Jump to ContentJump to Main Navigation
Images and Artefacts of the Ancient World$

Alan K. Bowman and Michael Brady

Print publication date: 2005

Print ISBN-13: 9780197262962

Published to British Academy Scholarship Online: January 2012

DOI: 10.5871/bacad/9780197262962.001.0001

Show Summary Details

Wooden Stilus Tablets from Roman Britain

Wooden Stilus Tablets from Roman Britain

(p.6) (p.7) Chapter 1 Wooden Stilus Tablets from Roman Britain
Images and Artefacts of the Ancient World

Alan K. Bowman

Roger S. O. Tomlin

British Academy

Abstract and Keywords

The imaging of ancient document papers presents several challenges, the nature of which is determined by the character of the text, the material on which it is written and the state of preservation. This chapter talks about the struggle to read and interpret Latin manuscripts from Roman Britain. These manuscripts come mainly in three forms: texts written in ink on thin wooden leaves, texts inscribed with metal stylus on wax-coated wooden stilus tablets, and texts incised on sheets of lead. This chapter focuses on the problems of imaging and signalling process of the texts found on the Vindolanda stilus tablets. These problems in interpreting ancient texts arise from the two identifiable sources of difficulty. The first one is the problem of seeing and identifying, in abraded and damaged documents what is aimed to be read. The second is the problem arising from the character of the text itself which determines the ability of the reader to decipher and interpret it.

Keywords:   ancient document papers, Latin manuscripts, Roman Britain, wooden stilus tablets, imaging, signalling process, Vindolanda stilus tablets, ancient texts, seeing, character of the text

I. The Nature of the Problem

The imaging of ancient documents presents a variety of challenges whose nature is determined principally by the character of the text, the material on which it is written, and the state of preservation.2 The authors of this paper, and Professor David Thomas (of the University of Durham), have been struggling to read and to interpret difficult Latin manuscripts from Roman Britain for the past quarter-century or more. These come mainly in three forms: texts written in ink on thin wooden leaves, texts incised with a metal stilus on wax-coated wooden stilus tablets, and texts (usually ‘curses’) inscribed on sheets of lead.3 It is the second category that concerns us in this paper, presenting, as it does, extremely challenging problems of imaging and signal processing.

There are at least two separately identifiable but linked sources of difficulty. The first and primary source is the problem of seeing and identifying, in abraded and damaged documents, what it is that we are trying to read, and it is this problem which primarily attracts the attention of the signal-processing expert. The other category of problems arises from what may be called the character of the text itself—those features which determine the ability of the reader to decipher and interpret it, assuming that we can actually see it as well as we need to. Although we can in principle separate these two aspects so that we would, in an ideal world, first achieve a perfect visual image and then read it correctly, in practice our ability to read even the ‘perfect visual image’ is less than perfect; it goes without saying that the likelihood of our achieving a correct reading diminishes the poorer the visual image is.

The difficulties can be well illustrated by the editorial history of a fragmentary text written in ink and discovered in 1974 among the writing tablets from Vindolanda.4 The text was originally read from an excellent infra-red photograph, made in 1975 by Alison Rutherford, in which the script and the individual letters in the top line were relatively clear (i.e. in it self it presented what seemed to be quite clear visual signals).5 Despite this, acomparison of the 1983 and the1994 transcripts shows that the best reading that we could achieve the first time round was woefully inadequate (Figure 1.1):

1983 transcript:6 c…io inmatura ad metalla

Wooden Stilus Tablets from Roman Britain

Figure 1.1 Tab. Vindol. I 25 = II 247.

Here, each letter as we transcribed it was a defensible reading and it was possible to reconstruct some known Latin words from it, if not any syntax or sense: inmatura would be a form of the adjective immaturus and the last two words could be understand as a possible reference to the known practice of condemning criminals to hard labour in the mines (metalla).

1994 transcript:7 Lepidinam tuam a me saluta

Eleven years later we were confident that we had the correct reading: ‘Greet your wife Lepid in a from me’. What made the difference here was an accumulation of contextual knowledge from other tablets, discovered and deciphered between 1985 and 1994, which gave us plenty of evidence for the existence of the woman called Lepidina, the wife of the commanding officer of the unit stationed at Vindolanda between c. AD 97 and 104, and of the formulae commonly used at the end of personal letters of this sort. The contextual information makes a crucial difference because in Latin cursive handwriting of this general type (known as Old Roman Cursive, and in vogue until the third century AD) the letter forms themselves are, in almost all cases, far from being completely unambiguous: for example, a and r are frequently made (p.8) similarly; c, p, t, and i are of ten difficult to tell apart; and these are by no means the only potential sources of confusion.8

Such confusions are only too easy to illustrate. One lead tablet was first transcribed a century ago, when it was interpreted as a letter written by one Christian to another in fourth-century Britain. To the first author of the present paper, as one generally familiar with the range of documentation from the Roman empire, the text as originally read looked very suspect but he was notable to explain it, except by vaguely wondering whether it might be a modern forgery. The second author achieved a better reading and interpretation by rotating the tablet through 180°: its Edwardian editor had read it upside down, and the new reading is self-evidently correct, relocating it to what is now a well-known genre of Latin texts.9 This reading experience has recently occurred in the Vindolanda tablets in a still more complex and enigmatic form: one of the ink tablets discovered in the 1990s contains a text which has two lines inserted upside down in relation to the rest of the writing on the leaf.10

A major part of the problem, particularly in the early stages of our work on this material, was simply its novelty. It was not completely unparalleled in texts from other parts of the Roman empire (particularly Egypt), but we lacked a large existing corpus of similar texts which would give us a broad enough basis of knowledge from which we could deduce or predict with a high level of confidence what we were likely to have in any new example. Papyrologists in general proceed on the principle that what they have in a new text should ideally be able to be paralleled in an existing text, and only when they have searched exhaustively and failed to find parallels do they feel comfortable about proposing a reading or interpretation that is completely novel. As the corpus of existing material accumulates, the possible parallels increase, and the decipherment and interpretation of new texts improve accordingly. The number of surviving Latin texts from Roman Britain (let alone the empire as a whole) as a proportion of the total ever written (or even of those produced in a few decades) must be minute.11

Even with the advantage of a steadily expanding knowledge base, collaboration on the Vindolanda tablets makes it painfully obvious that in many cases two supposed experts working independently can come up with completely different transcriptions of the same visual signals (for which there has to be an objectively correct interpretation, i.e. what was originally written), and that they must resolve these differences by negotiating their way towards an agreed version which approximates as closely as possible to the objectively correct transcription and interpretation.

In considering how computer-based image enhancement can help the expert reader, it seems obvious that there can in principle be no simple automated process which will solve problems that arise from a combination of damaged or abraded and inevitably imperfect human cognitive and knowledge-based skills. Our curiosity about developing an effective process that can help might profitably have been directed to the fairly substantial body of ink texts like the one described above. There are, however, already reasonably effective imaging techniques in existence for digitising and enhancing ink-written texts,12 and in fact the wooden stilus tablets—of which more than two hundred have been discovered at Vindolanda in excavations conducted since 1974, and many more lie unread in museums and older collections—present far more difficult and challenging problems, not least because they present them in three rather than two dimensions.

The cognitive and knowledge-based problems of the sort described above are in general common to ink texts and inscribed texts. Understanding them requires some introspection and analysis of the process of reading damaged, abraded, or degraded texts, which we take to be different from reading printed text on a page. First we identify the shapes of letter forms, which fall into a range of types and different hands; then we read individual letters and combinations of letters to the point where we can construct words, phrases, sentences, and finally whole texts.13 But of course we do not normally transcribe letter by letter in a completely neutral and automaton-like fashion, and then realise that we have transcribed a word or sentence; there is a point, very quickly or perhaps immediately reached, at which we bring into play our corpus of acquired linguistic, palaeographical, and historical information which in effect predisposes or even forces us to predict how we will identify, restore, or articulate letters or groups of letters. And this is a recursive process. We do the easy bits, then make hypotheses about the problematic bits and test them in the context which we think we have established by what we can read. The process of reading papyri was well described, even if somewhat intuitively and impressionistically, by Herbert You tie, probably the greatest papyrologist of the twentieth century, a generation or so ago and well before the existence of computers and imaging techniques:14

In the face of such discouragements [sc. the poor physical state of the papyrus], in whatever combination they may occur, the transcriber repeatedly finds that his most strenuous efforts to obtain a reading are frustrated. His only hope lies in supplementing his knowledge of handwriting with as full an understanding as he can get of the scribe’s purpose in writing the text. He tries (p.9) to take account of the text as a communication, as a message, as a linguistic pattern of meaning. He forms a concept of the writer’s intention and uses this to aid him in transcription. As his decipherment progresses, the amount of text that he has available for judging the writer’s intention increases, and as this increases he may be forced to revise his idea of the meaning or direction of the entire text, and as the meaning changes for him, he may revise his reading of portions of the text which he had previously thought to be well read. And so he constantly oscillates between the written text and his mental picture of its meaning, altering his view of one or both as his expanding knowledge of them seems to make necessary. Only when they at last cover each other is he able to feel that he has solved his problem. The tension between the script and its content is then relaxed: the two have become one.

To bring these skills into the purview of 21st-century technology, which we should surely try to do, we need to begin by analysing the process again and identifying the capabilities and the limitations of this technology.

Self-evidently, all the processes of cognition involved in reading a text cannot be replicated by a machine, but some of the processes can be made easier or brought into relationship with one another more effectively. Making alphabets for individual texts, which is a standard procedure followed by many palaeographers, is an act partly of vision, partly of cognition. In Latin cursive texts, a computer-based technique of pattern recognition for letters might tell us that a certain combination of strokes is most likely to belong to t, p, or c. (a similar group) or a and r, but it takes a combination of palaeographical and linguistic knowledge and expertise to perform the act of judgement which privileges one possibility over the others, when letter forms vary even in individual texts, as they tend to do frequently enough to cause confusion. Faced with a group of letters which could equally well be read as par or tra, the decision as to which is more likely cannot be made purely on visual criteria but requires the deployment of other skills and know-ledge. And it might be argued that a palaeographer could deploy them just as well without a computer, especially in a subject where the acquired knowledge is relatively limited. We believe, however, that this is too optimistic, and that much is to be gained by the systematic classification and presentation of comparative data, both visual and knowledge-based.15

The problems of visibility, however, are a different matter and here we think that signal processing really can help: once again, not in producing any kind of a fully automated ‘system’, but in producing techniques which enable us to see better the material text on which we are trying to perform the cognitive act of ‘reading’, bearing in mind the ways in which we can our selves deliberately manipulate the physical act of vision by varying angle and focal length. The stilus tablets offer an ideal corpus of material. Some few can be read wholly or partly without any artificial aid, save simple lighting; many others really do seem impossibly hard; others lie between these extremes. If we can read some of them, or only some bits of some of them, we ought to be able to use computer techniques to push forward the boundaries of legibility.

The stilus tablet is a thin, rectangular slab of wood, about the size of a postcard, with a hollowed-out centre which is filled with wax. The writing is incised with the sharp point of a metal stilus on the wax surface. The wax can be re-smoothed with the spatulate end of the stilus, obliterating the old text and allowing a new one to be written, or a new wax surface can be put on.16 In the great majority of cases the archaeological contexts in which these objects are found has caused the wax to perish. We are thus left only with the scratches made by the stilus where it penetrated the wax to the wood beneath.

Our starting point was the notion that we could describe and analyse the visual process and attempt to replicate it more efficiently by using a computer-based signal processing technique. The attempt to read these things in the old-fashioned, manual way can be based on photographs, which offer a static image, or on the originals. Achieving optimal visibility involves adjusting the angle of lighting (which would obviously require more than one photograph if one is not working from the original), varying the angle of vision so that different parts of the incision can be seen, and using other indicators such as alignment, depth of incision, the crossing of superimposed strokes. In fact, we observed that when we do try to read the originals we tend to use lighting from different angles and to tip and tilt the surface in relation to the eye, so that the text more or less ‘moves’ for us. This intuitive behaviour, when he ‘spotted’ it, prompted the way in which Brady came to our aid.17 What we would achieve, if only the human eye and brain were perfect, would be a series of optimal images of different bits of the text or, in actual fact, of different bits of the same strokes as they emerge more clearly under different lighting conditions and different angles of vision. But one of the respects in which the eye and brain are not perfect, as we conceive it, is that even where achieving such a series of optimal images is possible, it is impossible for the human eye to achieve a complete synthesis of them. This is where it seems that computer-based signal processing ought to be able to help. The examples which follow illustrate a variety of problems with different degrees of difficulty, the methods used, and the results we have obtained so far.

II. Two Sample Texts from Vindolanda

Tablet 797 (Figure 1.2)

This tablet is the most straightforward of the three presented here. It contains a single text written in two (p.10) columns on either side of a central recessed strip.18 We have succeeded in reading the last two(?) lines of column 1 on the left, and the five lines of column 2 on the right. We were able to do this on the basis of a small number of unprocessed scans, made with lighting from different angles. The tablet may therefore be placed at the relatively unproblematic end of the spectrum. The text itself is a letter, or rather the end of a letter, in which the writer offers a guarded apology to the recipient:

‘…angrywithme. […]some times I am surprised at why you are now angry with me. Farewell.’

Wooden Stilus Tablets from Roman Britain

Figure 1.2 Vindolanda stilus tablet 797.




nunc quid


mi irasce


ris aliqua

me irasce

ndo mir


[o]r vale

Tablet 836

This example gives a unique insight into the correspondence between the original state of the stilus tablet with its wax coating, and the more usual state in which tablets are found, without the wax, because this was the only case in which the wax coating survived the archaeological conditions (although it dissolved during the conservation process). It therefore offers a unique opportunity for illustrating and testing the imaging techniques which are being developed with the use of variable directional lighting and phase congruency.

Fortunately the tablet was photographed with the wax writing surface more or less intact, at two different scales, and we thus have two images of a single incised text on a relatively ‘noise’-free surface (Figure 1.3). When the wax was accidentally dissolved, it left the bare wooden surface with the incisions made by the stilus where it penetrated the wax. It can be seen that, as in so many tablets, there are incisions not just from the text which we could see on the surviving wax, but from at least one other text as well. We have photographs and digital scans of the tablet in this state, taken with directional illumination.

By using the images of the tablet with the wax coating we have been able to produce coherent readings of parts of the latest text written on this tablet. It should be emphasised that this text, the one on the wax surface, is relatively visible, but even so there are points where the readings of the two authors of this paper differ, and there is one area which offers particular difficulties and ambiguities. In the text presented below, letters printed in boldface are those which can be read with confidence; letters printed in ordinary type are read with some measure of conjecture; underlinings indicate traces of letters which cannot be identified with confidence:

  • binus bello suo salutem
  • [traces only]
  • ]acc__erunt in uecturas
  • de_arios octo reliquos solues
  • rios nouem qua__r_r___
  • ]sam dari debeb__
  • [interlinear addition?]
  • ]em libris
  • ]dus uale

Wooden Stilus Tablets from Roman Britain

Figure 1.3 (a) Vindolanda stilus tablet 836, before the wax dissolved. (b) Tablet 836 after the wax dissolved.

‘Albinus to his Bellus greetings…they have received for transport costs 8 denarii. You will pay the remaining 9 denarii…ought to be given(?)…nine pounds (?)…Farewell.’

(p.11) Notes:

  1. 1 There is a trace between the first and second l in bello which might or might not be a letter. The scratches on the wood show that this overlies an earlier text.

  2. 3 The correct reading is almost certainly acceperunt.

  3. 4 The word at the end of the line presents particular difficulty. Of the first three letters of solues only the o is certain. There is a clear high horizontal which has to be ignored if the first letter is read as s. The third letter might be p, and there is another apparent high horizontal which is discounted. The attraction of reading the word solues (from the verb soluere ‘to pay’) is obvious if the word ‘denarios’ occurs twice in lines 4–5.

By using the images shown in Figure 1.3, we can see which of the scratches belonging to this text actually penetrated the wax to the wood so as to be visible under the different angles of lighting. We can also attempt to deduce which scratches did not penetrate or are not visible, and which are the scratches that belong not to this text but to an earlier text. The transcription below prints those letters and words which we think we might have be enable to read without knowledge of the version on wax, but this is subject to some uncertainty in all lines except line 1 (where some letters appear more clearly on the wood than on the wax, no doubt because of the different quality of the photographs). The scratches which we have not been able to transcribe are confused and confusing; they probably belong to more than one text, and show significant differences when the tablet is illuminated from different angles:

  • albinus bello suo salutem
  • [traces only]
  • erunt in uecturas
  • arios octo
  • arios nouem qua
  • am debeb

III. Word by Word Decipherment

Tablet 974

Our third example illustrates the difficulty of working from conventional photographs or unprocessed images of stilus tablets. We have a surface on which the inscribed text is confused by discoloration, wood-grain, and traces of at least one other text, which may or may not have been aligned with it (Figure 1.4). Other figures, reproduced in Chapter 2 (Figure 2.1, 2.2, 2.5, 2.6, 2.13, and 2.14), illustrate various aspects of the process: image capture of the surface of the tablet at 0° and 180°, with many other angles being possible in principle (Figure 2.13(a)); removal of the wood-grain by a simple process of Fourier transform (Figure 2.5); then the composite image, in which 0° and 180° are combined and the wood-grain is removed (Figure 2.5); next the highlighted section with stroke-detection using shadows and highlights (Figure 2.13(b)); finally the application of phase congruency, which yields a variety of images (Figure 2.14). We must emphasise that they show different features of the text, and that in attempting to see and read the text, we actually negotiate our way round the different images and their constituent parts.

Wooden Stilus Tablets from Roman Britain

Figure 1.4 Vindolanda stilus tablet 974: (a) illumination from left; (b) illumination from right.

In thus negotiating our way, we feel somewhat like cells of the Earth, the computer in the Hitch Hiker’ Guide to the Galaxy which was so complex that it required an organic matrix. This is provided by the two of us, and in what follows we will reconstruct the way in which we progressively read and understand an incised text like Vindolanda tablet 974. It has been drawn here (Figure 1.5) for convenience like the page of a modern book, black letters on white, in a reversal of the original, which was white on black, with the letters scratched into white wood through black wax. This wax has since disappeared, and we are left with the scratches, euphemistically called ‘incised text’ How then do we ‘read’ them?

The great papyrologist Herbert You tie read papyri scientifically; the adverb is deliberately chosen, since he formed hypotheses and tested them.19 There is a story20 about another master of palaeography, M. R. James, who catalogued all the western manuscripts in Cambridge, but is now better known for his st Stories of an Antiquary. His own handwriting, appropriately enough, was illegible. His colleagues once received an (p.12) invitation to dinner: ‘we guessed that the time was 8 and not 3, as it appeared to be, but all we could tell about the day was that it was not Wednesday’.

Wooden Stilus Tablets from Roman Britain

Figure 1.5 Tablet 974, drawn by R. S. O. Tomlin.

As it happens, this was good palaeographical method. Not a letter-by-letter transcription like sleepwalkers, but a wakeful testing of possibilities in the light of other knowledge. If you know the letter is an invitation to dinner—something given—you can exclude 3 o’clock thanks to your knowledge of the English dinner-time, and since you know there are only seven days in a week, the absence of the distinctive letter ‘W’ is a compelling argument from silence, like the letter ‘S’ is in our own stilus tablets, for it is a tall, powerful, sinuous letter which extends over two lines. Its corresponding drawback is that it often persists from an earlier text, and is thus unrelated to what we are trying to read.

Figure 1.5, the drawing of Vindolanda tablet 974, is incomplete since it represents work in progress. We have been forging a chain of interlocking hypotheses, but the last link has not been closed. Our starting point was the photograph21 published by the excavators in their report, which gave us hope that the text could be read one day. In one of our copies there is evidence of this hope, the pencilled transcript of the last two lines. We might say, at this point, that one often starts from the end: the end of the whole text, the end of individual words. To illustrate this process, the lines in Figure 1.5 have been numbered from the bottom of the ‘page’. Later, when we were given the original of the published photograph, we were able to read lines 1–3, the last three lines. Circumstances favoured us: these lines were not ‘contaminated’ by traces of an earlier text. This gave us most of the alphabet (Figure 1.6). In theory we might have deduced it from the text alone, like Sherlock Holmes reading The Dancing Men cipher first time off, but fortunately for us we had already seen other examples of Roman handwriting, and indeed other stilus tablets of the same date with in the half-century AD 75–125.

Wooden Stilus Tablets from Roman Britain

Figure 1.6 Tablet 974: table of letter forms, drawn by R. S. O. Tomlin.

As we have said, letter forms can be ambiguous at first sight.22 In stilus-tablet texts ‘e’ and ‘u’, for example, can look the same since they are distinguished by the right-hand curve of the first down-stroke of ‘u’, which often disappears into the horizontal wood-grain. And although it was easy to read nutriui inline 2, there is a double down-stroke for ‘t’ which has not been drawn, but is visible in the photograph. It may be the trace of a previous text, or the writer may have written‘t’, then erased it, and written it again; whatever the case, the clear reading of the rest of the letters in this word leaves little doubt that nutriui is indeed the correct inter-pretation of these signals. In these three lines we could already see that the spelling of two words had been influenced by the spoken language: the writer wrote dece for decem (‘ten’), because the final ‘m’ was hardly sounded in speech, and he wrote serum for seruum (‘slave’) because ‘u’ and ‘w’ (literally ‘double u’) were indistinguishable in sound. Both these peculiarities armed us for the more difficult bits which lay ahead as we worked up the text, where we were aided by the improved images produced by our scientific colleagues. They reduced background ‘noise’, just as an observant drawing of an archaeological object is clearer than a photograph, and they distinguished deliberate stilus cuts from discoloration and casual damage.

At this point we could read the words ‘…thirty-five; and I have kept the slave for fifteen years’. We now had two hypotheses to test, as to what type of text it was: not a dinner invitation, clearly, but perhaps a deed of sale (vouching for the goods, one slave little used), or a deed of manumission (the freeing of the said slave). Working from the improved images, we could next read missione, inline 4. Naturally we then looked for manu in the line above, for ‘manumission’; but there was too little space after the word dedi (‘I have given’), and there was the recognisable descender of r, which is another helpful letter like ‘s’, and also ‘b’ and ‘q’, because it occupies so much vertical space. So we could now read et dedi permissione (m), ‘and I have given permission’, but then we were stuck. Moving forward in line 4, we could see that one text was written locally over another, indicating not a distinct text, but an erasure in the original, followed by a correction. The same letters were repeated, but they were ‘out of sync’ with each other. After quite a time, we decided the word must be uecturas, a word also found (p.13) in Vindolanda tablet 836 (above), where it apparently means ‘transport’ or perhaps ‘transportation charges’.

These references to ‘permission’ and to ‘transport’ made our text about a slave sound less of a legal document and more of a business letter. Perhaps the writer was sending a trusted slave on a journey. But what about ‘thirty-five’, and what has this got to do with ‘transport’? We cannot find ‘years’ (annorum) at the end of line 4—which might be the slave’s age—nor can we find ‘days’ (diebus)—which might be the length of the journey, there and back. Unfortunately there are two texts here, one on top of the other, and they are both fragmentary. So we are still baffled by them.

But working back from et dedi per at the end of line 5, we could now recognise our good friend serum again in line 6, and still more helpfully, the first ‘formula’ we had met, the word nomine, ‘by name’. This is like finding the letter ‘q’, which is bound to be followed by ‘u’. After nomine, you can expect a personal name. Here again, the descender of r was helpful; there was c rising above the line, and the name began with either e or u, followed again by either e or u. The name Verecundus was the obvious candidate. Although it is a Roman name, it is what German classical scholars call a ‘Deckname’, a cover-name, since it conceals a Celtic name-element, and was therefore popular in Brita in and Gaul. Two mistakes prevented us from recognising it more quickly than we did: at the beginning of line 5, we had read the preposition de (not -du), and after it uico, another formula: ‘from the village of…’. But we now realised that ‘de’ must be -du, and that the writer, once again, had omitted the final ‘-m’ of an accusative case-ending.

We now had a slave called Verecundus, who was apparently ciui, a ‘citizen’ of…somewhere. (We were assuming that ciui was an error for ciue, the accusative cive(m).) Thanks to the enhanced image, we could see that the place name began with a combination of a and m (m looks much the same as aa in this script), and that it ended with -nis. Mid-way, there was a b or d (another pair of letters difficult to distinguish when they are incomplete): ambianis, modern Amiens, satisfied us both, but we were uneasy that a slave should be ‘citizen’ of a town or tribe. We resigned ourselves to a search for parallels, still unfulfilled, and even wondered whether to read cum (‘with the Ambiani’) instead. But no; this seemed to be precluded by the traces visible.

At the beginning of line 6 was apparently the formulaic uico (‘village’), at least if we assumed that one stroke was surplus to requirement, that it was a descender from line7. Otherwise the reading would be uaco, which is not a Latin word at all. But fortunately it occurred to one of us that the last word inline 7 began with be-, and that the space and surviving traces (which have been over-restored in Figure 1.5 toillustrate what we thought) suited the reading bello. Before it there were two short words, fortunately short, since only the middle band of their incised letters survived: ciue meo, ‘my fellow-citizen’, was quite an easy reading, and this fitted bellouaco, which is the name of a Gallic tribe. Here the chain of hypotheses, as it were, closed for once. Neither of us knew this at the time, but when we resorted to works of reference that evening, we found that the Bellovaci and the Ambiani were neighbouring north-Gallic tribes, both in the region of modern Amiens.

This was cheering: it was a real coincidence, not one which we had made for ourselves. That said, we must now admit that the previous word, the first in line 7, batauorum, is read by only one of us. The other does not accept it. Is collateral evidence any help here? The Ninth Cohort of Batavians was the battalion which garrisoned Vindolanda, and batauorum occurs of ten in the Vindolanda tablets. But neither of us, the optimist nor the pessimist, can read cohortis (‘cohort’), which should accompany it, in line 8. Its absence is like that missing ‘W’ in the dinner invitation: Wednesday is out, and so is the battalion. Its absence casts doubt on batauorum as well.

This is almost as far as we have got, and it may be the end of the road. The further up the ‘page’, the worse it becomes. There are now two texts, if not three, and the band of lettering which survives has become narrower. It is also difficult to separate the lines of the different texts, whether by horizontal slope or vertical alignment, but these may be the only criteria. At the top of the ‘page’, optimistically emphasised in Figure 1.4, we think we can read uicesima, ‘one-twentieth’ or ‘5 per cent’. This might be another indication of the type of document, since one of its meanings is the tax levied on the manumission of slaves. Perhaps we will have to return to that hypothesis of manumission, after all.

We conclude with our working transcript and translation:

  • traces of 4–5 lines
  • Batauorum ciue meo Bello-
  • uaco ser(u)um nomine Verecun-
  • du(m) ciu(e) Ambianis et dedi per-
  • missione(m) et uecturas (over uecturas)…
  • triginta quinque et eum
  • ser(u)um nutriui annos
  • dece(m) quinque

…of the Batavians(?)…my fellow-citizen of the Bello-vaci [name and verb lost] a slave called Verecundus, citizen(?) at Amiens. And I have given permission and travel-expenses(?)…thirty-five; and I have kept that slave fifteen years.’

It is incomplete, and only ‘work in progress’. The emphasis should be on ‘progress’, since image enhancement has taken us further along an interesting road, and given us hope for the future. At least we are not alone. That story of the dinner invitation is capped by another about the handwriting of a legendary hostess, Lady Colefax: ‘the only hope of deciphering her invitations, someone said, was top in them on the wall and run past them’.23

(p.14) This too is good scientific method, a primitive way of combining subtly different images seen in quick succession from different angles. But since our own mathematics go no further than O levels passed during the Mesolithic, we must leave the explanations to Brady and his colleagues (below). The story is only worth quoting because it catches something of the excitement and the mystery of phase congruency, at least for us, when the text ‘moves’ and letters jump out of the shadows.


Bibliography references:

Bagnall, R. S. (1997), ‘Imaging of papyri: a strategic view’, in A. K. Bowman and M. Deegan (eds.), Imaging Documents, Literary and Linguistic Computing 12/3: 153–7.

Birley, A. R. and Birley, R. E. (1994), ‘Four new writing-tablets from Vindolanda’, Zpe 100: 431–46.

Birley, E., Birley, R. and Birley, A. (1993), The Early Wooden Forts: Reports on the Auxiliaries, the Writing Tablets, Inscriptions, Brands and Graffiti, Vindolanda Research Reports 2 (Bardon Mill).

Bowman, A. K. and Deegan, M., eds. (1997), Imaging Documents, Literary and Linguistic Computing 12/3.

Bowman, A. K. and Thomas, J. D. (1983), Vin-dolanda: the Latin Writing-Tablets, Britannia Monograph 4 (London).

Bowman, A. K. and Thomas, J. D. (1994), The Vindolanda Writing-Tablets (Tabulae Vindolandenses Ii) (London).

Bowman, A. K. and Thomas, J. D. (2003), The Vindolanda Writing-Tablets (Tabulae Vindolandenses Iii) (London).

Bowman, A. K., Brady, J. M. and Tomlin, R. S. O. (1997), ‘Imaging incised documents’, in A. K. Bowman and M. Deegan (eds.), Imaging Documents, Literary and Linguistic Computing 12/3: 169–76.

Crowther, C. W. (1997), ‘Imaging inscriptions’, in A. K. Bowman and M. Deegan(eds.), Imaging Documents, Literary and Linguistic Computing 12/3: 163–7.

Cunliffe, B. (1988), The Temple of Sulis Minerva at Bath, vol. 2, The Finds from the Sacred Spring, Oxford University Committee for Archaeology, Monograph 16 (Oxford).

Fink, R. O. (1971), Roman Military Records on Papyrus (Cleveland).

Hart-Davis, R., ed. (1978), The Lyttelton Hart-Davis Letters, vol. 1 (London).

Lalou, E., ed. (1992), Les tablettes à écrire de l’antiquité à l’époque moderne, Bibliologia 12 (Turnhout).

Marichal, R. (1992), ‘Les tablettes à écrire dans le monde romain’, in E. Lalou (ed.), Les tablettes à écrire de l’antiquité à l’époque moderne, Bibliologia 12 (Turnhout): 165–85.

Obbink, D. (1997), ‘Imagingthecarbonizedpapyrifrom Herculaneum’, in A. K. Bowman and M. Deegan (eds.), Imaging Documents, Literary and Linguistic Computing 12/3: 159–61.

Speidel, M. A. (1996), Die römische Schreibtafeln von Vindonissa, Veröffentlichungen der Gesellschaft pro Vindonissa Xxii (Brugg).

Terras, M. (2000), ‘Towards a reading of the Vindolanda stylus tablets: engineering science and the papyrologist’, Human It 2/3: 255–71.

Tomlin, R. S. O. (1988), ‘Thecursetablets’, in B. Cunliffe (ed.), The Temple of Sulis Minerva at Bath, vol. 2, The Finds from the Sacred Spring (Oxford): 59–277.

Tomlin, R. S. O. (1994), ‘Vinisius to Nigra: evidence from Oxford of Christianity in Roman Britain’, Zpe 100: 93–108.

Tomlin, R. S. O. (1998), ‘Romanmanuscripts from Carlisle: the ink-written tablets’, Britannia 29: 31–84.

Youtie, H. C. (1973), Scriptiunculae, vol. 1 (Amsterdam).


(2) For surveys of different types of material see Bowman and Deegan (1997), in particular the articles by Bagnall, Gagos, Crowther, and Bowman, Brady, and Tomlin. A range of material is available at http://www.csad.ox.ac.uk.

(3) See Bowman and Thomas (1983, 1994), and Tomlin (1988, 1998). These are far from being the only categories of written material from Roman Britain. This particular aspect of our imaging work does not address the problems presented by monument al inscriptions on stone, or by the graffiti and inscribed personal belongings classified as instrumentum domesticum.

(4) Tab. Vindol. I 25 = II 247. Readings by A. K. Bowman and J. D. Thomas.

(5) Figure 1. 1 is not the original photograph but a digital scan, made by Dr John Pearce in 2000, which offers a marginally better image.

(6) Bowman and Thomas (1983), no. 25.

(7) Bowman and Thomas (1994), no. 247.

(8) Bowman and Thomas (1983), 51–71; (1994), 47–61

(9) Tomlin (1994).

(10) Bowman and Thomas (2003), no. 649. Not recognised by A. R. Birley and R. E. Birley (1994), 431, no. 1, although they did note that there seemed to be an insertion in the text.

(11) Fink (1971), p. 242, noted that in the first 300 years of the Roman empire at least 225 million individual pay records would have been written for legionary soldiers, but that only two examples, each relating to a handful of soldiers, actually survive. One more has come to light since 1971.

(12) For some references see Obbink (1997).

(13) For tabulation of letter forms see Bowman and Thomas (1983), 54; (1994), 53; Tomlin(1988), 91–4. It should be noted that the exemplars are based on a relatively small sample of texts.

(14) You tie (1973), 26–7.

(15) For an outline see Terras (2000).

(16) Lalou (1992).

(17) We quote from the article ‘Wax tablet reveals secrets of Roman life’ in the Oxford Blueprint for April 2001, which in effect describes how the Professor of Information Engineering went bird-watching among the ancient historians.

(18) This was intended for the wax seals of the witnesses to a legal document, but the tablet was evidently not used for this purpose, or has been reused.

(19) See above, pp. 8–9.

(20) Told by George Lyttelton in Hart-Davis (1978), 14.

(21) Birley et al. (1993), plate xxi.

(22) The readings which follow are in bold for the sake of clarity.

(23) Rupert Hart-Davis in Hart-Davis (1978), 17.