The GOLEM Ontology for Narrative and Fiction

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The contribution illustrates the novel GOLEM ontology, which is intended to describe narrative and fictional elements at various levels. Specifically, it reuses classes and properties from DOLCE, CIDOC-CRM and LRMoo to provide modules describing 1) characters and their characteristics, 2) narrative events, 3) roles and relationships occurring in the different situations, 4) the various sequences in which events part of the same story can be ordered, 5) scholars’ interpretations of the narrative content.

The paper is well-written, easy to follow, well-grounded in the literature, and proposes a relevant, rich model that represents a valuable resource for the digital and computational humanities sector. The ontology is rigorously grounded in the semantics of DOLCE, and every modelling choice is well motivated. Furthermore, the presence of a repository, which contains the code, and of ontology descriptions both in a wiki and a PyLODE-generated web page, improves the understanding of the ontology, favouring its evaluation and reusability.

Nevertheless, some revisions are needed. In particular, the ontology needs an evaluation and to follow a development methodology more thoroughly. Despite this lack, the presented modules, which are well-grounded in foundational ontologies, are sound.  For these reasons, I recommend accepting the paper with revisions for its improvement, illustrated below.

 

Methodology:

Section 3.1

The authors state that the development process involved both a theory-driven and a bottom-up approach. Whereas the bottom-up approach is briefly illustrated, they don’t make explicit how the theories contributed to the modelling. Which theories are considered? Can they be aligned? In the case that an alignment was not possible, which theory was preferred and why? A brief description of this would improve the quality of the work.

The authors say that the ontological approach is proposed based on the reference Tomasi 2020. Nevertheless, as the reference is not included in the list, it is not possible to verify which approach they refer to. Please insert the reference.

Section 3.2

The development follows the practice of modular design. Nevertheless, well-known approaches for ontology development using an agile methodology are available (e.g., SAMOD [1] and NeON [2]). In particular, eXtreme design is developed for the reuse of ODPs, which the authors consider [3].  Specifically, they introduce scenarios and an evaluation phase for every module developed through competency questions over real examples, aspects that are missing in the current work. For these reasons, I recommend specifying how their method relates to similar approaches, and I suggest grounding the still missing evaluation on one of these methods.  

Section 3.3

On what basis was it possible to establish the requirements? Please briefly specify it

 

Additional recommendations for the evaluation

Besides an evaluation with competency questions, an overview of the ontology can be provided with tools as Ontometrics [4]. Additionally, the authors provided some aspects in line with the FAIR principles. I suggest evaluating the FAIRness of the ontology, for example, with the FOOPS! Tool [5]

 

Ontology

Narrative module, section 4.5: Characters are directly related to roles and the events in which they have such roles. Please clarify how it is possible to distinguish which role is played in which event when the character presents multiple roles in the same narrative sequence. For example, a story in which a character, in subsequent events, plays the role of “lover”, but then he/she turns into a murderer. In this scenario, both roles are described (dlp:modal-target) by the same G7 Narrative Sequence, the character plays both roles, and the character participates in both narrative events. How is it possible to understand in which narrative event (and narrative unit) the character has respectively the role of “lover” and of “murderer”?

 

Minor comments

Please add a table with the prefixes used.

Figure 3: The same class “social relationship” is presented as a gc: G4 Social relationship for the first instance and as a dlp:social-relationship for the second instance. Please indicate the introduced class and its alignment in both instances.

Figure 7: Multiple prefixes, namely “gc:” and “golem:”, probably referring to the same ontology, are used. Please update with a unique prefix or clarify their difference.

 

 

References

[1] Peroni, S. (2016). SAMOD: An agile methodology for the development of ontologies. Proceedings of the 13th OWL: Experiences and Directions Workshop and 5th OWL Reasoner Evaluation Workshop (OWLED-ORE 2016), 1579911 Bytes. https://doi.org/10.6084/M9.FIGSHARE.3189769

[2] Suárez-Figueroa, M. C., Gómez-Pérez, A., & Fernández-López, M. (2015). The NeOn Methodology framework: A scenario-based methodology for ontology development. Applied Ontology, 10(2), 107–145. https://doi.org/10.3233/AO-150145

[3] Presutti, V., Daga, E., Gangemi, A., & Blomqvist, E. (2009). eXtreme Design with Content Ontology Design Patterns. Proceedings of the 2009 International Conference on Ontology Patterns, 516, 83–97.

[4] https://ontometrics.informatik.uni-rostock.de/ontologymetrics/

[5] https://catalogue.fair-impact.eu/resources/foops

Author Response

Comment 1: The authors state that the development process involved both a theory-driven and a bottom-up approach. Whereas the bottom-up approach is briefly illustrated, they don’t make explicit how the theories contributed to the modelling. Which theories are considered? Can they be aligned? In the case that an alignment was not possible, which theory was preferred and why? A brief description of this would improve the quality of the work.

Response 1: Starting from line 341, we now clarify that “ Given the importance of scholarly debate and perspectivism in humanistic research, we focused on concepts that are broad enough and sufficiently expressive to serve theories based on different epistemological assumptions and definitions of key concepts.” Discussion of specific narrative theories is also included in the description of the respective modules, e.g. for the difference between “location” and “setting”, and for the modeling of narrative sequences.

 

Comment 2: The authors say that the ontological approach is proposed based on the reference Tomasi 2020. Nevertheless, as the reference is not included in the list, it is not possible to verify which approach they refer to. Please insert the reference.

Response 2: we added the reference.

 

Comment 3: Besides an evaluation with competency questions, an overview of the ontology can be provided with tools as Ontometrics. Additionally, the authors provided some aspects in line with the FAIR principles. I suggest evaluating the FAIRness of the ontology, for example, with the FOOPS! Tool.

Response 3: Thank you for this valuable suggestion. We accepted and implemented it by evaluating the ontology using both the Ontometrics and FOOPS! Tools. The Ontometrics results, including a selection of key structural metrics, have been compiled and added to the appendix in tabular form. Regarding FAIR compliance, we applied the FOOPS! tool. The initial overall FOOPS! score was 0.38; through several metadata improvements and structural refinements, we have raised it to 0.89. Further improvements are expected upon the completion of the ontology’s registration in metadata repositories. The application of both tools is now documented and cited in the revised manuscript (lines 365–369).

Comment 4: The development follows the practice of modular design. Nevertheless, well-known approaches for ontology development using an agile methodology are available (e.g., SAMOD [1] and NeON [2]). In particular, eXtreme design is developed for the reuse of ODPs, which the authors consider [3].  Specifically, they introduce scenarios and an evaluation phase for every module developed through competency questions over real examples, aspects that are missing in the current work. For these reasons, I recommend specifying how their method relates to similar approaches, and I suggest grounding the still missing evaluation on one of these methods.  

Response 4: We were not aware of these specific publications during the development phase and simply relied on examples from other articles. We acknowledge that it would have been more rigorous to follow one of the suggested methodologies, but to correctly describe our process, we now simply added a reference to the eXtreme design method, which is the closest one to the process we followed. See point 5 in section 3.1 (line 368) and line 378 in section 3.2.. Additionally, we conducted a more detailed evaluation with competency questions in another article that we now cite in line 370.

Comment 5: On what basis was it possible to establish the requirements? Please briefly specify it

Response 5: Starting from line 413, we clarify how the requirements are based on literary and narrative theories.

Comment 6: Narrative module, section 4.5: Characters are directly related to roles and the events in which they have such roles. Please clarify how it is possible to distinguish which role is played in which event when the character presents multiple roles in the same narrative sequence. For example, a story in which a character, in subsequent events, plays the role of “lover”, but then he/she turns into a murderer. In this scenario, both roles are described (dlp:modal-target) by the same G7 Narrative Sequence, the character plays both roles, and the character participates in both narrative events. How is it possible to understand in which narrative event (and narrative unit) the character has respectively the role of “lover” and of “murderer”?

Response 6: Starting from line 717, we now clarify that “To distinguish which role is played in which event when a character presents multiple roles in the same narrative sequence, it is necessary to specify a narrative sequence with a smaller scope. For example, in Figure 6, a sequence called "magic duel between Voldemort and Harry" could be used to specify the role of "dueller" for both characters.” We are aware that a simpler link between character, roles, and events may be desirable, but this would require a reification approach like that of Wikidata, or alternative methods like the temporal parts used in the Drammar ontology. For the uses we considered so far, the current design is good enough, but we don’t exclude that in the future we may reconsider this decision.

Comment 7: Please add a table with the prefixes used.

Response 7: added as Table 2 (line 410).

Comment 8: Figure 3: The same class “social relationship” is presented as a gc: G4 Social relationship for the first instance and as a dlp:social-relationship for the second instance. Please indicate the introduced class and its alignment in both instances.
Figure 7: Multiple prefixes, namely “gc:” and “golem:”, probably referring to the same ontology, are used. Please update with a unique prefix or clarify their difference.

Response 8: we edited all figures to improve clarity and correct mistakes.

 

 

Reviewer 2 Report

Comments and Suggestions for Authors

The article presents the ontology GOLEM, an original contribution aiming at representing fictional entities and integrating them with narrative theories. In doing so, GOLEM relies on CIDOC CRM and LRMoo to build its specific classes and aligns them with the foundational ontology DOLCE, contributing to reuse and interoperability.

The article is well written and structured, and the analysis of related works is wide and useful. Also the presented approach looks very promising, and I believe the article could constitute an important contribution to the special issue, provided some minor points are addressed. In the following the authors may find some specific comments aimed at improving the solidity of the article.

 

- P. 1: I don’t understand what the authors mean when they say that DH are addressing the challenge of translating complex concepts into minimal units of analysis and computer-readable formats through interoperability processes. In my opinion, interoperability is a challenge in itself, not a solution to a problem of translation; it is the challenge of exchanging information while preserving as much as possible its intended meaning.

- P. 2: it is not the adoption of OWL 2 that makes the ontologies compatible. Of course, using the same language may be helpful, but what makes the ontologies comparable (not necessarily compatible) is the fact that concepts and definitions are made explicit. Usually this happens when a formal axiomatization is provided, but not necessarily expressed in OWL 2, which, as a language, is not very expressive.

- P. 11: it is not clear to me the choice of aligning fandoms with conceptual object in CIDOC. Why are fandoms considered as non-material cultural entities, if they are communities of fans? Fans are agentive physical objects in DOLCE, aren’t they? If this is not the case, the authors should explain explicitly which is the relationship between the physical agents that are members of the fandom and their being subjects of discourse.

- P. 12: I believe the classification of non-character objects in DOLCE as social objects is correct and the example of the magic wand clearly explains the choice. However, I would add a line specifying that, depending on context (on the specific work) these can be classified either (more commonly) as non-agentive social objects or, like in the example, as agentive social objects. It is not clear, though, in the latter case, how they can be distinguished from characters. I would add a footnote for explanation.

- Pp. 12-13: I understand the difficulty of representing psychological features in DOLCE, as a specific category is missing. I’m not fully convinced by the choice of not aligning them with DOLCE, though. I agree that a quality like bravery can be “perceived and measured” only by observing the behaviour of the entity possessing the feature, but I don’t see a substantial difference with weight, a quality that can be also perceived and measured with some kind of observation. Moreover, I don’t see why Henry’s bravery should not be inhering in Henry like the red of this rose inheres specifically in this rose. I would like the authors either to reconsider the choice, or to explain better the difference between these features and features of a different kind. By the way, aligning also this category to DOLCE would enhance interoperability of the GOLEM ontology. Another suggestion I have for the authors is to consider the possibility of specifying the relation between character (as social object) and description, which now is called is_character_in/has_character, by using the relations specified in the original article on social roles and descriptions:

 

• Claudio Masolo, Laure Vieu, Emanuele Bottazzi, Carola Catenacci, Roberta Ferrario, Aldo Gangemi, and Nicola Guarino. 2004. Social roles and their descriptions. In Proceedings of the Ninth International Conference on Principles of Knowledge Representation and Reasoning (KR'04). AAAI Press, 267–277.

 

There, the relation between social concepts (which are non-agentive in that paper, but the explanation given by the authors suffices for the choice towards agentivity) and descriptions could be either of “use” or of “definition”, where the latter refers to the description where the concept (in this case, the character) is introduced for the first time and the former to the descriptions that talk about a concept/character that has previously been introduced in another description. This would provide the authors with a strategy to distinguish the original primitive work in which a character is introduced and the successive ones, including fan fictions.

• 15-16: for the prosecution of their work, the authors could consider to align psychological states with mental states in DOLCE. There are at least a couple of articles on ontology of mind and intentionality in DOLCE that could be interesting to read.

• 18: I would like the authors to provide an example of a variation of a narrative unit within different contexts. Also, I’m not sure what the authors mean by “the subject” of a narrative function or role. Is it the player? If this is the case, I would use “player” or add a footnote.

 

TYPOS:

- The use of British vs. American English should be made uniform throughout the article (see, for instance, categorised vs. categorized).

- P. 6: the reference(s) at the end of the page appear(s) to be broken.

- p. 12: real-word --> real-world

 

Author Response

Comment 1: - P. 1: I don’t understand what the authors mean when they say that DH are addressing the challenge of translating complex concepts into minimal units of analysis and computer-readable formats through interoperability processes. In my opinion, interoperability is a challenge in itself, not a solution to a problem of translation; it is the challenge of exchanging information while preserving as much as possible its intended meaning.

Response 1: Thank you for the clarification. The use of the term 'interoperability' in line 35 is indeed open to misinterpretation. Interoperability refers to the ability of different information systems, tools, or frameworks to work together while preserving the semantic integrity of the data they exchange. The process of transforming complex literary concepts into minimal, computable units is more accurately attributed to the capabilities of the declarative programming paradigm. This paradigm allows for the explicit formalization and axiomatization of knowledge domains, enabling their operational use in computational analysis. We have accepted your suggestion and revised the paragraph accordingly (lines 35–40).

Comment 2: - P. 2: it is not the adoption of OWL 2 that makes the ontologies compatible. Of course, using the same language may be helpful, but what makes the ontologies comparable (not necessarily compatible) is the fact that concepts and definitions are made explicit. Usually this happens when a formal axiomatization is provided, but not necessarily expressed in OWL 2, which, as a language, is not very expressive.

Response 2: Thanks for the clarification, we integrated it in line 70.

Comment 3: - P. 11: it is not clear to me the choice of aligning fandoms with conceptual object in CIDOC. Why are fandoms considered as non-material cultural entities, if they are communities of fans? Fans are agentive physical objects in DOLCE, aren’t they? If this is not the case, the authors should explain explicitly which is the relationship between the physical agents that are members of the fandom and their being subjects of discourse.

Response 3: We clarified this and added more references from line 462: “A fandom is constituted by debates over canon and fan-produced content, self-reflective engagement of fans who project personal and collective identities onto texts, and creative reinterpretations. Accordingly, we align G15_Fandom with CIDOC CRM’s E28_Conceptual_Object, as fandoms are non-material cultural entities that have become subjects of discourse regarding their identity and origins.”

Comment 4: - P. 12: I believe the classification of non-character objects in DOLCE as social objects is correct and the example of the magic wand clearly explains the choice. However, I would add a line specifying that, depending on context (on the specific work) these can be classified either (more commonly) as non-agentive social objects or, like in the example, as agentive social objects. It is not clear, though, in the latter case, how they can be distinguished from characters. I would add a footnote for explanation.

Response 4: Excellent suggestion, we have integrated it in line 495.

Comment 5: - Pp 12-13: I understand the difficulty of representing psychological features in DOLCE, as a specific category is missing. I’m not fully convinced by the choice of not aligning them with DOLCE, though. I agree that a quality like bravery can be “perceived and measured” only by observing the behaviour of the entity possessing the feature, but I don’t see a substantial difference with weight, a quality that can be also perceived and measured with some kind of observation. Moreover, I don’t see why Henry’s bravery should not be inhering in Henry like the red of this rose inheres specifically in this rose. I would like the authors either to reconsider the choice, or to explain better the difference between these features and features of a different kind. By the way, aligning also this category to DOLCE would enhance interoperability of the GOLEM ontology.

Response 5: Thanks for drawing our attention to an important issue—rooted in Aristotelian metaphysics—regarding the ontological relation of dependence between substances and accidens.  Indeed, at the level of 'being qua being', no fundamental ontological distinction is drawn between psychological, physical, or biographical attributes: they are all accidens and all follow the same principle of ontological dependence, which ties them to the substances they inhere in. That said, our modeling choice was also driven by practical considerations. The full DOLCE-Lite-Plus approach requires modeling qualities as individuals (tropes) and their values as regions, resulting in two layers of representation. In large-scale datasets, this leads to a significant increase in the number of triples, causing scalability and tractability issues. We now adopted the DOLCE-Ultralite (DUL) approach, which pragmatically simplifies this pattern. In DUL, entities can be directly linked to regions, omitting the explicit instantiation of qualities when the focus is on the value rather than on the trope. Consequently, we modeled G2 Feature, including G17 Character Feature and G18 Textual Feature, as dul:Region, representing values in conceptual spaces (e.g., bravery as a region in a psychological space). This choice improves reasoning efficiency and scalability in OWL-based systems, while preserving the option to reintroduce explicit qualities if needed in the future. Accordingly, we have revised the relevant paragraph (lines 502–515) to incorporate this clarification.

Comment 6: Another suggestion I have for the authors is to consider the possibility of specifying the relation between character (as social object) and description, which now is called is_character_in/has_character, by using the relations specified in the original article on social roles and descriptions… There, the relation between social concepts (which are non-agentive in that paper, but the explanation given by the authors suffices for the choice towards agentivity) and descriptions could be either of “use” or of “definition”, where the latter refers to the description where the concept (in this case, the character) is introduced for the first time and the former to the descriptions that talk about a concept/character that has previously been introduced in another description. This would provide the authors with a strategy to distinguish the original primitive work in which a character is introduced and the successive ones, including fan fictions.

Response 6: This is an interesting suggestion but probably not very useful for the domain of literature, especially if we consider oral traditions. We could specify that a character is defined by the first known Work in which it is introduced, but this kind of statements is often debated in philology and literary criticism. A phylogenetic analysis of character variants can be done in any case relying on the publication date of works.

Comment 7: 15-16: for the prosecution of their work, the authors could consider to align psychological states with mental states in DOLCE. There are at least a couple of articles on ontology of mind and intentionality in DOLCE that could be interesting to read.

Response 7: Thanks for the suggestion, we will consider it for the further improvement of the ontology in future work.

Comment 8: 18: I would like the authors to provide an example of a variation of a narrative unit within different contexts. Also, I’m not sure what the authors mean by “the subject” of a narrative function or role. Is it the player? If this is the case, I would use “player” or add a footnote.

Response 8: We clarified this in line 679: “An example of narrative unit is the minimal statement "Orpheus is struck by a thunderbolt," which can refer to one or more events, depending on the myth variant or media expression (Pannach 2023).”

We clarified the second point in line 693: “While the domain of G10_Narrative_Function is a narrative unit, the domain of G11_Narrative_Role is a character.”

Reviewer 3 Report

Comments and Suggestions for Authors

First, the article is well written and organized. The details of the description of the authors' GOLEM framework could use some additional examples and more careful explanation, especially where markers are only subtly distinct from each other, but I was able to gather the basic ideas without much trouble. If the authors are correct in claiming that existing frameworks for coding narrative forms lack the same nuances in their narrative ontologies, then the article's overarching claim to have created a better, more expressive, more supple system is likely correct, too.

Second, as a scholar of philosophy, digital studies, music, and other humanistic disciplines, I do not understand why an article like this should be published. It doesn't do any critical work, it doesn't arrive at any surprising insights, it doesn't advance our understanding of literature or film or even narrative generally. Instead, it is effectively an advertisement for an "ontology" or framework for coding features of diverse narratives. I am not convinced that such frameworks offer much value (see my third point), but even if they do, why should they be advertised in the form of an article in a Humanities journal? Presumably, whatever value the GOLEM system offers is to be found in actually using it, not in describing its features. More suitable for publication would be an exhibition of the actual use of this new framework, perhaps alongside some of the existing, purportedly less effective ontologies. Such demonstration by example might show how GOLEM helps us learn things and compare things that other systems cannot. That GOLEM offers new kinds of discovery and comparison is a central claim of the article, but without seeing it in use, that claim remains abstract and not of interest to a humanist like myself.

Third, to appreciate the value of the GOLEM framework, one must believe in a basic tenet of Digital Humanities, viz., that positivist capture of features of humanist artifacts allows us to see important things about how those artifacts function. My studied intuition is that what is worthy about narrative is precisely what inevitably falls through the cracks of a system like GOLEM. And the authors seem somewhat aware of this dilemma, as their claims for the superiority of GOLEM (versus other existing similar computational narrative ontologies) rely almost exclusively on the idea that GOLEM allows more subtle features to be captured with greater detail and nuance. That is, the authors seem to understand implicitly that what makes a narrative do what it does is extremely complex and subtle, and it lives in the most minute details of that narrative. But as such, no ontology can really get at the heart of narrative because computational ontology by its nature must use reproducible categories, categories that invariably miss what really constitutes the narrative in its unbounded richness. For example, the authors describe a marker that allows them to characterize the transition in the relationship between Ron and Hermione (in the Harry Potter novels) as moving from friendship to romance. But to call this a transition between two "states" of a relationship is actually to look past what makes this episode of the novel narratively effective, namely, the tension between friendship and romance that is irresolvable. It is because Ron and Hermione are not precisely friends nor entirely lovers that their relationship is dramatically interesting, but any attempt to put this equivocation or ambiguity into a category would leave out the fundamental feature, because ambiguity is definitionally what resists category. Or consider the holographic quality of good fictional texts or films: all the small details are also related to all the big themes, which means that (as Roland Barthes pointed out three-quarters of a century ago), the reader is responsible for making the meaning not out of piecemeal units of narrative but out of this mosaic of interrelated "Stoff." To try to capture what is meaningful in a narrative using a digital ontology is tantamount to catching water in a sieve: your sieve will end up wet, but almost all the water will have fallen through.

Author Response

Comment 1: … More suitable for publication would be an exhibition of the actual use of this new framework, perhaps alongside some of the existing, purportedly less effective ontologies. Such demonstration by example might show how GOLEM helps us learn things and compare things that other systems cannot. That GOLEM offers new kinds of discovery and comparison is a central claim of the article, but without seeing it in use, that claim remains abstract and not of interest to a humanist like myself.

Response 1: We agree that examples of actual uses of the ontology would be valuable but without first describing the whole model in detail, it would be hard to use it for scholars other than the authors.

Comment 2: Third, to appreciate the value of the GOLEM framework, one must believe in a basic tenet of Digital Humanities, viz., that positivist capture of features of humanist artifacts allows us to see important things about how those artifacts function. My studied intuition is that what is worthy about narrative is precisely what inevitably falls through the cracks of a system like GOLEM. …

Response 2: Knowledge organization and management is an established field of research, also with documented applications to the cultural heritage domain.