Project

Methodology: Ontology requirements specification document (ORSD)

The main functionalities of the ontology have been defined following the guidelines for Ontology Requirements Specification Document (ORSD) proposed in the paper "How to Write and Use the Ontology Requirements Specification Document" by Suárez-Figueroa et al.

The data used since the beginning in order to provide reliable examples and terminologies taken from concrete description of artworks have been extracted from the following catalogue of the MET museums:

• "Italian paintings. Vol. 2, Venetian School" by Federico Zeri;

By following the template proposed by the authors, we have identified the basic requirements to develop this ontology as exposed in the following table.

1. Purpose: represent the relationship between materials and techniques in art;
2. Scope: the ontology will define artworks at the level of denotative description taking under consideration all the elements that can be related to a material or technique.
3. Implementation Language: it is developed through Protégé which supports OWL and RDF.
4. Intended End-Users: public or private institutions which need to make available information about artworks in Linked Open Data (LOD).
5. Intended Uses: domain experts which wants to search for specific associations of materials and techniques; privates who wants to explore materials and techniques in arts.
6. Ontology Requirements:
   • Non-Functional Requirements: the ontology must be compliant with international standards.
   • Functional Requirements: Competency Questions (CQs) which in the end the user will be able to answer through the queries performed on a SPARQL endpoint.
     In this first draft the CQs have been written in natural language, since they are real questions between human beign, not yet formalized to be understandable by a machine. The expected answers are given according to the data contained in the Zeri catalogue (see above)

     CQ1

     Which technique has been used in the artwork "The Madonna and Child" by the workshop of Giovanni Bellini?

     Expected answer: Temper and oil

     CQ2

     From which material is artwork "The Madonna and Child" by the workshop of Giovanni Bellini made of?

     Expected answer: Wood

     CQ3

     Which are the most common materials on which is performed technique "oil"?

     Expected answer:canvas and wood

     CQ4

     Which are the most common materials in art?

     Expected answer: Canvas; wood

     CQ5

     Which artworks have changed support during their life?

     Expected answer: "Three Saints: Roch, Anthony Abbot, and Lucy" by Giovanni Battista Cima.
7. Pre-Glossary of Terms
   • Terms from Competency Questions: material, technique, artwork.
   • Terms from Answers: oil painting, canvas, wood.

Modelling

All the descriptions in the catalogues of the MET Museum have a well defined structure, with a terminology that can be used to make a first draft of the classes and properties of our ontology.
First, we started by analysing how materials and techniques were expressed in the text and which were their relationships. In the two examples, it is possible to see visually this formalization and its first transposition from real instances to general classes and properties.
As it can be seen, an important element that immediately comes out is that the material and the support are not the same entity, they can be different, i.e. the material on which the artwork is done can be mounted on a different support. From this we move to the definition of the first classes and relationships which are the core of the ontology.
The central realtion from which the ontology has been shaped is the one between technique and material which is defined through the concept that an artwork is realized by means of a technique which is performed on a material.

Previous works

Before moving to the proper design of the OM&TA ontology, it must be underlined that various ontologies have already tried to describe artworks in all their aspects, e.g. CIDOC-CRM. In particular, ArCo Ontology has already a module for denotative description which describes materials and techniques performed on the artworks.
The OM&TA ontology does not overlap with the previous two since it is domain-specific and, on the counterpart, needs to be extended through other ontologies as ArCo in order to describe all the other aspects and definitions of the artwork. As stated till know what changes with the OM&TA ontology is the relation between artwork, technique and material.
In the design of the OM&TA ontology classes will been integrated from the ArCo Ontology if needed equivalences will be pointed out.

Ontology Design and Implementation

The conceptual model

Having defined the very core classes, we move on by creating all the entities and the relationships needed to provide a proper representation of the external data of an artwork.

The first step was the creation of a conceptual model which tries to define the main elements and how they are related. Since materials and techniques are still at the level of denotative description, they are related to all the elements, as "Signature" or "Frame", which sometimes are added through a different technique or material from the one largely used for the artwork.

The Ontology

Starting from the very first theorization, then we proceded by the implementation of the actual ontology.

The elements described in the conceptual model, can now be organized into three main super-classes:

• Artwork_Component: it defines the elements that are somehow part of the artwork, but are not the artwork itself, as the marks, inscriptions etc.;
• External_Physical_Datum: information about elements shaped by the man but still external to the artwork (see again Sciolla's definition of materila)
• Technique: each artwork usually is not made from a single technique and some techniques have precise classifications, e.g. ornamental or decorative.

For completeness, two fundamental classes to define an artwork have been added: Title and Author.

In the image below it is possible to see a complete graphical representation of the final ontology, developed with the Graphical Framework for OWL Ontologies (Graffoo)

---

The ontology was designed on Protégé, an open-source ontology editor. Classes, object properties, data properties, named individuals, annotation properties, general axioms and namespace declarations are described in the documentation.

---

The documentation is accessible as an HTML page thanks to Live OWL Documentation Environment (LODE), which allows to automatically extract and visualize each element of an onotlogy from the OWL file.

Access the Ontology Documentation

Data Extraction and Dataset

Automatic Extraction of Materials and Techniques

Starting from Zeri's catalogue of paintins, a semi-automatic extraction was performed on the text in order to define the main elements: Title, Author, Material and Technique.

The steps to extract the data can be summed as follows:

1. Text preparation: provide a clean text, in a machine-readable format
2. Division into titles and paragraphs: breaking down the text into pieces makes easier to search more precise information. In this case, the division has been performed through Regular Expressions (REs), but other methods can be used according to the internal organization of the text (for example, nlp tools as the python library Spacy provide an automatic paragraph division).
3. Identification of artwork titles and authors: after the identification of titles, through Named Entity Recognition performed by Spacy, it was possible to extract names of authors, while name of artworks where extracted through REs.
4. Extraction of materials and techniques: in order to perform an automatic extraction we used as reference the two skos vocabularies:
   • Thesaurus per la definizione della materia per le opere e gli oggetti d'arte
   • Thesaurus per la definizione della tecnica per le opere e gli oggetti d'arte
   Since both of them where in italian, it was necessary to translate them in English. This operation was automated too through the python library deep translator
5. Cleaning and perfectioning: after the automatic extraction, some operations where needed in order to have reliable results:
   • Automatic translation's errors: some words weren't translated propertly, e.g., "blue" was inserted to the list as a translation for some blue stones. This cleaning step is an exception, since it would not be necessary by using the original skos Italian vocabulary on Italian texts, or by providing an English version of that one.
   • Since the vocabulary does not provide an organization of materials according to the kind of art, some "outsider" materials ended up in the final list. For example, "marble" is not a material appropriate for paintings, but some paintings represents "marble objects". An easy solution and a possible further development of this ontology would be to group materials from the vocabulary under their proper art category, like 'Painting_Material' and 'Sculpture_Material'.
     This is true also for "stucco" which has been correctly detected in the text, but should go under the "Decorative_Techniques" and not the main one.

Final Python code available here

Dataset and Knowlede Graph

In order to test as much classes as possible from the ontology, for 35 artworks additional information (as signature, inscription, support etc.) where manually extracted from the reference text and added to the dataset.

The intersection of the automatic and the manually extracted data has been used in order to populate the knowledge graph (python code available here) Download the Knowledge Graph

Results

Now we can translate the Competency Questions in SPARQL queries that can be performed using OM&TA ontology on a SPARQL endpoint.

The data have been loaded on a local graph database (Blazegraph) on which the queries have been performed to get the results and test the ontology.
The code used to populate the Knowledge Graph and load it into Blazegraph is accessible below

Kknowledge Graph's Python Code

Future developments

The OM&TA ontologies at its status describes the very core relations between the main elements needed to represent techniques and materials in art. Future developments will include the definition of subclasses for the materials and techniques in the art and the inclusion of the unique URIs of the single materials and techniques as defined by the two SKOS vocabularies.