a. Coverage studies
Thesaurus coverage was calculated for different versions of SILKNOW thesaurus. Data in this article is based upon SILKNOW thesaurus 1.8 version. To better understand how the developed thesaurus covers the resources that we have included in the project, we have performed an analysis of the final version of the thesaurus that will be available on the web, regarding coverage in different languages and number of unused terms.
4.1 Usage of the thesaurus concepts in the museums
The SILKNOW thesaurus was validated on textual data of online resources on the full thesaurus in all four languages it covers. The frequency of the individual thesaurus concepts that are present in specific online resources was calculated. English, French, Italian and Spanish translations of the thesaurus were each compared to resources in the corresponding language. The program for the calculation of coverage was written in Python. Preprocessing was carried out using the Natural Language Toolkit library which contains the Snowball Stemmer. It was used to convert all the terms and their synonyms from the thesaurus, as well as all the words from online resources, to their stem. This enabled us to compare two words and determine whether they have the same stem. The output of the coverage calculation is a table, where each row contains a term, its synonyms and the number of times the term was used in one of the resources. From this, the percentage of thesaurus concepts that can be found in the specific resources was calculated. The exact numbers we have arrived at are given in Table 4, and the associated percentages are shown in Figure 1.
The English part of the SILKNOW thesaurus contained at the moment of this experiment, 596 concepts. Out of these, 72 do not occur in any of the English museums that we have considered. These include: cannetillé, aceytuni, alberoni, alcarchofado, fibroin, pill, lammela, monture, etc.
Table 5
Coverage of SILKNOW thesaurus concepts in English museums.
Museum
|
Overlapping terms
|
Coverage of the thesaurus concepts
|
IMATEX
|
378
|
63.42%
|
Metropolitan Museum
|
413
|
69.30%
|
Museum of Fine Arts Boston
|
489
|
82.05%
|
Rhode Island School of Design
|
405
|
67.95%
|
Victoria and Albert Museum
|
501
|
84.06%
|
All museums together
|
524
|
87.92%
|
The French part of the SILKNOW thesaurus contained 611 concepts. Out of these, 85 do not occur in any of the French museums that we have considered. These include: artichaut, pélican, etoile, mante, ourdisseur, tercianela, nobleza, filoselle etc.
Table 6
Coverage of SILKNOW thesaurus concepts in French museums.
Museum
|
Overlapping terms
|
Coverage of the thesaurus concepts
|
Joconde
|
292
|
47.79%
|
Musée des Arts Décoratifs
|
382
|
62.52%
|
Musée des Tissus de Lyon
|
521
|
85.27%
|
All museums together
|
526
|
86.09%
|
The Italian part of the SILKNOW thesaurus contained 436 concepts. Out of these, 200 do not occur in any of the Italian museums that we have considered. These include: mantello, collarino, arazzo, restagno, piombo, tulle, sericina, etc.
Table 7
Coverage of SILKNOW thesaurus concepts in Italian museums.
Museum
|
Overlapping terms
|
Coverage of the thesaurus concepts
|
UNIPA
|
236
|
54.13%
|
All museums together
|
236
|
54.13%
|
The Spanish part of the SILKNOW thesaurus contained 629 concepts. Out of these, 113 do not occur in any of the Spanish museums that we have considered. These include: zarza, frappé, tusor, devanador, rasete, sericina, lisaje, estofar, etc.
Table 8
Coverage of SILKNOW thesaurus concepts in Spanish museums.
Museum
|
Overlapping terms
|
Coverage of the thesaurus concepts
|
CERES
|
396
|
74.30%
|
IMATEX
|
361
|
67.73%
|
All museums together
|
414
|
77.67%
|
To investigate which terms from the thesaurus are the most frequently used in our resources, we provide word-cloud visualization. As shown in Figure 2, some of the most frequently found thesaurus concepts in Metropolitan museums are: hard silk, spun silk, shot cloth, coarse silk, half silk and floss silk. The list of the most frequent concepts in the Museum of Fine Arts, Boston is very similar: hard silk, spun silk, shot cloth, coarse silk, tabby and bourette yarn (Figure 3).
Some of the most frequently found thesaurus concepts in the Rhode Island School of Design are hard: silk, spun silk, shot cloth, coarse silk and bourette yarn (Figure 4), while some of the most frequently found thesaurus concepts in the Victoria and Albert Museum are: hard silk, spun silk, shot cloth, coarse silk and half silk (Figure 5).
Interestingly, the most frequent thesaurus concepts that occur in French resources are quite different from the most frequent English concepts. As shown in Figure 6, some of the most frequently found thesaurus concepts in the Joconde database are: ruban, bande, soie, ruban brodé, ceinture and rayonne. These French concepts translate into English as: ribbon, loop, silk, embroidered ribbon, girdle and rayon. As shown in Figure 7, some of the most frequently found thesaurus concepts in the Musée des Arts Décoratifs are: pré-production des textiles, laize, soie, soie ouvrée, tissu and brocher. These French concepts translate into English as: fabric pre-production, width, silk, thrown silk, cloth and brocade. Notice that the range of frequencies of individual concepts is low, resulting in less variable font sizes.
Some of the most frequently found thesaurus concepts in the Musée des Tissus de Lyon are: velours de Lyon, satin de Lyon, crêpe de Lyon, soie, chaîne pièce, fond (Figure 8). In English these are: Lyon velvet, Lyon satin, Lyon crape, silk, main warp and ground. Some of the most frequently found thesaurus concepts in UNIPA are: fondo, filo, filato di seta, trama di fondo, trama lanciata and filo di giro (Figure 9). These Italian concepts translate into English as: ground, yarn, silk yarn, main weft, pattern weft, doup end.
Some of the most frequently found thesaurus concepts in CERES are: seda, seda cruda, seda salvaje, hilo de seda, tejido, tejer and motivo zoomórfico (Figure 10). These Spanish concepts translate into English as: silk, hard silk, wild silk, silk yarn, cloth, weave and zoomorphic or animal motif. As can be seen in Figure 16, some of the most frequently found thesaurus concepts in IMATEX are: tejer, tejido, seda cruda, seda salvaje, labrado and diseño (Figure 11). These Spanish concepts translate into English as: weave, cloth, hard silk, wild silk, figured weave and pattern design
4.2 Word and phrases in the museums not covered in SILKNOW thesaurus
For each online resource, a feature vector representing all its phrases was computed. The result was a set of n-grams with a maximum size of three words and a corresponding number of occurrences. From them, a subset was generated where all the concepts that can be found in the thesaurus were removed from the feature vector. This subset is referred to as online resource phrases without thesaurus overlap.
As shown in Figure 12, some of the most frequently found Metropolitan Museum phrases not included in the SILKNOW thesaurus are: fragment, sample, sampler, canvas, panel, border and gallon. These, as well as less frequent phrases, are mostly represented by common everyday words or easily understandable terms. From this we can presume that the SILKNOW thesaurus covers most domain specific terminology used in the Metropolitan Museum dataset.
Some of the most frequently found Boston Museum of Fine Arts phrases not included in the SILKNOW thesaurus are: white, yellow, black, green, red, coloured and flowers (Figure 13). These, and less frequent phrases, are mostly represented by common everyday words or easily understandable terms. From this, we can presume that the SILKNOW thesaurus covers most domain specific terminology used in the Boston Museum of Fine Arts dataset. Additionally, we can see that descriptive words like colours and materials are often used. Traditional floral and vegetal motifs are also evident from the high frequency of the phrases, flowers and leaves.
Some of the most frequently found Rhode Island School of Design phrases not included in the SILKNOW thesaurus are: metallic, fragment, beads, material, glass, sample, polyester and plastic (Figure 14). These, as well as less frequent phrases, are mostly represented by common everyday words or easily understandable terms. From this we can presume that the SILKNOW thesaurus covers most domain specific terminology used in the Rhode Island School of Design dataset. Additionally, we can see that different materials like metal, plastic, glass, nylon, polyester, crystal and leather are often used to describe the design of individual items. Some of the most frequently found Victoria and Albert Museum phrases not included in the SILKNOW thesaurus are: woven, yellow, red, white, green, border, century (Figure 15). These, as well as less frequent phrases, are mostly represented by common everyday words or easily understandable terms. From this, we can presume that the SILKNOW thesaurus covers most domain specific terminology used in the Victoria and Albert Museum dataset. Additionally, we can see that different materials, colours, techniques and time related words are used to describe, in detail, the design and motifs of individual items.
Some of the most frequently found Joconde phrases not included in the SILKNOW thesaurus are: couleurs, déposé, décor, fleurs, roses and tissé (Figure 16). These French phrases translate into English as: colours, deposit, decor, flowers, roses and woven, and are mostly represented by common everyday words or easily understandable terms. From this we can presume that the SILKNOW thesaurus covers most domain specific terminology used in the Joconde dataset. Additionally, floral motifs are also evident from the high frequency of the phrases fleurs and roses which translate as flowers and roses. As can be seen in Figure 17, some of the most frequently found Musée des Arts Décoratifs phrases not included in the SILKNOW thesaurus are: morceau, planche bois, échantillon, décor and couleurs. These French phrases translate into English as: piece, wooden board, sample, decor and colours. Data is mostly represented by common everyday words or easily understandable terms. From this we can presume that the SILKNOW thesaurus covers most domain specific terminology used in the Musée des Arts Décoratifs dataset.
As shown in Figure 18, some of the most frequently found Musée des Tissus de Lyon phrases not included in the SILKNOW thesaurus are: musée, fabricant, fleurs, saint, étoffe and maison. These French phrases translate into English as: museum, maker or manufacturer, flowers, saint, fabric and house. Data is mostly represented by common everyday words or easily understandable terms. From this, we can presume that the SILKNOW thesaurus covers most domain specific terminology used in the Musée des Tissus de Lyon dataset. Additionally, floral and sacred motifs are also evident from the high frequency of the phrases: fleurs, saint, sainte which translate as flowers, saint and holy. As can be seen in Figure 19, some of the most frequently found UNIPA phrases not included in the SILKNOW thesaurus are: trame, capi, colore, fili, riduzione, secolo and prodotto. These Italian phrases translate into English as: weave, leaders, colour, thread, reduction, century and product. Data is mostly represented by common everyday words or easily understandable terms. From this, we can presume that the SILKNOW thesaurus covers most domain specific terminology used in the UNIPA dataset. Additionally, sacred motifs are also evident from the high frequency of the phrases: chiesa, san, verde, tralci and foglie, which translate as: church, saint, green, vines or branches and leaves.
As can be seen in Figure 20, some of the most frequently found CERES phrases not included in the SILKNOW thesaurus are: color, decoración, bibliografía, vegetales, rojo and forma. These Spanish phrases translate into English as: colour, decor, bibliography, vegetation, red and shape. Data is mostly represented by common everyday words or easily understandable terms. From this, we can presume that the SILKNOW thesaurus covers most domain specific terminology used in the CERES dataset. Additionally, nature, floral and vegetal motifs are also evident from the high frequency of the phrases: paisaje, florales and vegetables which translate as landscape, floral and vegetation. As can be seen in Figure 21, some of the most frequently found IMATEX phrases not included in the SILKNOW thesaurus are: conserva, teixit, color, vellut, banda and peça. These Spanish and Catalan phrases translate into English as: preserves, fabric, colour, velvet, band and piece. Data is mostly represented by common everyday words or easily understandable terms. From this, we can presume that the SILKNOW thesaurus covers most domain specific terminology used in the IMATEX dataset.
4.3 SILKNOW thesaurus concepts not occuring in data
For each of the four languages there are some SILKNOW thesaurus concepts that were not found in data from online resources. These concepts might be missing because they are part of very domain specific terminology related to silk. Given that some of the concepts are tied only to specific time periods or places, it is unlikely for them to be used fully in every online resource, but they might still be included in the resources added in the future. The following are two examples of SILKNOW thesaurus concepts which were not found in online resources. The first example of the concept “damaras”, which is defined as: “A thin and lightweight taffeta, which has patterns of flowers. It comes from India.”, shows how a concept can be tied to a specific place. The second example of the concept “aceytuni”, which is defined as: “A medieval name for satin. Rich silk fabric that began to be manufactured in Chinese town of Zayton ...”, shows how a given concept can also be tied to a specific time period.
Some of the English thesaurus concepts not occurring in data from current online resources are tetramorph, zoomorphic, ternum, poult, tercianela and sericulture. This group includes 72 English thesaurus concepts out of a total of 596. Meaning that English thesaurus concepts that were not found in the data from online resources represent 12.08% of all English thesaurus concepts.
Some of the French thesaurus concepts not occurring in data from current online resources are denim, chrisme, etoile, pourpoint, mante and rhadamés. This group includes 85 French thesaurus concepts out of a total of 611. Meaning that French thesaurus concepts that were not found in data from online resources represent 13.91% of all French thesaurus concepts.
Some of the Italian thesaurus concepts not occurring in data from current online resources are casula, terno, collarino, tovaglia, arcata and batavia. This group includes 200 Italian thesaurus concepts out of a total of 436. Meaning that Italian thesaurus concepts that were not found in data from online resources represent 45.87% of all Italian thesaurus concepts.
Some of the Spanish thesaurus concepts not occurring in data from current online resources are: zarza, fular, muaré, rádium, armura and devanar. This group includes 113 Spanish thesaurus concepts out of a total of 629. Meaning that Spanish thesaurus concepts that were not found in data from online resources represent 17.97% of all Spanish thesaurus concepts.
We can see the percentage of unfound concepts is the lowest for English, followed by French, Spanish and Italian. This might be simply because there are four English online resources included in this study, compared to three French, two Spanish and one Italian. One would expect that including more resources will decrease the number of unused concepts.
b. Domain expert’s validation
The SILKNOW thesaurus was validated following a structured methodology that began with an extensive literature review, that is, a scope note was only approved after reviewing at least three authoritative sources. In this sense, the thesaurus has more than 278 references. The literature reviewed has been divided into:
-
General dictionaries: These include each reference source of each language (RAE in Spanish; Oxford in English; Dictionnaire de l’Académie française in French; Trenccani in Italian). These dictionaries were very useful to select PT.
-
Specialized dictionaries: These dictionaries cover the domain of silk heritage and were used to better define and select specific terms related to historical silk.
-
Specialized references: Specialized books served both because of their glossaries of specific terms, and because they improve knowledge of historical fabrics.
-
Historical sources: They helped the researchers to define historical terms, their evolution over time and their current use.
-
Thesauri: Other thesauri were barely used as they are quite generic.
-
Journal papers: Academic literature was reviewed due to its high degree of specialization.
-
Thesis: PhD thesis were also consulted.
The following figure shows the total of the literature review:
The second step was to review the SILKNOW ontology that as per May 2021 it has more than 40000 instances. In order to do so, statistics were provided by EURECOM. In this sense, it is important to remark how a multidisciplinary team can make the difference. This information on many occasions provided terms that appeared many times in data, but are not really specific within textile proper terminology, such as patterned fabric, a generic term that refers to any fabric that has decoration and figurative elements. On other occasions, reviewing literature led to adding terms that are not frequent in data collections but are quite specific such as tiraz.
Next steps involved discussion with other domain experts including weavers, art historians, historians and philologues. One of the most important steps was to discuss terms with the translators of the SILKNOW thesaurus, as mentioned, they were part of the SILKNOW thesaurus. Here it was necessary to confront not only terms in the four languages, but also to confront the existing literature in those languages.
Finally, the thesaurus was distributed in workshops held for museums with textile collections, which, on the one hand, showed how the main problem of textile conservation is the standardization of vocabulary and the lack of a specialized thesaurus. On the other hand, these museums and their textile specialists have constantly reviewed the terms, giving feedback to researchers, who are constantly improving the silk thesaurus.
This methodology guaranteed that the SILKNOW thesaurus was improved on many occasions, hence it is scientifically accurate and is the only existing open-access, interoperable, silk heritage thesaurus.