Semantic Web [Berners-Lee et al. 2001] advocates the representation of information on the Web in a structured, machine-interpretable format, thereby allowing automated reasoning, i.e. enabling computers to detect inconsistencies or infer new information from a set of asserted facts, and advanced added-value services on the uploaded data. The Semantic Web and related semantic technologies build on ontologies, which provide a means to formally define the basic terms and relations that comprise the vocabulary of a certain domain of interest. As such, they essentially represent the schema that the associated data descriptions should follow. To be usable in the uncontrolled Web environment, ontologies, as well as their associated data, need to be represented in a uniform, platform-independent, machine-interpretable and semantically-enriched format, using a generic language. This need is covered by the standard data model for representing information in the Semantic Web, RDF, which models data using triples (s, p, o), denoting that a subject (s) is associated with an object (o) via a property (p). The RDF Vocabulary Description Language, RDFS,[3] is an extension of RDF that enables defining domain-specific semantics, using simple constructs (classes, properties and individuals) and relationships (class and property subsumption). RDFS also supports some basic types of reasoning, via the transitivity of the subsumption hierarchies, and its propagation to the classification of individuals into classes/properties. For even more sophisticated semantics and reasoning, OWL and its more recent version, OWL2,[4] have become dominant. These allow more expressive semantics, such as the specification of custom properties for relations (e.g. transitivity or symmetry), existential or universal quantifiers, operations among classes (union, intersection, complement) and others. These languages have been widely used to model information in different domains, including health, law, culture, media, geography, e-governance, life sciences and cultural heritage (e.g. such as the ReDD-Observatory project; the Linked Jazz research project; GoodRelations ontology; the BBC Web; the Research Space project and the Open Government Data).[5]

The international standard for the representation and exchange of CH information is the Documentation of the International Council of Museums (CIDOC) Conceptual Reference Model (CRM). This ontology facilitates “the integration, mediation and interchange of heterogeneous cultural heritage information”  [Le Boeuf et al. 2015, 1]. The “GenderedCHContents” ontology builds upon the EDM, which collects, connects and enriches the descriptions provided by Europeana content providers. EDM is aligned to CIDOC-CRM in its definition of an event-centric model. In order to implement its tasks, the EDM uses two main categories. Firstly, it re-uses elements from other namespaces; specifically, from the RDF and the RDFS namespaces, the Open Archives Initiative Object Reuse and Exchange (OAI-ORE) namespace, the Simple Knowledge Organization System (SKOS) namespace, the Dublin Core namespaces for properties from the elements, terms and types namespaces, the W3C Data Catalog Vocabulary (DCAT) namespace, and the Creative Commons (CC) namespace.[6] Secondly, it uses the elements introduced by EDM [Europeana 2014, 8].

Despite the development of these ontologies, as Doerr comments, museums do not show an interest in using discrete schema elements to analyse iconographic representations [Doerr 2009, 474]. Similarly, Damiano and Lieto underline that there is a contradiction between the huge quantity of resources published on the Web in recent years, and the inadequate way that these resources are described. As they note:

the way they are described is far from meeting the requirements of content-based access required by the general public: neither their description in terms of editorial metadata nor the tags added by users seem adequate to describe the content of media resources, and fall short of providing an effective access to digital media. [Damiano and Lieto 2013, 76]

By developing the “GenderedCHContents” ontology, we suggest that a content-based description of artefacts is essential, especially in the field of CH, in order to facilitate the retrieval of gender-related concepts and encourage different interpretations of CH artefacts. As Damiano and Lieto further mention, media studies show that users, when tagging artworks, tend to describe the content of the artworks [Damiano and Lieto 2013, 76]. Therefore, a description of the content of CH artefacts could be useful to users in order for them to obtain a more holistic idea of the represented artworks or even question and rethink represented gender norms and beliefs.

Different story ontologies have also been developed recently, aiming to model narrative concepts. For example, the BBC’s “News Storyline” Ontology is a generic model that describes and organises the stories that news organisations present, but it is focused on the representation of news rather than on CH artwork stories.[7] On the other hand, the Contextus Project aims “to allow people to explore and analyse narratives [in heterogeneous media] through the annotation of those works with semantic descriptions”  [Describing Narrative in the Digital World n.d.]. Further, the “Stories ontology” was developed in collaboration with the BBC, “to create an ontology for narrative representation that could be applied across a diverse set of cases”  [Linked Open Vocabularies n.d.]. Moreover, within the same project, the “OntoMedia” ontology was developed in order to represent heterogeneous media by semantically describing their content. Thus, its scope is to enable users “to have easy access to the type of details that previously would have been impractical to search for due to the sheer effort it would have taken to collect and correlate”  [OntoMedia n.d.].

Another example, the Labyrinth system, which is targeted at CH dissemination and digital publishing, focuses on the notion of “cultural archetype” and is based on the “Archetype ontology” [Damiano and Lieto 2013, 77]. This ontology has been designed to support reasoning “on the relations among characters, actions and stories, while abstracting from different genres and media types”  [Damiano and Lieto 2013, 77]. Its role-based schema is closer to the aims of this study, since the focus of the “GenderedCHContents” ontology is centred on the concept Entity (Person / Animal / Object / Demigod / Deity). The current study, therefore, was influenced by Archetype’s aim but did not follow the same developing strategies in defining the basic concepts and properties because the central role here is found within entities and their gender relations rather than on the stories.

The “GenderedCHContents” ontology contains twenty-two classes which form the conceptual representations of the CH artefacts’ gendered content (see Figure 4). These extend the EDM by adding twenty-two new classes, defined as subclasses of E89_Propositional_Object, which EDM reuses from CIDOC CRM and it is equivalent to edm:InformationResource.

The class E89_Propositional_Object, as defined by CIDOC CRM, comprises “immaterial items” like “stories, plots, procedural prescriptions, algorithms, laws of physics or images that are, or represent in some sense, sets of propositions about real or imaginary things and that are documented as single units or serve as topic of discourse”. Moreover, it “comprises items that are ‘about’ something in the sense of a subject”  [Le Boeuf et al. 2015, 33–34]. Similarly, edm:InformationResource “is a resource whose essential characteristics can be conveyed in a single message” and it has a URI and a representation [Europeana 2014, 14].

We extended the EDM with a new class named Content (see Figure 4), which we defined as a subclass of E89_Propositional_Object, for it provides a description of the content of CH artefacts. This new class represents the thematic topics that each resource is related with. In the case of literary resources Content refers to “what is said in a literary work, as opposed to how it is said”  [Baldick 2008]; and in the case of visual resources, to describe what is depicted on them. The thematic topics are usually the results of an interpretation of the meaning of CH artefacts, which takes into account their historical and cultural context.

Although each content consists of a narrative, we decided not to include the term “narrative” in the class Content in order to emphasise that content contains a narrative which is expressed through some medium, in this case an image or a text. Thus, by using the term Content, we applied the more general concept, which denotes the connection with the form and context of CH artefacts.

We added three new classes, Non-representationalContent, RepresentationalContent and SignifiedConcepts, as subclasses of Content (see Figure 4). The first two are used to distinguish artefacts based on the structure of their contents. Non-representationalContent refers to abstract and non-figurative art or literary description which does not aim to reference reality, whereas RepresentationalContent refers to depictions of figures, etc. This distinction is not absolute, in the sense that an artefact may be associated to more than one of the subtypes of content.

The class SignifiedConcepts borrows the “term” signified from Semiotics [Saussure et al. 1986] and is used to connote meanings that entities may carry and, in this case, meanings related to gender symbolisms. We defined this class as a subclass of Content to show that every Entity, which is a subclass of Content, can potentially refer or stand for something other than itself (see Figure 5). This class enables the representation of metaphors related to women and, thus, an illustration of these symbolisms through the meanings that the concepts convey. For example, HesiodsTheogonyPandora isAssociatedWith Laziness (see Section Ontology Validation and Demonstration, Metaphors). We declared all other classes of our ontology (e.g. Entity, Body, EmbodiedActivity etc.) as subclasses of RepresentationalContent, since this superclass contains all the representational entities which convey descriptive characteristics about gender.

The core concept of this ontology is the class Entity, which is the “starting point” for detecting, describing and accessing figures of women within a content. Entity includes the subclass Person, which represents all women’s figures. Entity contains all entities that have distinct and independent existence, being and form. Therefore, Animal, Object, Deity and Demigod are also subclasses of this class. We declared Person as equivalent to foaf:Person that EDM reuses from Friend of a Friend (FOAF): “the Person class represents people. Something is a Person if it is a person. We don't nitpic about whether they're alive, dead, real, or imaginary”  [FOAF Vocabulary Specification n.d.].[14]

Unlike FOAF, our ontology differentiates between real and imaginary people by using two subclasses of Person: RealPerson and UnrealPerson. The first refers to people who lived in real life – there is historical evidence of their existence – whereas UnrealPerson refers to all characters and figures that are fictional, mythological and imaginary. With this grouping we aim to reveal different ideas about characteristics and roles that historical and imaginary entities represent in specific contents. The ontology can be further extended in the future by adding properties which can demonstrate similarities and differences between the two classes. We acknowledge that the categorisation of CH artefacts can be a complex process, in that for example, it is not historically clear if the ancient Greeks believed in their myths [Veyne 1988] and if Pandora was seen as a historical or as a mythological figure in Hesiod’s period, or even, whether she was perceived as a mythological figure in the classical period. The ontology offers the flexibility to support all these different presumptions by enabling an individual (e.g. Pandora) to belong to one of or both subclasses of Person (RealPerson and UnrealPerson), demonstrating the complexity of interpreting the content of CH resources.

Deity class refers to supernatural beings and conveys ideas about the sacred. Since “images of deities reflect and shape understandings of sexual difference and the cultural meanings attached to biological sex”, this class is used here as a resource that can illustrate beliefs and cultural visions, providing insights into social behaviour [O’Brien 2009, 191]. Moreover, Demigod is defined as a class that has partially divine and partially human characteristics. The Object class refers to all material things and the Animal class to all living organisms which feed on organic matter.

Further, we added the class Body and its subclass Posture. This refers to the position of the body, to gestures and to bodily attitudes, to represent “norms and ideals about standard body shapes [that] are culturally expressed in intentional and unintentional ways”  [O’Brien 2009, 82]. Body refers to the physical structure of a living entity. Also, since “cultural, economic and political factors are literally visible on the bodies of society members”  [O’Brien 2009, 86] we added the classes EmbodiedAppearance and EmbodiedActivity to examine and analyse bodily representations of entities and their actions.

EmbodiedAppearance, which refers to the way a body looks, has three subclasses: Accessories, Hairstyle, and Dress (see Figure 4). It designates “gender-differentiated body conceptions”, which reflect a “conception of men and women as essentially different”  [O’Brien 2009, 86]. This class aims to represent different appearances in order to demonstrate how human bodies can vary, aiming to question the “absolutist classifications of dominant sex/gender understandings”  [O’Brien 2009, 86–87]. Accessory refers to all things that complement a person’s appearance, Hairstyle to the way in which an entity’s hair is cut and arranged, and Dress to the clothing that entities wear. To describe any type of appearance that does not fall under one of the three subclasses (e.g. body painting), one can either directly associate it to the main class (EmbodiedAppearance), or extend the ontology by adding another relevant subclass.

EmbodiedActivity refers to the performance of an entity, “engaging in a set of behaviours that display an individual’s gender identity as female or male and that cause the individual to be so perceived during social interaction”  [O’Brien 2009, 366]. GenderedTraits pertains to the characteristics ascribed in the sources to a specific entity. It aims to represent these traits and relate them with an entity’s identities. GenderIdentity represents gender identities as described by or as performed in the resources. By creating the class GenderIdentity we aimed to include all categorisations that have been historically made and, based on these, to represent all the different relations between identities, roles, emotions and embodied activities. Further, the GenderRoles class refers to the distinct social roles assigned to men and women. “Gender role” is understood as a set of expected actions and dispositions ascribed to an individual on the basis of her or his assumed biological sex [O’Brien 2009, 371]. GenderedSkills refers to stereotypically gendered activities that genders may perform (such as cooking being female, fighting being male). Finally, the class GenderedEmotions represents the feelings that derive from an entity, which “might be symbolically gendered even if men and women do not manifest it differently”  [Anderson 2015].

The second step in developing the ontology was to determine the relationships within the ontology. For this purpose, we created sixteen object properties (see Figure 6).

All properties except isAssociatedWith, isSymbolFor and symbolises, have Entity as their domain. This choice underlines the focus on the Entity class to represent the gendered characteristics, roles, emotions, etc., which this class relates with. In order to represent the appearance of an entity, we added the hasAppearance property with EmbodiedAppearance as range. The hasAppearance has two subproperties: hasHairstyle and wears, with ranges Hairstyle and Dress respectively, to demonstrate the physical characteristics and appearance of the figure. The bears property also contributes to the descriptive representation of the entity, since it has Accessory or Entity as range, describing any accessories (e.g. bags) or living beings (e.g. babies) that someone may bear. The poses property, with Posture as range denotes movements of the body, gestures, body position and attitudes connected to the body. To describe an activity that an entity is engaged in, we added the property performsActivity with EmbodiedActivity as range. We also added the hasGenderRoleOf property, with GenderRoles as range, to describe an entity’s social role. To demonstrate the different GenderedSkills, GenderedEmotions, GenderedTraits and GenderIdentity of an entity we added the four following properties (which respectively have the above classes as their ranges): HasSkills, hasEmotions, isCharacterisedAs, hasGenderIdentity. The relation between different entities can be described using the relatesWith property, which we defined as symmetric since it is equivalent to its inverse [Antoniou et al. 2012, 106].

In order to represent the symbolisms that concepts entail, we created the symbolises property with Content as domain and SignifiedConcepts as range. This means that each entity that is part of a content potentially denotes a specific meaning. Further, in order to connect the symbol/signifier to the signified idea/meaning, we added two more properties which are inverse to each other: isSymbolFor and isSymbolisedBy. The latter has Entity as domain and Content as range (and vice versa for isSymbolFor). With this model, entities that have the role of symbols in specific contents and entities that are described by symbols are equally represented. In order to connect an entity that has a symbolic/metaphorical meaning with a concept that is related with the same meaning, we defined the isAssociatedWith property with SignifiedConcepts as domain and Entity as range. The ontology also reuses some of EDM’s object properties in order to demonstrate the subject of the resource and to relate it with other similar resources.

In addition to the object properties, we also added seven datatype properties (see Figure 7) that associate instances of the class Entity to certain data values (e.g. numbers, strings, or dates). The domain of all these properties is, therefore, Entity and the range is rdfs:Literal, which is a built-in RDF class.

Specifically, hasRelationToGenderNorms denotes whether a certain entity, through its embodied activity, appearance and role, complies with the gender norms ascribed to the individual’s gender within its historical framework. This categorisation can be made by consulting secondary sources which analyse stereotypical behaviours and gender norms in the period in which an entity is presented (e.g. for women’s roles in the ancient world, James and Dillon’s [James and Dillon 2015] and Cohen’s [Cohen 1994] works can be consulted). This property enables a representation of women’s figures that repudiate specific norms and roles. It takes either “complied” or “not complied” as its value. In this way, the ontology can be seen as a research resource for identifying the changes between following or changing the ascribed gender norms in different CH artefacts.

Another important aspect that we attempt to implement with this ontology is the association of a figure with its social circumstances in order to show the link between social, economical and political factors that may influence and partially determine a figure. For this reason, we added the properties hasSocialStatus and hasSocialClass, which describe an individual’s social position and reflect the roles, activities and identities that an individual holds within a social system. The acceptable values for hasSocialStatus are: “has status”, “no status”. This simplified model aims to be as inclusive as possible. Therefore, by denoting whether an entity has a status or not, we can group entities based on this characteristic. We acknowledge that in some situations this is not applicable or may be seen as an absolute classification. However, as a first step it would be interesting to gather figures that apply to this model and, as a second step, reconsider the use and naming of the specified values. The same applies to the property hasSocialClass since the only values it accepts are: “lower”, “middle”, “upper”, and “not applicable”. The limitation of this design is that these values classify an individual with values that represent social systems from the eighteenth century onwards, reflecting capitalist societies, whereas structures of earlier societies are not represented.[15] For this reason, we included “not applicable” as an acceptable value, to demonstrate that this is not the only hierarchical classification, and that it cannot be universally applied.

Another aspect captured by the ontology is the way that an entity is associated with health. We made this design choice because there is a broad literature referring to women and health, and how women are connected to mental disorders like hysteria, etc. in different historical periods (see, e.g. [Annandale 2008], [Tasca et al. 2012], [Gilman et al. 1993], [King 1998], and [King 2009]). The property hasStateOfHealth with acceptable values “healthy”, “mentioned illness”, “not mentioned” and “unhealthy”, aims to classify each individual according to a specific state of health as represented within the resource. Therefore, it enables the retrieval of cases where mentioned/recognisable illness or representations of women in unhealthy states are displayed. This will make it possible for researchers to identify the kind of illness that a figure is illustrated with and reflect on ideas about health and body.

The study takes into account the concept of “the space–time fixity constraint that binds activities or series of acts to a specific place and moment in chronological time”  [Schwanen et al. 2008, 2109]. It also takes into consideration attempts to deconstruct “dominant understanding(s) of time through, for example, drawing our attention to “women’s time” as embodied in daily life”  [Odih 1999, 10]. The term “relational time” was introduced in feminist discourse to describe “an experience of time that ‘exists’ in relation to an embeddedness in embodied social relations”. Also, since it is “mediated through significant others”, as Odih notes, “relational time is shared rather than personal and thus sensitive to the contextuality and particularity of interpersonal relations”  [Odih 1999, 10]. In order to represent the space-time in which a character/entity exists, as presented in the sources, we added two datatype properties: actsWithinSpace and performsWithinTime. The former enables the semantic annotation of space and has rdfs:Literal as its range. It, therefore, accepts any used-defined values that denote space. Some possible values are: “abstract”, to represent a non-representational space/place; “indoors” or “outdoors”; “private” or “public”; and “specified” or “non-specified” to denote whether there is a recognisable named place. The use of terms “public” and “private” aims to incorporate gendering theories of spaces[16] – when this information is transparent – and map the spaces where entities are presented. This is not an exhaustive list; depending of the content, new values can be added to describe space efficiently. Moreover, actsWithinSpace has actsWithinGenderedSpace as a subproperty to allow the semantic annotation of gendered spaces (i.e. women’s quarters in households, gender-specific restrictions in baths and temples, etc.) where this information is transparent within the artefact. Currently, performsWithinTime accepts the following values: “abstract”, when time is not specified; “ancient period” (3600 BC – 500 AD), “medieval period” (500 – 1500), “early modern period” (1500 – 1750) and “modern period” (1500 – present) to specify the historical time period in which an entity is presented; “day” and “night”; “narrative past”, “narrative present” and “narrative future” to represent cases where an entity is clearly presented within a narrative and her/his actions, descriptions etc. refer to an explicit past, present or future of narrative time. The ontology provides the flexibility to add new values for representing more fine-grained dates.