Proto-cuneiform: the archaic texts of Uruk

The corpus is about 6,000 clay tablets and fragments. The great majority come from Uruk in southern Iraq, where the German excavations in the Eanna temple precinct recovered them from 1928 onwards, almost always in secondary deposits such as rubbish pits and the fill of later buildings, which is why their dates rest on the shapes of the signs rather than on stratigraphy⁠[1]⁠[2]⁠[12]. The site counts on CDLI are about 190 Uruk V (numerical tablets without other signs), 1,776 Uruk IV and 3,094 Uruk III tablets from Uruk; 238 from Jemdet Nasr, excavated in 1926 and 1928 and divided between the Ashmolean Museum and the Iraq Museum; 398 from Umma, mostly looted; and small groups from Larsa, Tell Uqair, Kish and Eshnunna⁠[3]⁠[26]. The Uruk tablets are shared between the Vorderasiatisches Museum in Berlin and the Iraq Museum in Baghdad. A further 58 tablets of the Erlenmeyer collection, sold at auction in 1988, were bought by the State of Berlin⁠[3].

The editions are the series Archaische Texte aus Uruk (ATU), begun by Adam Falkenstein in 1936 and continued by Green, Nissen, Englund and Damerow, and Materialien zu den frühen Schriftzeugnissen des Vorderen Orients (MSVO), which re-edited Jemdet Nasr and published the scattered collections⁠[1]⁠[7]⁠[4]⁠[26]⁠[27]. Englund's 1998 chapter is the standard synthesis⁠[2]. Every tablet is on the Cuneiform Digital Library Initiative with a transliteration in ATF, a photograph and usually line art; as of May 2020 the Uruk-period set held 6,726 artifacts, 6,267 transliterated and 5,274 with at least one readable sign⁠[22]. The transliterations may be copied and reused with attribution; photographs belong to the museums⁠[20]⁠[19]. The tablets are small, mostly rectangular, and divided into cases by ruled lines, each case holding a numerical notation and one or more signs; totals are written on the reverse⁠[5].

The numbers are read. Damerow, Englund and Nissen showed that about sixty numerical signs belong to more than a dozen systems, each with its own factors: a sexagesimal system for discrete objects (1, 10, 60, 600, 3,600), a bisexagesimal system for rations (with 120 and 1,200), a grain-capacity system (the ŠE system) whose units stand in the ratios 5, 6 and 10, an area system and systems for time, and the same sign has a different value in each⁠[5]. The systems are fixed by the arithmetic: totals on the reverse must equal the entries on the obverse, and a correct assignment makes hundreds of tablets add up⁠[6]. The same method, applied to proto-Elamite, fixed its numerals too⁠[25].

The non-numerical signs number about 2,100 in the CDLI list, with a long tail: about 530 signs occur once and about 104 more than 100 times; EN, GAL and ŠE each occur more than 1,000 times⁠[9]⁠[8]⁠[28]. Many are pictorial (heads of animals, vessels, plants, body parts) and many are identified through their cuneiform descendants, so that signs can be glossed as barley, sheep, textiles, officials and institutions without being pronounced⁠[2]. The lexical lists give a second foothold. The Professions list (Lu A), the Tribute list (Word List C), and lists of cities, vessels, metals, fish and birds were copied in fixed order at Uruk and then at Fara, Abu Salabikh, Ebla and in Old Babylonian schools, where the entries carry Sumerian readings⁠[4]⁠[10]⁠[11]. Where a later copy preserves the archaic sequence, the Sumerian word is known even if the archaic sign was not written for it.

The script's structure is also known. Signs within a case are not in a fixed order; numerals precede the counted item; there is no grammar visible, no verbs, and no sentence⁠[9]⁠[14]. Phonetic spellings, where one sign is used for its sound rather than its meaning, are the test for language. Whittaker found no secure case in Uruk IV and a few plausible ones in Uruk III; the first unambiguous syllabic spellings are in the Archaic Ur texts of about 2900 to 2700 BC, mainly in personal names⁠[18]⁠[14].

The Sumerian hypothesis has been tested in three ways. The first is the search for rebus spellings, where a sign for one thing is used for a homophone. Steinkeller's review of ZATU and Wilcke's reading of the Professions list proposed such cases and sign values projected back from third-millennium Sumerian⁠[15]⁠[16]. Whittaker, following C. Johnson, checked the proposed cases in their contexts and found the meanings unclear or the sign a place or institution name, leaving no secure Uruk IV example⁠[18]. The second is onomastics. Englund collected about 440 probable slave names from the archaic texts and found none of the divine elements and none of the ur- and amar- patterns that dominate Early Dynastic and Ur III Sumerian names⁠[13]. The third is syntax: numerals and qualifiers precede nouns in the archaic texts where Sumerian puts them after, although Dahl notes that administrative formats may have a syntax of their own⁠[13]⁠[14]. None of these tests has been decisive, and Monaco and others continue to defend Sumerian readings⁠[17].

The token theory has been tested against the catalogue. Schmandt-Besserat argued that clay counters sealed in clay envelopes (bullae) and then impressed on their surfaces are the ancestors of numerical signs, and that complex tokens prefigure commodity signs⁠[29]. Zimansky's review found the sign-token links to rest on visual resemblance, the categories loose, and most token types rare; Englund noted there is no agreed criterion for calling a token numerical⁠[30]⁠[2]. The link from plain tokens to numerical notation is broadly accepted; the link from complex tokens to signs is not.

Computational work has so far aimed at the sign list and the corpus rather than at reading. Born and Kelley rebuilt the sign-frequency table for the whole CDLI corpus and showed, with an unequal-variances t-test, that the first third of the Tribute list uses sign pairs that are rare in administrative texts (mean 15 tablets against 87, p = 0.0005), which argues against the view that it was a primer of common signs⁠[22]⁠[31]. Born, Monroe, Kelley and Sarkar then learned a character inventory from sign images by unsupervised deep clustering, to check how far the editors' sign list is an artefact of their conventions⁠[23]⁠[24]. Sproat's simulation of how a concept-based symbol system becomes phonetic when repeatedly mapped onto a language with monosyllabic morphemes offers a model of the Uruk III to Early Dynastic transition rather than a test of it⁠[32].

Four positions are live. The first, held by most Assyriologists, is that proto-cuneiform was devised by and for Sumerian speakers: the sign shapes continue without a break into Early Dynastic Sumerian cuneiform, the lexical lists continue with Sumerian readings, the sexagesimal numeration matches Sumerian number words, and the few plausible Uruk III rebus spellings are Sumerian⁠[12]⁠[15]⁠[16]⁠[17]. On this view the texts are Sumerian in the sense that a Sumerian speaker would have read them aloud in Sumerian, even though the writing recorded no grammar.

The second, Englund's, is that the language of the archaic scribes is unknown. Sumerian speakers were probably present in the south, but the names, the syntax and the absence of phonetic spellings mean the Sumerian attribution is a convention inherited from Falkenstein rather than a finding, and the first scribes may have spoken a language that left no other record⁠[2]⁠[13]⁠[14]. Whittaker's own proposal that this earlier language was an Indo-European "Euphratic" has no support among specialists⁠[18].

The third, Damerow's, is that proto-cuneiform is proto-writing: a system of signs for quantities, commodities, persons, institutions and transactions that represented administrative knowledge directly, with little or no reference to speech, and that the question of its language is therefore wrongly posed for Uruk IV and only begins to have an answer as phonetic coding appears⁠[9]. This is compatible with Sumerian speakers having used it.

The fourth, Glassner's, is that writing was a deliberate invention tied to language from the start, and in his view to divination rather than bookkeeping; no divinatory text is known for 1,500 years after Uruk IV and the view has not been widely accepted⁠[33]⁠[14]⁠[3].

The Blau monuments stand at the join. Two inscribed stones bought near Uruk in 1886 and dated between about 3100 and 2700 BC, they record a land transaction in archaic signs; Krebernik's 2019 readings treat them as Sumerian and show individual archaic signs splitting into the distinct signs of later cuneiform⁠[34]⁠[35].

A single secure phonetic spelling in an Uruk IV text would show that the script was tied to a language, and the language it was tied to; a secure Sumerian rebus would settle the Sumerian question for Uruk IV as the Archaic Ur spellings settle it for 2900 BC⁠[18]⁠[14]. The place to look is personal names, because names are the first things any script spells phonetically. A systematic study of the roughly 440 archaic names against Early Dynastic Sumerian and Semitic name patterns, with a statistical null model rather than a list of resemblances, would turn Englund's argument into a test that could fail⁠[13].

A stable sign list would settle the inventory question. The gap between ZATU's roughly 800 signs and CDLI's 2,095 is a gap in conventions about variants and compounds, and image-based clustering of the kind Born and colleagues have begun can be checked against the editors' list sign by sign⁠[7]⁠[8]⁠[23]. The lexical lists offer a controlled experiment: for each archaic list with later copies, the proportion of entries whose later Sumerian reading is itself a plausible reading of the archaic sign can be measured, and compared with what chance predicts⁠[10]⁠[11].

New excavation would help most. The Uruk tablets come from fill, so an Uruk IV tablet in a sealed primary context, or a bulla and its tokens found together with a numerical tablet recording the same goods, would date the script and test the token theory directly⁠[2]⁠[29]. The comparison with proto-Elamite, which borrowed the system and then went its own way, is also a test: whatever structure the two scripts share is not a property of Sumerian⁠[25].