¹ Cf. also Brugmann, KVG 195, Anm. 3: ‘Solcher Wechsel hat vermutlich sehr verschiedene Anlässe gehabt. Möglicherweise ist s zum teil satzphonetisch, zum teil durch regressive Dissimilation ... geschwunden’. Schwyzer, Griech. Gram. 1.334, says: ‘Die satzphonetische Erklärung (*-os stegeti als *-os tegeti verstanden) ist jetzt aufgegeben’; no reference is given to those who held it. Siebs, KZ 37.277 ff., and others have assumed that this s was an old prefix, which seems to me no explanation, but a case of ‘obscurum per obscurius’. Hoenigswald, Lg. 28.182–5, seemed favorably inclined to the prefix theory, but rightly observed that Siebs ‘could find few plausible examples’ of such a prefixed s before initial vowels, where ‘one would expect to find it’, as well as before stops and semivowels. On the basis of the laryngeal hypothesis, he then undertook to make plausible, in about a dozen etymological groups, an original variation in Indo-Hittite (or Indo-European before the separation of Hittite and the disappearance of the laryngeals elsewhere) between initial s + laryngeal consonant (Indo-European simple s) and laryngeal without s (Indo-European zero), before vowels. Without undertaking to discuss the validity of his etymologies (some of which at least seem to me plausible), I need only say here that, if Hoenigswald was right, his cases would be precisely similar to the well-known cases of ‘s movable’ before stops and semivowels, and easily explainable in the same way. All that would be necessary is to assume that the law of Indo-European sentence saṃdhi, to be stated above, dates from ‘Indo-Hittite’ times, or from a stage of Indo-European before Hittite split off from it. Hoenigswald himself kindly informs me, by letter, that he agrees with this, and that he now regards my explanation of ‘s movable’ as convincing, in view of the Vedic evidence which I shall state.

² The relatively few cases in which the initial is another sibilant than s might be supposed to follow the analogy of s-clusters. But I think it would be more realistic to associate both cases with the general law, applying equally to internal samdhi, which allows (if it does not, in ordinary practice, require) degemination of any double consonant in Sanskrit before a consonantal semivowel. This will be taken up later in my paper.

³ The only other word in the RV showing du- for dus- before sibilant is duṣṭutí- (or dúṣṭuti-) ‘bad (hymn of) praise’, in which the sibilant is followed by a voiceless consonant and so is covered by the Prātiśākhya rule. Even Müller always (in the three occurrences) prints this duṣṭuti-, which is the more surprising since Müller otherwise, in sentence saṃdhi, seems regularly to ignore the Prātiśākhya rule, and prints final ḥ or sibilant before initial sibilant plus voiceless consonant, at least in the variant pādas of the Concordance, whereas Aufrecht omits the final ḥ or sibilant in such cases according to the rule (Ved. var. 2.456, §974). It may be assumed that the mss. generally support Aufrecht, and that Müller unwisely introduced the later Sanskrit saṃdhi, keeping the true Vedic saṃdhi of the Prātiśākhyas only in duṣṭuti-, as far as I have observed. We may be quite sure that the mss. read so, since Müller kept the reading, inconsistently with his usual practice.

⁴ Since m is also a semivowel, a strict preservation of the Indo-European phonemic rule about semivowels would require *ṣuvāsamát (Indo-European -sṃm- after heavy syllable). The AV does not preserve the rule as well as the RV, and its meter is much less perfect. Interpretation of the pāda as a jagatī line (... ṣuvāsamút) instead of a triṣṭubh would be a possible way of saving the situation. But at the moment I prefer to leave this question unsolved.

⁵ The unspecified reference must be, apparently, to AV 12.5.43 (prose), where the edition, however, reads chinátty asya, and Whitney's note in his translation fails to mention a different reading of the mss.; from his grammar, we conclude that this must be chináty asya. (I am grateful to Paul Thieme for identifying the passage from Whitney's Index verborum to the AV, which is not immediately accessible to me.) It should be noted that the Atharvan Prātiśākhya, unlike the others we know, does not prescribe dropping final s (or h) even before initial s plus voiceless consonant, though the AV mss. ‘in general, although with very numerous ... exceptions, practice’ this dropping (Whitney, Transl. l.cxxvi top). As we shall see, they even do the same sporadically before s plus semivowel, and show the like degemination of final plus initial other consonants when a semivowel follows and meter demands it. From Lanman's note 1, op.cit. cxxiii, it may be inferred that Whitney left his notes unfinished, particularly as regards some matters in which the mss. of AV are not consistent with the Prātiśākhya and some other matters classed as ‘orthographic’. From the case of AV 12.5.43 we have to conclude that Lanman did not always make good such omissions, though he did a heroic job of checking Whitney's vast and complicated apparatus.