Location and attendance
The meeting was hosted at Alphen an der Rijn (Netherlands) by Samson-Sitjthoff BV, part of a major Dutch printing and publishing conglomerate, whose Information Services division is directed by Jan Maasdam, chairman of the Dutch SGML users' group. The SIG itself was set up very recently by its current chairman, Martin Bryan (author of the only readable book on SGML) who works for a division of the SEMA group called Yard Software Systems but is also closely associated with SOBEMAP and MARKIT, their SGML parser. The SIG has about half a dozen active members drawn largely from major European software houses with an interest in the field. Its chief remit seems to be discussion of the interface between SGML and database design, but this was only the third meeting and the group has not yet felt the need to create a formal constitution or agenda.
Agenda for meeting
The agenda was as follows: Application of linguistic methods and tools to database management of SGML coded texts. Gert van der Steen from MID, a Dutch software house The TEI: an application of SGML in scholarly research. Me Performance comparisons of some UNIX-based RDBMS François Chahuneau of AIS, a French software house General discussion of a paper on conceptual modelling for a document database tabled at the previous meeting by Han Schouten of the TFDL (the Dutch Agricultural Ministry)
Van der Steen's presentation was overlong and rather rambling for the occasion, but raised some interesting points about the benign influence of computational linguistics in the development of SGML (a DTD -he said confidently- is a formal grammar) and the appropriateness of hierarchic database systems to it. His company is developing an "Integrated Publishing Management System" entirely dependent on SGML as transfer mechanism, which had necessitated a detailed specification for an ideal text retrieval system. He also described his own PARSPAT system which uses recognition of syntactic patterns as a database search mechanism (he has recently published a book on the unification of pattern-matching, recognition, parsing and transduction) and gave examples of its use for analysing the Brown Corpus and a database of 18th century Delft Estate inventories
My presentation simply outlined the structure of the TEI, gave some examples of the horrors of unchecked scholarly markup and discussed the relationship between text and databases.
Chahuneau 's company has the task of constructing a document database to support EEC legislative and other documents in nine languages in parallel. Its scale (15,000 pp in the printed annual form) and complexity set it apart from any other SGML applications I have yet come across. Because the database is constantly changing, sophisticated version control and integrity checks are essential to maintain all nine views of it in parallel. This ruled out any of the traditional text retrieval database systems; hence the case study of available UNIX RDBMS. Decisive factors in narrowing these down proved to be their degree of support for the 8-bit characters of ISO8859 (essential for the 9 languages); and the way in which the software implemented crucial database operations. Only three of the nine systems investigated allowed the manipulation of 8-bit characters as well as their storage. As to software performance, it seems that the ideal system would combine INGRES' query optimiser (which made an order of magnitude difference to the speed of join operations) with the SYBASE file-access engine (which had a similar effect on most other operations). An investigation of the various hardware platforms available showed perhaps unsurprisingly that although a RISC-architecture machine such as the new DEC Station gave enormous performance improvements, the low-end 386-based machine was a better price/performance option for development than any of the other available workstations.
Discussion of Schouten's paper was less focussed than it merited, largely due to the lateness of the hour. Schouten had advocated using a conceptual modelling technique known as NIAM, rather than the more usual Entity-Relationship model (which would have pleased Chahuneau and me) or a straight hierarchic datamodel derived from the DTD (which would have pleased van der Steen). He had not paid much attention to such specific problems as version control, which seemed to imply the need for a formalism superior to the DTD, the semantic adequacy of which was already in question.
Both Chahuneau and Bryan expressed interest in the work of the TEI, and a willingness to participate if invited. I took the opportunity of rehearsing some of the current Committee 4 arguments with them (both agreed that attributes were not formally necessary, but still extremely usefull). If we do decide to involve either of them, Bryan might be a better SOBEMAP representative than Gaspart, while Chahuneau or his nominee would be a good substitute for Dendien on committee 4. Either of these would be self-financing. It is important to stimulate SOBEMAP interest since they have EEC funding for their MARKIT product which is the only structured editor I have so far come across that runs in the MS-DOS environment we all know and loathe.
I met Jens Erlandsen, whose company TEXTware A/S based in Copenhagen is developing Gestorlex, which seems to be yet another SGML-based structured editor for dictionaries and other reference books. They also market a small free text browser of the Gofer-type, and are involved in an ESPRIT project to develop a multi-media publisher's workbench. The novelty in the latter is that they plan to implement Salton's "space-vector model" for the full text indexing requirement.
Kluwers, publishers of CHum, turn out to be a part of the same publishing empire as our hosts. They were represented by Drs van Wijnen, who seemed quite taken with the notion that CHum contributors might be amenable to supplying their material complete with SGML tags. She agreed that this was worth suggesting to the CHum editorial board at any rate.