¹ Thus A. Meillet, Caractères généraux des langues germaniques 27-55 (⁷1949), Les dialectes indo-européens 89-90 (1908); Karl Luick, Historische Grammatik der englischen Sprache §618, esp. Anm. 3 (1914 ff.)

² (1) R. C. Boer, Oergermaansch handboek §143 (²1924). (2) T. E. Karsten, Die Germanen 124 (1928). (3) W. Streitberg, Urgermanische Grammatik §122 (1896). (4) Karl Brugmann, Kurze vergleichende Grammatik der indogermanischen Sprachen §271.7 (1902). In Grundriss der vgl. Grammatik der idg. Sprachen §§792, 799 (²1897), he also assumes stops in gemination. (5) Richard Loewe, Germanische Sprachwissenschaft 1.70 (⁴1933). (6) E. Prokosch, A comparative Germanic grammar 58, 75 (1939). (7) Friedrich Kluge, Urgermanisch 48, 61 (4913).

³ R. Bethge, Laut- und Formenlehre der altgermanischen Dialekte l.§110 (Ferdinand Dieter, ed.; 1898); W. Wilmanns, Deutsche Grammatik l.§26 (³1911); Hermann Hirt, Handbuch des Urgermanischen 1.§55 (1931); Carl Karstien, Historische deutsche Grammatik 1.110 (1939).

⁴ Typical of many is Ernst Kieckers, Handbuch der vergleichenden gotischen Grammatik (1928), where four sets are listed on p. 12, but three on pp. 46-7.

^4a A brief structural analysis of PGmc. /b d g/ has been written by Bohumil Trnka, From Germanic to English: A chapter from the historical English phonology, Recueil linguistique de Bratislava 1.139-49 (1948). Since this journal is not easily available in this country, I am grateful to Paul L. Garvin for having called Trnka's article to my attention and for having lent me his copy of it after I had read a preliminary version of this paper at a meeting of the Linguistic Society in December 1951.

⁵ All Gothic forms cited may be found in Ernst Schulze, Gothisches Glossar (1848). Where necessary, Schulze's readings have been corrected to agree with those in Wilhelm Streitberg, Die gotische Bibel, 1. Teil (²1919).

⁶ Loanwords can be cited for the clusters /pt/ and /kt/: saraipta ‘Sarepta’ and laiktjo ‘lectio’. Comparative evidence suggests that Gothic had the geminates pp, ff, bb, and perhaps hh. Only Greek loans can be cited for pp and bb: filippus ‘Philip’, sabbato ‘sabbath’. For ff and hh the forms affalht ‘didst hide’ and uzuhhof ‘lifted’ might be cited; but these were probably /af-fálht/ and /uzuh+hó:f/. (I use the symbol + to indicate the phoneme of open juncture which seems to have existed in Gothic, as in other Germanic languages. Those who do not resort to juncture may read it as ‘beginning of a stem after a prefix’ or ‘beginning (or end) of a word or of part of a compound word.‘)

⁷ In a historical study such as this, dealing with written documents and reconstructions, it is obvious that so-called ‘phonetic’ transcriptions in square brackets do not mean the same thing as phonetic transcriptions based on speech that has been heard. Except in rare cases, the only phonetic details that we can obtain from written documents have to do with distinctive features. This is even more true of reconstructions: here we cannot possibly uncover any phonetic details, though I believe it is possible for us to reconstruct both the phonemes and the distinctive features which compose them. When, therefore, I cite Gothic (or PGmc.) [b] as a voiced labial stop and [ѣ] as a voiced labial spirant, I shall imply only the three distinctive features of voice, labial articulation, and occlusion. These three features can be read from the written documents and from the comparative evidence. Nothing can be read, and nothing will be implied in my transcription, as to whether [b] and [ѣ] were bilabial or labiodental, fortis or lenis, etc.

⁸ Following André Martinet, Word 8.13 (1952), I shall say that a number of phonemes sharing a place of articulation as one of their distinctive features form an order (e.g. Gothic /p f b/); and that a number of phonemes sharing a manner of articulation as one of their distinctive features form a series (e.g. Gothic /p t k/).

⁹ Edgar H. Sturtevant, The pronunciation of Greek and Latin §95 (²1940). Greek beta remained a stop only in gemination and after nasal.

¹⁰ On occasional spellings with postvocalic final b, bs, and also d, ds, z, see Wilhelm Streitberg, Indogermanische Forschungen 18.383-400 (1906). I interpret such spellings as morphophonemic, like modern German gab, gibt, gibst, beside geben, gaben, etc.

¹¹ E.g. in Max Hermann Jellinek, Geschichte der gotischen Sprache 34-5 (1926).

¹² There are no recorded examples of final lb (or lbs), column 6; the form halb which is sometimes cited (e.g. Prokosch, Comp. Gmc. gram. 75) does not exist. There is of course a postconsonantal alternation of b and j in Gothic, due to the effects of Verner's Law, as in pl. paurbun, sg. parf ‘need’. However, this alternation is not automatic, as is proved by a minimal contrast such as þarb ‘needy’ vs. þarf ‘needs’; and only automatic morphophonemic alternations (like giban, gaf) can be taken as evidence for the phonetics of Gothic.

¹³ Sturtevant, Pron. of Gk. and Lat. §§154, 199.

¹⁴ Franz Dietrich, Über die Aussprache des Gothischen während der Zeit seines Bestehens 71 (1862). The citation of such forms as evidence about biblical Gothic implies that the Gothic represented in these names had the same phonological structure as biblical Gothic. I shall make this assumption in all cases where it does not conflict with the internal evidence of biblical Gothic. Without this assumption the evidence of the names cannot be used at all.

¹⁵ One is tempted to add here that Gothic postvocalic b is usually transcribed in Latin with v, and Gothic postconsonantal b with b. Similarly, Latin November is transcribed in the Gothic calendar as naubaimbair, with postvocalic b for Latin v and postconsonantal b for Latin b. If these two types of evidence were reliable, they would be excellent corroboration of the internal Gothic evidence. However, the possibility of sound substitution in both cases is so great that such spellings prove little or nothing. Cf. the cautious remarks in Jellinek, Gesch. d. got. Sprache 36-7.

^15a I interpret the symbols q and h as representing respectively the clusters /kw/ and /hw/. See Lg. 24.83 fn. 28 (1948).

¹⁶ The absence of /nh/ in columns 3 and 4 reflects the change of PGmc /inx anx unx/ [į:x ą:x ų:x] to Gothic eih ah uh /i:h a:h u:h/. In column 7 the lack of evidence for /rk-+/ is surely fortuitous. On /hh/, see note 6 above.

¹⁷ Dietrich, Aussprache d. Got. 74. Ostrogothic examples in Ferdinand Wrede, Über die Sprache der Ostgoten in Italien 173-4 (1891).

¹⁸ It is of course true that the prototypes of g and h shared the feature ‘velar articulation’ in Pre-Gmc. times when the automatic alternation known as ‘grammatical change’ arose through Verner's Law. Cf. Gothic filhan ‘hide’ but fulgin ‘hidden’, taihun ‘ten’ but tiguns ‘tens’, etc. They were also still paired in Pre-Gothic times when the automatic alternation known as ‘spirantal dissimilation’ arose, whereby a spirant following an unstressed vowel was voiceless if the vowel was preceded by a voiced phoneme, but voiced if the vowel was preceded by a voiceless phoneme. Hence a given suffix has the shape -is- in agisa ‘fear’, rimisa ‘quietness’, but -iz- in hatiza ‘hate’, riqiza ‘darkness’; and, for the velars, -ah- in ainaha ‘only’ but -ag- in wulpaga ‘glorious’. (By the time of recorded Gothic this formerly automatic alternation had been mostly lost in the case of the velars: h appears only after voiced phonemes, but g appears after both voiceless and voiced. Cf. Jellinek, Gesch. d. got. Spr. 67.)

¹⁹ See Jellinek, Gesch. d. got. Spr. 32. The Gothic letter x is used twice (by error?) to transcribe Greek κ; otherwise it is used only to transcribe Greek χ (always in the abbreviations for Χριστός, and nine times in other Greek words).

²⁰ Cf. Sturtevant, Pron. of Gk. and Lat. §83; Jellinek, Gesch. d. got. Spr. 37-8.

²¹ Jellinek, Gesch. d. got. Spr. 72. But see also the skeptical remarks in W. van Helten, Indogermanische Forschungen 14.74-5 (1903).

²² Wilhelm Streitberg, Indogermanische Forschungen 24.181 (1909).

²³ In the more conservative manuscripts (Silver Codex, excluding Luke, and Ambrosian Codex B) it is assimilated only to a following /þ/, and even that not commonly. In other manuscripts, especially Ambrosian A, it is assimilated to a following /k s b d g m n 1 r/. Cf. Josef Janko, Prager deutsche Studien 8.64-5 (1908).

²⁴ Scholars who consider /h/ to have been a glottal spirant before vowels often insist that it must have been at least a weak velar spirant before consonants and finally, since otherwise it would not have been audible. Cf. Hermann Paul, Beiträge zur Gesch. d. dt. Spr. und Lit. 1.156 (1874); W. van Helten, Indogermanische Forschungen 14.75 (1903). However, we do not even need to go outside of Germanic to find examples of a perfectly audible pre-consonantal [h]: cf. the preaspirated stops of modern Icelandic described by Einar Haugen, Acta linguistica 2.98-107 (1941).

²⁵ Jellinek, Gesch. d. got. Spr. 35, considers this analysis, but rejects it. His reasoning is: Gothic x was copied from Greek X; Greek X stood for a voiceless palatovelar spirant; therefore Wulfila would surely have used it for any voiceless palatovelar spirant in Gothic; therefore Gothic cannot have had any voiceless palatovelar spirant, and post vocalic g before /+ s t/ must represent a stop. For two reasons this argument strikes me as most unconvincing. On the one hand, Wulfila seems to have had no qualms at all about borrowing a letter representing one type of sound in Greek and using it for quite a different type of sound in Gothic. Witness his borrowing of Greek Ψ for his þ symbol, and of Greek Θ for his h symbol. Secondly, Jellinek's argument assumes that a native speaker (Wulfila) is sufficiently aware of the allophones of his own language to use special symbols for them. This is of course contrary to nearly all our observations of the way in which native speakers write their own languages.

To test the hypothesis which (following many other investigators) I am proposing—that Wulfila's g stood for [x] before / + s t/—we would need to know how Wulfila transcribed Greek χ in final position, the only position where by our hypothesis both languages had [x]. Unfortunately no such forms occur. The name Ένώχ occurs once (Luke 3:37), but only with an ending in Gothic: gen. ainokis, translating του Ένώχ.

²⁶ Wrede, Sprache der Ostgoten 173-4.

²⁷ Dietrich, Aussprache d. Got. 73-4; Wrede, Sprache der Ostgoten 173 and note 3. With Wrede, we may interpret such spellings as Vitisclus beside Witigisclus as representing /wítigi:sl/ beside /wíti+gì:sl/, respectively. (Wrede of course does not present an analysis in terms of close and open juncture, but he clearly gives the 1891 equivalent.)

²⁸ There are exact parallels to Gothic kreks in West Germanic: OE Crēcas (pl.), OFris. Crēc(land), OHG Kriahi (pl.). On this loan, see Gustaf Kossinna, Zur Geschichte des Volksnamens Griechen, Festschrift zur 50jährigen Doktorjubelfeier Karl Weinholds 27-42 (1896).

The form marikreitum is dative plural. It occurs in only one passage (1. Timothy 2:9), but is attested in two manuscripts. I interpret the k as ‘initial’ because the word was presumably /mári-f-krì:tum/, with the stress and juncture of a compound. Cf. the treatment as a compound in OE meregrēot, OS (Heliand C) merigriotun, OHG merigrioz, literally (by popular etymology) ‘sea grit’.

²⁹ For the unvoicing of final stops, see Adolf Noreen, Altisländische und altnorwegische Grammatik (⁴1923) §220, final spirants §223.2; voicing of voiceless spirants §240.1 (f), §221.1 (þ), §230 (h).

³⁰ Pre-OIc. /b/ was also a stop in the cluster /mb/; cf. OIc. lamb ‘lamb’ etc. However, there are no recorded examples of final /mb/ > /mp/; cf. Noreen, Aisl. Gram. §§219 Anm. 1, 220 Anm. 2. To complete the evidence on /b d g/, we should add that Pre-OIc. /d/ seems to have been a stop also after /z/. Cf. Gothic huzd ‘treasure’, gazds ‘sting’, OIc. hodd, gaddr.

³¹ Cf. Noreen, Aisl. Gram. §266.1-3.

³² All runic forms and datings are from Noreen, Aisl. Gram. 374-93. They can easily be located with the aid of Noreen's special index, 462-6.

³³ bAriutiþ (with epenthetic a) represents PGmc. */bríwtid/. That Old Norse generalized the PGmc. 2nd and 3rd sg. endings /-iz, -id/ < PIE /-esi, -eti/ rather than /-is, -iþ/ < PIE /-ési, -éti/ is evident from the 2nd sg. ending -r < -r, which must come from /-iz/ rather than /-is/. Cf. runic bArutr ‘breaks’ (Björketorp, ca. 700), where this 2nd person ending has already spread to the 3rd person.—This interpolation is necessary, since otherwise the ending of bAriutiþ could be construed as representing PGmc. /-iþ/.

³⁴ Since runic r (< PGmc /z/) did not become /s/ through this change, it is clear that by this time it was no longer the voiced counterpart of /s/. I do not know what distinctive feature(s), if any, it shared with other phonemes. It fell together with /r/ in preliterary times. The earliest indication of this is the spelling Afatr ‘after’ on the Istaby stone (ca. 650), with r for inherited /r/. (Noreen's reading Afatr, Aisl. Gram. 380, is incorrect. Cf. the photograph in Wolfgang Krause, Runeninschriften im älteren Futhark [1937], plate 79).

³⁵ Although isolated spellings with f and þ for inherited /b d/ occur as early as the middle of the 6th century, inscriptions from around the year 700 (Björketorp, Eggjum, Stentoften) still distinguish /f þ/ from /b d/ almost without exception. (Cf. Noreen, Aisl. Gram. §221 Anm. 2.) I shall therefore refer to this voicing as an ‘8th-century’ change.

The earliest loss of /h/ in voiced surroundings (excluding the assimilation of /ht/ to /tt/ and the dissimilation of /hs/ to /ks/) is sigadur < */sígu+hàduz/ ‘Sigoþr’ (Svarteborg, shortly after 400)—a rather special case, since it involves loss of open juncture. Otherwise /h/ is not lost until ca. 700: Björketorp -sba < */spáhu/ (though the same inscription keeps /lh/, with epenthetic vowel, in JAlAh). Cf. Noreen, Aisl. Gram. §230.1.

Bohumil Trnka, Phonological remarks concerning the Scandinavian runic writing, Travaux du Cercle linguistique de Prague 8.292-6 (1939), argues convincingly that it was this voicing of voiceless spirants which caused the graphic contrasts p-f-b, t-p-d, k-h-g to be reduced in the younger futhark to b-f, t-þ, k-h. The first step in this ‘dephonologization of the voice of spirants’ was the change of /z/ to /r/; the remaining steps were the voicing of /f/ and /þ/, and the assimilation, dissimilation, and loss of non-initial /h/.

³⁶ This argument is not entirely decisive, since conceivably /h/ could have been lost before or independently of the 8th-century voicing. However, since /h/ became lost in precisely the same environments in which /f/ and /þ/ became voiced, I prefer to consider the loss of /h/ as part of the 8th century voicing.

³⁷ OIc. barg, sualg could theoretically have analogical g, but this is highly unlikely. Post-vocalically we find in class I strong verbs both regular hnē, mē, sē, stē and analogical hneig, meig, seig, steig; in class II both regular flō, lō, mō, smō and analogical flaug, laug, maug, smaug; in class III only regular brā; in class V both regular vā, frā, lā, þā and analogical lag; and in class VI only regular drō. (Noreen, Aisl. Gram. §§482, 486, 495, 497-8, 501.) This means that analogical formations certainly do occur—but only beside regular ones. It is therefore almost inconceivable that in barg and sualg only analogical forms should have survived.

³⁸ See Krause, Runeninschriften, plates 5 and 6. Following Krause (and many others), I use b d g in transcribing runic forms, rather than the that appear in many handbooks. These latter symbols are typographically inconvenient; they are also misleading, since they make it look as though the runic inscriptions offered evidence for spirantal articulation where Gothic does not. Prokosch, for example, seems to have been misled by them when he writes (Comp. Gmc. gram. 75): ‘Initial ѣ and đ appear as stops everywhere in Germanic, with the exception only of a few isolated forms in early Runic inscriptions, (barutR = brātr “breaks”, Björketorp; “daughters”, Tune ...), for which, however, other explanations might be given.‘ (See also a similar remark on p. 58.) These runic spellings are not isolated, but entirely regular; and they offer no information whatever on stop vs. spirant pronunciation. In this respect the futhark alphabet was quite phonemic, and hence provided no symbols to distinguish [ѣ ð] from their allophones [b d].

³⁹ Geminate consonants are never written double in the older runic inscriptions. Cf. Alexander Jóhannesson, Grammatik der urnordischen Runeninschriften §60 (1923).

⁴⁰ See Walde-Pokorny, Vergleichendes Wörterbuch der idg. Sprachen 1.335 (1930).

⁴¹ See Einar Haugen, First grammatical treatise 16, 34 (Language monograph No. 25, 1950). The First Grammarian uses the word in the meaning ‘shark’; it also meant ‘thole’ in Old Icelandic.

⁴² For convincing evidence that initial /g/ was a stop in Old Norse—and hence, as far as we can tell, in older runic—see Stefán Einarsson, Journal of English and Germanic philology 40.42-7 (1941).

⁴³ This may be illustrated by the following sets of correspondences: Gothic lamb, (silba), -swarb, land, skuld, waurd, gahugd, huzd, lagg, -balg, baurg; OIc. lamb, siof, suarf, land, skuld, orþ, -ugþ (~uþ), hodd, lang, belg = [belg], borg = [borg].

⁴⁴ E.g. Gothic -gulþ, fanþ ‘gold, found’, but OIc. gull, fann; and OIc. fīfl ‘giant’ < */fimfill/. See Noreen, Aisl. Gram. §§275, 298.2.

⁴⁵ For examples and dating of vowel syncope, see Noreen, Aisl. Gram. §§153-60.

⁴⁶ E.g. OE moþþe ‘moth’, Gothic aiþþau ‘or’, but OIc. motte, eþa (with /tt/ after stressed vowel, /þ/ after unstressed vowel). See Noreen, Aisl. Gram. §241. The cluster /þþ>/ was never re-established; the First Grammarian (Haugen 27) tells us explicitly that it did not exist in his dialect.

⁴⁷ For details of these developments, and for subsequent changes, see Noreen, Aisl. Gram. §238.1a, b. Forms such as vǫnþ, tǫiþ (for vaneþ, taleþ) are later analogical developments; see Andreas Heusler, Altisländisches Elementarbuch §317 (³1932).

⁴⁸ See Noreen, Aisl. Gram. §239.1b. The development is most clearly described in Heusler, Aisl. Elementarbuch §181.3: ‘Die Formen mit kt, x stellten oft den Reibelaut (3) der verwandten Formen her, und der wurde vor t, s automatisch stimmlos (χ): heilagt, heilags; sagt, dags usw. Dies waren die einzigen χ-Laute des Aisl.‘ Noreen's assumption of palatal and velar allophones of /g/ is not borne out by the modern evidence; see Einarrson, JEGP 40.40.

⁴⁹ Haugen 24-5.

⁵⁰ See Karl Luick, Historische Grammatik der englischen Sprache (1914 ff.) §639 (voicing of medial [f þ s]), §656 (loss of medial [h]), §651 (unvoicing of final [ѣ g]).

⁵¹ It might also be assumed that the spirantal allophones of /d/ became stops because /þ/ began to develop voiced allophones. For two reasons, however, this alternative analysis is unlikely. First, the change of [ð] to [d] presumably occurred before the Anglo-Saxon migration from the continent, since it is common to all the WGmc. dialects; whereas the voicing of medial voiceless spirants was apparently not completed at the time of our earliest OE documents. (See below under the discussion of the labials.) Secondly, final [ð] became a stop just like medial [ð], even though final [J?] did not become voiced and thereby ‘push’ [ð] into becoming a stop.

⁵² For the following, see Luick, Hist. Gram. §639.2. Other examples of voiceless [þ] following an unstressed vowel are the ordinal numerals sēofoþa, eahtopa, nigopa, etc.

⁵³ E.g. Epinal Glossary 223 giroefa ‘officer’, 183 gen. uulfes ‘wolf’, 142 uuf ‘owl’, 192 obaer ‘over’, 51 halbae ‘halves’, 577 staeb ‘staff’; Henry Sweet, ed., The oldest English texts (1885 = Early English Text Society, Vol. 83). Cf. Eduard Sievers, Altangelsächsisch F und B, Beiträge z. Gesch. d. dt. Spr. u. Lit. 11.542-5 (1886), and Anglia 13.15-6 (1891). Luick, Hist. Gram. §639 Anm. 4, §658, dates the voicing of medial [f s þ] as ‘spätestens im 6. Jahrhundert’, and interprets the f-b spelling contrasts of the early documents as representing labiodental vs. bilabial voiced spirants. However this may be, it is clear that the phonemic merger of /f/ and /b/ had not been completed at the time of these documents, but that it was completed shortly thereafter. This point is nicely made in Herbert Penzl, A phonemic change in early Old English, Lg. 20.84-7 (1944).

⁵⁴ Cf. Luick, Hist. Gram. §634.2; Eduard Sievers (ed. Karl Brunner), Altenglische Grammatik §193.3 (1942). For a different view (ræfsan < */ráfisjn/, ræpsan < */rábisjan/), see Albert Morey Sturtevant, Lg. 7.191 (1931).

⁵⁵ This is the usual interpretation; cf. Sievers-Brunner, Ae. Gram. §231.1. Luick, Hist. Gram. §§625.2, 631.1, 644.1, states that geminates which came into final position thereby lost their geminate quality, but still remained long.

⁵⁶ For details, see Luick, Hist. Gram. §637.

⁵⁷ To indicate the OE palatal stops, I use the traditional symbols ċ and ġ; the exact phonetic quality is uncertain. For an excellent analysis of palatalization in Old English, see Herbert Penzl, The phonemic split of Germanic k in Old English, Lg. 23.34-42 (1947). Penzl says that the ‘crucial innovation’ in the phonemic split of /k/ into /ċ/ and /k/ was ‘the appearance of velar [k] before the new palatal vowels developed through i-umlaut’ (42). This would mean that the phonemic split did not take place until early OE, when /ö/ and /ö:/ were unrounded in most dialects and thus merged with inherited /e/ and /e:/. However, if my own analysis is correct, the ‘crucial innovation’ was the loss of /j/ and final /i/, which put the [c] of [stréċċan, bénė] in contrast with the [k] of [hnékka, dránk]; and this happened before the time of the earliest documents.

⁵⁸ E.g. Pre-OE */gró:nja, gó:si/ > grēne, gēs ‘green, geese’. The evidence that OE initial g- plus back vowel or consonant represented a stop is inconclusive; it could conceivably have been a velar spirant. I assume stop articulation because that is what we find as far back as we can trace the pronunciation from modern times. Luick's assumption (Hist. Gram. §§633, 696) that both velar and palatal g, including g < /j/ (!), were stops initially, leaves me quite unconvinced. For the more generally accepted view (which I follow), see Sievers-Brunner, Ae. Gram. §206.8.

⁵⁹ Cf. Luick, Hist. Gram. §636; Sievers-Brunner, Ae. Gram. §§217-23. It is possible that a medial [x] still existed at the time of ‘breaking’, but changed to glottal (h] afterwards. This would give the following sequence of changes: *[séxan > séoxan > séohan > se:on], OE sēon ‘see’. However, this is slim evidence indeed, since there is no reason why ‘breaking’ could not have taken place before both [x] and [h].

⁶⁰ The cluster [xs] was later restored in such forms as nīehsta ‘next’ and stīhst ‘climbest’ (column 11); cf. Sievers-Brunner, Ae. Gram. §214.1 Anm. 2, §222. The cluster /kt/ in column 12 also arose as a result of vowel syncope: */é:akidæ/ > īecte ‘increased’.

⁶¹ The following description gives only the broad outlines; no attempt is made to trace the detailed usage of g, h, gh, kg, ch for the spirants, and g, c, cg, gc for the (palatal and velar) stops. Cf. Sievers-Brunner, Ae. Gram. §211-23. I use the letter g to represent the special OE symbol.

⁶² Except as noted below, all forms are from Edward H. Sehrt, Vollständiges Wörterbuch zum Heliand und zur altsächsischen Genesis (1925). The form uulf* ‘wolf’ (column 6) is assumed on the basis of Heliand uulbo, uuluo (gen. pl.) and uulbos, uuluos (ace. pl.). The infinitive hlahhian* ‘laugh’ (column 2) is assumed on the basis of Heliand pret, hlōgun and ppl. bihlagan, and MLG lachen. Cf. Agathe Lasch, Mittelniederdeutsche Grammatik §351 Anm. 2 (1914). The forms krampo ‘hook’ [hri]psod ‘reproves’, and unt ‘up to’ are from Elis Wadstein, Kleinere altsächsische Sprachdenkmäler (1899).

For an analysis of the graphemes and phonemes of the Heliand and Genesis manuscripts, see Carl Richard Page, The phonological system of the Old Saxon language, Cornell University M.A. thesis (typescript; Ithaca, 1952).

⁶³ See F. Holthausen, Altsächsisches Elementarbuch (²1921) §197 (voicing of [f]), §206 (voicing of [þ]), §223 (unvoicing of [b]), §234 (unvoicing of [g]).

⁶⁴ The abbreviations M, C, V, P, G refer to the Munich, Cotton, Vatican, and Prague manuscripts of the Heliand, and to the Genesis.

⁶⁵ See Holthausen, As. Elementarbuch §242.

⁶⁶ See Lasch, Mnd. Gram. §350.

⁶⁷ See Lasch, Mnd. Gram. §351. Holthausen, As. Elementarbuch §213, assumes that h represented [x] finally in all environments and not just after stressed short vowel. This is unlikely, as he shows in §214 by listing the many forms in which it was not written finally after consonant, long vowel, or unstressed short vowel.

⁶⁸ See Holthausen, As. Elementarbuch §§252, 252a; Lasch, Mnd. Gram. §§343-4.

⁶⁹ This is certainly what the OS spellings seem to indicate, though it does not agree entirely with the evidence of MLG. See Lasch, Mnd. Gram. §§341-2, 349.

⁷⁰ /g/ apparently also in the variant dative spelling dage and in genitive dages.

⁷¹ See Lasch, Mnd. Gram. §342, who also assumes an initial voiced spirant for Old Saxon. Holthausen, As. Elementarbuch §251, assumes an initial stop for all of West Germanic; this is hard to reconcile with the change of /g/ to /j/ before front vowels in Old English, Old Frisian, and Old Saxon, and with the widespread modern spirants before vowels and consonants (e.g. in Westphalian and Dutch).

⁷² For similar evidence in the smaller documents, see Holthausen, As. Elementarbuch §234.

⁷³ To complete the picture of the Old Saxon stops and spirants we must add one further detail which affects the distribution of phonemes and distinctive features, though not the basic phonological structure. Table 5 indicates a contrast in final position between voiceless and voiced stops: salt-geld, dranc-sprang. Variant spellings such as dump for dumb, dot for dod, and also hold for holt, uuilld for uuilt, indicate that this contrast was being given up (or perhaps had already largely been given up) during OS times. See Holthausen, As. Elementar-buch §§246, 248, 252, and 238 Anm. 1. The contrast was of course entirely gone by MLG times; see Lasch, Mnd. Gram. §227.

⁷⁴ In making the following analysis of Old High German I have benefited greatly from some spirited correspondence with W. Freeman Twaddell.

⁷⁵ For the following, see Otto Behaghel, Geschichte der deutschen Sprache 404-7 (⁶1928).

⁷⁶ See Wilhelm Braune (ed. Karl Helm), Althochdeutsche Grammatik §166 (⁵1936). Primus Lessiak, Beiträge zur Geschichte des deutschen Konsonantismus 130 (1933), considers the voicing of /þ/ to be parallel to that of /f/ (59-70) and /s/ (79, 95). However, the voicing of these latter two spirants seems clearly to have occurred during the OHG period, whereas the voicing of /þ/ just as clearly began before the OHG period.

⁷⁷ Cf. J. Fourquet, Les mutations consonantiques du germanique 101 (1948): ‘Le fait essentiel [of the HG sound shift] est l'évolution de la corrélation de sonorité vers une corrélation d'aspiration.’

⁷⁸ It would be interesting to know whether we are dealing here with a ‘drag-chain’ or a ‘push-chain’ (André Martinet, Word 8.11 [1952]). Did the aspiration of /p t k/ ‘drag’ along the unvoicing of /b d g/? Or did the unvoicing of /b d g/ ‘push’ the aspiration of /p t k/? I have presented it as a ‘drag-chain’ because the aspiration of /p t k/ covers a far wider area than the full unvoicing of /b d g/. Cf. Fourquet, Mutations consonantiques 85-6, 104-5.

⁷⁹ For these OHG spellings, see Braune-Helm, Ahd. Gram. §96 Anm. 1; and Josef Schatz, Althochdeutsche Grammatik 105-34 (1927). In most words these original long stops following long vowels, diphthongs, and consonants were analogically replaced by the short stops of related forms: 2nd sg. galaubis, 3rd sg. galaubit, and hence also infinitive galauben for galauppen. See Braune-Helm, Ahd. Gram. §§358, 359 Anm. 1. This analogical substitution of short stops must have been very widespread, because remarkably few of the long stops have survived.

⁸⁰ The change of [þ] to (voiceless lenis) [d] seems to be just about completed in the earliest Bavarian documents. See J. Schatz, Altbairische Grammatik 70 ff. (1907), Ahd. Gram. 97-8, 127 ff. Inherited /þþ/ > /dd/ > /tt/: feddhacho, feddah, fettah ‘wing’. See Braune-Helm, Ahd. Gram. §167 Anm. 10; Schatz, Ahd. Gram. 133.

⁸¹ Modern Swiss dialects prove that these fortis-lenis contrasts after long vowels, diphthongs, and consonants were real, and are not just a misinterpretation of clumsy scribal spellings: southern Swiss ərlöippə < */ar+láwbbjan/ ‘allow’, welpə < */hwálbbjan/ ‘arch’, ęrpə < */árbbjan/ ‘inherit’, contrasting with blībə < */blí:ban/ ‘stay’, xelbər < */kálbir/ ‘calves’, ‘die’. See Leo Jutz, Die alemannischen Mundarten 178-9 (1931); velar examples are given on 228-9.

⁸² Through the familiar exceptions to the HG sound shift, inherited /sp st sk tr ft xt/ gave OHG /sp st sk tr ft xt/, which I interpret as fortis stops and spirants.

⁸³ Leonard Bloomfield, Lg. 14.181 (1938): ‘Before the HG consonant shift, the language had the WGic unvoiced stops [p, t, k], the voiced (later unvoiced lenis) stops [b, d, g], and the unvoiced spirants [f, þ, s, h]—or was [þ] already voiced? The shift of initial [p-, t-, k-] consisted in their replacement by a phoneme of the second series plus a phoneme of the third series; apart from conditions peculiar to the cluster and from later changes, this produced phonemically [bf-, ds-, gh-].’ (In a footnote Bloomfield seems to change his [ds] to [dʒ].) I differ from this in interpreting the initial clusters as /pf tʒ kx/, with fortis stops and spirants. Although there was no fortis-lenis contrast in these clusters initially, there seems to have been such a contrast after /l r/ in the labial order. Pre-OHG /lpp rpp/ > OHG /lpf rpf/: gelpf ‘gay', scarpf ‘sharp’, with /lpf rpf/ surviving, beside (analogical?) /If rf/, into MHG gel(p)f, schar(p)f. In contrast with this, Pre-OHG /lp rp/ gave OHG /lbf rbf/: helphan ‘help’, uuerphan ‘throw’; but as early as the 9th century this became /If rf/: helfan, unerfan. See Braune-Helm, Ahd. Gram. §131 Anm. 5. Cf. also southern Swiss hęlffə, węrffə vs. šarpf; Jutz, Alem. Mdaa. 170-1.

⁸⁴ See Braune-Helm, Ahd. Gram. §§97, 132, 145, 160.

⁸⁵ Southern Swiss dialects seem to show a distribution of fortis and lenis that is even more archaic than that of the OHG documents. They have the fortis-lenis contrast not only medially (ɔffə ‘open’ vs. ofə ‘oven'), but also finally, even after long vowel (šlɔ:ff ‘sleep’ vs. grɔ:f ‘count') and after consonant (dɔ:rff ‘village’ vs. wolf ‘wolf‘). Cf. Jutz, Alem. Mdaa. 170-1, 180-1; the above forms from Wilhelm Wiget, Die Laute der Toggenburger Mundarten (1916).

Since these Swiss forms can be analyzed in terms of the two-way contrast /ff-f/, we may ask whether the three-way contrast /ff-f-v/ is really needed for OHG. The answer depends on how much symmetry we think there must be in a phonological system. In dealing with the stop series we could perfectly well say that fortis /p/ and /k/ were allophonically long after short vowels, but short after long vowels, diphthongs, or consonants, thus avoiding the assumption of /pp/ and /kk/. By this analysis crippa, brucca would be /kxrípa, brúka/. I have chosen not to make this analysis because of the evidence of the dentals. The contrast bittu, tritu, quidu forces us to analyze bittu as /bíttu/; therefore crippa, brucca are also /kxríppa, brúkka/, with long /pp kk/—even though in these orders there is no contrast with short /p k/. The situation with the spirants is exactly the same. After the change slāffan > slāfan had taken place, we could say that fortis /f/ was [ff] medially after short vowels, but [f] after long vowels, diphthongs, consonants, and in final position. It seems to me inconsistent, however, to analyze bittu [bittu] as /bíttu/, but offan [óffan] as /ófan/. I therefore analyze it as /óffan/, and put it in double contrast with ouan /óvan/.

One might of course analyze the fortis as /ff/ and the lenis as /f/, and write /hóff, hófes, skiff, skiff es/ for the scribes' hof, houes, scif, sciffes. This would again conflict with the analysis of the dental stops. It would also mean that we would write nom. haz, ioh, gen. hazzes, iohhes ‘hate, yoke’ as /haʒʒ, jóxx, háʒʒes, jóxxes/, and no simple /ʒ x/ would ever occur.

⁸⁶ On the phonetics of these two spirants see, most recently, Martin Joos, The medieval sibilants, Lg. 28.222-31 (1952).

⁸⁷ Modern standard German beissen /báisen/ ‘bite’ and weisen /váizen/ ‘show’ still reflect clearly the difference between the fortis /ʒ/ of OHG bīzan /bf:ʒan/ and the lenis /z/ of OHG wīsōn /wi:zo:n/.

⁸⁸ Edgar H. Sturtevant, The Indo-Hittite laryngeals §21 (1942).

⁸⁹ Zur Lautverschiebung, Beiträge z. Gesch. d. dt. Spr. u. Lit. 1.147-201 (1874), esp. 170-1.

⁹⁰ See above, note 18.

⁹¹ We must at least consider the possibility that this initial /h/ never was velar, but evolved directly from PIE /k/. Such an assumption would violate the working hypothesis which I believe we all use: that phonemic change involves only one distinctive feature at a time. For example, where we find that an earlier back rounded /u/ has become a front unrounded /i/, we never assume a direct change, but always back rounded /u/ to front rounded /ü/ to front unrounded /i/. This same hypothesis would force us to reject the theory that the velar stop /k/ changed directly to the glottal spirant /h/. Instead, we would assume the change of velar stop /k/ to velar spirant /x/ to glottal spirant /h/.

Evidence that PIE initial /k/ gave Pre-Gmc. initial /x/ is to be found in the prefix ga-, if we accept the etymology which connects this with Latin co- etc. This would mean that PIE */ko-/ gave Pre-Gmc. */xa-/ and then, by Verner's Law, Proto-Gmc. */ga,-/, proving that the spirant of */xa-/ shared with /g/ the feature of velar articulation.

⁹² OE cohhettan, ME coughen, MLG kuchen, MHG kūchen ‘cough’; OE ceahhettan, OHG kachazzen ‘laugh’; etc. The absence of inherited /xx/ in Gothic and Norse is crucial. For Gothic and runic Norse we have only negative evidence; for Old Icelandic we have the definite assurance of the First Grammarian. If /xx/ had survived in any of these last three dialects, it would have produced a three-way contrast in the velar order and have given it the same phonological structure as the labials and dentals.

⁹³ Proto-Gmc. intervocalic /x/ was still the voiceless counterpart of intervocalic /g/ at the time of the spirantal dissimilation in Gothic; see above, note 18. Since this phenomenon seems to belong to the prehistory of Gothic, this is added proof that we must set up intervocalic /x/ and not /h/ for Proto-Germanic. We may also posit Proto-Gmc. /x/ before /j/ in */hláxjanan/ ‘laugh’, and assume that it remained velar up to the time of the WGmc. consonant doubling. Cf. the /xx/ of OE hlæhhan, OHG hlahhen.

The necessity of assuming Proto-Gmc. medial /x/ before vowels is an apt illustration of Sturtevant's qualifying phrase (note 88 above) that a feature common to several daughter languages must be ascribed to the parent language ‘unless subsequent study proves this to be impossible’.

⁹⁴ [The foregoing article was offered as a contribution to the Edgerton number of Language (29:3), but reached the Editor too late to be included in that number.]