SOME QUESTIONS OF PRACTICE IN THE NOTATION OF RECONSTRUCTED IE FORMS1 THE CARL D. BUCK UNIVERSITY OF CHICAGO HE significance of our reconstructed forms, precisely what is intended by them, is an old question which some of us discussed briefly at the St. Louis Congress of Arts and Sciences of 1904 (cf. the publication, vol. III, pp. 35, 57), and which has been elaborated in the long article of E. Hermann, KZ 41. 1 ff. There is not only a difference in the views expressed by various scholars, but often in the case of the same one an apparent discrepancy between the interpretation of reconstruction. that is explicitly professed and that which must be inferred from his practice. For example Meillet, Introd. 24 ff., insists on the unreality and the purely formulaic character of the reconstructions, saying that these are nothing but convenient formulae for given correspondences (similarly Oertel, Lectures 128, and others). Yet throughout the work he is constantly, like any other scholar in the field, asserting or discussing the sound, form, or type that must be assumed for the parent speech to account for given correspondences, and whether this and that type is inherited from the parent speech or an innovation. Again, not believing in the existence of three guttural series in the parent speech, he does not recognize the 'plain velars' of other scholars, whereas from his professed principle one would expect him to have the least compunction in admitting them as convenient formulae for a well-known set of correspondences. The fact is, of course, that to him, and to all, the reconstructions, while mainly useful as formulae, are still something more than mere formulae of correspondences, they imply a certain. interpretation of these correspondences, a conviction or a provisional theory regarding their approximate common starting point. I believe that the difference in the principles professed is largely verbal, and that there is no real disagreement as to the purpose and the significance of the reconstructed forms. The reconstruction of the parent speech per se is not our object. It is neither possible nor important. We have no ambition to speak it, 1 Opening of a round table discussion at the meeting of the Linguistic Society of America, Dec. 29, 1925. nor shall we ever have opportunity to read it. The forms we reconstruct are at best disiecta membra, probably somewhat disparate in regard to the precise period reflected, and, even taken separately, of all degrees of approximation to reality. So that they are in one sense formulae in essence and object. On the other hand, it is not true that the reconstructions are merely convenient formulae of observed correspondences, or that the correspondence only is the object of our investigation (Meillet, Introd. 24 'la correspondance seul est donc objet de science'; but his whole procedure shows that he cannot really mean what his words seem to say). Rather must we assert that the correspondence by itself is of small consequence. It becomes important only when we interpret it in such a way as to indicate which sound (or form, or usage) in the correspondence is the more original, what has been the direction of the change, in short the history that lies back of the earliest recorded forms and reflects light on them. The reconstruction gives this interpretation in condensed form. It is not the ultimate object (which is the light reflected on the history) but it is a step nearer to it than the mere correspondence. It is the convenient intermediary. Such a formula as IE e=Skt. a, Grk e, Lat. e, etc. is clearly not a mere statement of the correspondence between the Sanskrit, Greek, and Latin vowels, but it conveys the assumption that it is not the Sanskrit a which most nearly reflects the sound of the parent speech, as was once believed, but the e of Greek and Latin. Without pretending to define the precise quality of this vowel in the parent speech, which would be impossible even for the Greek and Latin vowels, we mean that it was within the range of those vowel sounds that are commonly indicated by the letter e. This IE e, or an IE *esti is as near an approach to reality as can be expected from any written form. But there is great variation in the degree of approximation that is possible and that is intended by a given formula. Not all are so substantially accurate as an IE *esti. The formula IE = Skt. i, Grk. a, Lat. a etc. does not pretend to the same degree of accuracy regarding the vowel of the parent speech as that for IE e. Still it is more than a mere formula of correspondence, for it conveys the assumption (disputed by some) that the correspondence reflects an IE vowel that was different from either i or a. The formula IE dh=Skt. dh, Grk. 0, Lat. f (medial d, b), etc. originally implied the assumption that the IE sound in question was actually a voiced aspirate of the dental series. This is admittedly less certain than the assumption of IE t or d. But the formula may be retained not only by those who believe that the probability is still in favor of dh, but by others who would not give it more than an even chance. It might even be retained, in deference to custom, by those believing that the sound was a fricative. But if there should come to be a consensus of opinion that the sound was ƒ, for example, then the formula should be changed to accord. As it stands the IE dh is more formulaic than most of the reconstructions, yet even in its most liberal interpretation it is something more than a formula of correspondences, for it implies that the sound was of a different order from the t or d. Again, the formula IE q=Skt. k, c, Grk. x, Lat. c, etc. (kravis, κpéas, cruor) was introduced to represent the theory of a third guttural series intermediate between the labio-velars and the palatals. I am one of those who like Hermann (KZ 41. 46 ff.), Meillet (Introd. 66) and others believe that no language or group taken by itself requires the assumption of more than two guttural series, and who are more or less suspicious of the reality of an independent third guttural series. On the other hand none of the attempts to derive the gutturals of the correspondence from either palatals or labio-velars (or to derive the labio-velars from the 'plain velars', as tried by Reichelt) works out in any satisfying way. Hence even those doubtful of its reality may use the q as a convenient indication of this group and the problematical situation involved. So used this would be a case of a formula of correspondence. But it is perhaps preferable to use the indifferent k, as in other cases where one wishes to indicate a guttural whose appurtenance to one or the other guttural series is undetermined. After all, the 'plain velars' took care of only a part of the apparent anomalies in the distribution of the gutturals. Hermann, KZ 41. 62, has suggested that we use two different symbols according as we mean a real reconstruction or a formula. But more than two symbols would be necessary to indicate all the degrees of approach to reality (the absolute reconstructions would be relatively few), and it seems to me that the distinction would be both troublesome (as Hermann admits) and quite unnecessary, in fact attaching too much importance to the question of the parent speech per se. But the use of a special symbol, for example a double asterisk, may well be recommended for certain Proto-Indo-European reconstructions, which it is convenient to employ, but which are on a quite different plane (not merely disparate in small details) from the usual reconstructions. Thus **teres- as the common Proto-IE source of both IE *tres(Skt. tras-, Grk. 7péw) and *ters- (Lat. terreo), only these two being the true coordinates of IE *ei, *bheudh-, etc. Similarly **pele- as the common source of IE *pela- and *plē-. To sum up, the reconstructions are weighted formulae, they are formulae of correspondence plus as much interpretation as the evidence permits. Their purpose is not to furnish a picture of the parent speech for its own sake, but a background of the historical relations. The recognition of this purpose must affect our judgment of what constitutes the best formula. We are not concerned with small niceties of pronunciation in the parent speech, nor whether the forms set up are precisely coordinate in the period represented. We should aim at formulae. which best indicate the sound patterns (to use Sapir's term) and their relations. In particular, we should dispense with the notation of a 'glide' or transition from one sound to another, such transitions as a phonetic record of living speech will show to exist in a sequence of sounds, but which have no psychological value, no existence in consciousness, and are ignored in the alphabetic representation of actual speech. For the most part such fine points are ignored in our reconstructions, but there are some current reconstructions that are needlessly complicated by the notation of assumed transitions or other assumed changes in combination. As the antecedent of Lat. sessus Brugmann and others reconstruct an IE *set to-s, representing the belief that the sibilant which appears everywhere except in Sanskrit has developed already in the parent speech between contiguous dentals. However probable this may be, it is safer and simpler to write *set-to-s, or perhaps still better *sed-to-s as most plainly indicating the origin, which we know better than we know what stage of its phonetic development was finally reached in the parent speech. As the antecedent of Skt. buddha-s, Grk. Tvorós Brugmann set up IE *bhud dho-s, representing the belief that the change commonly known as Bartholomae's law had taken place in the parent speech. The evidence for such a change outside of Indo-Iranian is not conclusive, and it is certain that the mass of European forms resulting from the combination in question do not reflect it. The more serviceable IE reconstruction. is simply *bhudh-to-s. The climax of complicated reconstructions is reached in those that are set up for cases where Greek has a dental corresponding to a sibilant in other languages, e. g. not only *kpi- for Krišw, but also *gdhem- for xov. It is only Greek that has dentals where we expect a sibilant (alleged Celtic examples are wholly doubtful), and it is still possible that they represent a development of the sibilant in connection with a stop, subject to more special conditions that are obscure. But if we wish to assume provisionally the existence of an IE fricative different from s to cover this set of correspondences, we may be satisfied with one symbol, if, as suggested, we ignore Bartholomae's law in the reconstruction. We may use por 3 (as once suggested by Collitz) and write kš, ghs, etc. The relation between x0wv Skt. kṣās, and xauai, Av. zəmō, etc. is better shown by a ghšem- beside *ghem- than by *ĝdhem beside *ghem. In the question of the assimilation of nasals to a following consonant in the parent speech there is much that must be left undetermined (cf. Brugmann, Grd. I. 344). It might be well to ignore it even where the matter seems clear but takes care of itself, and for example write simply *penque not *penque, using only n or m as being the only nasals that are used independently and so belong to the original pattern. Similarly there is something to be said for writing *nisdo-s in accordance with its origin from the zero grade of *sed-, rather than the assimilated *nizdo-s. It is the usual practice in reconstructions to indicate the glides between i or u and a following vowel, using the same symbols as for IE consonantal i or u in general. In regard to the latter, to digress, Brugmann's notation i, had the merit of acceptability in international usage, avoiding controversy over the choice of y, j, v, w. But it is rightly felt that we should dispense with unnecessary diacritical signs. They add to the formidableness of our science to the uninitiate and to the cost of printing. There are many matters of this kind that deserve consideration, for example the simplest notation to indicate syllabic value of liquids and nasals, length of vowels, the different gutturals, etc. But here it does not seem possible as yet to agree on the best substitutes for current practice, whereas in the case of consonantal i and u the solution is simple and already adopted by several scholars. For the consonantal u the w, understood in its English value, is established in international usage in Semitic and other branches, and increasingly so in the Indo-European field. For the consonantal i there is still rivalry between j in its German, and y in its English value. It might plausibly be urged that we should compromise on j, w, as has been done by the International Phonetic Association for the transcription of modern languages. But there are several reasons for preferring y. The English y is the nearest modern representation of a pure consonantal i, such as we assume the IE sound to have been, while the German j has more friction. The y as a consonant sign is unambiguous, not representing a variety of |