Designing and Communicating Ontologies Visually

Tracking #: 2515-3729

Gilbert Paquette
Michel Héon

Responsible editor: 
Aldo Gangemi

Submission type: 
Full Paper
In this paper we discuss ways to support the Ontology Engineering process by providing a Visual Language for OWL 2 ontologies. We examine eight proposals for an ontology visual language stemming from Semantic Web re-search, including our own Visual Ontology Language that has evolved from our MOT semi-formal visual language. The MOT-OWL visual language implements the visual typing of ontology entities and links, and also the use of polysemy between these elements to increase the readability and manageability of the visual models. Then we compare this visu-al notation and other Visual Ontology Languages using principles from the Physics of Notation Theory. This compari-son helps us identify improvements that are being implemented in our more recent visual language, G-OWL, based on a systematic meta-modeling effort.
Full PDF Version: 

Major Revision

Solicited Reviews:
Click to Expand/Collapse
Review #1
Anonymous submitted on 29/Oct/2020
Review Comment:

Summary of the paper and main contributions

The present submission discusses some visual languages for ontology modeling, and in particular proposes an evaluation thereof, mainly based on nine criteria, adapted from the Physics of Notation Theory to the specificity of visual ontology languages. These criteria range from Completeness and Formality, which, roughly speaking, state that there should be a clear correspondence between objects of the W3C standard OWL2 and the visual symbols used in the graphical language (and vice-versa), to Perceptual Clarity (symbols should be clearly distinguishable with one another), Semantic Transparency (symbol appearance should suggest its meaning), Complexity Management (ability to deal with large ontologies), Totally Visual (no text needed to complement graphics), Parsimony/Polymorphism (use of limited amount of symbols), Cognitive Fit (fitting to the target users), and existence of an Editing Tool able to convert the graphical representation in the standard OWL2 syntaxes. Two additional criteria are also considered by the authors in their evaluation, i.e., the existence of a field experimentation and of a metamodel for the language.

The languages analyzed in the paper are:
- the MOT-OWL visual language (a previous proposal from one of the author of the present submission), which is described a bit more in details with respect to the other languages;
- UML-based visual languages, such as the Ontology Definition Metamodel and OWLGrEd;
- Non-UML visual Languages, such as GrOWL, the Graffoo notation, VOWL, Graphol, and G-OWL, an evolution of MOT-OWL proposed by the authors of the present submission.

Each language is evaluated with respect to the 11 mentioned criteria (the comparisons of the various languages are summarized in Table 1 in the paper). The authors finally present some guidelines on how to improve their G-OWL language, which turned out to be the most promising notation according to study presented in this submission.

Overall Evaluation

It is unquestionable in my opinion that a visual notation for ontology representation is a crucial tool for both ontology modeling and its understanding by domain experts or other users that are not specialized ontologists. At the same time, currently there is no standard visual language for ontologies, nor a graphical notation that has proved to be more suited than others or that succeeded in becoming more popular and widely adopted. Thus, the topic studied in this paper is certainly important for the community, and the paper is relevant for the semantic web journal. I also appreciate the idea of trying to evaluate languages according to the above mentioned criteria (in particular the ones adapted from the Physics of Notation Theory). I think that such criteria are very reasonable, and I agree that graphical languages for ontologies should satisfy them as much as possible.

Nonetheless, I found several weaknesses in the paper, which in the end provides a limited contribution to the field. The main reason in my opinion is because the assessments given by the authors are not always based on clear evidences, but are often presented through non-properly supported claims. I think that requirements like semantic transparency or cognitive fit (but not only these) should be analysed through a user evaluation study, which is missing in this submission. Also, the level at which some of the languages are presented (and the lack of adequate bibliographic references, such as for G-OWL) makes it difficult to assess Completeness and Formality of these languages. As the same authors say, the table summarizing the comparisons among visual notations is simply a "brief evaluation...based on the analysis of one or two documents for each notation". I found this a bit disappointing, in particular because the reader has no idea about how complex these documents are, and whether they are significative enough to highlight the main peculiarities of each language, and allow for a fair comparison among languages. Then, I am not sure that some of the diagrams in different languages presented in the paper as equivalent are indeed equivalent (see also the specific comments below). It is also a bit disappointing that the paper does not provide a single running example ontology, described according to all the notations investigated in this paper. This is done only for some cases.

Specific Comments

- When presenting the nine principles in Section 1.3, the authors should also mention which ones can be measured and how, and for the others you should explicitly say how they can be evaluated. There is something on this point only for Semantic Transparency, which is associated to a scale ranging from -1 to +1, but how this can be established is not discussed. Nothing in this respect is said for the for the other requirements. What about, e.g., for Perceptual Clarity or Cognitive Fit (to name a few)? How can it be said that a language is better than another with respect to these criteria?

- From the bibliographic references provided [29,30 or also 31], it is difficult to evaluate the real usage in the community of the MOT-OWL notation for representing ontologies. This language is around from a while, but it seems to me that it has been limitedly used in applications, and the present submission does not really provide enough new elements that should lead to raise interest on it (the paper does not even provide indication on how to download or access the GMOT-OWL editor). Which is the novelty here?

- I found the presentation of the GMOT-OWL confused, and some of the authors' claims make me wonder whether this language has really the ability of capturing OWL2, or even OWL (I mean, the ability of expressing every OWL ontology). Below I list some unclear aspects and problems that I found:

* which is the meaning of equality/inequality symbols connected to classes?
* the group in the right bottom corner of the picture is not explained. What does it represent?
* The meaning of the label R is unclear: there are several edges in a universal, existential or cardinality restriction, with different roles, but all with the label R. Why? This R stands for "ruled by", according to what the authors say, but the meaning of this for each edge is obscure.
* In a universal or existential or cardinality restriction, it seems that the class that is restricted (as Parent, in Figure 2) is necessarily equivalent to the class that is denoted by the restriction (and it is really weird that the "Equi" label is not used here). This is not always the case in OWL. How can a containment between such classes be expressed? This ability is crucial to specify, for example, the typing of a relation, or the mandatory participation of instances of a class in a relation (and not necessarily these two properties together).
* related to the same example (Figure 2): the authors say "the direction of the R links show “Parent” as a domain of “Property1” and “Person” as its range". However, the OWL translation (as well as the First order logic translation) of the diagram is not saying this. According to the OWL piece of code, there can be individuals occurring in the domain of hasChild that are not instances of Parent (it is sufficient that the filler of the role is not a Person). Similarly for the range. So, what exactly is represented in Figure 2? (BTW, “Property1” should be "hasChild")
* If a class is both the union and the intersection of other classes, must it be repeated twice in the diagram?

Some of the above considerations apply also to G-OWL (which is presented as a sort of evolution of GMOT-OWL). The paper does not present the syntax for this second language, so it is really difficult to make a clear assessment. In particular, the claim that G-OWL respects Formality and Completeness is not supported by adequate explanation, nor there is a bibliographic reference where this is proved.

- Figures are not clearly readable throughout the paper. In particular Fig. 1, 3, 4, 5, 7, 8, 12 and 13 should be enlarged and/or should have a better quality. Figure 11 remains unreadable also looking at the pdf (where it can be zoomed).

- Section 3.3: the authors consider the object-oriented nature of UML (which, for example, allows for the presence of methods in classes) as an obstacle for its usage for ontology representation, since ontology interpretations are based on set theory. I disagree with this statement. Methods in UML classes can be simply ignored. Then, classes in UML can be interpreted as set theoretic. For a formal correspondence between UML and set theory the authors can have a look at:

Daniela Berardi, Diego Calvanese, Giuseppe De Giacomo:
Reasoning on UML class diagrams. Artif. Intell. 168(1-2): 70-118 (2005)

- Below, as examples of what I mean when I say "non-properly supported claims"", I list some of such claims:
* page 10, referring to the Graffoo notation: "The shape and colors are a bit too similar to promote Perceptual Clarity"
* page 11: "Graphol respects well the Parsimony Principle"
* page 11, again referring to Graphol:" Metahuman is **understood more directly** in figure 14 [MOT-OWL representation] as the intersection of the sets of all Human and the set of persons (anonymous class) that have at least one ability of a Superpower."
* page 13: "MOT-OWL and G-OWL...have a Good Cognitive Fit"

How did they authors arrive to these conclusions? More in general, it remains unclear how several of the values given in Table 1 have been established.

- Comments on the Graphol diagram in Figure 12: "a Vilain is a class for which there exists an archenemy that is a super hero." This is not correct. There is no mandatory participation for the instances of the class Vilain in the is_archenemy_of relation (notice that there is no arrow from Vilain to the domain of is_archenemy_of). Thus, there can be villains without archenemies. This error is reflected in the representation of the ontology in MOT-OWL (Figure 13). More in general it seems to me that the authors do not distinguish between typing of roles and mandatory participation in their ontologies. Thus I am not sure the diagram in Figure 12 exactly corresponds to the Graphol diagram in Figure 12 (this comment is also connected to my previous observation on the MOT-OWL notation). Thus, it does not turn out clearly that in this case the MOT-OWL model is more easy to read with respect to the Graphol one, in particular because the former uses less nodes (as the authors affirm). As said, I doubts the two models are equivalent, and my guess is that to have the equivalence the number of nodes in the MOT-OWL diagrams should be incremented. BTW, in their current versions, the Graphol diagram has 26 nodes, the MOT-OWL 2 has 24 nodes. So nothing can be inferred by this situation even by ignoring the possible error.

Minor issues and Typos

- page 1: "OWL ontologies are schema that can process" I am not sure that process is the right term here. Do you mean that OWL ontologies can be expressed through RDF triples?

- page 2: "separating class, properties,..." -> "separating classes, properties,..."

- "A comparative survey of many other tools". Reference [9] is quite outdated (even in its 2004 revised version) and should be complemented with some more recent material.

- page 3: "for the use on ontologies " -> "for the use of ontologies "

- page 3: "has been used evaluate" -> "has been used to evaluate"

- page 5 "have at least one value in class "Person" by the property "hasChild" " -> please rephrase.

- page 5: "all of the individual" -> "all of the individuals"

- page 5: owl code in Fig. 2: there are unclosed tags

- page 5: "F,I,R,S for properties" -> "F,I,T,S for properties" (?)

- page 6: "in in overload" -> "in overload"

- page 7: in the last paragraphs of Section 2, Figure 4 and Figure 5 should be instead Figure 3 and Figure 4, respectively. Notice that references to the figures are wrong also in other points of the paper.

- page 7: "those Azaleas having exactly" -> "those individuals having exactly" notice that limitedly to Figure 6, a SingleColoredAzalea is not necessarily an Azalea.

- page 10: "are displayed to together" -> "are displayed together"

- page 11: "Shown on example of figure 12 are other kinds..." -> "Shown on example of Figure 12 there are other kinds..."

- page 11: "text like...are the type of restriction" please rephrase

- page 11: " a class for which there exist an archenemy..." -> " a class for which there exists an archenemy..."

- page 13: "To each semantic OWL object correspond a unique symbol...." wrong English sentence construction

- page 13: "one-t-one" -> "one-to-one"

- page 13: "Onto-Case4G-OWL Eclipse-based editor" Reference needed

- page 14: "can be linke to" -> "can be linked to"

- page 15: "can be specify" -> "can be specified"

Review #2
By Robert Pergl submitted on 05/May/2021
Minor Revision
Review Comment:

The authors' goal is to design a visual language for ontology engineering. Although this is not a completely novel idea, the authors argue based on a solid review and sound insights that there are significant gaps in today's state of the art of visual approaches to ontology engineering, while presenting a strong motivation to address them.

Methodologically, the effort is grounded in the Physics of Notation Theory, an established framework for evaluating, comparing, improving and designing visual notations. The MOT language is carefully designed with respect to its principles. Important additional aspects are taken into account, such as complexity management and modularisation of large ontologies.

The paper is well-written with a good balance of rigorous reasoning and explanatory narrative.

The designed language is purely graphical and I find it a great design choice to implement a bi-directional approach in the GMOT-OWL tool, which reportedly "provides translation back and forth to the textual OWL standard", as I agree with the authors that the preference of visual vs. textual representation varies among the users.

The language is then extensively and deeply compared with other approaches.

It would be informative for the interested readers if the authors mention under which license the tool is developed and a link to it would be nice.

Minor findings to help improve the paper:

p. 3: ...has been used TO evaluate...
p. 4: ...section2... -> ...section 2...
p. 4: ...Sections 3 will discuss... -> ...Section 3...
p. 4: ...actually in publication... -> ...currently...
p. 6: in...
p. 7, 3.1: ...Figure 4... -> ...Figure 5...
p. 7: hyphenation of unnecessary
p. 8: ...construct classes using Boolean operations... -> ...set operations...
p. 9: open-source editors for graphs... -> ...editor...
p. 11: hyphenation of modularization
p. 12: ...Figure 15... -> ...Figure 15...
p. 15: hyphenation of metamodels
p. 15: ...can be specify... -> ...specified...

Review #3
Anonymous submitted on 24/Aug/2021
Major Revision
Review Comment:

This paper presents a visual language for owl2 ontologies.
The authors compare eight proposals for visual languages in the context of the Semantic Web, including their proposals (MOT-OWL and G-OWL).
The comparison is based on the Physics of Notation Theory principles.
The eight visual models and their symbols are introduced using example visualizations.
The comparison of the eight notations is provided using a table, continued by a discussion of MOT-OWL and G-OWL in correspondence to the evaluation criteria.

The strengths of the proposed visual language are its total visibility of owl semantics and complexity management, allowing to reduce the number of visual primitives for large ontologies.

* Methodological application of PoNT to visual ontology languages for the evaluation criteria
* A visual language for large ontologies
* Incorporation of advanced visual methods for handling cognitive load (sub-model visualizations)

* The authors state that the research goal is to provide a completely visual language that exports to textual languages such as Turtle. However, this is a general requirement when visual modeling is addressed. From a research perspective, a formal bijective mapping definition is missing.

* Wrong figure references in the text make it very difficult to follow and always require double-checking.

* Missing formal description of how sub-models are generated
* Unclear research contribution, is it the comparison or the visual language (G-OWL)?

Detailed Review:
The authors start with a general introduction to the advantages of the Semantic Web for machine-readable representations of knowledge and the inherent difficulties for designing such semantically enriched structures.
Furthermore, the authors point out the need for user-friendly interfaces and visualization methods to support navigation, exploration, and understanding of ontologies.
This need for applications is grounded in the textual representation of ontologies. This textual representation is a linear representation of the data, which becomes more challenging to understand with growing size and complexity.

The authors then provide the specificity of visual ontology languages by applying the 9 PoNT principles to ontology visualizations are regroup them for the underlying purpose.

After defining the nine visual ontology paradigms, the authors continue with the definition of the individual visual languages and their graphical symbols (ODM, OWLGrEd, GrOWL Graffoo, VOWL, Graphol, MOT-OWL, and G-OWL). The authors then provide the comparison of the individual visual languages using a table and provide a discussion on the evaluation w.r.t MOT/G-OWL.

"Most of the new OWL2 features have been implemented in..."
> Which features are not implemented here?
Furthermore, this discussion focuses only on MOT and G-OWL neglecting the remaining visual languages.

Typos, Suggestions, Style, Other:

Grafoo in Table 1 should be Graffoo.

not to implement a strict one-t-one correspondence
> one-to-one

"First, we have to make an important distinction between two main orientations in the use of visual ontologies."
>What are visual ontologies? Do you mean "in the use of visualization for ontologies"?

In section2, we first present our own MOT-OWL ontology
> missing space bar between section and 2
Sections 3 will discuss proposals...
> Section 3, < singular.

Style suggestion:
when referring to specific figures or sections (e.g., In section 2,) use a capital letter, i.e., "In Section 2," ...
or "In Figure x,"

"We believe that computer scientists will prefer..."
>Generally, I would avoid "believe" statements.
Additionally, my opinion is that ontology engineers will prefer Protege to create complex owl constructs and eventually validate the textual representation output.
However, this is an assumption that needs to be validated.

"This object is pasted from the main model on figure 4 to the sub-model of figure 5 to which related OWL entities and links can be added."
> I think Figure 4=3 and 5=4<

This sub-model feature is important to handle the cognitive load. Unfortunately, the authors don't provide a formal definition of what criteria are used to compute a sub-model.

"Figure 4 displays an example of an ODM model for part of an ontology using some of the complex OWL-DL elements like class intersection and property restrictions."
> Do you mean Figure 5 <

"Figure 6 presents a MOT-OWL visual diagram equivalent to the bottom part of figure 6".
> Figure 6 vs Figure 6 ?, Do you mean maybe
"Figure 6 presents .... part of Figure 5" ?

"we present on figure 14 an equivalent graph in MOT-OWL" -> we present in Figure 14 ...
Figure 14 is however the G-OWL entity relation model.
Do you mean Figure 13???

"a Metahuman is understood more directly in figure 14 as the intersection of the sets of all Human... "
> Still, Figure 14 is the entity relation diagram; do you mean figure 13?

"We beleive that these improvements will increase the Parsimony of the notation and facilitate the complexity management of large ontologies."
> beleive -> believe
> Again believe statement; this needs to be validated or discussed in more detail.

"It corresponds to the ontologies for a science-fiction play presented earlier on figure 13(in Graphol) and figure 14(in MOT-OWL)."
> in figure x.
Again wrong figure number (12 us graphol and 13 is the Mot),
These wrong figure reference numbers make it very hard to follow the argumentation.

"The visual model on figure 17 requires fewer links "
> There is no figure 17 at all; do you mean figure 16?

Fig. 16.An equivalent G-OWL model to figure 13,14.
>> Again, wrong numbers; it is 12 and 13.

VOWL issues:
1) Figure 11 is displaying the protege plugin of VOWL and NOT WebVOWL.
2) The size of the circles is not determined by a node degree but can be scaled by the number of individuals assigned to the class (vowl spec 2)
3) "Information like disjointness or properties or types of properties like transitivity or symmetry are not displayed visually but listed in the sidebar."
>Partially true, disjointness is visuall represented
4) VOWL is not a totally visual modeling tool
> VOWL is a notation, not a tool.
WebVOWL has a visual editing functionality. However, I agree on the aspect that VOWL is not totally visual w.r.t., it does not display individuals themself and provides additional information in the sidebar.

Property types are visually displayed as annotations (small text under the label of a property)

Clarity suggestions:
When you refer to parts of the figure in the free text (e.g., in the lower part of figure x), I would suggest making a red box around the area you are talking about and give it a number. So it makes it more clear which part you mean directly.

Question w.r.t to the visual notations.
As far as I understand it, the blue dot on a class means that there is a sub-model behind it. But it is also indicated using the submodel sign in the top right corner.
So what is the blue dot mean on a class and on a property?

Overall suggestion:
It appears the document was written in word.
I would suggest using LaTeX to avoid reference issues when using figures. Additionally, you can use non-line-breaking spaces to avoid such cases line on page 10 (right side) "as shown in figure 11" -> as shown in Figure~\ref{fig:imageLabel}.

Final Notes:

Visualizing ontologies is an important aspect. MOT / G-OWL provide totally visual representations with additional cognitive handling for large ontologies.
Personally, I like these visual notations and expect them to be useful for a large audience.
It is not fully clear what the research contribution is: the comparison or G-OWL?
If this work addresses G-OWL, how does it differ from the previous work [29]? What are the extensions provided in this work?