________________
Semantic Randomness and the Comparative Method
Henry M. Hoenigswald
1. The prototypical phonological innovations are mergers, since by Polivanov's Law, phonemes can merge spontaneously but will split only in conjunction with a merger. In the prehistory of Sanskrit an Indo-European *e has merged with *o into a. The merger can be inferred from such facts as that (1) Skt. a will in some words be found to correspond to one Greek vowel (viz. e), while in other words Skt. a will correspond to another Gk. vowel (viz. a). Contrariwise, (2) Gk. h-, word-initially before a vowel, corresponds to a Skt. y- in some words (Gk. hepar, Skt. yakrt 'liver') but to a Skt. s- in others (Gk. hepta, Skt. sapta 'seven'). In the first case (1) it is Sanskrit that has innovated by eliminating a contrast; in the second case (2) it is Greek (phonetic properties are not at issue; only identities and differences). The languages, we say in such cases, are related.
. 2. Two related languages may be related in a special way: they may be ancestor (or older stage) and descendant (or later stage). When (3) one language has all the mergers while the other has none, the second is the ancestor of the first. To the extent that it is true that Sanskrit has only retentions but no innovations when matched against the modern Indo-Aryan languages, Sanskrit is their ancestor. In most-perhaps all-known histories, perfectly pure ancestor languages are not on record, but a collection of all the retentions may always be constructed. The result is known as a reconstruction. The process leading to a reconstruction as we have just described it is the comparative method (which is not 'comparison' in the ordinary sense but rather a technique to tell phonological innovation from retention).
3. Thus, the German Traum 'dream' (with tr-) corresponds to
