Proto-Indo-European Phonology

1. Introduction

1.1. Reconstruction of the PIE consonants in the 19th century

The descriptions of PIE phonology found in our handbooks date in great part from the nineteenth century. When we maintain these descriptions we accept the foundations on which they are based. At present our data have been increased by Hittite, and our linguistic methodology refined. In this monograph I shall examine the new data and some well-known but unexplained data in accordance with current linguistic methodology and propose a revised description of PIE phonology.

The phonology of PIE was first described and greatly developed in the course of the nineteenth century. Grimm, Rask, and Bopp did not even attempt to reconstruct the language from which the dialects developed, assuming it to be much like Skt. Schleicher, trying to escape such undue emphasis on Skt., reconstructed PIE on the basis of all the dialects. Yet with the nineteenth century emphasis on phonetics he and other Indo-Europeanists dealt primarily with the sounds of IE, not with their patterning. Thus the chief contribution of nineteenth century Indo-Europeanists consisted in determining the phonetic system of PIE. Only a few concerned themselves with its phonemic structure. Today this study rather than the phonetic description of PIE needs further development.

The earliest agreement in the description of PIE was on the consonants. After some debatable points were cleared up in the sixties and seventies of the nineteenth century, the consonants were pretty well established as we today reconstruct them. Since then the consonants [p b ph bh t d th dh k g kh gh k̑ g̑ k̑h g̑h kʷ gʷ kʷh gʷh s z m n ɲ ŋ r l w y] have been generally assumed, and often also four spirants [þ þh ð ðh]. The consonants over whose description there has been most dispute are [bh dh gh g̑h gʷh] ; they are usually described as aspirated stops, but occasionally as spirants.

The chief difficulties removed between 1860 and 1880 were those solved in the formulations known as Grassmann's Law, Verner's Law, and the Law of Palatals.

1.1a. Grassmann¹ in 1863 demonstrated that in Ind.-Ir. and Greek the first of two aspirates beginning successive syllables or a syllable that ends in an aspirate lost its aspiration: e.g. Skt. dadhā́ti, Gk. τίθητι ‘places’, from PIE dh_e-dhē-; Skt. bandh- ‘bind’, Gk. πενθερός ‘step-father’, but Mod. Eng. bind, from PIE bhendh-. The deaspiration occurred independently in the two dialects. After Grassmann's discovery the PIE aspirates could be reconstructed from reflexes in Ind.-Ir. and Gk. as well as from Gmc. and, less easily, some of the other dialects.

1.1b. Verner² in 1877 demonstrated that in Gmc., voiced spirants developed from voiceless spirants if the preceding syllable did not have the chief word accent in PIE: e.g. Goth. frawaírþan ‘spoil’, frawardjan ‘ruin’ from PIE wert- (compare Skt. vártate, vartáyati); OHG ziohan ‘pull’, pret. part. gizogan, from PIE deuk-. His demonstration indicated that the movable pitch accent of PIE had survived into early Proto-Germanic.

1.1c. About the time of Verner's discovery a number of linguists³ independently demonstrated that in PInd.-Ir. velar consonants were palatalized before i ī y e ē: e.g. Skt. ugrá ‘strong’, ójīya ‘stronger’; cakára ‘I have made’ from PIE kekora. After the origin of these Ind.-Ir. palatals was demonstrated, the group of PIE consonants was even more clearly defined.

1.1d. The difficulties that remained involved only a small proportion of the vocabulary; chief of these were the status of the voiceless aspirates, that of the four proposed spirants [þ þh ð ðh], and the relationship between velars, labio-velars, and palatals in PIE. Evidence for reconstructing PIE with voiceless aspirates and spirants other than s was drawn primarily from Ind.-Ir. and Gk.; the materials could be variously interpreted, and no interpretation has yet been widely accepted. The problem of the relationship between velars, labio-velars, and palatals is one of analysis. Evidence for the three groups of sounds can be found in the dialects. Interpretation of the evidence has varied with the methods employed in analyzing the data; if the data are analyzed phonetically, three groups of sounds must be assumed; if they are analyzed phonemically, one may set up two groups.

1.1e. With the solution of the chief problems in the PIE group of consonants went the assumption that the PIE accent was a variable pitch accent like that of Vedic Sanskrit. Verner demonstrated that if one assumed that the PIE (Vedic) accent had survived into PGmc., one could account for the development of voiced from voiceless spirants.

1.1f. We may note incidentally that Gmc., with all of its changes from PIE, has been involved in the clarification of many of the important problems of IE phonology; PGmc. maintained the PIE accentual system and consonantal system much longer than did IE dialects like Italic and Celtic, and even in some respects than Gk. For the apparently great number of changes described in Grimm's Law involved little change in the PIE consonant system. The PIE three-stop system was maintained in Gmc.; PIE p b bh > PGmc. f p ƀ, PIE t d dh > PGmc. þ t ð, PIE k g gh (kʷ gʷ gʷh) > PGmc. χ k ʒ (χw kw ʒw). In other dialects the PIE system was broken down; in Celtic a two-stop consonant system developed, e.g. PIE p b bh > PCelt. p b. A two-stop system developed also in Slavic and Baltic. Moreover from the evidence we have about the time of the Gmc. accent change we may assume that Gmc. maintained a variable pitch accent like that of PIE until the fourth century B.C.; at this time the Italic and Celtic dialects had an initial stress accent, Gk. a fixed pitch accent. Consequently PGmc. must be considered one of the most conservative IE dialects; investigation of PGmc. may therefore yield important information for a solution of some of the remaining phonological problems of PIE.

1.2. Reconstruction of the vowels

After the formulation of the Law of Palatals it was clear that the European dialects had maintained the PIE vowel system better than Ind.-Ir. For the palatals had developed before front vowels, such as the reflex of /e/, but not /o/ or /a/. From forms like Skt. 3d sg. pres. hánti ‘he strikes’, PIE gʷhen- (as opposed to the 3d pl. pres. ghnánti, PIE gʷhn-), and 3d sg. perf. jaghā́na, PIE gʷhegʷho-, it was apparent that until the Law of Palatals was in effect Skt. too had maintained the difference between PIE e and o. Before this discovery it was assumed that the PIE form should be patterned on Skt.: e.g. the Skt. 1st sg. perf. cakára ‘I have made’ was taken as the most conservative form, and the PIE 1st sg. perf. was reconstructed with the same vowel in the three syllables; now it is reconstructed with three vowels. And the Gk. (European) vowels e a o ē ā ō rather than the Ind.-Ir. a and ā were assumed for PIE.

We may note that in the solution of phonological problems two difficulties were often solved when a new phonetic relationship was recognized⁴; Verner settled the problem of the PIE accent while showing the interrelationships of the PGmc. voiced and voiceless spirants; Collitz and others established the conservatism of the Gk. vocalic system when they showed the interrelationships of the Ind.-Ir. palatals and velars. Similarly, Kurylowicz' demonstration that Hittite ḫ corresponded to hypothetical laryngeal consonants led to a clarification of the PIE ablaut relationships while settling the problem of the origin of Hittite ḫ. Brugmann's suggestion of IE vocalic n completed the data on the vowels; it led to the assumption of vocalic r l m as well as n.⁵ The only additional PIE sounds that have been suggested are the laryngeal consonants, which were first proposed by Saussure.

1.3. Formulations of the vocalic system

Two types of formulation of the PIE vocalic system have since been made from these data. Hirt's may be cited as an example of one of them. His system included: e o a ē ō ā; the diphthongs ei oi ai eu ou au ēi ōi āi ēu ōu āu; i u, ī ū, which he derives from unaccented diphthongal or bisyllabic formations; the reduced vowels ə ь ъ, and the syllabic liquids and nasals r̥ l ̥ m̥ n̥.⁶ Brugmann's system differs in including r̥̄ l ̥̄ m̥̄ n̥̄, and excluding ι ъ⁷. These systems are today maintained widely; Sturtevant in IHL describes his system as following ‘as closely as possible the one that was assumed by Brugmann and that underlies most of the current work on IE comparative grammar.’⁸

1.3a. In 1878 Saussure had already analyzed the PIE vocalic system quite differently on a structural basis.⁹ His system includes two vowels, e and o, and eight sonants, i u n m r l A O̮. The sonants entered into various diphthongal combinations, ei eu en, etc., and from eA developed Ā, from eO̮, Ō̮. None of the standard handbooks has adopted his system without modification; Meillet, perhaps, has remained closest to it.¹⁰ In addition to the consonants p t k kʷ b d g gʷ bh dh gh gʷh s, Meillet assumes: the vowels e a o ē ō ā; ə, which he says really is a sonant; a reduced vowel ^o; the sonants y w r l m n, which may be vocalic, the second element of diphthongs beginning with e o or a, and long when followed originally by ə. Meillet's system is inferior to Saussure's in that phonetic criteria outweigh functional criteria. To be sure Meillet has remained closer to Saussure than did Brugmann, who adopted Saussure's r̥̄ l ̥̄ m̥̄ n̥̄, but as phonetic not as structural units. With the ambiguous position of ə in his system, Meillet too wavers between a phonetic and a phonemic basis of classification.

1.3b. All of the vocalic systems proposed, except Saussure's, are very complex in spite of Hirt's preliminary assertion that the vocalic system of PIE is relatively simple. Among the problems are:

• A. the relationship of e a o and y w r l m n; should five vowels, e a o i u, be assumed, a series of diphthongs ey ew er el em en, etc. or even long diphthongs?
• B. what was the position of long vowels, ē ō ā, in the system; should one assume beside them one unaccented vowel, ə, or three? Is a further unaccented vowel to be assumed beside the PIE short vowels, e a o?
• C. should long r̥̄ l ̥̄ m̥̄ n̥̄ be assumed for PIE?

The answer to these problems depends largely on the assumptions made by the linguist. While Hirt admits that an analysis of the PIE vowels must be at least partially based on their systematic interrelationships, he bases his system primarily on phonetic criteria, and consequently he does not arrive at a phonemic analysis of PIE.

1.3c. Another difficulty with the systems sketched is that they do not account for various developments in the individual dialects. Among these are: the development of Gmc. lengthened y and w; the origin of Gmc. ē²; the development of IE y- to Gk. ζ and ʽ . No satisfactory explanation of these and other difficulties has been made on the basis of the PIE phonological systems hitherto proposed. It may be mentioned here that many of the problems in the development of PIE to the various dialects are found in words for which laryngeals have been suggested for IE.

1.4. Methodological differences between 19th century and current linguistic theory

Besides these external difficulties there is a fundamental theoretical difference between the linguistic elements used currently in linguistic study and those used by Brugmann and Hirt. The linguistic elements of Brugmann are based on phonetic criteria (Gdr. I.49), but those currently used are based on distribution as well. When Brugmann speaks of a Sprachlaut, he does not of course mean ‘sound’ in the narrow sense; he is following the lead of Sievers in choosing this term rather than the term ‘Sprachelement’ to refer to a class of sounds; [t] though articulated finally without an explosion would nonetheless be the same Sprachlaut as [t] articulated initially with an explosion. Still Brugmann's Sprachlaut differs from the phoneme used currently in linguistic study. To Brugmann an [ŋ], though found only before velars in PIE, was a different Sprachlaut from [n], though this was never found there. By modern linguistic theory PIE [n] and [ŋ] are classed in one phoneme.

1.4a. Current linguistic theory therefore demands a different analysis of PIE from Brugmann's. Adoption of a different analysis does not entail a rejection of Brugmann's facts. The data on which we construct PIE are unfortunately little more ample than Brugmann's. But if we reconstruct PIE with Brugmann's criteria, we compel contemporary linguists to devise their own phonemic system of PIE from an essentially phonetic system, or worse we give rise to misunderstandings. Moreover unsolved phonological problems will remain without hope of solution until we construct a phonemic system of PIE.

Current linguistics distinguishes sharply between speech and language, between sounds and phonemes. When reading Brugmann I feel that he was interested in reconstructing the IE speech. Later Indo-Europeanists were attempting to reconstruct the IE language, as is clear from such occasional statements on methodology as the following. In his IG Hirt had said:¹¹ ‘It really doesn't matter what was spoken in PIE; what matters are the ablaut relationships.’ Specht, KZ 59.98 (1932) explained that his transcription r̥̄ l ̥̄ was merely a convenient formula and was not meant to suggest the PIE pronunciation. But even these Indo-Europeanists failed to observe the rigorous distinction between phonetic and phonemic analysis that we should now demand.

1.4b. Therefore in our study today of the PIE phonological system we are faced with two fundamental and interrelated problems. A. We must reexamine in accordance with current phonemic theory the data known to Brugmann and his successors. B. These data have been augmented by material found in the Anatolian languages, especially Hittite. The attempts to account for this new material, to relate it to the data derived in the nineteenth century from the IE dialects, are known under the name of the laryngeal theory. The laryngeal theory demands further investigation, especially with regard to certain dialects such as Gmc.

1.4c. Compilations of the data on the PIE phonological system assembled in the nineteenth century and the beginning of the twentieth may be found in the standard works of Brugmann, Hirt, and Walde. Surveys of work published in connection with the laryngeal theory may be found in IHL, BHL, and SSP;¹² a resurvey here would therefore be superfluous.

Although evidence for the laryngeal theory in some IE dialects has yet to be investigated and although some suggestions made in connection with this theory have not been widely accepted, there is fundamental agreement on some of its propositions. These will be examined in chapter 3.

1.4d. Before examining these propositions, I shall suggest a phonemic system for the data drawn from the IE dialects other than Hittite. The system will be only provisional; it will hardly stand unmodified after the conclusions drawn from Hittite are added to those drawn from the IE dialects. But to make proper use of the Hittite data we have to systematize our other facts in terms of current linguistic theory. After systematizing them, I shall correlate with them the accepted propositions of the laryngeal theory, and then examine phonological problems with the provisional system. Finally we may be able to arrive at a phonological system taking as full account as possible of all of the data and systematizing them in terms of structural principles.

Footnotes

¹ Ueber die Aspiraten and ihr gleichzeitiges Vorhandensein im An- and Auslaut der Wurzeln, KZ 12.81-138.

² Eine Ausnahme der ersten Lautverschiebung, KZ 23. 97-130. Grimm and the other Germanists before Verner assumed that PIE /p t k/ had become PGmc. f θ χ but they did not account for the apparently aberrant development of some /p t k/ to PGmc. ƀ đ ʒ. By correlating such ƀ đ ʒ with the position of the PIE word accent Verner showed that they were not aberrant but followed a definite pattern.

³ See a summary like that in H. Hirt, Indogermanische Grammatik 2.20-1 (Heidelberg, 1921). A great number of articles were written about the subject, most of which agree in their results.

⁴ J. Whatmough, The Development of the Indo-European labiovelars with special reference to the dialects of ancient Italy, Mélanges Linguistiques offerts a M. Holger Pedersen 45-56 (Copenhagen, 1937), discusses at greater length the use, and necessity, of such argumentation in linguistics.

⁵ Brugmann's discovery was published in G. Curtius' Studien zur griechischen and lateinischen Grammatik 9.297ff. and 361ff. (Leipzig, 1877). In the zero grade of roots in en-, for example, Skt. and Gk. have an -a-, as in the to-forms, Skt. ha-tá, Gk. φα-τός from PIE /gʷhen-/; the normal grade is found in Skt. hánti ‘he kills’, Gk. θείνω ‘I kill’. Comparing these with the to-forms Skt. i-tá, Gk. ἰ-τός and Skt. é-ti ‘he goes’, Gk. εἰ̑-μι ‘I go’, one can assume for en a weakened form [n̥], of which the reflex in Skt. and Gk. is a, as for ey a weakened form [i]. Such a weakened form is also found in the second syllable of the word for ‘ten’: Skt. daśa, Gk. δέκα, Lat. decem, Goth. taíhun. H. Osthoff PBB 3.52 assumed syllabic r from the Skt. dat. pl. pitṛ́bhyas as opposed to the dat. sg. pitré.

⁶ This system is given in IG 2.101. In Hirt's Die Hauptprobleme der Indogermanischen Sprachenwissenschaft, herausgegeben und bearbeitet v. H. Arntz (Halle, 1939), ə is listed as the only weak vowel, and the following vowels are added: r̥̄ l ̥̄ m̥̄ n̥̄ r̥̄r l ̥̄l m̥̄m n̥̄n. IG 2.103, Hirt had said: ‘Es fehlen ganz die r̥̄, l ̥̄, m̥̄, n̥̄, und es ist auch keine Möglichkeit vorhanden, sie irgendwie anzusetzen. Die Lautverbindungen r̥̄r l ̥̄l m̥̄m n̥̄n, die zu den merkwürdigsten indogermanischen Irrformen geführt haben, sind hoffentlich ebenfalls beseitigt.’ I assume that the posthumous publication mistakenly printed a rejected version; the chaotic state of the manuscript Hirt left is indicated in the preface.

⁷ Gdr. I.92-3. Brugmann also posits the diphthongs əi and əu, the palatal and velar nasals as vowels, and aspirated and unaspirated forms of s and z.

⁸ E. H. Sturtevant, The Indo-Hittite Laryngeals 90-1 (Baltimore, 1942) (IHL). Sturtevant does not list diphthongs as unit phonemes, but fails to mention their place in his system.

⁹ Ferdinand de Saussure, Mémoire sur le système primitif des voyelles dans les langues indo-européennes (Leipzig, 1879).

¹⁰ A. Meillet, Introduction à l'étude comparative des langues indo-européennes⁸ 98-126 (Paris, 1937).

¹¹ IG 2.92: ‘Denn es kommt wirklich nicht darauf an, was im Indogermanischen gesprochen worden ist, sondern es kommt auf die Ablautsverhältnisse an.’

¹² H. Hendriksen, Untersuchungen über die Bedeutung des Hethitischen für die Laryngaitheorie, Det Kgl. Danske Videnskabernes Selskab, Historisk-filologiske Meddelelser 27.2 (Copenhagen, 1941). For a history of the theory see 4-11 (BHL). G. M. Messing, Selected Studies in Indo-European Phonology, Harvard Studies in Classical Philology 56-7. 161-232 (Cambridge, 1947) (SSP).