The Processing of Manuscripts

The Processing ofJ/anuscrpi ts 41 The Processing of Manuscripts
Manfred Th ler
Summary
We are a d, that this contribution, unlike the others in thsi volume, may em to be unduly abstract – or overly technical. Therefore, we would like to summ ize our argumentation rst, to show more clearly, how the highly technical considerations at the end are the consequence of a substanti , historical argument.
1) Techniques for the processing of digitalized manuscripts exist, which em to be promising.
2) Experiences have shown, that – unlike in the processing of imag in history of – there is not very much sense in an extended di ussion about the quality required: there exist quite clearcut criteria for the quality which is needed in the pr essing of text. And these criteria call for the proc sing of fairly high quality images for dealing with manu ripts.
3) Irrespective of !arge ale projects – like the one at the Archivo Gener de Indi – the application of such techniqu to typical medium size collections, like the on contained in city archiv , are nancially fe ible today.
4) To m e the most of such collections of manuscripts within historic r e ch and te hing, a workstation with a bundle of processing tecbniques is needed, which e, however, not automatic ly provided by modern image processing p kages.
5) To develop such collections of manuscripts om a purely pr rvation vehicle and a tool for the study of pal ographic properties further into a genuine repl ement of previous itori forms, a much closer connection between the bitmapped image of the m uscript and the transcribed ASCII text would be needed.
1. The problem
Images have in recent years drawn much attention in historical re arch: their im­ portance for a variety of subdisciplines in the eld of cultural and social history h been emph ized quite equently. Still, while their importance historic urce can scarcely be overestimated, traditional, written sources will never be superseded by them – and historical interpretation, we know it, is and will be b ed, first of all, upon the content of written documents, which, for most historical periods, have survived manu ripts.
Manuscripts, in many respects, share important qualities of images, when data pro­ cessing is concerned. Many manuscripts have survived in handwriting, which h to be interpreted, rather than simply transliterated character by clearly recognizable character, when their contents sh l be converted into a machine readahle dat et. Their can be vari­ ous re ons for that: be it, that they are heavily abbreviated, with abbreviations which are all but clear for many are , particularly of administrative writing, be it simply that the handwriting is hardly Iegihle – either because written by a calligraphically dissatisfactory person or because of damages having occured over the years.
Handwriting furthermore does not allow a clear separation of content and pr enta­ tion: in many are of historical research, particularly medieval diplomatic studies, the interpretation of the form of the letters, indications, by which author they have been
42 Mun d Thai/er
written, is just important the text these letters transmit to our interpretative abili­ ties. The form a Ietter h in print is usually irrelevant: in older manuscripts most o enly it not.
It arcely tonishing, therefore, that the advent of image processing techniques, described in the rst contribution of this volume, h very soon Iead to their application to manu ript material, focusing most oftenly upon the intensive study of a relatively small amount of material, in some c es also upon the systematic conversion of extremely !arge corpora of texts. Thsi treatment of manuscript sources by techniques drawn from image proc i ng1, we will call manuscript proce ing in this paper; a concept which we develop further, however, to include a nurober of techniques which are not usually connect with image processing.
What e the re ons why we would like to apply such techniqu ? The following are m t equently quoted:
• M uscript material can on the screen be repre nted in a quality, which is not wor than that of good photographic repr uctions. Indeed, with the exception of very few extremely expensive facsimile reproductions, digital reproductions of manuscript material will usually be better than photographic ones. These representations can not be damaged by careless handling. One of the rst purp es of manuscript processing therefore connected with the pre rvation of material.
• Di t repr entations !end themselves e ily to the application of various techniques of im e enhancement. By appropriate techniqu it is relatively e ily possible to m e manuscripts on the computer screen better readable than in the original. The second re n to start with image processing is therefore the wish to improve the legibility of damaged or from the beginning hardly readable portians of manuscripts.
• Palaeographic research, dealing with the properties of old handwriting, implies a very intensive handling of individual characters, for their systematic comparison accr s documents.
2. Experiences and Practical Background
paper not a project report; it shall, on the contrary, try to discuss the general p ibiliti of „manuscript processing“ a eld, where we expect great advances in the next few : nevertheless it h not been written without practical experiences and me of the recommendations implicit in the following pages can be understood only in the context of these experiences. Before proceeding further, we would therefore like to d cribe the pr tical background of the author and state explicitly some sumptions drawn from pr tical work, which underly all the remainder.
All experienc gained are derived from attempt to implement a gener data type „image“ into the DBMS KA€ which h been developed by the author since quite a number of years now. The experiences from the early Stages can be summarized follows:
1 We will not repeat y technical details bere: a good description of the b ic technolo , tagether with examples of early applications, is contain in: Kevin S. Kiernan, „Digital Image Proc i ng and the Beowulf Manuscript“ , Liter & Linguislic Computing, 6/1, 1991, pp. 20-27.

The ProcessingofManuscripts 43
Immediate Image Retrie . This describes the most basic methodolo that is, the ability of a software system to display the result of a query not just the description of an image, but the image itself. To do this within a research environment, a number of practical considerations apply:
• A hierarchical storage administration2 should be mandatory. This means, that each of the images out of the whole set of 20.000 or 100.000 administered ones can displayed in the form of a small snapshot without noticeable delay. Thsi is due to tbe fact, that such small snapshots are being kept on the f test medium within a storage hierarchy, while the bulk of the image data resides on slower media in the background, access to them possibly even implying the manual mounting of discrete storage media, like magnetooptical cartridges.
• E detail of an im e h to be available for zooming with an arbitrary ze of the zooming step. This is methodologically extremely important. If you provide imag and the possibility to Iook at only four or ve prede ned details, the editor of a CD­ ROM based system considered to be the most important, the editor is still the only one who controls what you are allowed to be interested in within an image. This do not change dramatically the way of your approach to the images, as compared to a printed book. Only when the user has the p sibility to control, what det ls sh l be zoomed – preferably zooming with g n ofinformation – access to the materi better than in the printed solution.
• Of course the usual cutting, p ting, mirroring and similar tools e also nec within the historic working process.
Im e Enhancement. Within the experimental system about 30 statistical operations for transforming and ltering image data have been implemented3. These methods – weil the usual false color techniques – can be applied to any image within the datab e and or/to any segment thereof. The application of these techniques within historic research h usu ly one or more of the following goals:
• Improving the readability of portions of manuscripts, which became unre able, cause either the writing itself or the material upon whicb it h been written h changed color, resulting in a reduction of contrast.
• The processing of documents, parts of which have been damaged by mould, humidity or similar re ons.
• The processing of items (inscriptions, coins, etc.) where letters which were cut into the surf e or are higher than it were partially destroyed by d age to the object; under some circumstances such material can be partially restored.
2 In this context it h been plemented for the t time, to the best of our knowledge, within a joint project of IBM Japan and the National Museum of Ethnology at . See Jung-Kook Hong and Sigeharu Sugita: A Color Image Datab e for Ethnolo Museum, in: Heinrich et al. (Edd.): Computers in the Hum ities and Social Sciences, K.G. Saur, 1991, 53-60.
3 These techniques are currently realized with the help of an image processing übrary, e Assistant, produced by IBM. At this moment another imple entation, using DECimage, a similar package by Digital Equipment, is checking whether our concern for portabiüty b been consistent enough.

44 Ma d Thai/er
age Bni ding. Many authors would describe this concept by binting at „Hypertext“ or me of the other „Hyper“ concepts. We would like to avoid thsi , as a matter of intel ctu strictne . In data processing there is the concept of nonlinear rcpresentations oftext: example o f t hose makes up the bulk of this paper, following below. „Hypertext“ is a phr coined by Theodor Nelson, for a very speci c and consistent model of a textual data type de ned on the basis of speci c tools out of the general realm of nonlinear – or non uenti – texts4 • This model h far never been fully implemented. Application Systems like Hypercard5 are no implementations of Hypertext, but systems to minister su ts of tbe general concept of nonlinear structures.
We therefore prefer to give a more precise de nition of our approach, than just a global – d wrang – reference to Hypertext. „Bound Images“ we call the administration of bit mapped data objects, which are nonlinearily related themselves, can at the same time p ts of arbitrarily complex network of tran ribed information, however. This done in a way, where e h portion of the transcription or de ription in text form, is explicitly related to an area within the bit mapped object. For the sake of completeness we d, that such areas in our implementation may overlap.
While ori nally geared to the a.dministration of collections of historic pictorial sourc , th tools have been applied in a nurober of c es to projects involving the handling of manuscript material, including two attempts at teaching such applications in university contexts. These experiments with teaching produced in some ways the most inter ting pr tical illustrations for the requirements which actually have to be ful lled, be re such techniques can be integrated into daily research.
On tbe one band , a group of students at the university of Bremen tried to use a digit repr ntation ofparts of a prominent source ofthat city to learn the reading and handling ofm iev manuscripts in practical work. At the beginning of the project it w assumed, that the wor ng group would spend some time to choose the m t economical way of ning the urce in a resolution which would be most appropriate for the hardware av lable and employ later the 30 or so techniques for enhancing images a ered at the time for improving the legibility of the manuscripts before transcribing them. The equipment av lable w de nitely insu cient: the scanner used h , e.g., been available only for a couple ofbours on a run after which a fairly complex proce to transport tbe digital images om a NeXT computer to a PS/2 80, running under the IBM UNIX implementation AIX b be start , involving the transfer of image les of about 5 – 6 Mbyte via diskette bet n th computers. The PS/2 80, though equipped with 12 Mbyte memory, w quite when handling synchroneously more than two images of more than 2 or 3 Mbyte.
4 For which bave to do with the very peculiar funding of this project, it is somewbat t good bibliographic references. Theodor H. Nelson: Literary Machin h since 1981 publi hed in various versions, usually by the author him lf. A good summary, which e ily available: Theodor H. Nelson: Managing Immense Storage, in: Byte 13/1, 1988, 22S-238; on the ncept see also Janet Fiderio: A Grand Vi on, in: Byte 13/10, 1988, 237-244; p ticularly 238.
5 di erence quite o enly ignor d in the HyperCard literature: a particularly bad ex­ am e in Carol Kaehler: HyperCard Power, Addison-W ley, 1988, 366-367.

The Processingo[Manuscripts 45
Indeed, most of the time of the participants w spent with the overcoming of fairly trivial hardware restrictions.
Wben it comes to the usefulne of image enhancement, this seminar proved incon­ clusive at best: the whole process w so cumbersorne, that such techniques were b ic ly never really tested. Still, the participants, even under tbis conditions, found the exercise u ful and rewarding. The re n: si ply scanning a manuscript di cult to re at 400 dpi and displaying it on a 1280 x 1024 pixel screen produced an implicit m ni cation, which made tbe m uscript e ier to read on tbe reen, Jong before any attempt at image enbancement took pl e. As a result of tbis, tbougb r tricting tbem lves to Jower olutions would bave been speeding up tbe processing, the usage of lower resolutions w never considered a serious Option by tbe participants of tbis course.
The second teaching exercise of that kind consisted in a seminar offered by the author at tbe university of Göttingen. The participants8 were ven a b ic „ß by de on type introduction to tbe principal techniques of image proc sing (immediate image retrieval, image enhancement, pattern recognition / compari n plus po ibilities of bypertext-like linking of images and connected or underlying texts). After a very sbort iutroduction into the bandling of tbe tools available, tbey wbere ked to cboose a source type tbey considered tbemselves familiar witb and to design an „optim n system for administering tbat type of sources on a computer, re izing witb tbe tools at band so mucb of such an optimal solution possible and specifying clearly wbicb technical building blocks were missing. Questions to the cap ities of the bardware were intentionally blocked: wbile some vague propos s were m e to nd out, bow far decisions like resolution cho en would inßuence tbe practicability of the rudimentary prototypes, all participants were told to concentrate on tbe de nition and construction a tool, which would tu ly improve the p biliti of historical work with their type of source and ignore all qu tions of affordability and the like. In comparison to the Bremen exercise tbe res urc available . were relatively ample: four 6000’s of IBM could be used, each of which w equipp with 16 Mbyte of memory and had cess to a network tot of rougbly 2.6 Gbyte of disk space. Acce to scanners w suboptimal, but manageable.
The ve participants decided eventually to work on prototypes of tbree systems for tbe bandling of manuscripts or related materials, b ed upon the available prototype of an image handling and processing system. In c es tbey u d sources witb wbich they b become familiar during tbeir regular bistorical training, most of tbem in advanced semin s dedicated explicitly to the study of tbis particular type of source. Tbe resulting systems were dedicated to tbe following are : (a) A modernized version of tbe cl sical cbartulary, wbere the digital image of a charter, Iogether with it’s tran ription sbould be administered by tbe computer baving image enhancement and similar tools available during tbe palaeograpbic study of tbe documents. (b) A system for tbe administration of bistorical maps, wbere indeces of topograpbical names and similar descriptors sbould give immediate access to tbe maps themselves. (c) A system for tbe bandling of political and religious pampblets om tbe reformation witb a double m in mind: on tbe one band to
G 3″1 – 5th year students of sto .

46 .Han rd Thai/er
study the relationships between the argument in the text and the illustration and on the other the way in which illustrations or parts of such became reused in other pamphlets.
Comp to some of the proposals, what might be adequate ressources for employing im relat computer techniques in research, the platform we quoted here seemed to be quite substanti : it is quite interesting therefore, that none of the systems proposed could have been realized with the available hardware and/or software. We should repeat, that what the students were asked w not „realize a system with the ressources you h e here“, but rather: �realize a system that really would make it worthwhile to use compute henceforth, when you handle such sources. lf such a system requires more r th have here, it certainly is not your fault.“ In all cases the reasoning behind the r uirements, which made it awkward or even impossible to realize what had been r w historically absolutely sound. Let us Iook at a few examples:
• p ticipants opted for the highest available reolution (400 dpi). For charters, it w gued, this was necessary, because any attempt at comparing the palaeographical properti of the texts needed the possibility to get close up Iooks of individual letters – d me nice examples for shedding light upon er ures (cf. the example in gur 1 and 2 on the facing page) and possible falsi cations would only work under such r utions. For maps it simply w not s nsible to accept a resolution, where top aphic names inscribed on the map would not be legible. For early modern p phlets, we might have accepted a slightly lower resolution most e ily, were it not, tbat in that c e the chances at comparing parts o f illustrations to nd out about reu e would have droppe dramatically. As a result of that resolution, the image cont n typic ly something between 8 and 12 Mbyte per image, which would ha been manageable on a 16 Mbyte machine. Unfortunately, however, the whole point of the exercise w the application of various advanced techniques, and that m e it n c to have more than one version of the image on the screen and there re in the memory at the same time.
• Commer systems most frequently are able to display and process one image at a time. H they handle two or three snychroneously, they tend to bo t about it. When the ware platform we used had been designed for experimental work, it was ­ sum , therefore, that having 9 – 12 images available for synchroneaus processing at any one time would be ample. As it turned out, using the m hine for real palaeo­ graphic work, with systematic comparisons between huge sets of characters, the snychron us handling of at the very least twenty, but probably for any but extremely trivi applications at least forty images w considered a conditio sine qua non.
• P ticul ly in the c e of historical maps, but also in the c of charters, the standard ofßatb scanners (usually the A4 format or its slightly !arger American cousins) h to considered insu cient . While the use of !arger atbeds (AO would probably go a long way) would theoretically be the e iest solution, it is quite improbable, that such m hines become affordable soon: therefore a whole set of possibilities for the simple r ombination of images scanned in parts into a complete picture have been for.

The Processing o[Man scripts 47
Figure1: Er ureof’A’inBE {A)HA Ibefore…
Figure2: … andalterenhancementofinkshadows.
48 Man d Thal/ r
We have described these experiences at quite some length, because, in our opinion, they illustrate two very important di erences between image processing in history, traditionally described in the area of pietonal sources, and the handling of manuscripts with techniques of image processing.
Proce ing manuscripts does not usually ha to be argued for. A system which displays a manuscript on the screen, which is weil or even better Iegihle than the ori nal, never h to be „sold“ to a historian. If we see a manuscript on the screen and do m t of what we could do with the origin , there is few doubt, that the technique such is v uable. If we can improve it beyond the state of the original, the soundness of doing so will go unchallenged.
B quality in manuscripts is not tolerated. On the other band, there is another fun­ dament di erence between manuscript systems and such, which display pictori al sources: when image – be it a work of art, or source for the study of the daily life of the middle es – displayed on the screen, a long discussion can start, whether the quality with which this image is displayed, will su ce for the purpo of the respective system. Such a di u ion can be long, protracted and highly stimulating – it will ways, however, be subjective, being totally dependent on the degree in which the individual application cl to make the recourse upon the original unneccessary. In manuscripts, the criteria are cruelly simple: the piece cannot be re on the screen weil in the original, the er w not worth the effort. (And it seems, that it is relatively y to use a picture illustration, without recognizing det ls of the relative size of an individual character in a p e of manuscript.) lf on the screen we can what the unaided eye can detect on the ori nal, the result is ceptable: but no historian will go for such a solution once she or he h realized, that in principle it is possible to on the screen much more than in the ori nal. There is a huge di erence between a very good image and an image displayed s y in a low quality; the mar n between a weil readable manu ript and a collection of irrelev t wiggles on the screen is extremely sm l.
With this restrictions in mind, ways clinging to such quality h been found appro­ priate in various tests, we will now Iook at a p sible pilot project and discu what it might in time and money. A final word of introductory explanation may be required: in a nu ber of ways we clearly deviate from the experiences reported by the impressive report about the project at the Archiv de Indi later in this volume. These di erences intention : our understanding of manuscript proce ing includes the notion of digit iz­ ing urc for conservation. While we see obvious implications of such a technique for archiv institutions, our primary metaphor is that of an edition, however: the proce ­ ing of corpora which are !arger than those handled by tradition editorial techniques, are still relatively small, however, and shall get the best av lable treatment. Therefore, legi­ bility being an obvious prerequisite, we require, furthermore, su cient quality to perform p ographic research on the material preparoo. The following descriptions build upon t projects which have been or rather are undertaken with the software platform devel­ oped by the author. In one c e the surviving material at the former concentration camp of Auschwitz-Birkenau shall be stored in digital form to .preserve it beyond the foreseeable di olution of the originals. In that c e the volume of images to be scanned in the near future is somewhere between 50,000 and 100,000 pages. On the other band our estimates

The Proces.sing ofManuscripts 49
are inßuenced by a project during. which roughly 2000 photographs of architectural draw­ ings shall be preserved in digital form for a study of the paper upon which th drawings have been sketched.
3. Step one: a digitalized archive
To discu the possibilities of manuscript processing more systematically after our initial considerations, we will rst examine the requirements of the following (so far hypo­ thetical) project:
The City archive of X has ]arge holdings of private charters from the later middle ages. About 20,000 pieces between, say, 1300 and 1500 exist. The collection is being used regularly by the loc university for traning in mediev palaeography. During recent yea some ofthe pieces seem to have gotten severly damaged, however. A solution looked for, which allows to preserve these 20.000 pieces without further damage, ma s to the u rs reproductions available, however, which do not hinder any kind of research. At a Iater stage, possibilities to edit such material and make it more widely available shall be found.
Almost l archives until very recently, and the v t majority of them even today, would for such a project more or less automatically Iook for a solution which tums the collection into a set of micro lms or micro che. We will not try to make a det l cost-e ectiven comparison between an approach at micro lming and digitalizing such a collection. Indeed, we sume that of todäy turning the collection into micro lms would be considerably cheaper, which may change during the next five and ll certainly change during the next ten years. We would rst of all like to prove, however, that the option of digitalizing such collections, while not cheap, is already within order of magnitude which makes it fe ible even today.
The b ic cost would of course consist of the computer needed: after the cautioning words we have given about the le sizes to expect, we have to envisage a work station which is powerful enough to proce images of ten – twenty Mbytes without getting stuck. Typically, that would be a UNIX workstation with about 32 Mbyte of central memory or more. At current rates, this would translate into roughly US $ 30,0 , t ing academic counts already into consideration. That price sho ld cover a local disk of a few hundred Mbytes, which, given the sizes of the images quoted, would be de nitely to small to begin with. At the current stage of technological development, we would therefore propose to add a lGbyte magnetic disk and a rewritable magnetooptical disk drive with two times 250 or two times 500 Mbyte per disk. Tagether this would add a Iittle bit l than US $ 10,000 to our con guration, with each Mbyte of magnetooptical medium, used for the long term storage of the scanned material costing about 50 US cents. (With a clear downward tendency.) To that we would have to add a black and white scanner; for practical reasons it usually turns out to be e cient, to have a PC with a modest hard disk, which is fully dedicated as scanner server. Scanner plus PC plus all nece ary connections betweenUNIXsystemandscanningunitshouldbeavailableforUS$5,000.Ifwe u me, that we have charters, which (as private charters) are relatively small and, a rule, t unto the atbed scanner, experience in the various projects quoted, but also in parated attempts at scanning, proposes two things: (a) we will not be able to get high quality reproductions at much below 10 Mbyte per charter of raw data and (b) considering the

50 Man ed Thai/er
various copying operations from scanner PC to optical background storage, we will not be able to k p up the scanning Operation with a speed much higher than 5 charters / hour. This means, that the media cost for our project will be roughly at five US $ a charter and roughly 100,000 US $ for the whole archive. For scanning the 20,000 charters we will need 4,000 hours, or, u ming 200 e ective 8 hour work days a year, roughly 2.5 years.
145,000 US $ and 2.5 person years, to convert late medieval private charters of a city archive into digital representation. This is a Iot of money; when we compare it with other steps to be taken for the preservation of endangered cultur property today, it loo some ofthe frightening magnitude it may have at first Iook, however. What si even more important : we have noticed, more than two thirds of the expected expenditure for equipment have been set aside for storage media.7 This, however, is precisely the area of development, where prices are currently going down most rapidly. That , we have every re onto u me,that,whatwepropo herewillbecomeeheaperbyabout25%every year. (An incre e in scanning speed is likely, but much barder to predict than the price development . )
While we have in any ways discussed a warst case gure now, we have on the other band le a few potential sources of cost unmentioned. So we have su ed so far, that the archive would have no objections to the private charters being prt e d rmly upon the ßatb scanner to have them absolutely Ievel for scanning. This could be replaced by a strate , by which a ovable scanner is mount d on a suitable mechanism over the original which than is pressed down by a gl s plate or another such device used in professional photography of documents. Such a solution would probably add between ten and twenty thou d US $ to our bill for equipment – we will not discuss it here, however, that is just the kind of cost which we also would incur in any micro lming operation at almost exactly the same rates.
So, that’s the price of a digital archive. What advantage will we gain over a i­ cro lming Operation? Of cour we could now quote the various p i bilities of computer supported techniques of image proce ing: image enhancement; the p sibility of improved handfing of parts of the image for palaeographic or similar studies; and so on. More im­ portant for an immediate co pari n, however, are a few advantages which are built right into the Jogic of digital image representation:
• Photographs deteriorate steadily, in an unpredictable step by step or rather shade by shade manner. Digital images can be compared bit by bit – it is clearly possible to say, whether they have the same quality they had some years ago or not.
• When micro lms are copied after some time, the copy will in some degree be wor than the original; copying them will usu ly involve quite a bit of manual handling. Di tal images can be copied by simple copy commands, the copy containing provably the same information as the origin .
1 T calculation may be open to doubt, we have given gures for the uncompressed images; we will not go into that detail, but propo , that we continu to use this gure, using the media cost w is not actually consumed after suitable compression for the addition of redundancy in our storage scheme for incre ed security.

The P cessingof. lantiScripts 51 • Storage media constantly bccoming eheaper each successive copy will be more compact
than the previous one.
We would like to take some pains at this stage to exp/ain, why have discu ed this example here at such lengtb and with those many nanci detai/s. Not, b ause think, that at this moment digitalizing manuscript materi is ready a nanci ly convincing ternative to micro lming, when we speak about preservation „only“, but to prove, far c culations can do that, tbat even t s would be p i ble already t ay within tbe archi / organisations of the more a uent European cities, without too sbatte ngly expensive requirements.
4. Using digit ized chives
We have in the l t section concerned ourselves only with the qu tion, how far the digitalization of substantial, but not overly huge corpora of manuscripts would be econom­ ically fe ible. One could Iet it rest at that and simply predict, that, the gures being they are presented, digitalizing manu ripts will eventu ly replace micro lming as a precaution to pre rve valuable material. We would, however, like to expand this argument further.
What we described so far, is only the rst step in a process. We propo , that it should be seen the initial step in the following idealtypische procedure:
Step 1 : A set of manuscript sourc ann� with the kind of r lution d cribed. According to the preceding technical discu ion, can sume that the administration of 10.000 – 20.000 pages of manuscript on PC’s of the upper or workstations of the lower range will be f rly easily possible within two or three years from now, manageable within the range of the typical two- to three-years project, dear to the heart ofvery many funding agencies.
Step 2: The documents which in that way are preserved permanently in photographic qu ity, are lo ed into a data b e: this data b e contains at the beginning just a seri of document identi cations (b ically archiv numbers) and the ann images of these documents.
Step 3: When working with such sources, the historian rst of l transcrib the manu ripts literally. For this purpo the following tools are at bis command:
• By a system of graphically displayed reference coordinates, the historian can bind individual transcribed phr es (or important abbreviations, or individu letters sig­ ni cant for the speci c scribe) to the transcription of these portians of the text.
• ls a reading di cult, it is possible
• to improve it by the means of the image enhancement techniques discu ed.
• but al to get a set of comparable portians of the manuscript transcribed so
far. Eilher by king for the graphical representation of c where the possible readings have been tran ribed at earlicr stages of the working proc s or asking for c es where such transcriptions are within a given degree of similarity to the graphical form to be transcribed now.
Step 4: As soon a continuous transcription exists, or parallel to it’s creation, tools exist to insert into the text symbolic markup, specifying e.g. persons mentioned or topographical entities referenced. While in the design plans of our project a more general

52 Ma d Thai/er
notion of markup is employed, speci c ly emph izing that there are situations, where the aphical repre ntation o(symbols may be important, SGML (Standard Gener ized Markup Language8) would be a good example for a fairly general markup language which could be employed r such purpo s. While we would like to stress, that we consider the qu tion of how such markup schemes should be constructed for other than printing purp to be anything but closed, one should point to the e orts of the Text Encoding Initiative9 to de ne common rules for the applications of such markup schem in specific are of the Humanities.
our context such markup is instrumental in converting the tran ribed text into a component of a structured data b e, into which each portion of the transcription is supposed to be parsed immediately after markup h been completed.
This, actu ly, would leave the notion of „digitalized im replace micro lms safety copies“ far behind. Indeed, what we describc here, is a potentially new technique for the dissemination of manuscript materials, being radically di erent from the notion of the cl sical printed edition. The di erences are:
A cl ical printed edition, which we will call a static edition henceforth, h two clearly identi able and completely separate stages. In stage one, an editor collects material, d tri to prepare it for printing. long that stage l ts, the material is useless for the community at !arge: it is not po ible to distribute the edition before it is fairly complete. On the contrary, in the process we described, which we call part of the concept of a dynamic edition, it would, su cient standardisation of the storage media being u med, be p i ble at any stage of the work to distribute it in precisely that stage.
a kind of corrollary to the previous Statement: a static edition is nished at a very precise moment. After that, changing and improving it becomes next to impo ible. A dynamic one is never nished – it will certainly be rather di cult to ensure, that such a dynamic repr ntation is not worsened by additional chang (I sume everybody is aw e of the great concem that very proble is creating for The ore Brunner of the Th urus Lingu Graecae10), but just because of this problem, we should not overlook the eat potential of continu ly improving editions.
The connection of a photographic reproduction with a tran ription of the text fur­ thermore opens up completely new uses for an edition: in our introductory example of section 4 we mentioned, that part of the re n for preserving the charters w their con­ stant use in the training of historians at the local university. A b e of numerous charters in high qu ity reproductions, partially transcribed, obviously would open up completely new p ibiliti for te hing.
. 8 l Gol b: The SGML Handbook, Oxford University Press, 1991; . Lou B d: What i� SGML d how do it help?, in: Daniel Gr �tein (Ed.): Modelling Hi�torical Data,
ripta Mercaturae V lag, 1991 (= Halbgraue Reihe zur historischen Fachinformatik), 65-79.
9 C. Mich l Sperberg-M ueen and Lou Burnard (Edd.): Guidelin for the Encoding and
Interch ge of Machine-Readable T t , Chicago aod Oxford·, D Version 1.0, 1990.
1° Cf. Theodore Brunner, Director’s Letter, in Thesaurus Linguae Graecae NPwsletter, # 15,
July 1989, pp. 1-2.

The ProcessingofManuscripts Fin ly: can you imagine a rinted , static edition of the quoted roughly 20,000 late
mediev private cbarters of a city archive?
Let us dwell a little bit more upon this choice between „static� versus „dynamic“ administration of „edited“ source material: one might wonder, whether this is not parti ly behind the controversy about the necessity to describe images in a controlled or an open vocabulary weil. A project which makes images av lable via a controlled vocabulary still very much like a printed edition: the work group responsible for the initi description is doing the „right“ thing; the user of the images later is very much in a p v e role,
„consuming“ the correctly prepared material. It may be, that this model can be defended in history of the art, where after all the nurober of items is relatively small. lf the new forms of history, focusing upon the strata in society below the small elites, which documented in fairly small corpora, shall ever be closely connected to the content of the sources, as traditional historiography has been, we will have to strive for types of edition, which make material available, before a huge editorial sta h invested heavily into each individual page.
We mentioned above, however, that actual pal ographical work asks for addition tools, which are hardly available in current software. To integrate such palaeographic capabilities, at le t the following tools would have to be available within the standard working environment:
• The capability to handle a !arge nurober of high qu ity images of individual characters synchroneously on the computer: „high number“ probably translates into something like thirty, or more precisely „sets of thirty“, where a „set“ can speedily be exchanged against another one, meaning, that something like two- or threehundred subimag have to be available at any one time for rapid swapping into the current work area.
• The capability to superimpose images above each other. To do tbis, however in a which lows:
• To abstract a character, say •r• to a b ic shape, which is described a t of graphically displayed relationships betwecn various parts of the „ideal“ Ietter. (Say the ratio between height and length of middle horizontal stroke; the gle between middle stroke and vertical line; the relative length of a bottom horizontal stroke, if existing; and so on.)
• Two or more of these „abstracted“ letters shall than be superimpo d, by Op­ erations which allow symbolic modi cations of one of them. (Say „superimp „fl“ and “ “ after scaling “ “ to the height of „fl“ and rotating “ “ into
vertical.)
• The whole operation would have to be controlled by a mixture of interactive
graphic commands ( “ pick the ‚center‘ of the Ietter“ ) and symbolic Operations {like the ones described before).

54
Manfred Thai/er
5. Are Manuscripts Images?
We started by saying, that „manu ript pr essing“ can be preliminarily de ned „the application of image processing tools“ to manuscripts. We changed this Statement slightly, however, by specifying later, that in at le t one of the current focal points of di ussion, there is a great di erence between im es and manuscripts: The „necc sary qu ity“ of image displayed on the screen, is very much open to debate, pending the deci on, whether you consider an image just a preliminary illustration of a data b e which shall provide access to a collection of such images stored in more conventional ways or an object of study right on the screcn. A manu ript is either re able on the screen in the ori n or it is not; if not, the exerci is futile. A manu ript on the screen can either be improved against the original, or it can not; if not, the solution is severly
restricted.
The ond major methodological conßict about images seems currently, whether it is more u ful to describe them in a closed universe of a predetermined terminology, or to low arbitrary, natural language desc ptions in a not standardized vocabulary. Again, in the c of manuscripts the decision is more e y: A „description“ of the manuscript, which cont ns a transcription has the possibility of eventually replacing a printed edition; any other h not.
We would not doubt, that there is still a Iot to be done beyond transcribing a m uscript: in the c of „describing“ a charter, e.g., we could construct a similar dis­ cussion, whether it is more useful to „describe“ the shape of a Chrismon in controlled or uncontrolled vocabulary – but somehow this conßict seems to be relatively minor, given the major decision made before. Incidentially, this single example is just one instance of a qu tion lur ng in the b kground of the whole concept of an „edition“, where a photo­ graphic reproduction is available all the time: IC an editor de ribes the writing habits of the ribe of a cha er, it would be obviously u ful to integrate into the running text small snapshots of appropriate char ters and low the user to expand, while reading the de­ scription, th e sna hots into a section of the ori nal manuscript giving the context of the individu char ter. That is, for the formulation of „descriptions“ of manuscripts we tu­ ly would n d a major new concept in dataproce ing, being an „enhanced string“, which actually would low a mixture of ASCII char ters and arbitrary portions of bitmaps to be ministered a fully integrated data type – that is not just being displayed in an appropriate way, but searched for, rted and administered in powerful data b es.
Here, we f l, a radic change of style is indicated. While so far, we have di ussed what would be needed to make a technolo useful for a historian under certain circum­ stanc , now we have to discuss, what technical t ls have to be forg for such a purpose. We therefore separate the discussion of this point into an appendix, which is addressed to the technologically highly sophisticated reader, presenting a bluprint for a technical solution for these problems.

The Processingof fanuscripts Appendix*
6. Design Proposal for a Nonlinear Data Type „Text“.
6.1. What a text „historical“?
Speaking on the most general Ievel, we consider a text to be �historic „, when it
describes a situation, where we do neither know for sure, what the situation h b n „in re ity“, nor cording to which rules it h been converted into a written report about re ity. On an intuitive Ievel this is exempli ed by c es, where two people with the s e graphic representation of their nam are mentioned in a set of documents, which p i bly could be two c es of the same �real“ individual being caught acting, which, however could al be homographic symbols for two completely different biologic entities. At a more sublime Ievel, a change in the color of the ink a given per n uses in an o cial corr pondence of the 19th century could be an indication of the origin supply of ink having dried up; or of a considerable rise of the author within the bureaucratic r ks. us just emph ize for non historians, that the second example is all but arti ci : indeed the di erent colors of comments to drafts for diplomatic documents are in the 191h century quite often the only identi ing mark, which diplomatic agent added which opinion.
What these introductory examples should demonstrate, is, that the text – the com­ puter interpretable representation of a written document – forms in historic re ch an intermediate layer between two other layers of information. On the one extreme we have abstract f tu knowledge about the v ious entfties described in a text, which lows the interpretation of it; on the other there are purely graphical characteristics of the written document, which may carry meaning, but need not do so.
That the second proble is a genuine markup proble is probably obvious: if we use a computer to prepare diplomatic drafts of the 19th century for printing, we ob ously n a way to describe a portion of the document being „written with blue pencil“. Which, at the time of the rst transcription is ex tly what it , a Iiter description of a graphic property, though during the proc of research it may well acquire a more abstract connotation, like author K. Simpson. This could of course be interpreted such properties being eminently tted to abstr t rules for markup, because at the time of producing the markup we have not yet the faintest idea what the nal representation in print, any, of the speci c graphic property is to be. Quite besides that at le t I cannot
very well see, how it should become possible to propose a nite Iist of such potenti ly signi cant graphical properties, there is a more b ic problem. We all the time have now been speaking about graphical properties which may represent some meaning. Which is another way of saying, that this graphic property is purely accidental. To bring it to a point: almost all the examples given in the discu ions on standardization during the l t few years dealt with how to tag a structure which is clearly understood and where the graphic representation is accidental. Historical work de s with structur in a text which
we want to discover, where the graphics we see may be all the clues we ever might get.
• The following de nition is – with minor modic cations also contained in the author’s paper Historical Information Science: Is there such a Thing? New Comments on an Old Idea. This paper h been presented in the fall of 1991 at the Academia dei Lincei in Roma; it is to appear soon in a volume edited by Tito Orlandi.

56 Manfred Thai/er
The re di culty behind this might be a mewhat imprecise de nition of what a „text� is to begin witb. To me it seems, that in almost all the contributions to relevant di u ions, a „text“ is seen either the starting point for research or the result of research: either what you get om a colleague to make your linguistic, stylometric or whatsoever an ysis om or what you are going to deliver to the printer and, potentially, at the s e time to other colleague. In the understanding of this document a text needs to be a considerably more dyn ic kind of thing, the fo ly treatable representation of the
current u mptions of a researcher about what his documents tually cont n.
Tbis on tbe one band means, that we have to provide facilities to mark up graphical attributes wbicb may quire substantial meaning; on the other tbere have to be provisions for a link between a text and a set of sumptions about portions of it. To go back to our initial e ple: wben we provide for marking a portion of a text representing the „n e of a per n“ , will have to provide for a link to some background data b e, wbicb cont ns d riptions of „rral“ persons, being repr ented in the text by kinds of con cting aphic repre ntations. State of the art data bases in history., actually carry this a step further, by providing parate links between the grapbical variat i n of a n e to an algorithm, which is supp dly able to lterout the „accidental“ orthographic variation of tbe name, before it is being linked to factu knowledge about the person this n e is a tag for.
So,abistoric textisintbisdocumentconsideredtobeanrepresentationof u mp­ tions about some bistoric re ity, cont ning on the one band descriptions of graphic properti , wbicb may require interpretation, at the other lin ges to repre ntations of knowledge, wbicb e connected to a speci ed portion of the text.
Tbis simple model h , bowever, to be extended into two directions. A „historical“ text is in our opinion metbing wbich b come to us under a consistent t of circumstanc : our interpretation of color notations c obviously be v id only witbin a corpus of materialswbichc eintoexistencewitbinonebureaucraticunit. Similarlythelangu eof a mediev cbronicle can be an yzed, at le t in the rst step, only within one copy of tbat chronicle, though it may have been transmitted, with minor variations, in a wbole f ily of texts. At the s e time, bowever, the reality de ribed by tbe proce of formulating a politic document can very orten be understood only, if two par lel sets of comments upon some drafts, by di erent brancbes of tbe bureaucr y, are interpreted synchroneously; and tbe „story“ told by a medieval cbronicle can only be analyzed, if it is seen complete p i ble: though metimes no sin e text exists, which cont ns l the parts of it. By rst approximation this means, we need a mechanism, to administer an integrated document an entity, wbich consists of ver layers of tr itions, e h consisting of me „text“ – i.e. a collection of words – which can only be interpreted in the context of me sumptions about tbe rul applicable to it. As, obviously, for some portions of the ctitious „true chronicle of x“ con ictingly tradited texts will exist, this Ieads directly to the requirement of a text repre ntation which lows a given portion of text to have more than one equ ly v id form. We bave, tberefore, also to provide for a mechanism, which allows a dynamic bandling of variants, which enables software, to treat one coherent representation of a text on a computer, if it would just consist of one manuscript il if it would consist of the logic sum of two or more manuscripts. The computer representation of a m hine

The Processing of Manuscripls 57
readable text should therefore in our opinion not only make it possible to handle variants, but to treat streams of tradition combined into a „text“ potentially equal.
Fin ly, we have to de ne the relationship between a „text“ as a running representation of a tradited document and a „text“ as converted into a database according to me abstract model. In our opinion these two repr ntations should be seen as very close to each other, allowing the database to inspect the natural language context out of wbicb its entities of attributes have been derived, and on the other band allowing the user of the machine readable text to jump from one portion of it to another portion which , irrespective of the langnage used, deals with the same abstract concept. More pragmatically: if you enter into a historical database a query Iike „When did the mon ,tery of St.X receive more tban ve solidi from a single tenant?“ we want to see at least the unstructured description of tbe relevant entries in the administrative records, if not tbe scanned image of the respective page, and when we encounter in our running text a peculiarly verb eulogy about a given benefactor, we would like to be abl , to get all other sources related to that benefactor, be the in the same source or not.
We are quite aware, that the requirements we just described can be met only in part by existing software: we think however, that tbere is small, if any, sense to concentrate completely upon the t k of bow to code data in such a way, that the can be h dled by present day software. As a matter of fact tagging a text exhaustively and completely, seems to create such an additional overbe , that I see spurious cbances at best, that any historian could be convinced to enter all needed tags by band. So what we de ne here is certainly no markup, wbich normally will and shall be entered by a historian: to pretend otherwise would in my opinion be nothing but ctitious. The sense of a standardiz text representation at le t in historic re arch – and decidedly there – can only be to create a means for the communication between software systems, not between human historians. As such, however, we need a medium that does take account of things to come and is bro enough to give software designers some re n, why they should invest into implementing components, which support such recommendations an xchange format. To do we need some foresight; which is why we start from a non-existing system.
6.2. A „historical text engine“
To allow us to do all the things speci ed in a coherent computing environment, would rst of all like to sketch how such an environment should behave.
We ume on the following pages, that all texts are treated as „informatioo strin “ .
A running text consists simply of a collection oflioearily ordered strings of this type; a data b e or knowledge representation consists of texts which are connected in a non-linear way. As every linear structure can be described a trivial c e of a non-linear one, running texts, (factual) data bases, full text b es, knowledge b es and, as we will see, collections of bit-mapped data objects are l to be considered speci c realizations of a general representation of information. To make that po. sible, we assume further, that all the nece ary string handling operations are taken care of by a „text engine“ which relies oo other software components to be provided with correct „information strings“ irrespective how they are administered. We will see, howev r, that links to other „information strin “ can be part of any of them.

58 Man d Thai/er
In any implementation of the following concepts, a „text engine“ could therefore be only realized in cl e connection with other dedicated ftware systems, which take care of administering the relationships between various information strings. These do not form part of the pre nt considerations. As the de nition of the various items of information to be handled requires references to them m�times, we will, however, just shortly de ne the three m t important tools of that type.
In our concept we did stress the similarity between a running text and a structured data b ; inde will later see, that we are also considering c es, where one collection of information strin can alternatively but synchroneously be interpreted a running text and a data b . To make that possibl , we sume that besides the text engine, which we cover here the following exist.
A text administrator. This is a very primitive program, which does not very much more, than performing I/0 on strictly sequentially stored collections of information strings. A text proce ing system in our concept would use such a text administrator to save and Ioad information strin from background media, which than are proce ed with the help of tools from the text engine. Whenever we use the term „textproce ing“ in the remainder of this paper, we refer to software, which p rforms typical t ks of current day textprocessing, including primitive full text retriev applications, by using the services of both, text administrator d text engine.
A data b engine. This is a f ily of software tools, which are responsible to
administer non linear collections of information strings. These software components are
r ponsible for the handling of problems resulting from the adaptation of current re­
trie concepts to handle context sensitivity of queries and uncertainty or ambiguity of
structural relationships.
·
A knowledge en ne. This is a family of ftware tools, which are responsible for the ministration of such conversion d transformation processes, which are built upon knowledge are b d upon dictionary-like structures or complex sets of rules. all information is suppo d to be evaluated dynamically, the ware components in turn use components of the text engine, when the need to handle information strings arises.
The „information strings“ which are treated by our sumed text engine in th�­ vironment sh ed with these other major modules, consist of linked lists of „uninterpreted items“ , which e st in an „interpretative environment“ . Whensoever an information string is handed to the text en ne, it is gu tueed, that the later is supplied with a full copy of an interpretative environment.
While more precise de nitions of interpretative environment and uninterpreted items will be given shortly, it makes their respective roles probably e ier to understand, when we describe them mewhat intuitively rst. As a rst approximation, we could consider the interpretative environment a table of mappings of abstract font commands into concrete printer Operations. So when we Iook at the text engine Operation „prepare output on a spe­ ci c printer“ the printing of the string starts with l such applicable parameters regarding printing and spacing, can be derived from the interpretative environment handed over with the string to be printed. After this, the uninterprcted items are inspected and item by item converted into such strings and/or printcr commands repre nt their output

The ProcessingofManuscripts 59
form in the light of the current interpretative nvironment. While this is a description of a current day printing proc s, in our opinion it should get further: the �ront“ of a text not only being relevant, when it is being printed, but also, when a string in font �A“ is compared to a string in font �a··.
A very important con uence of this separation between uninte.rpreted items and interpretative environment h however to be clari ed already now. As mentioned initially, we deal not immediately with the question of markup. We sume, however, that the inte al representation of a historical texts indeed needs some features, which are inherently like a symbolic markup: i.e., some information about how the text shall be pr d , which is interpreted only, when the text is being processed. This produces a subtle di culty, when we are speaking about non-linear structures of text, where individual parts of the text sh l be ce ible. we will see further, the interpretation of character i of a text may depend on some information, that is contained between character i- 1 and i – 90 of that text . So, if we want to interpret the i1h character correctly, we would have to know, that information relevant to that character occurs before it in the string. Therefore we assume, that a „string of information“ , we define it, is always ministered so, that it is only cessed at a point, where it can be guarantueed, that all information nec for it ’s interpretation is available. More formally, we speak of entrance points into a collection of strings, where a complete copy of the interpretative environment for the following character is available. All characters between two entrance points can only be correctly interpreted, when the text engine re s and interprets rst all parts of the string of information, which e situated between the ne est entrance point and the char ter in qu tion.
The import ce of this concept can scarcely be overestimated. Ind d, the need to provide a su ciently but not unneci arily !arge number of entrance points, is the m n reason, why we distinguish so sharply between a strictly sequential and linear text administrator,astrictlynonlineardatab eengine,which,however,can u me,that from e h of its items a path to the nearest applicable entrance point is de ned, and a knowledge engine, which bandies dictionaries of relatively small information strings, each of which h its own entrance point, the can be acces d completely at random.
60 Man d Thai/er
6.2.1 p of uninterpreted items
Strings of uninterpreted items are made up of ve di erent cl e s of items:
• B ic items.
• String qualities.
• String lin .
• St ng variants.
• Em ded structur .
The roJe of th e cl e s of constituents are in turn:
6.2.1.1 Basic items.
Th items carry the actu information derived from a historical source. In the most trivi c e, the consist of simple character codes. AU such items, however, are considered to have lo c ly the same rank. That is, a small bit map (e.g. for a non-deciphered Jangu e the Indus hieroghlyphs or a non-textual symbol, like a water mark in paper) or a plain ASCII character can both form distinct items of a „string“ in our sense. This u m ,thatthetextenginecontainstools,whichcansortandcomparealltypesofb ic items.
The following types of b ic items are de ned:
• ple te .
• Cb acter tokens.
• Bit mapped tokens.
• Pictu .
6.2.1.1.1 Simple characters.
Simple char ters are described by a sequence of n bytes per character, n defaulting to one in m t text engin . It is sumed in this paper, that char ters which represent lette , have one c only. For re ns which are ven further below, it is umed that is j another string quality which does not justify a special treatment. Each simple char ter repr nted by a numeric value, whlch indexes a table that contains a variable amount of information about the character. That information con sts of:
• Sorting p ition of the character within the table.
• ‚Binding‘ ofthe character. By this propl’rty we de ne its behavior in conjunction with
neighboring items to it’s left and right within the same string.
6.2.1.1.2 Character tokens.
A character token is represented by a traditional – henceforth called primitive – string of simple characters; in most real-world application starting with a common e ape char ter. Wbile being repr ented by a primitive string, the are conceptually, however, j the same simple characters: the degree of similarity between two text tokens – , e.g, exp by a table of rting values – is therefore completely independent of the string representation of the tokens. As the two primitive char ters „a“ and „A“ may or may not be considered identic , independent of the code values signed to them, in a historic text the two text tokens „\chrismon“ and „\cross“ may or may not be con dered identical or similar; there is, however, no inherent relationship created by both tokens starting with the primitive string „\c“.
The Proces� ng of Jlanuscripts 61
6.2.1.1.3 Bit mapped tokens.
Bit mapped tokens are tokens, for which all is valid, what h been said about the properties of text tokens. Bit mapped tokens do not consist of a sequence of primitive characters, however, but of a sequence of the form: escape_character-length-bitmap. A further di erence is, that their similarity is de ned not by a tabular listing of their relationships, but the decision rules for the compariosn of the bitmaps themselves.
6.2.1.1.4 Pictures.
Intuitively pictures are obviously the same as bitmaps: indeed, their internal repre· sentation is assumed to follow the same rules, just given in the prcceding ction. While a bit mapped token is assumed to be an atomic item of information, a picture sumed to be a po ibly structured entity, which may occur as part of a text, will more often be connected to it, however, by the mechanism de ribe in section 3.1.3.5 for text links.
6.2.1.2 String qualities.
As mentioned initially, each uninterpreted information string exists in an interpre­ tative environment. This is de ned by a number of assumptions, which are true for the rst information of the string. The information string contains, besid the b ic items discussed sofar, which carry the „real“ information, indications for a change in any of th e u mptions. This implies for the text engine, that all of its constituents are guarantueed to start the processing of a string only at weil de ned starting points, all Operations de ned on the strings parsing allong them. While this may em to be a distraction, we would like to emph ize it here, otherwise the concept of string quality cannot be understood.
Every string of information exists in an environment which de nes its
• modes,
• style,
• color,
• size and
• view.
It should be noted here, that these names have been cho en for intuitive plausibility, as have the ex ples bel . The exibility of the concepts, however, is to be derived from the abstr t de nitions given.
6.2.1.2.1 String modes.
String modes de ne the absence or the presence of a set of attributes. That is, a given item of information can have an attribute or can miss it. It is not possible, to have a mode in a certain degree. Every string of information inherits from its interpretative environment a set of default modes. If a certain mode is not de ned in the environment, it is assumed to be absent.
The most intuitive example of a string mode is the case of a character. we de ned before, that simple characters are assumed to be caseless, it would completely depend on the interpretative environment, whether the string
this is a string
would be interpreted by the text engine upper or lower case. By interpretation we mean in this and all following examples, the behaviour of all components: an „upper

62 JHan d Thai/er
c e“ string would be printed upper c e (if possible on the output device) but its being upperc e would also in uence comparison op rations (see below).
At any point in a string a mode can be activated or deactivated:
�his is a string
would always result in an upperc e string, irrespective of the sumptions of the
interpretative environment,
: �his is a string
always in a lowerc e one. The interpretation of
: � – is a string
is clear.
e h mode, which not explicitly de ned in the interpretative environment, is u med to be absent, their nurober is arbitrary and h not to be known by the text engine. Modes which are encountered in an information string, for which the text engine h no explicit instructions are therefore completely ignored. ln the c e of
: � is
the mode „german“ would in most search operations be ignored; in full text or data b applications it could, however, be used a selection criterion irrespective of the structure which de nes the relationship between this information string and all others in the currently administered data; in Anglosaxon text processing applications it could be interpret underlining.
6.2.1.2.2 String style.
While any item in a string of information can at the same time have an arbitrarily !arge nu ber of mod , it alway h precisely one style. Statistic ly spea ng, the style of an item handled on a strictly nomin Ievel: there are no sumptions about any rel ionship between two different styles expres d in the internal representation ofstyles.
The most intuitive example for the style of a text would be font information, in the example
this is a !Style: german keichenkettJStyie: basic I
Pie note, that this is not exactly the same than the previous example: while in the previous one, „z“ could acquire the mode case, without loosing the mode german, no part of Zeichenkette could acquire the style gothic without loosing the style german.
6.2.1.2.3 String color.
The quality of color is similar to that of style, by its values being mutually exclusive.
Its intuitive interpretation is probably obvious and the introductory remarks of this paper show a potential application. For a systematic interpretation, however, it is much more important, that this quality is supposed to represent statistically an ordinal Ievel. That is, it u med to be represented internally by ordinal numbers, which allow expressions of similarity. (A similarity to the implementation of the en concept in the ‚C‘ programming language is intentional.) The two strings:
s !Mode: +german I
ichenkett o e +german I
The ProcessingofManuscripts 63 I Co ; eorg Smith
and
!Color: light bluebeorge Smith
would in m t interpretative environments therefore be sumed to be clo r to e h
other, than the strings:
IC o ; eorge Smith and
!Color: li t redbeorge Smith
It uld , however, not be possible, to express the di erence in the degree of similarity between the two pairs of names. (This is a statistical stat ment and a de nition of the concept of color, not a Statement about artistic andjor biologic perception of colors.)
6.2.1.2.4 String size.
String size, too, has a pretty obvious application. It is similar to the concept of color, lows ditionally, however, to express a difference iu the degree of similarity between two strings of information being compared . In the three fragments:
Charter a: ISize: trinitatia …
20pt �chrismoJSize: JOpt �chrismo ze:
10pt I nomine individu�
lOpt I In nomine individu�
In nomine individu�
Charter b: ze:
trinitatis
..•
Charter c: ISize:
trinitatis
•..
the chrismon in a is more milar to the one in charter b therefore, th to the one in charter c; tbe proportion between the sizes of chrismon and main body of ript, however, identic between charters a and c, while both are dissimilar to b in preci ly the e degree.
6.2.1.2.5 String views.
The qu ities fo far – with the possible exception of color – may be seen an attempt to de ne cl i cal type tting attributes in a su ciently systematic way to low their interpretation on an intermediate Ievel between typographical repre ntation and conceptual understanding. We recapitulate: the color of a note in a diplomatic document may ultimately quire some meaningful, abstr t interpretation ; at tbe be nning of an editorial proce , however, it will be ex tly what it Iooks like: proof that Mr. X used a blue pencil.
The concept of string view, on the other band, h been introduced to handle phenom­ ena, which often occur in manuscripts, have, however, no generally cepted typographical conventions signed to them.
Typic examples would be portions of a text, which are Iegihle and obviously part of the original manuscript, but which later have been crossed out, additions being added at the same time, or manuscripts, which have been written by a number of scribes, some of which can be identi ed, while others can not clearly be distinguished. Obviously th e
40pt �chrismoJSize: 20pt I
64 , anfred Tha/ler
properties of a manu ript could in principle e covered y the string qualities given so far.
The tools provided far, did always a ume, however, that each of the qualities would exist one: a of binary qualities, exactly one nominal quality, exactly one ordinal and exactly one which allows comparisons of degrees of similarit . To eneralize this model,
weintroducetheconceptofatextview,whichisde ned View:T e,Name,n . it’s first argument, it accepts any of the previously identi ed text qu ities, i.e., mode, style, color or size. lt introduces a named text quality, which h the properties discussed so far. So our previous notations could be seen shorthand for a more gener text view notation. The following equivalences would hold:
n iew: mode,detault, nI
1 – – i e w : s tyl e , d e t a u l t , n I
!Style:
nI i : color, default, nI nI== iew: size,default,nI
n
The di erence between these two de nitions is, however, more important, when it comes to actually impi menting such a model. We sume, that a text engine optimizes the four default views with regard to speed of processing of individu information string. This means, that when one of the default views is encountered in an information string, it will be taken c e of by an extremely quick operation. When an explicitly named view is encountered, however, the text engine is allowed to reorganize the interpretative environ­ ment to allow for it. (All comparison operations have to allow for size sensitivity without l of e ciency; a comparison which h to allow for ve independent sensitivities for vievs of type dze, however, is allowed to be signi cantly le e cient than a comparison that b dies just the def ult size viev).
This di erentiation – and more so the sp e it is i gned – may be a re ection about the author’s b kground in tual program development: we consider this di erentiation to be extremely i portant, however, , on the other band we sume, that historic texts can be handled correctly only, if the nurober of vievs allowed is unlimited.
6.2.1.3 String .
While we consider string qualiti to be a more systematic description of cl sical textual properti , string links de ne the conditions for a embling individu information strings into !arger objects, like texts or data b es. B ically we consider it necessary to embed into a string reference points, from which it is po ible to branch to other strings. The intuitive example for this would be a footnote.
As we mentioned initially, we consider a text not so much to be something which primarily h to be printed, but a representation of the current knowledge about some historical phenomenon. All such points, where it shall be p sible to branch from a given point of reference within a text to somewhere else, are therefore m aningful only being connected to speci c Operations of the sumed text engine.

The ProcessingofManusc pts 65
These Operations are:
• brancbes,
• text references,
• data b e references,
• knowledge references a.nd
• bitmap references.
6.2.1.3.1 Branches
A branch is the m t simple string link. It consists of a pair of dress , connecting an arbitrary point within a string of information with the entra.nce point into another string. The intuitive example for it is a note in textproce ing. Branches pointing from an arbitr y point of an information string to the entrance point into another string, we will call exceptions d denote with the symbol I; branches from the entrance point of a string to an arbitrary point of a.notber information string, we will call reference and symbolize by1- I
In the c e of a footnote, these elements would be used follows:
in recent publications I this point is usually not discussed any more . . .
1- I Cf. John Smith:
The text engine r olves the arrows in these string links follows:
• exceptions are pla.in pointers to the beginning of another string of information, allow­ ing the interpretative environment to be initialized the standard way.
• references are simil pointers to the arbitrary point from which the exception did branch away. They can, however, only be traversed, if this point in the collection of strings h been re hed via a previous refernence from the corr ponding exception. In such c s the text engine st ks a copy of the state the interpretative environment h been in, when the exception w tivated. the reference reached by any other navigational operation witbin the collection ofstrings in question, it is not p sible to follow it to the spot of the exception.
6.2.1.3.2 Text Referenc
Textreferencesallowittobracketaspeci cportionoftexta.ndlogic lyto e mble suchportionsintoaspeci ccollectionoftexts. Anintuitiveexamplewithintextprocessing would be tbe creation of registers.
Text references consist of pairs of the form
I some string I
which we shall di uss „forward reference“ , „reference string“ and „b kward ref­ erence“ respectively.
A forw d reference consists of
• a mark pointing to the end of the reference string,
• a pointer to the next forward reference in the collection of strings with the same name
and
• a pointer to the nearest entrance point into the information string containing the next forward reference in front of it.

66 Man ed Thai/er
It enables a text en ne therefore, to navigate from one text reference immediately to the next; does not remove the necessity, however, to interpret the portion of the information string in front of the respective forward references to bring the interpretative environment into the state it h to be, when the reference string shall be interpreted correctly.
A backward reference contains the same information, does so with respect to the preciding text reference in the collection ofinformation strings in question, however.
6.2.1.3.3 Data Base References
Data b references have no precise equivalent in traditional applications. A speci c data b e pointer � is de ned by
• the data b which sh l be referenced,
• a procedure specifying for the data b e engine in question, how the reference shall
be converted into an information string.
When control returns to the text engine, a modi able copy of the information string created in this way is brought to its disp al and for purposes of textprocessing integrated transparently into the „text“ proper. Obvious, but trivial examples of usage would be the dynamic treatment of bibliographical and/or biographical information.
6.2.1.3.4 Knowledge References
Knowledge references are denoted by identical bracketing reference symbols of the form
� object text �
Object text refers to an information – like complex chronologies, historical currencies and similar, described in previous papers by this author – which needs to be trans· formed, before submitted to speci c cl e s of treatments. (Think once more of sorting or comparisons.)
The knowledge referenced is u med to consist of a formal de nition of the format a object text h to have to be proc e d plus a collection of suitable diction ies to apply specific transformation rul .
A speci c consists of
• a speci cation of the knowledge (b e) to be accessed plus
• a set of Operations of the text engine, which trigger the transformation of the object
text.
For l such Operations, the text engine does not process the object text itself, but the result of the conversion.
6.2.1.3.5 Bitmap References.
A bitmap reference consists of a p r of identic bracketing symbols of the form � interpretable equivalent �
It consists of
• a reference to a bit image, which either is administered a „picture“ de ned above
or a completely independent le and
• a set of Coordinates within the bitmap.
The ProcessingofManuscrpi ts 67
The text engine references the bit image, when suitable display units are av lable, uses the interpretable equivalent, however, when functions are being called, which require operations like arch or compari n.
6.2.1.4 String variants.
String variants have an obvious application: the administration .and proc sing of the apparatus criticus of a text. The following model for the integration of variance into a gener text representation model this only a starting point, however. We rather sume application the creation of dynamic text repre ntations, i.e. of text representations, which allow so ware to treat v ianc a text „equal“ . This could be envi ged by display modules, which low the u r to switch om variant a variant ß by hitting a function key, the displayed text in c where variants exist in both st es following the readin of manu ript a or ß.
It umed that – unless indicated otherwise by the interpretative environment – an information starts with a part that is present in all witne es for a given text. This assumption is changed, soon in the processing of the information string a block border is encountered. Block borders, which can be nested, Iimit parts of an information string, for which substitutable variant readings exist.
Generally a text with variants would therefore be represented
taxt present in all vitnesses !Block >I variant read s Block
text present in all vitnesses
The variant re ings consist of blocks of the form
ariant:
where nam et speci es in which witn s a given reading is pr nt.
The ( -, ) and ( , – ) symbo indicate, that l variant re ings are lin d a Iist, where e h atom h a pointer to the next variant re ing, but at the same time a pointer to the end of the r pective block to speed up proc sing.
The mechanism is probably best explained by two exampl . Lets Iook at the situation where manuscript a ves this is a text ot vords, while manuscript ß re s this is a string ot characters. In this c e we sume a repr entation which :
this is a Block
ot
When, to show nesting, we d manu ript which reads this shall be a sen­ tence, we get:
Name (-, )Itext of reading – – – -: a-r a-n
68
Ma d Tha/ler
a
!Block I
As th is m Iook mewhat complicated: Iet me remind the gentle reader, that we are d ribing here a structure, which intern ly sh l be able to handle something. Entering m kup into a text, which in turn is parsed to provide the required pointers, would be one way to hieve this.
6.2.1.5 Embedd structures.
While the p ng parts of th paper form presumably a general model for the representation of historic texts, this section may be more speci c to the activiti of the author. It de s with the problems from repre ntation, where a data b e, which follows a network model, is inte ated into a text . The idea is, that data b e queries e proc d , which are tr slat via structur information contained in a data dictionary into such navigation proc are required to l t various items of information for proce ing. Unlike in tradition data b es, that information is, however, not „extr ted“ from a data b . Inste of extr ted information from natural text, we sume, that the text a whole is repre nted on the m hine, various pointers – similar to the cl of such, which de ribed far – being „hidden“ within the running text, which de ne that a given p t of it is not just a series of characters, but at the same time the content of a structurally de ned field of a data b e.
The idea is, that we have a text, like the sentence On the 1
Winslow d George B min on appeared before tb court to cl m . . . . which by ftware b d upon our ide text engine can be handled for typical purp of textproce ing, other software b d upon this en ne can at the s e time, however, treat terms contained within it – like John Winslow – the value of the attribute n e of an incidence of the
: Variant
of M ch 1 764, John
The Processing ofManuscripts 69
entity peraon. Applications would be tasks like: Select all court records, where people who have been bo more than fty miles from a given city of reference appear inter­ ested p ty; select out of the running text all portians which are marked „quotations“; compute a set ofstylometric indices for the vocabulary used in such quotations.
ready mentioned, we do not claim in this ction, that our considerations here are general. We ume, that such models of the combination of structured representations of information and the covering text containing that information, c only be realized with datastructures that allow inconsistent information to be handled and avoid normalization procedur . As such models are few, we have so far only considered the problem of „hiding“ � E data structures within the text. In the c e of that system, the data are structured into a network whicb consists of entities11 called groups which among themselves can be linked by arbitrarily many relationships. Groups have an arbitrary nurober of attributes, called elements, which conceptually consist of variable length arrays allowing an arbitrary nu ber of values, called entries, for each attribute.
To hide a network like that within a collection of information strings, as we described them so far, we see two possibilities:
On the one band, the structural network of a data base could simply exist parallely to a text we described it here, all entries of the network consisting of pointers to the relevant portion of the text, tagether with length information. (Obviously these pointers would have to consist of the nearest entrance point of the text plus an offset to the relevant portion.) The vantage of such a construction is obvious: existing datab e software could be taken over more or less unchanged and all textual functions needed, are cont ned per de nition in a text engine, we de ned it so far. In our opinion this otherwise obvious model h , however, two major shortcomin : obviously updating the text could e ily corrupt l the addres s cont ned in the entri of the data b e. serious, as this proble is, we could handle it conceptually, by allowing for links with a slightly modi ed de nition, which would have to be inserted into the text at each point, which is the target of a data b referencing it. While this update proble is certainly tricky, it could eventually be m tered by the techniques hinted at.
Conceptu ly much more severe is a problem, which is introduced, when we consider even the most simple concept of data types. We already have introduced in section 3.1.3.4
above the concept of a knowledge pointer ( ). We did so, because of the necessity to treat „special kinds of text“ in a speci way. It is not nsible to sort calendar dates in alphanumeric f hion; d converting tempor notation related to the medieval saints into a notation tbat can meaningfully be sorted is no completely trivial task. This, however, is a typical data b e problem. Of course a nurober of solutions could exist: one would
be to include the fu nctional equivalence of a � into the data dictionary for this eld
11 We discuss here the proble of hiding data structures within a general text representation. �>.Etw’s groups are not entities in the sense of conventional DBMSs: to clarify, that they are not, we c l them groups; a precise de nition of the differences would go beyond the scope of this paper, we do not give it however, but mention that envisaging them entities will not be totally wrong. The same is the case for „elements“ and „entries“ introduced below.

70 Man d Thaller
– which me s a type of redundancy which dangerous. Another solution would be, to
Iet the Ipoint to the data dictionary, rather than to the relevant knowlledge
�
to begin with. This could mean, that we need di erent implementations for a 1 KB that occurs a text without a hidden structured data b e than the one u d in texts with such an underlying structure; a solution. Avoiding that, however would bring into the situation, where texts without an underlying dummy structure would not be allowed; not a b lution, but outright impossible one.
To avoid th situations, we propose therefore to discuss data models, which are per definition built into „natur texts“. This would seem unusual from the point of view of other di iplines, but in history – or rather in the source-orient model of historic dataproc sing – data b es, this author h pointed out elsewhere, are always not so much collections of information, which de ne their own reality, but an attempt to structure information that h been tradited in a coherent form, i.e., usu ly a text.
element pointers d entry pointe (denoted Gr :
In the c of € , we consider this p sible, by allowing the text engine to admin­ ister three dition cl e s of constitutents of a strin of information: group pointe ,
e
N , I
d IEntry: e I res ctively). They are de ned a generalization of the concept of a text reference introduced in tion 3.1.3.2. The symbol replacing the I compo­ nent of the I I I notation is sumed to indicate, that this generalization lows for an unlimited nu ber of references starting from each emb ded structur sym­ bol, while our textu referenc provided ways for precisely on physic reference.
6.2.2 The Interpretative Environment
The interpretative environment, which we have known intuitively so far a kind of re ement to the concept of a printer driver consists of two independent components.
There s a table which d rib for ea h information string to be proc d the text qu iti which it h , when any b ic item is being encounter by any component of the text engine. This part of the environment h to be complete: that is the re on, why introduced the concept of entrance point into an information string. This part of the interpretative environment is lo ed when sever a particular information string is pr nt to the text en ne.
The second part of the interpretative environment de ribes not an information string, but the status of the text engine itself: it consists of applicability information for each p b le textual qu ity and option ly a mapping of that quality to an input convention or an output property. This concept c b t interpreted, if we I k at the idea of c e n tivity in text operations. C e nsitivity would be modeled in our concept a state of the interpretative environment, where c e is an applicable mode (and char ters are during output mapped to di erent representations).
6.2.2.1 Applicability orModes
The applicability of a mode consists of a statement; if the status of this mode sh l be ch ked, when a b ic item is being encountered by the text engine. If during printing the m e german is applicable, a mapping of characters with this mode into another font (or

The P cessingo[Manuscripts 71 underlining) will be u d; if it is.not applicable, such mapping will be ignored. Mutatsi
mutandis this is so the c e, when comparsions are performed.
6.2.2.2 Applicability of Style.
The applicability of style consists of a Statement, if it shall be checked when a b ic item is being encountered by the text engine. For applicability style could be interpret like a mode: if it is applicable, any di erence in style make two b ic items un u ; there is no way to de ne a degree of di erence.
6.2.2.3 Applicability of Color.
The applicability of color de nes whether optional 1/0 mappings shall be used, d a range, within which otherwi identical b ic items have to be considered equal with reg d to color. This range can be given a pair of ab lute numeric values, or as a table which specifies for e h color in one information string, which colors in the other are cceptable.
6.2.2.4 Applicability of Size.
The applicability of size de nes whether optional 1/0 mappings shall be , and a range, within which otherwise identical b ic items have to be considered equal with regard to size. This range can be given a pair of a olute numeric values, or percentage of the !arger of two b ic items, within which a smaller one is still to be considered equ .
6.2.2.5 Applicability of Views.
As views are a more gener form of the other textual qualities, their applicability identical to that of the default modes of the four cl s ven above. The implementation con derations ven in 3.1.2.5 de ne the di erence between the applicability of any named view of the four types and the four default views of each type.
6.2.3 The Text En ne
What kind of tiviti our hypothetic text engine is perfor ing, w shown intu­ itively more or le durlog the preceding sections. Obviously it is related to the production of output; obviously it is performing compari n operations: our example of a „ger­ man nsitive“ compari n an equivalent to the option „c e nsitivity“ of current text proc i ng so ware impli much. Though the de nition of such a text en ne certainly not part of a discussion of text repr entation, we would like to include a brief, but somewhat more systematic, de nition th given so far. The re on for this is, that we sume, that recent discu ions on text representation b a tendency to concentrate a bit to much on printing. We would therefore like to describe operations, we think nec to make u of l the qualities d ribed in a more general way, i.e., in uencing any kind of operation the type of text we describe here h to undergo.
When ever we define in the following tions an ability the „text en ne shall have“, this is a abbreviated expre ion therefore for the following re oning: „We sume that processing historical data of requires a speci c ability. lf historical data are to be proc d by more than one ware system, we therefore need a standard for the encoding of the property calling for this ability.“

72
Man d Thaller
6.2.3.1 Texts Handled.
The text en ne sh l be able to handle texts, which are mixtures of byte coded and bit mapped items. All items a text consists of h to be proce able by components in qu on. Which d item h , h to be tr sparent for any applications progr mer ng pr id by the text engine.
6.2.3.2 Import / Export.
6.2.3.3 Comparison and Sorting.
The text en ne sh l have part of its interf e functions which e able to compare d rt ts of information strings, fully controlled by interpretative environment d above.
6.2.3.4 arc g.
The en ne sh l be able to u any informations string, which speci cally includes such cont n bit mapp items, a arch key in the administration of !arge string coll ti s, like diction ies.
6.2.3.5 1/0.
The t t en ne sh l be able to convert its internal repr entation into a form which
repr nts its v ious cl of items on output devices, which do not provide me s for the m t obvious kind of presentation. It shall cont n parsing functions, which conv rmats provided by input devic or ftw e supporting phisticated forms of inpnt a common intern pr entation.
The t t en ne sh l provide tools to import and export strings. This means, it shall le nvert its intern representation of an information string into a form that y d tinguish betw n di erent cl s of items, sp i cally between byte coded and bit mapp on , ftware components, which do not have the ability to handle both, tr t tho portions of a text, which they can handle. This export format, which is
d ri
e em text format h to be transferable on standard communication links.
Halbgraue Reihe
zur Historischen Fachinformatik
Herausgegeben von Manfred Thaller
Max-Planck-Institut für Geschichte
Serie A: Historische Quellenkunden Band 14
Erscheint gleichzeitig als:
MEDIUM AEVUM QUOTIDIANUM
HERAUSGEGEBEN VON GERHARD JARITZ
26
Manfred Thaller (Ed.)
Images and Manuscripts in Historical Computing
Max-Planck-Institut für Geschichte In Kommission bei SCRIPTA MERCATURAE VERLAG
St. Katharinen, 1992

© Max-Planck-Institut f Geschichte, Göttingen 1992 Print in Cermany
Druck: Konrad Pachnicke, Göttingen Umschlaggestaltung: B ta Werbeagentur, Göttingen
ISBN: 3-928134-53-1
lntroduction
Table of Contents
Manfred Tballer. . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . .. . . . .. . . . .. . 1 I. Basic De nitions
Image Processing and the (Art) Historical Di ipline
.Jörgen den Berg, Hans Brandhorst and Peter van Huisstede ……………. , ..5 II. Methodological Opinions
The Processing of Manuscripts
Manfred Tballer………………………………………………………..41 Pietonal Information Systems and the Teaching Imperative
FrankColsonandWendyHall………………….. ………….. ………….. The Open System Approach to Pictori Information Systems
WendyHallandFrankColson……………………… ……………………87 111. Projects and Case Studies
Tbe Digital Processing of Images in Archives and Libraries
PedroGonz lez………………………………… ……………………..97 High solution Images
AnthonyHamber……………………………………. ………………..123 A Supra-institutional Infrastructure for Image Proce ing in the Humanities?
penS.Ore. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135 D cribing the Indescribable
GerhardJaritzandBarbaraSchub……………………………………… 143
Full Text / Image DBMSs
RobertRowland………………… ………………………. ……………155

lntroduction
Manfred Th ler
This book is the product of a workshop held at the International Univer ty Institute in Firenze on November 151h, 1991. The intention of that workshop h b n to bring tagether people from m y di erent approaches to „im e pr ing“ p b le. The re on for this „collecting“ approach to the subject w a f ling, th t wbile proce ing in many ways h been the „hattest“ topic in Huma.niti computing 1n r ent years, it may be the le t weil de ned. It seems so much b der to say in this are wbat is speci cally important to historia.ns, tha.n to other people. In that situation it w feit, that a foruin would be helpful, which could sort out what of the various approach can be u ful in historic rese ch.
To solve this t k, the present volume h been produced: in m y ways, it re ts the di u ions which tually have been going on less, than the two comp ion volum on the workshops at Gl gow a.nd Trom do. This is intentional. On the one b d, the p ticipa.nts at the workshop in Firenze did strongly feel the need to have proj represented in the volume, which were not actu ly present at the workshop. On the other, the di u ions for quite me time were engaged in cl i ing what the metbodol i issues were. That is: what tu y e the topics for schol ly di u ion beyond the description of individu projects, when it comes to the processing of imag in historic rese ch?
The situation in the ea is made di cult, because some of the underlying umptions are connected with vigoraus re arch groups, who u fora of schol ly debate, which only slightly overlapping; , what is t itly sumed to hold true in one group of projects may be considered obviously wrang in a.nother one, that it cely d explicit refutation.
We hope, that we have been succes ful in bringing some of these hidden di erenc in opinion out into the open. We consider this extremely import t, because only th cl fication allows for a fair ev uation of projects which may have st ted om di erent sets of sumption. So importa.nt, indeed, that we would like to catalogue here me of the b ic di erences of opinion which exist between image proce ing projects. Tbe re er will rediscover them in many of the contributions; editor I think however, that sum izing tbem at tbe beginning may make the contributions- which, of course, have b n striving for impartiality – more e ily rccognizable parts of one coherent debate.
Three b ic di erences in opinion seem to exist today:
(1) im e processing a genuine and independent eld of Computer b d r cb in the Humanities, or is it an auxiliary too ? Many projects sume tacitly – d me do quite outspokenly- that imag� on the computer act illustrations to more conventional applications. To retrieval systems, illustrations in cat ogues and the like. Proj ts of this type tend to point out, that with currently e ily available equipment d currently clearly understood data proce ing technologies, the an ysis of imag , which c quite e ily be h dled illustrations today, is still costly and of uncertain promi . Wbich is the re n why they u me, that such analytical appro hes, if at all, should be undertaken

2
side efef cts of projects only, which focus upon the relatively simple administration of images. Their opponents think, in a nutshell, that while experiments may be need , their over J outcome is so promising, that even the more simple techniqu of today should be implemented only, if they can later be m e u ful for the advanced t hniques now only p ti ly fe ible.
(2) Connected to this is another con ict, which might be the m t constant one in Humaniti data processing during the l t d es, is particularly d isive, however, when it comes to image proce ing . Shall we concentrate on Ievels of pbistication, which are av lable for many on today’s equipment or shall we try to make use of the m t phisticated tools today, trusting that they will become available to an incre ingly !arge number of projects in the future? This specific battle h been fought since the earliest years of Humanities computing, and this editor h found bimself on th sides at difef rent stag . A „right� answer does not exist: the debate in image processing is probably one of the best occ sions to understand mutu ly, that both positions are full of merit. It is pointl to take permanently restrictions into consideration, which obviously will cease to exist a few years from now. It discredits l of us, if computing in history always promises r ults only on next years equipment and does not deliver here and now. Maybe, that is ind one of the more important t ks of the A ociation for History and Computin to provide a link between both worlds, Jending vision to those of us burdened down by the next funding deadline and di iplining the loftier projects by the question of when metbing be ordable for all of us.
(3) The third major underlying di erence is inherently connected to the previous ones. image such is beautiful, but not very u ful, before it is connected to a description. Sh l such de riptions be arbitrary, formulated in the tr ition ly clouded langnage of a histori , perfectly unsuitable for any phisticat t hnique of retrieval, maybe not even unambigously understandable to a fellow historian? Or shall they follow a prede ned cat ogue of narrow criteria, using a carefully controll vocabulary, for both of which it is mewbat unclear how they will rem n relevant for future r arch questions which have not been ked so far? – All the contributors to this volume have b n much to polite to pbr their opinions in this way: arcely any of them does not have a strong one with reg d to this problem.
More questions than answers. „Image proc sing“, whether appli to images proper or to di t ized manu ripts, ems indeed to be an area, where many methodological qu tions rem n open. Besides that, interestingly, it seems to be one of the most con­ uenti ones: a project like the di talization of the Archivo Gener de Indi will continue to in uence the conditions of historical work for d es in the next century. There e not only many open questions, it is worthwhile and necce ry to di uss them.
While everybody seems to have encountered im e processing in one form or the other alre y, preci knowledge about it seems to be relatively scarce. The volume st ts, therefore, with a general introduction into the eld by· J. v.d. Berg, H. Brandhorst and P. v. Hui tede. While most of the following contributions have been written to be self supporting possible, this introduction attempts to give l readers, particularly those

3
with only a vague notion of the techniques coucerned, a common ground upon which the more specialized discu ions may build.
The contributions that follow have been written to introduce speci c are , where handling of images is useful and can be integrated into a !arger context. All authors h e been ked in this part to clearly state their own opinion, to produce clearcut statements about their methodological position in the di ussions described above. Originally, four contributions were planned: the rst one, di u ing whether the more advanced techniqu of image processing can change the way in which images are analysed and handled by historians, could unfortunately not be included in this volume due to printing deadlin : we hope to present it part of follow up volumes or in one of the next i ues of History and Computing.
The paper of M. Thaller argues that anning and presenting corpora of manu ripts
on a work station can (a) save the origin s, (b) iutroduce new methods for palaeographic
training into university teaching, (c) provide tools for re ing damaged manu ripts, the
comparison of band writing and gener palaeographic studies. He further prop s to
build upon that a new understanding of editori work. A fairly long t hnical discussion
of the mechanisms needed to link images and transcriptions of manu ripts in a wider
context follows.
·
F. Colson and W. H l discuss the role of images in te hing systems in university education. They do so by a detailed description of the mechanism by which imag are integrated into Microcosm I HiDES te hing packages. Their considerations include the treatment of moving images; furthermore tbey enquire about relationships between image and text in typic stages in the di ogue between a teaching package and a user.
W. H l d F. Colson argue in the nal contribution to thi part the gener c of open systems, exemplifying their argument with a di u ion of the various degr which control about the choices a user h is certained in the ways in which navigation supported in a hyper-text oriented system containing images. In a outshell the di erence between „open“ and closed systems can be understood as the following: in an „open system“ the user can dynamic ly develop further the behaviour of an image-b ed or image-related system. On the contrary in static „editions“ the editor h ab lute control, the user none.
Following these general description of approaches, in the third part, several interna­ tion projects are presented, which describe in detail the decisions taken in implementing „re “ image proce ing b ed applications, some of them of almost frigthening magnitude. The contributors of this part were ked to provide a di erent kind of introduction to the subject than those to the previous two: all of them should discuss a relatively small topic, which, however, should be discus d with much greater detail than the relatively broad overviews of the rst two parts.
All the contributions growing out of the workshop came from projects, which had among their aims the immediate applicability of the tools developed within the next 12- 24 months. As a result they are focusing on corpora not much beyond 20.000 (color) and 100.000 (blw) images, which are supposed to be stored in resolutions manageable within 5MB I image (color) and 0.5 MB I image (blw). The participants of the workshop feit strongly, that this view should be augmented by a description of the rationale behind

4

the creation of a !arge scale projt t for the systematic conversion of a complete archive. The resulting paper, by P. Gonza!ez, describes the considerations which Iead to the design of the \rchivo General de Indi projt t and the experiences gained during the completed stages. That description is enhanced by a discussion of the stratrgies selected to make the raw bitmaps acce ible via suitable descriptions I transcriptions I keywords. A critical appraisal, which decisions would be made di erently after the developments in hardware tecbnology in recent years, augments the value of the de ription.
The participants of the workshop feit furthermore strongly, that their view described above sbould be augmented by a description of the techniques used for the handling of images in extremely high resolution. A. Hamber’s contribution, dealing with the V ari project, gives a very thorough introduction into the technical problems rncountered in handling images of extremely high quality and also explains the economic rationale behind an approach to start on purpose with the highest qu ity of images available today on prototypical hardware.
As these huge projects both were related to iustitutions which traditionally collect source material for historical studies, it seemed wise to include also a view on the roJe images would play in the data archives which traditionally have been of much importance in the considerations of the AHC. E.S. Ore discusses what implications this type of machine readable material should bave for tbe infrastructure of institutions speci c ly dedicated to Humanities computing.
Image systems which deal with the archiving of pictorial materi and manuscript systems have so far generally f rly „shallow“ descriptions. At le t in history, moreover, the rely quite frequently on pre-de nt’d terminologies. G. Jaritz and 8. Schuh describe how f and wby historical research needs a di erent approacb to grasp as much of the intemal structure and the content of image possible.
L t not le t R. Rowland, who acted host of the workshop at Firenze, describes tbe considerations which currently prepare the creation of another largesc e archival datab e, to cont n !arge ounts of material from the archives of the inquisition in Portugal. His contribution tries to explore the way in which the more recent developments of image processing can be embedded in the general services required for an archival system.
This series of workshop reports shall attempt to providr a broader b is for thor­ ough di ussions of current methodological questions. ‚fheir main virtue sh l be, that it is produced su ciently quick to become available, before developments in this eld of extremely quick development make them ob lete. We hope we have reached that goal: the editor has to apologize, however, that due to the necessity to bring this volume out in time, proofreading h by neccessity be not intensive it should have been. To which �nother shortcoming is ded: none of the persons engaged in the final production of this volume is a native speaker of English; so while we hope to have kept to the standards of what might be described „International“ or „Conti mtal“ English, the native speakers among the readers can only be ked for their tol rance.
Göttingrn, August 1992