The root texts of the Tibetan grammatical tradition are the Lung ston pa rtsa ba sum cu pa (widely known as the Sum cu pa) and the Rtags kyi ’jugs pa. Footnote 1 These are attributed to the putative inventor of the Tibetan script, Thonmi Sambhoṭa, a possibly legendary councillor of Emperor Khri Srong rtsan, alias Srong brtsan sgam po (c. 605–649). The Sum cu pa establishes, among other things, the norms for how one uses grammatical particles. Tibetan children and beginner language students often recite its verses or those of simpler primers, and this allows them to know, for example, that following a g or ng suffix, one should use the gi form of the genitive particle, and not the kyi or gyi form. As a clear set of widely known standards, these orthographic norms serve one well when editing Classical Tibetan texts. If a seventeenth-century commentarial text, for example, uses kyi where one should find gi, then one can be confident that this is not a variant, but an error. The problem is more difficult, however, when one considers the orthographic standards of Old Tibetan writing. Early Old Tibetan (mid-seventh to mid-eighth century), Middle Old Tibetan (late eighth to mid-ninth century), and Late Old Tibetan writings (late ninth to early twelfth century) do not follow Thonmi's norms. Indeed, texts of each of these three periods have their own orthographic norms that distinguish them.Footnote 2 One obvious reason that many Old Tibetan writings follow norms that differ from the later Tibetan grammatical tradition is that the Sum cu pa and the Rtags kyi ’jugs pa – and even more so the commentarial tradition that draws upon and reinterprets them – date to a later period.Footnote 3 Here Tibetan love for a glorious precedent, and a tendency to ancestralize their traditions, has produced an anomalous account, namely that of a language that has remained the same from its inception to the present. There is no need to debunk this position by pointing out its impossibility, nor to single out any credulous investigators who have too readily imbibed it; this ground is well trodden.Footnote 4 Rather, we must ask ourselves the more productive questions of which norms predominated prior to the emergence of Classical Tibetan or standard Written Tibetan – intentionally leaving these categories undefined – and how one can establish what these were. The norms of the later Tibetan grammatical tradition are convenient as a reference point for such a task, but it is from the evidence of Old Tibetan writings themselves that we must identify the prevailing orthographic standards. To this end, the present article examines officially sponsored sutras in order to uncover the orthographic norms relevant in Dunhuang from the 820s to the 840s, and to point out some of their relationships to earlier and later norms.
Drawing on the 1,650 copies of the Aparimitāyur-nāma mahāyāna-sūtra held in the British Library, one is able to contribute something towards a baseline of the standard orthography of the 820s to 840s. Selecting this particular sutra has many advantages. In the first place it is short and repetitive. This makes it fairly simple to transcribe explicits with which to build a searchable corpus. Second, about 30 per cent of the sutras were edited by rotating teams of three editors, implying that there was a quality control system and at least some concern with the orthography and content of these officially sponsored sutras. Third, the scribes were mostly Chinese who learned Tibetan as a foreign language, and did so, as we see from one interesting marginal note, with recourse to a grammatical treatise. Fourth, the scribes who copied these sutras and the editors who corrected them worked not only on other officially sponsored sutras such as the Chinese Aparimitāyur-nāma mahāyāna-sūtra, Tibetan Śatasāhasrikāprajñāpāramitā-sūtra, Chinese Mahāprajñāpāramitā-sūtra, but also on secular documents. This is evident from their jottings of letters and petitions, and from the character of the texts that they used as patches to repair torn sutras. The orthography represented in these sutra copies, therefore, is not particular to this genre, but represents the scribal practice of those who also wrote many of the secular and official Tibetan documents that lay alongside the sutras in Mogao Cave 17. The results, as one might expect, evince an orthography that is considerably less settled than what we find today. At the same time, one can distinguish some clear orthographic norms, some of which emerge as significant for dating Tibetan writing.
Orthography and the Tibetan grammatical tradition
It is beyond the scope of this article, and would in any case be redundant, to review or summarize the indigenous Tibetan grammatical tradition and to survey secondary scholarship on the evolution of Tibetan grammar. It will be helpful, however, briefly to contextualize the study of Old Tibetan orthography. To be sure, dozens of scholars – nearly everyone who has worked with Old Tibetan manuscripts or inscriptions – have made observations on the orthographies peculiar to the texts that they study. It is commonplace to note the customary use of the ma ya btags with i and e vowels (e.g. myi and mye), the preponderance of da drag (e.g. gsold), medial ’a (e.g. bka's), a’ suffixes (e.g. pa’), reverse gi gu (also known as gi log), alternation between aspirated and unaspirated consonants (e.g. chen pho and chen po), and the absence of a disambiguating wa zur (e.g. dags instead of dvags). In the first phases of scholarship on such writings, it was deemed sufficient to record the presence or absence of such features. This method was largely a result of the fact that a given study might be more focused on the content of a text rather than its form. As a result, orthography was only superficially treated. One consequence of this was (and in some corners still remains) the misapprehension that Old Tibetan orthography is a haphazard free-for-all. This faulty assumption has been extremely valuable to those philologists who have wielded it to grant themselves the remarkable privilege of changing the orthography of words that they do not know until they become words that they do.
Among those who have gone beyond mere documentation to consider the relevance of such forms to an underlying grammatical system, mention should first be made of Jacques Bacot. In his 1928 study of the Sum cu pa, Bacot noted the presence of a few ślokas in this text that describe orthographic norms different from those of standard Classical Tibetan. In particular, Bacot observed that śloka 8 prescribed the joining of the terminative particle to the preceding suffix without the insertion of the intersyllabic marker (tsheg), e.g. phyogsu rather than phyogs su. Bacot concluded that this must in fact have reflected the previous norm (Bacot Reference Bacot1928: 14, n. 2). In 1963 Roy Andrew Miller built on Bacot's work in his own analysis of the Sum cu pa. Miller perceived distinct temporal strata in the text, and singled out ślokas 8 and 13 as representing orthographic standards that predate standard Written Tibetan. Regarding the terminative in śloka 8, Miller ratified Bacot's findings, and further related them to Dunhuang manuscripts transcribed by F.W. Thomas. In these Miller encountered some instances that followed the prescriptions in śloka 8, e.g. sharu instead of shar ru, but also found many that adhered to the later norms, e.g. yar ru (Miller Reference Miller1963: 494). Miller made the further suggestion that śloka 8, when shorn of what he argues is an interpolated line, shows no awareness of the form tu (Miller Reference Miller1963: 496; Miller Reference Miller1993: 44–5). Here Miller again looked to Thomas's corpus and to other early texts to conclude that “the later orthographic convention governing tu was not employed in many instances”. Looking to śloka 13, Miller noted that, as in śloka 8, it prescribes adding only a vowel, rather than a sandhi-appropriate consonant separated by a tsheg. In this case it relates to the semi-final particle in forms such as byede rather than byed de. The verse mentions ste as a semi-final particle, but not te, and it does not mention de as a separate syllable (Miller Reference Miller1963: 494–5). (Classical Tibetan norms are te after n, r, l, and s; ste after g, ng, b, m, ’a, or vowel; and de after d.)
Bacot's and Miller's claims about ślokas 8 and 13 of the Sum cu pa reflecting an earlier set of norms can be tested against the transliterated corpus of Old Tibetan documents. As we shall explain below, one must be careful in this case to construct an orthographic baseline from texts that are reliably dated, rather than from those that float in time. For this purpose, I have chosen the officially sponsored Tibetan sutras produced from the 820s to the 840s, and in particular the Aparimitāyur-nāma mahāyāna-sūtra. In studying these sutras, along with the other officially sponsored sutras produced at the same time, I have also turned up some evidence that is relevant to the existence of early Tibetan grammars during this period. In the margin of a discarded folio of a pothī-format Śatasāhasrikāprajñāpāramitā-sūtra produced in Shazhou (Dunhuang), one finds the following jotting: “taking as a model The Thirty Letters, The Treatise That Teaches All Who Would Learn Letters, memorize it and recite it!” (yig slobs kun la stan pa'i mdo/ yi ge sum cu dpe zhag pa/ brtsongs la grus; PT 1382, 196r).Footnote 5 This is remarkable for the fact that it refers to a treatise – short title Yi ge sum cu, long title Yig slobs kun la stan pa'i mdo yi ge sum cu – with a title that recalls the famous Lung ston pa rtsa ba sum cu pa attributed to Thonmi Sambhoṭa. This does not constitute proof that the Sum cu pa existed at this time, but if the jotting dates to the time of the sutra-copying project, then the injunction bears witness to a book of grammar that was probably used, among other things, to train Chinese scribes in Tibetan language. As a didactic grammar of the Tibetan language, this could well be a predecessor to the Sum cu pa. In any case, the norms of this text, if they are at all reflected by the actual writings of the sutra scribes, were, as we shall see, significantly different from those of the Sum cu pa.
Digital humanities and Old Tibetan orthography
With the rise of digital humanities and the work of Old Tibetan Documents Online (OTDO), it is now possible to search a large corpus of reliable transliterations in a matter of seconds.Footnote 6 In many cases one can also easily check transliterations against high-quality digital images of the manuscripts, freely available online through the International Dunhuang Project and other resources.Footnote 7 The expansion of this transliterated corpus, in tandem with simple electronic search tools, has significantly broadened the horizons for the study of early Tibetan orthography. Besides focusing on narrow queries such as the morphology of the terminative and semi-final particles, one can make a quantitative analysis of the orthography of multiple features of Tibetan writing in a given document or group of documents. We have now gone beyond documentation and merely mentioning the presence or absence of certain orthographic features to their systematic documentation and quantification.Footnote 8 As a result, it is possible to determine the orthographic norms of a given text – provided it is of sufficient length – and thereby to gain some certainty about which sorts of editorial interventions are justified, and which are not. One also creates a basis for comparing the orthographic standards of one text with those of another.
Amid the undoubted advantages that digital humanities brings to the study of orthography, there are also some complications. In the first place, the researcher is limited by the extent and the bias of the transliterated corpus, and the manner in which s/he negotiates it. If “Old Tibetan” includes early, middle, and late Old Tibetan texts, and one searches the entirety of a large corpus (e.g. all of OTDO), then one can only make judgements about a range that includes hundreds of years of language use. These judgements will necessarily be quite broad and also quite superficial. Rather, one must study the texts individually or in coherent groups. Among further complications, the corpus is heterogeneous, not only in terms of media, date and location (e.g. central Tibetan inscriptions versus Dunhuang manuscripts), but also as regards genre and authorship. The Bsam yas Inscription (c. 779) and the Dbon zhang Inscription (823), for example, are both official documents issued by the Tibetan court. In terms of orthography, one can generally make the assumption that the forms one finds there were deemed “correct” at their respective times of carving. This is also largely true of official letters and administrative documents from Dunhuang, and copies of bureaucratic records such as the Old Tibetan Annals. On the other hand, the corpus contains Buddhist and non-Buddhist ritual texts written by people of diverse educational backgrounds. Orthography in official documents was not strictly morphemic, but also phonemic in the sense that one finds, to name just one example, alternation between aspirated and unaspirated voiceless consonants (e.g. chen pho, cen po, cen pho, and chen po). Phonemic spellings are generally more common in non-official texts, and one also finds simple errors that result from partial literacy, that is, the writer's distance from the orthographic standards of the day. In a few cases sloppy writing that employs non-standard ductus reinforces the impression of partial literacy.Footnote 9 Moreover, outside of official letters and contracts, most Tibetan Dunhuang manuscripts are undated. This is not to disqualify such writings as suitable objects of study; far from it, as one often learns more about the language by studying non-standard Tibetan writings. On the other hand, if one wishes to establish the “correct” orthography of writing in a given time and place, one must look to texts with high standards of literacy.
We see distinct orthographic changes from the earliest datable central Tibetan inscription, the Sri Pillar (aka Zhol Pillar, north and south faces) of c. 764 to the latest, the Dbon zhang Pillar of 823, so one obviously must periodize orthographic observations in order to avoid anachronism. Not only temporal, but also geographical considerations must be taken into account. If we speak of a “standard Dbon zhang Pillar orthography” for the decades on either side of 823, for example, can we also apply this to Tibetan documents written in far-away Dunhuang during this same period? The official documents written in or near Dunhuang offer a baseline for comparison that is more appropriate in terms of medium and regional provenance, but one in which many of the dates of the official documents, given as they are in the form of a given year of the twelve-year cycle, have an uncertainty of ±12 years. Besides that, these documents are usually short, so the data that one gleans from any one of them is not so indicative as it would be in a longer document. Some Dunhuang manuscripts also bring with them the further complication of being written by scribes whose first language was not Tibetan. The vast majority of the sutra scribes, for example, were Chinese. Some of the phonemic spellings that one finds here may therefore reflect a special situation relevant to the manner in which Tibetan phonetics were heard and then written by mid-ninth-century Chinese speakers in Dunhuang. This is a point that we shall explore below.
The question of “standard” orthography brings us to a genre that has received increased attention in recent years: officially commissioned sutras.Footnote 10 There are other groups of officially commissioned Chinese sutras from Dunhuang, but the group on which I focus were commissioned for the Tibetan emperor Khri Gtsug lde brtsan (r. 815–841) from the 820s to the 840s.Footnote 11 The project involved the production of thousands of copies of the Śatasāhasrikāprajñāpāramitā-sūtra (hereafter, SP) in Tibetan, the Mahāprajñāpāramitā-sūtra (hereafter, MP) in Chinese, and the Aparimitāyur-nāma mahāyāna-sūtra (hereafter, Ap) in Chinese and Tibetan. Some of these sutras were produced in eastern Tibet, but the vast majority were written in the scriptoria of Shazhou, mostly by Chinese scribes. There are several good reasons for focusing on this group of documents. The MP, SP, and the Chinese and Tibetan Ap constitute a large proportion of the Dunhuang manuscripts. There are over a thousand Tibetan Ap copies or fragments of copies in the British Library and a similar number in the Bibliothèque nationale de France.Footnote 12 Around 500 copies are also found in the libraries of the Dunhuang Museum, the Dunhuang Academy, the Jiuquan Museum, Zhangye Museum, and Lanzhou Library, and over 200 are conserved in the Institute of Oriental Manuscripts in St. Petersburg. Many are also kept in Japan, and four are kept in Taiwan.Footnote 13 Apart from their being so numerous, the Ap are ideal for the study of orthography because, unlike the much longer SP, each sutra is written by a single scribe, and is of a manageable, yet significant length (c. 2,500 syllables). The Ap are particularly useful for the study of orthography in that most intact copies contain colophons that list the name of the scribe. One-third of the sutra copies were edited, and the colophons of the edited sutras additionally contain the names of the editors, of which there were customarily three. Given the prestige associated with an officially commissioned sutra-copying project, the close regulation of textual production, and an editorial system that included three editors, one can assume that this material would display a general adherence to orthographic and lexicographic norms. As such, it represents an excellent source for information concerning what constituted an error in the minds of scribes and editors of the 820s to 840s in Dunhuang.
The data that I analyse is drawn from research that Lewis Doney and I carried out in 2013 and 2014, along with some initial work by Dongzhi Duojie. This consists of a study of all of the Tibetan Ap copies kept in the British Library, where 1,345 shelfmarks include copies or fragments of copies of Ap, and 1,208 of these are sub-shelfmarks under the shelfmark ITJ 310. I supplement this with data recorded from selected Ap copies held at the Bibliothèque nationale de France. These, as I shall describe below, are editors' copies that likely served as scribal and editorial exemplars. While the study recorded data regarding the codicology and format of the manuscripts, I focus here mainly on the orthography of the 188 transcribed explicits and, to a lesser extent, the editorial corrections. These transcribed explicits form a searchable sample from which to draw conclusions about orthographic norms. In order to have a fixed point of reference for the corrections throughout each sutra, we noted them not where they appeared in each copy, but on the corresponding line in a transliterated Ap copy. This was PT 3901, which had eight columns, to which I assigned the corresponding first eight letters of the alphabet. Thus a correction that we record in a given Ap copy may, for example, be at “d3”, which means that it was in the passage that appears in the third line of the fourth column of PT 3901. A transliteration of the latter, as well as a presentation of the data on which I draw here, is included in Dotson et al. Reference Dotson and Doneyforthcoming.
The Aparimitāyur-nāma mahāyāna-sūtra and its editing
The state of the Ap manuscripts offers a window into the editorial process. Stein found them in bundles, to which he assigned site numbers. These bundles consist either of unedited sutras in multiple-sutra or single-sutra rolls, or of mostly edited, single-sutra rolls. The production of the latter can be summarized as follows.Footnote 14 Scribes were issued with paper and ink, and the amounts were noted in a record. They copied sutras based on exemplars that the editors provided for them. These exemplars were copied either visually or aurally into large rolls comprising up to 45 sheets of paper and containing up to 15 sutras. Before or during submission to editors, they were “signed” or ascribed by the scribe to himself/herself or to a client scribe, and each sutra in the long roll was separated into an individual roll of one sutra each. A team of three editors then edited the sutras, usually with red ink. This, too, was completed with recourse to comparison with an editorial copy, which was compared (gtugs) with the sutra. In practice, the corrections come mostly from a single hand, and it is likely that the secondary editors looked over and ratified the main editor's work. The colophon gives their names, which are usually written in a single hand, probably that of the main editor. After the editing was completed, a sutra was finalized (gtan la phab), and submitted to a further office. Eventually, the finished products were organized into bundles, accounted for, and, as we know from PT 999, deposited in temples such as Longxing si, Dunhuang's main monastery.
Unlike the folia of pothī-format SP, or the panels of roll-format SP, the Ap very rarely include jottings. The paper, which is generally in good condition, has not been patched with cuttings from other manuscripts. It is also rare to find Ap torn or marked by editors to show that they are rejected and to be discarded. In this they differ from SP folia, which were rejected on account of missing lines and half-lines that editors had to insert. Such folia were replaced with pristine, rewritten folia.Footnote 15 The Ap editorial process, by contrast, appears to tolerate, or even expect, a final product that includes editorial corrections. This may be due to the nature of the text, whose purpose is to be copied, recited, etc. in order to achieve the aims enumerated therein, and which therefore might lend itself to more streamlined mass production (Iwao Reference Iwao, Popova and Yi2012: 103).
Our hope was that the editors would be pedants whose interventions extended to the correction of the sandhi of grammatical particles. In this we were disappointed, if not entirely surprised, that they were nothing of the sort. The overall impression is of an overburdened composition teacher, grading hundreds of essays in one sitting. The editors' primary concern seems to have been the completeness of the sutra. This is evident from the colophon of ITJ 310.1045: “additions and omissions having been corrected, it is finalized” (lhag chad bcos nas gtan phab bo). Indeed, completeness is emphasized in the vocabulary of Tibetan editing: a good copy has neither omission nor addition (chad lhag ma mchis), a dictum that applies on the level of phrases, lines and chapters, but also on the level of prefixes, suffixes, and words.Footnote 16 The ethic for completeness is further evident in the long insertions that one occasionally finds running all the way up the gutter.
As we are principally concerned here with orthography, it is necessary to have some idea of the text's content so that we may contextualize the “errors” that the editors correct, those that they let pass, and those that they introduce. I give here only the briefest of summaries, and refer the reader to published translations.Footnote 17 As a dhāraṇī sutra, the Ap is fairly simple, and it is repetitive. After a standard introduction setting the scene for the Buddha's sermon, the sutra extols the benefits of its dhāraṇī again and again (23 times in our example text) to punctuate statements about the virtues of the sutra, and the different rewards that one gains from reciting, copying, worshipping or causing others to do the same. Among the various benefits are long life, wealth, purification and rebirth in Sukāhvatī. The sutra ends with the gathered crowd rejoicing at the Buddha's teaching, and a statement that the Aparimitāyur-nāma mahāyāna-sūtra is complete. This is followed by a colophon giving the names of the scribe and, when applicable, the names of the editors.
The Ap's emphasis on producing physical copies has played a central role in discussions of the cult of the book in Buddhism.Footnote 18 The doxographical status of this sutra – or dhāraṇī sutra – has also been the object of discussion.Footnote 19 The Dunhuang versions differ from the two versions contained in the Derge Kanjur not only with respect to the title – the latter are the Aparimitāyurjñāna-nāma mahāyāna-sūtra (Tshe dang ye shes dpag tu med pa zhes bya ba theg pa chen po'i mdo) – but also in terms of the respective lengths of their dhāraṇī. In fact, some of the Tibetan Dunhuang Ap use a shorter dhāraṇī and some a longer one, as noted by Akira Fujieda and Daishun Ueyama (Reference Akira and Ueyama1962). The sutras' contents are otherwise identical, save for a few anomalies that shall not be discussed here.
The most challenging aspect of copying this text, from the scribe's perspective, is the repetition. Due to the nature of the text and the recurrence of the dhāraṇī, it is very easy to skip a short passage, either due to eyeskip by the scribe or by a reader giving dictation. The rendering of foreign place names (e.g. Dza’ ta'i tshal for Jetavana) and foreign terms (e.g. ka sha ni ’ga’ for kārṣāpaṇa, an Indian coin) is by comparison only a minor annoyance. The content and vocabulary of the sutra are elementary, apart from two sentences that make use of the contrastive or adversative use of the genitive, and which consistently gave the scribes problems. The second of these two sentences is as follows:
’dI lta ste rgya mtsho chen po bzhI'i chus yongs su gang ba'I thIgs pa re re nas bgrang bar nus gyI tshe dpag du myed pa'i mdo ‘di bsod nams gyI phung po'i tshad ni bgrang bar myI nus so / (Adversative gyI in italics. The corresponding lines in PT 3901 are f11–13.)Footnote 20
“One may be able to count each drop of water that fills the four great oceans, but one cannot count the extent of the heaps of merit of this Aparimitāyuḥ-sūtra.”
The most common error was to end the sentence with a second, redundant, myi nus gyi instead of closing it with myi nus so. In some cases, as in PT 3901, the editors missed this error. Another common mistake was to insert a negation in the first clause, e.g. myi bgrang bar nus gyi, in which case many a dutiful editor struck through the myi. A third, less grammatical but no less common, error in this particular passage was to write rgyal po chen po bzhi (“the four great kings”) instead of rgya mtsho chen po bzhi (“the four great oceans”). The four great kings occur elsewhere in the text and may also be an understandable substitution based on their comparative ubiquity in a Chinese milieu. Indeed a letter or draft letter, PT 1018, instructs the monk Bar ’byor to make beautiful carvings of the four great kings on the wooden boards (glegs shing) that form the “book covers” – possibly for a volume of the Śatasāhasrikāprajñāpāramitā-sūtra. Footnote 21
The editors' deletions and insertions tend to be found in predictable passages each time. The splashes of red are less a thorough correction of each error than a visible sign that the editorial duty has been performed. The most frequent corrections illustrate this well. With the exception of the dhāraṇī, which occurs throughout the sutra, I refer to the line numbers of the transliterated Ap copy PT 3901 to indicate in which passage the corrections are found.
dhāraṇī: de to dhe; insertion of na mo where left off
a7: de na to de nas
a9: stsogs pa to la stsogs pa
a11: bzhugs ’tsho to bzhugs shing ’tsho
b7, c11, e15, f2: gang zhig to gang la la zhig
b10: ’gyur ro to ’gyur to
c16: myI ’gyur te to myI ’gyur to
c16: myI ’gyur ro to myI ’gyur to
e6 and e12: mdo to mdo sde
f7: rin che to rin p(h)o che
f11: rgyal po chen po bzhi to rgya mtsho chen po bzhi
f13: nus kyi to nuso
Less frequently, editors corrected the sutra's title. In one case the scribe corrected himself at brgya dkar skad du (ITJ 310.111), but in another the same error goes uncorrected (ITJ 310.212). In several cases the Sanskrit title had to be corrected. In a few instances we have na ma ma ha na ya corrected to to na ma ma ha ya na (ITJ 310.788; ITJ 310.1119). This is an understandable mistake when one remembers that the dhāraṇī ends “ma ha na ya pa ri ba re sva ha”, and the scribe will go on to write this multiple times. The mistake is fully realized with this struck-through implicit at ITJ 310.1167: $/:/ rgya gar skad du ^a pa rI mi ta ^ yur nya na su bI ni sci ta ra dza ya ta tha ga ta ya tad thya tha ^om sa rba sang ska ra pa ri shud dha/ dhar ma te ga ga na sa mu dga’ te sba bha ba bi shud de ma ha na ya pa rI ba re sva hA/ / $/:/ rgya gar skad du ^a pa rI mi ta / ^a yur na ma ma ha ya na su tra /. The Tibetan title, too, often required correction: in ITJ 310.523, tshe dpa zhes bya ba’ is corrected to tshe dpag du myed pa zhes bya ba’.
Lest one assume that such errors are due only to the circumstance of having Chinese scribes write in Tibetan, a second language, one can point to comparable howlers in Chinese Ap copies. In the end title (weiti 尾題) of Or.8210/S.68, for example, the title is given as Foshuo wuliangshou yaozong jing 佛說无量壽要宗經 instead of Foshuo wuliangshou zongyao jing 佛說无量壽宗要經.
It is also instructive to examine what the editors have either introduced or left to stand. One consequence of the light-touch model of editing that operated here is that in nearly every sutra one can find errors that have been overlooked. In the sample sutra PT 3901, for example, ’jig rten is spelled ’jin rten in line a8, ri rab is rib rab in line d4, and the negation myed is misspelled myis at line f9. More interesting are those cases where editors have corrected a word, but done so in a way that allows for variation. To name a few examples taken from our data, we observe that in correcting bzhugs ’tsho’ to bzhugs shing ’tsho’, the editor did not strike through the suffix at the end of ’tsho’ (ITJ 310.104). Similarly, in this same passage in a separate sutra, an editor has corrected tso to ’tso, rather than ’tsho (ITJ 310.248). At a9, between yon tan dpag du myed pa and zhes bya, an editor inserted in red lastsogs pa, and not la stsogs pa or las stsogs pa (ITJ 310.343). In another copy, in the final sentence at g18, ’ga’ go is corrected to ’ga'o, rather than dga'o (ITJ 310.1030).
There is no doubt that editors sometimes introduced errors. To note one of the worst, in ITJ 310.554, an editor “corrects” sangs rgyas to sang rgyas at line f16, thus misspelling the word for Buddha.
There were a few intriguing hints of pedantry where editors have corrected grammatical particles. One editor corrected tshe'ang to tshe yang at b1 (ITJ 310.694). Another, in the passage from g1–g13, thrice corrected stobs gis to stobs gyis; and corrected rten gI to rten gyI at both e17 and g17 (ITJ 310.694). Incidentally, the scribe does use kyi(s) elsewhere in this manuscript, and the editor lets it stand. Another instance of orthographic correction, apart from the most common corrections listed above, is the correction of bud med to bud myed at e13 (ITJ 310.1116).
These examples serve to give one an idea of the sort of editing that was performed. Editorial thoroughness varied from one copy to another, and individual teams of editors had their own respective modus operandi in terms of which passages they tended to flag up. Their edits are helpful in gauging the difference between an acceptable variant and an unacceptable error. At the same time, and as a result of the fact that different teams of editors worked on these sutras, there is little global consistency in these insertions and corrections, and indeed there are some clear cases where one editorial team's correction is another team's error, as we shall see below in the context of the use of dga’ ’o versus dga go in the explicit. Stepping back from the content of their editing, the editors' red or black corrections are visual evidence that they have discharged their duties in the sutra-copying project. This is not to say that their relatively laissez-faire editorial style indicates an absence of orthographic norms. Indeed, to obtain a less impressionistic perspective on the orthography of these sutras it is more helpful to turn to quantitative data supplied by our transcribed explicits.
Explicit orthography
The final part of the sutra, just before the colophon, reads: “and all were truly pleased at what the bhagavān said. The Aparimitāyur-nāma mahāyāna-sūtra is complete”. One more or less “correct” version of this reads de thams chad / bcom ldan ’da's gyis gsungs pa la mngon bar dga’ ’o /:/ / tshe dpag du myed pa zhes bya ba theg pa chen po ’i mdo rdzogs so / /. To the uninitiated, there might seem to be little room for variation in this ending. I shall proceed through it slowly, therefore, to demonstrate exactly where the variation, and the relevance to orthography, lies. In 188 searchable explicits, we find omissions and additions, and we find variants or errors for nearly every syllable.
• “All that.” The explicit begins de thams chad. The word immediately preceding is ’khor. In a few cases (e.g. ITJ 310.150 and PT 3951), scribes omit de, and in one case it is inserted interline (ITJ 310.426). In one case, it is misspelled te (ITJ 310.1011). A more acceptable variant is the plural de dag, and we find 20 of these against 165 de. There is some interesting orthographic variation in the word “all”. Spelled thams cad in Classical Tibetan, we find three acceptable forms in our corpus. There are 86 thams cad, 79 thams chad, and 8 thams shad. One of the latter is spelled without tsheg, e.g., dethamsshad (ITJ 310.981). In three cases, we find the anusvāra, e.g. thaMs chad (ITJ 310.140 and 776) and thaMs shad (ITJ 310.258). We find subscribed suffixes, e.g. tham+s chad, twice (ITJ 310.1021 and PT 3812). Among misspellings, we have 2 thams can (ITJ 310.1044 and ITJ 1658), 1 thams cadu (ITJ 310.449), 1 thaMs shas (ITJ 310.649), 1 thams cas (ITJ 310.523), and 1 thams (ITJ 1687). We also find the correction thams bchad (ITJ 310.94). The most egregious misspelling is tha ced at PT 3648.
• “Bhagavān.” bcom ldan ’das may also be spelled with a medial ’a, e.g. bcom ldan ’da's. The aspirated bchom is also a variant. We find 170 bcom and 11 bchom. We also find 3 pcom, 2 brcom, and 1 mchom. In terms of corrections, we find 2 mbcom, and 1 bmcom. Twice we find an anusvāra, e.g. bcoM (ITJ 310.258 and 1094). We find the form ldan 185 times, and ldand twice. There is also 1 ltan (ITJ 1707), 1 ltaldan (ITJ 310.177), 1 ldag (ITJ 210.124), and once ldan is elided, e.g. bcom ’da's (ITJ 310.310). We also find 5 instances where the word is corrected to ldan, e.g. ltdan (ITJ 310.1084) or ldadn (ITJ 310.111). There are 116 ’da's, 70 ’das, 1 ’da (ITJ 310.352), and 1 ’da’ (ITJ 310.666). In one case, ’da is inserted interline, below the s suffix (ITJ 310.517). In another case, we find ’da’ sa for ’das (ITJ 1488).
• Ergative. Besides the ergative kyis or gyis, we also find here the genitive kyi or gyi. This error is not surprising, given the sentence structure. Still, the vast majority of scribes were correct, as we find 146 ergative against 35 genitive, and 5 instances where no particle is used, e.g. bcom ldan ’da's gsungs pa. As for the form or sandhi – and disregarding whether it is ergative or genitive or whether there is a gi gu or gi log (e.g. counting both kyI and kyis under the rubric “kyi(s)”), we find 57 kyi(s), 120 gyi(s), 3 gi(s), 2 yi(s) and 1 kis (ITJ 310.612).
• “What he said (gsungs pa la).” The verb gsungs is found as such 176 times, 10 times as gsung, and once as stsungs (ITJ 310.228). In two of the above cases we count gsung sa as gsungs. In 159 cases the nominal particle is pa, in 13 cases it is ba, in 1 case pa’ (ITJ 310.730), and in 2 cases it is absent. In one case it is an interlinear insertion (ITJ 310.94). In 167 cases we find la, and in 15 cases las. In one case we find pa pa instead of pa la (PT 3865).
• “Truly.” In 171 cases we find mngon bar, as opposed to 9 mngon par, 4 mgon bar, 1 mngond bar (ITJ 310.324), and 1 mngon du (ITJ 310.798). We once find mngor corrected to mngon (ITJ 310.517).
• “Were pleased.” In 38 cases we find the dga'o one would expect in Classical Tibetan. In 69 cases, a tsheg intervenes, e.g. dga’ ’o, in 11 cases we find dga’'o, and we find 8 dga ’o. In 43 cases, we find the unexpected dga’ go. Where the verb's suffix is left off we find 8 dga go, 1 dgago (ITJ 310.241), 2 dga ga'o (ITJ 310.248), and 1 dga ’go’ (ITJ 310.592). Intriguingly, we find 7 cases where the sentence final particle go is struck through and replaced with ’o. Once we find the misspelling gda'o (ITJ 1609).
• “The Aparimitāyur (tshe dpag du myed pa).” This part of the sutra's end is missing from 9 sutras of our sample, and is written twice in one. The sample size is thus different from above. We find 174 tshe, 4 tse, 1 dtshe, and 1 che. There are 167 dpag, 11 dphag, 1 dbag, and 1 pag. We never find the expected tu, but rather have 181 du, and 1 omission (e.g. dpag myed; ITJ 310.318). We find 179 of the expected myed, and not a single med. In a few cases, myid is corrected to myed, and in one case it is written myi. This, it should be noted, is in a semi-literate outlier, PT 3648, on which see fn. 29 below. Among the myed pa we include the correction myed pa'I mdo ’di b pa (ITJ 310.235). There are 160 pa, 11 pa’, 6 pa'i, and 1 omission, e.g. myed pa'I zhes (ITJ 310.971). In 7 cases, we find pa’i.
• “nāma (zhes bya ba).” We find 171 zhes, 2 shes, 2 zhe – one of which is zhes (ITJ 310.967) – and 1 omission (e.g. myed pa bya ba; ITJ 310.582). There are others that skip even more text, e.g. tshe dpag du myed pa'i mdo at ITJ 1697. We find 1 instance of the correction shzhes (ITJ 310.34). Where bya ba is expected we find 165 bya, 12 bye, 1 byas, and 2 missing (e.g. zhes theg, zhes ba). In a few cases this is corrected, byea (ITJ 1600), and byas (ITJ 310.967). Once we also find the reverse “correction”, bya bye (ITJ 310.772). The nominal particle is follows: 171 ba, 3 ba’, 2 ba ’i, and 1 pa.
• “Mahāyāna-sūtra (theg pa chen po ’i mdo).” In theg pa we find 168 theg, 4 thig, 2 thegs, 2 then, 1 them, and several omissions (e.g., zhes bya ba'i mdo). For the nominal particle there are 172 pa and 2 pe. We also find the corrections thoeg pa (ITJ 310.177) and thieg pa (PT 3520). For chen po'i the results are 158 chen, 11 ched, 8 cen, and 1 ced. There are 110 po'i, 30 po'I, 19 po ’i, 7 po ’I, 4 pho'i, 2 pho'I, 1 pho and 3 po. We find 178 mdo, 2 mdo’, 1 ’do, 1 mo, 1 ma, 1 mdo'o, and 1 omission (e.g. dga'o / / rdzogs so). We also find the correction bdmdo. In two instances we find the determinative “this sutra” mdo ’di, but in just as many cases we find this struck through.
• “Is complete.” We find 114 rdzogs so, 19 rdzogs so+’, 15 rdzogs s+ho, 4 rdzogsso, 1 rdzogso+’, 1 rdzogs+ho, 1 rdzogso’, 1 rdzogs stso, 3 rdzogs, and 1 skipped (mdo'o).
• “Amitābha. Praise to Amitābha” (^a myi ta phur / na mo ^a myi ta phur). Between the end of this explicit and the scribe's and editors' colophon, we find in 14 instances an invocation to Amitābha. This is the Tibetan rendering of the Chinese Amituofo, and as with the personal names of scribes and editors, it is given to variation.Footnote 22 We find 5 myi, 7 myI, 3 mi, 11 mI, and 1 ma, 1 mo, and 1 mye. There are 23 ta against 3 da, with one corrected to tia. There are 14 phur against 13 bur. We also find one alternate invocation: na mo dyi dzang bo sar, a homage to the bodhisattva Kṣitigarbha (Dizang 地藏).
From this sample it is possible to make a number of observations. Assuming that this is representative of these officially sponsored sutras, it provides a fairly good baseline for the orthographic principles of its time and place, namely the 820s to 840s in Dunhuang. Some results are not particularly surprising, but others help to establish a relative chronology of orthographic change in Middle Old Tibetan. To begin, medial ’a are licit. This is evident from the 116 ’da's as against 70 ’das in the word bcom ldan ’da's. The ma ya btags is normative with i and e vowels, e.g. myi and mye. The one exception to this comes in the name ^a mi ta phur, but this can be explained as a special case pertaining to the transcription of a foreign word. In the sutra's title given in the explicit, we find only myed and not a single med. Our sample sentences do not include any words where the da drag is expected, although we do find the odd over-eager scribe adding these at ldand and mngond. Had we applied the same methods to the SP, we would have found a sufficient proportion of pha rold du phyind pa to demonstrate that da drag are licit, if not expected.Footnote 23
The sample also demonstrates another common feature of early Tibetan writing, the alternation of n and d suffixes. This is attested in the ratio of 164 c(h)en : 12 c(h)ed, showing a clear preference for c(h)en. The gi log is also common. Making a ratio from the appearance of the i vowel in our sample explicits – mostly, but not exclusively, limited to ergative and genitive particles – we get 272 gi gu : 147 gi log. By comparison, the ratio is 100 : 86 in the north and south faces of the Sri Pillar (c. 764), and 229 : 287 in the Dbon zhang Inscription of 823.
Our sample provides fairly ample data on the variation between aspirated and unaspirated voiceless consonants. Here one must take account of whether or not the aspiration occurs in an initial position, word-internally, or otherwise. For example, in the initial position we find a ratio of 167 chen and ched to 9 cen and ced in the word chen po, and a ratio of 174 tshe : 4 tse. Word-internally, we have 169 po : 7 pho in c(h)en po/ c(h)ed po. These results are not far from Classical Tibetan standards, where one expects an aspirated initial and an unaspirated second syllable, e.g. chen po. Theg pa in fact fully accords with such standards, with 168 theg : 0 teg and 172 pa : 0 pha. If we consider the use of the sentence final particle in the same context, we find the second syllable mostly unaspirated: 118 rdzogso/ rdzogs so : 16 rdzogs s+ho/ rdzogs+ho. Footnote 24 Where there is significant divergence from Classical Tibetan standards is in the ratio 167 dpag : 11 dphag, where the consonant in question is not in the onset position. This is almost identical to the ratio of 170 bcom : 11 bchom. Our sample's relative alignment of aspiration with Classical Tibetan norms is further complicated by the ratio of 86 thams cad : 79 thams chad, where the former became the standard form. Compared with the other ratios, this is the most anomalous. Precisely this phenomenon has already been discussed, however, by Nathan Hill: “the use of aspirated spellings word-internally may be credited to a morphophonemic tendency in the orthography. Since these morphemes were most frequently spelled as aspirated, the aspirated spellings were generalized, despite the unaspirated pronunciation word-internally”.Footnote 25 In other words chad, as in the phrase chad lhag (“omissions and additions”), was sufficiently frequent to override the “correct” form thams cad with the “popular” form thams chad.
Our transcribed explicits also measure the fluidity of syllable margins. We find a ratio of 128 dga’ ’o/ dga’ go/ dga ’o/ dga go : 50 dga'o/ dga’'o/ dgago. The preference for separated syllables is more pronounced in the ratio of 148 rdzogs so/ rdzogs so+’/ rdzogs s+ho/ rdzogs stso : 7 rdzogso/ rdzogso’ rdzogso+’/ rdzogs+ho. Absent or excessive syllable margins, e.g. dethamsshad and gsung sa pa, respectively, are rare.
In the case of the ’i(s) form of the genitive and ergative particles, the preference for separated syllables is reversed. In our sample we find 26 separated p(h)o ’i or p(h)o ’I : 146 attached p(h)o'i or p(h)o'I. This latter ratio is 0 : 1 in all imperial-period inscriptions (e.g. 0 : 29 in the north and south faces of the Sri Pillar and 0 : 94 in the Dbon zhang Inscription). This appears to be part of a general movement from the mid-eighth century to the late tenth century whereby attached ’i(s) or ’I(s) gave way to separated ’i(s) or ’I(s). In PT 44, for example, which dates to either 966 or 978, the ratio of separated to attached ’i(s) or ’I(s) is 37 : 0.Footnote 26 In a Guiyijun administrative document, PT 1097, the ratio is 26 : 0. Of course there are outliers, and ratios are less stark in many documents, such as the mid-to-late ninth-century Guiyijun legal document PT 1081, which, in an admittedly small sample, has a ratio of 6 : 6. Moreover, the trend would reverse itself when separated ’i(s) and ’i(s) were replaced with yi(s), generally employed causa metri. Nevertheless, making allowance for such outliers, and understanding that the ratios did not flip all in the same place at the same time, the gradual separation of ’i(s) from the preceding syllable appears to be one of the few features one can use fairly reliably to distinguish Middle Old Tibetan from Late Old Tibetan orthography. Our sample includes one example each of yis and yI (ITJ 310.196 and ITJ 310.373, respectively) as ergative and genitive particles, suggesting that this convention may have been just edging its way into Shazhou in the 820s to 840s.
As for the sandhi of grammatical particles, the data turned up a few surprises. The first is the use of go as a sentence final particle following dga’. Here one expects ’o, and indeed ’o was far more prevalent. Additionally, several instances of dga’ go were corrected to dga’ ’o. This suggests that dga’ go was viewed, at least by some editors, as incorrect. For many of the Chinese scribes of Dunhuang, dga’ ’o and dga’ go must have been phonologically equivalent. We see this also in the spelling mthag dag for mtha’ dag, found in numerous Ap copies, a point that demonstrates that the sandhi in dga’ go was not the result of an agreement with the root letter ga. This adds some further support for Nathan Hill's conclusions concerning the status of the consonant ’a as a velar fricative.Footnote 27 There are several examples in the colophons, ranging from the spellings of personal names to the use of ’is following a suffix, that bear this out.
The forms of the genitive and ergative particles, like the use of attached versus separated ’i(s), appear to have been in a state of flux. In some texts one finds a distinct preference for gyi(s) where, following d, b, and s suffixes, one would expect kyi(s). This is true of the east face of the Sri Pillar (c. 784), the Bsam yas Inscription (c. 779), and many other texts (e.g. Old Tibetan Annals, Shangshu Paraphrase, PT 999, PT 1111, PT 1132). On the other hand, the Dbon zhang Inscription uses kyi(s) where it would be expected by Classical Tibetan standards. Furthermore, one finds idiosyncratic uses such as in the late Guiyijun administrative text PT 1082 where only gi(s) is used. In the Ap sample, the ergative following bcom ldan ’da's should, by the Classical Tibetan standards, be kyis. The results, however, are 120 gyi(s) : 57 kyi(s).
More decisive is the terminative particle. In the title of the text we invariably find dpag du, and not the Classical-Tibetan-mandated dpag tu. The ratio of the former to the latter is 181 : 0. In fact, this is representative of other Middle Old Tibetan texts, and may apply to Tibetan Dunhuang manuscripts more generally. Du is consistently used where Classical Tibetan standards expect tu in the Dbon zhang Inscription and in the vast majority of official imperial Tibetan letters, the Old Tibetan Annals, the Old Tibetan Chronicle, the Rāmāya ṇa, etc. Occurrences of tu in Old Tibetan documents are so rare as to be suspect: if they are not mistranscriptions or written in error by a sloppy or barely literate scribe, they should cause one to query the date and provenance of the document in which they are found. This is not, however, to claim that tu simply did not exist in Middle Old Tibetan. Indeed, one does find the occasional bona fide tu, as in the case of a few unequivocal rab tu interspersed with rab du in a pothī-format SP (type 1), PT 1321, r1. Here one must also take into account the graphic variations in the consonant t when it is joined with the u vowel. In most cases its descender is shortened such that tu can resemble du. The obvious way to reduce this from a matter of opinion to an informed and verifiable statement based on visible evidence is to compare closely any ambiguous du / tu of a given manuscript with unambiguous du / tu as they appear in the same manuscript within words such as dus or btus. So, while it is indicative, there are exceptions, and the presence of tu cannot be taken, by itself, as proof that a given manuscript is late (e.g. post-tenth century).
Two features, the anusvāra and the a’ suffix, are hardly attested. Only 3 of 153 thams c(h)ad/ thams shad use the anusvāra, and we find a similar ratio of 2 bcoM : 179 bc(h)om. Were one to include the colophon, the ratios would likely even out, as we find a preponderance of suM zhus. As for the a’ suffix, we could conceivably find it in our sample as gsungs pa’, myed pa’, and theg pa’. Only the former two are attested. This follows a more general pattern observed elsewhere, by which the a’ suffix is more common in nominalized verbs or in a prepausal position than it is in a noun such as theg pa. Even so, its occurrence in our sample is rare: we find 172 gsungs pa / gsungs ba against 1 gsungs pa’, and 160 myed pa to 11 myed pa’.
Editing the editors
Scribes used exemplars to copy sutras, and editors used exemplars to correct them. Both types of exemplars themselves were customarily written by editors, and it stands to reason that they should be something like a control from which to judge the sorts of variation that we've observed. Fortunately, our study and that of Nishioka allow us easily to find editors' copies in order to test this hypothesis. These are copies written by those who appear as editors in scores of sutra colophons. This is to be distinguished from the inverse, where a scribe of many sutras is found as the editor of a few. In the latter case, we assume that the scribe was promoted to an editor. Our representative sample of transcriptions from approximately every tenth shelfmark yielded only four editors' copies. We transcribed 48 additional editors' copies from the collections of the British Library and the Bibliothèque nationale de France. Among these are ten by the monk (ban de) Ci keng; eight by Shin dar; eight by Dpal gyi sgron ma; and six by Dpal mchog. There may have been more of these example copies, perhaps one for each team of scribes and editors. They were evidently included in the completed sutras when they were finalized and submitted, but their more immediate purpose was practical, and some show signs of wear. PT 3601, an exemplar written by the editor Shes rab, appears to have been kept by a scribe named Cang Zhig hing: we find in the colophon the phrase, “Cang ZhIg hing[’s] copy” (cang zhIg hing dpe’/ legs so/). Small panels are attached at the beginning and at the end, and the penultimate panel has been patched at the back. The panel following the sutra contains a key to a numbering system, perhaps that used by Cang Zhig hing's scriptorium in their tallies of how many folia were discarded due to scribal errors.Footnote 28 This suggests that it was a personal or scriptorium copy, not necessarily intended for submission.
Analysing these copies, and comparing the groups of copies attributed to a single editor, we find no false attributions. That is to say, all of the copies “signed” by Dpal gyi sgron ma, for example, are of the same hand, and the same principle holds true for all the 52 example copies in this group.Footnote 29 Apart from palaeographic features, one notes orthographic practices that are generally consistent across the copies written by a single editor. Ci keng, for example, places the shad punctuation mark after thams chad, and sometimes after gsung ba la, and prefers cen po to chen po; Shin dar consistently writes de dag thams cad instead of de thams cad; Dpal gyi sgron ma prefers dga’ go to dga’ ’o. This latter point is relevant to the matter of whether or not dga’ go was viewed as an incorrect form; to the editor Dpal gyi sgron ma it was not.
When taken together as a group, the editors' copies display an orthography remarkably similar to that of the representative sample of explicits. This justifies the grouping together of these explicits with those of the representative sample taken from approximately every tenth shelfmark. It also demonstrates that the sort of variation that we have observed, e.g. dga'o versus dga go, is not simply a product of poorly edited scribes, but represents variations beyond the simple dichotomy of correct and incorrect forms.
Conclusions
Our results are limited by the size and nature of our sample. As a fairly representative sample of the orthography employed by sutra scribes from the 820s to the 840s in Dunhuang, it flattens out and averages the data, drawing out the generalized standards of the sutra scribes. At the same time, certain scribes might consistently use orthographic forms that are in the statistical minority, or forms that some editors correct. This is true, for example, of Dpal gyi sgron ma's use of dga’ go. Such is the nature of how orthographic norms coalesce. In this case, however, it also reveals an editorial process in which some exemplars to be copied themselves contained forms that were on the borderline between acceptable variant and unacceptable error. These are relevant both to the process by which orthographic norms coalesced, and to the mode of editing applied to copies of the Ap in the sutra-copying project. The aggregate data, however, point to the existence of certain clear orthographic standards. These, together with studies of other Dunhuang documents and early Tibetan writings, help us to establish some of the prevailing orthographic norms of Middle Old Tibetan.
Our sample explicit contains information about syllable margins, aspiration, forms of genitive and ergative particles, d/n suffix variation, the use of my with i and e vowels, and the use of medial ’a, anusvāra, and a’ suffixes. The sample tells us nothing of several other interesting features, including the semi-final particle (ste, te, de), the presence or absence of the genitive preceding the pluralizer rnams, and the forms of the concessive (yang, kyang, ’ang), co-ordination (cing, zhing, shing), and quotation particles (ces, zhes). Two features, ’i(s) and tu, were identified as fairly reliable indicators for dating certain orthographies. Mid-eighth and early ninth-century documents from both central Tibet and Dunhuang tend to attach the ’i and ’is forms of the genitive and ergative particles to the preceding syllable. Tenth-century Dunhuang manuscripts, on the other hand, usually insert a tsheg to mark ’i(s) as a separate syllable. For the period from the late Tibetan Empire to the early Guiyijun, that is, mid-ninth to late ninth century, the data tends to be mixed. It remains to be determined when and where the ’i(s) rejoined the preceding syllable, and when yi(s) – probably by analogy with ’ang and yang – replaced the stand-alone form ’i(s). Here one must look to materials from Turfan, Tabo and elsewhere. Clues may also lie in the Dga’ thang ’bum pa texts from southern Tibet, which, while preferring attached ’i(s), do include some separated ’i(s) and show no knowledge of the form yi(s).Footnote 30
The absence of tu in our sample is striking. This would seem to accord with Miller's reading of the norms described in śloka 8 of the Sum cu pa. At the same time, we find unequivocal tu in copies of imperial-period SP and in a few other texts. If these are not phonemic misspellings – which is a real possibility, given the variation between voiced and voiceless consonants, e.g. te thams cad in ITJ 310.1011 – then they constitute evidence that this form was not unknown in Middle Old Tibetan. The uncertainty in this matter stems from both the graphic similarity of du and tu, and the alternation between voiced and voiceless consonants. In our sample, gyi(s) and kyi(s) are almost interchangeable following the s suffix. Morphemic norms seem to have developed in a piecemeal manner, with the norms governing the morphology of one case particle offering easy analogies for the standardization of another. Such seems to have happened, for example, with the innovation of yi(s) by analogy with the concessive particle yang, which enjoyed widespread use for centuries before yi(s) came into common use. The same principle stands behind forms such as thams shad, which probably owes its existence to an analogy with the morphology of the coordination particle (cing, zhing, shing), where shing follows the s suffix. To consider the emergence of tu, then, one might look to the use of unaspirated voiceless consonants in other case particles, such as ces, cing, kyi(s), kyang, and te. These are all attested in Middle Old Tibetan. The relative absence of tu in the corpus of searchable Old Tibetan documents might be taken to suggest, however, that these were secondary developments from their voiced counterparts and part of the same process that would eventually normalize tu. Indications that this process was still being settled include the preponderance of forms like ches and ching, but more especially the use of gyang, de, and gyi(s) where kyang, te, and kyi(s) are expected.
Obviously, the sample is not a basis from which to write a grammar of Middle Old Tibetan. It does, however, contribute towards establishing a baseline for the orthographic standards of literate, Middle Old Tibetan writing. As we know from the many jottings and patches that adorn the discarded leaves of Tibetan Dunhuang Śatasāhasrikāprajñāpāramitā-sūtra copies, and make these texts such a comparatively rich trove for the social and cultural history of the sutra-copying project, the scribes who wrote the sutras were the very same scribes who wrote official letters and contracts. Their flights of fancy include dhāraṇī, Mahāyoga tantra, proverbs and verses from the Rāmāya ṇa. The conclusions about their orthographic norms, therefore, are not to be cordoned off and held as relevant only to sutras. The study of the orthography that governed their writing, in addition to the palaeographic study of scribal hands, aids us in distinguishing between the production of such scribes and the work of those who came later.
