Modeling Linguistic Phenomena in Gallo-Italic Dialects: The Linguistic Phenomena Ontology

The paper introduces the Linguistic Phenomena Ontology (LiPh), an OWL extension of OntoLex-lemon. It aims to represent linguistic phenomena affecting the morphology and pronunciation of lexical expressions. The paper discusses use cases, possible SPARQL queries.

The ontology fills a certain gap in the Semantic Web community, where fine-grained modeling of linguistic change and contact phenomena is underexplored. By building on OntoLex-lemon and its etymological extension, the work ensures interoperability with existing linguistic LOD resources.
The paper defines competency questions, which is an important plus and provides formal OWL axioms, demonstrating their operationalization through queries. The prsented case study is relevant.

The work shows a good and clear built upon prior work (LiPh 1 and LiPh 2).

Regarding related work, maybe the authors may want to consider linguistic/lexical resources works in the sematic web community, like
Tchechmedjiev, A. et al (2017). Ontolex JeuxDeMots and Its Alignment to the Linguistic Linked Open Data Cloud. In ISWC

A main drawback to me relates to evaluation of the resource: the ontology is demonstrated with queries and examples, but lacks a systematic evaluation, like comparison with annotated linguistic corpora, expert validation of encoded phenomena, or performance testing (for example in specific reasoning tasks.

Also, the integration with existing resources is not pushed sufficiently (e.g. resources such as WordNet, Wikidata lexemes) to demonstrate broader interoperability and make a solid case of it.

I may have missed that, but a public release of the ontology (with documentation, examples, and perhaps a SPARQL endpoint) to support reproducibility and community adoption is missing. I did not find a link in the paper. This is a very important point that would condition acceptance in my view.

The paper proposes a big work and expands on previous existing ontological schema and publications.
Overall the work tackles interesting issues, there are nevertheless some modeling problems that need to be resolved, resulting in a conditional acceptance, given that the authors better motivate their choices and resolves some issues.
The work is indeed of interest and it well written, therefore I would highly recommend the authors to tackle the points mentioned below, to achieve a high quality publication.

==== Detailed Review ====

If the order of authors is alphabetical order please state it, I would also suggest to add a CrediT taxonomy, since the corresponding author is fourth in alphabetical order.

p. 6 - Figure 4: is "lenition t > d" a specific rule?, so, can I assume that there is a "super-rule" `:lenition`, and then subproperties `:lenition t > d`? Please clarify this, which seems to come from an existing ontology

p. 8 - Sec. 4.2: Competency Questions, when at domain modeling time, are usually broader and more general too, e.g. "what linguistic phenomena exist?", etc. Please rework them, or, if you want to keep these, specify that there are specific queries that you can resolve

p. 9 Sec. 4.3: "given an object property pρ dedicated...": this whole sentence should have a formula too, then explained in natural language as it is right now.
(and possibly an example)

p. 9 Sec. 4.4: "Specifically, it avoids providing semantically relevant descriptions–i.e., statements which affect reasoning–for phenomena and their occurrences, as doing so would lead to undecidability": what do you mean here? in the sentence above you just asserted that you could model `:lenition t > d` and `truncation` "as subproperties of liph:derives". This per se does not drive to any undecidability, you can have subproperties of properties which are transitive and reflexive, so what are you referring to here? please clarify.

p. 10 Sec. 4.4: regarding `liph:LinguisticPhenomenon` and `LinguisticPhenomenonOccurrence`: I see your necessity, but this is not the way to tackle it: you can solve this need modularising your ontology and using the punning technique. In OWL2 you can treat the same entity as both a class (of entities) and an individual, the only thing that you have to be careful to is that then, if you run the reasoner you could get some undesired inferences. But you can have a specific Linguistic Phenomenon as a class, and all individuals would be occurrences of that specific Linguistic Phenomenon. And then have the same URI of that Linguistic Phenomenon, but modeled as an individual, in a separate module dedicated to all the existing linguistic phenomena modeled as individuals, because you want to predicate something about them intensionally vs extensionally.

p. 15 Sec. 4.5: ":oi liph:source :x(i-1);" and followings: which syntax are you using here? the usage of parentheses in Turtle syntax is allowed but that does not seem proper Turtle.

p.16 Sec. 4.6: " Descriprion Logic" --> "Description Logic"

p.16 Sec 4.6: I'm not sure I completely understand "finite state phenomena", why do you need it?

p. 18 Sec 4.7: "In contrast with liph:linguisticPhenomenon, the reflexivity of liph:derives is appropriate, as it serves for modeling empty chains." --> I don't understand this sentence, could you elaborate please?

p. 18 Sec 4.7: it would be nice to have an endpoint with some toy data, or a version of the ontology with toy data to play with

Comments on the Ontology OWL file:

why is the `:LinguisticPhenomenonOccurrence`class axiomatised as a subClass of something that is:
`occurrenceOf only :LinguisticPhenomenon`
but
`occurrenceOf some owl:Thing` ?
Existential quantifier here means "it is an occurrence of something", but Universal quantifier adds: "if it is an occurrence of something, that something is a Linguistic Phenomenon".

In this article, the authors introduced the second version of an ontology, i.e. the Linguistic Phenomena Ontology (LiPh), which extends OntoLex-lemon to represent linguistic phenomena. Using the ontology, the authors also show a possible application for representing Gallo-Italic features according to the varieties that are spoken in Sicily.

I want to clarify that I am not a domain expert, since I am neither a linguist nor, as such, do I have the appropriate competencies to evaluate the domain-related aspects of the work introduced by the authors. However, as a non-expert on the topic, I praised the careful explanation the authors provided of the various linguistic aspects introduced in the paper, which is also accessible to non-experts. Thus, my entire review will focus on my areas of expertise, which include ontology engineering, and all the comments I provide concern this specific, but crucial, point of view, given the work presented.

Firstly, above all, as an article presenting a new ontology, even if a new version of an existing one, I would have expected a clear and detailed account of the ontology development methodology used and the related materials produced during its execution. In fact, in order to demonstrate to a reader that the ontology developed is robust against a specific set of requirements that the ontology should address, providing evidence of the various steps followed is always necessary material to substantiate the research conducted here. In the article, there is no mention of any methodological procedure followed for addressing the development, while, within the community, methodologies have been proposed and followed systematically for this kind of work, such as (just to mention a few):

* LOT - https://lot.linkeddata.es/;
* SAMOD - https://essepuntato.it/samod/ [full disclosure: I'm the author of this methodology];
* eXtreme Design - https://ceur-ws.org/Vol-516/pap21.pdf;
* UPON Lite - https://cacm.acm.org/research/a-lightweight-methodology-for-rapid-ontolo....

The good thing about applying such methodologies is that, as a consequence of their use, the ontology engineer is practically obliged to provide also a set of additional material which acts as documentation to present the ontology to others. For instance, what I would have expected to reach, even to an external repository (e.g. GitHub), is all the documents usually detailing:

* descriptive scenarios that the new ontology should address;
* a full list of competency questions (going beyond the two presented as examples in the article) acting as formal requirements for the development;
* some examples of use of the ontology (e.g. RDF files detailing complete real case scenarios like that in section 5.1);
* the translation of all the competency questions into appropriate SPARQL queries.

All of these seem missing from the article.

Other aspects have been addressed, even if only partially. First of all, the lack of appropriate human-readable documentation. At https://gallosiciliani.unict.it/ns/liph, there is a webpage that details the various entities in the ontology, but it does not provide enough detail for a human to understand what these entities are, how they interact, and how the ontology can be used. While producing human-readable documentation is indeed a huge amount of work, there are tools that, starting with the ontology's annotations, simplify the creation of such documentation. Just to mention a few:

* LODE: https://w3id.org/lode [full disclosure: I'm the developer of this tool]
* WIDOCO: https://github.com/dgarijo/WIDOCO
* PyLODE: https://github.com/rdflib/pyLODE

In addition, especially today, it is very important to demonstrate that the developed ontology adheres to well-known, shared practices. While assessing these by hand can be a huge manual work, there are tools that have been developed to at least provide an initial screening of what is missed, such as:

* OOPS!: https://oops.linkeddata.es/
* FOOPS!: https://foops.linkeddata.es/FAIR_validator.html

Finally, the Semantic Web community has provided several guidelines for writing an article about a resource (e.g. an ontology), as is the case with the present contribution. Following these guidelines, such as those at https://iswc2026.semanticweb.org/#/calls/resource, will greatly strengthen the article. While all these guidelines are of great importance, there are specific aspects that are more related to the ontology engineering parts that should be considered, such as (quoting from the URL above):

* Does the ontology have high-quality documentation? Are there tutorials available?
* Are reported evaluation results and metrics reproducible?
* Is there potential for extensibility to meet future requirements?
* Does the ontology include a clear explanation of how others use it and the data compliant with it?
* Does the ontology description clearly state what it can and cannot do, and the rationale for the exclusion of some functionality?
* Does the design of the ontology follow specific best practices?
* Did the authors perform an appropriate reuse or extension of suitable high-quality ontologies, such as extending upper ontologies and/or reusing ontology design patterns?
* Does the ontology provide an appropriate description (both human- and machine-readable), thus encouraging the adoption of FAIR principles?
* Is there a schema diagram in a given standard diagrammatic format (UML, E/R, GRAFFOO, etc.) for ontologies?
* Is the ontology registered in (community) registries (e.g. Linked Open Vocabularies and FAIRsharing)?
* Is the ontology registered in generic repositories such as FigShare, Zenodo or GitHub?
* Is the ontology (and related results) published at a persistent URI (PURL, DOI, w3id)?
* Is there a canonical citation associated with the ontology?
* Does the ontology provide a license specification?

As a final point in this respect, I tested accessing the ontology via CURL, and the content negotiation set is only partial, since it works correctly with HTML and Turtle only. There is no way, for instance, to get a representation of that ontology into other RDF formats, which are usually expected: RDF/XML, JSON-LD, N-triples.

In addition to these comments, there are other aspects that deserve some reflections. These are listed as follows.

#1. Avoiding affecting the semantics of existing models
In Table 1, the authors add specific axioms to two distinct entities defined in the OntoLex Lemon module, i.e. ontolex:Form and ontolex:writtenRep. In particular, the former is defined as a subclass of liph:LexicalObject, while the latter is defined as a subproperty of liph:writtenRep. While this can be interpreted as a legitimate choice by the article's authors, given the open-world assumption in this context, such modelling explicitly changes the semantics of ontolex:Form and ontolex:writtenRep beyond the original intention of the OntoLex Lemon module. As a general practice, this behaviour should be avoided by creating "proxy" entities that inherit from the OntoLex module and the LiPh ontology. For instance, instead of

ontolex:Form rdfs:subClassOf liph:LexicalObject

one should define a new entity (e.g. liph:Form) as follows:

liph:Form rdfs:subClassOf ontolex:Form
liph:Form rdfs:subClassOf liph:LexicalObject

In this way, liph:Form inherits all the semantic structure of ontolex:Form, and it is, at the same time, a subclass of liph:LexicalObject, but without affecting the semantics of ontolex:Form. A similar reasoning can be done for the data property liph:writtenRep.

#2. About reification
The authors claim that one of the most important additions in this new version of LiPh is the adoption of reification for handling the occurrences of linguistic phenomena, and they mention that the approach used is introduced in [25]. However, the reference points to a quite extensive book describing description logic and a reader may not know how it defines reification, also considering the different mechanisms that can be used for reifying depending on the situation (reifying a properties, reifying statements, reifying graphs, etc.). Thus, it would be important to introduce the reifying approach used in this context and explain how it differs from the previous version of the ontology.

3. Perdurants vs Endurants
There is a passage in the text where the authors refer to the concept "perdurant properties", which is a well-known aspect for expert ontologists, in particular in foundational ontologies. However, considering that the reader of the article may not be an expert in the topic, it would be appropriate to provide an explanation about what "perdurants" means, and how it differs from the other related concept, i.e. that of "endurants". Here, referring to existing foundational ontologies based on this characterisation (e.g. DOLCE) could add value to the discussion.

4. Related works
While the authors say that the ontology they propose is the first semantic artefact addressing the aspects they care about and introduce in the article, the related works section seems to focus mainly on OntoLex-lemon, which is then addressed in more detail in the following section. However, there may be other structured approaches that are not based on ontologies, and that may have been used in the past to define the aspects addressed by the LiPh (e.g. in JSON and CSV). In addition, given the approach used in the ontology, a brief introduction to some foundational ontologies that provide, for instance, a theoretical foundation the authors have followed to address part of the modelling (e.g. the occurrences of linguistic phenomena) would be appropriate in this section.

Silvio Peroni
https://orcid.org/0000-0003-0530-4305