1 From Parry-Lord to Shulman’s ‘play of formulas’

In this connection, a question Shulman seeks to explore in his book is just what constitutes a basic ‘text’ in early Buddhist literature. He thus wants to problematise the category of the ‘discourse’, that is, the sutta/sūtra, suggesting that we must look beyond this and explore how sūtras are the ‘end products of different dynamics of formula combination’ (Shulman 2021: 4). His study proceeds by considering four dimensions of early Buddhist literature: (1) the literary (chapters one and two), (2) the contemplative (chapter three), (3) the folkloric (chapter four), and (4) the oral (chapters five and six). Because these perspectives are ‘deeply intertwined,’ the literary and oral perspectives pervade the whole book.

Shulman’s whole study is premised on the idea that it is the formulas underlying sūtras that must be considered primary. The early sūtras should be seen as constructed, he argues, by early Buddhist preachers or reciters (bhāṇaka) through the creative use, application and adaptation of doctrinal and narrative formulas—what Shulman refers to as ‘the play of formulas’. In this respect he builds on an idea first proposed by Lance Cousins in his 1983 article, ‘Pāli Oral Literature’, namely, that the Parry-Lord theory of the composition and transmission of the Homeric epics might be adapted and applied to the composition and transmission of early Buddhist texts. As Shulman points out, scholars writing directly on the topic in the intervening years have mostly dismissed Cousins’ suggestion. Allon and Anālayo continue in that tradition, while McGovern (explicitly) and Shulman (more hesitantly)⁴ attempt to reconsider how the theory might be applied to early Buddhist texts, albeit in an adapted or refined form.

One sticking point has been that the Parry-Lord theory concerned verse formulas, while the early Buddhist formulas are prose. McGovern (2019: 454–459) attempts to address this point, and as Allon points out (even while rejecting the application of the theory to Buddhist texts), Buddhist ‘sutra prose […] is as highly structured as verse, if not more so’, and at times uses a kind of metrical prose (Allon 2021: 17–19).⁵ Another sticking point has been the notion of individual in-performance ‘improvisation’, when there is evidence to suggest that early Buddhist texts were memorized and recited communally as fixed texts. McGovern (2019: 455, 461) and Shulman (2021: 194–196) independently point out that Lord understands the term ‘improvisation’ in a somewhat restricted if not technical sense, and that it should not be seen as implying some kind of free performance; rather it occurs within, and is constrained by, a specific notion of ‘tradition’. And referring to the work of Ruth Finnegan on oral poetry, Shulman further points out that ‘improvisation’ (in the sense of the freedom to creatively adapt and enhance traditional material) can take place prior to actual performance, such that what finally emerges in performance is in fact something of a fixed text and composition. Again both McGovern and Shulman question the insistence that communal recitation of early Buddhist texts precludes ‘improvised’ or creative adaptations. As McGovern points out, communal recitation of an unfixed text led by a single reciter is quite possible when much of the material (e.g. the formulas) is fixed. And as Shulman points out, the memorization and communal recitation of fixed texts does not mean that there was not also individual recitation and adaptation.

In contrast to McGovern and Shulman, Allon in his study wishes to defend the position that the Buddhist tradition from an early stage was in principle concerned with the composition and transmission of fixed texts to be memorized for communal recitation. Yet he acknowledges that the differences we find between versions of what the tradition presents as the same text⁶ must mean that the texts were nonetheless in practice not fixed, in so far as they were deliberately adapted and changed:

[T]hose who transmitted these texts consciously reworked them to improve them, to make them more compelling, to make them better suit their purposes and their understanding of their audience’s expectations, to make them better reflect their understanding of the Buddha and his teaching, sometimes also creating new sutras on the basis of familiar elements.

Allon 2021: 112

Allon speculates that the authority to make changes must have lain with a few senior reciters, and remains resistant to the application of anything resembling the Parry-Lord theory to early Buddhist literature. And yet his acknowledgement of deliberate change is perhaps to move significantly towards the position of McGovern and Shulman, in so far as their position does not entail the claim that the period of ‘improvisation’ or ‘creative adaptation’ endured indefinitely (manifestly it did not as the texts became fixed), or that it was open and unrestricted in the sense that any monk, preacher, or reciter of Buddhist texts had the authority and freedom to make changes at will. After all, the Parry-Lord theory relates to specialist poets (‘the singers of tales’) who embody a particular tradition. Applying it in the context of early Buddhist literature is only to suggest that it relates to those who specialised in and oversaw the composition and transmission of the literary tradition. We might envisage a situation in which at the beginning of the tradition it was relevant more generally, but gradually became more and more restricted until the texts became fixed in the form of the various editions that come down to us.

Anālayo remains the most opposed to the application of the Parry-Lord theory to early Buddhist texts. His opposition is related to his resistance to the suggestion that the differences we find in the texts are to any significant degree the result of conscious and deliberate change and adaptation (see below). While he does not deny absolutely that the early texts show examples of intentional change, he sees these as restricted to minimal adaptation to accommodate different audiences and to the introduction of basic ‘commentary’ into a text (Anālayo 2022: 196). He does also acknowledge that ‘various tools of oral transmission and memorization in the form of repetitions, formulas, lists, abbreviations, etc., […] have a creative dimension of their own’ (Anālayo 2022: 196). But for Anālayo this creative dimension, rather than informing the beginnings of Buddhist oral transmission as something positive, emerges in its final phase, producing unwarranted additions and innovations, thereby betraying yet another way in which human memory is fallible, in this case by not being simply reduplicative but also constructive and creative (Anālayo 2022: 96, 136).

2 Playing with Formulas

Returning to Shulman’s book, in contrast to McGovern, Allon and Anālayo, Shulman makes a point of focusing his attention on a single tradition of early texts, namely the Pāli Nikāyas. His justification, in part, is that this allows him ‘to confront another central philological paradigm in contemporary Buddhist studies, which is that the best way to understand the early texts is to compare different versions of them’ (Shulman 2021: 13). Shulman suggests that what drives the comparison of different versions of a sūtra is typically the desire to decide which is earlier, and so closer in time to the words of the Buddha. While this has often been the motivation, it is clearly not universally so. In fact in places Shulman slips into generalizing about scholarly approaches at the same time as appealing to scholars whose approaches differ, as when he cites Jonathan Silk’s 2015 article ‘Establishing/Interpreting/Translating: Is It Just That Easy?’ to counter ‘the common philological approach that assumes that “parallel” texts […] can be compared to each other in order to identify more reliable, authentic, or “earlier” versions’ (Shulman 2021: 7). Or as when he scrutinizes ‘reading practices that are commonly employed by scholars of Buddhism’ when in fact his focus is Anālayo’s work (Shulman 2021: 150–156).

Nonetheless, looking at sūtras within a single tradition allows a better appreciation of their literary construction through the play of common formulas and reveals their intertextuality. It also allows Shulman to suggest that in some cases we might see different suttas within the Pāli Nikāyas as different versions of each other (Shulman 2021: 138, 160). In fact, Shulman does also consider parallel versions from the Chinese Āgamas and other sources. But in doing so his concern is to bring out how parallel versions belonging to different traditions are better understood, not as revealing how texts have ‘developed according to neat, linear hierarchies’ from a hypothetical, now lost, original (as seems to be assumed in Anālayo’s model of layers in the development of textual descriptions of the Buddha’s teaching, on which see below), but rather as alternative, equally possible, versions of each other, formed through the play of formulas. In this context Shulman invokes the botanical metaphor of the rhizome (borrowed from Deleuze and Guttari). Such an image contrasts with that of a tree which suggests the development of various branches from the same root and trunk; the rhizome (as in grass), on the other hand, ‘grows horizontally, branching off in numerous directions and sprouting roots at each new node or hub’, thus illustrating how ‘different elements of a text connect to each other and proliferate through new links of significance’ without reference to a genealogy or the implication of a hierarchy (Shulman 2021: 172–173). (It is not clear, however, that such an image entirely bypasses the notion of a first source.)

In this way Shulman is throughout his study concerned to expose the way in which formulas of various sorts inform the content and structure of sūtras to such a degree that they have a life of their own independent of the sūtras. In this he appeals to a suggestion first made by T.W. Rhys Davids in 1903, that the repeated doctrinal formulas may well be the oldest portions of early Buddhist texts. Shulman extends this to formulas more generally, including narrative formulas. But, as Allon (2021: 115–116) points out, while the doctrinal formulas may have been compiled independently (and we have some evidence for that in various versions of the Saṃgīti-sutta), it is difficult to see how the narrative formulas evolved apart from literary constructions that resembled sūtras, even if they were in effect subsequently added to a pool of formulas that the tradition could draw on.

Emphasising the priority of formulas over fully formed sūtras allows Shulman to understand the formulas as the basic resources of the earliest Buddhist preachers, yet in doing so there is perhaps a danger of creating a misleading opposition. I take it that Shulman would not wish to deny that what preachers preached were sūtras, not formulas, so sūtras still in some sense remain the basic expressions of Buddhist teaching. Moreover, while Shulman’s focus on the play of formulas may reveal something of the creative process of sūtra composition, in places it feels in tension with his concern to explore the literary dimensions of the sūtras. To fully explore their literary dimensions we must also consider individual sūtras in their own right, and not just as alternative rehearsals of the same formulaic material.

Given his uneasiness with the tendency to use comparative studies of versions of sūtras from different ancient schools of Buddhism as a basis for deciding which is earlier and which later, it is odd to find Shulman himself attempting to stratify on literary grounds different sūtras in the Pāli tradition that in part rehearse the same material; he suggests, for example, that a version he considers the more ‘polished’ fixes an ‘irregular’ earlier narration (Shulman 2021: 161–162). The problem is that such judgements are often subjective. Back in 1998 Collins observed (while, like Shulman, decrying the scholarly preoccupation with chronological stratification and neglect of literature):

Story motifs, especially in an oral culture, may often be found in other combinations in other contexts; but one must still analyse particular motifs in particular texts, and attempt to understand those particular texts in their given, as-redacted-to-us form.

Collins 1998: 480–481

For ‘story motifs’ here we might substitute ‘formulas’. Thus, in places Shulman forgets his own literary sensibilities and lapses into the idiom of chronological stratification. For example, in the case of the Alagaddūpama-sutta, he suggests that ‘there is, at best, a weak connection between the opening story and the doctrine’ that is the principal focus of the sūtra, namely, ‘not self’ (anattan). This leads him to raise the possibility of ‘borrowing from an earlier source’ (Shulman 2021: 164–165). In fact it seems possible to detect a clear thread running through the Alagaddūpama-sutta, namely, wrongly grasping what the Buddha has taught: the sūtra begins with Ariṭṭha’s misunderstanding of the Buddha’s teaching on the dangers of sexual activity (that it is not a problem when you engage in it without desire), prompting the Buddha to offer two similes illustrating the problem of not grasping things correctly (the similes of the snake and the raft), before moving on to discuss the doctrine of not-self and how some fail to grasp this correctly and accuse him of teaching a form of nihilism.

Nonetheless, Shulman is very effective in reminding us just how much narrative and storytelling there is in the sūtras. Again and again he draws our attention to how the texts employ story motifs. In this context he brings out how this often revolves around the figure of the Buddha in a manner intended ‘to produce a sense of beauty at his unique being, and to allow devotion to grow in his followers’ hearts’ (Shulman 2021: x); this functions in effect as a kind of ‘recollection of the Buddha’ (buddhānusmṛti/buddhānussati)—hence his title Visions of the Buddha. Yet in wanting to tie the poetic and literary dimensions of the texts to the figure of the Buddha (Shulman 2021: 47) and then contrasting this with the goal of preserving Buddhist teachings (Shulman 2021: 37), he is sometimes in danger of underplaying the manner in which the literary and narrative elements of the sūtras equally serve the purpose of preserving and communicating Buddhist doctrine. In this he overlooks the contribution of Steven Collins, who long ago bemoaned the scholarly failure to study the Pāli sūtras as literature, complaining of ‘the dour-faced and humorless positivism with which these texts are so often read’ (Collins 1998: 494). In his own work Collins was particularly concerned to bring out how ‘systematic thought’ might be expressed through ‘narrative thought’ (Collins 1998: 121). Thus for Collins, a sūtra such as the Cakkavattisīhanāda-sutta ‘tells a witty story, by turns pleasantly farcical and fearsomely imaginative, with some familiar doctrinal motifs in unexpected narrative settings; the whole parable being a disbelief-suspending morality tale’ (Collins 1998: 495).

3 Anālayo’s Theory of ‘textual memories’

Referring to the same article by Silk (2015) as did Shulman, Anālayo rejects the quest for the Ur-versions of the sermons of the Buddha, acknowledging that we cannot hope to recover the actual words spoken by the Buddha on a particular occasion. Nonetheless he suggests we must still think in terms of ‘a starting point’, albeit one that ‘can no longer be convincingly reconstructed’ (Anālayo 2022: 154). What we have access to now are the ‘textual descriptions’ presented by Buddhist tradition as the teachings of the Buddha; these textual descriptions are the product of ‘a prolonged period of textual transmission’; by employing the methods of historical-critical study, he argues, it is ‘possible to discern layers in the development of such descriptions’ (Anālayo 2022: 179). This does seem to assume a stemmatic model for the development of sūtras of the kind which Shulman especially calls into question.

Unlike Shulman and Allon, Anālayo doesn’t really offer a theory of oral composition, only one of transmission in terms of remembering, or as he puts it, ‘textual memories’ (Anālayo 2022: 195). Anālayo identifies among scholars of early Buddhist literature a ‘tendency to opt too easily for intentionality as the reason for change’; for him this is to overlook the ‘shortcomings of memory’ (Anālayo 2022: 183–184). In fact, he suggests, the variations we find are due to the ‘natural vagaries’ (Anālayo 2022: 169) or ‘natural fluctuations’ of oral tradition that do not require or indicate the intervention of ‘redactors’ or ‘editorial committees’ (187). For Anālayo, what we are in the main confronted with in the different versions of ancient Buddhist texts is evidence of ‘the limitations of human recall’ (Anālayo 2022: 191). (An exception to this rule is the intentional ‘improvement’ and ‘embellishment’ of stories illustrating Vinaya rules; see Anālayo 2022: 27–36.)

In chapter three of his book, entitled ‘Memory Errors’, Anālayo offers eleven instances of variation which he argues are not to be seen as examples of deliberate editorial intervention, but as examples of how shortcomings in human memory have introduced errors into the textual tradition (Anālayo 2022: 76–88). The problem here is that certainly in ten out of the eleven cases we might dispute that what he presents as an ‘error’ is really such:⁷ elements that appear puzzling or awkward might exist in a text for various reasons; labelling them ‘errors’ is to prejudge the matter. In one instance the ‘error’ is essentially manufactured by Anālayo’s English rendering of the following:

yaṃ vitakkaṃ ākaṅkhissati taṃ vitakkaṃ vitakkessati, yaṃ vitakkaṃ n’ ākaṅkhissati na taṃ vitakkaṃ vitakkessati. acchecchi taṇhaṃ, vāvattayi saṃyojanaṃ, sammā mānābhisamayā antam akāsi dukkhassā ti.

MN I 122.2–5

Anālayo sees a syntactical problem in the use of the future in combination with the aorist. To convey this ‘problem’ he renders the sentence as follows:

Whatever thought they will want, they will think that thought; whatever thought they will not want, they will not think that thought. They cut off craving, did away with [any] fetter, and by rightly penetrating conceit made an end of dukkha.

Anālayo 2022: 81; emphasis added

But the use of the future to emphatically affirm a present state of affairs is surely quite normal (cf. Coulson 1992: 207). That there is no problem here becomes apparent if we supply the implied logic of the sentence and translate (beginning with the previous sentence ayaṃ vuccati, bhikkhave, bhikkhu vasī vitakkapariyāyapathesu) as follows:

Monks, this monk is called a master of the ways and courses of thought. [It will turn out to be the case that] whatever thought he will want [to think], he will think that thought; whatever thought he will not want [to think], he will not think that thought. [In becoming a master of the ways and courses of thought,] he has cut off craving, done away with [any] fetter, and, by rightly penetrating conceit, made an end of dukkha.⁸

But there seems to me a more general problem with the way Anālayo approaches these ‘errors’ of transmission. In each case Anālayo argues that the ‘error’ has arisen in the Pāli transmission while the (Mūla-)Sarvāstivāda Āgama version preserves a text free of the error in question. This begins to look suspicious: are we to suppose that the reciters in the Theravāda tradition consistently suffered from worse memories than those in the Sarvāstivāda? One could equally argue that these examples reflect how the Pāli textual transmission became fixed relatively early, while the Sarvāstivāda versions show a later, deliberate intervention intended to improve the way a passage reads. Allon (2021: 3) notes that the versions ‘of the Sarvāstivādins and Mūlasarvāstivādins […] are generally far more elaborate than, say, the Pali versions’, suggesting that we might consider that they became fixed rather later. But perhaps the more significant difficulty for the claim that the differences between versions of a text result in the main from failures in memory is that it seems inadequate to explain the most common kinds of difference, namely, the differences in order and arrangement of material, and the inclusion or not of particular formulas. I will return to this presently.

Anālayo suggests that the starting point of oral transmission is ‘when someone says something’ (Anālayo 2022: 199). Anālayo does not tell us whom he understands this ‘someone’ to be in the case of the early sūtras. It would seem he is thinking of the Buddha himself. So, if I understand Anālayo’s theory correctly, the oral transmission of the Buddha’s teachings begins with the Buddha, and in some cases his immediate disciples, speaking words which their listeners then attempt to recall and memorize as best they can, given their failed attempt to employ Vedic techniques of verbatim memorization (Anālayo 2022: 64–65). The early sūtras take shape naturally as ‘textual memories’ based on these earliest recollections. These textual memories are then passed on by oral recitation, with only minimal and occasional intentional editorial intervention on the part of reciters and preachers. Yet in time slight and sometimes quite substantial variations naturally creep in as a result of the shortcomings of human memory. The plausibility of such a model seems to me to rest on three points:

1. That in failing to master and use the specialised techniques of Vedic memorization to pass down their texts, the first reciters of Buddhist texts were forced in the main to rely on their natural ability to recall what the Buddha and his disciples had said.

2. That the kinds of variation we find between different versions of a text are primarily of the sort that are best explained by reference to the shortcomings of human memory.

3. That early Buddhist sūtras are better characterised as ‘textual memories’ than deliberate authored compositions.

To my mind, each of these points is open to challenge. With regard to the first, in a developed oral culture with a tradition of handling complex and sophisticated bodies of learning without recourse to writing, to imagine that the only established mnemonic devices available were the Vedic techniques of verbatim memorization seems to me problematic. It is precisely here that the use of numbered lists (mātikā/mātṛkā; see Gethin 1992) and succinct doctrinal formulas as ways of embodying and transmitting learning come into play. That such devices predate Buddhism seems likely. Cousins (2015: 1–2) comments: ‘At the time of the Buddha, when writing was as yet unknown in Eastern India or at least not used for religious purposes, the creation of such a list [as that of the sixteen stages of mindfulness of breathing] must have been the equivalent of today’s authoring a book.’ This leads us back to Rhys Davids’ focus on the repeated doctrinal formulas and to Shulman’s play of formulas.

Regarding the kinds of variation we most commonly find in different versions of the texts, I can only side with Shulman, McGovern and, before them, Cousins: the rearrangement and reordering of formulaic material, the omission, addition, or substitution of one formula or another do not look like errors that have arisen in the course of trying to recall a fixed memorized text. In fact, they do not look like errors at all; they read as deliberate attempts to creatively explore and communicate the Buddha’s teaching. This relates to the third point: different versions of early Buddhist sūtras are better understood not as more or less accurate textual memories of what the Buddha said on one particular occasion, but as carefully considered compositions, each designed to communicate an understanding of his teachings and significance. (To be clear, this does not preclude their reflecting accurately what the Buddha taught.) This seems to me the best way to make sense of the kinds of variation we find in the versions of the Mahāsudassana-sutta (Gethin 2006) and Mahāpadāna-sutta, in the sūtras of the sīlakkhandhavagga (within the Pāli tradition; see Shulman 2021: 54–73), and in the sūtras across traditions setting out the different permutations of the gradual path (Gethin 2020).

4 Versions of the Satipaṭṭhāna-sutta

None of the four authors considers the example of the Satipaṭṭhāna-sutta directly in the works under review, but looked at from the perspective of Shulman’s ‘play of formulas’ (and of Cousins’ and McGovern’s and even Allon’s theories), it seems clear that it is problematic to think in terms an ‘original’ or ‘first’ Satipaṭṭhāna-sutta delivered by the Buddha on one single occasion long ago, which has been either well or not so well recollected in the various versions that come down to us. Even though the two surviving versions both choose to frame the sūtra as delivered by the Buddha on a particular occasion when he was staying in Kuru in a small town called Kammāsadhamma,⁹ it is rather unlikely that the Buddha expounded his teaching on the applications of mindfulness in the way set out in the sūtra on just one single occasion.¹⁰ What comes down to us, then, are various possible versions, all ‘authentic’. As Kuan (2008: 145–154) and Anālayo (2011: I, 73–97) have documented, the Pāli version offers six ways of ‘observing the body’ (kāyānupassanā), using familiar formulaic material found elsewhere in the early texts; (1) mindfulness of breathing, (2) knowledge of postures, (3) full awareness in undertaking routine activities; (4) reviewing the parts of the body, (5) reviewing the body as constituted by the elements, (6) visualising the body as a corpse in various stages of putrefaction. The Madhyama-āgama version (preserved in Chinese translation) suggests all these (although in a slightly different order and with some other variations), but adds several more ways, again drawing on formulaic material found elsewhere: (1) two methods for dealing with distracting thoughts (using formulas in part paralleled in Theravāda sources in the Vitakkasanthāna-sutta),¹¹ inserted near the beginning, which makes good sense; (2) observing the feelings that pervade the body in the attainment of each of the four dhyānas; (3) observing light, and (4) reviewing the sign of contemplation. These latter additions are placed after the practice of mindfulness of breathing, which again makes good sense if this is taken as essentially an exercise in calming the mind; the addition of the dhyānas also finds a parallel in the Pāli Kāyagatāsati-sutta. A further version of the Satipaṭṭhāna-sutta is found in a Chinese translation of the Ekottarika-āgama. This offers a somewhat reduced list, but adds awareness of the body’s orifices. Similar types of variation (additions and omissions of formulaic material) are found under observing dhammas (dhammānupassanā). Moreover, as Anālayo observes (2011, I, 80), reduced versions of ‘observing the body’ are also found within the Pāli tradition in the Paṭisambhidāmagga and Vibhaṅga. Are we really to read all this as reflecting reciters’ confused and unclear memories of what the Buddha taught under the heading of ‘observing the body’ and ‘observing dhammas’ in the context of the applications of mindfulness?¹²

If we imagine that the Buddha, himself educated in a culture in which learning was transmitted orally, made use of numbered lists and succinct doctrinal formulas in his teaching, then it is not impossible that the part of the play of formulas that involves these reflects something of his own style of teaching and that of his earliest and immediate disciples. We might then imagine that these doctrinal lists and formulas gradually became combined with narrative formulas to produce literary compositions—the first sūtras—which, by setting those teachings in compelling narrative frameworks, more effectively communicated the Buddha’s teachings. If we then imagine these sūtras are likely to become more fixed in form as time goes on, such that the authority to change and edit them rests only with recognised reciters and those who oversaw the early Buddhist ‘councils’, the saṃgitikāras,¹³ we introduce elements of Allon’s model also.

5 Conclusion

Shulman in places wishes to distance his own theory from that of Parry-Lord, emphasising that he is not attempting to revive it, yet to my mind his theory is best understood as an adaptation and refinement of it to make it more applicable to early Buddhist oral literature. These adaptations are meant to accommodate some degree of verbatim memorization in the process of composition and transmission. Yet the essentials of the Parry-Lord theory seem to me to remain in his ‘play of formulas’: the composition and transmission of early Buddhist texts is at every turn informed by stylized repetition and the use of narrative and doctrinal formulas. And yet their use is not merely mechanical, not merely formulaic; they are continually adapted to new circumstances, facilitating the application of the Buddha’s teaching to different situations.

Thus with regard to the three issues I highlighted at the beginning of this review article, it seems to me that the nature of the variations between versions of the same sūtra and the way the same formulas are used and repeated in different texts make it difficult to avoid the conclusion that in the earliest phase of composition and transmission sūtras were not taken as final fixed formulations. Following on from this, it also seems to me equally difficult to avoid the conclusion that the earliest reciters of Buddhist texts were in some sense ‘authors’ who, working within the constraints of tradition, creatively adapted doctrinal and narrative formulas and frames in order to better communicate the Buddha’s teaching. In part this may reflect the way the Buddha himself taught. Whether we call this creative process ‘improvisation’, whether this creative process took place ‘in performance’ or prior to it or in some way that combined elements of both are moot points. And, as I have said, none of this of itself need suggest that the sūtras as we have them do not contain the authentic word of the Buddha.

The three works under review adopt opposing positions on a number of issues. In some respects these might be reconciled to a degree, in others not. All three bring to the discussion a wealth of examples, details and ideas. None, in my view, succeeds in providing a finally satisfying model of how the oral composition and transmission of early Buddhist texts actually worked. Perhaps such a model must remain elusive. Still, whether or not we agree with Shulman’s book in all points, details, and formulations, his notion and analysis of ‘the play of formulas’ does bring us closer to an appreciation of the way in which the oral composition of early Buddhist texts might have operated, though some will continue to be reluctant to countenance this way of understanding the composition and transmission of early Buddhist texts.

Reading Allon’s, Shulman’s and Anālayo’s monographs together reaffirms my own sense that Lance Cousins was essentially right: the most plausible model for the composition and transmission of early Buddhist texts involves some form or adaptation of the Parry-Lord theory of oral literature. If this conclusion should seem to be born of a disciple’s desire to honour his teacher’s memory,¹⁴ I can only say that those who come up with good ideas deserve to be honoured.