STITCH

STITCH @ CATCH

Semantic Interoperability To access Cultural Heritage

NWO MPI VU KB

Integrated access to BnF Mandragore and KB Illuminated Manuscripts Collections

French: French

Introduction

Starting 2006, STITCH has collaborated with the French National Libary to build a demonstration web browser that would illustrate how semantic correspondances (alignment) between description vocabularies can help integrating collections coming from different backgrounds.
The object of this experiment consisted of two collections of illuminated manuscripts:Disclaimer: The prototype that is presented here is the result of preliminary investigations. Especially, the semantic correspondances it relies on have not been validated, the database contains only a sample of the original collections and the server may require patience. This work is still ongoing research, and shall not in any case be understood or used as a replacement for the original access means to the collections.

Help

Rationale
The illumination collections of both BnF and KB can be accessed from their respective portal using two options:Our experiment focuses on the subject browsing aspect of the orginal information systems. In these portals, it is impossible to access objects from KB using the BnF system and vice versa. The reason for this is that the system use different description vocabularies, namely Iconclass at KB, and the Mandragore content analysis vocabulary and classification at BnF.

To solve this semantic interoperability problem, our experiment uses semantic alignment links between the two vocabularies. These links built automatically (based for the moment on label comparison, like for "rodent: rabbit" and "rabbit"), are exploited to augment the number of items returned for a selected concept. If one selects the Mandragore subject "rabbit" in the browser, one will get all the documents from Mandragore indexed with this subject, but also the documents from KB that are indexed against the Iconclass "rodent: rabbit" one.

Hierarchical browsing of document sets
In the faceted subject browsing paradigm we have chosen for our demonstration, a user is typically presented with some view of the description vocabulary(ies) where he can click on subjects he wants documents about. In our browser, this is done on the left part of the screen (complete screen at the beginning of the browsing process). The vocabulary view presents the selected subject, the hierarchical path that leads from the root to this subject, and the concepts that specialize this subject. This enables to select more generic or more specific subjects.

On the right part of the screen, the user is presented with a representation of the set of documents that match the selected subject. In our case, these are thumbnails of the illuminations.

Access to different views
The STITCH demonstration introduces two modalities for browsing integrated collections using their initial vocabularies: single, and combined, that are adjusted to the different vocabularies available. Applying these to the specific BnF/KB case we obtain five views providing with different browsing "facets":

Mandragore Single View
Looking at the integrated collections through the lens of the Mandragore content analysis vocabulary. For example, one can view items described (directly or via an alignment) using "amphibiens" by selecting "Zoologie" (zoology) and then "Amphibiens" (amphibians).
Iconclass Single View (French)
Looking at the integrated collections through the lens of the Iconclass classification scheme used for Illuminated Manuscripts, using French labels found for Iconclass subjects. For example, one can view items described (directly or via an alignment) by "batraciens (amphibiens)" by selecting "Nature", "la terre, le monde en tant que corps céleste" ("earth, world as celestial body"), "animaux" (animals) and then "batraciens (amphibiens)" (batracians (amphibians)).
Iconclass Single View (English)
Same as above, using the English labels of Iconclass subjects (Iconclass comes in English, French, German and other languages).
Iconclass Single View (German)
Same as above, using the German labels of Iconclass subjects.
Combined View
Searching the integrated collections with the combination of Mandragore and Iconclass subject schemes. The results correspond to the conjunction of subjects selected in both hierarchies. An example would be searching both for Mandragore "amphibiens" and Iconclass "famille, descendance" (family, descendance), here.
In all cases, a basic facet is provided, "Collection", which allows to restrict the query for items coming from one collection or the other. This can be useful when too many items for one collection are returned for the selected subject, hence "hiding" items from the other collection.

Mapping configurations
When browsing, one can activate or de-activate the set of semantic mappings exploited by the system, clicking on the captions below the (blue) header information, as described below:

mandraBASE
Using all mappings produced with automatic alignment scripts.
mandraNewNONE
Using no mapping at all.

Accessing specific documents
Clicking on the thumbnails displayed on the right part of the screen, the user can access a specific document matching the current subject selection, like here. Some basic information is then displayed about the document, including a bigger picture and extra metadata information, like the title of the illumination, the date of creation and the original subject description, with elements coming from Mandragore or Iconclass concept schemes, depending on the provenance of the item. For Mandragore items, a link is provided that redirects towards the document such as found on the original BnF website.

Acknowledgements

Work at the BnF was coordinated by Thierry Cloarec and Frédéric Martin from the Digital Libary Department. We want to acknowledge contributions and interest from Jean-Pierre Aniel and Thierry Delcourt (Manuscript Department), Anila Angjeli and Françoise Bourdon (Documentary Normalization Office), Emmanuelle Bermes (Digital Library Department), Laure Mane (IT Department) and Michel Mingam (Rameau office). Jérôme Boillet and Patrick Le Boeuf were involved in facilitating early stages of our collaboration.

We would like to thank the Dutch institutes KNAW and RKD for providing the Iconclass resources, and Gerda Duijfjes-Vellekoop for helping with these.