This study introduces a tool and methodology for the creation of harmonized networks of ontologies, a precondition for the full exploitation of data in federated distributed systems. Bridge-Concepts are designed to alleviate well-known challenges in ontology mapping and to address network-specific issues, such as scalability and consistency. As standalone ontology entities, they function as data pipelines in hub-and-spoke structures. Designed with FAIR (Findable, Accessible, Interoperable, and Reusable) principles in mind, their rich informal characterization makes them user-friendly interfaces and candidates for a vocabulary tailored specifically for ontology usage. Bridge-Concepts form the central element of a network-specific alignment methodology based on pragmatic criteria that can be further improved by introducing high-level ontologies and automatic tools in the loop. This approach is based on an analysis of the limits of meaning-encoding within semantic artifacts.
Formal ontologies are one of the core knowledge representation technologies and the fundamental
infrastructure for the Semantic Web; however, their practical efectiveness in supporting interoperability
is impeded by the existence of a plurality of frameworks –even with overlapping, or equivalent,
domains of application. Not only is there a prevailing inclination among industrial stakeholders to
prefer ontologies developed internally (to exert greater control over proprietary data), but diferent
ontological frameworks can exhibit varying degrees of suitability with respect to specific pragmatical
goals, making a pluralistic approach actually desirable, especially in industrial contexts [
This study introduces a methodology and related tools (referred to as “Bridge-Concepts”) to ensure the comparability of core ontology entities employed by diferent ontologies in order to set up a FAIR-compliant network of partially aligned, harmonized ontologies. These minimal, pinpointed data pipelines are meant to support efective integration and interoperability among a plurality of knowledge bases. The discussion will proceed as follows: Section 2 provides a short introduction to relevant issues concerning meaning-encoding in semantic artifacts (2.1), as well as a general framework for the evaluation of alignments (2.2) and an overview of issues specific to ontology networks (2.3). In Section 3, Bridge-Concepts are introduced, following Bridge-Concept templates’ structure. Section 3.1 elaborates on Bridge-Concepts’ formal role, explaining how they support mediated alignments among ontologies, and how they address heterogeneity, while Section 3.2 presents their role as a “ontologyspecific vocabulary”, promoting FAIR-ness and alleviating issues related to lack of documentation. Section 3.3 explains how a contained set of Bridge-Concepts can establish a controlled, open network, touching on points related to Bridge-Concept engineering, framework consistency, and the possibility of improving the system by exploiting High-Level, Foundational Ontologies, and automatic tools.
Tool and methodology were developed in the context of two European Projects, OntoCommons
(https://ontocommons.eu/) and OntoTrans (https://ontotrans.eu/), and are part of larger toolbox building
on the Linked Open Terms (LOT) approach [
Computational ontologies serve as tools for data structuring, integration and retrieval. They play a
preeminent role in knowledge representation by providing schemas for knowledge bases (e.g., knowledge
graphs), thus supporting interoperability and knowledge discovery, among other things. They can be
understood as a representation of certain systems or as a systematization of domains of discourse [
Per the orthodox definition, ontologies are formal, explicit specifications of shared conceptualizations
[
Hence conceptualizations have to be shared: for ontologies to be efectively employed, all users have to converge towards (approximately) the same intended interpretations of the ontology entities, and, most importantly, of the assumed primitives. The criticality is accentuated by the fact that the domain which should be the target of the interpretation is not universally accepted by all interpreters, i.e., it is an empirical fact that diferent interpreters, and even the same interpreter across diferent contexts or timeframes, may subscribe to slightly divergent worldviews; moreover, even without getting into tangled issues concerning meaning indeterminacy, vagueness and referential failure are well-known and widespread phenomena. It is pivotal to recognize that users needn’t share exactly the same concepts, or a worldview across the board: in the same way as communication through natural languages is not compromised by speakers’ idiosyncrasies and borderline cases, ontologies are efective insofar as they deal successfully with most cases, and the remaining ones cause no significant practical frictions.
Euzenat [
Terminological heterogeneity occurs due to variations in identifiers and labels among ontology entities which purportedly refer to the same world entities. Diferences pertaining to identifiers are the standard if ontologies are developed separately; diferences pertaining to the labels can be due to the use of diferent natural languages (“gatto” [IT]; “cat” [EN]), cases of synonymy (“coat”; “jacket”), or preferences of specific communities. The latter can be tackled through synset analysis and more complex techniques employed by language models; however, interpretation remains the gold standard due to polysemy, jargons, vague or misleading labeling and the salience of contextual variants in practical scenarios.
Semantic/conceptual heterogeneity has to do with formal idiosyncrasies in modeling a given domain. This may manifest as the utilization of diverse concepts, or diferent choices in axiomatization, and is closely related to points discussed in Sections 1 & 2.1. Semantic heterogeneity is for instance exemplified by geometric theories adopting diferent primitives, leading to entirely diferent axiomatizations. This is not problematic when the resulting theories are logically equivalent, as direct correspondences between the entities involved could be established. However, the inherent incompleteness of ontologies, coupled with their worldview/use/goal-specific nature, makes this an exception rather than the rule. In interesting cases, reliance on interpretation is thus necessary. Therefore, informal elements are crucial.
Finally, semiotic/pragmatic heterogeneity, comes down to idiosyncrasies in interpretation proper, for ontology entities which appear to be (terminologically and semantically) similar, given context/user variance. The most relevant cases involve discrepancies among “false friends”, which can, if undetected, have significant consequences given erroneous alignments. The issues related to the absence of a univocal interpretation discussed above are greatly magnified if ontologies are brought outside their context – a precondition for the establishment of interoperability.
All these types of heterogeneity should be addressed in ontology harmonization. However, to respect pluralism and to allow for non-disruptive integration, this can only be done indirectly. Terminological heterogeneity can be resolved by (not) establishing identities among individual constants and equivalences among classes or properties. Semantic heterogeneity can be addressed by setting up semantic connections among ontology entities. Semiotic heterogeneity can only be tackled by producing links that take into account actual usage in practice and other informal documentation. This arguably requires collaboration with the stakeholders employing the ontology, i.e., domain experts, and close scrutiny of related knowledge bases. An ideal alignment between two ontologies would be such that information shared through mappings is indistinguishable from re-conceptualizations of the relevant systems/domains, and any form of redundancy is avoided. Needless to say, ideal alignments are purely a theoretical limit. Discarding the possibility of working directly on knowledge bases, which would undermine the benefits of employing semantic technologies while still incurring the aforementioned interpretative issues, such alignments could only be established among (locally) equally expressive ontologies, i.e., already (locally) semantically and semiotically harmonized ontologies. Nevertheless, this limit can be used as a reference point to establish a metric of alignments’ informativeness and error, as well as qualitative, and potentially also quantitative, criteria to guide alignment choices.
For the sake of simplicity, and in line with the examples that will be discussed below to present Bridge-Concepts, let us focus on classes () and individual constants (). A set of mappings from a certain ontology 1 to an ontology 2 is maximally information-preserving if, as a result of the mapping, all the individual constants 1, 2, ..., P 1 would be classified under all the classes 1, 2, ..., P 2 the referred-to entities (given a hypothetical intended interpretation) would be independently conceptualized under. Informativeness can thus be understood in terms of a simple proportion. Similarly, error can be defined as the ratio of individual constants that are categorized under classes that are either disjoint from, or subclasses of, the most specific class under which the referredto entities would be conceptualized in the target ontology. Extrapolating from common practice in ontology harmonization, the ideal goal is maximizing informativeness and minimizing error. However, in practical applications, other factors come into play. Balancing informativeness and error involves trade-ofs, to be evaluated in light of contextual factors. For example, it may be acceptable to tolerate a higher error to increase informativeness, provided that this does not cause practical disruptions. Additionally, scalability is a crucial factor to consider, especially when harmonizing a large number of ontologies. Thus, the aim shifts towards achieving an operationally optimal balance.
Harmonization is further complicated by contingent factors. Domain/Application-level ontologies are often axiomatized using languages of limited expressiveness. This can be due to the ontologies’ specialized intended use, but also to the need to reduce development costs. Ontologies often either lack documentation, or it can be not accessible/not clear. Specialized jargon is often employed, with examples and terminologies understandable only to those deeply involved in the ontology’s use. FAIRness sensibility is also a recent theme. Moreover, ontologies often contain outright mistakes (in conceptualization) or inconsistencies among documentation and axiomatization. This is partly due to the issues mentioned above, and partly because ontologies are “living artifacts”: they are shaped by use and undergo subtle changes throughout their lifecycle. In addition, most applications, especially in industrial contexts, require informative alignments with no errors concerning contextually salient links, i.e., they require high specificity in individuals’ discrimination. And this list is far from exhaustive.
More issues can be identified for the harmonization of large ontology networks. First, and foremost,
harmonization usually involves pairwise alignments through semantic links. Problematically, the
number of required alignments increases exponentially ( p2´1q ) with the number of ontologies involved.
Concatenations of alignments, i.e., alignments mediated by sets of ontologies/alignments, are often less
informative. This is especially true given heterogeneity and the diverse domains of application/coverage
granularity. Establishing well-documented and richly axiomatized Reference ontologies [
Another challenge for any approach has to do with usability. Given a set of ontologies and alignments, tractability and operational usability can easily be lost. Fortunately, stakeholders are usually interested in only a specific subset of the network. Therefore, it has to be possible to access and use only a fragment of the network, i.e., the network has to be modular. Due to stakeholders’ diverse needs and desiderata, lfexibility is also necessary when it comes to alignments’ required expressiveness, and the availability of alignments diferently balancing informativeness and error. Finally, the harmonized network ought to be plastic, update-friendly and reusable, allowing for the introduction of new ontologies, and adjustments to the harmonized ontologies and the relative knowledge bases.
To summarize: (i) Even with richly axiomatized foundational ontologies, the encoded meaning will never truly be fully transparent, especially to machines. (ii) Ontologies’ (and ontology entities’) adequacy should be evaluated based on whether they reduce ambiguity below break-points determined by contextual factors. (iii) Let the formal characterization of an ontology entity be the subset of its set of axioms involving said entity and ontological entities recursively related to the entity, and the informal characterization of an ontology entity encompass all the related annotations and documentation (including labels, descriptions, comments), and even contextual factors related to actual usage of the ontology; (iii.a) informal characterizations are at least as important as (and complementary to) formal characterizations for usability, and (iii.b) ultimately more salient when it comes to ontology harmonization, as the ontologies to be connected are often built referring to diferent goals/aims/stakeholders/use cases. (iv.a) Harmonization procedures must fully account for both formal and informal aspects to produce adequate results in non-trivial scenarios, and (iv.b) it shouldn’t be expected there to be a single “correct” alignment independently of context and pragmatic choices, but rather a plurality of approximations with diferent pros and cons. 2 (v) In establishing networks of ontologies, scalability, modularity and flexibility are of the utmost importance. (v.a) Scalability requires mappings’ partiality, and a strategy to ensure that the number of alignments does not depend exponentially on the number of ontologies involved, without an excessive loss of informativeness. (v.b) The resulting framework has to be modular and flexible as a precondition for its practical adequacy, given computational requirements and the diverse plethora of involved stakeholders.
Overall, and in a motto, both ontology engineering, and ontology harmonization are a matter of “fit, rather than match”. Under this respect, they can arguably be assimilated to other forms of representation and communication, where negotiation and friction-reduction occupy the center stage.
The proposed tool, and related ontology network harmonization methodology, has been designed around the points listed in Section 2.4. It revolves around the creation of a limited set of FAIR, well-documented, standalone ontology entities (Bridge-Concepts), establishing pinpointed mediated alignments among a set of ontologies at core network junctures, individuated bottom-up from the network’s overall structure (at a given time), taking into due account knowledge bases and use-cases’ salience.
The label, preferred label, or Internationalized Resource Identifier (IRI) title used to identify the Bridge-Concept.
Suggested Bridge-Concept IRI.
A value between: Class OR ObjectProperty OR DataProperty OR Individual.
The domain(s) the Bridge-Concept was built for. More can be included, possibly organized taxonomically. This serves as a first source of disambiguation for domain experts and users in general.
This provides a natural language, informal definition of the concept, intended to be easily understood by domain experts. Elucidations should align with common knowledge and domain resources, avoiding references to other ontology entities (i.e., they should be ontology neutral). Ideally, they should also remain ontologically neutral (avoid commitments beyond the domain they pertain to) and be concise, with the inclusion of diverse usage examples (a plurality of them, to avoid prototyping efects) and the explicit addressing of potential ambiguities, focusing on contextually salient cases relevant for (the expected) ontology usage.
Labels used to refer to the concept, categorized as follows: (i) preferred label – the primary label for referring to the concept, combining intuition and informativeness; (ii) alternative labels – multiple labels commonly used to address the concept, even if they have narrower or wider meanings; (iii) deprecated labels – labels that may be misleading or encourage misuse, but which are used in practice. (iv) Hidden labels can also be included to support queries.
Being engineered as standalone ontology entities, Bridge-Concepts initially lack a formal characterization. This feature, which might initially appear puzzling (especially considering the principles of semantic technologies), finds an immediate explanation in their role of “mediators” among diverse formal conceptualizations, as well as support in the discussion above concerning the limits of ontology entities’ formal characterizations, tractability and context-sensitivity. The emphasis is thus on BridgeConcepts’ informal characterization, which pivots on stakeholders’ domain expertise while avoiding non domain-specific commitments (unless strictly necessary), referring to salient gold standards, and 2The dificulties encountered in setting up efective automatic harmonization tools focusing on structural and terminological approaches can find an explanation in the points just discussed. It is thus important to investigate scalable manual alternatives.
It lists existing domain resources, such as standards, books, articles, and dictionaries considered during the development of the Bridge-Concept. The template includes static references to these resources and quotations of relevant content. Multiple resources can be reported; renown resources that have a high likelihood of having influenced users’ conceptualizations are given priority. These resources act as reference points in the engineering phase and (together with the related comments) help domain experts better understand the Bridge-Concept, enhancing conceptual clarity.
Comments in this section explain the motivations underlying engineering choices in the elucidation, drawing from domain resources and highlighting similarities and diferences; the discussion should aim at solving possible ambiguities not fully addressed in the elucidation.
This section includes the IRI of one of the ontologies encompassing ontology entities which are aligned to the Bridge-Concept. This part of the template is replicated for each ontology.
A list of IRIs of specific ontology entities (belonging to the target ontology) to which the Bridge-Conceptis connected to.
This section provides an extensive discussion (in natural language) of the mapping (between the Bridge-Concept and the target ontology entities) choices and the underlying rationale. It includes contextual information concerning the intentionally adopted trade ofs between informativeness and error, considerations about possible alternative mappings considered and the evidence gathered in support of the choices made, facilitating third-party evaluation and validation of the proposed connection, and contributing to the clarification of the Bridge-Concept.
A description of the kind of mapping established. E.g., strongly hierarchical (such as owl:EquivalentClass or rdfs:SubPropertyOf), weakly hierarchical (e.g., skos:narrower), of similarity (e.g., skos:related). The latter can be employed to enhance the framework’s querying capabilities, and for advanced analytic graph-based approaches.
Proposed mapping axiom(s) between the Bridge-Concept and the ontology entities
are provided in an OWL2 compliant syntax, such as Turtle, Manchester, RDF/XML,
Functional-Style, or OWL/XML. Notably, the mappings can be complex [
Before delving into the presentation of Bridge-Concepts’ two core roles, it might be beneficial to get acquainted with the tool. A template, divided in three parts depending on the core stakeholders addressed, is proposed for documenting bridge concepts. Notably, the template acts as a human interface, while being implementation-friendly in .owl file format. Examples produced in the context of the OntoCommons project in cooperation with practitioners are available at https://github.com/ OntoCommons/OntologyFramework, and the implementation schema, as well as practices to ensure FAIR-ness, are described in the already cited OntoCommons’ D 2.9.
The first part of the template (Table 1) is relevant for all users, and contains the core elements constituting the informal characterization of a Bridge-Concept. The second part (Table 2) holds particular significance for domain experts as it encompasses links with gold standards and other knowledge domain resources which better situate the concept. The third part (Table 3) delineates the formal mappings with existing ontology entities, along with pertinent information regarding the latter.
In order to fulfill their formal role, i.e., establish mediated alignments among ontologies, Bridge-Concepts themselves have to be semantically connected to the target ontologies through axioms. In the best case scenario, for each relevant ontology in the network, there should be one ontology entity equivalent to a given Bridge-Concept. In practice, this would only be feasible if all the involved ontologies covered approximately the same domain(s) and revolved around the same core concepts, with operationally and semantically consistent formal (and informal) characterizations. Usually, given an efective choice of Bridge-Concepts, it is possible to individuate both a super and a sub class/relations, without the need to make use of complex alignment axioms to establish suficiently informative mediated connections.
Let us consider a practical example (see Fig. 1), involving a Bridge-Concept developed in the context of the OntoCommons project, as well as two salient industrial ontologies part of the OntoCommons EcoSystem (OCES) connected to it, SAREF and IOF-Core.3 Specifically, we consider “ t, , , , u and relative axioms , and “ t , u and relative axioms . The engineered Bridge-Concept, with preferred label “Equipment”, was aligned to SAREF and to IOF-Core as follows: Ď , Ď , Ď . Given the ontology Network , such that “ Y Y tu and 1 “ Y Y , it trivially follows that e.g., Ď , whereas the first class belongs to SAREF, and the second to IOF-Core. Thus, SAREF and IOF-Core’s ontology entities are semantically linked through OntoCommons’ BridgeConcept “Equipment”, allowing the exchange of data, and individual constants to be imported. As per the example, Bridge-Concepts efectively function as a data pipeline. Intuitively, data “flows upwards”: all the data covered by the sub-ontology entities connected to the Bridge-Concept is made available to the entirety of the network. Conversely, reasoning “flows downwards”: the axioms characterizing super-ontology entities connected to the Bridge-Concept can be exploited by the entirety of the network: e.g., in the example all the axioms involving , , as well as BFO’s superclasses, constrain SAREF’s , and its subclasses, following the alignment.
In principle, two kinds of Bridge-Concepts could be distinguished, depending on whether the focus is establishing “vertical” connections among ontologies at diferent levels, or “horizontal” connections among ontologies at the same level – provided that the two functions are not mutually exclusive, and the classification is contextually dependent on the set of ontologies part of the network considered, as well as use-case-related factors. Vertical connections can be particularly useful for validation and 3See https://saref.etsi.org/ and https://spec.industrialontologies.org/iof/, respectively. avoiding double-counting, especially if High-Level ontologies are included in the network (as in the example). Notably, given standard High-Level ontologies’ architectures, few Bridge-Concepts could acts as seeds to ground a domain-level ontology on a top-level ontology capable of providing foundations [13]. Conversely, horizontal connections play a crucial role in ensuring eficient data sharing and the integration of specialized modules and related reasoning. Consequently, they are often regarded as the cornerstone of interoperability by stakeholders.
Notably, the semantic alignments supported by Bridge-Concepts facilitate the reduction of various kinds of heterogeneity. If informativeness and error is properly considered in the alignment process, they significantly contribute to addressing semiotic heterogeneity among the involved ontologies. Semantic heterogeneity is managed through the sharing of reasoning and the establishment of mediated correspondences between ontology entities, resulting in more comprehensive and multifaceted formal characterizations. Moreover, even terminological heterogeneity is addressed to the extent that entities formally linked through a mediated semantic correspondence collapse, rendering labeling/IRI variants inconsequential within the integrated ontology network (likewise, “false friends” are made readily discernible through a lack of mediated correspondences). Thus, in line with the outlined desiderata, these indirect solutions accommodate the pluralistic needs of stakeholders without necessitating any changes to the original ontologies, while establishing interoperability at the network level.
It is worth reiterating that, if necessary, Bridge-Concepts can be linked to ontology entities through complex axioms, in pursuit of optimal trade-ofs between informativeness and error. Additionally, while a simplified case involving classes was presented, in scenarios revolving around knowledge bases, focusing on object and data properties might prove beneficial, although this might pose greater challenges in the alignment phase. Furthermore, it is evident that any number of ontologies can be linked to a given Bridge-Concept. This aspect goes ways towards addressing two core issues in establishing a harmonized network of ontologies, namely scalability (specifically, the number of alignments) and lfexibility. These points will be elaborated on in Sec. 3.3; however, it is worth anticipating that BridgeConcepts’ standalone nature allows diferent stakeholders to select diferent sets of Bridge-Concepts (and ontologies), supporting spot connections on demand. Hence, through alignments, Bridge-Concepts are ultimately provided with network-specific, extendable and plastic formal characterizations. Moreover, similar to Reference Ontologies, sets of Bridge-Concepts can serve as hubs in a hub-and-spoke structure, ensuring that the number of alignments scales linearly with the number of ontologies to be harmonized, providing advantages with respect to standard approaches already with three ontologies involved.
As per the discussion above, informal characterizations are crucial when it comes to usability (and thus reusability) and the ultimate hinges for alignment procedures. This aspect arguably underscores the demand among stakeholders for controlled, shared vocabularies, alongside a renewed emphasis on documentation and the adoption of FAIR principles. With their focus on informal characterizations, and being FAIR-by-design, Bridge-Concepts can serve as vocabulary-entries tailored for ontology usage.
Bridge-Concepts are chiefly characterized through their elucidations, but also via labels, the specification of the more relevant domains, as well as through the connections with domain knowledge resources and existing ontologies’ concepts, which are explicitly discussed in their documentation. Elucidations of Bridge-Concepts are crafted to be easily comprehensible by domain experts, leveraging their expertise. They are not strictly definitions, as they do not provide necessary and suficient conditions referring to other concepts; instead, they are intended to guide stakeholders in making accurate intensional judgments. In essence, they must: (1) strike a balance between (1a) flexibility (i.e., they have to be intuitive given the assumed background knowledge, rather than a set of formal constraints that might seem obscure or noncore from the viewpoint endorsed by diferent stakeholders) and (1b) rigidity (i.e., they have to provide pragmatically well-defined boundaries, necessary for efective machine implementation); (2) maintain explicit, detailed connections (2a) with primary knowledge domain resources and (2b) ontology entities they are meant to be semantically connected with, depending on the importance of the relevant ontology in the network and/or for a use-case; (3) align with common sense, that being the ultimate foothold for interpretation negotiation.
To attain the desired level of detail, Bridge-Concepts’ elucidations should specifically address ambiguities relevant to their prospective usage. Unlike standard definitions, it can be informative for ontology use to specify that certain traits are not discriminatory, particularly when a concept is more coarse-grained. For example, when engineering a Bridge-Concept centered on atoms for chemical ontologies, it’s pertinent to specify whether both standalone and bonded entities are included, and whether they may have an unbalanced number of electrons with respect to their atomic number: in fact, core stakeholders take divergent stances on this specific matter, as testified both by knowledge domain resources and the characterizations (both formal and informal) of relevant ontology entities. At times, ofering specific examples and counterexamples can be efective. However, priority should be given to general principles to avoid prototyping efects, especially if the concepts are coarse-grained. Finally, brevity is a desirable characteristic, although achieving a harmonious balance among these requirements is a complex endeavor. A lengthy elucidation might increase the risk of stakeholders overlooking core points. It should be emphasized that the pragmatic aim is to efectively guide stakeholders, rather than precision itself. This necessitates an iterative process of refinement and adjustment. Elucidations should follow this standard internal organization: (1) introduction leveraging domain experts’ background knowledge; (2) informal description with implicit references to selected gold standards and ontology entities; (3) notes on the use of adjacent concepts in the domain; (4) resolution of ambiguities through the explicit individuation of traits and values commonly cited; (5) possible examples and counterexamples.
Moving on, the selection of the preferred label holds particular significance as it constitutes the initial and most prominent element influencing a user. Therefore, preferred labels should be designated as the final step in the Bridge-Concept engineering process. In certain instances, prioritizing clarity over immediacy by making labels explicit might be advantageous and prevent potential misunderstandings, especially if the relevant Bridge-Concepts were developed with specific objectives in mind. Finally, the annotations discussing connections and discrepancies with respect to knowledge domain resources and existing ontologies’ concepts (once a Bridge-Concept is aligned to the latter) should be comprehensive. Standards serve as reference points; thus, stakeholders might find these annotations more enlightening than the elucidations themselves.
A noteworthy aspect of Bridge-Concepts is their potential to serve as a standardized vocabulary. While primarily designed for pragmatic purposes, they ofer the prospect of evolving into recognized standards themselves, if they prove efective. It is essential to acknowledge that Bridge-Concepts aim to establish unifying connections rather than supplanting other resources, a fundamental concern within the domain of standards and meta-ontologies’ ecology, undermining long-term reusability and interoperability. Finally, like Bridge-Concepts are formally characterized through the alignment to ontology entities belonging to a set of ontologies making up the core of a network, said ontology entities’ documentation is indirectly enriched by their connection to Bridge-Concepts, addressing one of the core issues outlined in Section 2.4. Notably, Bridge-Concepts are FAIR-by-design, with all the sections in the template above being directly implementable in a machine-readable environment (.owl), either as elements of an ontology, or as annotations, following a standardized schema in line with W3C recommendations. They are meant to be associated with IRIs, and made available in maintained repositories and commonly employed portals. Thus, Bridge-Concepts can significantly enhance the FAIR-ness level of individual ontologies and the overall network.
In their role as ontology-specific vocabulary-entries, Bridge-Concepts can be assimilated to (degenerate) content ontology design patterns [14], and can fulfill some of their functions to enhance ontology design and reusability. Notably, an ontology deliberately incorporating an ontology entity equivalent to a Bridge-Concept will seamlessly integrate into the relevant ontology network. However, ontology design content patterns appear to be more efective for ontology design, ofering standardized, modular, and formal solutions that exemplify best practices. Conversely, Bridge-Concepts are arguably more appropriate for ontology harmonization, being engineered bottom-up for that very purpose. In fact, being standalone entities, they are more easily connectable and less formally committed.4
As shown by the example, individual Bridge-Concepts can serve as the foundation for mediated alignments among sets of ontologies, facilitating data and reasoning sharing while potentially significantly enhancing local clarity and FAIR-ness. However, establishing a fully harmonized network of ontologies typically necessitates a multitude of links. This leads to considerations regarding the selection of which Bridge-Concepts to engineer for a given set of ontologies (and given practical applications), as well as related issues concerning the framework’s maintenance. Indeed, scalability has been identified as one of the core challenges in establishing harmonized networks. While the target ontologies need only be partially aligned to support efective interoperability, the efort required to engineer a single Bridge-Concept (including both characterization and alignments) makes it mandatory to keep their number contained. Once again, an abundance of (potentially low-quality) Bridge-Concepts would be counterproductive, potentially diminishing their findability and reusability.
Delving into the selection procedure in detail exceeds the scope of this introductory paper. Nevertheless, several options can be outlined. One approach involves conducting a statistical analysis of the terms present in the ontologies to be harmonized, or in a subset forming the core of a potentially expandable network. Although purely terminological analysis is susceptible to the limitations outlined throughout the discussion, the frequency of terms can serve as a reliable indicator of the salience of underlying concepts, provided a suficiently large sample of ontologies is available. If the framework is to remain open and expandable, the results of the analysis can be supplemented with candidates directly selected by experts to mitigate deviations stemming from the idiosyncrasies of the initial set of core ontologies. A similar strategy has been employed within the context of the OntoTrans and 4It’s worth noting that specific use-cases might benefit from the utilization of semantically connected and inter-defined clusters of concepts or outright concept patterns to establish connections among multiple ontologies. However, exploring this topic further is beyond the scope of this brief, general introduction to Bridge-Concepts.
OntoCommons European Projects as detailed in [
When considering a plurality of Bridge-Concepts, issues related to network consistency arise: let ∆ be a set of ontologies 1, 2, ... and Δ “ 1 Y 2 Y ... Y its axioms, and and standalone Bridge-Concepts, and being the related alignment axioms. While the consistency of Δ Y and Δ Y is ensured by the consistency of the single ontologies, and of the Bridge-Concept-specific sets of alignment axioms (provided that the ontologies are not otherwise connected), nothing guarantees the consistency of Δ Y Y . While this inconsistency could suggest issues in Bridge-Concepts’ alignments, it could also stem from conceptual or architectural mistakes in the ontologies to be harmonized, which cannot, nor should, be rectified through the alignment itself. Ideally, a network of ontologies should strive for full consistency. However, since Bridge-Concepts are standalone, the resulting network is inherently modular, allowing for ways to manage inconsistency. Specifically, let be a network composed of a set of ontologies and a set of Bridge-Concepts with related mapping axioms ℬ; it is possible to select a consistent sub-network including only a certain set of ontologies such that Ď and a certain set of Bridge-Concepts ℬ such that ℬ Ď ℬ, picking the most relevant ontologies and connections. Likewise, stakeholders can leverage the resulting network’s modularity to suit their use cases, focusing on the part of the network that interests them. Nonetheless, extra caution should be exercised in data imports to prevent error escalation.
Networks can be improved through the inclusion of one of more High-Level ontologies, possibly independently formally aligned with each other. Aligning Bridge-Concepts to them first, can facilitate further alignments and prevent misalignments, providing a first form of validation. The benefits of leveraging foundational High-Level ontologies for ontology alignment are well-documented in the literature [15], and the prospect of including them through Bridge-Concepts accessible to domain experts can arguably be considered an additional advantage of the proposed tool. This strategy has already yielded positive results in the context of the OntoCommons project, with the creation of the OntoCommons EcoSystem (OCES). Among other things, in this context the inclusion of High-Level ontologies served to facilitate and partially validate Bridge-Concept-mediated alignments between SAREF and IOF-Core (which were presented as an example). In general, alignments among superclasses can facilitate the establishment of alignments among leaf classes. Consequently, it is worth mentioning that the links among ontology entities established by Bridge-Concepts might in turn be exploited as constraints for automatic alignment tools, with the potential of greatly increasing informativeness, with limited errors, further improving scalability [16][13]. 5Some of these options are explored in the already cited D 2.9 https://ec.europa.eu/research/participants/documents/ downloadPublic?documentIds=080166e503ae3f85&appId=PPGMS, including a tentative workflow and weighting formulas. Notably, it might be possible to leverage AI-based approaches to enhance the methodology and tailor it to specific contexts.
This paper ofered a concise introduction to a methodology to establish harmonized networks of ontologies by employing Bridge-Concepts, standalone ontology entities tailored to facilitate mediated semantic alignments at pivotal junctures and address issues related to documentation and FAIR-ness. The proposed framework is tailored for the establishment of large harmonized networks of ontologies for federated distributed systems, supporting spot-connections for data and reasoning-sharing, requiring no changes to the ontologies to be harmonized, and granting stakeholders the possibility to isolate specific network segments pertinent to their use cases. The approach is particularly suitable for industrial settings, involving a plurality of stakeholders and value-chains extending over a number of diferent domains, requiring high informativeness and reduced errors in data sharing, and where considerations about data control and the reduction of operational interference are of paramount concern.
Still, several points require further exploration: specifically, Bridge-Concepts’ reliance on manual alignments makes them susceptible to all the associated issues. Additionally, further extensive field testing is essential to identify potential bottlenecks in the procedures for selecting and engineering Bridge-Concepts. Moreover, potential issues may arise due to diachronic changes in ontologies – while the framework should be flexible enough to deal with them, maintenance costs have to be accounted for. Tentative answers to these challenges are outlined in the documents referenced in the introduction; a more extensive discussion is deferred to future publications.
This research would not have been possible without the funding from OntoCommons (GA N 958371) and OntoTrans (GA N 862136). The contents of this publication are solely the responsibility of the authors and do not necessarily represent the oficial views of the involved Consortia. We thank Claudio Masolo, the members of the Consortia, and FOMI’s reviewers & participants for the valuable comments. [13] P. Barcelos, et al., Inferring ontological categories of owl classes using foundational rules, in:
Proceedings of Formal Ontology in Information Systems 2023, 2023. [14] V. Presutti, A. Gangemi, Content Ontology Design Patterns as Practical Building Blocks for Web
Ontologies, in: Q. Li, et al. (Eds.), Conceptual Modeling, Springer Berlin, 2008, pp. 128–141. [15] D. Schmidt, C. T. dos Santos, R. Vieira, Analysing top-level and domain ontology alignments from matching systems, in: OM@ISWC, 2016. [16] P. Lambrix, Q. Liu, Using partial reference alignments to align ontologies, in: L. Aroyo (Ed.), The Semantic Web: Research and Applications, Springer Berlin, Berlin, 2009, pp. 188–202.