² Cf. especially Eduard Hermann, Lautgesetz und Analogie, and Leonard Bloomfield's review of the same, Lang. 8.220–33.

³ In fact the only words in Sanskrit beginning with saś- or the same combination of consonants with another vowel, are compounds in which the first element is the prefix sa- or the prefix su-, and vriddhi-derivatives of the same; in which, of course, a feeling for the prefix will keep the initial sibilant unchanged.

⁴ Notably, Skt. t, th, d, dh, n, s became ṭ, ṭh, ḍ, ḍh, ṇ, ṣ by the influence of preceding sounds, some of which were consonants and some were vowels.

⁵ But cf. the preceding footnote.

⁶ But conversely, in German the ch is palatal or velar according to the quality of the preceding vowel; this is natural when the ch is final in the word or in the syllable, as in (velar) Bach. When we find velar ch in Rache and palatal ch in rächen, the fact that the preceding vowel governs the quality in Rache indicates a certain semantic unity with the preceding, a feeling that the root is rach-, despite the syllabic division Ra-che.

⁷ The classical treatise on this subject is Maurice Grammont, La Dissimilation Consonantique dans les langues indo-européennes et dans les langues romanes (Dijon, 1895). Grammont goes too far in setting up formulations which operate almost as regular phonetic laws; rather we should limit ‘regular’ dissimilation to a few formulations in specified languages. His examples are virtually all of regressive character, though he professes that his ‘laws’ I-XVI are indifferently regressive or progressive. An examination of the progressive dissimilations which he cites shows that in virtually all there is some special factor to which the reversal of the natural direction of the influence may be attributed. Thus Eng. marble, from French marbre, may have been influenced by the numerous words ending in -ble in English; Ital. propio, from Latin proprius, lost the second r rather than the first by the influence of the many words which begin with pro-; Ital. frate, from Latin fratrem, probably lost the second r rather than the first, by the influence of fratello; etc., etc. Some of these influences are indeed suggested by Grammont, though he fails to see their significance.

⁸ The Real Nature of Dissimilation; TAPA 49.101–13; though this explanation, in less precise form perhaps, was current long before Carnoy's article, and was formulated by Grammont (16) in his first two principles.

⁹ J. Wackernagel, Archiv f. lat. Lex. 15.218–21.

¹⁰ Or the sound may be entirely lost, as in Span. postrado ‘prostrate, humbled’, from Latin prostratus.

¹¹ Cf. the discussion by Otto Jespersen, Language 396–411 (chapter on Sound Symbolism; New York, Henry Holt, 1922).

¹² It is to be noted especially that the apportionment of stetī to stō and of stitī to sistō, is not original, but is a relatively late development.

¹³ Cf. my Sounds of Latin §176.11. There are nearly 200 Latin words in -āris, with a preceding l somewhere in the word; other words in -āris are late remodelings of words in -ārius, or are from Greek.

¹⁴ M. Niedermann, Essais d'Étymologie et de Critique verbale latines 36–45, in Recueil de travaux No. 7, Univ. de Neuchâtel (1918).

¹⁵ Cf. my Sounds of Latin §166.II.

¹⁸ Lang. 8.18–26; also in Sounds of Latin §144.

¹⁷ Lang. 8.25.

¹⁸ Cf., for example, Wilhelm Braune, Althochdeutsche Grammatik² §103 (1891); V. Michels, Mittelhochdeutsches Elementarbuch³ §178 (1921); etc.

¹⁹ And the b, d, g which correspond, were ‘voiceless lenes’. The formulation given above, however, is not misleading, since the processes are as there stated.

²⁰ Calvin Thomas, Practical German Grammar⁴, App. II, §8.3.a (p. 411). E. C. Roedder, who called my attention to this passage, states that Thomas took this explanation from a paper read by Georg Hempl; but I cannot locate the latter.

²¹ This explanation was given briefly by C. Thomas, l.c., and by Hermann Paul, Deutsche Grammatik I. §200 (1916). A few other examples of the same shift, probably due to the same cause, are listed by Paul, op. cit. §202. The various forms of tausend in the older documents, are listed by J. Schatz, Althochdeutsche Grammatik §192, §193, §410, §414 (1927).

²² On dialectal variations, cf., for example, H. Paul, op. cit. §200; L. Sütterlin, Neuhochdeutsche Grammatik 242 (1924); O. Behaghel, Geschichte der deutschen Sprache⁵ §396 (1928).

²³ It would be unprofitable to list examples from the literature; such combinations would be more frequent in actual speech than in literary compositions.

²⁴ OHG zëhanzug '100', with variants in -zeg and -zech, may be understood as included with the multiples of '10'.

²⁵ At that time ‘8‘ was still ahto, with the final vowel.

²⁶ W. Braune, Ahd. Gr.² §278, cites OHG sibunto, niunto, zëhanto, with -to; H. Paul, Deut. Gram. I. §202, quotes MHG siebende, neunde, zehnde, with the voiced dental, coming down into the 18th century, when the d was finally ousted by the t. The probable variation in OHG times can hardly be taken as invalidating the argument which I give above.

²⁷ Cf. W. Braune, Ahd. Gram.² §306 Anm. 5.

²⁸ This is the interpretation of H. W. Sugden, The Grammar of Spenser's Faerie Queene §288; cf. G. O. Curme, Grammar of the English Language III, §49.4.c.(1).a, and Hereward T. Price, A History of Ablaut in the Strong Verbs from Caxton to the End of the Elizabethan Period 30 (Bonner Studien zur Englischen Philologie, Heft 3). As to another view, it seems to me hardly possible that note is here for ne wote, with the acquired meaning of ‘ could not’ (from 'does not know how'), as in French je ne saurais 'I should not know how to', hence ‘I cannot’.

²⁹ R. S. Conway, Italic Dialects No. 95; R. von Planta, Gram. d. osk.-umb. Dial. No. 127; C. D. Buck, Oscan-Umbrian Grammar No. 1.

³⁰ Conway No. 39, von Planta No. 28, Buck No. 3. The first occurrence is clearly written, except that the second letter may be u. The second occurrence has lost the first s and most of the preceding letter, at the end of the line, where moreover there is space for but two letters of the word; the su on the next line is clear.

³¹ Languages differ widely in their tolerance of homophones; Greek, Latin, German have relatively few, French, English, Chinese have many. Yet Chinese must often relieve the situation by adding a class-word or a synonym to the ambiguous word; occasionally English employs a similar device. On the avoidance of homophones, cf. A. Meillet, Sur les Effets de l'Homonymie dans les anciennes Langues indo-européennes, in Bibl. de l'Ecole des Hautes-Études 230.169–80 (1921); J. Gilliéron, Les Conséquences d'une Collision lexicale et la Latinisation des Mots français, ib. 54–74; and my article “No Tresspass” in Latin Linguistics, in Classical Studies in Honor of John C. Rolfe 143–61 (1931).