Foundational Ontologies meet Ontology Matching: A Survey

Tracking #: 2520-3734

Daniela Schmidt
Giancarlo Guizzardi
Adam Pease
Cassia Trojahn dos Santos
Renata Vieira 1

Responsible editor: 
Jérôme Euzenat

Submission type: 
Survey Article
Ontology matching is a research area aiming at finding ways to make different ontologies interoperable. Solutions to the problem have been proposed from many disciplines, including databases, natural language processing, and machine learning. The role of foundational ontologies for ontology matching is an important one, it is multifaceted and with room for development. This paper presents an overview of the different tasks involved in ontology matching that consider foundational ontologies. We discuss the weaknesses of existing proposals and highlight the challenges to be addressed in the future.
Full PDF Version: 

Major Revision

Solicited Reviews:
Click to Expand/Collapse
Review #1
Anonymous submitted on 07/Aug/2020
Minor Revision
Review Comment:

The paper is an overview of the main foundational ontologies together with ontology matching approaches regarding them in different aspects. The kind of work is a survey, and, from this perspective, it has the clear intent to introduce the topic to the interested scholars at any level of expertise and it has also the summative intent to bring a current overview of the topic to the Semantic Web community in general. The purpose of the survey is made clear in the introduction: to provide a complete account of the role of foundational ontologies in ontology matching and to outline the weaknesses so far emerged and the possibility of future improvements.

I appreciated the accurate and almost complete reporting of the main studies, approaches and initiatives related to the topic. However, in virtue of the very well claimed and circumscribed purpose of this article, I see room for improvement, especially in the systematization and structuring of the paper contents over the different themes brought to the attention of the reader. Hope that the following remarks could help authors go in this direction:

As an overall remark, many interesting subtopics emerge from reading the paper, e.g.,: the lack of comparability studies for the different formal languages used to axiomatized foundational ontologies, the issues of diverse granularity, philosophical vision, lack of vocabulary translations, and all of the overlooked aspects in ontology matching such as linking lexicon to semantics and considering other than equivalence relations only. Since one of the intent of the paper is a _revamping_ of topics that have been touched ten years ago, more of these topics could be anticipated in the introduction, so as to put all of these lively themes in the spotlight and give a sudden taste of the lively and open questions of the field that are going to be treated in the paper. An important aspect that I see very clearly is the lack of a critical view of foundational ontologies, and this aspect should be at least touched in this paper.

Abstract: since the focus is the role of foundational ontologies in ontology matching, I wonder whether it only relates to the weaknesses of matching ontologies or is the scope wider and inclusive of more general aspects afflicting ontologies in general (see also my comment above). Some of them are outlined in the Discussion section. It may be worth anticipating the main considerations emerged from this survey also in the abstract.

left column
- rows 45-50 (text in italic) these few rows are very dense of concepts that are kept at the implicit level and hence may not help beginners catch important aspects related to them, e.g., why it is important to distinguish between class or individual entities? Is it feasible to introduce some (formal) definitions or some (brief) examples to clarify this core passage?
- Row 50-51 “Such distinctions are however key aspects in many applications in Artificial Intelligence”, make an example of provide references to support this claim.
- Right column: row 28 – add “of” to the fragment integration of domain ontologies
- Rows 40-41 and 50-51 please check the two sentences as it seems they are redundant.
- Rows 44-45 please replace the example, since Author is not equivalent to Writer, there are differences between the two.
- Page 2 – left column-rows 2-6: is it only a matter of finding subsumption relations or also of how upper ontologies express/formalizes upper concepts?
Rows 11-15: please provide a reference for this claim or claim the lack of comprehensive studies in this sense.
- Row 25 why only considering the weaknesses and not the strengths also?
And btw: is there room for a comparison of the approaches presented (to see what strengths and/or weaknesses they may have in common, which approaches are good for which of the tasks considered and the like)? If yes, this aspect should be announced here and discussed in the Discussion Section.

Section 2 – Foundational ontologies.
In general, I found that there is room for improvements and better systematization of the aspects presented for each ontology, besides a descriptive approach . For example, some words could be spent to briefly introduce what are the main philosophical views and / or lenses through which upper ontologies were designed / classified/ described.
A table may help summarize all of the information that should be given for any of the foundational ontologies presented, for example: year of design/deployment, number of entities, main structure, axiomatization language and technology of deployment, available formats, official repository url, basic structural information (how more abstract concepts link to more concrete ones), philosophical perspective, maintenance, versioning, backward compatibility, and the like.
Some passages should be clarified, for example:
- row 25 right column, speaking of DOLCE: “an ontology of particulars which adopts a descriptive approach with a clear cognitive bias”, What is this cognitive bias? In what sense it shows a cognitive bias?
- row 2 page 3 speaking of GFO: “considers basic distinctions between individuals.” In what sense this ontology consider them and what are differences with other foundational ontologies in this respect? Please clarify this passage

Also for Section 3 a table could be useful to synthesize the main issues or tentative matching between foundational ontologies, and possibly many results may converge in the summary Table proposed above.

Section 4 begins by introducing a very important and delicate matter: providing a formal semantics to the lexical layer of ontologies. Since this is a very crucial point, more about it should be explained in this Section, maybe more formally and / or by providing examples.
In this Section also the content seems redundant and it should be better systematized, for example, rows 27-31 and 35-42 are redundant and, in general, the fact that WordNet was used to be matched with many foundational ontologies may be said more concisely.
Row 44: please provide a reference for OntoClean, row 13 page 5 add an “f” to foundational

Section 5: please explain the passage in rows 24-26, what do you mean by “low coverage of foundational ontologies in domain ontologies”? This also seems to be a crucial passage.
Rows 41-48 LOM matcher: please clarify whether the four approaches are used with foundational ontologies, since as it is written is a little bit confusing.

Section 6
This Section could be better structured by explicitly providing two subsections "Manual matching" and "Automatic matching". Many reflections about the effectiveness of the two approaches, their differences in performance and some other aspects emerging from their comparison may be worth anticipated here and discussed in the Discussion Section. This seems to be another central point of the paper (see the discussion) so probably this alignment modalities should be systematized better inside the paper.

Section 7 – Discussion Section.
Here the main issues of matching foundational ontologies and domain ontologies are described. The topics should be introduced in a less discursive and more systematic ways of the kind: problem at hand/ solution or future challenge. In this way, they could result better emphasized and more complete.
Here I also see many other themes than ontology matching, for example: the compatibility/comparability issues, natural language definitions vs logical statement issues and, in general, criticalities that are found at other level of the “ontological layer” rather than when matching them. As already remarked, the question should be dealt with in this paper to provide a clearer and more comprehensive framing of the topic.
The Discussion Section needs to be more robust and convincing, maybe by systematizing all of the topics, relying on the approaches presented and commenting them, adding wide-scope criticalities, and better structuring all of those aspects while discussing them, and by giving a more precise perspective of future challenge. This may help better connect the overall contents of the paper and reach the purposes declared in the Abstract and Introduction, provided they are also more connected with the necessary anticipations of the topics treated and emerged from the analysis carried out.

Review #2
Anonymous submitted on 28/Sep/2020
Major Revision
Review Comment:

This paper introduces a survey of ontology matching in the context of foundational ontologies. The proposed paper presents a short list of foundational ontologies, such as BFO, DOLCE, etc. After that a very abstracted description of matching approaches considering foundational ontologies. Even the proposed topic is an interesting topic; however, the introduced content does not provide sufficient detail or an acceptable conceptualization of the topic, and fails to provide a deeper understanding of the topic. Furthermore, this survey misses general guidelines the survey should follow as well as the main research questions that will be addressed during the survey.
More detailed comments:
- The main issue of the introduction is the missing of description for motivation for this work as well as an acceptable conceptualization to achieve the motivation. It seems to the reader that the background and related work are collected in a random way.
-The classification of ontologies to foundational and domain ontologies could be acceptable, however, it should also refer to core ontologies (Ansgar Scherp, Carsten Saathoff, Thomas Franz, Steffen Staab: Designing core ontologies. Appl. Ontology 6(3): 177-221 (2011))
Foundational ontologies:
- A short list of foundational ontologies has been introduced, without referring to any selection criteria, why only this list was selected?
-Missing important information, such as the current version of each ontology, exact number of classes (concepts), properties which helps understanding the matching problem
-The usage of different terminologies to describe the same entity. For example, for BFO “It has 34 terms”, for DOLCE, “includes roughly 100 terms”, but for PROTON, “It contains about 500 classes and 150 properties”, so, what is the relation between “term”, “classes”, and “properties”.
- Here, I have two recommendations: first, it would be good if a new section is added to introduce all these used terminologies through the paper to avoid such a situation. Second, it would be helpful if an example is added considering an entity which is defined differently in some of the mentioned ontologies to motivate the matching problem.

Sections 3-6:
- These four sections use the same style to present matching considering foundational ontologies (between foundational ontology; between foundational ontologies and wordnet; matching domain ontologies via foundational ontologies; and between domain and foundational ontologies). Actually, the introduced content of these sections is very weak and poorly described. It is hard to understand what the main goal of these sections is. Several issues and dimensions should be considered, especially when aligning foundational ontologies. For example, some of foundational ontologies are NOT frequently updated (BFO last version in 2015, GFO last version in 2010) with a small number of terms. In such a scenario, which kinds of matching approaches are fitting? Maybe manual alignment?
- It would be good if a use case is mentioned in each section? E.g. why we need to match between foundational ontologies and domain ontologies?
- A missing reference in Page 4?
- Why section 4 is mainly limited to match between foundational ontologies and WordNet?
- Even a recent work from the same authors has discussed matching between a number of foundational ontologies using existing matching task, however, it has not been included in this paper ( Daniela Schmidt, Cássia Trojahn, Renata Vieira: Matching BFO, DOLCE, GFO and SUMO: an Evaluation of OAEI 2018 Matching Systems. ONTOBRAS 2019)

Section 7: I see it would be good to first split it into four subsection discussing main issues within each case, and then a final subsection to gather and summarize

Review #3
By Michael.Uschold submitted on 08/Oct/2020
Minor Revision
Review Comment:

• There is an enormous body of work discussed and referenced. The overall structure of the paper is useful. Several of the references were of great interest to me, and I read them right away.
• The discussion section (7) at the end is very good.
• The work is competently described with many very good observations being made throughout the paper

• The problem the paper is supposed to be addressing is not clearly articulated. A lot of interesting things are raised w/o clear direction and conclusion. The reader is left w/o a clear answer to the question: “so what?”.
• Too few examples. Add a few more in key places, give ample details to clarify things.
• Most of the references are quite a few years ago
• Missing: UMBEL. Not necessary to include, but mention it anyway and indicate why not included.

• The authors acknowledge that there has been limited progress in the automated matching. It would be useful to explore this in more depth. Why has there been so little progress? What are the limiting factors? What avenues of research are looking promising? Do we think there will be breakthroughs before we have commercial fusion reactors?

Specific criteria:
(1) Suitability as introductory text, targeted at researchers, PhD students, or practitioners, to get started on the covered topic.
Comment: highly suitable, after clarifying a few things per my feedback.
(2) How comprehensive and how balanced is the presentation and coverage.
Comment: very comprehensive, reasonably balanced. More emphasis on commercial could help.
(3) Readability and clarity of the presentation.
Comment: generally good
(4) Importance of the covered material to the broader Semantic Web community.
Comment: if the problem could be cracked, it would be very important. So an update on where we are now is useful.

Px Ly Cz means Page x, Line y, column z

P1 L29-33 C2: what you are calling top-down vs. bottom up is not about up or down or more or less abstraction or level of detail. It is about WHEN the foundational ontology is used: it is either DURING or AFTER the domain ontology is built.

P2: why include OpenCyc if it is no longer being maintained?

P2 C1 P14-16: give examples of logical inconsistencies found.

P2 C1 P265-27 typo “of the” -> “of”
After reading sections 1-3, there is a lot of material, which for any give few sentences or paragraphs is interesting. However, none of it ties together and I have no idea where it is all going and why it is important to go there. There nothing to chew on, no punch line.

P4 C2 L32-35: it says that: “They argue that many NLP applications need to deal with events, actions,
states, processes, and other temporal entities that are usually represented by verbs.”
Are they correct? Why or why not?

P4 C2 L44: missing reference for OntoClean.
From personal conversation with Adam Pease, there was a lot of effort mapping SUMO to Wordnet. It seemed like the right things to do, but hardly anyone made use of it. Comment on the utility of mapping to Wordnet, in practice.

P2 C1 first paragraphs: very interesting discussion. Missing reference to DOLCE Zero.
Typo: “oundational” needs and ‘f’.

P5 C2 L35: How much did the F-measure increase? Is it significant? Did it make any difference in practice?

P5 C2 L41-42: “Very few concrete matching approaches however exploit foundational ontologies”. Why not? Should they? Is this worth researching?

P6: end of section 5. This is not at all clear. Please add some example to illustrate.
WE are told that: “LOM takes the source terms that are unmatched with the three first methods, collects the set of SUMO terms that their synsets map to, and then compares the SUMO term sets to their counterpart for each term in the target ontology.”
Tell us whether it worked, why or why not?

Please update the reference to the gist upper enterprise ontology ( [69]) to include the web site:
It is probably to late now, but it for the portion of the readership that is interested in applied ontology (as distinct from academic research), it would have been useful to have a separate section on gist. Gist is fairly unique among upper ontologies in that 1) it has been used in dozens of client projects 2) it is in production in a variety of actively evolving and 3) it is being actively developed by an open user community.
I have been using and developing gist for 10 years now, so I speak from experience. But of course, as an employee of Semantic Arts, I am also biased.

Separate from gist, it would be useful to indicate to what extent any of the foundational ontologies have been used in commercial deployments.