preterit third plurals, the medio-passive -ri is appended to certain present forms already characterized by a personal ending. While the ending -er is found in all active third persons preterit, the medio-passive -ri may be omitted without any change of meaning that has yet been detected. In the active ending the vowel precedes the consonant, but in the medio-passive ending it follows. It is possible, however, that the sharp distinction between the two endings is a Hittite innovation. For example, the final vowel of -ri may come from the ending -ti = IE -tai, which was discussed above.

We are now ready to consider h in certain of the medio-passive endings. Hittite iyahhari 'I go' and arhahari 'I come' may represent PIE **iohori and **shohori, which should yield IE *ööri and *röri. If the final i of the ending -ri is a Hittite innovation, we should assume rather PIE **iohor and **shohor, whence IE *ior, and *pār. In either case we have a satisfactory source for the ending of Lat. feror 'I am carried', OIr. labrur 'I speak', etc.

The present first personal endings without r, which we have had to reconstruct for Hittite, namely -ha and -haha, would similarly lead to IE -7; and this, I imagine, is the source of the anomalous first personal

I active ending of the IE thematic verbs. After -ōr had established itself as characteristic of the medio-passive, the originally equivalent -o was re-interpreted as an active.

The above conclusions bring two new arguments in favor of Forrer's theory that Hittite represents an earlier off-shoot from the parent stock than the IE languages strictly so called. 34

The Hittite medio-passive inflection has a far more primitive appearance than that of any IE language; we can still analyse a number of its forms into their elements. Some of these peculiarly lucid inflectional endings seem nearly identical with an early stage in the development of the IE medio-passive, while others (the preterit and part of the imperative endings) must result from an independent development in which IE had no share. No doubt such a state of affairs is conceivable in a language related to the parent speech precisely as are Skt. and Gk.; but it is easier to understand, if we can ascribe the traces of an early stage of IE to an early date of separation, and the features which are totally unlike IE to an independent development.

More cogent is the discovery in Hittite of an original sound which none

34 See Forrer, Mitteilungen der Deutschen Orient-Gesellschaft 61. 26 (1921). Cf. Kretschmer, Glotta 14. 300–19 (1925); Sturtevant, LANGUAGE, 2. 25–34.

of the IE languages have preserved. There is some difficulty in assuming even in a single language the total disappearance of a common sound without leaving a trace. Several sounds of the parent speech are lacking in each of the historical tongues, but each of them is represented more or less consistently by some other sound. Ancient Gk, had no voiced aspirates, but it had voiceless aspirates instead. In Modern Greek there are no aspirates at all, but spirants have taken their place. In Greek & was early lost between vowels and initially before vowels, and in Modern Greek there is nothing left to take its place; but s still survives under certain other conditions. Can we then assume that so rare a phenomenon as the total loss of a common sound has occurred independently in each of the eight branches of thd IE family?

Some will reply that the one speech sound which is frequently lost without leaving a trace is aspiration, and they will adduce the total loss of Latin h in all the Romance languages. Quite aside from the fact that Hittite and PIE h was probably much stronger than Lat. h (like German ch, or even stronger), it was also a far more frequent sound than Lat. h.

But even if we could admit the validity of the Romance parallel, it would but strengthen our case. Latin h was not lost independently by each of the Romance languages; it was lost once for all in Vulgar Latin. Just so PIE h was lost just once during the history of IE, while it was preserved intact by Hittite.

SOME LINGUISTIC NOTES ON THE MĪMĀNSĀ SYSTEM

FRANKLIN EDGERTON

YALE UNIVERSITY

Hindus of reflective tendencies have always been much given to linguistic analysis and speculation. All Indian systems deal more or less with problems of the nature of language, the relation of sound to sense, etc. When the time comes for a general History of Linguistic Theories, the Indian section will bulk large. While none of the recognized philosophic systems can be neglected, the three most important ones from this standpoint are, no doubt, the Vaiyākaraṇikas or grammatical school (most prominently represented by Pāṇini), the Naiyāyikas or followers of Nyāya, and the Mimānsakas or school of the Pūrva-Mimānsā.

The Mimānsā is in essence a code of legal logic, intended to interpret the laws of the Vedic ritual, and deduced—at least in theory-from the texts of the Veda. It holds that the Veda is eternal, uncreated, without beginning or end, and of absolute authority. Vedic commands constitute the whole duty of man, and it is the aim of the Mimārsā to interpret those commands systematically—to deduce logical principles by which the apparently unsystematic Vedic texts may be understood and applied in practice. The logical principles which the Mimāńsā worked out for the ritual code could be, and were in fact, applied to other codes of law, that is what we mean by law in the ordinary sense. This constitutes one of the reasons for the historic importance of the system.

The other chief reason is the linguistic theories and methods of the Mimāńsā. Some of the more spectacular of these, such as the doctrine of the eternality of words, have been repeatedly described in western writings. They have, in fact, attracted an amount of attention quite disproportionate to their prominence in the original sources.

It is easy

1 See e.g. Keith, The Karma-Mimāmsā 37ff., London, 1921. This handy little volume is the most convenient source of information on the system; but it deals almost entirely with its speculative side, whereas in most of the original texts the practical side bulks very much larger. Of this Keith tells us little, and what he tells is not always correct.

to understand this. And yet the more modestly practical part of the Mimārsā also contains not a little material for the historian of linguistic theories. And since this part is far greater in bulk, and also in importance, in the eyes of the Mimānsā writers themselves, it seems worth while to call attention to some of that material.

First, a principle of interpretation which is not only absolutely sound and scholarly, but of great practical value. It is one which many etymologists of the present day need to ponder. The Mimānsā distinguishes two methods of interpretation of words: yoga, or the meaning based on analysis and etymological derivation, and rūdhi, or conventionally establisht meaning. And it lays down the firm principle that 'rūdhi is stronger than yoga'. In other words, it is our duty to define the meaning of a word as used in actual texts or speech; and, if that is possible, to interpret the word always in that way, without regard to its derivation (real or apparent). It is not always possible to do this; and when it is impossible, we may make use of etymological interpretation, but only faute de mieux. This vigorous recognition of what we call ‘philological interpretation, in the true and original sense of that word, as distinguisht from etymological interpretation, is very creditable to the sound sense of the Mimānsakas. Many a western scholar has erred thru failure to live up to the principle that rūdhi overrules yogaor more precisely thru failure to investigate with sufficient thoroness the rūdhi of particular words, that is, their meaning in actual usage.

This rule is called the rathakāra-nyāya, from the stock example given of it in the Mimāńsā texts. A Vedic injunction says that 'a carpenter (rathakāra) shall lay the sacred fires in the rainy-season’. By rūdhi, a carpenter means a member of a particular low caste, which has no right to lay the sacred fires at all (a privilege reserved for Aryans). It might then be suggested that, by yoga or etymology, the word 'carpenter' means 'maker of wagons' (ratha-kāra), and so might be applied to an Aryan who makes wagons. This would seem to simplify the task of an interpreter of the Vedic text; we might expect the scholastic Mimānsā gladly to avail itself of such a dodge. But no: rūdhi prevails, ruat caelum, and tho it forces us to accept a difficult and rather far-fetched

? For instance, the Mimānsā Nyāya Prakāśa of Āpadeva never once refers to the doctrine of the eternality of the word. Yet this is the most widely-known and popular introductory text-book on the Mimānsā. My translation of it, with text, introduction, and glossary, will shortly be issued by the Yale University Press. This article is based entirely on it.

explanation, the word rathakāra cannot be assumed to have any except its standard meaning, namely, a member of a certain low caste.

Coming to matters of linguistic analysis, we find first that words are analyzed into root or stem, and ending. Let not the modern linguist sniff, however much he may deprecate the composition-theory of wordformation (which seems to be enjoying a certain revival at the moment, by the way). For that theory is not involved. The Mimānsā is thinking in psychological, not formal, terms. This is sufficiently shown by the fact, to be mentioned later, that it sometimes finds more than one ‘meaning' in a single unitary ending. It means only that in any inflected word there must be at least two semantic elements, the entity designated by the 'stem' or 'root', and the syntactic relationship designated by the ending. For an inflectional language, I see no possible scientific objection to this.

Furthermore, these two psychological elements cannot exist baldly and independently side by side; nor, obviously, can they be in a state of equal correlation (as in a copulative compound, meaning 'A and B'). One must be in dependence on the other. And according to the Mimārsā it is invariably the root or stem which is subordinate; the meaning of the ending is the principal element, upon which the stem-meaning depends. This might sound at first like mere perverse pedantry. But a very little reflection will show, I think, that if any such analysis is to be made at all, the Mimānsā is quite right. The meaning of the ending cannot depend on that of the base. The syntactic connexion of the noun, for instance, or the modality of the verb: it is these concepts which, logically and psychologically, are predominant, as bringing the word into relationship with other words. The base is connected with other words only thru the ending. And, as the Mimānsā says, what is dependent on one concept cannot be wrenched away from that and made dependent on another; 'otherwise in response to the command “bring the king's servant (rāja-puruşa)”, one might undertake to bring the king! Just as 'king's' is dependent on ‘servant', so the base-meaning of a word is dependent on the meaning of the ending.

This may not seem of very much practical importance; but in the code of logical principles evolved by the Mimānsā it does, in fact, repeatedly come into play, and proves its value in actual practice. It would take

3 The fact doubtless is that the particular rūdhi in question, in this instance, grew up after the date of the Vedic text quoted. Historic change in the meaning of words is not recognized by the Mimānsā. But tho the example may not be entirely a happy one, the principle is none the less sound and valuable.