In a long discussion on "accent and phonology in the Germanic languages" Eduard Sievers proposed a solution for the difference between the endings of Gothic harjis and hairdeis which illustrates his capabilities in analysis and the initial steps towards recognizing allophonic variation in language. The solution came to be known as Sievers's law, now occasionally Sievers-Edgerton's law, after two important articles by the late Franklin Edgerton: "Sievers's law and IE weak-grade vocalism," Language 10. 235-65 (1934) and "The Indo-European Semivowels," Language 19. 83-124 (1943). In its current form it proposes that [i],[y], and [iy] were allophones of PIE /y/, determined by its phonological environment; also that the other Proto-Indo-European resonants underwent the same allophonic variation.
Sievers recognized the "law", but did not describe the conditioning features thoroughly; nor did he apply it to the Indo-European resonants other than /y w/. His recognition was achieved before it was generally accepted that Proto-lndo-European has "several a vowels"; see the excerpt below from Saussure. The segment of Sievers's discussion included here illustrates the concern with accent, with more precise definition of other phonological entities, and with the relations between morphological and phonological structures that was general after Verner's article. It also reflects the increasing preoccupation with metrics.
Eduard Sievers (1850-1932) was probably the most brilliant of the neogrammarians. He had a remarkably fine capability for analyzing language. Among his legendary accomplishments was the identification of the Old English Genesis as a translation from Old Saxon before there was evidence for an Old Saxon poem on Genesis; it was later discovered in the Vatican. Another is his early recognition of the relationship between the accent and the phenomenon clarified by Verner. Like Verner he made his initial suggestion in a letter; this letter, written to Braune 24 March 1874, Sievers himself never published, nor did he proceed to a thorough formulation of the discovery as did Verner. Streitberg published a part of the letter in Germanisch, pp. 287-8:But the verbs in -ja? Should one think of the influence of the accent, for those in -áyati, -ā́yati originally have the accent after the stem syllable just as the preterite plurals do? According to the root vowel and the Slavic accentuation -énŭ the preterite participle also must have had an accent like *numánas. But how are accent and weakening related? The -da of the preterite participle of the weak verbs, such as *nas-i-dá-s is surely also pertinent; cf. Skt uktá etc., Gk plek-tós, etc.; also Gk -ikós = Goth. *ei-ga-s etc. A pity that we still do not have the beginnings of a sensible theory of accent; and who can understand German sound changes without an understanding of shifts of accent?
Fortunately Sievers made ample contributions to linguistics without pursuing this idea. A professor at 21, his long career at Leipzig made it one of the leading centers in linguistics.
The multiple recognition of the role of accent in language change is a further indication of the maturing of linguistics. It also illustrates that discoveries were not simply made by men of genius, but by capable workers in a developing field; and credit goes to those who formulated the discovery. The most notable multiple recognition at the time was that of the "law of palatals." During the 1870's it became clear to a number of linguists -- Ascoli, Thomsen, Verner, Tegnér, Saussure, Collitz, Schmidt -- that Sanskrit palatals were found in environments where the European languages had e. This observation led to the conclusion that more than one vowel must be assumed for late Proto-lndo-European and to the assumption of e a o where Schleicher posited a; it also contributed further to reducing the reliance on Sanskrit for comparative purposes, and instead to the reconstruction of Proto-Indo-European. Like the excerpt presented here, formulation of the "law of palatals" was only one of a great number of fine observations which were being published in the last quarter of the nineteenth century, often to be refined later, as Edgerton did for Sievers's law.
[125-31] A further reason against the assumption of general Germanic syncope of a I take from the inflection of the ja-stems. In order to clarify everything here, however, I must expand somewhat.
It is a question of the explanation of the groups of sounds: -ji and -ei in harjis, haírdeis and the corresponding verbal forms nasjis, sôkeis; on these compare Scherer, Zur Geschichte der Deutschen Sprache 113f.; Zimmer, Zeitschrift für deutsches Altertum 19.419; Amelung, ibid. 21. 230f.; Osthoff, Zeitschrift für vergleichende Sprachforschung 22.89f.
Scherer, whose view Zimmer and Amelung accept, as is known derives harjis and haírdeis from *harijas and *haírdijas through syncope of the a; in complete accordance with the views of Scherer, Zimmer and Amelung interpreted the latter forms as hárijàs, haírdìjas. This in turn presupposes the validity of the law of the Middle High German low accent, which I believe I have disproved for the original Germanic language; according to the principles proposed in Beiträge 4. 522ff., I can make no other assumption than that those forms, presupposing three syllables, were each pronounced hárijàs, haírdijàs. Why shouldn't both have developed similarly to harjis, *haírdjis, as the group of sounds ji is maintained undisturbed in the genitive singular neuter in reikjis, kunþjis etc. or in faírnjin etc.; or why isn't it *hareis, like haírdeis, after the analogy of naveis and gasteis from *navijiz and *gastijiz? [The ending -iz is supported by OE fêt, ON fœtr = *fôtiz; see above 111.] There is also a strong physiological question concerning the assumed loss of the vowel a between the consonants j and s; but I would not like to press this too strongly here, for the discussion necessary for its support might find little support.
I can come to terms still less with the view of Osthoff than with this conception, which one might call the general point of view and which one has to agree is consistent and logical in its point of view. A development of *hairdjas, *harjas through *hairdjs and *harjs to *hairdjis and harjis by means of the development of an auxiliary vowel from the j may indeed be represented graphically but not be made credible for the written language. If this a was actually lost after the j, then according to the laws developed in Lautphysiologie § 22 this j should have become the vowel i, and we would get only *haírdis, *haris. If one wanted to take refuge in the fact that j may not have been a semivowel, but rather a spirant or obstruent, then the development of an auxiliary i could not be conceived, nor its contraction with a completely non-homogeneous vowel. Finally, the objection that Scherer's hypothesis necessarily requires the dative form *haírdija does not hold up any better, for the development of a medial ija to ja is without question in sôkja and similar forms.
But if the assumption of harjis and haírdeisas general Germanic forms raises so many problems, one might simply ask whether they indeed have any claim to this status. North Germanic does not play a decisive role here; its forms niðr, hirðir = Goth. niþjis, haírdeis have the same phonological relationship as do ON biðr, sœkir = Goth. bidjis, sôkeis; hirðir, sœkir however are justified by means of analogies like ástir, nœmir = Goth. ansteis, nêmeis, whose i is certain in the Germanic period. Accordingly in the northern languages there is no sound law which hinders our equating hirðir directly with Goth. haírdeis.
It's quite different in West Germanic. The older Anglo-Saxon, Old Saxon and Old High German attest in weak verbs with short syllables instead of Goth. ji only i, e without lengthening of the preceding consonant. Compare, for example, from the Old Kentish psalter (Ed. Stevenson, London and Edinburgh, 1844) reces 2, 9, seleð 7, 8 etc., seles 15, 10 etc., cweceð 7, 13, ðeneð 7, 13, sites 7, 4 etc., swereð 14, 4, gestes 17, 44 etc., segeð 18, 2 (cf. J. Grimm, Grammatik I⁴, 822f.); Old Saxon fremis, frumid, haƀis, haƀid, hugis, hugid, letid, sagis, sagið, telid; Old High German examples, Grimm Grammatik I⁴, 788. [In Old High German, this law was broken through early by the sound shift, like many other things. The form of the infinitive, of the plural and subjunctive present is carried through everywhere where there was too great difference of the sound; accordingly, sezzis, deckis, like sazta, not *seʒʒiz, *dechis etc., = OE setes, þeces]. The j was merged here throughout to a simple vowel with i of the verbal ending, which was proved to be general Germanic through its causing umlaut in the strong verb at a very early time, before the beginning of the consonant lengthening. Differently among the nouns. Here we have nominatives and accusatives like OE hrycg, mecg, slecg, wecg, þrymm, neuter cynn, webb, bedd, nett, flett, OS hruggi, nt. bed, flet(ti), net(ti), siukki, kunni, webbi, OHG hrukki, nt. kunni, tenni, stukki, giuuiggi, âuuiggi, stuppi, uueppi, betti, antlutti, nezzi, uuizzi, etc.; further, adjectives like OE nytt, gesibb, OS middi, thriddi, luggi, OHG luggi, fluggi, âuulggi, sibbi, nuzzi, accordingly throughout lengthening of the consonant before the ending. I believe that this proves that a j was still present in West Germanic before the final vowel; and since analogy with the verb has just showed us that ji was not possible in West Germanic the final vowel must have been other than i. Where else might this questionable vowel arise other than from the thematic a? As the last general Germanic original form of the short syllable stems we must therefore not assume harjis, kuni, but only *harjəz, *kunjə; in these ə may designate the vowel sound that cannot be determined, which developed gradually under the influence of the j from the thematic vowel a². But also for the long syllable stems non-shortened forms with ia or iə must be assumed. For if the Germanic original form of the neuters had been, for example, rîki or even *rîkî, then the i would have been compelled to drop in Old English and Old Norse as in the imperatives OE sêc, ON sœk = Goth. sôkei, or in the feminines OE bend, hôeð, ON heið-r (with non-original r) = Goth. bandi, haiþi. Details are given about this below. [A further proof for the non-originality of the i in the nominative of the neuters is given by ON hey = Goth. havi. If havi were original Germanic, then the i would have been compelled to drop in North Germanic after short syllable without producing umlaut. Proto-Germanic *naviz regularly yielded ON há-r, as *favaz yielded fár; or in order to give an analogy for the medial position as well, as beside the verbs *haujan, þraujan = ON heyja (OE hêgan), þreyja the preterites haviða, þraviða, i.e. ON háða, þráða regularly occur. ON hey can accordingly stand only for Gmc. *hauja, *haujə (cf. Lappish avje, Thomson 131). On the other hand one may not adduce the analogy of mœr, þý = Proto-Norse *mavi-r, *þivi, for these forms actually have the ending Gmc -î, as will become clear later. But the inflection mœr, meyjar can warn us not to view the nominative-accusative hey prematurely as possible analogical formations to the other cases.] None of the forms discussed can be explained through analogical formation, for apart from them there is nowhere a pattern by which they could have been formed. There are clearly three groups: ja-stems that have remained short, with e in the nominative-accusative, e.g. here and the borrowing ele; those that have become long (through consonant gemination) without vocalic ending, hrycg, cynn; old long syllables with e, hyrde, rîce.
Examination of the genitive singular of the ja stems leads to similar results with regard to the non-originality of the Gothic forms. For in order to maintain haírdeis as common form, one must first of all seize upon the highly questionable assumption of a Proto-Germanic contraction of ie to ī in the penultimate (while the e of the genitive elsewhere did not become i, does not cause umlaut); then however one must explain all West Germanic forms as new formations (OE hyrdes, rîces, OS hirdies, rîkies, OHG hirtes, rîches). Only the North Germanic hirðis, ríkis with some difficulty be compared with the Gothic. Shouldn't one then rather admit that the Gothic haírdeis owes its development only to the specifically Gothic aversion to the sound e, with which was apparently also associated an effect from the nominative? Only in this way do the neuters, with their prevailing genitives in -jis, receive their due: kunþjis, reikjis, faírgunjis, andbahtjis, valdufnjis, gavaírþjis beside andbahteis, valdufneis, gavaírþeis, trausteis, fauramaþleis (see the list in Heyne, Ulfilas § 23). The lack of a nominative similar in sound helped to preserve the older forms here.
Accordingly: the i in Goth. harjis is a remainder of the thematic a; it did not develop from the derivative suffix i or j but was only conditioned in its color by these. The same remainder is found also in haírdeis, which we have to resolve first of all in a previous three-syllabic *herðiiz or *herðijiz, whose treatment corresponds completely with that of naveis, ansteis (cf. p. 125). Goth. naveis is particularly welcome as evidence that the contraction has nothing to do with quantity or with an accentual law dependent on quantity, as we objected above. For the language it's quite immaterial which of two similar contracted vowels had the accent; I need only recall the well-known elementary rules of Greek grammar.
The difference between the short and long syllable ja-stems accordingly consists only in the fact (as Scherer already recognized, though in my view without adequate justification) that the former had consonantal j, the latter vocalic, i.e. syllabic, i in their suffixes.
But where does this difference arise, if it does not depend on the low accent law? One would scarcely assume with no further evidence an earlier, general Germanic existence of this law in the Lachmann version, and a later complete reversal especially in West Germanic! On what should we base our suggestions? We will therefore have to go farther back and hold to the original Indo-European language.
If one may take confidence in the investigations of Benfey (Abhandlungen der Göttingen Gesellschaft der Wissenschaften 16. 91ff., 1871), the suffix ia was used interchangeably in the Veda as monosyllabic or bisyllabic. If one however examines the situation more precisely, a quite definite law stands out: unaccented (without svarita) i or u before a vowel is consonantal after a short syllable, vocalic after a long syllable, without regard for the other accentual situation of the word. Compare examples like the following:
[For the references, see Grassmann. Here I must withhold giving the proof for the above principle at length or discussing the regular exceptions which occur and the violations against it, which in part are not insignificant criteria for determining the age of Vedic hymns. I will only note here that that principle is only a segment of an extensive rhythmical law in oldest Sanskrit and Indo-European, particularly concerning the relationship of the vowels i, u and the semivowels y, v; for years I have been collecting material in support of it. Precise observation of these principles will be useful not only for metrics but also for grammar itself. It turns out, for example, that the lengthenings before r + consonant were still foreign to the living Vedic language; that ūr, īr always developed through ṛ, and the like.] Exceptions are the suffixes beginning with a consonant like -bhyās, -bhyām, -tva, to the extent that these (like word-initial consonants + y, v in general) were used interchangeably after long syllables (after short syllables only with consonantal y, v, i.e. monosyllabic); further certain short syllabic adjectives, especially verbal adjectives with bisyllabic suffix (Grassmann's Part. IV): gádhia, gúhia, gopayátia, carkṛ́tia, tújia, dábhia, dṛ́çia (mádia, yújia?), çásia, çrúitia, hávia (while for example the suffix of the so-called ya-class or the passive follows the rule).
I may report that Hübschmann has recently established the same laws for Old Bactrian, starting from another point of view, so that three languages may already be called upon as mutual witnesses for the great age of the phenomenon. In the remaining languages the old difference seems to have been eliminated early; at any rate none of them shows such an obvious observance of the law as do the three named. But scholars will doubtless succeed in finding remnants of the rule still in details. I'd like to direct attention to one such still: the Greek adjectives hágios and stúgios, which correspond to hádzomai i.e. *hagjomai and similar forms in the same way as do the Sanskrit verbal adjectives to the corresponding verbs.
Probably the most general formulation of the law discovered here may be given as follows: the vowel of a syllable of derivation is and remains heavier after a preceding long than after a preceding short (therefore ia, ua remain bisyllabic in the first case; in the second they become monosyllabic).