The University of Texas at Austin; College of Liberal Arts
Hans C. Boas, Director :: PCL 5.556, 1 University Station S5490 :: Austin, TX 78712 :: 512-471-4566
LRC Links: Home | About | Books Online | EIEOL | IE Doc. Center | IE Lexicon | IE Maps | IE Texts | Pub. Indices | SiteMap

Proto-Indo-European Phonology

Winfred P. Lehmann

2. The Phonological System of PIE

2.1. Criteria for describing the PIE phonological system

Indo-Europeanists agree in general on the description and distribution of the PIE sounds; these data are well-established. In formulating their phonemic systems from these data all Indo-Europeanists have used to a certain extent distribution of sounds as well as phonetic analysis. But up to the present they have relied more heavily on the latter, on phonetic rather than on structural criteria.

For a reconstructed language like PIE, whose dialects provide us with considerable phonological data, structural criteria can provide an adequate phonemic system; the description of the allophones of these phonemes, however, may be less exact than that of allophones of a language recorded from living speakers. Since we have no documents for PIE or for some of the IE dialects, such as PGmc., we have a better idea of the structural relationships of IE phonemes than of their phonetic relationships. For example, all Indo-Europeanists assume a series of PIE obstruents, usually written bh dh gh, that contrast with p t k and b d g. But they differ in their phonetic interpretation of this series; some assume that they were stops, others that they were spirants. From a structural point of view the phonetic interpretation is of secondary importance.

The following advantages are apparent in a system constructed with the aid of functional criteria.

2.1a. The criterion used here for determining the class to which a phoneme belongs is its use as syllabic. We then arrive at three classes of phonemes: 1. those which may not function as syllabics will be called obstruents; 2. those which may function only as syllabics, vowels; 3. those which function both as syllabics and as non-syllabics, resonants.

The following phonemic system may then be assumed for PIE, apart from laryngeals or reflexes of laryngeals:1

Obstruents:     p       t       k        
    b   bh   d   dh   g   gh     gʷh
Vowels:   e ē   a ā   o ō   e   ī   ū
Resonants:   y   w   l   r   m   n

2.1b. This system differs from Brugmann's system primarily in not including schwa, diphthongs, aspirated voiceless stops, or palatal stops. The unaccented vowels are central to the laryngeal theory, and will be treated later. The assumed diphthongs, and the resonants, will be discussed in this chapter. The problem of aspirated voiceless stops will be discussed in chapter 11. The problem of the back stops in PIE will not be discussed in detail since it is not central to this book. Brugmann of course assumes three series, palatals, velars, and labiovelars. Any attempt to assume two series, as well as to account for developments in all dialects from three PIE series, is confronted with numerous exceptions. One must accept the most plausible system, and assume it was disturbed by analogical changes. I accept Meillet's. (Cf. Introduction 91-5.) Meillet, pointing out that the velars are found especially before a, r, at the end of roots especially after u, and after s, assumes that in other positions, that is in the neighborhood of sounds with fronted articulation, the PIE velars were palatalized in the satem dialects. Kurylowicz, EI 1-26, on the other hand, assumes palatals and velars for PIE, and further that labiovelars are innovations in the centum dialects, resulting from a falling together of palatals and velars plus with velars before front vowels. In view of Meillet's more weighty linguistic arguments, such as his demonstration that the satem dialects form a central, innovating area, and have continued to undergo palatalization, I find his hypothesis more convincing than that of Kurylowicz.

2.1c. In the stage of PIE for which this phonemic system is assumed accent was not phonemic. The system had free pitch accent like that in Vedic. Most words had a syllable with high pitch; exceptions are enclitics, and verb-forms in some positions of the sentence.

At an earlier stage of PIE this pitch accent had been phonemic; during the period with phonemic pitch accent occurred the sound change of [e] to [o].

This stage of PIE with phonemic pitch accent was preceded by a stage with stress accent. During an early stage of pre-IE, stress accent had been phonemic, and then occurred the sound change of [e] to [e] and []. These shifts in stress are assumed from the ablaut interchanges found in the various IE dialects. Hirt and other Indo-Europeanists assumed them; yet they never indicated that accent, both stress and pitch, might be part of a linguistic system, but be non-phonemic; as a result the apparent vagaries of IE, the shifts from stress accent to pitch accent and back to stress accent, seemed embarassing in their waywardness. Today we find no difficulty in accepting a language with stress and pitch accent, of which either, or neither, may be phonemic. Presumably the various stages of IE had both pitch and stress accent simultaneously; in one period stress accent was phonemic, in another pitch accent, in another neither. These earlier stages of IE will be discussed below. The system assumed here is that of the stage of IE immediately before the division into dialects.

2.1d. Because of our lack of IE records it is difficult to determine the allophones of the various phonemes. From our evidence we have been able to deduce little allophonic variation of the obstruents other than /s/, and in position of articulation of the resonants other than /n/. /s/ had a voiced allophone when it stood before voiced obstruents, Av. miždəm, Goth. mizdō ‘reward’, a voiceless allophone elsewhere, Av. asti, Goth. ist ‘is’. /n/ had palatal and velar allophones when it stood before palatals and velars, Lat. quīnque, Skt. páñca ‘five’, a dental allophone elsewhere, Lat. ferunt, Skt. bháranti ‘they bear’.

None of the IE dialects contains evidence to contradict the assumption of a voiced allophone of /s/. Other, less general, developments in the IE dialects, especially of PIE clusters, may also give evidence of PIE allophonic variation. Thus, apparently an, excrescent spirant developed in the combination, dental stop plus dental stop; e.g. Skt. vittá, Av. vistō ‘known’, Gk. ἄιστος ‘unseen’, OIrish ro-fess ‘known’, OHG giwiss ‘certain’ from PIE /wyttos/ [witstos]. In some dialects the excrescent spirant fell into the /s/ phoneme, in others it remained subphonemic.

2.1e. There are further general combinatory changes of obstruents in PIE. Voiceless obstruents became voiced before voiced obstruents; voiced obstruents became voiceless before voiceless obstruents. Before the Skt. instrumental pl. ending -bhis, the final voiceless stop of vāc- ‘voice’ is voiced, vāgbhís; before the locative pl. ending -su, the final voiced stop of pā̆d- ‘foot’ is unvoiced, patsú.

Aspiration was a feature of an obstruent phoneme or of a cluster of obstruent phonemes. If one of a cluster of obstruent phonemes was an aspirate, the entire cluster was aspirated, and the last obstruent of the cluster was the member of the cluster marked with aspiration. Moreover any such cluster became voiced. The Skt. instrumental pl. of yudh- ‘fighter’ is yudbhís; the to-participle of labh- ‘seize’ is labdha.

2.1f. Other combinatory or cluster phenomena have not been so accurately described. Another combination giving evidence for an excrescent spirant is that of dentals and velars, e.g. Gk. χθών ‘earth’, Skt. kṣám, Hitt. te-kán. But the circumstances under which such spirantization takes place have not yet been determined; for not all combinations of dental and velar show such spirantization.2 Nor have the other possible PIE clusters been amply studied. When aberrant developments were found in some dialects, especially in Ind.-Ir. and Gk., the methodological principle adopted by former Indo-Europeanists was that of assuming another PIE phoneme. Because of the difficulties in words like Skt. kṣám, four new phonemes were assumed. To account for Gk. ζυγόν ‘yoke’ as opposed to ὅς ‘who’, another spirant phoneme was assumed. At least fourteen, and as many as thirty-six diphthongal phonemes were assumed for PIE. Some of these phonemes, e.g. the spirant /j/, are no longer generally accepted. But all Indo-Europeanists assume a series of diphthongal phonemes; and the four spirants /þ þh ð ðh/ are also generally assumed.

The PIE vowels give us little evidence to assume wide allophonic variation, with the possible exception of /o/, although with so few vowel phonemes there was latitude for such variation. Further study of vowels and obstruents may lead us to assume more allophonic variation, which rather than more phonemes may account for some of the aberrant developments in the dialects.

The resonants, however, show marked variation; it is here that the greatest unsolved problems have until recently remained in IE historical phonology.

2.2. The allophonic variation of the resonants; no diphthongs in PIE

Edgerton has analyzed the allophonic variation of the resonants and found a consistent pattern of positional variants for them. Choice of positional variant in PIE was determined by preceding and following phoneme, group of phonemes, or pause.3 Each resonant phoneme is composed of three allophones, one consonantal, one vocalic, and one vowel followed by consonant, e.g. /w/ [w u uw], /r/ [r r̥ r̥r], /n/ [n n̥ n̥n], etc. When resonants are found between vowels, their allophone is consonantal, e.g. [bherō] ‘I bear’. Between consonants, it is vocalic, e.g. [bhr̥to-] ‘borne’. Between consonant and vowel the allophone is consonantal if the preceding syllable is short, e.g. RV tuvi-grá < PIE [-gʷro-], and vowel followed by consonant if the preceding syllable is long, AV girati < PIE [gʷr̥reti]. The allophonic patterns are more complex when resonants are contiguous; Edgerton has worked out the following patterns. (The symbol t stands for any obstruent, y for the consonantal form of any resonant, i for the vocalic form of any resonant, k for a second obstruent, w and u for a second resonant, a for any vowel, ă for any short vowel, ā for any long vowel, and | for pause.)

one resonant   tit | it ti |        
    aya | ya ay |        
    ătya but ktiya, ātiya     | tiya
two resonants   ăywa but āyuwa    
    ătyut ătyu |   but ātiyut   ātiyu |
            ktiyut   ktiyu |
            | tiyut   | tiyu |
    | yut            
    ayut ayu |          
    | tiwa ātiwa   ktiwa      
either   yuwa or possibly iwa; pattern uncertain
three resonants   ăywit ăywi |   ăywiya      

Of the conclusions about the phonological interrelationships of the PIE phonemes which may be drawn from these patterns, one of interest here is that | (pause) is equal to a consonant; when, however, it precedes a resonant, it is equal to a consonant preceded by a vowel.

2.2a. Although the allophones of the resonants have now been determined, the relation of the resonants to the PIE vowels has not been clearly analyzed. When standing between vowels and obstruents, and occasionally when standing between vowels and vowels, they are designated as the second element of diphthongs. Thus for PIE is assumed a series of diphthongs composed of the vowels e a o ē ā ō, and sometimes ə, followed by the resonants.

It is difficult to find discussions supporting such an analysis. More common are statements like the following of Schrijnen:4 ‘with the discovery of the IE vowels went that of the IE diphthongs.’ He then assumes diphthongs with i and u, e.g. ei ai oi əi eu au ou əu. Meillet objects to such a restriction, and assumes diphthongs for y w r l m n with e a o, setting up thirty-six diphthongs for PIE.

Yet no evidence has been adduced to support such a fundamental assumption in drawing up the PIE phonemic system. For the reasons given below I assume that the PIE phonemic system did not include diphthongs at all.

2.2b. The word ‘diphthong’ is used from two points of view, the phonetic and the phonemic. Jones defines it phonetically as ‘an independent vowel glide not containing either a “peak” or a “valley” of prominence.’5 Our methods of arriving at phonetic descriptions of PIE phonemes are deductive; we can only determine them in general. Still we expect such forms to follow the same patterns as forms determined from written texts, or from oral communication.

Diphthongs usually undergo a different development from that of their components. None of the six MHG diphthongs developed in the same way as its components; ei developed to Mod. Germ. ai, ie to ī, but e and i remained; ou developed to au, uo to u, but o and u remained ; öu developed to eu, üe to ü, but ö, e and u remained. If PIE had a series of diphthongs we should expect at least some of them to undergo a development different from that of their components. But in almost all of the various IE dialects the components of the so called diphthongs develop to the same phonemes as do the components in other environments; e.g. PIE /a/ and /o/ became PGmc. /a/, PIE [i] became PGmc. /i/, PIE [ai] and [oil became PGmc. /ai/; PIE /e a o/ became Skt. /a/, PIE /n/ became Skt. /n/, PIE /en an on/ became Skt. /an/, etc. If the two phonemes had combined in PIE to form a diphthong such parallel patterns of development would be remarkable.

We also expect of diphthongal phonemes that they remain in various phonetic environments; e.g. the German diphthong /ai/ is found before consonant, Bein ‘leg’, before vowel, Eier ‘eggs’, and finally, Ei ‘egg’. The so-called PIE diphthongs did not remain constant in various environments; e.g. Skt. jáyāmi ‘I win’, jeṣyā́mi ‘I shall win’; agnáyas ‘fires’, agnés ‘of fire’; śátravas ‘enemies’, śátros ‘of the enemy’.

2.2c. But, since PIE is somewhat difficult to deal with phonetically, it is necessary to analyze the ai, au, etc. combinations phonemically. Phonemically we might define a diphthong as a combination of two (or more) phonemes, of which one is usually vocalic, the other(s) semi-vocalic; such combinations contrast minimally with unit syllabic phonemes.

One of our sources of evidence for determining the phonological contrasts of PIE syllabics is ablaut. Ablaut changes are variations of vowel phonemes in one morpheme which are found in all of the IE dialects and therefore are of IE origin. In any given morpheme we find an interchange of vowels that is parallel with vowel interchange in other morphemes; /bher-/ compounded with /-tó-/ becomes /bhrtó-/, /ghew-/ compounded with /-tó-/ becomes /ghwt-ó/, etc. By determining from the forms found in several dialects such variation in a given word, we can establish PIE phonemic contrasts of the time when ablaut first developed. Wide-spread ablaut patterns are illustrated by:

/bheydh-/   /bhoydh-/   /bhydh-/
/bhewg-/   /bhowg-/   /bhwg-/
/terp-/   /torp-/   /trp-/
/gʷem-/   /gʷom-/   /gʷm-/

From such patterns we note that the PIE phonemic system had contrasts between /y/ : /ey/, /y/ : /oy/, /ey/ : /oy/, etc. It might appear from such, apparently minimal, contrasts that /ey/ and /oy/ are phonemic units of the same order as /y/ [i], and therefore diphthongs. But if we use ablaut variations as a criterion to determine phonemes, we must also take into account ablauting forms in which the resonant precedes the vowel. Although these are not so widespread as the above-listed patterns, we find evidence for such patterns as:

/yek-/   /yok-/   /yk-/
/wek-/   /wok-/   /wk-/
/trep-/   /trop-/   /trp-/

We can draw two possible conclusions about the phonemic status of /ey/, /oy/, etc. Either we must assume 72 diphthongal phonemes; for /ye/ etc. like /ey/ etc. contrasts with /y/ [i]. Or we must construct our PIE phonemic system without diphthongal phonemes. Few Indo-Europeanists go to the extreme of assuming 72 diphthongs; Kent, who hesitantly does (S of L 33), suggests that combinations like /ye/ must ‘in a wider sense’ be called diphthongs.

In my opinion the IE ablaut changes provide evidence against the assumption of diphthongal phonemes. If a diphthongal phoneme /ew/ is assumed, phonemic analysis of PIE roots like /dhews-/ ‘spray, scatter like dust’ is difficult; for from it are made forms like Skt. dhvaṅsáyati ‘scatter’ (from PIE /dhwos-/) as well as forms like MHG tœ̄sen ‘scatter’ (from PIE /dhows-/). If we assumed that PIE [eu̯] was a diphthong, we would have to assume two roots. If we assume that PIE [eu̯] was a cluster of /e/ plus /w/, the variation in the PIE morpheme /dhews-/ and in other morphemes with so-called Schwebeablaut is no problem.

2.2d. Moreover, the question whether the combinations of vowel plus resonant were phonemic units might be tested by their historical developments. PIE /ghew-/ ‘pour’ is a so-called diphthongal root. The Skt. reduplicated present juhóti shows regular development of /ew/, as does the Gk. aorist ἔχευα. The Gk. present, however, is χέω. In χέω the /w/ was treated as an intervocalic /w/, e.g. βίος ‘life’ from PIE gʷiwos. If we assume an IE diphthongal phoneme for /ghew-/, we must assume it only before consonants; that is to say, PIE [eu̯t] varied with PIE [ewet]. Such problems are avoided if one assumes PIE clusters, e.g. /ew/ rather than diphthongal phonemes, e.g. /eu̯/. PIE /ew ow en on/ in morphemes such as /ghew-/ are parallel to PIE /ew ow en on/ in other contexts. The /en/ and /on/ of the PIE root /men-/ is of the same phonemic order as the /en : on/ in the PIE word for ‘knee’. I conclude that a PIE diphthongal phoneme /on/ is not present in the etyma of Gk. γόνυ ‘knee’ or μέμονα ‘I remember’.

It is noteworthy for setting up the relative chronology of the development of diphthongs in PIE and its dialects that the Gk. pre-vocalic eu in ἔχευ[σ is maintained while the PIE pre-vocalic ew in /ghewō/ is modified. In Gk. eu was a diphthong; it was maintained in forms like ἔχευα, where it had been ‘protected’ by s, even after -s- was lost.

2.2e. Sanskrit provides another illustration. Although PIE /ey/ and /oy/ fall together in Skt. e, we know from comparison with other dialects, and absence of secondary palatalization in Skt., that the perfect singular originally contained an o-vowel. We should expect e from /oy/ in the perfect 3d sg. of roots with a supposed PIE /ey/ diphthong, as we find it in Skt. bibhéda, from PIE /bheyd-/. The perfect made from the Skt. reflex of PIE /key/, however, contrasts markedly with bibhéda. If /oy/ had been a PIE diphthong, we should expect Skt. *cikéa; the form, however, is cikā́ya. The /o/ here has developed as elsewhere before w y l m n r in open syllables where diphthongs have never been assumed, e.g. Gk. γόνυ, Skt. jā́nu ‘knee’, etc. The same development may be noted in the following 3d sg. perf. forms: dudrā́va, susrā́va, babhā́ra, etc. PIE /oy/, /ow/, /or/, etc. cannot be considered phonemic units when we find them parallel in development with /-o-y-/, /-o-w-/, /-o-r-/, etc.

I conclude that PIE had no diphthongs, but rather clusters of vowel and resonant, of resonant and vowel. Although this conclusion could not be derived from arguments based on the resulting simplification of the PIE ablaut system, it can be tested by its effect on analysis of ablaut and other PIE phonological phenomena.

2.3. The structure of the PIE root, and the origin of the ablaut variations

If we assume that the phonemic system obtaining during the development of ablaut contained no diphthongs, we find that ablaut has its origin in two sound changes. Only one vowel was affected by these changes; this vowel may be written e. e was found in various phonemic environments. It was preceded and followed by obstruents, e.g. /pet-/ ‘fly’; it was preceded by an obstruent and followed by a resonant, e.g. /ghew-/ ‘pour’; it was preceded and followed by a resonant, e.g. /wey-/ ‘weave’, etc. Many of these environments were such that later combinatory changes obscured the original general sound changes. But on the whole the ablaut changes are obviously parallel regardless of environment. In spite of the form of the root we find /e/ in the indicative present singular forms of athematic verbs, /o/ in the perfect singular, etc. Before a given suffix we find a consistent treatment of roots : before /-tó-/ we find -, before /-os-/ we find /e/, before /-éyo-/ we find /o/, etc.

If we assume clusters of vowel plus resonant rather than diphthongs, we find complete parallelism between:

y-clusters   /bheydh-/   /bhoydh-/   /bhydh-/
    /yeg-/   /yog-/   /yg-/
w-clusters   /ghew-/   /ghow-/   /ghw-/
    /wekʷ-/   /wokʷ-/   /wkʷ-/
r-clusters   /bher-/   /bhor-/   /bhr-/
    /prek-/   /prok-/   /prk-/
as well as l-clusters, m-clusters, n-clusters, and roots that did not contain resonants, such as /pet-/, /pot-/, /pt-/.

The phonemic relation between these clusters which originally were quite parallel was later obscured by phonetic changes. If we had no forms from related dialects we might refuse to admit an original parallelism between Skt. bandh-, bandh-, badh- and Gk. πενθ-, πονθ-, παθ-, Skt. bodh-, bodh-, budh- and Goth. biud-, bauþ,-, bud-. Just so we should find it difficult to recognize the parallelism of modern German biegen, bog, gebogen and binden, band, gebunden if we did not have forms from other Gmc. dialects, such as Gothic biugan, baug, bugans and bindan, band, bundans.

2.3a. The ablaut of the roots given above is the result of the two following sound changes:

  1. /e/ when unaccented (not having chief stress accent) became reduced or was lost; the loss was accompanied by lengthening of the preceding vowel if this vowel was accented. The two processes may be written in the following formulae:
    • 1.a. /e/ unaccented became /e/ or -;
    • 1.b. /e/ under accent remained; in the pattern /éte/ > /ét-/ > /ḗt/, /e/ was lengthened.
  2. /e/ (/ē/) became /o/ (/ō/) under pitch accent, if /e/ (/ē/) lost the chief pitch accent, and received a secondary accent.

I conclude that in both changes the pattern of sound change is so consistent because the vowel affected had not entered into diphthongal combinations with any surrounding resonants.

2.3b. This conclusion is supported by the ablaut of roots with long vowels, or PIE /a/. The pattern of change here resembles the pattern of change in roots in /e/. For example, in forms where we expect /e/, we see for /dhē-/ a PIE /ē/, Gk. pres. τίθημι, Skt. dá-dhā-mi; where we espect /e/ or /-/, we find a PIE reduced vowel or zero, Gk. θετός, Skt. hitá, Skt. dadhmás. In the verb ‘give’ we find similar patterns: Gk. δίδωμι, Skt. dá-dā-mi; Skt. á-di-ta, Gk. ἔδοτο; Skt. devá-tta. For ablaut in a root with PIE /a/, see Gk. αἴθω ‘burn’, ἰθαρός ‘pure’. From the complete parallelism with other roots, it has been assumed that ‘original’ long vowels resulted from contraction of /e/ with laryngeal, most ‘original’ /a/ vowels from /e/ modified by a preceding laryngeal. (Cf. Chapter 3.) Such roots had undergone the same ablaut changes as had roots in which /e/ is preserved. If we reconstruct the roots as they were before the sound changes occurred which obscured the original structure, we assume for the above roots /dheʔ- deγ- Aed-/. These are parallel with such roots as /dher- dem- wed-/. That all PIE ablaut relations fall under the two formulations given above is illustrated in the following chart.

      1. a. 1. b. 2.
    /e/ /e/   /-/ /ē/ /o/
e   Goth. sitan nists sētum sat
    ‘sit’ ‘nest’ ‘we sat’ ‘he sat’
ey   Goth. in-weitan Gk. ἴδμεν Lith. véidas Gk. οἰ̑δα
    ‘pay homage’ ‘we know’ ‘face’ ‘I know’
ew   Goth. kiusan kusum   káusjan
    ‘choose’ ‘we chose’   ‘taste’
er   Gk. δέρκομαι ἔδρακον   δέδορκα
    ‘I see’ ‘I saw’   ‘I saw’
el   OIrish melim mlīth d.s.   MLG mol
    ‘grind’ ‘grinding’   ‘dust’
em   Lat. semel Gk. (σ)μία   Gk. ὁμός
    ‘once’ ‘one’   ‘same’
en   Gk. πένθος ἔπαθον   πέπονθα
    ‘grief’ ‘suffered’   ‘suffered’
  Skt. dádhāmi hitá, dadhmás   OHG tuon
    ‘I place’ ‘put’, ‘we place’   ‘do’
eA   Gk. ἴστᾱμι στατός    
    ‘stand’ ‘standing’    
  Gk. δω̑ρον δοτός, Skt. (devá-)tta    
    ‘gift’ ‘given’ ‘god-given’    
Ae   Gk. ἀπολαύω Lat. lucrum Dor. Gk. λᾱία  
    ‘enjoy’ ‘gain’ ‘booty’  
re   Lat. precēs Skt. pr̥ccháti Goth. frēhum Lat. procus
    ‘prayer’ ‘demand’ ‘we asked’ ‘wooer’
-ter-   Gk. πατέρα πατρός πατήρ ἀπάτορα
suffix   ‘father’ ‘of a father’ ‘father’ ‘fatherless’

2.3c. Various attempts have been made to check the validity of these phonetic changes by reconstructing the PIE forms in which they occurred. The greatest difficulty with such attempts has been the lack of forms which can be reconstructed from our attested forms for that stage of IE in which the ablaut changes occurred. PIE /bhérety/ can be assumed on the basis of Skt. bhárati ‘carries’, Gk. φέρει, Lat. fert, Goth. baíriþ, etc. But this form must be subsequent to the period of the IE ablaut changes. For the first ablaut change would have permitted only one /e/; in the second ablaut change, one /e/ would probably have become [o] and later /o/. Because early forms are lacking we must rely on forms derived on the basis of our reconstructions and linguistic analysis.

2.3d. Two types of structures have been posited for the time of the ablaut changes, bases and roots.

Hirt dealt primarily with bases. For him bases were the simplest possible PIE accented words. Since Hirt considered nouns older elements of the language than verbs, he constructed bases on the pattern of nouns. His monosyllabic heavy bases were constructed on the pattern of root nouns, e.g., dhē- ‘place’, Skt. apa-dhā́ ‘shutting up’, the bisyllabic heavy bases on the pattern of the ā-nouns like Gk. σκιᾱ́ ‘shade’. Assuming that the structure of these words was already established at the time of the ablaut changes, Hirt constructed with their help his ablaut theories, and assumed similar forms even when there is no evidence in any of the IE dialects for the base in question. (IG 2.103-7.)

Obviously there is only theoretical evidence for most of Hirt's bases. Nonetheless Hirt considered ‘bases’ less theoretical than the IE roots with which most Indo-Europeanists deal. Yet ‘roots’ as well as ‘bases’ may be useful in analyzing earlier stages of PIE.

2.3e. The usefulness of the term ‘root’ is somewhat reduced because various linguists hold various theories of the IE root. In general, the term ‘root’ is used for that part of IE words to which are added suffixes, determinatives, prefixes and endings. Roots then are morphemes found, with phonetic variation, throughout the forms of one paradigm (a paradigm in word-inflection and word-formation).

With varying assumptions varying roots will be posited. If, as often, roots are arrived at by subtraction, if they are assumed to be those portions of the IE word which are left after prefixes, suffixes, determinatives, and endings are removed, one finds roots very diversified in form; some are simple, e.g. wep- (WP 1.256), others complex, e.g. spyēw- (WP 2.683). Even Indo-Europeanists who like Walde assume such diversified roots assume various limitations for the structure of IE roots.

  1. A root cannot consist of a voiceless stop and a voiced aspirated stop; /bhet-/ and /tebh-/ are impossible.
  2. A root cannot begin and end with a voiced stop; /beg-/ is impossible.
  3. A root cannot contain two successive resonants; /teyw-/ is impossible.

If there seems to be such a root, e.g. /moyn-/ in Lat. commūnis < commoinis, the root must be /moy-/ and /n/ must be part of the suffix. (Introduction 157.)

Continuing such attempts to discover a definite structure for the IE root, Indo-Europeanists have delimited still further the shapes possible for roots. The theory that accounts most fully for the evidence found in Hittite and IE records is that of Benveniste.

2.3f. Benveniste assumes that all verbal roots, when accented, are composed of three phonemes, e.g. /teg ter wer Aer dheX XeX/.

Roots may be modified by means of suffixes, determinatives (which Benveniste calls enlargements) or infixes. Suffixes are distinguished from determinatives in having ablauting forms; determinatives have fixed and consonantic forms. But any modifications must be made in well-defined patterns. The accented root may receive only determinatives, not suffixes; when suffixes are added, the root is not accented. All such extended forms are referred to as bases.

Every verbal root may then have two extended forms:

I       II    
/pér-k-/   Lith. peršù ‘woo’   /pr-ék-/   Lat. precor ‘ask’
/tér-A-/   Hitt. tar-aḫ-zi ‘controls’   /tr-éA-/   Lat. intrāre ‘enter’.

When verbal, form I may undergo no further expansion; form II may be further expanded by means of a determinative. From the root /pel-/ a form I /pél-ʔ-/ and a form II /pl-éʔ-/ may be made; form II may be further expanded, as in /pl-eʔ-dh-/ Gk. πλήθω ‘fill’, but /pel-ʔ-dh-/ is impossible.

Roots expanded with a nasal infix are like form II plus determinative. Thus Skt. vr̥nóti, from /wr-n-éw-/, (form I /wér-w-/ form II /wr-éw-/) may be compared with Gk. πλήθω.

The theory then consists of five statements. (Orig. 170-1.)

  1. The IE root is monosyllabic, composed of the fundamental vowel /e/ between two different consonants.
  2. In this constant scheme: consonant plus vowel plus consonant, the consonants can be of any order, provided they are different; but the presence of both a voiceless stop and an aspirated voiced stop is impossible.
  3. From the root are made two ablauting forms by means of suffixes:
    • form I, root in full grade and accented, suffix zero;
    • form II, root zero, suffix in full grade and accented.
  4. To the suffix can be added one determinative, either after the suffix of form II, or, if /n/, inserted between the root element and suffix of form II.6
  5. Further addition of determinatives or suffixes points to a nominal base.

These then were the forms of the IE roots and bases at the time of the quantitative ablaut changes. Aberrant verb forms are either from nominal forms, e.g. Skt. tudáti, or they are secondary, e.g. Goth. háitan. One of the uses of Benveniste's theory is as a check for distinguishing between original and secondary ablaut changes. For after the phonetic changes of ablaut became phonemic and distinguished morphological categories, the contrasts were extended widely.

2.3g. To obtain a clear picture of PIE ablaut one must deal with such abstractions rather than with the reconstructions arrived at in the comparison of dialect forms. For the changes which produced ablaut, originally simple, in the early dialects resulted in the complex interchange of vowels which the handbooks call ablaut changes. Similar complication of a transparent interchange is found in the dialects; the Gmc. dialects offer a convenient illustration of how such patterns may be obscured by later developments. The ablaut relationships of the first five verb classes are transparent in our Gothic records of fifth century material. But our MHG material of the twelfth century presents a system in which these relationships have been obscured by phonetic changes.

I. Goth. steigan
‘climb’ Inf.
3d sg pret.
1st pl. pret.
pret. part.
  MHG stīgen steig stigen gestigen
II.   giutan
gáut gutum gutans
    giezen gōz guzen gegozen
III.   finþan
fanþ funþum funþans
    finden fand funden gefunden
IV.   baíran
bar bērum baúrans
    bërn bar bāren geboren
V.   giban
gaf gēbum gibans
    gëben gap gāben gegëben

If we did not have a body of material that gives us information about sound changes of the intervening centuries, we should only with difficulty be able to deduce the simple ablaut patterns of PGmc.

2.3h. Unfortunately we have for PIE no such body of material, and the ablaut changes have of necessity been arrived at deductively and slowly. Discussions of ablaut are still often obscured by a failure to distinguish between the PIE changes and later developments of the sounds produced in these changes. The most far-reaching error of the kind is the assumption of a ‘reduced grade’ beside ‘zero grade.’ It is quite clear that the so-called reduced grade is found only in restricted phonetic environments, that it correlates either with the position of the accent or with the phonetic surroundings. For syllables either directly following or preceding the main stress accent Hirt assumes zero grade, for syllables farther from the accent, the reduced grade. (IG 2.192-9.) Thus one of the important categories of reduced grade alternates with zero grade, depending on the accent. Another of the categories for reduced grade is the phonetic environment in which the resonants have the allophone, vowel plus resonant; Hirt calls this a translation of Sievers' Law into his terms. (IG 2.198.) Consequently reduced grade is a conditioned variant of zero grade, and from a phonemic point of view we need to deal with only one unaccented ablaut grade.

2.3i. Besides the effect of such sound changes the parallelism in ablaut was obscured even more by a development in the phonological system between PIE and the dialects, which we do not find between PGmc. and MHG, that is, a coalescence of phonological classes. PIE had at least three classes of phonemes, obstruents, resonants, vowels; the dialects only two, obstruents and vowels. Phonemic contrasts in the vocalic system when the vowels were distinct from the resonants and obstruents became quite confused when some allophones of the resonants coalesced with the vowels, others with the obstruents. In PIE a contrast between /pet ghew dher/ /pt ghw dhr/ had existed; the dialects show contrasts between /pet gheu dher/ /pt ghu dhr/. When the ablaut system first developed, /e/ could contrast only with /ē/; in the dialects /e/ has at least ten minimal contrasts.

The ablaut patterns then support the contention that there were no diphthongal phonemes in PIE.

2.4. The general principle of PIE metrics

A second result of adopting a PIE phonemic system without diphthongs is its clarification of IE metrics. In both Greek and Vedic poetry poetic rhythm is based on syllabic quantity, not patterning of accented and unaccented syllables. Both systems of prosody share so many essential features that we must derive them from an earlier common system. (Introduction 143-4.)

The verse of the Homeric poems consists of six feet, each divided into two parts, the accented and unaccented part. The accented part consists of a metrically long syllable, the unaccented part of one long syllable or two short. The unaccented part of the sixth foot, however, may be long or short.

Long syllables are of two types, those long by nature and those long by position. Syllables long by nature contain a long vowel or diphthong; long final vowels or diphthongs standing before a vowel, however, are short metrically. Syllables long by position contain a short vowel followed by two or more consonants.7

While the essential element in Vedic poetry is regularity in the number of syllables, the quantity of syllables also plays a role, if not as important as in the Gk. hexameter. The Vedic hymns show a great variety of poetic structure but usually have iambic rhythm. Prosodic rules similar to those cited above for Homeric verse are: 1. a vowel becomes long by position if followed by two consonants, that is, a syllable is long if it contains a short vowel followed by two consonants; 2. a vowel (long) is shortened before another, e.g. Skt. e from PIE [ai ei oi] and o from PIE [au eu ou] are short before a, that is, a syllable is short if it contains a long vowel followed by another vowel.

In both systems syllables with long vowel or diphthong, or with short vowel followed by consonant, are long. But if diphthongs stand before a vowel, they are short; and long vowels standing before other vowels are metrically short. These rules we may then assume for IE verse.

With the current phonemic analysis of PIE, the rules of prosody are a series of unrelated formulae. When, however, we assume that PIE ‘diphthongs’ and ‘long vowels’ were clusters of vowel and resonants or laryngeals, the general principle of IE metrics becomes obvious. CLOSED SYLLABLES WERE METRICALLY LONG, OPEN SYLLABLES WERE METRICALLY SHORT. Examples of metrically long syllables in PIE would then be /tet-t . . . tey-t . . . teX-t . . . tet|/; types of metrically short syllables would be /te-t . . . te-y . . . te-X . . . /.

With this principle it is apparent why the sixth unaccented syllable of Homeric lines could be short as well as long. For it is clear from Edgerton's studies that final pause fulfills the same function in PIE syllabification as does a consonant. (See 2.2.) Therefore every final syllable of the line was closed, and consequently metrically long.

2.5. The phonemic system of PIE and its use in further study

For the reasons given above diphthongal phonemes will not be admitted in the phonemic system used as basis for further study. The system is made up of twenty-eight segmental phonemes, which are of three orders, obstruents, resonants, and vowels, plus some as yet undiscussed segmental phonemes which are laryngeals or reflexes of laryngeals. The generally accepted allophones of these phonemes, especially in clusters, have been given above. Other allophonic variation and the allophones of the laryngeals may become apparent in the course of this investigation. Some IE phonological problems will now be examined with the help of the system. Such problems may be solved with it, and at the same time a more complete system developed.

Before these problems are examined, certain questions that have arisen in connection with the laryngeal theory must be reviewed. In chapter three the laryngeal theory will be discussed and the evidence for laryngeals in Hittite and the IE dialects will be examined. The points of agreement, and disagreement between various forms of the theory will be listed, and a preliminary answer to some of the problems raised by it will be presented.


1 The standard handbooks, e.g. Meillet's Introduction 82-126, give examples of these phonemes in various environments.

The laryngeals will be discussed in chapter 3. Discussions of the laryngeal theory vary widely in the symbols used to represent laryngeals; Sturtevant lists the various symbols, IHL 22-3. Each system of transcription has its drawbacks, as does the current system of transcription used for PIE. I use, with modifications, the symbols found most widely in American publications. As cover symbol for laryngeals I use X. As cover symbol for the a-colored laryngeals I use A. For the laryngeals I use four symbols: ʔ h x γ. h is the only one of these different from the symbols in IHL; Sturtevant there uses ʔ, but in later publications h.

IE forms are sometimes discussed below in terms of their structure, and cover symbols are used for other classes of phonemes: C for obstruent, R for resonant, e for vowel. CeRX thus refers to a structure consisting of obstruent, vowel, resonant, laryngeal. When further analysis is necessary, members of the various classes of phonemes will be used, see for example, 2.2.

2 Benveniste examines the words for which such spirants are assumed in Le problème du þ indo-européen, BSL 38.139-47 (1937), and offers no solution. For that proposed by L. L. Hammerich, Laryngeal Before Sonant, Det Kgl. Danske Videnskabernes Selskab, Historisk-filologiske Meddelelser 31.3 (Copenhagen, 1948) there is no evidence.

3 The Indo-European Semivowels, Language 19.83-124 (1943); on pages 108-9 he summarizes his findings in formulae similar to those given here.

4 J. Schrijnen, Einführung in das Studium der Indogermanischen Sprachwissenschaft, translated into German by W. Fischer, 251 (Heidelberg, 1921): ‘Mit der Entdeckung der indogermanischen Vokale ging die der indogermanischen Diphthonge Hand in Hand.’

5 D. Jones, An Outline of English Phonetics6 57 (New York, 1940).

6 Orig. p. 171 Benveniste concludes: ‘Au suffixe peut se joindre un seul élargissement, soit ajouté après le suffixe du thème I, soit inséré entre l'élément radical et le suffixe du thème II (infixation).’ In discussing the formation of roots on p. 153 he had said however: ‘De là se déduit ce principe qu'un thème à l'état I n'admet pas d'élargissement: seul l'état II en comporte.’ Since the statement on p. 153 is italicized, and is supported by the subsequent discussion, I follow it rather than the apparently contradictory statement on p. 171.

7 For exceptions see LaRoche, HU 1ff. Munro, Horn. Gram. 338ff., gives a concise description of Homeric prosody, Macdonnell, A Vedic Grammar for Students 436-7, of Vedic prosody.