¹ Introduction à l'étude comparative des langues indo-européennes⁵ 340 (Paris, 1922).

² Germanic forms quoted are Gothic unless otherwise identified; with respect to the crucial features, the dialects used agree among themselves and with proto-Germanic. The list is in no way exhaustive: it does not include Skt. t = Gmc. s (in clusters from IE t before t) or any reflex of the IE voiceless aspirates; nor does it include sets that involve Gothic specifically, such as where Gmc. b has been unvoiced to f in Gothic.

The components of the various sets are phonemes, obtained by descriptive analysis of phonetic records in the case of living languages or by interpretation of written sources in the case of dead languages. Since alphabet records are often in effect phonemic, we are mostly in the dark about subphonemic differences in earlier stages of a language. In the light of the discussion below, such knowledge, where we have it, may be helpful, though perhaps never crucial. For instance, it is helpful to know that the allophones of Gmc. b were spirants in the positions where set 8 occurs.

³ Of course, IE was not reconstructed from Vedic Sanskrit and Germanic alone. Actually, scholars made use of the fact that the other major IE languages are apt to show one and the same phoneme in 4 and 6 (not Greek, which has, roughly, ttttdth from 1 to 6; see note 6 below) but two different ones in 2 and 6. Still, such multilateral comparison may be viewed as a system of bilateral ones. This is important since it becomes increasingly clear (and not only for Romance; see R. A. Hall Jr., Lg. 26.24) that some of the well-known difficulties about the setting up of family relations between languages disappear as soon as these relationships are interpreted in terms of successive bifurcation rather than simultaneous, multiple split.

⁴ In terms of surrounding sets. Instances where two partially like sets differ in language A, but the environments appear as mutually exclusive only because they are different in language B, are the particular triumphs of the comparative method. Sets 2 and 3 differ only in Germanic, under conditions that are revealed only in Vedic Sanskrit (and, if we compare Germanic with other IE languages, only where the IE accent has left direct traces, as in Greek); hence the startling effect of Verner's discovery.

⁵ In this case, then, the criterion of mutually exclusive environments solely in terms of surrounding sets must be to some extent implemented by saying that if the structure of the environments has been reinterpreted through reconstruction or phonemicization, the distribution of the set in question must be reexamined in terms of the newly described environments. The general theory involved here is an interesting problem in itself. (Brugmann's *zd is unnecessary in any case.)

⁶ This peculiar result is of course due to the fact IE *s has been reassigned to different entities (Lat. r, Gk. 0) under what happen to be identical conditions. In a comparison between Greek and Sanskrit, the set Gk. t = Skt. d, corresponding to set 4 above, would be in a similar relation with th = dh (corresponding to set 6) as is 15 with regard to 13 + 14, reflecting the two independent dissimilations of aspirates in the two languages (see above, note 3). In the case of 15, we have reason to assume that IE *s between vowels was h at one time in Proto-Greek before it was lost entirely. If we had to reconstruct IE from that stage, we would find ourselves in a somewhat better position: 13 and 14 (sharing Lat. s), and 14 and 15 (sharing early Greek h) are partially alike, though 13 and 15 are not.

⁷ E.g. B. Bloch and G. L. Trager, Outline of Linguistic Analysis (Baltimore, 1942).

⁸ Some doubt it. J. Lohmann, in an otherwise very valuable discussion of Steinitz' reconstruction of the Finno-Ugric vowel system on the basis of phonemic interpretation, declares (DLZ 70.201) that here ‘Trubetzkoy's phonologie‘ has undergone its first (!) historical test, and adds that if it proves successful, all our earlier reconstructions, including the traditional work in Indo-European, will be worthless: a sorry misunderstanding of the continuity of linguistics and of the reviewer's own position.