less general statistical accumulations of facts actually occurring in the Veda, I have also pointed out that we are at liberty to found on these collected facts inductive observations which may assume the character of real rules. Thus, in our case, we can well understand why there should be none, or, at least, very few instances, where an unaccented word begins a pâda. We should not begin a verse with an enclitic particle in any other language either; and as in Sanskrit a verb at the beginning of a pâda receives ipso facto the accent, and as the same applies to vocatives, no chance is left for an unaccented word in that place except it be a particle. But the one particle that offends against this general observation is u, and the very word before which this u causes this metrical offence is loka. Can any argument be more tempting in favour of admitting an old form uloka instead of u loka? Lokám is preceded by u in i. 93, 6; ii. 30, 6; (asmín bhayá-sthe krinutam u lokám, make room for us, grant an escape to us, in this danger!) iv. 17, 17; vi. 23, 3; 7 (with urúm); 47, 8 (urúm nah lokám, or ulokám?); 73, 2; vii. 20, 2; 33, 5 (with urúm); 60,9 (with urúm); 84, 2 (with urúm); 99, 4 (with urúm); ix. 92, 5; X. 13, 2; 16, 4 (sukrítâm u lokám); 30, 7; 104, 10; 180, 3 (with urúm). Loké is preceded by u in iii. 29, 8; v. 1, 6; loka-krít, ix. 86, 21; X. 133, 1. In all remaining passages u loká is found at the beginning of a pâda: lokáh, iii. 37, 11; lokám, iii. 2, 9 (u lokám u dvé (íti) úpa gâmím îyatuh); v. 4, 11;

loka-kritnúm, viii. 15, 4; ix. 2, 8.

The only pas

sages in which loka occurs without being preceded by u, are lokám, vi. 47, 8 (see above); viii. 100, 12; X. 14, 9; 85, 20 (amrítasya); lokah, ix. 113, 9; lokẩn, x. 90, 14; loké, ix. 113, 72; x. 85, 24.

Considering all this, I feel as convinced as it is possible to be in such matters, that in all the passages where u loká occurs and where it means. space, carrière ouverte, freedom, we ought to read uroká; but in spite of this I could never bring myself to insert this word, of which neither the authors of the Brahmanas nor the writers of the Prâtisâkhyas or even later grammarians had any idea, into the text. On the contrary, I should here, too, consider it most useful to leave the traditional reading, and to add the corrections in the margin, in order that, if these conjectural emendations are in time considered as beyond the reach of doubt, they may be used as evidence in support of conjectures which, without such evidence, might seem intolerable in the eyes of timid critics.

There remains one difficulty about this hypothetical word uloká, which it is but fair to mention. If it is derived from uru, or, as Professor Bollensen suggests, from urvak or urvak, the change of va into o would require further support. Neither maghon for maghavan, nor durona for dura-vana are strictly analogous cases, because in each we have an a preceding the va or u. Strictly speaking, uroka presupposes uravaka, as slóka presupposes

sravaka, or óka, house, avaka (from av, not from uk). That, on the other hand, the u of uru is liable to disappear, is shown by passages such as i. 138, 3; vii. 39, 3, where the metre requires uru to be treated as one syllable; and possibly by ix. 96, 15, if the original reading was urur iva instead of urviva.

The most powerful instrument that has hitherto been applied to the emendation of Vedic texts, is the metre. Metre means measure, and uniform measure, and hence its importance for critical purposes, as second only to that of grammar. If our knowledge of the metrical system of the Vedic poets rests on a sound basis, any deviations from the general rule are rightly objected to; and if by a slight alteration they can be removed, and the metre be restored, we naturally feel inclined to adopt such emendations. Two safeguards, however, are needed in this kind of conjectural criticism. We ought to be quite certain that the anomaly is impossible, and we ought to be able to explain to a certain extent how the deviation from the original correct text could have occurred. As this subject has of late years received considerable attention, and as emendations of the Vedic texts, supported by metrical arguments, have been carried on on a very large scale, it becomes absolutely necessary to re-examine the grounds on which these emendations are supposed to rest. There are, in fact, but few hymns in which some verses or some words have not been challenged for metrical reasons, and I feel

bound, therefore, at the very beginning of my translation of the Rig-veda, to express my own opinion on this subject, and to give my reasons why in so many cases I allow metrical anomalies to remain which by some of the most learned and ingenious among Vedic scholars would be pronounced intolerable.

Even if the theory of the ancient metres had not been so carefully worked out by the authors of the Prâtisâkhyas and the Anukramanîs, an independent study of the Veda would have enabled us to discover the general rules by which the Vedic poets were guided in the composition of their works. Nor would it have been difficult to show how constantly these general principles are violated by the introduction of phonetic changes which in the later Sanskrit are called the euphonic changes of Sandhi, and according to which final vowels must be joined with initial vowels, and final consonants adapted to initial consonants, until at last each sentence becomes a continuous chain of closely linked syllables. It is far easier, as I remarked before, to discover the original and natural rhythm of the Vedic hymns by reading them in the Pada than in the Sanhitâ text, and after some practice our ear becomes sufficiently schooled to tell us at once how each line ought to be pronounced. We find, on the one hand, that the rules of Sandhi, instead of being generally binding, were treated by the Vedic poets as poetical licences only; and, on the other, that a greater

freedom of pronunciation was allowed even in the body of words than would be tolerated in the later Sanskrit. If a syllable was wanted to complete the metre, a semivowel might be pronounced as a vowel, many a long vowel might be protracted so as to count for two syllables, and short vowels might be inserted between certain consonants, of which no trace exists in the ordinary Sanskrit. If, on the contrary, there were too many syllables, then the rules of Sandhi were observed, or two short syllables contracted by rapid pronunciation into one; nay, in a few cases, a final m or s, it seems, might be omitted. It would be a mistake to suppose that the authors of the Prâtisâkhyas were not aware of this freedom allowed or required in the pronunciation of the Vedic hymns. Though they abstained from introducing into the text changes of pronunciation which even we ourselves would never tolerate, if inserted in the texts of Homer and Plautus, in the Pali verses of Buddha, or even in modern English poetry, the authors of the Prâtisâkhya were clearly aware that in many places one syllable had to be pronounced as two, or two as one. They were clearly aware that certain vowels, generally considered as long, had to be pronounced as short, but they did not change the text. They were clearly aware that in order to satisfy the demands of the metre, certain changes of pronunciation were indispensable. They knew it, but they did not change the text. And this shows that the text, as they