Faxian ’ ? : An Exercise in the Computer-Assisted Assessment of Attributions in the Chinese Buddhist Canon *

In the Taishō canon, the Mahāparinirvāṇa-sūtra 大般涅槃 經 T no. 7 is attributed to Faxian 法顯. However, on the basis of an examination of reports in the catalogues about various Chinese versions of the ‘mainstream’ Mahāparinirvāṇa-sūtra, Iwamatsu Asao 岩松浅夫 once questioned whether Faxian ever translated any such text. Iwamatsu argued further, on the basis of unspecified features of translation terminology and phraseology, that T no. 7 should instead be reascribed to Guṇabhadra 求那跋陀羅. This paper will examine the problem of the attribution of T no. 7 on the basis of a detailed examination of its language. Was the Mahāparinirvāṇa-sūtra 大 般涅槃經 T7 Translated by ‘Faxian’?: An Exercise in the ComputerAssisted Assessment of Attributions in the Chinese Buddhist Canon* * I am grateful to Hsue Yu-na 許尤娜 for conversations in Hamburg in late 2015 that stimulated me to look more closely into the problem treated in this paper.

Was the Mahāparinirvāṇa-sūtra 大 般涅槃經 T7 Translated by 'Faxian'?: An Exercise in the Computer-Assisted Assessment of Attributions in the Chinese Buddhist Canon * * I am grateful to Hsue Yu-na 許尤娜 for conversations in Hamburg in late 2015 that stimulated me to look more closely into the problem treated in this paper.

Introduction
I n the Taishō canon, the Mahāparinirvāṇa-sūtra 大般涅槃經 T no. 7 ('FX'-MPNS) is attributed to Faxian 法顯 (d. 418-423). 1 However, on the basis of an examination of reports in the catalogues about various Chinese versions of the mainstream Mahāparinirvāṇa-sūtra, Iwamatsu Asao 岩松浅夫 once questioned whether Faxian ever translated any such text. Iwamatsu argued further, on the basis of unspecified features of translation terminology and phraseology, that 'FX'-MPNS should instead be ascribed to Guṇabhadra 求 那跋陀羅 (394-468). 2 This paper evaluates Iwamatsu's hypothesis by examining the ascription of 'FX'-MPNS on the basis of internal stylistic evidence.
A cursory reading of 'FX'-MPNS in comparison to other Faxian ascriptions certainly seems initially to support the idea that 'FX'-MPNS at least cannot be by the same author as Faxian's other texts. For example, probably the most striking difference is the transcription of nirvāṇa, which is particularly telling given that both 'FX'-MPNS and the (Mahāyāna) Mahāparinirvāṇa-mahāsūtra 大般泥 洹經 T no. 376, also ascribed to Faxian, concern themselves centrally with the parinirvāṇa. Famously, Kumārajīva 鳩摩羅什 (350?-409?) seems to have coined the new transcription niepan 涅槃, whereas prior to Kumārajīva's time, other transcriptions were used, like nihuan 泥洹/泥垣, niyue 泥曰, etc. Kumārajīva's transcription seems largely to have supplanted the older transcriptions, and this term is therefore among the most famous watershed markers of chronology in the Chinese Buddhist canon. In this light, it is striking that the older transcription, nihuan 泥洹, is used copiously in T no. 376 and the Gaoseng Faxian zhuan 高僧法顯傳 T no. 2085-despite the fact that both texts were produced after Kumārajīva 3 -but never in 'FX'-MPNS; whereas niepan 涅槃 is copious in 'FX '-MPNS, but never used in T no. 2085, and only twice in T no. 376. Even more strikingly, in the remainder of the 'Faxian' corpus, banniepan 般涅槃 for parinirvāṇa only appears two times in the Mahāsāṅghika Vinaya 摩訶僧祇律 T no. 1425. 4 Similarly, further following the language associated with the particular theme and setting of the parinirvāṇa genre, 'FX'-MPNS transcribes Kuśinagara with the rare jiushina 鳩尸那, but T no. 376 and T no. 2085 both use juyi (cheng) 拘夷(城). 'FX'-MPNS transcribes the name of Cunda (a key personage) chuntuo 淳陀, whereas T no. 376 transcribes chuntuo 純陀. 5 'FX'-MPNS uses the rare transcription doupo 兜婆 for stūpa, which never appears in any other Faxian text, whereas other Faxian texts use ta 塔. For the śāla trees among or between which the Buddha passes into parinirvāṇa, 'FX'-MPNS uses the transcription suoluo 娑羅, whereas T no. 376 uses jiangu (lin) 堅固(林). Finally, for the verb 'weep' or 'lament', 'FX'-MPNS uses tiqi 涕泣, which is otherwise only ever found twice in T no. 1425 of the 'Faxian' corpus, whereas T no. 376 uses tiku 啼哭, which is conversely never found in 'FX'-MPNS.
Such anecdotal observations might suggest that Iwamatsu was correct, at least inasmuch as we should dissociate 'FX'-MPNS from  Especially in T no. 2085 (where Faxian was presumably the sole author), this perhaps reflects the fact that Faxian had his formative education before Kumārajīva's activity and was conservative in this wording. 4 The matter is complicated further by the fact that in the Faxian group's Vinaya translations, T no. 1425 and T no. 1437 include both transcriptions, though nihuan is still numerically dominant; T no. 1427 (a short text) includes one instance of niepan only. The instances in which niepan is used in T no. 376 are interesting precisely because they break this usual pattern. Both appear in verse: 1) 圓應神通眼/無量功德相/為眾生哀請/捨涅槃方便, T no. 376, 12: 1.858a29-b1; 2) 異法修無我/無量諸煩惱/異法修常存/佛性及涅槃, 885c12-13. 5 There is one apparent exception at T no. 376, 12: 1.858a9, but SYMP have the v.l. 純陀. 6 On T no. 745, see Tokiwa, Gokan, 55-56 and de Jong, 'Fa-hsien', 105-07 (who saw no reason to doubt that Faxian translated this text). 7 On T no. 2085, see Deeg,Liu,'Stories Written and Rewritten', ian's name. However, for various reasons, the assessment of ascriptions of Chinese Buddhist translations on the basis of style is a complex matter and requires that we marshal as much evidence as possible, as I will discuss in more detail below. Therefore, the best approach is to systematically compare the style of 'FX'-MPNS with other Faxian ascriptions and see whether or not any clear and significant commonalities and differences can be established. If we do find differences, we can then proceed to examine their possible significance, including whether they might point to a concrete alternative ascription.
This study therefore compares 'FX'-MPNS to other texts ascribed to Faxian. The other texts generally ascribed to Faxian at present are:  高僧法 顯傳 T no. 2085. 7 In the study of such questions, we should work conservatively to identify texts most certainly ascribable to the putative author of the text(s) under investigation, and take the style of those texts as a benchmark. In this light, we should note that there are reasons to be wary of taking T no. 1427 and T no. 1437 as direct representations of the Faxian style. Generally speaking, in the study of the Vinaya texts translated in the first decades of the fifth century, we need to be aware of extensive verbatim correspondences between them, which indicate heavy borrowing or recycling of wording. This problem potentially affects T no. 1427 and T no. 1437 particularly heavily, since they are both short texts (one fascicle each, compared to the forty fascicles of T no. 1425), so that the dilution affected by such verbatim borrowing is proportionally more intense. 8 I therefore provisionally exclude them from our benchmark corpus. 9 By contrast, I know of no particular reasons to doubt the ascriptions of T no. 376, T no. 745, T no. 1425, and T no. 2085, and in the course of research for this paper, I was unable to discover any. 10 I have therefore tentatively kept all these texts in the mix. This is a strictly methodological measure, and I do not intend by it to imply any judgment as to the reliability of the ascription of these texts to Faxian.
10 Special considerations apply to T no. 2085, Faxian's travelogue, which is quite different from the other texts in the 'Faxian' corpus. First, it is not a translation at all. This means that it is not a collective work in the sense they are; and that it belongs to an entirely different genre. Its idiom is closer to standard classical Chinese than almost any translation literature. We could naturally expect that many types of language that frequently recur in translation literature would not occur here-formulaic phrases of various types, common lists or pericopae for various doctrinal concepts, and so on. On the other hand, T no. 2085 is also the most likely source in which we might find preserved, undiluted, Faxian's own 'voice', and thereby, pinpoint traces of his individual contribution to the other more collective works.
These factors might lead us both to expect and to hope to find considerable stylistic differences between T no. 2085 and other 'Faxian' texts. In the event, however, my methods allow me to discover in T no. 2085 only a surprisingly small number of items of language that (possibly) are not content-related (e.g. do appear in other translation literature), and also appear in no other Faxian ascription: e.g. 毀壞 'destroy'; 貝多 'palm leaf'; 石柱 'stone pillar'; 彼土 'that country'; 挍飾 'ornamented(?)' (in varying orthography, this word is otherwise strongly associated with the Dharmarakṣa idiom); 佛處 'where the Buddha is/ was'"; 頂骨 '(Buddha's) skullbone, "uṣṇīṣa bone"'. At one fascicle, T no. 2085 is a relatively short text. Even allowing for this factor, however, these differences seem minimal. For the present, this means that despite differences in genre, idiom and compositional process, it is safe to leave it in the reference corpus for 'Faxian' style. 11 It is also a priori plausible that Faxian translated these texts. Faxian is supposed to have obtained in India manuscripts of the Mahāpari-nirvāṇa-mahāsūtra, the Mahāsāṅghika Vinaya, and the Sarvāstivāda Vinaya (among other texts); Glass,'Guṇabhadra',. It would make sense that he would have translated those texts upon his return to China. However, his name may also have been associated with the texts because he supplied the manuscripts; or the ascription to him could function as the (quasi-talismanic) guarantor of authenticity in the form of the living link with India. 12 The code repository for TACL may be found at: https://github.com/ ajenhl/tacl/. 13 The use of n-gram analysis for Chinese Buddhist texts has been pioneered by Ishii Kōsei. Ishii's methods differ somewhat from mine, but his ground-breaking work was an important source of inspiration. See Ishii, 'Daijō kishin ron'; Ishii, 'Shintai kan'yo bunken'. I also gratefully acknowledge the benefit to my work of email discussions with Professor Ishii, and his generosity in sharing with me some of his unpublished data.
14 Other studies using TACL are Radich, 'On the Sources' (part of a larger study with Radich, 'Tibetan Evidence'); Funayama, 'Da fangbian Fo bao'en jing'; Radich and Anālayo, 'Were the Ekottarika-āgama...'. For other studies using these tools, see Radich, 'Problems of Attribution'. For a little more discussion of TACL and its application, see Radich, 'On the Sources', 208.
This study was undertaken with the assistance of TACL ('Text Analysis for Corpus Linguistics'), a suite of computer tools I am currently developing in collaboration with Jamie Norrish. 12 As applied to the analysis of Chinese Buddhist texts, TACL allows a conceptually simple comparison of the n-grams 13 (strings of length n characters, where n is defined by the user), in two or more texts or corpora of any size, up to and including the entire canon, in either of two ways: (1) What n-grams are found only in A, and not in B (or vice versa)?
(2) What n-grams are found in both A and B? The tool generates full lists of n-grams matching these criteria, which the researcher can then examine in context, in conjunction with digital searches via the CBETA CBReader. 14 15 I did not especially target these texts in my searches. Rather, I searched equally over the whole corpus of ascriptions to Guṇabhadra in the Taishō. These texts emerged from such searches as most frequently containing phraseology linking them to 'FX'-MPNS.
The present study is intended in part as an introduction to TACL-assisted methods, and a showcase of their power to solve our research questions. For this reason, I have deliberately pursued a heuristic mode of exposition, which risks appearing somewhat mannered. To this end, I mimic the steps that such an investigation might take, beginning with the state of knowledge as we find it in the primary sources and the secondary literature, and 'walking the reader through' by steps to my final conclusions.

'FX'-MPNS is closer to 'Guṇabhadra' than to 'Faxian'
With the assistance of TACL, we can discover in 'FX'-MPNS numerous terms and phrases that never appear in any other text ascribed to Faxian. At the same time, many of these terms and phrases do appear in various 'Guṇabhadra' ascriptions. However, as I will discuss below, it turns out that these phrases are not evenly distributed, but appear most frequently in a particular subset of the Guṇabhadra corpus. For this reason, and because the evidence is copious and threatens to be overwhelming, I present here data for only a select subset of the Guṇabhadra corpus: 15 Saṃyuktāgama 雜阿含經 T no. 99; the Mahāyāna Aṅgulimāla-sūtra 央掘魔羅經 T no. 120; Guoqu xianzai yinguo jing 過去現在因果經 T no. 189 (abbreviated Guoqu); Pusa xing fangbian jingjie shentong bianhua jing 菩薩行方便境 界神通變化經 T no. 271; *Ratnakāraṇḍavyūha-sūtra 大方廣寶篋經 T no. 462; Laṅkāvatāra-sūtra 楞伽阿跋多羅寶經 T no. 670.
The terms and phrases in question are shown in Table 1. Throughout this paper, the translations or equivalents supplied for each item are approximate only (in any case, for some markers the meaning can shift somewhat depending on context, so that it is artificial to provide a single equivalent)-to aid readers in absorbing the information, and for the purposes of subsequent discussion about the types of language involved.     19 This phrase is surprisingly rare throughout the translation literature. The other text in which it is most concentrated is the *Ratnamegha T no. 658(8x). In all other translation texts, it only appears once. 20 I term markers like this 'juxtaposition markers'. They are constituted by the recurring combination of two or more habitual usages-here, for instance, a sentence marked with the final particle 也 and the habit of beginning a new sentence with 爾時 for 'at that time'. As 'wallpaper' (see p. 251), such markers may be particularly telling, though easily overlooked, and in application to some problems, they may combine to comprise a substantial set of evidence in their own right.
鬚髮自落袈裟著身 'his beard and hair fell out of their own accord, and a kāṣāya appeared on his body' By contrast, we also find a large number of words and phrases that occur in more than one text among T no. 376, T no. 745, T no. 1425, and T no. 2085, but not in 'FX'-MPNS (with very few exceptions, nearly all the items listed below occur at least ten times across the Faxian corpus as a whole). 25 25 I ask readers to be patient with the quantity of this evidence. I present it in full because an important part of the case I am presenting is that such copious evidence all points in the same direction; because I make use of the same evidence again below in a different connection; and because I believe the quantity of such evidence is significant methodologically.
草木 'trees and grasses, plants'  Thus, we find that 'FX'-MPNS and the remainder of the corpus ascribed to Faxian differ strikingly in the exact way they repeatedly phrase a wide range of terms and ideas. It is important to note that the above Tables include a wide range of types of language: ordinary nouns, verbs, and adjectives; words and phrases to do more specifically with Buddhism, in both its more technical aspects and in the more general 'worldview' that comes bundled with it; proper names; and recurring phrases, some betraying habitual preferences in conjunctions, pronouns and adverbs (in all lists in this paper, I have arranged markers very roughly into categories in this order). It is exactly this sort of recurring, diverse, and copious difference that adds up to a style, and these global differences between 'FX'-MPNS and other Faxian ascriptions indeed suggest that there was something fundamentally different about the compositional process behind each side of the comparison, and the person(s) responsible for them.
I believe it is safe to say that the application of these techniques shows us for the first time the quantities of such evidence to be found in a given body of text. TACL's first strength is the fine grain of the vision it bestows. It is as if we have been handed a microscope, which enables us to see features of the texts too fine to have been visible to the 'naked eye' of a human reader equipped only with ordinary philological acumen. The power of the tool is further increased by its scope. It is possible for TACL to work through the entire canon in a few minutes or hours, examining every fine detail of each text (if only details of a certain very narrowly circumscribed type), whereas the same task would take a human reader multiple years at best. Finally, an additional strength of these methods derives from the brute blindness of the machine. Buddhologists steeped in Buddhist problems and texts have tended overwhelmingly in prior studies to notice and exploit markers with an explicitly Buddhist colour-formulaic textual clichés (especially at the opening and closing of sūtras), doctrinal categories, proper names, and the like. By contrast, TACL does not know or care what kind of word or phrase an item is-it trades indifferently in all contiguous strings of characters. This enables us to expand our purview, as above, beyond such explicitly and saliently 'Buddhist' markers, to include a wide range of more ordinary language typically too nondescript to catch our attention. (I call such markers 'wallpaper'.) It is typical of work with TACL, as here, to discover that two texts or bodies of text are distinguished by a large number of such recurring fine-grained differences. This discovery is both exciting and challenging. On the one hand, it suggests that use of such internal evidence may eventually make possible much greater headway than we have achieved to date on questions of ascription, dating, and intertextual relations. At the same time, it also opens more than one new can of worms, each squirming with a lively knot of slippery problems.
One such problem is that it is difficult in many cases to differentiate with absolute clarity between content-related and stylistic material. For example, one area in which lexemes differ between texts because of content is Vinaya terminology (much of which appears for the first time in texts translated in Faxian's generation). 35 Some of this terminology also appears in T no. 376, mostly likely because some content in T no. 376 is also Vinaya-related.
However, against these considerations, we should note first that the above evidence includes a copious number of particular renderings of a wide range of items very common in sūtra literature. In total, we found over eighty items systematically differing from Faxian in the three fascicles of 'FX'-MPNS; and over 350 items systematically differing from 'FX'-MPNS in the Faxian reference corpus (a total of forty-eight fascicles). It is unlikely that such wide-ranging differences could be produced by accidents of content alone.
In the present case, we can also control for the possible confound of content by the fact that we find different translations or transcriptions for items identical in meaning: 'FX'-MPNS 兜婆 vs. FX 塔 for 35 Examples in the Faxian corpus include: 波夜提 pāyantika; 羯磨 karma (in the sense of monastic ritual); 某甲 'so-and-so, such-and-such a person'; 越比尼 'commit an infraction of the Vinaya'; 床褥 'bed, couch'; 革屣 'leather sandals'; 受具 'received [precepts, ordination] 36 These are reasonably common items in Buddhist discourse. The fact that they are systematically rendered differently on each side of our comparison strengthens the likelihood that we are dealing with various authors or translators. 37 A single person or group would be unlikely to switch between different renderings for such common terms, and if they were in the habit of alternating, we would expect to find both renderings occurring within single texts, rather than the clean split between texts that we see here.
In the present case, we also have an additional control against the possible confound of content. In addition to 'FX'-MPNS, the Chinese canon contains two other independent translations of the (Mainstream, non-Mahāyāna) Mahāparinirvāṇa-sūtra: the Fo bannihuan jing 佛般泥洹經 T no. 5 ascribed to Bo Fazu 白法祖; 38 and the anonymous Bannihuan jing 般泥洹經 T no. 6, which appears in the Taishō with a by-line dating the text to the E. Jin 東晉 (317-420), but which scholars have predominantly thought is probably by Zhi Qian. 39 In both T no. 5 and T no. 6 we find a large number of the exact markers listed in Table 2 above as distinguishing the Faxian 36 In 'FX'-MPNS, 頌 only occurs in the phrases 歌頌讚歎 and 歌唄讚頌. In other Faxian ascriptions 偈 never occurs in direct combination with 曰, as in 'FX'-MPNS; rather, it appears in the compound 偈頌, or with the verb of speech 言 (偈言), or with a verb of speech preceding, 說偈 (with no second verb of speech following), etc. 37 For a more extended application of this method, see Radich and Anālayo, 'Were the Ekottarika-āgama... '. 38 Iwamatsu and Park argued that T no. 5 is by Zhi Qian, but Nattier does not find these arguments convincing; Iwamatsu, 'Nehan gyō'; Park, 'New Attribution'; Nattier, Guide, 126, note 39, 127-28. 39 Nattier, Guide, 126-27. Nattier cites Ui, Yakukyōshi, 517-23. Iwamatsu, 'Nehan gyō', argues that T no. 6 was probably by Dharmarakṣa. Park, 'New Attribution', also treats the text as by Zhi Qian. corpus from 'FX'-MPNS: in T no. 5, approximately 122 items; 40 and in T no. 6, also 122 items. 41 Because these are parallel translations of the same text as 'FX'-MPNS, we can expect that differences in recurring wording between these texts and 'FX'-MPNS would primarily not inhere in content, but rather, in style. Although the number of markers of Faxian against 'FX'-MPNS is the same in both T no. 5 and T no. 6, this is something of a coincidence-only a little under two thirds of the markers (about 78) are shared between the two texts. Some of the language that is shared between the two texts could be accounted for by the fact that T no. 5 may be a revision of T no. 6; Nattier, Guide, 127. 42 This is naturally not to deny that there do indeed exist differences in details of content between T no. 5, T no. 6 and 'FX'-MPNS. The existence of such differences is well known. Careful analysis of the patterns of such difference (and contrasting commonalities) between these and other versions of the text (T no. 1(2), Pali, fragmentary Sanskrit, versions incorporated in the Vinayas) formed Nattier adduced strong reasons to think that T no. 6 is by Zhi Qian, and further, on the basis of relations between T no. 5 and the Fo mu bannihuan jing 佛母般泥洹經 T no. 145, that T no. 5 was 'likely... produced in the Wu kingdom in the third century CE'. 43 In showing the presence in T no. 5 and T no. 6 of markers more characteristic of Faxian than of 'FX'-MPNS, I therefore do not mean to suggest that either T no. 5 or T no. 6 should instead be ascribed to Faxian. Rather, my point is that even these two texts are closer to the style of the 'Faxian' corpus than 'FX'-MPNS, and this evidence therefore serves as an indication of the significant distance between 'FX'-MPNS and other Faxian texts. It also shows that differences in content cannot be responsible for this distance between 'FX'-MPNS and other Faxian texts.
To sum up the argument thus far: We have found over eighty terms and phrases recurring in 'FX'-MPNS, that never appear elsewhere in 'Faxian', but do repeatedly appear in Guṇabhadra. On the other hand, we also found over 350 items recurring in the remainder of the 'Faxian' corpus, which never occur in 'FX'-MPNS. We can exclude the possibility that these differences are based upon differences in content between 'FX'-MPNS and other 'Faxian' texts, because the same terms are sometimes translated differently on either side of the comparison, and because the markers otherwise characteristic of 'Faxian' do occur repeatedly in T no. 5 and T no. 6, which are parallel translations to 'FX'-MPNS. This evidence shows very strongly that 'FX'-MPNS is far closer, on stylistic grounds, to the Guṇabhadra corpus than it is to the Faxian corpus.

Complications
On the basis of the evidence surveyed thus far, it would be easy to leap to the conclusion that the above results resoundingly confirm Iwamatsu's hypothesis-'FX'-MPNS is stylistically closer to (some) the basis of a line of serious studies with historicist aspirations, such as Bareau, 'Les récits'; Waldschmidt, Die Überlieferung. 43 Nattier,Guide, One useful approach to such questions, suggested by Nattier, is to think in terms of 'rhetorical communities', identifiable by 'tracers' (distinctive terms of limited circulation), and divisible on occasion into further sub-groups. Such an approach has the advantage of shaking the problem of style loose from assumptions about named individuals (or even their ateliers). On the one hand, several such 'translators' could be members of a single 'rhetorical community'; while on the other, the corpus ascribed to a single 'translator' might comprise several separable 'rhetorical communities'. These two possibilities do not need to be mutually exclusive in a single case, since for various purposes, we might analyse a problem along a spectrum from coarse-to fine-grained. See Nattier,Guide,5,  texts ascribed to Guṇabhadra than those ascribed to Faxian, and we are therefore warranted in ascribing the text to Guṇabhadra. However, matters are in fact more complicated.
The study of ascriptions of Chinese Buddhist translations on the basis of stylistic evidence is complicated by the fact that translators often worked in teams, and the composition of those teams could shift over time. Insofar as we can show empirically that certain regular and consistent features are shared by a group of texts most firmly associated with the name of a given translator and his group, it is nevertheless still reasonable for us to seek to discriminate between works more or less typical of that 'author' and others. That is to say, we can reinterpret the names associated with texts in traditional ascriptions as labels for a translation group or atelier (for example, 'Faxian' = 'the Faxian group') and proceed from there. This is the approach taken here. 44 In the case of Faxian, however, these questions are further complicated by the fact that Faxian himself may not have been the person doing the principal work of actual 'translation' in the teams he worked in, but rather, the 'grunt work' of translation may have been done by Faxian's erstwhile travel companion, Baoyun 寶雲 (372?/376-449 Saṃyuktābhidharmahṛdaya T no. 1552Saṃyuktābhidharmahṛdaya T no. , T no. 2145 and the Śrīmālādevīsiṃhanāda-sūtra T no. 353, T no. 2145, 55: 9.67b3-5; and in the biography of Guṇabhadra, T no. 2145, 55: 14.105c14-20;GSZ, T no. 2059, 50: 3.344b3- a wider pattern indicating that Baoyun may have been the foremost Sanskrit-Chinese translator of his age. 46 We must therefore consider the possibility that Faxian, despite his extensive time in India, may not have actually been a real 'translator' (in our terms) after his return, and that any stylistic characteristics we can find in his corpus may in fact be the fingerprints of Baoyun (or someone like him). This is a particular problem for consideration of the present question. The Chu sanzang ji ji 出三藏記集 (CSZJJ) biography states that Baoyun did the main work of translation for at least some of the texts ascribed to Guṇabhadra, just as he did for Faxian. 47 Elsewhere, in a note to a list of thirteen texts, Sengyou writes, 'These texts... were all recited/read 宣出 by the Indian Mahāyāna Dharma Master Guṇabhadra...and translated 傳譯 by the śrāmaṇera Shi Baoyun and his disciple *Bodhidharmodgata 菩提法勇'. 48 Our primary sources also famously present evidence that Guṇabhadra himself may have been virtually incapable of speaking Chinese. 49 In other words, 50 Unlike CBETA, TACL has the capacity to search the Taishō apparatus for variant readings in other witnesses. Counts for a given word or phrase sometimes differ between witnesses. 51 This is likely to be in part because T no. 99, at fifty fascicles, is very large. See below. 52 T no. 270 features only 8 of the markers listed in Table 1. 53 T no. 353 also features only 8 of the markers listed in Table 1.
These are exactly the texts that also feature the largest concentration of the items in Table 1, which distinguish 'FX'-MPNS from Faxian. With the exception of T no. 270 (in which only eight items from Table 1 appear), this means that largely the same texts in the Guṇabhadra corpus are most like 'FX'-MPNS, and most like 'Faxian'. We must therefore consider the possibility that 'FX'-MPNS represents something more specific than a 'Guṇabhadra' text that was mis-ascribed by the tradition to 'Faxian'.
Thus, the markers in Table 1 are nearly twice as frequent in Guoqu as in any other 'Guṇabhadra' text. By this crude measure, 'FX'-MPNS lies closer to Guoqu than any other text in that corpus by a considerable margin.
It is also possible to find other evidence pointing in the same direction-phrasing shared by Guoqu and T no. 7, and entirely unique to them in all of the translation literature (in many cases, appearing more than once in one or both texts): 54 54 Some of these terms and phrases are also found in one other text-the Yinguo benqi jing 因果本起經, which was excluded from the canon, but preserved with an ascription to Guṇabhadra in the Fangshan stone canon (text no. 69 in Zhongguo Fojiao xiehui, Fangshan shi jing): 迦比羅斾兜 Kapilavastu; 自 傷貧乏 'beggaring themselves'; 非為小緣 'this is no trivial circumstance'. But these overlaps are to be explained by the fact that F69 is largely verbatim identical to about the first half of the first fascicle of Guoqu (T no. 189, 3: 1.620c15-623b27). Note that this makes F69 an important witness for the textual study of corresponding portions of Guoqu.  非為小緣 'this is no trivial circumstance [i.e. this is a fateful, weighty matter]' 58 55 尼連禪河側 (without 在) also has a telling distribution: 'FX '-MPNS, T no. 99;Guoqu, T no. 192;Mahāmāyā, T no. 1509. 56 The syntactically peculiar use of 見答 here may be a reflex (at what remove?) of an Indic passive; cf. the related 唯願見答, which is entirely unique to Guoqu. 57 In Guoqu, the reading 迦毘羅斾兜 in K hides this phrase from ordinary CBETA searches, but SYM and Shōgozō all record a v.l. identical to 'FX'-MPNS; in F69 (see note 54) we encounter the slight variant 迦毗羅斾兜. 58 Note also 迦比羅斾兜 Kapilavastu (note 18).
Other items shared between 'FX'-MPNS and Guoqu, though not entirely unique to these two texts, are still extremely rare, and provide additional evidence of close links between the two. 59 Where these rare pieces of phraseology appear in 'FX'-MPNS and Guoqu respectively, with one partial exception, in content and context that would indicate direct borrowing from one text to the other. 60 This means that they indicate, rather, some unusually close relation between the idiom of these two texts, and the person(s) who composed them.
At the same time, when we look further abroad, it turns out that one work outside the Guṇabhadra corpus has even closer links to 'FX'-MPNS than any of the Guṇabhadra works listed above, 61 excepting T no. 189-the *Mahāmāyā-sūtra 摩訶摩耶經 T no. 383 (hereafter 59 For example, the two texts share a verse, though the context differs in each text: 諸行無常/是生滅法/生滅滅已/寂滅為樂, T no. 7, 1: 3.204c23-24, T no. 189, 3: 1.623c21-22. This verse otherwise appears only in the anonymous Saṃyuktāgama T no. 100, *Dharmakṣema's , and the Mile da cheng Fo jing 彌勒大成佛經 T no. 456 ascribed to Kumārajīva. 60 The exception is a passage in Guoqu in which the Buddha refuses Māra's request, on the banks of the Nairañjanā River, to enter into parinirvāṇa, T no. 189, 3: 3.649a16-24. With the exception of a very few words, this passage is matched verbatim in a slightly longer and more repetitive passage at T no. 7, 1: 1.192a22-b12. However, even this long pericopae is set in a different larger context in each of the two texts: Guoqu is describing the initial encounter of Māra and the Buddha, at the beginning of the Buddha's teaching career; whereas 'FX'-MPNS is describing the reminiscence of this occasion forty-five years later, at the end of his career, when the Buddha agreed with Māra that he would enter parinirvāṇa three months later. 61 The Abhiniṣkramana-sūtra 佛本行集經 T no. 190, ascribed to *Jñānagupta, features the next largest gross number of Table 1  abbreviated Mahāmāyā), ascribed to Tanjing 曇景 (fl. ca. 479-502). This text features twenty-five items from Table 1, in a span of only two fascicles (12.5:1).
Utsuo argued that Mahāmāyā was composed in China, and further, that 'FX'-MPNS was among its principal sources. 62 Certainly, a close link between the two texts is corroborated by some very long and exact verbatim matches in phrasing. 63 However, not all the distinctive phraseology overlapping between the two texts can be accounted for by Mahāmāyā borrowing and reworking whole passages from 'FX'-MPNS, suggesting that the relation between the two texts might have some other dimension. These clues suggest that 'FX'-MPNS and Guoqu might belong together with Mahāmāyā in a group of texts sharing some quite specific interrelation.
As we will see immediately below, further investigation shows that in fact, these three texts share a considerable quantity of quite specific phraseology, and moreover, that the same characteristics are shared (to a lesser degree) by two more texts: the Buddhacarita 佛所行讚 T no. 192, ascribed to *Dharmakṣema, and the closely related Fo benxing jing 佛本行經 T no. 193, ascribed to none other than Baoyun.
was actually T no. 192 that was by Baoyun, and in presenting a full translation of the work, Willemen follows Ōminami in this regard. 66 Willemen also reports that Hikata Ryūshō believed T no. 193 was written after Zhi Qian and before Kumārajīva. 67 On the basis of a somewhat unconvincing computer-assisted analysis, Gotō argued that T no. 193 was translated by Dharmarakṣa 竺法護 (fl. ca. 284-306) rather than Baoyun; in the course of the same study, he appears to assume that T no. 192 is in fact by (Buddhabhadra and) Baoyun. 68 The evidence presented immediately below is ambiguous with regard to this question. It shows that T no. 192 and T no. 193 sport features that associate them closely with 'FX'-MPNS, Guoqu, and Mahāmāyā, but such features can be found in either text, and sometimes in both. This may be at least in part because one text could have been prepared in consultation with the other. This question, and the question of the ascription of both texts, deserves further study, but for present purposes, it will suffice to show the special relation enjoyed by both texts with the others in this group.  Guoqu,T no. 192/ T no. 193,and Mahāmāyā, but never in 'Guṇabhadra' Table 4 presents a sampling of phraseology distinguishing Guoqu from other texts ascribed to Guṇabhadra, but shared by texts in the group comprising 'FX Guoqu,T no. 192/193,and Mahāmāyā. For each item, I specify, after the item itself, whether it appears in T192, T193, or both.
Utsuo's work might give us reason to suspect that at least Mahāmāyā, in particular, shares such distinctive language with the other texts because it takes them as its sources. In fact, however, in very many cases, where these items occur in these texts, we do not generally find relations between contexts and content of the type that would show such borrowing. Moreover, as with earlier sets of evidence, we see recurring here all types of language. Again, these recurring features together constitute evidence of a style, which sets these four texts apart from 'Guṇabhadra' and ties them closely to one another. Further, much of this shared phraseology is otherwise rather rare in Chinese Buddhist translation literature as a whole. This suggests that these 'four texts ' (treating T no. 192 and T no. 193 together for the time being) are the product of the same close context or group. Future investigation should aim to discover whether these texts are linked by other features (including features of content), and whether more can be discovered about their context and links to other literature.

Conclusions
On the basis of the evidence presented above, we can conclude that the Mahāparinirvāṇa-sūtra T no. 7 is much closer to the style of certain texts ascribed to 'Guṇabhadra' than it is to 'Faxian'. Indeed, by the same yardstick, even 'FX'-MPNS's sister texts, T no. 5 and T no. 6, are closer to 'Faxian' than 'FX'-MPNS itself. We should, therefore, overturn the ascription to Faxian carried by 'FX'-MPNS in the Taishō.
At the same time, however, it is not safe to follow Iwamatsu and simply re-ascribe the text to 'Guṇabhadra'. In fact, markers distinguishing 'FX'-MPNS from the 'Faxian' corpus are found much more densely in the Guoqu xianzai yinguo jing than in any other 'Guṇabhadra' text. Further, a range of highly specific markers associate 'FX'-MPNS and Guoqu very closely with two further bodies of material, the *Mahāmāyā-sūtra, and the Buddhacarita T no. 192 and/or the Fo benxing jing T no. 193. Stylistically speaking, these four (or five) texts comprise a tightly interrelated group, which are also connected by common themes and content.
As we saw, historical evidence strongly suggests that Baoyun may have been the real translator in the production of several important works ascribed to both Faxian and Guṇabhadra. In this light, it is very tantalising to note that T no. 193 is one of only three texts ascribed to Baoyun in the present canon, 73 and among those texts, this is the ascription that is supported by the strongest external evidence. This might make it tempting to think that the 'FX'-MPNS- -Mahāmāyā group might have especially close links with Baoyun himself, or that the features discussed above, which unite those texts, comprise together a fingerprint of Baoyun's own style. In fact, however, the range of texts in which Baoyun is likely to have had a hand is much broader than only this group, and the problems involved in their study are considerable. 74 Those broader questions would take us well beyond the bounds of this study, but until they are resolved, we can say nothing reliable about the likelihood that Baoyun was involved in any or all of these texts. For the present, then, we can safely conclude only that 'FX'-MPNS is probably not by the exact same translator(s)/author(s) as the remaining core 'Faxian' texts (T no. 376, T no. 745, T no. 1425, and T no. 2085); and that our best indications tie it closely, rather, to Guoqu,T no. 192 and/or T no. 193,and Mahāmāyā. As mentioned at the outset, the above study was prepared with the aid of TACL, a suite of computer software tools designed for the discovery of evidence bearing on questions of style, attribution, and other intertextual relationships in the Chinese Buddhist canon. I hope that this study also demonstrated some of the promise and power of the careful use of those tools. It does not seem an overstatement to say that to date, without the aid of such tools, scholars in the field have been unaware of the full range, quantity and diversity of 73 The others are the Si tianwang jing 四天王經, T no. 590, ascribed to Baoyun in collaboration with Zhiyan 智嚴; and the Akṣayamati-nirdeśa included in the Mahāsaṃnipāta, 無盡意菩薩品 T no. 397 (12), also ascribed to Baoyun and Zhiyan. 74 I am currently preparing a systematic study of Baoyun's possible corpus and translation style and hope to take up these questions again in that work. evidence, like that examined here, that might exist in any given body of text. In comparison to the copious quantities of evidence discussed here, and the diverse range of types of language that can serve as distinctive markers on either side of a given comparison, I suggest that the handfuls of hand-picked (supposed) markers deployed in prior studies often now look impressionistic, scattershot and shaky. In this light, it will probably be necessary to re-examine even the small number of problematic ascriptions that have been critically studied on the basis of internal evidence in prior work.
At the same time, however, I believe that the present paper amply shows that these new tools promise to allow us to come to grips with such questions far more effectively than in the past. The mind boggles at the likely number of such problems that have probably slept for centuries beneath the surface of the canon, and the likely scale of the task of analysing the potential evidence, if it everywhere presents such an embarrassment of riches as here. If we can rise to the challenge, however, I also believe that such techniques might allow a profound and rigorous revision of the entire textual-evidential basis for many of our most important historical questions.
The total portion of each text comprised by such verbatim overlaps with other larger translations is large. For example, in T no. 1427, approx. 70% of the text is accounted for by verbatim matching strings of 8 characters or more in length with the four main Vinaya translations of the early fifth century (T no. 1421, T no. 1425, T no. 1428, and T no. 1435). In T no. 1437, the proportion of the same overlaps is approximately 67%. To give the reader some sense of the extent of this phenomenon, I have arranged each the two lists of overlaps from T no. 1427 above in the order in which they appear in the text.

Appendix II TACL methods used in this study
TACL includes a range of separate functions. For the convenience of the reader, I here provide a list of the basic functions deployed in this study, keyed to the places where they were used.
tacl difference: Finds all contiguous strings unique to each side of a comparison between two (or more) bodies of text. Examples: • 'FX'-MPNS versus other solid ascriptions to 'Faxian' (T no. 376, T no. 745, T no. 1425'Faxian' (T no. 376, T no. 745, T no. , T no. 2085): tacl search: Takes a list of multiple n-grams (sometimes many hundreds) and searches every text in the entire canon for all of them. Outputs a list and count of n-grams from that set found in every text. This allows the user to easily find places in the canon where a given set of n-grams are most (or least) concentrated. Examples: • Items from Table 2 in T no. 5 and T no. 6: 'In both T no. 5 and T no. 6 we find a large number of exactly the markers listed in Table 2 above as distinguishing the Faxian corpus from "FX"-MPNS...', p. 253-54, and fn. 40, fn. 41. • Items from Table 1 in 'Guṇabhadra': '...our markers of "FX"-lower limits for the length of n-grams in which they are interested, for maximum or minimum number of instances of n-grams, for the maximum or minimum number of works in which n-grams must appear, and so on; and that TACL also, unlike CBETA, searches the Taishō apparatus, and so, in principle, can take into account all the witnesses to a text consulted by the Taishō editors.
It should also be emphasised that TACL is only a tool or aid to human analysis. All the potential evidence that it finds must be subjected to further informed and careful analysis in context, as it occurs in the texts themselves. This phase of the analysis can only be performed by a competent human reader, and is just as difficult, and prone to error, as any other philological work. To the best of my ability, I have subjected all the evidence presented in this paper to such analysis.