The automation of manuscript cataloguing has to be able to handle descriptions at all the varying levels of detail and complexity discussed above. The interface for displaying these descriptions has to allow the user to search for manuscripts via a number of different paths, based on searchable and scrollable indexes (e.g. authors and texts, scribes and artists, owners and donors, iconographic subjects, etc.), and free-text searching; it has to be easily navigable, and visually acceptable to those who are familiar only with conventional printed media; it has to enable downloading and printing of catalogue entries; it should not be software-specific or require high-specification hardware; and it needs to co-exist with catalogues of other types of material, including post-medieval items.
Various automation options were examined over the course of a year, with these aims in mind: we examined proprietary systems such as Cairs, and experimented with the production of our own relational databases using the FoxPro package. The optimum solution which we have so far identified, however, is SGML, and this has formed the basis of most of our work during the project: for our collection-level, and minimal item-level, descriptions we have been using the Encoded Archival Description (EAD) ( Library of Congress 1996) and for detailed records we have extended the Text Encoding Initiative (TEI) ( Sperberg-McQueen and Burnard 1994) to improve its handling of manuscript descriptive information (metadata).
The EAD had reached its alpha version when we began encoding our finding aids, and it has readily proved itself suitable for providing the information which we had traditionally included in our printed versions of collection descriptions. It does not, however, provide enough specific elements at the item level to allow the marking up of catalogue records for individual manuscripts in as much depth as had been used in the most recent printed Bodleian catalogues. This can be circumvented by using the generic ODD (Other Descriptive Data) element, but this represents something of a evasion. It was therefore decided to use the EAD for information from collection level down to a minimal item level description, and then link from an EAD entry to a corresponding TEI record, in which much greater detail could be encoded: the EAD has several ways of linking to external files, of which the simplest is to use an entity reference. We plan to employ the same user interface for both DTDs: the user will not have to know which one applies at any given point.