One of the weaknesses of CDLI files has been their limited annotation of entries concerning seals and sealing in ancient Mesopotamia. Some years ago, Christina Tsouparopoulou, then a CDLI research associate at the Max Planck Institute for the History of Science, Berlin, entered to a dedicated relational database the records of the seal impressions found on texts from Drehem, ancient Puzrish-Dagan, that had formed the core of her 2008 Cambridge dissertation The Material Face of Bureaucracy: Writing, Sealing and Archiving Tablets for the Ur III State at Drehem. The 601 composite seals (that is, seals reconstructed based exclusively on their use on ancient documents), associated with some 1415 tablets registered in the CDLI catalogue, were described with a series of philological, archaeological and art historical qualifiers that Christina felt were essential to capture the detailed nuances in the (once) physical artifacts. I agreed, and agree, that in particular such elements as those describing the iconography, form, production technology, and practice of use of cylinder seals are essential for their in-depth study; but I had to be realistic, when Christina's position in Berlin ended, about the amount of resources we would be able to dedicate to a seals catalogue, and I thus decided to streamline it for continuation at CDLI, dispensing with its involved archaeological apparatus. Seals are now entered as any other CDLI artifact, but are treated as composite reconstructions, and, in the case of existing objects, as physical artifacts, though still essentially composites. These two categories are noted as “CDLI Seals n (composite)” and “CDLI Seals n (physical)”, whereby n is a six-digit number that, in ‘S123456', constitutes an artificial seal ID in our catalogue. It may be noted that specialists such as Rudi Mayr identified many instances of the re-use of the same physical seal. In these cases, each seal generation represents a new artifact, and therefore a new seal ID number with associated witnesses, even where the expert will identify two legends, one under the other, on the same seal.

The major interest, and given the nature of our files, capability of CDLI to contribute to seals research lies in our electronic transliterations and thus in the seals that, beginning in the ED IIIb period with some consistency, combined pictorial and geometric scenes with legends. These framed inscriptions as a rule named the seal owner, his profession or educational standing, his patronymic and, looking up in the Mesopotamian hierarchy, his administrative affiliations. They form a category of CDLI entries not originally envisioned at our conception, namely that of an ethereal composition that is related to the concept of an Urtext, but that in the case of seals and including brick stamps as the first cases of “printing presses,” is grounded in a physical artifact. Mesopotamian cylinder seals, as we know, are found in innumerable collections, with and without owner legends; I personally have no idea how many there are. In the great majority of exemplars with legends, the seals themselves appear to be lost, with exceedingly few matches having been made between actual seals and impressions on texts; the lost seals, however, are often fully reconstructable based on one, up to as many as 620 (Umma's Lukalla = S002932) impressions found on tablets in collections worldwide. More conventional composite texts now, or about to be registered in much the same way in the CDLI catalogue include royal (led by Daniel Foxvog), and, in time, lexical, and literary texts. Our ultimate internal justification for conventionalized composition entries is their use in creating cleaner data sets, since irregularities in basic text entry, lexicon and sign readings, are much more visible when things line up than when viewed individually; but their use in online research will be of greater interest to colleagues and will, hopefully, complement the available resources of the BDTNS (PI Manuel Molina cleansed BDTNS' seal legends some time ago, and these are searchable as separate entities in his files).

In the past several months then, I have spent quite a bit of my time, with the undying scripting support of Eunice Yuh-Jie Chen, graduate student of UCLA's Department of Computer Science, trying to remedy this failing, and can refer users to a prelimary CDLI page dedicated to the topic that presents some of the new search capabilities this work has made possible. In our main search page, the field “Seal ID” can be cross-referenced with other fields by typing in “S”, and in full search results pages, seal IDs are listed and color-coded as hyperlinks in the catalogue list to the left of associated images and transliterations; for instance, clicking on the ID S000003 in the entry for AUCT 1, 151, brings the user to a scroll listing of 29 witness texts, headed by their composite entry.. In raw numbers, CDLI catalogue now contains entries documenting ca. 27,550 Mesopotamian artifacts related to seals and sealing: 21,000 represent inscribed tablets whose seal impressions include owner legends; 5,250 are discrete composites derived from these impressions; and currently just 1050 are entered physical seals. As is to be expected, both in terms of its ancient administrative and legal apparatus, and in terms of the open access initiatives embraced since the commencement of data entry by Marcel Sigrist, Bram Jagersma and Remco de Maaijer, substantially furthered by BDTNS, then also by CDLI collaborators, an imposing majority of these entries derive from Ur III texts (26,000 total entries, and 5,100 composites--which can be noted to discussions of literacy and the administrative workforce in this short phase of Babylonian history; these files include the Irisagrig seal legends recently published by David Owen in Nisaba 15). Poorly accessible for harvest, aggregation and re-use have been the files of Old Babylonian, neo-Babylonian and neo-Assyrian specialists. The present work should preface a much larger initiative to order these hallmark administrative tools beginning with a full catalogue of all physical seals, focusing on seals with legends, and then inclusion in CDLI of transliterations of the other big sealing periods, if and when the artifacts and electronic files become available to us.

Eunice's working interface, scripted out of CDLI's full C-ATF file dump, was very simple. All identifed seal legends were isolated, duplicate legends merged and written into single, sortable lines in the form:

Expert users will notice many such mistakes and omissions in CDLI seal annotations. But I tend to finish stages of such bulk tasks in the database, and post the results online “to have something to look at.” An initial parsing of text proveniences has demonstrated to me, nevertheless, that the composite entries could, with exceptions of course, assume the same provenience as their witness artifacts. I suspect, further, that exact strings can be interpreted to act in a highly discriminatory fashion to keep discrete seal legends linked to single users, i.e., that the availability of better prosopographical tools will not result in very substantial numbers of differentiations in existing seal identifications. I cannot address the question of whether better iconographic annotations will lead to many such necessary disambiguations in future; probably not, but such legend-duplicates as those of the physical artifacts registered here suggest that we do have a problem. The same iconography issue can be seen in seal impressions. The legend usz-mu / dub-sar / dumu lugal-sa6-ga has in some instances legend cases that are of the same length (P104992, P125183 = S005503), in others with the first two cases short and the last one longer, framing a small figure P118486, P108308, P119870, P120363: human, P101491: a bird, P100596, P102644: a griffin!) standing behind a larger one (= S005502 & S005503, all from Umma).

A few random notes to “CDLI Seals” follow

As with the earlier general cleansing of CDLI's Ur III files (CDLN 2011:004), the seals work was not well served by current photo documentation; images recently done by Klaus Wagensonner, for instance of the Ryland texts (compare P130452 hand copy and photo!), have demonstrated the pattern recognition advantages of raking light imaging using digital cameras and RTI scanning technology (I can report on a bridge technique taught me by UCLA graduate student Michael Heinle to improve chances of reading difficult seal impressions: copy the online image to desktop, load in Photoshop and invert the colors). But even old hand copies, or more common flatbed or camera imaging, are a great improvement over nothing at all. It is, for example, common in transliteration publications that two-column legends are listed as a single column; this led to duplication entries that will in future need to be reduced to a single, two-column legend. In another instance, the common ensi2-seal of Ayakalla of Umma is found in sixty examples (both writings a-a-kal-la and a-kal-la) with col. i l. 2 lugal kal-ga, but only in five cases with nita kal-ga; all five cases, however, were edited by Ozaki & Sigrist in BPOA 1 and 7 BM and Yale texts with no image documentation. The nita kal-ga entries have been assigned to S005898-S005899 (a-a- and a- versions), but may on inspection turn out to be S000033. Or another still: P113690 is a very poorly written text, with the impression of a defectively cut seal legend; so, an accounting school text with a throw-away seal? We can check the Free Library of Philadelphia image of the seal impression, but the potential matches BPOA 2, 2325, and BPOA 6, 481, are from BM and Yale, respectively, and therefore not subject to online inspection; thus, S005327 may or may not represent the same ancient seal. Though improved by the apparatus of partial hand copies used by recent editors of Istanbul Ur III texts in the series MVN and UTI, the Arkeoloji Müzeleri texts are comparably wanting of image support.

The new entry of seal IDs excluded all or nearly all legends that seemed too ambivalent, i.e., any incomplete legend whose possible reconstructions permitted two or more possibilities among existing entries. These, currently 1652 incomplete seal legends were tagged with the general dummy marker ‘Sx' to enable new eyes with better analytical tools to locate them later. In one example, P133374 has now

Distinguishing uszur₃ from ušur₄, that often correspond to different glyphs in the inscriptions (including ušurx(|LAL₂.TUG₂|), can be vexing without image documentation; the only clear example of ušurx in a seal legend is i₇-pa-e₃ / dumu lugal-ušur_x(|LAL₂.TUG₂|) / nu-banda₃-gu₄ ^dszara₂ = S002538.

If “mu-ni,” son of Akalla gudu₄ of Nin-ura at Umma, is one person, then he had (at least) four different seals: