Go to:

AMPHORAS: A database on ancient wine jars

Note. This paper was given by P.M.W. Matheson and C.G. Koehler at the conference of the Association for Literary and Linguistic Computing in Toronto in June 1989. The paper documents the beginnings of our use of the computer data base for amphora studies.

Part 1: The Project (by C.G. Koehler)

              SLIDE (mosaic showing a ship with amphoras)

The containers in which the ancient Greeks shipped their wine and other produce offer unique evidence for the study of trade in antiquity. These large, undecorated jars with pointed bottom, which are today termed transport amphoras, serve to trace commercial exchange among Mediterranean nations for the three millennia stretching from their invention in Bronze Age Canaan to the last flourish of the Byzantine Empire.

          SLIDE (series of Knidian jars from the Agora at Athens)

These amphoras, for instance, represent the wine exports to Athens made by the city of Knidos on the coast of Asia Minor between the 3rd century B.C. and the 2nd century A.C. But amphoras do more than reveal ancient trading patterns. By means of the stamps impressed upon the handles of many jars, they also offer dates, often very precise ones, for archaeological contexts. Dating them involves identifying and assembling into chronological order the magistrates, or eponyms, whose names appear on the handles of some series.

                        SLIDE (Knidian stamp)

For example, in this Knidian stamp, bearing 3 lines of letters with a trident device in the bottom right, the name in the top line, mostly missing, has been restored as DIOKLHS---he is the eponym in whose term of office the jar was made. Below his name comes the name of the potter, or fabricant, CENOKLHS, whose factory produced the jar. And the last line says KNIDION: "this is a Knidian amphora." Note that the letters of each line are aligned under each other at the left-hand edge of the stamp, and that the handle of the trident extends to just under the letter delta in KNIDION. So with DIOKLHS et al. we get from stamped amphoras aids for political history, and in addition a sense that the state held an interest in export commodities and exercised a degree of control over the production of the jars in which they were shipped. Also, capacities of amphoras can indicate standard measures used in individual city-states, while their manufacture reveals something of the local exploitation of natural resources and technology. Thus amphoras and their stamps hold information of surprising scope for archaeology, economic history, prosopography, social history, and more. But the means of extracting it are complex; it has to be mined.


         SLIDE (Jars at the Athenian Agora in storage racks)

The jars of Greece in the Classical and Hellenistic period, 6th to 1st centuries B.C., are today perhaps the best documented of any in the Mediterranean, thanks to the research of Virginia R. Grace at the Agora Excavations of the American School of Classical Studies at Athens. As the marketplace of ancient Athens, the Agora has yielded over 30,000 jars and fragments, many of them stamped, and some of which you see here in the study display Dr. Grace designed. Over decades, she has amassed data on amphoras from excavations throughout the Mediterranean, on land and under water, organizing and cross-referencing some 150,000 objects on cards, with photographs, profile drawings and rubbings of stamped impressions, into a system which records stamps and jars by class---those from Knidos, which you've already met, and from Rhodes, Thasos, Chios, and many other states, each of which produced a distinctive series of amphoras.

Virginia Grace has for years constituted, in one person, a tripartite operation: she has created an archive; she has performed a service by answering queries concerning amphoras and their stamps from scholars around the world; and she has published groups of amphoras from excavations as well as studies on separate classes of jars. The challenge Philippa Matheson and I face in preparing to carry on her work is to continue all three functions in the face of a vastly increasing body of material and requests for information about it. We are, after three and a half years of our project, AMPHORAS, well launched upon the first objective, creation of an electronic form of Dr. Grace's files. Thus far we have concentrated upon the design of the database and upon data entry. We'll demonstrate how we've met the requirements of the data and what effect data entry has had upon the archive.

We have so organized ourselves that the files in Athens can remain normally active during our operations, so that we can all continue to provide information to scholars, as Virginia Grace has been doing for the last half century. We have contributed to the dissemination of information, our second objective, since parts of our database, into which records are being entered by site, are discrete hunks, and a few have already been shipped in hard copy, on disk, and by e-mail to excavators. Third, we are preparing for publication a manuscript of the stamps imported into ancient Corinth during the Greek period, which has served as a kind of pilot project for the design of the largest of our databases, the General Amphora File, or GAF. I'll pass over in silence (almost) the 16 fields of the GAF, since they differ little from any other archaeological ceramicist's: inventory number, class name, find spot with context date, plus descriptions, dimensions, photos, comments, and so forth. At the same time as the GAF we have created specialized study files, which contain information specific to individual classes of jars and are used for studies of, e.g., chronology, potters and eponyms. These will form the basis of core studies to be published on each class or group of amphora, making a great deal of new material available, pre-digested, which will allow archaeologists to publish their amphora stamps without such heavy dependence upon Dr. Grace or the archives.

Part 2: The Database (by Philippa M.W. Matheson)

        SLIDE (jars in place underwater in the Kyrenia wreck)

Setting out to put all this amphora information on computer could perhaps be viewed as getting it from its least accessible form (these are jars from a shipwreck on the sea-bottom off Kyrenia in Cyprus) into the most accessible form; it presents a number of challenges, of which I propose to deal with only one---how to represent the stamps on the handles of stamped amphoras. But first a brief digression on our use of the technology.

Ideally, we need a full range of optical imaging equipment, with the ability to scan, store and display for comparison, the mass of photographs and rubbings of stamps which form the basis of Dr. Grace's files. We are in fact in the process of purchasing a Sun386i which we hope eventually to use for this purpose. Add to that an integrated database system, which will allow the data records of the individual examples in the GAF, to be processed simultaneously with their images, together with the information from the Study Files which applies to each different type of stamp. While the technology for each separate stage of this ideal package already exists, we have been reluctant to commit ourselves to any cobbling together of the requisite elements, in the hopes that increasing standardization of database procedures, and better integration of visual data, will soon become available. Judging by what we have heard this week, we are not alone in this!

At the opposite end of the technological scale, we also need to make the project portable, both for data entry, which is being done on a variety of machines in 3 different countries, and in order to be able to disseminate the data to scholars in a form they can use. It came up on a recent debate on HUMANIST about how to get technologically-innocent academics hooked on computers, that word-processing, perhaps with limited use of a standard database system, is about as much as the average Classicist is prepared to tackle. While the identification of individual stamps by image comparison, and the core studies to be performed on the bulk of the amphora material, require sophisticated machine usage, the service side of the project requires the information we provide to be supplied in a standard form which a word-processor and/or dBase can use. Which means PCs, and we chose IBM over Apple (back in 1982) on the sole ground that the more open architecture made it easier to program the full epigraphical Greek character set we needed.

Coming from some experience with UNIX and the multirelational database system used by the ATHENIANS project (Prosopography of Ancient Athens) on a VAX 11-780, the descent to DOS was like a trip into the dark ages. Everything had to be programmed, beginning with the creation of a Greek character set in graphics for the screen which could be mixed with the normal character set, together with a database system that could use it, and finally a graphics font for use on several different dot matrix printers. The database program, called FERRET, was written in C---it uses only low-bit ascii characters in its database files, including the Greek, which is based on the APA/Beta format Greek transliteration scheme used by other large computer archives. We can thus transmit the database files freely by BITNET, and, as we shall see, convert them to dBase or other database systems when the need arises. FERRET data files can be edited by any word processor which reads and writes straight ascii: this includes the Greek, which appears in transliteration when used without the FERRET program's graphics.

As a humanist rather than a scientifically-minded business person, I assumed from the outset that it would be necessary to store carriage returns and control characters as data, that a limit of 1000 characters per field was none too many, and that fixed length fields were a waste of disk space we couldn't afford. FERRET has proved a very satisfactory tool, specially for data entry, but also for automatic data checking, and for basic selection and formatting procedures, such as those needed to send chunks of the database in text format to other scholars.

From time to time we have considered new database programs as they became available, including the various stages of dBase, to see if we could switch to a more standardized system with more features than the FERRET program, but so far it has always appeared more attractive to try to add to the original program, rather than sacrifice the features we already had, and could not implement in the commercial program. Using commercial software always seemed to involve drastic alterations in the data: those carefully-ambiguous phrases in the descriptions of clay or provenance would have to go, in order to accommodate fixed field lengths, and even after we had chopped and squashed and crammed the data into the smallest possible straight-jacket, the size of the data files would be up to 10 times their FERRET size. Last, but not least, there was the problem of accurate representation of the Greek stamp readings.

    KT 1130                                      Per: 5
                  'Epi~ Diokleu^v
                     trident r.
                 (S lunate)
    eth %trid        CENOKLHS fab DIOKLHS ep

Here is how the type of stamp which Carolyn Koehler showed you a moment ago appears in the KT study file of Knidian types. It is KT number 1130, one of the 2300 variations found on amphora stamps from Knidos. The period in the top right-hand corner is that assigned by Dr. Grace according to her system of amphora chronology: Period 5 falls in the second half of the 2nd c BC, a period of about 40 years following the destruction of Carthage and Corinth by the Romans in 146 BC. Any closer pin-pointing of the year in which the eponym DIOKLHS held office, based on evidence from archaeological contexts and/or name connections, will be included; further discussion of the type, such as the number of examples known etc., has been omitted from the slide, but is also part of the database structure. At the bottom, the salient parts of the stamp are given in a more computer-searchable form: the names of the fabricant and the eponym in capitals without diacriticals, the abbreviation "eth" to show the presence of the ethnic KNIDION on the stamp, and the per cent sign, indicating the presence of the trident device.

The reading represents the alignment of the lines on the original stamp at the left-hand edge, and the word "trident" begins where the handle of the trident in the actual stamp began, under the delta of KNIDION. This alignment is important, since another stamp with almost the same reading would have another KT number, and the placing of the device or the exact alignment of the lines may be an essential clue to identification, and so dating, of the stamp. Since this is the master "study" file, the reading is complete, a composite of known examples of the type. But for individual examples in the GAF, we must show exactly how much of that reading is preserved, and how much restored.

            SLIDE (photo of stamp C-1972-237 from Corinth)

This stamp, for instance, again on a Knidian amphora handle but this time found in the excavations at Corinth, was hastily set on the handle of the jar, and only some of the letters are wholly preserved. The stamp is round, and the letters run around the edge, with a device in the centre (an amphora), which also has a few letters on either side of it.

           SLIDE (composite drawing of a stamp of KT 0125)

In this drawing of stamp, which was pieced together from two other examples from the Athenian Agora, you can read the name of the fabricant, ANDROSQENHS, and the name of an eponym, APOLLWNIDAS, who held office a good deal earlier than DIOKLHS, in Period 4A. This dates him to the 20 years between 188 and 167 BC when the island of Rhodes had control over some large part of the mainland coast nearby. We know from other stamps, where APOLLWNIDAS is given his title of phrouarchos, that he was in fact a garrison commander in Knidos during this period. Note that the ethnic, KNIDION, appears here abbreviated to KNIDI, and is written backwards.

When the incompletely preserved example of this type is entered into the Corinth GAF, it looks like this:

    C-1972-237         Kor       (Co 0811)          Kn   KT 0125
                         [ 'Epi~  'Apoll]wni"da+ [ 'And]r+osqe"n+[euv]
                               [Kn] amphora i+di( (retr)
                         (S lunate, in O)
    ANDROSQENHS fab APOLLWNIDAS ep (phr) eth %amph in O
    Cxt: Temple Hill. Trench XIII. W Quarry Trench.
         NB564 p182 bsk 133
    Pht: K 1972-100-01*; K 1986-37-03*(CGK)

The "in O" at the bottom of the reading shows that it is a round stamp; the legend around the edge is shown in the top line, and the amphora device with the letters on either side of it is centred underneath. Though the letters of the ethnic are given the right way around, the "(retr)" in parentheses on that line shows that they run retrograde, or backward, on the stamp. The square brackets surround all the letters that are totally missing from this example of the stamp, and the ones that are only visible with the eye of faith are given with dots under them. And, in addition to all the other information about its context, photographic record etc, there is the all-important Knidian type number up in the top right, all-important because it is the link to the Knidian file where the dating information and official reading for this type of stamp is recorded.

Looking at all those brackets and dots, not to mention the fiddly business of aligning the lines of the stamp correctly in relation to each other, it will be clear that the data entry for stamp readings is a time-consuming process, full of potential for inaccuracies. But this is in fact exactly what computers are for: making exact copies of data and cross-referencing it from one place to another, so now the data enterers, instead of greeting each new Knidian stamp with the usual groan, simply put the KT nr in the Study field, skip the reading and the stamp description, and pass on to the remaining fields.

    KT 0125                                      Per: 4 A
                  'Epi~  'Apollwni"da  'Androsqe"neuv
                       Kn amphora idi( (retr)
                 (S lunate, in O)
    eth %amph in O   ANDROSQENHS fab APOLLWNIDAS ep

After all, a complete "canonical" copy of the reading for this type of stamp, with all the alignment carefully worked out, exists in the KT Study file.

A copy of this KT Study file exists also in dBase format, and the GAF files can be converted easily to and from dBase "delimited" text files. Thus when the data entry of 200 stamps or so has been completed, we load that new segment of the GAF file into dBase and run a quick program which finds all the records in the file where a KT number has been entered, and supplies the missing readings and stamp descriptions for them from the KT Study file by matching up the KT numbers in the two databases. Once the updated GAF files is back in FERRET format, it is a simple matter to find each reading which needs to have brackets and dots added and put them in in the appropriate places. This way all the readings of one Knidian type will have the same format, yet we have an accurate, publishable record of each individual example with the minimum amount of repetitive labour on the part of the data enterers.

Our main aim in all this was to speed up the data entry, while eliminating human error and avoiding the inconsistencies of treatment which employing 5 or 6 different people to do different parts of the data entry was bound to produce. What we had not fully anticipated was the salutory effect on the consistency of the original data which resulted from it.

In the first place it was necessary to make the KT Study file as close to accurate as our present state of knowledge allows, since any errors in the Study file would be reproduced in all the individual examples of that type. A critical examination of the readings as they were being entered showed up a number of minor inconsistencies in the primary recording of the KT types which had hitherto gone unnoticed (for instance: "why is this device called a bar in this stamp and a club in that one when they look very much the same?") and in some cases a whole type, represented by only one or two examples, simply melted away under scrutiny, when the examples were found to belong to other types, and that particular reading was left without foundation. In addition, the data enterers are astonished, and in some cases distressed, to find that changes in the Knidian type, although made with all the known examples in mind, have not always been recorded on the copy of the file cards from which they are working. We were pleased to find that 90 percent of the time, any discrepancies between the individual stamp and the official reading of the type occur only in letters which are either dotted or within square-brackets---the original observation of the stamps was close to impeccable, and subsequent improvements affect only the restoration of missing letters and/or devices. Other discrepancies are the result of simple errors in typing: the wrong KT number has been entered in the GAF database, resulting in one of those totally egregious and unmistakable errors which are so comforting about computers: when they do the wrong thing, they don't make a misleading approximation to the right one, as human beings tend to do, but come up with something gloriously inappropriate. As with many computer operations, a procedure intended only to streamline the data entry a bit, resulted also in increased accuracy in the original data and a check on the accuracy of the data entered in other fields.

Despite our focus here on the uncomputer-friendly aspects of our data, Virginia Grace's complex filing system might have been designed with computerization in mind. She did, in fact, consider the potential of computers for amphora studies nearly a quarter of a century ago, and rejected it then as too primitive .... There is an illustration in Donald Knuth's TeXBook, which no doubt many of you know, showing a small computer with a stop watch, timing the TeX lion as it runs around the track, which we prefer to think of in reverse: Virginia Grace is holding the stop watch, watching critically, while a panting computer races around desperately trying to catch up.