Since role(mixin)s are always defined in the scope of a relationship type, we use a
circle with the corresponding color of a relationship kind to bind a role(mixin) to that
relationship kind. For example, in Figure 2, we have that an Employment relationship
kind defines two roles, namely, Employer and Employee, whilst the Car Rental
relationship kind defines three role(mixin)s, namely, Lessor, Renter and Rental Car
(each of which is directly visually mapped to the kinds of entities that can play those
roles). Once more, we use visual variables to afford efficient parallel processing of
visual queries (e.g., inspecting a canvas for the role involved in a relationship and the
kinds of entities that can play them). The adjacency relation between a circle filled with
the color of relationship kind and a role(mixin) in this partition represents the
typelevel relation of mediation [
As previously discussed, role(mixin)s are relationally dependent types. As
consequence, the minimum cardinality constraint from the role to the relationship type
is at least one (and typically exactly 1) and the maximum cardinality constraint is
typically many (* in (Onto)UML). For this reason, we adopt these as the default
interpretation of the cardinality constraint in the mediation relation between a
role(mixin) and the corresponding relationship type (for example, in figure 2, we
should read that “every Employer is mediated by (participate in) one-to-many
employments”). Analogously, since relationships are existentially dependent entities
[
Whenever cardinality constraints in this mediation relation deviates from these
default values, we represent the deviating constraint the same identity color scheme to
represent the direction of the constraint plus brackets (“[ ]”) to represent the values
imposed by the constraint itself. For instance, in figure 2, we represent that a Rental
Car can participate in exactly one Car Rental6. Since the cardinality constraint, in this
case, refers to the association end connected to the relationship kind (Car Rental), we
use the color of the latter also for the constraint. Following the UML notation, we have
cardinality constraints represented in the form [a..b] (where b ≥ a) and that the most
typical combinations of value are [1..1] ([
Once more, we emphasize these cardinality constraints are only represented when they deviate from the default values. Although the instances of a role are always of a unique kind (hence, we have the role name represented in the color of the kind at hand), these roles at times do not subtype these kinds directly but a particular set of subtypes of that kind. In figure 2, for instance, an Employer must be not only an Organization but, more specifically, an Active Organization (and the same for Corporate Renter); an Employee must not only be a Person but an Adult and a Living Person (and the same for Personal Renter). We use parenthesis to the right of type name to represent the direct supertype of that type (within the specialization taxonomy of a given kind). In these examples from Figure 2, these direct supertypes are not subkinds of the kind of entities that play that role, but phases of that kind (see section 2.4).
Finally, we use a “Greek theater masks” sign to mark the role(mixin) region. This is meant to evoke the idea of roles as context dependent and contingent (as roles in a play are). Again, this visual representation choice addresses perceptual immediacy and discriminability.
Like roles and rolemixins, phases are anti-rigid (dynamic classification) types [
Finally, we make use of a sign showing “phases of the moon” to mark this region. This is meant to evoke the idea that being in a phase is a contingent property for entities of that kind but also the idea that phases are associated to other complementary phases (forming a phase partition).
Non-sortals are types that classify entities of multiple kinds. These include the socalled semi-rigid non-sortals called simply Mixins describing properties that are essential to some instances and accidental to others and rigid non-sortals called Categories. An example of the latter can be found in Figure 2: Physical Entity classifies both entities of the kind Person and entities of the kind Car. In this figure, we also have an example of the former, namely, Insurable Item, which can be thought to be essential for Cars but accidental for People (only living people are insurable) as well as for Organizations (only Active Organizations are insurable).
Contrary to subtype partitions in other regions, in the case of these partitions in the non-sortal region, we assume that, by default, these are disjoint (since all kinds are mutually disjoint) but not complete (e.g., there are of course other Physical Entities that are neither cars nor persons).
We believe that representing relationally independent non-sortals separately from
rolemixins is justified by a number of aspects that we have learned from our experience
and other people’s experience in using OntoUML. The first of these reasons has to do
with relational dependence. Because of this meta-property, rolemixins only appear in
models connected to relationships and this justifies their modeling in the relationally
dependent types region of the canvas. The second reason (related to the first) is that,
methodologically speaking, rolemixins appear in models in a modeling stage that is
much closer to how roles are modeled than to how other non-sortals are modeled.
Basically, rolemixins typically firstly appear in models considered as roles, i.e.,
dynamic types connected to a relational context. Then, in a second stage, one realizes
that these alleged “roles” could not be easily connected to a taxonomy of kinds. This is
exactly because they can be played by entities of multiple kinds, i.e., they are in fact
rolemixins. This modeling problem is so common that it is actually described as a
problem pattern in [
The sign we propose for the mixin region is a Venn diagram, in order to convey the notion that mixins aggregate intersections of multiple kinds. For the category region, we chose a hierarchy symbol, which conveys the abstraction/refactoring mechanism supplied by categories.
We show below an OntoUML model that is equivalent to the model of figure 2. This model can be automatically generated from the model of figure 2. As such, this OntoUML can clearly be taken as an alternative visualization of the original instantiation of the corresponding OMC. However, due to use of a standardized topology for the regions of the canvas, use of colors and default readings (e.g., cardinality constraints and generalization sets), we believe that the canvas can provide for an interesting alternative concrete representation for novice or non-technical users.
As a second illustrative example of an instantiation of the Ontology Model Canvas,
we propose the model of figure 4. This second example revisits the more complex
modeling case presented in [
7The only different between the model of Fig.4 and the one in [
In this case, we have three kinds of objects (Persons, Organizations and Projects) in this domain and three kinds of relationships involving them (Enrollment, Employment and Project Assignment). An Enrollment is a relationship kind mediating Students (also of the kind Person) and Universities (of the Organization subkind University).
In Figure 4, we can also see that Employment is a relationship binding Employers (which are Organizations) and Employees (who are of the kind Person). The model also shows that we have a particular subkind of Employment that is a University Employment. The role University Employee is a subtype of Employee, like for all Employees, this is a role played by people. Moreover, like all Employees, University Employees are mediated by Employment relations. However, University Employees are mediated by a particular subkind of Employment, namely, University Employment. Since a University Employee is an Employee, the default interpretation of the mediation relation between Employment and University Employee of a sub-relation of the mediation relation between Employment and Employee8. Previously, we have used parenthesis to indicate a direct subtyping constraint (e.g., Personal Renter is an Adult, Living Person). We here also use parenthesis to capture a subtyping constraint on the specialization of this mediation relation, i.e., the model captures that University Employees are mediated not by the general relation with Employment but by a specialization of that relation with a subkind of Employment, namely, University Employment. Analogously, the model in the canvas shows that a University Employer: is an Organization of a particular type, namely, of the subkind University; that it is mediated by a particular subkind of Employment, namely, University Employment.
Finally, we have that Project Assignment is a relationship involving a Project and a Project Party. Projects can exist without being associated to any Project Assignment (the zero-to-many [*] cardinality constraint on the association end connected to the latter that deviates from the default constraints for mediation relations) and Research Projects (a subkind of Project) is (again, optionally) mediated by a particular subkind of assignment (Project Assignments). Project Party is a rolemixin and, hence, it should be specialized into a partition of roles that assign different kinds to its instances. Here, we have that a Project Party can be either a Person Party (i.e., of the kind Person) or an Organization Party (i.e., of the kind Organization). As we previously mentioned, the default interpretation of subtypes that appear in a sequence of indentation is that they form a partition. For this reason, we have to explicitly represent when this is not case in a model. In this example, it is not the case that all the subtypes of Project Party are disjoint, in fact, (a) Research Student and (b) Researcher are Person Parties, and (c) Research Involved Organizations are Organization Parties in a project. The types (a-c) are, in turn, subtypes of Research Project Party (a rolemixin), which are Project Parties that are constrained to be associated in that role to a subkind of Project Assignment (Research Project Assignment). Finally, we would like to still represent that a subset of the subtypes of Project Party indeed form a partition of that type, namely, that Person Party and Organization Party tessellate the role Project Party. This is explicitly represented in the canvas notation by the use of braces (“{”) around the subtypes forming a partition.
In this paper, we elaborate on a first attempt to conceive an ontology modeling representation mechanism targeted at novice and non-technical users. This mechanism should, on one hand, have enough expressivity to represent important ontological distinctions that must be explicit represented in complex domains and, on the other hand, remain intuitive and relatively easy to learn to this class of users.
In order to target the first requirement, we ground our proposal on the Unified
Foundational Ontology (UFO) as well as on the accumulated experience of applying
the UFO-based conceptual modeling language OntoUML (and its associated
methodological and computational tools) in a number of projects in a diversity of
8 One should notice that here we are simply assuming that there is an inclusion constraint between sub
and super-relations, i.e., that the instances of the sub-relation are instances of the super-relation. We are not,
however, discriminating if these relations between relations are ones of subsetting, association specialization
or redefinition [
What we presented here is just a first sketch of the idea of an Ontology Model Canvas (OMC). Everything else remains to be done for this representation mechanism, including: extend it to represent other UFO categories (e.g., intrinsic properties, events, parthood), provide a formal definition of abstract syntax and semantics; build proper computational tool support and, perhaps, more importantly, to empirically validate this idea. After all, whether this representation mechanism is indeed easy to learn and employ by users is clearly an empirical question. We shall address all these issues in future work.
Acknowledgements. We thank João Paulo Almeida, Luiz Olavo Bonino, Ricardo Falbo and Renata Guizzardi for feedback on an earlier version of this paper.