Sie sind auf Seite 1von 21

Memorization and the Transmission of Sumerian Literary Compositions Author(s): Paul Delnero Reviewed work(s): Source: Journal of Near

Eastern Studies, Vol. 71, No. 2 (October 2012), pp. 189-208 Published by: The University of Chicago Press Stable URL: http://www.jstor.org/stable/10.1086/666645 . Accessed: 11/10/2012 10:06
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp

.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

The University of Chicago Press is collaborating with JSTOR to digitize, preserve and extend access to Journal of Near Eastern Studies.

http://www.jstor.org

Memorization and the Transmission of Sumerian Literary Compositions*


Paul Delnero, The Johns Hopkins University

Introduction It is widely recognized that nearly all preserved copies of Sumerian literary compositions were copied by apprentice scribes as part of their training in the Sumerian language. References to the use of Sumerian literary texts as tools for training scribes, which can be found in nearly every treatment of Sumerian literature published to date, appear as early as the second decade of the twentieth century, when it had become clear that many of the thousands of tablets found during the initial excavations at Nippur contained scribal exercises.1 H. Hilprecht, who had participated in the excavation that yielded most of these tablets, had al* Special thanks are due to Daniel Fleming, Christian Hess, Jacob Lauinger, Eleanor Robson, Christopher Woods, Martin Worthington, Gbor Zlyomi, and an anonymous reviewer for their insightful criticism of an earlier draft, which was invaluable in revising this article. 1 To cite only two more recent examples, see Jeremy Black and Gbor Zlyomi, Introduction to the Study of Sumerian, in Analysing Literary Sumerian: Corpus-based Approaches, ed. J. Ebeling and G. Cunningham (London, 2007), 3: The majority of such clay tablets (i.e. tablets containing Sumerian literary compositions) are the material debris of the educational process, as young Babylonian scribes learnt to speak and write Sumerian in scribal academies and training workshops; and Piotr Michalowski, Sumerian Literature: An Overview, in Civilizations of the Ancient Near East, vol. 4, ed. J. Sasson (New York, 1995), 2283: As is the case with most southern

ready begun to acknowledge the existence of a temple school with the announcement of a forthcoming volume of tablets from the Temple School of Nippur in 1910.2 By 1916, E. Chiera had demonstrated decisively, on the basis of the shape and format of the tablets on which these texts were inscribed, that many of the tablets from the Old Babylonian levels of Nippur, including tablets containing lists of personal names, thematic lexical lists, syllabaries, and grammatical texts, were produced by scribes-in-training.3 While Chiera did not include literary compositions in
literary texts of this period, the surviving tablets represent the curriculum of the scribal schools. 2 Hermann Hilprecht, The Earliest Version of the Babylonian Deluge Story and the Temple Library of Nippur, The Babylonian Expedition of the University of Pennsylvania, Series D: Researches and Treatises, vol. 5, pt. 1 (Philadelphia, 1910), 7 . The promised volume, Model Texts and Exercises from the Temple School of Nippur, scheduled to appear as Vol. 19, Part 1, of Series A of The Babylonian Expedition of Pennsylvania, was never published, but some of the copies from the volume were recently co-published by Niek Veldhuis and H. Hilprecht, Model Texts and Exercises from the Temple School of Nippur: BE 19, Archiv fr Orientfor schungen 50 (2003/2004): 2849. 3 Edward Chiera, Lists of Personal Names from the Temple School of Nippur: A Syllabary of Personal Names, Publications of the Babylonian Section, vol. 11, pt. 1 (Philadelphia, 1916), 1617 and 4148.

[JNES 71 no. 2 (2012)] 2012 by The University of Chicago. All rights reserved. 02229682012/7102001 $10.00.

189

190 F Journal of Near Eastern Studies

this group, it is clear from his later writings that he assumed they had also been copied as scribal exercises.4 Recently, with the recognition that many of the sources for these texts were the byproducts of scribal training, there has been a renewed interest in the educational context in which these documents were copied. H. Vanstiphout identified the royal hymn A Praise Poem of Lipit-Etar (Lipit-Eshtar B)5 as an elementary exercise based on the tablet format of the sources for this text as well as the content and structure of the composition itself.6 Many of the sources for Lipit-Eshtar B are written on tablets which are quite clearly exercise tablets, containing a model text copied by pupils on the same tablets. Additionally, the text contains a sequence of basic, paradigmatic grammatical forms that seems to have been contrived for purely pedagogical purposes. Building on Vanstiphouts pioneeering work, N. Veldhuis, S. Tinney, and E. Robson have been able to reconstruct the content and sequence of the scribal curriculum at Nippur (and probably also at Ur, Uruk, and Sippar) during the Old Babylonian Period.7 According to their reconstructions, the scribal curriculum would have comprised, in broad outline, an initial stage, in which syllabaries,
Cf., for example, E. Chiera, They Wrote on Clay: The Babylonian Tablets Speak Today (Chicago, 1938), 172: As a source for our knowledge of the important textbooks used in those times, we have accordingly, first, the original temple library containing all the classics and, then, the faulty copies of these texts made by the pupils in the school. 5 ETCSL no. 2.5.5.2. Unless otherwise indicated, all of the literary compositions cited in this study will be identified with the names and numbers assigned to these texts by the Electronic Text Corpus of Sumerian Literature (ETCSL) as they appear on the projects website (http:/ /etcsl.orinst.ox.ac.uk) and in Black and Zlyomi, Introduction to the Study of Sumerian, 351412. 6 H. L. J. Vanstiphout, Lipit-Eshtars Praise in the Edubba, Journal of Cuneiform Studies 30 (1978): 3361 and H. L. J. Vanstiphout, How did they Learn Sumerian? Journal of Cuneiform Studies 31 (1979): 11826. 7 N. Veldhuis, Elementary Education at Nippur: The Lists of Trees and Wooden Objects (Groningen, 1997); Steve Tinney, On the Curricular Setting of Sumerian Literature, Iraq 99 (1999): 15972; Eleanor Robson, The Tablet House: A Scribal School in Old Babylonian Nippur, Revue dAssyriologie 95 (2001): 3966; and E. Robson, More than Metrology: Mathematical Education in an Old Babylonian Scribal School, in Under one Sky: Astronomy and Mathematics in the Ancient Near East, ed. A. Imhausen and J. Steele, Alter Orient und Altes Testament 297 (Mnster, 2002). See also Paul Delnero, Sumerian Extract Tablets and Scribal Education, Journal of Cuneiform Studies 62 (2010): 5369 and Sumerian Literary Catalogues and the Scribal Curriculum, Zeitschrift fr Assyriologie 100 (2010): 3255.
4

lists, metrology, model contracts, and proverbs were learned; an intermediate stage, consisting of elementary literary compositions such as A Praise Poem of Lipit-Etar (Lipit-Eshtar B) and A Praise Poem of Enlil-bani (Enlil-Bani A)8; and a final stage, in which much of the known corpus of Sumerian literary compositions was copied.9 In addition to providing invaluable insight into the nature of scribal education during the Old Babylonian Period, the observation that copies of Sumerian literary compositions were produced by apprentice scribes has a direct bearing on how these texts are interpreted and understood by modern scholars. Since Sumerian literature was copied for didactic reasons, the motivation for copying each of these works was not to produce another perfectly accurate master copy to ensure the survival of the composition in written form, but instead to fulfill the requirements of a particular exercise. While the goal in each case may have been to produce an accurate copy, the lower level of competence of student scribes would have presumably increased the number of errors that occurred in the process. Another factor that would have contributed to the occurrence of errors was the means by which exercise tablets were compiled. The types of errors that occur in a resulting copy reflect the method that was used to compile any given text. In antiquity and throughout the Middle Ages, before the invention of movable type and the printing press, duplicate copies of texts were generally produced by expert scribes who copied directly from one or more available manuscripts. The errors that resulted from direct copying were typically visual and mechanical, including dittography (writing the same word twice, when it only occurs once in an original), haplography (the omission of a word that occurs in direct or near sequence to an identical or similar word), and parablepsis (the unintentional omission of everything between one occurrence of a word and the next occurrence of the same word). By contrast, the errors that occur in manuscripts copied by different means are qualitatively distinct, in many instances, from errors made while copying from another exemplar. Unlike errors that result from direct copying, copies produced with a printing press, for
ETCSL no. 2.5.8.1. For a more detailed list of this reconstruction of the scribal curriculum see either Robson, More than Metrology, 331 or Robson, Tablet House: 47, table 2.
8 9

Memorization and Transmission of Sumerian Literary Compositions F191

example, will be identical in all their details, but will also contain the same errors that were in the manuscript before it went to print. Given the extent to which the different methods of copying are associated with distinct types of errors, a similar correlation must exist between how Sumerian literature was copied and the variants that occur in the duplicates of texts of this type. But the means by which apprentice scribes compiled tablets is rarely considered in explanations of the cause or significance of textual variation in the duplicates of Sumerian literary compositions. One of the main reasons for this is that the more basic question of how these texts were copied has yet to be sufficiently addressed. Even in discussions of the transmission of Sumerian literature or of Old Babylonian scribal education, the issue of copying methods is rarely considered, and the few published statements on the subject are frequently contradictory. To cite a few examples, C. J. Gadd wrote in his study of the oldest schools: The pupils in the (Old Babylonian) scribal schools were in the habit of copying or having (Sumerian literary compositions such as The Debate between Date Palm and Tamarisk10) read out to them.11 While Gadd was not precise about how the scribes copied these texts, the notion that the pupils in scribal schools had compositions read out to them seems to imply copying from dictation. . Sjberg adopted a similar position in his seminal study of the Old Babylonian Edubba, citing selected lines from Sumerian literary texts about scribal education, like Schooldays12 and A Dialogue Between Two Unnamed Scribes,13 as evidence: The following passages (from the aforementioned compositions about scribal education) show that dictation was used as a method of instruction.14 A. L. Oppenheim and S. N. Kramer, on the other hand, suggested that apprentice scribes copied directly from master copies: The student copied (literary compositions) not only for practice purposes, but, at times, also to reproduce the original for his masters
ETCSL no. 5.3.7 . Cyril J. Gadd, Teachers and Students in the Oldest Schools (London, 1956), 39. 12 Samuel Noah Kramer, Schooldays: A Sumerian Composition Relating to the Education of a Scribe, Journal of the American Oriental Society 69 (1949): 199215. 13 ETCSL no. *5.4.01 (unpublished). 14 ke Sjberg, The Old Babylonian Eduba, in Sumerological Studies in Honor of Thorkild Jacobsen, ed. S. J. Lieberman, Assyriological Studies 20 (Chicago, 1975), 15979.
10 11

or his own use, this being the usual way of building up a collection,15 or, more explicitly, ... literary works were studied, copied, and redacted with zest and zeal, with care and understanding; almost all the literary works that have come down to us are known only from copies and redactions prepared in what might be described as the post-Sumerian edubbas.16 Finally, J. Black and his colleagues have suggested memorization, along with dictation, as a copying method: Examining the different sources for individual (literary) works shows us that compositions were passed on not through the copying of earlier manuscripts but through dictation, repetition, and memorization.17 As illustrated by these citations, the three primary methods of compiling a copy that were possible in antiquitycopying directly from another textual exemplar, copying from dictation, and copying from memoryhave all been proposed as methods employed to create duplicates of Sumerian literary compositions. Yet none of these explanations relies upon a systematic consideration of the evidence. Sjberg did cite textual references that might allude to the practice of copying from dictation, and Black and his colleagues called attention to errors in copies of Sumerian literary texts that appear to have resulted from mishearing and misremembering to demonstrate that scribes copied from both dictation and memory. But even in these instances, the evidence presented was confined to a relatively small number of examples, limiting the extent to which they can be representative. In the absence of a more thorough investigation, the question of which method or methods were used to copy Sumerian literature remains unresolved. In this study, the question of how Sumerian literary compositions were copied will be reconsidered in light of the types of variants that occur in their duplicates. To establish criteria for determining how the errors in the sources for these compositions were introduced, laboratory research on verbatim memory by cognitive psychologists and studies of memory errors in other text corpora will be examined. On the basis of the criteria derived from these studies, it will be argued that nearly all of the preserved copies of Sumerian literary
15 A. Leo Oppenheim, Ancient Mesopotamia: Portrait of a Dead Civilization (Chicago, 1977), 243. 16 S. N. Kramer, The Sumerians: Their History, Culture, and Character (Chicago, 1963), 169. 17 J. Black et al., The Literature of Ancient Sumer (Oxford, 2004), xlviii.

192 F Journal of Near Eastern Studies

works were copied from memory, and not by copying from dictation or from other exemplars. Memory Errors in Cognitive Psychology Research The errors that are likely to occur when copying from memory differ significantly from other types of copying mistakes. Reproducing a text entirely on the basis of how well it is stored in the mind, without being able to see or hear the text while copying, typically leads to the occurrence of errors that are causally linked to this process. Even when a text has been memorized well, it can rarely be reproduced perfectly, as certain details, if not entire words and phrases, will inevitably be forgotten and inaccurately recalled. Although a systematic study of the types of mistakes that can be expected to result from memory errors has not been undertaken for Sumerian texts, memory and the process by which the mind remembers and forgets has been investigated extensively for over a century by cognitive psychologists. The findings of these studies provide a valuable framework for identifying memory errors in the duplicates of Sumerian literary compositions. The first modern scientific studies of human memory were carried out in the late nineteenth century by H. Ebbinghaus, who examined rote memorization by testing subjects recall of lists of nonsense syllables after periods of learning and relearning these lists. The results of his experiments, which were published in 1885 in a book entitled ber das Gedchtnis, demonstrated that after twenty minutes of learning a list, subjects were likely to have forgotten more than half of its content, but that after one month of relearning the list each day, subjects could recall over three-quarters of it. The rate of retention observed by Ebbinghaus, which is sometimes called the curve of forgetting, established two of the fundamental attributes of human memory: details are forgotten rapidly, and the ability to recall material memorized by rote memorization improves steadily over time only after repeated intervals of relearning. The observation that memory weakens over time has inspired countless studies in cognitive psychology on how memory works. Since understanding why people forget is essential to determining how memory functions, thousands of experiments have been conducted that examine different types of forgetting. The title of D. L. Schacters seminal book, The Seven Sins

of Memory, alludes to what he identifies as the seven primary forms of forgetting.18 Schacter categorizes the first threetransience, absent-mindedness, and blockingas sins of omission, in contrast to the remaining fourmisattribution, bias, suggestibility, and persistencewhich he counts as sins of commission. Within the category of sins of omission, the most common type of forgetting, transience, denotes the phenomenon observed earlier by Ebbinghaus, that memory for detail declines with the passing of time and that most forgetting takes place during the earliest stages of learning;19 absent-mindedness refers to the failure to recall information that was either never encoded properly or was overlooked at the time of retrieval, typically because of external distractions; 20 and blocking is the so-called tip-of-the tongue phenomenon, when a person feels that he or she is on the verge of remembering something, but nevertheless cannot retrieve the information that is being sought.21 Of the three sins, transience and blocking are particularly useful for identifying the types of omissions likely to occur in texts copied from memory. Transience predicts that details in a narrative will inevitably be forgotten during recall: the number of details that are forgotten will increase in proportion to the amount of time that elapses between memorization and retrieval, and decrease in proportion to the number of times the text has been learned or memorized. Blocking, on the other hand, occurs frequently with proper nouns and numbers. One explanation for why lexical items of these two types can be difficult to recall is that they are typically expressed with words that bear little or no discernible semantic or conceptual relationship to the person, entity, or thing they denote, and are memorized more as sounds than as concepts.22 Since sounds are more difficult to remember than words that can be retrieved by association with other words belonging to the same conceptual field, proper nouns and numbers are more likely to be blocked during recall than words of other types. The sensation of having the word on the tip of the tongue is caused by the feeling of certainty which arises from remembering the conceptual category to which the word belongs without being able to remember the word itself. As
18 Daniel Schacter, The Seven Sins of Memory: How the Mind Forgets and Remembers (Boston, 2001). 19 Ibid., 1340. 20 Ibid., 4160. 21 Ibid., 6187 . 22 Ibid., 62.

Memorization and Transmission of Sumerian Literary Compositions F193

an example, people will often recall car before recalling the name of the car manufacturer Honda, because in contrast to car, which is a concept that is activated on a daily basis, Honda is little more than a meaningless and easily forgotten sound to many English speakers.23 Blocking therefore predicts that more specific (and less conceptual) lexical items like the names of people and places, as well as numbers and units of measurement, are more likely to be recalled incorrectly than more meaningful and familiar words or concepts. One of Schacters sins of commissionmisattributionis another particularly common type of memory error. Mistakes of misattribution are instances in which a lexical item is recalled erroneously as another word that is semantically, phonologically, and/or visually associated with it. Schacter identifies three types of misattribution: unconscious transference, remembering a specific detail from a different place, time, or (in this case) text in place of the correct detail;24 memory binding, when details or components from different sources are brought together as remembered aspects of the same experience;25 and memory conjunction errors, in which two or more separate pieces of information are combined into a single unit, such as when the words spaniel and varnish are recalled as spanish.26 Each of these three errors is caused by what cognitive psychologists call interference. Interference is defined as the process by which memory errors result from any form of previously acquired knowledge intruding or interfering with the recall of the information one is trying to remember. This phenomenon was first identified in a series of experiments published in 1900 by G. E. Mller and A. Pilzecker, who observed that subjects had much greater difficulty remembering the content of a list when asked to perform another task before recalling the list than when the list was recalled immediately after it was learned.27 This effect, which Mller and Pilzecker labelled retroactive inhibition, was then
Ibid., 6568. Ibid., 92. 25 Ibid., 94. 26 Ibid., 95. 27 Georg Elias Mller and A. Pilzecker, Experimentelle Beitrge zur Lehre vom Gedchnis, Zeitschrift fr Psychologie, Ergnzungsband 1 (1900). For a more complete survey of the earliest experiments on interference see L. M. Johnson, Similarity of Meaning as a Factor in Retroactive Inhibition, Journal of General Psychology 9 (1933): 37789, from which the summary presented here is largely drawn.
23 24

shown to increase (i.e., to lead to more forgetting) when subjects were asked to perform a similar task (such as learning another list) before recalling the original list.28 The effect increased even further when the content of the later (or interpolated) list was similar to the content of the initial list.29 The decisive breakthrough in understanding the causes of interference, however, only came in a 1931 experiment conducted by J. A. McGeoch and W. T. McDonald.30 Having subjects memorize different listsconsisting of either synonyms, antonyms, unrelated adjectives, meaningless syllables, or three-place numbersthey observed that recall was lower for the list of synonyms than for any of the other lists, indicating that retroactive inhibition is the highest when it becomes necessary to recall synonymous or semantically associated words. This observation was soon elaborated upon by A. W. Melton and J. M. Irwin, who observed that interference can occur proactively as well as retroactively, when previously learned items interfere with the recall of subsequently learned material, including items within the same list (inter-list intrusions).31 The results of these experiments demonstrate that memory errors often involve the confusion of one word for another word associated with it in meaning or form, especially when the similar word is learned before or after the correct word, as a result of proactive or retroactive interference. Building on the groundwork established by earlier studies, much of the more recent research on memory has focused on the causes of interference rather than its effects. In a classic study, J. Deese observed that certain words are more likely to be confused with associated words in recall tasks than others. Words with numerous associates have higher associative power than words with fewer associates, so that a word like chair, which has high associative power, is more likely to be incorrectly recalled as table or another of its associates (sit, legs, seat, desk, stool, etc.), than a word like butterfly, which has fewer
Jacqueline E. DeCamp, A Study of Retroactive Inhibition, Psychological Monographs 19 (1915): 169. 29 S. E. Robinson, Some Factors Determining the Degree of Retroactive Inhibition, Psychological Monographs 28 (1920): 157 . 30 John Alexander McGeoch and W. T. McDonald, Meaningful Relation and Retroactive Inhibition, American Journal of Psychology 43 (1931): 57988. 31 Arthur Weever Melton and J. M. Irwin, The Influence of Degree of Interpolated Learning on Retroactive Inhibition and the Overt Transfer of Specific Responses, American Journal of Psychology 53 (1940): 173203.
28

194 F Journal of Near Eastern Studies

associates with which it could be confused.32 As a consequence, more mistakes are made recalling lists of words with high associative power than are made recalling lists of words with low associative power. Deeses results, which were largely overlooked when they were published, were replicated almost forty years later by Roediger and McDermott, with the additional refinement that subjects were asked to recall words in the same order in which they appeared in the original list, rather than in any order, like the subjects in Deeses experiment.33 They found that items in the middle of the list were more frequently confused with their associates than items at the beginning and end of the list. Tendencies to recall the beginnings and ends of a list or text more effectively than its middle are observed frequently in memory experiments, and are known more generally as serial position effects. The two most common serial position effects are primacy, or the more accurate recall of earlier items, and recency, the more accurate recall of later items.34 To explain the cause(s) of the results obtained by Deese, Roediger, and McDermott (also known as the DRM paradigm), a theoretical model called fuzzy-trace theory has been proposed.35 According to this model, first put forward by C. J. Brainerd and J. Kingma,36 information is encoded in memory using
32 James Deese, On the Prediction of Occurrence of Particular Verbal Intrusions in Immediate Recall, Journal of Experimental Psychology 58 (1959): 1722. 33 Henry L. Roediger and K. B. McDermott, Creating False Memories: Remembering Words Not Presented in Lists, Journal of Experimental Psychology - Learning, Memory and Cognition 21 (1995): 80314. 34 This phenomenon is also clearly observable in the patterns of variation found in the copies of Sumerian literary sources. In general there tend to be significantly fewer variants and mistakes among the individual sources for the lines at the beginning of a text than toward the middle, where the number of textual variants and errors often increases substantially. It may also account for why many of the extracts of the metrological tables that were memorized by apprentice scribes tend to cluster toward the beginning as opposed to the middle or end of a series of tables. For a discussion of the distribution of extract tablets containing metrological tables and the place of such tables in the Old Babylonian scribal curriculum, see Robson, More than Metrology, 33945. 35 Special thanks are due to Peggy Intons-Peterson for bringing the literature on fuzzy-trace theory cited in this section to my attention. 36 Charles J. Brainerd and J. Kingma, Do Children Have to Remember to Reason? A Fuzzy-trace Theory of Transitivity Development, Developmental Review 4 (1984): 31177 . Further refinements of fuzzy-trace theory include C. J. Brainerd and V. F. Reyna,

two distinct processes: verbatim memory and gist memory. As the terms suggest, verbatim memory stores the exact content of a particular stimulus, and gist memory reduces the same information to its general sense or gist, in a process called gist extraction.37 Verbatim and gist memory represent two ends of a continuum of exactness, with the precise details of a stimulus at one end and the general sense or fuzzy traces at the other. What distinguishes fuzzy-trace theory from other memory models, which assume that memories are reconstructed from a combination of verbatim and gist traces, is the notion that the mind can choose to recall information using either verbatim or gist memory, depending on the nature of the task. By conceiving verbatim and gist recall as separate processes, fuzzytrace theory provides a convincing explanation for different types of interference, according to which words are confused with other associated words because the mind confuses the fuzzy traces, or associations, of a stimulus in gist memory, with the details of the same stimulus in verbatim memory. Learning related words before or after stimulus words strengthens the gist memory of those words at the expense of verbatim memory, accounting for why proactive and retroactive interference are frequent causes of memory errors.38 Furthermore, fuzzy-trace theory applies not only to semantically related words, but also to words that are phonologically associated.39 Phonological associates, like semantic associates, are stored separately in gist
Gist is the Grist: Fuzzy-trace Theory and the New Intuitionism, Developmental Review 10 (1990): 347; C. J. Brainerd and V. F. Reyna, Explaining Memory Free Reasoning, Psychological Science 3 (1992): 33239; V. F. Reyna and C. J. Brainerd, Fuzzy-trace Theory: An Interim Synthesis, Learning and Individual Differences 7 (1995): 175; and C. J. Brainerd and V. F. Reyna, Fuzzy-trace Theory and Childrens False Memories, Journal of Experimental Child Psychology 71 (1998): 81129, among many others. For a useful overview of fuzzy-trace theory and other models of memory accuracy, see especially Asher Koriat, M. Goldsmith, and A. Pansky, Toward a Psychology of Memory Accuracy, Annual Review of Psychology 51 (2000): 481537 and H. L. Roediger and K. B. McDermott, Distortions of Memory, in The Oxford Handbook of Memory, ed. E. Tulving and F. I. M. Craik (Oxford, 2000), 14962. 37 Brainerd and Reyna, Gist is Grist, 78. 38 For a similar application of fuzzy-trace theory as an explanation for interference and other types of memory errors see Jonathan W. Schooler, The Distinctions of False and Fuzzy Memories, Journal of Experimental Child Psychology 71 (1998): 13043. 39 M. S. Sommers and B. P. Lewis, Who Really Lives Next Door: Creating False Memories with Phonological Neighbors, Journal of Memory and Language 40 (1999): 83108.

Memorization and Transmission of Sumerian Literary Compositions F195

memory apart from the verbatim traces of the same word, and are thus more likely to be confused with similar-sounding words than words which are phonologically distinct. Most research on memory errors in cognitive psychology has been conducted using word and syllable lists, but many of the results apply equally to prose texts, with a few modifications. One of the earliest experiments testing the verbatim recall of continuous narratives, as opposed to lists, was carried out by F. Bartlett, who had subjects memorize and recall a short text called War of the Ghosts.40 The mistakes made in Bartletts experiment are the same types of errors that occur when lists of words or syllables are recalled. The most frequent errors were omissions, with subjects shortening the text considerably by leaving out content that had been forgotten as a result of transience. Moreover, proper names and numbers were omitted with particularly high frequency, a result which can be attributed to blocking.41 In addition to omissions, Bartlett identified certain types of memory errors that seemed to be specific to recalling prose texts, including the tendency to replace events that did not seem logical to the readers with material that was more concordant with their experience, and to substitute familiar expressions for more archaic language.42 The cognitive psychologist David Rubin, who studied memory errors as a way of explaining why the content of epics changes as a result of oral performance, has confirmed many of Bartletts results.43 To test the effects of long-term memory on verbatim recall, Rubin had subjects memorize and reproduce well known texts like the preamble to the constitution, Hamlets soliloquy, and popular Beatles songs. The results he obtained verify that interference is just as much a cause of false recall with prose narrative as it is with word and syllable lists, particularly when the source of interference is similar to the line or passage

being recalled, or contains a word with high associative frequency.44 Recalling the song Rocky Raccoon, for example, subjects incorrectly substituted words like everybody and legs with the semantically associated words most people and hands, and erroneously replaced words like Dan and revival with phonologically related words like Stan and survival.45 Subjects also transposed similar words and phrases from different sections of the song as a result of proactive and retroactive interference. For instance, subjects transposed the expression walked into for checked into in line 8 because walked into occurs in a similar context earlier in the song in line 6, and transposed grinning a grin for into his room in line 8, because grinning a grin occurs in a similar context in line 18.46 Finally, Rubin observed that subjects were much more successful when recalling the beginning of a text than the middle, confirming that the serial position effect known as primacy occurs for continuous narratives to the same extent as it does for lists.47 Rubins explanation for these types of mistakes is also compatible with fuzzy-trace theory. To account for the errors that occur in the recall of prose texts, Rubin argued that, in the process of learning narratives, people encode the meaning, sound, and appearance of each word and passage, and combine information from all three of these encoded sources when reproducing the text from memory.48 Since words are remembered as a combination of their auditory, semantic, and visual properties, mistakes occur most frequently when a word that is phonologically, semantically, and/or graphically similar interferes during recall. Memory Errors in Other Text Corpora While research on the errors that are likely to occur when copying from memory has yet to be conducted for the purpose of classifying variants in the duplicates of Sumerian texts, studies of this type have been carried out for at least three other text corpora: Old English poetry, Middle English romances, and the first printed editions of such William Shakespeare plays as Romeo and Juliet, the Merry Wives of Windsor, and
Rubin, Memory in Oral Traditions, 14755. Rubin and Hyman, Memorabeatlia, 210. 46 Ibid., 210. 47 Rubin, Memory in Oral Traditions, 17688. 48 Ibid., 9094.
44 45

Frederic Charles Bartlett, Remembering: A Study in Experimental and Social Psychology (Cambridge, 1932). 41 Bartlett, Remembering, 8182. 42 Ibid., 8489. 43 See, in particular, David C. Rubin, Memory in Oral Traditions: The Cognitive Psychology of Epic, Ballads, and Counting-out Rhymes (Oxford, 1995); but also D. C. Rubin, Very Long-Term Memory for Prose and Verse, Journal of Verbal Learning and Verbal Behavior 16 (1977): 61121; and D. C. Rubin and I. E. Hyman, Memorabeatlia: A Naturalistic Study of Long-term Memory, Memory and Cognition 18 (1990): 20514.
40

196 F Journal of Near Eastern Studies

Hamlet. The criteria for identifying memory errors in the copies of these texts accord completely with the results of experiments on memory conducted by cognitive psychologists. Moreover, the features that have been determined to be characteristic of memory errors for each individual text corpus are nearly identical across all three corpora, indicating that the criteria can be generalized with equal validity for compositions of different types. One of the earliest attempts to establish definitive criteria for identifying memory errors in copies of literary works was carried out by the Shakespeare scholar B. Shapin in 1944.49 Prior to Shapins study, an influential book was published by G. I. Duthie, who argued that some of the textual corruption and errors in earlier manuscripts, or quartos, of Shakespeares play Hamlet were the results of mistakes made by actors who had copied the play from memory.50 To test this theorywhich has become known as the memorial reconstruction hypothesisShapin memorized a play entitled Witch Hunt to determine whether the errors she made in recall were similar to the types of mistakes found in the so-called bad quarto of Hamlet. The errors she reported making include omissions, conflations, transpositions of clauses, and the anticipation of a word or line and its omission from its rightful place.51 Since similar mistakes occur in the bad quarto, Shapin concluded that Duthies argument for memorial reconstruction was essentially correct. The results of Shapins study have been corroborated by other scholars attempting to confirm or refute the memorial reconstruction hypothesis, including most recently by L. E. Maguire, who replicated Shapins experiment with a different memory experiment of her own.52 Beginning in 1979, the BBC filmed a series of Shakespeare plays to be broadcast on television. Since the plays were produced under time pressure, the participating actors were given an average of six weeks to memorize and rehearse the plays before they were filmed. Due to time and budget constraints, very little re-filming was done to correct mistakes that had been

made by the actors during performance, so the number and types of memory errors that occurred could be accurately determined by simply studying the broadcast recordings.53 By comparing the directors playtext used for each play with the filmed performances, Maguire catalogued and analyzed all of the errors the actors made for six plays (Hamlet, Julius Caesar, The Winters Tale, Antony and Cleopatra, Richard III, and Henry IV).54 The errors that occurred were numerous (55163 errors per play), but distributed more or less evenly across roles.55 The most common types of errors observed by Maguire were: 1) Grammatical errors - including changes of tense, changes of mood, alterations of singular to plural and vice-versa, and changes in the forms of possessive pronouns.56 2) Substitutions - usually of a correct word with a synonym (e.g., force for strength, holy for lordly, just for good, etc.) or with a similarsounding word (e.g., ought for oft, so bade for obeyd, and in sight for incite).57 3) Additions - typically of connective or emphatic conjunctions such as and and but, or yet and now.58 4) Omissions - often of conjunctions, inductions, explanations, and other small units of text, but also more substantial omissions of half lines, whole lines, and entire sequences of lines.59 5) Inversions - both locally within a line (e.g., either pluck back or push on for either push on or pluck back, we would, an if we could for we could, an if we would, etc.), and with entire lines that were moved erroneously from one place in the play to another.60

49 Betty Shapin, An Experiment in Memorial Reconstruction, Modern Language Review 39 (1944): 917 . 50 George Ian Duthie, The Bad Quarto of Hamlet (Cambridge, 1941). 51 Shapin, An Experiment in Memorial Reconstruction, 9. 52 Laurie E. Maguire, Shakespearean Suspect Texts: The Bad Quartos and Their Contexts (Cambridge, 1996).

Ibid., 135. Ibid., 136. 55 Ibid., 136. 56 Ibid., 136. 57 Ibid., 13639. Although the substitution of similar-sounding words was not included in the discussion of errors made by actors for the BBC recording, they are discussed in the section for aural error in Maguires analysis of the errors that occur in the quartos for different Shakepeare plays (pp. 19698). 58 Ibid., 13941. 59 Ibid., 14244. 60 Ibid., 14445. Maguire labels these types of errors transpositions, but the term inversion seems more precise to prevent confusion with transpositions in the sense of blocks of texts transferred from one point in the text to another as is the case with anticipations and preservations (discussed below).
53 54

Memorization and Transmission of Sumerian Literary Compositions F197

6) Numerical errors - such as three and twenty for three and thirty, a thousand for ten thousand, etc.61 7) Compound errors - two or more of any of these types of errors in combination, including substitution and inversion (e.g., Time as much again / My brother would be filld up for Time as long again / Would be filld up, my brother), as well as addition and omission (e.g., the wrongs that I have done thee for the wrongs I have done thee stir).62 All of these types of errors are caused by factors identified in experiments conducted by cognitive psychologists. Omissions are the inevitable consequence of transience, confirmed by Ebbinghaus curve of forgetting; substitutions, grammatical errors (in which the correct grammatical form is substituted with an analogous construction), and inversions are caused by proactive and retroactive interference, as predicted by the DRM paradigm and fuzzy-trace theory; the forgetting of numbers is the anticipated result of blocking; and the addition of conjunctions and stock phrases to convey what is perceived to be the style of a text can be attributed to what Bartlett termed the reproduction of style. 63 Furthermore, almost exactly the same types of errors identified by Maguire have been found in other text corpora thought to have been compiled from memory. In A. Jabbours study of the errors in the medieval sources for the Old English text Soul and Body, for example, he noted a proliferation of omissions, additions, inversions of word-order, and substitutions of synonyms and similar sounding words, as well as memorial skipsfrom one phrase to a similar phrase farther along.64 The same types of memory errors have also been observed by M. McGillivray, who studied the mistakes that occur in manuscripts of the Middle English romances Floris and Blaunchefore, King Horn, The Seege of Troy, and Sir Orfeo.65

In addition to grammatical variants,66 substitutions of similar expressions,67 omissions,68 and auditory errors,69 McGillivray identified another type of error called transpositions, which he defined as expressions interchanged within the same line or between different lines within the same text.70 Basing his definitions on a similar classification by H. R. Hoppe, who investigated the memory errors in a seemingly corrupt manuscript of another Shakespeare play, Romeo and Juliet,71 and a less elaborate typology for variants in medieval vernacular texts by J. Rychner,72 McGillivray distinguished four subtypes of transpositions: 8a) Anticipations - the insertion of words or phrases several lines or scenes before their proper place.73 8b) Preservations - words and lines used later than their proper place.74 8c) Repetitions - the replacement of words or phrases with expressions that occur earlier or later in the same text.75 8d) Borrowings - the transferring of material from a different text.76 All four types of transpositions are memorial transfer errors that involve the replacement,77 addition,78 or omission79 of words and phrases as a result of confusion with similar expressions or passages within the same text (or, in the case of borrowings, from another text).80 Like substitutions, grammatical errors, and inversions, transposition errors can also be explained as having been caused by cognitive interference. Anticipations are the expected result of proactive interference, in which encountering similar material
Ibid., 3637 . Ibid., 3739. 68 Ibid., 40. 69 Ibid., 4041. 70 Ibid., 47 . 71 Harry Hoppe, The Bad Quarto of Romeo and Juliet: A Bibliographical and Textual Study (Ithaca, 1948). 72 Jean Rychner, Contribution ltude des fabliaux (Geneva, 1960). 73 McGillivray, Memorization, 4849. 74 Ibid., 5051. McGillivray and Hoppe use the term recollection for this type of transposition, but preservation seems more precise. 75 Ibid., 52. 76 Ibid., 5253. 77 Ibid., 6367 . 78 Ibid., 6771. 79 Ibid., 7174. 80 Cf. McGillivray, Memorization, 63, 67, and 71.
66 67

Ibid., 14546. Ibid., 146. 63 Bartlett, Remembering, 81. 64 Alan Jabbour, Memorial Transmission in Old English Poetry, Chaucer Review 3 (1969): 18487 . 65 Murray McGillivray, Memorization in the Transmission of the Middle English Romances (New York, 1990).
61 62

198 F Journal of Near Eastern Studies

before learning new material leads to false recall, and preservations are analogous to errors caused by retroactive interference, in which material learned later is confused with similar material earlier in the text. Interference would also account for repetitions, which could be caused by either proactive or retroactive interference (depending on whether the incorrectly repeated passage occurs earlier or later within the same text), and for borrowings, which presuppose that the borrowed passage was transferred from a previously learned text as a result of proactive interference. Because the causes of transposition errors are evidently rooted in memory, it is unlikely that they were caused by other forms of textual reproduction. McGillivray considers such transfer errors particularly diagnostic: Memorial transfer, the movement of material from one part of a text to another part which is physically remote, but which is liable to confusion with it because of similarities of situations, content, or language, is a very secure indication that the entire text in which it occurs has at some stage in its transmission been copied from memory.81 Memory Errors in the Sources for Sumerian Literary Compositions Nearly all of the types of memory errors that occur in other text corpora, and in particular the eight identified by Maguire and McGillivray described above, are attested in duplicates of Sumerian literary compositions. To illustrate memory errors of each of these types, examples were selected from the sources for a group of ten literary compositions known as the Decad. This group comprises the following texts: 1) A: A praise poem of ulgi (ulgi A) = ETCSL no. 2.4.2.01 2) LiA: A praise poem of Lipit-Etar (Lipit-Etar A) = ETCSL no. 2.5.5.1
81 McGillivray, Memorization, 5 apud Linda Marie Zaerr and Mary Ellen Ryder, Psycholinguistic Theory and Modern Performance: Memory as a Key to Variants in Medieval Texts, Mosaic 26/3 (1993): 33, who corroborated the results of McGillivrays study by identifying similar types of memory errors in the sources for another Middle English romance, entitled The Tournament of Tottenham.

3) Al: The song of the hoe = ETCSL no. 5.5.4 4) InB: The exaltation of Inana (Inana B) = ETCSL no. 4.07.2 5) EnA: Enlil in the E-kur (Enlil A) = ETCSL no. 4.05.1 6) KH: The Ke temple hymn = ETCSL no. 4.80.2 7) ErH: Enkis journey to Nibru = ETCSL no. 1.1.4 8) IEb: Inana and Ebi = ETCSL no. 1.3.2 9) Nu: A hymn to Nungal (Nungal A) = ETCSL no. 4.28.1 10) GH: Gilgame and Huwawa (Version A) = ETCSL no. 1.8.1.5 These texts were selected because they are among the compositions for which there are the most preserved duplicates, and which appear to have been copied frequently as a group by apprentice scribes. Since these compositions were copied extensively during the Old Babylonian Period, each of the lines in these texts is preserved in an average of fifteen to thirty copies, providing abundant data for distinguishing intended forms from erroneous variants.82 Furthermore, since complete scores of all of the texts in this group could be consulted, it was not necessary to rely on composite texts in selecting the examples cited.83 At the time the scores used for this study were compiled, a total of 742 exemplars had been identified. Of these, 541 were from Nippur, out of which nearly half, or 204 exemplars, were from a private house at the site known as House F. This house, which was almost certainly a place where scribal training was conducted, contained a representative cross section of the literary compositions copied as part of the scribal curriculum
For a more detailed discussion of the use of the Decad as a corpus for studying Sumerian grammar and for further references to select treatments of the individual compositions in this group of texts, see P. Delnero, Pre-verbal /n/: Function, Distribution, and Stability, in Analysing Literary Sumerian: Corpus-based Approaches, ed. J. Ebeling and G. Cunningham (London, 2007): 11618 with notes 515, and P. Delnero, Variation in Sumerian Literary Compositions: A Case Study based on the Decad, Ph.D. dissertation (Philadelphia, 2006): 2224 with notes 1727 . For treatments of the Decad as a group of compositions and its use and significance in the Old Babylonian scribal curriculum see Tinney, On the Curricular Setting of Sumerian Literature; Robson, Tablet House; Black et al., The Literature of Ancient Sumer, 299301; and Delnero, Variation in Sumerian Literary Compositions, 22147 (with additional literature). 83 For these scores, see Delnero, Variation in Sumerian Literary Compositions, 18572473.
82

Memorization and Transmission of Sumerian Literary Compositions F199

throughout Mesopotamia at the time.84 Moreover, the duplicates from House F were probably produced by a small group of scribes during a short period of time before the house was abandoned in the tenth year of the ruler Samsuiluna (ca. 1739 b.c.).85 To eliminate variants that may have been the result of synchronic or diachronic factors, only examples that occur in sources from Nippur, and whenever possible, from House F, will be cited. All of the variants noted occur only in a single source. In addition to it being unambiguous that these variants are incorrect on the basis of the extent to which they deviate from the standard form expected in the context of the clause or passage, the large number of sources that contain the form from which they vary is a further indication that they are mistakes. Since the identification of certain variants as mistakes is essential to the argument being made in this study, some additional clarification about how mistakes can be distinguished from other types of variants, such as acceptable alternative writings, orthographic economy, and intentional grammatical alteration, is required. In discussing the transmission of pre-modern texts that have not been mechanically reproduced, the following assumption can and should be made: cultural attitudes toward reproducing a text so that a copy of the text is identical in every, or nearly every respect to a hypothetical Vorlage were almost certainly less strict in antiquity than they are today, when the technology exists for producing identical copies of a master text. The possibility must therefore be allowed that differences in the way certain words or forms are spelled and grammatical constructions are rendered are not necessarily errors, but acceptable alternative writings that would have been tolerated when conceptions of faithful reproduction were less rigid and more tolerant of alterations intentionally introduced by scribes. Recognizing that copies of Sumerian literary compositions were rarely, if ever, intended to be completely faithful reproductions of a single definitive version of the text, and that a certain degree of variability was unquestionably permitted and indeed inevitable, the objection could be raised as to whether some of the variants that have been identified as misFor a detailed discussion of the archaeology of House F and the reconstruction of the scribal curriculum on the basis of the tablets that were found there see Robson, Tablet House. 85 Ibid., 42.
84

takes are not mistakes at all, but free alterations resulting from a more general tolerance for variability. To address this issue, it is necessary to determine the extent to which variability was tolerated in the copying of Sumerian literary works and to establish criteria for distinguishing between free variants and scribal errors. Although this is not the place for a detailed presentation of the evidence, there are two good reasons for assuming that the content of the compositions copied during the Old Babylonian Period was relatively fixed (at least in the versions that were taught to apprentice scribes) and that constraints were normally placed on the amount of variability tolerated (again in the context of scribal training). First and more briefly, the content of the individual duplicates for compositions that survive in numerous copies from the same period is more often identical than it is different. For the ten compositions in the Decad, which in some instances have as many as thirty copies per line, an average of 9095 percent of the content of all of the duplicates for the line is completely identical in every orthographic and grammatical detail, and when variation occurs at all, there are typically fewer than five to ten textual variants for that line. The consistency with which multiple sources do in fact have the same content therefore suggests a certain degree of stability in the content of the version(s) that served as model texts in different scribal centers at the time the known copies of these compositions were compiled. Secondly, with the exception of a small number of duplicates which contain a substantial number of variants and whose content differs significantly in many other respects from other sources, there are few sources which consistently contain variant spellings and grammatical interpretations of particular words and forms. When a source contains an orthographic or grammatical variant that is not attested in any other source, there are rarely additional variants of a similar type in the same source. This is contrary to what would be expected if there was a large amount of tolerance for rendering forms freely. If scribes had been permitted to choose freely between two or more acceptable renderings of a form, such as the omission of grammatical elements that could have been deduced from the context, or the use of different cuneiform signs to render the same word, tendencies to make such choices should be observable throughout individual sources and not merely in isolated instances

200 F Journal of Near Eastern Studies

within a single source. If variation had been more generally tolerated, why would a scribe choose to render a form differently only once when it would have been permissible to do so with each and every similar form? Since free variants seem to be the exception and not the rule, it is more likely that a variant that occurs only once within the same source and among the other preserved sources for the line is an error and not an intentional alternate writing. These two observations have important implications for how variation in copies of Sumerian literary compositions is interpreted more generally. The first implies that something like a standard text was used as a model for many of the surviving duplicates, allowing for minor differences in orthography and grammar among the different versions of this standard text that served as models in different places. The second implies that scribes generally did not permit themselves a significant amount of flexibility in the rendering of the words and forms in the compositions they copiedwith minor exceptions, such as the use of nearly identical graphemes ( gi for gi4, uru for uru a , mu 2 for mu 3, etc.) and the omission of determinatives (such as na4 with za-gin3, and g i with al) and phonetic complements (such as ulu3 followed by lu). To summarize, then, what types of variants have been identified as mistakes: it is not being assumed that every textual variant, isolated or otherwise, is an error, since variants of other types, including synchronic variants, diachronic variants, idiosyncratic variants, variants that are the result of source relationship, and free variants, are all attested in duplicates of Sumerian literary compositions. Instead, it is only being claimed that scribal errors do occur and that it is (sometimes) possible to distinguish such errors from variants of other types. Thus, for the purposes of this study, a variant has only been classified as an error if it meets all of the following conditions: 1) the variant only occurs in one of the preserved sources for the line; 2) the source in which the variant occurs does not contain multiple variants of the same type; 3) variants of the same type are rare in Sumerian literary sources; and 4) whenever distinguishable, the form is overtly incorrect and differs substantially from the form that occurs in the other preserved sources.

In classifying memory errors, it is useful to make a distinction between conscious and unconscious mistakes. As shown above, the human mind, by its nature, is only able to store information temporarily, and is thus subject to a continual process of decay. As a consequence, mistakes made while copying from memory often result from the inability to recall specific details. In light of this, it is very likely that there were instances in which the scribes who produced these sources could not remember all of the details of the composition they were copying, and after unsuccessfully attempting to remember more clearly, either guessed and wrote something they thought could have been correct, or simply omitted what had been forgotten entirely. Since the acts of intentionally omitting, adding, or substituting occur consciously, mistakes of this type can be considered conscious memory errors. Variants involving conspicuous omissions, substitutions, or additions can be identified as mistakes that resulted in this manner. Examples of these types of errors from the Decad include those treated below. Conscious Memory Errors Omissions (1) IEb 14086: ba-ti (3 N-T 728) for ba-an-iin-ti (4,4)87
All of the examples cited in this study follow the line numbering of the scores of the Decad in Delnero, Variation in Sumerian Literary Compositions, 18572473. Since, in this system, only lines that occur in the majority of extant sources were numbered with whole numbers, and lines that occur in less than half of the preserved sources were given lettered numbers (e.g., 48a, 56b, etc.), the line numbers do not always correspond exactly to the line numberings for these compositions in the ETCSL corpus. For the equivalences of the line numbers used in this study and the corresponding ETCSL numbers, see Delnero, Pre-verbal /n/, 11920 n. 15 and Delnero, Sumerian Extract Tablets, 54041 n. 33. 87 The first form cited in each example is a variant form that only occurs in one of the preserved sources for the line in which it is attested. The museum number of the source in which this variant occurs is listed in parentheses directly after the citation of the variant form. The form following this citation is the form that is found in the majority of preserved sources for the same line. This form is followed by two numbers in parentheses. The first number refers to the total number of sources that contain this non-variant form, and the second number (after the comma) refers to the number of sources from this total which are from Nippur. So, for example, (11,7) indicates that there are a total number of eleven sources with the non-variant form, and out of these eleven, seven sources are
86

Memorization and Transmission of Sumerian Literary Compositions F201

(2) A 40: gub-ba me-en omitted (3 N-T 927, 525) (11,7) ar (3 N-T 721) for ki-si3-ga (3) InB 69: ki bi2-g ar (19,13) bi2-ib-g Substitutions (4) EnA 45: uri5(E.AB) ru2-a uri2(E. UNUG)ki (3 N-T 507) for ki uri3(E) ru2-a uri3(E) (9,6) (5) Nu 33: ku2 nu-til-le gal (CBS 3424b) for pe10 gal (6,6) (6) KH 112: EN.PAP-e ki am3-ma-MU3-le an-ma-MU3-e (UM 29-13-422) for enkum al2-le-e (11,6) e-ne KAxLI ki am3-ma-g Additions (7) EnA 125: nu-mu-ni-ib2-ra-ra (CBS 10475) for nu-ra-ra (3,2) a2-g a2 (N (8) KH 12: dub-ba-am3 mu-un-g a2-g a2 (6,3) 3534+N 3530) for dub-ba g Unconscious Memory Errors In addition to these types of errors, some of the mistakes that are found in the copies of the compositions in the Decad appear to have been committed without immediate or direct awareness that they had or could have been made. These types of errors can be classified as unconscious memory errors. While memory errors deemed to be conscious can be explained as instances in which the scribe actively tried, but failed, to remember the content of the text, these mistakes seem to have taken place passively, when the scribe was either not aware of having falsely recalled something, or was not concentrating fully on the task at hand, and in a moment of distraction unintentionally omitted, added, or replaced a correct sign or lexeme with another that was not correct. Another set of unconscious mistakes consists of the inversion of the order of specific elements or lexemes, and the transference of writings or forms that occur in a similar context to another where they are incorrect. These types of errors, which occur frequently
from Nippur. All of the variants cited occur only in sources from Nippur and are unique to the sources in which they are attested. Sources that are from or near House F at Nippur carry 3 N-T numbers. All of the other museum numbers (CBS, UM, and N) refer to tablets that are from unknown findspots at Nippur.

in experiments testing verbatim memory, and are also commonly attested in copies that were produced from memory of Old English compositions (e.g., the Anglo-Saxon Chronicle, or Middle English romances such as Sir Orfeo and the Seege of Troy) are attributed to what cognitive psychologists term proactive or retroactive interference, and are clearly diagnostic of memory errors; examples include inversions, anticipations, and preservations. Inversions (9) ErH 86: ni2-bi nam-KU ... ni2-bi nam-du8 (3 N-T 791) for ni2-bi nam-du8 ... ni2-bi nam-KU (9,3) (10) CBS 8533: (mu-ni-in-)si3 for (mu-ni-in-) sig7 (9,5) in EnA 96 and sig7-ga for si3-ga (2,1) in EnA 157 Anticipations (ant.) (11) GH 88: gir3-e3 (3 N-T 465) for gir3 (7,5) in ant. of e3 in kur-e3 and uruki-e3 (12) EnA 78: (an-ne2) us2-sa (CBS 14152) for (an-ne2) im-us2 (10,8) in ant. of (an-ne2) us2-sa (7,6) in EnA 144 -g a2 ... (13) A 84: (an-ne2) aga zi ma (sag gi-en) (UM 29-16-198+N 1519+N 1572, collective source with A and LiA) for (an-ne2 -g a2 ... gi-en) (7,3) in ant. of aga) ku3-ge (sag -g a2 ... gi-en) (11,6) (an-ne2 aga) zi ma (sag in LiA 24 Preservations (pres.) a2-g a2[-de3] (3 N-T (14) A 72: i3(NI)-e3-g a2-g a2(-de3) (10,4) pres. of (e2-a-) 826) for i3-g ni-e3 2-me-lam2(-)g ar (3 N-T 302) (15) InB 22: nig 2-me(-)g ar (9,6) pres. of (ni2) me-lam2 for nig (13,9) in InB 21 -g a2) e2-em-mi(16) LiA 24: (an-ne2 ... sag in-gi-en (UM 2913615) for (an-ne2 ... sag ga2) mu-ni-in-gi-en (11,7) pres. of (an-ne2 ... -g a2) e2-em-mi-in-gi-en (5,3) in A 84 sag Two of these examples, line 84 of A and line 24 of LiA, are of particular interest. Although cases in which forms are erroneously transferred from one part of a text to another are frequently influenced by similar forms that occur later or earlier in the same line (or in a different line in the same composition), these

202 F Journal of Near Eastern Studies

two examples involve the confusion of forms from different compositions. From catch-lines connecting A and LiA, and from collective tablets containing both of these texts, it is known that the two compositions were sometimes copied in sequence.88 Line 84 of A and line 24 of LiA are identical in all but two details. In A, aga crown is qualified with ku3-ge, pure, instead of zi mah true majestic, and the verbal chain preceding the verb gi-en, to establish, is he2em-mi-in- instead of mu-ni-in-. As shown in example 13, however, one source for A line 84 has zi mah instead of ku3-ge after aga, anticipating the form in LiA line 24; and in example 16, one of the sources has the verbal chain he2-em-mi-in- instead of mu-ni-in-, preserving the verbal form from A line 84. Both of these variants are instances of the memory errors McGillivray identified as borrowings, in which similar material is erroneously transferred from a different text; these are clearly indications that the sources with these mistakes were copied from memory. Visualizing Errors Other types of mistakes that could have occurred while copying from memory include visualizing errors,89
For lists of the sources connecting A and LiA see Tinney, On the Curricular Setting of Sumerian Literature, 16970; Delnero, Variation in Sumerian Literary Compositions, 3034; and Delnero, Sumerian Literary Catalogues, 33 n. 4 and 3435 with n. 7 . 89 To avoid confusion with visual errors that are caused by seeing the content of the source text incorrectly while copying directly from another written exemplar, the term visualizing is used throughout this study to refer to memory errors that are triggered by a visual aspect of the content of a text as it is being remembered. As described in the section above entitled Memory Errors in Cognitive Psychology Research, memory has both a visual and a mnemonic (phonological) component, and each of these two aspects of a stimulus are frequently combined during the act of recall. Just as a particular melody is recalled by hearing in the mind how it sounds, visual stimuli, such as landscapes, are recalled by seeing in the mind how they look. In neither case, however, does seeing or hearing take place in remembering in the same way as it occurs when actually seeing or hearing by direct perception. In the case of words and forms, which have both a visual and an aural component (the way they look when they are written and the way they sound when they are pronounced), both aspects are combined when the word is remembered. Although words (and other stimuli) are experienced differently in the mind than they are experienced directly through the senses, they can nonetheless be visually (and phonologically) confused with similar looking (or sounding) words in much the same way as they can when they are seen (or heard) directly. For this reason some of the same types of visual and phonological errors can occur while copying from memory as
88

such as the confusion of lexemes or signs that are visually similar, or the misreading of a sign with a value that is incorrect in a way affecting the rendering of an entire form. Examples of such errors include: Resemblance Errors a2 command (3 N-T 675) (17) Nu 78: a2-ag 2-g a2 in my arm (6,6) for a2-g (18) InB 1: (nin me) du10-ga (UM 29-15422+CBS 7847) for (nin-me) ar2-ra (12,8) Division Errors (19) ErH 24: im-ma-ti-a (HS 1447) for ni2(IM) u-ti-a (7,4) (20) Al 36: gial mu2-mu (3 N-T 919,463) for 2 al-mu2-mu2 (14,8) Phonetic Errors Phonetic errors (e.g., the confusion of signs with identical or similar phonological readings), the phonetic writing of multi-sign logograms or substantives, and the phonetic combination of two forms into a single form (also known as sandhi writings) are other errors of this type. Homophonous Signs (21) A 40: ne(-ba) (3 N-T 927,525) for ne3(-ba) (8,7) (22) IEb 120: (gurun im-)la (3 N-T 577+3 N-T 440) for (gurun im-)la2 (7,6) Syllabic Writings (23) A 73: (u-)ni-g in (3 N-T 826) for (u in2 (7,2) mu)-nig (24) InB 118: gi-ri2-na (3 N-T 469) for girin-na (14,9) (25) InB 79: (munus-bi) in-ga-am3-ma (3 N-T 781) for (munus-bi) in-ga-ma (7,3) Sandhi Writings (26) IEb 145: a2-tar-ri-a-ta (CBS 4586) for a2-ta ri-a-ta (10,8) (27) Nu 1: ug 3-g ir3-e 3 (CBS 3424b) for 3(UN) erim2(NE.RU)-e3 (7,5) ug
when copying directly from another source. Even though the errors are qualitatively identical, their underlying causes are completely different and should be distinguished when evaluating how mistakes were caused.

Memorization and Transmission of Sumerian Literary Compositions F203

Mechanical Errors The last group of variants relevant to this discussion comprises mechanical errors. These include dittography, haplography, metathesis, and parablepsis, as well as simple omissions, additions, or transpositions of single signs or lexemes. Although mechanical errors of these types are more familiar to the text criticism of biblical and classical manuscripts as mistakes that can occur in sources copied directly from another exemplar, their frequent occurrence in sources which also contain variants that are more diagnostic of memory errors (e.g., conspicuous omissions, anticipations, or preservations) demonstrates that they could have also occurred during copying from memory. One example of a source that contains both mechanical and memory errors is 3 N-T 728, a fourcolumn tablet from House F containing the entire composition of IEb. This source, which is in the Oriental Institute in Chicago, contains the writing rin-gi for gi-rin, a mechanical error involving the reversal of two signs, called metathesis, frequently attested in sources that were copied visually. The presence of numerous conspicuous omissions, like the writings ba-ti for ba-an-i-in-ti, mu-si-le for mu-un-si-il-le, the omission of the form an-ki-a, and the entire verbal chain before the verb gur5-gu2, however, show clearly that this source was copied from memory, and that the reversal of the signs gi and rin must also have been the result of a memory error. The same holds true for errors of dittography and haplography, which are common in sources copied directly from another exemplar, but are also attested in sources for compositions that contain diagnostic memory errors. Examples of mechanical errors include: Dittography (28) ErH 64: nundum-nundum (3 N-T 532) for nundum bur2-re (12,7) (29) GH 130: dlugal-ban3-ban3-da (3 N-T 777+3 N-T 778) for dlugal-ban3-da (11,7) Haplography (30) A 73: <kaskal> danna (3 N-T 826) for kaskal danna(KASKAL.BU) (12,6) (31) EnA 68: dur-an-ki <ki> (... tag) (HS 1576) for dur-an-ki ki (... tag) (6,4) Metathesis (32) IEb 55: rin-gi (3 N-T 728) for gi-rin (4,2)

(33) InB 84: an-na(-ku2-u3-de3-en) (UM 2915-422+CBS 7847) for na-an(-ku2-u3-de3-en) (12,9) Parablepsis (34) A 75: nibru<ki uri5>ki-ma (3 N-T 826) for nibruki uri5(E.AB)ki-ma (12,5) (35) IEb 83: e11<da-gin7> da-ga (3 N-T 577+3 N-T 440) for e11-da-gin7 da-ga (7,7) Omissions
d

(36) InB 115: da-<nun>-na (3 N-T 721) for a-nun-na (12,7) (37) A 39: a-ma-ab-<du11> (3 N-T 927, 525) for a-ma-ab-du11 (7,3)

Additions (38) InB 70: u4-di-de3 ba-te (3 N-T 917, 393) for u4-de3 ba-te(-en) (15,8) (39) GH 129: (u ki-a) li-bi2-in-um2 (3 N-T 777+3 N-T 778) for (u ki-a) bi2-in-um2 (5,4) Substitutions (40) InB 41: (dumu gal) dsuen- ra (CBS 12594) for (dumu gal) dsuen-na (9,3) (genitival construction; no dative) (41) KH 116: (su3-sa4) mi-ni-g ar (3 N-T 478) for (su3-sa4) mi-ni-ib-za (8,3) The Evidence for Memory Errors in Individual Sources The strongest evidence that Sumerian literary compositions were often copied from memory is the occurrence of numerous memory errors of more than one type within individual sources. While nearly all of the most common types of memory errors that occur in other text corpora and in experiments conducted by cognitive psychologists are also attested in the copies of Sumerian literary works, this in itself is not sufficient proof that the copies of these texts were necessarily copied from memory. As noted above, some classes of memory errors also occur when duplicates are produced by a different means, such as copying from dictation or copying directly from another exemplar. This is especially evident with phonetic errors which are just as likely to result from scribes mishearing a word or form being recited to them as it is when they are recalling compositions from memory, or from visual errors caused by scribes visually confusing graphically similar

204 F Journal of Near Eastern Studies

signs when copying texts directly from other tablets. Conversely, many of the mechanical errors (e.g., dittography, haplography, metathesis, and parablepsis) are errors typically associated with direct copying. There are, however, some types of mistakes more diagnostic of memory errors than others. McGillivrays observation that memorial transfer errors are the most reliable indication that a source was copied from memory is almost certainly correct. It is difficult to envision that all of the forms of memorial transfer together, such as anticipations, preservations, and borrowings, were caused by any other means than by copying from memory. A scribe copying directly from dictation or from another exemplar would have been more likely to confuse a word or form with a word or form that was similar to one he or she had just seen or heard than to substitute it for one that occurred in a similar context earlier or later in the same text. Additionally, omissions of entire words, phrases, and even lines are less likely to have occurred if the scribe had access to written or aural duplicates of the composition than if he or she had to rely entirely on memory to reconstruct the content of the text. The same holds true of conspicuous additions, which are difficult to explain as mistakes of the eye or ear that occurred immediately after seeing or hearing the text. Even more diagnostic of copying from memory than the extent of such errors, however, is the occurrence of large numbers of different types of memory errors within a single source. Unless a scribe is particularly careless and unskilled, the number of errors in a source copied directly from another written duplicate is generally small. Even the most attentive scribe would occasionally omit a word or form, or confuse graphically similar forms in the moment between looking at a source text and copying what has just been seen onto another tablet or manuscript. But it is rarer that a competent scribe would make numerous minor and major mistakes when copying from a clean and clearly written duplicate than he or she would from memory, since the scribe could compare the new copy with the other source to correct or avoid making mistakes. For different reasons, sources copied from dictation by inexperienced or careless scribes who did not know or understand the texts are more likely to contain numerous phonetic errors than errors of the types that occur primarily in duplicates copied visually or from memory. As a result, if a source contains numerous errors of different typesincluding diagnostic memory errors, memorial transfer errors and

significant omissionsin addition to phonetic errors, visualizing errors, and/or other mistakes that can also occur in sources copied from dictation or from another exemplar, the probability is higher that the source was copied from memory. One of the many examples of a source with a distribution of errors corresponding to the number and types of mistakes expected in a duplicate copied from memory is 3 N-T 577 + 3 N-T 440 (= NI7). The text is a well-preserved four-column tablet from House F at Nippur, which originally contained the entire composition of IEb. There are over seventy omissions in this source. In more than half of these instances, a single sign representing one or more grammatical elements is omitted. These include: the omission of (-)a(-) in the forms ma-za for ma-a-za (line 11), te-me-en for te-a-me-en (lines 2931), u for u-a (line 54), an-ki for an-ki-a (lines 66 and 88), a2-ta-ri-ta for a2-ta-ri-a-ta (line 145), and e3-za for e3-a-za (line 12); the omission of -(C)e in the forms kur for kur-re (lines 2930), ebiki for ebiki-e (line 140), ebiki for ebiki-ke4 (line 31), and gaz for gaz-e (line 73); and the omission of other elements in the forms ga-mifor ga-am3-mi- (lines 36 and 94), an-na for anna-ka (line 74), an-na-e3 for an-na-ka-e3 (line 85), u6-di for u6-di-de3 (line 121), gid2-da for gid2-da-bi (line 84), a2-ma for a2-ma-za (line 161), and kur-ku for na-kur-ku (line 165). Furthermore, there are twenty-one verbal forms in which the pre-verbal elements /b/ and /n/ were omitted.90 In addition to the omission of grammatical elements, however, there are also more significant omissions, including: the semantic omissions nu-za for nu-e-ga(-za) (line 9), giBU for gimu-bu-um-gin7 (line 70), na4 for na4-su (line 143), a3-tur3 for mu a3-tur3 i-an-dul3 (line (line 145), and the omission of g 122); the omission of two or more elements in the forms ga-zu for ga-am3-mi-ib-zu (line 35), e2-em-um2 for e2-em-mi-in-um2 (line 78), ga-bad for gaan-i(-in)-bad (line 81), ki-dar-ba for ki-in-dar-

The element /b/ was omitted in the verbal forms in lines 90, 95, 96, and 98; and /n/ was omitted in lines 6, 65, 79, 80, 82, 83, 84 (twice), 107, 108, 122, 141, 148, 149, 167, 171, and 174.
90

Memorization and Transmission of Sumerian Literary Compositions F205

ra-gin7 (line 82), and sig3 for sig3-ga-ke4 (line 164); and the omission of lines 27, 38, 104, 106, 175, and 176 entirely. While it is possible that in some cases (such as the omission of the pre-verbal elements /n/ and /b/ and nominal endings -a and -Ce) the omissions were intentional because the presence of the omitted elements could be easily inferred, the number and nature of omissions, many of which are substantial, suggest that they are the result of memory errors. Another indication that most of the omissions within source NI7 were the result of memory errors is the occurrence of large numbers of other types of mistakes. Grammatical variants such as: kur-re for kur-ra (line 84), im-da-an-ri for im in-a-zu for g in-a-za (line 13) da-ri (line 115), g and mu-de2-e for im-ma-de2-e (line 151); and phonetic writings such as: kur-e for kur-re (lines 6 and 149), e-en for KI.E.NE.DI.dMU3 (line 97), us2-a for us2 a2 for kin ga- (line 105), sa (line 103), kin-g I.DU for I.DU8 (line 74), -du-be2 for -dub2be2 (line 82), and im-la for im-la2 (line 120) all suggest that the scribe who copied this source had difficulties understanding the grammar and orthography of the composition, and as a consequence was unable to memorize the composition properly. The scribes inability to learn the content of the text is also clear from the semantic errors gal-zu for dumu-gal (line 22), im-si for im-sa2 (line 121), and ub2-TA?gin7-EN for ub2-dar AK (line 3), and the mechanical in2-na-g u10-ne for nig in2-namemory errors an nig gu10-ne in line 26, caused by an being erroneously carried over from the beginning of the preceding line, and e11 da-ga for e11-da-gin7 da-ga in line 83, where the scribe omitted all of the signs between the two occurrences of da (parablepsis). Since source NI7 is from House F at Nippur, and the literary sources from this house generally contain few substantial mistakes, it is unlikely that variants that occur in this source are synchronic or diachronic. Moreover, the quantity and nature of the variants that occur within the duplicate are consistent with the types of errors made by less advanced or experienced scribes. It is therefore very likely that the omissions and other erroneous writings in this source were the

result of mistakes made by a scribe who had learned the content of the composition poorly. Another example of duplicates of Sumerian literary compositions with a similar distribution of variation is a group of extract sources from Ur containing connected sections of IEb. Sources Ur1, Ur2, Ur5, and Ur7 are all Type III tablets, which are similar in format and contain extracts of approximately the same length from different sections of the text.91 The tablets, which are relatively small, landscape-shaped tablets with a longer width than length, each contain between thirty and thirty-four lines and probably belong to a series of sources with different sections of the entire composition. Since all four sources contain nearly the same number of lines and Ur 1 (which contains the first thirty-four lines of the text) ends with the same line with which Ur2 (which contains lines 34 to 65) begins, there must have been two additional tablets between Ur5 (which has lines 95 to 119) and Ur7 (which has the last thirty-two lines, lines 150181) containing lines 65 to 95 and 119 to 150. It is likely that this group of sources was copied during the same period of time by the same scribe. Like the series of tablets containing IEb, many of the literary sources from Ur are also groups of Type III extract tablets that had originally connected sections of entire compositions.92 Since most of the Type III sources from Ur with a secure provenience come from a private house known as No. 1 Broad Street, and since some of these tablets have colophons indicating that they were copied by Damqi-iliu, it is not unlikely that the group of sources with IEb was also written by this scribe. Each of the sources in this group also contain numerous phonetic variants, including the writings me for me3, ur5 for ur3, and du-du for du7-du7 in lines 3, 48, and 68 of Ur1; gu3 for gu2 in lines 45 and 49 of Ur2; -su for the verb -su3 (Ur2 line 47 and Ur5 line 106); and in for a-nig in (Ur2 line 46 and Ur5 line 105). a2-nig
91 Ur1 = UET 6/1: 12; Ur2 = UET 6/1: 13; Ur5 = UET 6/1: 15; and Ur7 = UET 6/1: 17 . 92 Examples include three tablets containing extracts of the composition Ewe and Wheat = ETCSL no. 5.3.2 (UET 6/1: 33, 34; and 35), and additional groups of sources for The Ur Lament = ETCSL no. 2.2.2 (UET 6/2: 136, 137, U16900N, 138, and 139), The Lamentation over the Destruction of Sumer and Ur = ETCSL no. 2.2.3 (UET 6/2: 124, 128, 129, 131, 132, and 133), and Bird and Fish = ETCSL no. 5.3.5 (UET 6/1: 38 and UET 6/1, 40 + UET 6/3, 617).

206 F Journal of Near Eastern Studies

But in addition to these phonetic writings, there are also a large number of grammatical variants, including mud-bi for mud- and an-ne2 for an-na in lines 2 and 13 of Ur1; ga-ba-ni-ib2-si-sa2 for ga-ba-ab-sa2 and mu-ni-ingub for bi2-in-gub in lines 39 and 59 of Ur2; ga-ba-i-in-gub for ga-ba-i-ib-gub and ebi2-bi for ebi2-gin7 in lines 9695 and 100 of Ur5; and mu-un-na-de2-e for im-ma-de2-e and nu-mu-unra-ab-ur3-ra-zu-e3 for nu-ur3-ra-zu-e3 in lines 151 and 157 of Ur7. These variants are unlikely to have resulted from mistakes made while copying from dictation. Since there are also numerous omissions in all of the sources in this group, including ga-ba-su for ga-ba-ni-ib-su3 and e-ka for e-erka-an in lines 47 and 54 of Ur2; kur for kur-re in lines 112 and 113 of Ur5; and u10 in line 169 of Ur7 mir for mir-g it is more probable that these duplicates were copied from memory, and that the phonetic variants were caused by a scribe who had difficulties remembering the orthography of the composition, incorrectly spelling certain forms with phonologically similar or identical writings of the correct forms. The occurrence of memorial transfer errors such as na-ma-ra-an-te-a-gin7 for na-ma-ab-te-a-gin7 in line 3334 of Ur1, resulting from the preservation of the infix -ra- in the verbal chain na-ma-ra-abAK-gin7 in line 32, and the form mu-un-na-an-ab-be2 for mu-na-ab-be2 in line 60 of Ur2, resulting from the preservation of the infix -an- of the verbal form mu-un-na-an-gub preceding it in the same line are further evidence that this group of sources was copied from memory. The pattern of variants most directly characteristic of sources copied from memory is not confined to the sources described in this section. Nearly all of the duplicates for the compositions in the Decad, whenever they are sufficiently preserved to contain a representative number of variants, have a distribution of variation that is consistent with the pattern most closely associated with sources copied from memory. Of the more than seven hundred sources for this group of texts, fewer than ten apparent exceptions

could be identified. One of these sources is NBC 7799 (= X3), an almost perfectly preserved four-column tablet of unknown provenience containing the entire composition of KH. Thirty of the variants in this source (approximately one third) are phonetic writings.93 Moreover, in addition to the phonetic variants, over half of the remaining variants in the source are omissions.94 The quantity of syllabic writings, together with the absence of numerous occurrences of the types of grammatical and semantic mistakes characteristic of sources copied from memory, may indicate that X3 for KH, among all the duplicates for the compositions in the Decad, is one of the few examples copied from dictation. Additionally, CBS 4586 + Ni 4199 (= NI4), a four-column source for IEb from the early seasons of excavation at Nippur, contains a similar distribution of phonetic writings, and could also have been copied from dictation. Finally, there is evidence that AO 9067 (= X3) and AO 6714 (= X4), two extract sources of unknown provenience for ErH, were copied from another exemplar, and not from dictation or memory. There are at least two significant indications that the two sources are directly related, and may even have been produced by the same scribe. One indication is that X3, which contains lines 182 of the composition, ends with the same line as X4, which contains exactly half the number of lines as this source (lines 4182). The other indication that the sources are related is that the lines which overlap in X3 and X4 share 26 variants, 21 of which are unique to these two sources. These include the consistent writing of den-ki-ka in place of d en-ki-ke4 (in lines 58, 66, and 70), the semantic variant ni2 ma for i7 ma (in line 59), the grammatical variant nam-ma-gub for am3-ma-gub (line 78), the
These include the syllabic writings sig7-a for sig7-ga (l. 85), du3-du3 for du7-du7 (l. 60), du3 for du10 (ll. 117118), de2 for de6 (l. 107), tug2-ba for tug2-ba13(ME) (l. 108), an-e for an-ne2 (l. 37), ar-sur-ra for gial-g ar-sur9-ra (l. 116); the phonetically and gial-g written logograms and diri-compounds a-ra-ta for arattaki (lines 1314) and a-lim for alim(ANE) (1. 46); and the sandhi writings men-na-da for men an-da (l. 66), a3-gi16 for a3 ki (l. 76), tu-daKIN for tu A.KIN (l. 110), i-in-ga-nu-tu and i-in-ga-nu-u3-tu 3-g al2 for for i-in-ga-an-u3-tu (ll. 19, 54, 70, and 122), and ug 3(UN) gal (l. 92). ug 94 -ur-e-ne for ur-sag The omissions include the forms ur-sag -e-ne (l.59), un- for mu-un- (l. 60), ama-la for ama gal-la ur-sag (l. 77), nam for na-nam (l. 52), im?-il2-il2 for mi-ni-ib-il2-il2(-i) (l. 4), du10 for e2 du10 (l. 23), den-lil2 for den-lil2-le (l. 37), du7-du7 for du7-du7-dam (l. 61), uumgal for uumgal-am3 (l. 76), ensi2 for ensi2-ke4 (l. 78), an-da-an-ti for am3-da-an-ti (l. 80), su3 for su3-ud (l. 74), and ka for ka-a (ll. 119120).
93

Memorization and Transmission of Sumerian Literary Compositions F207

al2 for mu-un-da-g al2 (line 80), omission mu-da-g and the omission of lines 49 and 54. 95 Since both X3 and X4 for ErH share a relatively large number of variants, many of which are unique to these two sources, but contain very few variants that occur in one of the two sources but not the other, and very few conspicuous errors of the type typically found in duplicates copied from memory, they were probably copied from another exemplar which contained these variant forms. The presence of possible exceptions does not detract, however, from the amount and nature of the evidence demonstrating that most of the preserved sources for Sumerian literary compositions were copied from memory. The exceptions are just thatexceptionsand since all but one of these sources are without clear provenience, and may have been written at a later date, it is possible that these were copied at a place and time in which scribes were not being trained to reproduce the compositions they were learning from memory. In the majority of preserved sources for the Decad, it can be deduced from the patterns of variation that scribes were typically trained by producing written copies of the texts they had learned from memory. Conclusion Proceeding from the widespread assumption that all or most of the preserved duplicates of Sumerian literary compositions are scribal exercises produced by apprentice scribes, the question of how these sources were copied has been addressed. By analyzing the types of errors that occur in experiments conducted by cognitive psychologists on verbatim memory and different forms of forgetting, and confirming that similar mistakes are also attested in copies of texts from other corpora that were also produced from memory, such as Middle English romances and quartos of Shakespeare plays, a list of the most common memory errors was compiled. This list comprises the following errors (in the order of most to least diagnostic): 1) Memorial transfer errors (anticipations, preservations, inversions, and borrowings) 2) Conspicuous omissions and additions
95 For a complete list of the variants shared by these two sources see Delnero, Variation in Sumerian Literary Compositions, 1510 15.

3) Substitutions (semantic, grammatical, and orthographic) 4) Phonetic errors 5) Visualizing errors 6) Mechanical errors (dittography, haplography, metathesis, parablepsis, minor omissions, additions, and substitutions) The means by which a source was copied is most evident not in the occurrence of isolated variants, which could be indicative of any one of several different types of copying, but instead in the distribution of different variant types within a source. Each method of copying results in distinct patterns of variation. Sources copied from dictation are likely to contain a relatively large number of incorrectly spelled, phonetic writings of the wordsforms the scribe was hearingbut fewer nonaural variants, such as grammatical mistakes, semantic substitutions, conspicuous omissions and additions, and transfer errors. By contrast, sources copied directly from another exemplar will contain more mechanical errors, visual mistakes,96 and minor omissions, but fewer mistakes in general, especially of more conspicuous types. Duplicates copied from memory, on the other hand, can be expected to contain patterns of variants similar to NI7 and Ur1, Ur2, Ur5, and Ur7 for IEb, discussed in the previous section. As these sources show, sources copied from memory contain not only memorial transfer errors and conspicuous omissionsthe two types of errors that are most diagnostic of memory errors but also numerous variants of other types, including phonetic writings, mechanical errors, and visual(izing) mistakes, all of which can also occur in sources copied from dictation or direct copying. The occurrence of more diagnostic memory errors, such as memorial transfer errors and conspicuous omissions/additions, together with numerous mistakes of all or some combination of the remaining types of mistakes within nearly all of the individual duplicates for the compositions in the Decad serves as evidence that these sources were copied from memory, and not by another means, such as by dictation or direct copying, for which different patterns of variants would be expected. The number of sources with the types and distribution of mistakes characteristic of memory errors strongly suggests that scribes first
96 For the difference between visual and visualizing errors see fn. 89 above.

208 F Journal of Near Eastern Studies

learned the texts they were taught by memorizing the compositions, and then produced written duplicates of those texts from memory. Although others have suggested that memorization played a critical role in the transmission of literary texts, this claim has always been made without citing any evidence that supports it. It has been the goal of this study to provide such

evidence. The role of memorization in the copying of literary compositions has critical implications for understanding how scribal education was conducted in ancient Mesopotamia and also for how the variants in the copies of Sumerian literary works are to be interpreted, but these are topics for another study.

Das könnte Ihnen auch gefallen