Assimilation may be regressive, progressive, or reciprocal. Regressive assimilation is where a sound is assimilated to that which follows it, as in ἕννυμι for ϝεσ-νυμι, from the root vas, or ποσσί, for ποδ-σι (ποδ-σϝ-ι), and γράμμα for γράφ-μα(τ). Progressive assimilation is the converse of this, as in στέλλω for στελ-yω, μᾶλλον for μαλ-ιον, mellis for melv-is, or the Æolic ἔστελλα for ἔστελ-σα. Regressive assimilation largely preponderates in our Aryan languages, progressive assimilation in the Ural-Altaic ones; and it is very possible that Sievers is right[202] in tracing this contrast to the difference of the accentuation, which in Ural-Altaic falls upon the first syllable of the word, while in the parent-Aryan it fell for the most part on the final syllable. Böhtlingk[203] says, very appositely: “An Indo-Germanic word is a real whole of such a kind that the speaker has uttered the whole word, as it were, in spirit, as soon as he has pronounced the first syllable. Only in this way can it be explained how a syllable (or sound) is modified in order to assist the pronunciation of the syllable (or sound) that follows it. A member of the Ural-Altaic race forces out the first syllable of a word—that part of it, namely, which has the accent—little caring for the fortune of the rest; on this he next strings in more or less rude fashion a few more significant syllables, only thinking of a remedy at the moment when he first feels the want of one.” As for reciprocal assimilation, an example of it may be found in the reduction of ai to e quoted above, where both sounds influence one another.

Assimilation may be either complete or partial. There are sounds which can never be thoroughly assimilated to each other, bn, for instance, can never at once become nn, only mn. Partial regressive assimilation meets us very frequently in the classical languages; e.g., λεκ-τός from the root λεγε, ἤνυσμαι from ἀνυτ-, δόγμα from δοκ-; partial progressive assimilation is rarer; e.g., πάσχω for πάσκω from παθ-σκω.

The changes dependent on the presence of a second sound, which are not due to assimilation, are necessarily produced by varying the time needed for pronunciation. Of these the most striking is metathesis. Metathesis must be referred rather to a mental than to a phonetic origin. Our thought and will outstrip our pronunciation, the result being that the sound which ought to follow is made to precede, or else the vocal organs are shaped prematurely for the formation of a sound which ought to be heard later, the consequence being that the sound which should come first has to come last. Metathesis, in fact, is similar to the rapidity, or rather relaxation, of thought which leads us sometimes to write or speak a word which belongs to a subsequent part of the sentence; and it may be of two kinds: either the place of two sounds may be simply inverted, or the second sound may be made to precede the first by two or three syllables. How easily the first case can happen is shown by the phonograph, where each syllable that has been uttered can be reproduced backward by merely turning the handle of the machine the wrong way. R and l are the most subject to metathesis, then the nasals; the other consonants vary according to their relationship to the vowels. More regular than metathesis are the insertion and omission of consonants, as in ἀν-δ-ρὸς, ἄ-μ-β-ροτος, τέτυφθε for τέτυφσθε, rêmus for resmus. Somewhat different are the insertion and omission of vowels, the first of which goes under the technical name of Swarabhakti. This name was imported from the Hindu grammarians by Johannes Schmidt,[204] to mark the growth of a short or reduced vowel from a liquid or nasal, when accompanied by another consonant. Thus ănman, “name,” became ănă-man, and then, by the loss of the first vowel and the compensatory lengthening of the second, nômen and nâmâ. Swarabhakti is, however, incompatible with the acute accent. We may find examples of it in the slow pronunciation which in English turns umbrella into umbĕrella, and Henry into Henĕry.[205] Prosthesis, or prothesis, the insertion of a short vowel at the beginning of a word before two consonants, is another illustration of Swarabhakti. There are many nations which find a difficulty in pronouncing two consonants at the beginning of a word. Thus the Bengali calls the English school yschool, the Arab says Iflatún for Platon, and the Ossete uses a for the same purpose. In other cases, one of the consonants is dropped altogether, as so frequently by children and systematically by the natives of Polynesia. In Latin inscriptions and MSS. later than the fourth century we find forms like istatuam, ispirito, just as in the Romanic tongues we have estar and espée (épée) for stare and spada, or in Welsh ysgol from schola, yspryd from spiritus. According to Wentrup,[206] a is often used as a prothetic vowel in Sicilian; Lithuanian has forms like iszkadà, German “schade,” and Basque and Hungarian prefix a similar aid to the pronunciation. No trace of a prothetic vowel can be found in Latin; in Greek, however, such vowels are very plentiful. Thus we have ἄσταχυς by the side of στάχυς, ἐχθές by the side of χθές, ἰγνύη by the side of γόνυ, Ὀβριαρευς by the side of Βριαρεύς. In Greek, too, as in other languages where prothesis occurs, the complementary vowel may be inserted before a liquid, more especially r, as well as before a strictly double consonant, e.g., ἀμύνω by the side of μύνη, ἐρυθρός by the side of ruber, ὀρέγω by the side of rego. Even the digamma may perhaps take the prefix as in the Homeric ἔεδνον. But it is probable that no other single consonant does so, the apparent exceptions being really explained by the loss of a consonant which once existed along with the one that is left. Ὀκέλλω, for instance, presupposes ὀ-κϝέλλω (Latin pellere), Ἀπόλλων presupposes Α-κϝολιων, “the son of the revolving one” (Sanskrit char, Greek πέλομαι). In other cases we are dealing not with a prothetic vowel, but with a part of the primitive root: ὄνομα, for example, is shown by the Irish aimn and Old Prussian emnes to be more original than the Sanskrit nâmâ or the Latin nomen, and to stand for an earlier an-man; and ὄνυξ, the Latin unguis, the Irish inga, is earlier in form than the Sanskrit nakha and the English nail (nagel)[207]. We may discover a tendency in Greek to adapt the prothetic vowel to that of the root, though it is hardly so regular as in Zend roots beginning with r, where we find i-rith for rith, but u-rud for rud. Sanskrit, like Latin, shows an inclination rather to drop initial vowels than to add them, but even in Sanskrit, Curtius has pointed out[208] the Vedic i-raj-yâmi from raj (rego) and i-radh, “to seek to obtain,” from râdh. As for the loss of a vowel, it is too familiar to every one to need any illustration.

More akin to metathesis is epenthesis, which closely resembles the Teutonic umlaut. Epenthesis is especially plentiful in Greek, where κτέν-yω becomes κτείνω, χερ-ιων χείρων, λόγοσι λόγοις, ἐλαν-ϝω ἐλαύνω, νερϝον νεῦρον. Probably λέγει for λεγειτ is to be explained as resulting from the epenthesis of ι (λεγειτ for λεγετι), just as λέγεις stands for an earlier λεγεσι. Epenthesis thus presupposes a mouillation or labialization in which the articulation of the consonant is absorbed, as it were, by that of the i and u. The greater the participation of the lips and tongue in the formation of these vowels, the greater will be the tendency towards epenthesis.

Lastly, we have to consider the lengthening of vowels, either by way of compensation or before certain consonants. By compensation is meant the additional force with which a vowel is pronounced after the loss of a consonant which followed or preceded it. Thus in Greek the loss of the digamma in βασιλεϝ-ος produced the Ionic βασιλῆος on the one side and the Attic βασίλεως on the other, just as the loss of the yod in πολιy-ος similarly produced πολῆος and πόλεως. So, too, πάνς became πᾶς, δαιμονς δαίμων, ἐφαν-σα ἔφηνα, rĕs-mus rémus, pĕds pês, exăgmen exâmen, măgior mâjor. In certain cases the vowel was raised into a diphthong, as in φέρουσι for φεροντι, τιθείς for τιθενς, ἔστειλα for έστελσα. But a vowel may also be lengthened before liquids, nasals, and spirants when combined with another consonant. If the grave or the circumflex accent fall upon the preceding vowel, the tendency is to lengthen the vowel at the expense of the sonant or spirant following. Hence it is, that whereas in our English tint, or hilt, where the vowel has the acute, the nasal and liquid are long; in kind and mild, on the other hand, where the vowel is circumflexed, it is the vowel (or rather the diphthong) that is long. The vowel, again, may be lengthened to compensate for the loss of a double letter. Thus in Latin we find vīlicus by the side of villicus, from villa, and whereas the grammarians lay down that when ll is followed by i, single l must be written, we find millia in the famous inscription of Ancyra. So, too, the inscriptions vary between Amulius and Amullius, Polio and Pollio, and good MSS. have loquella, medella, instead of loquēla, medēla.

There is another fact to be remembered when we are looking for the application of Grimm’s law—a fact which the law itself ought to bring to our minds. Different languages have different phonetic tendencies; the same sound is not equally affected by phonetic decay in two different dialects or modified in the same way; each language has phonetic laws and phænomena peculiar to itself. Thus, in Greek, σ between two vowels is lost, in Latin it becomes r; in Greek a nasal preserves, or perhaps introduces, the vowel a, in Latin it prefers the vowel e. Because τ between vowels becomes σ in Greek, or sr in Latin is changed into br (as in cerebrum for ceresrum, κέρας, śiras), we are not justified in expecting similar changes in other tongues. In fact we have only to look at the table of sound-changes, known as Grimm’s law, to see that it is just because two languages do not follow the same course of phonetic modification that a scientific philology is possible.

To speak of Grimm’s law being “suspended,” of “exceptions to Grimm’s law,” and the like, is only to show an ignorance of the principles of comparative philology. Grimm’s law is simply the statement of certain observed phonetic facts, which happen invariably, so far as we know, unless interfered with by other facts which, under given conditions, equally happen invariably. The accidental has little place in phonology, at all events in an illiterate and uncultivated age. Literature and education are no doubt disturbing forces: a writer may borrow a word without modifying its sound according to rule; and the word may be adopted into the common speech through the agency of the schoolmaster; but such words are mere aliens and strangers, never truly naturalized in their new home, and the philologist must treat them as such. Native words, as well as words which, though borrowed from abroad, have been borrowed by the people and so given a native stamp, undergo, and must undergo, all those changes and shiftings of sound which meet us in Grimm’s law, in the phonetic laws peculiar to individual languages, or in any other of the generalizations under which we sum up the phænomena of spoken utterance. False analogy, it is true, may divert a word from the path it would naturally have taken; one word may be assimilated to another regardless of its real etymology, or words whose real origin has been forgotten may be modified so as to convey a new meaning to the speaker. But, in such cases, the worst that could happen would be the loss of the true etymology; Grimm’s law would still hold good, and the originals of the existing sounds would be those demanded by the regular Lautverschiebung. So far as the present form of a word like Shotover (for château vert) is concerned, it is to the mere phonologist, as to the ordinary speaker, a compound of shot and over, and in comparing these two words with allied words in other languages the prescribed letter-change holds good. It is only the comparative philologist, who has to deal with the psychological as well as with the phonetic side of language, that needs to know more, and to determine that Shotover is not what it professes to be, but the product of a more or less conscious imagination. In most cases of analogy we have to do with mental as opposed to phonetic assimilation, and they fall, therefore, under sematology, the science of meanings, rather than under phonology, the science of sounds. No doubt we find instances of analogy, like the Greek accusative βεβαῶτα, modelled after the nominative βεβαώς,[209] or the Latin genitives diei, dierum, modelled after the accusative diem for diam, but such instances fall under the laws and conditions of that phonetic assimilation which has been already described. Let us hold fast to the fact that the generalizations, the chief of which are summed up in the formula known as Grimm’s law, are at once uniform and unvarying. If an etymology is suggested, which violates these generalizations, that etymology must be rejected, however plausible or attractive. It is upon the fixed character of these generalizations that the whole fabric of scientific philology rests.

Necessarily similar generalizations may be made in the case of other languages which, like the Aryan, can be grouped into single families of speech; nay, they must be made before we are justified in grouping them together, or in comparing and explaining their grammar and vocabulary. It is not always, however, that the changes of sound are so marked and violent as in the Indo-European. A group of allied languages may be as closely related to one another as the modern Romanic dialects of Europe, and various causes may have combined to give a stability and fixity to their phonology which has made it change but slightly in the course of centuries. This is the case with the Semitic dialects, whose laws of sound-change are extremely simple. Practically the sound shiftings are confined to the sibilants, where the equivalence of sounds is as follows:—

Assyrian.Hebrew.Ethiopic.Arabic.Aramaic.
s (sh)s (sh)s, ´ssh, s, th´s, s, th
´s´ss, ´ss, sh´s
tstststs, ds, dhsts, dh, ’e
zzzz, dhz, d[210]

One or two other general laws of phonetic change may be laid down for special members of the Semitic group; thus, in Assyrian, s before a dental becomes l, and kh is dropped when it answers to the Arabic and Ethiopic weak kh. In the Babylonian dialect, again, k took the place of g, and the n of the other dialects is sometimes replaced by r in Aramaic.