The problems of the transmission of the Avesta and the tools for Avestan text criticism I thank you very much the organizers of the conference for allowing me to speak here, especially because I m not going to show you a brilliant solution, but just to tell you my problems and the poor attempts I have made for their solution and to ask you for help. Allow me then beginning with a presentation with the specific problems of the Avestan transmission. The Avestan texts were composed somewhere in Eastern Iran probably between the second half of the 2 nd millennium b. C.E and the arrival of Alexander the Great in Iran. The invention of the Avesta script is dated much later, around the 6 th century C.E. or even later. The extant manuscripts are even later, however. Colophons mention manuscripts dated in the 10 th century, but our oldest manuscripts belong to the second half of the 13 th and 14 th century. They do not become frequent till the 17 th century. The manuscripts contain the transcription of the recitatives of several Zoroastrian rituals. [Dia2] There is a great typological variety of manuscripts depending on their contents, function and their place of publication. According to the type of rituals the Avestan manuscripts can be divided in two different types: 1. Manuscripts of the long liturgy that contain the Zoroastrian long liturgy in its different variants 2. Lectionaries or manuscripts including collections of ceremonies that can be celebrated outside the Fire Temple and by privates or priests of a lower degree. According to their function the manuscripts can be classified as: 1. Liturgical manuscripts that include not only the Avestan recitative of the long liturgy but also ritual directions in different languages depending on the time and place of the copy. 2. Exegetical manuscripts in which the Avestan recitative of the liturgy is translated in Pahlavi or Sanskrit. 3. Mix-manuscripts that include as well the ritual direction as the Pahlavi translation of the Avestan recitative Traditionally, it has been supposed that the liturgical manuscripts were dependent on the exegetical manuscripts with Pahlavi
translation. Today we believe that the situation is differethis view has changed. Descriptions of the rituals including ritual directions and Avestan recitative existed already in Sasanian times and were originally transmitted orally. This is, indeed, the ultimate source of our ritual manuscripts and the exegetical manuscripts derive from the liturgical ones via the mix-manuscripts. After the islamization of Iran, an important part of the Zoroastrian community migrated to Gujarat. Although the contacts between both communities of Zoroastrians have always been frequent, we can speak from two separated communities with independent transmissions. The last analysis of the Avestan transmission based on the manuscripts was done 125 years ago by the German scholar K. F. Geldner who is responsible of the last complete edition of the Avesta (Geldner 1886). Every analysis of the Avestan transmission done afterwards has been based on the data provided by Geldner. Several years ago, I decided to change this situation and started the Avestan Digital Archive (http://www.avesta-archive.com) [Dia3]. It aims digitizing, indexing and publishing on-line all available manuscripts of the Avesta. Since them we have located more than 300 manuscripts, digitized around 120 and indexed and published on-line around 35. Geldner had access to around 135 manuscripts. Through the direct work with manuscripts many common ideas about the transmission of the Avest have proved to be false. In fact, we have reasons to repeat the work done by Geldner in almost every steps of the editorial work. Today I m going to describe just the problems posed by Geldner s stemmatological analysis and to introduce to you the tools I have developed for trying another approach to this problem. [Dia4] The traditional view of the transmission of the Avesta recognizes three steps from the first writing down of the Avesta to the extant manuscripts: 1. the Sasanian redaction around the 6 th c.; 2. the Yazd-original or the hyparchetypes proposed by Hoffmann and Humbach around the 9 th or 10 th c.; and last, the manuscripts (from the 13 th c. on). Kellens has changed our perspective. He questioned the derivation of the manuscripts from the Sasanian archetype, and led them go back to a parallel ritual collection. However, the existence of the hyparchetypes has not been put into question till my talk in a conference celebrated in Salamanca, September 2009. The leading principle of the archetypes theory is the leading principle of Geldner s analysis of the manuscripts: the Lachmannian principle of
agreement in error. Beside the general criticisms of this method, there is a special feature of the Avestan transmission that makes it especially inappropriate for the analysis of the Avestan manuscripts. If my criticism results correct, then the evidence of the hyparchetypes will disappear and a new complete analysis of the relation between the Avestan manuscripts will be necessary. As mentioned, the Avestan manuscripts contain basically the recitatives of a series of liturgies that have been celebrated from the Sasanian times on and that are celebrated partly still today. Their use in the proper ceremonies was not allowed, but was mainly restricted to the priestly schools for teaching priests the right performance. The recitative of the rituals was known by heart and manuscripts were used only as a tool for the instruction of priests. This fact has conditioned enormously the way the manuscripts were produced. Geldner (1886, 1. xlvii) admitted, indeed, the influence of the ritual in the manuscripts [: The copyists knew the majority of their text by heart. The oral text, however, had become more corrupted than the written text and keeps constantly crossing the latter. Scribes who read and copied word for word from the text before them, ran less danger than those scribes who grasped the entire sentence and wrote it off before looking at the copy before them. But in fact, the mutual influence of the ritual performance and the production of manuscripts is much deeper. Ritual practice and copying of manuscripts are linked processes and influenced each other. In the production of the manuscripts are involved as well the process of copying from a written source as the influence of the ritual practice. Each manuscript moves between these two poles. Some manuscripts show, indeed, a clear influence of the oral-ritual text. In fact, some of them seem even to be copied not from a written source, but from an oral dictation or as a transcription of the ritual practice. [Dia 6] A clear example is the oldest known Yasna Sāde, B3, copied in the second half of the 16th century. Let us compare, for instance, Y1.3 in three different liturgical manuscripts: B3 (230) niuuaēδaīēme. asnīaēibiiō. ašahe. ratubiiō. hāuuanēašaoni. ašahe. raθβe. niuuaēδaīēme.. šāuuaŋh e. vī iiāeca. ašaoni. a ahe. raθβe. niuuaēδaīēme. miθrahe. vōuru.gaoiiaoitōiš. hazaŋhra.gaošahe.
baēuuar casmanō. aōxtō.nām[anō. y]azatahe. rāmanō. x v āštr[ahe.] K11 (110) niuuaeiδaiieimi. h āraiiemi. a niiaeibiiō. ašahe. ratubiiō. hāuuan ē. a aoni. ašahe. raθβe. niuuae. āuuaŋh ē. vī iiāi a. a aone. ašahe. raθβe. niuuae. miθrahe. vouru.gaōiiaōitōiš. hazaŋhara.gaōšahe. baeuuar. ašmanō. aoxtō.nāmanō. yazatahe. rāmanō. x ā trahe. Ave977 1 (4010) iuuaēδaiiemi. ha āraiiemi. a niiaēibiiō. a ahe. ratubiiō. [ ] hāuuan e. a aōne. a ahe. raθβe. iuuaēδaiiemi. ha āraiiemi. āuuaŋh e. vī iiāi a. a aōne. a ahe. raθβe. iuuaēδaiiemi. ha āraiiemi. miθrahe. vouru.gaōiiaōitōiš. hazaŋhrō.gaōšahe. baēuuar. ašmanō. aōxtō.nāmanō. yazatahe. rāmanō. x v ā trahe. The differences are obvious. Influence of the oral recitation is marked in red and it is obvious that it is much stronger in B3 than in the other 2. Actually, B3 might have been written down even without assistance of a written copy. Against the theory of one Sasanian archetype and of several hyparchetypes, it is likely that these rituals have been written down at different times and in different places. Furthermore, the influence of ritual practice is not limited to phonetic deviations from the old copies. Liturgical manuscripts do not pretend to be accurate copies of old originals, but authoritative guides for the right performance of the ritual. Accordingly, conscious changes introduced in a priestly school are reproduced, of course, in the manuscripts copied under the influence of this school disregarding the text found in the original written source of the copyist. The mayor or minor generalization of these new readings depends on the influence of the priestly school over other schools and priestly families. Thus, [Dia 7] in Y 30.1 the change of mazdāϑā into mazdā.ϑßā in all IndVS but B2 is obviously result of an analysis of the transmitted form as mazdā and the personal pronoun ϑßā and not a simple transmission error. It is a conscious decision. This new reading became, nevertheless, almost universal in India. It appears in all the liturgical manuscripts and even in some of the late exegetical 1 In the VS manuscripts this passage does not appear as such, but I have taken the first sentence from VS1.2 and the second from VS1.3.
manuscripts like F2 and T6. In M1 the copyist has written first the variant he knew from the ritual, mazdā.ϑßā, and then corrected it with the form that appeared in his original, mazdāϑā. Among the liturgical manuscripts only B2, the oldest Indian liturgical Videvdad (1626), retained the original mazdāϑā. It does not suppose, nevertheless, that all manuscripts but B2 are genealogically related. This variant became just trendy after 1626, date of B2. The intricate relationship between written copies and ritual practice, between unconscious mistakes in the written transmission and conscious corrections can be seen in following example [Dia 8]. In V3.41 the liturgical manuscripts appear divided in three groups and none of them corresponds with the text of the exegetical manuscripts: Original text spaiieti draošəm spaiieiti yātuγnīm aiieiti a auuaγnīm Exegetical manuscripts spaiieti draošəm spaiieiti a auuaγnīm A (B2, T46, L1, P1) spaiieti draošəm spaiieiti yātuγnīm aiieiti a auuaγnīm Liturgical manuscripts B (O2, L5, G106) spaiieti draošəm auuaγnīm spaiieiti yātuγnīm aiieiti a auuaγnīm C (L2, E4, B4, Ml630, FIRES1, G112 2 ) spaiieti draošəm spaiieiti auuaγnīm spaiieiti yātuγnīm aiieiti a auuaγnīm The origin of these variants is clear. The exegetical manuscripts attest a saut-du-même-au-même. Furthermore, in an unknown liturgical manuscript with the division of lines as shown here, one line was skipped: daēna. māzdaiia niš. narš. ā tauua nahe. baṇdəm. spaiieite. draošəm. aiieite. yātuγnīm. aiieiti. aš 2 Formerly (Cantera 2010) known as G42.
auuaγnīm. aiieite. na u aēm. The copyist wrote the beginning of the new line (auuaγnīm), than noticed his error, marked auuaγnīm with deletion dots and then continued writing the right text: daēna. māzdaiia niš. narš. ā tauua nahe. baṇdəm. spaiieite. draošəm. auuaγnīm. aiieite. yātuγnīm. aiieiti. ašauuaγnīm. spaiieite. nasu aēm. A later copyist did not notice the presence of the deletion dots and thus the text B arose. Then the pseudo-philological reflection of a school introduces spaiieiti before auuaγnīm in order to make this incorrect text understandable. That happens in the 17 th or 18 th century and from now on this variant becomes almost universal. In the reformist school of Nawsari this variant spread even into the exegetical manuscripts of the family of L4 and even a new Pahlavi translation is created for this invented text. In this fortunate case, the textual evidence allows us to distinguish between the variant of the "written" transmission that spreads into other manuscripts only by way of copying, and the variant created consciously by a school that contributed to its generalization through manuscripts and through priestly teaching and reciting. The new created variants (or at least some of them) were integrated in all manuscripts created in this school and, of course, in its ritual praxis. Errors in the written transmission entered the ritual performance and changes in the ritual performance entered the manuscripts. And if this school was authoritative enough, they spread to other schools. Hence, the manuscripts sharing this reading do not have necessarily any genetic relationship: they share just a more or less trendy variant within the Zoroastrian ritual community. In this context the textual homogeneity of the Avestan manuscripts can be, in fact, the reflection of a ritual homogeneity within the Zoroastrian community. Traveling priests and traveling manuscripts can be the responsible of such uniformity. This has important consequences for our view of the Avestan transmission. Regarding the prehistory of the written transmission, it is possible that the common readings that are adduced as proofs of the
different archetypes (only one in the case of the long liturgy, for example) can be, in fact, variants that have entered the ritual practice and have spread over wide areas through the influence of certain priestly schools. Hence, we would lack evidence for postulating such unlikely hyparchetypes for manuscripts of very different text types. Besides, Geldner s methodology is not suitable for the analysis of the relations between the manuscripts. If common errors can spread not only through the process of copy, but as well through the influence of the ritual practice, then the lists of 10 to 20 common errors adduced by Geldner as proof of a genetic relation cannot be the basis for establishing genetic relations between the manuscripts. Consequently, I have tried to find a method that can give a better answer to the problems posed by the Avestan transmission. Significant common errors cannot be the only basis for the analysis, since they can reflect not only a genetic relationship, but as well the belonging to the same sphere of influence of the ritual practice. Thus univocal relations are not adequate for the representation of the complex process of copy of the Avestan manuscripts. The ritual practice conditions the shape of a manuscript as well as its written source. A manuscript can be copied from an Iranian original in Nawsari, India. The resulting manuscript would share features with the Iranian manuscripts of the same text type, but also with other manuscripts copied in the same priestly school. After a review of the different approaches to the analysis of the genetic relations between manuscripts, I found the Coherence Based Genealogical Method developed at the Institut für Neutestamentliche Textforschung at the University of Münster in the cntext of the Editio Graeca Maior as the most appropriate model for the analysis for the relationship between Avestan manuscripts. Decisive was for me the fact that it combines the advantages of computational analyses with philological evaluation of the data. In the following I m not going to describe in detail the Coherence Based Genealogical Method and its methodological background 3, but simply to introduce briefly the Tools 3 There is an abundant bibliography on this method (Mink 2000; 2003; 2004; Wachtel 2008). Extensive online information is to be found at the page of the Institut für Neutestamentliche
that have been prepared on this basis for the analysis of the Avestan manuscripts. I have developed in the context of the Avestan Digital Archive a set of tools that I call TATEC (Tools for Avestan Text Criticism). We create Excel datasheets with transcriptions of the different manuscripts [Dia 9]: These datasheets are the basis for all the future comparison between the manuscripts and for a manual philological evaluation of each place of variation. The Tools present to the researcher the different readings of each place of variation and the researcher might evaluate the relevance of each reading and, if possible, the dependence of each reading in order to build local stemmata or sub-stemmata which are the base of the genealogical comparison in the Coherence Based Genealogical Method. Here is a sample of the interface for evaluating data [Dia 10]: Textforschung of Westfälische Wilhelms-Universität Münster (http://egora.unimuenster.de/intf/projekte/gsm_en.shtml).
We distinguish four different degrees of relevance: - r = reading is just a orthographic variation of another attested reading. Thus aetaδa is just a reading of aētaδa and is encoded accordingly as r-b that means that aetaδa is just a reading of (b) aētaδa. - 1 are banal variants of single letters, such as the usual variation between ī and ii or the confusion between ar, r, r, etc. - 2 are haplologies, dittographies, methatheses, accumulations of changes of level 1, etc. - 3 is reserved for aberrant variants that are quite unlikely to arise independently. The information of the dependency is encoded as well and stored in a database as you can see in following picture [Dia 11]:
The information encoded can be read as a local stemma in which var daθ mca is the original reading: (a) varədaϑəmca varədaϑəmca K1,M3,B1,D62,P10,P2 varədasəmca L4,T44,E10,G25 varədsīmca The information stored in the databases allows us to make queries about the index of agreement of the different manuscripts. They show a pure quantitative index of agreement between the different manuscripts, but also the number of agreements in readings that have been evaluated according to their relevance [Dia12]: G34
The utility of this tool is obvious. The manuscripts of the lines 3 to 13 belong to the same text type as K1, the manuscript that serves as basis for the comparison. All manuscripts of this type derive from two copies, L4 and K1 copied by the same writer, Mihrabān Kayxōsrow, in the years 1323 and 1324. The manuscripts that share a higher index of agreement with K1 than with L4 are copies from K1 (that is, B1, M3 and P10), the rest from L4. This is confirmed by the index of agreement of L4 [Dia 13]:
As you might see, the index of agreement of L4 with the manuscripts copied from K1 is higher than with its own copies. However, the agreements of level 3 help us to identify some copies of L4. All the numbers of agreement are active links that show the agreements shared by the compared manuscripts. Thus we can see that whereas all the agreements of level 3 of B1 with L4 are common with K1
, the situation is different in the case of the manuscript G34, a copy of L4: Here appears the reading a manō that characterizes all the copies of L4 and furthermore tarasca for jatarasca that is shared by both and T44, another copy of L4. Besides, the highest index of agreement of G34 is with L4, so that G34 is very likely a copy of L4:
This method for grouping manuscripts does not allow knowing the relative chronology of the witnesses. For the introduction of the chronological factor, we use queries about the amount of prior and posterior variants of a manuscript in comparison with the others. The numbers of prior and posterior variants are calculated on the basis of the local-stemmata stored in the database. Thus if we compare the number of prior variants of K1 with the other exegetical manuscripts of the Wīdēwdād and especially with the manuscripts of this group closer to it according to the pre-genealogical index of agreement (B1, M3 and P10) we get following results: - Prior variants of K1:
- posterior variants of K1: The percentages of prior variants of K1 in comparison with B1, M3 and P10 are by far higher than the percentages of the posterior variants, so that we might conclude that K1 is the oldest witness of this group. If we make the same query about M3, the results are less clear, but informative enough: - Prior variants of B1
- Posterior variants of B1 B1 seems to be younger than K1 but older than M3 and P10. The data are apparently less clear in the case of P10, but if we make a query about which are the prior variants of P10 we will discover that the picture is disturbed by the decision of the evaluator that almost all prior variants of P10 are the enclitic pronoun hē that appears in P10 as hē, but in M3 as he (with short e). This information can help the evaluator of the data for changing his former evaluation of these readings. Therefore, I consider very important that the numbers are hyperlinks that allow the researcher to discover potential anomalies and to correct them. The TATEC offers further possibilities like a search for omissions and additions or a tool for advanced collating, but I hope to have shown its most essential features. It is a set of tools that combines the evaluation of each variant by the specialist with quantitative analysis and assists him for the analysis of the relative chronology of each witness on the basis of the local stemmata built by him. The next step is the production of an algorithm that allows determining the different relations of dependency between the manuscripts combining the data of the quantitive analysis and of the local stemmata.
The result will probably offer a more realistic picture of the complex processes of the Avesta transmission than the simplistic stemmata produced by Geldner just on the basis of the agreement in error, but in any case I ll be glad to hear from you about other possible approaches to my problem.