Review Comment:
The paper describes and discusses the GOLEM ontology, which was mainly developed to model fictional characters and their reinterpretations in fan fictions. The paper evaluates the ontology by considering the case study of Greek Mythology in fan fictions, populating a dataset semi-manually and running SPARQL queries corresponding to 10 competency questions.
The first two sections (Introduction, Background & State of the Art) are well written and a pleasure to read.
The third section (Existing Ontological Models for Fictional Characters) is informative and quite comprehensive. However, given the paper's focus, I expected a more detailed and formal comparison of GOLEM with respect to the existing ontologies (maybe in a following section), which is lacking.
The fourth section (Modeling Fictional Character with the GOLEM Ontology) – arguably the central section of the paper — is too imprecise and incomplete to provide a detailed explanation of the GOLEM ontology (see detailed comments below).
The section on the case study is interesting but it lacks some details, especially regarding how the dataset and competency questions has been collected.
The space for adding details could be found by moving the SPARQL encoding of the CQs to an online file, while keeping the explanation of the rationale behind the encoding of the CQs in terms of the concepts and relations included in the GOLEM ontology.
My main concern regarding this paper is how to evaluate its novelty. The authors cite their paper "Pianzola F, Cheng L, Yang X and Panache F (2025) The golem ontology for narrative and fiction. Under review.", which, from the title, appears to be devoted to describing the GOLEM ontology. Without having additional details, it is unclear how the paper submitted to SWJ differs from the cited one, and whether these differences are significant enough. For instance, I don't know if the case study is also present in the other paper. If not, I'm indeed unsure whether this difference is sufficient for publication in the SWJ (see also below my comments on the case study).
Still regarding novelty and originality, the new categories in GOLEM modules described in the submitted paper are: G0_Character_Stoof, G1_Character, G2_Character_Feature, G3_Psychological_State, G4_Social_Relationship, G5_Narrative_Event, G6_Relationship_Role. The authors introduce these notions without precisely positioning their choices within the existing debates in literature and without a precise ontological characterisation: (1) the ontological distinction between G0_Character_Stoof and G1_Character is not clearly stated (see also comments below on that point); (2) G2_Character_Feature seems a trivial reification of a concept or (quality-)type, it is not clear why crm:E55_Type is not enough; (3) G3_Psychological_State, G4_Social_Relationship, G5_Narrative_Event, and G6_Relationship_Role seem restrictions of dlp categories to characters, no additional ontological analysis is provided.
From the methodological perspective, the authors import concepts from quite different ontologies (e.g., DOLCE and CIDOC-CRM). The difficulty of this integration and the extent to which the original meaning of the imported concepts is preserved in GOLEM are unclear. For instance, if I correctly understand, G1_Character is a subclass of both crm:E89_Propositional_Object and dlp:agentive-social-object. Did the authors checked whether the first concept is compatible with the second one?
Importantly, it is not clear in which language the ontology is implemented. The author say that "[t]he finalized controlled vocabulary is implemented as a SKOS (Simple Knowledge Organization System) concept scheme" but for the other modules of the ontology the linked website only contains graphs. Do the authors have an OWL implementation? Have they checked the consistency of the whole ontology? I think a clearer picture of the implementation aspects would be important.
Finally the pages linked in the SPARQL query gd: "https://data.golemlab.eu/entity/" and gt: are empty.
DETAILED COMMENTS
- p.6 "GOLEM is designed as an extension of CIDOC-CRM and LRMoo with alignment to DOLCE-Lite-Plus."
What exactly does "extension" mean? Does GOLEM import the whole CIDOC-CRM and LRMoo? What exactly does alignment to dlp mean? How are formal consistency and conceptual/ontological coherence guaranteed? Please add technical details.
- p.6 "G1 Character that extends CIDOC CRM’s E89 Propositional Object (...) G1 Character enriches this model by incorporating DOLCE’s social object framework. This dual classification acknowledges characters as socially constructed entities whose existence depends on people’s imagination within narrative communities." (and following)
In what sense does G1 Character extend E89 Propositional Object? If G1 is intended as a subclass of both E89 and dlp social agentive object, how can a propositional object be agentive? What is a dual classification? What does it mean "grounding characters in E28 while extending their classification to DOLCE’s socially constructed agents"? Please add technical details.
- p.6 "For significant objects showing some agency (e.g., magical artifacts), we employ G16 Object as a subclass of DOLCE’s social-object"
From an ontological point of view, what distinguishes G16 Object from DOLCE’s social-object?
- p.7 Figure 1 (erroneously referred as figure 3 in several places)
(A) If this is a class diagram I don't understand why several classes have been duplicated. The caption states "examples of related characters" but I can't see any examples or instances of characters.
(B) The prefixes of the names of the concepts/classes must be explained. In particular, I don't understand what gd: refers to.
(C) The intended semantics for the adopted graphical language has not been introduced. In particular, it is unclear why some boxes are divided into three and other into two. What is the semantics and the purpose of this division?
(D) Why is the reification of types necessary? Why are two subclasses of G1_Character (FanfictionCharacter and OriginalCharacter) not sufficient?
(E) Why features are not types?
(F) It is unclear to me why the instances of Character_Stoff do not have features; this would enable the relation features_are_also_found_on to be defined in terms of the features attributed to the stoffs and to the characters. Is features_are_also_found_on also defined between characters? If not, why?
(G) It is unclear whether a Character can be associated with only a work via is_character_in [this is also linked to CQ1 that is described as "cross-narrative identity tracking"]. Why are cardinality constraints not considered in the diagram?
If the work is unique, I don't understand the claim at p.1 "the same character often exists across disparate narratives, including canonical works, fan adaptations, and transmedia expansions, each offering conflicting or evolving traits, relationships, and backstories". Do the authors here refer to the stoff?
- p.7 "we introduce the G2 Feature class as an overarching category for attributes, with subclasses that reflect the diverse nature of these textual or hermeneutic elements, which can either be found in the narrative or be the outcome of reader interpretation."
In which sense features are textual or hermeneutic elements? I can't see how the reader interpretation is represented in the framework.
- p.7 "This epistemological distinction necessitated a departure from DOLCE’s conventional quality-based approach"
As far as I understand, the qualities of DOLCE can have an epistemological nature, e.g., colors. I don't see how the epistemological dimension of features is represented in GOLEM. Furthermore, according to table 1, the name is a feature of a character. A clarification on the nature of features would be useful (also explaining why clothes is included in physical description)
- p.8 "this fundamental characteristic can be radically transformed in alternative interpretations while the character is still being recognized as a variant of Harry Potter."
On the what basis the character is recognised as a variant of Harry Potter? Just because it is still named "Harry Potter"? I expected more discussion on this crucial point, which is widely debated in the philosophical literature on the identity of characters.
- p.8 "As illustrated in Figure 3, this approach allows conflicting or divergent traits to coexist within the same overarching character concept."
I can't see how figure 1 illustrates this.
- p.8 I would move section "Controlled Vocabulary Construction of Character Features" to the case study section. To me, it seems to be more about how the ontology is populated than the (structure of the) ontology itself.
- p.8 "The G17 Character Feature class captures qualitative character attributes through a bottom-up approach to controlled vocabulary construction."
Per se, character features do not imply a bottom-up approach. This seems to be an additional assumption of the proposed framework. Furthermore, I'm not sure to understand the strategy of developing a controlled vocabulary that covers all possible character features. In a bottom-up perspective, why not focus on the features necessary for the instances of the G1_Character class considered in the examples and adding new features as needed for new instances of G1_Character?
- p.10 Figure 3
(A) It is unclear to me whether narrative events only include characters and objects as participants. Consider, for example, the London in the Sherlock Holmes novels. Is London an instance of G16_object?
(B) Again, why must the types of events be reified rather than introducing different subclasses of G5?
(C) As far as I know dlp, relations dlp:follows, dlp:precedes, etc. have a temporal nature. It is unclear if the G7_Narrative_ Sequence class is characterised in terms of these relations, which do not encode causality. Furthermore, why are these relations defined between narrative events rather than time intervals?
(D) I don't understand why the duration and the temporal location of a narrative event are both dlp:time-intervals. It seems to me that duration is a quality of temporal locations / time intervals. Is it possible to have two different events with the same temporal location but different duration?
(E) Why psychological states do not have a temporal location and a duration (but they have temporal relations)?
(F) Why is the relation between characters and psychological states has_state rather than simply participation?
- p.10 "A psychological state represents a temporal mental condition, characterized by relatively stable qualities such as emotions, motivations, beliefs, and goals."
Are beliefs, goals, etc. qualities of psychological states or are they specific kinds of psychological states? I don't understand how they are represented.
- p.11 "By structuring competency questions (CQs) as SPARQL queries, we assess GOLEM’s capacity to represent character identities, relationships, and evolving traits, revealing its advantages over other frameworks."
First, I'm unsure if CQ1-CQ10 can be classified as competency questions. The authors did not collect them from domain experts; they just stated them. In this case, I would say they are examples of queries that can be done in the proposed framework.
Secondly, I don't see how answering these CGs reveals the advantages of GOLEM over other frameworks. A serious comparison would require proving that such CQs cannot be expressed and/or answered in other frameworks.
Thirdly, the reader would benefit from seeing some concrete examples of how "evolving traits" can be represented in the proposed framework. The authors later refer to "enabling roles like “lover” and “beloved” to be temporally bound" via Social Relationships. A more explicit example would be useful. Furthermore, it is unclear how the evolution of traits such as height, weight, personality, etc., which are represented by features, is captured.
|
