² Such as they are, they may simply result from the fact that the earlier definition was originally framed in French. Parry's earliest work was written in French; in Parry (1971) it appears in English translation.

³ This is a slight simplification: see Parry (1971: lx–lxi) for a detailed discussion of his position.

⁴ For a general comparativist critique see Finnegan (1977: ch. iii); for discussions in Indian contexts see Smith (1977 and 1989a), and Blackburn (1988: xvii–xxii, 23–6).

⁵ A preliminary version of some of the following material was presented at the Second International Seminar on Rajasthan, held under the auspices of the Rajasthan Studies Group in Udaipur in December, 1991.

⁶ The previous episode, that of the mare (11. 252–311) was the first which Parbü recorded for me, and it contains various confusions resulting from the unfamiliarity of the circumstances in which he had to perform.

⁷ Smith (1991: 267).

⁸ L. 316 = 1.377, 1.317 = 1.376, 1.324 = 1.379, 1.328=1. 364, 1.340 = 1.375, 1.344 = 1.365, 1.345 = 1.366, 1.357 = 1.371, 1.358 = 1.372, 1.359 = 1.373, 1.381 = 1.392, 1.382 = 1.393, 1.394 = 1.395.

⁹ See for example Hopkins (1901), Appendix A; Sharma (1964); Sen (1966); Brockington (1985) and the articles cited in the bibliography there; Smith (1987); Ingalls and Ingalls (1991).

¹⁰ My own computer, for example, is a very ordinary 486DX PC bought in 1993 and running the Linux operating system (a version of Unix); using the egrep program it can search texts at speeds of approximately 10,000 lines per second.

¹¹ Correction is always necessary since even under favourable circumstances OCR programs typically attain an accuracy of only 98% or 99%. (These may sound like high figures, but in normal English text they will lead to about one error in every ten words.)

¹² From ftp: //c c f tp.kyoto-su.ac. jp/pub/doc/sanskrit/mahabharata.

¹³ For some reason a full stop (period), not a space.

¹⁴ For the electronic text, and a detailed account of the process of correcting it, see http: //bombay.oriental.cam.ac.uk/index. html.

¹⁵ See Smith (1987).

¹⁶ In addition to the numbers and the system of parentheses, some less easily isolated formulaic patterns are identified in the notes to Annex 2: see for example the notes on 2d, 4b, 9d, lOd, 11c, 11d, 22b, 23b, 36d.

¹⁷ Of course, a literate poet using bardic epic as his model might be expected to produce verse showing comparable characteristics. In this context note that the Rāmāyaṇa reveals broadly similar preferences in its use of words for ‘he/she said’.

¹⁸ Duggan (1975), for example, finds himself defending one arbitrary cutoff point against another: he takes issue with Chasca for counting as formulae within the Cantar de mio Cid only ‘phrases which occur three times or more’ (p. 75), but his reason for doing so appears to be principally a desire to raise the resulting figure for ‘formula density’ above the 17% claimed by Chasca, as ‘I have conjectured that the threshold of semantic formula density below which one should conclude that a work was composed in writing is 20 per cent’ (p. 82).

¹⁹ That of Parbū's wife's brother Saṅkarjī, of village Sovaniyo near Pali, recorded in arthāv only in 1976. This performance also contains a different formula for Mirzā's wives tearing their hair.

²⁰ For instance, it is used in the course of the description of Pābūjī‘s battle against Ravana in Laṅka in a performance by the brothers Javārjī and Rānā of village Catāliyo (20°45', 73°20'), recorded sometime prior to 1976 by the Jodhpur branch of the Sangeet Natak Akademi (Akademi tapes, 83 6).

²¹ As initially suggested in Smith (1987).

²² Bearing in mind the purely literary background of the ‘classical’ textual criticism that forms the basis of that edition, it would be unsurprising if it ran into difficulties when used on a text whose origin and development were so largely oral.

²³ ‘Text’ has become such a problematical term that I should probably clarify what I mean: I am not referring to, say, ‘thetext’ of the epic of Pābūjī (which of course does not have ‘atext’), but rather to the text produced in any given oral performance by any given bard. Normally such texts are lost as they are performed, but occasionally one is committed to tape or to paper by a passing scholar.

²⁴ pp. 272–3 above.

²⁵ And, as noted above, some of them at least are employed by other bards also.

²⁶ In the case of Pābūjī arthāv performance, very free.

²⁷ As for pābū pāla ( – – – ∪), we may note (a) that, though it would perfectly fit the first cadence, it is never used in that position, (b) that its meaning is unclear (it has nothing to do with pāl ‘to protect’), and (c) that, though it is deficient by one mātrā, it is used in precisely the same types of position as all the other ‘4+4’ epithets. Taken together, these facts strongly suggest that the present form of the second word derives as a result of sound-change from an earlier trisyllabic form in – ∪ ∪ (e.g. *pāhala).

²⁸ In each case the formula too consists of a single word, but this is no attempt on my part to pull the wool over anybody's eyes: the formula ending the first pāda is an absolutive (with the sense ‘having encountered’), and those ending the second and third pādas are five-syllable vocative compound nouns. All three are thus complex forms comparable to phrases in less synthetic languages. See further my comments on ‘patterns of grammar’, pp. 278 9 below.

²⁹ Since writing this, I have seen Lauri and Anneli Honko's paper on multiforms in epic composition (Honko and Honko, n.d.), in which it is stated that formulae within the Siri epic of South Karnataka similarly vary from occurrence to occurrence: ‘the most interesting observation on formulas so far is their variability: even short two- or three-word formulas tend to develop an amazing number of variants. This discovery partly contradicts the postulated formulaic economy according to which formulas are there to save unnecessary variation and offer the best choice’ (p. 36).

³⁰ As well as that caused by presence/absence of fillers just mentioned: note eka and ghanī, present in 1. 314 but absent in 1. 1485.

³¹ 11. 314, 1415, 1472, 1485, 1489, 1518, 2707, 2709, 3160, 3172, 3259, 3692, 3706, 3971, 4040.

³² 1.11.86d, 1.86.15d, 2.43.23d, 2.68.20d, 3.13.78d, 3.152.11 d, 3.176.27b, 3.183.23d, 5.37.53d, 5.89.27d, 5.122.55b, 5.135.15d, 8.27.62f, 11.16.28d, 12.8.14d, 12.254.45b, 15.13.2d, 15.16.9d.

³³ kas tam nārcitum arhati (‘who would not honour him?’), 1.86.15d and 3.183.23d.

³⁴ There are other indications that in anuṢṭubh verses in the Mahābhārata the natural maximum unit for formulaic analysis may be the line, not t he pāda: see Annex 2, note to 5b, and note the 110 occurrences of atrāpy udāharantĪmam; itihāsam purātanam, 11 of atraivodāharanlīmam; itihāsam purātanam, etc.

³⁵ 2.19.16c, 2.53.21a, 3.46.11c, 3.48.36c, 3.84.7e, 3.259.13a, 5.51.6a, 5.149.6a, 5.196.3a.

³⁶ 5.46.7c, 5.56.29a, 12.29.10a.

³⁷ See pp. 273–4 above.

³⁸ pp. 276–7 above.

³⁹ See Smith (1989b).

⁴⁰ paga melai (‘sets foot’) is strictly not an intransitive verb, but it functions like one.

⁴¹ 11. 2639, 2677, 2679; 3370, 3376.

⁴² It appears from van Buitenen's translation of this stanza that he takes the word as a vocative addressed to Janamejaya by the narrator VaiŚampāyana; however, it is in fact a sandhi-form for the nominative bharataŚresthah, and so applies unquestionably to DuhŚāsana.

⁴³ 8.49.3d, 8.56.8b, 9.29.32b.

⁴⁴ Ignoring the 77 wholly iambic lines (∪ – ∪ – ∪ – ∪, ∪ – ∪ – ∪ – ∪) that occur: 5.36.14, 12.309.32–69. Hopkins (1901: 239) writes, ‘in general three final iambs are avoided’. Of wholly iambic lines he comments that they are ‘certainly later than other epic forms, of çloka’ (p. 238); presumably they form the origin for the yet later pañcacāmara metre, e.g. the Śiva stotra attributed to Ravana beginning

jaṭāṭavīgalajjalapravāhapālitasthale

gale 'valambya lambitāṃ bhujaṃgatuṇgamalikām

ḍamaḍḍamaḍḍamaḍḍamanninādavaḍḍamarv ayaṃ

cakāra caṇḍdatāṇḍavaṃ tanotu naḥ Ṣivaḥ Ṣivam

⁴⁵ Monier-Williams's dictionary under suyodhana- clearly interprets the names so that the polarity of compliment and contempt is reversed: ‘“fighting well,” euphemistic N[ame] of Duryodhana’. This is wrong: a v¯rttika to Pāṇini 3.3.130 explicitly accounts for the names in the senses I have given. If Monier-Williams's interpretation of the names had ever been current in India, attempts to search for distribution patterns based on sense would be pointless, but I know of no evidence pointing in this direction. The monumental Sanskrit-Sanskrit dictionary Vachaspatya of Taranatha Tarkavachaspati (Calcutta, 1873–83) defines suyodhana-correctly and evidently in full knowledge of the vārttika to Pāṇini's rule. (The Mahābhārata commentator Nilakantha does not comment on the name, at least not on the nine occurrences of it in Book 1.) Even among European dictionaries, Monier-Williams's interpretation seems to be unique.

⁴⁶ Of the 25 occurrences in triṣṭubhs, by contrast, 17 appear at the beginning of the line, the one position where the metrical value of the first syllable does not matter.

⁴⁷ Mahābhārata 3.240.44–7 (van Buitenen's translation).

⁴⁸ There is one significant difference from the position described by Parry, and that is that in both the Indian epics formulae are often used that are metrical equivalents (for example, kesara mātā and kesara ghoṛī in Pābūjī; suyodhana- and kurūdvaha- in the Mahābhārata). This contrasts with the ‘thrift’ that Parry found in the Homeric epic: Parry (1971: 276–9).

⁴⁹ See n. 27 above.

⁵⁰ Smith (1991: 282, n. 35).

⁵¹ ibid., 278, n. 25.

⁵² ibid., 24.

⁵³ Feyerabend (1978: 233–4). For further discussion in the context of the Pābūjī paṛ, see van de Wetering(1992).

⁵⁴ Feyerabend (1978: 240–1).

⁵⁵ This couplet really belongs in the preceding ‘Episode of the mare’, where it was accidentally omitted by Parbū. Pābūjī is offering Deval his protection against possible raids by Jindrāv Khĩcī in exchange for the mare Kesar Kāḷamī; in a further couplet, also omitted, he offers her the village of Goḷiyã-Mathāṇiyã to live in.

⁵⁶ Mirzā Khān was the wicked ruler of Pāṭaṇ.

⁵⁷ An expression used of the Rāṭhoṛ rulers of Marwar.

⁵⁸ Literally ‘in’ (mẽ).

⁵⁹ Pābūjī regularly says this when war is about to begin: Parbū's explanation is that he does so to test his companions. The literal meaning of the first line of his speech is ‘take the shelter of the brown hill’.

⁶⁰ Parbū here stumbled over which of three similar formulae to use, and finally selected the wrong one: he should have said, ‘you have made your name immortal in the land’.

⁶¹ The hide of the rhinoceros is believed to contain ready-made shields.

⁶² i.e. instead of putting us to death. According to Parbū, Pābūjī granted this wish.

⁶³ The Autumn festival of lights.

⁶⁴ i.e. the fast-day following Dīvāịī.

⁶⁵ A very holy pilgrimage-centre in Rajasthan.

⁶⁶ Ramdev is a very popular Rajasthani folk-deity: see Binford (1976).

⁶⁷ A name for the district in which Koịū is situated.

⁶⁸ The ‘five pīrs’ of Rajasthan are five celebrated folk-heroes. Parbū has omitted Gogo, but otherwise his list coincides with that found in a popular dūho (couplet) quoted by Māheśsvarī, Hīrālāl(Rājasthānī bhāṣā aur sāhitva (Vikram Saṃvat 1500–1650), Kalkatta, 1960, 272)Google Scholar: Pābū Harabhū Rāmade, Mãgailyā Mehā/pãcũ pīra padhārajyo, Gogāde jehā. For Haṛbū see Crooke (1920: 327–30); according to Naiṇasī, Meyo Mãgaḷiyo (Maiharāja Mãgaḷiyãṇī ro) was his father (Sākariya, 1960: 346–8).

⁶⁹ In one case (5.102.28d) ca gṛhān.

⁷⁰ The remaining five (which are all compounded) occur at the beginning of a second pāda: 1.92.5b, 3.61.52b, 3.286.9b, 8.5.1d, 12.66.36b.

⁷¹ = Bhīṣma.

⁷² = Dhṛtarāṣṭra.